]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/md/md.c
Linux 3.5-rc7
[mirror_ubuntu-artful-kernel.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);
0d9f4f13
JB
394
395 del_timer_sync(&mddev->safemode_timer);
409c57f3 396}
390ee602 397EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 398
fd01b88c 399void mddev_resume(struct mddev *mddev)
409c57f3
N
400{
401 mddev->suspended = 0;
402 wake_up(&mddev->sb_wait);
403 mddev->pers->quiesce(mddev, 0);
0fd018af 404
47525e59 405 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB
406 md_wakeup_thread(mddev->thread);
407 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 408}
390ee602 409EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 410
fd01b88c 411int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N
412{
413 return mddev->suspended;
414}
415EXPORT_SYMBOL(mddev_congested);
416
a2826aa9 417/*
e9c7469b 418 * Generic flush handling for md
a2826aa9
N
419 */
420
e9c7469b 421static void md_end_flush(struct bio *bio, int err)
a2826aa9 422{
3cb03002 423 struct md_rdev *rdev = bio->bi_private;
fd01b88c 424 struct mddev *mddev = rdev->mddev;
a2826aa9
N
425
426 rdev_dec_pending(rdev, mddev);
427
428 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 429 /* The pre-request flush has finished */
e804ac78 430 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N
431 }
432 bio_put(bio);
433}
434
a7a07e69
N
435static void md_submit_flush_data(struct work_struct *ws);
436
a035fc3e 437static void submit_flushes(struct work_struct *ws)
a2826aa9 438{
fd01b88c 439 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 440 struct md_rdev *rdev;
a2826aa9 441
a7a07e69
N
442 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
443 atomic_set(&mddev->flush_pending, 1);
a2826aa9 444 rcu_read_lock();
dafb20fa 445 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N
446 if (rdev->raid_disk >= 0 &&
447 !test_bit(Faulty, &rdev->flags)) {
448 /* Take two references, one is dropped
449 * when request finishes, one after
450 * we reclaim rcu_read_lock
451 */
452 struct bio *bi;
453 atomic_inc(&rdev->nr_pending);
454 atomic_inc(&rdev->nr_pending);
455 rcu_read_unlock();
b5e1b8ce 456 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
e9c7469b 457 bi->bi_end_io = md_end_flush;
a2826aa9
N
458 bi->bi_private = rdev;
459 bi->bi_bdev = rdev->bdev;
460 atomic_inc(&mddev->flush_pending);
e9c7469b 461 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N
462 rcu_read_lock();
463 rdev_dec_pending(rdev, mddev);
464 }
465 rcu_read_unlock();
a7a07e69
N
466 if (atomic_dec_and_test(&mddev->flush_pending))
467 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N
468}
469
e9c7469b 470static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 471{
fd01b88c 472 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 473 struct bio *bio = mddev->flush_bio;
a2826aa9 474
e9c7469b 475 if (bio->bi_size == 0)
a2826aa9
N
476 /* an empty barrier - all done */
477 bio_endio(bio, 0);
478 else {
e9c7469b 479 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 480 mddev->pers->make_request(mddev, bio);
a2826aa9 481 }
2b74e12e
N
482
483 mddev->flush_bio = NULL;
484 wake_up(&mddev->sb_wait);
a2826aa9
N
485}
486
fd01b88c 487void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N
488{
489 spin_lock_irq(&mddev->write_lock);
490 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 491 !mddev->flush_bio,
a2826aa9 492 mddev->write_lock, /*nothing*/);
e9c7469b 493 mddev->flush_bio = bio;
a2826aa9
N
494 spin_unlock_irq(&mddev->write_lock);
495
a035fc3e
N
496 INIT_WORK(&mddev->flush_work, submit_flushes);
497 queue_work(md_wq, &mddev->flush_work);
a2826aa9 498}
e9c7469b 499EXPORT_SYMBOL(md_flush_request);
409c57f3 500
97658cdd
N
501/* Support for plugging.
502 * This mirrors the plugging support in request_queue, but does not
503 * require having a whole queue or request structures.
504 * We allocate an md_plug_cb for each md device and each thread it gets
505 * plugged on. This links tot the private plug_handle structure in the
506 * personality data where we keep a count of the number of outstanding
507 * plugs so other code can see if a plug is active.
508 */
509struct md_plug_cb {
510 struct blk_plug_cb cb;
fd01b88c 511 struct mddev *mddev;
97658cdd
N
512};
513
514static void plugger_unplug(struct blk_plug_cb *cb)
515{
516 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
517 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
518 md_wakeup_thread(mdcb->mddev->thread);
519 kfree(mdcb);
520}
521
522/* Check that an unplug wakeup will come shortly.
523 * If not, wakeup the md thread immediately
524 */
fd01b88c 525int mddev_check_plugged(struct mddev *mddev)
97658cdd
N
526{
527 struct blk_plug *plug = current->plug;
528 struct md_plug_cb *mdcb;
529
530 if (!plug)
531 return 0;
532
533 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
534 if (mdcb->cb.callback == plugger_unplug &&
535 mdcb->mddev == mddev) {
536 /* Already on the list, move to top */
537 if (mdcb != list_first_entry(&plug->cb_list,
538 struct md_plug_cb,
539 cb.list))
540 list_move(&mdcb->cb.list, &plug->cb_list);
541 return 1;
542 }
543 }
544 /* Not currently on the callback list */
545 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
546 if (!mdcb)
547 return 0;
548
549 mdcb->mddev = mddev;
550 mdcb->cb.callback = plugger_unplug;
551 atomic_inc(&mddev->plug_cnt);
552 list_add(&mdcb->cb.list, &plug->cb_list);
553 return 1;
554}
555EXPORT_SYMBOL_GPL(mddev_check_plugged);
2ac87401 556
fd01b88c 557static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT
558{
559 atomic_inc(&mddev->active);
560 return mddev;
561}
562
5fd3a17e 563static void mddev_delayed_delete(struct work_struct *ws);
d3374825 564
fd01b88c 565static void mddev_put(struct mddev *mddev)
1da177e4 566{
a167f663
N
567 struct bio_set *bs = NULL;
568
1da177e4
LT
569 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
570 return;
d3374825 571 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N
572 mddev->ctime == 0 && !mddev->hold_active) {
573 /* Array is not configured at all, and not held active,
574 * so destroy it */
af8a2434 575 list_del_init(&mddev->all_mddevs);
a167f663
N
576 bs = mddev->bio_set;
577 mddev->bio_set = NULL;
d3374825 578 if (mddev->gendisk) {
e804ac78
TH
579 /* We did a probe so need to clean up. Call
580 * queue_work inside the spinlock so that
581 * flush_workqueue() after mddev_find will
582 * succeed in waiting for the work to be done.
d3374825
N
583 */
584 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 585 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N
586 } else
587 kfree(mddev);
588 }
589 spin_unlock(&all_mddevs_lock);
a167f663
N
590 if (bs)
591 bioset_free(bs);
1da177e4
LT
592}
593
fd01b88c 594void mddev_init(struct mddev *mddev)
fafd7fb0
N
595{
596 mutex_init(&mddev->open_mutex);
597 mutex_init(&mddev->reconfig_mutex);
598 mutex_init(&mddev->bitmap_info.mutex);
599 INIT_LIST_HEAD(&mddev->disks);
600 INIT_LIST_HEAD(&mddev->all_mddevs);
601 init_timer(&mddev->safemode_timer);
602 atomic_set(&mddev->active, 1);
603 atomic_set(&mddev->openers, 0);
604 atomic_set(&mddev->active_io, 0);
97658cdd 605 atomic_set(&mddev->plug_cnt, 0);
fafd7fb0
N
606 spin_lock_init(&mddev->write_lock);
607 atomic_set(&mddev->flush_pending, 0);
608 init_waitqueue_head(&mddev->sb_wait);
609 init_waitqueue_head(&mddev->recovery_wait);
610 mddev->reshape_position = MaxSector;
2c810cdd 611 mddev->reshape_backwards = 0;
fafd7fb0
N
612 mddev->resync_min = 0;
613 mddev->resync_max = MaxSector;
614 mddev->level = LEVEL_NONE;
615}
390ee602 616EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 617
fd01b88c 618static struct mddev * mddev_find(dev_t unit)
1da177e4 619{
fd01b88c 620 struct mddev *mddev, *new = NULL;
1da177e4 621
8f5f02c4
N
622 if (unit && MAJOR(unit) != MD_MAJOR)
623 unit &= ~((1<<MdpMinorShift)-1);
624
1da177e4
LT
625 retry:
626 spin_lock(&all_mddevs_lock);
efeb53c0
N
627
628 if (unit) {
629 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
630 if (mddev->unit == unit) {
631 mddev_get(mddev);
632 spin_unlock(&all_mddevs_lock);
633 kfree(new);
634 return mddev;
635 }
636
637 if (new) {
638 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 639 spin_unlock(&all_mddevs_lock);
efeb53c0
N
640 new->hold_active = UNTIL_IOCTL;
641 return new;
1da177e4 642 }
efeb53c0
N
643 } else if (new) {
644 /* find an unused unit number */
645 static int next_minor = 512;
646 int start = next_minor;
647 int is_free = 0;
648 int dev = 0;
649 while (!is_free) {
650 dev = MKDEV(MD_MAJOR, next_minor);
651 next_minor++;
652 if (next_minor > MINORMASK)
653 next_minor = 0;
654 if (next_minor == start) {
655 /* Oh dear, all in use. */
656 spin_unlock(&all_mddevs_lock);
657 kfree(new);
658 return NULL;
659 }
660
661 is_free = 1;
662 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
663 if (mddev->unit == dev) {
664 is_free = 0;
665 break;
666 }
667 }
668 new->unit = dev;
669 new->md_minor = MINOR(dev);
670 new->hold_active = UNTIL_STOP;
1da177e4
LT
671 list_add(&new->all_mddevs, &all_mddevs);
672 spin_unlock(&all_mddevs_lock);
673 return new;
674 }
675 spin_unlock(&all_mddevs_lock);
676
9ffae0cf 677 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT
678 if (!new)
679 return NULL;
680
1da177e4
LT
681 new->unit = unit;
682 if (MAJOR(unit) == MD_MAJOR)
683 new->md_minor = MINOR(unit);
684 else
685 new->md_minor = MINOR(unit) >> MdpMinorShift;
686
fafd7fb0 687 mddev_init(new);
1da177e4 688
1da177e4
LT
689 goto retry;
690}
691
fd01b88c 692static inline int mddev_lock(struct mddev * mddev)
1da177e4 693{
df5b89b3 694 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT
695}
696
fd01b88c 697static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW
698{
699 return mutex_is_locked(&mddev->reconfig_mutex);
700}
701
fd01b88c 702static inline int mddev_trylock(struct mddev * mddev)
1da177e4 703{
df5b89b3 704 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT
705}
706
b6eb127d
N
707static struct attribute_group md_redundancy_group;
708
fd01b88c 709static void mddev_unlock(struct mddev * mddev)
1da177e4 710{
a64c876f 711 if (mddev->to_remove) {
b6eb127d
N
712 /* These cannot be removed under reconfig_mutex as
713 * an access to the files will try to take reconfig_mutex
714 * while holding the file unremovable, which leads to
715 * a deadlock.
bb4f1e9d
N
716 * So hold set sysfs_active while the remove in happeing,
717 * and anything else which might set ->to_remove or my
718 * otherwise change the sysfs namespace will fail with
719 * -EBUSY if sysfs_active is still set.
720 * We set sysfs_active under reconfig_mutex and elsewhere
721 * test it under the same mutex to ensure its correct value
722 * is seen.
b6eb127d 723 */
a64c876f
N
724 struct attribute_group *to_remove = mddev->to_remove;
725 mddev->to_remove = NULL;
bb4f1e9d 726 mddev->sysfs_active = 1;
b6eb127d
N
727 mutex_unlock(&mddev->reconfig_mutex);
728
00bcb4ac
N
729 if (mddev->kobj.sd) {
730 if (to_remove != &md_redundancy_group)
731 sysfs_remove_group(&mddev->kobj, to_remove);
732 if (mddev->pers == NULL ||
733 mddev->pers->sync_request == NULL) {
734 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
735 if (mddev->sysfs_action)
736 sysfs_put(mddev->sysfs_action);
737 mddev->sysfs_action = NULL;
738 }
a64c876f 739 }
bb4f1e9d 740 mddev->sysfs_active = 0;
b6eb127d
N
741 } else
742 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 743
751e67ca
CD
744 /* As we've dropped the mutex we need a spinlock to
745 * make sure the thread doesn't disappear
01f96c0a
N
746 */
747 spin_lock(&pers_lock);
005eca5e 748 md_wakeup_thread(mddev->thread);
01f96c0a 749 spin_unlock(&pers_lock);
1da177e4
LT
750}
751
fd01b88c 752static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 753{
3cb03002 754 struct md_rdev *rdev;
1da177e4 755
dafb20fa 756 rdev_for_each(rdev, mddev)
1da177e4
LT
757 if (rdev->desc_nr == nr)
758 return rdev;
159ec1fc 759
1da177e4
LT
760 return NULL;
761}
762
fd01b88c 763static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
1da177e4 764{
3cb03002 765 struct md_rdev *rdev;
1da177e4 766
dafb20fa 767 rdev_for_each(rdev, mddev)
1da177e4
LT
768 if (rdev->bdev->bd_dev == dev)
769 return rdev;
159ec1fc 770
1da177e4
LT
771 return NULL;
772}
773
84fc4b56 774static struct md_personality *find_pers(int level, char *clevel)
2604b703 775{
84fc4b56 776 struct md_personality *pers;
d9d166c2
N
777 list_for_each_entry(pers, &pers_list, list) {
778 if (level != LEVEL_NONE && pers->level == level)
2604b703 779 return pers;
d9d166c2
N
780 if (strcmp(pers->name, clevel)==0)
781 return pers;
782 }
2604b703
N
783 return NULL;
784}
785
b73df2d3 786/* return the offset of the super block in 512byte sectors */
3cb03002 787static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 788{
57b2caa3 789 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 790 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT
791}
792
3cb03002 793static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT
794{
795 if (rdev->sb_page)
796 MD_BUG();
797
798 rdev->sb_page = alloc_page(GFP_KERNEL);
799 if (!rdev->sb_page) {
800 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 801 return -ENOMEM;
1da177e4
LT
802 }
803
804 return 0;
805}
806
545c8795 807void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT
808{
809 if (rdev->sb_page) {
2d1f3b5d 810 put_page(rdev->sb_page);
1da177e4
LT
811 rdev->sb_loaded = 0;
812 rdev->sb_page = NULL;
0f420358 813 rdev->sb_start = 0;
dd8ac336 814 rdev->sectors = 0;
1da177e4 815 }
2699b672
N
816 if (rdev->bb_page) {
817 put_page(rdev->bb_page);
818 rdev->bb_page = NULL;
819 }
4fa2f327
N
820 kfree(rdev->badblocks.page);
821 rdev->badblocks.page = NULL;
1da177e4 822}
545c8795 823EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 824
6712ecf8 825static void super_written(struct bio *bio, int error)
7bfa19f2 826{
3cb03002 827 struct md_rdev *rdev = bio->bi_private;
fd01b88c 828 struct mddev *mddev = rdev->mddev;
7bfa19f2 829
3a0f5bbb
N
830 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
831 printk("md: super_written gets error=%d, uptodate=%d\n",
832 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
833 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 834 md_error(mddev, rdev);
3a0f5bbb 835 }
7bfa19f2 836
a9701a30
N
837 if (atomic_dec_and_test(&mddev->pending_writes))
838 wake_up(&mddev->sb_wait);
f8b58edf 839 bio_put(bio);
7bfa19f2
N
840}
841
fd01b88c 842void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N
843 sector_t sector, int size, struct page *page)
844{
845 /* write first size bytes of page to sector of rdev
846 * Increment mddev->pending_writes before returning
847 * and decrement it on completion, waking up sb_wait
848 * if zero is reached.
849 * If an error occurred, call md_error
850 */
a167f663 851 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 852
a6ff7e08 853 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N
854 bio->bi_sector = sector;
855 bio_add_page(bio, page, size, 0);
856 bio->bi_private = rdev;
857 bio->bi_end_io = super_written;
a9701a30 858
7bfa19f2 859 atomic_inc(&mddev->pending_writes);
a5bf4df0 860 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N
861}
862
fd01b88c 863void md_super_wait(struct mddev *mddev)
a9701a30 864{
e9c7469b 865 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N
866 DEFINE_WAIT(wq);
867 for(;;) {
868 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
869 if (atomic_read(&mddev->pending_writes)==0)
870 break;
a9701a30
N
871 schedule();
872 }
873 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N
874}
875
6712ecf8 876static void bi_complete(struct bio *bio, int error)
1da177e4 877{
1da177e4 878 complete((struct completion*)bio->bi_private);
1da177e4
LT
879}
880
3cb03002 881int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 882 struct page *page, int rw, bool metadata_op)
1da177e4 883{
a167f663 884 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT
885 struct completion event;
886 int ret;
887
721a9602 888 rw |= REQ_SYNC;
1da177e4 889
a6ff7e08
JB
890 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
891 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB
892 if (metadata_op)
893 bio->bi_sector = sector + rdev->sb_start;
1fdd6fc9
N
894 else if (rdev->mddev->reshape_position != MaxSector &&
895 (rdev->mddev->reshape_backwards ==
896 (sector >= rdev->mddev->reshape_position)))
897 bio->bi_sector = sector + rdev->new_data_offset;
ccebd4c4
JB
898 else
899 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT
900 bio_add_page(bio, page, size, 0);
901 init_completion(&event);
902 bio->bi_private = &event;
903 bio->bi_end_io = bi_complete;
904 submit_bio(rw, bio);
905 wait_for_completion(&event);
906
907 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
908 bio_put(bio);
909 return ret;
910}
a8745db2 911EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 912
3cb03002 913static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT
914{
915 char b[BDEVNAME_SIZE];
916 if (!rdev->sb_page) {
917 MD_BUG();
918 return -EINVAL;
919 }
920 if (rdev->sb_loaded)
921 return 0;
922
923
ccebd4c4 924 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT
925 goto fail;
926 rdev->sb_loaded = 1;
927 return 0;
928
929fail:
930 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
931 bdevname(rdev->bdev,b));
932 return -EINVAL;
933}
934
935static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
936{
05710466
AN
937 return sb1->set_uuid0 == sb2->set_uuid0 &&
938 sb1->set_uuid1 == sb2->set_uuid1 &&
939 sb1->set_uuid2 == sb2->set_uuid2 &&
940 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT
941}
942
1da177e4
LT
943static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
944{
945 int ret;
946 mdp_super_t *tmp1, *tmp2;
947
948 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
949 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
950
951 if (!tmp1 || !tmp2) {
952 ret = 0;
35020f1a 953 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT
954 goto abort;
955 }
956
957 *tmp1 = *sb1;
958 *tmp2 = *sb2;
959
960 /*
961 * nr_disks is not constant
962 */
963 tmp1->nr_disks = 0;
964 tmp2->nr_disks = 0;
965
ce0c8e05 966 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 967abort:
990a8baf
JJ
968 kfree(tmp1);
969 kfree(tmp2);
1da177e4
LT
970 return ret;
971}
972
4d167f09
N
973
974static u32 md_csum_fold(u32 csum)
975{
976 csum = (csum & 0xffff) + (csum >> 16);
977 return (csum & 0xffff) + (csum >> 16);
978}
979
1da177e4
LT
980static unsigned int calc_sb_csum(mdp_super_t * sb)
981{
4d167f09
N
982 u64 newcsum = 0;
983 u32 *sb32 = (u32*)sb;
984 int i;
1da177e4
LT
985 unsigned int disk_csum, csum;
986
987 disk_csum = sb->sb_csum;
988 sb->sb_csum = 0;
4d167f09
N
989
990 for (i = 0; i < MD_SB_BYTES/4 ; i++)
991 newcsum += sb32[i];
992 csum = (newcsum & 0xffffffff) + (newcsum>>32);
993
994
995#ifdef CONFIG_ALPHA
996 /* This used to use csum_partial, which was wrong for several
997 * reasons including that different results are returned on
998 * different architectures. It isn't critical that we get exactly
999 * the same return value as before (we always csum_fold before
1000 * testing, and that removes any differences). However as we
1001 * know that csum_partial always returned a 16bit value on
1002 * alphas, do a fold to maximise conformity to previous behaviour.
1003 */
1004 sb->sb_csum = md_csum_fold(disk_csum);
1005#else
1da177e4 1006 sb->sb_csum = disk_csum;
4d167f09 1007#endif
1da177e4
LT
1008 return csum;
1009}
1010
1011
1012/*
1013 * Handle superblock details.
1014 * We want to be able to handle multiple superblock formats
1015 * so we have a common interface to them all, and an array of
1016 * different handlers.
1017 * We rely on user-space to write the initial superblock, and support
1018 * reading and updating of superblocks.
1019 * Interface methods are:
3cb03002 1020 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT
1021 * loads and validates a superblock on dev.
1022 * if refdev != NULL, compare superblocks on both devices
1023 * Return:
1024 * 0 - dev has a superblock that is compatible with refdev
1025 * 1 - dev has a superblock that is compatible and newer than refdev
1026 * so dev should be used as the refdev in future
1027 * -EINVAL superblock incompatible or invalid
1028 * -othererror e.g. -EIO
1029 *
fd01b88c 1030 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT
1031 * Verify that dev is acceptable into mddev.
1032 * The first time, mddev->raid_disks will be 0, and data from
1033 * dev should be merged in. Subsequent calls check that dev
1034 * is new enough. Return 0 or -EINVAL
1035 *
fd01b88c 1036 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT
1037 * Update the superblock for rdev with data in mddev
1038 * This does not write to disc.
1039 *
1040 */
1041
1042struct super_type {
0cd17fec
CW
1043 char *name;
1044 struct module *owner;
c6563a8c
N
1045 int (*load_super)(struct md_rdev *rdev,
1046 struct md_rdev *refdev,
0cd17fec 1047 int minor_version);
c6563a8c
N
1048 int (*validate_super)(struct mddev *mddev,
1049 struct md_rdev *rdev);
1050 void (*sync_super)(struct mddev *mddev,
1051 struct md_rdev *rdev);
3cb03002 1052 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 1053 sector_t num_sectors);
c6563a8c
N
1054 int (*allow_new_offset)(struct md_rdev *rdev,
1055 unsigned long long new_offset);
1da177e4
LT
1056};
1057
0894cc30
AN
1058/*
1059 * Check that the given mddev has no bitmap.
1060 *
1061 * This function is called from the run method of all personalities that do not
1062 * support bitmaps. It prints an error message and returns non-zero if mddev
1063 * has a bitmap. Otherwise, it returns 0.
1064 *
1065 */
fd01b88c 1066int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1067{
c3d9714e 1068 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN
1069 return 0;
1070 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1071 mdname(mddev), mddev->pers->name);
1072 return 1;
1073}
1074EXPORT_SYMBOL(md_check_no_bitmap);
1075
1da177e4
LT
1076/*
1077 * load_super for 0.90.0
1078 */
3cb03002 1079static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT
1080{
1081 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1082 mdp_super_t *sb;
1083 int ret;
1da177e4
LT
1084
1085 /*
0f420358 1086 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT
1087 * it's at the end of the disk.
1088 *
1089 * It also happens to be a multiple of 4Kb.
1090 */
57b2caa3 1091 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1092
0002b271 1093 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT
1094 if (ret) return ret;
1095
1096 ret = -EINVAL;
1097
1098 bdevname(rdev->bdev, b);
65a06f06 1099 sb = page_address(rdev->sb_page);
1da177e4
LT
1100
1101 if (sb->md_magic != MD_SB_MAGIC) {
1102 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1103 b);
1104 goto abort;
1105 }
1106
1107 if (sb->major_version != 0 ||
f6705578
N
1108 sb->minor_version < 90 ||
1109 sb->minor_version > 91) {
1da177e4
LT
1110 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1111 sb->major_version, sb->minor_version,
1112 b);
1113 goto abort;
1114 }
1115
1116 if (sb->raid_disks <= 0)
1117 goto abort;
1118
4d167f09 1119 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT
1120 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1121 b);
1122 goto abort;
1123 }
1124
1125 rdev->preferred_minor = sb->md_minor;
1126 rdev->data_offset = 0;
c6563a8c 1127 rdev->new_data_offset = 0;
0002b271 1128 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1129 rdev->badblocks.shift = -1;
1da177e4
LT
1130
1131 if (sb->level == LEVEL_MULTIPATH)
1132 rdev->desc_nr = -1;
1133 else
1134 rdev->desc_nr = sb->this_disk.number;
1135
9a7b2b0f 1136 if (!refdev) {
1da177e4 1137 ret = 1;
9a7b2b0f 1138 } else {
1da177e4 1139 __u64 ev1, ev2;
65a06f06 1140 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT
1141 if (!uuid_equal(refsb, sb)) {
1142 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1143 b, bdevname(refdev->bdev,b2));
1144 goto abort;
1145 }
1146 if (!sb_equal(refsb, sb)) {
1147 printk(KERN_WARNING "md: %s has same UUID"
1148 " but different superblock to %s\n",
1149 b, bdevname(refdev->bdev, b2));
1150 goto abort;
1151 }
1152 ev1 = md_event(sb);
1153 ev2 = md_event(refsb);
1154 if (ev1 > ev2)
1155 ret = 1;
1156 else
1157 ret = 0;
1158 }
8190e754 1159 rdev->sectors = rdev->sb_start;
27a7b260
N
1160 /* Limit to 4TB as metadata cannot record more than that */
1161 if (rdev->sectors >= (2ULL << 32))
1162 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1163
27a7b260 1164 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N
1165 /* "this cannot possibly happen" ... */
1166 ret = -EINVAL;
1167
1da177e4
LT
1168 abort:
1169 return ret;
1170}
1171
1172/*
1173 * validate_super for 0.90.0
1174 */
fd01b88c 1175static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
1176{
1177 mdp_disk_t *desc;
65a06f06 1178 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1179 __u64 ev1 = md_event(sb);
1da177e4 1180
41158c7e 1181 rdev->raid_disk = -1;
c5d79adb
N
1182 clear_bit(Faulty, &rdev->flags);
1183 clear_bit(In_sync, &rdev->flags);
1184 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1185
1da177e4
LT
1186 if (mddev->raid_disks == 0) {
1187 mddev->major_version = 0;
1188 mddev->minor_version = sb->minor_version;
1189 mddev->patch_version = sb->patch_version;
e691063a 1190 mddev->external = 0;
9d8f0363 1191 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT
1192 mddev->ctime = sb->ctime;
1193 mddev->utime = sb->utime;
1194 mddev->level = sb->level;
d9d166c2 1195 mddev->clevel[0] = 0;
1da177e4
LT
1196 mddev->layout = sb->layout;
1197 mddev->raid_disks = sb->raid_disks;
27a7b260 1198 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1199 mddev->events = ev1;
c3d9714e 1200 mddev->bitmap_info.offset = 0;
6409bb05
N
1201 mddev->bitmap_info.space = 0;
1202 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1203 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1204 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1205 mddev->reshape_backwards = 0;
1da177e4 1206
f6705578
N
1207 if (mddev->minor_version >= 91) {
1208 mddev->reshape_position = sb->reshape_position;
1209 mddev->delta_disks = sb->delta_disks;
1210 mddev->new_level = sb->new_level;
1211 mddev->new_layout = sb->new_layout;
664e7c41 1212 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N
1213 if (mddev->delta_disks < 0)
1214 mddev->reshape_backwards = 1;
f6705578
N
1215 } else {
1216 mddev->reshape_position = MaxSector;
1217 mddev->delta_disks = 0;
1218 mddev->new_level = mddev->level;
1219 mddev->new_layout = mddev->layout;
664e7c41 1220 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N
1221 }
1222
1da177e4
LT
1223 if (sb->state & (1<<MD_SB_CLEAN))
1224 mddev->recovery_cp = MaxSector;
1225 else {
1226 if (sb->events_hi == sb->cp_events_hi &&
1227 sb->events_lo == sb->cp_events_lo) {
1228 mddev->recovery_cp = sb->recovery_cp;
1229 } else
1230 mddev->recovery_cp = 0;
1231 }
1232
1233 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1234 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1235 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1236 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1237
1238 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N
1239
1240 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1241 mddev->bitmap_info.file == NULL) {
c3d9714e
N
1242 mddev->bitmap_info.offset =
1243 mddev->bitmap_info.default_offset;
6409bb05
N
1244 mddev->bitmap_info.space =
1245 mddev->bitmap_info.space;
1246 }
a654b9d8 1247
41158c7e 1248 } else if (mddev->pers == NULL) {
be6800a7
N
1249 /* Insist on good event counter while assembling, except
1250 * for spares (which don't need an event count) */
1da177e4 1251 ++ev1;
be6800a7
N
1252 if (sb->disks[rdev->desc_nr].state & (
1253 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1254 if (ev1 < mddev->events)
1255 return -EINVAL;
41158c7e
N
1256 } else if (mddev->bitmap) {
1257 /* if adding to array with a bitmap, then we can accept an
1258 * older device ... but not too old.
1259 */
41158c7e
N
1260 if (ev1 < mddev->bitmap->events_cleared)
1261 return 0;
07d84d10
N
1262 } else {
1263 if (ev1 < mddev->events)
1264 /* just a hot-add of a new device, leave raid_disk at -1 */
1265 return 0;
1266 }
41158c7e 1267
1da177e4 1268 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT
1269 desc = sb->disks + rdev->desc_nr;
1270
1271 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1272 set_bit(Faulty, &rdev->flags);
7c7546cc
N
1273 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1274 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1275 set_bit(In_sync, &rdev->flags);
1da177e4 1276 rdev->raid_disk = desc->raid_disk;
0261cd9f
N
1277 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1278 /* active but not in sync implies recovery up to
1279 * reshape position. We don't know exactly where
1280 * that is, so set to zero for now */
1281 if (mddev->minor_version >= 91) {
1282 rdev->recovery_offset = 0;
1283 rdev->raid_disk = desc->raid_disk;
1284 }
1da177e4 1285 }
8ddf9efe
N
1286 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1287 set_bit(WriteMostly, &rdev->flags);
41158c7e 1288 } else /* MULTIPATH are always insync */
b2d444d7 1289 set_bit(In_sync, &rdev->flags);
1da177e4
LT
1290 return 0;
1291}
1292
1293/*
1294 * sync_super for 0.90.0
1295 */
fd01b88c 1296static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
1297{
1298 mdp_super_t *sb;
3cb03002 1299 struct md_rdev *rdev2;
1da177e4 1300 int next_spare = mddev->raid_disks;
19133a42 1301
1da177e4
LT
1302
1303 /* make rdev->sb match mddev data..
1304 *
1305 * 1/ zero out disks
1306 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1307 * 3/ any empty disks < next_spare become removed
1308 *
1309 * disks[0] gets initialised to REMOVED because
1310 * we cannot be sure from other fields if it has
1311 * been initialised or not.
1312 */
1313 int i;
1314 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1315
61181565
N
1316 rdev->sb_size = MD_SB_BYTES;
1317
65a06f06 1318 sb = page_address(rdev->sb_page);
1da177e4
LT
1319
1320 memset(sb, 0, sizeof(*sb));
1321
1322 sb->md_magic = MD_SB_MAGIC;
1323 sb->major_version = mddev->major_version;
1da177e4
LT
1324 sb->patch_version = mddev->patch_version;
1325 sb->gvalid_words = 0; /* ignored */
1326 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1327 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1328 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1329 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1330
1331 sb->ctime = mddev->ctime;
1332 sb->level = mddev->level;
58c0fed4 1333 sb->size = mddev->dev_sectors / 2;
1da177e4
LT
1334 sb->raid_disks = mddev->raid_disks;
1335 sb->md_minor = mddev->md_minor;
e691063a 1336 sb->not_persistent = 0;
1da177e4
LT
1337 sb->utime = mddev->utime;
1338 sb->state = 0;
1339 sb->events_hi = (mddev->events>>32);
1340 sb->events_lo = (u32)mddev->events;
1341
f6705578
N
1342 if (mddev->reshape_position == MaxSector)
1343 sb->minor_version = 90;
1344 else {
1345 sb->minor_version = 91;
1346 sb->reshape_position = mddev->reshape_position;
1347 sb->new_level = mddev->new_level;
1348 sb->delta_disks = mddev->delta_disks;
1349 sb->new_layout = mddev->new_layout;
664e7c41 1350 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N
1351 }
1352 mddev->minor_version = sb->minor_version;
1da177e4
LT
1353 if (mddev->in_sync)
1354 {
1355 sb->recovery_cp = mddev->recovery_cp;
1356 sb->cp_events_hi = (mddev->events>>32);
1357 sb->cp_events_lo = (u32)mddev->events;
1358 if (mddev->recovery_cp == MaxSector)
1359 sb->state = (1<< MD_SB_CLEAN);
1360 } else
1361 sb->recovery_cp = 0;
1362
1363 sb->layout = mddev->layout;
9d8f0363 1364 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1365
c3d9714e 1366 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N
1367 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1368
1da177e4 1369 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1370 rdev_for_each(rdev2, mddev) {
1da177e4 1371 mdp_disk_t *d;
86e6ffdd 1372 int desc_nr;
0261cd9f
N
1373 int is_active = test_bit(In_sync, &rdev2->flags);
1374
1375 if (rdev2->raid_disk >= 0 &&
1376 sb->minor_version >= 91)
1377 /* we have nowhere to store the recovery_offset,
1378 * but if it is not below the reshape_position,
1379 * we can piggy-back on that.
1380 */
1381 is_active = 1;
1382 if (rdev2->raid_disk < 0 ||
1383 test_bit(Faulty, &rdev2->flags))
1384 is_active = 0;
1385 if (is_active)
86e6ffdd 1386 desc_nr = rdev2->raid_disk;
1da177e4 1387 else
86e6ffdd 1388 desc_nr = next_spare++;
19133a42 1389 rdev2->desc_nr = desc_nr;
1da177e4
LT
1390 d = &sb->disks[rdev2->desc_nr];
1391 nr_disks++;
1392 d->number = rdev2->desc_nr;
1393 d->major = MAJOR(rdev2->bdev->bd_dev);
1394 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1395 if (is_active)
1da177e4
LT
1396 d->raid_disk = rdev2->raid_disk;
1397 else
1398 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1399 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1400 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1401 else if (is_active) {
1da177e4 1402 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N
1403 if (test_bit(In_sync, &rdev2->flags))
1404 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT
1405 active++;
1406 working++;
1407 } else {
1408 d->state = 0;
1409 spare++;
1410 working++;
1411 }
8ddf9efe
N
1412 if (test_bit(WriteMostly, &rdev2->flags))
1413 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1414 }
1da177e4
LT
1415 /* now set the "removed" and "faulty" bits on any missing devices */
1416 for (i=0 ; i < mddev->raid_disks ; i++) {
1417 mdp_disk_t *d = &sb->disks[i];
1418 if (d->state == 0 && d->number == 0) {
1419 d->number = i;
1420 d->raid_disk = i;
1421 d->state = (1<<MD_DISK_REMOVED);
1422 d->state |= (1<<MD_DISK_FAULTY);
1423 failed++;
1424 }
1425 }
1426 sb->nr_disks = nr_disks;
1427 sb->active_disks = active;
1428 sb->working_disks = working;
1429 sb->failed_disks = failed;
1430 sb->spare_disks = spare;
1431
1432 sb->this_disk = sb->disks[rdev->desc_nr];
1433 sb->sb_csum = calc_sb_csum(sb);
1434}
1435
0cd17fec
CW
1436/*
1437 * rdev_size_change for 0.90.0
1438 */
1439static unsigned long long
3cb03002 1440super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1441{
58c0fed4 1442 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1443 return 0; /* component must fit device */
c3d9714e 1444 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1445 return 0; /* can't move bitmap */
57b2caa3 1446 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN
1447 if (!num_sectors || num_sectors > rdev->sb_start)
1448 num_sectors = rdev->sb_start;
27a7b260
N
1449 /* Limit to 4TB as metadata cannot record more than that.
1450 * 4TB == 2^32 KB, or 2*2^32 sectors.
1451 */
1452 if (num_sectors >= (2ULL << 32))
1453 num_sectors = (2ULL << 32) - 2;
0f420358 1454 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW
1455 rdev->sb_page);
1456 md_super_wait(rdev->mddev);
c26a44ed 1457 return num_sectors;
0cd17fec
CW
1458}
1459
c6563a8c
N
1460static int
1461super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1462{
1463 /* non-zero offset changes not possible with v0.90 */
1464 return new_offset == 0;
1465}
0cd17fec 1466
1da177e4
LT
1467/*
1468 * version 1 superblock
1469 */
1470
1c05b4bc 1471static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1472{
1c05b4bc
N
1473 __le32 disk_csum;
1474 u32 csum;
1da177e4
LT
1475 unsigned long long newcsum;
1476 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1477 __le32 *isuper = (__le32*)sb;
1da177e4
LT
1478 int i;
1479
1480 disk_csum = sb->sb_csum;
1481 sb->sb_csum = 0;
1482 newcsum = 0;
1483 for (i=0; size>=4; size -= 4 )
1484 newcsum += le32_to_cpu(*isuper++);
1485
1486 if (size == 2)
1c05b4bc 1487 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT
1488
1489 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1490 sb->sb_csum = disk_csum;
1491 return cpu_to_le32(csum);
1492}
1493
2699b672
N
1494static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1495 int acknowledged);
3cb03002 1496static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT
1497{
1498 struct mdp_superblock_1 *sb;
1499 int ret;
0f420358 1500 sector_t sb_start;
c6563a8c 1501 sector_t sectors;
1da177e4 1502 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1503 int bmask;
1da177e4
LT
1504
1505 /*
0f420358 1506 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT
1507 * It is always aligned to a 4K boundary and
1508 * depeding on minor_version, it can be:
1509 * 0: At least 8K, but less than 12K, from end of device
1510 * 1: At start of device
1511 * 2: 4K from start of device.
1512 */
1513 switch(minor_version) {
1514 case 0:
77304d2a 1515 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN
1516 sb_start -= 8*2;
1517 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT
1518 break;
1519 case 1:
0f420358 1520 sb_start = 0;
1da177e4
LT
1521 break;
1522 case 2:
0f420358 1523 sb_start = 8;
1da177e4
LT
1524 break;
1525 default:
1526 return -EINVAL;
1527 }
0f420358 1528 rdev->sb_start = sb_start;
1da177e4 1529
0002b271
N
1530 /* superblock is rarely larger than 1K, but it can be larger,
1531 * and it is safe to read 4k, so we do that
1532 */
1533 ret = read_disk_sb(rdev, 4096);
1da177e4
LT
1534 if (ret) return ret;
1535
1536
65a06f06 1537 sb = page_address(rdev->sb_page);
1da177e4
LT
1538
1539 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1540 sb->major_version != cpu_to_le32(1) ||
1541 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1542 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1543 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT
1544 return -EINVAL;
1545
1546 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1547 printk("md: invalid superblock checksum on %s\n",
1548 bdevname(rdev->bdev,b));
1549 return -EINVAL;
1550 }
1551 if (le64_to_cpu(sb->data_size) < 10) {
1552 printk("md: data_size too small on %s\n",
1553 bdevname(rdev->bdev,b));
1554 return -EINVAL;
1555 }
c6563a8c
N
1556 if (sb->pad0 ||
1557 sb->pad3[0] ||
1558 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1559 /* Some padding is non-zero, might be a new feature */
1560 return -EINVAL;
e11e93fa 1561
1da177e4
LT
1562 rdev->preferred_minor = 0xffff;
1563 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N
1564 rdev->new_data_offset = rdev->data_offset;
1565 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1566 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1567 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1568 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1569
0002b271 1570 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1571 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1572 if (rdev->sb_size & bmask)
a1801f85
N
1573 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1574
1575 if (minor_version
0f420358 1576 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1577 return -EINVAL;
c6563a8c
N
1578 if (minor_version
1579 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1580 return -EINVAL;
0002b271 1581
31b65a0d
N
1582 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1583 rdev->desc_nr = -1;
1584 else
1585 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1586
2699b672
N
1587 if (!rdev->bb_page) {
1588 rdev->bb_page = alloc_page(GFP_KERNEL);
1589 if (!rdev->bb_page)
1590 return -ENOMEM;
1591 }
1592 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1593 rdev->badblocks.count == 0) {
1594 /* need to load the bad block list.
1595 * Currently we limit it to one page.
1596 */
1597 s32 offset;
1598 sector_t bb_sector;
1599 u64 *bbp;
1600 int i;
1601 int sectors = le16_to_cpu(sb->bblog_size);
1602 if (sectors > (PAGE_SIZE / 512))
1603 return -EINVAL;
1604 offset = le32_to_cpu(sb->bblog_offset);
1605 if (offset == 0)
1606 return -EINVAL;
1607 bb_sector = (long long)offset;
1608 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1609 rdev->bb_page, READ, true))
1610 return -EIO;
1611 bbp = (u64 *)page_address(rdev->bb_page);
1612 rdev->badblocks.shift = sb->bblog_shift;
1613 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1614 u64 bb = le64_to_cpu(*bbp);
1615 int count = bb & (0x3ff);
1616 u64 sector = bb >> 10;
1617 sector <<= sb->bblog_shift;
1618 count <<= sb->bblog_shift;
1619 if (bb + 1 == 0)
1620 break;
1621 if (md_set_badblocks(&rdev->badblocks,
1622 sector, count, 1) == 0)
1623 return -EINVAL;
1624 }
1625 } else if (sb->bblog_offset == 0)
1626 rdev->badblocks.shift = -1;
1627
9a7b2b0f 1628 if (!refdev) {
8ed75463 1629 ret = 1;
9a7b2b0f 1630 } else {
1da177e4 1631 __u64 ev1, ev2;
65a06f06 1632 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT
1633
1634 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1635 sb->level != refsb->level ||
1636 sb->layout != refsb->layout ||
1637 sb->chunksize != refsb->chunksize) {
1638 printk(KERN_WARNING "md: %s has strangely different"
1639 " superblock to %s\n",
1640 bdevname(rdev->bdev,b),
1641 bdevname(refdev->bdev,b2));
1642 return -EINVAL;
1643 }
1644 ev1 = le64_to_cpu(sb->events);
1645 ev2 = le64_to_cpu(refsb->events);
1646
1647 if (ev1 > ev2)
8ed75463
N
1648 ret = 1;
1649 else
1650 ret = 0;
1da177e4 1651 }
c6563a8c
N
1652 if (minor_version) {
1653 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1654 sectors -= rdev->data_offset;
1655 } else
1656 sectors = rdev->sb_start;
1657 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1658 return -EINVAL;
dd8ac336 1659 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1660 return ret;
1da177e4
LT
1661}
1662
fd01b88c 1663static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1664{
65a06f06 1665 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1666 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1667
41158c7e 1668 rdev->raid_disk = -1;
c5d79adb
N
1669 clear_bit(Faulty, &rdev->flags);
1670 clear_bit(In_sync, &rdev->flags);
1671 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1672
1da177e4
LT
1673 if (mddev->raid_disks == 0) {
1674 mddev->major_version = 1;
1675 mddev->patch_version = 0;
e691063a 1676 mddev->external = 0;
9d8f0363 1677 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT
1678 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1679 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1680 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1681 mddev->clevel[0] = 0;
1da177e4
LT
1682 mddev->layout = le32_to_cpu(sb->layout);
1683 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1684 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1685 mddev->events = ev1;
c3d9714e 1686 mddev->bitmap_info.offset = 0;
6409bb05
N
1687 mddev->bitmap_info.space = 0;
1688 /* Default location for bitmap is 1K after superblock
1689 * using 3K - total of 4K
1690 */
c3d9714e 1691 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1692 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N
1693 mddev->reshape_backwards = 0;
1694
1da177e4
LT
1695 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1696 memcpy(mddev->uuid, sb->set_uuid, 16);
1697
1698 mddev->max_disks = (4096-256)/2;
a654b9d8 1699
71c0805c 1700 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1701 mddev->bitmap_info.file == NULL) {
c3d9714e
N
1702 mddev->bitmap_info.offset =
1703 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N
1704 /* Metadata doesn't record how much space is available.
1705 * For 1.0, we assume we can use up to the superblock
1706 * if before, else to 4K beyond superblock.
1707 * For others, assume no change is possible.
1708 */
1709 if (mddev->minor_version > 0)
1710 mddev->bitmap_info.space = 0;
1711 else if (mddev->bitmap_info.offset > 0)
1712 mddev->bitmap_info.space =
1713 8 - mddev->bitmap_info.offset;
1714 else
1715 mddev->bitmap_info.space =
1716 -mddev->bitmap_info.offset;
1717 }
e11e93fa 1718
f6705578
N
1719 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1720 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1721 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1722 mddev->new_level = le32_to_cpu(sb->new_level);
1723 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1724 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N
1725 if (mddev->delta_disks < 0 ||
1726 (mddev->delta_disks == 0 &&
1727 (le32_to_cpu(sb->feature_map)
1728 & MD_FEATURE_RESHAPE_BACKWARDS)))
1729 mddev->reshape_backwards = 1;
f6705578
N
1730 } else {
1731 mddev->reshape_position = MaxSector;
1732 mddev->delta_disks = 0;
1733 mddev->new_level = mddev->level;
1734 mddev->new_layout = mddev->layout;
664e7c41 1735 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N
1736 }
1737
41158c7e 1738 } else if (mddev->pers == NULL) {
be6800a7
N
1739 /* Insist of good event counter while assembling, except for
1740 * spares (which don't need an event count) */
1da177e4 1741 ++ev1;
be6800a7
N
1742 if (rdev->desc_nr >= 0 &&
1743 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1744 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1745 if (ev1 < mddev->events)
1746 return -EINVAL;
41158c7e
N
1747 } else if (mddev->bitmap) {
1748 /* If adding to array with a bitmap, then we can accept an
1749 * older device, but not too old.
1750 */
41158c7e
N
1751 if (ev1 < mddev->bitmap->events_cleared)
1752 return 0;
07d84d10
N
1753 } else {
1754 if (ev1 < mddev->events)
1755 /* just a hot-add of a new device, leave raid_disk at -1 */
1756 return 0;
1757 }
1da177e4
LT
1758 if (mddev->level != LEVEL_MULTIPATH) {
1759 int role;
3673f305
N
1760 if (rdev->desc_nr < 0 ||
1761 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1762 role = 0xffff;
1763 rdev->desc_nr = -1;
1764 } else
1765 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT
1766 switch(role) {
1767 case 0xffff: /* spare */
1da177e4
LT
1768 break;
1769 case 0xfffe: /* faulty */
b2d444d7 1770 set_bit(Faulty, &rdev->flags);
1da177e4
LT
1771 break;
1772 default:
5fd6c1dc
N
1773 if ((le32_to_cpu(sb->feature_map) &
1774 MD_FEATURE_RECOVERY_OFFSET))
1775 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1776 else
1777 set_bit(In_sync, &rdev->flags);
1da177e4
LT
1778 rdev->raid_disk = role;
1779 break;
1780 }
8ddf9efe
N
1781 if (sb->devflags & WriteMostly1)
1782 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N
1783 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1784 set_bit(Replacement, &rdev->flags);
41158c7e 1785 } else /* MULTIPATH are always insync */
b2d444d7 1786 set_bit(In_sync, &rdev->flags);
41158c7e 1787
1da177e4
LT
1788 return 0;
1789}
1790
fd01b88c 1791static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
1792{
1793 struct mdp_superblock_1 *sb;
3cb03002 1794 struct md_rdev *rdev2;
1da177e4
LT
1795 int max_dev, i;
1796 /* make rdev->sb match mddev and rdev data. */
1797
65a06f06 1798 sb = page_address(rdev->sb_page);
1da177e4
LT
1799
1800 sb->feature_map = 0;
1801 sb->pad0 = 0;
5fd6c1dc 1802 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT
1803 memset(sb->pad3, 0, sizeof(sb->pad3));
1804
1805 sb->utime = cpu_to_le64((__u64)mddev->utime);
1806 sb->events = cpu_to_le64(mddev->events);
1807 if (mddev->in_sync)
1808 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1809 else
1810 sb->resync_offset = cpu_to_le64(0);
1811
1c05b4bc 1812 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1813
f0ca340c 1814 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1815 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1816 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N
1817 sb->level = cpu_to_le32(mddev->level);
1818 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1819
aeb9b211
N
1820 if (test_bit(WriteMostly, &rdev->flags))
1821 sb->devflags |= WriteMostly1;
1822 else
1823 sb->devflags &= ~WriteMostly1;
c6563a8c
N
1824 sb->data_offset = cpu_to_le64(rdev->data_offset);
1825 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1826
c3d9714e
N
1827 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1828 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1829 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1830 }
5fd6c1dc
N
1831
1832 if (rdev->raid_disk >= 0 &&
97e4f42d 1833 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N
1834 sb->feature_map |=
1835 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1836 sb->recovery_offset =
1837 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc 1838 }
2d78f8c4
N
1839 if (test_bit(Replacement, &rdev->flags))
1840 sb->feature_map |=
1841 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1842
f6705578
N
1843 if (mddev->reshape_position != MaxSector) {
1844 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1845 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1846 sb->new_layout = cpu_to_le32(mddev->new_layout);
1847 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1848 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1849 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N
1850 if (mddev->delta_disks == 0 &&
1851 mddev->reshape_backwards)
1852 sb->feature_map
1853 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N
1854 if (rdev->new_data_offset != rdev->data_offset) {
1855 sb->feature_map
1856 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1857 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1858 - rdev->data_offset));
1859 }
f6705578 1860 }
a654b9d8 1861
2699b672
N
1862 if (rdev->badblocks.count == 0)
1863 /* Nothing to do for bad blocks*/ ;
1864 else if (sb->bblog_offset == 0)
1865 /* Cannot record bad blocks on this device */
1866 md_error(mddev, rdev);
1867 else {
1868 struct badblocks *bb = &rdev->badblocks;
1869 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1870 u64 *p = bb->page;
1871 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1872 if (bb->changed) {
1873 unsigned seq;
1874
1875retry:
1876 seq = read_seqbegin(&bb->lock);
1877
1878 memset(bbp, 0xff, PAGE_SIZE);
1879
1880 for (i = 0 ; i < bb->count ; i++) {
1881 u64 internal_bb = *p++;
1882 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1883 | BB_LEN(internal_bb));
1884 *bbp++ = cpu_to_le64(store_bb);
1885 }
d0962936 1886 bb->changed = 0;
2699b672
N
1887 if (read_seqretry(&bb->lock, seq))
1888 goto retry;
1889
1890 bb->sector = (rdev->sb_start +
1891 (int)le32_to_cpu(sb->bblog_offset));
1892 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N
1893 }
1894 }
1895
1da177e4 1896 max_dev = 0;
dafb20fa 1897 rdev_for_each(rdev2, mddev)
1da177e4
LT
1898 if (rdev2->desc_nr+1 > max_dev)
1899 max_dev = rdev2->desc_nr+1;
a778b73f 1900
70471daf
N
1901 if (max_dev > le32_to_cpu(sb->max_dev)) {
1902 int bmask;
a778b73f 1903 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N
1904 rdev->sb_size = max_dev * 2 + 256;
1905 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1906 if (rdev->sb_size & bmask)
1907 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N
1908 } else
1909 max_dev = le32_to_cpu(sb->max_dev);
1910
1da177e4
LT
1911 for (i=0; i<max_dev;i++)
1912 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1913
dafb20fa 1914 rdev_for_each(rdev2, mddev) {
1da177e4 1915 i = rdev2->desc_nr;
b2d444d7 1916 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1917 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1918 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1919 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1920 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1921 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT
1922 else
1923 sb->dev_roles[i] = cpu_to_le16(0xffff);
1924 }
1925
1da177e4
LT
1926 sb->sb_csum = calc_sb_1_csum(sb);
1927}
1928
0cd17fec 1929static unsigned long long
3cb03002 1930super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW
1931{
1932 struct mdp_superblock_1 *sb;
15f4a5fd 1933 sector_t max_sectors;
58c0fed4 1934 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1935 return 0; /* component must fit device */
c6563a8c
N
1936 if (rdev->data_offset != rdev->new_data_offset)
1937 return 0; /* too confusing */
0f420358 1938 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1939 /* minor versions 1 and 2; superblock before data */
77304d2a 1940 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN
1941 max_sectors -= rdev->data_offset;
1942 if (!num_sectors || num_sectors > max_sectors)
1943 num_sectors = max_sectors;
c3d9714e 1944 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW
1945 /* minor version 0 with bitmap we can't move */
1946 return 0;
1947 } else {
1948 /* minor version 0; superblock after data */
0f420358 1949 sector_t sb_start;
77304d2a 1950 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1951 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1952 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN
1953 if (!num_sectors || num_sectors > max_sectors)
1954 num_sectors = max_sectors;
0f420358 1955 rdev->sb_start = sb_start;
0cd17fec 1956 }
65a06f06 1957 sb = page_address(rdev->sb_page);
15f4a5fd 1958 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1959 sb->super_offset = rdev->sb_start;
0cd17fec 1960 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1961 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW
1962 rdev->sb_page);
1963 md_super_wait(rdev->mddev);
c26a44ed 1964 return num_sectors;
c6563a8c
N
1965
1966}
1967
1968static int
1969super_1_allow_new_offset(struct md_rdev *rdev,
1970 unsigned long long new_offset)
1971{
1972 /* All necessary checks on new >= old have been done */
1973 struct bitmap *bitmap;
1974 if (new_offset >= rdev->data_offset)
1975 return 1;
1976
1977 /* with 1.0 metadata, there is no metadata to tread on
1978 * so we can always move back */
1979 if (rdev->mddev->minor_version == 0)
1980 return 1;
1981
1982 /* otherwise we must be sure not to step on
1983 * any metadata, so stay:
1984 * 36K beyond start of superblock
1985 * beyond end of badblocks
1986 * beyond write-intent bitmap
1987 */
1988 if (rdev->sb_start + (32+4)*2 > new_offset)
1989 return 0;
1990 bitmap = rdev->mddev->bitmap;
1991 if (bitmap && !rdev->mddev->bitmap_info.file &&
1992 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1993 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N
1994 return 0;
1995 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1996 return 0;
1997
1998 return 1;
0cd17fec 1999}
1da177e4 2000
75c96f85 2001static struct super_type super_types[] = {
1da177e4
LT
2002 [0] = {
2003 .name = "0.90.0",
2004 .owner = THIS_MODULE,
0cd17fec
CW
2005 .load_super = super_90_load,
2006 .validate_super = super_90_validate,
2007 .sync_super = super_90_sync,
2008 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 2009 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT
2010 },
2011 [1] = {
2012 .name = "md-1",
2013 .owner = THIS_MODULE,
0cd17fec
CW
2014 .load_super = super_1_load,
2015 .validate_super = super_1_validate,
2016 .sync_super = super_1_sync,
2017 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 2018 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT
2019 },
2020};
1da177e4 2021
fd01b88c 2022static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB
2023{
2024 if (mddev->sync_super) {
2025 mddev->sync_super(mddev, rdev);
2026 return;
2027 }
2028
2029 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2030
2031 super_types[mddev->major_version].sync_super(mddev, rdev);
2032}
2033
fd01b88c 2034static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 2035{
3cb03002 2036 struct md_rdev *rdev, *rdev2;
1da177e4 2037
4b80991c
N
2038 rcu_read_lock();
2039 rdev_for_each_rcu(rdev, mddev1)
2040 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 2041 if (rdev->bdev->bd_contains ==
4b80991c
N
2042 rdev2->bdev->bd_contains) {
2043 rcu_read_unlock();
7dd5e7c3 2044 return 1;
4b80991c
N
2045 }
2046 rcu_read_unlock();
1da177e4
LT
2047 return 0;
2048}
2049
2050static LIST_HEAD(pending_raid_disks);
2051
ac5e7113
AN
2052/*
2053 * Try to register data integrity profile for an mddev
2054 *
2055 * This is called when an array is started and after a disk has been kicked
2056 * from the array. It only succeeds if all working and active component devices
2057 * are integrity capable with matching profiles.
2058 */
fd01b88c 2059int md_integrity_register(struct mddev *mddev)
ac5e7113 2060{
3cb03002 2061 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN
2062
2063 if (list_empty(&mddev->disks))
2064 return 0; /* nothing to do */
629acb6a
JB
2065 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2066 return 0; /* shouldn't register, or already is */
dafb20fa 2067 rdev_for_each(rdev, mddev) {
ac5e7113
AN
2068 /* skip spares and non-functional disks */
2069 if (test_bit(Faulty, &rdev->flags))
2070 continue;
2071 if (rdev->raid_disk < 0)
2072 continue;
ac5e7113
AN
2073 if (!reference) {
2074 /* Use the first rdev as the reference */
2075 reference = rdev;
2076 continue;
2077 }
2078 /* does this rdev's profile match the reference profile? */
2079 if (blk_integrity_compare(reference->bdev->bd_disk,
2080 rdev->bdev->bd_disk) < 0)
2081 return -EINVAL;
2082 }
89078d57
MP
2083 if (!reference || !bdev_get_integrity(reference->bdev))
2084 return 0;
ac5e7113
AN
2085 /*
2086 * All component devices are integrity capable and have matching
2087 * profiles, register the common profile for the md device.
2088 */
2089 if (blk_integrity_register(mddev->gendisk,
2090 bdev_get_integrity(reference->bdev)) != 0) {
2091 printk(KERN_ERR "md: failed to register integrity for %s\n",
2092 mdname(mddev));
2093 return -EINVAL;
2094 }
a91a2785
MP
2095 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2096 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2097 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2098 mdname(mddev));
2099 return -EINVAL;
2100 }
ac5e7113
AN
2101 return 0;
2102}
2103EXPORT_SYMBOL(md_integrity_register);
2104
2105/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 2106void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 2107{
3f9d99c1 2108 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 2109 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2110
ac5e7113 2111 if (!bi_mddev) /* nothing to do */
3f9d99c1 2112 return;
ac5e7113 2113 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2114 return;
ac5e7113
AN
2115 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2116 rdev->bdev->bd_disk) >= 0)
2117 return;
2118 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2119 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2120}
ac5e7113 2121EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2122
fd01b88c 2123static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
1da177e4 2124{
7dd5e7c3 2125 char b[BDEVNAME_SIZE];
f637b9f9 2126 struct kobject *ko;
1edf80d3 2127 char *s;
5e55e2f5 2128 int err;
1da177e4
LT
2129
2130 if (rdev->mddev) {
2131 MD_BUG();
2132 return -EINVAL;
2133 }
11e2ede0
DW
2134
2135 /* prevent duplicates */
2136 if (find_rdev(mddev, rdev->bdev->bd_dev))
2137 return -EEXIST;
2138
dd8ac336
AN
2139 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2140 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2141 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N
2142 if (mddev->pers) {
2143 /* Cannot change size, so fail
2144 * If mddev->level <= 0, then we don't care
2145 * about aligning sizes (e.g. linear)
2146 */
2147 if (mddev->level > 0)
2148 return -ENOSPC;
2149 } else
dd8ac336 2150 mddev->dev_sectors = rdev->sectors;
2bf071bf 2151 }
1da177e4
LT
2152
2153 /* Verify rdev->desc_nr is unique.
2154 * If it is -1, assign a free number, else
2155 * check number is not in use
2156 */
2157 if (rdev->desc_nr < 0) {
2158 int choice = 0;
2159 if (mddev->pers) choice = mddev->raid_disks;
2160 while (find_rdev_nr(mddev, choice))
2161 choice++;
2162 rdev->desc_nr = choice;
2163 } else {
2164 if (find_rdev_nr(mddev, rdev->desc_nr))
2165 return -EBUSY;
2166 }
de01dfad
N
2167 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2168 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2169 mdname(mddev), mddev->max_disks);
2170 return -EBUSY;
2171 }
19133a42 2172 bdevname(rdev->bdev,b);
649316b2 2173 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2174 *s = '!';
649316b2 2175
1da177e4 2176 rdev->mddev = mddev;
19133a42 2177 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2178
b2d6db58 2179 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2180 goto fail;
86e6ffdd 2181
0762b8bd 2182 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N
2183 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2184 /* failure here is OK */;
2185 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2186
4b80991c 2187 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2188 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N
2189
2190 /* May as well allow recovery to be retried once */
5389042f 2191 mddev->recovery_disabled++;
3f9d99c1 2192
1da177e4 2193 return 0;
5e55e2f5
N
2194
2195 fail:
2196 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2197 b, mdname(mddev));
2198 return err;
1da177e4
LT
2199}
2200
177a99b2 2201static void md_delayed_delete(struct work_struct *ws)
5792a285 2202{
3cb03002 2203 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2204 kobject_del(&rdev->kobj);
177a99b2 2205 kobject_put(&rdev->kobj);
5792a285
N
2206}
2207
3cb03002 2208static void unbind_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT
2209{
2210 char b[BDEVNAME_SIZE];
2211 if (!rdev->mddev) {
2212 MD_BUG();
2213 return;
2214 }
49731baa 2215 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2216 list_del_rcu(&rdev->same_set);
1da177e4
LT
2217 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2218 rdev->mddev = NULL;
86e6ffdd 2219 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N
2220 sysfs_put(rdev->sysfs_state);
2221 rdev->sysfs_state = NULL;
2230dfe4 2222 rdev->badblocks.count = 0;
5792a285 2223 /* We need to delay this, otherwise we can deadlock when
4b80991c
N
2224 * writing to 'remove' to "dev/state". We also need
2225 * to delay it due to rcu usage.
5792a285 2226 */
4b80991c 2227 synchronize_rcu();
177a99b2
N
2228 INIT_WORK(&rdev->del_work, md_delayed_delete);
2229 kobject_get(&rdev->kobj);
e804ac78 2230 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT
2231}
2232
2233/*
2234 * prevent the device from being mounted, repartitioned or
2235 * otherwise reused by a RAID array (or any other kernel
2236 * subsystem), by bd_claiming the device.
2237 */
3cb03002 2238static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT
2239{
2240 int err = 0;
2241 struct block_device *bdev;
2242 char b[BDEVNAME_SIZE];
2243
d4d77629 2244 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2245 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT
2246 if (IS_ERR(bdev)) {
2247 printk(KERN_ERR "md: could not open %s.\n",
2248 __bdevname(dev, b));
2249 return PTR_ERR(bdev);
2250 }
1da177e4
LT
2251 rdev->bdev = bdev;
2252 return err;
2253}
2254
3cb03002 2255static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT
2256{
2257 struct block_device *bdev = rdev->bdev;
2258 rdev->bdev = NULL;
2259 if (!bdev)
2260 MD_BUG();
e525fd89 2261 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT
2262}
2263
2264void md_autodetect_dev(dev_t dev);
2265
3cb03002 2266static void export_rdev(struct md_rdev * rdev)
1da177e4
LT
2267{
2268 char b[BDEVNAME_SIZE];
2269 printk(KERN_INFO "md: export_rdev(%s)\n",
2270 bdevname(rdev->bdev,b));
2271 if (rdev->mddev)
2272 MD_BUG();
545c8795 2273 md_rdev_clear(rdev);
1da177e4 2274#ifndef MODULE
d0fae18f
N
2275 if (test_bit(AutoDetected, &rdev->flags))
2276 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT
2277#endif
2278 unlock_rdev(rdev);
86e6ffdd 2279 kobject_put(&rdev->kobj);
1da177e4
LT
2280}
2281
3cb03002 2282static void kick_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT
2283{
2284 unbind_rdev_from_array(rdev);
2285 export_rdev(rdev);
2286}
2287
fd01b88c 2288static void export_array(struct mddev *mddev)
1da177e4 2289{
3cb03002 2290 struct md_rdev *rdev, *tmp;
1da177e4 2291
dafb20fa 2292 rdev_for_each_safe(rdev, tmp, mddev) {
1da177e4
LT
2293 if (!rdev->mddev) {
2294 MD_BUG();
2295 continue;
2296 }
2297 kick_rdev_from_array(rdev);
2298 }
2299 if (!list_empty(&mddev->disks))
2300 MD_BUG();
2301 mddev->raid_disks = 0;
2302 mddev->major_version = 0;
2303}
2304
2305static void print_desc(mdp_disk_t *desc)
2306{
2307 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2308 desc->major,desc->minor,desc->raid_disk,desc->state);
2309}
2310
cd2ac932 2311static void print_sb_90(mdp_super_t *sb)
1da177e4
LT
2312{
2313 int i;
2314
2315 printk(KERN_INFO
2316 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2317 sb->major_version, sb->minor_version, sb->patch_version,
2318 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2319 sb->ctime);
2320 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2321 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2322 sb->md_minor, sb->layout, sb->chunk_size);
2323 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2324 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2325 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2326 sb->failed_disks, sb->spare_disks,
2327 sb->sb_csum, (unsigned long)sb->events_lo);
2328
2329 printk(KERN_INFO);
2330 for (i = 0; i < MD_SB_DISKS; i++) {
2331 mdp_disk_t *desc;
2332
2333 desc = sb->disks + i;
2334 if (desc->number || desc->major || desc->minor ||
2335 desc->raid_disk || (desc->state && (desc->state != 4))) {
2336 printk(" D %2d: ", i);
2337 print_desc(desc);
2338 }
2339 }
2340 printk(KERN_INFO "md: THIS: ");
2341 print_desc(&sb->this_disk);
cd2ac932 2342}
1da177e4 2343
cd2ac932
CR
2344static void print_sb_1(struct mdp_superblock_1 *sb)
2345{
2346 __u8 *uuid;
2347
2348 uuid = sb->set_uuid;
ad361c98 2349 printk(KERN_INFO
7b75c2f8 2350 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2351 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR
2352 le32_to_cpu(sb->major_version),
2353 le32_to_cpu(sb->feature_map),
7b75c2f8 2354 uuid,
cd2ac932
CR
2355 sb->set_name,
2356 (unsigned long long)le64_to_cpu(sb->ctime)
2357 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2358
2359 uuid = sb->device_uuid;
ad361c98
JP
2360 printk(KERN_INFO
2361 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2362 " RO:%llu\n"
7b75c2f8 2363 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP
2364 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2365 "md: (MaxDev:%u) \n",
cd2ac932
CR
2366 le32_to_cpu(sb->level),
2367 (unsigned long long)le64_to_cpu(sb->size),
2368 le32_to_cpu(sb->raid_disks),
2369 le32_to_cpu(sb->layout),
2370 le32_to_cpu(sb->chunksize),
2371 (unsigned long long)le64_to_cpu(sb->data_offset),
2372 (unsigned long long)le64_to_cpu(sb->data_size),
2373 (unsigned long long)le64_to_cpu(sb->super_offset),
2374 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2375 le32_to_cpu(sb->dev_number),
7b75c2f8 2376 uuid,
cd2ac932
CR
2377 sb->devflags,
2378 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2379 (unsigned long long)le64_to_cpu(sb->events),
2380 (unsigned long long)le64_to_cpu(sb->resync_offset),
2381 le32_to_cpu(sb->sb_csum),
2382 le32_to_cpu(sb->max_dev)
2383 );
1da177e4
LT
2384}
2385
3cb03002 2386static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT
2387{
2388 char b[BDEVNAME_SIZE];
dd8ac336
AN
2389 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2390 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N
2391 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2392 rdev->desc_nr);
1da177e4 2393 if (rdev->sb_loaded) {
cd2ac932
CR
2394 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2395 switch (major_version) {
2396 case 0:
65a06f06 2397 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR
2398 break;
2399 case 1:
65a06f06 2400 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR
2401 break;
2402 }
1da177e4
LT
2403 } else
2404 printk(KERN_INFO "md: no rdev superblock!\n");
2405}
2406
5e56341d 2407static void md_print_devices(void)
1da177e4 2408{
159ec1fc 2409 struct list_head *tmp;
3cb03002 2410 struct md_rdev *rdev;
fd01b88c 2411 struct mddev *mddev;
1da177e4
LT
2412 char b[BDEVNAME_SIZE];
2413
2414 printk("\n");
2415 printk("md: **********************************\n");
2416 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2417 printk("md: **********************************\n");
29ac4aa3 2418 for_each_mddev(mddev, tmp) {
1da177e4 2419
32a7627c
N
2420 if (mddev->bitmap)
2421 bitmap_print_sb(mddev->bitmap);
2422 else
2423 printk("%s: ", mdname(mddev));
dafb20fa 2424 rdev_for_each(rdev, mddev)
1da177e4
LT
2425 printk("<%s>", bdevname(rdev->bdev,b));
2426 printk("\n");
2427
dafb20fa 2428 rdev_for_each(rdev, mddev)
cd2ac932 2429 print_rdev(rdev, mddev->major_version);
1da177e4
LT
2430 }
2431 printk("md: **********************************\n");
2432 printk("\n");
2433}
2434
2435
fd01b88c 2436static void sync_sbs(struct mddev * mddev, int nospares)
1da177e4 2437{
42543769
N
2438 /* Update each superblock (in-memory image), but
2439 * if we are allowed to, skip spares which already
2440 * have the right event counter, or have one earlier
2441 * (which would mean they aren't being marked as dirty
2442 * with the rest of the array)
2443 */
3cb03002 2444 struct md_rdev *rdev;
dafb20fa 2445 rdev_for_each(rdev, mddev) {
42543769
N
2446 if (rdev->sb_events == mddev->events ||
2447 (nospares &&
2448 rdev->raid_disk < 0 &&
42543769
N
2449 rdev->sb_events+1 == mddev->events)) {
2450 /* Don't update this superblock */
2451 rdev->sb_loaded = 2;
2452 } else {
076f968b 2453 sync_super(mddev, rdev);
42543769
N
2454 rdev->sb_loaded = 1;
2455 }
1da177e4
LT
2456 }
2457}
2458
fd01b88c 2459static void md_update_sb(struct mddev * mddev, int force_change)
1da177e4 2460{
3cb03002 2461 struct md_rdev *rdev;
06d91a5f 2462 int sync_req;
42543769 2463 int nospares = 0;
2699b672 2464 int any_badblocks_changed = 0;
1da177e4 2465
1da177e4 2466repeat:
3a3a5ddb 2467 /* First make sure individual recovery_offsets are correct */
dafb20fa 2468 rdev_for_each(rdev, mddev) {
3a3a5ddb
N
2469 if (rdev->raid_disk >= 0 &&
2470 mddev->delta_disks >= 0 &&
2471 !test_bit(In_sync, &rdev->flags) &&
2472 mddev->curr_resync_completed > rdev->recovery_offset)
2473 rdev->recovery_offset = mddev->curr_resync_completed;
2474
2475 }
bd52b746 2476 if (!mddev->persistent) {
070dc6dd 2477 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2478 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2479 if (!mddev->external) {
d97a41dc 2480 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2481 rdev_for_each(rdev, mddev) {
de393cde 2482 if (rdev->badblocks.changed) {
d0962936 2483 rdev->badblocks.changed = 0;
de393cde
N
2484 md_ack_all_badblocks(&rdev->badblocks);
2485 md_error(mddev, rdev);
2486 }
2487 clear_bit(Blocked, &rdev->flags);
2488 clear_bit(BlockedBadBlocks, &rdev->flags);
2489 wake_up(&rdev->blocked_wait);
2490 }
2491 }
3a3a5ddb
N
2492 wake_up(&mddev->sb_wait);
2493 return;
2494 }
2495
a9701a30 2496 spin_lock_irq(&mddev->write_lock);
84692195 2497
3a3a5ddb
N
2498 mddev->utime = get_seconds();
2499
850b2b42
N
2500 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2501 force_change = 1;
2502 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2503 /* just a clean<-> dirty transition, possibly leave spares alone,
2504 * though if events isn't the right even/odd, we will have to do
2505 * spares after all
2506 */
2507 nospares = 1;
2508 if (force_change)
2509 nospares = 0;
2510 if (mddev->degraded)
84692195
N
2511 /* If the array is degraded, then skipping spares is both
2512 * dangerous and fairly pointless.
2513 * Dangerous because a device that was removed from the array
2514 * might have a event_count that still looks up-to-date,
2515 * so it can be re-added without a resync.
2516 * Pointless because if there are any spares to skip,
2517 * then a recovery will happen and soon that array won't
2518 * be degraded any more and the spare can go back to sleep then.
2519 */
850b2b42 2520 nospares = 0;
84692195 2521
06d91a5f 2522 sync_req = mddev->in_sync;
42543769
N
2523
2524 /* If this is just a dirty<->clean transition, and the array is clean
2525 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2526 if (nospares
42543769 2527 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N
2528 && mddev->can_decrease_events
2529 && mddev->events != 1) {
42543769 2530 mddev->events--;
a8707c08
N
2531 mddev->can_decrease_events = 0;
2532 } else {
42543769
N
2533 /* otherwise we have to go forward and ... */
2534 mddev->events ++;
a8707c08 2535 mddev->can_decrease_events = nospares;
42543769 2536 }
1da177e4
LT
2537
2538 if (!mddev->events) {
2539 /*
2540 * oops, this 64-bit counter should never wrap.
2541 * Either we are in around ~1 trillion A.C., assuming
2542 * 1 reboot per second, or we have a bug:
2543 */
2544 MD_BUG();
2545 mddev->events --;
2546 }
2699b672 2547
dafb20fa 2548 rdev_for_each(rdev, mddev) {
2699b672
N
2549 if (rdev->badblocks.changed)
2550 any_badblocks_changed++;
de393cde
N
2551 if (test_bit(Faulty, &rdev->flags))
2552 set_bit(FaultRecorded, &rdev->flags);
2553 }
2699b672 2554
e691063a 2555 sync_sbs(mddev, nospares);
a9701a30 2556 spin_unlock_irq(&mddev->write_lock);
1da177e4 2557
36a4e1fe
N
2558 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2559 mdname(mddev), mddev->in_sync);
1da177e4 2560
4ad13663 2561 bitmap_update_sb(mddev->bitmap);
dafb20fa 2562 rdev_for_each(rdev, mddev) {
1da177e4 2563 char b[BDEVNAME_SIZE];
36a4e1fe 2564
42543769
N
2565 if (rdev->sb_loaded != 1)
2566 continue; /* no noise on spare devices */
1da177e4 2567
d70ed2e4
AW
2568 if (!test_bit(Faulty, &rdev->flags) &&
2569 rdev->saved_raid_disk == -1) {
7bfa19f2 2570 md_super_write(mddev,rdev,
0f420358 2571 rdev->sb_start, rdev->sb_size,
7bfa19f2 2572 rdev->sb_page);
36a4e1fe
N
2573 pr_debug("md: (write) %s's sb offset: %llu\n",
2574 bdevname(rdev->bdev, b),
2575 (unsigned long long)rdev->sb_start);
42543769 2576 rdev->sb_events = mddev->events;
2699b672
N
2577 if (rdev->badblocks.size) {
2578 md_super_write(mddev, rdev,
2579 rdev->badblocks.sector,
2580 rdev->badblocks.size << 9,
2581 rdev->bb_page);
2582 rdev->badblocks.size = 0;
2583 }
7bfa19f2 2584
d70ed2e4 2585 } else if (test_bit(Faulty, &rdev->flags))
36a4e1fe
N
2586 pr_debug("md: %s (skipping faulty)\n",
2587 bdevname(rdev->bdev, b));
d70ed2e4
AW
2588 else
2589 pr_debug("(skipping incremental s/r ");
2590
7bfa19f2 2591 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT
2592 /* only need to write one superblock... */
2593 break;
2594 }
a9701a30 2595 md_super_wait(mddev);
850b2b42 2596 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2597
a9701a30 2598 spin_lock_irq(&mddev->write_lock);
850b2b42
N
2599 if (mddev->in_sync != sync_req ||
2600 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2601 /* have to write it out again */
a9701a30 2602 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N
2603 goto repeat;
2604 }
850b2b42 2605 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2606 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2607 wake_up(&mddev->sb_wait);
acb180b0
N
2608 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2609 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2610
dafb20fa 2611 rdev_for_each(rdev, mddev) {
de393cde
N
2612 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2613 clear_bit(Blocked, &rdev->flags);
2614
2615 if (any_badblocks_changed)
2699b672 2616 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N
2617 clear_bit(BlockedBadBlocks, &rdev->flags);
2618 wake_up(&rdev->blocked_wait);
2619 }
1da177e4
LT
2620}
2621
7f6ce769 2622/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N
2623 * We want to accept with case. For this we use cmd_match.
2624 */
2625static int cmd_match(const char *cmd, const char *str)
2626{
2627 /* See if cmd, written into a sysfs file, matches
2628 * str. They must either be the same, or cmd can
2629 * have a trailing newline
2630 */
2631 while (*cmd && *str && *cmd == *str) {
2632 cmd++;
2633 str++;
2634 }
2635 if (*cmd == '\n')
2636 cmd++;
2637 if (*str || *cmd)
2638 return 0;
2639 return 1;
2640}
2641
86e6ffdd
N
2642struct rdev_sysfs_entry {
2643 struct attribute attr;
3cb03002
N
2644 ssize_t (*show)(struct md_rdev *, char *);
2645 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N
2646};
2647
2648static ssize_t
3cb03002 2649state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N
2650{
2651 char *sep = "";
20a49ff6 2652 size_t len = 0;
86e6ffdd 2653
de393cde
N
2654 if (test_bit(Faulty, &rdev->flags) ||
2655 rdev->badblocks.unacked_exist) {
86e6ffdd
N
2656 len+= sprintf(page+len, "%sfaulty",sep);
2657 sep = ",";
2658 }
b2d444d7 2659 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N
2660 len += sprintf(page+len, "%sin_sync",sep);
2661 sep = ",";
2662 }
f655675b
N
2663 if (test_bit(WriteMostly, &rdev->flags)) {
2664 len += sprintf(page+len, "%swrite_mostly",sep);
2665 sep = ",";
2666 }
de393cde 2667 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N
2668 (rdev->badblocks.unacked_exist
2669 && !test_bit(Faulty, &rdev->flags))) {
6bfe0b49
DW
2670 len += sprintf(page+len, "%sblocked", sep);
2671 sep = ",";
2672 }
b2d444d7
N
2673 if (!test_bit(Faulty, &rdev->flags) &&
2674 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N
2675 len += sprintf(page+len, "%sspare", sep);
2676 sep = ",";
2677 }
d7a9d443
N
2678 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2679 len += sprintf(page+len, "%swrite_error", sep);
2680 sep = ",";
2681 }
2d78f8c4
N
2682 if (test_bit(WantReplacement, &rdev->flags)) {
2683 len += sprintf(page+len, "%swant_replacement", sep);
2684 sep = ",";
2685 }
2686 if (test_bit(Replacement, &rdev->flags)) {
2687 len += sprintf(page+len, "%sreplacement", sep);
2688 sep = ",";
2689 }
2690
86e6ffdd
N
2691 return len+sprintf(page+len, "\n");
2692}
2693
45dc2de1 2694static ssize_t
3cb03002 2695state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N
2696{
2697 /* can write
de393cde 2698 * faulty - simulates an error
45dc2de1 2699 * remove - disconnects the device
f655675b
N
2700 * writemostly - sets write_mostly
2701 * -writemostly - clears write_mostly
de393cde
N
2702 * blocked - sets the Blocked flags
2703 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2704 * insync - sets Insync providing device isn't active
d7a9d443
N
2705 * write_error - sets WriteErrorSeen
2706 * -write_error - clears WriteErrorSeen
45dc2de1
N
2707 */
2708 int err = -EINVAL;
2709 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2710 md_error(rdev->mddev, rdev);
5ef56c8f
N
2711 if (test_bit(Faulty, &rdev->flags))
2712 err = 0;
2713 else
2714 err = -EBUSY;
45dc2de1
N
2715 } else if (cmd_match(buf, "remove")) {
2716 if (rdev->raid_disk >= 0)
2717 err = -EBUSY;
2718 else {
fd01b88c 2719 struct mddev *mddev = rdev->mddev;
45dc2de1 2720 kick_rdev_from_array(rdev);
3f9d7b0d
N
2721 if (mddev->pers)
2722 md_update_sb(mddev, 1);
45dc2de1
N
2723 md_new_event(mddev);
2724 err = 0;
2725 }
f655675b
N
2726 } else if (cmd_match(buf, "writemostly")) {
2727 set_bit(WriteMostly, &rdev->flags);
2728 err = 0;
2729 } else if (cmd_match(buf, "-writemostly")) {
2730 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW
2731 err = 0;
2732 } else if (cmd_match(buf, "blocked")) {
2733 set_bit(Blocked, &rdev->flags);
2734 err = 0;
2735 } else if (cmd_match(buf, "-blocked")) {
de393cde 2736 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2737 rdev->badblocks.unacked_exist) {
de393cde
N
2738 /* metadata handler doesn't understand badblocks,
2739 * so we need to fail the device
2740 */
2741 md_error(rdev->mddev, rdev);
2742 }
6bfe0b49 2743 clear_bit(Blocked, &rdev->flags);
de393cde 2744 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW
2745 wake_up(&rdev->blocked_wait);
2746 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2747 md_wakeup_thread(rdev->mddev->thread);
2748
6d56e278
N
2749 err = 0;
2750 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2751 set_bit(In_sync, &rdev->flags);
f655675b 2752 err = 0;
d7a9d443
N
2753 } else if (cmd_match(buf, "write_error")) {
2754 set_bit(WriteErrorSeen, &rdev->flags);
2755 err = 0;
2756 } else if (cmd_match(buf, "-write_error")) {
2757 clear_bit(WriteErrorSeen, &rdev->flags);
2758 err = 0;
2d78f8c4
N
2759 } else if (cmd_match(buf, "want_replacement")) {
2760 /* Any non-spare device that is not a replacement can
2761 * become want_replacement at any time, but we then need to
2762 * check if recovery is needed.
2763 */
2764 if (rdev->raid_disk >= 0 &&
2765 !test_bit(Replacement, &rdev->flags))
2766 set_bit(WantReplacement, &rdev->flags);
2767 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2768 md_wakeup_thread(rdev->mddev->thread);
2769 err = 0;
2770 } else if (cmd_match(buf, "-want_replacement")) {
2771 /* Clearing 'want_replacement' is always allowed.
2772 * Once replacements starts it is too late though.
2773 */
2774 err = 0;
2775 clear_bit(WantReplacement, &rdev->flags);
2776 } else if (cmd_match(buf, "replacement")) {
2777 /* Can only set a device as a replacement when array has not
2778 * yet been started. Once running, replacement is automatic
2779 * from spares, or by assigning 'slot'.
2780 */
2781 if (rdev->mddev->pers)
2782 err = -EBUSY;
2783 else {
2784 set_bit(Replacement, &rdev->flags);
2785 err = 0;
2786 }
2787 } else if (cmd_match(buf, "-replacement")) {
2788 /* Similarly, can only clear Replacement before start */
2789 if (rdev->mddev->pers)
2790 err = -EBUSY;
2791 else {
2792 clear_bit(Replacement, &rdev->flags);
2793 err = 0;
2794 }
45dc2de1 2795 }
00bcb4ac
N
2796 if (!err)
2797 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N
2798 return err ? err : len;
2799}
80ca3a44
N
2800static struct rdev_sysfs_entry rdev_state =
2801__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2802
4dbcdc75 2803static ssize_t
3cb03002 2804errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N
2805{
2806 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2807}
2808
2809static ssize_t
3cb03002 2810errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N
2811{
2812 char *e;
2813 unsigned long n = simple_strtoul(buf, &e, 10);
2814 if (*buf && (*e == 0 || *e == '\n')) {
2815 atomic_set(&rdev->corrected_errors, n);
2816 return len;
2817 }
2818 return -EINVAL;
2819}
2820static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2821__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2822
014236d2 2823static ssize_t
3cb03002 2824slot_show(struct md_rdev *rdev, char *page)
014236d2
N
2825{
2826 if (rdev->raid_disk < 0)
2827 return sprintf(page, "none\n");
2828 else
2829 return sprintf(page, "%d\n", rdev->raid_disk);
2830}
2831
2832static ssize_t
3cb03002 2833slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N
2834{
2835 char *e;
c303da6d 2836 int err;
014236d2
N
2837 int slot = simple_strtoul(buf, &e, 10);
2838 if (strncmp(buf, "none", 4)==0)
2839 slot = -1;
2840 else if (e==buf || (*e && *e!= '\n'))
2841 return -EINVAL;
6c2fce2e 2842 if (rdev->mddev->pers && slot == -1) {
c303da6d
N
2843 /* Setting 'slot' on an active array requires also
2844 * updating the 'rd%d' link, and communicating
2845 * with the personality with ->hot_*_disk.
2846 * For now we only support removing
2847 * failed/spare devices. This normally happens automatically,
2848 * but not when the metadata is externally managed.
2849 */
c303da6d
N
2850 if (rdev->raid_disk == -1)
2851 return -EEXIST;
2852 /* personality does all needed checks */
01393f3d 2853 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d
N
2854 return -EINVAL;
2855 err = rdev->mddev->pers->
b8321b68 2856 hot_remove_disk(rdev->mddev, rdev);
c303da6d
N
2857 if (err)
2858 return err;
36fad858 2859 sysfs_unlink_rdev(rdev->mddev, rdev);
b7103107 2860 rdev->raid_disk = -1;
c303da6d
N
2861 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2862 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2863 } else if (rdev->mddev->pers) {
6c2fce2e 2864 /* Activating a spare .. or possibly reactivating
6d56e278 2865 * if we ever get bitmaps working here.
6c2fce2e
NB
2866 */
2867
2868 if (rdev->raid_disk != -1)
2869 return -EBUSY;
2870
c6751b2b
N
2871 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2872 return -EBUSY;
2873
6c2fce2e
NB
2874 if (rdev->mddev->pers->hot_add_disk == NULL)
2875 return -EINVAL;
2876
ba1b41b6
N
2877 if (slot >= rdev->mddev->raid_disks &&
2878 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2879 return -ENOSPC;
2880
6c2fce2e
NB
2881 rdev->raid_disk = slot;
2882 if (test_bit(In_sync, &rdev->flags))
2883 rdev->saved_raid_disk = slot;
2884 else
2885 rdev->saved_raid_disk = -1;
d30519fc 2886 clear_bit(In_sync, &rdev->flags);
6c2fce2e
NB
2887 err = rdev->mddev->pers->
2888 hot_add_disk(rdev->mddev, rdev);
199050ea 2889 if (err) {
6c2fce2e 2890 rdev->raid_disk = -1;
6c2fce2e 2891 return err;
52664732 2892 } else
00bcb4ac 2893 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2894 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2895 /* failure here is OK */;
6c2fce2e 2896 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2897 } else {
ba1b41b6
N
2898 if (slot >= rdev->mddev->raid_disks &&
2899 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N
2900 return -ENOSPC;
2901 rdev->raid_disk = slot;
2902 /* assume it is working */
c5d79adb
N
2903 clear_bit(Faulty, &rdev->flags);
2904 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2905 set_bit(In_sync, &rdev->flags);
00bcb4ac 2906 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2907 }
014236d2
N
2908 return len;
2909}
2910
2911
2912static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2913__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2914
93c8cad0 2915static ssize_t
3cb03002 2916offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2917{
6961ece4 2918 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N
2919}
2920
2921static ssize_t
3cb03002 2922offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2923{
c6563a8c
N
2924 unsigned long long offset;
2925 if (strict_strtoull(buf, 10, &offset) < 0)
93c8cad0 2926 return -EINVAL;
8ed0a521 2927 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2928 return -EBUSY;
dd8ac336 2929 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N
2930 /* Must set offset before size, so overlap checks
2931 * can be sane */
2932 return -EBUSY;
93c8cad0
N
2933 rdev->data_offset = offset;
2934 return len;
2935}
2936
2937static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2938__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2939
c6563a8c
N
2940static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2941{
2942 return sprintf(page, "%llu\n",
2943 (unsigned long long)rdev->new_data_offset);
2944}
2945
2946static ssize_t new_offset_store(struct md_rdev *rdev,
2947 const char *buf, size_t len)
2948{
2949 unsigned long long new_offset;
2950 struct mddev *mddev = rdev->mddev;
2951
2952 if (strict_strtoull(buf, 10, &new_offset) < 0)
2953 return -EINVAL;
2954
2955 if (mddev->sync_thread)
2956 return -EBUSY;
2957 if (new_offset == rdev->data_offset)
2958 /* reset is always permitted */
2959 ;
2960 else if (new_offset > rdev->data_offset) {
2961 /* must not push array size beyond rdev_sectors */
2962 if (new_offset - rdev->data_offset
2963 + mddev->dev_sectors > rdev->sectors)
2964 return -E2BIG;
2965 }
2966 /* Metadata worries about other space details. */
2967
2968 /* decreasing the offset is inconsistent with a backwards
2969 * reshape.
2970 */
2971 if (new_offset < rdev->data_offset &&
2972 mddev->reshape_backwards)
2973 return -EINVAL;
2974 /* Increasing offset is inconsistent with forwards
2975 * reshape. reshape_direction should be set to
2976 * 'backwards' first.
2977 */
2978 if (new_offset > rdev->data_offset &&
2979 !mddev->reshape_backwards)
2980 return -EINVAL;
2981
2982 if (mddev->pers && mddev->persistent &&
2983 !super_types[mddev->major_version]
2984 .allow_new_offset(rdev, new_offset))
2985 return -E2BIG;
2986 rdev->new_data_offset = new_offset;
2987 if (new_offset > rdev->data_offset)
2988 mddev->reshape_backwards = 1;
2989 else if (new_offset < rdev->data_offset)
2990 mddev->reshape_backwards = 0;
2991
2992 return len;
2993}
2994static struct rdev_sysfs_entry rdev_new_offset =
2995__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2996
83303b61 2997static ssize_t
3cb03002 2998rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2999{
dd8ac336 3000 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N
3001}
3002
c5d79adb
N
3003static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3004{
3005 /* check if two start/length pairs overlap */
3006 if (s1+l1 <= s2)
3007 return 0;
3008 if (s2+l2 <= s1)
3009 return 0;
3010 return 1;
3011}
3012
b522adcd
DW
3013static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3014{
3015 unsigned long long blocks;
3016 sector_t new;
3017
3018 if (strict_strtoull(buf, 10, &blocks) < 0)
3019 return -EINVAL;
3020
3021 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3022 return -EINVAL; /* sector conversion overflow */
3023
3024 new = blocks * 2;
3025 if (new != blocks * 2)
3026 return -EINVAL; /* unsigned long long to sector_t overflow */
3027
3028 *sectors = new;
3029 return 0;
3030}
3031
83303b61 3032static ssize_t
3cb03002 3033rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 3034{
fd01b88c 3035 struct mddev *my_mddev = rdev->mddev;
dd8ac336 3036 sector_t oldsectors = rdev->sectors;
b522adcd 3037 sector_t sectors;
27c529bb 3038
b522adcd 3039 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 3040 return -EINVAL;
c6563a8c
N
3041 if (rdev->data_offset != rdev->new_data_offset)
3042 return -EINVAL; /* too confusing */
0cd17fec 3043 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 3044 if (my_mddev->persistent) {
dd8ac336
AN
3045 sectors = super_types[my_mddev->major_version].
3046 rdev_size_change(rdev, sectors);
3047 if (!sectors)
0cd17fec 3048 return -EBUSY;
dd8ac336 3049 } else if (!sectors)
77304d2a 3050 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 3051 rdev->data_offset;
0cd17fec 3052 }
dd8ac336 3053 if (sectors < my_mddev->dev_sectors)
7d3c6f87 3054 return -EINVAL; /* component must fit device */
0cd17fec 3055
dd8ac336
AN
3056 rdev->sectors = sectors;
3057 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N
3058 /* need to check that all other rdevs with the same ->bdev
3059 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 3060 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N
3061 * we have to change it back, we will have the lock again.
3062 */
fd01b88c 3063 struct mddev *mddev;
c5d79adb 3064 int overlap = 0;
159ec1fc 3065 struct list_head *tmp;
c5d79adb 3066
27c529bb 3067 mddev_unlock(my_mddev);
29ac4aa3 3068 for_each_mddev(mddev, tmp) {
3cb03002 3069 struct md_rdev *rdev2;
c5d79adb
N
3070
3071 mddev_lock(mddev);
dafb20fa 3072 rdev_for_each(rdev2, mddev)
f21e9ff7
N
3073 if (rdev->bdev == rdev2->bdev &&
3074 rdev != rdev2 &&
3075 overlaps(rdev->data_offset, rdev->sectors,
3076 rdev2->data_offset,
3077 rdev2->sectors)) {
c5d79adb
N
3078 overlap = 1;
3079 break;
3080 }
3081 mddev_unlock(mddev);
3082 if (overlap) {
3083 mddev_put(mddev);
3084 break;
3085 }
3086 }
27c529bb 3087 mddev_lock(my_mddev);
c5d79adb
N
3088 if (overlap) {
3089 /* Someone else could have slipped in a size
3090 * change here, but doing so is just silly.
dd8ac336 3091 * We put oldsectors back because we *know* it is
c5d79adb
N
3092 * safe, and trust userspace not to race with
3093 * itself
3094 */
dd8ac336 3095 rdev->sectors = oldsectors;
c5d79adb
N
3096 return -EBUSY;
3097 }
3098 }
83303b61
N
3099 return len;
3100}
3101
3102static struct rdev_sysfs_entry rdev_size =
80ca3a44 3103__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 3104
06e3c817 3105
3cb03002 3106static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW
3107{
3108 unsigned long long recovery_start = rdev->recovery_offset;
3109
3110 if (test_bit(In_sync, &rdev->flags) ||
3111 recovery_start == MaxSector)
3112 return sprintf(page, "none\n");
3113
3114 return sprintf(page, "%llu\n", recovery_start);
3115}
3116
3cb03002 3117static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW
3118{
3119 unsigned long long recovery_start;
3120
3121 if (cmd_match(buf, "none"))
3122 recovery_start = MaxSector;
3123 else if (strict_strtoull(buf, 10, &recovery_start))
3124 return -EINVAL;
3125
3126 if (rdev->mddev->pers &&
3127 rdev->raid_disk >= 0)
3128 return -EBUSY;
3129
3130 rdev->recovery_offset = recovery_start;
3131 if (recovery_start == MaxSector)
3132 set_bit(In_sync, &rdev->flags);
3133 else
3134 clear_bit(In_sync, &rdev->flags);
3135 return len;
3136}
3137
3138static struct rdev_sysfs_entry rdev_recovery_start =
3139__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3140
16c791a5
N
3141
3142static ssize_t
3143badblocks_show(struct badblocks *bb, char *page, int unack);
3144static ssize_t
3145badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3146
3cb03002 3147static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N
3148{
3149 return badblocks_show(&rdev->badblocks, page, 0);
3150}
3cb03002 3151static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 3152{
de393cde
N
3153 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3154 /* Maybe that ack was all we needed */
3155 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3156 wake_up(&rdev->blocked_wait);
3157 return rv;
16c791a5
N
3158}
3159static struct rdev_sysfs_entry rdev_bad_blocks =
3160__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3161
3162
3cb03002 3163static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N
3164{
3165 return badblocks_show(&rdev->badblocks, page, 1);
3166}
3cb03002 3167static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N
3168{
3169 return badblocks_store(&rdev->badblocks, page, len, 1);
3170}
3171static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3172__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3173
86e6ffdd
N
3174static struct attribute *rdev_default_attrs[] = {
3175 &rdev_state.attr,
4dbcdc75 3176 &rdev_errors.attr,
014236d2 3177 &rdev_slot.attr,
93c8cad0 3178 &rdev_offset.attr,
c6563a8c 3179 &rdev_new_offset.attr,
83303b61 3180 &rdev_size.attr,
06e3c817 3181 &rdev_recovery_start.attr,
16c791a5
N
3182 &rdev_bad_blocks.attr,
3183 &rdev_unack_bad_blocks.attr,
86e6ffdd
N
3184 NULL,
3185};
3186static ssize_t
3187rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3188{
3189 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3190 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 3191 struct mddev *mddev = rdev->mddev;
27c529bb 3192 ssize_t rv;
86e6ffdd
N
3193
3194 if (!entry->show)
3195 return -EIO;
27c529bb
N
3196
3197 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3198 if (!rv) {
3199 if (rdev->mddev == NULL)
3200 rv = -EBUSY;
3201 else
3202 rv = entry->show(rdev, page);
3203 mddev_unlock(mddev);
3204 }
3205 return rv;
86e6ffdd
N
3206}
3207
3208static ssize_t
3209rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3210 const char *page, size_t length)
3211{
3212 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3213 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3214 ssize_t rv;
fd01b88c 3215 struct mddev *mddev = rdev->mddev;
86e6ffdd
N
3216
3217 if (!entry->store)
3218 return -EIO;
67463acb
N
3219 if (!capable(CAP_SYS_ADMIN))
3220 return -EACCES;
27c529bb 3221 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3222 if (!rv) {
27c529bb
N
3223 if (rdev->mddev == NULL)
3224 rv = -EBUSY;
3225 else
3226 rv = entry->store(rdev, page, length);
6a51830e 3227 mddev_unlock(mddev);
ca388059
N
3228 }
3229 return rv;
86e6ffdd
N
3230}
3231
3232static void rdev_free(struct kobject *ko)
3233{
3cb03002 3234 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N
3235 kfree(rdev);
3236}
52cf25d0 3237static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N
3238 .show = rdev_attr_show,
3239 .store = rdev_attr_store,
3240};
3241static struct kobj_type rdev_ktype = {
3242 .release = rdev_free,
3243 .sysfs_ops = &rdev_sysfs_ops,
3244 .default_attrs = rdev_default_attrs,
3245};
3246
3cb03002 3247int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N
3248{
3249 rdev->desc_nr = -1;
3250 rdev->saved_raid_disk = -1;
3251 rdev->raid_disk = -1;
3252 rdev->flags = 0;
3253 rdev->data_offset = 0;
c6563a8c 3254 rdev->new_data_offset = 0;
e8bb9a83
N
3255 rdev->sb_events = 0;
3256 rdev->last_read_error.tv_sec = 0;
3257 rdev->last_read_error.tv_nsec = 0;
2699b672
N
3258 rdev->sb_loaded = 0;
3259 rdev->bb_page = NULL;
e8bb9a83
N
3260 atomic_set(&rdev->nr_pending, 0);
3261 atomic_set(&rdev->read_errors, 0);
3262 atomic_set(&rdev->corrected_errors, 0);
3263
3264 INIT_LIST_HEAD(&rdev->same_set);
3265 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N
3266
3267 /* Add space to store bad block list.
3268 * This reserves the space even on arrays where it cannot
3269 * be used - I wonder if that matters
3270 */
3271 rdev->badblocks.count = 0;
3272 rdev->badblocks.shift = 0;
3273 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3274 seqlock_init(&rdev->badblocks.lock);
3275 if (rdev->badblocks.page == NULL)
3276 return -ENOMEM;
3277
3278 return 0;
e8bb9a83
N
3279}
3280EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT
3281/*
3282 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3283 *
3284 * mark the device faulty if:
3285 *
3286 * - the device is nonexistent (zero size)
3287 * - the device has no valid superblock
3288 *
3289 * a faulty rdev _never_ has rdev->sb set.
3290 */
3cb03002 3291static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT
3292{
3293 char b[BDEVNAME_SIZE];
3294 int err;
3cb03002 3295 struct md_rdev *rdev;
1da177e4
LT
3296 sector_t size;
3297
9ffae0cf 3298 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT
3299 if (!rdev) {
3300 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3301 return ERR_PTR(-ENOMEM);
3302 }
1da177e4 3303
2230dfe4
N
3304 err = md_rdev_init(rdev);
3305 if (err)
3306 goto abort_free;
3307 err = alloc_disk_sb(rdev);
3308 if (err)
1da177e4
LT
3309 goto abort_free;
3310
c5d79adb 3311 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT
3312 if (err)
3313 goto abort_free;
3314
f9cb074b 3315 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3316
77304d2a 3317 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT
3318 if (!size) {
3319 printk(KERN_WARNING
3320 "md: %s has zero or unknown size, marking faulty!\n",
3321 bdevname(rdev->bdev,b));
3322 err = -EINVAL;
3323 goto abort_free;
3324 }
3325
3326 if (super_format >= 0) {
3327 err = super_types[super_format].
3328 load_super(rdev, NULL, super_minor);
3329 if (err == -EINVAL) {
df968c4e
N
3330 printk(KERN_WARNING
3331 "md: %s does not have a valid v%d.%d "
3332 "superblock, not importing!\n",
3333 bdevname(rdev->bdev,b),
3334 super_format, super_minor);
1da177e4
LT
3335 goto abort_free;
3336 }
3337 if (err < 0) {
3338 printk(KERN_WARNING
3339 "md: could not read %s's sb, not importing!\n",
3340 bdevname(rdev->bdev,b));
3341 goto abort_free;
3342 }
3343 }
9f2f3830
N
3344 if (super_format == -1)
3345 /* hot-add for 0.90, or non-persistent: so no badblocks */
3346 rdev->badblocks.shift = -1;
6bfe0b49 3347
1da177e4
LT
3348 return rdev;
3349
3350abort_free:
2699b672
N
3351 if (rdev->bdev)
3352 unlock_rdev(rdev);
545c8795 3353 md_rdev_clear(rdev);
1da177e4
LT
3354 kfree(rdev);
3355 return ERR_PTR(err);
3356}
3357
3358/*
3359 * Check a full RAID array for plausibility
3360 */
3361
3362
fd01b88c 3363static void analyze_sbs(struct mddev * mddev)
1da177e4
LT
3364{
3365 int i;
3cb03002 3366 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT
3367 char b[BDEVNAME_SIZE];
3368
3369 freshest = NULL;
dafb20fa 3370 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT
3371 switch (super_types[mddev->major_version].
3372 load_super(rdev, freshest, mddev->minor_version)) {
3373 case 1:
3374 freshest = rdev;
3375 break;
3376 case 0:
3377 break;
3378 default:
3379 printk( KERN_ERR \
3380 "md: fatal superblock inconsistency in %s"
3381 " -- removing from array\n",
3382 bdevname(rdev->bdev,b));
3383 kick_rdev_from_array(rdev);
3384 }
3385
3386
3387 super_types[mddev->major_version].
3388 validate_super(mddev, freshest);
3389
3390 i = 0;
dafb20fa 3391 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N
3392 if (mddev->max_disks &&
3393 (rdev->desc_nr >= mddev->max_disks ||
3394 i > mddev->max_disks)) {
de01dfad
N
3395 printk(KERN_WARNING
3396 "md: %s: %s: only %d devices permitted\n",
3397 mdname(mddev), bdevname(rdev->bdev, b),
3398 mddev->max_disks);
3399 kick_rdev_from_array(rdev);
3400 continue;
3401 }
1da177e4
LT
3402 if (rdev != freshest)
3403 if (super_types[mddev->major_version].
3404 validate_super(mddev, rdev)) {
3405 printk(KERN_WARNING "md: kicking non-fresh %s"
3406 " from array!\n",
3407 bdevname(rdev->bdev,b));
3408 kick_rdev_from_array(rdev);
3409 continue;
3410 }
3411 if (mddev->level == LEVEL_MULTIPATH) {
3412 rdev->desc_nr = i++;
3413 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3414 set_bit(In_sync, &rdev->flags);
5e5e3e78 3415 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N
3416 rdev->raid_disk = -1;
3417 clear_bit(In_sync, &rdev->flags);
1da177e4
LT
3418 }
3419 }
1da177e4
LT
3420}
3421
72e02075
N
3422/* Read a fixed-point number.
3423 * Numbers in sysfs attributes should be in "standard" units where
3424 * possible, so time should be in seconds.
3425 * However we internally use a a much smaller unit such as
3426 * milliseconds or jiffies.
3427 * This function takes a decimal number with a possible fractional
3428 * component, and produces an integer which is the result of
3429 * multiplying that number by 10^'scale'.
3430 * all without any floating-point arithmetic.
3431 */
3432int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3433{
3434 unsigned long result = 0;
3435 long decimals = -1;
3436 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3437 if (*cp == '.')
3438 decimals = 0;
3439 else if (decimals < scale) {
3440 unsigned int value;
3441 value = *cp - '0';
3442 result = result * 10 + value;
3443 if (decimals >= 0)
3444 decimals++;
3445 }
3446 cp++;
3447 }
3448 if (*cp == '\n')
3449 cp++;
3450 if (*cp)
3451 return -EINVAL;
3452 if (decimals < 0)
3453 decimals = 0;
3454 while (decimals < scale) {
3455 result *= 10;
3456 decimals ++;
3457 }
3458 *res = result;
3459 return 0;
3460}
3461
3462
19052c0e
N
3463static void md_safemode_timeout(unsigned long data);
3464
16f17b39 3465static ssize_t
fd01b88c 3466safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N
3467{
3468 int msec = (mddev->safemode_delay*1000)/HZ;
3469 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3470}
3471static ssize_t
fd01b88c 3472safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3473{
16f17b39 3474 unsigned long msec;
97ce0a7f 3475
72e02075 3476 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3477 return -EINVAL;
16f17b39
N
3478 if (msec == 0)
3479 mddev->safemode_delay = 0;
3480 else {
19052c0e 3481 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N
3482 mddev->safemode_delay = (msec*HZ)/1000;
3483 if (mddev->safemode_delay == 0)
3484 mddev->safemode_delay = 1;
19052c0e
N
3485 if (mddev->safemode_delay < old_delay)
3486 md_safemode_timeout((unsigned long)mddev);
16f17b39
N
3487 }
3488 return len;
3489}
3490static struct md_sysfs_entry md_safe_delay =
80ca3a44 3491__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3492
eae1701f 3493static ssize_t
fd01b88c 3494level_show(struct mddev *mddev, char *page)
eae1701f 3495{
84fc4b56 3496 struct md_personality *p = mddev->pers;
d9d166c2 3497 if (p)
eae1701f 3498 return sprintf(page, "%s\n", p->name);
d9d166c2
N
3499 else if (mddev->clevel[0])
3500 return sprintf(page, "%s\n", mddev->clevel);
3501 else if (mddev->level != LEVEL_NONE)
3502 return sprintf(page, "%d\n", mddev->level);
3503 else
3504 return 0;
eae1701f
N
3505}
3506
d9d166c2 3507static ssize_t
fd01b88c 3508level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3509{
f2859af6 3510 char clevel[16];
20a49ff6 3511 ssize_t rv = len;
84fc4b56 3512 struct md_personality *pers;
f2859af6 3513 long level;
245f46c2 3514 void *priv;
3cb03002 3515 struct md_rdev *rdev;
245f46c2
N
3516
3517 if (mddev->pers == NULL) {
3518 if (len == 0)
3519 return 0;
3520 if (len >= sizeof(mddev->clevel))
3521 return -ENOSPC;
3522 strncpy(mddev->clevel, buf, len);
3523 if (mddev->clevel[len-1] == '\n')
3524 len--;
3525 mddev->clevel[len] = 0;
3526 mddev->level = LEVEL_NONE;
3527 return rv;
3528 }
3529
3530 /* request to change the personality. Need to ensure:
3531 * - array is not engaged in resync/recovery/reshape
3532 * - old personality can be suspended
3533 * - new personality will access other array.
3534 */
3535
bb4f1e9d
N
3536 if (mddev->sync_thread ||
3537 mddev->reshape_position != MaxSector ||
3538 mddev->sysfs_active)
d9d166c2 3539 return -EBUSY;
245f46c2
N
3540
3541 if (!mddev->pers->quiesce) {
3542 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3543 mdname(mddev), mddev->pers->name);
3544 return -EINVAL;
3545 }
3546
3547 /* Now find the new personality */
f2859af6 3548 if (len == 0 || len >= sizeof(clevel))
245f46c2 3549 return -EINVAL;
f2859af6
DW
3550 strncpy(clevel, buf, len);
3551 if (clevel[len-1] == '\n')
d9d166c2 3552 len--;
f2859af6
DW
3553 clevel[len] = 0;
3554 if (strict_strtol(clevel, 10, &level))
3555 level = LEVEL_NONE;
245f46c2 3556
f2859af6
DW
3557 if (request_module("md-%s", clevel) != 0)
3558 request_module("md-level-%s", clevel);
245f46c2 3559 spin_lock(&pers_lock);
f2859af6 3560 pers = find_pers(level, clevel);
245f46c2
N
3561 if (!pers || !try_module_get(pers->owner)) {
3562 spin_unlock(&pers_lock);
f2859af6 3563 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N
3564 return -EINVAL;
3565 }
3566 spin_unlock(&pers_lock);
3567
3568 if (pers == mddev->pers) {
3569 /* Nothing to do! */
3570 module_put(pers->owner);
3571 return rv;
3572 }
3573 if (!pers->takeover) {
3574 module_put(pers->owner);
3575 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3576 mdname(mddev), clevel);
245f46c2
N
3577 return -EINVAL;
3578 }
3579
dafb20fa 3580 rdev_for_each(rdev, mddev)
e93f68a1
N
3581 rdev->new_raid_disk = rdev->raid_disk;
3582
245f46c2
N
3583 /* ->takeover must set new_* and/or delta_disks
3584 * if it succeeds, and may set them when it fails.
3585 */
3586 priv = pers->takeover(mddev);
3587 if (IS_ERR(priv)) {
3588 mddev->new_level = mddev->level;
3589 mddev->new_layout = mddev->layout;
664e7c41 3590 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N
3591 mddev->raid_disks -= mddev->delta_disks;
3592 mddev->delta_disks = 0;
2c810cdd 3593 mddev->reshape_backwards = 0;
245f46c2
N
3594 module_put(pers->owner);
3595 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3596 mdname(mddev), clevel);
245f46c2
N
3597 return PTR_ERR(priv);
3598 }
3599
3600 /* Looks like we have a winner */
3601 mddev_suspend(mddev);
3602 mddev->pers->stop(mddev);
a64c876f
N
3603
3604 if (mddev->pers->sync_request == NULL &&
3605 pers->sync_request != NULL) {
3606 /* need to add the md_redundancy_group */
3607 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3608 printk(KERN_WARNING
3609 "md: cannot register extra attributes for %s\n",
3610 mdname(mddev));
19fdb9ee 3611 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N
3612 }
3613 if (mddev->pers->sync_request != NULL &&
3614 pers->sync_request == NULL) {
3615 /* need to remove the md_redundancy_group */
3616 if (mddev->to_remove == NULL)
3617 mddev->to_remove = &md_redundancy_group;
3618 }
3619
54071b38
TM
3620 if (mddev->pers->sync_request == NULL &&
3621 mddev->external) {
3622 /* We are converting from a no-redundancy array
3623 * to a redundancy array and metadata is managed
3624 * externally so we need to be sure that writes
3625 * won't block due to a need to transition
3626 * clean->dirty
3627 * until external management is started.
3628 */
3629 mddev->in_sync = 0;
3630 mddev->safemode_delay = 0;
3631 mddev->safemode = 0;
3632 }
3633
dafb20fa 3634 rdev_for_each(rdev, mddev) {
e93f68a1
N
3635 if (rdev->raid_disk < 0)
3636 continue;
bf2cb0da 3637 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N
3638 rdev->new_raid_disk = -1;
3639 if (rdev->new_raid_disk == rdev->raid_disk)
3640 continue;
36fad858 3641 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3642 }
dafb20fa 3643 rdev_for_each(rdev, mddev) {
e93f68a1
N
3644 if (rdev->raid_disk < 0)
3645 continue;
3646 if (rdev->new_raid_disk == rdev->raid_disk)
3647 continue;
3648 rdev->raid_disk = rdev->new_raid_disk;
3649 if (rdev->raid_disk < 0)
3a981b03 3650 clear_bit(In_sync, &rdev->flags);
e93f68a1 3651 else {
36fad858
NK
3652 if (sysfs_link_rdev(mddev, rdev))
3653 printk(KERN_WARNING "md: cannot register rd%d"
3654 " for %s after level change\n",
3655 rdev->raid_disk, mdname(mddev));
3a981b03 3656 }
e93f68a1
N
3657 }
3658
3659 module_put(mddev->pers->owner);
245f46c2
N
3660 mddev->pers = pers;
3661 mddev->private = priv;
3662 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3663 mddev->level = mddev->new_level;
3664 mddev->layout = mddev->new_layout;
664e7c41 3665 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3666 mddev->delta_disks = 0;
2c810cdd 3667 mddev->reshape_backwards = 0;
fee68723 3668 mddev->degraded = 0;
9af204cf
TM
3669 if (mddev->pers->sync_request == NULL) {
3670 /* this is now an array without redundancy, so
3671 * it must always be in_sync
3672 */
3673 mddev->in_sync = 1;
3674 del_timer_sync(&mddev->safemode_timer);
3675 }
245f46c2 3676 pers->run(mddev);
245f46c2 3677 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3678 mddev_resume(mddev);
5cac7861 3679 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3680 md_new_event(mddev);
d9d166c2
N
3681 return rv;
3682}
3683
3684static struct md_sysfs_entry md_level =
80ca3a44 3685__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3686
d4dbd025
N
3687
3688static ssize_t
fd01b88c 3689layout_show(struct mddev *mddev, char *page)
d4dbd025
N
3690{
3691 /* just a number, not meaningful for all levels */
08a02ecd
N
3692 if (mddev->reshape_position != MaxSector &&
3693 mddev->layout != mddev->new_layout)
3694 return sprintf(page, "%d (%d)\n",
3695 mddev->new_layout, mddev->layout);
d4dbd025
N
3696 return sprintf(page, "%d\n", mddev->layout);
3697}
3698
3699static ssize_t
fd01b88c 3700layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N
3701{
3702 char *e;
3703 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N
3704
3705 if (!*buf || (*e && *e != '\n'))
3706 return -EINVAL;
3707
b3546035
N
3708 if (mddev->pers) {
3709 int err;
50ac168a 3710 if (mddev->pers->check_reshape == NULL)
b3546035 3711 return -EBUSY;
597a711b 3712 mddev->new_layout = n;
50ac168a 3713 err = mddev->pers->check_reshape(mddev);
597a711b
N
3714 if (err) {
3715 mddev->new_layout = mddev->layout;
b3546035 3716 return err;
597a711b 3717 }
b3546035 3718 } else {
08a02ecd 3719 mddev->new_layout = n;
b3546035
N
3720 if (mddev->reshape_position == MaxSector)
3721 mddev->layout = n;
3722 }
d4dbd025
N
3723 return len;
3724}
3725static struct md_sysfs_entry md_layout =
80ca3a44 3726__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N
3727
3728
eae1701f 3729static ssize_t
fd01b88c 3730raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3731{
bb636547
N
3732 if (mddev->raid_disks == 0)
3733 return 0;
08a02ecd
N
3734 if (mddev->reshape_position != MaxSector &&
3735 mddev->delta_disks != 0)
3736 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3737 mddev->raid_disks - mddev->delta_disks);
eae1701f
N
3738 return sprintf(page, "%d\n", mddev->raid_disks);
3739}
3740
fd01b88c 3741static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N
3742
3743static ssize_t
fd01b88c 3744raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3745{
da943b99
N
3746 char *e;
3747 int rv = 0;
3748 unsigned long n = simple_strtoul(buf, &e, 10);
3749
3750 if (!*buf || (*e && *e != '\n'))
3751 return -EINVAL;
3752
3753 if (mddev->pers)
3754 rv = update_raid_disks(mddev, n);
08a02ecd 3755 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3756 struct md_rdev *rdev;
08a02ecd 3757 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c
N
3758
3759 rdev_for_each(rdev, mddev) {
3760 if (olddisks < n &&
3761 rdev->data_offset < rdev->new_data_offset)
3762 return -EINVAL;
3763 if (olddisks > n &&
3764 rdev->data_offset > rdev->new_data_offset)
3765 return -EINVAL;
3766 }
08a02ecd
N
3767 mddev->delta_disks = n - olddisks;
3768 mddev->raid_disks = n;
2c810cdd 3769 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3770 } else
da943b99
N
3771 mddev->raid_disks = n;
3772 return rv ? rv : len;
3773}
3774static struct md_sysfs_entry md_raid_disks =
80ca3a44 3775__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3776
3b34380a 3777static ssize_t
fd01b88c 3778chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3779{
08a02ecd 3780 if (mddev->reshape_position != MaxSector &&
664e7c41
AN
3781 mddev->chunk_sectors != mddev->new_chunk_sectors)
3782 return sprintf(page, "%d (%d)\n",
3783 mddev->new_chunk_sectors << 9,
9d8f0363
AN
3784 mddev->chunk_sectors << 9);
3785 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N
3786}
3787
3788static ssize_t
fd01b88c 3789chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3790{
3b34380a
N
3791 char *e;
3792 unsigned long n = simple_strtoul(buf, &e, 10);
3793
3b34380a
N
3794 if (!*buf || (*e && *e != '\n'))
3795 return -EINVAL;
3796
b3546035
N
3797 if (mddev->pers) {
3798 int err;
50ac168a 3799 if (mddev->pers->check_reshape == NULL)
b3546035 3800 return -EBUSY;
597a711b 3801 mddev->new_chunk_sectors = n >> 9;
50ac168a 3802 err = mddev->pers->check_reshape(mddev);
597a711b
N
3803 if (err) {
3804 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3805 return err;
597a711b 3806 }
b3546035 3807 } else {
664e7c41 3808 mddev->new_chunk_sectors = n >> 9;
b3546035 3809 if (mddev->reshape_position == MaxSector)
9d8f0363 3810 mddev->chunk_sectors = n >> 9;
b3546035 3811 }
3b34380a
N
3812 return len;
3813}
3814static struct md_sysfs_entry md_chunk_size =
80ca3a44 3815__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3816
a94213b1 3817static ssize_t
fd01b88c 3818resync_start_show(struct mddev *mddev, char *page)
a94213b1 3819{
d1a7c503
N
3820 if (mddev->recovery_cp == MaxSector)
3821 return sprintf(page, "none\n");
a94213b1
N
3822 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3823}
3824
3825static ssize_t
fd01b88c 3826resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3827{
a94213b1
N
3828 char *e;
3829 unsigned long long n = simple_strtoull(buf, &e, 10);
3830
b098636c 3831 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3832 return -EBUSY;
06e3c817
DW
3833 if (cmd_match(buf, "none"))
3834 n = MaxSector;
3835 else if (!*buf || (*e && *e != '\n'))
a94213b1
N
3836 return -EINVAL;
3837
3838 mddev->recovery_cp = n;
3839 return len;
3840}
3841static struct md_sysfs_entry md_resync_start =
80ca3a44 3842__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3843
9e653b63
N
3844/*
3845 * The array state can be:
3846 *
3847 * clear
3848 * No devices, no size, no level
3849 * Equivalent to STOP_ARRAY ioctl
3850 * inactive
3851 * May have some settings, but array is not active
3852 * all IO results in error
3853 * When written, doesn't tear down array, but just stops it
3854 * suspended (not supported yet)
3855 * All IO requests will block. The array can be reconfigured.
910d8cb3 3856 * Writing this, if accepted, will block until array is quiescent
9e653b63
N
3857 * readonly
3858 * no resync can happen. no superblocks get written.
3859 * write requests fail
3860 * read-auto
3861 * like readonly, but behaves like 'clean' on a write request.
3862 *
3863 * clean - no pending writes, but otherwise active.
3864 * When written to inactive array, starts without resync
3865 * If a write request arrives then
3866 * if metadata is known, mark 'dirty' and switch to 'active'.
3867 * if not known, block and switch to write-pending
3868 * If written to an active array that has pending writes, then fails.
3869 * active
3870 * fully active: IO and resync can be happening.
3871 * When written to inactive array, starts with resync
3872 *
3873 * write-pending
3874 * clean, but writes are blocked waiting for 'active' to be written.
3875 *
3876 * active-idle
3877 * like active, but no writes have been seen for a while (100msec).
3878 *
3879 */
3880enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3881 write_pending, active_idle, bad_word};
05381954 3882static char *array_states[] = {
9e653b63
N
3883 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3884 "write-pending", "active-idle", NULL };
3885
3886static int match_word(const char *word, char **list)
3887{
3888 int n;
3889 for (n=0; list[n]; n++)
3890 if (cmd_match(word, list[n]))
3891 break;
3892 return n;
3893}
3894
3895static ssize_t
fd01b88c 3896array_state_show(struct mddev *mddev, char *page)
9e653b63
N
3897{
3898 enum array_state st = inactive;
3899
3900 if (mddev->pers)
3901 switch(mddev->ro) {
3902 case 1:
3903 st = readonly;
3904 break;
3905 case 2:
3906 st = read_auto;
3907 break;
3908 case 0:
3909 if (mddev->in_sync)
3910 st = clean;
070dc6dd 3911 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3912 st = write_pending;
9e653b63
N
3913 else if (mddev->safemode)
3914 st = active_idle;
3915 else
3916 st = active;
3917 }
3918 else {
3919 if (list_empty(&mddev->disks) &&
3920 mddev->raid_disks == 0 &&
58c0fed4 3921 mddev->dev_sectors == 0)
9e653b63
N
3922 st = clear;
3923 else
3924 st = inactive;
3925 }
3926 return sprintf(page, "%s\n", array_states[st]);
3927}
3928
fd01b88c
N
3929static int do_md_stop(struct mddev * mddev, int ro, int is_open);
3930static int md_set_readonly(struct mddev * mddev, int is_open);
3931static int do_md_run(struct mddev * mddev);
3932static int restart_array(struct mddev *mddev);
9e653b63
N
3933
3934static ssize_t
fd01b88c 3935array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N
3936{
3937 int err = -EINVAL;
3938 enum array_state st = match_word(buf, array_states);
3939 switch(st) {
3940 case bad_word:
3941 break;
3942 case clear:
3943 /* stopping an active array */
f2ea68cf 3944 if (atomic_read(&mddev->openers) > 0)
e691063a 3945 return -EBUSY;
df5b20cf 3946 err = do_md_stop(mddev, 0, 0);
9e653b63
N
3947 break;
3948 case inactive:
3949 /* stopping an active array */
3950 if (mddev->pers) {
f2ea68cf 3951 if (atomic_read(&mddev->openers) > 0)
9e653b63 3952 return -EBUSY;
df5b20cf 3953 err = do_md_stop(mddev, 2, 0);
e691063a
N
3954 } else
3955 err = 0; /* already inactive */
9e653b63
N
3956 break;
3957 case suspended:
3958 break; /* not supported yet */
3959 case readonly:
3960 if (mddev->pers)
a4bd82d0 3961 err = md_set_readonly(mddev, 0);
9e653b63
N
3962 else {
3963 mddev->ro = 1;
648b629e 3964 set_disk_ro(mddev->gendisk, 1);
9e653b63
N
3965 err = do_md_run(mddev);
3966 }
3967 break;
3968 case read_auto:
9e653b63 3969 if (mddev->pers) {
80268ee9 3970 if (mddev->ro == 0)
a4bd82d0 3971 err = md_set_readonly(mddev, 0);
80268ee9 3972 else if (mddev->ro == 1)
648b629e
N
3973 err = restart_array(mddev);
3974 if (err == 0) {
3975 mddev->ro = 2;
3976 set_disk_ro(mddev->gendisk, 0);
3977 }
9e653b63
N
3978 } else {
3979 mddev->ro = 2;
3980 err = do_md_run(mddev);
3981 }
3982 break;
3983 case clean:
3984 if (mddev->pers) {
3985 restart_array(mddev);
3986 spin_lock_irq(&mddev->write_lock);
3987 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N
3988 if (mddev->in_sync == 0) {
3989 mddev->in_sync = 1;
31a59e34
N
3990 if (mddev->safemode == 1)
3991 mddev->safemode = 0;
070dc6dd 3992 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N
3993 }
3994 err = 0;
3995 } else
3996 err = -EBUSY;
9e653b63 3997 spin_unlock_irq(&mddev->write_lock);
5bf29597
N
3998 } else
3999 err = -EINVAL;
9e653b63
N
4000 break;
4001 case active:
4002 if (mddev->pers) {
4003 restart_array(mddev);
070dc6dd 4004 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N
4005 wake_up(&mddev->sb_wait);
4006 err = 0;
4007 } else {
4008 mddev->ro = 0;
648b629e 4009 set_disk_ro(mddev->gendisk, 0);
9e653b63
N
4010 err = do_md_run(mddev);
4011 }
4012 break;
4013 case write_pending:
4014 case active_idle:
4015 /* these cannot be set */
4016 break;
4017 }
4018 if (err)
4019 return err;
0fd62b86 4020 else {
1d23f178
N
4021 if (mddev->hold_active == UNTIL_IOCTL)
4022 mddev->hold_active = 0;
00bcb4ac 4023 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 4024 return len;
0fd62b86 4025 }
9e653b63 4026}
80ca3a44
N
4027static struct md_sysfs_entry md_array_state =
4028__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 4029
1e50915f 4030static ssize_t
fd01b88c 4031max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB
4032 return sprintf(page, "%d\n",
4033 atomic_read(&mddev->max_corr_read_errors));
4034}
4035
4036static ssize_t
fd01b88c 4037max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB
4038{
4039 char *e;
4040 unsigned long n = simple_strtoul(buf, &e, 10);
4041
4042 if (*buf && (*e == 0 || *e == '\n')) {
4043 atomic_set(&mddev->max_corr_read_errors, n);
4044 return len;
4045 }
4046 return -EINVAL;
4047}
4048
4049static struct md_sysfs_entry max_corr_read_errors =
4050__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4051 max_corrected_read_errors_store);
4052
6d7ff738 4053static ssize_t
fd01b88c 4054null_show(struct mddev *mddev, char *page)
6d7ff738
N
4055{
4056 return -EINVAL;
4057}
4058
4059static ssize_t
fd01b88c 4060new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N
4061{
4062 /* buf must be %d:%d\n? giving major and minor numbers */
4063 /* The new device is added to the array.
4064 * If the array has a persistent superblock, we read the
4065 * superblock to initialise info and check validity.
4066 * Otherwise, only checking done is that in bind_rdev_to_array,
4067 * which mainly checks size.
4068 */
4069 char *e;
4070 int major = simple_strtoul(buf, &e, 10);
4071 int minor;
4072 dev_t dev;
3cb03002 4073 struct md_rdev *rdev;
6d7ff738
N
4074 int err;
4075
4076 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4077 return -EINVAL;
4078 minor = simple_strtoul(e+1, &e, 10);
4079 if (*e && *e != '\n')
4080 return -EINVAL;
4081 dev = MKDEV(major, minor);
4082 if (major != MAJOR(dev) ||
4083 minor != MINOR(dev))
4084 return -EOVERFLOW;
4085
4086
4087 if (mddev->persistent) {
4088 rdev = md_import_device(dev, mddev->major_version,
4089 mddev->minor_version);
4090 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N
4091 struct md_rdev *rdev0
4092 = list_entry(mddev->disks.next,
4093 struct md_rdev, same_set);
6d7ff738
N
4094 err = super_types[mddev->major_version]
4095 .load_super(rdev, rdev0, mddev->minor_version);
4096 if (err < 0)
4097 goto out;
4098 }
c5d79adb
N
4099 } else if (mddev->external)
4100 rdev = md_import_device(dev, -2, -1);
4101 else
6d7ff738
N
4102 rdev = md_import_device(dev, -1, -1);
4103
4104 if (IS_ERR(rdev))
4105 return PTR_ERR(rdev);
4106 err = bind_rdev_to_array(rdev, mddev);
4107 out:
4108 if (err)
4109 export_rdev(rdev);
4110 return err ? err : len;
4111}
4112
4113static struct md_sysfs_entry md_new_device =
80ca3a44 4114__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4115
9b1d1dac 4116static ssize_t
fd01b88c 4117bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC
4118{
4119 char *end;
4120 unsigned long chunk, end_chunk;
4121
4122 if (!mddev->bitmap)
4123 goto out;
4124 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4125 while (*buf) {
4126 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4127 if (buf == end) break;
4128 if (*end == '-') { /* range */
4129 buf = end + 1;
4130 end_chunk = simple_strtoul(buf, &end, 0);
4131 if (buf == end) break;
4132 }
4133 if (*end && !isspace(*end)) break;
4134 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4135 buf = skip_spaces(end);
9b1d1dac
PC
4136 }
4137 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4138out:
4139 return len;
4140}
4141
4142static struct md_sysfs_entry md_bitmap =
4143__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4144
a35b0d69 4145static ssize_t
fd01b88c 4146size_show(struct mddev *mddev, char *page)
a35b0d69 4147{
58c0fed4
AN
4148 return sprintf(page, "%llu\n",
4149 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N
4150}
4151
fd01b88c 4152static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N
4153
4154static ssize_t
fd01b88c 4155size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N
4156{
4157 /* If array is inactive, we can reduce the component size, but
4158 * not increase it (except from 0).
4159 * If array is active, we can try an on-line resize
4160 */
b522adcd
DW
4161 sector_t sectors;
4162 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4163
58c0fed4
AN
4164 if (err < 0)
4165 return err;
a35b0d69 4166 if (mddev->pers) {
58c0fed4 4167 err = update_size(mddev, sectors);
850b2b42 4168 md_update_sb(mddev, 1);
a35b0d69 4169 } else {
58c0fed4
AN
4170 if (mddev->dev_sectors == 0 ||
4171 mddev->dev_sectors > sectors)
4172 mddev->dev_sectors = sectors;
a35b0d69
N
4173 else
4174 err = -ENOSPC;
4175 }
4176 return err ? err : len;
4177}
4178
4179static struct md_sysfs_entry md_size =
80ca3a44 4180__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4181
8bb93aac
N
4182
4183/* Metdata version.
e691063a
N
4184 * This is one of
4185 * 'none' for arrays with no metadata (good luck...)
4186 * 'external' for arrays with externally managed metadata,
8bb93aac
N
4187 * or N.M for internally known formats
4188 */
4189static ssize_t
fd01b88c 4190metadata_show(struct mddev *mddev, char *page)
8bb93aac
N
4191{
4192 if (mddev->persistent)
4193 return sprintf(page, "%d.%d\n",
4194 mddev->major_version, mddev->minor_version);
e691063a
N
4195 else if (mddev->external)
4196 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N
4197 else
4198 return sprintf(page, "none\n");
4199}
4200
4201static ssize_t
fd01b88c 4202metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N
4203{
4204 int major, minor;
4205 char *e;
ea43ddd8
N
4206 /* Changing the details of 'external' metadata is
4207 * always permitted. Otherwise there must be
4208 * no devices attached to the array.
4209 */
4210 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4211 ;
4212 else if (!list_empty(&mddev->disks))
8bb93aac
N
4213 return -EBUSY;
4214
4215 if (cmd_match(buf, "none")) {
4216 mddev->persistent = 0;
e691063a
N
4217 mddev->external = 0;
4218 mddev->major_version = 0;
4219 mddev->minor_version = 90;
4220 return len;
4221 }
4222 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4223 size_t namelen = len-9;
e691063a
N
4224 if (namelen >= sizeof(mddev->metadata_type))
4225 namelen = sizeof(mddev->metadata_type)-1;
4226 strncpy(mddev->metadata_type, buf+9, namelen);
4227 mddev->metadata_type[namelen] = 0;
4228 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4229 mddev->metadata_type[--namelen] = 0;
4230 mddev->persistent = 0;
4231 mddev->external = 1;
8bb93aac
N
4232 mddev->major_version = 0;
4233 mddev->minor_version = 90;
4234 return len;
4235 }
4236 major = simple_strtoul(buf, &e, 10);
4237 if (e==buf || *e != '.')
4238 return -EINVAL;
4239 buf = e+1;
4240 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4241 if (e==buf || (*e && *e != '\n') )
8bb93aac 4242 return -EINVAL;
50511da3 4243 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N
4244 return -ENOENT;
4245 mddev->major_version = major;
4246 mddev->minor_version = minor;
4247 mddev->persistent = 1;
e691063a 4248 mddev->external = 0;
8bb93aac
N
4249 return len;
4250}
4251
4252static struct md_sysfs_entry md_metadata =
80ca3a44 4253__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4254
24dd469d 4255static ssize_t
fd01b88c 4256action_show(struct mddev *mddev, char *page)
24dd469d 4257{
7eec314d 4258 char *type = "idle";
b6a9ce68
N
4259 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4260 type = "frozen";
4261 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4262 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N
4263 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4264 type = "reshape";
4265 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N
4266 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4267 type = "resync";
4268 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4269 type = "check";
4270 else
4271 type = "repair";
72a23c21 4272 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N
4273 type = "recover";
4274 }
4275 return sprintf(page, "%s\n", type);
4276}
4277
fd01b88c 4278static void reap_sync_thread(struct mddev *mddev);
7ebc0be7 4279
24dd469d 4280static ssize_t
fd01b88c 4281action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4282{
7eec314d
N
4283 if (!mddev->pers || !mddev->pers->sync_request)
4284 return -EINVAL;
4285
b6a9ce68
N
4286 if (cmd_match(page, "frozen"))
4287 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4288 else
4289 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4290
4291 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N
4292 if (mddev->sync_thread) {
4293 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4294 reap_sync_thread(mddev);
7eec314d 4295 }
03c902e1
N
4296 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4297 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4298 return -EBUSY;
72a23c21
NB
4299 else if (cmd_match(page, "resync"))
4300 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4301 else if (cmd_match(page, "recover")) {
4302 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4303 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4304 } else if (cmd_match(page, "reshape")) {
16484bf5
N
4305 int err;
4306 if (mddev->pers->start_reshape == NULL)
4307 return -EINVAL;
4308 err = mddev->pers->start_reshape(mddev);
4309 if (err)
4310 return err;
a99ac971 4311 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4312 } else {
bce74dac 4313 if (cmd_match(page, "check"))
7eec314d 4314 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4315 else if (!cmd_match(page, "repair"))
7eec314d
N
4316 return -EINVAL;
4317 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4318 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4319 }
03c902e1 4320 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4321 md_wakeup_thread(mddev->thread);
00bcb4ac 4322 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N
4323 return len;
4324}
4325
9d88883e 4326static ssize_t
fd01b88c 4327mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N
4328{
4329 return sprintf(page, "%llu\n",
4330 (unsigned long long) mddev->resync_mismatches);
4331}
4332
80ca3a44
N
4333static struct md_sysfs_entry md_scan_mode =
4334__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 4335
96de1e66 4336
80ca3a44 4337static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4338
88202a0c 4339static ssize_t
fd01b88c 4340sync_min_show(struct mddev *mddev, char *page)
88202a0c
N
4341{
4342 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4343 mddev->sync_speed_min ? "local": "system");
4344}
4345
4346static ssize_t
fd01b88c 4347sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N
4348{
4349 int min;
4350 char *e;
4351 if (strncmp(buf, "system", 6)==0) {
4352 mddev->sync_speed_min = 0;
4353 return len;
4354 }
4355 min = simple_strtoul(buf, &e, 10);
4356 if (buf == e || (*e && *e != '\n') || min <= 0)
4357 return -EINVAL;
4358 mddev->sync_speed_min = min;
4359 return len;
4360}
4361
4362static struct md_sysfs_entry md_sync_min =
4363__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4364
4365static ssize_t
fd01b88c 4366sync_max_show(struct mddev *mddev, char *page)
88202a0c
N
4367{
4368 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4369 mddev->sync_speed_max ? "local": "system");
4370}
4371
4372static ssize_t
fd01b88c 4373sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N
4374{
4375 int max;
4376 char *e;
4377 if (strncmp(buf, "system", 6)==0) {
4378 mddev->sync_speed_max = 0;
4379 return len;
4380 }
4381 max = simple_strtoul(buf, &e, 10);
4382 if (buf == e || (*e && *e != '\n') || max <= 0)
4383 return -EINVAL;
4384 mddev->sync_speed_max = max;
4385 return len;
4386}
4387
4388static struct md_sysfs_entry md_sync_max =
4389__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4390
d7f3d291 4391static ssize_t
fd01b88c 4392degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP
4393{
4394 return sprintf(page, "%d\n", mddev->degraded);
4395}
4396static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4397
90b08710 4398static ssize_t
fd01b88c 4399sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS
4400{
4401 return sprintf(page, "%d\n", mddev->parallel_resync);
4402}
4403
4404static ssize_t
fd01b88c 4405sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS
4406{
4407 long n;
4408
4409 if (strict_strtol(buf, 10, &n))
4410 return -EINVAL;
4411
4412 if (n != 0 && n != 1)
4413 return -EINVAL;
4414
4415 mddev->parallel_resync = n;
4416
4417 if (mddev->sync_thread)
4418 wake_up(&resync_wait);
4419
4420 return len;
4421}
4422
4423/* force parallel resync, even with shared block devices */
4424static struct md_sysfs_entry md_sync_force_parallel =
4425__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4426 sync_force_parallel_show, sync_force_parallel_store);
4427
88202a0c 4428static ssize_t
fd01b88c 4429sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N
4430{
4431 unsigned long resync, dt, db;
d1a7c503
N
4432 if (mddev->curr_resync == 0)
4433 return sprintf(page, "none\n");
9687a60c
AN
4434 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4435 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4436 if (!dt) dt++;
9687a60c
AN
4437 db = resync - mddev->resync_mark_cnt;
4438 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N
4439}
4440
80ca3a44 4441static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N
4442
4443static ssize_t
fd01b88c 4444sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4445{
13ae864b 4446 unsigned long long max_sectors, resync;
88202a0c 4447
acb180b0
N
4448 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4449 return sprintf(page, "none\n");
4450
c804cdec
N
4451 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4452 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4453 max_sectors = mddev->resync_max_sectors;
88202a0c 4454 else
58c0fed4 4455 max_sectors = mddev->dev_sectors;
88202a0c 4456
acb180b0 4457 resync = mddev->curr_resync_completed;
13ae864b 4458 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N
4459}
4460
80ca3a44 4461static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4462
5e96ee65 4463static ssize_t
fd01b88c 4464min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB
4465{
4466 return sprintf(page, "%llu\n",
4467 (unsigned long long)mddev->resync_min);
4468}
4469static ssize_t
fd01b88c 4470min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB
4471{
4472 unsigned long long min;
4473 if (strict_strtoull(buf, 10, &min))
4474 return -EINVAL;
4475 if (min > mddev->resync_max)
4476 return -EINVAL;
4477 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4478 return -EBUSY;
4479
4480 /* Must be a multiple of chunk_size */
9d8f0363 4481 if (mddev->chunk_sectors) {
2ac06c33 4482 sector_t temp = min;
9d8f0363 4483 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB
4484 return -EINVAL;
4485 }
4486 mddev->resync_min = min;
4487
4488 return len;
4489}
4490
4491static struct md_sysfs_entry md_min_sync =
4492__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4493
c6207277 4494static ssize_t
fd01b88c 4495max_sync_show(struct mddev *mddev, char *page)
c6207277
N
4496{
4497 if (mddev->resync_max == MaxSector)
4498 return sprintf(page, "max\n");
4499 else
4500 return sprintf(page, "%llu\n",
4501 (unsigned long long)mddev->resync_max);
4502}
4503static ssize_t
fd01b88c 4504max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N
4505{
4506 if (strncmp(buf, "max", 3) == 0)
4507 mddev->resync_max = MaxSector;
4508 else {
5e96ee65
NB
4509 unsigned long long max;
4510 if (strict_strtoull(buf, 10, &max))
4511 return -EINVAL;
4512 if (max < mddev->resync_min)
c6207277
N
4513 return -EINVAL;
4514 if (max < mddev->resync_max &&
4d484a4a 4515 mddev->ro == 0 &&
c6207277
N
4516 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4517 return -EBUSY;
4518
4519 /* Must be a multiple of chunk_size */
9d8f0363 4520 if (mddev->chunk_sectors) {
2ac06c33 4521 sector_t temp = max;
9d8f0363 4522 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N
4523 return -EINVAL;
4524 }
4525 mddev->resync_max = max;
4526 }
4527 wake_up(&mddev->recovery_wait);
4528 return len;
4529}
4530
4531static struct md_sysfs_entry md_max_sync =
4532__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4533
e464eafd 4534static ssize_t
fd01b88c 4535suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N
4536{
4537 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4538}
4539
4540static ssize_t
fd01b88c 4541suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N
4542{
4543 char *e;
4544 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4545 unsigned long long old = mddev->suspend_lo;
e464eafd 4546
b8d966ef
N
4547 if (mddev->pers == NULL ||
4548 mddev->pers->quiesce == NULL)
e464eafd
N
4549 return -EINVAL;
4550 if (buf == e || (*e && *e != '\n'))
4551 return -EINVAL;
23ddff37
N
4552
4553 mddev->suspend_lo = new;
4554 if (new >= old)
4555 /* Shrinking suspended region */
e464eafd 4556 mddev->pers->quiesce(mddev, 2);
23ddff37
N
4557 else {
4558 /* Expanding suspended region - need to wait */
4559 mddev->pers->quiesce(mddev, 1);
4560 mddev->pers->quiesce(mddev, 0);
4561 }
4562 return len;
e464eafd
N
4563}
4564static struct md_sysfs_entry md_suspend_lo =
4565__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4566
4567
4568static ssize_t
fd01b88c 4569suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N
4570{
4571 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4572}
4573
4574static ssize_t
fd01b88c 4575suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N
4576{
4577 char *e;
4578 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4579 unsigned long long old = mddev->suspend_hi;
e464eafd 4580
b8d966ef
N
4581 if (mddev->pers == NULL ||
4582 mddev->pers->quiesce == NULL)
e464eafd
N
4583 return -EINVAL;
4584 if (buf == e || (*e && *e != '\n'))
4585 return -EINVAL;
23ddff37
N
4586
4587 mddev->suspend_hi = new;
4588 if (new <= old)
4589 /* Shrinking suspended region */
4590 mddev->pers->quiesce(mddev, 2);
4591 else {
4592 /* Expanding suspended region - need to wait */
e464eafd
N
4593 mddev->pers->quiesce(mddev, 1);
4594 mddev->pers->quiesce(mddev, 0);
23ddff37
N
4595 }
4596 return len;
e464eafd
N
4597}
4598static struct md_sysfs_entry md_suspend_hi =
4599__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4600
08a02ecd 4601static ssize_t
fd01b88c 4602reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N
4603{
4604 if (mddev->reshape_position != MaxSector)
4605 return sprintf(page, "%llu\n",
4606 (unsigned long long)mddev->reshape_position);
4607 strcpy(page, "none\n");
4608 return 5;
4609}
4610
4611static ssize_t
fd01b88c 4612reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4613{
c6563a8c 4614 struct md_rdev *rdev;
08a02ecd
N
4615 char *e;
4616 unsigned long long new = simple_strtoull(buf, &e, 10);
4617 if (mddev->pers)
4618 return -EBUSY;
4619 if (buf == e || (*e && *e != '\n'))
4620 return -EINVAL;
4621 mddev->reshape_position = new;
4622 mddev->delta_disks = 0;
2c810cdd 4623 mddev->reshape_backwards = 0;
08a02ecd
N
4624 mddev->new_level = mddev->level;
4625 mddev->new_layout = mddev->layout;
664e7c41 4626 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N
4627 rdev_for_each(rdev, mddev)
4628 rdev->new_data_offset = rdev->data_offset;
08a02ecd
N
4629 return len;
4630}
4631
4632static struct md_sysfs_entry md_reshape_position =
4633__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4634 reshape_position_store);
4635
2c810cdd
N
4636static ssize_t
4637reshape_direction_show(struct mddev *mddev, char *page)
4638{
4639 return sprintf(page, "%s\n",
4640 mddev->reshape_backwards ? "backwards" : "forwards");
4641}
4642
4643static ssize_t
4644reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4645{
4646 int backwards = 0;
4647 if (cmd_match(buf, "forwards"))
4648 backwards = 0;
4649 else if (cmd_match(buf, "backwards"))
4650 backwards = 1;
4651 else
4652 return -EINVAL;
4653 if (mddev->reshape_backwards == backwards)
4654 return len;
4655
4656 /* check if we are allowed to change */
4657 if (mddev->delta_disks)
4658 return -EBUSY;
4659
4660 if (mddev->persistent &&
4661 mddev->major_version == 0)
4662 return -EINVAL;
4663
4664 mddev->reshape_backwards = backwards;
4665 return len;
4666}
4667
4668static struct md_sysfs_entry md_reshape_direction =
4669__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4670 reshape_direction_store);
4671
b522adcd 4672static ssize_t
fd01b88c 4673array_size_show(struct mddev *mddev, char *page)
b522adcd
DW
4674{
4675 if (mddev->external_size)
4676 return sprintf(page, "%llu\n",
4677 (unsigned long long)mddev->array_sectors/2);
4678 else
4679 return sprintf(page, "default\n");
4680}
4681
4682static ssize_t
fd01b88c 4683array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW
4684{
4685 sector_t sectors;
4686
4687 if (strncmp(buf, "default", 7) == 0) {
4688 if (mddev->pers)
4689 sectors = mddev->pers->size(mddev, 0, 0);
4690 else
4691 sectors = mddev->array_sectors;
4692
4693 mddev->external_size = 0;
4694 } else {
4695 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4696 return -EINVAL;
4697 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4698 return -E2BIG;
b522adcd
DW
4699
4700 mddev->external_size = 1;
4701 }
4702
4703 mddev->array_sectors = sectors;
cbe6ef1d
N
4704 if (mddev->pers) {
4705 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4706 revalidate_disk(mddev->gendisk);
cbe6ef1d 4707 }
b522adcd
DW
4708 return len;
4709}
4710
4711static struct md_sysfs_entry md_array_size =
4712__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4713 array_size_store);
e464eafd 4714
eae1701f
N
4715static struct attribute *md_default_attrs[] = {
4716 &md_level.attr,
d4dbd025 4717 &md_layout.attr,
eae1701f 4718 &md_raid_disks.attr,
3b34380a 4719 &md_chunk_size.attr,
a35b0d69 4720 &md_size.attr,
a94213b1 4721 &md_resync_start.attr,
8bb93aac 4722 &md_metadata.attr,
6d7ff738 4723 &md_new_device.attr,
16f17b39 4724 &md_safe_delay.attr,
9e653b63 4725 &md_array_state.attr,
08a02ecd 4726 &md_reshape_position.attr,
2c810cdd 4727 &md_reshape_direction.attr,
b522adcd 4728 &md_array_size.attr,
1e50915f 4729 &max_corr_read_errors.attr,
411036fa
N
4730 NULL,
4731};
4732
4733static struct attribute *md_redundancy_attrs[] = {
24dd469d 4734 &md_scan_mode.attr,
9d88883e 4735 &md_mismatches.attr,
88202a0c
N
4736 &md_sync_min.attr,
4737 &md_sync_max.attr,
4738 &md_sync_speed.attr,
90b08710 4739 &md_sync_force_parallel.attr,
88202a0c 4740 &md_sync_completed.attr,
5e96ee65 4741 &md_min_sync.attr,
c6207277 4742 &md_max_sync.attr,
e464eafd
N
4743 &md_suspend_lo.attr,
4744 &md_suspend_hi.attr,
9b1d1dac 4745 &md_bitmap.attr,
d7f3d291 4746 &md_degraded.attr,
eae1701f
N
4747 NULL,
4748};
411036fa
N
4749static struct attribute_group md_redundancy_group = {
4750 .name = NULL,
4751 .attrs = md_redundancy_attrs,
4752};
4753
eae1701f
N
4754
4755static ssize_t
4756md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4757{
4758 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4759 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4760 ssize_t rv;
eae1701f
N
4761
4762 if (!entry->show)
4763 return -EIO;
af8a2434
N
4764 spin_lock(&all_mddevs_lock);
4765 if (list_empty(&mddev->all_mddevs)) {
4766 spin_unlock(&all_mddevs_lock);
4767 return -EBUSY;
4768 }
4769 mddev_get(mddev);
4770 spin_unlock(&all_mddevs_lock);
4771
5dc5cf7d
IM
4772 rv = mddev_lock(mddev);
4773 if (!rv) {
4774 rv = entry->show(mddev, page);
4775 mddev_unlock(mddev);
4776 }
af8a2434 4777 mddev_put(mddev);
96de1e66 4778 return rv;
eae1701f
N
4779}
4780
4781static ssize_t
4782md_attr_store(struct kobject *kobj, struct attribute *attr,
4783 const char *page, size_t length)
4784{
4785 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4786 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4787 ssize_t rv;
eae1701f
N
4788
4789 if (!entry->store)
4790 return -EIO;
67463acb
N
4791 if (!capable(CAP_SYS_ADMIN))
4792 return -EACCES;
af8a2434
N
4793 spin_lock(&all_mddevs_lock);
4794 if (list_empty(&mddev->all_mddevs)) {
4795 spin_unlock(&all_mddevs_lock);
4796 return -EBUSY;
4797 }
4798 mddev_get(mddev);
4799 spin_unlock(&all_mddevs_lock);
5dc5cf7d
IM
4800 rv = mddev_lock(mddev);
4801 if (!rv) {
4802 rv = entry->store(mddev, page, length);
4803 mddev_unlock(mddev);
4804 }
af8a2434 4805 mddev_put(mddev);
96de1e66 4806 return rv;
eae1701f
N
4807}
4808
4809static void md_free(struct kobject *ko)
4810{
fd01b88c 4811 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N
4812
4813 if (mddev->sysfs_state)
4814 sysfs_put(mddev->sysfs_state);
4815
4816 if (mddev->gendisk) {
4817 del_gendisk(mddev->gendisk);
4818 put_disk(mddev->gendisk);
4819 }
4820 if (mddev->queue)
4821 blk_cleanup_queue(mddev->queue);
4822
eae1701f
N
4823 kfree(mddev);
4824}
4825
52cf25d0 4826static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N
4827 .show = md_attr_show,
4828 .store = md_attr_store,
4829};
4830static struct kobj_type md_ktype = {
4831 .release = md_free,
4832 .sysfs_ops = &md_sysfs_ops,
4833 .default_attrs = md_default_attrs,
4834};
4835
1da177e4
LT
4836int mdp_major = 0;
4837
5fd3a17e
DW
4838static void mddev_delayed_delete(struct work_struct *ws)
4839{
fd01b88c 4840 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4841
43a70507 4842 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW
4843 kobject_del(&mddev->kobj);
4844 kobject_put(&mddev->kobj);
4845}
4846
efeb53c0 4847static int md_alloc(dev_t dev, char *name)
1da177e4 4848{
48c9c27b 4849 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4850 struct mddev *mddev = mddev_find(dev);
1da177e4 4851 struct gendisk *disk;
efeb53c0
N
4852 int partitioned;
4853 int shift;
4854 int unit;
3830c62f 4855 int error;
1da177e4
LT
4856
4857 if (!mddev)
efeb53c0
N
4858 return -ENODEV;
4859
4860 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4861 shift = partitioned ? MdpMinorShift : 0;
4862 unit = MINOR(mddev->unit) >> shift;
1da177e4 4863
e804ac78
TH
4864 /* wait for any previous instance of this device to be
4865 * completely removed (mddev_delayed_delete).
d3374825 4866 */
e804ac78 4867 flush_workqueue(md_misc_wq);
d3374825 4868
48c9c27b 4869 mutex_lock(&disks_mutex);
0909dc44
N
4870 error = -EEXIST;
4871 if (mddev->gendisk)
4872 goto abort;
efeb53c0
N
4873
4874 if (name) {
4875 /* Need to ensure that 'name' is not a duplicate.
4876 */
fd01b88c 4877 struct mddev *mddev2;
efeb53c0
N
4878 spin_lock(&all_mddevs_lock);
4879
4880 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4881 if (mddev2->gendisk &&
4882 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4883 spin_unlock(&all_mddevs_lock);
0909dc44 4884 goto abort;
efeb53c0
N
4885 }
4886 spin_unlock(&all_mddevs_lock);
1da177e4 4887 }
8b765398 4888
0909dc44 4889 error = -ENOMEM;
8b765398 4890 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N
4891 if (!mddev->queue)
4892 goto abort;
409c57f3
N
4893 mddev->queue->queuedata = mddev;
4894
409c57f3 4895 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4896 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4897
1da177e4
LT
4898 disk = alloc_disk(1 << shift);
4899 if (!disk) {
8b765398
N
4900 blk_cleanup_queue(mddev->queue);
4901 mddev->queue = NULL;
0909dc44 4902 goto abort;
1da177e4 4903 }
efeb53c0 4904 disk->major = MAJOR(mddev->unit);
1da177e4 4905 disk->first_minor = unit << shift;
efeb53c0
N
4906 if (name)
4907 strcpy(disk->disk_name, name);
4908 else if (partitioned)
1da177e4 4909 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4910 else
1da177e4 4911 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT
4912 disk->fops = &md_fops;
4913 disk->private_data = mddev;
4914 disk->queue = mddev->queue;
b0140891 4915 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4916 /* Allow extended partitions. This makes the
d3374825 4917 * 'mdp' device redundant, but we can't really
92850bbd
N
4918 * remove it now.
4919 */
4920 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4921 mddev->gendisk = disk;
b0140891
N
4922 /* As soon as we call add_disk(), another thread could get
4923 * through to md_open, so make sure it doesn't get too far
4924 */
4925 mutex_lock(&mddev->open_mutex);
4926 add_disk(disk);
4927
ed9e1982
TH
4928 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4929 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N
4930 if (error) {
4931 /* This isn't possible, but as kobject_init_and_add is marked
4932 * __must_check, we must do something with the result
4933 */
5e55e2f5
N
4934 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4935 disk->disk_name);
0909dc44
N
4936 error = 0;
4937 }
00bcb4ac
N
4938 if (mddev->kobj.sd &&
4939 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4940 printk(KERN_DEBUG "pointless warning\n");
b0140891 4941 mutex_unlock(&mddev->open_mutex);
0909dc44
N
4942 abort:
4943 mutex_unlock(&disks_mutex);
00bcb4ac 4944 if (!error && mddev->kobj.sd) {
3830c62f 4945 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4946 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4947 }
d3374825 4948 mddev_put(mddev);
0909dc44 4949 return error;
efeb53c0
N
4950}
4951
4952static struct kobject *md_probe(dev_t dev, int *part, void *data)
4953{
4954 md_alloc(dev, NULL);
1da177e4
LT
4955 return NULL;
4956}
4957
efeb53c0
N
4958static int add_named_array(const char *val, struct kernel_param *kp)
4959{
4960 /* val must be "md_*" where * is not all digits.
4961 * We allocate an array with a large free minor number, and
4962 * set the name to val. val must not already be an active name.
4963 */
4964 int len = strlen(val);
4965 char buf[DISK_NAME_LEN];
4966
4967 while (len && val[len-1] == '\n')
4968 len--;
4969 if (len >= DISK_NAME_LEN)
4970 return -E2BIG;
4971 strlcpy(buf, val, len+1);
4972 if (strncmp(buf, "md_", 3) != 0)
4973 return -EINVAL;
4974 return md_alloc(0, buf);
4975}
4976
1da177e4
LT
4977static void md_safemode_timeout(unsigned long data)
4978{
fd01b88c 4979 struct mddev *mddev = (struct mddev *) data;
1da177e4 4980
0fd62b86
NB
4981 if (!atomic_read(&mddev->writes_pending)) {
4982 mddev->safemode = 1;
4983 if (mddev->external)
00bcb4ac 4984 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4985 }
1da177e4
LT
4986 md_wakeup_thread(mddev->thread);
4987}
4988
6ff8d8ec 4989static int start_dirty_degraded;
1da177e4 4990
fd01b88c 4991int md_run(struct mddev *mddev)
1da177e4 4992{
2604b703 4993 int err;
3cb03002 4994 struct md_rdev *rdev;
84fc4b56 4995 struct md_personality *pers;
1da177e4 4996
a757e64c
N
4997 if (list_empty(&mddev->disks))
4998 /* cannot run an array with no devices.. */
1da177e4 4999 return -EINVAL;
1da177e4
LT
5000
5001 if (mddev->pers)
5002 return -EBUSY;
bb4f1e9d
N
5003 /* Cannot run until previous stop completes properly */
5004 if (mddev->sysfs_active)
5005 return -EBUSY;
b6eb127d 5006
1da177e4
LT
5007 /*
5008 * Analyze all RAID superblock(s)
5009 */
1ec4a939
N
5010 if (!mddev->raid_disks) {
5011 if (!mddev->persistent)
5012 return -EINVAL;
a757e64c 5013 analyze_sbs(mddev);
1ec4a939 5014 }
1da177e4 5015
d9d166c2
N
5016 if (mddev->level != LEVEL_NONE)
5017 request_module("md-level-%d", mddev->level);
5018 else if (mddev->clevel[0])
5019 request_module("md-%s", mddev->clevel);
1da177e4
LT
5020
5021 /*
5022 * Drop all container device buffers, from now on
5023 * the only valid external interface is through the md
5024 * device.
1da177e4 5025 */
dafb20fa 5026 rdev_for_each(rdev, mddev) {
b2d444d7 5027 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT
5028 continue;
5029 sync_blockdev(rdev->bdev);
f98393a6 5030 invalidate_bdev(rdev->bdev);
f0d76d70
N
5031
5032 /* perform some consistency tests on the device.
5033 * We don't want the data to overlap the metadata,
58c0fed4 5034 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 5035 */
a6ff7e08
JB
5036 if (rdev->meta_bdev) {
5037 /* Nothing to check */;
5038 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN
5039 if (mddev->dev_sectors &&
5040 rdev->data_offset + mddev->dev_sectors
0f420358 5041 > rdev->sb_start) {
f0d76d70
N
5042 printk("md: %s: data overlaps metadata\n",
5043 mdname(mddev));
5044 return -EINVAL;
5045 }
5046 } else {
0f420358 5047 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N
5048 > rdev->data_offset) {
5049 printk("md: %s: metadata overlaps data\n",
5050 mdname(mddev));
5051 return -EINVAL;
5052 }
5053 }
00bcb4ac 5054 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT
5055 }
5056
a167f663 5057 if (mddev->bio_set == NULL)
a519b26d 5058 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
fd01b88c 5059 sizeof(struct mddev *));
a167f663 5060
1da177e4 5061 spin_lock(&pers_lock);
d9d166c2 5062 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5063 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5064 spin_unlock(&pers_lock);
d9d166c2
N
5065 if (mddev->level != LEVEL_NONE)
5066 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5067 mddev->level);
5068 else
5069 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5070 mddev->clevel);
1da177e4
LT
5071 return -EINVAL;
5072 }
2604b703 5073 mddev->pers = pers;
1da177e4 5074 spin_unlock(&pers_lock);
34817e8c
N
5075 if (mddev->level != pers->level) {
5076 mddev->level = pers->level;
5077 mddev->new_level = pers->level;
5078 }
d9d166c2 5079 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5080
f6705578 5081 if (mddev->reshape_position != MaxSector &&
63c70c4f 5082 pers->start_reshape == NULL) {
f6705578
N
5083 /* This personality cannot handle reshaping... */
5084 mddev->pers = NULL;
5085 module_put(pers->owner);
5086 return -EINVAL;
5087 }
5088
7dd5e7c3
N
5089 if (pers->sync_request) {
5090 /* Warn if this is a potentially silly
5091 * configuration.
5092 */
5093 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5094 struct md_rdev *rdev2;
7dd5e7c3 5095 int warned = 0;
159ec1fc 5096
dafb20fa
N
5097 rdev_for_each(rdev, mddev)
5098 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N
5099 if (rdev < rdev2 &&
5100 rdev->bdev->bd_contains ==
5101 rdev2->bdev->bd_contains) {
5102 printk(KERN_WARNING
5103 "%s: WARNING: %s appears to be"
5104 " on the same physical disk as"
5105 " %s.\n",
5106 mdname(mddev),
5107 bdevname(rdev->bdev,b),
5108 bdevname(rdev2->bdev,b2));
5109 warned = 1;
5110 }
5111 }
159ec1fc 5112
7dd5e7c3
N
5113 if (warned)
5114 printk(KERN_WARNING
5115 "True protection against single-disk"
5116 " failure might be compromised.\n");
5117 }
5118
657390d2 5119 mddev->recovery = 0;
58c0fed4
AN
5120 /* may be over-ridden by personality */
5121 mddev->resync_max_sectors = mddev->dev_sectors;
5122
6ff8d8ec 5123 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5124
0f9552b5 5125 if (start_readonly && mddev->ro == 0)
f91de92e
N
5126 mddev->ro = 2; /* read-only, but switch on first write */
5127
b15c2e57 5128 err = mddev->pers->run(mddev);
13e53df3
AN
5129 if (err)
5130 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW
5131 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5132 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5133 " but 'external_size' not in effect?\n", __func__);
5134 printk(KERN_ERR
5135 "md: invalid array_size %llu > default size %llu\n",
5136 (unsigned long long)mddev->array_sectors / 2,
5137 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5138 err = -EINVAL;
5139 mddev->pers->stop(mddev);
5140 }
ef99bf48
N
5141 if (err == 0 && mddev->pers->sync_request &&
5142 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
b15c2e57
N
5143 err = bitmap_create(mddev);
5144 if (err) {
5145 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5146 mdname(mddev), err);
5147 mddev->pers->stop(mddev);
5148 }
5149 }
1da177e4 5150 if (err) {
1da177e4
LT
5151 module_put(mddev->pers->owner);
5152 mddev->pers = NULL;
32a7627c
N
5153 bitmap_destroy(mddev);
5154 return err;
1da177e4 5155 }
5e55e2f5 5156 if (mddev->pers->sync_request) {
00bcb4ac
N
5157 if (mddev->kobj.sd &&
5158 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N
5159 printk(KERN_WARNING
5160 "md: cannot register extra attributes for %s\n",
5161 mdname(mddev));
00bcb4ac 5162 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5163 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N
5164 mddev->ro = 0;
5165
1da177e4 5166 atomic_set(&mddev->writes_pending,0);
1e50915f
RB
5167 atomic_set(&mddev->max_corr_read_errors,
5168 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT
5169 mddev->safemode = 0;
5170 mddev->safemode_timer.function = md_safemode_timeout;
5171 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5172 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5173 mddev->in_sync = 1;
0ca69886
N
5174 smp_wmb();
5175 mddev->ready = 1;
dafb20fa 5176 rdev_for_each(rdev, mddev)
36fad858
NK
5177 if (rdev->raid_disk >= 0)
5178 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5179 /* failure here is OK */;
1da177e4
LT
5180
5181 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5182
850b2b42
N
5183 if (mddev->flags)
5184 md_update_sb(mddev, 0);
1da177e4 5185
d7603b7e 5186 md_new_event(mddev);
00bcb4ac
N
5187 sysfs_notify_dirent_safe(mddev->sysfs_state);
5188 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5189 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT
5190 return 0;
5191}
390ee602 5192EXPORT_SYMBOL_GPL(md_run);
1da177e4 5193
fd01b88c 5194static int do_md_run(struct mddev *mddev)
fe60b014
N
5195{
5196 int err;
5197
5198 err = md_run(mddev);
5199 if (err)
5200 goto out;
69e51b44
N
5201 err = bitmap_load(mddev);
5202 if (err) {
5203 bitmap_destroy(mddev);
5204 goto out;
5205 }
0fd018af
JB
5206
5207 md_wakeup_thread(mddev->thread);
5208 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5209
fe60b014
N
5210 set_capacity(mddev->gendisk, mddev->array_sectors);
5211 revalidate_disk(mddev->gendisk);
f0b4f7e2 5212 mddev->changed = 1;
fe60b014
N
5213 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5214out:
5215 return err;
5216}
5217
fd01b88c 5218static int restart_array(struct mddev *mddev)
1da177e4
LT
5219{
5220 struct gendisk *disk = mddev->gendisk;
1da177e4 5221
80fab1d7 5222 /* Complain if it has no devices */
1da177e4 5223 if (list_empty(&mddev->disks))
80fab1d7
AN
5224 return -ENXIO;
5225 if (!mddev->pers)
5226 return -EINVAL;
5227 if (!mddev->ro)
5228 return -EBUSY;
5229 mddev->safemode = 0;
5230 mddev->ro = 0;
5231 set_disk_ro(disk, 0);
5232 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5233 mdname(mddev));
5234 /* Kick recovery or resync if necessary */
5235 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5236 md_wakeup_thread(mddev->thread);
5237 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5238 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5239 return 0;
1da177e4
LT
5240}
5241
acc55e22
N
5242/* similar to deny_write_access, but accounts for our holding a reference
5243 * to the file ourselves */
5244static int deny_bitmap_write_access(struct file * file)
5245{
5246 struct inode *inode = file->f_mapping->host;
5247
5248 spin_lock(&inode->i_lock);
5249 if (atomic_read(&inode->i_writecount) > 1) {
5250 spin_unlock(&inode->i_lock);
5251 return -ETXTBSY;
5252 }
5253 atomic_set(&inode->i_writecount, -1);
5254 spin_unlock(&inode->i_lock);
5255
5256 return 0;
5257}
5258
43a70507 5259void restore_bitmap_write_access(struct file *file)
acc55e22
N
5260{
5261 struct inode *inode = file->f_mapping->host;
5262
5263 spin_lock(&inode->i_lock);
5264 atomic_set(&inode->i_writecount, 1);
5265 spin_unlock(&inode->i_lock);
5266}
5267
fd01b88c 5268static void md_clean(struct mddev *mddev)
6177b472
N
5269{
5270 mddev->array_sectors = 0;
5271 mddev->external_size = 0;
5272 mddev->dev_sectors = 0;
5273 mddev->raid_disks = 0;
5274 mddev->recovery_cp = 0;
5275 mddev->resync_min = 0;
5276 mddev->resync_max = MaxSector;
5277 mddev->reshape_position = MaxSector;
5278 mddev->external = 0;
5279 mddev->persistent = 0;
5280 mddev->level = LEVEL_NONE;
5281 mddev->clevel[0] = 0;
5282 mddev->flags = 0;
5283 mddev->ro = 0;
5284 mddev->metadata_type[0] = 0;
5285 mddev->chunk_sectors = 0;
5286 mddev->ctime = mddev->utime = 0;
5287 mddev->layout = 0;
5288 mddev->max_disks = 0;
5289 mddev->events = 0;
a8707c08 5290 mddev->can_decrease_events = 0;
6177b472 5291 mddev->delta_disks = 0;
2c810cdd 5292 mddev->reshape_backwards = 0;
6177b472
N
5293 mddev->new_level = LEVEL_NONE;
5294 mddev->new_layout = 0;
5295 mddev->new_chunk_sectors = 0;
5296 mddev->curr_resync = 0;
5297 mddev->resync_mismatches = 0;
5298 mddev->suspend_lo = mddev->suspend_hi = 0;
5299 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5300 mddev->recovery = 0;
5301 mddev->in_sync = 0;
f0b4f7e2 5302 mddev->changed = 0;
6177b472 5303 mddev->degraded = 0;
6177b472 5304 mddev->safemode = 0;
050b6615 5305 mddev->merge_check_needed = 0;
6177b472
N
5306 mddev->bitmap_info.offset = 0;
5307 mddev->bitmap_info.default_offset = 0;
6409bb05 5308 mddev->bitmap_info.default_space = 0;
6177b472
N
5309 mddev->bitmap_info.chunksize = 0;
5310 mddev->bitmap_info.daemon_sleep = 0;
5311 mddev->bitmap_info.max_write_behind = 0;
5312}
5313
fd01b88c 5314static void __md_stop_writes(struct mddev *mddev)
a047e125
N
5315{
5316 if (mddev->sync_thread) {
5317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5318 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 5319 reap_sync_thread(mddev);
a047e125
N
5320 }
5321
5322 del_timer_sync(&mddev->safemode_timer);
5323
5324 bitmap_flush(mddev);
5325 md_super_wait(mddev);
5326
5327 if (!mddev->in_sync || mddev->flags) {
5328 /* mark array as shutdown cleanly */
5329 mddev->in_sync = 1;
5330 md_update_sb(mddev, 1);
5331 }
5332}
defad61a 5333
fd01b88c 5334void md_stop_writes(struct mddev *mddev)
defad61a
N
5335{
5336 mddev_lock(mddev);
5337 __md_stop_writes(mddev);
5338 mddev_unlock(mddev);
5339}
390ee602 5340EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5341
fd01b88c 5342void md_stop(struct mddev *mddev)
6177b472 5343{
0ca69886 5344 mddev->ready = 0;
6177b472
N
5345 mddev->pers->stop(mddev);
5346 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5347 mddev->to_remove = &md_redundancy_group;
5348 module_put(mddev->pers->owner);
5349 mddev->pers = NULL;
cca9cf90 5350 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5351}
390ee602 5352EXPORT_SYMBOL_GPL(md_stop);
6177b472 5353
fd01b88c 5354static int md_set_readonly(struct mddev *mddev, int is_open)
a4bd82d0
N
5355{
5356 int err = 0;
5357 mutex_lock(&mddev->open_mutex);
5358 if (atomic_read(&mddev->openers) > is_open) {
5359 printk("md: %s still in use.\n",mdname(mddev));
5360 err = -EBUSY;
5361 goto out;
5362 }
5363 if (mddev->pers) {
defad61a 5364 __md_stop_writes(mddev);
a4bd82d0
N
5365
5366 err = -ENXIO;
5367 if (mddev->ro==1)
5368 goto out;
5369 mddev->ro = 1;
5370 set_disk_ro(mddev->gendisk, 1);
5371 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5372 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N
5373 err = 0;
5374 }
5375out:
5376 mutex_unlock(&mddev->open_mutex);
5377 return err;
5378}
5379
9e653b63
N
5380/* mode:
5381 * 0 - completely stop and dis-assemble array
9e653b63
N
5382 * 2 - stop but do not disassemble array
5383 */
fd01b88c 5384static int do_md_stop(struct mddev * mddev, int mode, int is_open)
1da177e4 5385{
1da177e4 5386 struct gendisk *disk = mddev->gendisk;
3cb03002 5387 struct md_rdev *rdev;
1da177e4 5388
c8c00a69 5389 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N
5390 if (atomic_read(&mddev->openers) > is_open ||
5391 mddev->sysfs_active) {
df5b20cf 5392 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N
5393 mutex_unlock(&mddev->open_mutex);
5394 return -EBUSY;
5395 }
1da177e4 5396
6e17b027 5397 if (mddev->pers) {
a4bd82d0
N
5398 if (mddev->ro)
5399 set_disk_ro(disk, 0);
409c57f3 5400
defad61a 5401 __md_stop_writes(mddev);
a4bd82d0
N
5402 md_stop(mddev);
5403 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5404 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5405
a4bd82d0 5406 /* tell userspace to handle 'inactive' */
00bcb4ac 5407 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5408
dafb20fa 5409 rdev_for_each(rdev, mddev)
36fad858
NK
5410 if (rdev->raid_disk >= 0)
5411 sysfs_unlink_rdev(mddev, rdev);
c4647292 5412
a4bd82d0 5413 set_capacity(disk, 0);
6e17b027 5414 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5415 mddev->changed = 1;
a4bd82d0 5416 revalidate_disk(disk);
0d4ca600 5417
a4bd82d0
N
5418 if (mddev->ro)
5419 mddev->ro = 0;
6e17b027
N
5420 } else
5421 mutex_unlock(&mddev->open_mutex);
1da177e4
LT
5422 /*
5423 * Free resources if final stop
5424 */
9e653b63 5425 if (mode == 0) {
1da177e4
LT
5426 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5427
978f946b 5428 bitmap_destroy(mddev);
c3d9714e
N
5429 if (mddev->bitmap_info.file) {
5430 restore_bitmap_write_access(mddev->bitmap_info.file);
5431 fput(mddev->bitmap_info.file);
5432 mddev->bitmap_info.file = NULL;
978f946b 5433 }
c3d9714e 5434 mddev->bitmap_info.offset = 0;
978f946b 5435
1da177e4
LT
5436 export_array(mddev);
5437
6177b472 5438 md_clean(mddev);
934d9c23 5439 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N
5440 if (mddev->hold_active == UNTIL_STOP)
5441 mddev->hold_active = 0;
a4bd82d0 5442 }
3f9d99c1 5443 blk_integrity_unregister(disk);
d7603b7e 5444 md_new_event(mddev);
00bcb4ac 5445 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5446 return 0;
1da177e4
LT
5447}
5448
fdee8ae4 5449#ifndef MODULE
fd01b88c 5450static void autorun_array(struct mddev *mddev)
1da177e4 5451{
3cb03002 5452 struct md_rdev *rdev;
1da177e4
LT
5453 int err;
5454
a757e64c 5455 if (list_empty(&mddev->disks))
1da177e4 5456 return;
1da177e4
LT
5457
5458 printk(KERN_INFO "md: running: ");
5459
dafb20fa 5460 rdev_for_each(rdev, mddev) {
1da177e4
LT
5461 char b[BDEVNAME_SIZE];
5462 printk("<%s>", bdevname(rdev->bdev,b));
5463 }
5464 printk("\n");
5465
d710e138 5466 err = do_md_run(mddev);
1da177e4
LT
5467 if (err) {
5468 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 5469 do_md_stop(mddev, 0, 0);
1da177e4
LT
5470 }
5471}
5472
5473/*
5474 * lets try to run arrays based on all disks that have arrived
5475 * until now. (those are in pending_raid_disks)
5476 *
5477 * the method: pick the first pending disk, collect all disks with
5478 * the same UUID, remove all from the pending list and put them into
5479 * the 'same_array' list. Then order this list based on superblock
5480 * update time (freshest comes first), kick out 'old' disks and
5481 * compare superblocks. If everything's fine then run it.
5482 *
5483 * If "unit" is allocated, then bump its reference count
5484 */
5485static void autorun_devices(int part)
5486{
3cb03002 5487 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5488 struct mddev *mddev;
1da177e4
LT
5489 char b[BDEVNAME_SIZE];
5490
5491 printk(KERN_INFO "md: autorun ...\n");
5492 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5493 int unit;
1da177e4 5494 dev_t dev;
ad01c9e3 5495 LIST_HEAD(candidates);
1da177e4 5496 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5497 struct md_rdev, same_set);
1da177e4
LT
5498
5499 printk(KERN_INFO "md: considering %s ...\n",
5500 bdevname(rdev0->bdev,b));
5501 INIT_LIST_HEAD(&candidates);
159ec1fc 5502 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT
5503 if (super_90_load(rdev, rdev0, 0) >= 0) {
5504 printk(KERN_INFO "md: adding %s ...\n",
5505 bdevname(rdev->bdev,b));
5506 list_move(&rdev->same_set, &candidates);
5507 }
5508 /*
5509 * now we have a set of devices, with all of them having
5510 * mostly sane superblocks. It's time to allocate the
5511 * mddev.
5512 */
e8703fe1
N
5513 if (part) {
5514 dev = MKDEV(mdp_major,
5515 rdev0->preferred_minor << MdpMinorShift);
5516 unit = MINOR(dev) >> MdpMinorShift;
5517 } else {
5518 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5519 unit = MINOR(dev);
5520 }
5521 if (rdev0->preferred_minor != unit) {
1da177e4
LT
5522 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5523 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5524 break;
5525 }
1da177e4
LT
5526
5527 md_probe(dev, NULL, NULL);
5528 mddev = mddev_find(dev);
9bbbca3a
NB
5529 if (!mddev || !mddev->gendisk) {
5530 if (mddev)
5531 mddev_put(mddev);
5532 printk(KERN_ERR
1da177e4
LT
5533 "md: cannot allocate memory for md drive.\n");
5534 break;
5535 }
5536 if (mddev_lock(mddev))
5537 printk(KERN_WARNING "md: %s locked, cannot run\n",
5538 mdname(mddev));
5539 else if (mddev->raid_disks || mddev->major_version
5540 || !list_empty(&mddev->disks)) {
5541 printk(KERN_WARNING
5542 "md: %s already running, cannot run %s\n",
5543 mdname(mddev), bdevname(rdev0->bdev,b));
5544 mddev_unlock(mddev);
5545 } else {
5546 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5547 mddev->persistent = 1;
159ec1fc 5548 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT
5549 list_del_init(&rdev->same_set);
5550 if (bind_rdev_to_array(rdev, mddev))
5551 export_rdev(rdev);
5552 }
5553 autorun_array(mddev);
5554 mddev_unlock(mddev);
5555 }
5556 /* on success, candidates will be empty, on error
5557 * it won't...
5558 */
159ec1fc 5559 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5560 list_del_init(&rdev->same_set);
1da177e4 5561 export_rdev(rdev);
4b80991c 5562 }
1da177e4
LT
5563 mddev_put(mddev);
5564 }
5565 printk(KERN_INFO "md: ... autorun DONE.\n");
5566}
fdee8ae4 5567#endif /* !MODULE */
1da177e4 5568
1da177e4
LT
5569static int get_version(void __user * arg)
5570{
5571 mdu_version_t ver;
5572
5573 ver.major = MD_MAJOR_VERSION;
5574 ver.minor = MD_MINOR_VERSION;
5575 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5576
5577 if (copy_to_user(arg, &ver, sizeof(ver)))
5578 return -EFAULT;
5579
5580 return 0;
5581}
5582
fd01b88c 5583static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT
5584{
5585 mdu_array_info_t info;
a9f326eb 5586 int nr,working,insync,failed,spare;
3cb03002 5587 struct md_rdev *rdev;
1da177e4 5588
a9f326eb 5589 nr=working=insync=failed=spare=0;
dafb20fa 5590 rdev_for_each(rdev, mddev) {
1da177e4 5591 nr++;
b2d444d7 5592 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT
5593 failed++;
5594 else {
5595 working++;
b2d444d7 5596 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5597 insync++;
1da177e4
LT
5598 else
5599 spare++;
5600 }
5601 }
5602
5603 info.major_version = mddev->major_version;
5604 info.minor_version = mddev->minor_version;
5605 info.patch_version = MD_PATCHLEVEL_VERSION;
5606 info.ctime = mddev->ctime;
5607 info.level = mddev->level;
58c0fed4
AN
5608 info.size = mddev->dev_sectors / 2;
5609 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5610 info.size = -1;
1da177e4
LT
5611 info.nr_disks = nr;
5612 info.raid_disks = mddev->raid_disks;
5613 info.md_minor = mddev->md_minor;
5614 info.not_persistent= !mddev->persistent;
5615
5616 info.utime = mddev->utime;
5617 info.state = 0;
5618 if (mddev->in_sync)
5619 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5620 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5621 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5622 info.active_disks = insync;
1da177e4
LT
5623 info.working_disks = working;
5624 info.failed_disks = failed;
5625 info.spare_disks = spare;
5626
5627 info.layout = mddev->layout;
9d8f0363 5628 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT
5629
5630 if (copy_to_user(arg, &info, sizeof(info)))
5631 return -EFAULT;
5632
5633 return 0;
5634}
5635
fd01b88c 5636static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N
5637{
5638 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5639 char *ptr, *buf = NULL;
5640 int err = -ENOMEM;
5641
b5470dc5
DW
5642 if (md_allow_write(mddev))
5643 file = kmalloc(sizeof(*file), GFP_NOIO);
5644 else
5645 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5646
32a7627c
N
5647 if (!file)
5648 goto out;
5649
5650 /* bitmap disabled, zero the first byte and copy out */
1ec885cd 5651 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
32a7627c
N
5652 file->pathname[0] = '\0';
5653 goto copy_out;
5654 }
5655
5656 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5657 if (!buf)
5658 goto out;
5659
1ec885cd
N
5660 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5661 buf, sizeof(file->pathname));
6bcfd601 5662 if (IS_ERR(ptr))
32a7627c
N
5663 goto out;
5664
5665 strcpy(file->pathname, ptr);
5666
5667copy_out:
5668 err = 0;
5669 if (copy_to_user(arg, file, sizeof(*file)))
5670 err = -EFAULT;
5671out:
5672 kfree(buf);
5673 kfree(file);
5674 return err;
5675}
5676
fd01b88c 5677static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT
5678{
5679 mdu_disk_info_t info;
3cb03002 5680 struct md_rdev *rdev;
1da177e4
LT
5681
5682 if (copy_from_user(&info, arg, sizeof(info)))
5683 return -EFAULT;
5684
26ef379f 5685 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT
5686 if (rdev) {
5687 info.major = MAJOR(rdev->bdev->bd_dev);
5688 info.minor = MINOR(rdev->bdev->bd_dev);
5689 info.raid_disk = rdev->raid_disk;
5690 info.state = 0;
b2d444d7 5691 if (test_bit(Faulty, &rdev->flags))
1da177e4 5692 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5693 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT
5694 info.state |= (1<<MD_DISK_ACTIVE);
5695 info.state |= (1<<MD_DISK_SYNC);
5696 }
8ddf9efe
N
5697 if (test_bit(WriteMostly, &rdev->flags))
5698 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT
5699 } else {
5700 info.major = info.minor = 0;
5701 info.raid_disk = -1;
5702 info.state = (1<<MD_DISK_REMOVED);
5703 }
5704
5705 if (copy_to_user(arg, &info, sizeof(info)))
5706 return -EFAULT;
5707
5708 return 0;
5709}
5710
fd01b88c 5711static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT
5712{
5713 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5714 struct md_rdev *rdev;
1da177e4
LT
5715 dev_t dev = MKDEV(info->major,info->minor);
5716
5717 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5718 return -EOVERFLOW;
5719
5720 if (!mddev->raid_disks) {
5721 int err;
5722 /* expecting a device which has a superblock */
5723 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5724 if (IS_ERR(rdev)) {
5725 printk(KERN_WARNING
5726 "md: md_import_device returned %ld\n",
5727 PTR_ERR(rdev));
5728 return PTR_ERR(rdev);
5729 }
5730 if (!list_empty(&mddev->disks)) {
3cb03002
N
5731 struct md_rdev *rdev0
5732 = list_entry(mddev->disks.next,
5733 struct md_rdev, same_set);
a9f326eb 5734 err = super_types[mddev->major_version]
1da177e4
LT
5735 .load_super(rdev, rdev0, mddev->minor_version);
5736 if (err < 0) {
5737 printk(KERN_WARNING
5738 "md: %s has different UUID to %s\n",
5739 bdevname(rdev->bdev,b),
5740 bdevname(rdev0->bdev,b2));
5741 export_rdev(rdev);
5742 return -EINVAL;
5743 }
5744 }
5745 err = bind_rdev_to_array(rdev, mddev);
5746 if (err)
5747 export_rdev(rdev);
5748 return err;
5749 }
5750
5751 /*
5752 * add_new_disk can be used once the array is assembled
5753 * to add "hot spares". They must already have a superblock
5754 * written
5755 */
5756 if (mddev->pers) {
5757 int err;
5758 if (!mddev->pers->hot_add_disk) {
5759 printk(KERN_WARNING
5760 "%s: personality does not support diskops!\n",
5761 mdname(mddev));
5762 return -EINVAL;
5763 }
7b1e35f6
N
5764 if (mddev->persistent)
5765 rdev = md_import_device(dev, mddev->major_version,
5766 mddev->minor_version);
5767 else
5768 rdev = md_import_device(dev, -1, -1);
1da177e4
LT
5769 if (IS_ERR(rdev)) {
5770 printk(KERN_WARNING
5771 "md: md_import_device returned %ld\n",
5772 PTR_ERR(rdev));
5773 return PTR_ERR(rdev);
5774 }
1a855a06 5775 /* set saved_raid_disk if appropriate */
41158c7e
N
5776 if (!mddev->persistent) {
5777 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5778 info->raid_disk < mddev->raid_disks) {
41158c7e 5779 rdev->raid_disk = info->raid_disk;
bf572541
N
5780 set_bit(In_sync, &rdev->flags);
5781 } else
41158c7e
N
5782 rdev->raid_disk = -1;
5783 } else
5784 super_types[mddev->major_version].
5785 validate_super(mddev, rdev);
bedd86b7 5786 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5787 rdev->raid_disk != info->raid_disk) {
bedd86b7
N
5788 /* This was a hot-add request, but events doesn't
5789 * match, so reject it.
5790 */
5791 export_rdev(rdev);
5792 return -EINVAL;
5793 }
5794
1a855a06
N
5795 if (test_bit(In_sync, &rdev->flags))
5796 rdev->saved_raid_disk = rdev->raid_disk;
5797 else
5798 rdev->saved_raid_disk = -1;
41158c7e 5799
b2d444d7 5800 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N
5801 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5802 set_bit(WriteMostly, &rdev->flags);
575a80fa
N
5803 else
5804 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5805
1da177e4
LT
5806 rdev->raid_disk = -1;
5807 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N
5808 if (!err && !mddev->pers->hot_remove_disk) {
5809 /* If there is hot_add_disk but no hot_remove_disk
5810 * then added disks for geometry changes,
5811 * and should be added immediately.
5812 */
5813 super_types[mddev->major_version].
5814 validate_super(mddev, rdev);
5815 err = mddev->pers->hot_add_disk(mddev, rdev);
5816 if (err)
5817 unbind_rdev_from_array(rdev);
5818 }
1da177e4
LT
5819 if (err)
5820 export_rdev(rdev);
52664732 5821 else
00bcb4ac 5822 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5823
17571284 5824 md_update_sb(mddev, 1);
72a23c21
NB
5825 if (mddev->degraded)
5826 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5827 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5828 if (!err)
5829 md_new_event(mddev);
005eca5e 5830 md_wakeup_thread(mddev->thread);
1da177e4
LT
5831 return err;
5832 }
5833
5834 /* otherwise, add_new_disk is only allowed
5835 * for major_version==0 superblocks
5836 */
5837 if (mddev->major_version != 0) {
5838 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5839 mdname(mddev));
5840 return -EINVAL;
5841 }
5842
5843 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5844 int err;
d710e138 5845 rdev = md_import_device(dev, -1, 0);
1da177e4
LT
5846 if (IS_ERR(rdev)) {
5847 printk(KERN_WARNING
5848 "md: error, md_import_device() returned %ld\n",
5849 PTR_ERR(rdev));
5850 return PTR_ERR(rdev);
5851 }
5852 rdev->desc_nr = info->number;
5853 if (info->raid_disk < mddev->raid_disks)
5854 rdev->raid_disk = info->raid_disk;
5855 else
5856 rdev->raid_disk = -1;
5857
1da177e4 5858 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N
5859 if (info->state & (1<<MD_DISK_SYNC))
5860 set_bit(In_sync, &rdev->flags);
1da177e4 5861
8ddf9efe
N
5862 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5863 set_bit(WriteMostly, &rdev->flags);
5864
1da177e4
LT
5865 if (!mddev->persistent) {
5866 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS
5867 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5868 } else
57b2caa3 5869 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5870 rdev->sectors = rdev->sb_start;
1da177e4 5871
2bf071bf
N
5872 err = bind_rdev_to_array(rdev, mddev);
5873 if (err) {
5874 export_rdev(rdev);
5875 return err;
5876 }
1da177e4
LT
5877 }
5878
5879 return 0;
5880}
5881
fd01b88c 5882static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT
5883{
5884 char b[BDEVNAME_SIZE];
3cb03002 5885 struct md_rdev *rdev;
1da177e4 5886
1da177e4
LT
5887 rdev = find_rdev(mddev, dev);
5888 if (!rdev)
5889 return -ENXIO;
5890
5891 if (rdev->raid_disk >= 0)
5892 goto busy;
5893
5894 kick_rdev_from_array(rdev);
850b2b42 5895 md_update_sb(mddev, 1);
d7603b7e 5896 md_new_event(mddev);
1da177e4
LT
5897
5898 return 0;
5899busy:
fdefa4d8 5900 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT
5901 bdevname(rdev->bdev,b), mdname(mddev));
5902 return -EBUSY;
5903}
5904
fd01b88c 5905static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT
5906{
5907 char b[BDEVNAME_SIZE];
5908 int err;
3cb03002 5909 struct md_rdev *rdev;
1da177e4
LT
5910
5911 if (!mddev->pers)
5912 return -ENODEV;
5913
5914 if (mddev->major_version != 0) {
5915 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5916 " version-0 superblocks.\n",
5917 mdname(mddev));
5918 return -EINVAL;
5919 }
5920 if (!mddev->pers->hot_add_disk) {
5921 printk(KERN_WARNING
5922 "%s: personality does not support diskops!\n",
5923 mdname(mddev));
5924 return -EINVAL;
5925 }
5926
d710e138 5927 rdev = md_import_device(dev, -1, 0);
1da177e4
LT
5928 if (IS_ERR(rdev)) {
5929 printk(KERN_WARNING
5930 "md: error, md_import_device() returned %ld\n",
5931 PTR_ERR(rdev));
5932 return -EINVAL;
5933 }
5934
5935 if (mddev->persistent)
57b2caa3 5936 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5937 else
77304d2a 5938 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5939
8190e754 5940 rdev->sectors = rdev->sb_start;
1da177e4 5941
b2d444d7 5942 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT
5943 printk(KERN_WARNING
5944 "md: can not hot-add faulty %s disk to %s!\n",
5945 bdevname(rdev->bdev,b), mdname(mddev));
5946 err = -EINVAL;
5947 goto abort_export;
5948 }
b2d444d7 5949 clear_bit(In_sync, &rdev->flags);
1da177e4 5950 rdev->desc_nr = -1;
5842730d 5951 rdev->saved_raid_disk = -1;
2bf071bf
N
5952 err = bind_rdev_to_array(rdev, mddev);
5953 if (err)
5954 goto abort_export;
1da177e4
LT
5955
5956 /*
5957 * The rest should better be atomic, we can have disk failures
5958 * noticed in interrupt contexts ...
5959 */
5960
1da177e4
LT
5961 rdev->raid_disk = -1;
5962
850b2b42 5963 md_update_sb(mddev, 1);
1da177e4
LT
5964
5965 /*
5966 * Kick recovery, maybe this spare has to be added to the
5967 * array immediately.
5968 */
5969 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5970 md_wakeup_thread(mddev->thread);
d7603b7e 5971 md_new_event(mddev);
1da177e4
LT
5972 return 0;
5973
1da177e4
LT
5974abort_export:
5975 export_rdev(rdev);
5976 return err;
5977}
5978
fd01b88c 5979static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N
5980{
5981 int err;
5982
36fa3063
N
5983 if (mddev->pers) {
5984 if (!mddev->pers->quiesce)
5985 return -EBUSY;
5986 if (mddev->recovery || mddev->sync_thread)
5987 return -EBUSY;
5988 /* we should be able to change the bitmap.. */
5989 }
32a7627c 5990
32a7627c 5991
36fa3063
N
5992 if (fd >= 0) {
5993 if (mddev->bitmap)
5994 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5995 mddev->bitmap_info.file = fget(fd);
32a7627c 5996
c3d9714e 5997 if (mddev->bitmap_info.file == NULL) {
36fa3063
N
5998 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5999 mdname(mddev));
6000 return -EBADF;
6001 }
6002
c3d9714e 6003 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N
6004 if (err) {
6005 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6006 mdname(mddev));
c3d9714e
N
6007 fput(mddev->bitmap_info.file);
6008 mddev->bitmap_info.file = NULL;
36fa3063
N
6009 return err;
6010 }
c3d9714e 6011 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N
6012 } else if (mddev->bitmap == NULL)
6013 return -ENOENT; /* cannot remove what isn't there */
6014 err = 0;
6015 if (mddev->pers) {
6016 mddev->pers->quiesce(mddev, 1);
69e51b44 6017 if (fd >= 0) {
36fa3063 6018 err = bitmap_create(mddev);
69e51b44
N
6019 if (!err)
6020 err = bitmap_load(mddev);
6021 }
d7375ab3 6022 if (fd < 0 || err) {
36fa3063 6023 bitmap_destroy(mddev);
d7375ab3
N
6024 fd = -1; /* make sure to put the file */
6025 }
36fa3063 6026 mddev->pers->quiesce(mddev, 0);
d7375ab3
N
6027 }
6028 if (fd < 0) {
c3d9714e
N
6029 if (mddev->bitmap_info.file) {
6030 restore_bitmap_write_access(mddev->bitmap_info.file);
6031 fput(mddev->bitmap_info.file);
acc55e22 6032 }
c3d9714e 6033 mddev->bitmap_info.file = NULL;
36fa3063
N
6034 }
6035
32a7627c
N
6036 return err;
6037}
6038
1da177e4
LT
6039/*
6040 * set_array_info is used two different ways
6041 * The original usage is when creating a new array.
6042 * In this usage, raid_disks is > 0 and it together with
6043 * level, size, not_persistent,layout,chunksize determine the
6044 * shape of the array.
6045 * This will always create an array with a type-0.90.0 superblock.
6046 * The newer usage is when assembling an array.
6047 * In this case raid_disks will be 0, and the major_version field is
6048 * use to determine which style super-blocks are to be found on the devices.
6049 * The minor and patch _version numbers are also kept incase the
6050 * super_block handler wishes to interpret them.
6051 */
fd01b88c 6052static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT
6053{
6054
6055 if (info->raid_disks == 0) {
6056 /* just setting version number for superblock loading */
6057 if (info->major_version < 0 ||
50511da3 6058 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT
6059 super_types[info->major_version].name == NULL) {
6060 /* maybe try to auto-load a module? */
6061 printk(KERN_INFO
6062 "md: superblock version %d not known\n",
6063 info->major_version);
6064 return -EINVAL;
6065 }
6066 mddev->major_version = info->major_version;
6067 mddev->minor_version = info->minor_version;
6068 mddev->patch_version = info->patch_version;
3f9d7b0d 6069 mddev->persistent = !info->not_persistent;
cbd19983
N
6070 /* ensure mddev_put doesn't delete this now that there
6071 * is some minimal configuration.
6072 */
6073 mddev->ctime = get_seconds();
1da177e4
LT
6074 return 0;
6075 }
6076 mddev->major_version = MD_MAJOR_VERSION;
6077 mddev->minor_version = MD_MINOR_VERSION;
6078 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6079 mddev->ctime = get_seconds();
6080
6081 mddev->level = info->level;
17115e03 6082 mddev->clevel[0] = 0;
58c0fed4 6083 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT
6084 mddev->raid_disks = info->raid_disks;
6085 /* don't set md_minor, it is determined by which /dev/md* was
6086 * openned
6087 */
6088 if (info->state & (1<<MD_SB_CLEAN))
6089 mddev->recovery_cp = MaxSector;
6090 else
6091 mddev->recovery_cp = 0;
6092 mddev->persistent = ! info->not_persistent;
e691063a 6093 mddev->external = 0;
1da177e4
LT
6094
6095 mddev->layout = info->layout;
9d8f0363 6096 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT
6097
6098 mddev->max_disks = MD_SB_DISKS;
6099
e691063a
N
6100 if (mddev->persistent)
6101 mddev->flags = 0;
850b2b42 6102 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6103
c3d9714e 6104 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6105 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6106 mddev->bitmap_info.offset = 0;
b2a2703c 6107
f6705578
N
6108 mddev->reshape_position = MaxSector;
6109
1da177e4
LT
6110 /*
6111 * Generate a 128 bit UUID
6112 */
6113 get_random_bytes(mddev->uuid, 16);
6114
f6705578 6115 mddev->new_level = mddev->level;
664e7c41 6116 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N
6117 mddev->new_layout = mddev->layout;
6118 mddev->delta_disks = 0;
2c810cdd 6119 mddev->reshape_backwards = 0;
f6705578 6120
1da177e4
LT
6121 return 0;
6122}
6123
fd01b88c 6124void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6125{
b522adcd
DW
6126 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6127
6128 if (mddev->external_size)
6129 return;
6130
1f403624
DW
6131 mddev->array_sectors = array_sectors;
6132}
6133EXPORT_SYMBOL(md_set_array_sectors);
6134
fd01b88c 6135static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6136{
3cb03002 6137 struct md_rdev *rdev;
a35b0d69 6138 int rv;
d71f9f88 6139 int fit = (num_sectors == 0);
a35b0d69
N
6140
6141 if (mddev->pers->resize == NULL)
6142 return -EINVAL;
d71f9f88
AN
6143 /* The "num_sectors" is the number of sectors of each device that
6144 * is used. This can only make sense for arrays with redundancy.
6145 * linear and raid0 always use whatever space is available. We can only
6146 * consider changing this number if no resync or reconstruction is
6147 * happening, and if the new size is acceptable. It must fit before the
0f420358 6148 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN
6149 * of each device. If num_sectors is zero, we find the largest size
6150 * that fits.
a35b0d69
N
6151 */
6152 if (mddev->sync_thread)
6153 return -EBUSY;
a4a6125a 6154
dafb20fa 6155 rdev_for_each(rdev, mddev) {
dd8ac336 6156 sector_t avail = rdev->sectors;
01ab5662 6157
d71f9f88
AN
6158 if (fit && (num_sectors == 0 || num_sectors > avail))
6159 num_sectors = avail;
6160 if (avail < num_sectors)
a35b0d69
N
6161 return -ENOSPC;
6162 }
d71f9f88 6163 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N
6164 if (!rv)
6165 revalidate_disk(mddev->gendisk);
a35b0d69
N
6166 return rv;
6167}
6168
fd01b88c 6169static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N
6170{
6171 int rv;
c6563a8c 6172 struct md_rdev *rdev;
da943b99 6173 /* change the number of raid disks */
63c70c4f 6174 if (mddev->pers->check_reshape == NULL)
da943b99
N
6175 return -EINVAL;
6176 if (raid_disks <= 0 ||
233fca36 6177 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6178 return -EINVAL;
63c70c4f 6179 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 6180 return -EBUSY;
c6563a8c
N
6181
6182 rdev_for_each(rdev, mddev) {
6183 if (mddev->raid_disks < raid_disks &&
6184 rdev->data_offset < rdev->new_data_offset)
6185 return -EINVAL;
6186 if (mddev->raid_disks > raid_disks &&
6187 rdev->data_offset > rdev->new_data_offset)
6188 return -EINVAL;
6189 }
6190
63c70c4f 6191 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N
6192 if (mddev->delta_disks < 0)
6193 mddev->reshape_backwards = 1;
6194 else if (mddev->delta_disks > 0)
6195 mddev->reshape_backwards = 0;
63c70c4f
N
6196
6197 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6198 if (rv < 0) {
de171cb9 6199 mddev->delta_disks = 0;
2c810cdd
N
6200 mddev->reshape_backwards = 0;
6201 }
da943b99
N
6202 return rv;
6203}
6204
6205
1da177e4
LT
6206/*
6207 * update_array_info is used to change the configuration of an
6208 * on-line array.
6209 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6210 * fields in the info are checked against the array.
6211 * Any differences that cannot be handled will cause an error.
6212 * Normally, only one change can be managed at a time.
6213 */
fd01b88c 6214static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT
6215{
6216 int rv = 0;
6217 int cnt = 0;
36fa3063
N
6218 int state = 0;
6219
6220 /* calculate expected state,ignoring low bits */
c3d9714e 6221 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6222 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT
6223
6224 if (mddev->major_version != info->major_version ||
6225 mddev->minor_version != info->minor_version ||
6226/* mddev->patch_version != info->patch_version || */
6227 mddev->ctime != info->ctime ||
6228 mddev->level != info->level ||
6229/* mddev->layout != info->layout || */
6230 !mddev->persistent != info->not_persistent||
9d8f0363 6231 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N
6232 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6233 ((state^info->state) & 0xfffffe00)
6234 )
1da177e4
LT
6235 return -EINVAL;
6236 /* Check there is only one change */
58c0fed4
AN
6237 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6238 cnt++;
6239 if (mddev->raid_disks != info->raid_disks)
6240 cnt++;
6241 if (mddev->layout != info->layout)
6242 cnt++;
6243 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6244 cnt++;
6245 if (cnt == 0)
6246 return 0;
6247 if (cnt > 1)
6248 return -EINVAL;
1da177e4
LT
6249
6250 if (mddev->layout != info->layout) {
6251 /* Change layout
6252 * we don't need to do anything at the md level, the
6253 * personality will take care of it all.
6254 */
50ac168a 6255 if (mddev->pers->check_reshape == NULL)
1da177e4 6256 return -EINVAL;
597a711b
N
6257 else {
6258 mddev->new_layout = info->layout;
50ac168a 6259 rv = mddev->pers->check_reshape(mddev);
597a711b
N
6260 if (rv)
6261 mddev->new_layout = mddev->layout;
6262 return rv;
6263 }
1da177e4 6264 }
58c0fed4 6265 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6266 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6267
da943b99
N
6268 if (mddev->raid_disks != info->raid_disks)
6269 rv = update_raid_disks(mddev, info->raid_disks);
6270
36fa3063
N
6271 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6272 if (mddev->pers->quiesce == NULL)
6273 return -EINVAL;
6274 if (mddev->recovery || mddev->sync_thread)
6275 return -EBUSY;
6276 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6277 /* add the bitmap */
6278 if (mddev->bitmap)
6279 return -EEXIST;
c3d9714e 6280 if (mddev->bitmap_info.default_offset == 0)
36fa3063 6281 return -EINVAL;
c3d9714e
N
6282 mddev->bitmap_info.offset =
6283 mddev->bitmap_info.default_offset;
6409bb05
N
6284 mddev->bitmap_info.space =
6285 mddev->bitmap_info.default_space;
36fa3063
N
6286 mddev->pers->quiesce(mddev, 1);
6287 rv = bitmap_create(mddev);
69e51b44
N
6288 if (!rv)
6289 rv = bitmap_load(mddev);
36fa3063
N
6290 if (rv)
6291 bitmap_destroy(mddev);
6292 mddev->pers->quiesce(mddev, 0);
6293 } else {
6294 /* remove the bitmap */
6295 if (!mddev->bitmap)
6296 return -ENOENT;
1ec885cd 6297 if (mddev->bitmap->storage.file)
36fa3063
N
6298 return -EINVAL;
6299 mddev->pers->quiesce(mddev, 1);
6300 bitmap_destroy(mddev);
6301 mddev->pers->quiesce(mddev, 0);
c3d9714e 6302 mddev->bitmap_info.offset = 0;
36fa3063
N
6303 }
6304 }
850b2b42 6305 md_update_sb(mddev, 1);
1da177e4
LT
6306 return rv;
6307}
6308
fd01b88c 6309static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6310{
3cb03002 6311 struct md_rdev *rdev;
1da177e4
LT
6312
6313 if (mddev->pers == NULL)
6314 return -ENODEV;
6315
6316 rdev = find_rdev(mddev, dev);
6317 if (!rdev)
6318 return -ENODEV;
6319
6320 md_error(mddev, rdev);
5ef56c8f
N
6321 if (!test_bit(Faulty, &rdev->flags))
6322 return -EBUSY;
1da177e4
LT
6323 return 0;
6324}
6325
2f9618ce
AN
6326/*
6327 * We have a problem here : there is no easy way to give a CHS
6328 * virtual geometry. We currently pretend that we have a 2 heads
6329 * 4 sectors (with a BIG number of cylinders...). This drives
6330 * dosfs just mad... ;-)
6331 */
a885c8c4
CH
6332static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6333{
fd01b88c 6334 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH
6335
6336 geo->heads = 2;
6337 geo->sectors = 4;
49ce6cea 6338 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH
6339 return 0;
6340}
6341
a39907fa 6342static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT
6343 unsigned int cmd, unsigned long arg)
6344{
6345 int err = 0;
6346 void __user *argp = (void __user *)arg;
fd01b88c 6347 struct mddev *mddev = NULL;
e2218350 6348 int ro;
1da177e4 6349
506c9e44
N
6350 switch (cmd) {
6351 case RAID_VERSION:
6352 case GET_ARRAY_INFO:
6353 case GET_DISK_INFO:
6354 break;
6355 default:
6356 if (!capable(CAP_SYS_ADMIN))
6357 return -EACCES;
6358 }
1da177e4
LT
6359
6360 /*
6361 * Commands dealing with the RAID driver but not any
6362 * particular array:
6363 */
6364 switch (cmd)
6365 {
6366 case RAID_VERSION:
6367 err = get_version(argp);
6368 goto done;
6369
6370 case PRINT_RAID_DEBUG:
6371 err = 0;
6372 md_print_devices();
6373 goto done;
6374
6375#ifndef MODULE
6376 case RAID_AUTORUN:
6377 err = 0;
6378 autostart_arrays(arg);
6379 goto done;
6380#endif
6381 default:;
6382 }
6383
6384 /*
6385 * Commands creating/starting a new array:
6386 */
6387
a39907fa 6388 mddev = bdev->bd_disk->private_data;
1da177e4
LT
6389
6390 if (!mddev) {
6391 BUG();
6392 goto abort;
6393 }
6394
1da177e4
LT
6395 err = mddev_lock(mddev);
6396 if (err) {
6397 printk(KERN_INFO
6398 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6399 err, cmd);
6400 goto abort;
6401 }
6402
6403 switch (cmd)
6404 {
6405 case SET_ARRAY_INFO:
6406 {
6407 mdu_array_info_t info;
6408 if (!arg)
6409 memset(&info, 0, sizeof(info));
6410 else if (copy_from_user(&info, argp, sizeof(info))) {
6411 err = -EFAULT;
6412 goto abort_unlock;
6413 }
6414 if (mddev->pers) {
6415 err = update_array_info(mddev, &info);
6416 if (err) {
6417 printk(KERN_WARNING "md: couldn't update"
6418 " array info. %d\n", err);
6419 goto abort_unlock;
6420 }
6421 goto done_unlock;
6422 }
6423 if (!list_empty(&mddev->disks)) {
6424 printk(KERN_WARNING
6425 "md: array %s already has disks!\n",
6426 mdname(mddev));
6427 err = -EBUSY;
6428 goto abort_unlock;
6429 }
6430 if (mddev->raid_disks) {
6431 printk(KERN_WARNING
6432 "md: array %s already initialised!\n",
6433 mdname(mddev));
6434 err = -EBUSY;
6435 goto abort_unlock;
6436 }
6437 err = set_array_info(mddev, &info);
6438 if (err) {
6439 printk(KERN_WARNING "md: couldn't set"
6440 " array info. %d\n", err);
6441 goto abort_unlock;
6442 }
6443 }
6444 goto done_unlock;
6445
6446 default:;
6447 }
6448
6449 /*
6450 * Commands querying/configuring an existing array:
6451 */
32a7627c 6452 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6453 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N
6454 if ((!mddev->raid_disks && !mddev->external)
6455 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6456 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6457 && cmd != GET_BITMAP_FILE) {
1da177e4
LT
6458 err = -ENODEV;
6459 goto abort_unlock;
6460 }
6461
6462 /*
6463 * Commands even a read-only array can execute:
6464 */
6465 switch (cmd)
6466 {
6467 case GET_ARRAY_INFO:
6468 err = get_array_info(mddev, argp);
6469 goto done_unlock;
6470
32a7627c 6471 case GET_BITMAP_FILE:
87162a28 6472 err = get_bitmap_file(mddev, argp);
32a7627c
N
6473 goto done_unlock;
6474
1da177e4
LT
6475 case GET_DISK_INFO:
6476 err = get_disk_info(mddev, argp);
6477 goto done_unlock;
6478
6479 case RESTART_ARRAY_RW:
6480 err = restart_array(mddev);
6481 goto done_unlock;
6482
6483 case STOP_ARRAY:
d710e138 6484 err = do_md_stop(mddev, 0, 1);
1da177e4
LT
6485 goto done_unlock;
6486
6487 case STOP_ARRAY_RO:
a4bd82d0 6488 err = md_set_readonly(mddev, 1);
1da177e4
LT
6489 goto done_unlock;
6490
e2218350
DW
6491 case BLKROSET:
6492 if (get_user(ro, (int __user *)(arg))) {
6493 err = -EFAULT;
6494 goto done_unlock;
6495 }
6496 err = -EINVAL;
6497
6498 /* if the bdev is going readonly the value of mddev->ro
6499 * does not matter, no writes are coming
6500 */
6501 if (ro)
6502 goto done_unlock;
6503
6504 /* are we are already prepared for writes? */
6505 if (mddev->ro != 1)
6506 goto done_unlock;
6507
6508 /* transitioning to readauto need only happen for
6509 * arrays that call md_write_start
6510 */
6511 if (mddev->pers) {
6512 err = restart_array(mddev);
6513 if (err == 0) {
6514 mddev->ro = 2;
6515 set_disk_ro(mddev->gendisk, 0);
6516 }
6517 }
6518 goto done_unlock;
1da177e4
LT
6519 }
6520
6521 /*
6522 * The remaining ioctls are changing the state of the
f91de92e
N
6523 * superblock, so we do not allow them on read-only arrays.
6524 * However non-MD ioctls (e.g. get-size) will still come through
6525 * here and hit the 'default' below, so only disallow
6526 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6527 */
bb57fc64 6528 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N
6529 if (mddev->ro == 2) {
6530 mddev->ro = 0;
00bcb4ac 6531 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB
6532 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6533 md_wakeup_thread(mddev->thread);
f91de92e
N
6534 } else {
6535 err = -EROFS;
6536 goto abort_unlock;
6537 }
1da177e4
LT
6538 }
6539
6540 switch (cmd)
6541 {
6542 case ADD_NEW_DISK:
6543 {
6544 mdu_disk_info_t info;
6545 if (copy_from_user(&info, argp, sizeof(info)))
6546 err = -EFAULT;
6547 else
6548 err = add_new_disk(mddev, &info);
6549 goto done_unlock;
6550 }
6551
6552 case HOT_REMOVE_DISK:
6553 err = hot_remove_disk(mddev, new_decode_dev(arg));
6554 goto done_unlock;
6555
6556 case HOT_ADD_DISK:
6557 err = hot_add_disk(mddev, new_decode_dev(arg));
6558 goto done_unlock;
6559
6560 case SET_DISK_FAULTY:
6561 err = set_disk_faulty(mddev, new_decode_dev(arg));
6562 goto done_unlock;
6563
6564 case RUN_ARRAY:
d710e138 6565 err = do_md_run(mddev);
1da177e4
LT
6566 goto done_unlock;
6567
32a7627c
N
6568 case SET_BITMAP_FILE:
6569 err = set_bitmap_file(mddev, (int)arg);
6570 goto done_unlock;
6571
1da177e4 6572 default:
1da177e4
LT
6573 err = -EINVAL;
6574 goto abort_unlock;
6575 }
6576
6577done_unlock:
6578abort_unlock:
d3374825
N
6579 if (mddev->hold_active == UNTIL_IOCTL &&
6580 err != -EINVAL)
6581 mddev->hold_active = 0;
1da177e4
LT
6582 mddev_unlock(mddev);
6583
6584 return err;
6585done:
6586 if (err)
6587 MD_BUG();
6588abort:
6589 return err;
6590}
aa98aa31
AB
6591#ifdef CONFIG_COMPAT
6592static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6593 unsigned int cmd, unsigned long arg)
6594{
6595 switch (cmd) {
6596 case HOT_REMOVE_DISK:
6597 case HOT_ADD_DISK:
6598 case SET_DISK_FAULTY:
6599 case SET_BITMAP_FILE:
6600 /* These take in integer arg, do not convert */
6601 break;
6602 default:
6603 arg = (unsigned long)compat_ptr(arg);
6604 break;
6605 }
6606
6607 return md_ioctl(bdev, mode, cmd, arg);
6608}
6609#endif /* CONFIG_COMPAT */
1da177e4 6610
a39907fa 6611static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT
6612{
6613 /*
6614 * Succeed if we can lock the mddev, which confirms that
6615 * it isn't being stopped right now.
6616 */
fd01b88c 6617 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT
6618 int err;
6619
0c098220
YL
6620 if (!mddev)
6621 return -ENODEV;
6622
d3374825
N
6623 if (mddev->gendisk != bdev->bd_disk) {
6624 /* we are racing with mddev_put which is discarding this
6625 * bd_disk.
6626 */
6627 mddev_put(mddev);
6628 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6629 flush_workqueue(md_misc_wq);
d3374825
N
6630 /* Then retry the open from the top */
6631 return -ERESTARTSYS;
6632 }
6633 BUG_ON(mddev != bdev->bd_disk->private_data);
6634
c8c00a69 6635 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT
6636 goto out;
6637
6638 err = 0;
f2ea68cf 6639 atomic_inc(&mddev->openers);
c8c00a69 6640 mutex_unlock(&mddev->open_mutex);
1da177e4 6641
f0b4f7e2 6642 check_disk_change(bdev);
1da177e4
LT
6643 out:
6644 return err;
6645}
6646
a39907fa 6647static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6648{
fd01b88c 6649 struct mddev *mddev = disk->private_data;
1da177e4 6650
52e5f9d1 6651 BUG_ON(!mddev);
f2ea68cf 6652 atomic_dec(&mddev->openers);
1da177e4
LT
6653 mddev_put(mddev);
6654
6655 return 0;
6656}
f0b4f7e2
N
6657
6658static int md_media_changed(struct gendisk *disk)
6659{
fd01b88c 6660 struct mddev *mddev = disk->private_data;
f0b4f7e2
N
6661
6662 return mddev->changed;
6663}
6664
6665static int md_revalidate(struct gendisk *disk)
6666{
fd01b88c 6667 struct mddev *mddev = disk->private_data;
f0b4f7e2
N
6668
6669 mddev->changed = 0;
6670 return 0;
6671}
83d5cde4 6672static const struct block_device_operations md_fops =
1da177e4
LT
6673{
6674 .owner = THIS_MODULE,
a39907fa
AV
6675 .open = md_open,
6676 .release = md_release,
b492b852 6677 .ioctl = md_ioctl,
aa98aa31
AB
6678#ifdef CONFIG_COMPAT
6679 .compat_ioctl = md_compat_ioctl,
6680#endif
a885c8c4 6681 .getgeo = md_getgeo,
f0b4f7e2
N
6682 .media_changed = md_media_changed,
6683 .revalidate_disk= md_revalidate,
1da177e4
LT
6684};
6685
75c96f85 6686static int md_thread(void * arg)
1da177e4 6687{
2b8bf345 6688 struct md_thread *thread = arg;
1da177e4 6689
1da177e4
LT
6690 /*
6691 * md_thread is a 'system-thread', it's priority should be very
6692 * high. We avoid resource deadlocks individually in each
6693 * raid personality. (RAID5 does preallocation) We also use RR and
6694 * the very same RT priority as kswapd, thus we will never get
6695 * into a priority inversion deadlock.
6696 *
6697 * we definitely have to have equal or higher priority than
6698 * bdflush, otherwise bdflush will deadlock if there are too
6699 * many dirty RAID5 blocks.
6700 */
1da177e4 6701
6985c43f 6702 allow_signal(SIGKILL);
a6fb0934 6703 while (!kthread_should_stop()) {
1da177e4 6704
93588e22
N
6705 /* We need to wait INTERRUPTIBLE so that
6706 * we don't add to the load-average.
6707 * That means we need to be sure no signals are
6708 * pending
6709 */
6710 if (signal_pending(current))
6711 flush_signals(current);
6712
6713 wait_event_interruptible_timeout
6714 (thread->wqueue,
6715 test_bit(THREAD_WAKEUP, &thread->flags)
6716 || kthread_should_stop(),
6717 thread->timeout);
1da177e4 6718
6c987910
N
6719 clear_bit(THREAD_WAKEUP, &thread->flags);
6720 if (!kthread_should_stop())
589a594b 6721 thread->run(thread->mddev);
1da177e4 6722 }
a6fb0934 6723
1da177e4
LT
6724 return 0;
6725}
6726
2b8bf345 6727void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT
6728{
6729 if (thread) {
36a4e1fe 6730 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT
6731 set_bit(THREAD_WAKEUP, &thread->flags);
6732 wake_up(&thread->wqueue);
6733 }
6734}
6735
2b8bf345 6736struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
1da177e4
LT
6737 const char *name)
6738{
2b8bf345 6739 struct md_thread *thread;
1da177e4 6740
2b8bf345 6741 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT
6742 if (!thread)
6743 return NULL;
6744
1da177e4
LT
6745 init_waitqueue_head(&thread->wqueue);
6746
1da177e4
LT
6747 thread->run = run;
6748 thread->mddev = mddev;
32a7627c 6749 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N
6750 thread->tsk = kthread_run(md_thread, thread,
6751 "%s_%s",
6752 mdname(thread->mddev),
0232605d 6753 name);
a6fb0934 6754 if (IS_ERR(thread->tsk)) {
1da177e4
LT
6755 kfree(thread);
6756 return NULL;
6757 }
1da177e4
LT
6758 return thread;
6759}
6760
2b8bf345 6761void md_unregister_thread(struct md_thread **threadp)
1da177e4 6762{
2b8bf345 6763 struct md_thread *thread = *threadp;
e0cf8f04
N
6764 if (!thread)
6765 return;
36a4e1fe 6766 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N
6767 /* Locking ensures that mddev_unlock does not wake_up a
6768 * non-existent thread
6769 */
6770 spin_lock(&pers_lock);
6771 *threadp = NULL;
6772 spin_unlock(&pers_lock);
a6fb0934
N
6773
6774 kthread_stop(thread->tsk);
1da177e4
LT
6775 kfree(thread);
6776}
6777
fd01b88c 6778void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
6779{
6780 if (!mddev) {
6781 MD_BUG();
6782 return;
6783 }
6784
b2d444d7 6785 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6786 return;
6bfe0b49 6787
de393cde 6788 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT
6789 return;
6790 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB
6791 if (mddev->degraded)
6792 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6793 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT
6794 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6795 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6796 md_wakeup_thread(mddev->thread);
768a418d 6797 if (mddev->event_work.func)
e804ac78 6798 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6799 md_new_event_inintr(mddev);
1da177e4
LT
6800}
6801
6802/* seq_file implementation /proc/mdstat */
6803
6804static void status_unused(struct seq_file *seq)
6805{
6806 int i = 0;
3cb03002 6807 struct md_rdev *rdev;
1da177e4
LT
6808
6809 seq_printf(seq, "unused devices: ");
6810
159ec1fc 6811 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT
6812 char b[BDEVNAME_SIZE];
6813 i++;
6814 seq_printf(seq, "%s ",
6815 bdevname(rdev->bdev,b));
6816 }
6817 if (!i)
6818 seq_printf(seq, "<none>");
6819
6820 seq_printf(seq, "\n");
6821}
6822
6823
fd01b88c 6824static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6825{
dd71cf6b
N
6826 sector_t max_sectors, resync, res;
6827 unsigned long dt, db;
6828 sector_t rt;
4588b42e
N
6829 int scale;
6830 unsigned int per_milli;
1da177e4 6831
dd71cf6b 6832 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4 6833
c804cdec
N
6834 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6835 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 6836 max_sectors = mddev->resync_max_sectors;
1da177e4 6837 else
dd71cf6b 6838 max_sectors = mddev->dev_sectors;
1da177e4
LT
6839
6840 /*
6841 * Should not happen.
6842 */
dd71cf6b 6843 if (!max_sectors) {
1da177e4
LT
6844 MD_BUG();
6845 return;
6846 }
4588b42e 6847 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6848 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N
6849 * u32, as those are the requirements for sector_div.
6850 * Thus 'scale' must be at least 10
6851 */
6852 scale = 10;
6853 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6854 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N
6855 scale++;
6856 }
6857 res = (resync>>scale)*1000;
dd71cf6b 6858 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N
6859
6860 per_milli = res;
1da177e4 6861 {
4588b42e 6862 int i, x = per_milli/50, y = 20-x;
1da177e4
LT
6863 seq_printf(seq, "[");
6864 for (i = 0; i < x; i++)
6865 seq_printf(seq, "=");
6866 seq_printf(seq, ">");
6867 for (i = 0; i < y; i++)
6868 seq_printf(seq, ".");
6869 seq_printf(seq, "] ");
6870 }
4588b42e 6871 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N
6872 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6873 "reshape" :
61df9d91
N
6874 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6875 "check" :
6876 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6877 "resync" : "recovery"))),
6878 per_milli/10, per_milli % 10,
dd71cf6b
N
6879 (unsigned long long) resync/2,
6880 (unsigned long long) max_sectors/2);
1da177e4
LT
6881
6882 /*
1da177e4
LT
6883 * dt: time from mark until now
6884 * db: blocks written from mark until now
6885 * rt: remaining time
dd71cf6b
N
6886 *
6887 * rt is a sector_t, so could be 32bit or 64bit.
6888 * So we divide before multiply in case it is 32bit and close
6889 * to the limit.
25985edc 6890 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N
6891 * near the end of resync when the number of remaining sectors
6892 * is close to 'db'.
6893 * We then divide rt by 32 after multiplying by db to compensate.
6894 * The '+1' avoids division by zero if db is very small.
1da177e4
LT
6895 */
6896 dt = ((jiffies - mddev->resync_mark) / HZ);
6897 if (!dt) dt++;
ff4e8d9a
N
6898 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6899 - mddev->resync_mark_cnt;
1da177e4 6900
dd71cf6b
N
6901 rt = max_sectors - resync; /* number of remaining sectors */
6902 sector_div(rt, db/32+1);
6903 rt *= dt;
6904 rt >>= 5;
6905
6906 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6907 ((unsigned long)rt % 60)/6);
1da177e4 6908
ff4e8d9a 6909 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT
6910}
6911
6912static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6913{
6914 struct list_head *tmp;
6915 loff_t l = *pos;
fd01b88c 6916 struct mddev *mddev;
1da177e4
LT
6917
6918 if (l >= 0x10000)
6919 return NULL;
6920 if (!l--)
6921 /* header */
6922 return (void*)1;
6923
6924 spin_lock(&all_mddevs_lock);
6925 list_for_each(tmp,&all_mddevs)
6926 if (!l--) {
fd01b88c 6927 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT
6928 mddev_get(mddev);
6929 spin_unlock(&all_mddevs_lock);
6930 return mddev;
6931 }
6932 spin_unlock(&all_mddevs_lock);
6933 if (!l--)
6934 return (void*)2;/* tail */
6935 return NULL;
6936}
6937
6938static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6939{
6940 struct list_head *tmp;
fd01b88c 6941 struct mddev *next_mddev, *mddev = v;
1da177e4
LT
6942
6943 ++*pos;
6944 if (v == (void*)2)
6945 return NULL;
6946
6947 spin_lock(&all_mddevs_lock);
6948 if (v == (void*)1)
6949 tmp = all_mddevs.next;
6950 else
6951 tmp = mddev->all_mddevs.next;
6952 if (tmp != &all_mddevs)
fd01b88c 6953 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT
6954 else {
6955 next_mddev = (void*)2;
6956 *pos = 0x10000;
6957 }
6958 spin_unlock(&all_mddevs_lock);
6959
6960 if (v != (void*)1)
6961 mddev_put(mddev);
6962 return next_mddev;
6963
6964}
6965
6966static void md_seq_stop(struct seq_file *seq, void *v)
6967{
fd01b88c 6968 struct mddev *mddev = v;
1da177e4
LT
6969
6970 if (mddev && v != (void*)1 && v != (void*)2)
6971 mddev_put(mddev);
6972}
6973
6974static int md_seq_show(struct seq_file *seq, void *v)
6975{
fd01b88c 6976 struct mddev *mddev = v;
dd8ac336 6977 sector_t sectors;
3cb03002 6978 struct md_rdev *rdev;
1da177e4
LT
6979
6980 if (v == (void*)1) {
84fc4b56 6981 struct md_personality *pers;
1da177e4
LT
6982 seq_printf(seq, "Personalities : ");
6983 spin_lock(&pers_lock);
2604b703
N
6984 list_for_each_entry(pers, &pers_list, list)
6985 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT
6986
6987 spin_unlock(&pers_lock);
6988 seq_printf(seq, "\n");
f1514638 6989 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT
6990 return 0;
6991 }
6992 if (v == (void*)2) {
6993 status_unused(seq);
6994 return 0;
6995 }
6996
5dc5cf7d 6997 if (mddev_lock(mddev) < 0)
1da177e4 6998 return -EINTR;
5dc5cf7d 6999
1da177e4
LT
7000 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7001 seq_printf(seq, "%s : %sactive", mdname(mddev),
7002 mddev->pers ? "" : "in");
7003 if (mddev->pers) {
f91de92e 7004 if (mddev->ro==1)
1da177e4 7005 seq_printf(seq, " (read-only)");
f91de92e 7006 if (mddev->ro==2)
52720ae7 7007 seq_printf(seq, " (auto-read-only)");
1da177e4
LT
7008 seq_printf(seq, " %s", mddev->pers->name);
7009 }
7010
dd8ac336 7011 sectors = 0;
dafb20fa 7012 rdev_for_each(rdev, mddev) {
1da177e4
LT
7013 char b[BDEVNAME_SIZE];
7014 seq_printf(seq, " %s[%d]",
7015 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N
7016 if (test_bit(WriteMostly, &rdev->flags))
7017 seq_printf(seq, "(W)");
b2d444d7 7018 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT
7019 seq_printf(seq, "(F)");
7020 continue;
2d78f8c4
N
7021 }
7022 if (rdev->raid_disk < 0)
b325a32e 7023 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N
7024 if (test_bit(Replacement, &rdev->flags))
7025 seq_printf(seq, "(R)");
dd8ac336 7026 sectors += rdev->sectors;
1da177e4
LT
7027 }
7028
7029 if (!list_empty(&mddev->disks)) {
7030 if (mddev->pers)
7031 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN
7032 (unsigned long long)
7033 mddev->array_sectors / 2);
1da177e4
LT
7034 else
7035 seq_printf(seq, "\n %llu blocks",
dd8ac336 7036 (unsigned long long)sectors / 2);
1da177e4 7037 }
1cd6bf19
N
7038 if (mddev->persistent) {
7039 if (mddev->major_version != 0 ||
7040 mddev->minor_version != 90) {
7041 seq_printf(seq," super %d.%d",
7042 mddev->major_version,
7043 mddev->minor_version);
7044 }
e691063a
N
7045 } else if (mddev->external)
7046 seq_printf(seq, " super external:%s",
7047 mddev->metadata_type);
7048 else
1cd6bf19 7049 seq_printf(seq, " super non-persistent");
1da177e4
LT
7050
7051 if (mddev->pers) {
d710e138 7052 mddev->pers->status(seq, mddev);
1da177e4 7053 seq_printf(seq, "\n ");
8e1b39d6
N
7054 if (mddev->pers->sync_request) {
7055 if (mddev->curr_resync > 2) {
d710e138 7056 status_resync(seq, mddev);
8e1b39d6
N
7057 seq_printf(seq, "\n ");
7058 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
7059 seq_printf(seq, "\tresync=DELAYED\n ");
7060 else if (mddev->recovery_cp < MaxSector)
7061 seq_printf(seq, "\tresync=PENDING\n ");
7062 }
32a7627c
N
7063 } else
7064 seq_printf(seq, "\n ");
7065
57148964 7066 bitmap_status(seq, mddev->bitmap);
1da177e4
LT
7067
7068 seq_printf(seq, "\n");
7069 }
7070 mddev_unlock(mddev);
7071
7072 return 0;
7073}
7074
110518bc 7075static const struct seq_operations md_seq_ops = {
1da177e4
LT
7076 .start = md_seq_start,
7077 .next = md_seq_next,
7078 .stop = md_seq_stop,
7079 .show = md_seq_show,
7080};
7081
7082static int md_seq_open(struct inode *inode, struct file *file)
7083{
f1514638 7084 struct seq_file *seq;
1da177e4
LT
7085 int error;
7086
7087 error = seq_open(file, &md_seq_ops);
d7603b7e 7088 if (error)
f1514638
KS
7089 return error;
7090
7091 seq = file->private_data;
7092 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT
7093 return error;
7094}
7095
d7603b7e
N
7096static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7097{
f1514638 7098 struct seq_file *seq = filp->private_data;
d7603b7e
N
7099 int mask;
7100
7101 poll_wait(filp, &md_event_waiters, wait);
7102
7103 /* always allow read */
7104 mask = POLLIN | POLLRDNORM;
7105
f1514638 7106 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N
7107 mask |= POLLERR | POLLPRI;
7108 return mask;
7109}
7110
fa027c2a 7111static const struct file_operations md_seq_fops = {
e24650c2 7112 .owner = THIS_MODULE,
1da177e4
LT
7113 .open = md_seq_open,
7114 .read = seq_read,
7115 .llseek = seq_lseek,
c3f94b40 7116 .release = seq_release_private,
d7603b7e 7117 .poll = mdstat_poll,
1da177e4
LT
7118};
7119
84fc4b56 7120int register_md_personality(struct md_personality *p)
1da177e4 7121{
1da177e4 7122 spin_lock(&pers_lock);
2604b703
N
7123 list_add_tail(&p->list, &pers_list);
7124 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT
7125 spin_unlock(&pers_lock);
7126 return 0;
7127}
7128
84fc4b56 7129int unregister_md_personality(struct md_personality *p)
1da177e4 7130{
2604b703 7131 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7132 spin_lock(&pers_lock);
2604b703 7133 list_del_init(&p->list);
1da177e4
LT
7134 spin_unlock(&pers_lock);
7135 return 0;
7136}
7137
fd01b88c 7138static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7139{
3cb03002 7140 struct md_rdev * rdev;
1da177e4 7141 int idle;
eea1bf38 7142 int curr_events;
1da177e4
LT
7143
7144 idle = 1;
4b80991c
N
7145 rcu_read_lock();
7146 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7147 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N
7148 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7149 (int)part_stat_read(&disk->part0, sectors[1]) -
7150 atomic_read(&disk->sync_io);
713f6ab1
N
7151 /* sync IO will cause sync_io to increase before the disk_stats
7152 * as sync_io is counted when a request starts, and
7153 * disk_stats is counted when it completes.
7154 * So resync activity will cause curr_events to be smaller than
7155 * when there was no such activity.
7156 * non-sync IO will cause disk_stat to increase without
7157 * increasing sync_io so curr_events will (eventually)
7158 * be larger than it was before. Once it becomes
7159 * substantially larger, the test below will cause
7160 * the array to appear non-idle, and resync will slow
7161 * down.
7162 * If there is a lot of outstanding resync activity when
7163 * we set last_event to curr_events, then all that activity
7164 * completing might cause the array to appear non-idle
7165 * and resync will be slowed down even though there might
7166 * not have been non-resync activity. This will only
7167 * happen once though. 'last_events' will soon reflect
7168 * the state where there is little or no outstanding
7169 * resync requests, and further resync activity will
7170 * always make curr_events less than last_events.
c0e48521 7171 *
1da177e4 7172 */
eea1bf38 7173 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT
7174 rdev->last_events = curr_events;
7175 idle = 0;
7176 }
7177 }
4b80991c 7178 rcu_read_unlock();
1da177e4
LT
7179 return idle;
7180}
7181
fd01b88c 7182void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT
7183{
7184 /* another "blocks" (512byte) blocks have been synced */
7185 atomic_sub(blocks, &mddev->recovery_active);
7186 wake_up(&mddev->recovery_wait);
7187 if (!ok) {
dfc70645 7188 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT
7189 md_wakeup_thread(mddev->thread);
7190 // stop recovery, signal do_sync ....
7191 }
7192}
7193
7194
06d91a5f
N
7195/* md_write_start(mddev, bi)
7196 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N
7197 * in superblock) before writing, schedule a superblock update
7198 * and wait for it to complete.
06d91a5f 7199 */
fd01b88c 7200void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7201{
0fd62b86 7202 int did_change = 0;
06d91a5f 7203 if (bio_data_dir(bi) != WRITE)
3d310eb7 7204 return;
06d91a5f 7205
f91de92e
N
7206 BUG_ON(mddev->ro == 1);
7207 if (mddev->ro == 2) {
7208 /* need to switch to read/write */
7209 mddev->ro = 0;
7210 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7211 md_wakeup_thread(mddev->thread);
25156198 7212 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7213 did_change = 1;
f91de92e 7214 }
06d91a5f 7215 atomic_inc(&mddev->writes_pending);
31a59e34
N
7216 if (mddev->safemode == 1)
7217 mddev->safemode = 0;
06d91a5f 7218 if (mddev->in_sync) {
a9701a30 7219 spin_lock_irq(&mddev->write_lock);
3d310eb7
N
7220 if (mddev->in_sync) {
7221 mddev->in_sync = 0;
850b2b42 7222 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7223 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7224 md_wakeup_thread(mddev->thread);
0fd62b86 7225 did_change = 1;
3d310eb7 7226 }
a9701a30 7227 spin_unlock_irq(&mddev->write_lock);
06d91a5f 7228 }
0fd62b86 7229 if (did_change)
00bcb4ac 7230 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7231 wait_event(mddev->sb_wait,
09a44cc1 7232 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT
7233}
7234
fd01b88c 7235void md_write_end(struct mddev *mddev)
1da177e4
LT
7236{
7237 if (atomic_dec_and_test(&mddev->writes_pending)) {
7238 if (mddev->safemode == 2)
7239 md_wakeup_thread(mddev->thread);
16f17b39 7240 else if (mddev->safemode_delay)
1da177e4
LT
7241 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7242 }
7243}
7244
2a2275d6
N
7245/* md_allow_write(mddev)
7246 * Calling this ensures that the array is marked 'active' so that writes
7247 * may proceed without blocking. It is important to call this before
7248 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7249 * Must be called with mddev_lock held.
b5470dc5
DW
7250 *
7251 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7252 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7253 */
fd01b88c 7254int md_allow_write(struct mddev *mddev)
2a2275d6
N
7255{
7256 if (!mddev->pers)
b5470dc5 7257 return 0;
2a2275d6 7258 if (mddev->ro)
b5470dc5 7259 return 0;
1a0fd497 7260 if (!mddev->pers->sync_request)
b5470dc5 7261 return 0;
2a2275d6
N
7262
7263 spin_lock_irq(&mddev->write_lock);
7264 if (mddev->in_sync) {
7265 mddev->in_sync = 0;
7266 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7267 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N
7268 if (mddev->safemode_delay &&
7269 mddev->safemode == 0)
7270 mddev->safemode = 1;
7271 spin_unlock_irq(&mddev->write_lock);
7272 md_update_sb(mddev, 0);
00bcb4ac 7273 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N
7274 } else
7275 spin_unlock_irq(&mddev->write_lock);
b5470dc5 7276
070dc6dd 7277 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW
7278 return -EAGAIN;
7279 else
7280 return 0;
2a2275d6
N
7281}
7282EXPORT_SYMBOL_GPL(md_allow_write);
7283
1da177e4
LT
7284#define SYNC_MARKS 10
7285#define SYNC_MARK_STEP (3*HZ)
fd01b88c 7286void md_do_sync(struct mddev *mddev)
1da177e4 7287{
fd01b88c 7288 struct mddev *mddev2;
1da177e4
LT
7289 unsigned int currspeed = 0,
7290 window;
57afd89f 7291 sector_t max_sectors,j, io_sectors;
1da177e4
LT
7292 unsigned long mark[SYNC_MARKS];
7293 sector_t mark_cnt[SYNC_MARKS];
7294 int last_mark,m;
7295 struct list_head *tmp;
7296 sector_t last_check;
57afd89f 7297 int skipped = 0;
3cb03002 7298 struct md_rdev *rdev;
61df9d91 7299 char *desc;
7c2c57c9 7300 struct blk_plug plug;
1da177e4
LT
7301
7302 /* just incase thread restarts... */
7303 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7304 return;
5fd6c1dc
N
7305 if (mddev->ro) /* never try to sync a read-only array */
7306 return;
1da177e4 7307
61df9d91
N
7308 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7309 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7310 desc = "data-check";
7311 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7312 desc = "requested-resync";
7313 else
7314 desc = "resync";
7315 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7316 desc = "reshape";
7317 else
7318 desc = "recovery";
7319
1da177e4
LT
7320 /* we overload curr_resync somewhat here.
7321 * 0 == not engaged in resync at all
7322 * 2 == checking that there is no conflict with another sync
7323 * 1 == like 2, but have yielded to allow conflicting resync to
7324 * commense
7325 * other == active in resync - this many blocks
7326 *
7327 * Before starting a resync we must have set curr_resync to
7328 * 2, and then checked that every "conflicting" array has curr_resync
7329 * less than ours. When we find one that is the same or higher
7330 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7331 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7332 * This will mean we have to start checking from the beginning again.
7333 *
7334 */
7335
7336 do {
7337 mddev->curr_resync = 2;
7338
7339 try_again:
404e4b43 7340 if (kthread_should_stop())
6985c43f 7341 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N
7342
7343 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7344 goto skip;
29ac4aa3 7345 for_each_mddev(mddev2, tmp) {
1da177e4
LT
7346 if (mddev2 == mddev)
7347 continue;
90b08710
BS
7348 if (!mddev->parallel_resync
7349 && mddev2->curr_resync
7350 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT
7351 DEFINE_WAIT(wq);
7352 if (mddev < mddev2 && mddev->curr_resync == 2) {
7353 /* arbitrarily yield */
7354 mddev->curr_resync = 1;
7355 wake_up(&resync_wait);
7356 }
7357 if (mddev > mddev2 && mddev->curr_resync == 1)
7358 /* no need to wait here, we can wait the next
7359 * time 'round when curr_resync == 2
7360 */
7361 continue;
9744197c
N
7362 /* We need to wait 'interruptible' so as not to
7363 * contribute to the load average, and not to
7364 * be caught by 'softlockup'
7365 */
7366 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7367 if (!kthread_should_stop() &&
8712e553 7368 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N
7369 printk(KERN_INFO "md: delaying %s of %s"
7370 " until %s has finished (they"
1da177e4 7371 " share one or more physical units)\n",
61df9d91 7372 desc, mdname(mddev), mdname(mddev2));
1da177e4 7373 mddev_put(mddev2);
9744197c
N
7374 if (signal_pending(current))
7375 flush_signals(current);
1da177e4
LT
7376 schedule();
7377 finish_wait(&resync_wait, &wq);
7378 goto try_again;
7379 }
7380 finish_wait(&resync_wait, &wq);
7381 }
7382 }
7383 } while (mddev->curr_resync < 2);
7384
5fd6c1dc 7385 j = 0;
9d88883e 7386 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7387 /* resync follows the size requested by the personality,
57afd89f 7388 * which defaults to physical size, but can be virtual size
1da177e4
LT
7389 */
7390 max_sectors = mddev->resync_max_sectors;
9d88883e 7391 mddev->resync_mismatches = 0;
5fd6c1dc 7392 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB
7393 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7394 j = mddev->resync_min;
7395 else if (!mddev->bitmap)
5fd6c1dc 7396 j = mddev->recovery_cp;
5e96ee65 7397
ccfcc3c1 7398 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7399 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7400 else {
1da177e4 7401 /* recovery follows the physical size of devices */
58c0fed4 7402 max_sectors = mddev->dev_sectors;
5fd6c1dc 7403 j = MaxSector;
4e59ca7d 7404 rcu_read_lock();
dafb20fa 7405 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N
7406 if (rdev->raid_disk >= 0 &&
7407 !test_bit(Faulty, &rdev->flags) &&
7408 !test_bit(In_sync, &rdev->flags) &&
7409 rdev->recovery_offset < j)
7410 j = rdev->recovery_offset;
4e59ca7d 7411 rcu_read_unlock();
5fd6c1dc 7412 }
1da177e4 7413
61df9d91
N
7414 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7415 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7416 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7417 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N
7418 "(but not more than %d KB/sec) for %s.\n",
7419 speed_max(mddev), desc);
1da177e4 7420
eea1bf38 7421 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7422
57afd89f 7423 io_sectors = 0;
1da177e4
LT
7424 for (m = 0; m < SYNC_MARKS; m++) {
7425 mark[m] = jiffies;
57afd89f 7426 mark_cnt[m] = io_sectors;
1da177e4
LT
7427 }
7428 last_mark = 0;
7429 mddev->resync_mark = mark[last_mark];
7430 mddev->resync_mark_cnt = mark_cnt[last_mark];
7431
7432 /*
7433 * Tune reconstruction:
7434 */
7435 window = 32*(PAGE_SIZE/512);
ac42450c
JB
7436 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7437 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT
7438
7439 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT
7440 last_check = 0;
7441
7442 if (j>2) {
7443 printk(KERN_INFO
61df9d91
N
7444 "md: resuming %s of %s from checkpoint.\n",
7445 desc, mdname(mddev));
1da177e4
LT
7446 mddev->curr_resync = j;
7447 }
75d3da43 7448 mddev->curr_resync_completed = j;
1da177e4 7449
7c2c57c9 7450 blk_start_plug(&plug);
1da177e4 7451 while (j < max_sectors) {
57afd89f 7452 sector_t sectors;
1da177e4 7453
57afd89f 7454 skipped = 0;
97e4f42d 7455
7a91ee1f
N
7456 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7457 ((mddev->curr_resync > mddev->curr_resync_completed &&
7458 (mddev->curr_resync - mddev->curr_resync_completed)
7459 > (max_sectors >> 4)) ||
7460 (j - mddev->curr_resync_completed)*2
7461 >= mddev->resync_max - mddev->curr_resync_completed
7462 )) {
97e4f42d 7463 /* time to update curr_resync_completed */
97e4f42d
N
7464 wait_event(mddev->recovery_wait,
7465 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7466 mddev->curr_resync_completed = j;
070dc6dd 7467 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7468 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7469 }
acb180b0 7470
e62e58a5
N
7471 while (j >= mddev->resync_max && !kthread_should_stop()) {
7472 /* As this condition is controlled by user-space,
7473 * we can block indefinitely, so use '_interruptible'
7474 * to avoid triggering warnings.
7475 */
7476 flush_signals(current); /* just in case */
7477 wait_event_interruptible(mddev->recovery_wait,
7478 mddev->resync_max > j
7479 || kthread_should_stop());
7480 }
acb180b0
N
7481
7482 if (kthread_should_stop())
7483 goto interrupted;
7484
57afd89f 7485 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7486 currspeed < speed_min(mddev));
57afd89f 7487 if (sectors == 0) {
dfc70645 7488 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT
7489 goto out;
7490 }
57afd89f
N
7491
7492 if (!skipped) { /* actual IO requested */
7493 io_sectors += sectors;
7494 atomic_add(sectors, &mddev->recovery_active);
7495 }
7496
e875ecea
N
7497 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7498 break;
7499
1da177e4
LT
7500 j += sectors;
7501 if (j>1) mddev->curr_resync = j;
ff4e8d9a 7502 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7503 if (last_check == 0)
e875ecea 7504 /* this is the earliest that rebuild will be
d7603b7e
N
7505 * visible in /proc/mdstat
7506 */
7507 md_new_event(mddev);
57afd89f
N
7508
7509 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT
7510 continue;
7511
57afd89f 7512 last_check = io_sectors;
1da177e4
LT
7513 repeat:
7514 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7515 /* step marks */
7516 int next = (last_mark+1) % SYNC_MARKS;
7517
7518 mddev->resync_mark = mark[next];
7519 mddev->resync_mark_cnt = mark_cnt[next];
7520 mark[next] = jiffies;
57afd89f 7521 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT
7522 last_mark = next;
7523 }
7524
7525
c6207277
N
7526 if (kthread_should_stop())
7527 goto interrupted;
7528
1da177e4
LT
7529
7530 /*
7531 * this loop exits only if either when we are slower than
7532 * the 'hard' speed limit, or the system was IO-idle for
7533 * a jiffy.
7534 * the system might be non-idle CPU-wise, but we only care
7535 * about not overloading the IO subsystem. (things like an
7536 * e2fsck being done on the RAID array should execute fast)
7537 */
1da177e4
LT
7538 cond_resched();
7539
57afd89f
N
7540 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7541 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7542
88202a0c
N
7543 if (currspeed > speed_min(mddev)) {
7544 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7545 !is_mddev_idle(mddev, 0)) {
c0e48521 7546 msleep(500);
1da177e4
LT
7547 goto repeat;
7548 }
7549 }
7550 }
61df9d91 7551 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT
7552 /*
7553 * this also signals 'finished resyncing' to md_stop
7554 */
7555 out:
7c2c57c9 7556 blk_finish_plug(&plug);
1da177e4
LT
7557 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7558
7559 /* tell personality that we are finished */
57afd89f 7560 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7561
dfc70645 7562 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N
7563 mddev->curr_resync > 2) {
7564 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7565 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7566 if (mddev->curr_resync >= mddev->recovery_cp) {
7567 printk(KERN_INFO
61df9d91
N
7568 "md: checkpointing %s of %s.\n",
7569 desc, mdname(mddev));
db91ff55
N
7570 mddev->recovery_cp =
7571 mddev->curr_resync_completed;
5fd6c1dc
N
7572 }
7573 } else
7574 mddev->recovery_cp = MaxSector;
7575 } else {
7576 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7577 mddev->curr_resync = MaxSector;
4e59ca7d 7578 rcu_read_lock();
dafb20fa 7579 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7580 if (rdev->raid_disk >= 0 &&
70fffd0b 7581 mddev->delta_disks >= 0 &&
5fd6c1dc
N
7582 !test_bit(Faulty, &rdev->flags) &&
7583 !test_bit(In_sync, &rdev->flags) &&
7584 rdev->recovery_offset < mddev->curr_resync)
7585 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7586 rcu_read_unlock();
5fd6c1dc 7587 }
1da177e4 7588 }
db91ff55 7589 skip:
17571284 7590 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7591
c07b70ad
N
7592 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7593 /* We completed so min/max setting can be forgotten if used. */
7594 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7595 mddev->resync_min = 0;
7596 mddev->resync_max = MaxSector;
7597 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7598 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT
7599 mddev->curr_resync = 0;
7600 wake_up(&resync_wait);
7601 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7602 md_wakeup_thread(mddev->thread);
c6207277
N
7603 return;
7604
7605 interrupted:
7606 /*
7607 * got a signal, exit.
7608 */
7609 printk(KERN_INFO
7610 "md: md_do_sync() got signal ... exiting\n");
7611 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7612 goto out;
7613
1da177e4 7614}
29269553 7615EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7616
fd01b88c 7617static int remove_and_add_spares(struct mddev *mddev)
b4c4c7b8 7618{
3cb03002 7619 struct md_rdev *rdev;
b4c4c7b8 7620 int spares = 0;
f2a371c5 7621 int removed = 0;
b4c4c7b8 7622
97e4f42d
N
7623 mddev->curr_resync_completed = 0;
7624
dafb20fa 7625 rdev_for_each(rdev, mddev)
b4c4c7b8 7626 if (rdev->raid_disk >= 0 &&
6bfe0b49 7627 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N
7628 (test_bit(Faulty, &rdev->flags) ||
7629 ! test_bit(In_sync, &rdev->flags)) &&
7630 atomic_read(&rdev->nr_pending)==0) {
7631 if (mddev->pers->hot_remove_disk(
b8321b68 7632 mddev, rdev) == 0) {
36fad858 7633 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7634 rdev->raid_disk = -1;
f2a371c5 7635 removed++;
b4c4c7b8
N
7636 }
7637 }
f2a371c5
N
7638 if (removed)
7639 sysfs_notify(&mddev->kobj, NULL,
7640 "degraded");
7641
b4c4c7b8 7642
dafb20fa 7643 rdev_for_each(rdev, mddev) {
7bfec5f3
N
7644 if (rdev->raid_disk >= 0 &&
7645 !test_bit(In_sync, &rdev->flags) &&
7646 !test_bit(Faulty, &rdev->flags))
7647 spares++;
7648 if (rdev->raid_disk < 0
7649 && !test_bit(Faulty, &rdev->flags)) {
7650 rdev->recovery_offset = 0;
7651 if (mddev->pers->
7652 hot_add_disk(mddev, rdev) == 0) {
7653 if (sysfs_link_rdev(mddev, rdev))
7654 /* failure here is OK */;
dfc70645 7655 spares++;
7bfec5f3
N
7656 md_new_event(mddev);
7657 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8 7658 }
dfc70645 7659 }
b4c4c7b8
N
7660 }
7661 return spares;
7662}
7ebc0be7 7663
fd01b88c 7664static void reap_sync_thread(struct mddev *mddev)
7ebc0be7 7665{
3cb03002 7666 struct md_rdev *rdev;
7ebc0be7
N
7667
7668 /* resync has finished, collect result */
01f96c0a 7669 md_unregister_thread(&mddev->sync_thread);
7ebc0be7
N
7670 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7671 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7672 /* success...*/
7673 /* activate any spares */
7674 if (mddev->pers->spare_active(mddev))
7675 sysfs_notify(&mddev->kobj, NULL,
7676 "degraded");
7677 }
7678 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7679 mddev->pers->finish_reshape)
7680 mddev->pers->finish_reshape(mddev);
7ebc0be7 7681
d70ed2e4
AW
7682 /* If array is no-longer degraded, then any saved_raid_disk
7683 * information must be scrapped. Also if any device is now
7684 * In_sync we must scrape the saved_raid_disk for that device
7685 * do the superblock for an incrementally recovered device
7686 * written out.
7ebc0be7 7687 */
dafb20fa 7688 rdev_for_each(rdev, mddev)
d70ed2e4
AW
7689 if (!mddev->degraded ||
7690 test_bit(In_sync, &rdev->flags))
7ebc0be7
N
7691 rdev->saved_raid_disk = -1;
7692
d70ed2e4 7693 md_update_sb(mddev, 1);
7ebc0be7
N
7694 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7695 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7696 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7697 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7698 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7699 /* flag recovery needed just to double check */
7700 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7701 sysfs_notify_dirent_safe(mddev->sysfs_action);
7702 md_new_event(mddev);
768e587e
JB
7703 if (mddev->event_work.func)
7704 queue_work(md_misc_wq, &mddev->event_work);
7ebc0be7
N
7705}
7706
1da177e4
LT
7707/*
7708 * This routine is regularly called by all per-raid-array threads to
7709 * deal with generic issues like resync and super-block update.
7710 * Raid personalities that don't have a thread (linear/raid0) do not
7711 * need this as they never do any recovery or update the superblock.
7712 *
7713 * It does not do any resync itself, but rather "forks" off other threads
7714 * to do that as needed.
7715 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7716 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7717 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT
7718 * and wakeups up this thread which will reap the thread and finish up.
7719 * This thread also removes any faulty devices (with nr_pending == 0).
7720 *
7721 * The overall approach is:
7722 * 1/ if the superblock needs updating, update it.
7723 * 2/ If a recovery thread is running, don't do anything else.
7724 * 3/ If recovery has finished, clean up, possibly marking spares active.
7725 * 4/ If there are any faulty devices, remove them.
7726 * 5/ If array is degraded, try to add spares devices
7727 * 6/ If array has spares or is not in-sync, start a resync thread.
7728 */
fd01b88c 7729void md_check_recovery(struct mddev *mddev)
1da177e4 7730{
68866e42
JB
7731 if (mddev->suspended)
7732 return;
7733
5f40402d 7734 if (mddev->bitmap)
aa5cbd10 7735 bitmap_daemon_work(mddev);
1da177e4 7736
fca4d848 7737 if (signal_pending(current)) {
31a59e34 7738 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N
7739 printk(KERN_INFO "md: %s in immediate safe mode\n",
7740 mdname(mddev));
7741 mddev->safemode = 2;
7742 }
7743 flush_signals(current);
7744 }
7745
c89a8eee
N
7746 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7747 return;
1da177e4 7748 if ( ! (
126925c0 7749 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7750 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7751 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7752 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N
7753 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7754 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT
7755 ))
7756 return;
fca4d848 7757
df5b89b3 7758 if (mddev_trylock(mddev)) {
b4c4c7b8 7759 int spares = 0;
fca4d848 7760
c89a8eee
N
7761 if (mddev->ro) {
7762 /* Only thing we do on a ro array is remove
7763 * failed devices.
7764 */
3cb03002 7765 struct md_rdev *rdev;
dafb20fa 7766 rdev_for_each(rdev, mddev)
a8c42c7f
N
7767 if (rdev->raid_disk >= 0 &&
7768 !test_bit(Blocked, &rdev->flags) &&
7769 test_bit(Faulty, &rdev->flags) &&
7770 atomic_read(&rdev->nr_pending)==0) {
7771 if (mddev->pers->hot_remove_disk(
b8321b68 7772 mddev, rdev) == 0) {
36fad858 7773 sysfs_unlink_rdev(mddev, rdev);
a8c42c7f
N
7774 rdev->raid_disk = -1;
7775 }
7776 }
c89a8eee
N
7777 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7778 goto unlock;
7779 }
7780
31a59e34 7781 if (!mddev->external) {
0fd62b86 7782 int did_change = 0;
31a59e34
N
7783 spin_lock_irq(&mddev->write_lock);
7784 if (mddev->safemode &&
7785 !atomic_read(&mddev->writes_pending) &&
7786 !mddev->in_sync &&
7787 mddev->recovery_cp == MaxSector) {
7788 mddev->in_sync = 1;
0fd62b86 7789 did_change = 1;
070dc6dd 7790 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N
7791 }
7792 if (mddev->safemode == 1)
7793 mddev->safemode = 0;
7794 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7795 if (did_change)
00bcb4ac 7796 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7797 }
fca4d848 7798
850b2b42
N
7799 if (mddev->flags)
7800 md_update_sb(mddev, 0);
06d91a5f 7801
1da177e4
LT
7802 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7803 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7804 /* resync/recovery still happening */
7805 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7806 goto unlock;
7807 }
7808 if (mddev->sync_thread) {
7ebc0be7 7809 reap_sync_thread(mddev);
1da177e4
LT
7810 goto unlock;
7811 }
72a23c21
NB
7812 /* Set RUNNING before clearing NEEDED to avoid
7813 * any transients in the value of "sync_action".
7814 */
7815 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
24dd469d
N
7816 /* Clear some bits that don't mean anything, but
7817 * might be left set
7818 */
24dd469d
N
7819 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7820 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7821
ed209584
N
7822 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7823 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5fd6c1dc 7824 goto unlock;
1da177e4
LT
7825 /* no recovery is running.
7826 * remove any failed drives, then
7827 * add spares if possible.
7828 * Spare are also removed and re-added, to allow
7829 * the personality to fail the re-add.
7830 */
1da177e4 7831
b4c4c7b8 7832 if (mddev->reshape_position != MaxSector) {
50ac168a
N
7833 if (mddev->pers->check_reshape == NULL ||
7834 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N
7835 /* Cannot proceed */
7836 goto unlock;
7837 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7838 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7839 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N
7840 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7841 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7842 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7843 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N
7844 } else if (mddev->recovery_cp < MaxSector) {
7845 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7846 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N
7847 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7848 /* nothing to be done ... */
1da177e4 7849 goto unlock;
24dd469d 7850
1da177e4 7851 if (mddev->pers->sync_request) {
ef99bf48 7852 if (spares) {
a654b9d8
N
7853 /* We are adding a device or devices to an array
7854 * which has the bitmap stored on all devices.
7855 * So make sure all bitmap pages get written
7856 */
7857 bitmap_write_all(mddev->bitmap);
7858 }
1da177e4
LT
7859 mddev->sync_thread = md_register_thread(md_do_sync,
7860 mddev,
0da3c619 7861 "resync");
1da177e4
LT
7862 if (!mddev->sync_thread) {
7863 printk(KERN_ERR "%s: could not start resync"
7864 " thread...\n",
7865 mdname(mddev));
7866 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N
7867 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7868 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7869 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7870 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7871 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7872 } else
1da177e4 7873 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7874 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7875 md_new_event(mddev);
1da177e4
LT
7876 }
7877 unlock:
72a23c21
NB
7878 if (!mddev->sync_thread) {
7879 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7880 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7881 &mddev->recovery))
0c3573f1 7882 if (mddev->sysfs_action)
00bcb4ac 7883 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7884 }
1da177e4
LT
7885 mddev_unlock(mddev);
7886 }
7887}
7888
fd01b88c 7889void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7890{
00bcb4ac 7891 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7892 wait_event_timeout(rdev->blocked_wait,
de393cde
N
7893 !test_bit(Blocked, &rdev->flags) &&
7894 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW
7895 msecs_to_jiffies(5000));
7896 rdev_dec_pending(rdev, mddev);
7897}
7898EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7899
c6563a8c
N
7900void md_finish_reshape(struct mddev *mddev)
7901{
7902 /* called be personality module when reshape completes. */
7903 struct md_rdev *rdev;
7904
7905 rdev_for_each(rdev, mddev) {
7906 if (rdev->data_offset > rdev->new_data_offset)
7907 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7908 else
7909 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7910 rdev->data_offset = rdev->new_data_offset;
7911 }
7912}
7913EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N
7914
7915/* Bad block management.
7916 * We can record which blocks on each device are 'bad' and so just
7917 * fail those blocks, or that stripe, rather than the whole device.
7918 * Entries in the bad-block table are 64bits wide. This comprises:
7919 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7920 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7921 * A 'shift' can be set so that larger blocks are tracked and
7922 * consequently larger devices can be covered.
7923 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7924 *
7925 * Locking of the bad-block table uses a seqlock so md_is_badblock
7926 * might need to retry if it is very unlucky.
7927 * We will sometimes want to check for bad blocks in a bi_end_io function,
7928 * so we use the write_seqlock_irq variant.
7929 *
7930 * When looking for a bad block we specify a range and want to
7931 * know if any block in the range is bad. So we binary-search
7932 * to the last range that starts at-or-before the given endpoint,
7933 * (or "before the sector after the target range")
7934 * then see if it ends after the given start.
7935 * We return
7936 * 0 if there are no known bad blocks in the range
7937 * 1 if there are known bad block which are all acknowledged
7938 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7939 * plus the start/length of the first bad section we overlap.
7940 */
7941int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7942 sector_t *first_bad, int *bad_sectors)
7943{
7944 int hi;
7945 int lo = 0;
7946 u64 *p = bb->page;
7947 int rv = 0;
7948 sector_t target = s + sectors;
7949 unsigned seq;
7950
7951 if (bb->shift > 0) {
7952 /* round the start down, and the end up */
7953 s >>= bb->shift;
7954 target += (1<<bb->shift) - 1;
7955 target >>= bb->shift;
7956 sectors = target - s;
7957 }
7958 /* 'target' is now the first block after the bad range */
7959
7960retry:
7961 seq = read_seqbegin(&bb->lock);
7962
7963 hi = bb->count;
7964
7965 /* Binary search between lo and hi for 'target'
7966 * i.e. for the last range that starts before 'target'
7967 */
7968 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7969 * are known not to be the last range before target.
7970 * VARIANT: hi-lo is the number of possible
7971 * ranges, and decreases until it reaches 1
7972 */
7973 while (hi - lo > 1) {
7974 int mid = (lo + hi) / 2;
7975 sector_t a = BB_OFFSET(p[mid]);
7976 if (a < target)
7977 /* This could still be the one, earlier ranges
7978 * could not. */
7979 lo = mid;
7980 else
7981 /* This and later ranges are definitely out. */
7982 hi = mid;
7983 }
7984 /* 'lo' might be the last that started before target, but 'hi' isn't */
7985 if (hi > lo) {
7986 /* need to check all range that end after 's' to see if
7987 * any are unacknowledged.
7988 */
7989 while (lo >= 0 &&
7990 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7991 if (BB_OFFSET(p[lo]) < target) {
7992 /* starts before the end, and finishes after
7993 * the start, so they must overlap
7994 */
7995 if (rv != -1 && BB_ACK(p[lo]))
7996 rv = 1;
7997 else
7998 rv = -1;
7999 *first_bad = BB_OFFSET(p[lo]);
8000 *bad_sectors = BB_LEN(p[lo]);
8001 }
8002 lo--;
8003 }
8004 }
8005
8006 if (read_seqretry(&bb->lock, seq))
8007 goto retry;
8008
8009 return rv;
8010}
8011EXPORT_SYMBOL_GPL(md_is_badblock);
8012
8013/*
8014 * Add a range of bad blocks to the table.
8015 * This might extend the table, or might contract it
8016 * if two adjacent ranges can be merged.
8017 * We binary-search to find the 'insertion' point, then
8018 * decide how best to handle it.
8019 */
8020static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8021 int acknowledged)
8022{
8023 u64 *p;
8024 int lo, hi;
8025 int rv = 1;
8026
8027 if (bb->shift < 0)
8028 /* badblocks are disabled */
8029 return 0;
8030
8031 if (bb->shift) {
8032 /* round the start down, and the end up */
8033 sector_t next = s + sectors;
8034 s >>= bb->shift;
8035 next += (1<<bb->shift) - 1;
8036 next >>= bb->shift;
8037 sectors = next - s;
8038 }
8039
8040 write_seqlock_irq(&bb->lock);
8041
8042 p = bb->page;
8043 lo = 0;
8044 hi = bb->count;
8045 /* Find the last range that starts at-or-before 's' */
8046 while (hi - lo > 1) {
8047 int mid = (lo + hi) / 2;
8048 sector_t a = BB_OFFSET(p[mid]);
8049 if (a <= s)
8050 lo = mid;
8051 else
8052 hi = mid;
8053 }
8054 if (hi > lo && BB_OFFSET(p[lo]) > s)
8055 hi = lo;
8056
8057 if (hi > lo) {
8058 /* we found a range that might merge with the start
8059 * of our new range
8060 */
8061 sector_t a = BB_OFFSET(p[lo]);
8062 sector_t e = a + BB_LEN(p[lo]);
8063 int ack = BB_ACK(p[lo]);
8064 if (e >= s) {
8065 /* Yes, we can merge with a previous range */
8066 if (s == a && s + sectors >= e)
8067 /* new range covers old */
8068 ack = acknowledged;
8069 else
8070 ack = ack && acknowledged;
8071
8072 if (e < s + sectors)
8073 e = s + sectors;
8074 if (e - a <= BB_MAX_LEN) {
8075 p[lo] = BB_MAKE(a, e-a, ack);
8076 s = e;
8077 } else {
8078 /* does not all fit in one range,
8079 * make p[lo] maximal
8080 */
8081 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8082 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8083 s = a + BB_MAX_LEN;
8084 }
8085 sectors = e - s;
8086 }
8087 }
8088 if (sectors && hi < bb->count) {
8089 /* 'hi' points to the first range that starts after 's'.
8090 * Maybe we can merge with the start of that range */
8091 sector_t a = BB_OFFSET(p[hi]);
8092 sector_t e = a + BB_LEN(p[hi]);
8093 int ack = BB_ACK(p[hi]);
8094 if (a <= s + sectors) {
8095 /* merging is possible */
8096 if (e <= s + sectors) {
8097 /* full overlap */
8098 e = s + sectors;
8099 ack = acknowledged;
8100 } else
8101 ack = ack && acknowledged;
8102
8103 a = s;
8104 if (e - a <= BB_MAX_LEN) {
8105 p[hi] = BB_MAKE(a, e-a, ack);
8106 s = e;
8107 } else {
8108 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8109 s = a + BB_MAX_LEN;
8110 }
8111 sectors = e - s;
8112 lo = hi;
8113 hi++;
8114 }
8115 }
8116 if (sectors == 0 && hi < bb->count) {
8117 /* we might be able to combine lo and hi */
8118 /* Note: 's' is at the end of 'lo' */
8119 sector_t a = BB_OFFSET(p[hi]);
8120 int lolen = BB_LEN(p[lo]);
8121 int hilen = BB_LEN(p[hi]);
8122 int newlen = lolen + hilen - (s - a);
8123 if (s >= a && newlen < BB_MAX_LEN) {
8124 /* yes, we can combine them */
8125 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8126 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8127 memmove(p + hi, p + hi + 1,
8128 (bb->count - hi - 1) * 8);
8129 bb->count--;
8130 }
8131 }
8132 while (sectors) {
8133 /* didn't merge (it all).
8134 * Need to add a range just before 'hi' */
8135 if (bb->count >= MD_MAX_BADBLOCKS) {
8136 /* No room for more */
8137 rv = 0;
8138 break;
8139 } else {
8140 int this_sectors = sectors;
8141 memmove(p + hi + 1, p + hi,
8142 (bb->count - hi) * 8);
8143 bb->count++;
8144
8145 if (this_sectors > BB_MAX_LEN)
8146 this_sectors = BB_MAX_LEN;
8147 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8148 sectors -= this_sectors;
8149 s += this_sectors;
8150 }
8151 }
8152
8153 bb->changed = 1;
de393cde
N
8154 if (!acknowledged)
8155 bb->unacked_exist = 1;
2230dfe4
N
8156 write_sequnlock_irq(&bb->lock);
8157
8158 return rv;
8159}
8160
3cb03002 8161int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8162 int is_new)
2230dfe4 8163{
c6563a8c
N
8164 int rv;
8165 if (is_new)
8166 s += rdev->new_data_offset;
8167 else
8168 s += rdev->data_offset;
8169 rv = md_set_badblocks(&rdev->badblocks,
8170 s, sectors, 0);
2230dfe4
N
8171 if (rv) {
8172 /* Make sure they get written out promptly */
8bd2f0a0 8173 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N
8174 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8175 md_wakeup_thread(rdev->mddev->thread);
8176 }
8177 return rv;
8178}
8179EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8180
8181/*
8182 * Remove a range of bad blocks from the table.
8183 * This may involve extending the table if we spilt a region,
8184 * but it must not fail. So if the table becomes full, we just
8185 * drop the remove request.
8186 */
8187static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8188{
8189 u64 *p;
8190 int lo, hi;
8191 sector_t target = s + sectors;
8192 int rv = 0;
8193
8194 if (bb->shift > 0) {
8195 /* When clearing we round the start up and the end down.
8196 * This should not matter as the shift should align with
8197 * the block size and no rounding should ever be needed.
8198 * However it is better the think a block is bad when it
8199 * isn't than to think a block is not bad when it is.
8200 */
8201 s += (1<<bb->shift) - 1;
8202 s >>= bb->shift;
8203 target >>= bb->shift;
8204 sectors = target - s;
8205 }
8206
8207 write_seqlock_irq(&bb->lock);
8208
8209 p = bb->page;
8210 lo = 0;
8211 hi = bb->count;
8212 /* Find the last range that starts before 'target' */
8213 while (hi - lo > 1) {
8214 int mid = (lo + hi) / 2;
8215 sector_t a = BB_OFFSET(p[mid]);
8216 if (a < target)
8217 lo = mid;
8218 else
8219 hi = mid;
8220 }
8221 if (hi > lo) {
8222 /* p[lo] is the last range that could overlap the
8223 * current range. Earlier ranges could also overlap,
8224 * but only this one can overlap the end of the range.
8225 */
8226 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8227 /* Partial overlap, leave the tail of this range */
8228 int ack = BB_ACK(p[lo]);
8229 sector_t a = BB_OFFSET(p[lo]);
8230 sector_t end = a + BB_LEN(p[lo]);
8231
8232 if (a < s) {
8233 /* we need to split this range */
8234 if (bb->count >= MD_MAX_BADBLOCKS) {
8235 rv = 0;
8236 goto out;
8237 }
8238 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8239 bb->count++;
8240 p[lo] = BB_MAKE(a, s-a, ack);
8241 lo++;
8242 }
8243 p[lo] = BB_MAKE(target, end - target, ack);
8244 /* there is no longer an overlap */
8245 hi = lo;
8246 lo--;
8247 }
8248 while (lo >= 0 &&
8249 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8250 /* This range does overlap */
8251 if (BB_OFFSET(p[lo]) < s) {
8252 /* Keep the early parts of this range. */
8253 int ack = BB_ACK(p[lo]);
8254 sector_t start = BB_OFFSET(p[lo]);
8255 p[lo] = BB_MAKE(start, s - start, ack);
8256 /* now low doesn't overlap, so.. */
8257 break;
8258 }
8259 lo--;
8260 }
8261 /* 'lo' is strictly before, 'hi' is strictly after,
8262 * anything between needs to be discarded
8263 */
8264 if (hi - lo > 1) {
8265 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8266 bb->count -= (hi - lo - 1);
8267 }
8268 }
8269
8270 bb->changed = 1;
8271out:
8272 write_sequnlock_irq(&bb->lock);
8273 return rv;
8274}
8275
c6563a8c
N
8276int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8277 int is_new)
2230dfe4 8278{
c6563a8c
N
8279 if (is_new)
8280 s += rdev->new_data_offset;
8281 else
8282 s += rdev->data_offset;
2230dfe4 8283 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8284 s, sectors);
2230dfe4
N
8285}
8286EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8287
8288/*
8289 * Acknowledge all bad blocks in a list.
8290 * This only succeeds if ->changed is clear. It is used by
8291 * in-kernel metadata updates
8292 */
8293void md_ack_all_badblocks(struct badblocks *bb)
8294{
8295 if (bb->page == NULL || bb->changed)
8296 /* no point even trying */
8297 return;
8298 write_seqlock_irq(&bb->lock);
8299
ecb178bb 8300 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N
8301 u64 *p = bb->page;
8302 int i;
8303 for (i = 0; i < bb->count ; i++) {
8304 if (!BB_ACK(p[i])) {
8305 sector_t start = BB_OFFSET(p[i]);
8306 int len = BB_LEN(p[i]);
8307 p[i] = BB_MAKE(start, len, 1);
8308 }
8309 }
de393cde 8310 bb->unacked_exist = 0;
2230dfe4
N
8311 }
8312 write_sequnlock_irq(&bb->lock);
8313}
8314EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8315
16c791a5
N
8316/* sysfs access to bad-blocks list.
8317 * We present two files.
8318 * 'bad-blocks' lists sector numbers and lengths of ranges that
8319 * are recorded as bad. The list is truncated to fit within
8320 * the one-page limit of sysfs.
8321 * Writing "sector length" to this file adds an acknowledged
8322 * bad block list.
8323 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8324 * been acknowledged. Writing to this file adds bad blocks
8325 * without acknowledging them. This is largely for testing.
8326 */
8327
8328static ssize_t
8329badblocks_show(struct badblocks *bb, char *page, int unack)
8330{
8331 size_t len;
8332 int i;
8333 u64 *p = bb->page;
8334 unsigned seq;
8335
8336 if (bb->shift < 0)
8337 return 0;
8338
8339retry:
8340 seq = read_seqbegin(&bb->lock);
8341
8342 len = 0;
8343 i = 0;
8344
8345 while (len < PAGE_SIZE && i < bb->count) {
8346 sector_t s = BB_OFFSET(p[i]);
8347 unsigned int length = BB_LEN(p[i]);
8348 int ack = BB_ACK(p[i]);
8349 i++;
8350
8351 if (unack && ack)
8352 continue;
8353
8354 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8355 (unsigned long long)s << bb->shift,
8356 length << bb->shift);
8357 }
de393cde
N
8358 if (unack && len == 0)
8359 bb->unacked_exist = 0;
16c791a5
N
8360
8361 if (read_seqretry(&bb->lock, seq))
8362 goto retry;
8363
8364 return len;
8365}
8366
8367#define DO_DEBUG 1
8368
8369static ssize_t
8370badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8371{
8372 unsigned long long sector;
8373 int length;
8374 char newline;
8375#ifdef DO_DEBUG
8376 /* Allow clearing via sysfs *only* for testing/debugging.
8377 * Normally only a successful write may clear a badblock
8378 */
8379 int clear = 0;
8380 if (page[0] == '-') {
8381 clear = 1;
8382 page++;
8383 }
8384#endif /* DO_DEBUG */
8385
8386 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8387 case 3:
8388 if (newline != '\n')
8389 return -EINVAL;
8390 case 2:
8391 if (length <= 0)
8392 return -EINVAL;
8393 break;
8394 default:
8395 return -EINVAL;
8396 }
8397
8398#ifdef DO_DEBUG
8399 if (clear) {
8400 md_clear_badblocks(bb, sector, length);
8401 return len;
8402 }
8403#endif /* DO_DEBUG */
8404 if (md_set_badblocks(bb, sector, length, !unack))
8405 return len;
8406 else
8407 return -ENOSPC;
8408}
8409
75c96f85
AB
8410static int md_notify_reboot(struct notifier_block *this,
8411 unsigned long code, void *x)
1da177e4
LT
8412{
8413 struct list_head *tmp;
fd01b88c 8414 struct mddev *mddev;
2dba6a91 8415 int need_delay = 0;
1da177e4 8416
c744a65c
N
8417 for_each_mddev(mddev, tmp) {
8418 if (mddev_trylock(mddev)) {
30b8aa91
N
8419 if (mddev->pers)
8420 __md_stop_writes(mddev);
c744a65c
N
8421 mddev->safemode = 2;
8422 mddev_unlock(mddev);
2dba6a91 8423 }
c744a65c 8424 need_delay = 1;
1da177e4 8425 }
c744a65c
N
8426 /*
8427 * certain more exotic SCSI devices are known to be
8428 * volatile wrt too early system reboots. While the
8429 * right place to handle this issue is the given
8430 * driver, we do want to have a safe RAID driver ...
8431 */
8432 if (need_delay)
8433 mdelay(1000*1);
8434
1da177e4
LT
8435 return NOTIFY_DONE;
8436}
8437
75c96f85 8438static struct notifier_block md_notifier = {
1da177e4
LT
8439 .notifier_call = md_notify_reboot,
8440 .next = NULL,
8441 .priority = INT_MAX, /* before any real devices */
8442};
8443
8444static void md_geninit(void)
8445{
36a4e1fe 8446 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8447
c7705f34 8448 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT
8449}
8450
75c96f85 8451static int __init md_init(void)
1da177e4 8452{
e804ac78
TH
8453 int ret = -ENOMEM;
8454
ada609ee 8455 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH
8456 if (!md_wq)
8457 goto err_wq;
8458
8459 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8460 if (!md_misc_wq)
8461 goto err_misc_wq;
8462
8463 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8464 goto err_md;
8465
8466 if ((ret = register_blkdev(0, "mdp")) < 0)
8467 goto err_mdp;
8468 mdp_major = ret;
8469
3dbd8c2e 8470 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N
8471 md_probe, NULL, NULL);
8472 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT
8473 md_probe, NULL, NULL);
8474
1da177e4 8475 register_reboot_notifier(&md_notifier);
0b4d4147 8476 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT
8477
8478 md_geninit();
d710e138 8479 return 0;
1da177e4 8480
e804ac78
TH
8481err_mdp:
8482 unregister_blkdev(MD_MAJOR, "md");
8483err_md:
8484 destroy_workqueue(md_misc_wq);
8485err_misc_wq:
8486 destroy_workqueue(md_wq);
8487err_wq:
8488 return ret;
8489}
1da177e4
LT
8490
8491#ifndef MODULE
8492
8493/*
8494 * Searches all registered partitions for autorun RAID arrays
8495 * at boot time.
8496 */
4d936ec1
ME
8497
8498static LIST_HEAD(all_detected_devices);
8499struct detected_devices_node {
8500 struct list_head list;
8501 dev_t dev;
8502};
1da177e4
LT
8503
8504void md_autodetect_dev(dev_t dev)
8505{
4d936ec1
ME
8506 struct detected_devices_node *node_detected_dev;
8507
8508 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8509 if (node_detected_dev) {
8510 node_detected_dev->dev = dev;
8511 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8512 } else {
8513 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8514 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8515 }
1da177e4
LT
8516}
8517
8518
8519static void autostart_arrays(int part)
8520{
3cb03002 8521 struct md_rdev *rdev;
4d936ec1
ME
8522 struct detected_devices_node *node_detected_dev;
8523 dev_t dev;
8524 int i_scanned, i_passed;
1da177e4 8525
4d936ec1
ME
8526 i_scanned = 0;
8527 i_passed = 0;
1da177e4 8528
4d936ec1 8529 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8530
4d936ec1
ME
8531 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8532 i_scanned++;
8533 node_detected_dev = list_entry(all_detected_devices.next,
8534 struct detected_devices_node, list);
8535 list_del(&node_detected_dev->list);
8536 dev = node_detected_dev->dev;
8537 kfree(node_detected_dev);
df968c4e 8538 rdev = md_import_device(dev,0, 90);
1da177e4
LT
8539 if (IS_ERR(rdev))
8540 continue;
8541
b2d444d7 8542 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT
8543 MD_BUG();
8544 continue;
8545 }
d0fae18f 8546 set_bit(AutoDetected, &rdev->flags);
1da177e4 8547 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8548 i_passed++;
1da177e4 8549 }
4d936ec1
ME
8550
8551 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8552 i_scanned, i_passed);
1da177e4
LT
8553
8554 autorun_devices(part);
8555}
8556
fdee8ae4 8557#endif /* !MODULE */
1da177e4
LT
8558
8559static __exit void md_exit(void)
8560{
fd01b88c 8561 struct mddev *mddev;
1da177e4 8562 struct list_head *tmp;
8ab5e4c1 8563
3dbd8c2e 8564 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8565 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8566
3dbd8c2e 8567 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT
8568 unregister_blkdev(mdp_major, "mdp");
8569 unregister_reboot_notifier(&md_notifier);
8570 unregister_sysctl_table(raid_table_header);
8571 remove_proc_entry("mdstat", NULL);
29ac4aa3 8572 for_each_mddev(mddev, tmp) {
1da177e4 8573 export_array(mddev);
d3374825 8574 mddev->hold_active = 0;
1da177e4 8575 }
e804ac78
TH
8576 destroy_workqueue(md_misc_wq);
8577 destroy_workqueue(md_wq);
1da177e4
LT
8578}
8579
685784aa 8580subsys_initcall(md_init);
1da177e4
LT
8581module_exit(md_exit)
8582
f91de92e
N
8583static int get_ro(char *buffer, struct kernel_param *kp)
8584{
8585 return sprintf(buffer, "%d", start_readonly);
8586}
8587static int set_ro(const char *val, struct kernel_param *kp)
8588{
8589 char *e;
8590 int num = simple_strtoul(val, &e, 10);
8591 if (*val && (*e == '\0' || *e == '\n')) {
8592 start_readonly = num;
4dbcdc75 8593 return 0;
f91de92e
N
8594 }
8595 return -EINVAL;
8596}
8597
80ca3a44
N
8598module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8599module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8600
efeb53c0 8601module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8602
1da177e4
LT
8603EXPORT_SYMBOL(register_md_personality);
8604EXPORT_SYMBOL(unregister_md_personality);
8605EXPORT_SYMBOL(md_error);
8606EXPORT_SYMBOL(md_done_sync);
8607EXPORT_SYMBOL(md_write_start);
8608EXPORT_SYMBOL(md_write_end);
1da177e4
LT
8609EXPORT_SYMBOL(md_register_thread);
8610EXPORT_SYMBOL(md_unregister_thread);
8611EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT
8612EXPORT_SYMBOL(md_check_recovery);
8613MODULE_LICENSE("GPL");
0efb9e61 8614MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8615MODULE_ALIAS("md");
72008652 8616MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);