projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: change hot_remove_disk to take an rdev rather than a number.
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
2a48fc0a 39#include <linux/mutex.h>
1da177e4 40#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 41#include <linux/poll.h>
16f17b39 42#include <linux/ctype.h>
e7d2860b 43#include <linux/string.h>
fb4d8c76
N		 44#include <linux/hdreg.h>
45#include <linux/proc_fs.h>
46#include <linux/random.h>
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 */
5a7bbad2 336static void 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;
49077326 340 int cpu;
e91ece55 341 unsigned int sectors;
49077326 342
0ca69886
N		 343 if (mddev == NULL || mddev->pers == NULL
344 || !mddev->ready) {
409c57f3 345 bio_io_error(bio);
5a7bbad2 346 return;
409c57f3 347 }
0ca69886 348 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 349 rcu_read_lock();
e9c7469b 350 if (mddev->suspended) {
409c57f3
N		 351 DEFINE_WAIT(__wait);
352 for (;;) {
353 prepare_to_wait(&mddev->sb_wait, &__wait,
354 TASK_UNINTERRUPTIBLE);
e9c7469b 355 if (!mddev->suspended)
409c57f3
N		 356 break;
357 rcu_read_unlock();
358 schedule();
359 rcu_read_lock();
360 }
361 finish_wait(&mddev->sb_wait, &__wait);
362 }
363 atomic_inc(&mddev->active_io);
364 rcu_read_unlock();
49077326 365
e91ece55
CM		 366 /*
367 * save the sectors now since our bio can
368 * go away inside make_request
369 */
370 sectors = bio_sectors(bio);
5a7bbad2 371 mddev->pers->make_request(mddev, bio);
49077326
N		 372
373 cpu = part_stat_lock();
374 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 375 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N		 376 part_stat_unlock();
377
409c57f3
N		 378 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
379 wake_up(&mddev->sb_wait);
409c57f3
N		 380}
381
9e35b99c
N		 382/* mddev_suspend makes sure no new requests are submitted
383 * to the device, and that any requests that have been submitted
384 * are completely handled.
385 * Once ->stop is called and completes, the module will be completely
386 * unused.
387 */
fd01b88c 388void mddev_suspend(struct mddev *mddev)
409c57f3
N		 389{
390 BUG_ON(mddev->suspended);
391 mddev->suspended = 1;
392 synchronize_rcu();
393 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
394 mddev->pers->quiesce(mddev, 1);
409c57f3 395}
390ee602 396EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 397
fd01b88c 398void mddev_resume(struct mddev *mddev)
409c57f3
N		 399{
400 mddev->suspended = 0;
401 wake_up(&mddev->sb_wait);
402 mddev->pers->quiesce(mddev, 0);
0fd018af
JB		 403
404 md_wakeup_thread(mddev->thread);
405 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 406}
390ee602 407EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 408
fd01b88c 409int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N		 410{
411 return mddev->suspended;
412}
413EXPORT_SYMBOL(mddev_congested);
414
a2826aa9 415/*
e9c7469b 416 * Generic flush handling for md
a2826aa9
N		 417 */
418
e9c7469b 419static void md_end_flush(struct bio *bio, int err)
a2826aa9 420{
3cb03002 421 struct md_rdev *rdev = bio->bi_private;
fd01b88c 422 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 423
424 rdev_dec_pending(rdev, mddev);
425
426 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 427 /* The pre-request flush has finished */
e804ac78 428 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 429 }
430 bio_put(bio);
431}
432
a7a07e69
N		 433static void md_submit_flush_data(struct work_struct *ws);
434
a035fc3e 435static void submit_flushes(struct work_struct *ws)
a2826aa9 436{
fd01b88c 437 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 438 struct md_rdev *rdev;
a2826aa9 439
a7a07e69
N		 440 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
441 atomic_set(&mddev->flush_pending, 1);
a2826aa9
N		 442 rcu_read_lock();
443 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
444 if (rdev->raid_disk >= 0 &&
445 !test_bit(Faulty, &rdev->flags)) {
446 /* Take two references, one is dropped
447 * when request finishes, one after
448 * we reclaim rcu_read_lock
449 */
450 struct bio *bi;
451 atomic_inc(&rdev->nr_pending);
452 atomic_inc(&rdev->nr_pending);
453 rcu_read_unlock();
a167f663 454 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
e9c7469b 455 bi->bi_end_io = md_end_flush;
a2826aa9
N		 456 bi->bi_private = rdev;
457 bi->bi_bdev = rdev->bdev;
458 atomic_inc(&mddev->flush_pending);
e9c7469b 459 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 460 rcu_read_lock();
461 rdev_dec_pending(rdev, mddev);
462 }
463 rcu_read_unlock();
a7a07e69
N		 464 if (atomic_dec_and_test(&mddev->flush_pending))
465 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 466}
467
e9c7469b 468static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 469{
fd01b88c 470 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 471 struct bio *bio = mddev->flush_bio;
a2826aa9 472
e9c7469b 473 if (bio->bi_size == 0)
a2826aa9
N		 474 /* an empty barrier - all done */
475 bio_endio(bio, 0);
476 else {
e9c7469b 477 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 478 mddev->pers->make_request(mddev, bio);
a2826aa9 479 }
2b74e12e
N		 480
481 mddev->flush_bio = NULL;
482 wake_up(&mddev->sb_wait);
a2826aa9
N		 483}
484
fd01b88c 485void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N		 486{
487 spin_lock_irq(&mddev->write_lock);
488 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 489 !mddev->flush_bio,
a2826aa9 490 mddev->write_lock, /*nothing*/);
e9c7469b 491 mddev->flush_bio = bio;
a2826aa9
N		 492 spin_unlock_irq(&mddev->write_lock);
493
a035fc3e
N		 494 INIT_WORK(&mddev->flush_work, submit_flushes);
495 queue_work(md_wq, &mddev->flush_work);
a2826aa9 496}
e9c7469b 497EXPORT_SYMBOL(md_flush_request);
409c57f3 498
97658cdd
N		 499/* Support for plugging.
500 * This mirrors the plugging support in request_queue, but does not
501 * require having a whole queue or request structures.
502 * We allocate an md_plug_cb for each md device and each thread it gets
503 * plugged on. This links tot the private plug_handle structure in the
504 * personality data where we keep a count of the number of outstanding
505 * plugs so other code can see if a plug is active.
506 */
507struct md_plug_cb {
508 struct blk_plug_cb cb;
fd01b88c 509 struct mddev *mddev;
97658cdd
N		 510};
511
512static void plugger_unplug(struct blk_plug_cb *cb)
513{
514 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
515 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
516 md_wakeup_thread(mdcb->mddev->thread);
517 kfree(mdcb);
518}
519
520/* Check that an unplug wakeup will come shortly.
521 * If not, wakeup the md thread immediately
522 */
fd01b88c 523int mddev_check_plugged(struct mddev *mddev)
97658cdd
N		 524{
525 struct blk_plug *plug = current->plug;
526 struct md_plug_cb *mdcb;
527
528 if (!plug)
529 return 0;
530
531 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
532 if (mdcb->cb.callback == plugger_unplug &&
533 mdcb->mddev == mddev) {
534 /* Already on the list, move to top */
535 if (mdcb != list_first_entry(&plug->cb_list,
536 struct md_plug_cb,
537 cb.list))
538 list_move(&mdcb->cb.list, &plug->cb_list);
539 return 1;
540 }
541 }
542 /* Not currently on the callback list */
543 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
544 if (!mdcb)
545 return 0;
546
547 mdcb->mddev = mddev;
548 mdcb->cb.callback = plugger_unplug;
549 atomic_inc(&mddev->plug_cnt);
550 list_add(&mdcb->cb.list, &plug->cb_list);
551 return 1;
552}
553EXPORT_SYMBOL_GPL(mddev_check_plugged);
2ac87401 554
fd01b88c 555static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 556{
557 atomic_inc(&mddev->active);
558 return mddev;
559}
560
5fd3a17e 561static void mddev_delayed_delete(struct work_struct *ws);
d3374825 562
fd01b88c 563static void mddev_put(struct mddev *mddev)
1da177e4 564{
a167f663
N		 565 struct bio_set *bs = NULL;
566
1da177e4
LT		 567 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
568 return;
d3374825 569 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 570 mddev->ctime == 0 && !mddev->hold_active) {
571 /* Array is not configured at all, and not held active,
572 * so destroy it */
af8a2434 573 list_del_init(&mddev->all_mddevs);
a167f663
N		 574 bs = mddev->bio_set;
575 mddev->bio_set = NULL;
d3374825 576 if (mddev->gendisk) {
e804ac78
TH		 577 /* We did a probe so need to clean up. Call
578 * queue_work inside the spinlock so that
579 * flush_workqueue() after mddev_find will
580 * succeed in waiting for the work to be done.
d3374825
N		 581 */
582 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 583 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 584 } else
585 kfree(mddev);
586 }
587 spin_unlock(&all_mddevs_lock);
a167f663
N		 588 if (bs)
589 bioset_free(bs);
1da177e4
LT		 590}
591
fd01b88c 592void mddev_init(struct mddev *mddev)
fafd7fb0
N		 593{
594 mutex_init(&mddev->open_mutex);
595 mutex_init(&mddev->reconfig_mutex);
596 mutex_init(&mddev->bitmap_info.mutex);
597 INIT_LIST_HEAD(&mddev->disks);
598 INIT_LIST_HEAD(&mddev->all_mddevs);
599 init_timer(&mddev->safemode_timer);
600 atomic_set(&mddev->active, 1);
601 atomic_set(&mddev->openers, 0);
602 atomic_set(&mddev->active_io, 0);
97658cdd 603 atomic_set(&mddev->plug_cnt, 0);
fafd7fb0
N		 604 spin_lock_init(&mddev->write_lock);
605 atomic_set(&mddev->flush_pending, 0);
606 init_waitqueue_head(&mddev->sb_wait);
607 init_waitqueue_head(&mddev->recovery_wait);
608 mddev->reshape_position = MaxSector;
609 mddev->resync_min = 0;
610 mddev->resync_max = MaxSector;
611 mddev->level = LEVEL_NONE;
612}
390ee602 613EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 614
fd01b88c 615static struct mddev * mddev_find(dev_t unit)
1da177e4 616{
fd01b88c 617 struct mddev *mddev, *new = NULL;
1da177e4 618
8f5f02c4
N		 619 if (unit && MAJOR(unit) != MD_MAJOR)
620 unit &= ~((1<<MdpMinorShift)-1);
621
1da177e4
LT		 622 retry:
623 spin_lock(&all_mddevs_lock);
efeb53c0
N		 624
625 if (unit) {
626 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
627 if (mddev->unit == unit) {
628 mddev_get(mddev);
629 spin_unlock(&all_mddevs_lock);
630 kfree(new);
631 return mddev;
632 }
633
634 if (new) {
635 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 636 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 637 new->hold_active = UNTIL_IOCTL;
638 return new;
1da177e4 639 }
efeb53c0
N		 640 } else if (new) {
641 /* find an unused unit number */
642 static int next_minor = 512;
643 int start = next_minor;
644 int is_free = 0;
645 int dev = 0;
646 while (!is_free) {
647 dev = MKDEV(MD_MAJOR, next_minor);
648 next_minor++;
649 if (next_minor > MINORMASK)
650 next_minor = 0;
651 if (next_minor == start) {
652 /* Oh dear, all in use. */
653 spin_unlock(&all_mddevs_lock);
654 kfree(new);
655 return NULL;
656 }
657
658 is_free = 1;
659 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
660 if (mddev->unit == dev) {
661 is_free = 0;
662 break;
663 }
664 }
665 new->unit = dev;
666 new->md_minor = MINOR(dev);
667 new->hold_active = UNTIL_STOP;
1da177e4
LT		 668 list_add(&new->all_mddevs, &all_mddevs);
669 spin_unlock(&all_mddevs_lock);
670 return new;
671 }
672 spin_unlock(&all_mddevs_lock);
673
9ffae0cf 674 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 675 if (!new)
676 return NULL;
677
1da177e4
LT		 678 new->unit = unit;
679 if (MAJOR(unit) == MD_MAJOR)
680 new->md_minor = MINOR(unit);
681 else
682 new->md_minor = MINOR(unit) >> MdpMinorShift;
683
fafd7fb0 684 mddev_init(new);
1da177e4 685
1da177e4
LT		 686 goto retry;
687}
688
fd01b88c 689static inline int mddev_lock(struct mddev * mddev)
1da177e4 690{
df5b89b3 691 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 692}
693
fd01b88c 694static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW		 695{
696 return mutex_is_locked(&mddev->reconfig_mutex);
697}
698
fd01b88c 699static inline int mddev_trylock(struct mddev * mddev)
1da177e4 700{
df5b89b3 701 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 702}
703
b6eb127d
N		 704static struct attribute_group md_redundancy_group;
705
fd01b88c 706static void mddev_unlock(struct mddev * mddev)
1da177e4 707{
a64c876f 708 if (mddev->to_remove) {
b6eb127d
N		 709 /* These cannot be removed under reconfig_mutex as
710 * an access to the files will try to take reconfig_mutex
711 * while holding the file unremovable, which leads to
712 * a deadlock.
bb4f1e9d
N		 713 * So hold set sysfs_active while the remove in happeing,
714 * and anything else which might set ->to_remove or my
715 * otherwise change the sysfs namespace will fail with
716 * -EBUSY if sysfs_active is still set.
717 * We set sysfs_active under reconfig_mutex and elsewhere
718 * test it under the same mutex to ensure its correct value
719 * is seen.
b6eb127d 720 */
a64c876f
N		 721 struct attribute_group *to_remove = mddev->to_remove;
722 mddev->to_remove = NULL;
bb4f1e9d 723 mddev->sysfs_active = 1;
b6eb127d
N		 724 mutex_unlock(&mddev->reconfig_mutex);
725
00bcb4ac
N		 726 if (mddev->kobj.sd) {
727 if (to_remove != &md_redundancy_group)
728 sysfs_remove_group(&mddev->kobj, to_remove);
729 if (mddev->pers == NULL ||
730 mddev->pers->sync_request == NULL) {
731 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
732 if (mddev->sysfs_action)
733 sysfs_put(mddev->sysfs_action);
734 mddev->sysfs_action = NULL;
735 }
a64c876f 736 }
bb4f1e9d 737 mddev->sysfs_active = 0;
b6eb127d
N		 738 } else
739 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 740
751e67ca
CD		 741 /* As we've dropped the mutex we need a spinlock to
742 * make sure the thread doesn't disappear
01f96c0a
N		 743 */
744 spin_lock(&pers_lock);
005eca5e 745 md_wakeup_thread(mddev->thread);
01f96c0a 746 spin_unlock(&pers_lock);
1da177e4
LT		 747}
748
fd01b88c 749static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 750{
3cb03002 751 struct md_rdev *rdev;
1da177e4 752
159ec1fc 753 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 754 if (rdev->desc_nr == nr)
755 return rdev;
159ec1fc 756
1da177e4
LT		 757 return NULL;
758}
759
fd01b88c 760static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
1da177e4 761{
3cb03002 762 struct md_rdev *rdev;
1da177e4 763
159ec1fc 764 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 765 if (rdev->bdev->bd_dev == dev)
766 return rdev;
159ec1fc 767
1da177e4
LT		 768 return NULL;
769}
770
84fc4b56 771static struct md_personality *find_pers(int level, char *clevel)
2604b703 772{
84fc4b56 773 struct md_personality *pers;
d9d166c2
N		 774 list_for_each_entry(pers, &pers_list, list) {
775 if (level != LEVEL_NONE && pers->level == level)
2604b703 776 return pers;
d9d166c2
N		 777 if (strcmp(pers->name, clevel)==0)
778 return pers;
779 }
2604b703
N		 780 return NULL;
781}
782
b73df2d3 783/* return the offset of the super block in 512byte sectors */
3cb03002 784static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 785{
57b2caa3 786 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 787 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 788}
789
3cb03002 790static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 791{
792 if (rdev->sb_page)
793 MD_BUG();
794
795 rdev->sb_page = alloc_page(GFP_KERNEL);
796 if (!rdev->sb_page) {
797 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 798 return -ENOMEM;
1da177e4
LT		 799 }
800
801 return 0;
802}
803
3cb03002 804static void free_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 805{
806 if (rdev->sb_page) {
2d1f3b5d 807 put_page(rdev->sb_page);
1da177e4
LT		 808 rdev->sb_loaded = 0;
809 rdev->sb_page = NULL;
0f420358 810 rdev->sb_start = 0;
dd8ac336 811 rdev->sectors = 0;
1da177e4 812 }
2699b672
N		 813 if (rdev->bb_page) {
814 put_page(rdev->bb_page);
815 rdev->bb_page = NULL;
816 }
1da177e4
LT		 817}
818
819
6712ecf8 820static void super_written(struct bio *bio, int error)
7bfa19f2 821{
3cb03002 822 struct md_rdev *rdev = bio->bi_private;
fd01b88c 823 struct mddev *mddev = rdev->mddev;
7bfa19f2 824
3a0f5bbb
N		 825 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
826 printk("md: super_written gets error=%d, uptodate=%d\n",
827 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
828 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 829 md_error(mddev, rdev);
3a0f5bbb 830 }
7bfa19f2 831
a9701a30
N		 832 if (atomic_dec_and_test(&mddev->pending_writes))
833 wake_up(&mddev->sb_wait);
f8b58edf 834 bio_put(bio);
7bfa19f2
N		 835}
836
fd01b88c 837void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 838 sector_t sector, int size, struct page *page)
839{
840 /* write first size bytes of page to sector of rdev
841 * Increment mddev->pending_writes before returning
842 * and decrement it on completion, waking up sb_wait
843 * if zero is reached.
844 * If an error occurred, call md_error
845 */
a167f663 846 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 847
a6ff7e08 848 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 849 bio->bi_sector = sector;
850 bio_add_page(bio, page, size, 0);
851 bio->bi_private = rdev;
852 bio->bi_end_io = super_written;
a9701a30 853
7bfa19f2 854 atomic_inc(&mddev->pending_writes);
a5bf4df0 855 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 856}
857
fd01b88c 858void md_super_wait(struct mddev *mddev)
a9701a30 859{
e9c7469b 860 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 861 DEFINE_WAIT(wq);
862 for(;;) {
863 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
864 if (atomic_read(&mddev->pending_writes)==0)
865 break;
a9701a30
N		 866 schedule();
867 }
868 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 869}
870
6712ecf8 871static void bi_complete(struct bio *bio, int error)
1da177e4 872{
1da177e4 873 complete((struct completion*)bio->bi_private);
1da177e4
LT		 874}
875
3cb03002 876int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 877 struct page *page, int rw, bool metadata_op)
1da177e4 878{
a167f663 879 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 880 struct completion event;
881 int ret;
882
721a9602 883 rw |= REQ_SYNC;
1da177e4 884
a6ff7e08
JB		 885 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
886 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 887 if (metadata_op)
888 bio->bi_sector = sector + rdev->sb_start;
889 else
890 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 891 bio_add_page(bio, page, size, 0);
892 init_completion(&event);
893 bio->bi_private = &event;
894 bio->bi_end_io = bi_complete;
895 submit_bio(rw, bio);
896 wait_for_completion(&event);
897
898 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
899 bio_put(bio);
900 return ret;
901}
a8745db2 902EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 903
3cb03002 904static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT		 905{
906 char b[BDEVNAME_SIZE];
907 if (!rdev->sb_page) {
908 MD_BUG();
909 return -EINVAL;
910 }
911 if (rdev->sb_loaded)
912 return 0;
913
914
ccebd4c4 915 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 916 goto fail;
917 rdev->sb_loaded = 1;
918 return 0;
919
920fail:
921 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
922 bdevname(rdev->bdev,b));
923 return -EINVAL;
924}
925
926static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
927{
05710466
AN		 928 return sb1->set_uuid0 == sb2->set_uuid0 &&
929 sb1->set_uuid1 == sb2->set_uuid1 &&
930 sb1->set_uuid2 == sb2->set_uuid2 &&
931 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 932}
933
1da177e4
LT		 934static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
935{
936 int ret;
937 mdp_super_t *tmp1, *tmp2;
938
939 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
940 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
941
942 if (!tmp1 || !tmp2) {
943 ret = 0;
35020f1a 944 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 945 goto abort;
946 }
947
948 *tmp1 = *sb1;
949 *tmp2 = *sb2;
950
951 /*
952 * nr_disks is not constant
953 */
954 tmp1->nr_disks = 0;
955 tmp2->nr_disks = 0;
956
ce0c8e05 957 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 958abort:
990a8baf
JJ		 959 kfree(tmp1);
960 kfree(tmp2);
1da177e4
LT		 961 return ret;
962}
963
4d167f09
N		 964
965static u32 md_csum_fold(u32 csum)
966{
967 csum = (csum & 0xffff) + (csum >> 16);
968 return (csum & 0xffff) + (csum >> 16);
969}
970
1da177e4
LT		 971static unsigned int calc_sb_csum(mdp_super_t * sb)
972{
4d167f09
N		 973 u64 newcsum = 0;
974 u32 *sb32 = (u32*)sb;
975 int i;
1da177e4
LT		 976 unsigned int disk_csum, csum;
977
978 disk_csum = sb->sb_csum;
979 sb->sb_csum = 0;
4d167f09
N		 980
981 for (i = 0; i < MD_SB_BYTES/4 ; i++)
982 newcsum += sb32[i];
983 csum = (newcsum & 0xffffffff) + (newcsum>>32);
984
985
986#ifdef CONFIG_ALPHA
987 /* This used to use csum_partial, which was wrong for several
988 * reasons including that different results are returned on
989 * different architectures. It isn't critical that we get exactly
990 * the same return value as before (we always csum_fold before
991 * testing, and that removes any differences). However as we
992 * know that csum_partial always returned a 16bit value on
993 * alphas, do a fold to maximise conformity to previous behaviour.
994 */
995 sb->sb_csum = md_csum_fold(disk_csum);
996#else
1da177e4 997 sb->sb_csum = disk_csum;
4d167f09 998#endif
1da177e4
LT		 999 return csum;
1000}
1001
1002
1003/*
1004 * Handle superblock details.
1005 * We want to be able to handle multiple superblock formats
1006 * so we have a common interface to them all, and an array of
1007 * different handlers.
1008 * We rely on user-space to write the initial superblock, and support
1009 * reading and updating of superblocks.
1010 * Interface methods are:
3cb03002 1011 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1012 * loads and validates a superblock on dev.
1013 * if refdev != NULL, compare superblocks on both devices
1014 * Return:
1015 * 0 - dev has a superblock that is compatible with refdev
1016 * 1 - dev has a superblock that is compatible and newer than refdev
1017 * so dev should be used as the refdev in future
1018 * -EINVAL superblock incompatible or invalid
1019 * -othererror e.g. -EIO
1020 *
fd01b88c 1021 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1022 * Verify that dev is acceptable into mddev.
1023 * The first time, mddev->raid_disks will be 0, and data from
1024 * dev should be merged in. Subsequent calls check that dev
1025 * is new enough. Return 0 or -EINVAL
1026 *
fd01b88c 1027 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1028 * Update the superblock for rdev with data in mddev
1029 * This does not write to disc.
1030 *
1031 */
1032
1033struct super_type {
0cd17fec
CW		 1034 char *name;
1035 struct module *owner;
3cb03002 1036 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
0cd17fec 1037 int minor_version);
fd01b88c
N		 1038 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1039 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
3cb03002 1040 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 1041 sector_t num_sectors);
1da177e4
LT		 1042};
1043
0894cc30
AN		 1044/*
1045 * Check that the given mddev has no bitmap.
1046 *
1047 * This function is called from the run method of all personalities that do not
1048 * support bitmaps. It prints an error message and returns non-zero if mddev
1049 * has a bitmap. Otherwise, it returns 0.
1050 *
1051 */
fd01b88c 1052int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1053{
c3d9714e 1054 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 1055 return 0;
1056 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1057 mdname(mddev), mddev->pers->name);
1058 return 1;
1059}
1060EXPORT_SYMBOL(md_check_no_bitmap);
1061
1da177e4
LT		 1062/*
1063 * load_super for 0.90.0
1064 */
3cb03002 1065static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1066{
1067 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1068 mdp_super_t *sb;
1069 int ret;
1da177e4
LT		 1070
1071 /*
0f420358 1072 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1073 * it's at the end of the disk.
1074 *
1075 * It also happens to be a multiple of 4Kb.
1076 */
57b2caa3 1077 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1078
0002b271 1079 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1080 if (ret) return ret;
1081
1082 ret = -EINVAL;
1083
1084 bdevname(rdev->bdev, b);
65a06f06 1085 sb = page_address(rdev->sb_page);
1da177e4
LT		 1086
1087 if (sb->md_magic != MD_SB_MAGIC) {
1088 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1089 b);
1090 goto abort;
1091 }
1092
1093 if (sb->major_version != 0 ||
f6705578
N		 1094 sb->minor_version < 90 ||
1095 sb->minor_version > 91) {
1da177e4
LT		 1096 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1097 sb->major_version, sb->minor_version,
1098 b);
1099 goto abort;
1100 }
1101
1102 if (sb->raid_disks <= 0)
1103 goto abort;
1104
4d167f09 1105 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1106 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1107 b);
1108 goto abort;
1109 }
1110
1111 rdev->preferred_minor = sb->md_minor;
1112 rdev->data_offset = 0;
0002b271 1113 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1114 rdev->badblocks.shift = -1;
1da177e4
LT		 1115
1116 if (sb->level == LEVEL_MULTIPATH)
1117 rdev->desc_nr = -1;
1118 else
1119 rdev->desc_nr = sb->this_disk.number;
1120
9a7b2b0f 1121 if (!refdev) {
1da177e4 1122 ret = 1;
9a7b2b0f 1123 } else {
1da177e4 1124 __u64 ev1, ev2;
65a06f06 1125 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1126 if (!uuid_equal(refsb, sb)) {
1127 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1128 b, bdevname(refdev->bdev,b2));
1129 goto abort;
1130 }
1131 if (!sb_equal(refsb, sb)) {
1132 printk(KERN_WARNING "md: %s has same UUID"
1133 " but different superblock to %s\n",
1134 b, bdevname(refdev->bdev, b2));
1135 goto abort;
1136 }
1137 ev1 = md_event(sb);
1138 ev2 = md_event(refsb);
1139 if (ev1 > ev2)
1140 ret = 1;
1141 else
1142 ret = 0;
1143 }
8190e754 1144 rdev->sectors = rdev->sb_start;
27a7b260
N		 1145 /* Limit to 4TB as metadata cannot record more than that */
1146 if (rdev->sectors >= (2ULL << 32))
1147 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1148
27a7b260 1149 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1150 /* "this cannot possibly happen" ... */
1151 ret = -EINVAL;
1152
1da177e4
LT		 1153 abort:
1154 return ret;
1155}
1156
1157/*
1158 * validate_super for 0.90.0
1159 */
fd01b88c 1160static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1161{
1162 mdp_disk_t *desc;
65a06f06 1163 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1164 __u64 ev1 = md_event(sb);
1da177e4 1165
41158c7e 1166 rdev->raid_disk = -1;
c5d79adb
N		 1167 clear_bit(Faulty, &rdev->flags);
1168 clear_bit(In_sync, &rdev->flags);
1169 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1170
1da177e4
LT		 1171 if (mddev->raid_disks == 0) {
1172 mddev->major_version = 0;
1173 mddev->minor_version = sb->minor_version;
1174 mddev->patch_version = sb->patch_version;
e691063a 1175 mddev->external = 0;
9d8f0363 1176 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1177 mddev->ctime = sb->ctime;
1178 mddev->utime = sb->utime;
1179 mddev->level = sb->level;
d9d166c2 1180 mddev->clevel[0] = 0;
1da177e4
LT		 1181 mddev->layout = sb->layout;
1182 mddev->raid_disks = sb->raid_disks;
27a7b260 1183 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1184 mddev->events = ev1;
c3d9714e
N		 1185 mddev->bitmap_info.offset = 0;
1186 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1187
f6705578
N		 1188 if (mddev->minor_version >= 91) {
1189 mddev->reshape_position = sb->reshape_position;
1190 mddev->delta_disks = sb->delta_disks;
1191 mddev->new_level = sb->new_level;
1192 mddev->new_layout = sb->new_layout;
664e7c41 1193 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1194 } else {
1195 mddev->reshape_position = MaxSector;
1196 mddev->delta_disks = 0;
1197 mddev->new_level = mddev->level;
1198 mddev->new_layout = mddev->layout;
664e7c41 1199 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1200 }
1201
1da177e4
LT		 1202 if (sb->state & (1<<MD_SB_CLEAN))
1203 mddev->recovery_cp = MaxSector;
1204 else {
1205 if (sb->events_hi == sb->cp_events_hi &&
1206 sb->events_lo == sb->cp_events_lo) {
1207 mddev->recovery_cp = sb->recovery_cp;
1208 } else
1209 mddev->recovery_cp = 0;
1210 }
1211
1212 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1213 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1214 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1215 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1216
1217 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1218
1219 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1220 mddev->bitmap_info.file == NULL)
1221 mddev->bitmap_info.offset =
1222 mddev->bitmap_info.default_offset;
a654b9d8 1223
41158c7e 1224 } else if (mddev->pers == NULL) {
be6800a7
N		 1225 /* Insist on good event counter while assembling, except
1226 * for spares (which don't need an event count) */
1da177e4 1227 ++ev1;
be6800a7
N		 1228 if (sb->disks[rdev->desc_nr].state & (
1229 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1230 if (ev1 < mddev->events)
1231 return -EINVAL;
41158c7e
N		 1232 } else if (mddev->bitmap) {
1233 /* if adding to array with a bitmap, then we can accept an
1234 * older device ... but not too old.
1235 */
41158c7e
N		 1236 if (ev1 < mddev->bitmap->events_cleared)
1237 return 0;
07d84d10
N		 1238 } else {
1239 if (ev1 < mddev->events)
1240 /* just a hot-add of a new device, leave raid_disk at -1 */
1241 return 0;
1242 }
41158c7e 1243
1da177e4 1244 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1245 desc = sb->disks + rdev->desc_nr;
1246
1247 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1248 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1249 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1250 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1251 set_bit(In_sync, &rdev->flags);
1da177e4 1252 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1253 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1254 /* active but not in sync implies recovery up to
1255 * reshape position. We don't know exactly where
1256 * that is, so set to zero for now */
1257 if (mddev->minor_version >= 91) {
1258 rdev->recovery_offset = 0;
1259 rdev->raid_disk = desc->raid_disk;
1260 }
1da177e4 1261 }
8ddf9efe
N		 1262 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1263 set_bit(WriteMostly, &rdev->flags);
41158c7e 1264 } else /* MULTIPATH are always insync */
b2d444d7 1265 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1266 return 0;
1267}
1268
1269/*
1270 * sync_super for 0.90.0
1271 */
fd01b88c 1272static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1273{
1274 mdp_super_t *sb;
3cb03002 1275 struct md_rdev *rdev2;
1da177e4 1276 int next_spare = mddev->raid_disks;
19133a42 1277
1da177e4
LT		 1278
1279 /* make rdev->sb match mddev data..
1280 *
1281 * 1/ zero out disks
1282 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1283 * 3/ any empty disks < next_spare become removed
1284 *
1285 * disks[0] gets initialised to REMOVED because
1286 * we cannot be sure from other fields if it has
1287 * been initialised or not.
1288 */
1289 int i;
1290 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1291
61181565
N		 1292 rdev->sb_size = MD_SB_BYTES;
1293
65a06f06 1294 sb = page_address(rdev->sb_page);
1da177e4
LT		 1295
1296 memset(sb, 0, sizeof(*sb));
1297
1298 sb->md_magic = MD_SB_MAGIC;
1299 sb->major_version = mddev->major_version;
1da177e4
LT		 1300 sb->patch_version = mddev->patch_version;
1301 sb->gvalid_words = 0; /* ignored */
1302 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1303 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1304 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1305 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1306
1307 sb->ctime = mddev->ctime;
1308 sb->level = mddev->level;
58c0fed4 1309 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1310 sb->raid_disks = mddev->raid_disks;
1311 sb->md_minor = mddev->md_minor;
e691063a 1312 sb->not_persistent = 0;
1da177e4
LT		 1313 sb->utime = mddev->utime;
1314 sb->state = 0;
1315 sb->events_hi = (mddev->events>>32);
1316 sb->events_lo = (u32)mddev->events;
1317
f6705578
N		 1318 if (mddev->reshape_position == MaxSector)
1319 sb->minor_version = 90;
1320 else {
1321 sb->minor_version = 91;
1322 sb->reshape_position = mddev->reshape_position;
1323 sb->new_level = mddev->new_level;
1324 sb->delta_disks = mddev->delta_disks;
1325 sb->new_layout = mddev->new_layout;
664e7c41 1326 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1327 }
1328 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1329 if (mddev->in_sync)
1330 {
1331 sb->recovery_cp = mddev->recovery_cp;
1332 sb->cp_events_hi = (mddev->events>>32);
1333 sb->cp_events_lo = (u32)mddev->events;
1334 if (mddev->recovery_cp == MaxSector)
1335 sb->state = (1<< MD_SB_CLEAN);
1336 } else
1337 sb->recovery_cp = 0;
1338
1339 sb->layout = mddev->layout;
9d8f0363 1340 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1341
c3d9714e 1342 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1343 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1344
1da177e4 1345 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1346 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1347 mdp_disk_t *d;
86e6ffdd 1348 int desc_nr;
0261cd9f
N		 1349 int is_active = test_bit(In_sync, &rdev2->flags);
1350
1351 if (rdev2->raid_disk >= 0 &&
1352 sb->minor_version >= 91)
1353 /* we have nowhere to store the recovery_offset,
1354 * but if it is not below the reshape_position,
1355 * we can piggy-back on that.
1356 */
1357 is_active = 1;
1358 if (rdev2->raid_disk < 0 ||
1359 test_bit(Faulty, &rdev2->flags))
1360 is_active = 0;
1361 if (is_active)
86e6ffdd 1362 desc_nr = rdev2->raid_disk;
1da177e4 1363 else
86e6ffdd 1364 desc_nr = next_spare++;
19133a42 1365 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1366 d = &sb->disks[rdev2->desc_nr];
1367 nr_disks++;
1368 d->number = rdev2->desc_nr;
1369 d->major = MAJOR(rdev2->bdev->bd_dev);
1370 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1371 if (is_active)
1da177e4
LT		 1372 d->raid_disk = rdev2->raid_disk;
1373 else
1374 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1375 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1376 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1377 else if (is_active) {
1da177e4 1378 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1379 if (test_bit(In_sync, &rdev2->flags))
1380 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1381 active++;
1382 working++;
1383 } else {
1384 d->state = 0;
1385 spare++;
1386 working++;
1387 }
8ddf9efe
N		 1388 if (test_bit(WriteMostly, &rdev2->flags))
1389 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1390 }
1da177e4
LT		 1391 /* now set the "removed" and "faulty" bits on any missing devices */
1392 for (i=0 ; i < mddev->raid_disks ; i++) {
1393 mdp_disk_t *d = &sb->disks[i];
1394 if (d->state == 0 && d->number == 0) {
1395 d->number = i;
1396 d->raid_disk = i;
1397 d->state = (1<<MD_DISK_REMOVED);
1398 d->state |= (1<<MD_DISK_FAULTY);
1399 failed++;
1400 }
1401 }
1402 sb->nr_disks = nr_disks;
1403 sb->active_disks = active;
1404 sb->working_disks = working;
1405 sb->failed_disks = failed;
1406 sb->spare_disks = spare;
1407
1408 sb->this_disk = sb->disks[rdev->desc_nr];
1409 sb->sb_csum = calc_sb_csum(sb);
1410}
1411
0cd17fec
CW		 1412/*
1413 * rdev_size_change for 0.90.0
1414 */
1415static unsigned long long
3cb03002 1416super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1417{
58c0fed4 1418 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1419 return 0; /* component must fit device */
c3d9714e 1420 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1421 return 0; /* can't move bitmap */
57b2caa3 1422 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1423 if (!num_sectors || num_sectors > rdev->sb_start)
1424 num_sectors = rdev->sb_start;
27a7b260
N		 1425 /* Limit to 4TB as metadata cannot record more than that.
1426 * 4TB == 2^32 KB, or 2*2^32 sectors.
1427 */
1428 if (num_sectors >= (2ULL << 32))
1429 num_sectors = (2ULL << 32) - 2;
0f420358 1430 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1431 rdev->sb_page);
1432 md_super_wait(rdev->mddev);
c26a44ed 1433 return num_sectors;
0cd17fec
CW		 1434}
1435
1436
1da177e4
LT		 1437/*
1438 * version 1 superblock
1439 */
1440
1c05b4bc 1441static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1442{
1c05b4bc
N		 1443 __le32 disk_csum;
1444 u32 csum;
1da177e4
LT		 1445 unsigned long long newcsum;
1446 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1447 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1448 int i;
1449
1450 disk_csum = sb->sb_csum;
1451 sb->sb_csum = 0;
1452 newcsum = 0;
1453 for (i=0; size>=4; size -= 4 )
1454 newcsum += le32_to_cpu(*isuper++);
1455
1456 if (size == 2)
1c05b4bc 1457 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1458
1459 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1460 sb->sb_csum = disk_csum;
1461 return cpu_to_le32(csum);
1462}
1463
2699b672
N		 1464static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1465 int acknowledged);
3cb03002 1466static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1467{
1468 struct mdp_superblock_1 *sb;
1469 int ret;
0f420358 1470 sector_t sb_start;
1da177e4 1471 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1472 int bmask;
1da177e4
LT		 1473
1474 /*
0f420358 1475 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1476 * It is always aligned to a 4K boundary and
1477 * depeding on minor_version, it can be:
1478 * 0: At least 8K, but less than 12K, from end of device
1479 * 1: At start of device
1480 * 2: 4K from start of device.
1481 */
1482 switch(minor_version) {
1483 case 0:
77304d2a 1484 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1485 sb_start -= 8*2;
1486 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1487 break;
1488 case 1:
0f420358 1489 sb_start = 0;
1da177e4
LT		 1490 break;
1491 case 2:
0f420358 1492 sb_start = 8;
1da177e4
LT		 1493 break;
1494 default:
1495 return -EINVAL;
1496 }
0f420358 1497 rdev->sb_start = sb_start;
1da177e4 1498
0002b271
N		 1499 /* superblock is rarely larger than 1K, but it can be larger,
1500 * and it is safe to read 4k, so we do that
1501 */
1502 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1503 if (ret) return ret;
1504
1505
65a06f06 1506 sb = page_address(rdev->sb_page);
1da177e4
LT		 1507
1508 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1509 sb->major_version != cpu_to_le32(1) ||
1510 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1511 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1512 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1513 return -EINVAL;
1514
1515 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1516 printk("md: invalid superblock checksum on %s\n",
1517 bdevname(rdev->bdev,b));
1518 return -EINVAL;
1519 }
1520 if (le64_to_cpu(sb->data_size) < 10) {
1521 printk("md: data_size too small on %s\n",
1522 bdevname(rdev->bdev,b));
1523 return -EINVAL;
1524 }
e11e93fa 1525
1da177e4
LT		 1526 rdev->preferred_minor = 0xffff;
1527 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1528 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1529
0002b271 1530 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1531 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1532 if (rdev->sb_size & bmask)
a1801f85
N		 1533 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1534
1535 if (minor_version
0f420358 1536 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1537 return -EINVAL;
0002b271 1538
31b65a0d
N		 1539 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1540 rdev->desc_nr = -1;
1541 else
1542 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1543
2699b672
N		 1544 if (!rdev->bb_page) {
1545 rdev->bb_page = alloc_page(GFP_KERNEL);
1546 if (!rdev->bb_page)
1547 return -ENOMEM;
1548 }
1549 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1550 rdev->badblocks.count == 0) {
1551 /* need to load the bad block list.
1552 * Currently we limit it to one page.
1553 */
1554 s32 offset;
1555 sector_t bb_sector;
1556 u64 *bbp;
1557 int i;
1558 int sectors = le16_to_cpu(sb->bblog_size);
1559 if (sectors > (PAGE_SIZE / 512))
1560 return -EINVAL;
1561 offset = le32_to_cpu(sb->bblog_offset);
1562 if (offset == 0)
1563 return -EINVAL;
1564 bb_sector = (long long)offset;
1565 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1566 rdev->bb_page, READ, true))
1567 return -EIO;
1568 bbp = (u64 *)page_address(rdev->bb_page);
1569 rdev->badblocks.shift = sb->bblog_shift;
1570 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1571 u64 bb = le64_to_cpu(*bbp);
1572 int count = bb & (0x3ff);
1573 u64 sector = bb >> 10;
1574 sector <<= sb->bblog_shift;
1575 count <<= sb->bblog_shift;
1576 if (bb + 1 == 0)
1577 break;
1578 if (md_set_badblocks(&rdev->badblocks,
1579 sector, count, 1) == 0)
1580 return -EINVAL;
1581 }
1582 } else if (sb->bblog_offset == 0)
1583 rdev->badblocks.shift = -1;
1584
9a7b2b0f 1585 if (!refdev) {
8ed75463 1586 ret = 1;
9a7b2b0f 1587 } else {
1da177e4 1588 __u64 ev1, ev2;
65a06f06 1589 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1590
1591 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1592 sb->level != refsb->level ||
1593 sb->layout != refsb->layout ||
1594 sb->chunksize != refsb->chunksize) {
1595 printk(KERN_WARNING "md: %s has strangely different"
1596 " superblock to %s\n",
1597 bdevname(rdev->bdev,b),
1598 bdevname(refdev->bdev,b2));
1599 return -EINVAL;
1600 }
1601 ev1 = le64_to_cpu(sb->events);
1602 ev2 = le64_to_cpu(refsb->events);
1603
1604 if (ev1 > ev2)
8ed75463
N		 1605 ret = 1;
1606 else
1607 ret = 0;
1da177e4 1608 }
a1801f85 1609 if (minor_version)
77304d2a 1610 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 1611 le64_to_cpu(sb->data_offset);
1da177e4 1612 else
dd8ac336
AN		 1613 rdev->sectors = rdev->sb_start;
1614 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1615 return -EINVAL;
dd8ac336 1616 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1617 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1618 return -EINVAL;
8ed75463 1619 return ret;
1da177e4
LT		 1620}
1621
fd01b88c 1622static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1623{
65a06f06 1624 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1625 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1626
41158c7e 1627 rdev->raid_disk = -1;
c5d79adb
N		 1628 clear_bit(Faulty, &rdev->flags);
1629 clear_bit(In_sync, &rdev->flags);
1630 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1631
1da177e4
LT		 1632 if (mddev->raid_disks == 0) {
1633 mddev->major_version = 1;
1634 mddev->patch_version = 0;
e691063a 1635 mddev->external = 0;
9d8f0363 1636 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1637 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1638 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1639 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1640 mddev->clevel[0] = 0;
1da177e4
LT		 1641 mddev->layout = le32_to_cpu(sb->layout);
1642 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1643 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1644 mddev->events = ev1;
c3d9714e
N		 1645 mddev->bitmap_info.offset = 0;
1646 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1647
1648 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1649 memcpy(mddev->uuid, sb->set_uuid, 16);
1650
1651 mddev->max_disks = (4096-256)/2;
a654b9d8 1652
71c0805c 1653 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1654 mddev->bitmap_info.file == NULL )
1655 mddev->bitmap_info.offset =
1656 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1657
f6705578
N		 1658 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1659 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1660 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1661 mddev->new_level = le32_to_cpu(sb->new_level);
1662 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1663 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1664 } else {
1665 mddev->reshape_position = MaxSector;
1666 mddev->delta_disks = 0;
1667 mddev->new_level = mddev->level;
1668 mddev->new_layout = mddev->layout;
664e7c41 1669 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1670 }
1671
41158c7e 1672 } else if (mddev->pers == NULL) {
be6800a7
N		 1673 /* Insist of good event counter while assembling, except for
1674 * spares (which don't need an event count) */
1da177e4 1675 ++ev1;
be6800a7
N		 1676 if (rdev->desc_nr >= 0 &&
1677 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1678 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1679 if (ev1 < mddev->events)
1680 return -EINVAL;
41158c7e
N		 1681 } else if (mddev->bitmap) {
1682 /* If adding to array with a bitmap, then we can accept an
1683 * older device, but not too old.
1684 */
41158c7e
N		 1685 if (ev1 < mddev->bitmap->events_cleared)
1686 return 0;
07d84d10
N		 1687 } else {
1688 if (ev1 < mddev->events)
1689 /* just a hot-add of a new device, leave raid_disk at -1 */
1690 return 0;
1691 }
1da177e4
LT		 1692 if (mddev->level != LEVEL_MULTIPATH) {
1693 int role;
3673f305
N		 1694 if (rdev->desc_nr < 0 ||
1695 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1696 role = 0xffff;
1697 rdev->desc_nr = -1;
1698 } else
1699 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1700 switch(role) {
1701 case 0xffff: /* spare */
1da177e4
LT		 1702 break;
1703 case 0xfffe: /* faulty */
b2d444d7 1704 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1705 break;
1706 default:
5fd6c1dc
N		 1707 if ((le32_to_cpu(sb->feature_map) &
1708 MD_FEATURE_RECOVERY_OFFSET))
1709 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1710 else
1711 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1712 rdev->raid_disk = role;
1713 break;
1714 }
8ddf9efe
N		 1715 if (sb->devflags & WriteMostly1)
1716 set_bit(WriteMostly, &rdev->flags);
41158c7e 1717 } else /* MULTIPATH are always insync */
b2d444d7 1718 set_bit(In_sync, &rdev->flags);
41158c7e 1719
1da177e4
LT		 1720 return 0;
1721}
1722
fd01b88c 1723static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1724{
1725 struct mdp_superblock_1 *sb;
3cb03002 1726 struct md_rdev *rdev2;
1da177e4
LT		 1727 int max_dev, i;
1728 /* make rdev->sb match mddev and rdev data. */
1729
65a06f06 1730 sb = page_address(rdev->sb_page);
1da177e4
LT		 1731
1732 sb->feature_map = 0;
1733 sb->pad0 = 0;
5fd6c1dc 1734 sb->recovery_offset = cpu_to_le64(0);
1da177e4 1735 memset(sb->pad1, 0, sizeof(sb->pad1));
1da177e4
LT		 1736 memset(sb->pad3, 0, sizeof(sb->pad3));
1737
1738 sb->utime = cpu_to_le64((__u64)mddev->utime);
1739 sb->events = cpu_to_le64(mddev->events);
1740 if (mddev->in_sync)
1741 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1742 else
1743 sb->resync_offset = cpu_to_le64(0);
1744
1c05b4bc 1745 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1746
f0ca340c 1747 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1748 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1749 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1750 sb->level = cpu_to_le32(mddev->level);
1751 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1752
aeb9b211
N		 1753 if (test_bit(WriteMostly, &rdev->flags))
1754 sb->devflags |= WriteMostly1;
1755 else
1756 sb->devflags &= ~WriteMostly1;
1757
c3d9714e
N		 1758 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1759 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1760 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1761 }
5fd6c1dc
N		 1762
1763 if (rdev->raid_disk >= 0 &&
97e4f42d 1764 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1765 sb->feature_map |=
1766 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1767 sb->recovery_offset =
1768 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc
N		 1769 }
1770
f6705578
N		 1771 if (mddev->reshape_position != MaxSector) {
1772 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1773 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1774 sb->new_layout = cpu_to_le32(mddev->new_layout);
1775 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1776 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1777 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1778 }
a654b9d8 1779
2699b672
N		 1780 if (rdev->badblocks.count == 0)
1781 /* Nothing to do for bad blocks*/ ;
1782 else if (sb->bblog_offset == 0)
1783 /* Cannot record bad blocks on this device */
1784 md_error(mddev, rdev);
1785 else {
1786 struct badblocks *bb = &rdev->badblocks;
1787 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1788 u64 *p = bb->page;
1789 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1790 if (bb->changed) {
1791 unsigned seq;
1792
1793retry:
1794 seq = read_seqbegin(&bb->lock);
1795
1796 memset(bbp, 0xff, PAGE_SIZE);
1797
1798 for (i = 0 ; i < bb->count ; i++) {
1799 u64 internal_bb = *p++;
1800 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1801 | BB_LEN(internal_bb));
1802 *bbp++ = cpu_to_le64(store_bb);
1803 }
1804 if (read_seqretry(&bb->lock, seq))
1805 goto retry;
1806
1807 bb->sector = (rdev->sb_start +
1808 (int)le32_to_cpu(sb->bblog_offset));
1809 bb->size = le16_to_cpu(sb->bblog_size);
1810 bb->changed = 0;
1811 }
1812 }
1813
1da177e4 1814 max_dev = 0;
159ec1fc 1815 list_for_each_entry(rdev2, &mddev->disks, same_set)
1da177e4
LT		 1816 if (rdev2->desc_nr+1 > max_dev)
1817 max_dev = rdev2->desc_nr+1;
a778b73f 1818
70471daf
N		 1819 if (max_dev > le32_to_cpu(sb->max_dev)) {
1820 int bmask;
a778b73f 1821 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1822 rdev->sb_size = max_dev * 2 + 256;
1823 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1824 if (rdev->sb_size & bmask)
1825 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1826 } else
1827 max_dev = le32_to_cpu(sb->max_dev);
1828
1da177e4
LT		 1829 for (i=0; i<max_dev;i++)
1830 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1831
159ec1fc 1832 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1833 i = rdev2->desc_nr;
b2d444d7 1834 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1835 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1836 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1837 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1838 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1839 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1840 else
1841 sb->dev_roles[i] = cpu_to_le16(0xffff);
1842 }
1843
1da177e4
LT		 1844 sb->sb_csum = calc_sb_1_csum(sb);
1845}
1846
0cd17fec 1847static unsigned long long
3cb03002 1848super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1849{
1850 struct mdp_superblock_1 *sb;
15f4a5fd 1851 sector_t max_sectors;
58c0fed4 1852 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1853 return 0; /* component must fit device */
0f420358 1854 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1855 /* minor versions 1 and 2; superblock before data */
77304d2a 1856 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1857 max_sectors -= rdev->data_offset;
1858 if (!num_sectors || num_sectors > max_sectors)
1859 num_sectors = max_sectors;
c3d9714e 1860 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1861 /* minor version 0 with bitmap we can't move */
1862 return 0;
1863 } else {
1864 /* minor version 0; superblock after data */
0f420358 1865 sector_t sb_start;
77304d2a 1866 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1867 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1868 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1869 if (!num_sectors || num_sectors > max_sectors)
1870 num_sectors = max_sectors;
0f420358 1871 rdev->sb_start = sb_start;
0cd17fec 1872 }
65a06f06 1873 sb = page_address(rdev->sb_page);
15f4a5fd 1874 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1875 sb->super_offset = rdev->sb_start;
0cd17fec 1876 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1877 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1878 rdev->sb_page);
1879 md_super_wait(rdev->mddev);
c26a44ed 1880 return num_sectors;
0cd17fec 1881}
1da177e4 1882
75c96f85 1883static struct super_type super_types[] = {
1da177e4
LT		 1884 [0] = {
1885 .name = "0.90.0",
1886 .owner = THIS_MODULE,
0cd17fec
CW		 1887 .load_super = super_90_load,
1888 .validate_super = super_90_validate,
1889 .sync_super = super_90_sync,
1890 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1891 },
1892 [1] = {
1893 .name = "md-1",
1894 .owner = THIS_MODULE,
0cd17fec
CW		 1895 .load_super = super_1_load,
1896 .validate_super = super_1_validate,
1897 .sync_super = super_1_sync,
1898 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1899 },
1900};
1da177e4 1901
fd01b88c 1902static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1903{
1904 if (mddev->sync_super) {
1905 mddev->sync_super(mddev, rdev);
1906 return;
1907 }
1908
1909 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1910
1911 super_types[mddev->major_version].sync_super(mddev, rdev);
1912}
1913
fd01b88c 1914static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1915{
3cb03002 1916 struct md_rdev *rdev, *rdev2;
1da177e4 1917
4b80991c
N		 1918 rcu_read_lock();
1919 rdev_for_each_rcu(rdev, mddev1)
1920 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1921 if (rdev->bdev->bd_contains ==
4b80991c
N		 1922 rdev2->bdev->bd_contains) {
1923 rcu_read_unlock();
7dd5e7c3 1924 return 1;
4b80991c
N		 1925 }
1926 rcu_read_unlock();
1da177e4
LT		 1927 return 0;
1928}
1929
1930static LIST_HEAD(pending_raid_disks);
1931
ac5e7113
AN		 1932/*
1933 * Try to register data integrity profile for an mddev
1934 *
1935 * This is called when an array is started and after a disk has been kicked
1936 * from the array. It only succeeds if all working and active component devices
1937 * are integrity capable with matching profiles.
1938 */
fd01b88c 1939int md_integrity_register(struct mddev *mddev)
ac5e7113 1940{
3cb03002 1941 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1942
1943 if (list_empty(&mddev->disks))
1944 return 0; /* nothing to do */
629acb6a
JB		 1945 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1946 return 0; /* shouldn't register, or already is */
ac5e7113
AN		 1947 list_for_each_entry(rdev, &mddev->disks, same_set) {
1948 /* skip spares and non-functional disks */
1949 if (test_bit(Faulty, &rdev->flags))
1950 continue;
1951 if (rdev->raid_disk < 0)
1952 continue;
ac5e7113
AN		 1953 if (!reference) {
1954 /* Use the first rdev as the reference */
1955 reference = rdev;
1956 continue;
1957 }
1958 /* does this rdev's profile match the reference profile? */
1959 if (blk_integrity_compare(reference->bdev->bd_disk,
1960 rdev->bdev->bd_disk) < 0)
1961 return -EINVAL;
1962 }
89078d57
MP		 1963 if (!reference || !bdev_get_integrity(reference->bdev))
1964 return 0;
ac5e7113
AN		 1965 /*
1966 * All component devices are integrity capable and have matching
1967 * profiles, register the common profile for the md device.
1968 */
1969 if (blk_integrity_register(mddev->gendisk,
1970 bdev_get_integrity(reference->bdev)) != 0) {
1971 printk(KERN_ERR "md: failed to register integrity for %s\n",
1972 mdname(mddev));
1973 return -EINVAL;
1974 }
a91a2785
MP		 1975 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1976 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1977 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1978 mdname(mddev));
1979 return -EINVAL;
1980 }
ac5e7113
AN		 1981 return 0;
1982}
1983EXPORT_SYMBOL(md_integrity_register);
1984
1985/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 1986void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 1987{
3f9d99c1 1988 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1989 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1990
ac5e7113 1991 if (!bi_mddev) /* nothing to do */
3f9d99c1 1992 return;
ac5e7113 1993 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1994 return;
ac5e7113
AN		 1995 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1996 rdev->bdev->bd_disk) >= 0)
1997 return;
1998 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1999 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2000}
ac5e7113 2001EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2002
fd01b88c 2003static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
1da177e4 2004{
7dd5e7c3 2005 char b[BDEVNAME_SIZE];
f637b9f9 2006 struct kobject *ko;
1edf80d3 2007 char *s;
5e55e2f5 2008 int err;
1da177e4
LT		 2009
2010 if (rdev->mddev) {
2011 MD_BUG();
2012 return -EINVAL;
2013 }
11e2ede0
DW		 2014
2015 /* prevent duplicates */
2016 if (find_rdev(mddev, rdev->bdev->bd_dev))
2017 return -EEXIST;
2018
dd8ac336
AN		 2019 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2020 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2021 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2022 if (mddev->pers) {
2023 /* Cannot change size, so fail
2024 * If mddev->level <= 0, then we don't care
2025 * about aligning sizes (e.g. linear)
2026 */
2027 if (mddev->level > 0)
2028 return -ENOSPC;
2029 } else
dd8ac336 2030 mddev->dev_sectors = rdev->sectors;
2bf071bf 2031 }
1da177e4
LT		 2032
2033 /* Verify rdev->desc_nr is unique.
2034 * If it is -1, assign a free number, else
2035 * check number is not in use
2036 */
2037 if (rdev->desc_nr < 0) {
2038 int choice = 0;
2039 if (mddev->pers) choice = mddev->raid_disks;
2040 while (find_rdev_nr(mddev, choice))
2041 choice++;
2042 rdev->desc_nr = choice;
2043 } else {
2044 if (find_rdev_nr(mddev, rdev->desc_nr))
2045 return -EBUSY;
2046 }
de01dfad
N		 2047 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2048 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2049 mdname(mddev), mddev->max_disks);
2050 return -EBUSY;
2051 }
19133a42 2052 bdevname(rdev->bdev,b);
649316b2 2053 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2054 *s = '!';
649316b2 2055
1da177e4 2056 rdev->mddev = mddev;
19133a42 2057 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2058
b2d6db58 2059 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2060 goto fail;
86e6ffdd 2061
0762b8bd 2062 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2063 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2064 /* failure here is OK */;
2065 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2066
4b80991c 2067 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2068 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2069
2070 /* May as well allow recovery to be retried once */
5389042f 2071 mddev->recovery_disabled++;
3f9d99c1 2072
1da177e4 2073 return 0;
5e55e2f5
N		 2074
2075 fail:
2076 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2077 b, mdname(mddev));
2078 return err;
1da177e4
LT		 2079}
2080
177a99b2 2081static void md_delayed_delete(struct work_struct *ws)
5792a285 2082{
3cb03002 2083 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2084 kobject_del(&rdev->kobj);
177a99b2 2085 kobject_put(&rdev->kobj);
5792a285
N		 2086}
2087
3cb03002 2088static void unbind_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2089{
2090 char b[BDEVNAME_SIZE];
2091 if (!rdev->mddev) {
2092 MD_BUG();
2093 return;
2094 }
49731baa 2095 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2096 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2097 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2098 rdev->mddev = NULL;
86e6ffdd 2099 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2100 sysfs_put(rdev->sysfs_state);
2101 rdev->sysfs_state = NULL;
2230dfe4
N		 2102 kfree(rdev->badblocks.page);
2103 rdev->badblocks.count = 0;
2104 rdev->badblocks.page = NULL;
5792a285 2105 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2106 * writing to 'remove' to "dev/state". We also need
2107 * to delay it due to rcu usage.
5792a285 2108 */
4b80991c 2109 synchronize_rcu();
177a99b2
N		 2110 INIT_WORK(&rdev->del_work, md_delayed_delete);
2111 kobject_get(&rdev->kobj);
e804ac78 2112 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2113}
2114
2115/*
2116 * prevent the device from being mounted, repartitioned or
2117 * otherwise reused by a RAID array (or any other kernel
2118 * subsystem), by bd_claiming the device.
2119 */
3cb03002 2120static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2121{
2122 int err = 0;
2123 struct block_device *bdev;
2124 char b[BDEVNAME_SIZE];
2125
d4d77629 2126 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2127 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2128 if (IS_ERR(bdev)) {
2129 printk(KERN_ERR "md: could not open %s.\n",
2130 __bdevname(dev, b));
2131 return PTR_ERR(bdev);
2132 }
1da177e4
LT		 2133 rdev->bdev = bdev;
2134 return err;
2135}
2136
3cb03002 2137static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2138{
2139 struct block_device *bdev = rdev->bdev;
2140 rdev->bdev = NULL;
2141 if (!bdev)
2142 MD_BUG();
e525fd89 2143 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2144}
2145
2146void md_autodetect_dev(dev_t dev);
2147
3cb03002 2148static void export_rdev(struct md_rdev * rdev)
1da177e4
LT		 2149{
2150 char b[BDEVNAME_SIZE];
2151 printk(KERN_INFO "md: export_rdev(%s)\n",
2152 bdevname(rdev->bdev,b));
2153 if (rdev->mddev)
2154 MD_BUG();
2155 free_disk_sb(rdev);
1da177e4 2156#ifndef MODULE
d0fae18f
N		 2157 if (test_bit(AutoDetected, &rdev->flags))
2158 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2159#endif
2160 unlock_rdev(rdev);
86e6ffdd 2161 kobject_put(&rdev->kobj);
1da177e4
LT		 2162}
2163
3cb03002 2164static void kick_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2165{
2166 unbind_rdev_from_array(rdev);
2167 export_rdev(rdev);
2168}
2169
fd01b88c 2170static void export_array(struct mddev *mddev)
1da177e4 2171{
3cb03002 2172 struct md_rdev *rdev, *tmp;
1da177e4 2173
d089c6af 2174 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 2175 if (!rdev->mddev) {
2176 MD_BUG();
2177 continue;
2178 }
2179 kick_rdev_from_array(rdev);
2180 }
2181 if (!list_empty(&mddev->disks))
2182 MD_BUG();
2183 mddev->raid_disks = 0;
2184 mddev->major_version = 0;
2185}
2186
2187static void print_desc(mdp_disk_t *desc)
2188{
2189 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2190 desc->major,desc->minor,desc->raid_disk,desc->state);
2191}
2192
cd2ac932 2193static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2194{
2195 int i;
2196
2197 printk(KERN_INFO
2198 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2199 sb->major_version, sb->minor_version, sb->patch_version,
2200 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2201 sb->ctime);
2202 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2203 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2204 sb->md_minor, sb->layout, sb->chunk_size);
2205 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2206 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2207 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2208 sb->failed_disks, sb->spare_disks,
2209 sb->sb_csum, (unsigned long)sb->events_lo);
2210
2211 printk(KERN_INFO);
2212 for (i = 0; i < MD_SB_DISKS; i++) {
2213 mdp_disk_t *desc;
2214
2215 desc = sb->disks + i;
2216 if (desc->number || desc->major || desc->minor ||
2217 desc->raid_disk || (desc->state && (desc->state != 4))) {
2218 printk(" D %2d: ", i);
2219 print_desc(desc);
2220 }
2221 }
2222 printk(KERN_INFO "md: THIS: ");
2223 print_desc(&sb->this_disk);
cd2ac932 2224}
1da177e4 2225
cd2ac932
CR		 2226static void print_sb_1(struct mdp_superblock_1 *sb)
2227{
2228 __u8 *uuid;
2229
2230 uuid = sb->set_uuid;
ad361c98 2231 printk(KERN_INFO
7b75c2f8 2232 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2233 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2234 le32_to_cpu(sb->major_version),
2235 le32_to_cpu(sb->feature_map),
7b75c2f8 2236 uuid,
cd2ac932
CR		 2237 sb->set_name,
2238 (unsigned long long)le64_to_cpu(sb->ctime)
2239 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2240
2241 uuid = sb->device_uuid;
ad361c98
JP		 2242 printk(KERN_INFO
2243 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2244 " RO:%llu\n"
7b75c2f8 2245 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2246 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2247 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2248 le32_to_cpu(sb->level),
2249 (unsigned long long)le64_to_cpu(sb->size),
2250 le32_to_cpu(sb->raid_disks),
2251 le32_to_cpu(sb->layout),
2252 le32_to_cpu(sb->chunksize),
2253 (unsigned long long)le64_to_cpu(sb->data_offset),
2254 (unsigned long long)le64_to_cpu(sb->data_size),
2255 (unsigned long long)le64_to_cpu(sb->super_offset),
2256 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2257 le32_to_cpu(sb->dev_number),
7b75c2f8 2258 uuid,
cd2ac932
CR		 2259 sb->devflags,
2260 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2261 (unsigned long long)le64_to_cpu(sb->events),
2262 (unsigned long long)le64_to_cpu(sb->resync_offset),
2263 le32_to_cpu(sb->sb_csum),
2264 le32_to_cpu(sb->max_dev)
2265 );
1da177e4
LT		 2266}
2267
3cb03002 2268static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT		 2269{
2270 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2271 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2272 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2273 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2274 rdev->desc_nr);
1da177e4 2275 if (rdev->sb_loaded) {
cd2ac932
CR		 2276 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2277 switch (major_version) {
2278 case 0:
65a06f06 2279 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR		 2280 break;
2281 case 1:
65a06f06 2282 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR		 2283 break;
2284 }
1da177e4
LT		 2285 } else
2286 printk(KERN_INFO "md: no rdev superblock!\n");
2287}
2288
5e56341d 2289static void md_print_devices(void)
1da177e4 2290{
159ec1fc 2291 struct list_head *tmp;
3cb03002 2292 struct md_rdev *rdev;
fd01b88c 2293 struct mddev *mddev;
1da177e4
LT		 2294 char b[BDEVNAME_SIZE];
2295
2296 printk("\n");
2297 printk("md: **********************************\n");
2298 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2299 printk("md: **********************************\n");
29ac4aa3 2300 for_each_mddev(mddev, tmp) {
1da177e4 2301
32a7627c
N		 2302 if (mddev->bitmap)
2303 bitmap_print_sb(mddev->bitmap);
2304 else
2305 printk("%s: ", mdname(mddev));
159ec1fc 2306 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 2307 printk("<%s>", bdevname(rdev->bdev,b));
2308 printk("\n");
2309
159ec1fc 2310 list_for_each_entry(rdev, &mddev->disks, same_set)
cd2ac932 2311 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2312 }
2313 printk("md: **********************************\n");
2314 printk("\n");
2315}
2316
2317
fd01b88c 2318static void sync_sbs(struct mddev * mddev, int nospares)
1da177e4 2319{
42543769
N		 2320 /* Update each superblock (in-memory image), but
2321 * if we are allowed to, skip spares which already
2322 * have the right event counter, or have one earlier
2323 * (which would mean they aren't being marked as dirty
2324 * with the rest of the array)
2325 */
3cb03002 2326 struct md_rdev *rdev;
159ec1fc 2327 list_for_each_entry(rdev, &mddev->disks, same_set) {
42543769
N		 2328 if (rdev->sb_events == mddev->events ||
2329 (nospares &&
2330 rdev->raid_disk < 0 &&
42543769
N		 2331 rdev->sb_events+1 == mddev->events)) {
2332 /* Don't update this superblock */
2333 rdev->sb_loaded = 2;
2334 } else {
076f968b 2335 sync_super(mddev, rdev);
42543769
N		 2336 rdev->sb_loaded = 1;
2337 }
1da177e4
LT		 2338 }
2339}
2340
fd01b88c 2341static void md_update_sb(struct mddev * mddev, int force_change)
1da177e4 2342{
3cb03002 2343 struct md_rdev *rdev;
06d91a5f 2344 int sync_req;
42543769 2345 int nospares = 0;
2699b672 2346 int any_badblocks_changed = 0;
1da177e4 2347
1da177e4 2348repeat:
3a3a5ddb
N		 2349 /* First make sure individual recovery_offsets are correct */
2350 list_for_each_entry(rdev, &mddev->disks, same_set) {
2351 if (rdev->raid_disk >= 0 &&
2352 mddev->delta_disks >= 0 &&
2353 !test_bit(In_sync, &rdev->flags) &&
2354 mddev->curr_resync_completed > rdev->recovery_offset)
2355 rdev->recovery_offset = mddev->curr_resync_completed;
2356
2357 }
bd52b746 2358 if (!mddev->persistent) {
070dc6dd 2359 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2360 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2361 if (!mddev->external) {
d97a41dc 2362 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
de393cde
N		 2363 list_for_each_entry(rdev, &mddev->disks, same_set) {
2364 if (rdev->badblocks.changed) {
2365 md_ack_all_badblocks(&rdev->badblocks);
2366 md_error(mddev, rdev);
2367 }
2368 clear_bit(Blocked, &rdev->flags);
2369 clear_bit(BlockedBadBlocks, &rdev->flags);
2370 wake_up(&rdev->blocked_wait);
2371 }
2372 }
3a3a5ddb
N		 2373 wake_up(&mddev->sb_wait);
2374 return;
2375 }
2376
a9701a30 2377 spin_lock_irq(&mddev->write_lock);
84692195 2378
3a3a5ddb
N		 2379 mddev->utime = get_seconds();
2380
850b2b42
N		 2381 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2382 force_change = 1;
2383 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2384 /* just a clean<-> dirty transition, possibly leave spares alone,
2385 * though if events isn't the right even/odd, we will have to do
2386 * spares after all
2387 */
2388 nospares = 1;
2389 if (force_change)
2390 nospares = 0;
2391 if (mddev->degraded)
84692195
N		 2392 /* If the array is degraded, then skipping spares is both
2393 * dangerous and fairly pointless.
2394 * Dangerous because a device that was removed from the array
2395 * might have a event_count that still looks up-to-date,
2396 * so it can be re-added without a resync.
2397 * Pointless because if there are any spares to skip,
2398 * then a recovery will happen and soon that array won't
2399 * be degraded any more and the spare can go back to sleep then.
2400 */
850b2b42 2401 nospares = 0;
84692195 2402
06d91a5f 2403 sync_req = mddev->in_sync;
42543769
N		 2404
2405 /* If this is just a dirty<->clean transition, and the array is clean
2406 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2407 if (nospares
42543769 2408 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2409 && mddev->can_decrease_events
2410 && mddev->events != 1) {
42543769 2411 mddev->events--;
a8707c08
N		 2412 mddev->can_decrease_events = 0;
2413 } else {
42543769
N		 2414 /* otherwise we have to go forward and ... */
2415 mddev->events ++;
a8707c08 2416 mddev->can_decrease_events = nospares;
42543769 2417 }
1da177e4
LT		 2418
2419 if (!mddev->events) {
2420 /*
2421 * oops, this 64-bit counter should never wrap.
2422 * Either we are in around ~1 trillion A.C., assuming
2423 * 1 reboot per second, or we have a bug:
2424 */
2425 MD_BUG();
2426 mddev->events --;
2427 }
2699b672 2428
de393cde 2429 list_for_each_entry(rdev, &mddev->disks, same_set) {
2699b672
N		 2430 if (rdev->badblocks.changed)
2431 any_badblocks_changed++;
de393cde
N		 2432 if (test_bit(Faulty, &rdev->flags))
2433 set_bit(FaultRecorded, &rdev->flags);
2434 }
2699b672 2435
e691063a 2436 sync_sbs(mddev, nospares);
a9701a30 2437 spin_unlock_irq(&mddev->write_lock);
1da177e4 2438
36a4e1fe
N		 2439 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2440 mdname(mddev), mddev->in_sync);
1da177e4 2441
4ad13663 2442 bitmap_update_sb(mddev->bitmap);
159ec1fc 2443 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 2444 char b[BDEVNAME_SIZE];
36a4e1fe 2445
42543769
N		 2446 if (rdev->sb_loaded != 1)
2447 continue; /* no noise on spare devices */
1da177e4 2448
d70ed2e4
AW		 2449 if (!test_bit(Faulty, &rdev->flags) &&
2450 rdev->saved_raid_disk == -1) {
7bfa19f2 2451 md_super_write(mddev,rdev,
0f420358 2452 rdev->sb_start, rdev->sb_size,
7bfa19f2 2453 rdev->sb_page);
36a4e1fe
N		 2454 pr_debug("md: (write) %s's sb offset: %llu\n",
2455 bdevname(rdev->bdev, b),
2456 (unsigned long long)rdev->sb_start);
42543769 2457 rdev->sb_events = mddev->events;
2699b672
N		 2458 if (rdev->badblocks.size) {
2459 md_super_write(mddev, rdev,
2460 rdev->badblocks.sector,
2461 rdev->badblocks.size << 9,
2462 rdev->bb_page);
2463 rdev->badblocks.size = 0;
2464 }
7bfa19f2 2465
d70ed2e4 2466 } else if (test_bit(Faulty, &rdev->flags))
36a4e1fe
N		 2467 pr_debug("md: %s (skipping faulty)\n",
2468 bdevname(rdev->bdev, b));
d70ed2e4
AW		 2469 else
2470 pr_debug("(skipping incremental s/r ");
2471
7bfa19f2 2472 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2473 /* only need to write one superblock... */
2474 break;
2475 }
a9701a30 2476 md_super_wait(mddev);
850b2b42 2477 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2478
a9701a30 2479 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2480 if (mddev->in_sync != sync_req ||
2481 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2482 /* have to write it out again */
a9701a30 2483 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2484 goto repeat;
2485 }
850b2b42 2486 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2487 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2488 wake_up(&mddev->sb_wait);
acb180b0
N		 2489 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2490 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2491
de393cde
N		 2492 list_for_each_entry(rdev, &mddev->disks, same_set) {
2493 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2494 clear_bit(Blocked, &rdev->flags);
2495
2496 if (any_badblocks_changed)
2699b672 2497 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2498 clear_bit(BlockedBadBlocks, &rdev->flags);
2499 wake_up(&rdev->blocked_wait);
2500 }
1da177e4
LT		 2501}
2502
7f6ce769 2503/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2504 * We want to accept with case. For this we use cmd_match.
2505 */
2506static int cmd_match(const char *cmd, const char *str)
2507{
2508 /* See if cmd, written into a sysfs file, matches
2509 * str. They must either be the same, or cmd can
2510 * have a trailing newline
2511 */
2512 while (*cmd && *str && *cmd == *str) {
2513 cmd++;
2514 str++;
2515 }
2516 if (*cmd == '\n')
2517 cmd++;
2518 if (*str || *cmd)
2519 return 0;
2520 return 1;
2521}
2522
86e6ffdd
N		 2523struct rdev_sysfs_entry {
2524 struct attribute attr;
3cb03002
N		 2525 ssize_t (*show)(struct md_rdev *, char *);
2526 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2527};
2528
2529static ssize_t
3cb03002 2530state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2531{
2532 char *sep = "";
20a49ff6 2533 size_t len = 0;
86e6ffdd 2534
de393cde
N		 2535 if (test_bit(Faulty, &rdev->flags) ||
2536 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2537 len+= sprintf(page+len, "%sfaulty",sep);
2538 sep = ",";
2539 }
b2d444d7 2540 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2541 len += sprintf(page+len, "%sin_sync",sep);
2542 sep = ",";
2543 }
f655675b
N		 2544 if (test_bit(WriteMostly, &rdev->flags)) {
2545 len += sprintf(page+len, "%swrite_mostly",sep);
2546 sep = ",";
2547 }
de393cde 2548 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N		 2549 (rdev->badblocks.unacked_exist
2550 && !test_bit(Faulty, &rdev->flags))) {
6bfe0b49
DW		 2551 len += sprintf(page+len, "%sblocked", sep);
2552 sep = ",";
2553 }
b2d444d7
N		 2554 if (!test_bit(Faulty, &rdev->flags) &&
2555 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2556 len += sprintf(page+len, "%sspare", sep);
2557 sep = ",";
2558 }
d7a9d443
N		 2559 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2560 len += sprintf(page+len, "%swrite_error", sep);
2561 sep = ",";
2562 }
86e6ffdd
N		 2563 return len+sprintf(page+len, "\n");
2564}
2565
45dc2de1 2566static ssize_t
3cb03002 2567state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2568{
2569 /* can write
de393cde 2570 * faulty - simulates an error
45dc2de1 2571 * remove - disconnects the device
f655675b
N		 2572 * writemostly - sets write_mostly
2573 * -writemostly - clears write_mostly
de393cde
N		 2574 * blocked - sets the Blocked flags
2575 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2576 * insync - sets Insync providing device isn't active
d7a9d443
N		 2577 * write_error - sets WriteErrorSeen
2578 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2579 */
2580 int err = -EINVAL;
2581 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2582 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2583 if (test_bit(Faulty, &rdev->flags))
2584 err = 0;
2585 else
2586 err = -EBUSY;
45dc2de1
N		 2587 } else if (cmd_match(buf, "remove")) {
2588 if (rdev->raid_disk >= 0)
2589 err = -EBUSY;
2590 else {
fd01b88c 2591 struct mddev *mddev = rdev->mddev;
45dc2de1 2592 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2593 if (mddev->pers)
2594 md_update_sb(mddev, 1);
45dc2de1
N		 2595 md_new_event(mddev);
2596 err = 0;
2597 }
f655675b
N		 2598 } else if (cmd_match(buf, "writemostly")) {
2599 set_bit(WriteMostly, &rdev->flags);
2600 err = 0;
2601 } else if (cmd_match(buf, "-writemostly")) {
2602 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2603 err = 0;
2604 } else if (cmd_match(buf, "blocked")) {
2605 set_bit(Blocked, &rdev->flags);
2606 err = 0;
2607 } else if (cmd_match(buf, "-blocked")) {
de393cde 2608 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2609 rdev->badblocks.unacked_exist) {
de393cde
N		 2610 /* metadata handler doesn't understand badblocks,
2611 * so we need to fail the device
2612 */
2613 md_error(rdev->mddev, rdev);
2614 }
6bfe0b49 2615 clear_bit(Blocked, &rdev->flags);
de393cde 2616 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2617 wake_up(&rdev->blocked_wait);
2618 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2619 md_wakeup_thread(rdev->mddev->thread);
2620
6d56e278
N		 2621 err = 0;
2622 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2623 set_bit(In_sync, &rdev->flags);
f655675b 2624 err = 0;
d7a9d443
N		 2625 } else if (cmd_match(buf, "write_error")) {
2626 set_bit(WriteErrorSeen, &rdev->flags);
2627 err = 0;
2628 } else if (cmd_match(buf, "-write_error")) {
2629 clear_bit(WriteErrorSeen, &rdev->flags);
2630 err = 0;
45dc2de1 2631 }
00bcb4ac
N		 2632 if (!err)
2633 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2634 return err ? err : len;
2635}
80ca3a44
N		 2636static struct rdev_sysfs_entry rdev_state =
2637__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2638
4dbcdc75 2639static ssize_t
3cb03002 2640errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2641{
2642 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2643}
2644
2645static ssize_t
3cb03002 2646errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2647{
2648 char *e;
2649 unsigned long n = simple_strtoul(buf, &e, 10);
2650 if (*buf && (*e == 0 || *e == '\n')) {
2651 atomic_set(&rdev->corrected_errors, n);
2652 return len;
2653 }
2654 return -EINVAL;
2655}
2656static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2657__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2658
014236d2 2659static ssize_t
3cb03002 2660slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2661{
2662 if (rdev->raid_disk < 0)
2663 return sprintf(page, "none\n");
2664 else
2665 return sprintf(page, "%d\n", rdev->raid_disk);
2666}
2667
2668static ssize_t
3cb03002 2669slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2670{
2671 char *e;
c303da6d 2672 int err;
014236d2
N		 2673 int slot = simple_strtoul(buf, &e, 10);
2674 if (strncmp(buf, "none", 4)==0)
2675 slot = -1;
2676 else if (e==buf || (*e && *e!= '\n'))
2677 return -EINVAL;
6c2fce2e 2678 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2679 /* Setting 'slot' on an active array requires also
2680 * updating the 'rd%d' link, and communicating
2681 * with the personality with ->hot_*_disk.
2682 * For now we only support removing
2683 * failed/spare devices. This normally happens automatically,
2684 * but not when the metadata is externally managed.
2685 */
c303da6d
N		 2686 if (rdev->raid_disk == -1)
2687 return -EEXIST;
2688 /* personality does all needed checks */
01393f3d 2689 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d
N		 2690 return -EINVAL;
2691 err = rdev->mddev->pers->
b8321b68 2692 hot_remove_disk(rdev->mddev, rdev);
c303da6d
N		 2693 if (err)
2694 return err;
36fad858 2695 sysfs_unlink_rdev(rdev->mddev, rdev);
b7103107 2696 rdev->raid_disk = -1;
c303da6d
N		 2697 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2698 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2699 } else if (rdev->mddev->pers) {
6c2fce2e 2700 /* Activating a spare .. or possibly reactivating
6d56e278 2701 * if we ever get bitmaps working here.
6c2fce2e
NB		 2702 */
2703
2704 if (rdev->raid_disk != -1)
2705 return -EBUSY;
2706
c6751b2b
N		 2707 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2708 return -EBUSY;
2709
6c2fce2e
NB		 2710 if (rdev->mddev->pers->hot_add_disk == NULL)
2711 return -EINVAL;
2712
ba1b41b6
N		 2713 if (slot >= rdev->mddev->raid_disks &&
2714 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2715 return -ENOSPC;
2716
6c2fce2e
NB		 2717 rdev->raid_disk = slot;
2718 if (test_bit(In_sync, &rdev->flags))
2719 rdev->saved_raid_disk = slot;
2720 else
2721 rdev->saved_raid_disk = -1;
d30519fc 2722 clear_bit(In_sync, &rdev->flags);
6c2fce2e
NB		 2723 err = rdev->mddev->pers->
2724 hot_add_disk(rdev->mddev, rdev);
199050ea 2725 if (err) {
6c2fce2e 2726 rdev->raid_disk = -1;
6c2fce2e 2727 return err;
52664732 2728 } else
00bcb4ac 2729 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2730 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2731 /* failure here is OK */;
6c2fce2e 2732 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2733 } else {
ba1b41b6
N		 2734 if (slot >= rdev->mddev->raid_disks &&
2735 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2736 return -ENOSPC;
2737 rdev->raid_disk = slot;
2738 /* assume it is working */
c5d79adb
N		 2739 clear_bit(Faulty, &rdev->flags);
2740 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2741 set_bit(In_sync, &rdev->flags);
00bcb4ac 2742 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2743 }
014236d2
N		 2744 return len;
2745}
2746
2747
2748static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2749__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2750
93c8cad0 2751static ssize_t
3cb03002 2752offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2753{
6961ece4 2754 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2755}
2756
2757static ssize_t
3cb03002 2758offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0
N		 2759{
2760 char *e;
2761 unsigned long long offset = simple_strtoull(buf, &e, 10);
2762 if (e==buf || (*e && *e != '\n'))
2763 return -EINVAL;
8ed0a521 2764 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2765 return -EBUSY;
dd8ac336 2766 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2767 /* Must set offset before size, so overlap checks
2768 * can be sane */
2769 return -EBUSY;
93c8cad0
N		 2770 rdev->data_offset = offset;
2771 return len;
2772}
2773
2774static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2775__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2776
83303b61 2777static ssize_t
3cb03002 2778rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2779{
dd8ac336 2780 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2781}
2782
c5d79adb
N		 2783static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2784{
2785 /* check if two start/length pairs overlap */
2786 if (s1+l1 <= s2)
2787 return 0;
2788 if (s2+l2 <= s1)
2789 return 0;
2790 return 1;
2791}
2792
b522adcd
DW		 2793static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2794{
2795 unsigned long long blocks;
2796 sector_t new;
2797
2798 if (strict_strtoull(buf, 10, &blocks) < 0)
2799 return -EINVAL;
2800
2801 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2802 return -EINVAL; /* sector conversion overflow */
2803
2804 new = blocks * 2;
2805 if (new != blocks * 2)
2806 return -EINVAL; /* unsigned long long to sector_t overflow */
2807
2808 *sectors = new;
2809 return 0;
2810}
2811
83303b61 2812static ssize_t
3cb03002 2813rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2814{
fd01b88c 2815 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2816 sector_t oldsectors = rdev->sectors;
b522adcd 2817 sector_t sectors;
27c529bb 2818
b522adcd 2819 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2820 return -EINVAL;
0cd17fec 2821 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2822 if (my_mddev->persistent) {
dd8ac336
AN		 2823 sectors = super_types[my_mddev->major_version].
2824 rdev_size_change(rdev, sectors);
2825 if (!sectors)
0cd17fec 2826 return -EBUSY;
dd8ac336 2827 } else if (!sectors)
77304d2a 2828 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2829 rdev->data_offset;
0cd17fec 2830 }
dd8ac336 2831 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2832 return -EINVAL; /* component must fit device */
0cd17fec 2833
dd8ac336
AN		 2834 rdev->sectors = sectors;
2835 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2836 /* need to check that all other rdevs with the same ->bdev
2837 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2838 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2839 * we have to change it back, we will have the lock again.
2840 */
fd01b88c 2841 struct mddev *mddev;
c5d79adb 2842 int overlap = 0;
159ec1fc 2843 struct list_head *tmp;
c5d79adb 2844
27c529bb 2845 mddev_unlock(my_mddev);
29ac4aa3 2846 for_each_mddev(mddev, tmp) {
3cb03002 2847 struct md_rdev *rdev2;
c5d79adb
N		 2848
2849 mddev_lock(mddev);
159ec1fc 2850 list_for_each_entry(rdev2, &mddev->disks, same_set)
f21e9ff7
N		 2851 if (rdev->bdev == rdev2->bdev &&
2852 rdev != rdev2 &&
2853 overlaps(rdev->data_offset, rdev->sectors,
2854 rdev2->data_offset,
2855 rdev2->sectors)) {
c5d79adb
N		 2856 overlap = 1;
2857 break;
2858 }
2859 mddev_unlock(mddev);
2860 if (overlap) {
2861 mddev_put(mddev);
2862 break;
2863 }
2864 }
27c529bb 2865 mddev_lock(my_mddev);
c5d79adb
N		 2866 if (overlap) {
2867 /* Someone else could have slipped in a size
2868 * change here, but doing so is just silly.
dd8ac336 2869 * We put oldsectors back because we *know* it is
c5d79adb
N		 2870 * safe, and trust userspace not to race with
2871 * itself
2872 */
dd8ac336 2873 rdev->sectors = oldsectors;
c5d79adb
N		 2874 return -EBUSY;
2875 }
2876 }
83303b61
N		 2877 return len;
2878}
2879
2880static struct rdev_sysfs_entry rdev_size =
80ca3a44 2881__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2882
06e3c817 2883
3cb03002 2884static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2885{
2886 unsigned long long recovery_start = rdev->recovery_offset;
2887
2888 if (test_bit(In_sync, &rdev->flags) ||
2889 recovery_start == MaxSector)
2890 return sprintf(page, "none\n");
2891
2892 return sprintf(page, "%llu\n", recovery_start);
2893}
2894
3cb03002 2895static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2896{
2897 unsigned long long recovery_start;
2898
2899 if (cmd_match(buf, "none"))
2900 recovery_start = MaxSector;
2901 else if (strict_strtoull(buf, 10, &recovery_start))
2902 return -EINVAL;
2903
2904 if (rdev->mddev->pers &&
2905 rdev->raid_disk >= 0)
2906 return -EBUSY;
2907
2908 rdev->recovery_offset = recovery_start;
2909 if (recovery_start == MaxSector)
2910 set_bit(In_sync, &rdev->flags);
2911 else
2912 clear_bit(In_sync, &rdev->flags);
2913 return len;
2914}
2915
2916static struct rdev_sysfs_entry rdev_recovery_start =
2917__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2918
16c791a5
N		 2919
2920static ssize_t
2921badblocks_show(struct badblocks *bb, char *page, int unack);
2922static ssize_t
2923badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2924
3cb03002 2925static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2926{
2927 return badblocks_show(&rdev->badblocks, page, 0);
2928}
3cb03002 2929static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2930{
de393cde
N		 2931 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2932 /* Maybe that ack was all we needed */
2933 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2934 wake_up(&rdev->blocked_wait);
2935 return rv;
16c791a5
N		 2936}
2937static struct rdev_sysfs_entry rdev_bad_blocks =
2938__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2939
2940
3cb03002 2941static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2942{
2943 return badblocks_show(&rdev->badblocks, page, 1);
2944}
3cb03002 2945static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2946{
2947 return badblocks_store(&rdev->badblocks, page, len, 1);
2948}
2949static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2950__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2951
86e6ffdd
N		 2952static struct attribute *rdev_default_attrs[] = {
2953 &rdev_state.attr,
4dbcdc75 2954 &rdev_errors.attr,
014236d2 2955 &rdev_slot.attr,
93c8cad0 2956 &rdev_offset.attr,
83303b61 2957 &rdev_size.attr,
06e3c817 2958 &rdev_recovery_start.attr,
16c791a5
N		 2959 &rdev_bad_blocks.attr,
2960 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 2961 NULL,
2962};
2963static ssize_t
2964rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2965{
2966 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2967 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 2968 struct mddev *mddev = rdev->mddev;
27c529bb 2969 ssize_t rv;
86e6ffdd
N		 2970
2971 if (!entry->show)
2972 return -EIO;
27c529bb
N		 2973
2974 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2975 if (!rv) {
2976 if (rdev->mddev == NULL)
2977 rv = -EBUSY;
2978 else
2979 rv = entry->show(rdev, page);
2980 mddev_unlock(mddev);
2981 }
2982 return rv;
86e6ffdd
N		 2983}
2984
2985static ssize_t
2986rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2987 const char *page, size_t length)
2988{
2989 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2990 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 2991 ssize_t rv;
fd01b88c 2992 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 2993
2994 if (!entry->store)
2995 return -EIO;
67463acb
N		 2996 if (!capable(CAP_SYS_ADMIN))
2997 return -EACCES;
27c529bb 2998 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2999 if (!rv) {
27c529bb
N		 3000 if (rdev->mddev == NULL)
3001 rv = -EBUSY;
3002 else
3003 rv = entry->store(rdev, page, length);
6a51830e 3004 mddev_unlock(mddev);
ca388059
N		 3005 }
3006 return rv;
86e6ffdd
N		 3007}
3008
3009static void rdev_free(struct kobject *ko)
3010{
3cb03002 3011 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3012 kfree(rdev);
3013}
52cf25d0 3014static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3015 .show = rdev_attr_show,
3016 .store = rdev_attr_store,
3017};
3018static struct kobj_type rdev_ktype = {
3019 .release = rdev_free,
3020 .sysfs_ops = &rdev_sysfs_ops,
3021 .default_attrs = rdev_default_attrs,
3022};
3023
3cb03002 3024int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3025{
3026 rdev->desc_nr = -1;
3027 rdev->saved_raid_disk = -1;
3028 rdev->raid_disk = -1;
3029 rdev->flags = 0;
3030 rdev->data_offset = 0;
3031 rdev->sb_events = 0;
3032 rdev->last_read_error.tv_sec = 0;
3033 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3034 rdev->sb_loaded = 0;
3035 rdev->bb_page = NULL;
e8bb9a83
N		 3036 atomic_set(&rdev->nr_pending, 0);
3037 atomic_set(&rdev->read_errors, 0);
3038 atomic_set(&rdev->corrected_errors, 0);
3039
3040 INIT_LIST_HEAD(&rdev->same_set);
3041 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3042
3043 /* Add space to store bad block list.
3044 * This reserves the space even on arrays where it cannot
3045 * be used - I wonder if that matters
3046 */
3047 rdev->badblocks.count = 0;
3048 rdev->badblocks.shift = 0;
3049 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3050 seqlock_init(&rdev->badblocks.lock);
3051 if (rdev->badblocks.page == NULL)
3052 return -ENOMEM;
3053
3054 return 0;
e8bb9a83
N		 3055}
3056EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3057/*
3058 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3059 *
3060 * mark the device faulty if:
3061 *
3062 * - the device is nonexistent (zero size)
3063 * - the device has no valid superblock
3064 *
3065 * a faulty rdev _never_ has rdev->sb set.
3066 */
3cb03002 3067static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3068{
3069 char b[BDEVNAME_SIZE];
3070 int err;
3cb03002 3071 struct md_rdev *rdev;
1da177e4
LT		 3072 sector_t size;
3073
9ffae0cf 3074 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3075 if (!rdev) {
3076 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3077 return ERR_PTR(-ENOMEM);
3078 }
1da177e4 3079
2230dfe4
N		 3080 err = md_rdev_init(rdev);
3081 if (err)
3082 goto abort_free;
3083 err = alloc_disk_sb(rdev);
3084 if (err)
1da177e4
LT		 3085 goto abort_free;
3086
c5d79adb 3087 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3088 if (err)
3089 goto abort_free;
3090
f9cb074b 3091 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3092
77304d2a 3093 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 3094 if (!size) {
3095 printk(KERN_WARNING
3096 "md: %s has zero or unknown size, marking faulty!\n",
3097 bdevname(rdev->bdev,b));
3098 err = -EINVAL;
3099 goto abort_free;
3100 }
3101
3102 if (super_format >= 0) {
3103 err = super_types[super_format].
3104 load_super(rdev, NULL, super_minor);
3105 if (err == -EINVAL) {
df968c4e
N		 3106 printk(KERN_WARNING
3107 "md: %s does not have a valid v%d.%d "
3108 "superblock, not importing!\n",
3109 bdevname(rdev->bdev,b),
3110 super_format, super_minor);
1da177e4
LT		 3111 goto abort_free;
3112 }
3113 if (err < 0) {
3114 printk(KERN_WARNING
3115 "md: could not read %s's sb, not importing!\n",
3116 bdevname(rdev->bdev,b));
3117 goto abort_free;
3118 }
3119 }
9f2f3830
N		 3120 if (super_format == -1)
3121 /* hot-add for 0.90, or non-persistent: so no badblocks */
3122 rdev->badblocks.shift = -1;
6bfe0b49 3123
1da177e4
LT		 3124 return rdev;
3125
3126abort_free:
2699b672
N		 3127 if (rdev->bdev)
3128 unlock_rdev(rdev);
3129 free_disk_sb(rdev);
2230dfe4 3130 kfree(rdev->badblocks.page);
1da177e4
LT		 3131 kfree(rdev);
3132 return ERR_PTR(err);
3133}
3134
3135/*
3136 * Check a full RAID array for plausibility
3137 */
3138
3139
fd01b88c 3140static void analyze_sbs(struct mddev * mddev)
1da177e4
LT		 3141{
3142 int i;
3cb03002 3143 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3144 char b[BDEVNAME_SIZE];
3145
3146 freshest = NULL;
d089c6af 3147 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 3148 switch (super_types[mddev->major_version].
3149 load_super(rdev, freshest, mddev->minor_version)) {
3150 case 1:
3151 freshest = rdev;
3152 break;
3153 case 0:
3154 break;
3155 default:
3156 printk( KERN_ERR \
3157 "md: fatal superblock inconsistency in %s"
3158 " -- removing from array\n",
3159 bdevname(rdev->bdev,b));
3160 kick_rdev_from_array(rdev);
3161 }
3162
3163
3164 super_types[mddev->major_version].
3165 validate_super(mddev, freshest);
3166
3167 i = 0;
d089c6af 3168 rdev_for_each(rdev, tmp, mddev) {
233fca36
N		 3169 if (mddev->max_disks &&
3170 (rdev->desc_nr >= mddev->max_disks ||
3171 i > mddev->max_disks)) {
de01dfad
N		 3172 printk(KERN_WARNING
3173 "md: %s: %s: only %d devices permitted\n",
3174 mdname(mddev), bdevname(rdev->bdev, b),
3175 mddev->max_disks);
3176 kick_rdev_from_array(rdev);
3177 continue;
3178 }
1da177e4
LT		 3179 if (rdev != freshest)
3180 if (super_types[mddev->major_version].
3181 validate_super(mddev, rdev)) {
3182 printk(KERN_WARNING "md: kicking non-fresh %s"
3183 " from array!\n",
3184 bdevname(rdev->bdev,b));
3185 kick_rdev_from_array(rdev);
3186 continue;
3187 }
3188 if (mddev->level == LEVEL_MULTIPATH) {
3189 rdev->desc_nr = i++;
3190 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3191 set_bit(In_sync, &rdev->flags);
5e5e3e78 3192 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3193 rdev->raid_disk = -1;
3194 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3195 }
3196 }
1da177e4
LT		 3197}
3198
72e02075
N		 3199/* Read a fixed-point number.
3200 * Numbers in sysfs attributes should be in "standard" units where
3201 * possible, so time should be in seconds.
3202 * However we internally use a a much smaller unit such as
3203 * milliseconds or jiffies.
3204 * This function takes a decimal number with a possible fractional
3205 * component, and produces an integer which is the result of
3206 * multiplying that number by 10^'scale'.
3207 * all without any floating-point arithmetic.
3208 */
3209int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3210{
3211 unsigned long result = 0;
3212 long decimals = -1;
3213 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3214 if (*cp == '.')
3215 decimals = 0;
3216 else if (decimals < scale) {
3217 unsigned int value;
3218 value = *cp - '0';
3219 result = result * 10 + value;
3220 if (decimals >= 0)
3221 decimals++;
3222 }
3223 cp++;
3224 }
3225 if (*cp == '\n')
3226 cp++;
3227 if (*cp)
3228 return -EINVAL;
3229 if (decimals < 0)
3230 decimals = 0;
3231 while (decimals < scale) {
3232 result *= 10;
3233 decimals ++;
3234 }
3235 *res = result;
3236 return 0;
3237}
3238
3239
19052c0e
N		 3240static void md_safemode_timeout(unsigned long data);
3241
16f17b39 3242static ssize_t
fd01b88c 3243safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3244{
3245 int msec = (mddev->safemode_delay*1000)/HZ;
3246 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3247}
3248static ssize_t
fd01b88c 3249safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3250{
16f17b39 3251 unsigned long msec;
97ce0a7f 3252
72e02075 3253 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3254 return -EINVAL;
16f17b39
N		 3255 if (msec == 0)
3256 mddev->safemode_delay = 0;
3257 else {
19052c0e 3258 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 3259 mddev->safemode_delay = (msec*HZ)/1000;
3260 if (mddev->safemode_delay == 0)
3261 mddev->safemode_delay = 1;
19052c0e
N		 3262 if (mddev->safemode_delay < old_delay)
3263 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 3264 }
3265 return len;
3266}
3267static struct md_sysfs_entry md_safe_delay =
80ca3a44 3268__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3269
eae1701f 3270static ssize_t
fd01b88c 3271level_show(struct mddev *mddev, char *page)
eae1701f 3272{
84fc4b56 3273 struct md_personality *p = mddev->pers;
d9d166c2 3274 if (p)
eae1701f 3275 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 3276 else if (mddev->clevel[0])
3277 return sprintf(page, "%s\n", mddev->clevel);
3278 else if (mddev->level != LEVEL_NONE)
3279 return sprintf(page, "%d\n", mddev->level);
3280 else
3281 return 0;
eae1701f
N		 3282}
3283
d9d166c2 3284static ssize_t
fd01b88c 3285level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3286{
f2859af6 3287 char clevel[16];
20a49ff6 3288 ssize_t rv = len;
84fc4b56 3289 struct md_personality *pers;
f2859af6 3290 long level;
245f46c2 3291 void *priv;
3cb03002 3292 struct md_rdev *rdev;
245f46c2
N		 3293
3294 if (mddev->pers == NULL) {
3295 if (len == 0)
3296 return 0;
3297 if (len >= sizeof(mddev->clevel))
3298 return -ENOSPC;
3299 strncpy(mddev->clevel, buf, len);
3300 if (mddev->clevel[len-1] == '\n')
3301 len--;
3302 mddev->clevel[len] = 0;
3303 mddev->level = LEVEL_NONE;
3304 return rv;
3305 }
3306
3307 /* request to change the personality. Need to ensure:
3308 * - array is not engaged in resync/recovery/reshape
3309 * - old personality can be suspended
3310 * - new personality will access other array.
3311 */
3312
bb4f1e9d
N		 3313 if (mddev->sync_thread ||
3314 mddev->reshape_position != MaxSector ||
3315 mddev->sysfs_active)
d9d166c2 3316 return -EBUSY;
245f46c2
N		 3317
3318 if (!mddev->pers->quiesce) {
3319 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3320 mdname(mddev), mddev->pers->name);
3321 return -EINVAL;
3322 }
3323
3324 /* Now find the new personality */
f2859af6 3325 if (len == 0 || len >= sizeof(clevel))
245f46c2 3326 return -EINVAL;
f2859af6
DW		 3327 strncpy(clevel, buf, len);
3328 if (clevel[len-1] == '\n')
d9d166c2 3329 len--;
f2859af6
DW		 3330 clevel[len] = 0;
3331 if (strict_strtol(clevel, 10, &level))
3332 level = LEVEL_NONE;
245f46c2 3333
f2859af6
DW		 3334 if (request_module("md-%s", clevel) != 0)
3335 request_module("md-level-%s", clevel);
245f46c2 3336 spin_lock(&pers_lock);
f2859af6 3337 pers = find_pers(level, clevel);
245f46c2
N		 3338 if (!pers || !try_module_get(pers->owner)) {
3339 spin_unlock(&pers_lock);
f2859af6 3340 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3341 return -EINVAL;
3342 }
3343 spin_unlock(&pers_lock);
3344
3345 if (pers == mddev->pers) {
3346 /* Nothing to do! */
3347 module_put(pers->owner);
3348 return rv;
3349 }
3350 if (!pers->takeover) {
3351 module_put(pers->owner);
3352 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3353 mdname(mddev), clevel);
245f46c2
N		 3354 return -EINVAL;
3355 }
3356
e93f68a1
N		 3357 list_for_each_entry(rdev, &mddev->disks, same_set)
3358 rdev->new_raid_disk = rdev->raid_disk;
3359
245f46c2
N		 3360 /* ->takeover must set new_* and/or delta_disks
3361 * if it succeeds, and may set them when it fails.
3362 */
3363 priv = pers->takeover(mddev);
3364 if (IS_ERR(priv)) {
3365 mddev->new_level = mddev->level;
3366 mddev->new_layout = mddev->layout;
664e7c41 3367 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3368 mddev->raid_disks -= mddev->delta_disks;
3369 mddev->delta_disks = 0;
3370 module_put(pers->owner);
3371 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3372 mdname(mddev), clevel);
245f46c2
N		 3373 return PTR_ERR(priv);
3374 }
3375
3376 /* Looks like we have a winner */
3377 mddev_suspend(mddev);
3378 mddev->pers->stop(mddev);
a64c876f
N		 3379
3380 if (mddev->pers->sync_request == NULL &&
3381 pers->sync_request != NULL) {
3382 /* need to add the md_redundancy_group */
3383 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3384 printk(KERN_WARNING
3385 "md: cannot register extra attributes for %s\n",
3386 mdname(mddev));
19fdb9ee 3387 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3388 }
3389 if (mddev->pers->sync_request != NULL &&
3390 pers->sync_request == NULL) {
3391 /* need to remove the md_redundancy_group */
3392 if (mddev->to_remove == NULL)
3393 mddev->to_remove = &md_redundancy_group;
3394 }
3395
54071b38
TM		 3396 if (mddev->pers->sync_request == NULL &&
3397 mddev->external) {
3398 /* We are converting from a no-redundancy array
3399 * to a redundancy array and metadata is managed
3400 * externally so we need to be sure that writes
3401 * won't block due to a need to transition
3402 * clean->dirty
3403 * until external management is started.
3404 */
3405 mddev->in_sync = 0;
3406 mddev->safemode_delay = 0;
3407 mddev->safemode = 0;
3408 }
3409
e93f68a1 3410 list_for_each_entry(rdev, &mddev->disks, same_set) {
e93f68a1
N		 3411 if (rdev->raid_disk < 0)
3412 continue;
bf2cb0da 3413 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3414 rdev->new_raid_disk = -1;
3415 if (rdev->new_raid_disk == rdev->raid_disk)
3416 continue;
36fad858 3417 sysfs_unlink_rdev(mddev, rdev);
e93f68a1
N		 3418 }
3419 list_for_each_entry(rdev, &mddev->disks, same_set) {
3420 if (rdev->raid_disk < 0)
3421 continue;
3422 if (rdev->new_raid_disk == rdev->raid_disk)
3423 continue;
3424 rdev->raid_disk = rdev->new_raid_disk;
3425 if (rdev->raid_disk < 0)
3a981b03 3426 clear_bit(In_sync, &rdev->flags);
e93f68a1 3427 else {
36fad858
NK		 3428 if (sysfs_link_rdev(mddev, rdev))
3429 printk(KERN_WARNING "md: cannot register rd%d"
3430 " for %s after level change\n",
3431 rdev->raid_disk, mdname(mddev));
3a981b03 3432 }
e93f68a1
N		 3433 }
3434
3435 module_put(mddev->pers->owner);
245f46c2
N		 3436 mddev->pers = pers;
3437 mddev->private = priv;
3438 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3439 mddev->level = mddev->new_level;
3440 mddev->layout = mddev->new_layout;
664e7c41 3441 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3442 mddev->delta_disks = 0;
fee68723 3443 mddev->degraded = 0;
9af204cf
TM		 3444 if (mddev->pers->sync_request == NULL) {
3445 /* this is now an array without redundancy, so
3446 * it must always be in_sync
3447 */
3448 mddev->in_sync = 1;
3449 del_timer_sync(&mddev->safemode_timer);
3450 }
245f46c2
N		 3451 pers->run(mddev);
3452 mddev_resume(mddev);
3453 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3454 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3455 md_wakeup_thread(mddev->thread);
5cac7861 3456 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3457 md_new_event(mddev);
d9d166c2
N		 3458 return rv;
3459}
3460
3461static struct md_sysfs_entry md_level =
80ca3a44 3462__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3463
d4dbd025
N		 3464
3465static ssize_t
fd01b88c 3466layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3467{
3468 /* just a number, not meaningful for all levels */
08a02ecd
N		 3469 if (mddev->reshape_position != MaxSector &&
3470 mddev->layout != mddev->new_layout)
3471 return sprintf(page, "%d (%d)\n",
3472 mddev->new_layout, mddev->layout);
d4dbd025
N		 3473 return sprintf(page, "%d\n", mddev->layout);
3474}
3475
3476static ssize_t
fd01b88c 3477layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3478{
3479 char *e;
3480 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3481
3482 if (!*buf || (*e && *e != '\n'))
3483 return -EINVAL;
3484
b3546035
N		 3485 if (mddev->pers) {
3486 int err;
50ac168a 3487 if (mddev->pers->check_reshape == NULL)
b3546035 3488 return -EBUSY;
597a711b 3489 mddev->new_layout = n;
50ac168a 3490 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3491 if (err) {
3492 mddev->new_layout = mddev->layout;
b3546035 3493 return err;
597a711b 3494 }
b3546035 3495 } else {
08a02ecd 3496 mddev->new_layout = n;
b3546035
N		 3497 if (mddev->reshape_position == MaxSector)
3498 mddev->layout = n;
3499 }
d4dbd025
N		 3500 return len;
3501}
3502static struct md_sysfs_entry md_layout =
80ca3a44 3503__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3504
3505
eae1701f 3506static ssize_t
fd01b88c 3507raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3508{
bb636547
N		 3509 if (mddev->raid_disks == 0)
3510 return 0;
08a02ecd
N		 3511 if (mddev->reshape_position != MaxSector &&
3512 mddev->delta_disks != 0)
3513 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3514 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3515 return sprintf(page, "%d\n", mddev->raid_disks);
3516}
3517
fd01b88c 3518static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3519
3520static ssize_t
fd01b88c 3521raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3522{
da943b99
N		 3523 char *e;
3524 int rv = 0;
3525 unsigned long n = simple_strtoul(buf, &e, 10);
3526
3527 if (!*buf || (*e && *e != '\n'))
3528 return -EINVAL;
3529
3530 if (mddev->pers)
3531 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3532 else if (mddev->reshape_position != MaxSector) {
3533 int olddisks = mddev->raid_disks - mddev->delta_disks;
3534 mddev->delta_disks = n - olddisks;
3535 mddev->raid_disks = n;
3536 } else
da943b99
N		 3537 mddev->raid_disks = n;
3538 return rv ? rv : len;
3539}
3540static struct md_sysfs_entry md_raid_disks =
80ca3a44 3541__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3542
3b34380a 3543static ssize_t
fd01b88c 3544chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3545{
08a02ecd 3546 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3547 mddev->chunk_sectors != mddev->new_chunk_sectors)
3548 return sprintf(page, "%d (%d)\n",
3549 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3550 mddev->chunk_sectors << 9);
3551 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3552}
3553
3554static ssize_t
fd01b88c 3555chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3556{
3b34380a
N		 3557 char *e;
3558 unsigned long n = simple_strtoul(buf, &e, 10);
3559
3b34380a
N		 3560 if (!*buf || (*e && *e != '\n'))
3561 return -EINVAL;
3562
b3546035
N		 3563 if (mddev->pers) {
3564 int err;
50ac168a 3565 if (mddev->pers->check_reshape == NULL)
b3546035 3566 return -EBUSY;
597a711b 3567 mddev->new_chunk_sectors = n >> 9;
50ac168a 3568 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3569 if (err) {
3570 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3571 return err;
597a711b 3572 }
b3546035 3573 } else {
664e7c41 3574 mddev->new_chunk_sectors = n >> 9;
b3546035 3575 if (mddev->reshape_position == MaxSector)
9d8f0363 3576 mddev->chunk_sectors = n >> 9;
b3546035 3577 }
3b34380a
N		 3578 return len;
3579}
3580static struct md_sysfs_entry md_chunk_size =
80ca3a44 3581__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3582
a94213b1 3583static ssize_t
fd01b88c 3584resync_start_show(struct mddev *mddev, char *page)
a94213b1 3585{
d1a7c503
N		 3586 if (mddev->recovery_cp == MaxSector)
3587 return sprintf(page, "none\n");
a94213b1
N		 3588 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3589}
3590
3591static ssize_t
fd01b88c 3592resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3593{
a94213b1
N		 3594 char *e;
3595 unsigned long long n = simple_strtoull(buf, &e, 10);
3596
b098636c 3597 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3598 return -EBUSY;
06e3c817
DW		 3599 if (cmd_match(buf, "none"))
3600 n = MaxSector;
3601 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3602 return -EINVAL;
3603
3604 mddev->recovery_cp = n;
3605 return len;
3606}
3607static struct md_sysfs_entry md_resync_start =
80ca3a44 3608__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3609
9e653b63
N		 3610/*
3611 * The array state can be:
3612 *
3613 * clear
3614 * No devices, no size, no level
3615 * Equivalent to STOP_ARRAY ioctl
3616 * inactive
3617 * May have some settings, but array is not active
3618 * all IO results in error
3619 * When written, doesn't tear down array, but just stops it
3620 * suspended (not supported yet)
3621 * All IO requests will block. The array can be reconfigured.
910d8cb3 3622 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3623 * readonly
3624 * no resync can happen. no superblocks get written.
3625 * write requests fail
3626 * read-auto
3627 * like readonly, but behaves like 'clean' on a write request.
3628 *
3629 * clean - no pending writes, but otherwise active.
3630 * When written to inactive array, starts without resync
3631 * If a write request arrives then
3632 * if metadata is known, mark 'dirty' and switch to 'active'.
3633 * if not known, block and switch to write-pending
3634 * If written to an active array that has pending writes, then fails.
3635 * active
3636 * fully active: IO and resync can be happening.
3637 * When written to inactive array, starts with resync
3638 *
3639 * write-pending
3640 * clean, but writes are blocked waiting for 'active' to be written.
3641 *
3642 * active-idle
3643 * like active, but no writes have been seen for a while (100msec).
3644 *
3645 */
3646enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3647 write_pending, active_idle, bad_word};
05381954 3648static char *array_states[] = {
9e653b63
N		 3649 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3650 "write-pending", "active-idle", NULL };
3651
3652static int match_word(const char *word, char **list)
3653{
3654 int n;
3655 for (n=0; list[n]; n++)
3656 if (cmd_match(word, list[n]))
3657 break;
3658 return n;
3659}
3660
3661static ssize_t
fd01b88c 3662array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3663{
3664 enum array_state st = inactive;
3665
3666 if (mddev->pers)
3667 switch(mddev->ro) {
3668 case 1:
3669 st = readonly;
3670 break;
3671 case 2:
3672 st = read_auto;
3673 break;
3674 case 0:
3675 if (mddev->in_sync)
3676 st = clean;
070dc6dd 3677 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3678 st = write_pending;
9e653b63
N		 3679 else if (mddev->safemode)
3680 st = active_idle;
3681 else
3682 st = active;
3683 }
3684 else {
3685 if (list_empty(&mddev->disks) &&
3686 mddev->raid_disks == 0 &&
58c0fed4 3687 mddev->dev_sectors == 0)
9e653b63
N		 3688 st = clear;
3689 else
3690 st = inactive;
3691 }
3692 return sprintf(page, "%s\n", array_states[st]);
3693}
3694
fd01b88c
N		 3695static int do_md_stop(struct mddev * mddev, int ro, int is_open);
3696static int md_set_readonly(struct mddev * mddev, int is_open);
3697static int do_md_run(struct mddev * mddev);
3698static int restart_array(struct mddev *mddev);
9e653b63
N		 3699
3700static ssize_t
fd01b88c 3701array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N		 3702{
3703 int err = -EINVAL;
3704 enum array_state st = match_word(buf, array_states);
3705 switch(st) {
3706 case bad_word:
3707 break;
3708 case clear:
3709 /* stopping an active array */
f2ea68cf 3710 if (atomic_read(&mddev->openers) > 0)
e691063a 3711 return -EBUSY;
df5b20cf 3712 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3713 break;
3714 case inactive:
3715 /* stopping an active array */
3716 if (mddev->pers) {
f2ea68cf 3717 if (atomic_read(&mddev->openers) > 0)
9e653b63 3718 return -EBUSY;
df5b20cf 3719 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3720 } else
3721 err = 0; /* already inactive */
9e653b63
N		 3722 break;
3723 case suspended:
3724 break; /* not supported yet */
3725 case readonly:
3726 if (mddev->pers)
a4bd82d0 3727 err = md_set_readonly(mddev, 0);
9e653b63
N		 3728 else {
3729 mddev->ro = 1;
648b629e 3730 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3731 err = do_md_run(mddev);
3732 }
3733 break;
3734 case read_auto:
9e653b63 3735 if (mddev->pers) {
80268ee9 3736 if (mddev->ro == 0)
a4bd82d0 3737 err = md_set_readonly(mddev, 0);
80268ee9 3738 else if (mddev->ro == 1)
648b629e
N		 3739 err = restart_array(mddev);
3740 if (err == 0) {
3741 mddev->ro = 2;
3742 set_disk_ro(mddev->gendisk, 0);
3743 }
9e653b63
N		 3744 } else {
3745 mddev->ro = 2;
3746 err = do_md_run(mddev);
3747 }
3748 break;
3749 case clean:
3750 if (mddev->pers) {
3751 restart_array(mddev);
3752 spin_lock_irq(&mddev->write_lock);
3753 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3754 if (mddev->in_sync == 0) {
3755 mddev->in_sync = 1;
31a59e34
N		 3756 if (mddev->safemode == 1)
3757 mddev->safemode = 0;
070dc6dd 3758 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3759 }
3760 err = 0;
3761 } else
3762 err = -EBUSY;
9e653b63 3763 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3764 } else
3765 err = -EINVAL;
9e653b63
N		 3766 break;
3767 case active:
3768 if (mddev->pers) {
3769 restart_array(mddev);
070dc6dd 3770 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3771 wake_up(&mddev->sb_wait);
3772 err = 0;
3773 } else {
3774 mddev->ro = 0;
648b629e 3775 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3776 err = do_md_run(mddev);
3777 }
3778 break;
3779 case write_pending:
3780 case active_idle:
3781 /* these cannot be set */
3782 break;
3783 }
3784 if (err)
3785 return err;
0fd62b86 3786 else {
1d23f178
N		 3787 if (mddev->hold_active == UNTIL_IOCTL)
3788 mddev->hold_active = 0;
00bcb4ac 3789 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3790 return len;
0fd62b86 3791 }
9e653b63 3792}
80ca3a44
N		 3793static struct md_sysfs_entry md_array_state =
3794__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3795
1e50915f 3796static ssize_t
fd01b88c 3797max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3798 return sprintf(page, "%d\n",
3799 atomic_read(&mddev->max_corr_read_errors));
3800}
3801
3802static ssize_t
fd01b88c 3803max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3804{
3805 char *e;
3806 unsigned long n = simple_strtoul(buf, &e, 10);
3807
3808 if (*buf && (*e == 0 || *e == '\n')) {
3809 atomic_set(&mddev->max_corr_read_errors, n);
3810 return len;
3811 }
3812 return -EINVAL;
3813}
3814
3815static struct md_sysfs_entry max_corr_read_errors =
3816__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3817 max_corrected_read_errors_store);
3818
6d7ff738 3819static ssize_t
fd01b88c 3820null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3821{
3822 return -EINVAL;
3823}
3824
3825static ssize_t
fd01b88c 3826new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3827{
3828 /* buf must be %d:%d\n? giving major and minor numbers */
3829 /* The new device is added to the array.
3830 * If the array has a persistent superblock, we read the
3831 * superblock to initialise info and check validity.
3832 * Otherwise, only checking done is that in bind_rdev_to_array,
3833 * which mainly checks size.
3834 */
3835 char *e;
3836 int major = simple_strtoul(buf, &e, 10);
3837 int minor;
3838 dev_t dev;
3cb03002 3839 struct md_rdev *rdev;
6d7ff738
N		 3840 int err;
3841
3842 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3843 return -EINVAL;
3844 minor = simple_strtoul(e+1, &e, 10);
3845 if (*e && *e != '\n')
3846 return -EINVAL;
3847 dev = MKDEV(major, minor);
3848 if (major != MAJOR(dev) ||
3849 minor != MINOR(dev))
3850 return -EOVERFLOW;
3851
3852
3853 if (mddev->persistent) {
3854 rdev = md_import_device(dev, mddev->major_version,
3855 mddev->minor_version);
3856 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 3857 struct md_rdev *rdev0
3858 = list_entry(mddev->disks.next,
3859 struct md_rdev, same_set);
6d7ff738
N		 3860 err = super_types[mddev->major_version]
3861 .load_super(rdev, rdev0, mddev->minor_version);
3862 if (err < 0)
3863 goto out;
3864 }
c5d79adb
N		 3865 } else if (mddev->external)
3866 rdev = md_import_device(dev, -2, -1);
3867 else
6d7ff738
N		 3868 rdev = md_import_device(dev, -1, -1);
3869
3870 if (IS_ERR(rdev))
3871 return PTR_ERR(rdev);
3872 err = bind_rdev_to_array(rdev, mddev);
3873 out:
3874 if (err)
3875 export_rdev(rdev);
3876 return err ? err : len;
3877}
3878
3879static struct md_sysfs_entry md_new_device =
80ca3a44 3880__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3881
9b1d1dac 3882static ssize_t
fd01b88c 3883bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 3884{
3885 char *end;
3886 unsigned long chunk, end_chunk;
3887
3888 if (!mddev->bitmap)
3889 goto out;
3890 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3891 while (*buf) {
3892 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3893 if (buf == end) break;
3894 if (*end == '-') { /* range */
3895 buf = end + 1;
3896 end_chunk = simple_strtoul(buf, &end, 0);
3897 if (buf == end) break;
3898 }
3899 if (*end && !isspace(*end)) break;
3900 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3901 buf = skip_spaces(end);
9b1d1dac
PC		 3902 }
3903 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3904out:
3905 return len;
3906}
3907
3908static struct md_sysfs_entry md_bitmap =
3909__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3910
a35b0d69 3911static ssize_t
fd01b88c 3912size_show(struct mddev *mddev, char *page)
a35b0d69 3913{
58c0fed4
AN		 3914 return sprintf(page, "%llu\n",
3915 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3916}
3917
fd01b88c 3918static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 3919
3920static ssize_t
fd01b88c 3921size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 3922{
3923 /* If array is inactive, we can reduce the component size, but
3924 * not increase it (except from 0).
3925 * If array is active, we can try an on-line resize
3926 */
b522adcd
DW		 3927 sector_t sectors;
3928 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3929
58c0fed4
AN		 3930 if (err < 0)
3931 return err;
a35b0d69 3932 if (mddev->pers) {
58c0fed4 3933 err = update_size(mddev, sectors);
850b2b42 3934 md_update_sb(mddev, 1);
a35b0d69 3935 } else {
58c0fed4
AN		 3936 if (mddev->dev_sectors == 0 ||
3937 mddev->dev_sectors > sectors)
3938 mddev->dev_sectors = sectors;
a35b0d69
N		 3939 else
3940 err = -ENOSPC;
3941 }
3942 return err ? err : len;
3943}
3944
3945static struct md_sysfs_entry md_size =
80ca3a44 3946__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3947
8bb93aac
N		 3948
3949/* Metdata version.
e691063a
N		 3950 * This is one of
3951 * 'none' for arrays with no metadata (good luck...)
3952 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 3953 * or N.M for internally known formats
3954 */
3955static ssize_t
fd01b88c 3956metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 3957{
3958 if (mddev->persistent)
3959 return sprintf(page, "%d.%d\n",
3960 mddev->major_version, mddev->minor_version);
e691063a
N		 3961 else if (mddev->external)
3962 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 3963 else
3964 return sprintf(page, "none\n");
3965}
3966
3967static ssize_t
fd01b88c 3968metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 3969{
3970 int major, minor;
3971 char *e;
ea43ddd8
N		 3972 /* Changing the details of 'external' metadata is
3973 * always permitted. Otherwise there must be
3974 * no devices attached to the array.
3975 */
3976 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3977 ;
3978 else if (!list_empty(&mddev->disks))
8bb93aac
N		 3979 return -EBUSY;
3980
3981 if (cmd_match(buf, "none")) {
3982 mddev->persistent = 0;
e691063a
N		 3983 mddev->external = 0;
3984 mddev->major_version = 0;
3985 mddev->minor_version = 90;
3986 return len;
3987 }
3988 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 3989 size_t namelen = len-9;
e691063a
N		 3990 if (namelen >= sizeof(mddev->metadata_type))
3991 namelen = sizeof(mddev->metadata_type)-1;
3992 strncpy(mddev->metadata_type, buf+9, namelen);
3993 mddev->metadata_type[namelen] = 0;
3994 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3995 mddev->metadata_type[--namelen] = 0;
3996 mddev->persistent = 0;
3997 mddev->external = 1;
8bb93aac
N		 3998 mddev->major_version = 0;
3999 mddev->minor_version = 90;
4000 return len;
4001 }
4002 major = simple_strtoul(buf, &e, 10);
4003 if (e==buf || *e != '.')
4004 return -EINVAL;
4005 buf = e+1;
4006 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4007 if (e==buf || (*e && *e != '\n') )
8bb93aac 4008 return -EINVAL;
50511da3 4009 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 4010 return -ENOENT;
4011 mddev->major_version = major;
4012 mddev->minor_version = minor;
4013 mddev->persistent = 1;
e691063a 4014 mddev->external = 0;
8bb93aac
N		 4015 return len;
4016}
4017
4018static struct md_sysfs_entry md_metadata =
80ca3a44 4019__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4020
24dd469d 4021static ssize_t
fd01b88c 4022action_show(struct mddev *mddev, char *page)
24dd469d 4023{
7eec314d 4024 char *type = "idle";
b6a9ce68
N		 4025 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4026 type = "frozen";
4027 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4028 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 4029 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4030 type = "reshape";
4031 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 4032 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4033 type = "resync";
4034 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4035 type = "check";
4036 else
4037 type = "repair";
72a23c21 4038 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 4039 type = "recover";
4040 }
4041 return sprintf(page, "%s\n", type);
4042}
4043
fd01b88c 4044static void reap_sync_thread(struct mddev *mddev);
7ebc0be7 4045
24dd469d 4046static ssize_t
fd01b88c 4047action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4048{
7eec314d
N		 4049 if (!mddev->pers || !mddev->pers->sync_request)
4050 return -EINVAL;
4051
b6a9ce68
N		 4052 if (cmd_match(page, "frozen"))
4053 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4054 else
4055 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4056
4057 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 4058 if (mddev->sync_thread) {
4059 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4060 reap_sync_thread(mddev);
7eec314d 4061 }
03c902e1
N		 4062 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4063 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4064 return -EBUSY;
72a23c21
NB		 4065 else if (cmd_match(page, "resync"))
4066 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4067 else if (cmd_match(page, "recover")) {
4068 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4069 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4070 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4071 int err;
4072 if (mddev->pers->start_reshape == NULL)
4073 return -EINVAL;
4074 err = mddev->pers->start_reshape(mddev);
4075 if (err)
4076 return err;
a99ac971 4077 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4078 } else {
bce74dac 4079 if (cmd_match(page, "check"))
7eec314d 4080 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4081 else if (!cmd_match(page, "repair"))
7eec314d
N		 4082 return -EINVAL;
4083 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4084 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4085 }
03c902e1 4086 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4087 md_wakeup_thread(mddev->thread);
00bcb4ac 4088 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4089 return len;
4090}
4091
9d88883e 4092static ssize_t
fd01b88c 4093mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4094{
4095 return sprintf(page, "%llu\n",
4096 (unsigned long long) mddev->resync_mismatches);
4097}
4098
80ca3a44
N		 4099static struct md_sysfs_entry md_scan_mode =
4100__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 4101
96de1e66 4102
80ca3a44 4103static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4104
88202a0c 4105static ssize_t
fd01b88c 4106sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4107{
4108 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4109 mddev->sync_speed_min ? "local": "system");
4110}
4111
4112static ssize_t
fd01b88c 4113sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4114{
4115 int min;
4116 char *e;
4117 if (strncmp(buf, "system", 6)==0) {
4118 mddev->sync_speed_min = 0;
4119 return len;
4120 }
4121 min = simple_strtoul(buf, &e, 10);
4122 if (buf == e || (*e && *e != '\n') || min <= 0)
4123 return -EINVAL;
4124 mddev->sync_speed_min = min;
4125 return len;
4126}
4127
4128static struct md_sysfs_entry md_sync_min =
4129__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4130
4131static ssize_t
fd01b88c 4132sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4133{
4134 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4135 mddev->sync_speed_max ? "local": "system");
4136}
4137
4138static ssize_t
fd01b88c 4139sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4140{
4141 int max;
4142 char *e;
4143 if (strncmp(buf, "system", 6)==0) {
4144 mddev->sync_speed_max = 0;
4145 return len;
4146 }
4147 max = simple_strtoul(buf, &e, 10);
4148 if (buf == e || (*e && *e != '\n') || max <= 0)
4149 return -EINVAL;
4150 mddev->sync_speed_max = max;
4151 return len;
4152}
4153
4154static struct md_sysfs_entry md_sync_max =
4155__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4156
d7f3d291 4157static ssize_t
fd01b88c 4158degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4159{
4160 return sprintf(page, "%d\n", mddev->degraded);
4161}
4162static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4163
90b08710 4164static ssize_t
fd01b88c 4165sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4166{
4167 return sprintf(page, "%d\n", mddev->parallel_resync);
4168}
4169
4170static ssize_t
fd01b88c 4171sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4172{
4173 long n;
4174
4175 if (strict_strtol(buf, 10, &n))
4176 return -EINVAL;
4177
4178 if (n != 0 && n != 1)
4179 return -EINVAL;
4180
4181 mddev->parallel_resync = n;
4182
4183 if (mddev->sync_thread)
4184 wake_up(&resync_wait);
4185
4186 return len;
4187}
4188
4189/* force parallel resync, even with shared block devices */
4190static struct md_sysfs_entry md_sync_force_parallel =
4191__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4192 sync_force_parallel_show, sync_force_parallel_store);
4193
88202a0c 4194static ssize_t
fd01b88c 4195sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4196{
4197 unsigned long resync, dt, db;
d1a7c503
N		 4198 if (mddev->curr_resync == 0)
4199 return sprintf(page, "none\n");
9687a60c
AN		 4200 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4201 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4202 if (!dt) dt++;
9687a60c
AN		 4203 db = resync - mddev->resync_mark_cnt;
4204 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4205}
4206
80ca3a44 4207static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4208
4209static ssize_t
fd01b88c 4210sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4211{
13ae864b 4212 unsigned long long max_sectors, resync;
88202a0c 4213
acb180b0
N		 4214 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4215 return sprintf(page, "none\n");
4216
88202a0c 4217 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 4218 max_sectors = mddev->resync_max_sectors;
88202a0c 4219 else
58c0fed4 4220 max_sectors = mddev->dev_sectors;
88202a0c 4221
acb180b0 4222 resync = mddev->curr_resync_completed;
13ae864b 4223 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4224}
4225
80ca3a44 4226static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4227
5e96ee65 4228static ssize_t
fd01b88c 4229min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4230{
4231 return sprintf(page, "%llu\n",
4232 (unsigned long long)mddev->resync_min);
4233}
4234static ssize_t
fd01b88c 4235min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4236{
4237 unsigned long long min;
4238 if (strict_strtoull(buf, 10, &min))
4239 return -EINVAL;
4240 if (min > mddev->resync_max)
4241 return -EINVAL;
4242 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4243 return -EBUSY;
4244
4245 /* Must be a multiple of chunk_size */
9d8f0363 4246 if (mddev->chunk_sectors) {
2ac06c33 4247 sector_t temp = min;
9d8f0363 4248 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 4249 return -EINVAL;
4250 }
4251 mddev->resync_min = min;
4252
4253 return len;
4254}
4255
4256static struct md_sysfs_entry md_min_sync =
4257__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4258
c6207277 4259static ssize_t
fd01b88c 4260max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4261{
4262 if (mddev->resync_max == MaxSector)
4263 return sprintf(page, "max\n");
4264 else
4265 return sprintf(page, "%llu\n",
4266 (unsigned long long)mddev->resync_max);
4267}
4268static ssize_t
fd01b88c 4269max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N		 4270{
4271 if (strncmp(buf, "max", 3) == 0)
4272 mddev->resync_max = MaxSector;
4273 else {
5e96ee65
NB		 4274 unsigned long long max;
4275 if (strict_strtoull(buf, 10, &max))
4276 return -EINVAL;
4277 if (max < mddev->resync_min)
c6207277
N		 4278 return -EINVAL;
4279 if (max < mddev->resync_max &&
4d484a4a 4280 mddev->ro == 0 &&
c6207277
N		 4281 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4282 return -EBUSY;
4283
4284 /* Must be a multiple of chunk_size */
9d8f0363 4285 if (mddev->chunk_sectors) {
2ac06c33 4286 sector_t temp = max;
9d8f0363 4287 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4288 return -EINVAL;
4289 }
4290 mddev->resync_max = max;
4291 }
4292 wake_up(&mddev->recovery_wait);
4293 return len;
4294}
4295
4296static struct md_sysfs_entry md_max_sync =
4297__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4298
e464eafd 4299static ssize_t
fd01b88c 4300suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4301{
4302 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4303}
4304
4305static ssize_t
fd01b88c 4306suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4307{
4308 char *e;
4309 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4310 unsigned long long old = mddev->suspend_lo;
e464eafd 4311
b8d966ef
N		 4312 if (mddev->pers == NULL ||
4313 mddev->pers->quiesce == NULL)
e464eafd
N		 4314 return -EINVAL;
4315 if (buf == e || (*e && *e != '\n'))
4316 return -EINVAL;
23ddff37
N		 4317
4318 mddev->suspend_lo = new;
4319 if (new >= old)
4320 /* Shrinking suspended region */
e464eafd 4321 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4322 else {
4323 /* Expanding suspended region - need to wait */
4324 mddev->pers->quiesce(mddev, 1);
4325 mddev->pers->quiesce(mddev, 0);
4326 }
4327 return len;
e464eafd
N		 4328}
4329static struct md_sysfs_entry md_suspend_lo =
4330__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4331
4332
4333static ssize_t
fd01b88c 4334suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4335{
4336 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4337}
4338
4339static ssize_t
fd01b88c 4340suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4341{
4342 char *e;
4343 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4344 unsigned long long old = mddev->suspend_hi;
e464eafd 4345
b8d966ef
N		 4346 if (mddev->pers == NULL ||
4347 mddev->pers->quiesce == NULL)
e464eafd
N		 4348 return -EINVAL;
4349 if (buf == e || (*e && *e != '\n'))
4350 return -EINVAL;
23ddff37
N		 4351
4352 mddev->suspend_hi = new;
4353 if (new <= old)
4354 /* Shrinking suspended region */
4355 mddev->pers->quiesce(mddev, 2);
4356 else {
4357 /* Expanding suspended region - need to wait */
e464eafd
N		 4358 mddev->pers->quiesce(mddev, 1);
4359 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4360 }
4361 return len;
e464eafd
N		 4362}
4363static struct md_sysfs_entry md_suspend_hi =
4364__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4365
08a02ecd 4366static ssize_t
fd01b88c 4367reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4368{
4369 if (mddev->reshape_position != MaxSector)
4370 return sprintf(page, "%llu\n",
4371 (unsigned long long)mddev->reshape_position);
4372 strcpy(page, "none\n");
4373 return 5;
4374}
4375
4376static ssize_t
fd01b88c 4377reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd
N		 4378{
4379 char *e;
4380 unsigned long long new = simple_strtoull(buf, &e, 10);
4381 if (mddev->pers)
4382 return -EBUSY;
4383 if (buf == e || (*e && *e != '\n'))
4384 return -EINVAL;
4385 mddev->reshape_position = new;
4386 mddev->delta_disks = 0;
4387 mddev->new_level = mddev->level;
4388 mddev->new_layout = mddev->layout;
664e7c41 4389 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4390 return len;
4391}
4392
4393static struct md_sysfs_entry md_reshape_position =
4394__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4395 reshape_position_store);
4396
b522adcd 4397static ssize_t
fd01b88c 4398array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4399{
4400 if (mddev->external_size)
4401 return sprintf(page, "%llu\n",
4402 (unsigned long long)mddev->array_sectors/2);
4403 else
4404 return sprintf(page, "default\n");
4405}
4406
4407static ssize_t
fd01b88c 4408array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4409{
4410 sector_t sectors;
4411
4412 if (strncmp(buf, "default", 7) == 0) {
4413 if (mddev->pers)
4414 sectors = mddev->pers->size(mddev, 0, 0);
4415 else
4416 sectors = mddev->array_sectors;
4417
4418 mddev->external_size = 0;
4419 } else {
4420 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4421 return -EINVAL;
4422 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4423 return -E2BIG;
b522adcd
DW		 4424
4425 mddev->external_size = 1;
4426 }
4427
4428 mddev->array_sectors = sectors;
cbe6ef1d
N		 4429 if (mddev->pers) {
4430 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4431 revalidate_disk(mddev->gendisk);
cbe6ef1d 4432 }
b522adcd
DW		 4433 return len;
4434}
4435
4436static struct md_sysfs_entry md_array_size =
4437__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4438 array_size_store);
e464eafd 4439
eae1701f
N		 4440static struct attribute *md_default_attrs[] = {
4441 &md_level.attr,
d4dbd025 4442 &md_layout.attr,
eae1701f 4443 &md_raid_disks.attr,
3b34380a 4444 &md_chunk_size.attr,
a35b0d69 4445 &md_size.attr,
a94213b1 4446 &md_resync_start.attr,
8bb93aac 4447 &md_metadata.attr,
6d7ff738 4448 &md_new_device.attr,
16f17b39 4449 &md_safe_delay.attr,
9e653b63 4450 &md_array_state.attr,
08a02ecd 4451 &md_reshape_position.attr,
b522adcd 4452 &md_array_size.attr,
1e50915f 4453 &max_corr_read_errors.attr,
411036fa
N		 4454 NULL,
4455};
4456
4457static struct attribute *md_redundancy_attrs[] = {
24dd469d 4458 &md_scan_mode.attr,
9d88883e 4459 &md_mismatches.attr,
88202a0c
N		 4460 &md_sync_min.attr,
4461 &md_sync_max.attr,
4462 &md_sync_speed.attr,
90b08710 4463 &md_sync_force_parallel.attr,
88202a0c 4464 &md_sync_completed.attr,
5e96ee65 4465 &md_min_sync.attr,
c6207277 4466 &md_max_sync.attr,
e464eafd
N		 4467 &md_suspend_lo.attr,
4468 &md_suspend_hi.attr,
9b1d1dac 4469 &md_bitmap.attr,
d7f3d291 4470 &md_degraded.attr,
eae1701f
N		 4471 NULL,
4472};
411036fa
N		 4473static struct attribute_group md_redundancy_group = {
4474 .name = NULL,
4475 .attrs = md_redundancy_attrs,
4476};
4477
eae1701f
N		 4478
4479static ssize_t
4480md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4481{
4482 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4483 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4484 ssize_t rv;
eae1701f
N		 4485
4486 if (!entry->show)
4487 return -EIO;
af8a2434
N		 4488 spin_lock(&all_mddevs_lock);
4489 if (list_empty(&mddev->all_mddevs)) {
4490 spin_unlock(&all_mddevs_lock);
4491 return -EBUSY;
4492 }
4493 mddev_get(mddev);
4494 spin_unlock(&all_mddevs_lock);
4495
5dc5cf7d
IM		 4496 rv = mddev_lock(mddev);
4497 if (!rv) {
4498 rv = entry->show(mddev, page);
4499 mddev_unlock(mddev);
4500 }
af8a2434 4501 mddev_put(mddev);
96de1e66 4502 return rv;
eae1701f
N		 4503}
4504
4505static ssize_t
4506md_attr_store(struct kobject *kobj, struct attribute *attr,
4507 const char *page, size_t length)
4508{
4509 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4510 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4511 ssize_t rv;
eae1701f
N		 4512
4513 if (!entry->store)
4514 return -EIO;
67463acb
N		 4515 if (!capable(CAP_SYS_ADMIN))
4516 return -EACCES;
af8a2434
N		 4517 spin_lock(&all_mddevs_lock);
4518 if (list_empty(&mddev->all_mddevs)) {
4519 spin_unlock(&all_mddevs_lock);
4520 return -EBUSY;
4521 }
4522 mddev_get(mddev);
4523 spin_unlock(&all_mddevs_lock);
5dc5cf7d
IM		 4524 rv = mddev_lock(mddev);
4525 if (!rv) {
4526 rv = entry->store(mddev, page, length);
4527 mddev_unlock(mddev);
4528 }
af8a2434 4529 mddev_put(mddev);
96de1e66 4530 return rv;
eae1701f
N		 4531}
4532
4533static void md_free(struct kobject *ko)
4534{
fd01b88c 4535 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4536
4537 if (mddev->sysfs_state)
4538 sysfs_put(mddev->sysfs_state);
4539
4540 if (mddev->gendisk) {
4541 del_gendisk(mddev->gendisk);
4542 put_disk(mddev->gendisk);
4543 }
4544 if (mddev->queue)
4545 blk_cleanup_queue(mddev->queue);
4546
eae1701f
N		 4547 kfree(mddev);
4548}
4549
52cf25d0 4550static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4551 .show = md_attr_show,
4552 .store = md_attr_store,
4553};
4554static struct kobj_type md_ktype = {
4555 .release = md_free,
4556 .sysfs_ops = &md_sysfs_ops,
4557 .default_attrs = md_default_attrs,
4558};
4559
1da177e4
LT		 4560int mdp_major = 0;
4561
5fd3a17e
DW		 4562static void mddev_delayed_delete(struct work_struct *ws)
4563{
fd01b88c 4564 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4565
43a70507 4566 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4567 kobject_del(&mddev->kobj);
4568 kobject_put(&mddev->kobj);
4569}
4570
efeb53c0 4571static int md_alloc(dev_t dev, char *name)
1da177e4 4572{
48c9c27b 4573 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4574 struct mddev *mddev = mddev_find(dev);
1da177e4 4575 struct gendisk *disk;
efeb53c0
N		 4576 int partitioned;
4577 int shift;
4578 int unit;
3830c62f 4579 int error;
1da177e4
LT		 4580
4581 if (!mddev)
efeb53c0
N		 4582 return -ENODEV;
4583
4584 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4585 shift = partitioned ? MdpMinorShift : 0;
4586 unit = MINOR(mddev->unit) >> shift;
1da177e4 4587
e804ac78
TH		 4588 /* wait for any previous instance of this device to be
4589 * completely removed (mddev_delayed_delete).
d3374825 4590 */
e804ac78 4591 flush_workqueue(md_misc_wq);
d3374825 4592
48c9c27b 4593 mutex_lock(&disks_mutex);
0909dc44
N		 4594 error = -EEXIST;
4595 if (mddev->gendisk)
4596 goto abort;
efeb53c0
N		 4597
4598 if (name) {
4599 /* Need to ensure that 'name' is not a duplicate.
4600 */
fd01b88c 4601 struct mddev *mddev2;
efeb53c0
N		 4602 spin_lock(&all_mddevs_lock);
4603
4604 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4605 if (mddev2->gendisk &&
4606 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4607 spin_unlock(&all_mddevs_lock);
0909dc44 4608 goto abort;
efeb53c0
N		 4609 }
4610 spin_unlock(&all_mddevs_lock);
1da177e4 4611 }
8b765398 4612
0909dc44 4613 error = -ENOMEM;
8b765398 4614 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4615 if (!mddev->queue)
4616 goto abort;
409c57f3
N		 4617 mddev->queue->queuedata = mddev;
4618
409c57f3 4619 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4620
1da177e4
LT		 4621 disk = alloc_disk(1 << shift);
4622 if (!disk) {
8b765398
N		 4623 blk_cleanup_queue(mddev->queue);
4624 mddev->queue = NULL;
0909dc44 4625 goto abort;
1da177e4 4626 }
efeb53c0 4627 disk->major = MAJOR(mddev->unit);
1da177e4 4628 disk->first_minor = unit << shift;
efeb53c0
N		 4629 if (name)
4630 strcpy(disk->disk_name, name);
4631 else if (partitioned)
1da177e4 4632 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4633 else
1da177e4 4634 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4635 disk->fops = &md_fops;
4636 disk->private_data = mddev;
4637 disk->queue = mddev->queue;
b0140891 4638 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4639 /* Allow extended partitions. This makes the
d3374825 4640 * 'mdp' device redundant, but we can't really
92850bbd
N		 4641 * remove it now.
4642 */
4643 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4644 mddev->gendisk = disk;
b0140891
N		 4645 /* As soon as we call add_disk(), another thread could get
4646 * through to md_open, so make sure it doesn't get too far
4647 */
4648 mutex_lock(&mddev->open_mutex);
4649 add_disk(disk);
4650
ed9e1982
TH		 4651 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4652 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4653 if (error) {
4654 /* This isn't possible, but as kobject_init_and_add is marked
4655 * __must_check, we must do something with the result
4656 */
5e55e2f5
N		 4657 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4658 disk->disk_name);
0909dc44
N		 4659 error = 0;
4660 }
00bcb4ac
N		 4661 if (mddev->kobj.sd &&
4662 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4663 printk(KERN_DEBUG "pointless warning\n");
b0140891 4664 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4665 abort:
4666 mutex_unlock(&disks_mutex);
00bcb4ac 4667 if (!error && mddev->kobj.sd) {
3830c62f 4668 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4669 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4670 }
d3374825 4671 mddev_put(mddev);
0909dc44 4672 return error;
efeb53c0
N		 4673}
4674
4675static struct kobject *md_probe(dev_t dev, int *part, void *data)
4676{
4677 md_alloc(dev, NULL);
1da177e4
LT		 4678 return NULL;
4679}
4680
efeb53c0
N		 4681static int add_named_array(const char *val, struct kernel_param *kp)
4682{
4683 /* val must be "md_*" where * is not all digits.
4684 * We allocate an array with a large free minor number, and
4685 * set the name to val. val must not already be an active name.
4686 */
4687 int len = strlen(val);
4688 char buf[DISK_NAME_LEN];
4689
4690 while (len && val[len-1] == '\n')
4691 len--;
4692 if (len >= DISK_NAME_LEN)
4693 return -E2BIG;
4694 strlcpy(buf, val, len+1);
4695 if (strncmp(buf, "md_", 3) != 0)
4696 return -EINVAL;
4697 return md_alloc(0, buf);
4698}
4699
1da177e4
LT		 4700static void md_safemode_timeout(unsigned long data)
4701{
fd01b88c 4702 struct mddev *mddev = (struct mddev *) data;
1da177e4 4703
0fd62b86
NB		 4704 if (!atomic_read(&mddev->writes_pending)) {
4705 mddev->safemode = 1;
4706 if (mddev->external)
00bcb4ac 4707 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4708 }
1da177e4
LT		 4709 md_wakeup_thread(mddev->thread);
4710}
4711
6ff8d8ec 4712static int start_dirty_degraded;
1da177e4 4713
fd01b88c 4714int md_run(struct mddev *mddev)
1da177e4 4715{
2604b703 4716 int err;
3cb03002 4717 struct md_rdev *rdev;
84fc4b56 4718 struct md_personality *pers;
1da177e4 4719
a757e64c
N		 4720 if (list_empty(&mddev->disks))
4721 /* cannot run an array with no devices.. */
1da177e4 4722 return -EINVAL;
1da177e4
LT		 4723
4724 if (mddev->pers)
4725 return -EBUSY;
bb4f1e9d
N		 4726 /* Cannot run until previous stop completes properly */
4727 if (mddev->sysfs_active)
4728 return -EBUSY;
b6eb127d 4729
1da177e4
LT		 4730 /*
4731 * Analyze all RAID superblock(s)
4732 */
1ec4a939
N		 4733 if (!mddev->raid_disks) {
4734 if (!mddev->persistent)
4735 return -EINVAL;
a757e64c 4736 analyze_sbs(mddev);
1ec4a939 4737 }
1da177e4 4738
d9d166c2
N		 4739 if (mddev->level != LEVEL_NONE)
4740 request_module("md-level-%d", mddev->level);
4741 else if (mddev->clevel[0])
4742 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4743
4744 /*
4745 * Drop all container device buffers, from now on
4746 * the only valid external interface is through the md
4747 * device.
1da177e4 4748 */
159ec1fc 4749 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4750 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4751 continue;
4752 sync_blockdev(rdev->bdev);
f98393a6 4753 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4754
4755 /* perform some consistency tests on the device.
4756 * We don't want the data to overlap the metadata,
58c0fed4 4757 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4758 */
a6ff7e08
JB		 4759 if (rdev->meta_bdev) {
4760 /* Nothing to check */;
4761 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4762 if (mddev->dev_sectors &&
4763 rdev->data_offset + mddev->dev_sectors
0f420358 4764 > rdev->sb_start) {
f0d76d70
N		 4765 printk("md: %s: data overlaps metadata\n",
4766 mdname(mddev));
4767 return -EINVAL;
4768 }
4769 } else {
0f420358 4770 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4771 > rdev->data_offset) {
4772 printk("md: %s: metadata overlaps data\n",
4773 mdname(mddev));
4774 return -EINVAL;
4775 }
4776 }
00bcb4ac 4777 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4778 }
4779
a167f663 4780 if (mddev->bio_set == NULL)
a519b26d 4781 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
fd01b88c 4782 sizeof(struct mddev *));
a167f663 4783
1da177e4 4784 spin_lock(&pers_lock);
d9d166c2 4785 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4786 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4787 spin_unlock(&pers_lock);
d9d166c2
N		 4788 if (mddev->level != LEVEL_NONE)
4789 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4790 mddev->level);
4791 else
4792 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4793 mddev->clevel);
1da177e4
LT		 4794 return -EINVAL;
4795 }
2604b703 4796 mddev->pers = pers;
1da177e4 4797 spin_unlock(&pers_lock);
34817e8c
N		 4798 if (mddev->level != pers->level) {
4799 mddev->level = pers->level;
4800 mddev->new_level = pers->level;
4801 }
d9d166c2 4802 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4803
f6705578 4804 if (mddev->reshape_position != MaxSector &&
63c70c4f 4805 pers->start_reshape == NULL) {
f6705578
N		 4806 /* This personality cannot handle reshaping... */
4807 mddev->pers = NULL;
4808 module_put(pers->owner);
4809 return -EINVAL;
4810 }
4811
7dd5e7c3
N		 4812 if (pers->sync_request) {
4813 /* Warn if this is a potentially silly
4814 * configuration.
4815 */
4816 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 4817 struct md_rdev *rdev2;
7dd5e7c3 4818 int warned = 0;
159ec1fc
CR		 4819
4820 list_for_each_entry(rdev, &mddev->disks, same_set)
4821 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4822 if (rdev < rdev2 &&
4823 rdev->bdev->bd_contains ==
4824 rdev2->bdev->bd_contains) {
4825 printk(KERN_WARNING
4826 "%s: WARNING: %s appears to be"
4827 " on the same physical disk as"
4828 " %s.\n",
4829 mdname(mddev),
4830 bdevname(rdev->bdev,b),
4831 bdevname(rdev2->bdev,b2));
4832 warned = 1;
4833 }
4834 }
159ec1fc 4835
7dd5e7c3
N		 4836 if (warned)
4837 printk(KERN_WARNING
4838 "True protection against single-disk"
4839 " failure might be compromised.\n");
4840 }
4841
657390d2 4842 mddev->recovery = 0;
58c0fed4
AN		 4843 /* may be over-ridden by personality */
4844 mddev->resync_max_sectors = mddev->dev_sectors;
4845
6ff8d8ec 4846 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4847
0f9552b5 4848 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4849 mddev->ro = 2; /* read-only, but switch on first write */
4850
b15c2e57 4851 err = mddev->pers->run(mddev);
13e53df3
AN		 4852 if (err)
4853 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4854 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4855 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4856 " but 'external_size' not in effect?\n", __func__);
4857 printk(KERN_ERR
4858 "md: invalid array_size %llu > default size %llu\n",
4859 (unsigned long long)mddev->array_sectors / 2,
4860 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4861 err = -EINVAL;
4862 mddev->pers->stop(mddev);
4863 }
4864 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4865 err = bitmap_create(mddev);
4866 if (err) {
4867 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4868 mdname(mddev), err);
4869 mddev->pers->stop(mddev);
4870 }
4871 }
1da177e4 4872 if (err) {
1da177e4
LT		 4873 module_put(mddev->pers->owner);
4874 mddev->pers = NULL;
32a7627c
N		 4875 bitmap_destroy(mddev);
4876 return err;
1da177e4 4877 }
5e55e2f5 4878 if (mddev->pers->sync_request) {
00bcb4ac
N		 4879 if (mddev->kobj.sd &&
4880 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4881 printk(KERN_WARNING
4882 "md: cannot register extra attributes for %s\n",
4883 mdname(mddev));
00bcb4ac 4884 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4885 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4886 mddev->ro = 0;
4887
1da177e4 4888 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4889 atomic_set(&mddev->max_corr_read_errors,
4890 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4891 mddev->safemode = 0;
4892 mddev->safemode_timer.function = md_safemode_timeout;
4893 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4894 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4895 mddev->in_sync = 1;
0ca69886
N		 4896 smp_wmb();
4897 mddev->ready = 1;
159ec1fc 4898 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 4899 if (rdev->raid_disk >= 0)
4900 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 4901 /* failure here is OK */;
1da177e4
LT		 4902
4903 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4904
850b2b42
N		 4905 if (mddev->flags)
4906 md_update_sb(mddev, 0);
1da177e4 4907
d7603b7e 4908 md_new_event(mddev);
00bcb4ac
N		 4909 sysfs_notify_dirent_safe(mddev->sysfs_state);
4910 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4911 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4912 return 0;
4913}
390ee602 4914EXPORT_SYMBOL_GPL(md_run);
1da177e4 4915
fd01b88c 4916static int do_md_run(struct mddev *mddev)
fe60b014
N		 4917{
4918 int err;
4919
4920 err = md_run(mddev);
4921 if (err)
4922 goto out;
69e51b44
N		 4923 err = bitmap_load(mddev);
4924 if (err) {
4925 bitmap_destroy(mddev);
4926 goto out;
4927 }
0fd018af
JB		 4928
4929 md_wakeup_thread(mddev->thread);
4930 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4931
fe60b014
N		 4932 set_capacity(mddev->gendisk, mddev->array_sectors);
4933 revalidate_disk(mddev->gendisk);
f0b4f7e2 4934 mddev->changed = 1;
fe60b014
N		 4935 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4936out:
4937 return err;
4938}
4939
fd01b88c 4940static int restart_array(struct mddev *mddev)
1da177e4
LT		 4941{
4942 struct gendisk *disk = mddev->gendisk;
1da177e4 4943
80fab1d7 4944 /* Complain if it has no devices */
1da177e4 4945 if (list_empty(&mddev->disks))
80fab1d7
AN		 4946 return -ENXIO;
4947 if (!mddev->pers)
4948 return -EINVAL;
4949 if (!mddev->ro)
4950 return -EBUSY;
4951 mddev->safemode = 0;
4952 mddev->ro = 0;
4953 set_disk_ro(disk, 0);
4954 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4955 mdname(mddev));
4956 /* Kick recovery or resync if necessary */
4957 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4958 md_wakeup_thread(mddev->thread);
4959 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 4960 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 4961 return 0;
1da177e4
LT		 4962}
4963
acc55e22
N		 4964/* similar to deny_write_access, but accounts for our holding a reference
4965 * to the file ourselves */
4966static int deny_bitmap_write_access(struct file * file)
4967{
4968 struct inode *inode = file->f_mapping->host;
4969
4970 spin_lock(&inode->i_lock);
4971 if (atomic_read(&inode->i_writecount) > 1) {
4972 spin_unlock(&inode->i_lock);
4973 return -ETXTBSY;
4974 }
4975 atomic_set(&inode->i_writecount, -1);
4976 spin_unlock(&inode->i_lock);
4977
4978 return 0;
4979}
4980
43a70507 4981void restore_bitmap_write_access(struct file *file)
acc55e22
N		 4982{
4983 struct inode *inode = file->f_mapping->host;
4984
4985 spin_lock(&inode->i_lock);
4986 atomic_set(&inode->i_writecount, 1);
4987 spin_unlock(&inode->i_lock);
4988}
4989
fd01b88c 4990static void md_clean(struct mddev *mddev)
6177b472
N		 4991{
4992 mddev->array_sectors = 0;
4993 mddev->external_size = 0;
4994 mddev->dev_sectors = 0;
4995 mddev->raid_disks = 0;
4996 mddev->recovery_cp = 0;
4997 mddev->resync_min = 0;
4998 mddev->resync_max = MaxSector;
4999 mddev->reshape_position = MaxSector;
5000 mddev->external = 0;
5001 mddev->persistent = 0;
5002 mddev->level = LEVEL_NONE;
5003 mddev->clevel[0] = 0;
5004 mddev->flags = 0;
5005 mddev->ro = 0;
5006 mddev->metadata_type[0] = 0;
5007 mddev->chunk_sectors = 0;
5008 mddev->ctime = mddev->utime = 0;
5009 mddev->layout = 0;
5010 mddev->max_disks = 0;
5011 mddev->events = 0;
a8707c08 5012 mddev->can_decrease_events = 0;
6177b472
N		 5013 mddev->delta_disks = 0;
5014 mddev->new_level = LEVEL_NONE;
5015 mddev->new_layout = 0;
5016 mddev->new_chunk_sectors = 0;
5017 mddev->curr_resync = 0;
5018 mddev->resync_mismatches = 0;
5019 mddev->suspend_lo = mddev->suspend_hi = 0;
5020 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5021 mddev->recovery = 0;
5022 mddev->in_sync = 0;
f0b4f7e2 5023 mddev->changed = 0;
6177b472 5024 mddev->degraded = 0;
6177b472
N		 5025 mddev->safemode = 0;
5026 mddev->bitmap_info.offset = 0;
5027 mddev->bitmap_info.default_offset = 0;
5028 mddev->bitmap_info.chunksize = 0;
5029 mddev->bitmap_info.daemon_sleep = 0;
5030 mddev->bitmap_info.max_write_behind = 0;
5031}
5032
fd01b88c 5033static void __md_stop_writes(struct mddev *mddev)
a047e125
N		 5034{
5035 if (mddev->sync_thread) {
5036 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5037 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 5038 reap_sync_thread(mddev);
a047e125
N		 5039 }
5040
5041 del_timer_sync(&mddev->safemode_timer);
5042
5043 bitmap_flush(mddev);
5044 md_super_wait(mddev);
5045
5046 if (!mddev->in_sync || mddev->flags) {
5047 /* mark array as shutdown cleanly */
5048 mddev->in_sync = 1;
5049 md_update_sb(mddev, 1);
5050 }
5051}
defad61a 5052
fd01b88c 5053void md_stop_writes(struct mddev *mddev)
defad61a
N		 5054{
5055 mddev_lock(mddev);
5056 __md_stop_writes(mddev);
5057 mddev_unlock(mddev);
5058}
390ee602 5059EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5060
fd01b88c 5061void md_stop(struct mddev *mddev)
6177b472 5062{
0ca69886 5063 mddev->ready = 0;
6177b472
N		 5064 mddev->pers->stop(mddev);
5065 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5066 mddev->to_remove = &md_redundancy_group;
5067 module_put(mddev->pers->owner);
5068 mddev->pers = NULL;
cca9cf90 5069 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5070}
390ee602 5071EXPORT_SYMBOL_GPL(md_stop);
6177b472 5072
fd01b88c 5073static int md_set_readonly(struct mddev *mddev, int is_open)
a4bd82d0
N		 5074{
5075 int err = 0;
5076 mutex_lock(&mddev->open_mutex);
5077 if (atomic_read(&mddev->openers) > is_open) {
5078 printk("md: %s still in use.\n",mdname(mddev));
5079 err = -EBUSY;
5080 goto out;
5081 }
5082 if (mddev->pers) {
defad61a 5083 __md_stop_writes(mddev);
a4bd82d0
N		 5084
5085 err = -ENXIO;
5086 if (mddev->ro==1)
5087 goto out;
5088 mddev->ro = 1;
5089 set_disk_ro(mddev->gendisk, 1);
5090 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5091 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5092 err = 0;
5093 }
5094out:
5095 mutex_unlock(&mddev->open_mutex);
5096 return err;
5097}
5098
9e653b63
N		 5099/* mode:
5100 * 0 - completely stop and dis-assemble array
9e653b63
N		 5101 * 2 - stop but do not disassemble array
5102 */
fd01b88c 5103static int do_md_stop(struct mddev * mddev, int mode, int is_open)
1da177e4 5104{
1da177e4 5105 struct gendisk *disk = mddev->gendisk;
3cb03002 5106 struct md_rdev *rdev;
1da177e4 5107
c8c00a69 5108 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 5109 if (atomic_read(&mddev->openers) > is_open ||
5110 mddev->sysfs_active) {
df5b20cf 5111 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5112 mutex_unlock(&mddev->open_mutex);
5113 return -EBUSY;
5114 }
1da177e4 5115
6e17b027 5116 if (mddev->pers) {
a4bd82d0
N		 5117 if (mddev->ro)
5118 set_disk_ro(disk, 0);
409c57f3 5119
defad61a 5120 __md_stop_writes(mddev);
a4bd82d0
N		 5121 md_stop(mddev);
5122 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5123 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5124
a4bd82d0 5125 /* tell userspace to handle 'inactive' */
00bcb4ac 5126 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5127
a4bd82d0 5128 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 5129 if (rdev->raid_disk >= 0)
5130 sysfs_unlink_rdev(mddev, rdev);
c4647292 5131
a4bd82d0 5132 set_capacity(disk, 0);
6e17b027 5133 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5134 mddev->changed = 1;
a4bd82d0 5135 revalidate_disk(disk);
0d4ca600 5136
a4bd82d0
N		 5137 if (mddev->ro)
5138 mddev->ro = 0;
6e17b027
N		 5139 } else
5140 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5141 /*
5142 * Free resources if final stop
5143 */
9e653b63 5144 if (mode == 0) {
1da177e4
LT		 5145 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5146
978f946b 5147 bitmap_destroy(mddev);
c3d9714e
N		 5148 if (mddev->bitmap_info.file) {
5149 restore_bitmap_write_access(mddev->bitmap_info.file);
5150 fput(mddev->bitmap_info.file);
5151 mddev->bitmap_info.file = NULL;
978f946b 5152 }
c3d9714e 5153 mddev->bitmap_info.offset = 0;
978f946b 5154
1da177e4
LT		 5155 export_array(mddev);
5156
6177b472 5157 md_clean(mddev);
934d9c23 5158 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5159 if (mddev->hold_active == UNTIL_STOP)
5160 mddev->hold_active = 0;
a4bd82d0 5161 }
3f9d99c1 5162 blk_integrity_unregister(disk);
d7603b7e 5163 md_new_event(mddev);
00bcb4ac 5164 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5165 return 0;
1da177e4
LT		 5166}
5167
fdee8ae4 5168#ifndef MODULE
fd01b88c 5169static void autorun_array(struct mddev *mddev)
1da177e4 5170{
3cb03002 5171 struct md_rdev *rdev;
1da177e4
LT		 5172 int err;
5173
a757e64c 5174 if (list_empty(&mddev->disks))
1da177e4 5175 return;
1da177e4
LT		 5176
5177 printk(KERN_INFO "md: running: ");
5178
159ec1fc 5179 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 5180 char b[BDEVNAME_SIZE];
5181 printk("<%s>", bdevname(rdev->bdev,b));
5182 }
5183 printk("\n");
5184
d710e138 5185 err = do_md_run(mddev);
1da177e4
LT		 5186 if (err) {
5187 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 5188 do_md_stop(mddev, 0, 0);
1da177e4
LT		 5189 }
5190}
5191
5192/*
5193 * lets try to run arrays based on all disks that have arrived
5194 * until now. (those are in pending_raid_disks)
5195 *
5196 * the method: pick the first pending disk, collect all disks with
5197 * the same UUID, remove all from the pending list and put them into
5198 * the 'same_array' list. Then order this list based on superblock
5199 * update time (freshest comes first), kick out 'old' disks and
5200 * compare superblocks. If everything's fine then run it.
5201 *
5202 * If "unit" is allocated, then bump its reference count
5203 */
5204static void autorun_devices(int part)
5205{
3cb03002 5206 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5207 struct mddev *mddev;
1da177e4
LT		 5208 char b[BDEVNAME_SIZE];
5209
5210 printk(KERN_INFO "md: autorun ...\n");
5211 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5212 int unit;
1da177e4 5213 dev_t dev;
ad01c9e3 5214 LIST_HEAD(candidates);
1da177e4 5215 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5216 struct md_rdev, same_set);
1da177e4
LT		 5217
5218 printk(KERN_INFO "md: considering %s ...\n",
5219 bdevname(rdev0->bdev,b));
5220 INIT_LIST_HEAD(&candidates);
159ec1fc 5221 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5222 if (super_90_load(rdev, rdev0, 0) >= 0) {
5223 printk(KERN_INFO "md: adding %s ...\n",
5224 bdevname(rdev->bdev,b));
5225 list_move(&rdev->same_set, &candidates);
5226 }
5227 /*
5228 * now we have a set of devices, with all of them having
5229 * mostly sane superblocks. It's time to allocate the
5230 * mddev.
5231 */
e8703fe1
N		 5232 if (part) {
5233 dev = MKDEV(mdp_major,
5234 rdev0->preferred_minor << MdpMinorShift);
5235 unit = MINOR(dev) >> MdpMinorShift;
5236 } else {
5237 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5238 unit = MINOR(dev);
5239 }
5240 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5241 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5242 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5243 break;
5244 }
1da177e4
LT		 5245
5246 md_probe(dev, NULL, NULL);
5247 mddev = mddev_find(dev);
9bbbca3a
NB		 5248 if (!mddev || !mddev->gendisk) {
5249 if (mddev)
5250 mddev_put(mddev);
5251 printk(KERN_ERR
1da177e4
LT		 5252 "md: cannot allocate memory for md drive.\n");
5253 break;
5254 }
5255 if (mddev_lock(mddev))
5256 printk(KERN_WARNING "md: %s locked, cannot run\n",
5257 mdname(mddev));
5258 else if (mddev->raid_disks || mddev->major_version
5259 || !list_empty(&mddev->disks)) {
5260 printk(KERN_WARNING
5261 "md: %s already running, cannot run %s\n",
5262 mdname(mddev), bdevname(rdev0->bdev,b));
5263 mddev_unlock(mddev);
5264 } else {
5265 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5266 mddev->persistent = 1;
159ec1fc 5267 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5268 list_del_init(&rdev->same_set);
5269 if (bind_rdev_to_array(rdev, mddev))
5270 export_rdev(rdev);
5271 }
5272 autorun_array(mddev);
5273 mddev_unlock(mddev);
5274 }
5275 /* on success, candidates will be empty, on error
5276 * it won't...
5277 */
159ec1fc 5278 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5279 list_del_init(&rdev->same_set);
1da177e4 5280 export_rdev(rdev);
4b80991c 5281 }
1da177e4
LT		 5282 mddev_put(mddev);
5283 }
5284 printk(KERN_INFO "md: ... autorun DONE.\n");
5285}
fdee8ae4 5286#endif /* !MODULE */
1da177e4 5287
1da177e4
LT		 5288static int get_version(void __user * arg)
5289{
5290 mdu_version_t ver;
5291
5292 ver.major = MD_MAJOR_VERSION;
5293 ver.minor = MD_MINOR_VERSION;
5294 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5295
5296 if (copy_to_user(arg, &ver, sizeof(ver)))
5297 return -EFAULT;
5298
5299 return 0;
5300}
5301
fd01b88c 5302static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5303{
5304 mdu_array_info_t info;
a9f326eb 5305 int nr,working,insync,failed,spare;
3cb03002 5306 struct md_rdev *rdev;
1da177e4 5307
a9f326eb 5308 nr=working=insync=failed=spare=0;
159ec1fc 5309 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 5310 nr++;
b2d444d7 5311 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5312 failed++;
5313 else {
5314 working++;
b2d444d7 5315 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5316 insync++;
1da177e4
LT		 5317 else
5318 spare++;
5319 }
5320 }
5321
5322 info.major_version = mddev->major_version;
5323 info.minor_version = mddev->minor_version;
5324 info.patch_version = MD_PATCHLEVEL_VERSION;
5325 info.ctime = mddev->ctime;
5326 info.level = mddev->level;
58c0fed4
AN		 5327 info.size = mddev->dev_sectors / 2;
5328 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5329 info.size = -1;
1da177e4
LT		 5330 info.nr_disks = nr;
5331 info.raid_disks = mddev->raid_disks;
5332 info.md_minor = mddev->md_minor;
5333 info.not_persistent= !mddev->persistent;
5334
5335 info.utime = mddev->utime;
5336 info.state = 0;
5337 if (mddev->in_sync)
5338 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5339 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5340 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5341 info.active_disks = insync;
1da177e4
LT		 5342 info.working_disks = working;
5343 info.failed_disks = failed;
5344 info.spare_disks = spare;
5345
5346 info.layout = mddev->layout;
9d8f0363 5347 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5348
5349 if (copy_to_user(arg, &info, sizeof(info)))
5350 return -EFAULT;
5351
5352 return 0;
5353}
5354
fd01b88c 5355static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5356{
5357 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5358 char *ptr, *buf = NULL;
5359 int err = -ENOMEM;
5360
b5470dc5
DW		 5361 if (md_allow_write(mddev))
5362 file = kmalloc(sizeof(*file), GFP_NOIO);
5363 else
5364 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5365
32a7627c
N		 5366 if (!file)
5367 goto out;
5368
5369 /* bitmap disabled, zero the first byte and copy out */
5370 if (!mddev->bitmap || !mddev->bitmap->file) {
5371 file->pathname[0] = '\0';
5372 goto copy_out;
5373 }
5374
5375 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5376 if (!buf)
5377 goto out;
5378
6bcfd601
CH		 5379 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5380 if (IS_ERR(ptr))
32a7627c
N		 5381 goto out;
5382
5383 strcpy(file->pathname, ptr);
5384
5385copy_out:
5386 err = 0;
5387 if (copy_to_user(arg, file, sizeof(*file)))
5388 err = -EFAULT;
5389out:
5390 kfree(buf);
5391 kfree(file);
5392 return err;
5393}
5394
fd01b88c 5395static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5396{
5397 mdu_disk_info_t info;
3cb03002 5398 struct md_rdev *rdev;
1da177e4
LT		 5399
5400 if (copy_from_user(&info, arg, sizeof(info)))
5401 return -EFAULT;
5402
26ef379f 5403 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5404 if (rdev) {
5405 info.major = MAJOR(rdev->bdev->bd_dev);
5406 info.minor = MINOR(rdev->bdev->bd_dev);
5407 info.raid_disk = rdev->raid_disk;
5408 info.state = 0;
b2d444d7 5409 if (test_bit(Faulty, &rdev->flags))
1da177e4 5410 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5411 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5412 info.state |= (1<<MD_DISK_ACTIVE);
5413 info.state |= (1<<MD_DISK_SYNC);
5414 }
8ddf9efe
N		 5415 if (test_bit(WriteMostly, &rdev->flags))
5416 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5417 } else {
5418 info.major = info.minor = 0;
5419 info.raid_disk = -1;
5420 info.state = (1<<MD_DISK_REMOVED);
5421 }
5422
5423 if (copy_to_user(arg, &info, sizeof(info)))
5424 return -EFAULT;
5425
5426 return 0;
5427}
5428
fd01b88c 5429static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT		 5430{
5431 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5432 struct md_rdev *rdev;
1da177e4
LT		 5433 dev_t dev = MKDEV(info->major,info->minor);
5434
5435 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5436 return -EOVERFLOW;
5437
5438 if (!mddev->raid_disks) {
5439 int err;
5440 /* expecting a device which has a superblock */
5441 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5442 if (IS_ERR(rdev)) {
5443 printk(KERN_WARNING
5444 "md: md_import_device returned %ld\n",
5445 PTR_ERR(rdev));
5446 return PTR_ERR(rdev);
5447 }
5448 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5449 struct md_rdev *rdev0
5450 = list_entry(mddev->disks.next,
5451 struct md_rdev, same_set);
a9f326eb 5452 err = super_types[mddev->major_version]
1da177e4
LT		 5453 .load_super(rdev, rdev0, mddev->minor_version);
5454 if (err < 0) {
5455 printk(KERN_WARNING
5456 "md: %s has different UUID to %s\n",
5457 bdevname(rdev->bdev,b),
5458 bdevname(rdev0->bdev,b2));
5459 export_rdev(rdev);
5460 return -EINVAL;
5461 }
5462 }
5463 err = bind_rdev_to_array(rdev, mddev);
5464 if (err)
5465 export_rdev(rdev);
5466 return err;
5467 }
5468
5469 /*
5470 * add_new_disk can be used once the array is assembled
5471 * to add "hot spares". They must already have a superblock
5472 * written
5473 */
5474 if (mddev->pers) {
5475 int err;
5476 if (!mddev->pers->hot_add_disk) {
5477 printk(KERN_WARNING
5478 "%s: personality does not support diskops!\n",
5479 mdname(mddev));
5480 return -EINVAL;
5481 }
7b1e35f6
N		 5482 if (mddev->persistent)
5483 rdev = md_import_device(dev, mddev->major_version,
5484 mddev->minor_version);
5485 else
5486 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5487 if (IS_ERR(rdev)) {
5488 printk(KERN_WARNING
5489 "md: md_import_device returned %ld\n",
5490 PTR_ERR(rdev));
5491 return PTR_ERR(rdev);
5492 }
1a855a06 5493 /* set saved_raid_disk if appropriate */
41158c7e
N		 5494 if (!mddev->persistent) {
5495 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5496 info->raid_disk < mddev->raid_disks) {
41158c7e 5497 rdev->raid_disk = info->raid_disk;
bf572541
N		 5498 set_bit(In_sync, &rdev->flags);
5499 } else
41158c7e
N		 5500 rdev->raid_disk = -1;
5501 } else
5502 super_types[mddev->major_version].
5503 validate_super(mddev, rdev);
bedd86b7
N		 5504 if ((info->state & (1<<MD_DISK_SYNC)) &&
5505 (!test_bit(In_sync, &rdev->flags) ||
5506 rdev->raid_disk != info->raid_disk)) {
5507 /* This was a hot-add request, but events doesn't
5508 * match, so reject it.
5509 */
5510 export_rdev(rdev);
5511 return -EINVAL;
5512 }
5513
1a855a06
N		 5514 if (test_bit(In_sync, &rdev->flags))
5515 rdev->saved_raid_disk = rdev->raid_disk;
5516 else
5517 rdev->saved_raid_disk = -1;
41158c7e 5518
b2d444d7 5519 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5520 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5521 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5522 else
5523 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5524
1da177e4
LT		 5525 rdev->raid_disk = -1;
5526 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5527 if (!err && !mddev->pers->hot_remove_disk) {
5528 /* If there is hot_add_disk but no hot_remove_disk
5529 * then added disks for geometry changes,
5530 * and should be added immediately.
5531 */
5532 super_types[mddev->major_version].
5533 validate_super(mddev, rdev);
5534 err = mddev->pers->hot_add_disk(mddev, rdev);
5535 if (err)
5536 unbind_rdev_from_array(rdev);
5537 }
1da177e4
LT		 5538 if (err)
5539 export_rdev(rdev);
52664732 5540 else
00bcb4ac 5541 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5542
17571284 5543 md_update_sb(mddev, 1);
72a23c21
NB		 5544 if (mddev->degraded)
5545 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5546 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5547 if (!err)
5548 md_new_event(mddev);
005eca5e 5549 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5550 return err;
5551 }
5552
5553 /* otherwise, add_new_disk is only allowed
5554 * for major_version==0 superblocks
5555 */
5556 if (mddev->major_version != 0) {
5557 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5558 mdname(mddev));
5559 return -EINVAL;
5560 }
5561
5562 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5563 int err;
d710e138 5564 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5565 if (IS_ERR(rdev)) {
5566 printk(KERN_WARNING
5567 "md: error, md_import_device() returned %ld\n",
5568 PTR_ERR(rdev));
5569 return PTR_ERR(rdev);
5570 }
5571 rdev->desc_nr = info->number;
5572 if (info->raid_disk < mddev->raid_disks)
5573 rdev->raid_disk = info->raid_disk;
5574 else
5575 rdev->raid_disk = -1;
5576
1da177e4 5577 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5578 if (info->state & (1<<MD_DISK_SYNC))
5579 set_bit(In_sync, &rdev->flags);
1da177e4 5580
8ddf9efe
N		 5581 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5582 set_bit(WriteMostly, &rdev->flags);
5583
1da177e4
LT		 5584 if (!mddev->persistent) {
5585 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5586 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5587 } else
57b2caa3 5588 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5589 rdev->sectors = rdev->sb_start;
1da177e4 5590
2bf071bf
N		 5591 err = bind_rdev_to_array(rdev, mddev);
5592 if (err) {
5593 export_rdev(rdev);
5594 return err;
5595 }
1da177e4
LT		 5596 }
5597
5598 return 0;
5599}
5600
fd01b88c 5601static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5602{
5603 char b[BDEVNAME_SIZE];
3cb03002 5604 struct md_rdev *rdev;
1da177e4 5605
1da177e4
LT		 5606 rdev = find_rdev(mddev, dev);
5607 if (!rdev)
5608 return -ENXIO;
5609
5610 if (rdev->raid_disk >= 0)
5611 goto busy;
5612
5613 kick_rdev_from_array(rdev);
850b2b42 5614 md_update_sb(mddev, 1);
d7603b7e 5615 md_new_event(mddev);
1da177e4
LT		 5616
5617 return 0;
5618busy:
fdefa4d8 5619 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5620 bdevname(rdev->bdev,b), mdname(mddev));
5621 return -EBUSY;
5622}
5623
fd01b88c 5624static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5625{
5626 char b[BDEVNAME_SIZE];
5627 int err;
3cb03002 5628 struct md_rdev *rdev;
1da177e4
LT		 5629
5630 if (!mddev->pers)
5631 return -ENODEV;
5632
5633 if (mddev->major_version != 0) {
5634 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5635 " version-0 superblocks.\n",
5636 mdname(mddev));
5637 return -EINVAL;
5638 }
5639 if (!mddev->pers->hot_add_disk) {
5640 printk(KERN_WARNING
5641 "%s: personality does not support diskops!\n",
5642 mdname(mddev));
5643 return -EINVAL;
5644 }
5645
d710e138 5646 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5647 if (IS_ERR(rdev)) {
5648 printk(KERN_WARNING
5649 "md: error, md_import_device() returned %ld\n",
5650 PTR_ERR(rdev));
5651 return -EINVAL;
5652 }
5653
5654 if (mddev->persistent)
57b2caa3 5655 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5656 else
77304d2a 5657 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5658
8190e754 5659 rdev->sectors = rdev->sb_start;
1da177e4 5660
b2d444d7 5661 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5662 printk(KERN_WARNING
5663 "md: can not hot-add faulty %s disk to %s!\n",
5664 bdevname(rdev->bdev,b), mdname(mddev));
5665 err = -EINVAL;
5666 goto abort_export;
5667 }
b2d444d7 5668 clear_bit(In_sync, &rdev->flags);
1da177e4 5669 rdev->desc_nr = -1;
5842730d 5670 rdev->saved_raid_disk = -1;
2bf071bf
N		 5671 err = bind_rdev_to_array(rdev, mddev);
5672 if (err)
5673 goto abort_export;
1da177e4
LT		 5674
5675 /*
5676 * The rest should better be atomic, we can have disk failures
5677 * noticed in interrupt contexts ...
5678 */
5679
1da177e4
LT		 5680 rdev->raid_disk = -1;
5681
850b2b42 5682 md_update_sb(mddev, 1);
1da177e4
LT		 5683
5684 /*
5685 * Kick recovery, maybe this spare has to be added to the
5686 * array immediately.
5687 */
5688 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5689 md_wakeup_thread(mddev->thread);
d7603b7e 5690 md_new_event(mddev);
1da177e4
LT		 5691 return 0;
5692
1da177e4
LT		 5693abort_export:
5694 export_rdev(rdev);
5695 return err;
5696}
5697
fd01b88c 5698static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N		 5699{
5700 int err;
5701
36fa3063
N		 5702 if (mddev->pers) {
5703 if (!mddev->pers->quiesce)
5704 return -EBUSY;
5705 if (mddev->recovery || mddev->sync_thread)
5706 return -EBUSY;
5707 /* we should be able to change the bitmap.. */
5708 }
32a7627c 5709
32a7627c 5710
36fa3063
N		 5711 if (fd >= 0) {
5712 if (mddev->bitmap)
5713 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5714 mddev->bitmap_info.file = fget(fd);
32a7627c 5715
c3d9714e 5716 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5717 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5718 mdname(mddev));
5719 return -EBADF;
5720 }
5721
c3d9714e 5722 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5723 if (err) {
5724 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5725 mdname(mddev));
c3d9714e
N		 5726 fput(mddev->bitmap_info.file);
5727 mddev->bitmap_info.file = NULL;
36fa3063
N		 5728 return err;
5729 }
c3d9714e 5730 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5731 } else if (mddev->bitmap == NULL)
5732 return -ENOENT; /* cannot remove what isn't there */
5733 err = 0;
5734 if (mddev->pers) {
5735 mddev->pers->quiesce(mddev, 1);
69e51b44 5736 if (fd >= 0) {
36fa3063 5737 err = bitmap_create(mddev);
69e51b44
N		 5738 if (!err)
5739 err = bitmap_load(mddev);
5740 }
d7375ab3 5741 if (fd < 0 || err) {
36fa3063 5742 bitmap_destroy(mddev);
d7375ab3
N		 5743 fd = -1; /* make sure to put the file */
5744 }
36fa3063 5745 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5746 }
5747 if (fd < 0) {
c3d9714e
N		 5748 if (mddev->bitmap_info.file) {
5749 restore_bitmap_write_access(mddev->bitmap_info.file);
5750 fput(mddev->bitmap_info.file);
acc55e22 5751 }
c3d9714e 5752 mddev->bitmap_info.file = NULL;
36fa3063
N		 5753 }
5754
32a7627c
N		 5755 return err;
5756}
5757
1da177e4
LT		 5758/*
5759 * set_array_info is used two different ways
5760 * The original usage is when creating a new array.
5761 * In this usage, raid_disks is > 0 and it together with
5762 * level, size, not_persistent,layout,chunksize determine the
5763 * shape of the array.
5764 * This will always create an array with a type-0.90.0 superblock.
5765 * The newer usage is when assembling an array.
5766 * In this case raid_disks will be 0, and the major_version field is
5767 * use to determine which style super-blocks are to be found on the devices.
5768 * The minor and patch _version numbers are also kept incase the
5769 * super_block handler wishes to interpret them.
5770 */
fd01b88c 5771static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT		 5772{
5773
5774 if (info->raid_disks == 0) {
5775 /* just setting version number for superblock loading */
5776 if (info->major_version < 0 ||
50511da3 5777 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5778 super_types[info->major_version].name == NULL) {
5779 /* maybe try to auto-load a module? */
5780 printk(KERN_INFO
5781 "md: superblock version %d not known\n",
5782 info->major_version);
5783 return -EINVAL;
5784 }
5785 mddev->major_version = info->major_version;
5786 mddev->minor_version = info->minor_version;
5787 mddev->patch_version = info->patch_version;
3f9d7b0d 5788 mddev->persistent = !info->not_persistent;
cbd19983
N		 5789 /* ensure mddev_put doesn't delete this now that there
5790 * is some minimal configuration.
5791 */
5792 mddev->ctime = get_seconds();
1da177e4
LT		 5793 return 0;
5794 }
5795 mddev->major_version = MD_MAJOR_VERSION;
5796 mddev->minor_version = MD_MINOR_VERSION;
5797 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5798 mddev->ctime = get_seconds();
5799
5800 mddev->level = info->level;
17115e03 5801 mddev->clevel[0] = 0;
58c0fed4 5802 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5803 mddev->raid_disks = info->raid_disks;
5804 /* don't set md_minor, it is determined by which /dev/md* was
5805 * openned
5806 */
5807 if (info->state & (1<<MD_SB_CLEAN))
5808 mddev->recovery_cp = MaxSector;
5809 else
5810 mddev->recovery_cp = 0;
5811 mddev->persistent = ! info->not_persistent;
e691063a 5812 mddev->external = 0;
1da177e4
LT		 5813
5814 mddev->layout = info->layout;
9d8f0363 5815 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5816
5817 mddev->max_disks = MD_SB_DISKS;
5818
e691063a
N		 5819 if (mddev->persistent)
5820 mddev->flags = 0;
850b2b42 5821 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5822
c3d9714e
N		 5823 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5824 mddev->bitmap_info.offset = 0;
b2a2703c 5825
f6705578
N		 5826 mddev->reshape_position = MaxSector;
5827
1da177e4
LT		 5828 /*
5829 * Generate a 128 bit UUID
5830 */
5831 get_random_bytes(mddev->uuid, 16);
5832
f6705578 5833 mddev->new_level = mddev->level;
664e7c41 5834 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5835 mddev->new_layout = mddev->layout;
5836 mddev->delta_disks = 0;
5837
1da177e4
LT		 5838 return 0;
5839}
5840
fd01b88c 5841void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 5842{
b522adcd
DW		 5843 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5844
5845 if (mddev->external_size)
5846 return;
5847
1f403624
DW		 5848 mddev->array_sectors = array_sectors;
5849}
5850EXPORT_SYMBOL(md_set_array_sectors);
5851
fd01b88c 5852static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 5853{
3cb03002 5854 struct md_rdev *rdev;
a35b0d69 5855 int rv;
d71f9f88 5856 int fit = (num_sectors == 0);
a35b0d69
N		 5857
5858 if (mddev->pers->resize == NULL)
5859 return -EINVAL;
d71f9f88
AN		 5860 /* The "num_sectors" is the number of sectors of each device that
5861 * is used. This can only make sense for arrays with redundancy.
5862 * linear and raid0 always use whatever space is available. We can only
5863 * consider changing this number if no resync or reconstruction is
5864 * happening, and if the new size is acceptable. It must fit before the
0f420358 5865 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5866 * of each device. If num_sectors is zero, we find the largest size
5867 * that fits.
a35b0d69
N		 5868 */
5869 if (mddev->sync_thread)
5870 return -EBUSY;
dba034ee
N		 5871 if (mddev->bitmap)
5872 /* Sorry, cannot grow a bitmap yet, just remove it,
5873 * grow, and re-add.
5874 */
5875 return -EBUSY;
159ec1fc 5876 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5877 sector_t avail = rdev->sectors;
01ab5662 5878
d71f9f88
AN		 5879 if (fit && (num_sectors == 0 || num_sectors > avail))
5880 num_sectors = avail;
5881 if (avail < num_sectors)
a35b0d69
N		 5882 return -ENOSPC;
5883 }
d71f9f88 5884 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5885 if (!rv)
5886 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5887 return rv;
5888}
5889
fd01b88c 5890static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 5891{
5892 int rv;
5893 /* change the number of raid disks */
63c70c4f 5894 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5895 return -EINVAL;
5896 if (raid_disks <= 0 ||
233fca36 5897 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5898 return -EINVAL;
63c70c4f 5899 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5900 return -EBUSY;
63c70c4f
N		 5901 mddev->delta_disks = raid_disks - mddev->raid_disks;
5902
5903 rv = mddev->pers->check_reshape(mddev);
de171cb9
N		 5904 if (rv < 0)
5905 mddev->delta_disks = 0;
da943b99
N		 5906 return rv;
5907}
5908
5909
1da177e4
LT		 5910/*
5911 * update_array_info is used to change the configuration of an
5912 * on-line array.
5913 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5914 * fields in the info are checked against the array.
5915 * Any differences that cannot be handled will cause an error.
5916 * Normally, only one change can be managed at a time.
5917 */
fd01b88c 5918static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 5919{
5920 int rv = 0;
5921 int cnt = 0;
36fa3063
N		 5922 int state = 0;
5923
5924 /* calculate expected state,ignoring low bits */
c3d9714e 5925 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5926 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5927
5928 if (mddev->major_version != info->major_version ||
5929 mddev->minor_version != info->minor_version ||
5930/* mddev->patch_version != info->patch_version || */
5931 mddev->ctime != info->ctime ||
5932 mddev->level != info->level ||
5933/* mddev->layout != info->layout || */
5934 !mddev->persistent != info->not_persistent||
9d8f0363 5935 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5936 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5937 ((state^info->state) & 0xfffffe00)
5938 )
1da177e4
LT		 5939 return -EINVAL;
5940 /* Check there is only one change */
58c0fed4
AN		 5941 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5942 cnt++;
5943 if (mddev->raid_disks != info->raid_disks)
5944 cnt++;
5945 if (mddev->layout != info->layout)
5946 cnt++;
5947 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5948 cnt++;
5949 if (cnt == 0)
5950 return 0;
5951 if (cnt > 1)
5952 return -EINVAL;
1da177e4
LT		 5953
5954 if (mddev->layout != info->layout) {
5955 /* Change layout
5956 * we don't need to do anything at the md level, the
5957 * personality will take care of it all.
5958 */
50ac168a 5959 if (mddev->pers->check_reshape == NULL)
1da177e4 5960 return -EINVAL;
597a711b
N		 5961 else {
5962 mddev->new_layout = info->layout;
50ac168a 5963 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 5964 if (rv)
5965 mddev->new_layout = mddev->layout;
5966 return rv;
5967 }
1da177e4 5968 }
58c0fed4 5969 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 5970 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 5971
da943b99
N		 5972 if (mddev->raid_disks != info->raid_disks)
5973 rv = update_raid_disks(mddev, info->raid_disks);
5974
36fa3063
N		 5975 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5976 if (mddev->pers->quiesce == NULL)
5977 return -EINVAL;
5978 if (mddev->recovery || mddev->sync_thread)
5979 return -EBUSY;
5980 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5981 /* add the bitmap */
5982 if (mddev->bitmap)
5983 return -EEXIST;
c3d9714e 5984 if (mddev->bitmap_info.default_offset == 0)
36fa3063 5985 return -EINVAL;
c3d9714e
N		 5986 mddev->bitmap_info.offset =
5987 mddev->bitmap_info.default_offset;
36fa3063
N		 5988 mddev->pers->quiesce(mddev, 1);
5989 rv = bitmap_create(mddev);
69e51b44
N		 5990 if (!rv)
5991 rv = bitmap_load(mddev);
36fa3063
N		 5992 if (rv)
5993 bitmap_destroy(mddev);
5994 mddev->pers->quiesce(mddev, 0);
5995 } else {
5996 /* remove the bitmap */
5997 if (!mddev->bitmap)
5998 return -ENOENT;
5999 if (mddev->bitmap->file)
6000 return -EINVAL;
6001 mddev->pers->quiesce(mddev, 1);
6002 bitmap_destroy(mddev);
6003 mddev->pers->quiesce(mddev, 0);
c3d9714e 6004 mddev->bitmap_info.offset = 0;
36fa3063
N		 6005 }
6006 }
850b2b42 6007 md_update_sb(mddev, 1);
1da177e4
LT		 6008 return rv;
6009}
6010
fd01b88c 6011static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6012{
3cb03002 6013 struct md_rdev *rdev;
1da177e4
LT		 6014
6015 if (mddev->pers == NULL)
6016 return -ENODEV;
6017
6018 rdev = find_rdev(mddev, dev);
6019 if (!rdev)
6020 return -ENODEV;
6021
6022 md_error(mddev, rdev);
5ef56c8f
N		 6023 if (!test_bit(Faulty, &rdev->flags))
6024 return -EBUSY;
1da177e4
LT		 6025 return 0;
6026}
6027
2f9618ce
AN		 6028/*
6029 * We have a problem here : there is no easy way to give a CHS
6030 * virtual geometry. We currently pretend that we have a 2 heads
6031 * 4 sectors (with a BIG number of cylinders...). This drives
6032 * dosfs just mad... ;-)
6033 */
a885c8c4
CH		 6034static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6035{
fd01b88c 6036 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6037
6038 geo->heads = 2;
6039 geo->sectors = 4;
49ce6cea 6040 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6041 return 0;
6042}
6043
a39907fa 6044static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6045 unsigned int cmd, unsigned long arg)
6046{
6047 int err = 0;
6048 void __user *argp = (void __user *)arg;
fd01b88c 6049 struct mddev *mddev = NULL;
e2218350 6050 int ro;
1da177e4 6051
506c9e44
N		 6052 switch (cmd) {
6053 case RAID_VERSION:
6054 case GET_ARRAY_INFO:
6055 case GET_DISK_INFO:
6056 break;
6057 default:
6058 if (!capable(CAP_SYS_ADMIN))
6059 return -EACCES;
6060 }
1da177e4
LT		 6061
6062 /*
6063 * Commands dealing with the RAID driver but not any
6064 * particular array:
6065 */
6066 switch (cmd)
6067 {
6068 case RAID_VERSION:
6069 err = get_version(argp);
6070 goto done;
6071
6072 case PRINT_RAID_DEBUG:
6073 err = 0;
6074 md_print_devices();
6075 goto done;
6076
6077#ifndef MODULE
6078 case RAID_AUTORUN:
6079 err = 0;
6080 autostart_arrays(arg);
6081 goto done;
6082#endif
6083 default:;
6084 }
6085
6086 /*
6087 * Commands creating/starting a new array:
6088 */
6089
a39907fa 6090 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6091
6092 if (!mddev) {
6093 BUG();
6094 goto abort;
6095 }
6096
1da177e4
LT		 6097 err = mddev_lock(mddev);
6098 if (err) {
6099 printk(KERN_INFO
6100 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6101 err, cmd);
6102 goto abort;
6103 }
6104
6105 switch (cmd)
6106 {
6107 case SET_ARRAY_INFO:
6108 {
6109 mdu_array_info_t info;
6110 if (!arg)
6111 memset(&info, 0, sizeof(info));
6112 else if (copy_from_user(&info, argp, sizeof(info))) {
6113 err = -EFAULT;
6114 goto abort_unlock;
6115 }
6116 if (mddev->pers) {
6117 err = update_array_info(mddev, &info);
6118 if (err) {
6119 printk(KERN_WARNING "md: couldn't update"
6120 " array info. %d\n", err);
6121 goto abort_unlock;
6122 }
6123 goto done_unlock;
6124 }
6125 if (!list_empty(&mddev->disks)) {
6126 printk(KERN_WARNING
6127 "md: array %s already has disks!\n",
6128 mdname(mddev));
6129 err = -EBUSY;
6130 goto abort_unlock;
6131 }
6132 if (mddev->raid_disks) {
6133 printk(KERN_WARNING
6134 "md: array %s already initialised!\n",
6135 mdname(mddev));
6136 err = -EBUSY;
6137 goto abort_unlock;
6138 }
6139 err = set_array_info(mddev, &info);
6140 if (err) {
6141 printk(KERN_WARNING "md: couldn't set"
6142 " array info. %d\n", err);
6143 goto abort_unlock;
6144 }
6145 }
6146 goto done_unlock;
6147
6148 default:;
6149 }
6150
6151 /*
6152 * Commands querying/configuring an existing array:
6153 */
32a7627c 6154 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6155 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6156 if ((!mddev->raid_disks && !mddev->external)
6157 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6158 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6159 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6160 err = -ENODEV;
6161 goto abort_unlock;
6162 }
6163
6164 /*
6165 * Commands even a read-only array can execute:
6166 */
6167 switch (cmd)
6168 {
6169 case GET_ARRAY_INFO:
6170 err = get_array_info(mddev, argp);
6171 goto done_unlock;
6172
32a7627c 6173 case GET_BITMAP_FILE:
87162a28 6174 err = get_bitmap_file(mddev, argp);
32a7627c
N		 6175 goto done_unlock;
6176
1da177e4
LT		 6177 case GET_DISK_INFO:
6178 err = get_disk_info(mddev, argp);
6179 goto done_unlock;
6180
6181 case RESTART_ARRAY_RW:
6182 err = restart_array(mddev);
6183 goto done_unlock;
6184
6185 case STOP_ARRAY:
d710e138 6186 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 6187 goto done_unlock;
6188
6189 case STOP_ARRAY_RO:
a4bd82d0 6190 err = md_set_readonly(mddev, 1);
1da177e4
LT		 6191 goto done_unlock;
6192
e2218350
DW		 6193 case BLKROSET:
6194 if (get_user(ro, (int __user *)(arg))) {
6195 err = -EFAULT;
6196 goto done_unlock;
6197 }
6198 err = -EINVAL;
6199
6200 /* if the bdev is going readonly the value of mddev->ro
6201 * does not matter, no writes are coming
6202 */
6203 if (ro)
6204 goto done_unlock;
6205
6206 /* are we are already prepared for writes? */
6207 if (mddev->ro != 1)
6208 goto done_unlock;
6209
6210 /* transitioning to readauto need only happen for
6211 * arrays that call md_write_start
6212 */
6213 if (mddev->pers) {
6214 err = restart_array(mddev);
6215 if (err == 0) {
6216 mddev->ro = 2;
6217 set_disk_ro(mddev->gendisk, 0);
6218 }
6219 }
6220 goto done_unlock;
1da177e4
LT		 6221 }
6222
6223 /*
6224 * The remaining ioctls are changing the state of the
f91de92e
N		 6225 * superblock, so we do not allow them on read-only arrays.
6226 * However non-MD ioctls (e.g. get-size) will still come through
6227 * here and hit the 'default' below, so only disallow
6228 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6229 */
bb57fc64 6230 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6231 if (mddev->ro == 2) {
6232 mddev->ro = 0;
00bcb4ac 6233 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 6234 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6235 md_wakeup_thread(mddev->thread);
f91de92e
N		 6236 } else {
6237 err = -EROFS;
6238 goto abort_unlock;
6239 }
1da177e4
LT		 6240 }
6241
6242 switch (cmd)
6243 {
6244 case ADD_NEW_DISK:
6245 {
6246 mdu_disk_info_t info;
6247 if (copy_from_user(&info, argp, sizeof(info)))
6248 err = -EFAULT;
6249 else
6250 err = add_new_disk(mddev, &info);
6251 goto done_unlock;
6252 }
6253
6254 case HOT_REMOVE_DISK:
6255 err = hot_remove_disk(mddev, new_decode_dev(arg));
6256 goto done_unlock;
6257
6258 case HOT_ADD_DISK:
6259 err = hot_add_disk(mddev, new_decode_dev(arg));
6260 goto done_unlock;
6261
6262 case SET_DISK_FAULTY:
6263 err = set_disk_faulty(mddev, new_decode_dev(arg));
6264 goto done_unlock;
6265
6266 case RUN_ARRAY:
d710e138 6267 err = do_md_run(mddev);
1da177e4
LT		 6268 goto done_unlock;
6269
32a7627c
N		 6270 case SET_BITMAP_FILE:
6271 err = set_bitmap_file(mddev, (int)arg);
6272 goto done_unlock;
6273
1da177e4 6274 default:
1da177e4
LT		 6275 err = -EINVAL;
6276 goto abort_unlock;
6277 }
6278
6279done_unlock:
6280abort_unlock:
d3374825
N		 6281 if (mddev->hold_active == UNTIL_IOCTL &&
6282 err != -EINVAL)
6283 mddev->hold_active = 0;
1da177e4
LT		 6284 mddev_unlock(mddev);
6285
6286 return err;
6287done:
6288 if (err)
6289 MD_BUG();
6290abort:
6291 return err;
6292}
aa98aa31
AB		 6293#ifdef CONFIG_COMPAT
6294static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6295 unsigned int cmd, unsigned long arg)
6296{
6297 switch (cmd) {
6298 case HOT_REMOVE_DISK:
6299 case HOT_ADD_DISK:
6300 case SET_DISK_FAULTY:
6301 case SET_BITMAP_FILE:
6302 /* These take in integer arg, do not convert */
6303 break;
6304 default:
6305 arg = (unsigned long)compat_ptr(arg);
6306 break;
6307 }
6308
6309 return md_ioctl(bdev, mode, cmd, arg);
6310}
6311#endif /* CONFIG_COMPAT */
1da177e4 6312
a39907fa 6313static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6314{
6315 /*
6316 * Succeed if we can lock the mddev, which confirms that
6317 * it isn't being stopped right now.
6318 */
fd01b88c 6319 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6320 int err;
6321
d3374825
N		 6322 if (mddev->gendisk != bdev->bd_disk) {
6323 /* we are racing with mddev_put which is discarding this
6324 * bd_disk.
6325 */
6326 mddev_put(mddev);
6327 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6328 flush_workqueue(md_misc_wq);
d3374825
N		 6329 /* Then retry the open from the top */
6330 return -ERESTARTSYS;
6331 }
6332 BUG_ON(mddev != bdev->bd_disk->private_data);
6333
c8c00a69 6334 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6335 goto out;
6336
6337 err = 0;
f2ea68cf 6338 atomic_inc(&mddev->openers);
c8c00a69 6339 mutex_unlock(&mddev->open_mutex);
1da177e4 6340
f0b4f7e2 6341 check_disk_change(bdev);
1da177e4
LT		 6342 out:
6343 return err;
6344}
6345
a39907fa 6346static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6347{
fd01b88c 6348 struct mddev *mddev = disk->private_data;
1da177e4 6349
52e5f9d1 6350 BUG_ON(!mddev);
f2ea68cf 6351 atomic_dec(&mddev->openers);
1da177e4
LT		 6352 mddev_put(mddev);
6353
6354 return 0;
6355}
f0b4f7e2
N		 6356
6357static int md_media_changed(struct gendisk *disk)
6358{
fd01b88c 6359 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6360
6361 return mddev->changed;
6362}
6363
6364static int md_revalidate(struct gendisk *disk)
6365{
fd01b88c 6366 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6367
6368 mddev->changed = 0;
6369 return 0;
6370}
83d5cde4 6371static const struct block_device_operations md_fops =
1da177e4
LT		 6372{
6373 .owner = THIS_MODULE,
a39907fa
AV		 6374 .open = md_open,
6375 .release = md_release,
b492b852 6376 .ioctl = md_ioctl,
aa98aa31
AB		 6377#ifdef CONFIG_COMPAT
6378 .compat_ioctl = md_compat_ioctl,
6379#endif
a885c8c4 6380 .getgeo = md_getgeo,
f0b4f7e2
N		 6381 .media_changed = md_media_changed,
6382 .revalidate_disk= md_revalidate,
1da177e4
LT		 6383};
6384
75c96f85 6385static int md_thread(void * arg)
1da177e4 6386{
2b8bf345 6387 struct md_thread *thread = arg;
1da177e4 6388
1da177e4
LT		 6389 /*
6390 * md_thread is a 'system-thread', it's priority should be very
6391 * high. We avoid resource deadlocks individually in each
6392 * raid personality. (RAID5 does preallocation) We also use RR and
6393 * the very same RT priority as kswapd, thus we will never get
6394 * into a priority inversion deadlock.
6395 *
6396 * we definitely have to have equal or higher priority than
6397 * bdflush, otherwise bdflush will deadlock if there are too
6398 * many dirty RAID5 blocks.
6399 */
1da177e4 6400
6985c43f 6401 allow_signal(SIGKILL);
a6fb0934 6402 while (!kthread_should_stop()) {
1da177e4 6403
93588e22
N		 6404 /* We need to wait INTERRUPTIBLE so that
6405 * we don't add to the load-average.
6406 * That means we need to be sure no signals are
6407 * pending
6408 */
6409 if (signal_pending(current))
6410 flush_signals(current);
6411
6412 wait_event_interruptible_timeout
6413 (thread->wqueue,
6414 test_bit(THREAD_WAKEUP, &thread->flags)
6415 || kthread_should_stop(),
6416 thread->timeout);
1da177e4 6417
6c987910
N		 6418 clear_bit(THREAD_WAKEUP, &thread->flags);
6419 if (!kthread_should_stop())
589a594b 6420 thread->run(thread->mddev);
1da177e4 6421 }
a6fb0934 6422
1da177e4
LT		 6423 return 0;
6424}
6425
2b8bf345 6426void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6427{
6428 if (thread) {
36a4e1fe 6429 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6430 set_bit(THREAD_WAKEUP, &thread->flags);
6431 wake_up(&thread->wqueue);
6432 }
6433}
6434
2b8bf345 6435struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
1da177e4
LT		 6436 const char *name)
6437{
2b8bf345 6438 struct md_thread *thread;
1da177e4 6439
2b8bf345 6440 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6441 if (!thread)
6442 return NULL;
6443
1da177e4
LT		 6444 init_waitqueue_head(&thread->wqueue);
6445
1da177e4
LT		 6446 thread->run = run;
6447 thread->mddev = mddev;
32a7627c 6448 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6449 thread->tsk = kthread_run(md_thread, thread,
6450 "%s_%s",
6451 mdname(thread->mddev),
6452 name ?: mddev->pers->name);
a6fb0934 6453 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6454 kfree(thread);
6455 return NULL;
6456 }
1da177e4
LT		 6457 return thread;
6458}
6459
2b8bf345 6460void md_unregister_thread(struct md_thread **threadp)
1da177e4 6461{
2b8bf345 6462 struct md_thread *thread = *threadp;
e0cf8f04
N		 6463 if (!thread)
6464 return;
36a4e1fe 6465 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6466 /* Locking ensures that mddev_unlock does not wake_up a
6467 * non-existent thread
6468 */
6469 spin_lock(&pers_lock);
6470 *threadp = NULL;
6471 spin_unlock(&pers_lock);
a6fb0934
N		 6472
6473 kthread_stop(thread->tsk);
1da177e4
LT		 6474 kfree(thread);
6475}
6476
fd01b88c 6477void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6478{
6479 if (!mddev) {
6480 MD_BUG();
6481 return;
6482 }
6483
b2d444d7 6484 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6485 return;
6bfe0b49 6486
de393cde 6487 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6488 return;
6489 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6490 if (mddev->degraded)
6491 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6492 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6493 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6494 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6495 md_wakeup_thread(mddev->thread);
768a418d 6496 if (mddev->event_work.func)
e804ac78 6497 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6498 md_new_event_inintr(mddev);
1da177e4
LT		 6499}
6500
6501/* seq_file implementation /proc/mdstat */
6502
6503static void status_unused(struct seq_file *seq)
6504{
6505 int i = 0;
3cb03002 6506 struct md_rdev *rdev;
1da177e4
LT		 6507
6508 seq_printf(seq, "unused devices: ");
6509
159ec1fc 6510 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6511 char b[BDEVNAME_SIZE];
6512 i++;
6513 seq_printf(seq, "%s ",
6514 bdevname(rdev->bdev,b));
6515 }
6516 if (!i)
6517 seq_printf(seq, "<none>");
6518
6519 seq_printf(seq, "\n");
6520}
6521
6522
fd01b88c 6523static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6524{
dd71cf6b
N		 6525 sector_t max_sectors, resync, res;
6526 unsigned long dt, db;
6527 sector_t rt;
4588b42e
N		 6528 int scale;
6529 unsigned int per_milli;
1da177e4 6530
dd71cf6b 6531 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6532
6533 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6534 max_sectors = mddev->resync_max_sectors;
1da177e4 6535 else
dd71cf6b 6536 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6537
6538 /*
6539 * Should not happen.
6540 */
dd71cf6b 6541 if (!max_sectors) {
1da177e4
LT		 6542 MD_BUG();
6543 return;
6544 }
4588b42e 6545 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6546 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6547 * u32, as those are the requirements for sector_div.
6548 * Thus 'scale' must be at least 10
6549 */
6550 scale = 10;
6551 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6552 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6553 scale++;
6554 }
6555 res = (resync>>scale)*1000;
dd71cf6b 6556 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6557
6558 per_milli = res;
1da177e4 6559 {
4588b42e 6560 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6561 seq_printf(seq, "[");
6562 for (i = 0; i < x; i++)
6563 seq_printf(seq, "=");
6564 seq_printf(seq, ">");
6565 for (i = 0; i < y; i++)
6566 seq_printf(seq, ".");
6567 seq_printf(seq, "] ");
6568 }
4588b42e 6569 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6570 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6571 "reshape" :
61df9d91
N		 6572 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6573 "check" :
6574 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6575 "resync" : "recovery"))),
6576 per_milli/10, per_milli % 10,
dd71cf6b
N		 6577 (unsigned long long) resync/2,
6578 (unsigned long long) max_sectors/2);
1da177e4
LT		 6579
6580 /*
1da177e4
LT		 6581 * dt: time from mark until now
6582 * db: blocks written from mark until now
6583 * rt: remaining time
dd71cf6b
N		 6584 *
6585 * rt is a sector_t, so could be 32bit or 64bit.
6586 * So we divide before multiply in case it is 32bit and close
6587 * to the limit.
25985edc 6588 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6589 * near the end of resync when the number of remaining sectors
6590 * is close to 'db'.
6591 * We then divide rt by 32 after multiplying by db to compensate.
6592 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6593 */
6594 dt = ((jiffies - mddev->resync_mark) / HZ);
6595 if (!dt) dt++;
ff4e8d9a
N		 6596 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6597 - mddev->resync_mark_cnt;
1da177e4 6598
dd71cf6b
N		 6599 rt = max_sectors - resync; /* number of remaining sectors */
6600 sector_div(rt, db/32+1);
6601 rt *= dt;
6602 rt >>= 5;
6603
6604 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6605 ((unsigned long)rt % 60)/6);
1da177e4 6606
ff4e8d9a 6607 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6608}
6609
6610static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6611{
6612 struct list_head *tmp;
6613 loff_t l = *pos;
fd01b88c 6614 struct mddev *mddev;
1da177e4
LT		 6615
6616 if (l >= 0x10000)
6617 return NULL;
6618 if (!l--)
6619 /* header */
6620 return (void*)1;
6621
6622 spin_lock(&all_mddevs_lock);
6623 list_for_each(tmp,&all_mddevs)
6624 if (!l--) {
fd01b88c 6625 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6626 mddev_get(mddev);
6627 spin_unlock(&all_mddevs_lock);
6628 return mddev;
6629 }
6630 spin_unlock(&all_mddevs_lock);
6631 if (!l--)
6632 return (void*)2;/* tail */
6633 return NULL;
6634}
6635
6636static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6637{
6638 struct list_head *tmp;
fd01b88c 6639 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6640
6641 ++*pos;
6642 if (v == (void*)2)
6643 return NULL;
6644
6645 spin_lock(&all_mddevs_lock);
6646 if (v == (void*)1)
6647 tmp = all_mddevs.next;
6648 else
6649 tmp = mddev->all_mddevs.next;
6650 if (tmp != &all_mddevs)
fd01b88c 6651 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 6652 else {
6653 next_mddev = (void*)2;
6654 *pos = 0x10000;
6655 }
6656 spin_unlock(&all_mddevs_lock);
6657
6658 if (v != (void*)1)
6659 mddev_put(mddev);
6660 return next_mddev;
6661
6662}
6663
6664static void md_seq_stop(struct seq_file *seq, void *v)
6665{
fd01b88c 6666 struct mddev *mddev = v;
1da177e4
LT		 6667
6668 if (mddev && v != (void*)1 && v != (void*)2)
6669 mddev_put(mddev);
6670}
6671
6672static int md_seq_show(struct seq_file *seq, void *v)
6673{
fd01b88c 6674 struct mddev *mddev = v;
dd8ac336 6675 sector_t sectors;
3cb03002 6676 struct md_rdev *rdev;
32a7627c 6677 struct bitmap *bitmap;
1da177e4
LT		 6678
6679 if (v == (void*)1) {
84fc4b56 6680 struct md_personality *pers;
1da177e4
LT		 6681 seq_printf(seq, "Personalities : ");
6682 spin_lock(&pers_lock);
2604b703
N		 6683 list_for_each_entry(pers, &pers_list, list)
6684 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6685
6686 spin_unlock(&pers_lock);
6687 seq_printf(seq, "\n");
f1514638 6688 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6689 return 0;
6690 }
6691 if (v == (void*)2) {
6692 status_unused(seq);
6693 return 0;
6694 }
6695
5dc5cf7d 6696 if (mddev_lock(mddev) < 0)
1da177e4 6697 return -EINTR;
5dc5cf7d 6698
1da177e4
LT		 6699 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6700 seq_printf(seq, "%s : %sactive", mdname(mddev),
6701 mddev->pers ? "" : "in");
6702 if (mddev->pers) {
f91de92e 6703 if (mddev->ro==1)
1da177e4 6704 seq_printf(seq, " (read-only)");
f91de92e 6705 if (mddev->ro==2)
52720ae7 6706 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6707 seq_printf(seq, " %s", mddev->pers->name);
6708 }
6709
dd8ac336 6710 sectors = 0;
159ec1fc 6711 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 6712 char b[BDEVNAME_SIZE];
6713 seq_printf(seq, " %s[%d]",
6714 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6715 if (test_bit(WriteMostly, &rdev->flags))
6716 seq_printf(seq, "(W)");
b2d444d7 6717 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6718 seq_printf(seq, "(F)");
6719 continue;
b325a32e
N		 6720 } else if (rdev->raid_disk < 0)
6721 seq_printf(seq, "(S)"); /* spare */
dd8ac336 6722 sectors += rdev->sectors;
1da177e4
LT		 6723 }
6724
6725 if (!list_empty(&mddev->disks)) {
6726 if (mddev->pers)
6727 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6728 (unsigned long long)
6729 mddev->array_sectors / 2);
1da177e4
LT		 6730 else
6731 seq_printf(seq, "\n %llu blocks",
dd8ac336 6732 (unsigned long long)sectors / 2);
1da177e4 6733 }
1cd6bf19
N		 6734 if (mddev->persistent) {
6735 if (mddev->major_version != 0 ||
6736 mddev->minor_version != 90) {
6737 seq_printf(seq," super %d.%d",
6738 mddev->major_version,
6739 mddev->minor_version);
6740 }
e691063a
N		 6741 } else if (mddev->external)
6742 seq_printf(seq, " super external:%s",
6743 mddev->metadata_type);
6744 else
1cd6bf19 6745 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6746
6747 if (mddev->pers) {
d710e138 6748 mddev->pers->status(seq, mddev);
1da177e4 6749 seq_printf(seq, "\n ");
8e1b39d6
N		 6750 if (mddev->pers->sync_request) {
6751 if (mddev->curr_resync > 2) {
d710e138 6752 status_resync(seq, mddev);
8e1b39d6
N		 6753 seq_printf(seq, "\n ");
6754 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6755 seq_printf(seq, "\tresync=DELAYED\n ");
6756 else if (mddev->recovery_cp < MaxSector)
6757 seq_printf(seq, "\tresync=PENDING\n ");
6758 }
32a7627c
N		 6759 } else
6760 seq_printf(seq, "\n ");
6761
6762 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6763 unsigned long chunk_kb;
6764 unsigned long flags;
32a7627c 6765 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6766 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6767 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6768 "%lu%s chunk",
6769 bitmap->pages - bitmap->missing_pages,
6770 bitmap->pages,
6771 (bitmap->pages - bitmap->missing_pages)
6772 << (PAGE_SHIFT - 10),
42a04b50 6773 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6774 chunk_kb ? "KB" : "B");
78d742d8
N		 6775 if (bitmap->file) {
6776 seq_printf(seq, ", file: ");
c32c2f63 6777 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6778 }
78d742d8 6779
32a7627c
N		 6780 seq_printf(seq, "\n");
6781 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6782 }
6783
6784 seq_printf(seq, "\n");
6785 }
6786 mddev_unlock(mddev);
6787
6788 return 0;
6789}
6790
110518bc 6791static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6792 .start = md_seq_start,
6793 .next = md_seq_next,
6794 .stop = md_seq_stop,
6795 .show = md_seq_show,
6796};
6797
6798static int md_seq_open(struct inode *inode, struct file *file)
6799{
f1514638 6800 struct seq_file *seq;
1da177e4
LT		 6801 int error;
6802
6803 error = seq_open(file, &md_seq_ops);
d7603b7e 6804 if (error)
f1514638
KS		 6805 return error;
6806
6807 seq = file->private_data;
6808 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6809 return error;
6810}
6811
d7603b7e
N		 6812static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6813{
f1514638 6814 struct seq_file *seq = filp->private_data;
d7603b7e
N		 6815 int mask;
6816
6817 poll_wait(filp, &md_event_waiters, wait);
6818
6819 /* always allow read */
6820 mask = POLLIN | POLLRDNORM;
6821
f1514638 6822 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 6823 mask |= POLLERR | POLLPRI;
6824 return mask;
6825}
6826
fa027c2a 6827static const struct file_operations md_seq_fops = {
e24650c2 6828 .owner = THIS_MODULE,
1da177e4
LT		 6829 .open = md_seq_open,
6830 .read = seq_read,
6831 .llseek = seq_lseek,
c3f94b40 6832 .release = seq_release_private,
d7603b7e 6833 .poll = mdstat_poll,
1da177e4
LT		 6834};
6835
84fc4b56 6836int register_md_personality(struct md_personality *p)
1da177e4 6837{
1da177e4 6838 spin_lock(&pers_lock);
2604b703
N		 6839 list_add_tail(&p->list, &pers_list);
6840 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6841 spin_unlock(&pers_lock);
6842 return 0;
6843}
6844
84fc4b56 6845int unregister_md_personality(struct md_personality *p)
1da177e4 6846{
2604b703 6847 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6848 spin_lock(&pers_lock);
2604b703 6849 list_del_init(&p->list);
1da177e4
LT		 6850 spin_unlock(&pers_lock);
6851 return 0;
6852}
6853
fd01b88c 6854static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 6855{
3cb03002 6856 struct md_rdev * rdev;
1da177e4 6857 int idle;
eea1bf38 6858 int curr_events;
1da177e4
LT		 6859
6860 idle = 1;
4b80991c
N		 6861 rcu_read_lock();
6862 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6863 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6864 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6865 (int)part_stat_read(&disk->part0, sectors[1]) -
6866 atomic_read(&disk->sync_io);
713f6ab1
N		 6867 /* sync IO will cause sync_io to increase before the disk_stats
6868 * as sync_io is counted when a request starts, and
6869 * disk_stats is counted when it completes.
6870 * So resync activity will cause curr_events to be smaller than
6871 * when there was no such activity.
6872 * non-sync IO will cause disk_stat to increase without
6873 * increasing sync_io so curr_events will (eventually)
6874 * be larger than it was before. Once it becomes
6875 * substantially larger, the test below will cause
6876 * the array to appear non-idle, and resync will slow
6877 * down.
6878 * If there is a lot of outstanding resync activity when
6879 * we set last_event to curr_events, then all that activity
6880 * completing might cause the array to appear non-idle
6881 * and resync will be slowed down even though there might
6882 * not have been non-resync activity. This will only
6883 * happen once though. 'last_events' will soon reflect
6884 * the state where there is little or no outstanding
6885 * resync requests, and further resync activity will
6886 * always make curr_events less than last_events.
c0e48521 6887 *
1da177e4 6888 */
eea1bf38 6889 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6890 rdev->last_events = curr_events;
6891 idle = 0;
6892 }
6893 }
4b80991c 6894 rcu_read_unlock();
1da177e4
LT		 6895 return idle;
6896}
6897
fd01b88c 6898void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 6899{
6900 /* another "blocks" (512byte) blocks have been synced */
6901 atomic_sub(blocks, &mddev->recovery_active);
6902 wake_up(&mddev->recovery_wait);
6903 if (!ok) {
dfc70645 6904 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6905 md_wakeup_thread(mddev->thread);
6906 // stop recovery, signal do_sync ....
6907 }
6908}
6909
6910
06d91a5f
N		 6911/* md_write_start(mddev, bi)
6912 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6913 * in superblock) before writing, schedule a superblock update
6914 * and wait for it to complete.
06d91a5f 6915 */
fd01b88c 6916void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 6917{
0fd62b86 6918 int did_change = 0;
06d91a5f 6919 if (bio_data_dir(bi) != WRITE)
3d310eb7 6920 return;
06d91a5f 6921
f91de92e
N		 6922 BUG_ON(mddev->ro == 1);
6923 if (mddev->ro == 2) {
6924 /* need to switch to read/write */
6925 mddev->ro = 0;
6926 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6927 md_wakeup_thread(mddev->thread);
25156198 6928 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6929 did_change = 1;
f91de92e 6930 }
06d91a5f 6931 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6932 if (mddev->safemode == 1)
6933 mddev->safemode = 0;
06d91a5f 6934 if (mddev->in_sync) {
a9701a30 6935 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6936 if (mddev->in_sync) {
6937 mddev->in_sync = 0;
850b2b42 6938 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6939 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6940 md_wakeup_thread(mddev->thread);
0fd62b86 6941 did_change = 1;
3d310eb7 6942 }
a9701a30 6943 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6944 }
0fd62b86 6945 if (did_change)
00bcb4ac 6946 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 6947 wait_event(mddev->sb_wait,
09a44cc1 6948 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 6949}
6950
fd01b88c 6951void md_write_end(struct mddev *mddev)
1da177e4
LT		 6952{
6953 if (atomic_dec_and_test(&mddev->writes_pending)) {
6954 if (mddev->safemode == 2)
6955 md_wakeup_thread(mddev->thread);
16f17b39 6956 else if (mddev->safemode_delay)
1da177e4
LT		 6957 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6958 }
6959}
6960
2a2275d6
N		 6961/* md_allow_write(mddev)
6962 * Calling this ensures that the array is marked 'active' so that writes
6963 * may proceed without blocking. It is important to call this before
6964 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6965 * Must be called with mddev_lock held.
b5470dc5
DW		 6966 *
6967 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6968 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 6969 */
fd01b88c 6970int md_allow_write(struct mddev *mddev)
2a2275d6
N		 6971{
6972 if (!mddev->pers)
b5470dc5 6973 return 0;
2a2275d6 6974 if (mddev->ro)
b5470dc5 6975 return 0;
1a0fd497 6976 if (!mddev->pers->sync_request)
b5470dc5 6977 return 0;
2a2275d6
N		 6978
6979 spin_lock_irq(&mddev->write_lock);
6980 if (mddev->in_sync) {
6981 mddev->in_sync = 0;
6982 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6983 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 6984 if (mddev->safemode_delay &&
6985 mddev->safemode == 0)
6986 mddev->safemode = 1;
6987 spin_unlock_irq(&mddev->write_lock);
6988 md_update_sb(mddev, 0);
00bcb4ac 6989 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 6990 } else
6991 spin_unlock_irq(&mddev->write_lock);
b5470dc5 6992
070dc6dd 6993 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 6994 return -EAGAIN;
6995 else
6996 return 0;
2a2275d6
N		 6997}
6998EXPORT_SYMBOL_GPL(md_allow_write);
6999
1da177e4
LT		 7000#define SYNC_MARKS 10
7001#define SYNC_MARK_STEP (3*HZ)
fd01b88c 7002void md_do_sync(struct mddev *mddev)
1da177e4 7003{
fd01b88c 7004 struct mddev *mddev2;
1da177e4
LT		 7005 unsigned int currspeed = 0,
7006 window;
57afd89f 7007 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 7008 unsigned long mark[SYNC_MARKS];
7009 sector_t mark_cnt[SYNC_MARKS];
7010 int last_mark,m;
7011 struct list_head *tmp;
7012 sector_t last_check;
57afd89f 7013 int skipped = 0;
3cb03002 7014 struct md_rdev *rdev;
61df9d91 7015 char *desc;
1da177e4
LT		 7016
7017 /* just incase thread restarts... */
7018 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7019 return;
5fd6c1dc
N		 7020 if (mddev->ro) /* never try to sync a read-only array */
7021 return;
1da177e4 7022
61df9d91
N		 7023 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7024 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7025 desc = "data-check";
7026 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7027 desc = "requested-resync";
7028 else
7029 desc = "resync";
7030 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7031 desc = "reshape";
7032 else
7033 desc = "recovery";
7034
1da177e4
LT		 7035 /* we overload curr_resync somewhat here.
7036 * 0 == not engaged in resync at all
7037 * 2 == checking that there is no conflict with another sync
7038 * 1 == like 2, but have yielded to allow conflicting resync to
7039 * commense
7040 * other == active in resync - this many blocks
7041 *
7042 * Before starting a resync we must have set curr_resync to
7043 * 2, and then checked that every "conflicting" array has curr_resync
7044 * less than ours. When we find one that is the same or higher
7045 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7046 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7047 * This will mean we have to start checking from the beginning again.
7048 *
7049 */
7050
7051 do {
7052 mddev->curr_resync = 2;
7053
7054 try_again:
404e4b43 7055 if (kthread_should_stop())
6985c43f 7056 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7057
7058 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7059 goto skip;
29ac4aa3 7060 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7061 if (mddev2 == mddev)
7062 continue;
90b08710
BS		 7063 if (!mddev->parallel_resync
7064 && mddev2->curr_resync
7065 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7066 DEFINE_WAIT(wq);
7067 if (mddev < mddev2 && mddev->curr_resync == 2) {
7068 /* arbitrarily yield */
7069 mddev->curr_resync = 1;
7070 wake_up(&resync_wait);
7071 }
7072 if (mddev > mddev2 && mddev->curr_resync == 1)
7073 /* no need to wait here, we can wait the next
7074 * time 'round when curr_resync == 2
7075 */
7076 continue;
9744197c
N		 7077 /* We need to wait 'interruptible' so as not to
7078 * contribute to the load average, and not to
7079 * be caught by 'softlockup'
7080 */
7081 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7082 if (!kthread_should_stop() &&
8712e553 7083 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7084 printk(KERN_INFO "md: delaying %s of %s"
7085 " until %s has finished (they"
1da177e4 7086 " share one or more physical units)\n",
61df9d91 7087 desc, mdname(mddev), mdname(mddev2));
1da177e4 7088 mddev_put(mddev2);
9744197c
N		 7089 if (signal_pending(current))
7090 flush_signals(current);
1da177e4
LT		 7091 schedule();
7092 finish_wait(&resync_wait, &wq);
7093 goto try_again;
7094 }
7095 finish_wait(&resync_wait, &wq);
7096 }
7097 }
7098 } while (mddev->curr_resync < 2);
7099
5fd6c1dc 7100 j = 0;
9d88883e 7101 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7102 /* resync follows the size requested by the personality,
57afd89f 7103 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7104 */
7105 max_sectors = mddev->resync_max_sectors;
9d88883e 7106 mddev->resync_mismatches = 0;
5fd6c1dc 7107 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7108 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7109 j = mddev->resync_min;
7110 else if (!mddev->bitmap)
5fd6c1dc 7111 j = mddev->recovery_cp;
5e96ee65 7112
ccfcc3c1 7113 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 7114 max_sectors = mddev->dev_sectors;
5fd6c1dc 7115 else {
1da177e4 7116 /* recovery follows the physical size of devices */
58c0fed4 7117 max_sectors = mddev->dev_sectors;
5fd6c1dc 7118 j = MaxSector;
4e59ca7d
DW		 7119 rcu_read_lock();
7120 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 7121 if (rdev->raid_disk >= 0 &&
7122 !test_bit(Faulty, &rdev->flags) &&
7123 !test_bit(In_sync, &rdev->flags) &&
7124 rdev->recovery_offset < j)
7125 j = rdev->recovery_offset;
4e59ca7d 7126 rcu_read_unlock();
5fd6c1dc 7127 }
1da177e4 7128
61df9d91
N		 7129 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7130 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7131 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7132 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7133 "(but not more than %d KB/sec) for %s.\n",
7134 speed_max(mddev), desc);
1da177e4 7135
eea1bf38 7136 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7137
57afd89f 7138 io_sectors = 0;
1da177e4
LT		 7139 for (m = 0; m < SYNC_MARKS; m++) {
7140 mark[m] = jiffies;
57afd89f 7141 mark_cnt[m] = io_sectors;
1da177e4
LT		 7142 }
7143 last_mark = 0;
7144 mddev->resync_mark = mark[last_mark];
7145 mddev->resync_mark_cnt = mark_cnt[last_mark];
7146
7147 /*
7148 * Tune reconstruction:
7149 */
7150 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7151 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7152 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7153
7154 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7155 last_check = 0;
7156
7157 if (j>2) {
7158 printk(KERN_INFO
61df9d91
N		 7159 "md: resuming %s of %s from checkpoint.\n",
7160 desc, mdname(mddev));
1da177e4
LT		 7161 mddev->curr_resync = j;
7162 }
75d3da43 7163 mddev->curr_resync_completed = j;
1da177e4
LT		 7164
7165 while (j < max_sectors) {
57afd89f 7166 sector_t sectors;
1da177e4 7167
57afd89f 7168 skipped = 0;
97e4f42d 7169
7a91ee1f
N		 7170 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7171 ((mddev->curr_resync > mddev->curr_resync_completed &&
7172 (mddev->curr_resync - mddev->curr_resync_completed)
7173 > (max_sectors >> 4)) ||
7174 (j - mddev->curr_resync_completed)*2
7175 >= mddev->resync_max - mddev->curr_resync_completed
7176 )) {
97e4f42d 7177 /* time to update curr_resync_completed */
97e4f42d
N		 7178 wait_event(mddev->recovery_wait,
7179 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7180 mddev->curr_resync_completed = j;
070dc6dd 7181 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7182 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7183 }
acb180b0 7184
e62e58a5
N		 7185 while (j >= mddev->resync_max && !kthread_should_stop()) {
7186 /* As this condition is controlled by user-space,
7187 * we can block indefinitely, so use '_interruptible'
7188 * to avoid triggering warnings.
7189 */
7190 flush_signals(current); /* just in case */
7191 wait_event_interruptible(mddev->recovery_wait,
7192 mddev->resync_max > j
7193 || kthread_should_stop());
7194 }
acb180b0
N		 7195
7196 if (kthread_should_stop())
7197 goto interrupted;
7198
57afd89f 7199 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7200 currspeed < speed_min(mddev));
57afd89f 7201 if (sectors == 0) {
dfc70645 7202 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7203 goto out;
7204 }
57afd89f
N		 7205
7206 if (!skipped) { /* actual IO requested */
7207 io_sectors += sectors;
7208 atomic_add(sectors, &mddev->recovery_active);
7209 }
7210
e875ecea
N		 7211 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7212 break;
7213
1da177e4
LT		 7214 j += sectors;
7215 if (j>1) mddev->curr_resync = j;
ff4e8d9a 7216 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7217 if (last_check == 0)
e875ecea 7218 /* this is the earliest that rebuild will be
d7603b7e
N		 7219 * visible in /proc/mdstat
7220 */
7221 md_new_event(mddev);
57afd89f
N		 7222
7223 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7224 continue;
7225
57afd89f 7226 last_check = io_sectors;
1da177e4
LT		 7227 repeat:
7228 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7229 /* step marks */
7230 int next = (last_mark+1) % SYNC_MARKS;
7231
7232 mddev->resync_mark = mark[next];
7233 mddev->resync_mark_cnt = mark_cnt[next];
7234 mark[next] = jiffies;
57afd89f 7235 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7236 last_mark = next;
7237 }
7238
7239
c6207277
N		 7240 if (kthread_should_stop())
7241 goto interrupted;
7242
1da177e4
LT		 7243
7244 /*
7245 * this loop exits only if either when we are slower than
7246 * the 'hard' speed limit, or the system was IO-idle for
7247 * a jiffy.
7248 * the system might be non-idle CPU-wise, but we only care
7249 * about not overloading the IO subsystem. (things like an
7250 * e2fsck being done on the RAID array should execute fast)
7251 */
1da177e4
LT		 7252 cond_resched();
7253
57afd89f
N		 7254 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7255 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7256
88202a0c
N		 7257 if (currspeed > speed_min(mddev)) {
7258 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7259 !is_mddev_idle(mddev, 0)) {
c0e48521 7260 msleep(500);
1da177e4
LT		 7261 goto repeat;
7262 }
7263 }
7264 }
61df9d91 7265 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7266 /*
7267 * this also signals 'finished resyncing' to md_stop
7268 */
7269 out:
1da177e4
LT		 7270 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7271
7272 /* tell personality that we are finished */
57afd89f 7273 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7274
dfc70645 7275 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7276 mddev->curr_resync > 2) {
7277 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7278 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7279 if (mddev->curr_resync >= mddev->recovery_cp) {
7280 printk(KERN_INFO
61df9d91
N		 7281 "md: checkpointing %s of %s.\n",
7282 desc, mdname(mddev));
5fd6c1dc
N		 7283 mddev->recovery_cp = mddev->curr_resync;
7284 }
7285 } else
7286 mddev->recovery_cp = MaxSector;
7287 } else {
7288 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7289 mddev->curr_resync = MaxSector;
4e59ca7d
DW		 7290 rcu_read_lock();
7291 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc 7292 if (rdev->raid_disk >= 0 &&
70fffd0b 7293 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7294 !test_bit(Faulty, &rdev->flags) &&
7295 !test_bit(In_sync, &rdev->flags) &&
7296 rdev->recovery_offset < mddev->curr_resync)
7297 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7298 rcu_read_unlock();
5fd6c1dc 7299 }
1da177e4 7300 }
17571284 7301 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7302
1da177e4 7303 skip:
c07b70ad
N		 7304 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7305 /* We completed so min/max setting can be forgotten if used. */
7306 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7307 mddev->resync_min = 0;
7308 mddev->resync_max = MaxSector;
7309 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7310 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7311 mddev->curr_resync = 0;
7312 wake_up(&resync_wait);
7313 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7314 md_wakeup_thread(mddev->thread);
c6207277
N		 7315 return;
7316
7317 interrupted:
7318 /*
7319 * got a signal, exit.
7320 */
7321 printk(KERN_INFO
7322 "md: md_do_sync() got signal ... exiting\n");
7323 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7324 goto out;
7325
1da177e4 7326}
29269553 7327EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7328
fd01b88c 7329static int remove_and_add_spares(struct mddev *mddev)
b4c4c7b8 7330{
3cb03002 7331 struct md_rdev *rdev;
b4c4c7b8
N		 7332 int spares = 0;
7333
97e4f42d
N		 7334 mddev->curr_resync_completed = 0;
7335
159ec1fc 7336 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 7337 if (rdev->raid_disk >= 0 &&
6bfe0b49 7338 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7339 (test_bit(Faulty, &rdev->flags) ||
7340 ! test_bit(In_sync, &rdev->flags)) &&
7341 atomic_read(&rdev->nr_pending)==0) {
7342 if (mddev->pers->hot_remove_disk(
b8321b68 7343 mddev, rdev) == 0) {
36fad858 7344 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8
N		 7345 rdev->raid_disk = -1;
7346 }
7347 }
7348
5389042f 7349 if (mddev->degraded) {
159ec1fc 7350 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7351 if (rdev->raid_disk >= 0 &&
e5427135 7352 !test_bit(In_sync, &rdev->flags) &&
de393cde 7353 !test_bit(Faulty, &rdev->flags))
dfc70645 7354 spares++;
b4c4c7b8
N		 7355 if (rdev->raid_disk < 0
7356 && !test_bit(Faulty, &rdev->flags)) {
7357 rdev->recovery_offset = 0;
199050ea
NB		 7358 if (mddev->pers->
7359 hot_add_disk(mddev, rdev) == 0) {
36fad858 7360 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 7361 /* failure here is OK */;
b4c4c7b8
N		 7362 spares++;
7363 md_new_event(mddev);
93be75ff 7364 set_bit(MD_CHANGE_DEVS, &mddev->flags);
60fc1370 7365 }
b4c4c7b8 7366 }
dfc70645 7367 }
b4c4c7b8
N		 7368 }
7369 return spares;
7370}
7ebc0be7 7371
fd01b88c 7372static void reap_sync_thread(struct mddev *mddev)
7ebc0be7 7373{
3cb03002 7374 struct md_rdev *rdev;
7ebc0be7
N		 7375
7376 /* resync has finished, collect result */
01f96c0a 7377 md_unregister_thread(&mddev->sync_thread);
7ebc0be7
N		 7378 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7379 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7380 /* success...*/
7381 /* activate any spares */
7382 if (mddev->pers->spare_active(mddev))
7383 sysfs_notify(&mddev->kobj, NULL,
7384 "degraded");
7385 }
7386 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7387 mddev->pers->finish_reshape)
7388 mddev->pers->finish_reshape(mddev);
7ebc0be7 7389
d70ed2e4
AW		 7390 /* If array is no-longer degraded, then any saved_raid_disk
7391 * information must be scrapped. Also if any device is now
7392 * In_sync we must scrape the saved_raid_disk for that device
7393 * do the superblock for an incrementally recovered device
7394 * written out.
7ebc0be7 7395 */
d70ed2e4
AW		 7396 list_for_each_entry(rdev, &mddev->disks, same_set)
7397 if (!mddev->degraded ||
7398 test_bit(In_sync, &rdev->flags))
7ebc0be7
N		 7399 rdev->saved_raid_disk = -1;
7400
d70ed2e4 7401 md_update_sb(mddev, 1);
7ebc0be7
N		 7402 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7403 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7404 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7405 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7406 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7407 /* flag recovery needed just to double check */
7408 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7409 sysfs_notify_dirent_safe(mddev->sysfs_action);
7410 md_new_event(mddev);
768e587e
JB		 7411 if (mddev->event_work.func)
7412 queue_work(md_misc_wq, &mddev->event_work);
7ebc0be7
N		 7413}
7414
1da177e4
LT		 7415/*
7416 * This routine is regularly called by all per-raid-array threads to
7417 * deal with generic issues like resync and super-block update.
7418 * Raid personalities that don't have a thread (linear/raid0) do not
7419 * need this as they never do any recovery or update the superblock.
7420 *
7421 * It does not do any resync itself, but rather "forks" off other threads
7422 * to do that as needed.
7423 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7424 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7425 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7426 * and wakeups up this thread which will reap the thread and finish up.
7427 * This thread also removes any faulty devices (with nr_pending == 0).
7428 *
7429 * The overall approach is:
7430 * 1/ if the superblock needs updating, update it.
7431 * 2/ If a recovery thread is running, don't do anything else.
7432 * 3/ If recovery has finished, clean up, possibly marking spares active.
7433 * 4/ If there are any faulty devices, remove them.
7434 * 5/ If array is degraded, try to add spares devices
7435 * 6/ If array has spares or is not in-sync, start a resync thread.
7436 */
fd01b88c 7437void md_check_recovery(struct mddev *mddev)
1da177e4 7438{
68866e42
JB		 7439 if (mddev->suspended)
7440 return;
7441
5f40402d 7442 if (mddev->bitmap)
aa5cbd10 7443 bitmap_daemon_work(mddev);
1da177e4 7444
fca4d848 7445 if (signal_pending(current)) {
31a59e34 7446 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7447 printk(KERN_INFO "md: %s in immediate safe mode\n",
7448 mdname(mddev));
7449 mddev->safemode = 2;
7450 }
7451 flush_signals(current);
7452 }
7453
c89a8eee
N		 7454 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7455 return;
1da177e4 7456 if ( ! (
126925c0 7457 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7458 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7459 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7460 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7461 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7462 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7463 ))
7464 return;
fca4d848 7465
df5b89b3 7466 if (mddev_trylock(mddev)) {
b4c4c7b8 7467 int spares = 0;
fca4d848 7468
c89a8eee
N		 7469 if (mddev->ro) {
7470 /* Only thing we do on a ro array is remove
7471 * failed devices.
7472 */
3cb03002 7473 struct md_rdev *rdev;
a8c42c7f
N		 7474 list_for_each_entry(rdev, &mddev->disks, same_set)
7475 if (rdev->raid_disk >= 0 &&
7476 !test_bit(Blocked, &rdev->flags) &&
7477 test_bit(Faulty, &rdev->flags) &&
7478 atomic_read(&rdev->nr_pending)==0) {
7479 if (mddev->pers->hot_remove_disk(
b8321b68 7480 mddev, rdev) == 0) {
36fad858 7481 sysfs_unlink_rdev(mddev, rdev);
a8c42c7f
N		 7482 rdev->raid_disk = -1;
7483 }
7484 }
c89a8eee
N		 7485 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7486 goto unlock;
7487 }
7488
31a59e34 7489 if (!mddev->external) {
0fd62b86 7490 int did_change = 0;
31a59e34
N		 7491 spin_lock_irq(&mddev->write_lock);
7492 if (mddev->safemode &&
7493 !atomic_read(&mddev->writes_pending) &&
7494 !mddev->in_sync &&
7495 mddev->recovery_cp == MaxSector) {
7496 mddev->in_sync = 1;
0fd62b86 7497 did_change = 1;
070dc6dd 7498 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7499 }
7500 if (mddev->safemode == 1)
7501 mddev->safemode = 0;
7502 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7503 if (did_change)
00bcb4ac 7504 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7505 }
fca4d848 7506
850b2b42
N		 7507 if (mddev->flags)
7508 md_update_sb(mddev, 0);
06d91a5f 7509
1da177e4
LT		 7510 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7511 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7512 /* resync/recovery still happening */
7513 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7514 goto unlock;
7515 }
7516 if (mddev->sync_thread) {
7ebc0be7 7517 reap_sync_thread(mddev);
1da177e4
LT		 7518 goto unlock;
7519 }
72a23c21
NB		 7520 /* Set RUNNING before clearing NEEDED to avoid
7521 * any transients in the value of "sync_action".
7522 */
7523 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7524 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7525 /* Clear some bits that don't mean anything, but
7526 * might be left set
7527 */
24dd469d
N		 7528 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7529 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7530
5fd6c1dc
N		 7531 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7532 goto unlock;
1da177e4
LT		 7533 /* no recovery is running.
7534 * remove any failed drives, then
7535 * add spares if possible.
7536 * Spare are also removed and re-added, to allow
7537 * the personality to fail the re-add.
7538 */
1da177e4 7539
b4c4c7b8 7540 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7541 if (mddev->pers->check_reshape == NULL ||
7542 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7543 /* Cannot proceed */
7544 goto unlock;
7545 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7546 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7547 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7548 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7549 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7550 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7551 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7552 } else if (mddev->recovery_cp < MaxSector) {
7553 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7554 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7555 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7556 /* nothing to be done ... */
1da177e4 7557 goto unlock;
24dd469d 7558
1da177e4 7559 if (mddev->pers->sync_request) {
a654b9d8
N		 7560 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7561 /* We are adding a device or devices to an array
7562 * which has the bitmap stored on all devices.
7563 * So make sure all bitmap pages get written
7564 */
7565 bitmap_write_all(mddev->bitmap);
7566 }
1da177e4
LT		 7567 mddev->sync_thread = md_register_thread(md_do_sync,
7568 mddev,
0da3c619 7569 "resync");
1da177e4
LT		 7570 if (!mddev->sync_thread) {
7571 printk(KERN_ERR "%s: could not start resync"
7572 " thread...\n",
7573 mdname(mddev));
7574 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7575 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7576 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7577 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7578 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7579 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7580 } else
1da177e4 7581 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7582 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7583 md_new_event(mddev);
1da177e4
LT		 7584 }
7585 unlock:
72a23c21
NB		 7586 if (!mddev->sync_thread) {
7587 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7588 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7589 &mddev->recovery))
0c3573f1 7590 if (mddev->sysfs_action)
00bcb4ac 7591 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7592 }
1da177e4
LT		 7593 mddev_unlock(mddev);
7594 }
7595}
7596
fd01b88c 7597void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7598{
00bcb4ac 7599 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7600 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7601 !test_bit(Blocked, &rdev->flags) &&
7602 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7603 msecs_to_jiffies(5000));
7604 rdev_dec_pending(rdev, mddev);
7605}
7606EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7607
2230dfe4
N		 7608
7609/* Bad block management.
7610 * We can record which blocks on each device are 'bad' and so just
7611 * fail those blocks, or that stripe, rather than the whole device.
7612 * Entries in the bad-block table are 64bits wide. This comprises:
7613 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7614 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7615 * A 'shift' can be set so that larger blocks are tracked and
7616 * consequently larger devices can be covered.
7617 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7618 *
7619 * Locking of the bad-block table uses a seqlock so md_is_badblock
7620 * might need to retry if it is very unlucky.
7621 * We will sometimes want to check for bad blocks in a bi_end_io function,
7622 * so we use the write_seqlock_irq variant.
7623 *
7624 * When looking for a bad block we specify a range and want to
7625 * know if any block in the range is bad. So we binary-search
7626 * to the last range that starts at-or-before the given endpoint,
7627 * (or "before the sector after the target range")
7628 * then see if it ends after the given start.
7629 * We return
7630 * 0 if there are no known bad blocks in the range
7631 * 1 if there are known bad block which are all acknowledged
7632 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7633 * plus the start/length of the first bad section we overlap.
7634 */
7635int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7636 sector_t *first_bad, int *bad_sectors)
7637{
7638 int hi;
7639 int lo = 0;
7640 u64 *p = bb->page;
7641 int rv = 0;
7642 sector_t target = s + sectors;
7643 unsigned seq;
7644
7645 if (bb->shift > 0) {
7646 /* round the start down, and the end up */
7647 s >>= bb->shift;
7648 target += (1<<bb->shift) - 1;
7649 target >>= bb->shift;
7650 sectors = target - s;
7651 }
7652 /* 'target' is now the first block after the bad range */
7653
7654retry:
7655 seq = read_seqbegin(&bb->lock);
7656
7657 hi = bb->count;
7658
7659 /* Binary search between lo and hi for 'target'
7660 * i.e. for the last range that starts before 'target'
7661 */
7662 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7663 * are known not to be the last range before target.
7664 * VARIANT: hi-lo is the number of possible
7665 * ranges, and decreases until it reaches 1
7666 */
7667 while (hi - lo > 1) {
7668 int mid = (lo + hi) / 2;
7669 sector_t a = BB_OFFSET(p[mid]);
7670 if (a < target)
7671 /* This could still be the one, earlier ranges
7672 * could not. */
7673 lo = mid;
7674 else
7675 /* This and later ranges are definitely out. */
7676 hi = mid;
7677 }
7678 /* 'lo' might be the last that started before target, but 'hi' isn't */
7679 if (hi > lo) {
7680 /* need to check all range that end after 's' to see if
7681 * any are unacknowledged.
7682 */
7683 while (lo >= 0 &&
7684 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7685 if (BB_OFFSET(p[lo]) < target) {
7686 /* starts before the end, and finishes after
7687 * the start, so they must overlap
7688 */
7689 if (rv != -1 && BB_ACK(p[lo]))
7690 rv = 1;
7691 else
7692 rv = -1;
7693 *first_bad = BB_OFFSET(p[lo]);
7694 *bad_sectors = BB_LEN(p[lo]);
7695 }
7696 lo--;
7697 }
7698 }
7699
7700 if (read_seqretry(&bb->lock, seq))
7701 goto retry;
7702
7703 return rv;
7704}
7705EXPORT_SYMBOL_GPL(md_is_badblock);
7706
7707/*
7708 * Add a range of bad blocks to the table.
7709 * This might extend the table, or might contract it
7710 * if two adjacent ranges can be merged.
7711 * We binary-search to find the 'insertion' point, then
7712 * decide how best to handle it.
7713 */
7714static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7715 int acknowledged)
7716{
7717 u64 *p;
7718 int lo, hi;
7719 int rv = 1;
7720
7721 if (bb->shift < 0)
7722 /* badblocks are disabled */
7723 return 0;
7724
7725 if (bb->shift) {
7726 /* round the start down, and the end up */
7727 sector_t next = s + sectors;
7728 s >>= bb->shift;
7729 next += (1<<bb->shift) - 1;
7730 next >>= bb->shift;
7731 sectors = next - s;
7732 }
7733
7734 write_seqlock_irq(&bb->lock);
7735
7736 p = bb->page;
7737 lo = 0;
7738 hi = bb->count;
7739 /* Find the last range that starts at-or-before 's' */
7740 while (hi - lo > 1) {
7741 int mid = (lo + hi) / 2;
7742 sector_t a = BB_OFFSET(p[mid]);
7743 if (a <= s)
7744 lo = mid;
7745 else
7746 hi = mid;
7747 }
7748 if (hi > lo && BB_OFFSET(p[lo]) > s)
7749 hi = lo;
7750
7751 if (hi > lo) {
7752 /* we found a range that might merge with the start
7753 * of our new range
7754 */
7755 sector_t a = BB_OFFSET(p[lo]);
7756 sector_t e = a + BB_LEN(p[lo]);
7757 int ack = BB_ACK(p[lo]);
7758 if (e >= s) {
7759 /* Yes, we can merge with a previous range */
7760 if (s == a && s + sectors >= e)
7761 /* new range covers old */
7762 ack = acknowledged;
7763 else
7764 ack = ack && acknowledged;
7765
7766 if (e < s + sectors)
7767 e = s + sectors;
7768 if (e - a <= BB_MAX_LEN) {
7769 p[lo] = BB_MAKE(a, e-a, ack);
7770 s = e;
7771 } else {
7772 /* does not all fit in one range,
7773 * make p[lo] maximal
7774 */
7775 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7776 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7777 s = a + BB_MAX_LEN;
7778 }
7779 sectors = e - s;
7780 }
7781 }
7782 if (sectors && hi < bb->count) {
7783 /* 'hi' points to the first range that starts after 's'.
7784 * Maybe we can merge with the start of that range */
7785 sector_t a = BB_OFFSET(p[hi]);
7786 sector_t e = a + BB_LEN(p[hi]);
7787 int ack = BB_ACK(p[hi]);
7788 if (a <= s + sectors) {
7789 /* merging is possible */
7790 if (e <= s + sectors) {
7791 /* full overlap */
7792 e = s + sectors;
7793 ack = acknowledged;
7794 } else
7795 ack = ack && acknowledged;
7796
7797 a = s;
7798 if (e - a <= BB_MAX_LEN) {
7799 p[hi] = BB_MAKE(a, e-a, ack);
7800 s = e;
7801 } else {
7802 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7803 s = a + BB_MAX_LEN;
7804 }
7805 sectors = e - s;
7806 lo = hi;
7807 hi++;
7808 }
7809 }
7810 if (sectors == 0 && hi < bb->count) {
7811 /* we might be able to combine lo and hi */
7812 /* Note: 's' is at the end of 'lo' */
7813 sector_t a = BB_OFFSET(p[hi]);
7814 int lolen = BB_LEN(p[lo]);
7815 int hilen = BB_LEN(p[hi]);
7816 int newlen = lolen + hilen - (s - a);
7817 if (s >= a && newlen < BB_MAX_LEN) {
7818 /* yes, we can combine them */
7819 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7820 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7821 memmove(p + hi, p + hi + 1,
7822 (bb->count - hi - 1) * 8);
7823 bb->count--;
7824 }
7825 }
7826 while (sectors) {
7827 /* didn't merge (it all).
7828 * Need to add a range just before 'hi' */
7829 if (bb->count >= MD_MAX_BADBLOCKS) {
7830 /* No room for more */
7831 rv = 0;
7832 break;
7833 } else {
7834 int this_sectors = sectors;
7835 memmove(p + hi + 1, p + hi,
7836 (bb->count - hi) * 8);
7837 bb->count++;
7838
7839 if (this_sectors > BB_MAX_LEN)
7840 this_sectors = BB_MAX_LEN;
7841 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7842 sectors -= this_sectors;
7843 s += this_sectors;
7844 }
7845 }
7846
7847 bb->changed = 1;
de393cde
N		 7848 if (!acknowledged)
7849 bb->unacked_exist = 1;
2230dfe4
N		 7850 write_sequnlock_irq(&bb->lock);
7851
7852 return rv;
7853}
7854
3cb03002 7855int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
2230dfe4
N		 7856 int acknowledged)
7857{
7858 int rv = md_set_badblocks(&rdev->badblocks,
7859 s + rdev->data_offset, sectors, acknowledged);
7860 if (rv) {
7861 /* Make sure they get written out promptly */
8bd2f0a0 7862 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 7863 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7864 md_wakeup_thread(rdev->mddev->thread);
7865 }
7866 return rv;
7867}
7868EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7869
7870/*
7871 * Remove a range of bad blocks from the table.
7872 * This may involve extending the table if we spilt a region,
7873 * but it must not fail. So if the table becomes full, we just
7874 * drop the remove request.
7875 */
7876static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7877{
7878 u64 *p;
7879 int lo, hi;
7880 sector_t target = s + sectors;
7881 int rv = 0;
7882
7883 if (bb->shift > 0) {
7884 /* When clearing we round the start up and the end down.
7885 * This should not matter as the shift should align with
7886 * the block size and no rounding should ever be needed.
7887 * However it is better the think a block is bad when it
7888 * isn't than to think a block is not bad when it is.
7889 */
7890 s += (1<<bb->shift) - 1;
7891 s >>= bb->shift;
7892 target >>= bb->shift;
7893 sectors = target - s;
7894 }
7895
7896 write_seqlock_irq(&bb->lock);
7897
7898 p = bb->page;
7899 lo = 0;
7900 hi = bb->count;
7901 /* Find the last range that starts before 'target' */
7902 while (hi - lo > 1) {
7903 int mid = (lo + hi) / 2;
7904 sector_t a = BB_OFFSET(p[mid]);
7905 if (a < target)
7906 lo = mid;
7907 else
7908 hi = mid;
7909 }
7910 if (hi > lo) {
7911 /* p[lo] is the last range that could overlap the
7912 * current range. Earlier ranges could also overlap,
7913 * but only this one can overlap the end of the range.
7914 */
7915 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7916 /* Partial overlap, leave the tail of this range */
7917 int ack = BB_ACK(p[lo]);
7918 sector_t a = BB_OFFSET(p[lo]);
7919 sector_t end = a + BB_LEN(p[lo]);
7920
7921 if (a < s) {
7922 /* we need to split this range */
7923 if (bb->count >= MD_MAX_BADBLOCKS) {
7924 rv = 0;
7925 goto out;
7926 }
7927 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7928 bb->count++;
7929 p[lo] = BB_MAKE(a, s-a, ack);
7930 lo++;
7931 }
7932 p[lo] = BB_MAKE(target, end - target, ack);
7933 /* there is no longer an overlap */
7934 hi = lo;
7935 lo--;
7936 }
7937 while (lo >= 0 &&
7938 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7939 /* This range does overlap */
7940 if (BB_OFFSET(p[lo]) < s) {
7941 /* Keep the early parts of this range. */
7942 int ack = BB_ACK(p[lo]);
7943 sector_t start = BB_OFFSET(p[lo]);
7944 p[lo] = BB_MAKE(start, s - start, ack);
7945 /* now low doesn't overlap, so.. */
7946 break;
7947 }
7948 lo--;
7949 }
7950 /* 'lo' is strictly before, 'hi' is strictly after,
7951 * anything between needs to be discarded
7952 */
7953 if (hi - lo > 1) {
7954 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
7955 bb->count -= (hi - lo - 1);
7956 }
7957 }
7958
7959 bb->changed = 1;
7960out:
7961 write_sequnlock_irq(&bb->lock);
7962 return rv;
7963}
7964
3cb03002 7965int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
2230dfe4
N		 7966{
7967 return md_clear_badblocks(&rdev->badblocks,
7968 s + rdev->data_offset,
7969 sectors);
7970}
7971EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
7972
7973/*
7974 * Acknowledge all bad blocks in a list.
7975 * This only succeeds if ->changed is clear. It is used by
7976 * in-kernel metadata updates
7977 */
7978void md_ack_all_badblocks(struct badblocks *bb)
7979{
7980 if (bb->page == NULL || bb->changed)
7981 /* no point even trying */
7982 return;
7983 write_seqlock_irq(&bb->lock);
7984
7985 if (bb->changed == 0) {
7986 u64 *p = bb->page;
7987 int i;
7988 for (i = 0; i < bb->count ; i++) {
7989 if (!BB_ACK(p[i])) {
7990 sector_t start = BB_OFFSET(p[i]);
7991 int len = BB_LEN(p[i]);
7992 p[i] = BB_MAKE(start, len, 1);
7993 }
7994 }
de393cde 7995 bb->unacked_exist = 0;
2230dfe4
N		 7996 }
7997 write_sequnlock_irq(&bb->lock);
7998}
7999EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8000
16c791a5
N		 8001/* sysfs access to bad-blocks list.
8002 * We present two files.
8003 * 'bad-blocks' lists sector numbers and lengths of ranges that
8004 * are recorded as bad. The list is truncated to fit within
8005 * the one-page limit of sysfs.
8006 * Writing "sector length" to this file adds an acknowledged
8007 * bad block list.
8008 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8009 * been acknowledged. Writing to this file adds bad blocks
8010 * without acknowledging them. This is largely for testing.
8011 */
8012
8013static ssize_t
8014badblocks_show(struct badblocks *bb, char *page, int unack)
8015{
8016 size_t len;
8017 int i;
8018 u64 *p = bb->page;
8019 unsigned seq;
8020
8021 if (bb->shift < 0)
8022 return 0;
8023
8024retry:
8025 seq = read_seqbegin(&bb->lock);
8026
8027 len = 0;
8028 i = 0;
8029
8030 while (len < PAGE_SIZE && i < bb->count) {
8031 sector_t s = BB_OFFSET(p[i]);
8032 unsigned int length = BB_LEN(p[i]);
8033 int ack = BB_ACK(p[i]);
8034 i++;
8035
8036 if (unack && ack)
8037 continue;
8038
8039 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8040 (unsigned long long)s << bb->shift,
8041 length << bb->shift);
8042 }
de393cde
N		 8043 if (unack && len == 0)
8044 bb->unacked_exist = 0;
16c791a5
N		 8045
8046 if (read_seqretry(&bb->lock, seq))
8047 goto retry;
8048
8049 return len;
8050}
8051
8052#define DO_DEBUG 1
8053
8054static ssize_t
8055badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8056{
8057 unsigned long long sector;
8058 int length;
8059 char newline;
8060#ifdef DO_DEBUG
8061 /* Allow clearing via sysfs *only* for testing/debugging.
8062 * Normally only a successful write may clear a badblock
8063 */
8064 int clear = 0;
8065 if (page[0] == '-') {
8066 clear = 1;
8067 page++;
8068 }
8069#endif /* DO_DEBUG */
8070
8071 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8072 case 3:
8073 if (newline != '\n')
8074 return -EINVAL;
8075 case 2:
8076 if (length <= 0)
8077 return -EINVAL;
8078 break;
8079 default:
8080 return -EINVAL;
8081 }
8082
8083#ifdef DO_DEBUG
8084 if (clear) {
8085 md_clear_badblocks(bb, sector, length);
8086 return len;
8087 }
8088#endif /* DO_DEBUG */
8089 if (md_set_badblocks(bb, sector, length, !unack))
8090 return len;
8091 else
8092 return -ENOSPC;
8093}
8094
75c96f85
AB		 8095static int md_notify_reboot(struct notifier_block *this,
8096 unsigned long code, void *x)
1da177e4
LT		 8097{
8098 struct list_head *tmp;
fd01b88c 8099 struct mddev *mddev;
2dba6a91 8100 int need_delay = 0;
1da177e4
LT		 8101
8102 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
8103
8104 printk(KERN_INFO "md: stopping all md devices.\n");
8105
2dba6a91 8106 for_each_mddev(mddev, tmp) {
c71d4887 8107 if (mddev_trylock(mddev)) {
2b25000b
N		 8108 /* Force a switch to readonly even array
8109 * appears to still be in use. Hence
8110 * the '100'.
8111 */
a4bd82d0 8112 md_set_readonly(mddev, 100);
c71d4887
NB		 8113 mddev_unlock(mddev);
8114 }
2dba6a91
DB		 8115 need_delay = 1;
8116 }
1da177e4
LT		 8117 /*
8118 * certain more exotic SCSI devices are known to be
8119 * volatile wrt too early system reboots. While the
8120 * right place to handle this issue is the given
8121 * driver, we do want to have a safe RAID driver ...
8122 */
2dba6a91
DB		 8123 if (need_delay)
8124 mdelay(1000*1);
1da177e4
LT		 8125 }
8126 return NOTIFY_DONE;
8127}
8128
75c96f85 8129static struct notifier_block md_notifier = {
1da177e4
LT		 8130 .notifier_call = md_notify_reboot,
8131 .next = NULL,
8132 .priority = INT_MAX, /* before any real devices */
8133};
8134
8135static void md_geninit(void)
8136{
36a4e1fe 8137 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8138
c7705f34 8139 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8140}
8141
75c96f85 8142static int __init md_init(void)
1da177e4 8143{
e804ac78
TH		 8144 int ret = -ENOMEM;
8145
ada609ee 8146 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8147 if (!md_wq)
8148 goto err_wq;
8149
8150 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8151 if (!md_misc_wq)
8152 goto err_misc_wq;
8153
8154 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8155 goto err_md;
8156
8157 if ((ret = register_blkdev(0, "mdp")) < 0)
8158 goto err_mdp;
8159 mdp_major = ret;
8160
3dbd8c2e 8161 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8162 md_probe, NULL, NULL);
8163 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8164 md_probe, NULL, NULL);
8165
1da177e4 8166 register_reboot_notifier(&md_notifier);
0b4d4147 8167 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8168
8169 md_geninit();
d710e138 8170 return 0;
1da177e4 8171
e804ac78
TH		 8172err_mdp:
8173 unregister_blkdev(MD_MAJOR, "md");
8174err_md:
8175 destroy_workqueue(md_misc_wq);
8176err_misc_wq:
8177 destroy_workqueue(md_wq);
8178err_wq:
8179 return ret;
8180}
1da177e4
LT		 8181
8182#ifndef MODULE
8183
8184/*
8185 * Searches all registered partitions for autorun RAID arrays
8186 * at boot time.
8187 */
4d936ec1
ME		 8188
8189static LIST_HEAD(all_detected_devices);
8190struct detected_devices_node {
8191 struct list_head list;
8192 dev_t dev;
8193};
1da177e4
LT		 8194
8195void md_autodetect_dev(dev_t dev)
8196{
4d936ec1
ME		 8197 struct detected_devices_node *node_detected_dev;
8198
8199 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8200 if (node_detected_dev) {
8201 node_detected_dev->dev = dev;
8202 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8203 } else {
8204 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8205 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8206 }
1da177e4
LT		 8207}
8208
8209
8210static void autostart_arrays(int part)
8211{
3cb03002 8212 struct md_rdev *rdev;
4d936ec1
ME		 8213 struct detected_devices_node *node_detected_dev;
8214 dev_t dev;
8215 int i_scanned, i_passed;
1da177e4 8216
4d936ec1
ME		 8217 i_scanned = 0;
8218 i_passed = 0;
1da177e4 8219
4d936ec1 8220 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8221
4d936ec1
ME		 8222 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8223 i_scanned++;
8224 node_detected_dev = list_entry(all_detected_devices.next,
8225 struct detected_devices_node, list);
8226 list_del(&node_detected_dev->list);
8227 dev = node_detected_dev->dev;
8228 kfree(node_detected_dev);
df968c4e 8229 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8230 if (IS_ERR(rdev))
8231 continue;
8232
b2d444d7 8233 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8234 MD_BUG();
8235 continue;
8236 }
d0fae18f 8237 set_bit(AutoDetected, &rdev->flags);
1da177e4 8238 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8239 i_passed++;
1da177e4 8240 }
4d936ec1
ME		 8241
8242 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8243 i_scanned, i_passed);
1da177e4
LT		 8244
8245 autorun_devices(part);
8246}
8247
fdee8ae4 8248#endif /* !MODULE */
1da177e4
LT		 8249
8250static __exit void md_exit(void)
8251{
fd01b88c 8252 struct mddev *mddev;
1da177e4 8253 struct list_head *tmp;
8ab5e4c1 8254
3dbd8c2e 8255 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8256 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8257
3dbd8c2e 8258 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8259 unregister_blkdev(mdp_major, "mdp");
8260 unregister_reboot_notifier(&md_notifier);
8261 unregister_sysctl_table(raid_table_header);
8262 remove_proc_entry("mdstat", NULL);
29ac4aa3 8263 for_each_mddev(mddev, tmp) {
1da177e4 8264 export_array(mddev);
d3374825 8265 mddev->hold_active = 0;
1da177e4 8266 }
e804ac78
TH		 8267 destroy_workqueue(md_misc_wq);
8268 destroy_workqueue(md_wq);
1da177e4
LT		 8269}
8270
685784aa 8271subsys_initcall(md_init);
1da177e4
LT		 8272module_exit(md_exit)
8273
f91de92e
N		 8274static int get_ro(char *buffer, struct kernel_param *kp)
8275{
8276 return sprintf(buffer, "%d", start_readonly);
8277}
8278static int set_ro(const char *val, struct kernel_param *kp)
8279{
8280 char *e;
8281 int num = simple_strtoul(val, &e, 10);
8282 if (*val && (*e == '\0' || *e == '\n')) {
8283 start_readonly = num;
4dbcdc75 8284 return 0;
f91de92e
N		 8285 }
8286 return -EINVAL;
8287}
8288
80ca3a44
N		 8289module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8290module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8291
efeb53c0 8292module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8293
1da177e4
LT		 8294EXPORT_SYMBOL(register_md_personality);
8295EXPORT_SYMBOL(unregister_md_personality);
8296EXPORT_SYMBOL(md_error);
8297EXPORT_SYMBOL(md_done_sync);
8298EXPORT_SYMBOL(md_write_start);
8299EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8300EXPORT_SYMBOL(md_register_thread);
8301EXPORT_SYMBOL(md_unregister_thread);
8302EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 8303EXPORT_SYMBOL(md_check_recovery);
8304MODULE_LICENSE("GPL");
0efb9e61 8305MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8306MODULE_ALIAS("md");
72008652 8307MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)