projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
fs: move code out of buffer.c
[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>
ff01bb48 39#include <linux/fs.h>
d7603b7e 40#include <linux/poll.h>
16f17b39 41#include <linux/ctype.h>
e7d2860b 42#include <linux/string.h>
fb4d8c76
N		 43#include <linux/hdreg.h>
44#include <linux/proc_fs.h>
45#include <linux/random.h>
056075c7 46#include <linux/module.h>
fb4d8c76 47#include <linux/reboot.h>
32a7627c 48#include <linux/file.h>
aa98aa31 49#include <linux/compat.h>
25570727 50#include <linux/delay.h>
bff61975
N		 51#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
ef740c37 55#include "bitmap.h"
1da177e4 56
1da177e4 57#ifndef MODULE
d710e138 58static void autostart_arrays(int part);
1da177e4
LT		 59#endif
60
01f96c0a
N		 61/* pers_list is a list of registered personalities protected
62 * by pers_lock.
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
65 */
2604b703 66static LIST_HEAD(pers_list);
1da177e4
LT		 67static DEFINE_SPINLOCK(pers_lock);
68
5e56341d
AB		 69static void md_print_devices(void);
70
90b08710 71static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 72static struct workqueue_struct *md_wq;
73static struct workqueue_struct *md_misc_wq;
90b08710 74
5e56341d
AB		 75#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76
1e50915f
RB		 77/*
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
81 */
82#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 83/*
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 87 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
90 * idle IO detection.
91 *
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 93 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 94 */
95
96static int sysctl_speed_limit_min = 1000;
97static int sysctl_speed_limit_max = 200000;
fd01b88c 98static inline int speed_min(struct mddev *mddev)
88202a0c
N		 99{
100 return mddev->sync_speed_min ?
101 mddev->sync_speed_min : sysctl_speed_limit_min;
102}
103
fd01b88c 104static inline int speed_max(struct mddev *mddev)
88202a0c
N		 105{
106 return mddev->sync_speed_max ?
107 mddev->sync_speed_max : sysctl_speed_limit_max;
108}
1da177e4
LT		 109
110static struct ctl_table_header *raid_table_header;
111
112static ctl_table raid_table[] = {
113 {
1da177e4
LT		 114 .procname = "speed_limit_min",
115 .data = &sysctl_speed_limit_min,
116 .maxlen = sizeof(int),
80ca3a44 117 .mode = S_IRUGO|S_IWUSR,
6d456111 118 .proc_handler = proc_dointvec,
1da177e4
LT		 119 },
120 {
1da177e4
LT		 121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
80ca3a44 124 .mode = S_IRUGO|S_IWUSR,
6d456111 125 .proc_handler = proc_dointvec,
1da177e4 126 },
894d2491 127 { }
1da177e4
LT		 128};
129
130static ctl_table raid_dir_table[] = {
131 {
1da177e4
LT		 132 .procname = "raid",
133 .maxlen = 0,
80ca3a44 134 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 135 .child = raid_table,
136 },
894d2491 137 { }
1da177e4
LT		 138};
139
140static ctl_table raid_root_table[] = {
141 {
1da177e4
LT		 142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
146 },
894d2491 147 { }
1da177e4
LT		 148};
149
83d5cde4 150static const struct block_device_operations md_fops;
1da177e4 151
f91de92e
N		 152static int start_readonly;
153
a167f663
N		 154/* bio_clone_mddev
155 * like bio_clone, but with a local bio set
156 */
157
158static void mddev_bio_destructor(struct bio *bio)
159{
fd01b88c 160 struct mddev *mddev, **mddevp;
a167f663
N		 161
162 mddevp = (void*)bio;
163 mddev = mddevp[-1];
164
165 bio_free(bio, mddev->bio_set);
166}
167
168struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 169 struct mddev *mddev)
a167f663
N		 170{
171 struct bio *b;
fd01b88c 172 struct mddev **mddevp;
a167f663
N		 173
174 if (!mddev || !mddev->bio_set)
175 return bio_alloc(gfp_mask, nr_iovecs);
176
177 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
178 mddev->bio_set);
179 if (!b)
180 return NULL;
181 mddevp = (void*)b;
182 mddevp[-1] = mddev;
183 b->bi_destructor = mddev_bio_destructor;
184 return b;
185}
186EXPORT_SYMBOL_GPL(bio_alloc_mddev);
187
188struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 189 struct mddev *mddev)
a167f663
N		 190{
191 struct bio *b;
fd01b88c 192 struct mddev **mddevp;
a167f663
N		 193
194 if (!mddev || !mddev->bio_set)
195 return bio_clone(bio, gfp_mask);
196
197 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
198 mddev->bio_set);
199 if (!b)
200 return NULL;
201 mddevp = (void*)b;
202 mddevp[-1] = mddev;
203 b->bi_destructor = mddev_bio_destructor;
204 __bio_clone(b, bio);
205 if (bio_integrity(bio)) {
206 int ret;
207
208 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
209
210 if (ret < 0) {
211 bio_put(b);
212 return NULL;
213 }
214 }
215
216 return b;
217}
218EXPORT_SYMBOL_GPL(bio_clone_mddev);
219
d2eb35ac
N		 220void md_trim_bio(struct bio *bio, int offset, int size)
221{
222 /* 'bio' is a cloned bio which we need to trim to match
223 * the given offset and size.
224 * This requires adjusting bi_sector, bi_size, and bi_io_vec
225 */
226 int i;
227 struct bio_vec *bvec;
228 int sofar = 0;
229
230 size <<= 9;
231 if (offset == 0 && size == bio->bi_size)
232 return;
233
234 bio->bi_sector += offset;
235 bio->bi_size = size;
236 offset <<= 9;
237 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
238
239 while (bio->bi_idx < bio->bi_vcnt &&
240 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
241 /* remove this whole bio_vec */
242 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
243 bio->bi_idx++;
244 }
245 if (bio->bi_idx < bio->bi_vcnt) {
246 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
247 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
248 }
249 /* avoid any complications with bi_idx being non-zero*/
250 if (bio->bi_idx) {
251 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
252 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
253 bio->bi_vcnt -= bio->bi_idx;
254 bio->bi_idx = 0;
255 }
256 /* Make sure vcnt and last bv are not too big */
257 bio_for_each_segment(bvec, bio, i) {
258 if (sofar + bvec->bv_len > size)
259 bvec->bv_len = size - sofar;
260 if (bvec->bv_len == 0) {
261 bio->bi_vcnt = i;
262 break;
263 }
264 sofar += bvec->bv_len;
265 }
266}
267EXPORT_SYMBOL_GPL(md_trim_bio);
268
d7603b7e
N		 269/*
270 * We have a system wide 'event count' that is incremented
271 * on any 'interesting' event, and readers of /proc/mdstat
272 * can use 'poll' or 'select' to find out when the event
273 * count increases.
274 *
275 * Events are:
276 * start array, stop array, error, add device, remove device,
277 * start build, activate spare
278 */
2989ddbd 279static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 280static atomic_t md_event_count;
fd01b88c 281void md_new_event(struct mddev *mddev)
d7603b7e
N		 282{
283 atomic_inc(&md_event_count);
284 wake_up(&md_event_waiters);
285}
29269553 286EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 287
c331eb04
N		 288/* Alternate version that can be called from interrupts
289 * when calling sysfs_notify isn't needed.
290 */
fd01b88c 291static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 292{
293 atomic_inc(&md_event_count);
294 wake_up(&md_event_waiters);
295}
296
1da177e4
LT		 297/*
298 * Enables to iterate over all existing md arrays
299 * all_mddevs_lock protects this list.
300 */
301static LIST_HEAD(all_mddevs);
302static DEFINE_SPINLOCK(all_mddevs_lock);
303
304
305/*
306 * iterates through all used mddevs in the system.
307 * We take care to grab the all_mddevs_lock whenever navigating
308 * the list, and to always hold a refcount when unlocked.
309 * Any code which breaks out of this loop while own
310 * a reference to the current mddev and must mddev_put it.
311 */
fd01b88c 312#define for_each_mddev(_mddev,_tmp) \
1da177e4
LT		 313 \
314 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 315 _tmp = all_mddevs.next; \
316 _mddev = NULL;}); \
317 ({ if (_tmp != &all_mddevs) \
318 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 319 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 320 if (_mddev) mddev_put(_mddev); \
321 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
322 _tmp != &all_mddevs;}); \
1da177e4 323 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 324 _tmp = _tmp->next;}) \
1da177e4
LT		 325 )
326
327
409c57f3
N		 328/* Rather than calling directly into the personality make_request function,
329 * IO requests come here first so that we can check if the device is
330 * being suspended pending a reconfiguration.
331 * We hold a refcount over the call to ->make_request. By the time that
332 * call has finished, the bio has been linked into some internal structure
333 * and so is visible to ->quiesce(), so we don't need the refcount any more.
334 */
5a7bbad2 335static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 336{
49077326 337 const int rw = bio_data_dir(bio);
fd01b88c 338 struct mddev *mddev = q->queuedata;
49077326 339 int cpu;
e91ece55 340 unsigned int sectors;
49077326 341
0ca69886
N		 342 if (mddev == NULL || mddev->pers == NULL
343 || !mddev->ready) {
409c57f3 344 bio_io_error(bio);
5a7bbad2 345 return;
409c57f3 346 }
0ca69886 347 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 348 rcu_read_lock();
e9c7469b 349 if (mddev->suspended) {
409c57f3
N		 350 DEFINE_WAIT(__wait);
351 for (;;) {
352 prepare_to_wait(&mddev->sb_wait, &__wait,
353 TASK_UNINTERRUPTIBLE);
e9c7469b 354 if (!mddev->suspended)
409c57f3
N		 355 break;
356 rcu_read_unlock();
357 schedule();
358 rcu_read_lock();
359 }
360 finish_wait(&mddev->sb_wait, &__wait);
361 }
362 atomic_inc(&mddev->active_io);
363 rcu_read_unlock();
49077326 364
e91ece55
CM		 365 /*
366 * save the sectors now since our bio can
367 * go away inside make_request
368 */
369 sectors = bio_sectors(bio);
5a7bbad2 370 mddev->pers->make_request(mddev, bio);
49077326
N		 371
372 cpu = part_stat_lock();
373 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 374 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N		 375 part_stat_unlock();
376
409c57f3
N		 377 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
378 wake_up(&mddev->sb_wait);
409c57f3
N		 379}
380
9e35b99c
N		 381/* mddev_suspend makes sure no new requests are submitted
382 * to the device, and that any requests that have been submitted
383 * are completely handled.
384 * Once ->stop is called and completes, the module will be completely
385 * unused.
386 */
fd01b88c 387void mddev_suspend(struct mddev *mddev)
409c57f3
N		 388{
389 BUG_ON(mddev->suspended);
390 mddev->suspended = 1;
391 synchronize_rcu();
392 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
393 mddev->pers->quiesce(mddev, 1);
409c57f3 394}
390ee602 395EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 396
fd01b88c 397void mddev_resume(struct mddev *mddev)
409c57f3
N		 398{
399 mddev->suspended = 0;
400 wake_up(&mddev->sb_wait);
401 mddev->pers->quiesce(mddev, 0);
0fd018af
JB		 402
403 md_wakeup_thread(mddev->thread);
404 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 405}
390ee602 406EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 407
fd01b88c 408int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N		 409{
410 return mddev->suspended;
411}
412EXPORT_SYMBOL(mddev_congested);
413
a2826aa9 414/*
e9c7469b 415 * Generic flush handling for md
a2826aa9
N		 416 */
417
e9c7469b 418static void md_end_flush(struct bio *bio, int err)
a2826aa9 419{
3cb03002 420 struct md_rdev *rdev = bio->bi_private;
fd01b88c 421 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 422
423 rdev_dec_pending(rdev, mddev);
424
425 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 426 /* The pre-request flush has finished */
e804ac78 427 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 428 }
429 bio_put(bio);
430}
431
a7a07e69
N		 432static void md_submit_flush_data(struct work_struct *ws);
433
a035fc3e 434static void submit_flushes(struct work_struct *ws)
a2826aa9 435{
fd01b88c 436 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 437 struct md_rdev *rdev;
a2826aa9 438
a7a07e69
N		 439 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
440 atomic_set(&mddev->flush_pending, 1);
a2826aa9
N		 441 rcu_read_lock();
442 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
443 if (rdev->raid_disk >= 0 &&
444 !test_bit(Faulty, &rdev->flags)) {
445 /* Take two references, one is dropped
446 * when request finishes, one after
447 * we reclaim rcu_read_lock
448 */
449 struct bio *bi;
450 atomic_inc(&rdev->nr_pending);
451 atomic_inc(&rdev->nr_pending);
452 rcu_read_unlock();
a167f663 453 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
e9c7469b 454 bi->bi_end_io = md_end_flush;
a2826aa9
N		 455 bi->bi_private = rdev;
456 bi->bi_bdev = rdev->bdev;
457 atomic_inc(&mddev->flush_pending);
e9c7469b 458 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 459 rcu_read_lock();
460 rdev_dec_pending(rdev, mddev);
461 }
462 rcu_read_unlock();
a7a07e69
N		 463 if (atomic_dec_and_test(&mddev->flush_pending))
464 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 465}
466
e9c7469b 467static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 468{
fd01b88c 469 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 470 struct bio *bio = mddev->flush_bio;
a2826aa9 471
e9c7469b 472 if (bio->bi_size == 0)
a2826aa9
N		 473 /* an empty barrier - all done */
474 bio_endio(bio, 0);
475 else {
e9c7469b 476 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 477 mddev->pers->make_request(mddev, bio);
a2826aa9 478 }
2b74e12e
N		 479
480 mddev->flush_bio = NULL;
481 wake_up(&mddev->sb_wait);
a2826aa9
N		 482}
483
fd01b88c 484void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N		 485{
486 spin_lock_irq(&mddev->write_lock);
487 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 488 !mddev->flush_bio,
a2826aa9 489 mddev->write_lock, /*nothing*/);
e9c7469b 490 mddev->flush_bio = bio;
a2826aa9
N		 491 spin_unlock_irq(&mddev->write_lock);
492
a035fc3e
N		 493 INIT_WORK(&mddev->flush_work, submit_flushes);
494 queue_work(md_wq, &mddev->flush_work);
a2826aa9 495}
e9c7469b 496EXPORT_SYMBOL(md_flush_request);
409c57f3 497
97658cdd
N		 498/* Support for plugging.
499 * This mirrors the plugging support in request_queue, but does not
500 * require having a whole queue or request structures.
501 * We allocate an md_plug_cb for each md device and each thread it gets
502 * plugged on. This links tot the private plug_handle structure in the
503 * personality data where we keep a count of the number of outstanding
504 * plugs so other code can see if a plug is active.
505 */
506struct md_plug_cb {
507 struct blk_plug_cb cb;
fd01b88c 508 struct mddev *mddev;
97658cdd
N		 509};
510
511static void plugger_unplug(struct blk_plug_cb *cb)
512{
513 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
514 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
515 md_wakeup_thread(mdcb->mddev->thread);
516 kfree(mdcb);
517}
518
519/* Check that an unplug wakeup will come shortly.
520 * If not, wakeup the md thread immediately
521 */
fd01b88c 522int mddev_check_plugged(struct mddev *mddev)
97658cdd
N		 523{
524 struct blk_plug *plug = current->plug;
525 struct md_plug_cb *mdcb;
526
527 if (!plug)
528 return 0;
529
530 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
531 if (mdcb->cb.callback == plugger_unplug &&
532 mdcb->mddev == mddev) {
533 /* Already on the list, move to top */
534 if (mdcb != list_first_entry(&plug->cb_list,
535 struct md_plug_cb,
536 cb.list))
537 list_move(&mdcb->cb.list, &plug->cb_list);
538 return 1;
539 }
540 }
541 /* Not currently on the callback list */
542 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
543 if (!mdcb)
544 return 0;
545
546 mdcb->mddev = mddev;
547 mdcb->cb.callback = plugger_unplug;
548 atomic_inc(&mddev->plug_cnt);
549 list_add(&mdcb->cb.list, &plug->cb_list);
550 return 1;
551}
552EXPORT_SYMBOL_GPL(mddev_check_plugged);
2ac87401 553
fd01b88c 554static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 555{
556 atomic_inc(&mddev->active);
557 return mddev;
558}
559
5fd3a17e 560static void mddev_delayed_delete(struct work_struct *ws);
d3374825 561
fd01b88c 562static void mddev_put(struct mddev *mddev)
1da177e4 563{
a167f663
N		 564 struct bio_set *bs = NULL;
565
1da177e4
LT		 566 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
567 return;
d3374825 568 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 569 mddev->ctime == 0 && !mddev->hold_active) {
570 /* Array is not configured at all, and not held active,
571 * so destroy it */
af8a2434 572 list_del_init(&mddev->all_mddevs);
a167f663
N		 573 bs = mddev->bio_set;
574 mddev->bio_set = NULL;
d3374825 575 if (mddev->gendisk) {
e804ac78
TH		 576 /* We did a probe so need to clean up. Call
577 * queue_work inside the spinlock so that
578 * flush_workqueue() after mddev_find will
579 * succeed in waiting for the work to be done.
d3374825
N		 580 */
581 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 582 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 583 } else
584 kfree(mddev);
585 }
586 spin_unlock(&all_mddevs_lock);
a167f663
N		 587 if (bs)
588 bioset_free(bs);
1da177e4
LT		 589}
590
fd01b88c 591void mddev_init(struct mddev *mddev)
fafd7fb0
N		 592{
593 mutex_init(&mddev->open_mutex);
594 mutex_init(&mddev->reconfig_mutex);
595 mutex_init(&mddev->bitmap_info.mutex);
596 INIT_LIST_HEAD(&mddev->disks);
597 INIT_LIST_HEAD(&mddev->all_mddevs);
598 init_timer(&mddev->safemode_timer);
599 atomic_set(&mddev->active, 1);
600 atomic_set(&mddev->openers, 0);
601 atomic_set(&mddev->active_io, 0);
97658cdd 602 atomic_set(&mddev->plug_cnt, 0);
fafd7fb0
N		 603 spin_lock_init(&mddev->write_lock);
604 atomic_set(&mddev->flush_pending, 0);
605 init_waitqueue_head(&mddev->sb_wait);
606 init_waitqueue_head(&mddev->recovery_wait);
607 mddev->reshape_position = MaxSector;
608 mddev->resync_min = 0;
609 mddev->resync_max = MaxSector;
610 mddev->level = LEVEL_NONE;
611}
390ee602 612EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 613
fd01b88c 614static struct mddev * mddev_find(dev_t unit)
1da177e4 615{
fd01b88c 616 struct mddev *mddev, *new = NULL;
1da177e4 617
8f5f02c4
N		 618 if (unit && MAJOR(unit) != MD_MAJOR)
619 unit &= ~((1<<MdpMinorShift)-1);
620
1da177e4
LT		 621 retry:
622 spin_lock(&all_mddevs_lock);
efeb53c0
N		 623
624 if (unit) {
625 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
626 if (mddev->unit == unit) {
627 mddev_get(mddev);
628 spin_unlock(&all_mddevs_lock);
629 kfree(new);
630 return mddev;
631 }
632
633 if (new) {
634 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 635 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 636 new->hold_active = UNTIL_IOCTL;
637 return new;
1da177e4 638 }
efeb53c0
N		 639 } else if (new) {
640 /* find an unused unit number */
641 static int next_minor = 512;
642 int start = next_minor;
643 int is_free = 0;
644 int dev = 0;
645 while (!is_free) {
646 dev = MKDEV(MD_MAJOR, next_minor);
647 next_minor++;
648 if (next_minor > MINORMASK)
649 next_minor = 0;
650 if (next_minor == start) {
651 /* Oh dear, all in use. */
652 spin_unlock(&all_mddevs_lock);
653 kfree(new);
654 return NULL;
655 }
656
657 is_free = 1;
658 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
659 if (mddev->unit == dev) {
660 is_free = 0;
661 break;
662 }
663 }
664 new->unit = dev;
665 new->md_minor = MINOR(dev);
666 new->hold_active = UNTIL_STOP;
1da177e4
LT		 667 list_add(&new->all_mddevs, &all_mddevs);
668 spin_unlock(&all_mddevs_lock);
669 return new;
670 }
671 spin_unlock(&all_mddevs_lock);
672
9ffae0cf 673 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 674 if (!new)
675 return NULL;
676
1da177e4
LT		 677 new->unit = unit;
678 if (MAJOR(unit) == MD_MAJOR)
679 new->md_minor = MINOR(unit);
680 else
681 new->md_minor = MINOR(unit) >> MdpMinorShift;
682
fafd7fb0 683 mddev_init(new);
1da177e4 684
1da177e4
LT		 685 goto retry;
686}
687
fd01b88c 688static inline int mddev_lock(struct mddev * mddev)
1da177e4 689{
df5b89b3 690 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 691}
692
fd01b88c 693static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW		 694{
695 return mutex_is_locked(&mddev->reconfig_mutex);
696}
697
fd01b88c 698static inline int mddev_trylock(struct mddev * mddev)
1da177e4 699{
df5b89b3 700 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 701}
702
b6eb127d
N		 703static struct attribute_group md_redundancy_group;
704
fd01b88c 705static void mddev_unlock(struct mddev * mddev)
1da177e4 706{
a64c876f 707 if (mddev->to_remove) {
b6eb127d
N		 708 /* These cannot be removed under reconfig_mutex as
709 * an access to the files will try to take reconfig_mutex
710 * while holding the file unremovable, which leads to
711 * a deadlock.
bb4f1e9d
N		 712 * So hold set sysfs_active while the remove in happeing,
713 * and anything else which might set ->to_remove or my
714 * otherwise change the sysfs namespace will fail with
715 * -EBUSY if sysfs_active is still set.
716 * We set sysfs_active under reconfig_mutex and elsewhere
717 * test it under the same mutex to ensure its correct value
718 * is seen.
b6eb127d 719 */
a64c876f
N		 720 struct attribute_group *to_remove = mddev->to_remove;
721 mddev->to_remove = NULL;
bb4f1e9d 722 mddev->sysfs_active = 1;
b6eb127d
N		 723 mutex_unlock(&mddev->reconfig_mutex);
724
00bcb4ac
N		 725 if (mddev->kobj.sd) {
726 if (to_remove != &md_redundancy_group)
727 sysfs_remove_group(&mddev->kobj, to_remove);
728 if (mddev->pers == NULL ||
729 mddev->pers->sync_request == NULL) {
730 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
731 if (mddev->sysfs_action)
732 sysfs_put(mddev->sysfs_action);
733 mddev->sysfs_action = NULL;
734 }
a64c876f 735 }
bb4f1e9d 736 mddev->sysfs_active = 0;
b6eb127d
N		 737 } else
738 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 739
751e67ca
CD		 740 /* As we've dropped the mutex we need a spinlock to
741 * make sure the thread doesn't disappear
01f96c0a
N		 742 */
743 spin_lock(&pers_lock);
005eca5e 744 md_wakeup_thread(mddev->thread);
01f96c0a 745 spin_unlock(&pers_lock);
1da177e4
LT		 746}
747
fd01b88c 748static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 749{
3cb03002 750 struct md_rdev *rdev;
1da177e4 751
159ec1fc 752 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 753 if (rdev->desc_nr == nr)
754 return rdev;
159ec1fc 755
1da177e4
LT		 756 return NULL;
757}
758
fd01b88c 759static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
1da177e4 760{
3cb03002 761 struct md_rdev *rdev;
1da177e4 762
159ec1fc 763 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 764 if (rdev->bdev->bd_dev == dev)
765 return rdev;
159ec1fc 766
1da177e4
LT		 767 return NULL;
768}
769
84fc4b56 770static struct md_personality *find_pers(int level, char *clevel)
2604b703 771{
84fc4b56 772 struct md_personality *pers;
d9d166c2
N		 773 list_for_each_entry(pers, &pers_list, list) {
774 if (level != LEVEL_NONE && pers->level == level)
2604b703 775 return pers;
d9d166c2
N		 776 if (strcmp(pers->name, clevel)==0)
777 return pers;
778 }
2604b703
N		 779 return NULL;
780}
781
b73df2d3 782/* return the offset of the super block in 512byte sectors */
3cb03002 783static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 784{
57b2caa3 785 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 786 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 787}
788
3cb03002 789static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 790{
791 if (rdev->sb_page)
792 MD_BUG();
793
794 rdev->sb_page = alloc_page(GFP_KERNEL);
795 if (!rdev->sb_page) {
796 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 797 return -ENOMEM;
1da177e4
LT		 798 }
799
800 return 0;
801}
802
3cb03002 803static void free_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 804{
805 if (rdev->sb_page) {
2d1f3b5d 806 put_page(rdev->sb_page);
1da177e4
LT		 807 rdev->sb_loaded = 0;
808 rdev->sb_page = NULL;
0f420358 809 rdev->sb_start = 0;
dd8ac336 810 rdev->sectors = 0;
1da177e4 811 }
2699b672
N		 812 if (rdev->bb_page) {
813 put_page(rdev->bb_page);
814 rdev->bb_page = NULL;
815 }
1da177e4
LT		 816}
817
818
6712ecf8 819static void super_written(struct bio *bio, int error)
7bfa19f2 820{
3cb03002 821 struct md_rdev *rdev = bio->bi_private;
fd01b88c 822 struct mddev *mddev = rdev->mddev;
7bfa19f2 823
3a0f5bbb
N		 824 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
825 printk("md: super_written gets error=%d, uptodate=%d\n",
826 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
827 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 828 md_error(mddev, rdev);
3a0f5bbb 829 }
7bfa19f2 830
a9701a30
N		 831 if (atomic_dec_and_test(&mddev->pending_writes))
832 wake_up(&mddev->sb_wait);
f8b58edf 833 bio_put(bio);
7bfa19f2
N		 834}
835
fd01b88c 836void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 837 sector_t sector, int size, struct page *page)
838{
839 /* write first size bytes of page to sector of rdev
840 * Increment mddev->pending_writes before returning
841 * and decrement it on completion, waking up sb_wait
842 * if zero is reached.
843 * If an error occurred, call md_error
844 */
a167f663 845 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 846
a6ff7e08 847 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 848 bio->bi_sector = sector;
849 bio_add_page(bio, page, size, 0);
850 bio->bi_private = rdev;
851 bio->bi_end_io = super_written;
a9701a30 852
7bfa19f2 853 atomic_inc(&mddev->pending_writes);
a5bf4df0 854 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 855}
856
fd01b88c 857void md_super_wait(struct mddev *mddev)
a9701a30 858{
e9c7469b 859 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 860 DEFINE_WAIT(wq);
861 for(;;) {
862 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
863 if (atomic_read(&mddev->pending_writes)==0)
864 break;
a9701a30
N		 865 schedule();
866 }
867 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 868}
869
6712ecf8 870static void bi_complete(struct bio *bio, int error)
1da177e4 871{
1da177e4 872 complete((struct completion*)bio->bi_private);
1da177e4
LT		 873}
874
3cb03002 875int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 876 struct page *page, int rw, bool metadata_op)
1da177e4 877{
a167f663 878 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 879 struct completion event;
880 int ret;
881
721a9602 882 rw |= REQ_SYNC;
1da177e4 883
a6ff7e08
JB		 884 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
885 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 886 if (metadata_op)
887 bio->bi_sector = sector + rdev->sb_start;
888 else
889 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 890 bio_add_page(bio, page, size, 0);
891 init_completion(&event);
892 bio->bi_private = &event;
893 bio->bi_end_io = bi_complete;
894 submit_bio(rw, bio);
895 wait_for_completion(&event);
896
897 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
898 bio_put(bio);
899 return ret;
900}
a8745db2 901EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 902
3cb03002 903static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT		 904{
905 char b[BDEVNAME_SIZE];
906 if (!rdev->sb_page) {
907 MD_BUG();
908 return -EINVAL;
909 }
910 if (rdev->sb_loaded)
911 return 0;
912
913
ccebd4c4 914 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 915 goto fail;
916 rdev->sb_loaded = 1;
917 return 0;
918
919fail:
920 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
921 bdevname(rdev->bdev,b));
922 return -EINVAL;
923}
924
925static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
926{
05710466
AN		 927 return sb1->set_uuid0 == sb2->set_uuid0 &&
928 sb1->set_uuid1 == sb2->set_uuid1 &&
929 sb1->set_uuid2 == sb2->set_uuid2 &&
930 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 931}
932
1da177e4
LT		 933static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
934{
935 int ret;
936 mdp_super_t *tmp1, *tmp2;
937
938 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
939 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
940
941 if (!tmp1 || !tmp2) {
942 ret = 0;
35020f1a 943 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 944 goto abort;
945 }
946
947 *tmp1 = *sb1;
948 *tmp2 = *sb2;
949
950 /*
951 * nr_disks is not constant
952 */
953 tmp1->nr_disks = 0;
954 tmp2->nr_disks = 0;
955
ce0c8e05 956 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 957abort:
990a8baf
JJ		 958 kfree(tmp1);
959 kfree(tmp2);
1da177e4
LT		 960 return ret;
961}
962
4d167f09
N		 963
964static u32 md_csum_fold(u32 csum)
965{
966 csum = (csum & 0xffff) + (csum >> 16);
967 return (csum & 0xffff) + (csum >> 16);
968}
969
1da177e4
LT		 970static unsigned int calc_sb_csum(mdp_super_t * sb)
971{
4d167f09
N		 972 u64 newcsum = 0;
973 u32 *sb32 = (u32*)sb;
974 int i;
1da177e4
LT		 975 unsigned int disk_csum, csum;
976
977 disk_csum = sb->sb_csum;
978 sb->sb_csum = 0;
4d167f09
N		 979
980 for (i = 0; i < MD_SB_BYTES/4 ; i++)
981 newcsum += sb32[i];
982 csum = (newcsum & 0xffffffff) + (newcsum>>32);
983
984
985#ifdef CONFIG_ALPHA
986 /* This used to use csum_partial, which was wrong for several
987 * reasons including that different results are returned on
988 * different architectures. It isn't critical that we get exactly
989 * the same return value as before (we always csum_fold before
990 * testing, and that removes any differences). However as we
991 * know that csum_partial always returned a 16bit value on
992 * alphas, do a fold to maximise conformity to previous behaviour.
993 */
994 sb->sb_csum = md_csum_fold(disk_csum);
995#else
1da177e4 996 sb->sb_csum = disk_csum;
4d167f09 997#endif
1da177e4
LT		 998 return csum;
999}
1000
1001
1002/*
1003 * Handle superblock details.
1004 * We want to be able to handle multiple superblock formats
1005 * so we have a common interface to them all, and an array of
1006 * different handlers.
1007 * We rely on user-space to write the initial superblock, and support
1008 * reading and updating of superblocks.
1009 * Interface methods are:
3cb03002 1010 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1011 * loads and validates a superblock on dev.
1012 * if refdev != NULL, compare superblocks on both devices
1013 * Return:
1014 * 0 - dev has a superblock that is compatible with refdev
1015 * 1 - dev has a superblock that is compatible and newer than refdev
1016 * so dev should be used as the refdev in future
1017 * -EINVAL superblock incompatible or invalid
1018 * -othererror e.g. -EIO
1019 *
fd01b88c 1020 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1021 * Verify that dev is acceptable into mddev.
1022 * The first time, mddev->raid_disks will be 0, and data from
1023 * dev should be merged in. Subsequent calls check that dev
1024 * is new enough. Return 0 or -EINVAL
1025 *
fd01b88c 1026 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1027 * Update the superblock for rdev with data in mddev
1028 * This does not write to disc.
1029 *
1030 */
1031
1032struct super_type {
0cd17fec
CW		 1033 char *name;
1034 struct module *owner;
3cb03002 1035 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
0cd17fec 1036 int minor_version);
fd01b88c
N		 1037 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1038 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
3cb03002 1039 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 1040 sector_t num_sectors);
1da177e4
LT		 1041};
1042
0894cc30
AN		 1043/*
1044 * Check that the given mddev has no bitmap.
1045 *
1046 * This function is called from the run method of all personalities that do not
1047 * support bitmaps. It prints an error message and returns non-zero if mddev
1048 * has a bitmap. Otherwise, it returns 0.
1049 *
1050 */
fd01b88c 1051int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1052{
c3d9714e 1053 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 1054 return 0;
1055 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1056 mdname(mddev), mddev->pers->name);
1057 return 1;
1058}
1059EXPORT_SYMBOL(md_check_no_bitmap);
1060
1da177e4
LT		 1061/*
1062 * load_super for 0.90.0
1063 */
3cb03002 1064static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1065{
1066 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1067 mdp_super_t *sb;
1068 int ret;
1da177e4
LT		 1069
1070 /*
0f420358 1071 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1072 * it's at the end of the disk.
1073 *
1074 * It also happens to be a multiple of 4Kb.
1075 */
57b2caa3 1076 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1077
0002b271 1078 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1079 if (ret) return ret;
1080
1081 ret = -EINVAL;
1082
1083 bdevname(rdev->bdev, b);
65a06f06 1084 sb = page_address(rdev->sb_page);
1da177e4
LT		 1085
1086 if (sb->md_magic != MD_SB_MAGIC) {
1087 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1088 b);
1089 goto abort;
1090 }
1091
1092 if (sb->major_version != 0 ||
f6705578
N		 1093 sb->minor_version < 90 ||
1094 sb->minor_version > 91) {
1da177e4
LT		 1095 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1096 sb->major_version, sb->minor_version,
1097 b);
1098 goto abort;
1099 }
1100
1101 if (sb->raid_disks <= 0)
1102 goto abort;
1103
4d167f09 1104 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1105 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1106 b);
1107 goto abort;
1108 }
1109
1110 rdev->preferred_minor = sb->md_minor;
1111 rdev->data_offset = 0;
0002b271 1112 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1113 rdev->badblocks.shift = -1;
1da177e4
LT		 1114
1115 if (sb->level == LEVEL_MULTIPATH)
1116 rdev->desc_nr = -1;
1117 else
1118 rdev->desc_nr = sb->this_disk.number;
1119
9a7b2b0f 1120 if (!refdev) {
1da177e4 1121 ret = 1;
9a7b2b0f 1122 } else {
1da177e4 1123 __u64 ev1, ev2;
65a06f06 1124 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1125 if (!uuid_equal(refsb, sb)) {
1126 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1127 b, bdevname(refdev->bdev,b2));
1128 goto abort;
1129 }
1130 if (!sb_equal(refsb, sb)) {
1131 printk(KERN_WARNING "md: %s has same UUID"
1132 " but different superblock to %s\n",
1133 b, bdevname(refdev->bdev, b2));
1134 goto abort;
1135 }
1136 ev1 = md_event(sb);
1137 ev2 = md_event(refsb);
1138 if (ev1 > ev2)
1139 ret = 1;
1140 else
1141 ret = 0;
1142 }
8190e754 1143 rdev->sectors = rdev->sb_start;
27a7b260
N		 1144 /* Limit to 4TB as metadata cannot record more than that */
1145 if (rdev->sectors >= (2ULL << 32))
1146 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1147
27a7b260 1148 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1149 /* "this cannot possibly happen" ... */
1150 ret = -EINVAL;
1151
1da177e4
LT		 1152 abort:
1153 return ret;
1154}
1155
1156/*
1157 * validate_super for 0.90.0
1158 */
fd01b88c 1159static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1160{
1161 mdp_disk_t *desc;
65a06f06 1162 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1163 __u64 ev1 = md_event(sb);
1da177e4 1164
41158c7e 1165 rdev->raid_disk = -1;
c5d79adb
N		 1166 clear_bit(Faulty, &rdev->flags);
1167 clear_bit(In_sync, &rdev->flags);
1168 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1169
1da177e4
LT		 1170 if (mddev->raid_disks == 0) {
1171 mddev->major_version = 0;
1172 mddev->minor_version = sb->minor_version;
1173 mddev->patch_version = sb->patch_version;
e691063a 1174 mddev->external = 0;
9d8f0363 1175 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1176 mddev->ctime = sb->ctime;
1177 mddev->utime = sb->utime;
1178 mddev->level = sb->level;
d9d166c2 1179 mddev->clevel[0] = 0;
1da177e4
LT		 1180 mddev->layout = sb->layout;
1181 mddev->raid_disks = sb->raid_disks;
27a7b260 1182 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1183 mddev->events = ev1;
c3d9714e
N		 1184 mddev->bitmap_info.offset = 0;
1185 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1186
f6705578
N		 1187 if (mddev->minor_version >= 91) {
1188 mddev->reshape_position = sb->reshape_position;
1189 mddev->delta_disks = sb->delta_disks;
1190 mddev->new_level = sb->new_level;
1191 mddev->new_layout = sb->new_layout;
664e7c41 1192 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1193 } else {
1194 mddev->reshape_position = MaxSector;
1195 mddev->delta_disks = 0;
1196 mddev->new_level = mddev->level;
1197 mddev->new_layout = mddev->layout;
664e7c41 1198 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1199 }
1200
1da177e4
LT		 1201 if (sb->state & (1<<MD_SB_CLEAN))
1202 mddev->recovery_cp = MaxSector;
1203 else {
1204 if (sb->events_hi == sb->cp_events_hi &&
1205 sb->events_lo == sb->cp_events_lo) {
1206 mddev->recovery_cp = sb->recovery_cp;
1207 } else
1208 mddev->recovery_cp = 0;
1209 }
1210
1211 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1212 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1213 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1214 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1215
1216 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1217
1218 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1219 mddev->bitmap_info.file == NULL)
1220 mddev->bitmap_info.offset =
1221 mddev->bitmap_info.default_offset;
a654b9d8 1222
41158c7e 1223 } else if (mddev->pers == NULL) {
be6800a7
N		 1224 /* Insist on good event counter while assembling, except
1225 * for spares (which don't need an event count) */
1da177e4 1226 ++ev1;
be6800a7
N		 1227 if (sb->disks[rdev->desc_nr].state & (
1228 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1229 if (ev1 < mddev->events)
1230 return -EINVAL;
41158c7e
N		 1231 } else if (mddev->bitmap) {
1232 /* if adding to array with a bitmap, then we can accept an
1233 * older device ... but not too old.
1234 */
41158c7e
N		 1235 if (ev1 < mddev->bitmap->events_cleared)
1236 return 0;
07d84d10
N		 1237 } else {
1238 if (ev1 < mddev->events)
1239 /* just a hot-add of a new device, leave raid_disk at -1 */
1240 return 0;
1241 }
41158c7e 1242
1da177e4 1243 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1244 desc = sb->disks + rdev->desc_nr;
1245
1246 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1247 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1248 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1249 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1250 set_bit(In_sync, &rdev->flags);
1da177e4 1251 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1252 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1253 /* active but not in sync implies recovery up to
1254 * reshape position. We don't know exactly where
1255 * that is, so set to zero for now */
1256 if (mddev->minor_version >= 91) {
1257 rdev->recovery_offset = 0;
1258 rdev->raid_disk = desc->raid_disk;
1259 }
1da177e4 1260 }
8ddf9efe
N		 1261 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1262 set_bit(WriteMostly, &rdev->flags);
41158c7e 1263 } else /* MULTIPATH are always insync */
b2d444d7 1264 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1265 return 0;
1266}
1267
1268/*
1269 * sync_super for 0.90.0
1270 */
fd01b88c 1271static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1272{
1273 mdp_super_t *sb;
3cb03002 1274 struct md_rdev *rdev2;
1da177e4 1275 int next_spare = mddev->raid_disks;
19133a42 1276
1da177e4
LT		 1277
1278 /* make rdev->sb match mddev data..
1279 *
1280 * 1/ zero out disks
1281 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1282 * 3/ any empty disks < next_spare become removed
1283 *
1284 * disks[0] gets initialised to REMOVED because
1285 * we cannot be sure from other fields if it has
1286 * been initialised or not.
1287 */
1288 int i;
1289 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1290
61181565
N		 1291 rdev->sb_size = MD_SB_BYTES;
1292
65a06f06 1293 sb = page_address(rdev->sb_page);
1da177e4
LT		 1294
1295 memset(sb, 0, sizeof(*sb));
1296
1297 sb->md_magic = MD_SB_MAGIC;
1298 sb->major_version = mddev->major_version;
1da177e4
LT		 1299 sb->patch_version = mddev->patch_version;
1300 sb->gvalid_words = 0; /* ignored */
1301 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1302 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1303 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1304 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1305
1306 sb->ctime = mddev->ctime;
1307 sb->level = mddev->level;
58c0fed4 1308 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1309 sb->raid_disks = mddev->raid_disks;
1310 sb->md_minor = mddev->md_minor;
e691063a 1311 sb->not_persistent = 0;
1da177e4
LT		 1312 sb->utime = mddev->utime;
1313 sb->state = 0;
1314 sb->events_hi = (mddev->events>>32);
1315 sb->events_lo = (u32)mddev->events;
1316
f6705578
N		 1317 if (mddev->reshape_position == MaxSector)
1318 sb->minor_version = 90;
1319 else {
1320 sb->minor_version = 91;
1321 sb->reshape_position = mddev->reshape_position;
1322 sb->new_level = mddev->new_level;
1323 sb->delta_disks = mddev->delta_disks;
1324 sb->new_layout = mddev->new_layout;
664e7c41 1325 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1326 }
1327 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1328 if (mddev->in_sync)
1329 {
1330 sb->recovery_cp = mddev->recovery_cp;
1331 sb->cp_events_hi = (mddev->events>>32);
1332 sb->cp_events_lo = (u32)mddev->events;
1333 if (mddev->recovery_cp == MaxSector)
1334 sb->state = (1<< MD_SB_CLEAN);
1335 } else
1336 sb->recovery_cp = 0;
1337
1338 sb->layout = mddev->layout;
9d8f0363 1339 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1340
c3d9714e 1341 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1342 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1343
1da177e4 1344 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1345 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1346 mdp_disk_t *d;
86e6ffdd 1347 int desc_nr;
0261cd9f
N		 1348 int is_active = test_bit(In_sync, &rdev2->flags);
1349
1350 if (rdev2->raid_disk >= 0 &&
1351 sb->minor_version >= 91)
1352 /* we have nowhere to store the recovery_offset,
1353 * but if it is not below the reshape_position,
1354 * we can piggy-back on that.
1355 */
1356 is_active = 1;
1357 if (rdev2->raid_disk < 0 ||
1358 test_bit(Faulty, &rdev2->flags))
1359 is_active = 0;
1360 if (is_active)
86e6ffdd 1361 desc_nr = rdev2->raid_disk;
1da177e4 1362 else
86e6ffdd 1363 desc_nr = next_spare++;
19133a42 1364 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1365 d = &sb->disks[rdev2->desc_nr];
1366 nr_disks++;
1367 d->number = rdev2->desc_nr;
1368 d->major = MAJOR(rdev2->bdev->bd_dev);
1369 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1370 if (is_active)
1da177e4
LT		 1371 d->raid_disk = rdev2->raid_disk;
1372 else
1373 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1374 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1375 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1376 else if (is_active) {
1da177e4 1377 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1378 if (test_bit(In_sync, &rdev2->flags))
1379 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1380 active++;
1381 working++;
1382 } else {
1383 d->state = 0;
1384 spare++;
1385 working++;
1386 }
8ddf9efe
N		 1387 if (test_bit(WriteMostly, &rdev2->flags))
1388 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1389 }
1da177e4
LT		 1390 /* now set the "removed" and "faulty" bits on any missing devices */
1391 for (i=0 ; i < mddev->raid_disks ; i++) {
1392 mdp_disk_t *d = &sb->disks[i];
1393 if (d->state == 0 && d->number == 0) {
1394 d->number = i;
1395 d->raid_disk = i;
1396 d->state = (1<<MD_DISK_REMOVED);
1397 d->state |= (1<<MD_DISK_FAULTY);
1398 failed++;
1399 }
1400 }
1401 sb->nr_disks = nr_disks;
1402 sb->active_disks = active;
1403 sb->working_disks = working;
1404 sb->failed_disks = failed;
1405 sb->spare_disks = spare;
1406
1407 sb->this_disk = sb->disks[rdev->desc_nr];
1408 sb->sb_csum = calc_sb_csum(sb);
1409}
1410
0cd17fec
CW		 1411/*
1412 * rdev_size_change for 0.90.0
1413 */
1414static unsigned long long
3cb03002 1415super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1416{
58c0fed4 1417 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1418 return 0; /* component must fit device */
c3d9714e 1419 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1420 return 0; /* can't move bitmap */
57b2caa3 1421 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1422 if (!num_sectors || num_sectors > rdev->sb_start)
1423 num_sectors = rdev->sb_start;
27a7b260
N		 1424 /* Limit to 4TB as metadata cannot record more than that.
1425 * 4TB == 2^32 KB, or 2*2^32 sectors.
1426 */
1427 if (num_sectors >= (2ULL << 32))
1428 num_sectors = (2ULL << 32) - 2;
0f420358 1429 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1430 rdev->sb_page);
1431 md_super_wait(rdev->mddev);
c26a44ed 1432 return num_sectors;
0cd17fec
CW		 1433}
1434
1435
1da177e4
LT		 1436/*
1437 * version 1 superblock
1438 */
1439
1c05b4bc 1440static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1441{
1c05b4bc
N		 1442 __le32 disk_csum;
1443 u32 csum;
1da177e4
LT		 1444 unsigned long long newcsum;
1445 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1446 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1447 int i;
1448
1449 disk_csum = sb->sb_csum;
1450 sb->sb_csum = 0;
1451 newcsum = 0;
1452 for (i=0; size>=4; size -= 4 )
1453 newcsum += le32_to_cpu(*isuper++);
1454
1455 if (size == 2)
1c05b4bc 1456 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1457
1458 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1459 sb->sb_csum = disk_csum;
1460 return cpu_to_le32(csum);
1461}
1462
2699b672
N		 1463static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1464 int acknowledged);
3cb03002 1465static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1466{
1467 struct mdp_superblock_1 *sb;
1468 int ret;
0f420358 1469 sector_t sb_start;
1da177e4 1470 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1471 int bmask;
1da177e4
LT		 1472
1473 /*
0f420358 1474 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1475 * It is always aligned to a 4K boundary and
1476 * depeding on minor_version, it can be:
1477 * 0: At least 8K, but less than 12K, from end of device
1478 * 1: At start of device
1479 * 2: 4K from start of device.
1480 */
1481 switch(minor_version) {
1482 case 0:
77304d2a 1483 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1484 sb_start -= 8*2;
1485 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1486 break;
1487 case 1:
0f420358 1488 sb_start = 0;
1da177e4
LT		 1489 break;
1490 case 2:
0f420358 1491 sb_start = 8;
1da177e4
LT		 1492 break;
1493 default:
1494 return -EINVAL;
1495 }
0f420358 1496 rdev->sb_start = sb_start;
1da177e4 1497
0002b271
N		 1498 /* superblock is rarely larger than 1K, but it can be larger,
1499 * and it is safe to read 4k, so we do that
1500 */
1501 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1502 if (ret) return ret;
1503
1504
65a06f06 1505 sb = page_address(rdev->sb_page);
1da177e4
LT		 1506
1507 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1508 sb->major_version != cpu_to_le32(1) ||
1509 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1510 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1511 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1512 return -EINVAL;
1513
1514 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1515 printk("md: invalid superblock checksum on %s\n",
1516 bdevname(rdev->bdev,b));
1517 return -EINVAL;
1518 }
1519 if (le64_to_cpu(sb->data_size) < 10) {
1520 printk("md: data_size too small on %s\n",
1521 bdevname(rdev->bdev,b));
1522 return -EINVAL;
1523 }
e11e93fa 1524
1da177e4
LT		 1525 rdev->preferred_minor = 0xffff;
1526 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1527 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1528
0002b271 1529 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1530 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1531 if (rdev->sb_size & bmask)
a1801f85
N		 1532 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1533
1534 if (minor_version
0f420358 1535 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1536 return -EINVAL;
0002b271 1537
31b65a0d
N		 1538 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1539 rdev->desc_nr = -1;
1540 else
1541 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1542
2699b672
N		 1543 if (!rdev->bb_page) {
1544 rdev->bb_page = alloc_page(GFP_KERNEL);
1545 if (!rdev->bb_page)
1546 return -ENOMEM;
1547 }
1548 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1549 rdev->badblocks.count == 0) {
1550 /* need to load the bad block list.
1551 * Currently we limit it to one page.
1552 */
1553 s32 offset;
1554 sector_t bb_sector;
1555 u64 *bbp;
1556 int i;
1557 int sectors = le16_to_cpu(sb->bblog_size);
1558 if (sectors > (PAGE_SIZE / 512))
1559 return -EINVAL;
1560 offset = le32_to_cpu(sb->bblog_offset);
1561 if (offset == 0)
1562 return -EINVAL;
1563 bb_sector = (long long)offset;
1564 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1565 rdev->bb_page, READ, true))
1566 return -EIO;
1567 bbp = (u64 *)page_address(rdev->bb_page);
1568 rdev->badblocks.shift = sb->bblog_shift;
1569 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1570 u64 bb = le64_to_cpu(*bbp);
1571 int count = bb & (0x3ff);
1572 u64 sector = bb >> 10;
1573 sector <<= sb->bblog_shift;
1574 count <<= sb->bblog_shift;
1575 if (bb + 1 == 0)
1576 break;
1577 if (md_set_badblocks(&rdev->badblocks,
1578 sector, count, 1) == 0)
1579 return -EINVAL;
1580 }
1581 } else if (sb->bblog_offset == 0)
1582 rdev->badblocks.shift = -1;
1583
9a7b2b0f 1584 if (!refdev) {
8ed75463 1585 ret = 1;
9a7b2b0f 1586 } else {
1da177e4 1587 __u64 ev1, ev2;
65a06f06 1588 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1589
1590 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1591 sb->level != refsb->level ||
1592 sb->layout != refsb->layout ||
1593 sb->chunksize != refsb->chunksize) {
1594 printk(KERN_WARNING "md: %s has strangely different"
1595 " superblock to %s\n",
1596 bdevname(rdev->bdev,b),
1597 bdevname(refdev->bdev,b2));
1598 return -EINVAL;
1599 }
1600 ev1 = le64_to_cpu(sb->events);
1601 ev2 = le64_to_cpu(refsb->events);
1602
1603 if (ev1 > ev2)
8ed75463
N		 1604 ret = 1;
1605 else
1606 ret = 0;
1da177e4 1607 }
a1801f85 1608 if (minor_version)
77304d2a 1609 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 1610 le64_to_cpu(sb->data_offset);
1da177e4 1611 else
dd8ac336
AN		 1612 rdev->sectors = rdev->sb_start;
1613 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1614 return -EINVAL;
dd8ac336 1615 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1616 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1617 return -EINVAL;
8ed75463 1618 return ret;
1da177e4
LT		 1619}
1620
fd01b88c 1621static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1622{
65a06f06 1623 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1624 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1625
41158c7e 1626 rdev->raid_disk = -1;
c5d79adb
N		 1627 clear_bit(Faulty, &rdev->flags);
1628 clear_bit(In_sync, &rdev->flags);
1629 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1630
1da177e4
LT		 1631 if (mddev->raid_disks == 0) {
1632 mddev->major_version = 1;
1633 mddev->patch_version = 0;
e691063a 1634 mddev->external = 0;
9d8f0363 1635 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1636 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1637 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1638 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1639 mddev->clevel[0] = 0;
1da177e4
LT		 1640 mddev->layout = le32_to_cpu(sb->layout);
1641 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1642 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1643 mddev->events = ev1;
c3d9714e
N		 1644 mddev->bitmap_info.offset = 0;
1645 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1646
1647 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1648 memcpy(mddev->uuid, sb->set_uuid, 16);
1649
1650 mddev->max_disks = (4096-256)/2;
a654b9d8 1651
71c0805c 1652 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1653 mddev->bitmap_info.file == NULL )
1654 mddev->bitmap_info.offset =
1655 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1656
f6705578
N		 1657 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1658 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1659 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1660 mddev->new_level = le32_to_cpu(sb->new_level);
1661 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1662 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1663 } else {
1664 mddev->reshape_position = MaxSector;
1665 mddev->delta_disks = 0;
1666 mddev->new_level = mddev->level;
1667 mddev->new_layout = mddev->layout;
664e7c41 1668 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1669 }
1670
41158c7e 1671 } else if (mddev->pers == NULL) {
be6800a7
N		 1672 /* Insist of good event counter while assembling, except for
1673 * spares (which don't need an event count) */
1da177e4 1674 ++ev1;
be6800a7
N		 1675 if (rdev->desc_nr >= 0 &&
1676 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1677 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1678 if (ev1 < mddev->events)
1679 return -EINVAL;
41158c7e
N		 1680 } else if (mddev->bitmap) {
1681 /* If adding to array with a bitmap, then we can accept an
1682 * older device, but not too old.
1683 */
41158c7e
N		 1684 if (ev1 < mddev->bitmap->events_cleared)
1685 return 0;
07d84d10
N		 1686 } else {
1687 if (ev1 < mddev->events)
1688 /* just a hot-add of a new device, leave raid_disk at -1 */
1689 return 0;
1690 }
1da177e4
LT		 1691 if (mddev->level != LEVEL_MULTIPATH) {
1692 int role;
3673f305
N		 1693 if (rdev->desc_nr < 0 ||
1694 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1695 role = 0xffff;
1696 rdev->desc_nr = -1;
1697 } else
1698 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1699 switch(role) {
1700 case 0xffff: /* spare */
1da177e4
LT		 1701 break;
1702 case 0xfffe: /* faulty */
b2d444d7 1703 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1704 break;
1705 default:
5fd6c1dc
N		 1706 if ((le32_to_cpu(sb->feature_map) &
1707 MD_FEATURE_RECOVERY_OFFSET))
1708 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1709 else
1710 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1711 rdev->raid_disk = role;
1712 break;
1713 }
8ddf9efe
N		 1714 if (sb->devflags & WriteMostly1)
1715 set_bit(WriteMostly, &rdev->flags);
41158c7e 1716 } else /* MULTIPATH are always insync */
b2d444d7 1717 set_bit(In_sync, &rdev->flags);
41158c7e 1718
1da177e4
LT		 1719 return 0;
1720}
1721
fd01b88c 1722static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1723{
1724 struct mdp_superblock_1 *sb;
3cb03002 1725 struct md_rdev *rdev2;
1da177e4
LT		 1726 int max_dev, i;
1727 /* make rdev->sb match mddev and rdev data. */
1728
65a06f06 1729 sb = page_address(rdev->sb_page);
1da177e4
LT		 1730
1731 sb->feature_map = 0;
1732 sb->pad0 = 0;
5fd6c1dc 1733 sb->recovery_offset = cpu_to_le64(0);
1da177e4 1734 memset(sb->pad1, 0, sizeof(sb->pad1));
1da177e4
LT		 1735 memset(sb->pad3, 0, sizeof(sb->pad3));
1736
1737 sb->utime = cpu_to_le64((__u64)mddev->utime);
1738 sb->events = cpu_to_le64(mddev->events);
1739 if (mddev->in_sync)
1740 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1741 else
1742 sb->resync_offset = cpu_to_le64(0);
1743
1c05b4bc 1744 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1745
f0ca340c 1746 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1747 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1748 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1749 sb->level = cpu_to_le32(mddev->level);
1750 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1751
aeb9b211
N		 1752 if (test_bit(WriteMostly, &rdev->flags))
1753 sb->devflags |= WriteMostly1;
1754 else
1755 sb->devflags &= ~WriteMostly1;
1756
c3d9714e
N		 1757 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1758 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1759 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1760 }
5fd6c1dc
N		 1761
1762 if (rdev->raid_disk >= 0 &&
97e4f42d 1763 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1764 sb->feature_map |=
1765 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1766 sb->recovery_offset =
1767 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc
N		 1768 }
1769
f6705578
N		 1770 if (mddev->reshape_position != MaxSector) {
1771 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1772 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1773 sb->new_layout = cpu_to_le32(mddev->new_layout);
1774 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1775 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1776 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1777 }
a654b9d8 1778
2699b672
N		 1779 if (rdev->badblocks.count == 0)
1780 /* Nothing to do for bad blocks*/ ;
1781 else if (sb->bblog_offset == 0)
1782 /* Cannot record bad blocks on this device */
1783 md_error(mddev, rdev);
1784 else {
1785 struct badblocks *bb = &rdev->badblocks;
1786 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1787 u64 *p = bb->page;
1788 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1789 if (bb->changed) {
1790 unsigned seq;
1791
1792retry:
1793 seq = read_seqbegin(&bb->lock);
1794
1795 memset(bbp, 0xff, PAGE_SIZE);
1796
1797 for (i = 0 ; i < bb->count ; i++) {
1798 u64 internal_bb = *p++;
1799 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1800 | BB_LEN(internal_bb));
1801 *bbp++ = cpu_to_le64(store_bb);
1802 }
1803 if (read_seqretry(&bb->lock, seq))
1804 goto retry;
1805
1806 bb->sector = (rdev->sb_start +
1807 (int)le32_to_cpu(sb->bblog_offset));
1808 bb->size = le16_to_cpu(sb->bblog_size);
1809 bb->changed = 0;
1810 }
1811 }
1812
1da177e4 1813 max_dev = 0;
159ec1fc 1814 list_for_each_entry(rdev2, &mddev->disks, same_set)
1da177e4
LT		 1815 if (rdev2->desc_nr+1 > max_dev)
1816 max_dev = rdev2->desc_nr+1;
a778b73f 1817
70471daf
N		 1818 if (max_dev > le32_to_cpu(sb->max_dev)) {
1819 int bmask;
a778b73f 1820 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1821 rdev->sb_size = max_dev * 2 + 256;
1822 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1823 if (rdev->sb_size & bmask)
1824 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1825 } else
1826 max_dev = le32_to_cpu(sb->max_dev);
1827
1da177e4
LT		 1828 for (i=0; i<max_dev;i++)
1829 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1830
159ec1fc 1831 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1832 i = rdev2->desc_nr;
b2d444d7 1833 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1834 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1835 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1836 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1837 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1838 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1839 else
1840 sb->dev_roles[i] = cpu_to_le16(0xffff);
1841 }
1842
1da177e4
LT		 1843 sb->sb_csum = calc_sb_1_csum(sb);
1844}
1845
0cd17fec 1846static unsigned long long
3cb03002 1847super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1848{
1849 struct mdp_superblock_1 *sb;
15f4a5fd 1850 sector_t max_sectors;
58c0fed4 1851 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1852 return 0; /* component must fit device */
0f420358 1853 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1854 /* minor versions 1 and 2; superblock before data */
77304d2a 1855 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1856 max_sectors -= rdev->data_offset;
1857 if (!num_sectors || num_sectors > max_sectors)
1858 num_sectors = max_sectors;
c3d9714e 1859 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1860 /* minor version 0 with bitmap we can't move */
1861 return 0;
1862 } else {
1863 /* minor version 0; superblock after data */
0f420358 1864 sector_t sb_start;
77304d2a 1865 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1866 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1867 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1868 if (!num_sectors || num_sectors > max_sectors)
1869 num_sectors = max_sectors;
0f420358 1870 rdev->sb_start = sb_start;
0cd17fec 1871 }
65a06f06 1872 sb = page_address(rdev->sb_page);
15f4a5fd 1873 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1874 sb->super_offset = rdev->sb_start;
0cd17fec 1875 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1876 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1877 rdev->sb_page);
1878 md_super_wait(rdev->mddev);
c26a44ed 1879 return num_sectors;
0cd17fec 1880}
1da177e4 1881
75c96f85 1882static struct super_type super_types[] = {
1da177e4
LT		 1883 [0] = {
1884 .name = "0.90.0",
1885 .owner = THIS_MODULE,
0cd17fec
CW		 1886 .load_super = super_90_load,
1887 .validate_super = super_90_validate,
1888 .sync_super = super_90_sync,
1889 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1890 },
1891 [1] = {
1892 .name = "md-1",
1893 .owner = THIS_MODULE,
0cd17fec
CW		 1894 .load_super = super_1_load,
1895 .validate_super = super_1_validate,
1896 .sync_super = super_1_sync,
1897 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1898 },
1899};
1da177e4 1900
fd01b88c 1901static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1902{
1903 if (mddev->sync_super) {
1904 mddev->sync_super(mddev, rdev);
1905 return;
1906 }
1907
1908 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1909
1910 super_types[mddev->major_version].sync_super(mddev, rdev);
1911}
1912
fd01b88c 1913static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1914{
3cb03002 1915 struct md_rdev *rdev, *rdev2;
1da177e4 1916
4b80991c
N		 1917 rcu_read_lock();
1918 rdev_for_each_rcu(rdev, mddev1)
1919 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1920 if (rdev->bdev->bd_contains ==
4b80991c
N		 1921 rdev2->bdev->bd_contains) {
1922 rcu_read_unlock();
7dd5e7c3 1923 return 1;
4b80991c
N		 1924 }
1925 rcu_read_unlock();
1da177e4
LT		 1926 return 0;
1927}
1928
1929static LIST_HEAD(pending_raid_disks);
1930
ac5e7113
AN		 1931/*
1932 * Try to register data integrity profile for an mddev
1933 *
1934 * This is called when an array is started and after a disk has been kicked
1935 * from the array. It only succeeds if all working and active component devices
1936 * are integrity capable with matching profiles.
1937 */
fd01b88c 1938int md_integrity_register(struct mddev *mddev)
ac5e7113 1939{
3cb03002 1940 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1941
1942 if (list_empty(&mddev->disks))
1943 return 0; /* nothing to do */
629acb6a
JB		 1944 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1945 return 0; /* shouldn't register, or already is */
ac5e7113
AN		 1946 list_for_each_entry(rdev, &mddev->disks, same_set) {
1947 /* skip spares and non-functional disks */
1948 if (test_bit(Faulty, &rdev->flags))
1949 continue;
1950 if (rdev->raid_disk < 0)
1951 continue;
ac5e7113
AN		 1952 if (!reference) {
1953 /* Use the first rdev as the reference */
1954 reference = rdev;
1955 continue;
1956 }
1957 /* does this rdev's profile match the reference profile? */
1958 if (blk_integrity_compare(reference->bdev->bd_disk,
1959 rdev->bdev->bd_disk) < 0)
1960 return -EINVAL;
1961 }
89078d57
MP		 1962 if (!reference || !bdev_get_integrity(reference->bdev))
1963 return 0;
ac5e7113
AN		 1964 /*
1965 * All component devices are integrity capable and have matching
1966 * profiles, register the common profile for the md device.
1967 */
1968 if (blk_integrity_register(mddev->gendisk,
1969 bdev_get_integrity(reference->bdev)) != 0) {
1970 printk(KERN_ERR "md: failed to register integrity for %s\n",
1971 mdname(mddev));
1972 return -EINVAL;
1973 }
a91a2785
MP		 1974 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1975 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1976 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1977 mdname(mddev));
1978 return -EINVAL;
1979 }
ac5e7113
AN		 1980 return 0;
1981}
1982EXPORT_SYMBOL(md_integrity_register);
1983
1984/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 1985void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 1986{
3f9d99c1 1987 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1988 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1989
ac5e7113 1990 if (!bi_mddev) /* nothing to do */
3f9d99c1 1991 return;
ac5e7113 1992 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1993 return;
ac5e7113
AN		 1994 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1995 rdev->bdev->bd_disk) >= 0)
1996 return;
1997 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1998 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 1999}
ac5e7113 2000EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2001
fd01b88c 2002static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
1da177e4 2003{
7dd5e7c3 2004 char b[BDEVNAME_SIZE];
f637b9f9 2005 struct kobject *ko;
1edf80d3 2006 char *s;
5e55e2f5 2007 int err;
1da177e4
LT		 2008
2009 if (rdev->mddev) {
2010 MD_BUG();
2011 return -EINVAL;
2012 }
11e2ede0
DW		 2013
2014 /* prevent duplicates */
2015 if (find_rdev(mddev, rdev->bdev->bd_dev))
2016 return -EEXIST;
2017
dd8ac336
AN		 2018 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2019 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2020 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2021 if (mddev->pers) {
2022 /* Cannot change size, so fail
2023 * If mddev->level <= 0, then we don't care
2024 * about aligning sizes (e.g. linear)
2025 */
2026 if (mddev->level > 0)
2027 return -ENOSPC;
2028 } else
dd8ac336 2029 mddev->dev_sectors = rdev->sectors;
2bf071bf 2030 }
1da177e4
LT		 2031
2032 /* Verify rdev->desc_nr is unique.
2033 * If it is -1, assign a free number, else
2034 * check number is not in use
2035 */
2036 if (rdev->desc_nr < 0) {
2037 int choice = 0;
2038 if (mddev->pers) choice = mddev->raid_disks;
2039 while (find_rdev_nr(mddev, choice))
2040 choice++;
2041 rdev->desc_nr = choice;
2042 } else {
2043 if (find_rdev_nr(mddev, rdev->desc_nr))
2044 return -EBUSY;
2045 }
de01dfad
N		 2046 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2047 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2048 mdname(mddev), mddev->max_disks);
2049 return -EBUSY;
2050 }
19133a42 2051 bdevname(rdev->bdev,b);
649316b2 2052 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2053 *s = '!';
649316b2 2054
1da177e4 2055 rdev->mddev = mddev;
19133a42 2056 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2057
b2d6db58 2058 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2059 goto fail;
86e6ffdd 2060
0762b8bd 2061 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2062 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2063 /* failure here is OK */;
2064 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2065
4b80991c 2066 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2067 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2068
2069 /* May as well allow recovery to be retried once */
5389042f 2070 mddev->recovery_disabled++;
3f9d99c1 2071
1da177e4 2072 return 0;
5e55e2f5
N		 2073
2074 fail:
2075 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2076 b, mdname(mddev));
2077 return err;
1da177e4
LT		 2078}
2079
177a99b2 2080static void md_delayed_delete(struct work_struct *ws)
5792a285 2081{
3cb03002 2082 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2083 kobject_del(&rdev->kobj);
177a99b2 2084 kobject_put(&rdev->kobj);
5792a285
N		 2085}
2086
3cb03002 2087static void unbind_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2088{
2089 char b[BDEVNAME_SIZE];
2090 if (!rdev->mddev) {
2091 MD_BUG();
2092 return;
2093 }
49731baa 2094 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2095 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2096 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2097 rdev->mddev = NULL;
86e6ffdd 2098 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2099 sysfs_put(rdev->sysfs_state);
2100 rdev->sysfs_state = NULL;
2230dfe4
N		 2101 kfree(rdev->badblocks.page);
2102 rdev->badblocks.count = 0;
2103 rdev->badblocks.page = NULL;
5792a285 2104 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2105 * writing to 'remove' to "dev/state". We also need
2106 * to delay it due to rcu usage.
5792a285 2107 */
4b80991c 2108 synchronize_rcu();
177a99b2
N		 2109 INIT_WORK(&rdev->del_work, md_delayed_delete);
2110 kobject_get(&rdev->kobj);
e804ac78 2111 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2112}
2113
2114/*
2115 * prevent the device from being mounted, repartitioned or
2116 * otherwise reused by a RAID array (or any other kernel
2117 * subsystem), by bd_claiming the device.
2118 */
3cb03002 2119static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2120{
2121 int err = 0;
2122 struct block_device *bdev;
2123 char b[BDEVNAME_SIZE];
2124
d4d77629 2125 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2126 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2127 if (IS_ERR(bdev)) {
2128 printk(KERN_ERR "md: could not open %s.\n",
2129 __bdevname(dev, b));
2130 return PTR_ERR(bdev);
2131 }
1da177e4
LT		 2132 rdev->bdev = bdev;
2133 return err;
2134}
2135
3cb03002 2136static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2137{
2138 struct block_device *bdev = rdev->bdev;
2139 rdev->bdev = NULL;
2140 if (!bdev)
2141 MD_BUG();
e525fd89 2142 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2143}
2144
2145void md_autodetect_dev(dev_t dev);
2146
3cb03002 2147static void export_rdev(struct md_rdev * rdev)
1da177e4
LT		 2148{
2149 char b[BDEVNAME_SIZE];
2150 printk(KERN_INFO "md: export_rdev(%s)\n",
2151 bdevname(rdev->bdev,b));
2152 if (rdev->mddev)
2153 MD_BUG();
2154 free_disk_sb(rdev);
1da177e4 2155#ifndef MODULE
d0fae18f
N		 2156 if (test_bit(AutoDetected, &rdev->flags))
2157 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2158#endif
2159 unlock_rdev(rdev);
86e6ffdd 2160 kobject_put(&rdev->kobj);
1da177e4
LT		 2161}
2162
3cb03002 2163static void kick_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2164{
2165 unbind_rdev_from_array(rdev);
2166 export_rdev(rdev);
2167}
2168
fd01b88c 2169static void export_array(struct mddev *mddev)
1da177e4 2170{
3cb03002 2171 struct md_rdev *rdev, *tmp;
1da177e4 2172
d089c6af 2173 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 2174 if (!rdev->mddev) {
2175 MD_BUG();
2176 continue;
2177 }
2178 kick_rdev_from_array(rdev);
2179 }
2180 if (!list_empty(&mddev->disks))
2181 MD_BUG();
2182 mddev->raid_disks = 0;
2183 mddev->major_version = 0;
2184}
2185
2186static void print_desc(mdp_disk_t *desc)
2187{
2188 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2189 desc->major,desc->minor,desc->raid_disk,desc->state);
2190}
2191
cd2ac932 2192static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2193{
2194 int i;
2195
2196 printk(KERN_INFO
2197 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2198 sb->major_version, sb->minor_version, sb->patch_version,
2199 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2200 sb->ctime);
2201 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2202 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2203 sb->md_minor, sb->layout, sb->chunk_size);
2204 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2205 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2206 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2207 sb->failed_disks, sb->spare_disks,
2208 sb->sb_csum, (unsigned long)sb->events_lo);
2209
2210 printk(KERN_INFO);
2211 for (i = 0; i < MD_SB_DISKS; i++) {
2212 mdp_disk_t *desc;
2213
2214 desc = sb->disks + i;
2215 if (desc->number || desc->major || desc->minor ||
2216 desc->raid_disk || (desc->state && (desc->state != 4))) {
2217 printk(" D %2d: ", i);
2218 print_desc(desc);
2219 }
2220 }
2221 printk(KERN_INFO "md: THIS: ");
2222 print_desc(&sb->this_disk);
cd2ac932 2223}
1da177e4 2224
cd2ac932
CR		 2225static void print_sb_1(struct mdp_superblock_1 *sb)
2226{
2227 __u8 *uuid;
2228
2229 uuid = sb->set_uuid;
ad361c98 2230 printk(KERN_INFO
7b75c2f8 2231 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2232 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2233 le32_to_cpu(sb->major_version),
2234 le32_to_cpu(sb->feature_map),
7b75c2f8 2235 uuid,
cd2ac932
CR		 2236 sb->set_name,
2237 (unsigned long long)le64_to_cpu(sb->ctime)
2238 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2239
2240 uuid = sb->device_uuid;
ad361c98
JP		 2241 printk(KERN_INFO
2242 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2243 " RO:%llu\n"
7b75c2f8 2244 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2245 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2246 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2247 le32_to_cpu(sb->level),
2248 (unsigned long long)le64_to_cpu(sb->size),
2249 le32_to_cpu(sb->raid_disks),
2250 le32_to_cpu(sb->layout),
2251 le32_to_cpu(sb->chunksize),
2252 (unsigned long long)le64_to_cpu(sb->data_offset),
2253 (unsigned long long)le64_to_cpu(sb->data_size),
2254 (unsigned long long)le64_to_cpu(sb->super_offset),
2255 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2256 le32_to_cpu(sb->dev_number),
7b75c2f8 2257 uuid,
cd2ac932
CR		 2258 sb->devflags,
2259 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2260 (unsigned long long)le64_to_cpu(sb->events),
2261 (unsigned long long)le64_to_cpu(sb->resync_offset),
2262 le32_to_cpu(sb->sb_csum),
2263 le32_to_cpu(sb->max_dev)
2264 );
1da177e4
LT		 2265}
2266
3cb03002 2267static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT		 2268{
2269 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2270 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2271 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2272 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2273 rdev->desc_nr);
1da177e4 2274 if (rdev->sb_loaded) {
cd2ac932
CR		 2275 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2276 switch (major_version) {
2277 case 0:
65a06f06 2278 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR		 2279 break;
2280 case 1:
65a06f06 2281 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR		 2282 break;
2283 }
1da177e4
LT		 2284 } else
2285 printk(KERN_INFO "md: no rdev superblock!\n");
2286}
2287
5e56341d 2288static void md_print_devices(void)
1da177e4 2289{
159ec1fc 2290 struct list_head *tmp;
3cb03002 2291 struct md_rdev *rdev;
fd01b88c 2292 struct mddev *mddev;
1da177e4
LT		 2293 char b[BDEVNAME_SIZE];
2294
2295 printk("\n");
2296 printk("md: **********************************\n");
2297 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2298 printk("md: **********************************\n");
29ac4aa3 2299 for_each_mddev(mddev, tmp) {
1da177e4 2300
32a7627c
N		 2301 if (mddev->bitmap)
2302 bitmap_print_sb(mddev->bitmap);
2303 else
2304 printk("%s: ", mdname(mddev));
159ec1fc 2305 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 2306 printk("<%s>", bdevname(rdev->bdev,b));
2307 printk("\n");
2308
159ec1fc 2309 list_for_each_entry(rdev, &mddev->disks, same_set)
cd2ac932 2310 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2311 }
2312 printk("md: **********************************\n");
2313 printk("\n");
2314}
2315
2316
fd01b88c 2317static void sync_sbs(struct mddev * mddev, int nospares)
1da177e4 2318{
42543769
N		 2319 /* Update each superblock (in-memory image), but
2320 * if we are allowed to, skip spares which already
2321 * have the right event counter, or have one earlier
2322 * (which would mean they aren't being marked as dirty
2323 * with the rest of the array)
2324 */
3cb03002 2325 struct md_rdev *rdev;
159ec1fc 2326 list_for_each_entry(rdev, &mddev->disks, same_set) {
42543769
N		 2327 if (rdev->sb_events == mddev->events ||
2328 (nospares &&
2329 rdev->raid_disk < 0 &&
42543769
N		 2330 rdev->sb_events+1 == mddev->events)) {
2331 /* Don't update this superblock */
2332 rdev->sb_loaded = 2;
2333 } else {
076f968b 2334 sync_super(mddev, rdev);
42543769
N		 2335 rdev->sb_loaded = 1;
2336 }
1da177e4
LT		 2337 }
2338}
2339
fd01b88c 2340static void md_update_sb(struct mddev * mddev, int force_change)
1da177e4 2341{
3cb03002 2342 struct md_rdev *rdev;
06d91a5f 2343 int sync_req;
42543769 2344 int nospares = 0;
2699b672 2345 int any_badblocks_changed = 0;
1da177e4 2346
1da177e4 2347repeat:
3a3a5ddb
N		 2348 /* First make sure individual recovery_offsets are correct */
2349 list_for_each_entry(rdev, &mddev->disks, same_set) {
2350 if (rdev->raid_disk >= 0 &&
2351 mddev->delta_disks >= 0 &&
2352 !test_bit(In_sync, &rdev->flags) &&
2353 mddev->curr_resync_completed > rdev->recovery_offset)
2354 rdev->recovery_offset = mddev->curr_resync_completed;
2355
2356 }
bd52b746 2357 if (!mddev->persistent) {
070dc6dd 2358 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2359 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2360 if (!mddev->external) {
d97a41dc 2361 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
de393cde
N		 2362 list_for_each_entry(rdev, &mddev->disks, same_set) {
2363 if (rdev->badblocks.changed) {
2364 md_ack_all_badblocks(&rdev->badblocks);
2365 md_error(mddev, rdev);
2366 }
2367 clear_bit(Blocked, &rdev->flags);
2368 clear_bit(BlockedBadBlocks, &rdev->flags);
2369 wake_up(&rdev->blocked_wait);
2370 }
2371 }
3a3a5ddb
N		 2372 wake_up(&mddev->sb_wait);
2373 return;
2374 }
2375
a9701a30 2376 spin_lock_irq(&mddev->write_lock);
84692195 2377
3a3a5ddb
N		 2378 mddev->utime = get_seconds();
2379
850b2b42
N		 2380 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2381 force_change = 1;
2382 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2383 /* just a clean<-> dirty transition, possibly leave spares alone,
2384 * though if events isn't the right even/odd, we will have to do
2385 * spares after all
2386 */
2387 nospares = 1;
2388 if (force_change)
2389 nospares = 0;
2390 if (mddev->degraded)
84692195
N		 2391 /* If the array is degraded, then skipping spares is both
2392 * dangerous and fairly pointless.
2393 * Dangerous because a device that was removed from the array
2394 * might have a event_count that still looks up-to-date,
2395 * so it can be re-added without a resync.
2396 * Pointless because if there are any spares to skip,
2397 * then a recovery will happen and soon that array won't
2398 * be degraded any more and the spare can go back to sleep then.
2399 */
850b2b42 2400 nospares = 0;
84692195 2401
06d91a5f 2402 sync_req = mddev->in_sync;
42543769
N		 2403
2404 /* If this is just a dirty<->clean transition, and the array is clean
2405 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2406 if (nospares
42543769 2407 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2408 && mddev->can_decrease_events
2409 && mddev->events != 1) {
42543769 2410 mddev->events--;
a8707c08
N		 2411 mddev->can_decrease_events = 0;
2412 } else {
42543769
N		 2413 /* otherwise we have to go forward and ... */
2414 mddev->events ++;
a8707c08 2415 mddev->can_decrease_events = nospares;
42543769 2416 }
1da177e4
LT		 2417
2418 if (!mddev->events) {
2419 /*
2420 * oops, this 64-bit counter should never wrap.
2421 * Either we are in around ~1 trillion A.C., assuming
2422 * 1 reboot per second, or we have a bug:
2423 */
2424 MD_BUG();
2425 mddev->events --;
2426 }
2699b672 2427
de393cde 2428 list_for_each_entry(rdev, &mddev->disks, same_set) {
2699b672
N		 2429 if (rdev->badblocks.changed)
2430 any_badblocks_changed++;
de393cde
N		 2431 if (test_bit(Faulty, &rdev->flags))
2432 set_bit(FaultRecorded, &rdev->flags);
2433 }
2699b672 2434
e691063a 2435 sync_sbs(mddev, nospares);
a9701a30 2436 spin_unlock_irq(&mddev->write_lock);
1da177e4 2437
36a4e1fe
N		 2438 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2439 mdname(mddev), mddev->in_sync);
1da177e4 2440
4ad13663 2441 bitmap_update_sb(mddev->bitmap);
159ec1fc 2442 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 2443 char b[BDEVNAME_SIZE];
36a4e1fe 2444
42543769
N		 2445 if (rdev->sb_loaded != 1)
2446 continue; /* no noise on spare devices */
1da177e4 2447
d70ed2e4
AW		 2448 if (!test_bit(Faulty, &rdev->flags) &&
2449 rdev->saved_raid_disk == -1) {
7bfa19f2 2450 md_super_write(mddev,rdev,
0f420358 2451 rdev->sb_start, rdev->sb_size,
7bfa19f2 2452 rdev->sb_page);
36a4e1fe
N		 2453 pr_debug("md: (write) %s's sb offset: %llu\n",
2454 bdevname(rdev->bdev, b),
2455 (unsigned long long)rdev->sb_start);
42543769 2456 rdev->sb_events = mddev->events;
2699b672
N		 2457 if (rdev->badblocks.size) {
2458 md_super_write(mddev, rdev,
2459 rdev->badblocks.sector,
2460 rdev->badblocks.size << 9,
2461 rdev->bb_page);
2462 rdev->badblocks.size = 0;
2463 }
7bfa19f2 2464
d70ed2e4 2465 } else if (test_bit(Faulty, &rdev->flags))
36a4e1fe
N		 2466 pr_debug("md: %s (skipping faulty)\n",
2467 bdevname(rdev->bdev, b));
d70ed2e4
AW		 2468 else
2469 pr_debug("(skipping incremental s/r ");
2470
7bfa19f2 2471 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2472 /* only need to write one superblock... */
2473 break;
2474 }
a9701a30 2475 md_super_wait(mddev);
850b2b42 2476 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2477
a9701a30 2478 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2479 if (mddev->in_sync != sync_req ||
2480 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2481 /* have to write it out again */
a9701a30 2482 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2483 goto repeat;
2484 }
850b2b42 2485 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2486 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2487 wake_up(&mddev->sb_wait);
acb180b0
N		 2488 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2489 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2490
de393cde
N		 2491 list_for_each_entry(rdev, &mddev->disks, same_set) {
2492 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2493 clear_bit(Blocked, &rdev->flags);
2494
2495 if (any_badblocks_changed)
2699b672 2496 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2497 clear_bit(BlockedBadBlocks, &rdev->flags);
2498 wake_up(&rdev->blocked_wait);
2499 }
1da177e4
LT		 2500}
2501
7f6ce769 2502/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2503 * We want to accept with case. For this we use cmd_match.
2504 */
2505static int cmd_match(const char *cmd, const char *str)
2506{
2507 /* See if cmd, written into a sysfs file, matches
2508 * str. They must either be the same, or cmd can
2509 * have a trailing newline
2510 */
2511 while (*cmd && *str && *cmd == *str) {
2512 cmd++;
2513 str++;
2514 }
2515 if (*cmd == '\n')
2516 cmd++;
2517 if (*str || *cmd)
2518 return 0;
2519 return 1;
2520}
2521
86e6ffdd
N		 2522struct rdev_sysfs_entry {
2523 struct attribute attr;
3cb03002
N		 2524 ssize_t (*show)(struct md_rdev *, char *);
2525 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2526};
2527
2528static ssize_t
3cb03002 2529state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2530{
2531 char *sep = "";
20a49ff6 2532 size_t len = 0;
86e6ffdd 2533
de393cde
N		 2534 if (test_bit(Faulty, &rdev->flags) ||
2535 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2536 len+= sprintf(page+len, "%sfaulty",sep);
2537 sep = ",";
2538 }
b2d444d7 2539 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2540 len += sprintf(page+len, "%sin_sync",sep);
2541 sep = ",";
2542 }
f655675b
N		 2543 if (test_bit(WriteMostly, &rdev->flags)) {
2544 len += sprintf(page+len, "%swrite_mostly",sep);
2545 sep = ",";
2546 }
de393cde 2547 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N		 2548 (rdev->badblocks.unacked_exist
2549 && !test_bit(Faulty, &rdev->flags))) {
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;
af8a2434
N		 4492 spin_lock(&all_mddevs_lock);
4493 if (list_empty(&mddev->all_mddevs)) {
4494 spin_unlock(&all_mddevs_lock);
4495 return -EBUSY;
4496 }
4497 mddev_get(mddev);
4498 spin_unlock(&all_mddevs_lock);
4499
5dc5cf7d
IM		 4500 rv = mddev_lock(mddev);
4501 if (!rv) {
4502 rv = entry->show(mddev, page);
4503 mddev_unlock(mddev);
4504 }
af8a2434 4505 mddev_put(mddev);
96de1e66 4506 return rv;
eae1701f
N		 4507}
4508
4509static ssize_t
4510md_attr_store(struct kobject *kobj, struct attribute *attr,
4511 const char *page, size_t length)
4512{
4513 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4514 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4515 ssize_t rv;
eae1701f
N		 4516
4517 if (!entry->store)
4518 return -EIO;
67463acb
N		 4519 if (!capable(CAP_SYS_ADMIN))
4520 return -EACCES;
af8a2434
N		 4521 spin_lock(&all_mddevs_lock);
4522 if (list_empty(&mddev->all_mddevs)) {
4523 spin_unlock(&all_mddevs_lock);
4524 return -EBUSY;
4525 }
4526 mddev_get(mddev);
4527 spin_unlock(&all_mddevs_lock);
5dc5cf7d
IM		 4528 rv = mddev_lock(mddev);
4529 if (!rv) {
4530 rv = entry->store(mddev, page, length);
4531 mddev_unlock(mddev);
4532 }
af8a2434 4533 mddev_put(mddev);
96de1e66 4534 return rv;
eae1701f
N		 4535}
4536
4537static void md_free(struct kobject *ko)
4538{
fd01b88c 4539 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4540
4541 if (mddev->sysfs_state)
4542 sysfs_put(mddev->sysfs_state);
4543
4544 if (mddev->gendisk) {
4545 del_gendisk(mddev->gendisk);
4546 put_disk(mddev->gendisk);
4547 }
4548 if (mddev->queue)
4549 blk_cleanup_queue(mddev->queue);
4550
eae1701f
N		 4551 kfree(mddev);
4552}
4553
52cf25d0 4554static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4555 .show = md_attr_show,
4556 .store = md_attr_store,
4557};
4558static struct kobj_type md_ktype = {
4559 .release = md_free,
4560 .sysfs_ops = &md_sysfs_ops,
4561 .default_attrs = md_default_attrs,
4562};
4563
1da177e4
LT		 4564int mdp_major = 0;
4565
5fd3a17e
DW		 4566static void mddev_delayed_delete(struct work_struct *ws)
4567{
fd01b88c 4568 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4569
43a70507 4570 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4571 kobject_del(&mddev->kobj);
4572 kobject_put(&mddev->kobj);
4573}
4574
efeb53c0 4575static int md_alloc(dev_t dev, char *name)
1da177e4 4576{
48c9c27b 4577 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4578 struct mddev *mddev = mddev_find(dev);
1da177e4 4579 struct gendisk *disk;
efeb53c0
N		 4580 int partitioned;
4581 int shift;
4582 int unit;
3830c62f 4583 int error;
1da177e4
LT		 4584
4585 if (!mddev)
efeb53c0
N		 4586 return -ENODEV;
4587
4588 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4589 shift = partitioned ? MdpMinorShift : 0;
4590 unit = MINOR(mddev->unit) >> shift;
1da177e4 4591
e804ac78
TH		 4592 /* wait for any previous instance of this device to be
4593 * completely removed (mddev_delayed_delete).
d3374825 4594 */
e804ac78 4595 flush_workqueue(md_misc_wq);
d3374825 4596
48c9c27b 4597 mutex_lock(&disks_mutex);
0909dc44
N		 4598 error = -EEXIST;
4599 if (mddev->gendisk)
4600 goto abort;
efeb53c0
N		 4601
4602 if (name) {
4603 /* Need to ensure that 'name' is not a duplicate.
4604 */
fd01b88c 4605 struct mddev *mddev2;
efeb53c0
N		 4606 spin_lock(&all_mddevs_lock);
4607
4608 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4609 if (mddev2->gendisk &&
4610 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4611 spin_unlock(&all_mddevs_lock);
0909dc44 4612 goto abort;
efeb53c0
N		 4613 }
4614 spin_unlock(&all_mddevs_lock);
1da177e4 4615 }
8b765398 4616
0909dc44 4617 error = -ENOMEM;
8b765398 4618 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4619 if (!mddev->queue)
4620 goto abort;
409c57f3
N		 4621 mddev->queue->queuedata = mddev;
4622
409c57f3 4623 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4624
1da177e4
LT		 4625 disk = alloc_disk(1 << shift);
4626 if (!disk) {
8b765398
N		 4627 blk_cleanup_queue(mddev->queue);
4628 mddev->queue = NULL;
0909dc44 4629 goto abort;
1da177e4 4630 }
efeb53c0 4631 disk->major = MAJOR(mddev->unit);
1da177e4 4632 disk->first_minor = unit << shift;
efeb53c0
N		 4633 if (name)
4634 strcpy(disk->disk_name, name);
4635 else if (partitioned)
1da177e4 4636 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4637 else
1da177e4 4638 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4639 disk->fops = &md_fops;
4640 disk->private_data = mddev;
4641 disk->queue = mddev->queue;
b0140891 4642 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4643 /* Allow extended partitions. This makes the
d3374825 4644 * 'mdp' device redundant, but we can't really
92850bbd
N		 4645 * remove it now.
4646 */
4647 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4648 mddev->gendisk = disk;
b0140891
N		 4649 /* As soon as we call add_disk(), another thread could get
4650 * through to md_open, so make sure it doesn't get too far
4651 */
4652 mutex_lock(&mddev->open_mutex);
4653 add_disk(disk);
4654
ed9e1982
TH		 4655 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4656 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4657 if (error) {
4658 /* This isn't possible, but as kobject_init_and_add is marked
4659 * __must_check, we must do something with the result
4660 */
5e55e2f5
N		 4661 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4662 disk->disk_name);
0909dc44
N		 4663 error = 0;
4664 }
00bcb4ac
N		 4665 if (mddev->kobj.sd &&
4666 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4667 printk(KERN_DEBUG "pointless warning\n");
b0140891 4668 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4669 abort:
4670 mutex_unlock(&disks_mutex);
00bcb4ac 4671 if (!error && mddev->kobj.sd) {
3830c62f 4672 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4673 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4674 }
d3374825 4675 mddev_put(mddev);
0909dc44 4676 return error;
efeb53c0
N		 4677}
4678
4679static struct kobject *md_probe(dev_t dev, int *part, void *data)
4680{
4681 md_alloc(dev, NULL);
1da177e4
LT		 4682 return NULL;
4683}
4684
efeb53c0
N		 4685static int add_named_array(const char *val, struct kernel_param *kp)
4686{
4687 /* val must be "md_*" where * is not all digits.
4688 * We allocate an array with a large free minor number, and
4689 * set the name to val. val must not already be an active name.
4690 */
4691 int len = strlen(val);
4692 char buf[DISK_NAME_LEN];
4693
4694 while (len && val[len-1] == '\n')
4695 len--;
4696 if (len >= DISK_NAME_LEN)
4697 return -E2BIG;
4698 strlcpy(buf, val, len+1);
4699 if (strncmp(buf, "md_", 3) != 0)
4700 return -EINVAL;
4701 return md_alloc(0, buf);
4702}
4703
1da177e4
LT		 4704static void md_safemode_timeout(unsigned long data)
4705{
fd01b88c 4706 struct mddev *mddev = (struct mddev *) data;
1da177e4 4707
0fd62b86
NB		 4708 if (!atomic_read(&mddev->writes_pending)) {
4709 mddev->safemode = 1;
4710 if (mddev->external)
00bcb4ac 4711 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4712 }
1da177e4
LT		 4713 md_wakeup_thread(mddev->thread);
4714}
4715
6ff8d8ec 4716static int start_dirty_degraded;
1da177e4 4717
fd01b88c 4718int md_run(struct mddev *mddev)
1da177e4 4719{
2604b703 4720 int err;
3cb03002 4721 struct md_rdev *rdev;
84fc4b56 4722 struct md_personality *pers;
1da177e4 4723
a757e64c
N		 4724 if (list_empty(&mddev->disks))
4725 /* cannot run an array with no devices.. */
1da177e4 4726 return -EINVAL;
1da177e4
LT		 4727
4728 if (mddev->pers)
4729 return -EBUSY;
bb4f1e9d
N		 4730 /* Cannot run until previous stop completes properly */
4731 if (mddev->sysfs_active)
4732 return -EBUSY;
b6eb127d 4733
1da177e4
LT		 4734 /*
4735 * Analyze all RAID superblock(s)
4736 */
1ec4a939
N		 4737 if (!mddev->raid_disks) {
4738 if (!mddev->persistent)
4739 return -EINVAL;
a757e64c 4740 analyze_sbs(mddev);
1ec4a939 4741 }
1da177e4 4742
d9d166c2
N		 4743 if (mddev->level != LEVEL_NONE)
4744 request_module("md-level-%d", mddev->level);
4745 else if (mddev->clevel[0])
4746 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4747
4748 /*
4749 * Drop all container device buffers, from now on
4750 * the only valid external interface is through the md
4751 * device.
1da177e4 4752 */
159ec1fc 4753 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4754 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4755 continue;
4756 sync_blockdev(rdev->bdev);
f98393a6 4757 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4758
4759 /* perform some consistency tests on the device.
4760 * We don't want the data to overlap the metadata,
58c0fed4 4761 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4762 */
a6ff7e08
JB		 4763 if (rdev->meta_bdev) {
4764 /* Nothing to check */;
4765 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4766 if (mddev->dev_sectors &&
4767 rdev->data_offset + mddev->dev_sectors
0f420358 4768 > rdev->sb_start) {
f0d76d70
N		 4769 printk("md: %s: data overlaps metadata\n",
4770 mdname(mddev));
4771 return -EINVAL;
4772 }
4773 } else {
0f420358 4774 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4775 > rdev->data_offset) {
4776 printk("md: %s: metadata overlaps data\n",
4777 mdname(mddev));
4778 return -EINVAL;
4779 }
4780 }
00bcb4ac 4781 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4782 }
4783
a167f663 4784 if (mddev->bio_set == NULL)
a519b26d 4785 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
fd01b88c 4786 sizeof(struct mddev *));
a167f663 4787
1da177e4 4788 spin_lock(&pers_lock);
d9d166c2 4789 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4790 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4791 spin_unlock(&pers_lock);
d9d166c2
N		 4792 if (mddev->level != LEVEL_NONE)
4793 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4794 mddev->level);
4795 else
4796 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4797 mddev->clevel);
1da177e4
LT		 4798 return -EINVAL;
4799 }
2604b703 4800 mddev->pers = pers;
1da177e4 4801 spin_unlock(&pers_lock);
34817e8c
N		 4802 if (mddev->level != pers->level) {
4803 mddev->level = pers->level;
4804 mddev->new_level = pers->level;
4805 }
d9d166c2 4806 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4807
f6705578 4808 if (mddev->reshape_position != MaxSector &&
63c70c4f 4809 pers->start_reshape == NULL) {
f6705578
N		 4810 /* This personality cannot handle reshaping... */
4811 mddev->pers = NULL;
4812 module_put(pers->owner);
4813 return -EINVAL;
4814 }
4815
7dd5e7c3
N		 4816 if (pers->sync_request) {
4817 /* Warn if this is a potentially silly
4818 * configuration.
4819 */
4820 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 4821 struct md_rdev *rdev2;
7dd5e7c3 4822 int warned = 0;
159ec1fc
CR		 4823
4824 list_for_each_entry(rdev, &mddev->disks, same_set)
4825 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4826 if (rdev < rdev2 &&
4827 rdev->bdev->bd_contains ==
4828 rdev2->bdev->bd_contains) {
4829 printk(KERN_WARNING
4830 "%s: WARNING: %s appears to be"
4831 " on the same physical disk as"
4832 " %s.\n",
4833 mdname(mddev),
4834 bdevname(rdev->bdev,b),
4835 bdevname(rdev2->bdev,b2));
4836 warned = 1;
4837 }
4838 }
159ec1fc 4839
7dd5e7c3
N		 4840 if (warned)
4841 printk(KERN_WARNING
4842 "True protection against single-disk"
4843 " failure might be compromised.\n");
4844 }
4845
657390d2 4846 mddev->recovery = 0;
58c0fed4
AN		 4847 /* may be over-ridden by personality */
4848 mddev->resync_max_sectors = mddev->dev_sectors;
4849
6ff8d8ec 4850 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4851
0f9552b5 4852 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4853 mddev->ro = 2; /* read-only, but switch on first write */
4854
b15c2e57 4855 err = mddev->pers->run(mddev);
13e53df3
AN		 4856 if (err)
4857 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4858 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4859 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4860 " but 'external_size' not in effect?\n", __func__);
4861 printk(KERN_ERR
4862 "md: invalid array_size %llu > default size %llu\n",
4863 (unsigned long long)mddev->array_sectors / 2,
4864 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4865 err = -EINVAL;
4866 mddev->pers->stop(mddev);
4867 }
4868 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4869 err = bitmap_create(mddev);
4870 if (err) {
4871 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4872 mdname(mddev), err);
4873 mddev->pers->stop(mddev);
4874 }
4875 }
1da177e4 4876 if (err) {
1da177e4
LT		 4877 module_put(mddev->pers->owner);
4878 mddev->pers = NULL;
32a7627c
N		 4879 bitmap_destroy(mddev);
4880 return err;
1da177e4 4881 }
5e55e2f5 4882 if (mddev->pers->sync_request) {
00bcb4ac
N		 4883 if (mddev->kobj.sd &&
4884 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4885 printk(KERN_WARNING
4886 "md: cannot register extra attributes for %s\n",
4887 mdname(mddev));
00bcb4ac 4888 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4889 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4890 mddev->ro = 0;
4891
1da177e4 4892 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4893 atomic_set(&mddev->max_corr_read_errors,
4894 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4895 mddev->safemode = 0;
4896 mddev->safemode_timer.function = md_safemode_timeout;
4897 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4898 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4899 mddev->in_sync = 1;
0ca69886
N		 4900 smp_wmb();
4901 mddev->ready = 1;
159ec1fc 4902 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 4903 if (rdev->raid_disk >= 0)
4904 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 4905 /* failure here is OK */;
1da177e4
LT		 4906
4907 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4908
850b2b42
N		 4909 if (mddev->flags)
4910 md_update_sb(mddev, 0);
1da177e4 4911
d7603b7e 4912 md_new_event(mddev);
00bcb4ac
N		 4913 sysfs_notify_dirent_safe(mddev->sysfs_state);
4914 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4915 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4916 return 0;
4917}
390ee602 4918EXPORT_SYMBOL_GPL(md_run);
1da177e4 4919
fd01b88c 4920static int do_md_run(struct mddev *mddev)
fe60b014
N		 4921{
4922 int err;
4923
4924 err = md_run(mddev);
4925 if (err)
4926 goto out;
69e51b44
N		 4927 err = bitmap_load(mddev);
4928 if (err) {
4929 bitmap_destroy(mddev);
4930 goto out;
4931 }
0fd018af
JB		 4932
4933 md_wakeup_thread(mddev->thread);
4934 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4935
fe60b014
N		 4936 set_capacity(mddev->gendisk, mddev->array_sectors);
4937 revalidate_disk(mddev->gendisk);
f0b4f7e2 4938 mddev->changed = 1;
fe60b014
N		 4939 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4940out:
4941 return err;
4942}
4943
fd01b88c 4944static int restart_array(struct mddev *mddev)
1da177e4
LT		 4945{
4946 struct gendisk *disk = mddev->gendisk;
1da177e4 4947
80fab1d7 4948 /* Complain if it has no devices */
1da177e4 4949 if (list_empty(&mddev->disks))
80fab1d7
AN		 4950 return -ENXIO;
4951 if (!mddev->pers)
4952 return -EINVAL;
4953 if (!mddev->ro)
4954 return -EBUSY;
4955 mddev->safemode = 0;
4956 mddev->ro = 0;
4957 set_disk_ro(disk, 0);
4958 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4959 mdname(mddev));
4960 /* Kick recovery or resync if necessary */
4961 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4962 md_wakeup_thread(mddev->thread);
4963 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 4964 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 4965 return 0;
1da177e4
LT		 4966}
4967
acc55e22
N		 4968/* similar to deny_write_access, but accounts for our holding a reference
4969 * to the file ourselves */
4970static int deny_bitmap_write_access(struct file * file)
4971{
4972 struct inode *inode = file->f_mapping->host;
4973
4974 spin_lock(&inode->i_lock);
4975 if (atomic_read(&inode->i_writecount) > 1) {
4976 spin_unlock(&inode->i_lock);
4977 return -ETXTBSY;
4978 }
4979 atomic_set(&inode->i_writecount, -1);
4980 spin_unlock(&inode->i_lock);
4981
4982 return 0;
4983}
4984
43a70507 4985void restore_bitmap_write_access(struct file *file)
acc55e22
N		 4986{
4987 struct inode *inode = file->f_mapping->host;
4988
4989 spin_lock(&inode->i_lock);
4990 atomic_set(&inode->i_writecount, 1);
4991 spin_unlock(&inode->i_lock);
4992}
4993
fd01b88c 4994static void md_clean(struct mddev *mddev)
6177b472
N		 4995{
4996 mddev->array_sectors = 0;
4997 mddev->external_size = 0;
4998 mddev->dev_sectors = 0;
4999 mddev->raid_disks = 0;
5000 mddev->recovery_cp = 0;
5001 mddev->resync_min = 0;
5002 mddev->resync_max = MaxSector;
5003 mddev->reshape_position = MaxSector;
5004 mddev->external = 0;
5005 mddev->persistent = 0;
5006 mddev->level = LEVEL_NONE;
5007 mddev->clevel[0] = 0;
5008 mddev->flags = 0;
5009 mddev->ro = 0;
5010 mddev->metadata_type[0] = 0;
5011 mddev->chunk_sectors = 0;
5012 mddev->ctime = mddev->utime = 0;
5013 mddev->layout = 0;
5014 mddev->max_disks = 0;
5015 mddev->events = 0;
a8707c08 5016 mddev->can_decrease_events = 0;
6177b472
N		 5017 mddev->delta_disks = 0;
5018 mddev->new_level = LEVEL_NONE;
5019 mddev->new_layout = 0;
5020 mddev->new_chunk_sectors = 0;
5021 mddev->curr_resync = 0;
5022 mddev->resync_mismatches = 0;
5023 mddev->suspend_lo = mddev->suspend_hi = 0;
5024 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5025 mddev->recovery = 0;
5026 mddev->in_sync = 0;
f0b4f7e2 5027 mddev->changed = 0;
6177b472 5028 mddev->degraded = 0;
6177b472
N		 5029 mddev->safemode = 0;
5030 mddev->bitmap_info.offset = 0;
5031 mddev->bitmap_info.default_offset = 0;
5032 mddev->bitmap_info.chunksize = 0;
5033 mddev->bitmap_info.daemon_sleep = 0;
5034 mddev->bitmap_info.max_write_behind = 0;
5035}
5036
fd01b88c 5037static void __md_stop_writes(struct mddev *mddev)
a047e125
N		 5038{
5039 if (mddev->sync_thread) {
5040 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5041 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 5042 reap_sync_thread(mddev);
a047e125
N		 5043 }
5044
5045 del_timer_sync(&mddev->safemode_timer);
5046
5047 bitmap_flush(mddev);
5048 md_super_wait(mddev);
5049
5050 if (!mddev->in_sync || mddev->flags) {
5051 /* mark array as shutdown cleanly */
5052 mddev->in_sync = 1;
5053 md_update_sb(mddev, 1);
5054 }
5055}
defad61a 5056
fd01b88c 5057void md_stop_writes(struct mddev *mddev)
defad61a
N		 5058{
5059 mddev_lock(mddev);
5060 __md_stop_writes(mddev);
5061 mddev_unlock(mddev);
5062}
390ee602 5063EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5064
fd01b88c 5065void md_stop(struct mddev *mddev)
6177b472 5066{
0ca69886 5067 mddev->ready = 0;
6177b472
N		 5068 mddev->pers->stop(mddev);
5069 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5070 mddev->to_remove = &md_redundancy_group;
5071 module_put(mddev->pers->owner);
5072 mddev->pers = NULL;
cca9cf90 5073 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5074}
390ee602 5075EXPORT_SYMBOL_GPL(md_stop);
6177b472 5076
fd01b88c 5077static int md_set_readonly(struct mddev *mddev, int is_open)
a4bd82d0
N		 5078{
5079 int err = 0;
5080 mutex_lock(&mddev->open_mutex);
5081 if (atomic_read(&mddev->openers) > is_open) {
5082 printk("md: %s still in use.\n",mdname(mddev));
5083 err = -EBUSY;
5084 goto out;
5085 }
5086 if (mddev->pers) {
defad61a 5087 __md_stop_writes(mddev);
a4bd82d0
N		 5088
5089 err = -ENXIO;
5090 if (mddev->ro==1)
5091 goto out;
5092 mddev->ro = 1;
5093 set_disk_ro(mddev->gendisk, 1);
5094 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5095 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5096 err = 0;
5097 }
5098out:
5099 mutex_unlock(&mddev->open_mutex);
5100 return err;
5101}
5102
9e653b63
N		 5103/* mode:
5104 * 0 - completely stop and dis-assemble array
9e653b63
N		 5105 * 2 - stop but do not disassemble array
5106 */
fd01b88c 5107static int do_md_stop(struct mddev * mddev, int mode, int is_open)
1da177e4 5108{
1da177e4 5109 struct gendisk *disk = mddev->gendisk;
3cb03002 5110 struct md_rdev *rdev;
1da177e4 5111
c8c00a69 5112 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 5113 if (atomic_read(&mddev->openers) > is_open ||
5114 mddev->sysfs_active) {
df5b20cf 5115 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5116 mutex_unlock(&mddev->open_mutex);
5117 return -EBUSY;
5118 }
1da177e4 5119
6e17b027 5120 if (mddev->pers) {
a4bd82d0
N		 5121 if (mddev->ro)
5122 set_disk_ro(disk, 0);
409c57f3 5123
defad61a 5124 __md_stop_writes(mddev);
a4bd82d0
N		 5125 md_stop(mddev);
5126 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5127 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5128
a4bd82d0 5129 /* tell userspace to handle 'inactive' */
00bcb4ac 5130 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5131
a4bd82d0 5132 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 5133 if (rdev->raid_disk >= 0)
5134 sysfs_unlink_rdev(mddev, rdev);
c4647292 5135
a4bd82d0 5136 set_capacity(disk, 0);
6e17b027 5137 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5138 mddev->changed = 1;
a4bd82d0 5139 revalidate_disk(disk);
0d4ca600 5140
a4bd82d0
N		 5141 if (mddev->ro)
5142 mddev->ro = 0;
6e17b027
N		 5143 } else
5144 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5145 /*
5146 * Free resources if final stop
5147 */
9e653b63 5148 if (mode == 0) {
1da177e4
LT		 5149 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5150
978f946b 5151 bitmap_destroy(mddev);
c3d9714e
N		 5152 if (mddev->bitmap_info.file) {
5153 restore_bitmap_write_access(mddev->bitmap_info.file);
5154 fput(mddev->bitmap_info.file);
5155 mddev->bitmap_info.file = NULL;
978f946b 5156 }
c3d9714e 5157 mddev->bitmap_info.offset = 0;
978f946b 5158
1da177e4
LT		 5159 export_array(mddev);
5160
6177b472 5161 md_clean(mddev);
934d9c23 5162 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5163 if (mddev->hold_active == UNTIL_STOP)
5164 mddev->hold_active = 0;
a4bd82d0 5165 }
3f9d99c1 5166 blk_integrity_unregister(disk);
d7603b7e 5167 md_new_event(mddev);
00bcb4ac 5168 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5169 return 0;
1da177e4
LT		 5170}
5171
fdee8ae4 5172#ifndef MODULE
fd01b88c 5173static void autorun_array(struct mddev *mddev)
1da177e4 5174{
3cb03002 5175 struct md_rdev *rdev;
1da177e4
LT		 5176 int err;
5177
a757e64c 5178 if (list_empty(&mddev->disks))
1da177e4 5179 return;
1da177e4
LT		 5180
5181 printk(KERN_INFO "md: running: ");
5182
159ec1fc 5183 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 5184 char b[BDEVNAME_SIZE];
5185 printk("<%s>", bdevname(rdev->bdev,b));
5186 }
5187 printk("\n");
5188
d710e138 5189 err = do_md_run(mddev);
1da177e4
LT		 5190 if (err) {
5191 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 5192 do_md_stop(mddev, 0, 0);
1da177e4
LT		 5193 }
5194}
5195
5196/*
5197 * lets try to run arrays based on all disks that have arrived
5198 * until now. (those are in pending_raid_disks)
5199 *
5200 * the method: pick the first pending disk, collect all disks with
5201 * the same UUID, remove all from the pending list and put them into
5202 * the 'same_array' list. Then order this list based on superblock
5203 * update time (freshest comes first), kick out 'old' disks and
5204 * compare superblocks. If everything's fine then run it.
5205 *
5206 * If "unit" is allocated, then bump its reference count
5207 */
5208static void autorun_devices(int part)
5209{
3cb03002 5210 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5211 struct mddev *mddev;
1da177e4
LT		 5212 char b[BDEVNAME_SIZE];
5213
5214 printk(KERN_INFO "md: autorun ...\n");
5215 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5216 int unit;
1da177e4 5217 dev_t dev;
ad01c9e3 5218 LIST_HEAD(candidates);
1da177e4 5219 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5220 struct md_rdev, same_set);
1da177e4
LT		 5221
5222 printk(KERN_INFO "md: considering %s ...\n",
5223 bdevname(rdev0->bdev,b));
5224 INIT_LIST_HEAD(&candidates);
159ec1fc 5225 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5226 if (super_90_load(rdev, rdev0, 0) >= 0) {
5227 printk(KERN_INFO "md: adding %s ...\n",
5228 bdevname(rdev->bdev,b));
5229 list_move(&rdev->same_set, &candidates);
5230 }
5231 /*
5232 * now we have a set of devices, with all of them having
5233 * mostly sane superblocks. It's time to allocate the
5234 * mddev.
5235 */
e8703fe1
N		 5236 if (part) {
5237 dev = MKDEV(mdp_major,
5238 rdev0->preferred_minor << MdpMinorShift);
5239 unit = MINOR(dev) >> MdpMinorShift;
5240 } else {
5241 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5242 unit = MINOR(dev);
5243 }
5244 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5245 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5246 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5247 break;
5248 }
1da177e4
LT		 5249
5250 md_probe(dev, NULL, NULL);
5251 mddev = mddev_find(dev);
9bbbca3a
NB		 5252 if (!mddev || !mddev->gendisk) {
5253 if (mddev)
5254 mddev_put(mddev);
5255 printk(KERN_ERR
1da177e4
LT		 5256 "md: cannot allocate memory for md drive.\n");
5257 break;
5258 }
5259 if (mddev_lock(mddev))
5260 printk(KERN_WARNING "md: %s locked, cannot run\n",
5261 mdname(mddev));
5262 else if (mddev->raid_disks || mddev->major_version
5263 || !list_empty(&mddev->disks)) {
5264 printk(KERN_WARNING
5265 "md: %s already running, cannot run %s\n",
5266 mdname(mddev), bdevname(rdev0->bdev,b));
5267 mddev_unlock(mddev);
5268 } else {
5269 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5270 mddev->persistent = 1;
159ec1fc 5271 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5272 list_del_init(&rdev->same_set);
5273 if (bind_rdev_to_array(rdev, mddev))
5274 export_rdev(rdev);
5275 }
5276 autorun_array(mddev);
5277 mddev_unlock(mddev);
5278 }
5279 /* on success, candidates will be empty, on error
5280 * it won't...
5281 */
159ec1fc 5282 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5283 list_del_init(&rdev->same_set);
1da177e4 5284 export_rdev(rdev);
4b80991c 5285 }
1da177e4
LT		 5286 mddev_put(mddev);
5287 }
5288 printk(KERN_INFO "md: ... autorun DONE.\n");
5289}
fdee8ae4 5290#endif /* !MODULE */
1da177e4 5291
1da177e4
LT		 5292static int get_version(void __user * arg)
5293{
5294 mdu_version_t ver;
5295
5296 ver.major = MD_MAJOR_VERSION;
5297 ver.minor = MD_MINOR_VERSION;
5298 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5299
5300 if (copy_to_user(arg, &ver, sizeof(ver)))
5301 return -EFAULT;
5302
5303 return 0;
5304}
5305
fd01b88c 5306static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5307{
5308 mdu_array_info_t info;
a9f326eb 5309 int nr,working,insync,failed,spare;
3cb03002 5310 struct md_rdev *rdev;
1da177e4 5311
a9f326eb 5312 nr=working=insync=failed=spare=0;
159ec1fc 5313 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 5314 nr++;
b2d444d7 5315 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5316 failed++;
5317 else {
5318 working++;
b2d444d7 5319 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5320 insync++;
1da177e4
LT		 5321 else
5322 spare++;
5323 }
5324 }
5325
5326 info.major_version = mddev->major_version;
5327 info.minor_version = mddev->minor_version;
5328 info.patch_version = MD_PATCHLEVEL_VERSION;
5329 info.ctime = mddev->ctime;
5330 info.level = mddev->level;
58c0fed4
AN		 5331 info.size = mddev->dev_sectors / 2;
5332 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5333 info.size = -1;
1da177e4
LT		 5334 info.nr_disks = nr;
5335 info.raid_disks = mddev->raid_disks;
5336 info.md_minor = mddev->md_minor;
5337 info.not_persistent= !mddev->persistent;
5338
5339 info.utime = mddev->utime;
5340 info.state = 0;
5341 if (mddev->in_sync)
5342 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5343 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5344 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5345 info.active_disks = insync;
1da177e4
LT		 5346 info.working_disks = working;
5347 info.failed_disks = failed;
5348 info.spare_disks = spare;
5349
5350 info.layout = mddev->layout;
9d8f0363 5351 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5352
5353 if (copy_to_user(arg, &info, sizeof(info)))
5354 return -EFAULT;
5355
5356 return 0;
5357}
5358
fd01b88c 5359static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5360{
5361 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5362 char *ptr, *buf = NULL;
5363 int err = -ENOMEM;
5364
b5470dc5
DW		 5365 if (md_allow_write(mddev))
5366 file = kmalloc(sizeof(*file), GFP_NOIO);
5367 else
5368 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5369
32a7627c
N		 5370 if (!file)
5371 goto out;
5372
5373 /* bitmap disabled, zero the first byte and copy out */
5374 if (!mddev->bitmap || !mddev->bitmap->file) {
5375 file->pathname[0] = '\0';
5376 goto copy_out;
5377 }
5378
5379 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5380 if (!buf)
5381 goto out;
5382
6bcfd601
CH		 5383 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5384 if (IS_ERR(ptr))
32a7627c
N		 5385 goto out;
5386
5387 strcpy(file->pathname, ptr);
5388
5389copy_out:
5390 err = 0;
5391 if (copy_to_user(arg, file, sizeof(*file)))
5392 err = -EFAULT;
5393out:
5394 kfree(buf);
5395 kfree(file);
5396 return err;
5397}
5398
fd01b88c 5399static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5400{
5401 mdu_disk_info_t info;
3cb03002 5402 struct md_rdev *rdev;
1da177e4
LT		 5403
5404 if (copy_from_user(&info, arg, sizeof(info)))
5405 return -EFAULT;
5406
26ef379f 5407 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5408 if (rdev) {
5409 info.major = MAJOR(rdev->bdev->bd_dev);
5410 info.minor = MINOR(rdev->bdev->bd_dev);
5411 info.raid_disk = rdev->raid_disk;
5412 info.state = 0;
b2d444d7 5413 if (test_bit(Faulty, &rdev->flags))
1da177e4 5414 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5415 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5416 info.state |= (1<<MD_DISK_ACTIVE);
5417 info.state |= (1<<MD_DISK_SYNC);
5418 }
8ddf9efe
N		 5419 if (test_bit(WriteMostly, &rdev->flags))
5420 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5421 } else {
5422 info.major = info.minor = 0;
5423 info.raid_disk = -1;
5424 info.state = (1<<MD_DISK_REMOVED);
5425 }
5426
5427 if (copy_to_user(arg, &info, sizeof(info)))
5428 return -EFAULT;
5429
5430 return 0;
5431}
5432
fd01b88c 5433static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT		 5434{
5435 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5436 struct md_rdev *rdev;
1da177e4
LT		 5437 dev_t dev = MKDEV(info->major,info->minor);
5438
5439 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5440 return -EOVERFLOW;
5441
5442 if (!mddev->raid_disks) {
5443 int err;
5444 /* expecting a device which has a superblock */
5445 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5446 if (IS_ERR(rdev)) {
5447 printk(KERN_WARNING
5448 "md: md_import_device returned %ld\n",
5449 PTR_ERR(rdev));
5450 return PTR_ERR(rdev);
5451 }
5452 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5453 struct md_rdev *rdev0
5454 = list_entry(mddev->disks.next,
5455 struct md_rdev, same_set);
a9f326eb 5456 err = super_types[mddev->major_version]
1da177e4
LT		 5457 .load_super(rdev, rdev0, mddev->minor_version);
5458 if (err < 0) {
5459 printk(KERN_WARNING
5460 "md: %s has different UUID to %s\n",
5461 bdevname(rdev->bdev,b),
5462 bdevname(rdev0->bdev,b2));
5463 export_rdev(rdev);
5464 return -EINVAL;
5465 }
5466 }
5467 err = bind_rdev_to_array(rdev, mddev);
5468 if (err)
5469 export_rdev(rdev);
5470 return err;
5471 }
5472
5473 /*
5474 * add_new_disk can be used once the array is assembled
5475 * to add "hot spares". They must already have a superblock
5476 * written
5477 */
5478 if (mddev->pers) {
5479 int err;
5480 if (!mddev->pers->hot_add_disk) {
5481 printk(KERN_WARNING
5482 "%s: personality does not support diskops!\n",
5483 mdname(mddev));
5484 return -EINVAL;
5485 }
7b1e35f6
N		 5486 if (mddev->persistent)
5487 rdev = md_import_device(dev, mddev->major_version,
5488 mddev->minor_version);
5489 else
5490 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5491 if (IS_ERR(rdev)) {
5492 printk(KERN_WARNING
5493 "md: md_import_device returned %ld\n",
5494 PTR_ERR(rdev));
5495 return PTR_ERR(rdev);
5496 }
1a855a06 5497 /* set saved_raid_disk if appropriate */
41158c7e
N		 5498 if (!mddev->persistent) {
5499 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5500 info->raid_disk < mddev->raid_disks) {
41158c7e 5501 rdev->raid_disk = info->raid_disk;
bf572541
N		 5502 set_bit(In_sync, &rdev->flags);
5503 } else
41158c7e
N		 5504 rdev->raid_disk = -1;
5505 } else
5506 super_types[mddev->major_version].
5507 validate_super(mddev, rdev);
bedd86b7
N		 5508 if ((info->state & (1<<MD_DISK_SYNC)) &&
5509 (!test_bit(In_sync, &rdev->flags) ||
5510 rdev->raid_disk != info->raid_disk)) {
5511 /* This was a hot-add request, but events doesn't
5512 * match, so reject it.
5513 */
5514 export_rdev(rdev);
5515 return -EINVAL;
5516 }
5517
1a855a06
N		 5518 if (test_bit(In_sync, &rdev->flags))
5519 rdev->saved_raid_disk = rdev->raid_disk;
5520 else
5521 rdev->saved_raid_disk = -1;
41158c7e 5522
b2d444d7 5523 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5524 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5525 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5526 else
5527 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5528
1da177e4
LT		 5529 rdev->raid_disk = -1;
5530 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5531 if (!err && !mddev->pers->hot_remove_disk) {
5532 /* If there is hot_add_disk but no hot_remove_disk
5533 * then added disks for geometry changes,
5534 * and should be added immediately.
5535 */
5536 super_types[mddev->major_version].
5537 validate_super(mddev, rdev);
5538 err = mddev->pers->hot_add_disk(mddev, rdev);
5539 if (err)
5540 unbind_rdev_from_array(rdev);
5541 }
1da177e4
LT		 5542 if (err)
5543 export_rdev(rdev);
52664732 5544 else
00bcb4ac 5545 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5546
17571284 5547 md_update_sb(mddev, 1);
72a23c21
NB		 5548 if (mddev->degraded)
5549 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5550 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5551 if (!err)
5552 md_new_event(mddev);
005eca5e 5553 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5554 return err;
5555 }
5556
5557 /* otherwise, add_new_disk is only allowed
5558 * for major_version==0 superblocks
5559 */
5560 if (mddev->major_version != 0) {
5561 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5562 mdname(mddev));
5563 return -EINVAL;
5564 }
5565
5566 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5567 int err;
d710e138 5568 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5569 if (IS_ERR(rdev)) {
5570 printk(KERN_WARNING
5571 "md: error, md_import_device() returned %ld\n",
5572 PTR_ERR(rdev));
5573 return PTR_ERR(rdev);
5574 }
5575 rdev->desc_nr = info->number;
5576 if (info->raid_disk < mddev->raid_disks)
5577 rdev->raid_disk = info->raid_disk;
5578 else
5579 rdev->raid_disk = -1;
5580
1da177e4 5581 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5582 if (info->state & (1<<MD_DISK_SYNC))
5583 set_bit(In_sync, &rdev->flags);
1da177e4 5584
8ddf9efe
N		 5585 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5586 set_bit(WriteMostly, &rdev->flags);
5587
1da177e4
LT		 5588 if (!mddev->persistent) {
5589 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5590 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5591 } else
57b2caa3 5592 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5593 rdev->sectors = rdev->sb_start;
1da177e4 5594
2bf071bf
N		 5595 err = bind_rdev_to_array(rdev, mddev);
5596 if (err) {
5597 export_rdev(rdev);
5598 return err;
5599 }
1da177e4
LT		 5600 }
5601
5602 return 0;
5603}
5604
fd01b88c 5605static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5606{
5607 char b[BDEVNAME_SIZE];
3cb03002 5608 struct md_rdev *rdev;
1da177e4 5609
1da177e4
LT		 5610 rdev = find_rdev(mddev, dev);
5611 if (!rdev)
5612 return -ENXIO;
5613
5614 if (rdev->raid_disk >= 0)
5615 goto busy;
5616
5617 kick_rdev_from_array(rdev);
850b2b42 5618 md_update_sb(mddev, 1);
d7603b7e 5619 md_new_event(mddev);
1da177e4
LT		 5620
5621 return 0;
5622busy:
fdefa4d8 5623 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5624 bdevname(rdev->bdev,b), mdname(mddev));
5625 return -EBUSY;
5626}
5627
fd01b88c 5628static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5629{
5630 char b[BDEVNAME_SIZE];
5631 int err;
3cb03002 5632 struct md_rdev *rdev;
1da177e4
LT		 5633
5634 if (!mddev->pers)
5635 return -ENODEV;
5636
5637 if (mddev->major_version != 0) {
5638 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5639 " version-0 superblocks.\n",
5640 mdname(mddev));
5641 return -EINVAL;
5642 }
5643 if (!mddev->pers->hot_add_disk) {
5644 printk(KERN_WARNING
5645 "%s: personality does not support diskops!\n",
5646 mdname(mddev));
5647 return -EINVAL;
5648 }
5649
d710e138 5650 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5651 if (IS_ERR(rdev)) {
5652 printk(KERN_WARNING
5653 "md: error, md_import_device() returned %ld\n",
5654 PTR_ERR(rdev));
5655 return -EINVAL;
5656 }
5657
5658 if (mddev->persistent)
57b2caa3 5659 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5660 else
77304d2a 5661 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5662
8190e754 5663 rdev->sectors = rdev->sb_start;
1da177e4 5664
b2d444d7 5665 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5666 printk(KERN_WARNING
5667 "md: can not hot-add faulty %s disk to %s!\n",
5668 bdevname(rdev->bdev,b), mdname(mddev));
5669 err = -EINVAL;
5670 goto abort_export;
5671 }
b2d444d7 5672 clear_bit(In_sync, &rdev->flags);
1da177e4 5673 rdev->desc_nr = -1;
5842730d 5674 rdev->saved_raid_disk = -1;
2bf071bf
N		 5675 err = bind_rdev_to_array(rdev, mddev);
5676 if (err)
5677 goto abort_export;
1da177e4
LT		 5678
5679 /*
5680 * The rest should better be atomic, we can have disk failures
5681 * noticed in interrupt contexts ...
5682 */
5683
1da177e4
LT		 5684 rdev->raid_disk = -1;
5685
850b2b42 5686 md_update_sb(mddev, 1);
1da177e4
LT		 5687
5688 /*
5689 * Kick recovery, maybe this spare has to be added to the
5690 * array immediately.
5691 */
5692 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5693 md_wakeup_thread(mddev->thread);
d7603b7e 5694 md_new_event(mddev);
1da177e4
LT		 5695 return 0;
5696
1da177e4
LT		 5697abort_export:
5698 export_rdev(rdev);
5699 return err;
5700}
5701
fd01b88c 5702static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N		 5703{
5704 int err;
5705
36fa3063
N		 5706 if (mddev->pers) {
5707 if (!mddev->pers->quiesce)
5708 return -EBUSY;
5709 if (mddev->recovery || mddev->sync_thread)
5710 return -EBUSY;
5711 /* we should be able to change the bitmap.. */
5712 }
32a7627c 5713
32a7627c 5714
36fa3063
N		 5715 if (fd >= 0) {
5716 if (mddev->bitmap)
5717 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5718 mddev->bitmap_info.file = fget(fd);
32a7627c 5719
c3d9714e 5720 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5721 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5722 mdname(mddev));
5723 return -EBADF;
5724 }
5725
c3d9714e 5726 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5727 if (err) {
5728 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5729 mdname(mddev));
c3d9714e
N		 5730 fput(mddev->bitmap_info.file);
5731 mddev->bitmap_info.file = NULL;
36fa3063
N		 5732 return err;
5733 }
c3d9714e 5734 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5735 } else if (mddev->bitmap == NULL)
5736 return -ENOENT; /* cannot remove what isn't there */
5737 err = 0;
5738 if (mddev->pers) {
5739 mddev->pers->quiesce(mddev, 1);
69e51b44 5740 if (fd >= 0) {
36fa3063 5741 err = bitmap_create(mddev);
69e51b44
N		 5742 if (!err)
5743 err = bitmap_load(mddev);
5744 }
d7375ab3 5745 if (fd < 0 || err) {
36fa3063 5746 bitmap_destroy(mddev);
d7375ab3
N		 5747 fd = -1; /* make sure to put the file */
5748 }
36fa3063 5749 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5750 }
5751 if (fd < 0) {
c3d9714e
N		 5752 if (mddev->bitmap_info.file) {
5753 restore_bitmap_write_access(mddev->bitmap_info.file);
5754 fput(mddev->bitmap_info.file);
acc55e22 5755 }
c3d9714e 5756 mddev->bitmap_info.file = NULL;
36fa3063
N		 5757 }
5758
32a7627c
N		 5759 return err;
5760}
5761
1da177e4
LT		 5762/*
5763 * set_array_info is used two different ways
5764 * The original usage is when creating a new array.
5765 * In this usage, raid_disks is > 0 and it together with
5766 * level, size, not_persistent,layout,chunksize determine the
5767 * shape of the array.
5768 * This will always create an array with a type-0.90.0 superblock.
5769 * The newer usage is when assembling an array.
5770 * In this case raid_disks will be 0, and the major_version field is
5771 * use to determine which style super-blocks are to be found on the devices.
5772 * The minor and patch _version numbers are also kept incase the
5773 * super_block handler wishes to interpret them.
5774 */
fd01b88c 5775static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT		 5776{
5777
5778 if (info->raid_disks == 0) {
5779 /* just setting version number for superblock loading */
5780 if (info->major_version < 0 ||
50511da3 5781 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5782 super_types[info->major_version].name == NULL) {
5783 /* maybe try to auto-load a module? */
5784 printk(KERN_INFO
5785 "md: superblock version %d not known\n",
5786 info->major_version);
5787 return -EINVAL;
5788 }
5789 mddev->major_version = info->major_version;
5790 mddev->minor_version = info->minor_version;
5791 mddev->patch_version = info->patch_version;
3f9d7b0d 5792 mddev->persistent = !info->not_persistent;
cbd19983
N		 5793 /* ensure mddev_put doesn't delete this now that there
5794 * is some minimal configuration.
5795 */
5796 mddev->ctime = get_seconds();
1da177e4
LT		 5797 return 0;
5798 }
5799 mddev->major_version = MD_MAJOR_VERSION;
5800 mddev->minor_version = MD_MINOR_VERSION;
5801 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5802 mddev->ctime = get_seconds();
5803
5804 mddev->level = info->level;
17115e03 5805 mddev->clevel[0] = 0;
58c0fed4 5806 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5807 mddev->raid_disks = info->raid_disks;
5808 /* don't set md_minor, it is determined by which /dev/md* was
5809 * openned
5810 */
5811 if (info->state & (1<<MD_SB_CLEAN))
5812 mddev->recovery_cp = MaxSector;
5813 else
5814 mddev->recovery_cp = 0;
5815 mddev->persistent = ! info->not_persistent;
e691063a 5816 mddev->external = 0;
1da177e4
LT		 5817
5818 mddev->layout = info->layout;
9d8f0363 5819 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5820
5821 mddev->max_disks = MD_SB_DISKS;
5822
e691063a
N		 5823 if (mddev->persistent)
5824 mddev->flags = 0;
850b2b42 5825 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5826
c3d9714e
N		 5827 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5828 mddev->bitmap_info.offset = 0;
b2a2703c 5829
f6705578
N		 5830 mddev->reshape_position = MaxSector;
5831
1da177e4
LT		 5832 /*
5833 * Generate a 128 bit UUID
5834 */
5835 get_random_bytes(mddev->uuid, 16);
5836
f6705578 5837 mddev->new_level = mddev->level;
664e7c41 5838 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5839 mddev->new_layout = mddev->layout;
5840 mddev->delta_disks = 0;
5841
1da177e4
LT		 5842 return 0;
5843}
5844
fd01b88c 5845void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 5846{
b522adcd
DW		 5847 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5848
5849 if (mddev->external_size)
5850 return;
5851
1f403624
DW		 5852 mddev->array_sectors = array_sectors;
5853}
5854EXPORT_SYMBOL(md_set_array_sectors);
5855
fd01b88c 5856static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 5857{
3cb03002 5858 struct md_rdev *rdev;
a35b0d69 5859 int rv;
d71f9f88 5860 int fit = (num_sectors == 0);
a35b0d69
N		 5861
5862 if (mddev->pers->resize == NULL)
5863 return -EINVAL;
d71f9f88
AN		 5864 /* The "num_sectors" is the number of sectors of each device that
5865 * is used. This can only make sense for arrays with redundancy.
5866 * linear and raid0 always use whatever space is available. We can only
5867 * consider changing this number if no resync or reconstruction is
5868 * happening, and if the new size is acceptable. It must fit before the
0f420358 5869 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5870 * of each device. If num_sectors is zero, we find the largest size
5871 * that fits.
a35b0d69
N		 5872 */
5873 if (mddev->sync_thread)
5874 return -EBUSY;
dba034ee
N		 5875 if (mddev->bitmap)
5876 /* Sorry, cannot grow a bitmap yet, just remove it,
5877 * grow, and re-add.
5878 */
5879 return -EBUSY;
159ec1fc 5880 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5881 sector_t avail = rdev->sectors;
01ab5662 5882
d71f9f88
AN		 5883 if (fit && (num_sectors == 0 || num_sectors > avail))
5884 num_sectors = avail;
5885 if (avail < num_sectors)
a35b0d69
N		 5886 return -ENOSPC;
5887 }
d71f9f88 5888 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5889 if (!rv)
5890 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5891 return rv;
5892}
5893
fd01b88c 5894static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 5895{
5896 int rv;
5897 /* change the number of raid disks */
63c70c4f 5898 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5899 return -EINVAL;
5900 if (raid_disks <= 0 ||
233fca36 5901 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5902 return -EINVAL;
63c70c4f 5903 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5904 return -EBUSY;
63c70c4f
N		 5905 mddev->delta_disks = raid_disks - mddev->raid_disks;
5906
5907 rv = mddev->pers->check_reshape(mddev);
de171cb9
N		 5908 if (rv < 0)
5909 mddev->delta_disks = 0;
da943b99
N		 5910 return rv;
5911}
5912
5913
1da177e4
LT		 5914/*
5915 * update_array_info is used to change the configuration of an
5916 * on-line array.
5917 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5918 * fields in the info are checked against the array.
5919 * Any differences that cannot be handled will cause an error.
5920 * Normally, only one change can be managed at a time.
5921 */
fd01b88c 5922static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 5923{
5924 int rv = 0;
5925 int cnt = 0;
36fa3063
N		 5926 int state = 0;
5927
5928 /* calculate expected state,ignoring low bits */
c3d9714e 5929 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5930 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5931
5932 if (mddev->major_version != info->major_version ||
5933 mddev->minor_version != info->minor_version ||
5934/* mddev->patch_version != info->patch_version || */
5935 mddev->ctime != info->ctime ||
5936 mddev->level != info->level ||
5937/* mddev->layout != info->layout || */
5938 !mddev->persistent != info->not_persistent||
9d8f0363 5939 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5940 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5941 ((state^info->state) & 0xfffffe00)
5942 )
1da177e4
LT		 5943 return -EINVAL;
5944 /* Check there is only one change */
58c0fed4
AN		 5945 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5946 cnt++;
5947 if (mddev->raid_disks != info->raid_disks)
5948 cnt++;
5949 if (mddev->layout != info->layout)
5950 cnt++;
5951 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5952 cnt++;
5953 if (cnt == 0)
5954 return 0;
5955 if (cnt > 1)
5956 return -EINVAL;
1da177e4
LT		 5957
5958 if (mddev->layout != info->layout) {
5959 /* Change layout
5960 * we don't need to do anything at the md level, the
5961 * personality will take care of it all.
5962 */
50ac168a 5963 if (mddev->pers->check_reshape == NULL)
1da177e4 5964 return -EINVAL;
597a711b
N		 5965 else {
5966 mddev->new_layout = info->layout;
50ac168a 5967 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 5968 if (rv)
5969 mddev->new_layout = mddev->layout;
5970 return rv;
5971 }
1da177e4 5972 }
58c0fed4 5973 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 5974 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 5975
da943b99
N		 5976 if (mddev->raid_disks != info->raid_disks)
5977 rv = update_raid_disks(mddev, info->raid_disks);
5978
36fa3063
N		 5979 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5980 if (mddev->pers->quiesce == NULL)
5981 return -EINVAL;
5982 if (mddev->recovery || mddev->sync_thread)
5983 return -EBUSY;
5984 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5985 /* add the bitmap */
5986 if (mddev->bitmap)
5987 return -EEXIST;
c3d9714e 5988 if (mddev->bitmap_info.default_offset == 0)
36fa3063 5989 return -EINVAL;
c3d9714e
N		 5990 mddev->bitmap_info.offset =
5991 mddev->bitmap_info.default_offset;
36fa3063
N		 5992 mddev->pers->quiesce(mddev, 1);
5993 rv = bitmap_create(mddev);
69e51b44
N		 5994 if (!rv)
5995 rv = bitmap_load(mddev);
36fa3063
N		 5996 if (rv)
5997 bitmap_destroy(mddev);
5998 mddev->pers->quiesce(mddev, 0);
5999 } else {
6000 /* remove the bitmap */
6001 if (!mddev->bitmap)
6002 return -ENOENT;
6003 if (mddev->bitmap->file)
6004 return -EINVAL;
6005 mddev->pers->quiesce(mddev, 1);
6006 bitmap_destroy(mddev);
6007 mddev->pers->quiesce(mddev, 0);
c3d9714e 6008 mddev->bitmap_info.offset = 0;
36fa3063
N		 6009 }
6010 }
850b2b42 6011 md_update_sb(mddev, 1);
1da177e4
LT		 6012 return rv;
6013}
6014
fd01b88c 6015static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6016{
3cb03002 6017 struct md_rdev *rdev;
1da177e4
LT		 6018
6019 if (mddev->pers == NULL)
6020 return -ENODEV;
6021
6022 rdev = find_rdev(mddev, dev);
6023 if (!rdev)
6024 return -ENODEV;
6025
6026 md_error(mddev, rdev);
5ef56c8f
N		 6027 if (!test_bit(Faulty, &rdev->flags))
6028 return -EBUSY;
1da177e4
LT		 6029 return 0;
6030}
6031
2f9618ce
AN		 6032/*
6033 * We have a problem here : there is no easy way to give a CHS
6034 * virtual geometry. We currently pretend that we have a 2 heads
6035 * 4 sectors (with a BIG number of cylinders...). This drives
6036 * dosfs just mad... ;-)
6037 */
a885c8c4
CH		 6038static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6039{
fd01b88c 6040 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6041
6042 geo->heads = 2;
6043 geo->sectors = 4;
49ce6cea 6044 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6045 return 0;
6046}
6047
a39907fa 6048static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6049 unsigned int cmd, unsigned long arg)
6050{
6051 int err = 0;
6052 void __user *argp = (void __user *)arg;
fd01b88c 6053 struct mddev *mddev = NULL;
e2218350 6054 int ro;
1da177e4
LT		 6055
6056 if (!capable(CAP_SYS_ADMIN))
6057 return -EACCES;
6058
6059 /*
6060 * Commands dealing with the RAID driver but not any
6061 * particular array:
6062 */
6063 switch (cmd)
6064 {
6065 case RAID_VERSION:
6066 err = get_version(argp);
6067 goto done;
6068
6069 case PRINT_RAID_DEBUG:
6070 err = 0;
6071 md_print_devices();
6072 goto done;
6073
6074#ifndef MODULE
6075 case RAID_AUTORUN:
6076 err = 0;
6077 autostart_arrays(arg);
6078 goto done;
6079#endif
6080 default:;
6081 }
6082
6083 /*
6084 * Commands creating/starting a new array:
6085 */
6086
a39907fa 6087 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6088
6089 if (!mddev) {
6090 BUG();
6091 goto abort;
6092 }
6093
1da177e4
LT		 6094 err = mddev_lock(mddev);
6095 if (err) {
6096 printk(KERN_INFO
6097 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6098 err, cmd);
6099 goto abort;
6100 }
6101
6102 switch (cmd)
6103 {
6104 case SET_ARRAY_INFO:
6105 {
6106 mdu_array_info_t info;
6107 if (!arg)
6108 memset(&info, 0, sizeof(info));
6109 else if (copy_from_user(&info, argp, sizeof(info))) {
6110 err = -EFAULT;
6111 goto abort_unlock;
6112 }
6113 if (mddev->pers) {
6114 err = update_array_info(mddev, &info);
6115 if (err) {
6116 printk(KERN_WARNING "md: couldn't update"
6117 " array info. %d\n", err);
6118 goto abort_unlock;
6119 }
6120 goto done_unlock;
6121 }
6122 if (!list_empty(&mddev->disks)) {
6123 printk(KERN_WARNING
6124 "md: array %s already has disks!\n",
6125 mdname(mddev));
6126 err = -EBUSY;
6127 goto abort_unlock;
6128 }
6129 if (mddev->raid_disks) {
6130 printk(KERN_WARNING
6131 "md: array %s already initialised!\n",
6132 mdname(mddev));
6133 err = -EBUSY;
6134 goto abort_unlock;
6135 }
6136 err = set_array_info(mddev, &info);
6137 if (err) {
6138 printk(KERN_WARNING "md: couldn't set"
6139 " array info. %d\n", err);
6140 goto abort_unlock;
6141 }
6142 }
6143 goto done_unlock;
6144
6145 default:;
6146 }
6147
6148 /*
6149 * Commands querying/configuring an existing array:
6150 */
32a7627c 6151 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6152 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6153 if ((!mddev->raid_disks && !mddev->external)
6154 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6155 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6156 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6157 err = -ENODEV;
6158 goto abort_unlock;
6159 }
6160
6161 /*
6162 * Commands even a read-only array can execute:
6163 */
6164 switch (cmd)
6165 {
6166 case GET_ARRAY_INFO:
6167 err = get_array_info(mddev, argp);
6168 goto done_unlock;
6169
32a7627c 6170 case GET_BITMAP_FILE:
87162a28 6171 err = get_bitmap_file(mddev, argp);
32a7627c
N		 6172 goto done_unlock;
6173
1da177e4
LT		 6174 case GET_DISK_INFO:
6175 err = get_disk_info(mddev, argp);
6176 goto done_unlock;
6177
6178 case RESTART_ARRAY_RW:
6179 err = restart_array(mddev);
6180 goto done_unlock;
6181
6182 case STOP_ARRAY:
d710e138 6183 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 6184 goto done_unlock;
6185
6186 case STOP_ARRAY_RO:
a4bd82d0 6187 err = md_set_readonly(mddev, 1);
1da177e4
LT		 6188 goto done_unlock;
6189
e2218350
DW		 6190 case BLKROSET:
6191 if (get_user(ro, (int __user *)(arg))) {
6192 err = -EFAULT;
6193 goto done_unlock;
6194 }
6195 err = -EINVAL;
6196
6197 /* if the bdev is going readonly the value of mddev->ro
6198 * does not matter, no writes are coming
6199 */
6200 if (ro)
6201 goto done_unlock;
6202
6203 /* are we are already prepared for writes? */
6204 if (mddev->ro != 1)
6205 goto done_unlock;
6206
6207 /* transitioning to readauto need only happen for
6208 * arrays that call md_write_start
6209 */
6210 if (mddev->pers) {
6211 err = restart_array(mddev);
6212 if (err == 0) {
6213 mddev->ro = 2;
6214 set_disk_ro(mddev->gendisk, 0);
6215 }
6216 }
6217 goto done_unlock;
1da177e4
LT		 6218 }
6219
6220 /*
6221 * The remaining ioctls are changing the state of the
f91de92e
N		 6222 * superblock, so we do not allow them on read-only arrays.
6223 * However non-MD ioctls (e.g. get-size) will still come through
6224 * here and hit the 'default' below, so only disallow
6225 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6226 */
bb57fc64 6227 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6228 if (mddev->ro == 2) {
6229 mddev->ro = 0;
00bcb4ac 6230 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 6231 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6232 md_wakeup_thread(mddev->thread);
f91de92e
N		 6233 } else {
6234 err = -EROFS;
6235 goto abort_unlock;
6236 }
1da177e4
LT		 6237 }
6238
6239 switch (cmd)
6240 {
6241 case ADD_NEW_DISK:
6242 {
6243 mdu_disk_info_t info;
6244 if (copy_from_user(&info, argp, sizeof(info)))
6245 err = -EFAULT;
6246 else
6247 err = add_new_disk(mddev, &info);
6248 goto done_unlock;
6249 }
6250
6251 case HOT_REMOVE_DISK:
6252 err = hot_remove_disk(mddev, new_decode_dev(arg));
6253 goto done_unlock;
6254
6255 case HOT_ADD_DISK:
6256 err = hot_add_disk(mddev, new_decode_dev(arg));
6257 goto done_unlock;
6258
6259 case SET_DISK_FAULTY:
6260 err = set_disk_faulty(mddev, new_decode_dev(arg));
6261 goto done_unlock;
6262
6263 case RUN_ARRAY:
d710e138 6264 err = do_md_run(mddev);
1da177e4
LT		 6265 goto done_unlock;
6266
32a7627c
N		 6267 case SET_BITMAP_FILE:
6268 err = set_bitmap_file(mddev, (int)arg);
6269 goto done_unlock;
6270
1da177e4 6271 default:
1da177e4
LT		 6272 err = -EINVAL;
6273 goto abort_unlock;
6274 }
6275
6276done_unlock:
6277abort_unlock:
d3374825
N		 6278 if (mddev->hold_active == UNTIL_IOCTL &&
6279 err != -EINVAL)
6280 mddev->hold_active = 0;
1da177e4
LT		 6281 mddev_unlock(mddev);
6282
6283 return err;
6284done:
6285 if (err)
6286 MD_BUG();
6287abort:
6288 return err;
6289}
aa98aa31
AB		 6290#ifdef CONFIG_COMPAT
6291static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6292 unsigned int cmd, unsigned long arg)
6293{
6294 switch (cmd) {
6295 case HOT_REMOVE_DISK:
6296 case HOT_ADD_DISK:
6297 case SET_DISK_FAULTY:
6298 case SET_BITMAP_FILE:
6299 /* These take in integer arg, do not convert */
6300 break;
6301 default:
6302 arg = (unsigned long)compat_ptr(arg);
6303 break;
6304 }
6305
6306 return md_ioctl(bdev, mode, cmd, arg);
6307}
6308#endif /* CONFIG_COMPAT */
1da177e4 6309
a39907fa 6310static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6311{
6312 /*
6313 * Succeed if we can lock the mddev, which confirms that
6314 * it isn't being stopped right now.
6315 */
fd01b88c 6316 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6317 int err;
6318
d3374825
N		 6319 if (mddev->gendisk != bdev->bd_disk) {
6320 /* we are racing with mddev_put which is discarding this
6321 * bd_disk.
6322 */
6323 mddev_put(mddev);
6324 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6325 flush_workqueue(md_misc_wq);
d3374825
N		 6326 /* Then retry the open from the top */
6327 return -ERESTARTSYS;
6328 }
6329 BUG_ON(mddev != bdev->bd_disk->private_data);
6330
c8c00a69 6331 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6332 goto out;
6333
6334 err = 0;
f2ea68cf 6335 atomic_inc(&mddev->openers);
c8c00a69 6336 mutex_unlock(&mddev->open_mutex);
1da177e4 6337
f0b4f7e2 6338 check_disk_change(bdev);
1da177e4
LT		 6339 out:
6340 return err;
6341}
6342
a39907fa 6343static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6344{
fd01b88c 6345 struct mddev *mddev = disk->private_data;
1da177e4 6346
52e5f9d1 6347 BUG_ON(!mddev);
f2ea68cf 6348 atomic_dec(&mddev->openers);
1da177e4
LT		 6349 mddev_put(mddev);
6350
6351 return 0;
6352}
f0b4f7e2
N		 6353
6354static int md_media_changed(struct gendisk *disk)
6355{
fd01b88c 6356 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6357
6358 return mddev->changed;
6359}
6360
6361static int md_revalidate(struct gendisk *disk)
6362{
fd01b88c 6363 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6364
6365 mddev->changed = 0;
6366 return 0;
6367}
83d5cde4 6368static const struct block_device_operations md_fops =
1da177e4
LT		 6369{
6370 .owner = THIS_MODULE,
a39907fa
AV		 6371 .open = md_open,
6372 .release = md_release,
b492b852 6373 .ioctl = md_ioctl,
aa98aa31
AB		 6374#ifdef CONFIG_COMPAT
6375 .compat_ioctl = md_compat_ioctl,
6376#endif
a885c8c4 6377 .getgeo = md_getgeo,
f0b4f7e2
N		 6378 .media_changed = md_media_changed,
6379 .revalidate_disk= md_revalidate,
1da177e4
LT		 6380};
6381
75c96f85 6382static int md_thread(void * arg)
1da177e4 6383{
2b8bf345 6384 struct md_thread *thread = arg;
1da177e4 6385
1da177e4
LT		 6386 /*
6387 * md_thread is a 'system-thread', it's priority should be very
6388 * high. We avoid resource deadlocks individually in each
6389 * raid personality. (RAID5 does preallocation) We also use RR and
6390 * the very same RT priority as kswapd, thus we will never get
6391 * into a priority inversion deadlock.
6392 *
6393 * we definitely have to have equal or higher priority than
6394 * bdflush, otherwise bdflush will deadlock if there are too
6395 * many dirty RAID5 blocks.
6396 */
1da177e4 6397
6985c43f 6398 allow_signal(SIGKILL);
a6fb0934 6399 while (!kthread_should_stop()) {
1da177e4 6400
93588e22
N		 6401 /* We need to wait INTERRUPTIBLE so that
6402 * we don't add to the load-average.
6403 * That means we need to be sure no signals are
6404 * pending
6405 */
6406 if (signal_pending(current))
6407 flush_signals(current);
6408
6409 wait_event_interruptible_timeout
6410 (thread->wqueue,
6411 test_bit(THREAD_WAKEUP, &thread->flags)
6412 || kthread_should_stop(),
6413 thread->timeout);
1da177e4 6414
6c987910
N		 6415 clear_bit(THREAD_WAKEUP, &thread->flags);
6416 if (!kthread_should_stop())
589a594b 6417 thread->run(thread->mddev);
1da177e4 6418 }
a6fb0934 6419
1da177e4
LT		 6420 return 0;
6421}
6422
2b8bf345 6423void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6424{
6425 if (thread) {
36a4e1fe 6426 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6427 set_bit(THREAD_WAKEUP, &thread->flags);
6428 wake_up(&thread->wqueue);
6429 }
6430}
6431
2b8bf345 6432struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
1da177e4
LT		 6433 const char *name)
6434{
2b8bf345 6435 struct md_thread *thread;
1da177e4 6436
2b8bf345 6437 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6438 if (!thread)
6439 return NULL;
6440
1da177e4
LT		 6441 init_waitqueue_head(&thread->wqueue);
6442
1da177e4
LT		 6443 thread->run = run;
6444 thread->mddev = mddev;
32a7627c 6445 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6446 thread->tsk = kthread_run(md_thread, thread,
6447 "%s_%s",
6448 mdname(thread->mddev),
6449 name ?: mddev->pers->name);
a6fb0934 6450 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6451 kfree(thread);
6452 return NULL;
6453 }
1da177e4
LT		 6454 return thread;
6455}
6456
2b8bf345 6457void md_unregister_thread(struct md_thread **threadp)
1da177e4 6458{
2b8bf345 6459 struct md_thread *thread = *threadp;
e0cf8f04
N		 6460 if (!thread)
6461 return;
36a4e1fe 6462 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6463 /* Locking ensures that mddev_unlock does not wake_up a
6464 * non-existent thread
6465 */
6466 spin_lock(&pers_lock);
6467 *threadp = NULL;
6468 spin_unlock(&pers_lock);
a6fb0934
N		 6469
6470 kthread_stop(thread->tsk);
1da177e4
LT		 6471 kfree(thread);
6472}
6473
fd01b88c 6474void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6475{
6476 if (!mddev) {
6477 MD_BUG();
6478 return;
6479 }
6480
b2d444d7 6481 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6482 return;
6bfe0b49 6483
de393cde 6484 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6485 return;
6486 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6487 if (mddev->degraded)
6488 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6489 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6490 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6491 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6492 md_wakeup_thread(mddev->thread);
768a418d 6493 if (mddev->event_work.func)
e804ac78 6494 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6495 md_new_event_inintr(mddev);
1da177e4
LT		 6496}
6497
6498/* seq_file implementation /proc/mdstat */
6499
6500static void status_unused(struct seq_file *seq)
6501{
6502 int i = 0;
3cb03002 6503 struct md_rdev *rdev;
1da177e4
LT		 6504
6505 seq_printf(seq, "unused devices: ");
6506
159ec1fc 6507 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6508 char b[BDEVNAME_SIZE];
6509 i++;
6510 seq_printf(seq, "%s ",
6511 bdevname(rdev->bdev,b));
6512 }
6513 if (!i)
6514 seq_printf(seq, "<none>");
6515
6516 seq_printf(seq, "\n");
6517}
6518
6519
fd01b88c 6520static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6521{
dd71cf6b
N		 6522 sector_t max_sectors, resync, res;
6523 unsigned long dt, db;
6524 sector_t rt;
4588b42e
N		 6525 int scale;
6526 unsigned int per_milli;
1da177e4 6527
dd71cf6b 6528 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6529
6530 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6531 max_sectors = mddev->resync_max_sectors;
1da177e4 6532 else
dd71cf6b 6533 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6534
6535 /*
6536 * Should not happen.
6537 */
dd71cf6b 6538 if (!max_sectors) {
1da177e4
LT		 6539 MD_BUG();
6540 return;
6541 }
4588b42e 6542 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6543 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6544 * u32, as those are the requirements for sector_div.
6545 * Thus 'scale' must be at least 10
6546 */
6547 scale = 10;
6548 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6549 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6550 scale++;
6551 }
6552 res = (resync>>scale)*1000;
dd71cf6b 6553 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6554
6555 per_milli = res;
1da177e4 6556 {
4588b42e 6557 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6558 seq_printf(seq, "[");
6559 for (i = 0; i < x; i++)
6560 seq_printf(seq, "=");
6561 seq_printf(seq, ">");
6562 for (i = 0; i < y; i++)
6563 seq_printf(seq, ".");
6564 seq_printf(seq, "] ");
6565 }
4588b42e 6566 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6567 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6568 "reshape" :
61df9d91
N		 6569 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6570 "check" :
6571 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6572 "resync" : "recovery"))),
6573 per_milli/10, per_milli % 10,
dd71cf6b
N		 6574 (unsigned long long) resync/2,
6575 (unsigned long long) max_sectors/2);
1da177e4
LT		 6576
6577 /*
1da177e4
LT		 6578 * dt: time from mark until now
6579 * db: blocks written from mark until now
6580 * rt: remaining time
dd71cf6b
N		 6581 *
6582 * rt is a sector_t, so could be 32bit or 64bit.
6583 * So we divide before multiply in case it is 32bit and close
6584 * to the limit.
25985edc 6585 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6586 * near the end of resync when the number of remaining sectors
6587 * is close to 'db'.
6588 * We then divide rt by 32 after multiplying by db to compensate.
6589 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6590 */
6591 dt = ((jiffies - mddev->resync_mark) / HZ);
6592 if (!dt) dt++;
ff4e8d9a
N		 6593 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6594 - mddev->resync_mark_cnt;
1da177e4 6595
dd71cf6b
N		 6596 rt = max_sectors - resync; /* number of remaining sectors */
6597 sector_div(rt, db/32+1);
6598 rt *= dt;
6599 rt >>= 5;
6600
6601 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6602 ((unsigned long)rt % 60)/6);
1da177e4 6603
ff4e8d9a 6604 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6605}
6606
6607static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6608{
6609 struct list_head *tmp;
6610 loff_t l = *pos;
fd01b88c 6611 struct mddev *mddev;
1da177e4
LT		 6612
6613 if (l >= 0x10000)
6614 return NULL;
6615 if (!l--)
6616 /* header */
6617 return (void*)1;
6618
6619 spin_lock(&all_mddevs_lock);
6620 list_for_each(tmp,&all_mddevs)
6621 if (!l--) {
fd01b88c 6622 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6623 mddev_get(mddev);
6624 spin_unlock(&all_mddevs_lock);
6625 return mddev;
6626 }
6627 spin_unlock(&all_mddevs_lock);
6628 if (!l--)
6629 return (void*)2;/* tail */
6630 return NULL;
6631}
6632
6633static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6634{
6635 struct list_head *tmp;
fd01b88c 6636 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6637
6638 ++*pos;
6639 if (v == (void*)2)
6640 return NULL;
6641
6642 spin_lock(&all_mddevs_lock);
6643 if (v == (void*)1)
6644 tmp = all_mddevs.next;
6645 else
6646 tmp = mddev->all_mddevs.next;
6647 if (tmp != &all_mddevs)
fd01b88c 6648 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 6649 else {
6650 next_mddev = (void*)2;
6651 *pos = 0x10000;
6652 }
6653 spin_unlock(&all_mddevs_lock);
6654
6655 if (v != (void*)1)
6656 mddev_put(mddev);
6657 return next_mddev;
6658
6659}
6660
6661static void md_seq_stop(struct seq_file *seq, void *v)
6662{
fd01b88c 6663 struct mddev *mddev = v;
1da177e4
LT		 6664
6665 if (mddev && v != (void*)1 && v != (void*)2)
6666 mddev_put(mddev);
6667}
6668
6669static int md_seq_show(struct seq_file *seq, void *v)
6670{
fd01b88c 6671 struct mddev *mddev = v;
dd8ac336 6672 sector_t sectors;
3cb03002 6673 struct md_rdev *rdev;
32a7627c 6674 struct bitmap *bitmap;
1da177e4
LT		 6675
6676 if (v == (void*)1) {
84fc4b56 6677 struct md_personality *pers;
1da177e4
LT		 6678 seq_printf(seq, "Personalities : ");
6679 spin_lock(&pers_lock);
2604b703
N		 6680 list_for_each_entry(pers, &pers_list, list)
6681 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6682
6683 spin_unlock(&pers_lock);
6684 seq_printf(seq, "\n");
f1514638 6685 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6686 return 0;
6687 }
6688 if (v == (void*)2) {
6689 status_unused(seq);
6690 return 0;
6691 }
6692
5dc5cf7d 6693 if (mddev_lock(mddev) < 0)
1da177e4 6694 return -EINTR;
5dc5cf7d 6695
1da177e4
LT		 6696 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6697 seq_printf(seq, "%s : %sactive", mdname(mddev),
6698 mddev->pers ? "" : "in");
6699 if (mddev->pers) {
f91de92e 6700 if (mddev->ro==1)
1da177e4 6701 seq_printf(seq, " (read-only)");
f91de92e 6702 if (mddev->ro==2)
52720ae7 6703 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6704 seq_printf(seq, " %s", mddev->pers->name);
6705 }
6706
dd8ac336 6707 sectors = 0;
159ec1fc 6708 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 6709 char b[BDEVNAME_SIZE];
6710 seq_printf(seq, " %s[%d]",
6711 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6712 if (test_bit(WriteMostly, &rdev->flags))
6713 seq_printf(seq, "(W)");
b2d444d7 6714 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6715 seq_printf(seq, "(F)");
6716 continue;
b325a32e
N		 6717 } else if (rdev->raid_disk < 0)
6718 seq_printf(seq, "(S)"); /* spare */
dd8ac336 6719 sectors += rdev->sectors;
1da177e4
LT		 6720 }
6721
6722 if (!list_empty(&mddev->disks)) {
6723 if (mddev->pers)
6724 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6725 (unsigned long long)
6726 mddev->array_sectors / 2);
1da177e4
LT		 6727 else
6728 seq_printf(seq, "\n %llu blocks",
dd8ac336 6729 (unsigned long long)sectors / 2);
1da177e4 6730 }
1cd6bf19
N		 6731 if (mddev->persistent) {
6732 if (mddev->major_version != 0 ||
6733 mddev->minor_version != 90) {
6734 seq_printf(seq," super %d.%d",
6735 mddev->major_version,
6736 mddev->minor_version);
6737 }
e691063a
N		 6738 } else if (mddev->external)
6739 seq_printf(seq, " super external:%s",
6740 mddev->metadata_type);
6741 else
1cd6bf19 6742 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6743
6744 if (mddev->pers) {
d710e138 6745 mddev->pers->status(seq, mddev);
1da177e4 6746 seq_printf(seq, "\n ");
8e1b39d6
N		 6747 if (mddev->pers->sync_request) {
6748 if (mddev->curr_resync > 2) {
d710e138 6749 status_resync(seq, mddev);
8e1b39d6
N		 6750 seq_printf(seq, "\n ");
6751 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6752 seq_printf(seq, "\tresync=DELAYED\n ");
6753 else if (mddev->recovery_cp < MaxSector)
6754 seq_printf(seq, "\tresync=PENDING\n ");
6755 }
32a7627c
N		 6756 } else
6757 seq_printf(seq, "\n ");
6758
6759 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6760 unsigned long chunk_kb;
6761 unsigned long flags;
32a7627c 6762 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6763 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6764 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6765 "%lu%s chunk",
6766 bitmap->pages - bitmap->missing_pages,
6767 bitmap->pages,
6768 (bitmap->pages - bitmap->missing_pages)
6769 << (PAGE_SHIFT - 10),
42a04b50 6770 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6771 chunk_kb ? "KB" : "B");
78d742d8
N		 6772 if (bitmap->file) {
6773 seq_printf(seq, ", file: ");
c32c2f63 6774 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6775 }
78d742d8 6776
32a7627c
N		 6777 seq_printf(seq, "\n");
6778 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6779 }
6780
6781 seq_printf(seq, "\n");
6782 }
6783 mddev_unlock(mddev);
6784
6785 return 0;
6786}
6787
110518bc 6788static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6789 .start = md_seq_start,
6790 .next = md_seq_next,
6791 .stop = md_seq_stop,
6792 .show = md_seq_show,
6793};
6794
6795static int md_seq_open(struct inode *inode, struct file *file)
6796{
f1514638 6797 struct seq_file *seq;
1da177e4
LT		 6798 int error;
6799
6800 error = seq_open(file, &md_seq_ops);
d7603b7e 6801 if (error)
f1514638
KS		 6802 return error;
6803
6804 seq = file->private_data;
6805 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6806 return error;
6807}
6808
d7603b7e
N		 6809static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6810{
f1514638 6811 struct seq_file *seq = filp->private_data;
d7603b7e
N		 6812 int mask;
6813
6814 poll_wait(filp, &md_event_waiters, wait);
6815
6816 /* always allow read */
6817 mask = POLLIN | POLLRDNORM;
6818
f1514638 6819 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 6820 mask |= POLLERR | POLLPRI;
6821 return mask;
6822}
6823
fa027c2a 6824static const struct file_operations md_seq_fops = {
e24650c2 6825 .owner = THIS_MODULE,
1da177e4
LT		 6826 .open = md_seq_open,
6827 .read = seq_read,
6828 .llseek = seq_lseek,
c3f94b40 6829 .release = seq_release_private,
d7603b7e 6830 .poll = mdstat_poll,
1da177e4
LT		 6831};
6832
84fc4b56 6833int register_md_personality(struct md_personality *p)
1da177e4 6834{
1da177e4 6835 spin_lock(&pers_lock);
2604b703
N		 6836 list_add_tail(&p->list, &pers_list);
6837 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6838 spin_unlock(&pers_lock);
6839 return 0;
6840}
6841
84fc4b56 6842int unregister_md_personality(struct md_personality *p)
1da177e4 6843{
2604b703 6844 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6845 spin_lock(&pers_lock);
2604b703 6846 list_del_init(&p->list);
1da177e4
LT		 6847 spin_unlock(&pers_lock);
6848 return 0;
6849}
6850
fd01b88c 6851static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 6852{
3cb03002 6853 struct md_rdev * rdev;
1da177e4 6854 int idle;
eea1bf38 6855 int curr_events;
1da177e4
LT		 6856
6857 idle = 1;
4b80991c
N		 6858 rcu_read_lock();
6859 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6860 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6861 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6862 (int)part_stat_read(&disk->part0, sectors[1]) -
6863 atomic_read(&disk->sync_io);
713f6ab1
N		 6864 /* sync IO will cause sync_io to increase before the disk_stats
6865 * as sync_io is counted when a request starts, and
6866 * disk_stats is counted when it completes.
6867 * So resync activity will cause curr_events to be smaller than
6868 * when there was no such activity.
6869 * non-sync IO will cause disk_stat to increase without
6870 * increasing sync_io so curr_events will (eventually)
6871 * be larger than it was before. Once it becomes
6872 * substantially larger, the test below will cause
6873 * the array to appear non-idle, and resync will slow
6874 * down.
6875 * If there is a lot of outstanding resync activity when
6876 * we set last_event to curr_events, then all that activity
6877 * completing might cause the array to appear non-idle
6878 * and resync will be slowed down even though there might
6879 * not have been non-resync activity. This will only
6880 * happen once though. 'last_events' will soon reflect
6881 * the state where there is little or no outstanding
6882 * resync requests, and further resync activity will
6883 * always make curr_events less than last_events.
c0e48521 6884 *
1da177e4 6885 */
eea1bf38 6886 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6887 rdev->last_events = curr_events;
6888 idle = 0;
6889 }
6890 }
4b80991c 6891 rcu_read_unlock();
1da177e4
LT		 6892 return idle;
6893}
6894
fd01b88c 6895void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 6896{
6897 /* another "blocks" (512byte) blocks have been synced */
6898 atomic_sub(blocks, &mddev->recovery_active);
6899 wake_up(&mddev->recovery_wait);
6900 if (!ok) {
dfc70645 6901 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6902 md_wakeup_thread(mddev->thread);
6903 // stop recovery, signal do_sync ....
6904 }
6905}
6906
6907
06d91a5f
N		 6908/* md_write_start(mddev, bi)
6909 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6910 * in superblock) before writing, schedule a superblock update
6911 * and wait for it to complete.
06d91a5f 6912 */
fd01b88c 6913void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 6914{
0fd62b86 6915 int did_change = 0;
06d91a5f 6916 if (bio_data_dir(bi) != WRITE)
3d310eb7 6917 return;
06d91a5f 6918
f91de92e
N		 6919 BUG_ON(mddev->ro == 1);
6920 if (mddev->ro == 2) {
6921 /* need to switch to read/write */
6922 mddev->ro = 0;
6923 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6924 md_wakeup_thread(mddev->thread);
25156198 6925 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6926 did_change = 1;
f91de92e 6927 }
06d91a5f 6928 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6929 if (mddev->safemode == 1)
6930 mddev->safemode = 0;
06d91a5f 6931 if (mddev->in_sync) {
a9701a30 6932 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6933 if (mddev->in_sync) {
6934 mddev->in_sync = 0;
850b2b42 6935 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6936 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6937 md_wakeup_thread(mddev->thread);
0fd62b86 6938 did_change = 1;
3d310eb7 6939 }
a9701a30 6940 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6941 }
0fd62b86 6942 if (did_change)
00bcb4ac 6943 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 6944 wait_event(mddev->sb_wait,
09a44cc1 6945 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 6946}
6947
fd01b88c 6948void md_write_end(struct mddev *mddev)
1da177e4
LT		 6949{
6950 if (atomic_dec_and_test(&mddev->writes_pending)) {
6951 if (mddev->safemode == 2)
6952 md_wakeup_thread(mddev->thread);
16f17b39 6953 else if (mddev->safemode_delay)
1da177e4
LT		 6954 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6955 }
6956}
6957
2a2275d6
N		 6958/* md_allow_write(mddev)
6959 * Calling this ensures that the array is marked 'active' so that writes
6960 * may proceed without blocking. It is important to call this before
6961 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6962 * Must be called with mddev_lock held.
b5470dc5
DW		 6963 *
6964 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6965 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 6966 */
fd01b88c 6967int md_allow_write(struct mddev *mddev)
2a2275d6
N		 6968{
6969 if (!mddev->pers)
b5470dc5 6970 return 0;
2a2275d6 6971 if (mddev->ro)
b5470dc5 6972 return 0;
1a0fd497 6973 if (!mddev->pers->sync_request)
b5470dc5 6974 return 0;
2a2275d6
N		 6975
6976 spin_lock_irq(&mddev->write_lock);
6977 if (mddev->in_sync) {
6978 mddev->in_sync = 0;
6979 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6980 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 6981 if (mddev->safemode_delay &&
6982 mddev->safemode == 0)
6983 mddev->safemode = 1;
6984 spin_unlock_irq(&mddev->write_lock);
6985 md_update_sb(mddev, 0);
00bcb4ac 6986 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 6987 } else
6988 spin_unlock_irq(&mddev->write_lock);
b5470dc5 6989
070dc6dd 6990 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 6991 return -EAGAIN;
6992 else
6993 return 0;
2a2275d6
N		 6994}
6995EXPORT_SYMBOL_GPL(md_allow_write);
6996
1da177e4
LT		 6997#define SYNC_MARKS 10
6998#define SYNC_MARK_STEP (3*HZ)
fd01b88c 6999void md_do_sync(struct mddev *mddev)
1da177e4 7000{
fd01b88c 7001 struct mddev *mddev2;
1da177e4
LT		 7002 unsigned int currspeed = 0,
7003 window;
57afd89f 7004 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 7005 unsigned long mark[SYNC_MARKS];
7006 sector_t mark_cnt[SYNC_MARKS];
7007 int last_mark,m;
7008 struct list_head *tmp;
7009 sector_t last_check;
57afd89f 7010 int skipped = 0;
3cb03002 7011 struct md_rdev *rdev;
61df9d91 7012 char *desc;
1da177e4
LT		 7013
7014 /* just incase thread restarts... */
7015 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7016 return;
5fd6c1dc
N		 7017 if (mddev->ro) /* never try to sync a read-only array */
7018 return;
1da177e4 7019
61df9d91
N		 7020 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7021 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7022 desc = "data-check";
7023 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7024 desc = "requested-resync";
7025 else
7026 desc = "resync";
7027 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7028 desc = "reshape";
7029 else
7030 desc = "recovery";
7031
1da177e4
LT		 7032 /* we overload curr_resync somewhat here.
7033 * 0 == not engaged in resync at all
7034 * 2 == checking that there is no conflict with another sync
7035 * 1 == like 2, but have yielded to allow conflicting resync to
7036 * commense
7037 * other == active in resync - this many blocks
7038 *
7039 * Before starting a resync we must have set curr_resync to
7040 * 2, and then checked that every "conflicting" array has curr_resync
7041 * less than ours. When we find one that is the same or higher
7042 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7043 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7044 * This will mean we have to start checking from the beginning again.
7045 *
7046 */
7047
7048 do {
7049 mddev->curr_resync = 2;
7050
7051 try_again:
404e4b43 7052 if (kthread_should_stop())
6985c43f 7053 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7054
7055 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7056 goto skip;
29ac4aa3 7057 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7058 if (mddev2 == mddev)
7059 continue;
90b08710
BS		 7060 if (!mddev->parallel_resync
7061 && mddev2->curr_resync
7062 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7063 DEFINE_WAIT(wq);
7064 if (mddev < mddev2 && mddev->curr_resync == 2) {
7065 /* arbitrarily yield */
7066 mddev->curr_resync = 1;
7067 wake_up(&resync_wait);
7068 }
7069 if (mddev > mddev2 && mddev->curr_resync == 1)
7070 /* no need to wait here, we can wait the next
7071 * time 'round when curr_resync == 2
7072 */
7073 continue;
9744197c
N		 7074 /* We need to wait 'interruptible' so as not to
7075 * contribute to the load average, and not to
7076 * be caught by 'softlockup'
7077 */
7078 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7079 if (!kthread_should_stop() &&
8712e553 7080 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7081 printk(KERN_INFO "md: delaying %s of %s"
7082 " until %s has finished (they"
1da177e4 7083 " share one or more physical units)\n",
61df9d91 7084 desc, mdname(mddev), mdname(mddev2));
1da177e4 7085 mddev_put(mddev2);
9744197c
N		 7086 if (signal_pending(current))
7087 flush_signals(current);
1da177e4
LT		 7088 schedule();
7089 finish_wait(&resync_wait, &wq);
7090 goto try_again;
7091 }
7092 finish_wait(&resync_wait, &wq);
7093 }
7094 }
7095 } while (mddev->curr_resync < 2);
7096
5fd6c1dc 7097 j = 0;
9d88883e 7098 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7099 /* resync follows the size requested by the personality,
57afd89f 7100 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7101 */
7102 max_sectors = mddev->resync_max_sectors;
9d88883e 7103 mddev->resync_mismatches = 0;
5fd6c1dc 7104 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7105 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7106 j = mddev->resync_min;
7107 else if (!mddev->bitmap)
5fd6c1dc 7108 j = mddev->recovery_cp;
5e96ee65 7109
ccfcc3c1 7110 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 7111 max_sectors = mddev->dev_sectors;
5fd6c1dc 7112 else {
1da177e4 7113 /* recovery follows the physical size of devices */
58c0fed4 7114 max_sectors = mddev->dev_sectors;
5fd6c1dc 7115 j = MaxSector;
4e59ca7d
DW		 7116 rcu_read_lock();
7117 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 7118 if (rdev->raid_disk >= 0 &&
7119 !test_bit(Faulty, &rdev->flags) &&
7120 !test_bit(In_sync, &rdev->flags) &&
7121 rdev->recovery_offset < j)
7122 j = rdev->recovery_offset;
4e59ca7d 7123 rcu_read_unlock();
5fd6c1dc 7124 }
1da177e4 7125
61df9d91
N		 7126 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7127 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7128 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7129 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7130 "(but not more than %d KB/sec) for %s.\n",
7131 speed_max(mddev), desc);
1da177e4 7132
eea1bf38 7133 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7134
57afd89f 7135 io_sectors = 0;
1da177e4
LT		 7136 for (m = 0; m < SYNC_MARKS; m++) {
7137 mark[m] = jiffies;
57afd89f 7138 mark_cnt[m] = io_sectors;
1da177e4
LT		 7139 }
7140 last_mark = 0;
7141 mddev->resync_mark = mark[last_mark];
7142 mddev->resync_mark_cnt = mark_cnt[last_mark];
7143
7144 /*
7145 * Tune reconstruction:
7146 */
7147 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7148 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7149 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7150
7151 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7152 last_check = 0;
7153
7154 if (j>2) {
7155 printk(KERN_INFO
61df9d91
N		 7156 "md: resuming %s of %s from checkpoint.\n",
7157 desc, mdname(mddev));
1da177e4
LT		 7158 mddev->curr_resync = j;
7159 }
75d3da43 7160 mddev->curr_resync_completed = j;
1da177e4
LT		 7161
7162 while (j < max_sectors) {
57afd89f 7163 sector_t sectors;
1da177e4 7164
57afd89f 7165 skipped = 0;
97e4f42d 7166
7a91ee1f
N		 7167 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7168 ((mddev->curr_resync > mddev->curr_resync_completed &&
7169 (mddev->curr_resync - mddev->curr_resync_completed)
7170 > (max_sectors >> 4)) ||
7171 (j - mddev->curr_resync_completed)*2
7172 >= mddev->resync_max - mddev->curr_resync_completed
7173 )) {
97e4f42d 7174 /* time to update curr_resync_completed */
97e4f42d
N		 7175 wait_event(mddev->recovery_wait,
7176 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7177 mddev->curr_resync_completed = j;
070dc6dd 7178 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7179 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7180 }
acb180b0 7181
e62e58a5
N		 7182 while (j >= mddev->resync_max && !kthread_should_stop()) {
7183 /* As this condition is controlled by user-space,
7184 * we can block indefinitely, so use '_interruptible'
7185 * to avoid triggering warnings.
7186 */
7187 flush_signals(current); /* just in case */
7188 wait_event_interruptible(mddev->recovery_wait,
7189 mddev->resync_max > j
7190 || kthread_should_stop());
7191 }
acb180b0
N		 7192
7193 if (kthread_should_stop())
7194 goto interrupted;
7195
57afd89f 7196 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7197 currspeed < speed_min(mddev));
57afd89f 7198 if (sectors == 0) {
dfc70645 7199 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7200 goto out;
7201 }
57afd89f
N		 7202
7203 if (!skipped) { /* actual IO requested */
7204 io_sectors += sectors;
7205 atomic_add(sectors, &mddev->recovery_active);
7206 }
7207
e875ecea
N		 7208 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7209 break;
7210
1da177e4
LT		 7211 j += sectors;
7212 if (j>1) mddev->curr_resync = j;
ff4e8d9a 7213 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7214 if (last_check == 0)
e875ecea 7215 /* this is the earliest that rebuild will be
d7603b7e
N		 7216 * visible in /proc/mdstat
7217 */
7218 md_new_event(mddev);
57afd89f
N		 7219
7220 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7221 continue;
7222
57afd89f 7223 last_check = io_sectors;
1da177e4
LT		 7224 repeat:
7225 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7226 /* step marks */
7227 int next = (last_mark+1) % SYNC_MARKS;
7228
7229 mddev->resync_mark = mark[next];
7230 mddev->resync_mark_cnt = mark_cnt[next];
7231 mark[next] = jiffies;
57afd89f 7232 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7233 last_mark = next;
7234 }
7235
7236
c6207277
N		 7237 if (kthread_should_stop())
7238 goto interrupted;
7239
1da177e4
LT		 7240
7241 /*
7242 * this loop exits only if either when we are slower than
7243 * the 'hard' speed limit, or the system was IO-idle for
7244 * a jiffy.
7245 * the system might be non-idle CPU-wise, but we only care
7246 * about not overloading the IO subsystem. (things like an
7247 * e2fsck being done on the RAID array should execute fast)
7248 */
1da177e4
LT		 7249 cond_resched();
7250
57afd89f
N		 7251 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7252 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7253
88202a0c
N		 7254 if (currspeed > speed_min(mddev)) {
7255 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7256 !is_mddev_idle(mddev, 0)) {
c0e48521 7257 msleep(500);
1da177e4
LT		 7258 goto repeat;
7259 }
7260 }
7261 }
61df9d91 7262 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7263 /*
7264 * this also signals 'finished resyncing' to md_stop
7265 */
7266 out:
1da177e4
LT		 7267 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7268
7269 /* tell personality that we are finished */
57afd89f 7270 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7271
dfc70645 7272 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7273 mddev->curr_resync > 2) {
7274 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7275 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7276 if (mddev->curr_resync >= mddev->recovery_cp) {
7277 printk(KERN_INFO
61df9d91
N		 7278 "md: checkpointing %s of %s.\n",
7279 desc, mdname(mddev));
5fd6c1dc
N		 7280 mddev->recovery_cp = mddev->curr_resync;
7281 }
7282 } else
7283 mddev->recovery_cp = MaxSector;
7284 } else {
7285 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7286 mddev->curr_resync = MaxSector;
4e59ca7d
DW		 7287 rcu_read_lock();
7288 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc 7289 if (rdev->raid_disk >= 0 &&
70fffd0b 7290 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7291 !test_bit(Faulty, &rdev->flags) &&
7292 !test_bit(In_sync, &rdev->flags) &&
7293 rdev->recovery_offset < mddev->curr_resync)
7294 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7295 rcu_read_unlock();
5fd6c1dc 7296 }
1da177e4 7297 }
17571284 7298 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7299
1da177e4 7300 skip:
c07b70ad
N		 7301 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7302 /* We completed so min/max setting can be forgotten if used. */
7303 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7304 mddev->resync_min = 0;
7305 mddev->resync_max = MaxSector;
7306 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7307 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7308 mddev->curr_resync = 0;
7309 wake_up(&resync_wait);
7310 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7311 md_wakeup_thread(mddev->thread);
c6207277
N		 7312 return;
7313
7314 interrupted:
7315 /*
7316 * got a signal, exit.
7317 */
7318 printk(KERN_INFO
7319 "md: md_do_sync() got signal ... exiting\n");
7320 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7321 goto out;
7322
1da177e4 7323}
29269553 7324EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7325
fd01b88c 7326static int remove_and_add_spares(struct mddev *mddev)
b4c4c7b8 7327{
3cb03002 7328 struct md_rdev *rdev;
b4c4c7b8
N		 7329 int spares = 0;
7330
97e4f42d
N		 7331 mddev->curr_resync_completed = 0;
7332
159ec1fc 7333 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 7334 if (rdev->raid_disk >= 0 &&
6bfe0b49 7335 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7336 (test_bit(Faulty, &rdev->flags) ||
7337 ! test_bit(In_sync, &rdev->flags)) &&
7338 atomic_read(&rdev->nr_pending)==0) {
7339 if (mddev->pers->hot_remove_disk(
7340 mddev, rdev->raid_disk)==0) {
36fad858 7341 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8
N		 7342 rdev->raid_disk = -1;
7343 }
7344 }
7345
5389042f 7346 if (mddev->degraded) {
159ec1fc 7347 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7348 if (rdev->raid_disk >= 0 &&
e5427135 7349 !test_bit(In_sync, &rdev->flags) &&
de393cde 7350 !test_bit(Faulty, &rdev->flags))
dfc70645 7351 spares++;
b4c4c7b8
N		 7352 if (rdev->raid_disk < 0
7353 && !test_bit(Faulty, &rdev->flags)) {
7354 rdev->recovery_offset = 0;
199050ea
NB		 7355 if (mddev->pers->
7356 hot_add_disk(mddev, rdev) == 0) {
36fad858 7357 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 7358 /* failure here is OK */;
b4c4c7b8
N		 7359 spares++;
7360 md_new_event(mddev);
93be75ff 7361 set_bit(MD_CHANGE_DEVS, &mddev->flags);
60fc1370 7362 }
b4c4c7b8 7363 }
dfc70645 7364 }
b4c4c7b8
N		 7365 }
7366 return spares;
7367}
7ebc0be7 7368
fd01b88c 7369static void reap_sync_thread(struct mddev *mddev)
7ebc0be7 7370{
3cb03002 7371 struct md_rdev *rdev;
7ebc0be7
N		 7372
7373 /* resync has finished, collect result */
01f96c0a 7374 md_unregister_thread(&mddev->sync_thread);
7ebc0be7
N		 7375 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7376 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7377 /* success...*/
7378 /* activate any spares */
7379 if (mddev->pers->spare_active(mddev))
7380 sysfs_notify(&mddev->kobj, NULL,
7381 "degraded");
7382 }
7383 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7384 mddev->pers->finish_reshape)
7385 mddev->pers->finish_reshape(mddev);
7ebc0be7 7386
d70ed2e4
AW		 7387 /* If array is no-longer degraded, then any saved_raid_disk
7388 * information must be scrapped. Also if any device is now
7389 * In_sync we must scrape the saved_raid_disk for that device
7390 * do the superblock for an incrementally recovered device
7391 * written out.
7ebc0be7 7392 */
d70ed2e4
AW		 7393 list_for_each_entry(rdev, &mddev->disks, same_set)
7394 if (!mddev->degraded ||
7395 test_bit(In_sync, &rdev->flags))
7ebc0be7
N		 7396 rdev->saved_raid_disk = -1;
7397
d70ed2e4 7398 md_update_sb(mddev, 1);
7ebc0be7
N		 7399 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7400 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7401 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7402 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7403 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7404 /* flag recovery needed just to double check */
7405 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7406 sysfs_notify_dirent_safe(mddev->sysfs_action);
7407 md_new_event(mddev);
768e587e
JB		 7408 if (mddev->event_work.func)
7409 queue_work(md_misc_wq, &mddev->event_work);
7ebc0be7
N		 7410}
7411
1da177e4
LT		 7412/*
7413 * This routine is regularly called by all per-raid-array threads to
7414 * deal with generic issues like resync and super-block update.
7415 * Raid personalities that don't have a thread (linear/raid0) do not
7416 * need this as they never do any recovery or update the superblock.
7417 *
7418 * It does not do any resync itself, but rather "forks" off other threads
7419 * to do that as needed.
7420 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7421 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7422 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7423 * and wakeups up this thread which will reap the thread and finish up.
7424 * This thread also removes any faulty devices (with nr_pending == 0).
7425 *
7426 * The overall approach is:
7427 * 1/ if the superblock needs updating, update it.
7428 * 2/ If a recovery thread is running, don't do anything else.
7429 * 3/ If recovery has finished, clean up, possibly marking spares active.
7430 * 4/ If there are any faulty devices, remove them.
7431 * 5/ If array is degraded, try to add spares devices
7432 * 6/ If array has spares or is not in-sync, start a resync thread.
7433 */
fd01b88c 7434void md_check_recovery(struct mddev *mddev)
1da177e4 7435{
68866e42
JB		 7436 if (mddev->suspended)
7437 return;
7438
5f40402d 7439 if (mddev->bitmap)
aa5cbd10 7440 bitmap_daemon_work(mddev);
1da177e4 7441
fca4d848 7442 if (signal_pending(current)) {
31a59e34 7443 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7444 printk(KERN_INFO "md: %s in immediate safe mode\n",
7445 mdname(mddev));
7446 mddev->safemode = 2;
7447 }
7448 flush_signals(current);
7449 }
7450
c89a8eee
N		 7451 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7452 return;
1da177e4 7453 if ( ! (
126925c0 7454 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7455 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7456 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7457 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7458 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7459 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7460 ))
7461 return;
fca4d848 7462
df5b89b3 7463 if (mddev_trylock(mddev)) {
b4c4c7b8 7464 int spares = 0;
fca4d848 7465
c89a8eee
N		 7466 if (mddev->ro) {
7467 /* Only thing we do on a ro array is remove
7468 * failed devices.
7469 */
3cb03002 7470 struct md_rdev *rdev;
a8c42c7f
N		 7471 list_for_each_entry(rdev, &mddev->disks, same_set)
7472 if (rdev->raid_disk >= 0 &&
7473 !test_bit(Blocked, &rdev->flags) &&
7474 test_bit(Faulty, &rdev->flags) &&
7475 atomic_read(&rdev->nr_pending)==0) {
7476 if (mddev->pers->hot_remove_disk(
7477 mddev, rdev->raid_disk)==0) {
36fad858 7478 sysfs_unlink_rdev(mddev, rdev);
a8c42c7f
N		 7479 rdev->raid_disk = -1;
7480 }
7481 }
c89a8eee
N		 7482 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7483 goto unlock;
7484 }
7485
31a59e34 7486 if (!mddev->external) {
0fd62b86 7487 int did_change = 0;
31a59e34
N		 7488 spin_lock_irq(&mddev->write_lock);
7489 if (mddev->safemode &&
7490 !atomic_read(&mddev->writes_pending) &&
7491 !mddev->in_sync &&
7492 mddev->recovery_cp == MaxSector) {
7493 mddev->in_sync = 1;
0fd62b86 7494 did_change = 1;
070dc6dd 7495 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7496 }
7497 if (mddev->safemode == 1)
7498 mddev->safemode = 0;
7499 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7500 if (did_change)
00bcb4ac 7501 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7502 }
fca4d848 7503
850b2b42
N		 7504 if (mddev->flags)
7505 md_update_sb(mddev, 0);
06d91a5f 7506
1da177e4
LT		 7507 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7508 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7509 /* resync/recovery still happening */
7510 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7511 goto unlock;
7512 }
7513 if (mddev->sync_thread) {
7ebc0be7 7514 reap_sync_thread(mddev);
1da177e4
LT		 7515 goto unlock;
7516 }
72a23c21
NB		 7517 /* Set RUNNING before clearing NEEDED to avoid
7518 * any transients in the value of "sync_action".
7519 */
7520 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7521 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7522 /* Clear some bits that don't mean anything, but
7523 * might be left set
7524 */
24dd469d
N		 7525 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7526 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7527
5fd6c1dc
N		 7528 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7529 goto unlock;
1da177e4
LT		 7530 /* no recovery is running.
7531 * remove any failed drives, then
7532 * add spares if possible.
7533 * Spare are also removed and re-added, to allow
7534 * the personality to fail the re-add.
7535 */
1da177e4 7536
b4c4c7b8 7537 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7538 if (mddev->pers->check_reshape == NULL ||
7539 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7540 /* Cannot proceed */
7541 goto unlock;
7542 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7543 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7544 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7545 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7546 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7547 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7548 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7549 } else if (mddev->recovery_cp < MaxSector) {
7550 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7551 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7552 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7553 /* nothing to be done ... */
1da177e4 7554 goto unlock;
24dd469d 7555
1da177e4 7556 if (mddev->pers->sync_request) {
a654b9d8
N		 7557 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7558 /* We are adding a device or devices to an array
7559 * which has the bitmap stored on all devices.
7560 * So make sure all bitmap pages get written
7561 */
7562 bitmap_write_all(mddev->bitmap);
7563 }
1da177e4
LT		 7564 mddev->sync_thread = md_register_thread(md_do_sync,
7565 mddev,
0da3c619 7566 "resync");
1da177e4
LT		 7567 if (!mddev->sync_thread) {
7568 printk(KERN_ERR "%s: could not start resync"
7569 " thread...\n",
7570 mdname(mddev));
7571 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7572 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7573 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7574 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7575 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7576 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7577 } else
1da177e4 7578 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7579 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7580 md_new_event(mddev);
1da177e4
LT		 7581 }
7582 unlock:
72a23c21
NB		 7583 if (!mddev->sync_thread) {
7584 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7585 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7586 &mddev->recovery))
0c3573f1 7587 if (mddev->sysfs_action)
00bcb4ac 7588 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7589 }
1da177e4
LT		 7590 mddev_unlock(mddev);
7591 }
7592}
7593
fd01b88c 7594void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7595{
00bcb4ac 7596 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7597 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7598 !test_bit(Blocked, &rdev->flags) &&
7599 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7600 msecs_to_jiffies(5000));
7601 rdev_dec_pending(rdev, mddev);
7602}
7603EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7604
2230dfe4
N		 7605
7606/* Bad block management.
7607 * We can record which blocks on each device are 'bad' and so just
7608 * fail those blocks, or that stripe, rather than the whole device.
7609 * Entries in the bad-block table are 64bits wide. This comprises:
7610 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7611 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7612 * A 'shift' can be set so that larger blocks are tracked and
7613 * consequently larger devices can be covered.
7614 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7615 *
7616 * Locking of the bad-block table uses a seqlock so md_is_badblock
7617 * might need to retry if it is very unlucky.
7618 * We will sometimes want to check for bad blocks in a bi_end_io function,
7619 * so we use the write_seqlock_irq variant.
7620 *
7621 * When looking for a bad block we specify a range and want to
7622 * know if any block in the range is bad. So we binary-search
7623 * to the last range that starts at-or-before the given endpoint,
7624 * (or "before the sector after the target range")
7625 * then see if it ends after the given start.
7626 * We return
7627 * 0 if there are no known bad blocks in the range
7628 * 1 if there are known bad block which are all acknowledged
7629 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7630 * plus the start/length of the first bad section we overlap.
7631 */
7632int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7633 sector_t *first_bad, int *bad_sectors)
7634{
7635 int hi;
7636 int lo = 0;
7637 u64 *p = bb->page;
7638 int rv = 0;
7639 sector_t target = s + sectors;
7640 unsigned seq;
7641
7642 if (bb->shift > 0) {
7643 /* round the start down, and the end up */
7644 s >>= bb->shift;
7645 target += (1<<bb->shift) - 1;
7646 target >>= bb->shift;
7647 sectors = target - s;
7648 }
7649 /* 'target' is now the first block after the bad range */
7650
7651retry:
7652 seq = read_seqbegin(&bb->lock);
7653
7654 hi = bb->count;
7655
7656 /* Binary search between lo and hi for 'target'
7657 * i.e. for the last range that starts before 'target'
7658 */
7659 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7660 * are known not to be the last range before target.
7661 * VARIANT: hi-lo is the number of possible
7662 * ranges, and decreases until it reaches 1
7663 */
7664 while (hi - lo > 1) {
7665 int mid = (lo + hi) / 2;
7666 sector_t a = BB_OFFSET(p[mid]);
7667 if (a < target)
7668 /* This could still be the one, earlier ranges
7669 * could not. */
7670 lo = mid;
7671 else
7672 /* This and later ranges are definitely out. */
7673 hi = mid;
7674 }
7675 /* 'lo' might be the last that started before target, but 'hi' isn't */
7676 if (hi > lo) {
7677 /* need to check all range that end after 's' to see if
7678 * any are unacknowledged.
7679 */
7680 while (lo >= 0 &&
7681 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7682 if (BB_OFFSET(p[lo]) < target) {
7683 /* starts before the end, and finishes after
7684 * the start, so they must overlap
7685 */
7686 if (rv != -1 && BB_ACK(p[lo]))
7687 rv = 1;
7688 else
7689 rv = -1;
7690 *first_bad = BB_OFFSET(p[lo]);
7691 *bad_sectors = BB_LEN(p[lo]);
7692 }
7693 lo--;
7694 }
7695 }
7696
7697 if (read_seqretry(&bb->lock, seq))
7698 goto retry;
7699
7700 return rv;
7701}
7702EXPORT_SYMBOL_GPL(md_is_badblock);
7703
7704/*
7705 * Add a range of bad blocks to the table.
7706 * This might extend the table, or might contract it
7707 * if two adjacent ranges can be merged.
7708 * We binary-search to find the 'insertion' point, then
7709 * decide how best to handle it.
7710 */
7711static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7712 int acknowledged)
7713{
7714 u64 *p;
7715 int lo, hi;
7716 int rv = 1;
7717
7718 if (bb->shift < 0)
7719 /* badblocks are disabled */
7720 return 0;
7721
7722 if (bb->shift) {
7723 /* round the start down, and the end up */
7724 sector_t next = s + sectors;
7725 s >>= bb->shift;
7726 next += (1<<bb->shift) - 1;
7727 next >>= bb->shift;
7728 sectors = next - s;
7729 }
7730
7731 write_seqlock_irq(&bb->lock);
7732
7733 p = bb->page;
7734 lo = 0;
7735 hi = bb->count;
7736 /* Find the last range that starts at-or-before 's' */
7737 while (hi - lo > 1) {
7738 int mid = (lo + hi) / 2;
7739 sector_t a = BB_OFFSET(p[mid]);
7740 if (a <= s)
7741 lo = mid;
7742 else
7743 hi = mid;
7744 }
7745 if (hi > lo && BB_OFFSET(p[lo]) > s)
7746 hi = lo;
7747
7748 if (hi > lo) {
7749 /* we found a range that might merge with the start
7750 * of our new range
7751 */
7752 sector_t a = BB_OFFSET(p[lo]);
7753 sector_t e = a + BB_LEN(p[lo]);
7754 int ack = BB_ACK(p[lo]);
7755 if (e >= s) {
7756 /* Yes, we can merge with a previous range */
7757 if (s == a && s + sectors >= e)
7758 /* new range covers old */
7759 ack = acknowledged;
7760 else
7761 ack = ack && acknowledged;
7762
7763 if (e < s + sectors)
7764 e = s + sectors;
7765 if (e - a <= BB_MAX_LEN) {
7766 p[lo] = BB_MAKE(a, e-a, ack);
7767 s = e;
7768 } else {
7769 /* does not all fit in one range,
7770 * make p[lo] maximal
7771 */
7772 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7773 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7774 s = a + BB_MAX_LEN;
7775 }
7776 sectors = e - s;
7777 }
7778 }
7779 if (sectors && hi < bb->count) {
7780 /* 'hi' points to the first range that starts after 's'.
7781 * Maybe we can merge with the start of that range */
7782 sector_t a = BB_OFFSET(p[hi]);
7783 sector_t e = a + BB_LEN(p[hi]);
7784 int ack = BB_ACK(p[hi]);
7785 if (a <= s + sectors) {
7786 /* merging is possible */
7787 if (e <= s + sectors) {
7788 /* full overlap */
7789 e = s + sectors;
7790 ack = acknowledged;
7791 } else
7792 ack = ack && acknowledged;
7793
7794 a = s;
7795 if (e - a <= BB_MAX_LEN) {
7796 p[hi] = BB_MAKE(a, e-a, ack);
7797 s = e;
7798 } else {
7799 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7800 s = a + BB_MAX_LEN;
7801 }
7802 sectors = e - s;
7803 lo = hi;
7804 hi++;
7805 }
7806 }
7807 if (sectors == 0 && hi < bb->count) {
7808 /* we might be able to combine lo and hi */
7809 /* Note: 's' is at the end of 'lo' */
7810 sector_t a = BB_OFFSET(p[hi]);
7811 int lolen = BB_LEN(p[lo]);
7812 int hilen = BB_LEN(p[hi]);
7813 int newlen = lolen + hilen - (s - a);
7814 if (s >= a && newlen < BB_MAX_LEN) {
7815 /* yes, we can combine them */
7816 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7817 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7818 memmove(p + hi, p + hi + 1,
7819 (bb->count - hi - 1) * 8);
7820 bb->count--;
7821 }
7822 }
7823 while (sectors) {
7824 /* didn't merge (it all).
7825 * Need to add a range just before 'hi' */
7826 if (bb->count >= MD_MAX_BADBLOCKS) {
7827 /* No room for more */
7828 rv = 0;
7829 break;
7830 } else {
7831 int this_sectors = sectors;
7832 memmove(p + hi + 1, p + hi,
7833 (bb->count - hi) * 8);
7834 bb->count++;
7835
7836 if (this_sectors > BB_MAX_LEN)
7837 this_sectors = BB_MAX_LEN;
7838 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7839 sectors -= this_sectors;
7840 s += this_sectors;
7841 }
7842 }
7843
7844 bb->changed = 1;
de393cde
N		 7845 if (!acknowledged)
7846 bb->unacked_exist = 1;
2230dfe4
N		 7847 write_sequnlock_irq(&bb->lock);
7848
7849 return rv;
7850}
7851
3cb03002 7852int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
2230dfe4
N		 7853 int acknowledged)
7854{
7855 int rv = md_set_badblocks(&rdev->badblocks,
7856 s + rdev->data_offset, sectors, acknowledged);
7857 if (rv) {
7858 /* Make sure they get written out promptly */
8bd2f0a0 7859 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 7860 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7861 md_wakeup_thread(rdev->mddev->thread);
7862 }
7863 return rv;
7864}
7865EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7866
7867/*
7868 * Remove a range of bad blocks from the table.
7869 * This may involve extending the table if we spilt a region,
7870 * but it must not fail. So if the table becomes full, we just
7871 * drop the remove request.
7872 */
7873static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7874{
7875 u64 *p;
7876 int lo, hi;
7877 sector_t target = s + sectors;
7878 int rv = 0;
7879
7880 if (bb->shift > 0) {
7881 /* When clearing we round the start up and the end down.
7882 * This should not matter as the shift should align with
7883 * the block size and no rounding should ever be needed.
7884 * However it is better the think a block is bad when it
7885 * isn't than to think a block is not bad when it is.
7886 */
7887 s += (1<<bb->shift) - 1;
7888 s >>= bb->shift;
7889 target >>= bb->shift;
7890 sectors = target - s;
7891 }
7892
7893 write_seqlock_irq(&bb->lock);
7894
7895 p = bb->page;
7896 lo = 0;
7897 hi = bb->count;
7898 /* Find the last range that starts before 'target' */
7899 while (hi - lo > 1) {
7900 int mid = (lo + hi) / 2;
7901 sector_t a = BB_OFFSET(p[mid]);
7902 if (a < target)
7903 lo = mid;
7904 else
7905 hi = mid;
7906 }
7907 if (hi > lo) {
7908 /* p[lo] is the last range that could overlap the
7909 * current range. Earlier ranges could also overlap,
7910 * but only this one can overlap the end of the range.
7911 */
7912 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7913 /* Partial overlap, leave the tail of this range */
7914 int ack = BB_ACK(p[lo]);
7915 sector_t a = BB_OFFSET(p[lo]);
7916 sector_t end = a + BB_LEN(p[lo]);
7917
7918 if (a < s) {
7919 /* we need to split this range */
7920 if (bb->count >= MD_MAX_BADBLOCKS) {
7921 rv = 0;
7922 goto out;
7923 }
7924 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7925 bb->count++;
7926 p[lo] = BB_MAKE(a, s-a, ack);
7927 lo++;
7928 }
7929 p[lo] = BB_MAKE(target, end - target, ack);
7930 /* there is no longer an overlap */
7931 hi = lo;
7932 lo--;
7933 }
7934 while (lo >= 0 &&
7935 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7936 /* This range does overlap */
7937 if (BB_OFFSET(p[lo]) < s) {
7938 /* Keep the early parts of this range. */
7939 int ack = BB_ACK(p[lo]);
7940 sector_t start = BB_OFFSET(p[lo]);
7941 p[lo] = BB_MAKE(start, s - start, ack);
7942 /* now low doesn't overlap, so.. */
7943 break;
7944 }
7945 lo--;
7946 }
7947 /* 'lo' is strictly before, 'hi' is strictly after,
7948 * anything between needs to be discarded
7949 */
7950 if (hi - lo > 1) {
7951 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
7952 bb->count -= (hi - lo - 1);
7953 }
7954 }
7955
7956 bb->changed = 1;
7957out:
7958 write_sequnlock_irq(&bb->lock);
7959 return rv;
7960}
7961
3cb03002 7962int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
2230dfe4
N		 7963{
7964 return md_clear_badblocks(&rdev->badblocks,
7965 s + rdev->data_offset,
7966 sectors);
7967}
7968EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
7969
7970/*
7971 * Acknowledge all bad blocks in a list.
7972 * This only succeeds if ->changed is clear. It is used by
7973 * in-kernel metadata updates
7974 */
7975void md_ack_all_badblocks(struct badblocks *bb)
7976{
7977 if (bb->page == NULL || bb->changed)
7978 /* no point even trying */
7979 return;
7980 write_seqlock_irq(&bb->lock);
7981
7982 if (bb->changed == 0) {
7983 u64 *p = bb->page;
7984 int i;
7985 for (i = 0; i < bb->count ; i++) {
7986 if (!BB_ACK(p[i])) {
7987 sector_t start = BB_OFFSET(p[i]);
7988 int len = BB_LEN(p[i]);
7989 p[i] = BB_MAKE(start, len, 1);
7990 }
7991 }
de393cde 7992 bb->unacked_exist = 0;
2230dfe4
N		 7993 }
7994 write_sequnlock_irq(&bb->lock);
7995}
7996EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
7997
16c791a5
N		 7998/* sysfs access to bad-blocks list.
7999 * We present two files.
8000 * 'bad-blocks' lists sector numbers and lengths of ranges that
8001 * are recorded as bad. The list is truncated to fit within
8002 * the one-page limit of sysfs.
8003 * Writing "sector length" to this file adds an acknowledged
8004 * bad block list.
8005 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8006 * been acknowledged. Writing to this file adds bad blocks
8007 * without acknowledging them. This is largely for testing.
8008 */
8009
8010static ssize_t
8011badblocks_show(struct badblocks *bb, char *page, int unack)
8012{
8013 size_t len;
8014 int i;
8015 u64 *p = bb->page;
8016 unsigned seq;
8017
8018 if (bb->shift < 0)
8019 return 0;
8020
8021retry:
8022 seq = read_seqbegin(&bb->lock);
8023
8024 len = 0;
8025 i = 0;
8026
8027 while (len < PAGE_SIZE && i < bb->count) {
8028 sector_t s = BB_OFFSET(p[i]);
8029 unsigned int length = BB_LEN(p[i]);
8030 int ack = BB_ACK(p[i]);
8031 i++;
8032
8033 if (unack && ack)
8034 continue;
8035
8036 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8037 (unsigned long long)s << bb->shift,
8038 length << bb->shift);
8039 }
de393cde
N		 8040 if (unack && len == 0)
8041 bb->unacked_exist = 0;
16c791a5
N		 8042
8043 if (read_seqretry(&bb->lock, seq))
8044 goto retry;
8045
8046 return len;
8047}
8048
8049#define DO_DEBUG 1
8050
8051static ssize_t
8052badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8053{
8054 unsigned long long sector;
8055 int length;
8056 char newline;
8057#ifdef DO_DEBUG
8058 /* Allow clearing via sysfs *only* for testing/debugging.
8059 * Normally only a successful write may clear a badblock
8060 */
8061 int clear = 0;
8062 if (page[0] == '-') {
8063 clear = 1;
8064 page++;
8065 }
8066#endif /* DO_DEBUG */
8067
8068 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8069 case 3:
8070 if (newline != '\n')
8071 return -EINVAL;
8072 case 2:
8073 if (length <= 0)
8074 return -EINVAL;
8075 break;
8076 default:
8077 return -EINVAL;
8078 }
8079
8080#ifdef DO_DEBUG
8081 if (clear) {
8082 md_clear_badblocks(bb, sector, length);
8083 return len;
8084 }
8085#endif /* DO_DEBUG */
8086 if (md_set_badblocks(bb, sector, length, !unack))
8087 return len;
8088 else
8089 return -ENOSPC;
8090}
8091
75c96f85
AB		 8092static int md_notify_reboot(struct notifier_block *this,
8093 unsigned long code, void *x)
1da177e4
LT		 8094{
8095 struct list_head *tmp;
fd01b88c 8096 struct mddev *mddev;
2dba6a91 8097 int need_delay = 0;
1da177e4
LT		 8098
8099 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
8100
8101 printk(KERN_INFO "md: stopping all md devices.\n");
8102
2dba6a91 8103 for_each_mddev(mddev, tmp) {
c71d4887 8104 if (mddev_trylock(mddev)) {
2b25000b
N		 8105 /* Force a switch to readonly even array
8106 * appears to still be in use. Hence
8107 * the '100'.
8108 */
a4bd82d0 8109 md_set_readonly(mddev, 100);
c71d4887
NB		 8110 mddev_unlock(mddev);
8111 }
2dba6a91
DB		 8112 need_delay = 1;
8113 }
1da177e4
LT		 8114 /*
8115 * certain more exotic SCSI devices are known to be
8116 * volatile wrt too early system reboots. While the
8117 * right place to handle this issue is the given
8118 * driver, we do want to have a safe RAID driver ...
8119 */
2dba6a91
DB		 8120 if (need_delay)
8121 mdelay(1000*1);
1da177e4
LT		 8122 }
8123 return NOTIFY_DONE;
8124}
8125
75c96f85 8126static struct notifier_block md_notifier = {
1da177e4
LT		 8127 .notifier_call = md_notify_reboot,
8128 .next = NULL,
8129 .priority = INT_MAX, /* before any real devices */
8130};
8131
8132static void md_geninit(void)
8133{
36a4e1fe 8134 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8135
c7705f34 8136 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8137}
8138
75c96f85 8139static int __init md_init(void)
1da177e4 8140{
e804ac78
TH		 8141 int ret = -ENOMEM;
8142
ada609ee 8143 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8144 if (!md_wq)
8145 goto err_wq;
8146
8147 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8148 if (!md_misc_wq)
8149 goto err_misc_wq;
8150
8151 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8152 goto err_md;
8153
8154 if ((ret = register_blkdev(0, "mdp")) < 0)
8155 goto err_mdp;
8156 mdp_major = ret;
8157
3dbd8c2e 8158 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8159 md_probe, NULL, NULL);
8160 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8161 md_probe, NULL, NULL);
8162
1da177e4 8163 register_reboot_notifier(&md_notifier);
0b4d4147 8164 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8165
8166 md_geninit();
d710e138 8167 return 0;
1da177e4 8168
e804ac78
TH		 8169err_mdp:
8170 unregister_blkdev(MD_MAJOR, "md");
8171err_md:
8172 destroy_workqueue(md_misc_wq);
8173err_misc_wq:
8174 destroy_workqueue(md_wq);
8175err_wq:
8176 return ret;
8177}
1da177e4
LT		 8178
8179#ifndef MODULE
8180
8181/*
8182 * Searches all registered partitions for autorun RAID arrays
8183 * at boot time.
8184 */
4d936ec1
ME		 8185
8186static LIST_HEAD(all_detected_devices);
8187struct detected_devices_node {
8188 struct list_head list;
8189 dev_t dev;
8190};
1da177e4
LT		 8191
8192void md_autodetect_dev(dev_t dev)
8193{
4d936ec1
ME		 8194 struct detected_devices_node *node_detected_dev;
8195
8196 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8197 if (node_detected_dev) {
8198 node_detected_dev->dev = dev;
8199 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8200 } else {
8201 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8202 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8203 }
1da177e4
LT		 8204}
8205
8206
8207static void autostart_arrays(int part)
8208{
3cb03002 8209 struct md_rdev *rdev;
4d936ec1
ME		 8210 struct detected_devices_node *node_detected_dev;
8211 dev_t dev;
8212 int i_scanned, i_passed;
1da177e4 8213
4d936ec1
ME		 8214 i_scanned = 0;
8215 i_passed = 0;
1da177e4 8216
4d936ec1 8217 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8218
4d936ec1
ME		 8219 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8220 i_scanned++;
8221 node_detected_dev = list_entry(all_detected_devices.next,
8222 struct detected_devices_node, list);
8223 list_del(&node_detected_dev->list);
8224 dev = node_detected_dev->dev;
8225 kfree(node_detected_dev);
df968c4e 8226 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8227 if (IS_ERR(rdev))
8228 continue;
8229
b2d444d7 8230 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8231 MD_BUG();
8232 continue;
8233 }
d0fae18f 8234 set_bit(AutoDetected, &rdev->flags);
1da177e4 8235 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8236 i_passed++;
1da177e4 8237 }
4d936ec1
ME		 8238
8239 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8240 i_scanned, i_passed);
1da177e4
LT		 8241
8242 autorun_devices(part);
8243}
8244
fdee8ae4 8245#endif /* !MODULE */
1da177e4
LT		 8246
8247static __exit void md_exit(void)
8248{
fd01b88c 8249 struct mddev *mddev;
1da177e4 8250 struct list_head *tmp;
8ab5e4c1 8251
3dbd8c2e 8252 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8253 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8254
3dbd8c2e 8255 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8256 unregister_blkdev(mdp_major, "mdp");
8257 unregister_reboot_notifier(&md_notifier);
8258 unregister_sysctl_table(raid_table_header);
8259 remove_proc_entry("mdstat", NULL);
29ac4aa3 8260 for_each_mddev(mddev, tmp) {
1da177e4 8261 export_array(mddev);
d3374825 8262 mddev->hold_active = 0;
1da177e4 8263 }
e804ac78
TH		 8264 destroy_workqueue(md_misc_wq);
8265 destroy_workqueue(md_wq);
1da177e4
LT		 8266}
8267
685784aa 8268subsys_initcall(md_init);
1da177e4
LT		 8269module_exit(md_exit)
8270
f91de92e
N		 8271static int get_ro(char *buffer, struct kernel_param *kp)
8272{
8273 return sprintf(buffer, "%d", start_readonly);
8274}
8275static int set_ro(const char *val, struct kernel_param *kp)
8276{
8277 char *e;
8278 int num = simple_strtoul(val, &e, 10);
8279 if (*val && (*e == '\0' || *e == '\n')) {
8280 start_readonly = num;
4dbcdc75 8281 return 0;
f91de92e
N		 8282 }
8283 return -EINVAL;
8284}
8285
80ca3a44
N		 8286module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8287module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8288
efeb53c0 8289module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8290
1da177e4
LT		 8291EXPORT_SYMBOL(register_md_personality);
8292EXPORT_SYMBOL(unregister_md_personality);
8293EXPORT_SYMBOL(md_error);
8294EXPORT_SYMBOL(md_done_sync);
8295EXPORT_SYMBOL(md_write_start);
8296EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8297EXPORT_SYMBOL(md_register_thread);
8298EXPORT_SYMBOL(md_unregister_thread);
8299EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 8300EXPORT_SYMBOL(md_check_recovery);
8301MODULE_LICENSE("GPL");
0efb9e61 8302MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8303MODULE_ALIAS("md");
72008652 8304MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux kernel - Deliverables
This page took 1.552965 seconds and 5 git commands to generate.