projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: refine interpretation of "hold_active == UNTIL_IOCTL".
[deliverable/linux.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
2a48fc0a 39#include <linux/mutex.h>
1da177e4 40#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 41#include <linux/poll.h>
16f17b39 42#include <linux/ctype.h>
e7d2860b 43#include <linux/string.h>
fb4d8c76
N		 44#include <linux/hdreg.h>
45#include <linux/proc_fs.h>
46#include <linux/random.h>
056075c7 47#include <linux/module.h>
fb4d8c76 48#include <linux/reboot.h>
32a7627c 49#include <linux/file.h>
aa98aa31 50#include <linux/compat.h>
25570727 51#include <linux/delay.h>
bff61975
N		 52#include <linux/raid/md_p.h>
53#include <linux/raid/md_u.h>
5a0e3ad6 54#include <linux/slab.h>
43b2e5d8 55#include "md.h"
ef740c37 56#include "bitmap.h"
1da177e4 57
1da177e4 58#ifndef MODULE
d710e138 59static void autostart_arrays(int part);
1da177e4
LT		 60#endif
61
01f96c0a
N		 62/* pers_list is a list of registered personalities protected
63 * by pers_lock.
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
66 */
2604b703 67static LIST_HEAD(pers_list);
1da177e4
LT		 68static DEFINE_SPINLOCK(pers_lock);
69
5e56341d
AB		 70static void md_print_devices(void);
71
90b08710 72static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 73static struct workqueue_struct *md_wq;
74static struct workqueue_struct *md_misc_wq;
90b08710 75
5e56341d
AB		 76#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
77
1e50915f
RB		 78/*
79 * Default number of read corrections we'll attempt on an rdev
80 * before ejecting it from the array. We divide the read error
81 * count by 2 for every hour elapsed between read errors.
82 */
83#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 84/*
85 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
86 * is 1000 KB/sec, so the extra system load does not show up that much.
87 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 88 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 89 * subsystem is idle. There is also an 'absolute maximum' reconstruction
90 * speed limit - in case reconstruction slows down your system despite
91 * idle IO detection.
92 *
93 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 94 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 95 */
96
97static int sysctl_speed_limit_min = 1000;
98static int sysctl_speed_limit_max = 200000;
fd01b88c 99static inline int speed_min(struct mddev *mddev)
88202a0c
N		 100{
101 return mddev->sync_speed_min ?
102 mddev->sync_speed_min : sysctl_speed_limit_min;
103}
104
fd01b88c 105static inline int speed_max(struct mddev *mddev)
88202a0c
N		 106{
107 return mddev->sync_speed_max ?
108 mddev->sync_speed_max : sysctl_speed_limit_max;
109}
1da177e4
LT		 110
111static struct ctl_table_header *raid_table_header;
112
113static ctl_table raid_table[] = {
114 {
1da177e4
LT		 115 .procname = "speed_limit_min",
116 .data = &sysctl_speed_limit_min,
117 .maxlen = sizeof(int),
80ca3a44 118 .mode = S_IRUGO|S_IWUSR,
6d456111 119 .proc_handler = proc_dointvec,
1da177e4
LT		 120 },
121 {
1da177e4
LT		 122 .procname = "speed_limit_max",
123 .data = &sysctl_speed_limit_max,
124 .maxlen = sizeof(int),
80ca3a44 125 .mode = S_IRUGO|S_IWUSR,
6d456111 126 .proc_handler = proc_dointvec,
1da177e4 127 },
894d2491 128 { }
1da177e4
LT		 129};
130
131static ctl_table raid_dir_table[] = {
132 {
1da177e4
LT		 133 .procname = "raid",
134 .maxlen = 0,
80ca3a44 135 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 136 .child = raid_table,
137 },
894d2491 138 { }
1da177e4
LT		 139};
140
141static ctl_table raid_root_table[] = {
142 {
1da177e4
LT		 143 .procname = "dev",
144 .maxlen = 0,
145 .mode = 0555,
146 .child = raid_dir_table,
147 },
894d2491 148 { }
1da177e4
LT		 149};
150
83d5cde4 151static const struct block_device_operations md_fops;
1da177e4 152
f91de92e
N		 153static int start_readonly;
154
a167f663
N		 155/* bio_clone_mddev
156 * like bio_clone, but with a local bio set
157 */
158
159static void mddev_bio_destructor(struct bio *bio)
160{
fd01b88c 161 struct mddev *mddev, **mddevp;
a167f663
N		 162
163 mddevp = (void*)bio;
164 mddev = mddevp[-1];
165
166 bio_free(bio, mddev->bio_set);
167}
168
169struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 170 struct mddev *mddev)
a167f663
N		 171{
172 struct bio *b;
fd01b88c 173 struct mddev **mddevp;
a167f663
N		 174
175 if (!mddev || !mddev->bio_set)
176 return bio_alloc(gfp_mask, nr_iovecs);
177
178 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
179 mddev->bio_set);
180 if (!b)
181 return NULL;
182 mddevp = (void*)b;
183 mddevp[-1] = mddev;
184 b->bi_destructor = mddev_bio_destructor;
185 return b;
186}
187EXPORT_SYMBOL_GPL(bio_alloc_mddev);
188
189struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 190 struct mddev *mddev)
a167f663
N		 191{
192 struct bio *b;
fd01b88c 193 struct mddev **mddevp;
a167f663
N		 194
195 if (!mddev || !mddev->bio_set)
196 return bio_clone(bio, gfp_mask);
197
198 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
199 mddev->bio_set);
200 if (!b)
201 return NULL;
202 mddevp = (void*)b;
203 mddevp[-1] = mddev;
204 b->bi_destructor = mddev_bio_destructor;
205 __bio_clone(b, bio);
206 if (bio_integrity(bio)) {
207 int ret;
208
209 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
210
211 if (ret < 0) {
212 bio_put(b);
213 return NULL;
214 }
215 }
216
217 return b;
218}
219EXPORT_SYMBOL_GPL(bio_clone_mddev);
220
d2eb35ac
N		 221void md_trim_bio(struct bio *bio, int offset, int size)
222{
223 /* 'bio' is a cloned bio which we need to trim to match
224 * the given offset and size.
225 * This requires adjusting bi_sector, bi_size, and bi_io_vec
226 */
227 int i;
228 struct bio_vec *bvec;
229 int sofar = 0;
230
231 size <<= 9;
232 if (offset == 0 && size == bio->bi_size)
233 return;
234
235 bio->bi_sector += offset;
236 bio->bi_size = size;
237 offset <<= 9;
238 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
239
240 while (bio->bi_idx < bio->bi_vcnt &&
241 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
242 /* remove this whole bio_vec */
243 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
244 bio->bi_idx++;
245 }
246 if (bio->bi_idx < bio->bi_vcnt) {
247 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
248 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
249 }
250 /* avoid any complications with bi_idx being non-zero*/
251 if (bio->bi_idx) {
252 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
253 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
254 bio->bi_vcnt -= bio->bi_idx;
255 bio->bi_idx = 0;
256 }
257 /* Make sure vcnt and last bv are not too big */
258 bio_for_each_segment(bvec, bio, i) {
259 if (sofar + bvec->bv_len > size)
260 bvec->bv_len = size - sofar;
261 if (bvec->bv_len == 0) {
262 bio->bi_vcnt = i;
263 break;
264 }
265 sofar += bvec->bv_len;
266 }
267}
268EXPORT_SYMBOL_GPL(md_trim_bio);
269
d7603b7e
N		 270/*
271 * We have a system wide 'event count' that is incremented
272 * on any 'interesting' event, and readers of /proc/mdstat
273 * can use 'poll' or 'select' to find out when the event
274 * count increases.
275 *
276 * Events are:
277 * start array, stop array, error, add device, remove device,
278 * start build, activate spare
279 */
2989ddbd 280static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 281static atomic_t md_event_count;
fd01b88c 282void md_new_event(struct mddev *mddev)
d7603b7e
N		 283{
284 atomic_inc(&md_event_count);
285 wake_up(&md_event_waiters);
286}
29269553 287EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 288
c331eb04
N		 289/* Alternate version that can be called from interrupts
290 * when calling sysfs_notify isn't needed.
291 */
fd01b88c 292static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 293{
294 atomic_inc(&md_event_count);
295 wake_up(&md_event_waiters);
296}
297
1da177e4
LT		 298/*
299 * Enables to iterate over all existing md arrays
300 * all_mddevs_lock protects this list.
301 */
302static LIST_HEAD(all_mddevs);
303static DEFINE_SPINLOCK(all_mddevs_lock);
304
305
306/*
307 * iterates through all used mddevs in the system.
308 * We take care to grab the all_mddevs_lock whenever navigating
309 * the list, and to always hold a refcount when unlocked.
310 * Any code which breaks out of this loop while own
311 * a reference to the current mddev and must mddev_put it.
312 */
fd01b88c 313#define for_each_mddev(_mddev,_tmp) \
1da177e4
LT		 314 \
315 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 316 _tmp = all_mddevs.next; \
317 _mddev = NULL;}); \
318 ({ if (_tmp != &all_mddevs) \
319 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 320 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 321 if (_mddev) mddev_put(_mddev); \
322 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
323 _tmp != &all_mddevs;}); \
1da177e4 324 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 325 _tmp = _tmp->next;}) \
1da177e4
LT		 326 )
327
328
409c57f3
N		 329/* Rather than calling directly into the personality make_request function,
330 * IO requests come here first so that we can check if the device is
331 * being suspended pending a reconfiguration.
332 * We hold a refcount over the call to ->make_request. By the time that
333 * call has finished, the bio has been linked into some internal structure
334 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 */
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 */
1da177e4 573 list_del(&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
N		 2548 if (test_bit(Blocked, &rdev->flags) ||
2549 rdev->badblocks.unacked_exist) {
6bfe0b49
DW		 2550 len += sprintf(page+len, "%sblocked", sep);
2551 sep = ",";
2552 }
b2d444d7
N		 2553 if (!test_bit(Faulty, &rdev->flags) &&
2554 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2555 len += sprintf(page+len, "%sspare", sep);
2556 sep = ",";
2557 }
d7a9d443
N		 2558 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2559 len += sprintf(page+len, "%swrite_error", sep);
2560 sep = ",";
2561 }
86e6ffdd
N		 2562 return len+sprintf(page+len, "\n");
2563}
2564
45dc2de1 2565static ssize_t
3cb03002 2566state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2567{
2568 /* can write
de393cde 2569 * faulty - simulates an error
45dc2de1 2570 * remove - disconnects the device
f655675b
N		 2571 * writemostly - sets write_mostly
2572 * -writemostly - clears write_mostly
de393cde
N		 2573 * blocked - sets the Blocked flags
2574 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2575 * insync - sets Insync providing device isn't active
d7a9d443
N		 2576 * write_error - sets WriteErrorSeen
2577 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2578 */
2579 int err = -EINVAL;
2580 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2581 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2582 if (test_bit(Faulty, &rdev->flags))
2583 err = 0;
2584 else
2585 err = -EBUSY;
45dc2de1
N		 2586 } else if (cmd_match(buf, "remove")) {
2587 if (rdev->raid_disk >= 0)
2588 err = -EBUSY;
2589 else {
fd01b88c 2590 struct mddev *mddev = rdev->mddev;
45dc2de1 2591 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2592 if (mddev->pers)
2593 md_update_sb(mddev, 1);
45dc2de1
N		 2594 md_new_event(mddev);
2595 err = 0;
2596 }
f655675b
N		 2597 } else if (cmd_match(buf, "writemostly")) {
2598 set_bit(WriteMostly, &rdev->flags);
2599 err = 0;
2600 } else if (cmd_match(buf, "-writemostly")) {
2601 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2602 err = 0;
2603 } else if (cmd_match(buf, "blocked")) {
2604 set_bit(Blocked, &rdev->flags);
2605 err = 0;
2606 } else if (cmd_match(buf, "-blocked")) {
de393cde 2607 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2608 rdev->badblocks.unacked_exist) {
de393cde
N		 2609 /* metadata handler doesn't understand badblocks,
2610 * so we need to fail the device
2611 */
2612 md_error(rdev->mddev, rdev);
2613 }
6bfe0b49 2614 clear_bit(Blocked, &rdev->flags);
de393cde 2615 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2616 wake_up(&rdev->blocked_wait);
2617 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2618 md_wakeup_thread(rdev->mddev->thread);
2619
6d56e278
N		 2620 err = 0;
2621 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2622 set_bit(In_sync, &rdev->flags);
f655675b 2623 err = 0;
d7a9d443
N		 2624 } else if (cmd_match(buf, "write_error")) {
2625 set_bit(WriteErrorSeen, &rdev->flags);
2626 err = 0;
2627 } else if (cmd_match(buf, "-write_error")) {
2628 clear_bit(WriteErrorSeen, &rdev->flags);
2629 err = 0;
45dc2de1 2630 }
00bcb4ac
N		 2631 if (!err)
2632 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2633 return err ? err : len;
2634}
80ca3a44
N		 2635static struct rdev_sysfs_entry rdev_state =
2636__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2637
4dbcdc75 2638static ssize_t
3cb03002 2639errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2640{
2641 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2642}
2643
2644static ssize_t
3cb03002 2645errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2646{
2647 char *e;
2648 unsigned long n = simple_strtoul(buf, &e, 10);
2649 if (*buf && (*e == 0 || *e == '\n')) {
2650 atomic_set(&rdev->corrected_errors, n);
2651 return len;
2652 }
2653 return -EINVAL;
2654}
2655static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2656__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2657
014236d2 2658static ssize_t
3cb03002 2659slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2660{
2661 if (rdev->raid_disk < 0)
2662 return sprintf(page, "none\n");
2663 else
2664 return sprintf(page, "%d\n", rdev->raid_disk);
2665}
2666
2667static ssize_t
3cb03002 2668slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2669{
2670 char *e;
c303da6d 2671 int err;
014236d2
N		 2672 int slot = simple_strtoul(buf, &e, 10);
2673 if (strncmp(buf, "none", 4)==0)
2674 slot = -1;
2675 else if (e==buf || (*e && *e!= '\n'))
2676 return -EINVAL;
6c2fce2e 2677 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2678 /* Setting 'slot' on an active array requires also
2679 * updating the 'rd%d' link, and communicating
2680 * with the personality with ->hot_*_disk.
2681 * For now we only support removing
2682 * failed/spare devices. This normally happens automatically,
2683 * but not when the metadata is externally managed.
2684 */
c303da6d
N		 2685 if (rdev->raid_disk == -1)
2686 return -EEXIST;
2687 /* personality does all needed checks */
01393f3d 2688 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d
N		 2689 return -EINVAL;
2690 err = rdev->mddev->pers->
2691 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2692 if (err)
2693 return err;
36fad858 2694 sysfs_unlink_rdev(rdev->mddev, rdev);
b7103107 2695 rdev->raid_disk = -1;
c303da6d
N		 2696 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2697 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2698 } else if (rdev->mddev->pers) {
3cb03002 2699 struct md_rdev *rdev2;
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
159ec1fc 2713 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
6c2fce2e
NB		 2714 if (rdev2->raid_disk == slot)
2715 return -EEXIST;
2716
ba1b41b6
N		 2717 if (slot >= rdev->mddev->raid_disks &&
2718 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2719 return -ENOSPC;
2720
6c2fce2e
NB		 2721 rdev->raid_disk = slot;
2722 if (test_bit(In_sync, &rdev->flags))
2723 rdev->saved_raid_disk = slot;
2724 else
2725 rdev->saved_raid_disk = -1;
d30519fc 2726 clear_bit(In_sync, &rdev->flags);
6c2fce2e
NB		 2727 err = rdev->mddev->pers->
2728 hot_add_disk(rdev->mddev, rdev);
199050ea 2729 if (err) {
6c2fce2e 2730 rdev->raid_disk = -1;
6c2fce2e 2731 return err;
52664732 2732 } else
00bcb4ac 2733 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2734 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2735 /* failure here is OK */;
6c2fce2e 2736 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2737 } else {
ba1b41b6
N		 2738 if (slot >= rdev->mddev->raid_disks &&
2739 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2740 return -ENOSPC;
2741 rdev->raid_disk = slot;
2742 /* assume it is working */
c5d79adb
N		 2743 clear_bit(Faulty, &rdev->flags);
2744 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2745 set_bit(In_sync, &rdev->flags);
00bcb4ac 2746 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2747 }
014236d2
N		 2748 return len;
2749}
2750
2751
2752static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2753__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2754
93c8cad0 2755static ssize_t
3cb03002 2756offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2757{
6961ece4 2758 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2759}
2760
2761static ssize_t
3cb03002 2762offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0
N		 2763{
2764 char *e;
2765 unsigned long long offset = simple_strtoull(buf, &e, 10);
2766 if (e==buf || (*e && *e != '\n'))
2767 return -EINVAL;
8ed0a521 2768 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2769 return -EBUSY;
dd8ac336 2770 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2771 /* Must set offset before size, so overlap checks
2772 * can be sane */
2773 return -EBUSY;
93c8cad0
N		 2774 rdev->data_offset = offset;
2775 return len;
2776}
2777
2778static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2779__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2780
83303b61 2781static ssize_t
3cb03002 2782rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2783{
dd8ac336 2784 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2785}
2786
c5d79adb
N		 2787static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2788{
2789 /* check if two start/length pairs overlap */
2790 if (s1+l1 <= s2)
2791 return 0;
2792 if (s2+l2 <= s1)
2793 return 0;
2794 return 1;
2795}
2796
b522adcd
DW		 2797static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2798{
2799 unsigned long long blocks;
2800 sector_t new;
2801
2802 if (strict_strtoull(buf, 10, &blocks) < 0)
2803 return -EINVAL;
2804
2805 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2806 return -EINVAL; /* sector conversion overflow */
2807
2808 new = blocks * 2;
2809 if (new != blocks * 2)
2810 return -EINVAL; /* unsigned long long to sector_t overflow */
2811
2812 *sectors = new;
2813 return 0;
2814}
2815
83303b61 2816static ssize_t
3cb03002 2817rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2818{
fd01b88c 2819 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2820 sector_t oldsectors = rdev->sectors;
b522adcd 2821 sector_t sectors;
27c529bb 2822
b522adcd 2823 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2824 return -EINVAL;
0cd17fec 2825 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2826 if (my_mddev->persistent) {
dd8ac336
AN		 2827 sectors = super_types[my_mddev->major_version].
2828 rdev_size_change(rdev, sectors);
2829 if (!sectors)
0cd17fec 2830 return -EBUSY;
dd8ac336 2831 } else if (!sectors)
77304d2a 2832 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2833 rdev->data_offset;
0cd17fec 2834 }
dd8ac336 2835 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2836 return -EINVAL; /* component must fit device */
0cd17fec 2837
dd8ac336
AN		 2838 rdev->sectors = sectors;
2839 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2840 /* need to check that all other rdevs with the same ->bdev
2841 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2842 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2843 * we have to change it back, we will have the lock again.
2844 */
fd01b88c 2845 struct mddev *mddev;
c5d79adb 2846 int overlap = 0;
159ec1fc 2847 struct list_head *tmp;
c5d79adb 2848
27c529bb 2849 mddev_unlock(my_mddev);
29ac4aa3 2850 for_each_mddev(mddev, tmp) {
3cb03002 2851 struct md_rdev *rdev2;
c5d79adb
N		 2852
2853 mddev_lock(mddev);
159ec1fc 2854 list_for_each_entry(rdev2, &mddev->disks, same_set)
f21e9ff7
N		 2855 if (rdev->bdev == rdev2->bdev &&
2856 rdev != rdev2 &&
2857 overlaps(rdev->data_offset, rdev->sectors,
2858 rdev2->data_offset,
2859 rdev2->sectors)) {
c5d79adb
N		 2860 overlap = 1;
2861 break;
2862 }
2863 mddev_unlock(mddev);
2864 if (overlap) {
2865 mddev_put(mddev);
2866 break;
2867 }
2868 }
27c529bb 2869 mddev_lock(my_mddev);
c5d79adb
N		 2870 if (overlap) {
2871 /* Someone else could have slipped in a size
2872 * change here, but doing so is just silly.
dd8ac336 2873 * We put oldsectors back because we *know* it is
c5d79adb
N		 2874 * safe, and trust userspace not to race with
2875 * itself
2876 */
dd8ac336 2877 rdev->sectors = oldsectors;
c5d79adb
N		 2878 return -EBUSY;
2879 }
2880 }
83303b61
N		 2881 return len;
2882}
2883
2884static struct rdev_sysfs_entry rdev_size =
80ca3a44 2885__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2886
06e3c817 2887
3cb03002 2888static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2889{
2890 unsigned long long recovery_start = rdev->recovery_offset;
2891
2892 if (test_bit(In_sync, &rdev->flags) ||
2893 recovery_start == MaxSector)
2894 return sprintf(page, "none\n");
2895
2896 return sprintf(page, "%llu\n", recovery_start);
2897}
2898
3cb03002 2899static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2900{
2901 unsigned long long recovery_start;
2902
2903 if (cmd_match(buf, "none"))
2904 recovery_start = MaxSector;
2905 else if (strict_strtoull(buf, 10, &recovery_start))
2906 return -EINVAL;
2907
2908 if (rdev->mddev->pers &&
2909 rdev->raid_disk >= 0)
2910 return -EBUSY;
2911
2912 rdev->recovery_offset = recovery_start;
2913 if (recovery_start == MaxSector)
2914 set_bit(In_sync, &rdev->flags);
2915 else
2916 clear_bit(In_sync, &rdev->flags);
2917 return len;
2918}
2919
2920static struct rdev_sysfs_entry rdev_recovery_start =
2921__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2922
16c791a5
N		 2923
2924static ssize_t
2925badblocks_show(struct badblocks *bb, char *page, int unack);
2926static ssize_t
2927badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2928
3cb03002 2929static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2930{
2931 return badblocks_show(&rdev->badblocks, page, 0);
2932}
3cb03002 2933static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2934{
de393cde
N		 2935 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2936 /* Maybe that ack was all we needed */
2937 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2938 wake_up(&rdev->blocked_wait);
2939 return rv;
16c791a5
N		 2940}
2941static struct rdev_sysfs_entry rdev_bad_blocks =
2942__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2943
2944
3cb03002 2945static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2946{
2947 return badblocks_show(&rdev->badblocks, page, 1);
2948}
3cb03002 2949static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2950{
2951 return badblocks_store(&rdev->badblocks, page, len, 1);
2952}
2953static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2954__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2955
86e6ffdd
N		 2956static struct attribute *rdev_default_attrs[] = {
2957 &rdev_state.attr,
4dbcdc75 2958 &rdev_errors.attr,
014236d2 2959 &rdev_slot.attr,
93c8cad0 2960 &rdev_offset.attr,
83303b61 2961 &rdev_size.attr,
06e3c817 2962 &rdev_recovery_start.attr,
16c791a5
N		 2963 &rdev_bad_blocks.attr,
2964 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 2965 NULL,
2966};
2967static ssize_t
2968rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2969{
2970 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2971 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 2972 struct mddev *mddev = rdev->mddev;
27c529bb 2973 ssize_t rv;
86e6ffdd
N		 2974
2975 if (!entry->show)
2976 return -EIO;
27c529bb
N		 2977
2978 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2979 if (!rv) {
2980 if (rdev->mddev == NULL)
2981 rv = -EBUSY;
2982 else
2983 rv = entry->show(rdev, page);
2984 mddev_unlock(mddev);
2985 }
2986 return rv;
86e6ffdd
N		 2987}
2988
2989static ssize_t
2990rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2991 const char *page, size_t length)
2992{
2993 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2994 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 2995 ssize_t rv;
fd01b88c 2996 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 2997
2998 if (!entry->store)
2999 return -EIO;
67463acb
N		 3000 if (!capable(CAP_SYS_ADMIN))
3001 return -EACCES;
27c529bb 3002 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3003 if (!rv) {
27c529bb
N		 3004 if (rdev->mddev == NULL)
3005 rv = -EBUSY;
3006 else
3007 rv = entry->store(rdev, page, length);
6a51830e 3008 mddev_unlock(mddev);
ca388059
N		 3009 }
3010 return rv;
86e6ffdd
N		 3011}
3012
3013static void rdev_free(struct kobject *ko)
3014{
3cb03002 3015 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3016 kfree(rdev);
3017}
52cf25d0 3018static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3019 .show = rdev_attr_show,
3020 .store = rdev_attr_store,
3021};
3022static struct kobj_type rdev_ktype = {
3023 .release = rdev_free,
3024 .sysfs_ops = &rdev_sysfs_ops,
3025 .default_attrs = rdev_default_attrs,
3026};
3027
3cb03002 3028int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3029{
3030 rdev->desc_nr = -1;
3031 rdev->saved_raid_disk = -1;
3032 rdev->raid_disk = -1;
3033 rdev->flags = 0;
3034 rdev->data_offset = 0;
3035 rdev->sb_events = 0;
3036 rdev->last_read_error.tv_sec = 0;
3037 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3038 rdev->sb_loaded = 0;
3039 rdev->bb_page = NULL;
e8bb9a83
N		 3040 atomic_set(&rdev->nr_pending, 0);
3041 atomic_set(&rdev->read_errors, 0);
3042 atomic_set(&rdev->corrected_errors, 0);
3043
3044 INIT_LIST_HEAD(&rdev->same_set);
3045 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3046
3047 /* Add space to store bad block list.
3048 * This reserves the space even on arrays where it cannot
3049 * be used - I wonder if that matters
3050 */
3051 rdev->badblocks.count = 0;
3052 rdev->badblocks.shift = 0;
3053 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3054 seqlock_init(&rdev->badblocks.lock);
3055 if (rdev->badblocks.page == NULL)
3056 return -ENOMEM;
3057
3058 return 0;
e8bb9a83
N		 3059}
3060EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3061/*
3062 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3063 *
3064 * mark the device faulty if:
3065 *
3066 * - the device is nonexistent (zero size)
3067 * - the device has no valid superblock
3068 *
3069 * a faulty rdev _never_ has rdev->sb set.
3070 */
3cb03002 3071static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3072{
3073 char b[BDEVNAME_SIZE];
3074 int err;
3cb03002 3075 struct md_rdev *rdev;
1da177e4
LT		 3076 sector_t size;
3077
9ffae0cf 3078 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3079 if (!rdev) {
3080 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3081 return ERR_PTR(-ENOMEM);
3082 }
1da177e4 3083
2230dfe4
N		 3084 err = md_rdev_init(rdev);
3085 if (err)
3086 goto abort_free;
3087 err = alloc_disk_sb(rdev);
3088 if (err)
1da177e4
LT		 3089 goto abort_free;
3090
c5d79adb 3091 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3092 if (err)
3093 goto abort_free;
3094
f9cb074b 3095 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3096
77304d2a 3097 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 3098 if (!size) {
3099 printk(KERN_WARNING
3100 "md: %s has zero or unknown size, marking faulty!\n",
3101 bdevname(rdev->bdev,b));
3102 err = -EINVAL;
3103 goto abort_free;
3104 }
3105
3106 if (super_format >= 0) {
3107 err = super_types[super_format].
3108 load_super(rdev, NULL, super_minor);
3109 if (err == -EINVAL) {
df968c4e
N		 3110 printk(KERN_WARNING
3111 "md: %s does not have a valid v%d.%d "
3112 "superblock, not importing!\n",
3113 bdevname(rdev->bdev,b),
3114 super_format, super_minor);
1da177e4
LT		 3115 goto abort_free;
3116 }
3117 if (err < 0) {
3118 printk(KERN_WARNING
3119 "md: could not read %s's sb, not importing!\n",
3120 bdevname(rdev->bdev,b));
3121 goto abort_free;
3122 }
3123 }
9f2f3830
N		 3124 if (super_format == -1)
3125 /* hot-add for 0.90, or non-persistent: so no badblocks */
3126 rdev->badblocks.shift = -1;
6bfe0b49 3127
1da177e4
LT		 3128 return rdev;
3129
3130abort_free:
2699b672
N		 3131 if (rdev->bdev)
3132 unlock_rdev(rdev);
3133 free_disk_sb(rdev);
2230dfe4 3134 kfree(rdev->badblocks.page);
1da177e4
LT		 3135 kfree(rdev);
3136 return ERR_PTR(err);
3137}
3138
3139/*
3140 * Check a full RAID array for plausibility
3141 */
3142
3143
fd01b88c 3144static void analyze_sbs(struct mddev * mddev)
1da177e4
LT		 3145{
3146 int i;
3cb03002 3147 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3148 char b[BDEVNAME_SIZE];
3149
3150 freshest = NULL;
d089c6af 3151 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 3152 switch (super_types[mddev->major_version].
3153 load_super(rdev, freshest, mddev->minor_version)) {
3154 case 1:
3155 freshest = rdev;
3156 break;
3157 case 0:
3158 break;
3159 default:
3160 printk( KERN_ERR \
3161 "md: fatal superblock inconsistency in %s"
3162 " -- removing from array\n",
3163 bdevname(rdev->bdev,b));
3164 kick_rdev_from_array(rdev);
3165 }
3166
3167
3168 super_types[mddev->major_version].
3169 validate_super(mddev, freshest);
3170
3171 i = 0;
d089c6af 3172 rdev_for_each(rdev, tmp, mddev) {
233fca36
N		 3173 if (mddev->max_disks &&
3174 (rdev->desc_nr >= mddev->max_disks ||
3175 i > mddev->max_disks)) {
de01dfad
N		 3176 printk(KERN_WARNING
3177 "md: %s: %s: only %d devices permitted\n",
3178 mdname(mddev), bdevname(rdev->bdev, b),
3179 mddev->max_disks);
3180 kick_rdev_from_array(rdev);
3181 continue;
3182 }
1da177e4
LT		 3183 if (rdev != freshest)
3184 if (super_types[mddev->major_version].
3185 validate_super(mddev, rdev)) {
3186 printk(KERN_WARNING "md: kicking non-fresh %s"
3187 " from array!\n",
3188 bdevname(rdev->bdev,b));
3189 kick_rdev_from_array(rdev);
3190 continue;
3191 }
3192 if (mddev->level == LEVEL_MULTIPATH) {
3193 rdev->desc_nr = i++;
3194 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3195 set_bit(In_sync, &rdev->flags);
5e5e3e78 3196 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3197 rdev->raid_disk = -1;
3198 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3199 }
3200 }
1da177e4
LT		 3201}
3202
72e02075
N		 3203/* Read a fixed-point number.
3204 * Numbers in sysfs attributes should be in "standard" units where
3205 * possible, so time should be in seconds.
3206 * However we internally use a a much smaller unit such as
3207 * milliseconds or jiffies.
3208 * This function takes a decimal number with a possible fractional
3209 * component, and produces an integer which is the result of
3210 * multiplying that number by 10^'scale'.
3211 * all without any floating-point arithmetic.
3212 */
3213int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3214{
3215 unsigned long result = 0;
3216 long decimals = -1;
3217 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3218 if (*cp == '.')
3219 decimals = 0;
3220 else if (decimals < scale) {
3221 unsigned int value;
3222 value = *cp - '0';
3223 result = result * 10 + value;
3224 if (decimals >= 0)
3225 decimals++;
3226 }
3227 cp++;
3228 }
3229 if (*cp == '\n')
3230 cp++;
3231 if (*cp)
3232 return -EINVAL;
3233 if (decimals < 0)
3234 decimals = 0;
3235 while (decimals < scale) {
3236 result *= 10;
3237 decimals ++;
3238 }
3239 *res = result;
3240 return 0;
3241}
3242
3243
19052c0e
N		 3244static void md_safemode_timeout(unsigned long data);
3245
16f17b39 3246static ssize_t
fd01b88c 3247safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3248{
3249 int msec = (mddev->safemode_delay*1000)/HZ;
3250 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3251}
3252static ssize_t
fd01b88c 3253safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3254{
16f17b39 3255 unsigned long msec;
97ce0a7f 3256
72e02075 3257 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3258 return -EINVAL;
16f17b39
N		 3259 if (msec == 0)
3260 mddev->safemode_delay = 0;
3261 else {
19052c0e 3262 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 3263 mddev->safemode_delay = (msec*HZ)/1000;
3264 if (mddev->safemode_delay == 0)
3265 mddev->safemode_delay = 1;
19052c0e
N		 3266 if (mddev->safemode_delay < old_delay)
3267 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 3268 }
3269 return len;
3270}
3271static struct md_sysfs_entry md_safe_delay =
80ca3a44 3272__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3273
eae1701f 3274static ssize_t
fd01b88c 3275level_show(struct mddev *mddev, char *page)
eae1701f 3276{
84fc4b56 3277 struct md_personality *p = mddev->pers;
d9d166c2 3278 if (p)
eae1701f 3279 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 3280 else if (mddev->clevel[0])
3281 return sprintf(page, "%s\n", mddev->clevel);
3282 else if (mddev->level != LEVEL_NONE)
3283 return sprintf(page, "%d\n", mddev->level);
3284 else
3285 return 0;
eae1701f
N		 3286}
3287
d9d166c2 3288static ssize_t
fd01b88c 3289level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3290{
f2859af6 3291 char clevel[16];
20a49ff6 3292 ssize_t rv = len;
84fc4b56 3293 struct md_personality *pers;
f2859af6 3294 long level;
245f46c2 3295 void *priv;
3cb03002 3296 struct md_rdev *rdev;
245f46c2
N		 3297
3298 if (mddev->pers == NULL) {
3299 if (len == 0)
3300 return 0;
3301 if (len >= sizeof(mddev->clevel))
3302 return -ENOSPC;
3303 strncpy(mddev->clevel, buf, len);
3304 if (mddev->clevel[len-1] == '\n')
3305 len--;
3306 mddev->clevel[len] = 0;
3307 mddev->level = LEVEL_NONE;
3308 return rv;
3309 }
3310
3311 /* request to change the personality. Need to ensure:
3312 * - array is not engaged in resync/recovery/reshape
3313 * - old personality can be suspended
3314 * - new personality will access other array.
3315 */
3316
bb4f1e9d
N		 3317 if (mddev->sync_thread ||
3318 mddev->reshape_position != MaxSector ||
3319 mddev->sysfs_active)
d9d166c2 3320 return -EBUSY;
245f46c2
N		 3321
3322 if (!mddev->pers->quiesce) {
3323 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3324 mdname(mddev), mddev->pers->name);
3325 return -EINVAL;
3326 }
3327
3328 /* Now find the new personality */
f2859af6 3329 if (len == 0 || len >= sizeof(clevel))
245f46c2 3330 return -EINVAL;
f2859af6
DW		 3331 strncpy(clevel, buf, len);
3332 if (clevel[len-1] == '\n')
d9d166c2 3333 len--;
f2859af6
DW		 3334 clevel[len] = 0;
3335 if (strict_strtol(clevel, 10, &level))
3336 level = LEVEL_NONE;
245f46c2 3337
f2859af6
DW		 3338 if (request_module("md-%s", clevel) != 0)
3339 request_module("md-level-%s", clevel);
245f46c2 3340 spin_lock(&pers_lock);
f2859af6 3341 pers = find_pers(level, clevel);
245f46c2
N		 3342 if (!pers || !try_module_get(pers->owner)) {
3343 spin_unlock(&pers_lock);
f2859af6 3344 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3345 return -EINVAL;
3346 }
3347 spin_unlock(&pers_lock);
3348
3349 if (pers == mddev->pers) {
3350 /* Nothing to do! */
3351 module_put(pers->owner);
3352 return rv;
3353 }
3354 if (!pers->takeover) {
3355 module_put(pers->owner);
3356 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3357 mdname(mddev), clevel);
245f46c2
N		 3358 return -EINVAL;
3359 }
3360
e93f68a1
N		 3361 list_for_each_entry(rdev, &mddev->disks, same_set)
3362 rdev->new_raid_disk = rdev->raid_disk;
3363
245f46c2
N		 3364 /* ->takeover must set new_* and/or delta_disks
3365 * if it succeeds, and may set them when it fails.
3366 */
3367 priv = pers->takeover(mddev);
3368 if (IS_ERR(priv)) {
3369 mddev->new_level = mddev->level;
3370 mddev->new_layout = mddev->layout;
664e7c41 3371 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3372 mddev->raid_disks -= mddev->delta_disks;
3373 mddev->delta_disks = 0;
3374 module_put(pers->owner);
3375 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3376 mdname(mddev), clevel);
245f46c2
N		 3377 return PTR_ERR(priv);
3378 }
3379
3380 /* Looks like we have a winner */
3381 mddev_suspend(mddev);
3382 mddev->pers->stop(mddev);
a64c876f
N		 3383
3384 if (mddev->pers->sync_request == NULL &&
3385 pers->sync_request != NULL) {
3386 /* need to add the md_redundancy_group */
3387 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3388 printk(KERN_WARNING
3389 "md: cannot register extra attributes for %s\n",
3390 mdname(mddev));
19fdb9ee 3391 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3392 }
3393 if (mddev->pers->sync_request != NULL &&
3394 pers->sync_request == NULL) {
3395 /* need to remove the md_redundancy_group */
3396 if (mddev->to_remove == NULL)
3397 mddev->to_remove = &md_redundancy_group;
3398 }
3399
54071b38
TM		 3400 if (mddev->pers->sync_request == NULL &&
3401 mddev->external) {
3402 /* We are converting from a no-redundancy array
3403 * to a redundancy array and metadata is managed
3404 * externally so we need to be sure that writes
3405 * won't block due to a need to transition
3406 * clean->dirty
3407 * until external management is started.
3408 */
3409 mddev->in_sync = 0;
3410 mddev->safemode_delay = 0;
3411 mddev->safemode = 0;
3412 }
3413
e93f68a1 3414 list_for_each_entry(rdev, &mddev->disks, same_set) {
e93f68a1
N		 3415 if (rdev->raid_disk < 0)
3416 continue;
bf2cb0da 3417 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3418 rdev->new_raid_disk = -1;
3419 if (rdev->new_raid_disk == rdev->raid_disk)
3420 continue;
36fad858 3421 sysfs_unlink_rdev(mddev, rdev);
e93f68a1
N		 3422 }
3423 list_for_each_entry(rdev, &mddev->disks, same_set) {
3424 if (rdev->raid_disk < 0)
3425 continue;
3426 if (rdev->new_raid_disk == rdev->raid_disk)
3427 continue;
3428 rdev->raid_disk = rdev->new_raid_disk;
3429 if (rdev->raid_disk < 0)
3a981b03 3430 clear_bit(In_sync, &rdev->flags);
e93f68a1 3431 else {
36fad858
NK		 3432 if (sysfs_link_rdev(mddev, rdev))
3433 printk(KERN_WARNING "md: cannot register rd%d"
3434 " for %s after level change\n",
3435 rdev->raid_disk, mdname(mddev));
3a981b03 3436 }
e93f68a1
N		 3437 }
3438
3439 module_put(mddev->pers->owner);
245f46c2
N		 3440 mddev->pers = pers;
3441 mddev->private = priv;
3442 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3443 mddev->level = mddev->new_level;
3444 mddev->layout = mddev->new_layout;
664e7c41 3445 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3446 mddev->delta_disks = 0;
fee68723 3447 mddev->degraded = 0;
9af204cf
TM		 3448 if (mddev->pers->sync_request == NULL) {
3449 /* this is now an array without redundancy, so
3450 * it must always be in_sync
3451 */
3452 mddev->in_sync = 1;
3453 del_timer_sync(&mddev->safemode_timer);
3454 }
245f46c2
N		 3455 pers->run(mddev);
3456 mddev_resume(mddev);
3457 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3458 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3459 md_wakeup_thread(mddev->thread);
5cac7861 3460 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3461 md_new_event(mddev);
d9d166c2
N		 3462 return rv;
3463}
3464
3465static struct md_sysfs_entry md_level =
80ca3a44 3466__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3467
d4dbd025
N		 3468
3469static ssize_t
fd01b88c 3470layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3471{
3472 /* just a number, not meaningful for all levels */
08a02ecd
N		 3473 if (mddev->reshape_position != MaxSector &&
3474 mddev->layout != mddev->new_layout)
3475 return sprintf(page, "%d (%d)\n",
3476 mddev->new_layout, mddev->layout);
d4dbd025
N		 3477 return sprintf(page, "%d\n", mddev->layout);
3478}
3479
3480static ssize_t
fd01b88c 3481layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3482{
3483 char *e;
3484 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3485
3486 if (!*buf || (*e && *e != '\n'))
3487 return -EINVAL;
3488
b3546035
N		 3489 if (mddev->pers) {
3490 int err;
50ac168a 3491 if (mddev->pers->check_reshape == NULL)
b3546035 3492 return -EBUSY;
597a711b 3493 mddev->new_layout = n;
50ac168a 3494 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3495 if (err) {
3496 mddev->new_layout = mddev->layout;
b3546035 3497 return err;
597a711b 3498 }
b3546035 3499 } else {
08a02ecd 3500 mddev->new_layout = n;
b3546035
N		 3501 if (mddev->reshape_position == MaxSector)
3502 mddev->layout = n;
3503 }
d4dbd025
N		 3504 return len;
3505}
3506static struct md_sysfs_entry md_layout =
80ca3a44 3507__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3508
3509
eae1701f 3510static ssize_t
fd01b88c 3511raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3512{
bb636547
N		 3513 if (mddev->raid_disks == 0)
3514 return 0;
08a02ecd
N		 3515 if (mddev->reshape_position != MaxSector &&
3516 mddev->delta_disks != 0)
3517 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3518 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3519 return sprintf(page, "%d\n", mddev->raid_disks);
3520}
3521
fd01b88c 3522static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3523
3524static ssize_t
fd01b88c 3525raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3526{
da943b99
N		 3527 char *e;
3528 int rv = 0;
3529 unsigned long n = simple_strtoul(buf, &e, 10);
3530
3531 if (!*buf || (*e && *e != '\n'))
3532 return -EINVAL;
3533
3534 if (mddev->pers)
3535 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3536 else if (mddev->reshape_position != MaxSector) {
3537 int olddisks = mddev->raid_disks - mddev->delta_disks;
3538 mddev->delta_disks = n - olddisks;
3539 mddev->raid_disks = n;
3540 } else
da943b99
N		 3541 mddev->raid_disks = n;
3542 return rv ? rv : len;
3543}
3544static struct md_sysfs_entry md_raid_disks =
80ca3a44 3545__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3546
3b34380a 3547static ssize_t
fd01b88c 3548chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3549{
08a02ecd 3550 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3551 mddev->chunk_sectors != mddev->new_chunk_sectors)
3552 return sprintf(page, "%d (%d)\n",
3553 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3554 mddev->chunk_sectors << 9);
3555 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3556}
3557
3558static ssize_t
fd01b88c 3559chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3560{
3b34380a
N		 3561 char *e;
3562 unsigned long n = simple_strtoul(buf, &e, 10);
3563
3b34380a
N		 3564 if (!*buf || (*e && *e != '\n'))
3565 return -EINVAL;
3566
b3546035
N		 3567 if (mddev->pers) {
3568 int err;
50ac168a 3569 if (mddev->pers->check_reshape == NULL)
b3546035 3570 return -EBUSY;
597a711b 3571 mddev->new_chunk_sectors = n >> 9;
50ac168a 3572 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3573 if (err) {
3574 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3575 return err;
597a711b 3576 }
b3546035 3577 } else {
664e7c41 3578 mddev->new_chunk_sectors = n >> 9;
b3546035 3579 if (mddev->reshape_position == MaxSector)
9d8f0363 3580 mddev->chunk_sectors = n >> 9;
b3546035 3581 }
3b34380a
N		 3582 return len;
3583}
3584static struct md_sysfs_entry md_chunk_size =
80ca3a44 3585__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3586
a94213b1 3587static ssize_t
fd01b88c 3588resync_start_show(struct mddev *mddev, char *page)
a94213b1 3589{
d1a7c503
N		 3590 if (mddev->recovery_cp == MaxSector)
3591 return sprintf(page, "none\n");
a94213b1
N		 3592 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3593}
3594
3595static ssize_t
fd01b88c 3596resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3597{
a94213b1
N		 3598 char *e;
3599 unsigned long long n = simple_strtoull(buf, &e, 10);
3600
b098636c 3601 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3602 return -EBUSY;
06e3c817
DW		 3603 if (cmd_match(buf, "none"))
3604 n = MaxSector;
3605 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3606 return -EINVAL;
3607
3608 mddev->recovery_cp = n;
3609 return len;
3610}
3611static struct md_sysfs_entry md_resync_start =
80ca3a44 3612__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3613
9e653b63
N		 3614/*
3615 * The array state can be:
3616 *
3617 * clear
3618 * No devices, no size, no level
3619 * Equivalent to STOP_ARRAY ioctl
3620 * inactive
3621 * May have some settings, but array is not active
3622 * all IO results in error
3623 * When written, doesn't tear down array, but just stops it
3624 * suspended (not supported yet)
3625 * All IO requests will block. The array can be reconfigured.
910d8cb3 3626 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3627 * readonly
3628 * no resync can happen. no superblocks get written.
3629 * write requests fail
3630 * read-auto
3631 * like readonly, but behaves like 'clean' on a write request.
3632 *
3633 * clean - no pending writes, but otherwise active.
3634 * When written to inactive array, starts without resync
3635 * If a write request arrives then
3636 * if metadata is known, mark 'dirty' and switch to 'active'.
3637 * if not known, block and switch to write-pending
3638 * If written to an active array that has pending writes, then fails.
3639 * active
3640 * fully active: IO and resync can be happening.
3641 * When written to inactive array, starts with resync
3642 *
3643 * write-pending
3644 * clean, but writes are blocked waiting for 'active' to be written.
3645 *
3646 * active-idle
3647 * like active, but no writes have been seen for a while (100msec).
3648 *
3649 */
3650enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3651 write_pending, active_idle, bad_word};
05381954 3652static char *array_states[] = {
9e653b63
N		 3653 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3654 "write-pending", "active-idle", NULL };
3655
3656static int match_word(const char *word, char **list)
3657{
3658 int n;
3659 for (n=0; list[n]; n++)
3660 if (cmd_match(word, list[n]))
3661 break;
3662 return n;
3663}
3664
3665static ssize_t
fd01b88c 3666array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3667{
3668 enum array_state st = inactive;
3669
3670 if (mddev->pers)
3671 switch(mddev->ro) {
3672 case 1:
3673 st = readonly;
3674 break;
3675 case 2:
3676 st = read_auto;
3677 break;
3678 case 0:
3679 if (mddev->in_sync)
3680 st = clean;
070dc6dd 3681 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3682 st = write_pending;
9e653b63
N		 3683 else if (mddev->safemode)
3684 st = active_idle;
3685 else
3686 st = active;
3687 }
3688 else {
3689 if (list_empty(&mddev->disks) &&
3690 mddev->raid_disks == 0 &&
58c0fed4 3691 mddev->dev_sectors == 0)
9e653b63
N		 3692 st = clear;
3693 else
3694 st = inactive;
3695 }
3696 return sprintf(page, "%s\n", array_states[st]);
3697}
3698
fd01b88c
N		 3699static int do_md_stop(struct mddev * mddev, int ro, int is_open);
3700static int md_set_readonly(struct mddev * mddev, int is_open);
3701static int do_md_run(struct mddev * mddev);
3702static int restart_array(struct mddev *mddev);
9e653b63
N		 3703
3704static ssize_t
fd01b88c 3705array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N		 3706{
3707 int err = -EINVAL;
3708 enum array_state st = match_word(buf, array_states);
3709 switch(st) {
3710 case bad_word:
3711 break;
3712 case clear:
3713 /* stopping an active array */
f2ea68cf 3714 if (atomic_read(&mddev->openers) > 0)
e691063a 3715 return -EBUSY;
df5b20cf 3716 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3717 break;
3718 case inactive:
3719 /* stopping an active array */
3720 if (mddev->pers) {
f2ea68cf 3721 if (atomic_read(&mddev->openers) > 0)
9e653b63 3722 return -EBUSY;
df5b20cf 3723 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3724 } else
3725 err = 0; /* already inactive */
9e653b63
N		 3726 break;
3727 case suspended:
3728 break; /* not supported yet */
3729 case readonly:
3730 if (mddev->pers)
a4bd82d0 3731 err = md_set_readonly(mddev, 0);
9e653b63
N		 3732 else {
3733 mddev->ro = 1;
648b629e 3734 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3735 err = do_md_run(mddev);
3736 }
3737 break;
3738 case read_auto:
9e653b63 3739 if (mddev->pers) {
80268ee9 3740 if (mddev->ro == 0)
a4bd82d0 3741 err = md_set_readonly(mddev, 0);
80268ee9 3742 else if (mddev->ro == 1)
648b629e
N		 3743 err = restart_array(mddev);
3744 if (err == 0) {
3745 mddev->ro = 2;
3746 set_disk_ro(mddev->gendisk, 0);
3747 }
9e653b63
N		 3748 } else {
3749 mddev->ro = 2;
3750 err = do_md_run(mddev);
3751 }
3752 break;
3753 case clean:
3754 if (mddev->pers) {
3755 restart_array(mddev);
3756 spin_lock_irq(&mddev->write_lock);
3757 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3758 if (mddev->in_sync == 0) {
3759 mddev->in_sync = 1;
31a59e34
N		 3760 if (mddev->safemode == 1)
3761 mddev->safemode = 0;
070dc6dd 3762 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3763 }
3764 err = 0;
3765 } else
3766 err = -EBUSY;
9e653b63 3767 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3768 } else
3769 err = -EINVAL;
9e653b63
N		 3770 break;
3771 case active:
3772 if (mddev->pers) {
3773 restart_array(mddev);
070dc6dd 3774 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3775 wake_up(&mddev->sb_wait);
3776 err = 0;
3777 } else {
3778 mddev->ro = 0;
648b629e 3779 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3780 err = do_md_run(mddev);
3781 }
3782 break;
3783 case write_pending:
3784 case active_idle:
3785 /* these cannot be set */
3786 break;
3787 }
3788 if (err)
3789 return err;
0fd62b86 3790 else {
1d23f178
N		 3791 if (mddev->hold_active == UNTIL_IOCTL)
3792 mddev->hold_active = 0;
00bcb4ac 3793 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3794 return len;
0fd62b86 3795 }
9e653b63 3796}
80ca3a44
N		 3797static struct md_sysfs_entry md_array_state =
3798__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3799
1e50915f 3800static ssize_t
fd01b88c 3801max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3802 return sprintf(page, "%d\n",
3803 atomic_read(&mddev->max_corr_read_errors));
3804}
3805
3806static ssize_t
fd01b88c 3807max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3808{
3809 char *e;
3810 unsigned long n = simple_strtoul(buf, &e, 10);
3811
3812 if (*buf && (*e == 0 || *e == '\n')) {
3813 atomic_set(&mddev->max_corr_read_errors, n);
3814 return len;
3815 }
3816 return -EINVAL;
3817}
3818
3819static struct md_sysfs_entry max_corr_read_errors =
3820__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3821 max_corrected_read_errors_store);
3822
6d7ff738 3823static ssize_t
fd01b88c 3824null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3825{
3826 return -EINVAL;
3827}
3828
3829static ssize_t
fd01b88c 3830new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3831{
3832 /* buf must be %d:%d\n? giving major and minor numbers */
3833 /* The new device is added to the array.
3834 * If the array has a persistent superblock, we read the
3835 * superblock to initialise info and check validity.
3836 * Otherwise, only checking done is that in bind_rdev_to_array,
3837 * which mainly checks size.
3838 */
3839 char *e;
3840 int major = simple_strtoul(buf, &e, 10);
3841 int minor;
3842 dev_t dev;
3cb03002 3843 struct md_rdev *rdev;
6d7ff738
N		 3844 int err;
3845
3846 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3847 return -EINVAL;
3848 minor = simple_strtoul(e+1, &e, 10);
3849 if (*e && *e != '\n')
3850 return -EINVAL;
3851 dev = MKDEV(major, minor);
3852 if (major != MAJOR(dev) ||
3853 minor != MINOR(dev))
3854 return -EOVERFLOW;
3855
3856
3857 if (mddev->persistent) {
3858 rdev = md_import_device(dev, mddev->major_version,
3859 mddev->minor_version);
3860 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 3861 struct md_rdev *rdev0
3862 = list_entry(mddev->disks.next,
3863 struct md_rdev, same_set);
6d7ff738
N		 3864 err = super_types[mddev->major_version]
3865 .load_super(rdev, rdev0, mddev->minor_version);
3866 if (err < 0)
3867 goto out;
3868 }
c5d79adb
N		 3869 } else if (mddev->external)
3870 rdev = md_import_device(dev, -2, -1);
3871 else
6d7ff738
N		 3872 rdev = md_import_device(dev, -1, -1);
3873
3874 if (IS_ERR(rdev))
3875 return PTR_ERR(rdev);
3876 err = bind_rdev_to_array(rdev, mddev);
3877 out:
3878 if (err)
3879 export_rdev(rdev);
3880 return err ? err : len;
3881}
3882
3883static struct md_sysfs_entry md_new_device =
80ca3a44 3884__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3885
9b1d1dac 3886static ssize_t
fd01b88c 3887bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 3888{
3889 char *end;
3890 unsigned long chunk, end_chunk;
3891
3892 if (!mddev->bitmap)
3893 goto out;
3894 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3895 while (*buf) {
3896 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3897 if (buf == end) break;
3898 if (*end == '-') { /* range */
3899 buf = end + 1;
3900 end_chunk = simple_strtoul(buf, &end, 0);
3901 if (buf == end) break;
3902 }
3903 if (*end && !isspace(*end)) break;
3904 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3905 buf = skip_spaces(end);
9b1d1dac
PC		 3906 }
3907 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3908out:
3909 return len;
3910}
3911
3912static struct md_sysfs_entry md_bitmap =
3913__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3914
a35b0d69 3915static ssize_t
fd01b88c 3916size_show(struct mddev *mddev, char *page)
a35b0d69 3917{
58c0fed4
AN		 3918 return sprintf(page, "%llu\n",
3919 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3920}
3921
fd01b88c 3922static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 3923
3924static ssize_t
fd01b88c 3925size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 3926{
3927 /* If array is inactive, we can reduce the component size, but
3928 * not increase it (except from 0).
3929 * If array is active, we can try an on-line resize
3930 */
b522adcd
DW		 3931 sector_t sectors;
3932 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3933
58c0fed4
AN		 3934 if (err < 0)
3935 return err;
a35b0d69 3936 if (mddev->pers) {
58c0fed4 3937 err = update_size(mddev, sectors);
850b2b42 3938 md_update_sb(mddev, 1);
a35b0d69 3939 } else {
58c0fed4
AN		 3940 if (mddev->dev_sectors == 0 ||
3941 mddev->dev_sectors > sectors)
3942 mddev->dev_sectors = sectors;
a35b0d69
N		 3943 else
3944 err = -ENOSPC;
3945 }
3946 return err ? err : len;
3947}
3948
3949static struct md_sysfs_entry md_size =
80ca3a44 3950__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3951
8bb93aac
N		 3952
3953/* Metdata version.
e691063a
N		 3954 * This is one of
3955 * 'none' for arrays with no metadata (good luck...)
3956 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 3957 * or N.M for internally known formats
3958 */
3959static ssize_t
fd01b88c 3960metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 3961{
3962 if (mddev->persistent)
3963 return sprintf(page, "%d.%d\n",
3964 mddev->major_version, mddev->minor_version);
e691063a
N		 3965 else if (mddev->external)
3966 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 3967 else
3968 return sprintf(page, "none\n");
3969}
3970
3971static ssize_t
fd01b88c 3972metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 3973{
3974 int major, minor;
3975 char *e;
ea43ddd8
N		 3976 /* Changing the details of 'external' metadata is
3977 * always permitted. Otherwise there must be
3978 * no devices attached to the array.
3979 */
3980 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3981 ;
3982 else if (!list_empty(&mddev->disks))
8bb93aac
N		 3983 return -EBUSY;
3984
3985 if (cmd_match(buf, "none")) {
3986 mddev->persistent = 0;
e691063a
N		 3987 mddev->external = 0;
3988 mddev->major_version = 0;
3989 mddev->minor_version = 90;
3990 return len;
3991 }
3992 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 3993 size_t namelen = len-9;
e691063a
N		 3994 if (namelen >= sizeof(mddev->metadata_type))
3995 namelen = sizeof(mddev->metadata_type)-1;
3996 strncpy(mddev->metadata_type, buf+9, namelen);
3997 mddev->metadata_type[namelen] = 0;
3998 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3999 mddev->metadata_type[--namelen] = 0;
4000 mddev->persistent = 0;
4001 mddev->external = 1;
8bb93aac
N		 4002 mddev->major_version = 0;
4003 mddev->minor_version = 90;
4004 return len;
4005 }
4006 major = simple_strtoul(buf, &e, 10);
4007 if (e==buf || *e != '.')
4008 return -EINVAL;
4009 buf = e+1;
4010 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4011 if (e==buf || (*e && *e != '\n') )
8bb93aac 4012 return -EINVAL;
50511da3 4013 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 4014 return -ENOENT;
4015 mddev->major_version = major;
4016 mddev->minor_version = minor;
4017 mddev->persistent = 1;
e691063a 4018 mddev->external = 0;
8bb93aac
N		 4019 return len;
4020}
4021
4022static struct md_sysfs_entry md_metadata =
80ca3a44 4023__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4024
24dd469d 4025static ssize_t
fd01b88c 4026action_show(struct mddev *mddev, char *page)
24dd469d 4027{
7eec314d 4028 char *type = "idle";
b6a9ce68
N		 4029 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4030 type = "frozen";
4031 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4032 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 4033 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4034 type = "reshape";
4035 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 4036 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4037 type = "resync";
4038 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4039 type = "check";
4040 else
4041 type = "repair";
72a23c21 4042 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 4043 type = "recover";
4044 }
4045 return sprintf(page, "%s\n", type);
4046}
4047
fd01b88c 4048static void reap_sync_thread(struct mddev *mddev);
7ebc0be7 4049
24dd469d 4050static ssize_t
fd01b88c 4051action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4052{
7eec314d
N		 4053 if (!mddev->pers || !mddev->pers->sync_request)
4054 return -EINVAL;
4055
b6a9ce68
N		 4056 if (cmd_match(page, "frozen"))
4057 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4058 else
4059 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4060
4061 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 4062 if (mddev->sync_thread) {
4063 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4064 reap_sync_thread(mddev);
7eec314d 4065 }
03c902e1
N		 4066 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4067 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4068 return -EBUSY;
72a23c21
NB		 4069 else if (cmd_match(page, "resync"))
4070 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4071 else if (cmd_match(page, "recover")) {
4072 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4073 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4074 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4075 int err;
4076 if (mddev->pers->start_reshape == NULL)
4077 return -EINVAL;
4078 err = mddev->pers->start_reshape(mddev);
4079 if (err)
4080 return err;
a99ac971 4081 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4082 } else {
bce74dac 4083 if (cmd_match(page, "check"))
7eec314d 4084 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4085 else if (!cmd_match(page, "repair"))
7eec314d
N		 4086 return -EINVAL;
4087 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4088 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4089 }
03c902e1 4090 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4091 md_wakeup_thread(mddev->thread);
00bcb4ac 4092 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4093 return len;
4094}
4095
9d88883e 4096static ssize_t
fd01b88c 4097mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4098{
4099 return sprintf(page, "%llu\n",
4100 (unsigned long long) mddev->resync_mismatches);
4101}
4102
80ca3a44
N		 4103static struct md_sysfs_entry md_scan_mode =
4104__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 4105
96de1e66 4106
80ca3a44 4107static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4108
88202a0c 4109static ssize_t
fd01b88c 4110sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4111{
4112 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4113 mddev->sync_speed_min ? "local": "system");
4114}
4115
4116static ssize_t
fd01b88c 4117sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4118{
4119 int min;
4120 char *e;
4121 if (strncmp(buf, "system", 6)==0) {
4122 mddev->sync_speed_min = 0;
4123 return len;
4124 }
4125 min = simple_strtoul(buf, &e, 10);
4126 if (buf == e || (*e && *e != '\n') || min <= 0)
4127 return -EINVAL;
4128 mddev->sync_speed_min = min;
4129 return len;
4130}
4131
4132static struct md_sysfs_entry md_sync_min =
4133__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4134
4135static ssize_t
fd01b88c 4136sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4137{
4138 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4139 mddev->sync_speed_max ? "local": "system");
4140}
4141
4142static ssize_t
fd01b88c 4143sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4144{
4145 int max;
4146 char *e;
4147 if (strncmp(buf, "system", 6)==0) {
4148 mddev->sync_speed_max = 0;
4149 return len;
4150 }
4151 max = simple_strtoul(buf, &e, 10);
4152 if (buf == e || (*e && *e != '\n') || max <= 0)
4153 return -EINVAL;
4154 mddev->sync_speed_max = max;
4155 return len;
4156}
4157
4158static struct md_sysfs_entry md_sync_max =
4159__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4160
d7f3d291 4161static ssize_t
fd01b88c 4162degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4163{
4164 return sprintf(page, "%d\n", mddev->degraded);
4165}
4166static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4167
90b08710 4168static ssize_t
fd01b88c 4169sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4170{
4171 return sprintf(page, "%d\n", mddev->parallel_resync);
4172}
4173
4174static ssize_t
fd01b88c 4175sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4176{
4177 long n;
4178
4179 if (strict_strtol(buf, 10, &n))
4180 return -EINVAL;
4181
4182 if (n != 0 && n != 1)
4183 return -EINVAL;
4184
4185 mddev->parallel_resync = n;
4186
4187 if (mddev->sync_thread)
4188 wake_up(&resync_wait);
4189
4190 return len;
4191}
4192
4193/* force parallel resync, even with shared block devices */
4194static struct md_sysfs_entry md_sync_force_parallel =
4195__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4196 sync_force_parallel_show, sync_force_parallel_store);
4197
88202a0c 4198static ssize_t
fd01b88c 4199sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4200{
4201 unsigned long resync, dt, db;
d1a7c503
N		 4202 if (mddev->curr_resync == 0)
4203 return sprintf(page, "none\n");
9687a60c
AN		 4204 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4205 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4206 if (!dt) dt++;
9687a60c
AN		 4207 db = resync - mddev->resync_mark_cnt;
4208 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4209}
4210
80ca3a44 4211static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4212
4213static ssize_t
fd01b88c 4214sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4215{
13ae864b 4216 unsigned long long max_sectors, resync;
88202a0c 4217
acb180b0
N		 4218 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4219 return sprintf(page, "none\n");
4220
88202a0c 4221 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 4222 max_sectors = mddev->resync_max_sectors;
88202a0c 4223 else
58c0fed4 4224 max_sectors = mddev->dev_sectors;
88202a0c 4225
acb180b0 4226 resync = mddev->curr_resync_completed;
13ae864b 4227 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4228}
4229
80ca3a44 4230static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4231
5e96ee65 4232static ssize_t
fd01b88c 4233min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4234{
4235 return sprintf(page, "%llu\n",
4236 (unsigned long long)mddev->resync_min);
4237}
4238static ssize_t
fd01b88c 4239min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4240{
4241 unsigned long long min;
4242 if (strict_strtoull(buf, 10, &min))
4243 return -EINVAL;
4244 if (min > mddev->resync_max)
4245 return -EINVAL;
4246 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4247 return -EBUSY;
4248
4249 /* Must be a multiple of chunk_size */
9d8f0363 4250 if (mddev->chunk_sectors) {
2ac06c33 4251 sector_t temp = min;
9d8f0363 4252 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 4253 return -EINVAL;
4254 }
4255 mddev->resync_min = min;
4256
4257 return len;
4258}
4259
4260static struct md_sysfs_entry md_min_sync =
4261__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4262
c6207277 4263static ssize_t
fd01b88c 4264max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4265{
4266 if (mddev->resync_max == MaxSector)
4267 return sprintf(page, "max\n");
4268 else
4269 return sprintf(page, "%llu\n",
4270 (unsigned long long)mddev->resync_max);
4271}
4272static ssize_t
fd01b88c 4273max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N		 4274{
4275 if (strncmp(buf, "max", 3) == 0)
4276 mddev->resync_max = MaxSector;
4277 else {
5e96ee65
NB		 4278 unsigned long long max;
4279 if (strict_strtoull(buf, 10, &max))
4280 return -EINVAL;
4281 if (max < mddev->resync_min)
c6207277
N		 4282 return -EINVAL;
4283 if (max < mddev->resync_max &&
4d484a4a 4284 mddev->ro == 0 &&
c6207277
N		 4285 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4286 return -EBUSY;
4287
4288 /* Must be a multiple of chunk_size */
9d8f0363 4289 if (mddev->chunk_sectors) {
2ac06c33 4290 sector_t temp = max;
9d8f0363 4291 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4292 return -EINVAL;
4293 }
4294 mddev->resync_max = max;
4295 }
4296 wake_up(&mddev->recovery_wait);
4297 return len;
4298}
4299
4300static struct md_sysfs_entry md_max_sync =
4301__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4302
e464eafd 4303static ssize_t
fd01b88c 4304suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4305{
4306 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4307}
4308
4309static ssize_t
fd01b88c 4310suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4311{
4312 char *e;
4313 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4314 unsigned long long old = mddev->suspend_lo;
e464eafd 4315
b8d966ef
N		 4316 if (mddev->pers == NULL ||
4317 mddev->pers->quiesce == NULL)
e464eafd
N		 4318 return -EINVAL;
4319 if (buf == e || (*e && *e != '\n'))
4320 return -EINVAL;
23ddff37
N		 4321
4322 mddev->suspend_lo = new;
4323 if (new >= old)
4324 /* Shrinking suspended region */
e464eafd 4325 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4326 else {
4327 /* Expanding suspended region - need to wait */
4328 mddev->pers->quiesce(mddev, 1);
4329 mddev->pers->quiesce(mddev, 0);
4330 }
4331 return len;
e464eafd
N		 4332}
4333static struct md_sysfs_entry md_suspend_lo =
4334__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4335
4336
4337static ssize_t
fd01b88c 4338suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4339{
4340 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4341}
4342
4343static ssize_t
fd01b88c 4344suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4345{
4346 char *e;
4347 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4348 unsigned long long old = mddev->suspend_hi;
e464eafd 4349
b8d966ef
N		 4350 if (mddev->pers == NULL ||
4351 mddev->pers->quiesce == NULL)
e464eafd
N		 4352 return -EINVAL;
4353 if (buf == e || (*e && *e != '\n'))
4354 return -EINVAL;
23ddff37
N		 4355
4356 mddev->suspend_hi = new;
4357 if (new <= old)
4358 /* Shrinking suspended region */
4359 mddev->pers->quiesce(mddev, 2);
4360 else {
4361 /* Expanding suspended region - need to wait */
e464eafd
N		 4362 mddev->pers->quiesce(mddev, 1);
4363 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4364 }
4365 return len;
e464eafd
N		 4366}
4367static struct md_sysfs_entry md_suspend_hi =
4368__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4369
08a02ecd 4370static ssize_t
fd01b88c 4371reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4372{
4373 if (mddev->reshape_position != MaxSector)
4374 return sprintf(page, "%llu\n",
4375 (unsigned long long)mddev->reshape_position);
4376 strcpy(page, "none\n");
4377 return 5;
4378}
4379
4380static ssize_t
fd01b88c 4381reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd
N		 4382{
4383 char *e;
4384 unsigned long long new = simple_strtoull(buf, &e, 10);
4385 if (mddev->pers)
4386 return -EBUSY;
4387 if (buf == e || (*e && *e != '\n'))
4388 return -EINVAL;
4389 mddev->reshape_position = new;
4390 mddev->delta_disks = 0;
4391 mddev->new_level = mddev->level;
4392 mddev->new_layout = mddev->layout;
664e7c41 4393 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4394 return len;
4395}
4396
4397static struct md_sysfs_entry md_reshape_position =
4398__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4399 reshape_position_store);
4400
b522adcd 4401static ssize_t
fd01b88c 4402array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4403{
4404 if (mddev->external_size)
4405 return sprintf(page, "%llu\n",
4406 (unsigned long long)mddev->array_sectors/2);
4407 else
4408 return sprintf(page, "default\n");
4409}
4410
4411static ssize_t
fd01b88c 4412array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4413{
4414 sector_t sectors;
4415
4416 if (strncmp(buf, "default", 7) == 0) {
4417 if (mddev->pers)
4418 sectors = mddev->pers->size(mddev, 0, 0);
4419 else
4420 sectors = mddev->array_sectors;
4421
4422 mddev->external_size = 0;
4423 } else {
4424 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4425 return -EINVAL;
4426 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4427 return -E2BIG;
b522adcd
DW		 4428
4429 mddev->external_size = 1;
4430 }
4431
4432 mddev->array_sectors = sectors;
cbe6ef1d
N		 4433 if (mddev->pers) {
4434 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4435 revalidate_disk(mddev->gendisk);
cbe6ef1d 4436 }
b522adcd
DW		 4437 return len;
4438}
4439
4440static struct md_sysfs_entry md_array_size =
4441__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4442 array_size_store);
e464eafd 4443
eae1701f
N		 4444static struct attribute *md_default_attrs[] = {
4445 &md_level.attr,
d4dbd025 4446 &md_layout.attr,
eae1701f 4447 &md_raid_disks.attr,
3b34380a 4448 &md_chunk_size.attr,
a35b0d69 4449 &md_size.attr,
a94213b1 4450 &md_resync_start.attr,
8bb93aac 4451 &md_metadata.attr,
6d7ff738 4452 &md_new_device.attr,
16f17b39 4453 &md_safe_delay.attr,
9e653b63 4454 &md_array_state.attr,
08a02ecd 4455 &md_reshape_position.attr,
b522adcd 4456 &md_array_size.attr,
1e50915f 4457 &max_corr_read_errors.attr,
411036fa
N		 4458 NULL,
4459};
4460
4461static struct attribute *md_redundancy_attrs[] = {
24dd469d 4462 &md_scan_mode.attr,
9d88883e 4463 &md_mismatches.attr,
88202a0c
N		 4464 &md_sync_min.attr,
4465 &md_sync_max.attr,
4466 &md_sync_speed.attr,
90b08710 4467 &md_sync_force_parallel.attr,
88202a0c 4468 &md_sync_completed.attr,
5e96ee65 4469 &md_min_sync.attr,
c6207277 4470 &md_max_sync.attr,
e464eafd
N		 4471 &md_suspend_lo.attr,
4472 &md_suspend_hi.attr,
9b1d1dac 4473 &md_bitmap.attr,
d7f3d291 4474 &md_degraded.attr,
eae1701f
N		 4475 NULL,
4476};
411036fa
N		 4477static struct attribute_group md_redundancy_group = {
4478 .name = NULL,
4479 .attrs = md_redundancy_attrs,
4480};
4481
eae1701f
N		 4482
4483static ssize_t
4484md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4485{
4486 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4487 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4488 ssize_t rv;
eae1701f
N		 4489
4490 if (!entry->show)
4491 return -EIO;
5dc5cf7d
IM		 4492 rv = mddev_lock(mddev);
4493 if (!rv) {
4494 rv = entry->show(mddev, page);
4495 mddev_unlock(mddev);
4496 }
96de1e66 4497 return rv;
eae1701f
N		 4498}
4499
4500static ssize_t
4501md_attr_store(struct kobject *kobj, struct attribute *attr,
4502 const char *page, size_t length)
4503{
4504 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4505 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4506 ssize_t rv;
eae1701f
N		 4507
4508 if (!entry->store)
4509 return -EIO;
67463acb
N		 4510 if (!capable(CAP_SYS_ADMIN))
4511 return -EACCES;
5dc5cf7d
IM		 4512 rv = mddev_lock(mddev);
4513 if (!rv) {
4514 rv = entry->store(mddev, page, length);
4515 mddev_unlock(mddev);
4516 }
96de1e66 4517 return rv;
eae1701f
N		 4518}
4519
4520static void md_free(struct kobject *ko)
4521{
fd01b88c 4522 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4523
4524 if (mddev->sysfs_state)
4525 sysfs_put(mddev->sysfs_state);
4526
4527 if (mddev->gendisk) {
4528 del_gendisk(mddev->gendisk);
4529 put_disk(mddev->gendisk);
4530 }
4531 if (mddev->queue)
4532 blk_cleanup_queue(mddev->queue);
4533
eae1701f
N		 4534 kfree(mddev);
4535}
4536
52cf25d0 4537static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4538 .show = md_attr_show,
4539 .store = md_attr_store,
4540};
4541static struct kobj_type md_ktype = {
4542 .release = md_free,
4543 .sysfs_ops = &md_sysfs_ops,
4544 .default_attrs = md_default_attrs,
4545};
4546
1da177e4
LT		 4547int mdp_major = 0;
4548
5fd3a17e
DW		 4549static void mddev_delayed_delete(struct work_struct *ws)
4550{
fd01b88c 4551 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4552
43a70507 4553 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4554 kobject_del(&mddev->kobj);
4555 kobject_put(&mddev->kobj);
4556}
4557
efeb53c0 4558static int md_alloc(dev_t dev, char *name)
1da177e4 4559{
48c9c27b 4560 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4561 struct mddev *mddev = mddev_find(dev);
1da177e4 4562 struct gendisk *disk;
efeb53c0
N		 4563 int partitioned;
4564 int shift;
4565 int unit;
3830c62f 4566 int error;
1da177e4
LT		 4567
4568 if (!mddev)
efeb53c0
N		 4569 return -ENODEV;
4570
4571 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4572 shift = partitioned ? MdpMinorShift : 0;
4573 unit = MINOR(mddev->unit) >> shift;
1da177e4 4574
e804ac78
TH		 4575 /* wait for any previous instance of this device to be
4576 * completely removed (mddev_delayed_delete).
d3374825 4577 */
e804ac78 4578 flush_workqueue(md_misc_wq);
d3374825 4579
48c9c27b 4580 mutex_lock(&disks_mutex);
0909dc44
N		 4581 error = -EEXIST;
4582 if (mddev->gendisk)
4583 goto abort;
efeb53c0
N		 4584
4585 if (name) {
4586 /* Need to ensure that 'name' is not a duplicate.
4587 */
fd01b88c 4588 struct mddev *mddev2;
efeb53c0
N		 4589 spin_lock(&all_mddevs_lock);
4590
4591 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4592 if (mddev2->gendisk &&
4593 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4594 spin_unlock(&all_mddevs_lock);
0909dc44 4595 goto abort;
efeb53c0
N		 4596 }
4597 spin_unlock(&all_mddevs_lock);
1da177e4 4598 }
8b765398 4599
0909dc44 4600 error = -ENOMEM;
8b765398 4601 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4602 if (!mddev->queue)
4603 goto abort;
409c57f3
N		 4604 mddev->queue->queuedata = mddev;
4605
409c57f3 4606 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4607
1da177e4
LT		 4608 disk = alloc_disk(1 << shift);
4609 if (!disk) {
8b765398
N		 4610 blk_cleanup_queue(mddev->queue);
4611 mddev->queue = NULL;
0909dc44 4612 goto abort;
1da177e4 4613 }
efeb53c0 4614 disk->major = MAJOR(mddev->unit);
1da177e4 4615 disk->first_minor = unit << shift;
efeb53c0
N		 4616 if (name)
4617 strcpy(disk->disk_name, name);
4618 else if (partitioned)
1da177e4 4619 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4620 else
1da177e4 4621 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4622 disk->fops = &md_fops;
4623 disk->private_data = mddev;
4624 disk->queue = mddev->queue;
b0140891 4625 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4626 /* Allow extended partitions. This makes the
d3374825 4627 * 'mdp' device redundant, but we can't really
92850bbd
N		 4628 * remove it now.
4629 */
4630 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4631 mddev->gendisk = disk;
b0140891
N		 4632 /* As soon as we call add_disk(), another thread could get
4633 * through to md_open, so make sure it doesn't get too far
4634 */
4635 mutex_lock(&mddev->open_mutex);
4636 add_disk(disk);
4637
ed9e1982
TH		 4638 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4639 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4640 if (error) {
4641 /* This isn't possible, but as kobject_init_and_add is marked
4642 * __must_check, we must do something with the result
4643 */
5e55e2f5
N		 4644 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4645 disk->disk_name);
0909dc44
N		 4646 error = 0;
4647 }
00bcb4ac
N		 4648 if (mddev->kobj.sd &&
4649 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4650 printk(KERN_DEBUG "pointless warning\n");
b0140891 4651 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4652 abort:
4653 mutex_unlock(&disks_mutex);
00bcb4ac 4654 if (!error && mddev->kobj.sd) {
3830c62f 4655 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4656 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4657 }
d3374825 4658 mddev_put(mddev);
0909dc44 4659 return error;
efeb53c0
N		 4660}
4661
4662static struct kobject *md_probe(dev_t dev, int *part, void *data)
4663{
4664 md_alloc(dev, NULL);
1da177e4
LT		 4665 return NULL;
4666}
4667
efeb53c0
N		 4668static int add_named_array(const char *val, struct kernel_param *kp)
4669{
4670 /* val must be "md_*" where * is not all digits.
4671 * We allocate an array with a large free minor number, and
4672 * set the name to val. val must not already be an active name.
4673 */
4674 int len = strlen(val);
4675 char buf[DISK_NAME_LEN];
4676
4677 while (len && val[len-1] == '\n')
4678 len--;
4679 if (len >= DISK_NAME_LEN)
4680 return -E2BIG;
4681 strlcpy(buf, val, len+1);
4682 if (strncmp(buf, "md_", 3) != 0)
4683 return -EINVAL;
4684 return md_alloc(0, buf);
4685}
4686
1da177e4
LT		 4687static void md_safemode_timeout(unsigned long data)
4688{
fd01b88c 4689 struct mddev *mddev = (struct mddev *) data;
1da177e4 4690
0fd62b86
NB		 4691 if (!atomic_read(&mddev->writes_pending)) {
4692 mddev->safemode = 1;
4693 if (mddev->external)
00bcb4ac 4694 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4695 }
1da177e4
LT		 4696 md_wakeup_thread(mddev->thread);
4697}
4698
6ff8d8ec 4699static int start_dirty_degraded;
1da177e4 4700
fd01b88c 4701int md_run(struct mddev *mddev)
1da177e4 4702{
2604b703 4703 int err;
3cb03002 4704 struct md_rdev *rdev;
84fc4b56 4705 struct md_personality *pers;
1da177e4 4706
a757e64c
N		 4707 if (list_empty(&mddev->disks))
4708 /* cannot run an array with no devices.. */
1da177e4 4709 return -EINVAL;
1da177e4
LT		 4710
4711 if (mddev->pers)
4712 return -EBUSY;
bb4f1e9d
N		 4713 /* Cannot run until previous stop completes properly */
4714 if (mddev->sysfs_active)
4715 return -EBUSY;
b6eb127d 4716
1da177e4
LT		 4717 /*
4718 * Analyze all RAID superblock(s)
4719 */
1ec4a939
N		 4720 if (!mddev->raid_disks) {
4721 if (!mddev->persistent)
4722 return -EINVAL;
a757e64c 4723 analyze_sbs(mddev);
1ec4a939 4724 }
1da177e4 4725
d9d166c2
N		 4726 if (mddev->level != LEVEL_NONE)
4727 request_module("md-level-%d", mddev->level);
4728 else if (mddev->clevel[0])
4729 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4730
4731 /*
4732 * Drop all container device buffers, from now on
4733 * the only valid external interface is through the md
4734 * device.
1da177e4 4735 */
159ec1fc 4736 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4737 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4738 continue;
4739 sync_blockdev(rdev->bdev);
f98393a6 4740 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4741
4742 /* perform some consistency tests on the device.
4743 * We don't want the data to overlap the metadata,
58c0fed4 4744 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4745 */
a6ff7e08
JB		 4746 if (rdev->meta_bdev) {
4747 /* Nothing to check */;
4748 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4749 if (mddev->dev_sectors &&
4750 rdev->data_offset + mddev->dev_sectors
0f420358 4751 > rdev->sb_start) {
f0d76d70
N		 4752 printk("md: %s: data overlaps metadata\n",
4753 mdname(mddev));
4754 return -EINVAL;
4755 }
4756 } else {
0f420358 4757 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4758 > rdev->data_offset) {
4759 printk("md: %s: metadata overlaps data\n",
4760 mdname(mddev));
4761 return -EINVAL;
4762 }
4763 }
00bcb4ac 4764 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4765 }
4766
a167f663 4767 if (mddev->bio_set == NULL)
a519b26d 4768 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
fd01b88c 4769 sizeof(struct mddev *));
a167f663 4770
1da177e4 4771 spin_lock(&pers_lock);
d9d166c2 4772 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4773 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4774 spin_unlock(&pers_lock);
d9d166c2
N		 4775 if (mddev->level != LEVEL_NONE)
4776 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4777 mddev->level);
4778 else
4779 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4780 mddev->clevel);
1da177e4
LT		 4781 return -EINVAL;
4782 }
2604b703 4783 mddev->pers = pers;
1da177e4 4784 spin_unlock(&pers_lock);
34817e8c
N		 4785 if (mddev->level != pers->level) {
4786 mddev->level = pers->level;
4787 mddev->new_level = pers->level;
4788 }
d9d166c2 4789 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4790
f6705578 4791 if (mddev->reshape_position != MaxSector &&
63c70c4f 4792 pers->start_reshape == NULL) {
f6705578
N		 4793 /* This personality cannot handle reshaping... */
4794 mddev->pers = NULL;
4795 module_put(pers->owner);
4796 return -EINVAL;
4797 }
4798
7dd5e7c3
N		 4799 if (pers->sync_request) {
4800 /* Warn if this is a potentially silly
4801 * configuration.
4802 */
4803 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 4804 struct md_rdev *rdev2;
7dd5e7c3 4805 int warned = 0;
159ec1fc
CR		 4806
4807 list_for_each_entry(rdev, &mddev->disks, same_set)
4808 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4809 if (rdev < rdev2 &&
4810 rdev->bdev->bd_contains ==
4811 rdev2->bdev->bd_contains) {
4812 printk(KERN_WARNING
4813 "%s: WARNING: %s appears to be"
4814 " on the same physical disk as"
4815 " %s.\n",
4816 mdname(mddev),
4817 bdevname(rdev->bdev,b),
4818 bdevname(rdev2->bdev,b2));
4819 warned = 1;
4820 }
4821 }
159ec1fc 4822
7dd5e7c3
N		 4823 if (warned)
4824 printk(KERN_WARNING
4825 "True protection against single-disk"
4826 " failure might be compromised.\n");
4827 }
4828
657390d2 4829 mddev->recovery = 0;
58c0fed4
AN		 4830 /* may be over-ridden by personality */
4831 mddev->resync_max_sectors = mddev->dev_sectors;
4832
6ff8d8ec 4833 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4834
0f9552b5 4835 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4836 mddev->ro = 2; /* read-only, but switch on first write */
4837
b15c2e57 4838 err = mddev->pers->run(mddev);
13e53df3
AN		 4839 if (err)
4840 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4841 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4842 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4843 " but 'external_size' not in effect?\n", __func__);
4844 printk(KERN_ERR
4845 "md: invalid array_size %llu > default size %llu\n",
4846 (unsigned long long)mddev->array_sectors / 2,
4847 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4848 err = -EINVAL;
4849 mddev->pers->stop(mddev);
4850 }
4851 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4852 err = bitmap_create(mddev);
4853 if (err) {
4854 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4855 mdname(mddev), err);
4856 mddev->pers->stop(mddev);
4857 }
4858 }
1da177e4 4859 if (err) {
1da177e4
LT		 4860 module_put(mddev->pers->owner);
4861 mddev->pers = NULL;
32a7627c
N		 4862 bitmap_destroy(mddev);
4863 return err;
1da177e4 4864 }
5e55e2f5 4865 if (mddev->pers->sync_request) {
00bcb4ac
N		 4866 if (mddev->kobj.sd &&
4867 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4868 printk(KERN_WARNING
4869 "md: cannot register extra attributes for %s\n",
4870 mdname(mddev));
00bcb4ac 4871 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4872 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4873 mddev->ro = 0;
4874
1da177e4 4875 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4876 atomic_set(&mddev->max_corr_read_errors,
4877 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4878 mddev->safemode = 0;
4879 mddev->safemode_timer.function = md_safemode_timeout;
4880 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4881 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4882 mddev->in_sync = 1;
0ca69886
N		 4883 smp_wmb();
4884 mddev->ready = 1;
159ec1fc 4885 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 4886 if (rdev->raid_disk >= 0)
4887 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 4888 /* failure here is OK */;
1da177e4
LT		 4889
4890 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4891
850b2b42
N		 4892 if (mddev->flags)
4893 md_update_sb(mddev, 0);
1da177e4 4894
d7603b7e 4895 md_new_event(mddev);
00bcb4ac
N		 4896 sysfs_notify_dirent_safe(mddev->sysfs_state);
4897 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4898 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4899 return 0;
4900}
390ee602 4901EXPORT_SYMBOL_GPL(md_run);
1da177e4 4902
fd01b88c 4903static int do_md_run(struct mddev *mddev)
fe60b014
N		 4904{
4905 int err;
4906
4907 err = md_run(mddev);
4908 if (err)
4909 goto out;
69e51b44
N		 4910 err = bitmap_load(mddev);
4911 if (err) {
4912 bitmap_destroy(mddev);
4913 goto out;
4914 }
0fd018af
JB		 4915
4916 md_wakeup_thread(mddev->thread);
4917 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4918
fe60b014
N		 4919 set_capacity(mddev->gendisk, mddev->array_sectors);
4920 revalidate_disk(mddev->gendisk);
f0b4f7e2 4921 mddev->changed = 1;
fe60b014
N		 4922 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4923out:
4924 return err;
4925}
4926
fd01b88c 4927static int restart_array(struct mddev *mddev)
1da177e4
LT		 4928{
4929 struct gendisk *disk = mddev->gendisk;
1da177e4 4930
80fab1d7 4931 /* Complain if it has no devices */
1da177e4 4932 if (list_empty(&mddev->disks))
80fab1d7
AN		 4933 return -ENXIO;
4934 if (!mddev->pers)
4935 return -EINVAL;
4936 if (!mddev->ro)
4937 return -EBUSY;
4938 mddev->safemode = 0;
4939 mddev->ro = 0;
4940 set_disk_ro(disk, 0);
4941 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4942 mdname(mddev));
4943 /* Kick recovery or resync if necessary */
4944 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4945 md_wakeup_thread(mddev->thread);
4946 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 4947 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 4948 return 0;
1da177e4
LT		 4949}
4950
acc55e22
N		 4951/* similar to deny_write_access, but accounts for our holding a reference
4952 * to the file ourselves */
4953static int deny_bitmap_write_access(struct file * file)
4954{
4955 struct inode *inode = file->f_mapping->host;
4956
4957 spin_lock(&inode->i_lock);
4958 if (atomic_read(&inode->i_writecount) > 1) {
4959 spin_unlock(&inode->i_lock);
4960 return -ETXTBSY;
4961 }
4962 atomic_set(&inode->i_writecount, -1);
4963 spin_unlock(&inode->i_lock);
4964
4965 return 0;
4966}
4967
43a70507 4968void restore_bitmap_write_access(struct file *file)
acc55e22
N		 4969{
4970 struct inode *inode = file->f_mapping->host;
4971
4972 spin_lock(&inode->i_lock);
4973 atomic_set(&inode->i_writecount, 1);
4974 spin_unlock(&inode->i_lock);
4975}
4976
fd01b88c 4977static void md_clean(struct mddev *mddev)
6177b472
N		 4978{
4979 mddev->array_sectors = 0;
4980 mddev->external_size = 0;
4981 mddev->dev_sectors = 0;
4982 mddev->raid_disks = 0;
4983 mddev->recovery_cp = 0;
4984 mddev->resync_min = 0;
4985 mddev->resync_max = MaxSector;
4986 mddev->reshape_position = MaxSector;
4987 mddev->external = 0;
4988 mddev->persistent = 0;
4989 mddev->level = LEVEL_NONE;
4990 mddev->clevel[0] = 0;
4991 mddev->flags = 0;
4992 mddev->ro = 0;
4993 mddev->metadata_type[0] = 0;
4994 mddev->chunk_sectors = 0;
4995 mddev->ctime = mddev->utime = 0;
4996 mddev->layout = 0;
4997 mddev->max_disks = 0;
4998 mddev->events = 0;
a8707c08 4999 mddev->can_decrease_events = 0;
6177b472
N		 5000 mddev->delta_disks = 0;
5001 mddev->new_level = LEVEL_NONE;
5002 mddev->new_layout = 0;
5003 mddev->new_chunk_sectors = 0;
5004 mddev->curr_resync = 0;
5005 mddev->resync_mismatches = 0;
5006 mddev->suspend_lo = mddev->suspend_hi = 0;
5007 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5008 mddev->recovery = 0;
5009 mddev->in_sync = 0;
f0b4f7e2 5010 mddev->changed = 0;
6177b472 5011 mddev->degraded = 0;
6177b472
N		 5012 mddev->safemode = 0;
5013 mddev->bitmap_info.offset = 0;
5014 mddev->bitmap_info.default_offset = 0;
5015 mddev->bitmap_info.chunksize = 0;
5016 mddev->bitmap_info.daemon_sleep = 0;
5017 mddev->bitmap_info.max_write_behind = 0;
5018}
5019
fd01b88c 5020static void __md_stop_writes(struct mddev *mddev)
a047e125
N		 5021{
5022 if (mddev->sync_thread) {
5023 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5024 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 5025 reap_sync_thread(mddev);
a047e125
N		 5026 }
5027
5028 del_timer_sync(&mddev->safemode_timer);
5029
5030 bitmap_flush(mddev);
5031 md_super_wait(mddev);
5032
5033 if (!mddev->in_sync || mddev->flags) {
5034 /* mark array as shutdown cleanly */
5035 mddev->in_sync = 1;
5036 md_update_sb(mddev, 1);
5037 }
5038}
defad61a 5039
fd01b88c 5040void md_stop_writes(struct mddev *mddev)
defad61a
N		 5041{
5042 mddev_lock(mddev);
5043 __md_stop_writes(mddev);
5044 mddev_unlock(mddev);
5045}
390ee602 5046EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5047
fd01b88c 5048void md_stop(struct mddev *mddev)
6177b472 5049{
0ca69886 5050 mddev->ready = 0;
6177b472
N		 5051 mddev->pers->stop(mddev);
5052 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5053 mddev->to_remove = &md_redundancy_group;
5054 module_put(mddev->pers->owner);
5055 mddev->pers = NULL;
cca9cf90 5056 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5057}
390ee602 5058EXPORT_SYMBOL_GPL(md_stop);
6177b472 5059
fd01b88c 5060static int md_set_readonly(struct mddev *mddev, int is_open)
a4bd82d0
N		 5061{
5062 int err = 0;
5063 mutex_lock(&mddev->open_mutex);
5064 if (atomic_read(&mddev->openers) > is_open) {
5065 printk("md: %s still in use.\n",mdname(mddev));
5066 err = -EBUSY;
5067 goto out;
5068 }
5069 if (mddev->pers) {
defad61a 5070 __md_stop_writes(mddev);
a4bd82d0
N		 5071
5072 err = -ENXIO;
5073 if (mddev->ro==1)
5074 goto out;
5075 mddev->ro = 1;
5076 set_disk_ro(mddev->gendisk, 1);
5077 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5078 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5079 err = 0;
5080 }
5081out:
5082 mutex_unlock(&mddev->open_mutex);
5083 return err;
5084}
5085
9e653b63
N		 5086/* mode:
5087 * 0 - completely stop and dis-assemble array
9e653b63
N		 5088 * 2 - stop but do not disassemble array
5089 */
fd01b88c 5090static int do_md_stop(struct mddev * mddev, int mode, int is_open)
1da177e4 5091{
1da177e4 5092 struct gendisk *disk = mddev->gendisk;
3cb03002 5093 struct md_rdev *rdev;
1da177e4 5094
c8c00a69 5095 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 5096 if (atomic_read(&mddev->openers) > is_open ||
5097 mddev->sysfs_active) {
df5b20cf 5098 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5099 mutex_unlock(&mddev->open_mutex);
5100 return -EBUSY;
5101 }
1da177e4 5102
6e17b027 5103 if (mddev->pers) {
a4bd82d0
N		 5104 if (mddev->ro)
5105 set_disk_ro(disk, 0);
409c57f3 5106
defad61a 5107 __md_stop_writes(mddev);
a4bd82d0
N		 5108 md_stop(mddev);
5109 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5110 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5111
a4bd82d0 5112 /* tell userspace to handle 'inactive' */
00bcb4ac 5113 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5114
a4bd82d0 5115 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 5116 if (rdev->raid_disk >= 0)
5117 sysfs_unlink_rdev(mddev, rdev);
c4647292 5118
a4bd82d0 5119 set_capacity(disk, 0);
6e17b027 5120 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5121 mddev->changed = 1;
a4bd82d0 5122 revalidate_disk(disk);
0d4ca600 5123
a4bd82d0
N		 5124 if (mddev->ro)
5125 mddev->ro = 0;
6e17b027
N		 5126 } else
5127 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5128 /*
5129 * Free resources if final stop
5130 */
9e653b63 5131 if (mode == 0) {
1da177e4
LT		 5132 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5133
978f946b 5134 bitmap_destroy(mddev);
c3d9714e
N		 5135 if (mddev->bitmap_info.file) {
5136 restore_bitmap_write_access(mddev->bitmap_info.file);
5137 fput(mddev->bitmap_info.file);
5138 mddev->bitmap_info.file = NULL;
978f946b 5139 }
c3d9714e 5140 mddev->bitmap_info.offset = 0;
978f946b 5141
1da177e4
LT		 5142 export_array(mddev);
5143
6177b472 5144 md_clean(mddev);
934d9c23 5145 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5146 if (mddev->hold_active == UNTIL_STOP)
5147 mddev->hold_active = 0;
a4bd82d0 5148 }
3f9d99c1 5149 blk_integrity_unregister(disk);
d7603b7e 5150 md_new_event(mddev);
00bcb4ac 5151 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5152 return 0;
1da177e4
LT		 5153}
5154
fdee8ae4 5155#ifndef MODULE
fd01b88c 5156static void autorun_array(struct mddev *mddev)
1da177e4 5157{
3cb03002 5158 struct md_rdev *rdev;
1da177e4
LT		 5159 int err;
5160
a757e64c 5161 if (list_empty(&mddev->disks))
1da177e4 5162 return;
1da177e4
LT		 5163
5164 printk(KERN_INFO "md: running: ");
5165
159ec1fc 5166 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 5167 char b[BDEVNAME_SIZE];
5168 printk("<%s>", bdevname(rdev->bdev,b));
5169 }
5170 printk("\n");
5171
d710e138 5172 err = do_md_run(mddev);
1da177e4
LT		 5173 if (err) {
5174 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 5175 do_md_stop(mddev, 0, 0);
1da177e4
LT		 5176 }
5177}
5178
5179/*
5180 * lets try to run arrays based on all disks that have arrived
5181 * until now. (those are in pending_raid_disks)
5182 *
5183 * the method: pick the first pending disk, collect all disks with
5184 * the same UUID, remove all from the pending list and put them into
5185 * the 'same_array' list. Then order this list based on superblock
5186 * update time (freshest comes first), kick out 'old' disks and
5187 * compare superblocks. If everything's fine then run it.
5188 *
5189 * If "unit" is allocated, then bump its reference count
5190 */
5191static void autorun_devices(int part)
5192{
3cb03002 5193 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5194 struct mddev *mddev;
1da177e4
LT		 5195 char b[BDEVNAME_SIZE];
5196
5197 printk(KERN_INFO "md: autorun ...\n");
5198 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5199 int unit;
1da177e4 5200 dev_t dev;
ad01c9e3 5201 LIST_HEAD(candidates);
1da177e4 5202 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5203 struct md_rdev, same_set);
1da177e4
LT		 5204
5205 printk(KERN_INFO "md: considering %s ...\n",
5206 bdevname(rdev0->bdev,b));
5207 INIT_LIST_HEAD(&candidates);
159ec1fc 5208 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5209 if (super_90_load(rdev, rdev0, 0) >= 0) {
5210 printk(KERN_INFO "md: adding %s ...\n",
5211 bdevname(rdev->bdev,b));
5212 list_move(&rdev->same_set, &candidates);
5213 }
5214 /*
5215 * now we have a set of devices, with all of them having
5216 * mostly sane superblocks. It's time to allocate the
5217 * mddev.
5218 */
e8703fe1
N		 5219 if (part) {
5220 dev = MKDEV(mdp_major,
5221 rdev0->preferred_minor << MdpMinorShift);
5222 unit = MINOR(dev) >> MdpMinorShift;
5223 } else {
5224 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5225 unit = MINOR(dev);
5226 }
5227 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5228 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5229 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5230 break;
5231 }
1da177e4
LT		 5232
5233 md_probe(dev, NULL, NULL);
5234 mddev = mddev_find(dev);
9bbbca3a
NB		 5235 if (!mddev || !mddev->gendisk) {
5236 if (mddev)
5237 mddev_put(mddev);
5238 printk(KERN_ERR
1da177e4
LT		 5239 "md: cannot allocate memory for md drive.\n");
5240 break;
5241 }
5242 if (mddev_lock(mddev))
5243 printk(KERN_WARNING "md: %s locked, cannot run\n",
5244 mdname(mddev));
5245 else if (mddev->raid_disks || mddev->major_version
5246 || !list_empty(&mddev->disks)) {
5247 printk(KERN_WARNING
5248 "md: %s already running, cannot run %s\n",
5249 mdname(mddev), bdevname(rdev0->bdev,b));
5250 mddev_unlock(mddev);
5251 } else {
5252 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5253 mddev->persistent = 1;
159ec1fc 5254 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5255 list_del_init(&rdev->same_set);
5256 if (bind_rdev_to_array(rdev, mddev))
5257 export_rdev(rdev);
5258 }
5259 autorun_array(mddev);
5260 mddev_unlock(mddev);
5261 }
5262 /* on success, candidates will be empty, on error
5263 * it won't...
5264 */
159ec1fc 5265 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5266 list_del_init(&rdev->same_set);
1da177e4 5267 export_rdev(rdev);
4b80991c 5268 }
1da177e4
LT		 5269 mddev_put(mddev);
5270 }
5271 printk(KERN_INFO "md: ... autorun DONE.\n");
5272}
fdee8ae4 5273#endif /* !MODULE */
1da177e4 5274
1da177e4
LT		 5275static int get_version(void __user * arg)
5276{
5277 mdu_version_t ver;
5278
5279 ver.major = MD_MAJOR_VERSION;
5280 ver.minor = MD_MINOR_VERSION;
5281 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5282
5283 if (copy_to_user(arg, &ver, sizeof(ver)))
5284 return -EFAULT;
5285
5286 return 0;
5287}
5288
fd01b88c 5289static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5290{
5291 mdu_array_info_t info;
a9f326eb 5292 int nr,working,insync,failed,spare;
3cb03002 5293 struct md_rdev *rdev;
1da177e4 5294
a9f326eb 5295 nr=working=insync=failed=spare=0;
159ec1fc 5296 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 5297 nr++;
b2d444d7 5298 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5299 failed++;
5300 else {
5301 working++;
b2d444d7 5302 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5303 insync++;
1da177e4
LT		 5304 else
5305 spare++;
5306 }
5307 }
5308
5309 info.major_version = mddev->major_version;
5310 info.minor_version = mddev->minor_version;
5311 info.patch_version = MD_PATCHLEVEL_VERSION;
5312 info.ctime = mddev->ctime;
5313 info.level = mddev->level;
58c0fed4
AN		 5314 info.size = mddev->dev_sectors / 2;
5315 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5316 info.size = -1;
1da177e4
LT		 5317 info.nr_disks = nr;
5318 info.raid_disks = mddev->raid_disks;
5319 info.md_minor = mddev->md_minor;
5320 info.not_persistent= !mddev->persistent;
5321
5322 info.utime = mddev->utime;
5323 info.state = 0;
5324 if (mddev->in_sync)
5325 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5326 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5327 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5328 info.active_disks = insync;
1da177e4
LT		 5329 info.working_disks = working;
5330 info.failed_disks = failed;
5331 info.spare_disks = spare;
5332
5333 info.layout = mddev->layout;
9d8f0363 5334 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5335
5336 if (copy_to_user(arg, &info, sizeof(info)))
5337 return -EFAULT;
5338
5339 return 0;
5340}
5341
fd01b88c 5342static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5343{
5344 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5345 char *ptr, *buf = NULL;
5346 int err = -ENOMEM;
5347
b5470dc5
DW		 5348 if (md_allow_write(mddev))
5349 file = kmalloc(sizeof(*file), GFP_NOIO);
5350 else
5351 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5352
32a7627c
N		 5353 if (!file)
5354 goto out;
5355
5356 /* bitmap disabled, zero the first byte and copy out */
5357 if (!mddev->bitmap || !mddev->bitmap->file) {
5358 file->pathname[0] = '\0';
5359 goto copy_out;
5360 }
5361
5362 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5363 if (!buf)
5364 goto out;
5365
6bcfd601
CH		 5366 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5367 if (IS_ERR(ptr))
32a7627c
N		 5368 goto out;
5369
5370 strcpy(file->pathname, ptr);
5371
5372copy_out:
5373 err = 0;
5374 if (copy_to_user(arg, file, sizeof(*file)))
5375 err = -EFAULT;
5376out:
5377 kfree(buf);
5378 kfree(file);
5379 return err;
5380}
5381
fd01b88c 5382static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5383{
5384 mdu_disk_info_t info;
3cb03002 5385 struct md_rdev *rdev;
1da177e4
LT		 5386
5387 if (copy_from_user(&info, arg, sizeof(info)))
5388 return -EFAULT;
5389
26ef379f 5390 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5391 if (rdev) {
5392 info.major = MAJOR(rdev->bdev->bd_dev);
5393 info.minor = MINOR(rdev->bdev->bd_dev);
5394 info.raid_disk = rdev->raid_disk;
5395 info.state = 0;
b2d444d7 5396 if (test_bit(Faulty, &rdev->flags))
1da177e4 5397 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5398 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5399 info.state |= (1<<MD_DISK_ACTIVE);
5400 info.state |= (1<<MD_DISK_SYNC);
5401 }
8ddf9efe
N		 5402 if (test_bit(WriteMostly, &rdev->flags))
5403 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5404 } else {
5405 info.major = info.minor = 0;
5406 info.raid_disk = -1;
5407 info.state = (1<<MD_DISK_REMOVED);
5408 }
5409
5410 if (copy_to_user(arg, &info, sizeof(info)))
5411 return -EFAULT;
5412
5413 return 0;
5414}
5415
fd01b88c 5416static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT		 5417{
5418 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5419 struct md_rdev *rdev;
1da177e4
LT		 5420 dev_t dev = MKDEV(info->major,info->minor);
5421
5422 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5423 return -EOVERFLOW;
5424
5425 if (!mddev->raid_disks) {
5426 int err;
5427 /* expecting a device which has a superblock */
5428 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5429 if (IS_ERR(rdev)) {
5430 printk(KERN_WARNING
5431 "md: md_import_device returned %ld\n",
5432 PTR_ERR(rdev));
5433 return PTR_ERR(rdev);
5434 }
5435 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5436 struct md_rdev *rdev0
5437 = list_entry(mddev->disks.next,
5438 struct md_rdev, same_set);
a9f326eb 5439 err = super_types[mddev->major_version]
1da177e4
LT		 5440 .load_super(rdev, rdev0, mddev->minor_version);
5441 if (err < 0) {
5442 printk(KERN_WARNING
5443 "md: %s has different UUID to %s\n",
5444 bdevname(rdev->bdev,b),
5445 bdevname(rdev0->bdev,b2));
5446 export_rdev(rdev);
5447 return -EINVAL;
5448 }
5449 }
5450 err = bind_rdev_to_array(rdev, mddev);
5451 if (err)
5452 export_rdev(rdev);
5453 return err;
5454 }
5455
5456 /*
5457 * add_new_disk can be used once the array is assembled
5458 * to add "hot spares". They must already have a superblock
5459 * written
5460 */
5461 if (mddev->pers) {
5462 int err;
5463 if (!mddev->pers->hot_add_disk) {
5464 printk(KERN_WARNING
5465 "%s: personality does not support diskops!\n",
5466 mdname(mddev));
5467 return -EINVAL;
5468 }
7b1e35f6
N		 5469 if (mddev->persistent)
5470 rdev = md_import_device(dev, mddev->major_version,
5471 mddev->minor_version);
5472 else
5473 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5474 if (IS_ERR(rdev)) {
5475 printk(KERN_WARNING
5476 "md: md_import_device returned %ld\n",
5477 PTR_ERR(rdev));
5478 return PTR_ERR(rdev);
5479 }
1a855a06 5480 /* set saved_raid_disk if appropriate */
41158c7e
N		 5481 if (!mddev->persistent) {
5482 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5483 info->raid_disk < mddev->raid_disks) {
41158c7e 5484 rdev->raid_disk = info->raid_disk;
bf572541
N		 5485 set_bit(In_sync, &rdev->flags);
5486 } else
41158c7e
N		 5487 rdev->raid_disk = -1;
5488 } else
5489 super_types[mddev->major_version].
5490 validate_super(mddev, rdev);
bedd86b7
N		 5491 if ((info->state & (1<<MD_DISK_SYNC)) &&
5492 (!test_bit(In_sync, &rdev->flags) ||
5493 rdev->raid_disk != info->raid_disk)) {
5494 /* This was a hot-add request, but events doesn't
5495 * match, so reject it.
5496 */
5497 export_rdev(rdev);
5498 return -EINVAL;
5499 }
5500
1a855a06
N		 5501 if (test_bit(In_sync, &rdev->flags))
5502 rdev->saved_raid_disk = rdev->raid_disk;
5503 else
5504 rdev->saved_raid_disk = -1;
41158c7e 5505
b2d444d7 5506 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5507 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5508 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5509 else
5510 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5511
1da177e4
LT		 5512 rdev->raid_disk = -1;
5513 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5514 if (!err && !mddev->pers->hot_remove_disk) {
5515 /* If there is hot_add_disk but no hot_remove_disk
5516 * then added disks for geometry changes,
5517 * and should be added immediately.
5518 */
5519 super_types[mddev->major_version].
5520 validate_super(mddev, rdev);
5521 err = mddev->pers->hot_add_disk(mddev, rdev);
5522 if (err)
5523 unbind_rdev_from_array(rdev);
5524 }
1da177e4
LT		 5525 if (err)
5526 export_rdev(rdev);
52664732 5527 else
00bcb4ac 5528 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5529
17571284 5530 md_update_sb(mddev, 1);
72a23c21
NB		 5531 if (mddev->degraded)
5532 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5533 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5534 if (!err)
5535 md_new_event(mddev);
005eca5e 5536 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5537 return err;
5538 }
5539
5540 /* otherwise, add_new_disk is only allowed
5541 * for major_version==0 superblocks
5542 */
5543 if (mddev->major_version != 0) {
5544 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5545 mdname(mddev));
5546 return -EINVAL;
5547 }
5548
5549 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5550 int err;
d710e138 5551 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5552 if (IS_ERR(rdev)) {
5553 printk(KERN_WARNING
5554 "md: error, md_import_device() returned %ld\n",
5555 PTR_ERR(rdev));
5556 return PTR_ERR(rdev);
5557 }
5558 rdev->desc_nr = info->number;
5559 if (info->raid_disk < mddev->raid_disks)
5560 rdev->raid_disk = info->raid_disk;
5561 else
5562 rdev->raid_disk = -1;
5563
1da177e4 5564 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5565 if (info->state & (1<<MD_DISK_SYNC))
5566 set_bit(In_sync, &rdev->flags);
1da177e4 5567
8ddf9efe
N		 5568 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5569 set_bit(WriteMostly, &rdev->flags);
5570
1da177e4
LT		 5571 if (!mddev->persistent) {
5572 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5573 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5574 } else
57b2caa3 5575 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5576 rdev->sectors = rdev->sb_start;
1da177e4 5577
2bf071bf
N		 5578 err = bind_rdev_to_array(rdev, mddev);
5579 if (err) {
5580 export_rdev(rdev);
5581 return err;
5582 }
1da177e4
LT		 5583 }
5584
5585 return 0;
5586}
5587
fd01b88c 5588static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5589{
5590 char b[BDEVNAME_SIZE];
3cb03002 5591 struct md_rdev *rdev;
1da177e4 5592
1da177e4
LT		 5593 rdev = find_rdev(mddev, dev);
5594 if (!rdev)
5595 return -ENXIO;
5596
5597 if (rdev->raid_disk >= 0)
5598 goto busy;
5599
5600 kick_rdev_from_array(rdev);
850b2b42 5601 md_update_sb(mddev, 1);
d7603b7e 5602 md_new_event(mddev);
1da177e4
LT		 5603
5604 return 0;
5605busy:
fdefa4d8 5606 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5607 bdevname(rdev->bdev,b), mdname(mddev));
5608 return -EBUSY;
5609}
5610
fd01b88c 5611static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5612{
5613 char b[BDEVNAME_SIZE];
5614 int err;
3cb03002 5615 struct md_rdev *rdev;
1da177e4
LT		 5616
5617 if (!mddev->pers)
5618 return -ENODEV;
5619
5620 if (mddev->major_version != 0) {
5621 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5622 " version-0 superblocks.\n",
5623 mdname(mddev));
5624 return -EINVAL;
5625 }
5626 if (!mddev->pers->hot_add_disk) {
5627 printk(KERN_WARNING
5628 "%s: personality does not support diskops!\n",
5629 mdname(mddev));
5630 return -EINVAL;
5631 }
5632
d710e138 5633 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5634 if (IS_ERR(rdev)) {
5635 printk(KERN_WARNING
5636 "md: error, md_import_device() returned %ld\n",
5637 PTR_ERR(rdev));
5638 return -EINVAL;
5639 }
5640
5641 if (mddev->persistent)
57b2caa3 5642 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5643 else
77304d2a 5644 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5645
8190e754 5646 rdev->sectors = rdev->sb_start;
1da177e4 5647
b2d444d7 5648 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5649 printk(KERN_WARNING
5650 "md: can not hot-add faulty %s disk to %s!\n",
5651 bdevname(rdev->bdev,b), mdname(mddev));
5652 err = -EINVAL;
5653 goto abort_export;
5654 }
b2d444d7 5655 clear_bit(In_sync, &rdev->flags);
1da177e4 5656 rdev->desc_nr = -1;
5842730d 5657 rdev->saved_raid_disk = -1;
2bf071bf
N		 5658 err = bind_rdev_to_array(rdev, mddev);
5659 if (err)
5660 goto abort_export;
1da177e4
LT		 5661
5662 /*
5663 * The rest should better be atomic, we can have disk failures
5664 * noticed in interrupt contexts ...
5665 */
5666
1da177e4
LT		 5667 rdev->raid_disk = -1;
5668
850b2b42 5669 md_update_sb(mddev, 1);
1da177e4
LT		 5670
5671 /*
5672 * Kick recovery, maybe this spare has to be added to the
5673 * array immediately.
5674 */
5675 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5676 md_wakeup_thread(mddev->thread);
d7603b7e 5677 md_new_event(mddev);
1da177e4
LT		 5678 return 0;
5679
1da177e4
LT		 5680abort_export:
5681 export_rdev(rdev);
5682 return err;
5683}
5684
fd01b88c 5685static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N		 5686{
5687 int err;
5688
36fa3063
N		 5689 if (mddev->pers) {
5690 if (!mddev->pers->quiesce)
5691 return -EBUSY;
5692 if (mddev->recovery || mddev->sync_thread)
5693 return -EBUSY;
5694 /* we should be able to change the bitmap.. */
5695 }
32a7627c 5696
32a7627c 5697
36fa3063
N		 5698 if (fd >= 0) {
5699 if (mddev->bitmap)
5700 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5701 mddev->bitmap_info.file = fget(fd);
32a7627c 5702
c3d9714e 5703 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5704 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5705 mdname(mddev));
5706 return -EBADF;
5707 }
5708
c3d9714e 5709 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5710 if (err) {
5711 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5712 mdname(mddev));
c3d9714e
N		 5713 fput(mddev->bitmap_info.file);
5714 mddev->bitmap_info.file = NULL;
36fa3063
N		 5715 return err;
5716 }
c3d9714e 5717 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5718 } else if (mddev->bitmap == NULL)
5719 return -ENOENT; /* cannot remove what isn't there */
5720 err = 0;
5721 if (mddev->pers) {
5722 mddev->pers->quiesce(mddev, 1);
69e51b44 5723 if (fd >= 0) {
36fa3063 5724 err = bitmap_create(mddev);
69e51b44
N		 5725 if (!err)
5726 err = bitmap_load(mddev);
5727 }
d7375ab3 5728 if (fd < 0 || err) {
36fa3063 5729 bitmap_destroy(mddev);
d7375ab3
N		 5730 fd = -1; /* make sure to put the file */
5731 }
36fa3063 5732 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5733 }
5734 if (fd < 0) {
c3d9714e
N		 5735 if (mddev->bitmap_info.file) {
5736 restore_bitmap_write_access(mddev->bitmap_info.file);
5737 fput(mddev->bitmap_info.file);
acc55e22 5738 }
c3d9714e 5739 mddev->bitmap_info.file = NULL;
36fa3063
N		 5740 }
5741
32a7627c
N		 5742 return err;
5743}
5744
1da177e4
LT		 5745/*
5746 * set_array_info is used two different ways
5747 * The original usage is when creating a new array.
5748 * In this usage, raid_disks is > 0 and it together with
5749 * level, size, not_persistent,layout,chunksize determine the
5750 * shape of the array.
5751 * This will always create an array with a type-0.90.0 superblock.
5752 * The newer usage is when assembling an array.
5753 * In this case raid_disks will be 0, and the major_version field is
5754 * use to determine which style super-blocks are to be found on the devices.
5755 * The minor and patch _version numbers are also kept incase the
5756 * super_block handler wishes to interpret them.
5757 */
fd01b88c 5758static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT		 5759{
5760
5761 if (info->raid_disks == 0) {
5762 /* just setting version number for superblock loading */
5763 if (info->major_version < 0 ||
50511da3 5764 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5765 super_types[info->major_version].name == NULL) {
5766 /* maybe try to auto-load a module? */
5767 printk(KERN_INFO
5768 "md: superblock version %d not known\n",
5769 info->major_version);
5770 return -EINVAL;
5771 }
5772 mddev->major_version = info->major_version;
5773 mddev->minor_version = info->minor_version;
5774 mddev->patch_version = info->patch_version;
3f9d7b0d 5775 mddev->persistent = !info->not_persistent;
cbd19983
N		 5776 /* ensure mddev_put doesn't delete this now that there
5777 * is some minimal configuration.
5778 */
5779 mddev->ctime = get_seconds();
1da177e4
LT		 5780 return 0;
5781 }
5782 mddev->major_version = MD_MAJOR_VERSION;
5783 mddev->minor_version = MD_MINOR_VERSION;
5784 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5785 mddev->ctime = get_seconds();
5786
5787 mddev->level = info->level;
17115e03 5788 mddev->clevel[0] = 0;
58c0fed4 5789 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5790 mddev->raid_disks = info->raid_disks;
5791 /* don't set md_minor, it is determined by which /dev/md* was
5792 * openned
5793 */
5794 if (info->state & (1<<MD_SB_CLEAN))
5795 mddev->recovery_cp = MaxSector;
5796 else
5797 mddev->recovery_cp = 0;
5798 mddev->persistent = ! info->not_persistent;
e691063a 5799 mddev->external = 0;
1da177e4
LT		 5800
5801 mddev->layout = info->layout;
9d8f0363 5802 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5803
5804 mddev->max_disks = MD_SB_DISKS;
5805
e691063a
N		 5806 if (mddev->persistent)
5807 mddev->flags = 0;
850b2b42 5808 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5809
c3d9714e
N		 5810 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5811 mddev->bitmap_info.offset = 0;
b2a2703c 5812
f6705578
N		 5813 mddev->reshape_position = MaxSector;
5814
1da177e4
LT		 5815 /*
5816 * Generate a 128 bit UUID
5817 */
5818 get_random_bytes(mddev->uuid, 16);
5819
f6705578 5820 mddev->new_level = mddev->level;
664e7c41 5821 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5822 mddev->new_layout = mddev->layout;
5823 mddev->delta_disks = 0;
5824
1da177e4
LT		 5825 return 0;
5826}
5827
fd01b88c 5828void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 5829{
b522adcd
DW		 5830 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5831
5832 if (mddev->external_size)
5833 return;
5834
1f403624
DW		 5835 mddev->array_sectors = array_sectors;
5836}
5837EXPORT_SYMBOL(md_set_array_sectors);
5838
fd01b88c 5839static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 5840{
3cb03002 5841 struct md_rdev *rdev;
a35b0d69 5842 int rv;
d71f9f88 5843 int fit = (num_sectors == 0);
a35b0d69
N		 5844
5845 if (mddev->pers->resize == NULL)
5846 return -EINVAL;
d71f9f88
AN		 5847 /* The "num_sectors" is the number of sectors of each device that
5848 * is used. This can only make sense for arrays with redundancy.
5849 * linear and raid0 always use whatever space is available. We can only
5850 * consider changing this number if no resync or reconstruction is
5851 * happening, and if the new size is acceptable. It must fit before the
0f420358 5852 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5853 * of each device. If num_sectors is zero, we find the largest size
5854 * that fits.
a35b0d69
N		 5855 */
5856 if (mddev->sync_thread)
5857 return -EBUSY;
dba034ee
N		 5858 if (mddev->bitmap)
5859 /* Sorry, cannot grow a bitmap yet, just remove it,
5860 * grow, and re-add.
5861 */
5862 return -EBUSY;
159ec1fc 5863 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5864 sector_t avail = rdev->sectors;
01ab5662 5865
d71f9f88
AN		 5866 if (fit && (num_sectors == 0 || num_sectors > avail))
5867 num_sectors = avail;
5868 if (avail < num_sectors)
a35b0d69
N		 5869 return -ENOSPC;
5870 }
d71f9f88 5871 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5872 if (!rv)
5873 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5874 return rv;
5875}
5876
fd01b88c 5877static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 5878{
5879 int rv;
5880 /* change the number of raid disks */
63c70c4f 5881 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5882 return -EINVAL;
5883 if (raid_disks <= 0 ||
233fca36 5884 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5885 return -EINVAL;
63c70c4f 5886 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5887 return -EBUSY;
63c70c4f
N		 5888 mddev->delta_disks = raid_disks - mddev->raid_disks;
5889
5890 rv = mddev->pers->check_reshape(mddev);
de171cb9
N		 5891 if (rv < 0)
5892 mddev->delta_disks = 0;
da943b99
N		 5893 return rv;
5894}
5895
5896
1da177e4
LT		 5897/*
5898 * update_array_info is used to change the configuration of an
5899 * on-line array.
5900 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5901 * fields in the info are checked against the array.
5902 * Any differences that cannot be handled will cause an error.
5903 * Normally, only one change can be managed at a time.
5904 */
fd01b88c 5905static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 5906{
5907 int rv = 0;
5908 int cnt = 0;
36fa3063
N		 5909 int state = 0;
5910
5911 /* calculate expected state,ignoring low bits */
c3d9714e 5912 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5913 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5914
5915 if (mddev->major_version != info->major_version ||
5916 mddev->minor_version != info->minor_version ||
5917/* mddev->patch_version != info->patch_version || */
5918 mddev->ctime != info->ctime ||
5919 mddev->level != info->level ||
5920/* mddev->layout != info->layout || */
5921 !mddev->persistent != info->not_persistent||
9d8f0363 5922 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5923 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5924 ((state^info->state) & 0xfffffe00)
5925 )
1da177e4
LT		 5926 return -EINVAL;
5927 /* Check there is only one change */
58c0fed4
AN		 5928 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5929 cnt++;
5930 if (mddev->raid_disks != info->raid_disks)
5931 cnt++;
5932 if (mddev->layout != info->layout)
5933 cnt++;
5934 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5935 cnt++;
5936 if (cnt == 0)
5937 return 0;
5938 if (cnt > 1)
5939 return -EINVAL;
1da177e4
LT		 5940
5941 if (mddev->layout != info->layout) {
5942 /* Change layout
5943 * we don't need to do anything at the md level, the
5944 * personality will take care of it all.
5945 */
50ac168a 5946 if (mddev->pers->check_reshape == NULL)
1da177e4 5947 return -EINVAL;
597a711b
N		 5948 else {
5949 mddev->new_layout = info->layout;
50ac168a 5950 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 5951 if (rv)
5952 mddev->new_layout = mddev->layout;
5953 return rv;
5954 }
1da177e4 5955 }
58c0fed4 5956 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 5957 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 5958
da943b99
N		 5959 if (mddev->raid_disks != info->raid_disks)
5960 rv = update_raid_disks(mddev, info->raid_disks);
5961
36fa3063
N		 5962 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5963 if (mddev->pers->quiesce == NULL)
5964 return -EINVAL;
5965 if (mddev->recovery || mddev->sync_thread)
5966 return -EBUSY;
5967 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5968 /* add the bitmap */
5969 if (mddev->bitmap)
5970 return -EEXIST;
c3d9714e 5971 if (mddev->bitmap_info.default_offset == 0)
36fa3063 5972 return -EINVAL;
c3d9714e
N		 5973 mddev->bitmap_info.offset =
5974 mddev->bitmap_info.default_offset;
36fa3063
N		 5975 mddev->pers->quiesce(mddev, 1);
5976 rv = bitmap_create(mddev);
69e51b44
N		 5977 if (!rv)
5978 rv = bitmap_load(mddev);
36fa3063
N		 5979 if (rv)
5980 bitmap_destroy(mddev);
5981 mddev->pers->quiesce(mddev, 0);
5982 } else {
5983 /* remove the bitmap */
5984 if (!mddev->bitmap)
5985 return -ENOENT;
5986 if (mddev->bitmap->file)
5987 return -EINVAL;
5988 mddev->pers->quiesce(mddev, 1);
5989 bitmap_destroy(mddev);
5990 mddev->pers->quiesce(mddev, 0);
c3d9714e 5991 mddev->bitmap_info.offset = 0;
36fa3063
N		 5992 }
5993 }
850b2b42 5994 md_update_sb(mddev, 1);
1da177e4
LT		 5995 return rv;
5996}
5997
fd01b88c 5998static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 5999{
3cb03002 6000 struct md_rdev *rdev;
1da177e4
LT		 6001
6002 if (mddev->pers == NULL)
6003 return -ENODEV;
6004
6005 rdev = find_rdev(mddev, dev);
6006 if (!rdev)
6007 return -ENODEV;
6008
6009 md_error(mddev, rdev);
5ef56c8f
N		 6010 if (!test_bit(Faulty, &rdev->flags))
6011 return -EBUSY;
1da177e4
LT		 6012 return 0;
6013}
6014
2f9618ce
AN		 6015/*
6016 * We have a problem here : there is no easy way to give a CHS
6017 * virtual geometry. We currently pretend that we have a 2 heads
6018 * 4 sectors (with a BIG number of cylinders...). This drives
6019 * dosfs just mad... ;-)
6020 */
a885c8c4
CH		 6021static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6022{
fd01b88c 6023 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6024
6025 geo->heads = 2;
6026 geo->sectors = 4;
49ce6cea 6027 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6028 return 0;
6029}
6030
a39907fa 6031static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6032 unsigned int cmd, unsigned long arg)
6033{
6034 int err = 0;
6035 void __user *argp = (void __user *)arg;
fd01b88c 6036 struct mddev *mddev = NULL;
e2218350 6037 int ro;
1da177e4
LT		 6038
6039 if (!capable(CAP_SYS_ADMIN))
6040 return -EACCES;
6041
6042 /*
6043 * Commands dealing with the RAID driver but not any
6044 * particular array:
6045 */
6046 switch (cmd)
6047 {
6048 case RAID_VERSION:
6049 err = get_version(argp);
6050 goto done;
6051
6052 case PRINT_RAID_DEBUG:
6053 err = 0;
6054 md_print_devices();
6055 goto done;
6056
6057#ifndef MODULE
6058 case RAID_AUTORUN:
6059 err = 0;
6060 autostart_arrays(arg);
6061 goto done;
6062#endif
6063 default:;
6064 }
6065
6066 /*
6067 * Commands creating/starting a new array:
6068 */
6069
a39907fa 6070 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6071
6072 if (!mddev) {
6073 BUG();
6074 goto abort;
6075 }
6076
1da177e4
LT		 6077 err = mddev_lock(mddev);
6078 if (err) {
6079 printk(KERN_INFO
6080 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6081 err, cmd);
6082 goto abort;
6083 }
6084
6085 switch (cmd)
6086 {
6087 case SET_ARRAY_INFO:
6088 {
6089 mdu_array_info_t info;
6090 if (!arg)
6091 memset(&info, 0, sizeof(info));
6092 else if (copy_from_user(&info, argp, sizeof(info))) {
6093 err = -EFAULT;
6094 goto abort_unlock;
6095 }
6096 if (mddev->pers) {
6097 err = update_array_info(mddev, &info);
6098 if (err) {
6099 printk(KERN_WARNING "md: couldn't update"
6100 " array info. %d\n", err);
6101 goto abort_unlock;
6102 }
6103 goto done_unlock;
6104 }
6105 if (!list_empty(&mddev->disks)) {
6106 printk(KERN_WARNING
6107 "md: array %s already has disks!\n",
6108 mdname(mddev));
6109 err = -EBUSY;
6110 goto abort_unlock;
6111 }
6112 if (mddev->raid_disks) {
6113 printk(KERN_WARNING
6114 "md: array %s already initialised!\n",
6115 mdname(mddev));
6116 err = -EBUSY;
6117 goto abort_unlock;
6118 }
6119 err = set_array_info(mddev, &info);
6120 if (err) {
6121 printk(KERN_WARNING "md: couldn't set"
6122 " array info. %d\n", err);
6123 goto abort_unlock;
6124 }
6125 }
6126 goto done_unlock;
6127
6128 default:;
6129 }
6130
6131 /*
6132 * Commands querying/configuring an existing array:
6133 */
32a7627c 6134 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6135 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6136 if ((!mddev->raid_disks && !mddev->external)
6137 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6138 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6139 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6140 err = -ENODEV;
6141 goto abort_unlock;
6142 }
6143
6144 /*
6145 * Commands even a read-only array can execute:
6146 */
6147 switch (cmd)
6148 {
6149 case GET_ARRAY_INFO:
6150 err = get_array_info(mddev, argp);
6151 goto done_unlock;
6152
32a7627c 6153 case GET_BITMAP_FILE:
87162a28 6154 err = get_bitmap_file(mddev, argp);
32a7627c
N		 6155 goto done_unlock;
6156
1da177e4
LT		 6157 case GET_DISK_INFO:
6158 err = get_disk_info(mddev, argp);
6159 goto done_unlock;
6160
6161 case RESTART_ARRAY_RW:
6162 err = restart_array(mddev);
6163 goto done_unlock;
6164
6165 case STOP_ARRAY:
d710e138 6166 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 6167 goto done_unlock;
6168
6169 case STOP_ARRAY_RO:
a4bd82d0 6170 err = md_set_readonly(mddev, 1);
1da177e4
LT		 6171 goto done_unlock;
6172
e2218350
DW		 6173 case BLKROSET:
6174 if (get_user(ro, (int __user *)(arg))) {
6175 err = -EFAULT;
6176 goto done_unlock;
6177 }
6178 err = -EINVAL;
6179
6180 /* if the bdev is going readonly the value of mddev->ro
6181 * does not matter, no writes are coming
6182 */
6183 if (ro)
6184 goto done_unlock;
6185
6186 /* are we are already prepared for writes? */
6187 if (mddev->ro != 1)
6188 goto done_unlock;
6189
6190 /* transitioning to readauto need only happen for
6191 * arrays that call md_write_start
6192 */
6193 if (mddev->pers) {
6194 err = restart_array(mddev);
6195 if (err == 0) {
6196 mddev->ro = 2;
6197 set_disk_ro(mddev->gendisk, 0);
6198 }
6199 }
6200 goto done_unlock;
1da177e4
LT		 6201 }
6202
6203 /*
6204 * The remaining ioctls are changing the state of the
f91de92e
N		 6205 * superblock, so we do not allow them on read-only arrays.
6206 * However non-MD ioctls (e.g. get-size) will still come through
6207 * here and hit the 'default' below, so only disallow
6208 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6209 */
bb57fc64 6210 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6211 if (mddev->ro == 2) {
6212 mddev->ro = 0;
00bcb4ac 6213 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 6214 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6215 md_wakeup_thread(mddev->thread);
f91de92e
N		 6216 } else {
6217 err = -EROFS;
6218 goto abort_unlock;
6219 }
1da177e4
LT		 6220 }
6221
6222 switch (cmd)
6223 {
6224 case ADD_NEW_DISK:
6225 {
6226 mdu_disk_info_t info;
6227 if (copy_from_user(&info, argp, sizeof(info)))
6228 err = -EFAULT;
6229 else
6230 err = add_new_disk(mddev, &info);
6231 goto done_unlock;
6232 }
6233
6234 case HOT_REMOVE_DISK:
6235 err = hot_remove_disk(mddev, new_decode_dev(arg));
6236 goto done_unlock;
6237
6238 case HOT_ADD_DISK:
6239 err = hot_add_disk(mddev, new_decode_dev(arg));
6240 goto done_unlock;
6241
6242 case SET_DISK_FAULTY:
6243 err = set_disk_faulty(mddev, new_decode_dev(arg));
6244 goto done_unlock;
6245
6246 case RUN_ARRAY:
d710e138 6247 err = do_md_run(mddev);
1da177e4
LT		 6248 goto done_unlock;
6249
32a7627c
N		 6250 case SET_BITMAP_FILE:
6251 err = set_bitmap_file(mddev, (int)arg);
6252 goto done_unlock;
6253
1da177e4 6254 default:
1da177e4
LT		 6255 err = -EINVAL;
6256 goto abort_unlock;
6257 }
6258
6259done_unlock:
6260abort_unlock:
d3374825
N		 6261 if (mddev->hold_active == UNTIL_IOCTL &&
6262 err != -EINVAL)
6263 mddev->hold_active = 0;
1da177e4
LT		 6264 mddev_unlock(mddev);
6265
6266 return err;
6267done:
6268 if (err)
6269 MD_BUG();
6270abort:
6271 return err;
6272}
aa98aa31
AB		 6273#ifdef CONFIG_COMPAT
6274static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6275 unsigned int cmd, unsigned long arg)
6276{
6277 switch (cmd) {
6278 case HOT_REMOVE_DISK:
6279 case HOT_ADD_DISK:
6280 case SET_DISK_FAULTY:
6281 case SET_BITMAP_FILE:
6282 /* These take in integer arg, do not convert */
6283 break;
6284 default:
6285 arg = (unsigned long)compat_ptr(arg);
6286 break;
6287 }
6288
6289 return md_ioctl(bdev, mode, cmd, arg);
6290}
6291#endif /* CONFIG_COMPAT */
1da177e4 6292
a39907fa 6293static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6294{
6295 /*
6296 * Succeed if we can lock the mddev, which confirms that
6297 * it isn't being stopped right now.
6298 */
fd01b88c 6299 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6300 int err;
6301
d3374825
N		 6302 if (mddev->gendisk != bdev->bd_disk) {
6303 /* we are racing with mddev_put which is discarding this
6304 * bd_disk.
6305 */
6306 mddev_put(mddev);
6307 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6308 flush_workqueue(md_misc_wq);
d3374825
N		 6309 /* Then retry the open from the top */
6310 return -ERESTARTSYS;
6311 }
6312 BUG_ON(mddev != bdev->bd_disk->private_data);
6313
c8c00a69 6314 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6315 goto out;
6316
6317 err = 0;
f2ea68cf 6318 atomic_inc(&mddev->openers);
c8c00a69 6319 mutex_unlock(&mddev->open_mutex);
1da177e4 6320
f0b4f7e2 6321 check_disk_change(bdev);
1da177e4
LT		 6322 out:
6323 return err;
6324}
6325
a39907fa 6326static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6327{
fd01b88c 6328 struct mddev *mddev = disk->private_data;
1da177e4 6329
52e5f9d1 6330 BUG_ON(!mddev);
f2ea68cf 6331 atomic_dec(&mddev->openers);
1da177e4
LT		 6332 mddev_put(mddev);
6333
6334 return 0;
6335}
f0b4f7e2
N		 6336
6337static int md_media_changed(struct gendisk *disk)
6338{
fd01b88c 6339 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6340
6341 return mddev->changed;
6342}
6343
6344static int md_revalidate(struct gendisk *disk)
6345{
fd01b88c 6346 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6347
6348 mddev->changed = 0;
6349 return 0;
6350}
83d5cde4 6351static const struct block_device_operations md_fops =
1da177e4
LT		 6352{
6353 .owner = THIS_MODULE,
a39907fa
AV		 6354 .open = md_open,
6355 .release = md_release,
b492b852 6356 .ioctl = md_ioctl,
aa98aa31
AB		 6357#ifdef CONFIG_COMPAT
6358 .compat_ioctl = md_compat_ioctl,
6359#endif
a885c8c4 6360 .getgeo = md_getgeo,
f0b4f7e2
N		 6361 .media_changed = md_media_changed,
6362 .revalidate_disk= md_revalidate,
1da177e4
LT		 6363};
6364
75c96f85 6365static int md_thread(void * arg)
1da177e4 6366{
2b8bf345 6367 struct md_thread *thread = arg;
1da177e4 6368
1da177e4
LT		 6369 /*
6370 * md_thread is a 'system-thread', it's priority should be very
6371 * high. We avoid resource deadlocks individually in each
6372 * raid personality. (RAID5 does preallocation) We also use RR and
6373 * the very same RT priority as kswapd, thus we will never get
6374 * into a priority inversion deadlock.
6375 *
6376 * we definitely have to have equal or higher priority than
6377 * bdflush, otherwise bdflush will deadlock if there are too
6378 * many dirty RAID5 blocks.
6379 */
1da177e4 6380
6985c43f 6381 allow_signal(SIGKILL);
a6fb0934 6382 while (!kthread_should_stop()) {
1da177e4 6383
93588e22
N		 6384 /* We need to wait INTERRUPTIBLE so that
6385 * we don't add to the load-average.
6386 * That means we need to be sure no signals are
6387 * pending
6388 */
6389 if (signal_pending(current))
6390 flush_signals(current);
6391
6392 wait_event_interruptible_timeout
6393 (thread->wqueue,
6394 test_bit(THREAD_WAKEUP, &thread->flags)
6395 || kthread_should_stop(),
6396 thread->timeout);
1da177e4 6397
6c987910
N		 6398 clear_bit(THREAD_WAKEUP, &thread->flags);
6399 if (!kthread_should_stop())
589a594b 6400 thread->run(thread->mddev);
1da177e4 6401 }
a6fb0934 6402
1da177e4
LT		 6403 return 0;
6404}
6405
2b8bf345 6406void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6407{
6408 if (thread) {
36a4e1fe 6409 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6410 set_bit(THREAD_WAKEUP, &thread->flags);
6411 wake_up(&thread->wqueue);
6412 }
6413}
6414
2b8bf345 6415struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
1da177e4
LT		 6416 const char *name)
6417{
2b8bf345 6418 struct md_thread *thread;
1da177e4 6419
2b8bf345 6420 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6421 if (!thread)
6422 return NULL;
6423
1da177e4
LT		 6424 init_waitqueue_head(&thread->wqueue);
6425
1da177e4
LT		 6426 thread->run = run;
6427 thread->mddev = mddev;
32a7627c 6428 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6429 thread->tsk = kthread_run(md_thread, thread,
6430 "%s_%s",
6431 mdname(thread->mddev),
6432 name ?: mddev->pers->name);
a6fb0934 6433 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6434 kfree(thread);
6435 return NULL;
6436 }
1da177e4
LT		 6437 return thread;
6438}
6439
2b8bf345 6440void md_unregister_thread(struct md_thread **threadp)
1da177e4 6441{
2b8bf345 6442 struct md_thread *thread = *threadp;
e0cf8f04
N		 6443 if (!thread)
6444 return;
36a4e1fe 6445 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6446 /* Locking ensures that mddev_unlock does not wake_up a
6447 * non-existent thread
6448 */
6449 spin_lock(&pers_lock);
6450 *threadp = NULL;
6451 spin_unlock(&pers_lock);
a6fb0934
N		 6452
6453 kthread_stop(thread->tsk);
1da177e4
LT		 6454 kfree(thread);
6455}
6456
fd01b88c 6457void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6458{
6459 if (!mddev) {
6460 MD_BUG();
6461 return;
6462 }
6463
b2d444d7 6464 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6465 return;
6bfe0b49 6466
de393cde 6467 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6468 return;
6469 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6470 if (mddev->degraded)
6471 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6472 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6473 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6474 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6475 md_wakeup_thread(mddev->thread);
768a418d 6476 if (mddev->event_work.func)
e804ac78 6477 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6478 md_new_event_inintr(mddev);
1da177e4
LT		 6479}
6480
6481/* seq_file implementation /proc/mdstat */
6482
6483static void status_unused(struct seq_file *seq)
6484{
6485 int i = 0;
3cb03002 6486 struct md_rdev *rdev;
1da177e4
LT		 6487
6488 seq_printf(seq, "unused devices: ");
6489
159ec1fc 6490 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6491 char b[BDEVNAME_SIZE];
6492 i++;
6493 seq_printf(seq, "%s ",
6494 bdevname(rdev->bdev,b));
6495 }
6496 if (!i)
6497 seq_printf(seq, "<none>");
6498
6499 seq_printf(seq, "\n");
6500}
6501
6502
fd01b88c 6503static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6504{
dd71cf6b
N		 6505 sector_t max_sectors, resync, res;
6506 unsigned long dt, db;
6507 sector_t rt;
4588b42e
N		 6508 int scale;
6509 unsigned int per_milli;
1da177e4 6510
dd71cf6b 6511 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6512
6513 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6514 max_sectors = mddev->resync_max_sectors;
1da177e4 6515 else
dd71cf6b 6516 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6517
6518 /*
6519 * Should not happen.
6520 */
dd71cf6b 6521 if (!max_sectors) {
1da177e4
LT		 6522 MD_BUG();
6523 return;
6524 }
4588b42e 6525 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6526 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6527 * u32, as those are the requirements for sector_div.
6528 * Thus 'scale' must be at least 10
6529 */
6530 scale = 10;
6531 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6532 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6533 scale++;
6534 }
6535 res = (resync>>scale)*1000;
dd71cf6b 6536 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6537
6538 per_milli = res;
1da177e4 6539 {
4588b42e 6540 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6541 seq_printf(seq, "[");
6542 for (i = 0; i < x; i++)
6543 seq_printf(seq, "=");
6544 seq_printf(seq, ">");
6545 for (i = 0; i < y; i++)
6546 seq_printf(seq, ".");
6547 seq_printf(seq, "] ");
6548 }
4588b42e 6549 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6550 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6551 "reshape" :
61df9d91
N		 6552 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6553 "check" :
6554 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6555 "resync" : "recovery"))),
6556 per_milli/10, per_milli % 10,
dd71cf6b
N		 6557 (unsigned long long) resync/2,
6558 (unsigned long long) max_sectors/2);
1da177e4
LT		 6559
6560 /*
1da177e4
LT		 6561 * dt: time from mark until now
6562 * db: blocks written from mark until now
6563 * rt: remaining time
dd71cf6b
N		 6564 *
6565 * rt is a sector_t, so could be 32bit or 64bit.
6566 * So we divide before multiply in case it is 32bit and close
6567 * to the limit.
25985edc 6568 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6569 * near the end of resync when the number of remaining sectors
6570 * is close to 'db'.
6571 * We then divide rt by 32 after multiplying by db to compensate.
6572 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6573 */
6574 dt = ((jiffies - mddev->resync_mark) / HZ);
6575 if (!dt) dt++;
ff4e8d9a
N		 6576 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6577 - mddev->resync_mark_cnt;
1da177e4 6578
dd71cf6b
N		 6579 rt = max_sectors - resync; /* number of remaining sectors */
6580 sector_div(rt, db/32+1);
6581 rt *= dt;
6582 rt >>= 5;
6583
6584 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6585 ((unsigned long)rt % 60)/6);
1da177e4 6586
ff4e8d9a 6587 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6588}
6589
6590static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6591{
6592 struct list_head *tmp;
6593 loff_t l = *pos;
fd01b88c 6594 struct mddev *mddev;
1da177e4
LT		 6595
6596 if (l >= 0x10000)
6597 return NULL;
6598 if (!l--)
6599 /* header */
6600 return (void*)1;
6601
6602 spin_lock(&all_mddevs_lock);
6603 list_for_each(tmp,&all_mddevs)
6604 if (!l--) {
fd01b88c 6605 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6606 mddev_get(mddev);
6607 spin_unlock(&all_mddevs_lock);
6608 return mddev;
6609 }
6610 spin_unlock(&all_mddevs_lock);
6611 if (!l--)
6612 return (void*)2;/* tail */
6613 return NULL;
6614}
6615
6616static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6617{
6618 struct list_head *tmp;
fd01b88c 6619 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6620
6621 ++*pos;
6622 if (v == (void*)2)
6623 return NULL;
6624
6625 spin_lock(&all_mddevs_lock);
6626 if (v == (void*)1)
6627 tmp = all_mddevs.next;
6628 else
6629 tmp = mddev->all_mddevs.next;
6630 if (tmp != &all_mddevs)
fd01b88c 6631 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 6632 else {
6633 next_mddev = (void*)2;
6634 *pos = 0x10000;
6635 }
6636 spin_unlock(&all_mddevs_lock);
6637
6638 if (v != (void*)1)
6639 mddev_put(mddev);
6640 return next_mddev;
6641
6642}
6643
6644static void md_seq_stop(struct seq_file *seq, void *v)
6645{
fd01b88c 6646 struct mddev *mddev = v;
1da177e4
LT		 6647
6648 if (mddev && v != (void*)1 && v != (void*)2)
6649 mddev_put(mddev);
6650}
6651
6652static int md_seq_show(struct seq_file *seq, void *v)
6653{
fd01b88c 6654 struct mddev *mddev = v;
dd8ac336 6655 sector_t sectors;
3cb03002 6656 struct md_rdev *rdev;
32a7627c 6657 struct bitmap *bitmap;
1da177e4
LT		 6658
6659 if (v == (void*)1) {
84fc4b56 6660 struct md_personality *pers;
1da177e4
LT		 6661 seq_printf(seq, "Personalities : ");
6662 spin_lock(&pers_lock);
2604b703
N		 6663 list_for_each_entry(pers, &pers_list, list)
6664 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6665
6666 spin_unlock(&pers_lock);
6667 seq_printf(seq, "\n");
f1514638 6668 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6669 return 0;
6670 }
6671 if (v == (void*)2) {
6672 status_unused(seq);
6673 return 0;
6674 }
6675
5dc5cf7d 6676 if (mddev_lock(mddev) < 0)
1da177e4 6677 return -EINTR;
5dc5cf7d 6678
1da177e4
LT		 6679 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6680 seq_printf(seq, "%s : %sactive", mdname(mddev),
6681 mddev->pers ? "" : "in");
6682 if (mddev->pers) {
f91de92e 6683 if (mddev->ro==1)
1da177e4 6684 seq_printf(seq, " (read-only)");
f91de92e 6685 if (mddev->ro==2)
52720ae7 6686 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6687 seq_printf(seq, " %s", mddev->pers->name);
6688 }
6689
dd8ac336 6690 sectors = 0;
159ec1fc 6691 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 6692 char b[BDEVNAME_SIZE];
6693 seq_printf(seq, " %s[%d]",
6694 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6695 if (test_bit(WriteMostly, &rdev->flags))
6696 seq_printf(seq, "(W)");
b2d444d7 6697 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6698 seq_printf(seq, "(F)");
6699 continue;
b325a32e
N		 6700 } else if (rdev->raid_disk < 0)
6701 seq_printf(seq, "(S)"); /* spare */
dd8ac336 6702 sectors += rdev->sectors;
1da177e4
LT		 6703 }
6704
6705 if (!list_empty(&mddev->disks)) {
6706 if (mddev->pers)
6707 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6708 (unsigned long long)
6709 mddev->array_sectors / 2);
1da177e4
LT		 6710 else
6711 seq_printf(seq, "\n %llu blocks",
dd8ac336 6712 (unsigned long long)sectors / 2);
1da177e4 6713 }
1cd6bf19
N		 6714 if (mddev->persistent) {
6715 if (mddev->major_version != 0 ||
6716 mddev->minor_version != 90) {
6717 seq_printf(seq," super %d.%d",
6718 mddev->major_version,
6719 mddev->minor_version);
6720 }
e691063a
N		 6721 } else if (mddev->external)
6722 seq_printf(seq, " super external:%s",
6723 mddev->metadata_type);
6724 else
1cd6bf19 6725 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6726
6727 if (mddev->pers) {
d710e138 6728 mddev->pers->status(seq, mddev);
1da177e4 6729 seq_printf(seq, "\n ");
8e1b39d6
N		 6730 if (mddev->pers->sync_request) {
6731 if (mddev->curr_resync > 2) {
d710e138 6732 status_resync(seq, mddev);
8e1b39d6
N		 6733 seq_printf(seq, "\n ");
6734 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6735 seq_printf(seq, "\tresync=DELAYED\n ");
6736 else if (mddev->recovery_cp < MaxSector)
6737 seq_printf(seq, "\tresync=PENDING\n ");
6738 }
32a7627c
N		 6739 } else
6740 seq_printf(seq, "\n ");
6741
6742 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6743 unsigned long chunk_kb;
6744 unsigned long flags;
32a7627c 6745 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6746 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6747 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6748 "%lu%s chunk",
6749 bitmap->pages - bitmap->missing_pages,
6750 bitmap->pages,
6751 (bitmap->pages - bitmap->missing_pages)
6752 << (PAGE_SHIFT - 10),
42a04b50 6753 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6754 chunk_kb ? "KB" : "B");
78d742d8
N		 6755 if (bitmap->file) {
6756 seq_printf(seq, ", file: ");
c32c2f63 6757 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6758 }
78d742d8 6759
32a7627c
N		 6760 seq_printf(seq, "\n");
6761 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6762 }
6763
6764 seq_printf(seq, "\n");
6765 }
6766 mddev_unlock(mddev);
6767
6768 return 0;
6769}
6770
110518bc 6771static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6772 .start = md_seq_start,
6773 .next = md_seq_next,
6774 .stop = md_seq_stop,
6775 .show = md_seq_show,
6776};
6777
6778static int md_seq_open(struct inode *inode, struct file *file)
6779{
f1514638 6780 struct seq_file *seq;
1da177e4
LT		 6781 int error;
6782
6783 error = seq_open(file, &md_seq_ops);
d7603b7e 6784 if (error)
f1514638
KS		 6785 return error;
6786
6787 seq = file->private_data;
6788 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6789 return error;
6790}
6791
d7603b7e
N		 6792static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6793{
f1514638 6794 struct seq_file *seq = filp->private_data;
d7603b7e
N		 6795 int mask;
6796
6797 poll_wait(filp, &md_event_waiters, wait);
6798
6799 /* always allow read */
6800 mask = POLLIN | POLLRDNORM;
6801
f1514638 6802 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 6803 mask |= POLLERR | POLLPRI;
6804 return mask;
6805}
6806
fa027c2a 6807static const struct file_operations md_seq_fops = {
e24650c2 6808 .owner = THIS_MODULE,
1da177e4
LT		 6809 .open = md_seq_open,
6810 .read = seq_read,
6811 .llseek = seq_lseek,
c3f94b40 6812 .release = seq_release_private,
d7603b7e 6813 .poll = mdstat_poll,
1da177e4
LT		 6814};
6815
84fc4b56 6816int register_md_personality(struct md_personality *p)
1da177e4 6817{
1da177e4 6818 spin_lock(&pers_lock);
2604b703
N		 6819 list_add_tail(&p->list, &pers_list);
6820 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6821 spin_unlock(&pers_lock);
6822 return 0;
6823}
6824
84fc4b56 6825int unregister_md_personality(struct md_personality *p)
1da177e4 6826{
2604b703 6827 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6828 spin_lock(&pers_lock);
2604b703 6829 list_del_init(&p->list);
1da177e4
LT		 6830 spin_unlock(&pers_lock);
6831 return 0;
6832}
6833
fd01b88c 6834static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 6835{
3cb03002 6836 struct md_rdev * rdev;
1da177e4 6837 int idle;
eea1bf38 6838 int curr_events;
1da177e4
LT		 6839
6840 idle = 1;
4b80991c
N		 6841 rcu_read_lock();
6842 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6843 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6844 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6845 (int)part_stat_read(&disk->part0, sectors[1]) -
6846 atomic_read(&disk->sync_io);
713f6ab1
N		 6847 /* sync IO will cause sync_io to increase before the disk_stats
6848 * as sync_io is counted when a request starts, and
6849 * disk_stats is counted when it completes.
6850 * So resync activity will cause curr_events to be smaller than
6851 * when there was no such activity.
6852 * non-sync IO will cause disk_stat to increase without
6853 * increasing sync_io so curr_events will (eventually)
6854 * be larger than it was before. Once it becomes
6855 * substantially larger, the test below will cause
6856 * the array to appear non-idle, and resync will slow
6857 * down.
6858 * If there is a lot of outstanding resync activity when
6859 * we set last_event to curr_events, then all that activity
6860 * completing might cause the array to appear non-idle
6861 * and resync will be slowed down even though there might
6862 * not have been non-resync activity. This will only
6863 * happen once though. 'last_events' will soon reflect
6864 * the state where there is little or no outstanding
6865 * resync requests, and further resync activity will
6866 * always make curr_events less than last_events.
c0e48521 6867 *
1da177e4 6868 */
eea1bf38 6869 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6870 rdev->last_events = curr_events;
6871 idle = 0;
6872 }
6873 }
4b80991c 6874 rcu_read_unlock();
1da177e4
LT		 6875 return idle;
6876}
6877
fd01b88c 6878void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 6879{
6880 /* another "blocks" (512byte) blocks have been synced */
6881 atomic_sub(blocks, &mddev->recovery_active);
6882 wake_up(&mddev->recovery_wait);
6883 if (!ok) {
dfc70645 6884 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6885 md_wakeup_thread(mddev->thread);
6886 // stop recovery, signal do_sync ....
6887 }
6888}
6889
6890
06d91a5f
N		 6891/* md_write_start(mddev, bi)
6892 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6893 * in superblock) before writing, schedule a superblock update
6894 * and wait for it to complete.
06d91a5f 6895 */
fd01b88c 6896void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 6897{
0fd62b86 6898 int did_change = 0;
06d91a5f 6899 if (bio_data_dir(bi) != WRITE)
3d310eb7 6900 return;
06d91a5f 6901
f91de92e
N		 6902 BUG_ON(mddev->ro == 1);
6903 if (mddev->ro == 2) {
6904 /* need to switch to read/write */
6905 mddev->ro = 0;
6906 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6907 md_wakeup_thread(mddev->thread);
25156198 6908 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6909 did_change = 1;
f91de92e 6910 }
06d91a5f 6911 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6912 if (mddev->safemode == 1)
6913 mddev->safemode = 0;
06d91a5f 6914 if (mddev->in_sync) {
a9701a30 6915 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6916 if (mddev->in_sync) {
6917 mddev->in_sync = 0;
850b2b42 6918 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6919 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6920 md_wakeup_thread(mddev->thread);
0fd62b86 6921 did_change = 1;
3d310eb7 6922 }
a9701a30 6923 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6924 }
0fd62b86 6925 if (did_change)
00bcb4ac 6926 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 6927 wait_event(mddev->sb_wait,
09a44cc1 6928 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 6929}
6930
fd01b88c 6931void md_write_end(struct mddev *mddev)
1da177e4
LT		 6932{
6933 if (atomic_dec_and_test(&mddev->writes_pending)) {
6934 if (mddev->safemode == 2)
6935 md_wakeup_thread(mddev->thread);
16f17b39 6936 else if (mddev->safemode_delay)
1da177e4
LT		 6937 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6938 }
6939}
6940
2a2275d6
N		 6941/* md_allow_write(mddev)
6942 * Calling this ensures that the array is marked 'active' so that writes
6943 * may proceed without blocking. It is important to call this before
6944 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6945 * Must be called with mddev_lock held.
b5470dc5
DW		 6946 *
6947 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6948 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 6949 */
fd01b88c 6950int md_allow_write(struct mddev *mddev)
2a2275d6
N		 6951{
6952 if (!mddev->pers)
b5470dc5 6953 return 0;
2a2275d6 6954 if (mddev->ro)
b5470dc5 6955 return 0;
1a0fd497 6956 if (!mddev->pers->sync_request)
b5470dc5 6957 return 0;
2a2275d6
N		 6958
6959 spin_lock_irq(&mddev->write_lock);
6960 if (mddev->in_sync) {
6961 mddev->in_sync = 0;
6962 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6963 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 6964 if (mddev->safemode_delay &&
6965 mddev->safemode == 0)
6966 mddev->safemode = 1;
6967 spin_unlock_irq(&mddev->write_lock);
6968 md_update_sb(mddev, 0);
00bcb4ac 6969 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 6970 } else
6971 spin_unlock_irq(&mddev->write_lock);
b5470dc5 6972
070dc6dd 6973 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 6974 return -EAGAIN;
6975 else
6976 return 0;
2a2275d6
N		 6977}
6978EXPORT_SYMBOL_GPL(md_allow_write);
6979
1da177e4
LT		 6980#define SYNC_MARKS 10
6981#define SYNC_MARK_STEP (3*HZ)
fd01b88c 6982void md_do_sync(struct mddev *mddev)
1da177e4 6983{
fd01b88c 6984 struct mddev *mddev2;
1da177e4
LT		 6985 unsigned int currspeed = 0,
6986 window;
57afd89f 6987 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 6988 unsigned long mark[SYNC_MARKS];
6989 sector_t mark_cnt[SYNC_MARKS];
6990 int last_mark,m;
6991 struct list_head *tmp;
6992 sector_t last_check;
57afd89f 6993 int skipped = 0;
3cb03002 6994 struct md_rdev *rdev;
61df9d91 6995 char *desc;
1da177e4
LT		 6996
6997 /* just incase thread restarts... */
6998 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6999 return;
5fd6c1dc
N		 7000 if (mddev->ro) /* never try to sync a read-only array */
7001 return;
1da177e4 7002
61df9d91
N		 7003 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7004 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7005 desc = "data-check";
7006 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7007 desc = "requested-resync";
7008 else
7009 desc = "resync";
7010 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7011 desc = "reshape";
7012 else
7013 desc = "recovery";
7014
1da177e4
LT		 7015 /* we overload curr_resync somewhat here.
7016 * 0 == not engaged in resync at all
7017 * 2 == checking that there is no conflict with another sync
7018 * 1 == like 2, but have yielded to allow conflicting resync to
7019 * commense
7020 * other == active in resync - this many blocks
7021 *
7022 * Before starting a resync we must have set curr_resync to
7023 * 2, and then checked that every "conflicting" array has curr_resync
7024 * less than ours. When we find one that is the same or higher
7025 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7026 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7027 * This will mean we have to start checking from the beginning again.
7028 *
7029 */
7030
7031 do {
7032 mddev->curr_resync = 2;
7033
7034 try_again:
404e4b43 7035 if (kthread_should_stop())
6985c43f 7036 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7037
7038 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7039 goto skip;
29ac4aa3 7040 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7041 if (mddev2 == mddev)
7042 continue;
90b08710
BS		 7043 if (!mddev->parallel_resync
7044 && mddev2->curr_resync
7045 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7046 DEFINE_WAIT(wq);
7047 if (mddev < mddev2 && mddev->curr_resync == 2) {
7048 /* arbitrarily yield */
7049 mddev->curr_resync = 1;
7050 wake_up(&resync_wait);
7051 }
7052 if (mddev > mddev2 && mddev->curr_resync == 1)
7053 /* no need to wait here, we can wait the next
7054 * time 'round when curr_resync == 2
7055 */
7056 continue;
9744197c
N		 7057 /* We need to wait 'interruptible' so as not to
7058 * contribute to the load average, and not to
7059 * be caught by 'softlockup'
7060 */
7061 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7062 if (!kthread_should_stop() &&
8712e553 7063 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7064 printk(KERN_INFO "md: delaying %s of %s"
7065 " until %s has finished (they"
1da177e4 7066 " share one or more physical units)\n",
61df9d91 7067 desc, mdname(mddev), mdname(mddev2));
1da177e4 7068 mddev_put(mddev2);
9744197c
N		 7069 if (signal_pending(current))
7070 flush_signals(current);
1da177e4
LT		 7071 schedule();
7072 finish_wait(&resync_wait, &wq);
7073 goto try_again;
7074 }
7075 finish_wait(&resync_wait, &wq);
7076 }
7077 }
7078 } while (mddev->curr_resync < 2);
7079
5fd6c1dc 7080 j = 0;
9d88883e 7081 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7082 /* resync follows the size requested by the personality,
57afd89f 7083 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7084 */
7085 max_sectors = mddev->resync_max_sectors;
9d88883e 7086 mddev->resync_mismatches = 0;
5fd6c1dc 7087 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7088 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7089 j = mddev->resync_min;
7090 else if (!mddev->bitmap)
5fd6c1dc 7091 j = mddev->recovery_cp;
5e96ee65 7092
ccfcc3c1 7093 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 7094 max_sectors = mddev->dev_sectors;
5fd6c1dc 7095 else {
1da177e4 7096 /* recovery follows the physical size of devices */
58c0fed4 7097 max_sectors = mddev->dev_sectors;
5fd6c1dc 7098 j = MaxSector;
4e59ca7d
DW		 7099 rcu_read_lock();
7100 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 7101 if (rdev->raid_disk >= 0 &&
7102 !test_bit(Faulty, &rdev->flags) &&
7103 !test_bit(In_sync, &rdev->flags) &&
7104 rdev->recovery_offset < j)
7105 j = rdev->recovery_offset;
4e59ca7d 7106 rcu_read_unlock();
5fd6c1dc 7107 }
1da177e4 7108
61df9d91
N		 7109 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7110 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7111 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7112 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7113 "(but not more than %d KB/sec) for %s.\n",
7114 speed_max(mddev), desc);
1da177e4 7115
eea1bf38 7116 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7117
57afd89f 7118 io_sectors = 0;
1da177e4
LT		 7119 for (m = 0; m < SYNC_MARKS; m++) {
7120 mark[m] = jiffies;
57afd89f 7121 mark_cnt[m] = io_sectors;
1da177e4
LT		 7122 }
7123 last_mark = 0;
7124 mddev->resync_mark = mark[last_mark];
7125 mddev->resync_mark_cnt = mark_cnt[last_mark];
7126
7127 /*
7128 * Tune reconstruction:
7129 */
7130 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7131 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7132 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7133
7134 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7135 last_check = 0;
7136
7137 if (j>2) {
7138 printk(KERN_INFO
61df9d91
N		 7139 "md: resuming %s of %s from checkpoint.\n",
7140 desc, mdname(mddev));
1da177e4
LT		 7141 mddev->curr_resync = j;
7142 }
75d3da43 7143 mddev->curr_resync_completed = j;
1da177e4
LT		 7144
7145 while (j < max_sectors) {
57afd89f 7146 sector_t sectors;
1da177e4 7147
57afd89f 7148 skipped = 0;
97e4f42d 7149
7a91ee1f
N		 7150 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7151 ((mddev->curr_resync > mddev->curr_resync_completed &&
7152 (mddev->curr_resync - mddev->curr_resync_completed)
7153 > (max_sectors >> 4)) ||
7154 (j - mddev->curr_resync_completed)*2
7155 >= mddev->resync_max - mddev->curr_resync_completed
7156 )) {
97e4f42d 7157 /* time to update curr_resync_completed */
97e4f42d
N		 7158 wait_event(mddev->recovery_wait,
7159 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7160 mddev->curr_resync_completed = j;
070dc6dd 7161 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7162 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7163 }
acb180b0 7164
e62e58a5
N		 7165 while (j >= mddev->resync_max && !kthread_should_stop()) {
7166 /* As this condition is controlled by user-space,
7167 * we can block indefinitely, so use '_interruptible'
7168 * to avoid triggering warnings.
7169 */
7170 flush_signals(current); /* just in case */
7171 wait_event_interruptible(mddev->recovery_wait,
7172 mddev->resync_max > j
7173 || kthread_should_stop());
7174 }
acb180b0
N		 7175
7176 if (kthread_should_stop())
7177 goto interrupted;
7178
57afd89f 7179 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7180 currspeed < speed_min(mddev));
57afd89f 7181 if (sectors == 0) {
dfc70645 7182 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7183 goto out;
7184 }
57afd89f
N		 7185
7186 if (!skipped) { /* actual IO requested */
7187 io_sectors += sectors;
7188 atomic_add(sectors, &mddev->recovery_active);
7189 }
7190
e875ecea
N		 7191 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7192 break;
7193
1da177e4
LT		 7194 j += sectors;
7195 if (j>1) mddev->curr_resync = j;
ff4e8d9a 7196 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7197 if (last_check == 0)
e875ecea 7198 /* this is the earliest that rebuild will be
d7603b7e
N		 7199 * visible in /proc/mdstat
7200 */
7201 md_new_event(mddev);
57afd89f
N		 7202
7203 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7204 continue;
7205
57afd89f 7206 last_check = io_sectors;
1da177e4
LT		 7207 repeat:
7208 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7209 /* step marks */
7210 int next = (last_mark+1) % SYNC_MARKS;
7211
7212 mddev->resync_mark = mark[next];
7213 mddev->resync_mark_cnt = mark_cnt[next];
7214 mark[next] = jiffies;
57afd89f 7215 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7216 last_mark = next;
7217 }
7218
7219
c6207277
N		 7220 if (kthread_should_stop())
7221 goto interrupted;
7222
1da177e4
LT		 7223
7224 /*
7225 * this loop exits only if either when we are slower than
7226 * the 'hard' speed limit, or the system was IO-idle for
7227 * a jiffy.
7228 * the system might be non-idle CPU-wise, but we only care
7229 * about not overloading the IO subsystem. (things like an
7230 * e2fsck being done on the RAID array should execute fast)
7231 */
1da177e4
LT		 7232 cond_resched();
7233
57afd89f
N		 7234 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7235 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7236
88202a0c
N		 7237 if (currspeed > speed_min(mddev)) {
7238 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7239 !is_mddev_idle(mddev, 0)) {
c0e48521 7240 msleep(500);
1da177e4
LT		 7241 goto repeat;
7242 }
7243 }
7244 }
61df9d91 7245 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7246 /*
7247 * this also signals 'finished resyncing' to md_stop
7248 */
7249 out:
1da177e4
LT		 7250 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7251
7252 /* tell personality that we are finished */
57afd89f 7253 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7254
dfc70645 7255 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7256 mddev->curr_resync > 2) {
7257 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7258 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7259 if (mddev->curr_resync >= mddev->recovery_cp) {
7260 printk(KERN_INFO
61df9d91
N		 7261 "md: checkpointing %s of %s.\n",
7262 desc, mdname(mddev));
5fd6c1dc
N		 7263 mddev->recovery_cp = mddev->curr_resync;
7264 }
7265 } else
7266 mddev->recovery_cp = MaxSector;
7267 } else {
7268 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7269 mddev->curr_resync = MaxSector;
4e59ca7d
DW		 7270 rcu_read_lock();
7271 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc 7272 if (rdev->raid_disk >= 0 &&
70fffd0b 7273 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7274 !test_bit(Faulty, &rdev->flags) &&
7275 !test_bit(In_sync, &rdev->flags) &&
7276 rdev->recovery_offset < mddev->curr_resync)
7277 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7278 rcu_read_unlock();
5fd6c1dc 7279 }
1da177e4 7280 }
17571284 7281 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7282
1da177e4 7283 skip:
c07b70ad
N		 7284 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7285 /* We completed so min/max setting can be forgotten if used. */
7286 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7287 mddev->resync_min = 0;
7288 mddev->resync_max = MaxSector;
7289 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7290 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7291 mddev->curr_resync = 0;
7292 wake_up(&resync_wait);
7293 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7294 md_wakeup_thread(mddev->thread);
c6207277
N		 7295 return;
7296
7297 interrupted:
7298 /*
7299 * got a signal, exit.
7300 */
7301 printk(KERN_INFO
7302 "md: md_do_sync() got signal ... exiting\n");
7303 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7304 goto out;
7305
1da177e4 7306}
29269553 7307EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7308
fd01b88c 7309static int remove_and_add_spares(struct mddev *mddev)
b4c4c7b8 7310{
3cb03002 7311 struct md_rdev *rdev;
b4c4c7b8
N		 7312 int spares = 0;
7313
97e4f42d
N		 7314 mddev->curr_resync_completed = 0;
7315
159ec1fc 7316 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 7317 if (rdev->raid_disk >= 0 &&
6bfe0b49 7318 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7319 (test_bit(Faulty, &rdev->flags) ||
7320 ! test_bit(In_sync, &rdev->flags)) &&
7321 atomic_read(&rdev->nr_pending)==0) {
7322 if (mddev->pers->hot_remove_disk(
7323 mddev, rdev->raid_disk)==0) {
36fad858 7324 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8
N		 7325 rdev->raid_disk = -1;
7326 }
7327 }
7328
5389042f 7329 if (mddev->degraded) {
159ec1fc 7330 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7331 if (rdev->raid_disk >= 0 &&
e5427135 7332 !test_bit(In_sync, &rdev->flags) &&
de393cde 7333 !test_bit(Faulty, &rdev->flags))
dfc70645 7334 spares++;
b4c4c7b8
N		 7335 if (rdev->raid_disk < 0
7336 && !test_bit(Faulty, &rdev->flags)) {
7337 rdev->recovery_offset = 0;
199050ea
NB		 7338 if (mddev->pers->
7339 hot_add_disk(mddev, rdev) == 0) {
36fad858 7340 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 7341 /* failure here is OK */;
b4c4c7b8
N		 7342 spares++;
7343 md_new_event(mddev);
93be75ff 7344 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7345 } else
7346 break;
7347 }
dfc70645 7348 }
b4c4c7b8
N		 7349 }
7350 return spares;
7351}
7ebc0be7 7352
fd01b88c 7353static void reap_sync_thread(struct mddev *mddev)
7ebc0be7 7354{
3cb03002 7355 struct md_rdev *rdev;
7ebc0be7
N		 7356
7357 /* resync has finished, collect result */
01f96c0a 7358 md_unregister_thread(&mddev->sync_thread);
7ebc0be7
N		 7359 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7360 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7361 /* success...*/
7362 /* activate any spares */
7363 if (mddev->pers->spare_active(mddev))
7364 sysfs_notify(&mddev->kobj, NULL,
7365 "degraded");
7366 }
7367 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7368 mddev->pers->finish_reshape)
7369 mddev->pers->finish_reshape(mddev);
7ebc0be7 7370
d70ed2e4
AW		 7371 /* If array is no-longer degraded, then any saved_raid_disk
7372 * information must be scrapped. Also if any device is now
7373 * In_sync we must scrape the saved_raid_disk for that device
7374 * do the superblock for an incrementally recovered device
7375 * written out.
7ebc0be7 7376 */
d70ed2e4
AW		 7377 list_for_each_entry(rdev, &mddev->disks, same_set)
7378 if (!mddev->degraded ||
7379 test_bit(In_sync, &rdev->flags))
7ebc0be7
N		 7380 rdev->saved_raid_disk = -1;
7381
d70ed2e4 7382 md_update_sb(mddev, 1);
7ebc0be7
N		 7383 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7384 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7385 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7386 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7387 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7388 /* flag recovery needed just to double check */
7389 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7390 sysfs_notify_dirent_safe(mddev->sysfs_action);
7391 md_new_event(mddev);
768e587e
JB		 7392 if (mddev->event_work.func)
7393 queue_work(md_misc_wq, &mddev->event_work);
7ebc0be7
N		 7394}
7395
1da177e4
LT		 7396/*
7397 * This routine is regularly called by all per-raid-array threads to
7398 * deal with generic issues like resync and super-block update.
7399 * Raid personalities that don't have a thread (linear/raid0) do not
7400 * need this as they never do any recovery or update the superblock.
7401 *
7402 * It does not do any resync itself, but rather "forks" off other threads
7403 * to do that as needed.
7404 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7405 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7406 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7407 * and wakeups up this thread which will reap the thread and finish up.
7408 * This thread also removes any faulty devices (with nr_pending == 0).
7409 *
7410 * The overall approach is:
7411 * 1/ if the superblock needs updating, update it.
7412 * 2/ If a recovery thread is running, don't do anything else.
7413 * 3/ If recovery has finished, clean up, possibly marking spares active.
7414 * 4/ If there are any faulty devices, remove them.
7415 * 5/ If array is degraded, try to add spares devices
7416 * 6/ If array has spares or is not in-sync, start a resync thread.
7417 */
fd01b88c 7418void md_check_recovery(struct mddev *mddev)
1da177e4 7419{
68866e42
JB		 7420 if (mddev->suspended)
7421 return;
7422
5f40402d 7423 if (mddev->bitmap)
aa5cbd10 7424 bitmap_daemon_work(mddev);
1da177e4 7425
fca4d848 7426 if (signal_pending(current)) {
31a59e34 7427 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7428 printk(KERN_INFO "md: %s in immediate safe mode\n",
7429 mdname(mddev));
7430 mddev->safemode = 2;
7431 }
7432 flush_signals(current);
7433 }
7434
c89a8eee
N		 7435 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7436 return;
1da177e4 7437 if ( ! (
126925c0 7438 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7439 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7440 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7441 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7442 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7443 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7444 ))
7445 return;
fca4d848 7446
df5b89b3 7447 if (mddev_trylock(mddev)) {
b4c4c7b8 7448 int spares = 0;
fca4d848 7449
c89a8eee
N		 7450 if (mddev->ro) {
7451 /* Only thing we do on a ro array is remove
7452 * failed devices.
7453 */
3cb03002 7454 struct md_rdev *rdev;
a8c42c7f
N		 7455 list_for_each_entry(rdev, &mddev->disks, same_set)
7456 if (rdev->raid_disk >= 0 &&
7457 !test_bit(Blocked, &rdev->flags) &&
7458 test_bit(Faulty, &rdev->flags) &&
7459 atomic_read(&rdev->nr_pending)==0) {
7460 if (mddev->pers->hot_remove_disk(
7461 mddev, rdev->raid_disk)==0) {
36fad858 7462 sysfs_unlink_rdev(mddev, rdev);
a8c42c7f
N		 7463 rdev->raid_disk = -1;
7464 }
7465 }
c89a8eee
N		 7466 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7467 goto unlock;
7468 }
7469
31a59e34 7470 if (!mddev->external) {
0fd62b86 7471 int did_change = 0;
31a59e34
N		 7472 spin_lock_irq(&mddev->write_lock);
7473 if (mddev->safemode &&
7474 !atomic_read(&mddev->writes_pending) &&
7475 !mddev->in_sync &&
7476 mddev->recovery_cp == MaxSector) {
7477 mddev->in_sync = 1;
0fd62b86 7478 did_change = 1;
070dc6dd 7479 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7480 }
7481 if (mddev->safemode == 1)
7482 mddev->safemode = 0;
7483 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7484 if (did_change)
00bcb4ac 7485 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7486 }
fca4d848 7487
850b2b42
N		 7488 if (mddev->flags)
7489 md_update_sb(mddev, 0);
06d91a5f 7490
1da177e4
LT		 7491 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7492 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7493 /* resync/recovery still happening */
7494 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7495 goto unlock;
7496 }
7497 if (mddev->sync_thread) {
7ebc0be7 7498 reap_sync_thread(mddev);
1da177e4
LT		 7499 goto unlock;
7500 }
72a23c21
NB		 7501 /* Set RUNNING before clearing NEEDED to avoid
7502 * any transients in the value of "sync_action".
7503 */
7504 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7505 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7506 /* Clear some bits that don't mean anything, but
7507 * might be left set
7508 */
24dd469d
N		 7509 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7510 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7511
5fd6c1dc
N		 7512 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7513 goto unlock;
1da177e4
LT		 7514 /* no recovery is running.
7515 * remove any failed drives, then
7516 * add spares if possible.
7517 * Spare are also removed and re-added, to allow
7518 * the personality to fail the re-add.
7519 */
1da177e4 7520
b4c4c7b8 7521 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7522 if (mddev->pers->check_reshape == NULL ||
7523 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7524 /* Cannot proceed */
7525 goto unlock;
7526 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7527 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7528 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7529 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7530 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7531 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7532 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7533 } else if (mddev->recovery_cp < MaxSector) {
7534 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7535 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7536 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7537 /* nothing to be done ... */
1da177e4 7538 goto unlock;
24dd469d 7539
1da177e4 7540 if (mddev->pers->sync_request) {
a654b9d8
N		 7541 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7542 /* We are adding a device or devices to an array
7543 * which has the bitmap stored on all devices.
7544 * So make sure all bitmap pages get written
7545 */
7546 bitmap_write_all(mddev->bitmap);
7547 }
1da177e4
LT		 7548 mddev->sync_thread = md_register_thread(md_do_sync,
7549 mddev,
0da3c619 7550 "resync");
1da177e4
LT		 7551 if (!mddev->sync_thread) {
7552 printk(KERN_ERR "%s: could not start resync"
7553 " thread...\n",
7554 mdname(mddev));
7555 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7556 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7557 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7558 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7559 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7560 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7561 } else
1da177e4 7562 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7563 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7564 md_new_event(mddev);
1da177e4
LT		 7565 }
7566 unlock:
72a23c21
NB		 7567 if (!mddev->sync_thread) {
7568 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7569 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7570 &mddev->recovery))
0c3573f1 7571 if (mddev->sysfs_action)
00bcb4ac 7572 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7573 }
1da177e4
LT		 7574 mddev_unlock(mddev);
7575 }
7576}
7577
fd01b88c 7578void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7579{
00bcb4ac 7580 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7581 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7582 !test_bit(Blocked, &rdev->flags) &&
7583 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7584 msecs_to_jiffies(5000));
7585 rdev_dec_pending(rdev, mddev);
7586}
7587EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7588
2230dfe4
N		 7589
7590/* Bad block management.
7591 * We can record which blocks on each device are 'bad' and so just
7592 * fail those blocks, or that stripe, rather than the whole device.
7593 * Entries in the bad-block table are 64bits wide. This comprises:
7594 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7595 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7596 * A 'shift' can be set so that larger blocks are tracked and
7597 * consequently larger devices can be covered.
7598 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7599 *
7600 * Locking of the bad-block table uses a seqlock so md_is_badblock
7601 * might need to retry if it is very unlucky.
7602 * We will sometimes want to check for bad blocks in a bi_end_io function,
7603 * so we use the write_seqlock_irq variant.
7604 *
7605 * When looking for a bad block we specify a range and want to
7606 * know if any block in the range is bad. So we binary-search
7607 * to the last range that starts at-or-before the given endpoint,
7608 * (or "before the sector after the target range")
7609 * then see if it ends after the given start.
7610 * We return
7611 * 0 if there are no known bad blocks in the range
7612 * 1 if there are known bad block which are all acknowledged
7613 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7614 * plus the start/length of the first bad section we overlap.
7615 */
7616int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7617 sector_t *first_bad, int *bad_sectors)
7618{
7619 int hi;
7620 int lo = 0;
7621 u64 *p = bb->page;
7622 int rv = 0;
7623 sector_t target = s + sectors;
7624 unsigned seq;
7625
7626 if (bb->shift > 0) {
7627 /* round the start down, and the end up */
7628 s >>= bb->shift;
7629 target += (1<<bb->shift) - 1;
7630 target >>= bb->shift;
7631 sectors = target - s;
7632 }
7633 /* 'target' is now the first block after the bad range */
7634
7635retry:
7636 seq = read_seqbegin(&bb->lock);
7637
7638 hi = bb->count;
7639
7640 /* Binary search between lo and hi for 'target'
7641 * i.e. for the last range that starts before 'target'
7642 */
7643 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7644 * are known not to be the last range before target.
7645 * VARIANT: hi-lo is the number of possible
7646 * ranges, and decreases until it reaches 1
7647 */
7648 while (hi - lo > 1) {
7649 int mid = (lo + hi) / 2;
7650 sector_t a = BB_OFFSET(p[mid]);
7651 if (a < target)
7652 /* This could still be the one, earlier ranges
7653 * could not. */
7654 lo = mid;
7655 else
7656 /* This and later ranges are definitely out. */
7657 hi = mid;
7658 }
7659 /* 'lo' might be the last that started before target, but 'hi' isn't */
7660 if (hi > lo) {
7661 /* need to check all range that end after 's' to see if
7662 * any are unacknowledged.
7663 */
7664 while (lo >= 0 &&
7665 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7666 if (BB_OFFSET(p[lo]) < target) {
7667 /* starts before the end, and finishes after
7668 * the start, so they must overlap
7669 */
7670 if (rv != -1 && BB_ACK(p[lo]))
7671 rv = 1;
7672 else
7673 rv = -1;
7674 *first_bad = BB_OFFSET(p[lo]);
7675 *bad_sectors = BB_LEN(p[lo]);
7676 }
7677 lo--;
7678 }
7679 }
7680
7681 if (read_seqretry(&bb->lock, seq))
7682 goto retry;
7683
7684 return rv;
7685}
7686EXPORT_SYMBOL_GPL(md_is_badblock);
7687
7688/*
7689 * Add a range of bad blocks to the table.
7690 * This might extend the table, or might contract it
7691 * if two adjacent ranges can be merged.
7692 * We binary-search to find the 'insertion' point, then
7693 * decide how best to handle it.
7694 */
7695static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7696 int acknowledged)
7697{
7698 u64 *p;
7699 int lo, hi;
7700 int rv = 1;
7701
7702 if (bb->shift < 0)
7703 /* badblocks are disabled */
7704 return 0;
7705
7706 if (bb->shift) {
7707 /* round the start down, and the end up */
7708 sector_t next = s + sectors;
7709 s >>= bb->shift;
7710 next += (1<<bb->shift) - 1;
7711 next >>= bb->shift;
7712 sectors = next - s;
7713 }
7714
7715 write_seqlock_irq(&bb->lock);
7716
7717 p = bb->page;
7718 lo = 0;
7719 hi = bb->count;
7720 /* Find the last range that starts at-or-before 's' */
7721 while (hi - lo > 1) {
7722 int mid = (lo + hi) / 2;
7723 sector_t a = BB_OFFSET(p[mid]);
7724 if (a <= s)
7725 lo = mid;
7726 else
7727 hi = mid;
7728 }
7729 if (hi > lo && BB_OFFSET(p[lo]) > s)
7730 hi = lo;
7731
7732 if (hi > lo) {
7733 /* we found a range that might merge with the start
7734 * of our new range
7735 */
7736 sector_t a = BB_OFFSET(p[lo]);
7737 sector_t e = a + BB_LEN(p[lo]);
7738 int ack = BB_ACK(p[lo]);
7739 if (e >= s) {
7740 /* Yes, we can merge with a previous range */
7741 if (s == a && s + sectors >= e)
7742 /* new range covers old */
7743 ack = acknowledged;
7744 else
7745 ack = ack && acknowledged;
7746
7747 if (e < s + sectors)
7748 e = s + sectors;
7749 if (e - a <= BB_MAX_LEN) {
7750 p[lo] = BB_MAKE(a, e-a, ack);
7751 s = e;
7752 } else {
7753 /* does not all fit in one range,
7754 * make p[lo] maximal
7755 */
7756 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7757 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7758 s = a + BB_MAX_LEN;
7759 }
7760 sectors = e - s;
7761 }
7762 }
7763 if (sectors && hi < bb->count) {
7764 /* 'hi' points to the first range that starts after 's'.
7765 * Maybe we can merge with the start of that range */
7766 sector_t a = BB_OFFSET(p[hi]);
7767 sector_t e = a + BB_LEN(p[hi]);
7768 int ack = BB_ACK(p[hi]);
7769 if (a <= s + sectors) {
7770 /* merging is possible */
7771 if (e <= s + sectors) {
7772 /* full overlap */
7773 e = s + sectors;
7774 ack = acknowledged;
7775 } else
7776 ack = ack && acknowledged;
7777
7778 a = s;
7779 if (e - a <= BB_MAX_LEN) {
7780 p[hi] = BB_MAKE(a, e-a, ack);
7781 s = e;
7782 } else {
7783 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7784 s = a + BB_MAX_LEN;
7785 }
7786 sectors = e - s;
7787 lo = hi;
7788 hi++;
7789 }
7790 }
7791 if (sectors == 0 && hi < bb->count) {
7792 /* we might be able to combine lo and hi */
7793 /* Note: 's' is at the end of 'lo' */
7794 sector_t a = BB_OFFSET(p[hi]);
7795 int lolen = BB_LEN(p[lo]);
7796 int hilen = BB_LEN(p[hi]);
7797 int newlen = lolen + hilen - (s - a);
7798 if (s >= a && newlen < BB_MAX_LEN) {
7799 /* yes, we can combine them */
7800 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7801 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7802 memmove(p + hi, p + hi + 1,
7803 (bb->count - hi - 1) * 8);
7804 bb->count--;
7805 }
7806 }
7807 while (sectors) {
7808 /* didn't merge (it all).
7809 * Need to add a range just before 'hi' */
7810 if (bb->count >= MD_MAX_BADBLOCKS) {
7811 /* No room for more */
7812 rv = 0;
7813 break;
7814 } else {
7815 int this_sectors = sectors;
7816 memmove(p + hi + 1, p + hi,
7817 (bb->count - hi) * 8);
7818 bb->count++;
7819
7820 if (this_sectors > BB_MAX_LEN)
7821 this_sectors = BB_MAX_LEN;
7822 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7823 sectors -= this_sectors;
7824 s += this_sectors;
7825 }
7826 }
7827
7828 bb->changed = 1;
de393cde
N		 7829 if (!acknowledged)
7830 bb->unacked_exist = 1;
2230dfe4
N		 7831 write_sequnlock_irq(&bb->lock);
7832
7833 return rv;
7834}
7835
3cb03002 7836int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
2230dfe4
N		 7837 int acknowledged)
7838{
7839 int rv = md_set_badblocks(&rdev->badblocks,
7840 s + rdev->data_offset, sectors, acknowledged);
7841 if (rv) {
7842 /* Make sure they get written out promptly */
7843 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7844 md_wakeup_thread(rdev->mddev->thread);
7845 }
7846 return rv;
7847}
7848EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7849
7850/*
7851 * Remove a range of bad blocks from the table.
7852 * This may involve extending the table if we spilt a region,
7853 * but it must not fail. So if the table becomes full, we just
7854 * drop the remove request.
7855 */
7856static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7857{
7858 u64 *p;
7859 int lo, hi;
7860 sector_t target = s + sectors;
7861 int rv = 0;
7862
7863 if (bb->shift > 0) {
7864 /* When clearing we round the start up and the end down.
7865 * This should not matter as the shift should align with
7866 * the block size and no rounding should ever be needed.
7867 * However it is better the think a block is bad when it
7868 * isn't than to think a block is not bad when it is.
7869 */
7870 s += (1<<bb->shift) - 1;
7871 s >>= bb->shift;
7872 target >>= bb->shift;
7873 sectors = target - s;
7874 }
7875
7876 write_seqlock_irq(&bb->lock);
7877
7878 p = bb->page;
7879 lo = 0;
7880 hi = bb->count;
7881 /* Find the last range that starts before 'target' */
7882 while (hi - lo > 1) {
7883 int mid = (lo + hi) / 2;
7884 sector_t a = BB_OFFSET(p[mid]);
7885 if (a < target)
7886 lo = mid;
7887 else
7888 hi = mid;
7889 }
7890 if (hi > lo) {
7891 /* p[lo] is the last range that could overlap the
7892 * current range. Earlier ranges could also overlap,
7893 * but only this one can overlap the end of the range.
7894 */
7895 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7896 /* Partial overlap, leave the tail of this range */
7897 int ack = BB_ACK(p[lo]);
7898 sector_t a = BB_OFFSET(p[lo]);
7899 sector_t end = a + BB_LEN(p[lo]);
7900
7901 if (a < s) {
7902 /* we need to split this range */
7903 if (bb->count >= MD_MAX_BADBLOCKS) {
7904 rv = 0;
7905 goto out;
7906 }
7907 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7908 bb->count++;
7909 p[lo] = BB_MAKE(a, s-a, ack);
7910 lo++;
7911 }
7912 p[lo] = BB_MAKE(target, end - target, ack);
7913 /* there is no longer an overlap */
7914 hi = lo;
7915 lo--;
7916 }
7917 while (lo >= 0 &&
7918 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7919 /* This range does overlap */
7920 if (BB_OFFSET(p[lo]) < s) {
7921 /* Keep the early parts of this range. */
7922 int ack = BB_ACK(p[lo]);
7923 sector_t start = BB_OFFSET(p[lo]);
7924 p[lo] = BB_MAKE(start, s - start, ack);
7925 /* now low doesn't overlap, so.. */
7926 break;
7927 }
7928 lo--;
7929 }
7930 /* 'lo' is strictly before, 'hi' is strictly after,
7931 * anything between needs to be discarded
7932 */
7933 if (hi - lo > 1) {
7934 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
7935 bb->count -= (hi - lo - 1);
7936 }
7937 }
7938
7939 bb->changed = 1;
7940out:
7941 write_sequnlock_irq(&bb->lock);
7942 return rv;
7943}
7944
3cb03002 7945int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
2230dfe4
N		 7946{
7947 return md_clear_badblocks(&rdev->badblocks,
7948 s + rdev->data_offset,
7949 sectors);
7950}
7951EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
7952
7953/*
7954 * Acknowledge all bad blocks in a list.
7955 * This only succeeds if ->changed is clear. It is used by
7956 * in-kernel metadata updates
7957 */
7958void md_ack_all_badblocks(struct badblocks *bb)
7959{
7960 if (bb->page == NULL || bb->changed)
7961 /* no point even trying */
7962 return;
7963 write_seqlock_irq(&bb->lock);
7964
7965 if (bb->changed == 0) {
7966 u64 *p = bb->page;
7967 int i;
7968 for (i = 0; i < bb->count ; i++) {
7969 if (!BB_ACK(p[i])) {
7970 sector_t start = BB_OFFSET(p[i]);
7971 int len = BB_LEN(p[i]);
7972 p[i] = BB_MAKE(start, len, 1);
7973 }
7974 }
de393cde 7975 bb->unacked_exist = 0;
2230dfe4
N		 7976 }
7977 write_sequnlock_irq(&bb->lock);
7978}
7979EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
7980
16c791a5
N		 7981/* sysfs access to bad-blocks list.
7982 * We present two files.
7983 * 'bad-blocks' lists sector numbers and lengths of ranges that
7984 * are recorded as bad. The list is truncated to fit within
7985 * the one-page limit of sysfs.
7986 * Writing "sector length" to this file adds an acknowledged
7987 * bad block list.
7988 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7989 * been acknowledged. Writing to this file adds bad blocks
7990 * without acknowledging them. This is largely for testing.
7991 */
7992
7993static ssize_t
7994badblocks_show(struct badblocks *bb, char *page, int unack)
7995{
7996 size_t len;
7997 int i;
7998 u64 *p = bb->page;
7999 unsigned seq;
8000
8001 if (bb->shift < 0)
8002 return 0;
8003
8004retry:
8005 seq = read_seqbegin(&bb->lock);
8006
8007 len = 0;
8008 i = 0;
8009
8010 while (len < PAGE_SIZE && i < bb->count) {
8011 sector_t s = BB_OFFSET(p[i]);
8012 unsigned int length = BB_LEN(p[i]);
8013 int ack = BB_ACK(p[i]);
8014 i++;
8015
8016 if (unack && ack)
8017 continue;
8018
8019 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8020 (unsigned long long)s << bb->shift,
8021 length << bb->shift);
8022 }
de393cde
N		 8023 if (unack && len == 0)
8024 bb->unacked_exist = 0;
16c791a5
N		 8025
8026 if (read_seqretry(&bb->lock, seq))
8027 goto retry;
8028
8029 return len;
8030}
8031
8032#define DO_DEBUG 1
8033
8034static ssize_t
8035badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8036{
8037 unsigned long long sector;
8038 int length;
8039 char newline;
8040#ifdef DO_DEBUG
8041 /* Allow clearing via sysfs *only* for testing/debugging.
8042 * Normally only a successful write may clear a badblock
8043 */
8044 int clear = 0;
8045 if (page[0] == '-') {
8046 clear = 1;
8047 page++;
8048 }
8049#endif /* DO_DEBUG */
8050
8051 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8052 case 3:
8053 if (newline != '\n')
8054 return -EINVAL;
8055 case 2:
8056 if (length <= 0)
8057 return -EINVAL;
8058 break;
8059 default:
8060 return -EINVAL;
8061 }
8062
8063#ifdef DO_DEBUG
8064 if (clear) {
8065 md_clear_badblocks(bb, sector, length);
8066 return len;
8067 }
8068#endif /* DO_DEBUG */
8069 if (md_set_badblocks(bb, sector, length, !unack))
8070 return len;
8071 else
8072 return -ENOSPC;
8073}
8074
75c96f85
AB		 8075static int md_notify_reboot(struct notifier_block *this,
8076 unsigned long code, void *x)
1da177e4
LT		 8077{
8078 struct list_head *tmp;
fd01b88c 8079 struct mddev *mddev;
2dba6a91 8080 int need_delay = 0;
1da177e4
LT		 8081
8082 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
8083
8084 printk(KERN_INFO "md: stopping all md devices.\n");
8085
2dba6a91 8086 for_each_mddev(mddev, tmp) {
c71d4887 8087 if (mddev_trylock(mddev)) {
2b25000b
N		 8088 /* Force a switch to readonly even array
8089 * appears to still be in use. Hence
8090 * the '100'.
8091 */
a4bd82d0 8092 md_set_readonly(mddev, 100);
c71d4887
NB		 8093 mddev_unlock(mddev);
8094 }
2dba6a91
DB		 8095 need_delay = 1;
8096 }
1da177e4
LT		 8097 /*
8098 * certain more exotic SCSI devices are known to be
8099 * volatile wrt too early system reboots. While the
8100 * right place to handle this issue is the given
8101 * driver, we do want to have a safe RAID driver ...
8102 */
2dba6a91
DB		 8103 if (need_delay)
8104 mdelay(1000*1);
1da177e4
LT		 8105 }
8106 return NOTIFY_DONE;
8107}
8108
75c96f85 8109static struct notifier_block md_notifier = {
1da177e4
LT		 8110 .notifier_call = md_notify_reboot,
8111 .next = NULL,
8112 .priority = INT_MAX, /* before any real devices */
8113};
8114
8115static void md_geninit(void)
8116{
36a4e1fe 8117 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8118
c7705f34 8119 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8120}
8121
75c96f85 8122static int __init md_init(void)
1da177e4 8123{
e804ac78
TH		 8124 int ret = -ENOMEM;
8125
ada609ee 8126 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8127 if (!md_wq)
8128 goto err_wq;
8129
8130 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8131 if (!md_misc_wq)
8132 goto err_misc_wq;
8133
8134 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8135 goto err_md;
8136
8137 if ((ret = register_blkdev(0, "mdp")) < 0)
8138 goto err_mdp;
8139 mdp_major = ret;
8140
3dbd8c2e 8141 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8142 md_probe, NULL, NULL);
8143 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8144 md_probe, NULL, NULL);
8145
1da177e4 8146 register_reboot_notifier(&md_notifier);
0b4d4147 8147 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8148
8149 md_geninit();
d710e138 8150 return 0;
1da177e4 8151
e804ac78
TH		 8152err_mdp:
8153 unregister_blkdev(MD_MAJOR, "md");
8154err_md:
8155 destroy_workqueue(md_misc_wq);
8156err_misc_wq:
8157 destroy_workqueue(md_wq);
8158err_wq:
8159 return ret;
8160}
1da177e4
LT		 8161
8162#ifndef MODULE
8163
8164/*
8165 * Searches all registered partitions for autorun RAID arrays
8166 * at boot time.
8167 */
4d936ec1
ME		 8168
8169static LIST_HEAD(all_detected_devices);
8170struct detected_devices_node {
8171 struct list_head list;
8172 dev_t dev;
8173};
1da177e4
LT		 8174
8175void md_autodetect_dev(dev_t dev)
8176{
4d936ec1
ME		 8177 struct detected_devices_node *node_detected_dev;
8178
8179 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8180 if (node_detected_dev) {
8181 node_detected_dev->dev = dev;
8182 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8183 } else {
8184 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8185 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8186 }
1da177e4
LT		 8187}
8188
8189
8190static void autostart_arrays(int part)
8191{
3cb03002 8192 struct md_rdev *rdev;
4d936ec1
ME		 8193 struct detected_devices_node *node_detected_dev;
8194 dev_t dev;
8195 int i_scanned, i_passed;
1da177e4 8196
4d936ec1
ME		 8197 i_scanned = 0;
8198 i_passed = 0;
1da177e4 8199
4d936ec1 8200 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8201
4d936ec1
ME		 8202 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8203 i_scanned++;
8204 node_detected_dev = list_entry(all_detected_devices.next,
8205 struct detected_devices_node, list);
8206 list_del(&node_detected_dev->list);
8207 dev = node_detected_dev->dev;
8208 kfree(node_detected_dev);
df968c4e 8209 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8210 if (IS_ERR(rdev))
8211 continue;
8212
b2d444d7 8213 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8214 MD_BUG();
8215 continue;
8216 }
d0fae18f 8217 set_bit(AutoDetected, &rdev->flags);
1da177e4 8218 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8219 i_passed++;
1da177e4 8220 }
4d936ec1
ME		 8221
8222 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8223 i_scanned, i_passed);
1da177e4
LT		 8224
8225 autorun_devices(part);
8226}
8227
fdee8ae4 8228#endif /* !MODULE */
1da177e4
LT		 8229
8230static __exit void md_exit(void)
8231{
fd01b88c 8232 struct mddev *mddev;
1da177e4 8233 struct list_head *tmp;
8ab5e4c1 8234
3dbd8c2e 8235 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8236 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8237
3dbd8c2e 8238 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8239 unregister_blkdev(mdp_major, "mdp");
8240 unregister_reboot_notifier(&md_notifier);
8241 unregister_sysctl_table(raid_table_header);
8242 remove_proc_entry("mdstat", NULL);
29ac4aa3 8243 for_each_mddev(mddev, tmp) {
1da177e4 8244 export_array(mddev);
d3374825 8245 mddev->hold_active = 0;
1da177e4 8246 }
e804ac78
TH		 8247 destroy_workqueue(md_misc_wq);
8248 destroy_workqueue(md_wq);
1da177e4
LT		 8249}
8250
685784aa 8251subsys_initcall(md_init);
1da177e4
LT		 8252module_exit(md_exit)
8253
f91de92e
N		 8254static int get_ro(char *buffer, struct kernel_param *kp)
8255{
8256 return sprintf(buffer, "%d", start_readonly);
8257}
8258static int set_ro(const char *val, struct kernel_param *kp)
8259{
8260 char *e;
8261 int num = simple_strtoul(val, &e, 10);
8262 if (*val && (*e == '\0' || *e == '\n')) {
8263 start_readonly = num;
4dbcdc75 8264 return 0;
f91de92e
N		 8265 }
8266 return -EINVAL;
8267}
8268
80ca3a44
N		 8269module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8270module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8271
efeb53c0 8272module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8273
1da177e4
LT		 8274EXPORT_SYMBOL(register_md_personality);
8275EXPORT_SYMBOL(unregister_md_personality);
8276EXPORT_SYMBOL(md_error);
8277EXPORT_SYMBOL(md_done_sync);
8278EXPORT_SYMBOL(md_write_start);
8279EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8280EXPORT_SYMBOL(md_register_thread);
8281EXPORT_SYMBOL(md_unregister_thread);
8282EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 8283EXPORT_SYMBOL(md_check_recovery);
8284MODULE_LICENSE("GPL");
0efb9e61 8285MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8286MODULE_ALIAS("md");
72008652 8287MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux kernel - Deliverables
This page took 1.424967 seconds and 5 git commands to generate.