projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md/raid5: recognise replacements when assembling array.
[linux-2.6-block.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
2a48fc0a 39#include <linux/mutex.h>
1da177e4 40#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 41#include <linux/poll.h>
16f17b39 42#include <linux/ctype.h>
e7d2860b 43#include <linux/string.h>
fb4d8c76
N		 44#include <linux/hdreg.h>
45#include <linux/proc_fs.h>
46#include <linux/random.h>
056075c7 47#include <linux/module.h>
fb4d8c76 48#include <linux/reboot.h>
32a7627c 49#include <linux/file.h>
aa98aa31 50#include <linux/compat.h>
25570727 51#include <linux/delay.h>
bff61975
N		 52#include <linux/raid/md_p.h>
53#include <linux/raid/md_u.h>
5a0e3ad6 54#include <linux/slab.h>
43b2e5d8 55#include "md.h"
ef740c37 56#include "bitmap.h"
1da177e4 57
1da177e4 58#ifndef MODULE
d710e138 59static void autostart_arrays(int part);
1da177e4
LT		 60#endif
61
01f96c0a
N		 62/* pers_list is a list of registered personalities protected
63 * by pers_lock.
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
66 */
2604b703 67static LIST_HEAD(pers_list);
1da177e4
LT		 68static DEFINE_SPINLOCK(pers_lock);
69
5e56341d
AB		 70static void md_print_devices(void);
71
90b08710 72static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 73static struct workqueue_struct *md_wq;
74static struct workqueue_struct *md_misc_wq;
90b08710 75
5e56341d
AB		 76#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
77
1e50915f
RB		 78/*
79 * Default number of read corrections we'll attempt on an rdev
80 * before ejecting it from the array. We divide the read error
81 * count by 2 for every hour elapsed between read errors.
82 */
83#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 84/*
85 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
86 * is 1000 KB/sec, so the extra system load does not show up that much.
87 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 88 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 89 * subsystem is idle. There is also an 'absolute maximum' reconstruction
90 * speed limit - in case reconstruction slows down your system despite
91 * idle IO detection.
92 *
93 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 94 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 95 */
96
97static int sysctl_speed_limit_min = 1000;
98static int sysctl_speed_limit_max = 200000;
fd01b88c 99static inline int speed_min(struct mddev *mddev)
88202a0c
N		 100{
101 return mddev->sync_speed_min ?
102 mddev->sync_speed_min : sysctl_speed_limit_min;
103}
104
fd01b88c 105static inline int speed_max(struct mddev *mddev)
88202a0c
N		 106{
107 return mddev->sync_speed_max ?
108 mddev->sync_speed_max : sysctl_speed_limit_max;
109}
1da177e4
LT		 110
111static struct ctl_table_header *raid_table_header;
112
113static ctl_table raid_table[] = {
114 {
1da177e4
LT		 115 .procname = "speed_limit_min",
116 .data = &sysctl_speed_limit_min,
117 .maxlen = sizeof(int),
80ca3a44 118 .mode = S_IRUGO|S_IWUSR,
6d456111 119 .proc_handler = proc_dointvec,
1da177e4
LT		 120 },
121 {
1da177e4
LT		 122 .procname = "speed_limit_max",
123 .data = &sysctl_speed_limit_max,
124 .maxlen = sizeof(int),
80ca3a44 125 .mode = S_IRUGO|S_IWUSR,
6d456111 126 .proc_handler = proc_dointvec,
1da177e4 127 },
894d2491 128 { }
1da177e4
LT		 129};
130
131static ctl_table raid_dir_table[] = {
132 {
1da177e4
LT		 133 .procname = "raid",
134 .maxlen = 0,
80ca3a44 135 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 136 .child = raid_table,
137 },
894d2491 138 { }
1da177e4
LT		 139};
140
141static ctl_table raid_root_table[] = {
142 {
1da177e4
LT		 143 .procname = "dev",
144 .maxlen = 0,
145 .mode = 0555,
146 .child = raid_dir_table,
147 },
894d2491 148 { }
1da177e4
LT		 149};
150
83d5cde4 151static const struct block_device_operations md_fops;
1da177e4 152
f91de92e
N		 153static int start_readonly;
154
a167f663
N		 155/* bio_clone_mddev
156 * like bio_clone, but with a local bio set
157 */
158
159static void mddev_bio_destructor(struct bio *bio)
160{
fd01b88c 161 struct mddev *mddev, **mddevp;
a167f663
N		 162
163 mddevp = (void*)bio;
164 mddev = mddevp[-1];
165
166 bio_free(bio, mddev->bio_set);
167}
168
169struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 170 struct mddev *mddev)
a167f663
N		 171{
172 struct bio *b;
fd01b88c 173 struct mddev **mddevp;
a167f663
N		 174
175 if (!mddev || !mddev->bio_set)
176 return bio_alloc(gfp_mask, nr_iovecs);
177
178 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
179 mddev->bio_set);
180 if (!b)
181 return NULL;
182 mddevp = (void*)b;
183 mddevp[-1] = mddev;
184 b->bi_destructor = mddev_bio_destructor;
185 return b;
186}
187EXPORT_SYMBOL_GPL(bio_alloc_mddev);
188
189struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 190 struct mddev *mddev)
a167f663
N		 191{
192 struct bio *b;
fd01b88c 193 struct mddev **mddevp;
a167f663
N		 194
195 if (!mddev || !mddev->bio_set)
196 return bio_clone(bio, gfp_mask);
197
198 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
199 mddev->bio_set);
200 if (!b)
201 return NULL;
202 mddevp = (void*)b;
203 mddevp[-1] = mddev;
204 b->bi_destructor = mddev_bio_destructor;
205 __bio_clone(b, bio);
206 if (bio_integrity(bio)) {
207 int ret;
208
209 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
210
211 if (ret < 0) {
212 bio_put(b);
213 return NULL;
214 }
215 }
216
217 return b;
218}
219EXPORT_SYMBOL_GPL(bio_clone_mddev);
220
d2eb35ac
N		 221void md_trim_bio(struct bio *bio, int offset, int size)
222{
223 /* 'bio' is a cloned bio which we need to trim to match
224 * the given offset and size.
225 * This requires adjusting bi_sector, bi_size, and bi_io_vec
226 */
227 int i;
228 struct bio_vec *bvec;
229 int sofar = 0;
230
231 size <<= 9;
232 if (offset == 0 && size == bio->bi_size)
233 return;
234
235 bio->bi_sector += offset;
236 bio->bi_size = size;
237 offset <<= 9;
238 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
239
240 while (bio->bi_idx < bio->bi_vcnt &&
241 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
242 /* remove this whole bio_vec */
243 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
244 bio->bi_idx++;
245 }
246 if (bio->bi_idx < bio->bi_vcnt) {
247 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
248 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
249 }
250 /* avoid any complications with bi_idx being non-zero*/
251 if (bio->bi_idx) {
252 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
253 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
254 bio->bi_vcnt -= bio->bi_idx;
255 bio->bi_idx = 0;
256 }
257 /* Make sure vcnt and last bv are not too big */
258 bio_for_each_segment(bvec, bio, i) {
259 if (sofar + bvec->bv_len > size)
260 bvec->bv_len = size - sofar;
261 if (bvec->bv_len == 0) {
262 bio->bi_vcnt = i;
263 break;
264 }
265 sofar += bvec->bv_len;
266 }
267}
268EXPORT_SYMBOL_GPL(md_trim_bio);
269
d7603b7e
N		 270/*
271 * We have a system wide 'event count' that is incremented
272 * on any 'interesting' event, and readers of /proc/mdstat
273 * can use 'poll' or 'select' to find out when the event
274 * count increases.
275 *
276 * Events are:
277 * start array, stop array, error, add device, remove device,
278 * start build, activate spare
279 */
2989ddbd 280static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 281static atomic_t md_event_count;
fd01b88c 282void md_new_event(struct mddev *mddev)
d7603b7e
N		 283{
284 atomic_inc(&md_event_count);
285 wake_up(&md_event_waiters);
286}
29269553 287EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 288
c331eb04
N		 289/* Alternate version that can be called from interrupts
290 * when calling sysfs_notify isn't needed.
291 */
fd01b88c 292static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 293{
294 atomic_inc(&md_event_count);
295 wake_up(&md_event_waiters);
296}
297
1da177e4
LT		 298/*
299 * Enables to iterate over all existing md arrays
300 * all_mddevs_lock protects this list.
301 */
302static LIST_HEAD(all_mddevs);
303static DEFINE_SPINLOCK(all_mddevs_lock);
304
305
306/*
307 * iterates through all used mddevs in the system.
308 * We take care to grab the all_mddevs_lock whenever navigating
309 * the list, and to always hold a refcount when unlocked.
310 * Any code which breaks out of this loop while own
311 * a reference to the current mddev and must mddev_put it.
312 */
fd01b88c 313#define for_each_mddev(_mddev,_tmp) \
1da177e4
LT		 314 \
315 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 316 _tmp = all_mddevs.next; \
317 _mddev = NULL;}); \
318 ({ if (_tmp != &all_mddevs) \
319 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 320 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 321 if (_mddev) mddev_put(_mddev); \
322 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
323 _tmp != &all_mddevs;}); \
1da177e4 324 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 325 _tmp = _tmp->next;}) \
1da177e4
LT		 326 )
327
328
409c57f3
N		 329/* Rather than calling directly into the personality make_request function,
330 * IO requests come here first so that we can check if the device is
331 * being suspended pending a reconfiguration.
332 * We hold a refcount over the call to ->make_request. By the time that
333 * call has finished, the bio has been linked into some internal structure
334 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 */
5a7bbad2 336static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 337{
49077326 338 const int rw = bio_data_dir(bio);
fd01b88c 339 struct mddev *mddev = q->queuedata;
49077326 340 int cpu;
e91ece55 341 unsigned int sectors;
49077326 342
0ca69886
N		 343 if (mddev == NULL || mddev->pers == NULL
344 || !mddev->ready) {
409c57f3 345 bio_io_error(bio);
5a7bbad2 346 return;
409c57f3 347 }
0ca69886 348 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 349 rcu_read_lock();
e9c7469b 350 if (mddev->suspended) {
409c57f3
N		 351 DEFINE_WAIT(__wait);
352 for (;;) {
353 prepare_to_wait(&mddev->sb_wait, &__wait,
354 TASK_UNINTERRUPTIBLE);
e9c7469b 355 if (!mddev->suspended)
409c57f3
N		 356 break;
357 rcu_read_unlock();
358 schedule();
359 rcu_read_lock();
360 }
361 finish_wait(&mddev->sb_wait, &__wait);
362 }
363 atomic_inc(&mddev->active_io);
364 rcu_read_unlock();
49077326 365
e91ece55
CM		 366 /*
367 * save the sectors now since our bio can
368 * go away inside make_request
369 */
370 sectors = bio_sectors(bio);
5a7bbad2 371 mddev->pers->make_request(mddev, bio);
49077326
N		 372
373 cpu = part_stat_lock();
374 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 375 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N		 376 part_stat_unlock();
377
409c57f3
N		 378 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
379 wake_up(&mddev->sb_wait);
409c57f3
N		 380}
381
9e35b99c
N		 382/* mddev_suspend makes sure no new requests are submitted
383 * to the device, and that any requests that have been submitted
384 * are completely handled.
385 * Once ->stop is called and completes, the module will be completely
386 * unused.
387 */
fd01b88c 388void mddev_suspend(struct mddev *mddev)
409c57f3
N		 389{
390 BUG_ON(mddev->suspended);
391 mddev->suspended = 1;
392 synchronize_rcu();
393 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
394 mddev->pers->quiesce(mddev, 1);
409c57f3 395}
390ee602 396EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 397
fd01b88c 398void mddev_resume(struct mddev *mddev)
409c57f3
N		 399{
400 mddev->suspended = 0;
401 wake_up(&mddev->sb_wait);
402 mddev->pers->quiesce(mddev, 0);
0fd018af
JB		 403
404 md_wakeup_thread(mddev->thread);
405 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 406}
390ee602 407EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 408
fd01b88c 409int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N		 410{
411 return mddev->suspended;
412}
413EXPORT_SYMBOL(mddev_congested);
414
a2826aa9 415/*
e9c7469b 416 * Generic flush handling for md
a2826aa9
N		 417 */
418
e9c7469b 419static void md_end_flush(struct bio *bio, int err)
a2826aa9 420{
3cb03002 421 struct md_rdev *rdev = bio->bi_private;
fd01b88c 422 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 423
424 rdev_dec_pending(rdev, mddev);
425
426 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 427 /* The pre-request flush has finished */
e804ac78 428 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 429 }
430 bio_put(bio);
431}
432
a7a07e69
N		 433static void md_submit_flush_data(struct work_struct *ws);
434
a035fc3e 435static void submit_flushes(struct work_struct *ws)
a2826aa9 436{
fd01b88c 437 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 438 struct md_rdev *rdev;
a2826aa9 439
a7a07e69
N		 440 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
441 atomic_set(&mddev->flush_pending, 1);
a2826aa9
N		 442 rcu_read_lock();
443 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
444 if (rdev->raid_disk >= 0 &&
445 !test_bit(Faulty, &rdev->flags)) {
446 /* Take two references, one is dropped
447 * when request finishes, one after
448 * we reclaim rcu_read_lock
449 */
450 struct bio *bi;
451 atomic_inc(&rdev->nr_pending);
452 atomic_inc(&rdev->nr_pending);
453 rcu_read_unlock();
a167f663 454 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
e9c7469b 455 bi->bi_end_io = md_end_flush;
a2826aa9
N		 456 bi->bi_private = rdev;
457 bi->bi_bdev = rdev->bdev;
458 atomic_inc(&mddev->flush_pending);
e9c7469b 459 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 460 rcu_read_lock();
461 rdev_dec_pending(rdev, mddev);
462 }
463 rcu_read_unlock();
a7a07e69
N		 464 if (atomic_dec_and_test(&mddev->flush_pending))
465 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 466}
467
e9c7469b 468static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 469{
fd01b88c 470 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 471 struct bio *bio = mddev->flush_bio;
a2826aa9 472
e9c7469b 473 if (bio->bi_size == 0)
a2826aa9
N		 474 /* an empty barrier - all done */
475 bio_endio(bio, 0);
476 else {
e9c7469b 477 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 478 mddev->pers->make_request(mddev, bio);
a2826aa9 479 }
2b74e12e
N		 480
481 mddev->flush_bio = NULL;
482 wake_up(&mddev->sb_wait);
a2826aa9
N		 483}
484
fd01b88c 485void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N		 486{
487 spin_lock_irq(&mddev->write_lock);
488 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 489 !mddev->flush_bio,
a2826aa9 490 mddev->write_lock, /*nothing*/);
e9c7469b 491 mddev->flush_bio = bio;
a2826aa9
N		 492 spin_unlock_irq(&mddev->write_lock);
493
a035fc3e
N		 494 INIT_WORK(&mddev->flush_work, submit_flushes);
495 queue_work(md_wq, &mddev->flush_work);
a2826aa9 496}
e9c7469b 497EXPORT_SYMBOL(md_flush_request);
409c57f3 498
97658cdd
N		 499/* Support for plugging.
500 * This mirrors the plugging support in request_queue, but does not
501 * require having a whole queue or request structures.
502 * We allocate an md_plug_cb for each md device and each thread it gets
503 * plugged on. This links tot the private plug_handle structure in the
504 * personality data where we keep a count of the number of outstanding
505 * plugs so other code can see if a plug is active.
506 */
507struct md_plug_cb {
508 struct blk_plug_cb cb;
fd01b88c 509 struct mddev *mddev;
97658cdd
N		 510};
511
512static void plugger_unplug(struct blk_plug_cb *cb)
513{
514 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
515 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
516 md_wakeup_thread(mdcb->mddev->thread);
517 kfree(mdcb);
518}
519
520/* Check that an unplug wakeup will come shortly.
521 * If not, wakeup the md thread immediately
522 */
fd01b88c 523int mddev_check_plugged(struct mddev *mddev)
97658cdd
N		 524{
525 struct blk_plug *plug = current->plug;
526 struct md_plug_cb *mdcb;
527
528 if (!plug)
529 return 0;
530
531 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
532 if (mdcb->cb.callback == plugger_unplug &&
533 mdcb->mddev == mddev) {
534 /* Already on the list, move to top */
535 if (mdcb != list_first_entry(&plug->cb_list,
536 struct md_plug_cb,
537 cb.list))
538 list_move(&mdcb->cb.list, &plug->cb_list);
539 return 1;
540 }
541 }
542 /* Not currently on the callback list */
543 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
544 if (!mdcb)
545 return 0;
546
547 mdcb->mddev = mddev;
548 mdcb->cb.callback = plugger_unplug;
549 atomic_inc(&mddev->plug_cnt);
550 list_add(&mdcb->cb.list, &plug->cb_list);
551 return 1;
552}
553EXPORT_SYMBOL_GPL(mddev_check_plugged);
2ac87401 554
fd01b88c 555static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 556{
557 atomic_inc(&mddev->active);
558 return mddev;
559}
560
5fd3a17e 561static void mddev_delayed_delete(struct work_struct *ws);
d3374825 562
fd01b88c 563static void mddev_put(struct mddev *mddev)
1da177e4 564{
a167f663
N		 565 struct bio_set *bs = NULL;
566
1da177e4
LT		 567 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
568 return;
d3374825 569 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 570 mddev->ctime == 0 && !mddev->hold_active) {
571 /* Array is not configured at all, and not held active,
572 * so destroy it */
af8a2434 573 list_del_init(&mddev->all_mddevs);
a167f663
N		 574 bs = mddev->bio_set;
575 mddev->bio_set = NULL;
d3374825 576 if (mddev->gendisk) {
e804ac78
TH		 577 /* We did a probe so need to clean up. Call
578 * queue_work inside the spinlock so that
579 * flush_workqueue() after mddev_find will
580 * succeed in waiting for the work to be done.
d3374825
N		 581 */
582 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 583 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 584 } else
585 kfree(mddev);
586 }
587 spin_unlock(&all_mddevs_lock);
a167f663
N		 588 if (bs)
589 bioset_free(bs);
1da177e4
LT		 590}
591
fd01b88c 592void mddev_init(struct mddev *mddev)
fafd7fb0
N		 593{
594 mutex_init(&mddev->open_mutex);
595 mutex_init(&mddev->reconfig_mutex);
596 mutex_init(&mddev->bitmap_info.mutex);
597 INIT_LIST_HEAD(&mddev->disks);
598 INIT_LIST_HEAD(&mddev->all_mddevs);
599 init_timer(&mddev->safemode_timer);
600 atomic_set(&mddev->active, 1);
601 atomic_set(&mddev->openers, 0);
602 atomic_set(&mddev->active_io, 0);
97658cdd 603 atomic_set(&mddev->plug_cnt, 0);
fafd7fb0
N		 604 spin_lock_init(&mddev->write_lock);
605 atomic_set(&mddev->flush_pending, 0);
606 init_waitqueue_head(&mddev->sb_wait);
607 init_waitqueue_head(&mddev->recovery_wait);
608 mddev->reshape_position = MaxSector;
609 mddev->resync_min = 0;
610 mddev->resync_max = MaxSector;
611 mddev->level = LEVEL_NONE;
612}
390ee602 613EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 614
fd01b88c 615static struct mddev * mddev_find(dev_t unit)
1da177e4 616{
fd01b88c 617 struct mddev *mddev, *new = NULL;
1da177e4 618
8f5f02c4
N		 619 if (unit && MAJOR(unit) != MD_MAJOR)
620 unit &= ~((1<<MdpMinorShift)-1);
621
1da177e4
LT		 622 retry:
623 spin_lock(&all_mddevs_lock);
efeb53c0
N		 624
625 if (unit) {
626 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
627 if (mddev->unit == unit) {
628 mddev_get(mddev);
629 spin_unlock(&all_mddevs_lock);
630 kfree(new);
631 return mddev;
632 }
633
634 if (new) {
635 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 636 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 637 new->hold_active = UNTIL_IOCTL;
638 return new;
1da177e4 639 }
efeb53c0
N		 640 } else if (new) {
641 /* find an unused unit number */
642 static int next_minor = 512;
643 int start = next_minor;
644 int is_free = 0;
645 int dev = 0;
646 while (!is_free) {
647 dev = MKDEV(MD_MAJOR, next_minor);
648 next_minor++;
649 if (next_minor > MINORMASK)
650 next_minor = 0;
651 if (next_minor == start) {
652 /* Oh dear, all in use. */
653 spin_unlock(&all_mddevs_lock);
654 kfree(new);
655 return NULL;
656 }
657
658 is_free = 1;
659 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
660 if (mddev->unit == dev) {
661 is_free = 0;
662 break;
663 }
664 }
665 new->unit = dev;
666 new->md_minor = MINOR(dev);
667 new->hold_active = UNTIL_STOP;
1da177e4
LT		 668 list_add(&new->all_mddevs, &all_mddevs);
669 spin_unlock(&all_mddevs_lock);
670 return new;
671 }
672 spin_unlock(&all_mddevs_lock);
673
9ffae0cf 674 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 675 if (!new)
676 return NULL;
677
1da177e4
LT		 678 new->unit = unit;
679 if (MAJOR(unit) == MD_MAJOR)
680 new->md_minor = MINOR(unit);
681 else
682 new->md_minor = MINOR(unit) >> MdpMinorShift;
683
fafd7fb0 684 mddev_init(new);
1da177e4 685
1da177e4
LT		 686 goto retry;
687}
688
fd01b88c 689static inline int mddev_lock(struct mddev * mddev)
1da177e4 690{
df5b89b3 691 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 692}
693
fd01b88c 694static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW		 695{
696 return mutex_is_locked(&mddev->reconfig_mutex);
697}
698
fd01b88c 699static inline int mddev_trylock(struct mddev * mddev)
1da177e4 700{
df5b89b3 701 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 702}
703
b6eb127d
N		 704static struct attribute_group md_redundancy_group;
705
fd01b88c 706static void mddev_unlock(struct mddev * mddev)
1da177e4 707{
a64c876f 708 if (mddev->to_remove) {
b6eb127d
N		 709 /* These cannot be removed under reconfig_mutex as
710 * an access to the files will try to take reconfig_mutex
711 * while holding the file unremovable, which leads to
712 * a deadlock.
bb4f1e9d
N		 713 * So hold set sysfs_active while the remove in happeing,
714 * and anything else which might set ->to_remove or my
715 * otherwise change the sysfs namespace will fail with
716 * -EBUSY if sysfs_active is still set.
717 * We set sysfs_active under reconfig_mutex and elsewhere
718 * test it under the same mutex to ensure its correct value
719 * is seen.
b6eb127d 720 */
a64c876f
N		 721 struct attribute_group *to_remove = mddev->to_remove;
722 mddev->to_remove = NULL;
bb4f1e9d 723 mddev->sysfs_active = 1;
b6eb127d
N		 724 mutex_unlock(&mddev->reconfig_mutex);
725
00bcb4ac
N		 726 if (mddev->kobj.sd) {
727 if (to_remove != &md_redundancy_group)
728 sysfs_remove_group(&mddev->kobj, to_remove);
729 if (mddev->pers == NULL ||
730 mddev->pers->sync_request == NULL) {
731 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
732 if (mddev->sysfs_action)
733 sysfs_put(mddev->sysfs_action);
734 mddev->sysfs_action = NULL;
735 }
a64c876f 736 }
bb4f1e9d 737 mddev->sysfs_active = 0;
b6eb127d
N		 738 } else
739 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 740
751e67ca
CD		 741 /* As we've dropped the mutex we need a spinlock to
742 * make sure the thread doesn't disappear
01f96c0a
N		 743 */
744 spin_lock(&pers_lock);
005eca5e 745 md_wakeup_thread(mddev->thread);
01f96c0a 746 spin_unlock(&pers_lock);
1da177e4
LT		 747}
748
fd01b88c 749static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 750{
3cb03002 751 struct md_rdev *rdev;
1da177e4 752
159ec1fc 753 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 754 if (rdev->desc_nr == nr)
755 return rdev;
159ec1fc 756
1da177e4
LT		 757 return NULL;
758}
759
fd01b88c 760static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
1da177e4 761{
3cb03002 762 struct md_rdev *rdev;
1da177e4 763
159ec1fc 764 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 765 if (rdev->bdev->bd_dev == dev)
766 return rdev;
159ec1fc 767
1da177e4
LT		 768 return NULL;
769}
770
84fc4b56 771static struct md_personality *find_pers(int level, char *clevel)
2604b703 772{
84fc4b56 773 struct md_personality *pers;
d9d166c2
N		 774 list_for_each_entry(pers, &pers_list, list) {
775 if (level != LEVEL_NONE && pers->level == level)
2604b703 776 return pers;
d9d166c2
N		 777 if (strcmp(pers->name, clevel)==0)
778 return pers;
779 }
2604b703
N		 780 return NULL;
781}
782
b73df2d3 783/* return the offset of the super block in 512byte sectors */
3cb03002 784static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 785{
57b2caa3 786 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 787 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 788}
789
3cb03002 790static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 791{
792 if (rdev->sb_page)
793 MD_BUG();
794
795 rdev->sb_page = alloc_page(GFP_KERNEL);
796 if (!rdev->sb_page) {
797 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 798 return -ENOMEM;
1da177e4
LT		 799 }
800
801 return 0;
802}
803
3cb03002 804static void free_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 805{
806 if (rdev->sb_page) {
2d1f3b5d 807 put_page(rdev->sb_page);
1da177e4
LT		 808 rdev->sb_loaded = 0;
809 rdev->sb_page = NULL;
0f420358 810 rdev->sb_start = 0;
dd8ac336 811 rdev->sectors = 0;
1da177e4 812 }
2699b672
N		 813 if (rdev->bb_page) {
814 put_page(rdev->bb_page);
815 rdev->bb_page = NULL;
816 }
1da177e4
LT		 817}
818
819
6712ecf8 820static void super_written(struct bio *bio, int error)
7bfa19f2 821{
3cb03002 822 struct md_rdev *rdev = bio->bi_private;
fd01b88c 823 struct mddev *mddev = rdev->mddev;
7bfa19f2 824
3a0f5bbb
N		 825 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
826 printk("md: super_written gets error=%d, uptodate=%d\n",
827 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
828 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 829 md_error(mddev, rdev);
3a0f5bbb 830 }
7bfa19f2 831
a9701a30
N		 832 if (atomic_dec_and_test(&mddev->pending_writes))
833 wake_up(&mddev->sb_wait);
f8b58edf 834 bio_put(bio);
7bfa19f2
N		 835}
836
fd01b88c 837void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 838 sector_t sector, int size, struct page *page)
839{
840 /* write first size bytes of page to sector of rdev
841 * Increment mddev->pending_writes before returning
842 * and decrement it on completion, waking up sb_wait
843 * if zero is reached.
844 * If an error occurred, call md_error
845 */
a167f663 846 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 847
a6ff7e08 848 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 849 bio->bi_sector = sector;
850 bio_add_page(bio, page, size, 0);
851 bio->bi_private = rdev;
852 bio->bi_end_io = super_written;
a9701a30 853
7bfa19f2 854 atomic_inc(&mddev->pending_writes);
a5bf4df0 855 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 856}
857
fd01b88c 858void md_super_wait(struct mddev *mddev)
a9701a30 859{
e9c7469b 860 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 861 DEFINE_WAIT(wq);
862 for(;;) {
863 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
864 if (atomic_read(&mddev->pending_writes)==0)
865 break;
a9701a30
N		 866 schedule();
867 }
868 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 869}
870
6712ecf8 871static void bi_complete(struct bio *bio, int error)
1da177e4 872{
1da177e4 873 complete((struct completion*)bio->bi_private);
1da177e4
LT		 874}
875
3cb03002 876int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 877 struct page *page, int rw, bool metadata_op)
1da177e4 878{
a167f663 879 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 880 struct completion event;
881 int ret;
882
721a9602 883 rw |= REQ_SYNC;
1da177e4 884
a6ff7e08
JB		 885 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
886 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 887 if (metadata_op)
888 bio->bi_sector = sector + rdev->sb_start;
889 else
890 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 891 bio_add_page(bio, page, size, 0);
892 init_completion(&event);
893 bio->bi_private = &event;
894 bio->bi_end_io = bi_complete;
895 submit_bio(rw, bio);
896 wait_for_completion(&event);
897
898 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
899 bio_put(bio);
900 return ret;
901}
a8745db2 902EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 903
3cb03002 904static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT		 905{
906 char b[BDEVNAME_SIZE];
907 if (!rdev->sb_page) {
908 MD_BUG();
909 return -EINVAL;
910 }
911 if (rdev->sb_loaded)
912 return 0;
913
914
ccebd4c4 915 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 916 goto fail;
917 rdev->sb_loaded = 1;
918 return 0;
919
920fail:
921 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
922 bdevname(rdev->bdev,b));
923 return -EINVAL;
924}
925
926static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
927{
05710466
AN		 928 return sb1->set_uuid0 == sb2->set_uuid0 &&
929 sb1->set_uuid1 == sb2->set_uuid1 &&
930 sb1->set_uuid2 == sb2->set_uuid2 &&
931 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 932}
933
1da177e4
LT		 934static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
935{
936 int ret;
937 mdp_super_t *tmp1, *tmp2;
938
939 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
940 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
941
942 if (!tmp1 || !tmp2) {
943 ret = 0;
35020f1a 944 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 945 goto abort;
946 }
947
948 *tmp1 = *sb1;
949 *tmp2 = *sb2;
950
951 /*
952 * nr_disks is not constant
953 */
954 tmp1->nr_disks = 0;
955 tmp2->nr_disks = 0;
956
ce0c8e05 957 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 958abort:
990a8baf
JJ		 959 kfree(tmp1);
960 kfree(tmp2);
1da177e4
LT		 961 return ret;
962}
963
4d167f09
N		 964
965static u32 md_csum_fold(u32 csum)
966{
967 csum = (csum & 0xffff) + (csum >> 16);
968 return (csum & 0xffff) + (csum >> 16);
969}
970
1da177e4
LT		 971static unsigned int calc_sb_csum(mdp_super_t * sb)
972{
4d167f09
N		 973 u64 newcsum = 0;
974 u32 *sb32 = (u32*)sb;
975 int i;
1da177e4
LT		 976 unsigned int disk_csum, csum;
977
978 disk_csum = sb->sb_csum;
979 sb->sb_csum = 0;
4d167f09
N		 980
981 for (i = 0; i < MD_SB_BYTES/4 ; i++)
982 newcsum += sb32[i];
983 csum = (newcsum & 0xffffffff) + (newcsum>>32);
984
985
986#ifdef CONFIG_ALPHA
987 /* This used to use csum_partial, which was wrong for several
988 * reasons including that different results are returned on
989 * different architectures. It isn't critical that we get exactly
990 * the same return value as before (we always csum_fold before
991 * testing, and that removes any differences). However as we
992 * know that csum_partial always returned a 16bit value on
993 * alphas, do a fold to maximise conformity to previous behaviour.
994 */
995 sb->sb_csum = md_csum_fold(disk_csum);
996#else
1da177e4 997 sb->sb_csum = disk_csum;
4d167f09 998#endif
1da177e4
LT		 999 return csum;
1000}
1001
1002
1003/*
1004 * Handle superblock details.
1005 * We want to be able to handle multiple superblock formats
1006 * so we have a common interface to them all, and an array of
1007 * different handlers.
1008 * We rely on user-space to write the initial superblock, and support
1009 * reading and updating of superblocks.
1010 * Interface methods are:
3cb03002 1011 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1012 * loads and validates a superblock on dev.
1013 * if refdev != NULL, compare superblocks on both devices
1014 * Return:
1015 * 0 - dev has a superblock that is compatible with refdev
1016 * 1 - dev has a superblock that is compatible and newer than refdev
1017 * so dev should be used as the refdev in future
1018 * -EINVAL superblock incompatible or invalid
1019 * -othererror e.g. -EIO
1020 *
fd01b88c 1021 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1022 * Verify that dev is acceptable into mddev.
1023 * The first time, mddev->raid_disks will be 0, and data from
1024 * dev should be merged in. Subsequent calls check that dev
1025 * is new enough. Return 0 or -EINVAL
1026 *
fd01b88c 1027 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1028 * Update the superblock for rdev with data in mddev
1029 * This does not write to disc.
1030 *
1031 */
1032
1033struct super_type {
0cd17fec
CW		 1034 char *name;
1035 struct module *owner;
3cb03002 1036 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
0cd17fec 1037 int minor_version);
fd01b88c
N		 1038 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1039 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
3cb03002 1040 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 1041 sector_t num_sectors);
1da177e4
LT		 1042};
1043
0894cc30
AN		 1044/*
1045 * Check that the given mddev has no bitmap.
1046 *
1047 * This function is called from the run method of all personalities that do not
1048 * support bitmaps. It prints an error message and returns non-zero if mddev
1049 * has a bitmap. Otherwise, it returns 0.
1050 *
1051 */
fd01b88c 1052int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1053{
c3d9714e 1054 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 1055 return 0;
1056 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1057 mdname(mddev), mddev->pers->name);
1058 return 1;
1059}
1060EXPORT_SYMBOL(md_check_no_bitmap);
1061
1da177e4
LT		 1062/*
1063 * load_super for 0.90.0
1064 */
3cb03002 1065static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1066{
1067 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1068 mdp_super_t *sb;
1069 int ret;
1da177e4
LT		 1070
1071 /*
0f420358 1072 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1073 * it's at the end of the disk.
1074 *
1075 * It also happens to be a multiple of 4Kb.
1076 */
57b2caa3 1077 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1078
0002b271 1079 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1080 if (ret) return ret;
1081
1082 ret = -EINVAL;
1083
1084 bdevname(rdev->bdev, b);
65a06f06 1085 sb = page_address(rdev->sb_page);
1da177e4
LT		 1086
1087 if (sb->md_magic != MD_SB_MAGIC) {
1088 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1089 b);
1090 goto abort;
1091 }
1092
1093 if (sb->major_version != 0 ||
f6705578
N		 1094 sb->minor_version < 90 ||
1095 sb->minor_version > 91) {
1da177e4
LT		 1096 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1097 sb->major_version, sb->minor_version,
1098 b);
1099 goto abort;
1100 }
1101
1102 if (sb->raid_disks <= 0)
1103 goto abort;
1104
4d167f09 1105 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1106 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1107 b);
1108 goto abort;
1109 }
1110
1111 rdev->preferred_minor = sb->md_minor;
1112 rdev->data_offset = 0;
0002b271 1113 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1114 rdev->badblocks.shift = -1;
1da177e4
LT		 1115
1116 if (sb->level == LEVEL_MULTIPATH)
1117 rdev->desc_nr = -1;
1118 else
1119 rdev->desc_nr = sb->this_disk.number;
1120
9a7b2b0f 1121 if (!refdev) {
1da177e4 1122 ret = 1;
9a7b2b0f 1123 } else {
1da177e4 1124 __u64 ev1, ev2;
65a06f06 1125 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1126 if (!uuid_equal(refsb, sb)) {
1127 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1128 b, bdevname(refdev->bdev,b2));
1129 goto abort;
1130 }
1131 if (!sb_equal(refsb, sb)) {
1132 printk(KERN_WARNING "md: %s has same UUID"
1133 " but different superblock to %s\n",
1134 b, bdevname(refdev->bdev, b2));
1135 goto abort;
1136 }
1137 ev1 = md_event(sb);
1138 ev2 = md_event(refsb);
1139 if (ev1 > ev2)
1140 ret = 1;
1141 else
1142 ret = 0;
1143 }
8190e754 1144 rdev->sectors = rdev->sb_start;
27a7b260
N		 1145 /* Limit to 4TB as metadata cannot record more than that */
1146 if (rdev->sectors >= (2ULL << 32))
1147 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1148
27a7b260 1149 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1150 /* "this cannot possibly happen" ... */
1151 ret = -EINVAL;
1152
1da177e4
LT		 1153 abort:
1154 return ret;
1155}
1156
1157/*
1158 * validate_super for 0.90.0
1159 */
fd01b88c 1160static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1161{
1162 mdp_disk_t *desc;
65a06f06 1163 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1164 __u64 ev1 = md_event(sb);
1da177e4 1165
41158c7e 1166 rdev->raid_disk = -1;
c5d79adb
N		 1167 clear_bit(Faulty, &rdev->flags);
1168 clear_bit(In_sync, &rdev->flags);
1169 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1170
1da177e4
LT		 1171 if (mddev->raid_disks == 0) {
1172 mddev->major_version = 0;
1173 mddev->minor_version = sb->minor_version;
1174 mddev->patch_version = sb->patch_version;
e691063a 1175 mddev->external = 0;
9d8f0363 1176 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1177 mddev->ctime = sb->ctime;
1178 mddev->utime = sb->utime;
1179 mddev->level = sb->level;
d9d166c2 1180 mddev->clevel[0] = 0;
1da177e4
LT		 1181 mddev->layout = sb->layout;
1182 mddev->raid_disks = sb->raid_disks;
27a7b260 1183 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1184 mddev->events = ev1;
c3d9714e
N		 1185 mddev->bitmap_info.offset = 0;
1186 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1187
f6705578
N		 1188 if (mddev->minor_version >= 91) {
1189 mddev->reshape_position = sb->reshape_position;
1190 mddev->delta_disks = sb->delta_disks;
1191 mddev->new_level = sb->new_level;
1192 mddev->new_layout = sb->new_layout;
664e7c41 1193 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1194 } else {
1195 mddev->reshape_position = MaxSector;
1196 mddev->delta_disks = 0;
1197 mddev->new_level = mddev->level;
1198 mddev->new_layout = mddev->layout;
664e7c41 1199 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1200 }
1201
1da177e4
LT		 1202 if (sb->state & (1<<MD_SB_CLEAN))
1203 mddev->recovery_cp = MaxSector;
1204 else {
1205 if (sb->events_hi == sb->cp_events_hi &&
1206 sb->events_lo == sb->cp_events_lo) {
1207 mddev->recovery_cp = sb->recovery_cp;
1208 } else
1209 mddev->recovery_cp = 0;
1210 }
1211
1212 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1213 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1214 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1215 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1216
1217 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1218
1219 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1220 mddev->bitmap_info.file == NULL)
1221 mddev->bitmap_info.offset =
1222 mddev->bitmap_info.default_offset;
a654b9d8 1223
41158c7e 1224 } else if (mddev->pers == NULL) {
be6800a7
N		 1225 /* Insist on good event counter while assembling, except
1226 * for spares (which don't need an event count) */
1da177e4 1227 ++ev1;
be6800a7
N		 1228 if (sb->disks[rdev->desc_nr].state & (
1229 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1230 if (ev1 < mddev->events)
1231 return -EINVAL;
41158c7e
N		 1232 } else if (mddev->bitmap) {
1233 /* if adding to array with a bitmap, then we can accept an
1234 * older device ... but not too old.
1235 */
41158c7e
N		 1236 if (ev1 < mddev->bitmap->events_cleared)
1237 return 0;
07d84d10
N		 1238 } else {
1239 if (ev1 < mddev->events)
1240 /* just a hot-add of a new device, leave raid_disk at -1 */
1241 return 0;
1242 }
41158c7e 1243
1da177e4 1244 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1245 desc = sb->disks + rdev->desc_nr;
1246
1247 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1248 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1249 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1250 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1251 set_bit(In_sync, &rdev->flags);
1da177e4 1252 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1253 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1254 /* active but not in sync implies recovery up to
1255 * reshape position. We don't know exactly where
1256 * that is, so set to zero for now */
1257 if (mddev->minor_version >= 91) {
1258 rdev->recovery_offset = 0;
1259 rdev->raid_disk = desc->raid_disk;
1260 }
1da177e4 1261 }
8ddf9efe
N		 1262 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1263 set_bit(WriteMostly, &rdev->flags);
41158c7e 1264 } else /* MULTIPATH are always insync */
b2d444d7 1265 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1266 return 0;
1267}
1268
1269/*
1270 * sync_super for 0.90.0
1271 */
fd01b88c 1272static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1273{
1274 mdp_super_t *sb;
3cb03002 1275 struct md_rdev *rdev2;
1da177e4 1276 int next_spare = mddev->raid_disks;
19133a42 1277
1da177e4
LT		 1278
1279 /* make rdev->sb match mddev data..
1280 *
1281 * 1/ zero out disks
1282 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1283 * 3/ any empty disks < next_spare become removed
1284 *
1285 * disks[0] gets initialised to REMOVED because
1286 * we cannot be sure from other fields if it has
1287 * been initialised or not.
1288 */
1289 int i;
1290 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1291
61181565
N		 1292 rdev->sb_size = MD_SB_BYTES;
1293
65a06f06 1294 sb = page_address(rdev->sb_page);
1da177e4
LT		 1295
1296 memset(sb, 0, sizeof(*sb));
1297
1298 sb->md_magic = MD_SB_MAGIC;
1299 sb->major_version = mddev->major_version;
1da177e4
LT		 1300 sb->patch_version = mddev->patch_version;
1301 sb->gvalid_words = 0; /* ignored */
1302 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1303 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1304 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1305 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1306
1307 sb->ctime = mddev->ctime;
1308 sb->level = mddev->level;
58c0fed4 1309 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1310 sb->raid_disks = mddev->raid_disks;
1311 sb->md_minor = mddev->md_minor;
e691063a 1312 sb->not_persistent = 0;
1da177e4
LT		 1313 sb->utime = mddev->utime;
1314 sb->state = 0;
1315 sb->events_hi = (mddev->events>>32);
1316 sb->events_lo = (u32)mddev->events;
1317
f6705578
N		 1318 if (mddev->reshape_position == MaxSector)
1319 sb->minor_version = 90;
1320 else {
1321 sb->minor_version = 91;
1322 sb->reshape_position = mddev->reshape_position;
1323 sb->new_level = mddev->new_level;
1324 sb->delta_disks = mddev->delta_disks;
1325 sb->new_layout = mddev->new_layout;
664e7c41 1326 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1327 }
1328 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1329 if (mddev->in_sync)
1330 {
1331 sb->recovery_cp = mddev->recovery_cp;
1332 sb->cp_events_hi = (mddev->events>>32);
1333 sb->cp_events_lo = (u32)mddev->events;
1334 if (mddev->recovery_cp == MaxSector)
1335 sb->state = (1<< MD_SB_CLEAN);
1336 } else
1337 sb->recovery_cp = 0;
1338
1339 sb->layout = mddev->layout;
9d8f0363 1340 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1341
c3d9714e 1342 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1343 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1344
1da177e4 1345 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1346 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1347 mdp_disk_t *d;
86e6ffdd 1348 int desc_nr;
0261cd9f
N		 1349 int is_active = test_bit(In_sync, &rdev2->flags);
1350
1351 if (rdev2->raid_disk >= 0 &&
1352 sb->minor_version >= 91)
1353 /* we have nowhere to store the recovery_offset,
1354 * but if it is not below the reshape_position,
1355 * we can piggy-back on that.
1356 */
1357 is_active = 1;
1358 if (rdev2->raid_disk < 0 ||
1359 test_bit(Faulty, &rdev2->flags))
1360 is_active = 0;
1361 if (is_active)
86e6ffdd 1362 desc_nr = rdev2->raid_disk;
1da177e4 1363 else
86e6ffdd 1364 desc_nr = next_spare++;
19133a42 1365 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1366 d = &sb->disks[rdev2->desc_nr];
1367 nr_disks++;
1368 d->number = rdev2->desc_nr;
1369 d->major = MAJOR(rdev2->bdev->bd_dev);
1370 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1371 if (is_active)
1da177e4
LT		 1372 d->raid_disk = rdev2->raid_disk;
1373 else
1374 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1375 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1376 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1377 else if (is_active) {
1da177e4 1378 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1379 if (test_bit(In_sync, &rdev2->flags))
1380 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1381 active++;
1382 working++;
1383 } else {
1384 d->state = 0;
1385 spare++;
1386 working++;
1387 }
8ddf9efe
N		 1388 if (test_bit(WriteMostly, &rdev2->flags))
1389 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1390 }
1da177e4
LT		 1391 /* now set the "removed" and "faulty" bits on any missing devices */
1392 for (i=0 ; i < mddev->raid_disks ; i++) {
1393 mdp_disk_t *d = &sb->disks[i];
1394 if (d->state == 0 && d->number == 0) {
1395 d->number = i;
1396 d->raid_disk = i;
1397 d->state = (1<<MD_DISK_REMOVED);
1398 d->state |= (1<<MD_DISK_FAULTY);
1399 failed++;
1400 }
1401 }
1402 sb->nr_disks = nr_disks;
1403 sb->active_disks = active;
1404 sb->working_disks = working;
1405 sb->failed_disks = failed;
1406 sb->spare_disks = spare;
1407
1408 sb->this_disk = sb->disks[rdev->desc_nr];
1409 sb->sb_csum = calc_sb_csum(sb);
1410}
1411
0cd17fec
CW		 1412/*
1413 * rdev_size_change for 0.90.0
1414 */
1415static unsigned long long
3cb03002 1416super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1417{
58c0fed4 1418 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1419 return 0; /* component must fit device */
c3d9714e 1420 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1421 return 0; /* can't move bitmap */
57b2caa3 1422 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1423 if (!num_sectors || num_sectors > rdev->sb_start)
1424 num_sectors = rdev->sb_start;
27a7b260
N		 1425 /* Limit to 4TB as metadata cannot record more than that.
1426 * 4TB == 2^32 KB, or 2*2^32 sectors.
1427 */
1428 if (num_sectors >= (2ULL << 32))
1429 num_sectors = (2ULL << 32) - 2;
0f420358 1430 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1431 rdev->sb_page);
1432 md_super_wait(rdev->mddev);
c26a44ed 1433 return num_sectors;
0cd17fec
CW		 1434}
1435
1436
1da177e4
LT		 1437/*
1438 * version 1 superblock
1439 */
1440
1c05b4bc 1441static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1442{
1c05b4bc
N		 1443 __le32 disk_csum;
1444 u32 csum;
1da177e4
LT		 1445 unsigned long long newcsum;
1446 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1447 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1448 int i;
1449
1450 disk_csum = sb->sb_csum;
1451 sb->sb_csum = 0;
1452 newcsum = 0;
1453 for (i=0; size>=4; size -= 4 )
1454 newcsum += le32_to_cpu(*isuper++);
1455
1456 if (size == 2)
1c05b4bc 1457 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1458
1459 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1460 sb->sb_csum = disk_csum;
1461 return cpu_to_le32(csum);
1462}
1463
2699b672
N		 1464static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1465 int acknowledged);
3cb03002 1466static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1467{
1468 struct mdp_superblock_1 *sb;
1469 int ret;
0f420358 1470 sector_t sb_start;
1da177e4 1471 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1472 int bmask;
1da177e4
LT		 1473
1474 /*
0f420358 1475 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1476 * It is always aligned to a 4K boundary and
1477 * depeding on minor_version, it can be:
1478 * 0: At least 8K, but less than 12K, from end of device
1479 * 1: At start of device
1480 * 2: 4K from start of device.
1481 */
1482 switch(minor_version) {
1483 case 0:
77304d2a 1484 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1485 sb_start -= 8*2;
1486 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1487 break;
1488 case 1:
0f420358 1489 sb_start = 0;
1da177e4
LT		 1490 break;
1491 case 2:
0f420358 1492 sb_start = 8;
1da177e4
LT		 1493 break;
1494 default:
1495 return -EINVAL;
1496 }
0f420358 1497 rdev->sb_start = sb_start;
1da177e4 1498
0002b271
N		 1499 /* superblock is rarely larger than 1K, but it can be larger,
1500 * and it is safe to read 4k, so we do that
1501 */
1502 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1503 if (ret) return ret;
1504
1505
65a06f06 1506 sb = page_address(rdev->sb_page);
1da177e4
LT		 1507
1508 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1509 sb->major_version != cpu_to_le32(1) ||
1510 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1511 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1512 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1513 return -EINVAL;
1514
1515 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1516 printk("md: invalid superblock checksum on %s\n",
1517 bdevname(rdev->bdev,b));
1518 return -EINVAL;
1519 }
1520 if (le64_to_cpu(sb->data_size) < 10) {
1521 printk("md: data_size too small on %s\n",
1522 bdevname(rdev->bdev,b));
1523 return -EINVAL;
1524 }
e11e93fa 1525
1da177e4
LT		 1526 rdev->preferred_minor = 0xffff;
1527 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1528 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1529
0002b271 1530 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1531 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1532 if (rdev->sb_size & bmask)
a1801f85
N		 1533 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1534
1535 if (minor_version
0f420358 1536 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1537 return -EINVAL;
0002b271 1538
31b65a0d
N		 1539 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1540 rdev->desc_nr = -1;
1541 else
1542 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1543
2699b672
N		 1544 if (!rdev->bb_page) {
1545 rdev->bb_page = alloc_page(GFP_KERNEL);
1546 if (!rdev->bb_page)
1547 return -ENOMEM;
1548 }
1549 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1550 rdev->badblocks.count == 0) {
1551 /* need to load the bad block list.
1552 * Currently we limit it to one page.
1553 */
1554 s32 offset;
1555 sector_t bb_sector;
1556 u64 *bbp;
1557 int i;
1558 int sectors = le16_to_cpu(sb->bblog_size);
1559 if (sectors > (PAGE_SIZE / 512))
1560 return -EINVAL;
1561 offset = le32_to_cpu(sb->bblog_offset);
1562 if (offset == 0)
1563 return -EINVAL;
1564 bb_sector = (long long)offset;
1565 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1566 rdev->bb_page, READ, true))
1567 return -EIO;
1568 bbp = (u64 *)page_address(rdev->bb_page);
1569 rdev->badblocks.shift = sb->bblog_shift;
1570 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1571 u64 bb = le64_to_cpu(*bbp);
1572 int count = bb & (0x3ff);
1573 u64 sector = bb >> 10;
1574 sector <<= sb->bblog_shift;
1575 count <<= sb->bblog_shift;
1576 if (bb + 1 == 0)
1577 break;
1578 if (md_set_badblocks(&rdev->badblocks,
1579 sector, count, 1) == 0)
1580 return -EINVAL;
1581 }
1582 } else if (sb->bblog_offset == 0)
1583 rdev->badblocks.shift = -1;
1584
9a7b2b0f 1585 if (!refdev) {
8ed75463 1586 ret = 1;
9a7b2b0f 1587 } else {
1da177e4 1588 __u64 ev1, ev2;
65a06f06 1589 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1590
1591 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1592 sb->level != refsb->level ||
1593 sb->layout != refsb->layout ||
1594 sb->chunksize != refsb->chunksize) {
1595 printk(KERN_WARNING "md: %s has strangely different"
1596 " superblock to %s\n",
1597 bdevname(rdev->bdev,b),
1598 bdevname(refdev->bdev,b2));
1599 return -EINVAL;
1600 }
1601 ev1 = le64_to_cpu(sb->events);
1602 ev2 = le64_to_cpu(refsb->events);
1603
1604 if (ev1 > ev2)
8ed75463
N		 1605 ret = 1;
1606 else
1607 ret = 0;
1da177e4 1608 }
a1801f85 1609 if (minor_version)
77304d2a 1610 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 1611 le64_to_cpu(sb->data_offset);
1da177e4 1612 else
dd8ac336
AN		 1613 rdev->sectors = rdev->sb_start;
1614 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1615 return -EINVAL;
dd8ac336 1616 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1617 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1618 return -EINVAL;
8ed75463 1619 return ret;
1da177e4
LT		 1620}
1621
fd01b88c 1622static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1623{
65a06f06 1624 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1625 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1626
41158c7e 1627 rdev->raid_disk = -1;
c5d79adb
N		 1628 clear_bit(Faulty, &rdev->flags);
1629 clear_bit(In_sync, &rdev->flags);
1630 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1631
1da177e4
LT		 1632 if (mddev->raid_disks == 0) {
1633 mddev->major_version = 1;
1634 mddev->patch_version = 0;
e691063a 1635 mddev->external = 0;
9d8f0363 1636 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1637 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1638 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1639 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1640 mddev->clevel[0] = 0;
1da177e4
LT		 1641 mddev->layout = le32_to_cpu(sb->layout);
1642 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1643 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1644 mddev->events = ev1;
c3d9714e
N		 1645 mddev->bitmap_info.offset = 0;
1646 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1647
1648 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1649 memcpy(mddev->uuid, sb->set_uuid, 16);
1650
1651 mddev->max_disks = (4096-256)/2;
a654b9d8 1652
71c0805c 1653 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1654 mddev->bitmap_info.file == NULL )
1655 mddev->bitmap_info.offset =
1656 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1657
f6705578
N		 1658 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1659 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1660 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1661 mddev->new_level = le32_to_cpu(sb->new_level);
1662 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1663 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1664 } else {
1665 mddev->reshape_position = MaxSector;
1666 mddev->delta_disks = 0;
1667 mddev->new_level = mddev->level;
1668 mddev->new_layout = mddev->layout;
664e7c41 1669 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1670 }
1671
41158c7e 1672 } else if (mddev->pers == NULL) {
be6800a7
N		 1673 /* Insist of good event counter while assembling, except for
1674 * spares (which don't need an event count) */
1da177e4 1675 ++ev1;
be6800a7
N		 1676 if (rdev->desc_nr >= 0 &&
1677 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1678 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1679 if (ev1 < mddev->events)
1680 return -EINVAL;
41158c7e
N		 1681 } else if (mddev->bitmap) {
1682 /* If adding to array with a bitmap, then we can accept an
1683 * older device, but not too old.
1684 */
41158c7e
N		 1685 if (ev1 < mddev->bitmap->events_cleared)
1686 return 0;
07d84d10
N		 1687 } else {
1688 if (ev1 < mddev->events)
1689 /* just a hot-add of a new device, leave raid_disk at -1 */
1690 return 0;
1691 }
1da177e4
LT		 1692 if (mddev->level != LEVEL_MULTIPATH) {
1693 int role;
3673f305
N		 1694 if (rdev->desc_nr < 0 ||
1695 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1696 role = 0xffff;
1697 rdev->desc_nr = -1;
1698 } else
1699 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1700 switch(role) {
1701 case 0xffff: /* spare */
1da177e4
LT		 1702 break;
1703 case 0xfffe: /* faulty */
b2d444d7 1704 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1705 break;
1706 default:
5fd6c1dc
N		 1707 if ((le32_to_cpu(sb->feature_map) &
1708 MD_FEATURE_RECOVERY_OFFSET))
1709 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1710 else
1711 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1712 rdev->raid_disk = role;
1713 break;
1714 }
8ddf9efe
N		 1715 if (sb->devflags & WriteMostly1)
1716 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1717 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1718 set_bit(Replacement, &rdev->flags);
41158c7e 1719 } else /* MULTIPATH are always insync */
b2d444d7 1720 set_bit(In_sync, &rdev->flags);
41158c7e 1721
1da177e4
LT		 1722 return 0;
1723}
1724
fd01b88c 1725static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1726{
1727 struct mdp_superblock_1 *sb;
3cb03002 1728 struct md_rdev *rdev2;
1da177e4
LT		 1729 int max_dev, i;
1730 /* make rdev->sb match mddev and rdev data. */
1731
65a06f06 1732 sb = page_address(rdev->sb_page);
1da177e4
LT		 1733
1734 sb->feature_map = 0;
1735 sb->pad0 = 0;
5fd6c1dc 1736 sb->recovery_offset = cpu_to_le64(0);
1da177e4 1737 memset(sb->pad1, 0, sizeof(sb->pad1));
1da177e4
LT		 1738 memset(sb->pad3, 0, sizeof(sb->pad3));
1739
1740 sb->utime = cpu_to_le64((__u64)mddev->utime);
1741 sb->events = cpu_to_le64(mddev->events);
1742 if (mddev->in_sync)
1743 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1744 else
1745 sb->resync_offset = cpu_to_le64(0);
1746
1c05b4bc 1747 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1748
f0ca340c 1749 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1750 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1751 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1752 sb->level = cpu_to_le32(mddev->level);
1753 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1754
aeb9b211
N		 1755 if (test_bit(WriteMostly, &rdev->flags))
1756 sb->devflags |= WriteMostly1;
1757 else
1758 sb->devflags &= ~WriteMostly1;
1759
c3d9714e
N		 1760 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1761 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1762 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1763 }
5fd6c1dc
N		 1764
1765 if (rdev->raid_disk >= 0 &&
97e4f42d 1766 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1767 sb->feature_map |=
1768 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1769 sb->recovery_offset =
1770 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc 1771 }
2d78f8c4
N		 1772 if (test_bit(Replacement, &rdev->flags))
1773 sb->feature_map |=
1774 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1775
f6705578
N		 1776 if (mddev->reshape_position != MaxSector) {
1777 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1778 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1779 sb->new_layout = cpu_to_le32(mddev->new_layout);
1780 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1781 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1782 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1783 }
a654b9d8 1784
2699b672
N		 1785 if (rdev->badblocks.count == 0)
1786 /* Nothing to do for bad blocks*/ ;
1787 else if (sb->bblog_offset == 0)
1788 /* Cannot record bad blocks on this device */
1789 md_error(mddev, rdev);
1790 else {
1791 struct badblocks *bb = &rdev->badblocks;
1792 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1793 u64 *p = bb->page;
1794 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1795 if (bb->changed) {
1796 unsigned seq;
1797
1798retry:
1799 seq = read_seqbegin(&bb->lock);
1800
1801 memset(bbp, 0xff, PAGE_SIZE);
1802
1803 for (i = 0 ; i < bb->count ; i++) {
1804 u64 internal_bb = *p++;
1805 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1806 | BB_LEN(internal_bb));
1807 *bbp++ = cpu_to_le64(store_bb);
1808 }
1809 if (read_seqretry(&bb->lock, seq))
1810 goto retry;
1811
1812 bb->sector = (rdev->sb_start +
1813 (int)le32_to_cpu(sb->bblog_offset));
1814 bb->size = le16_to_cpu(sb->bblog_size);
1815 bb->changed = 0;
1816 }
1817 }
1818
1da177e4 1819 max_dev = 0;
159ec1fc 1820 list_for_each_entry(rdev2, &mddev->disks, same_set)
1da177e4
LT		 1821 if (rdev2->desc_nr+1 > max_dev)
1822 max_dev = rdev2->desc_nr+1;
a778b73f 1823
70471daf
N		 1824 if (max_dev > le32_to_cpu(sb->max_dev)) {
1825 int bmask;
a778b73f 1826 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1827 rdev->sb_size = max_dev * 2 + 256;
1828 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1829 if (rdev->sb_size & bmask)
1830 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1831 } else
1832 max_dev = le32_to_cpu(sb->max_dev);
1833
1da177e4
LT		 1834 for (i=0; i<max_dev;i++)
1835 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1836
159ec1fc 1837 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1838 i = rdev2->desc_nr;
b2d444d7 1839 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1840 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1841 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1842 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1843 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1844 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1845 else
1846 sb->dev_roles[i] = cpu_to_le16(0xffff);
1847 }
1848
1da177e4
LT		 1849 sb->sb_csum = calc_sb_1_csum(sb);
1850}
1851
0cd17fec 1852static unsigned long long
3cb03002 1853super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1854{
1855 struct mdp_superblock_1 *sb;
15f4a5fd 1856 sector_t max_sectors;
58c0fed4 1857 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1858 return 0; /* component must fit device */
0f420358 1859 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1860 /* minor versions 1 and 2; superblock before data */
77304d2a 1861 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1862 max_sectors -= rdev->data_offset;
1863 if (!num_sectors || num_sectors > max_sectors)
1864 num_sectors = max_sectors;
c3d9714e 1865 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1866 /* minor version 0 with bitmap we can't move */
1867 return 0;
1868 } else {
1869 /* minor version 0; superblock after data */
0f420358 1870 sector_t sb_start;
77304d2a 1871 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1872 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1873 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1874 if (!num_sectors || num_sectors > max_sectors)
1875 num_sectors = max_sectors;
0f420358 1876 rdev->sb_start = sb_start;
0cd17fec 1877 }
65a06f06 1878 sb = page_address(rdev->sb_page);
15f4a5fd 1879 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1880 sb->super_offset = rdev->sb_start;
0cd17fec 1881 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1882 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1883 rdev->sb_page);
1884 md_super_wait(rdev->mddev);
c26a44ed 1885 return num_sectors;
0cd17fec 1886}
1da177e4 1887
75c96f85 1888static struct super_type super_types[] = {
1da177e4
LT		 1889 [0] = {
1890 .name = "0.90.0",
1891 .owner = THIS_MODULE,
0cd17fec
CW		 1892 .load_super = super_90_load,
1893 .validate_super = super_90_validate,
1894 .sync_super = super_90_sync,
1895 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1896 },
1897 [1] = {
1898 .name = "md-1",
1899 .owner = THIS_MODULE,
0cd17fec
CW		 1900 .load_super = super_1_load,
1901 .validate_super = super_1_validate,
1902 .sync_super = super_1_sync,
1903 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1904 },
1905};
1da177e4 1906
fd01b88c 1907static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1908{
1909 if (mddev->sync_super) {
1910 mddev->sync_super(mddev, rdev);
1911 return;
1912 }
1913
1914 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1915
1916 super_types[mddev->major_version].sync_super(mddev, rdev);
1917}
1918
fd01b88c 1919static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1920{
3cb03002 1921 struct md_rdev *rdev, *rdev2;
1da177e4 1922
4b80991c
N		 1923 rcu_read_lock();
1924 rdev_for_each_rcu(rdev, mddev1)
1925 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1926 if (rdev->bdev->bd_contains ==
4b80991c
N		 1927 rdev2->bdev->bd_contains) {
1928 rcu_read_unlock();
7dd5e7c3 1929 return 1;
4b80991c
N		 1930 }
1931 rcu_read_unlock();
1da177e4
LT		 1932 return 0;
1933}
1934
1935static LIST_HEAD(pending_raid_disks);
1936
ac5e7113
AN		 1937/*
1938 * Try to register data integrity profile for an mddev
1939 *
1940 * This is called when an array is started and after a disk has been kicked
1941 * from the array. It only succeeds if all working and active component devices
1942 * are integrity capable with matching profiles.
1943 */
fd01b88c 1944int md_integrity_register(struct mddev *mddev)
ac5e7113 1945{
3cb03002 1946 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1947
1948 if (list_empty(&mddev->disks))
1949 return 0; /* nothing to do */
629acb6a
JB		 1950 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1951 return 0; /* shouldn't register, or already is */
ac5e7113
AN		 1952 list_for_each_entry(rdev, &mddev->disks, same_set) {
1953 /* skip spares and non-functional disks */
1954 if (test_bit(Faulty, &rdev->flags))
1955 continue;
1956 if (rdev->raid_disk < 0)
1957 continue;
ac5e7113
AN		 1958 if (!reference) {
1959 /* Use the first rdev as the reference */
1960 reference = rdev;
1961 continue;
1962 }
1963 /* does this rdev's profile match the reference profile? */
1964 if (blk_integrity_compare(reference->bdev->bd_disk,
1965 rdev->bdev->bd_disk) < 0)
1966 return -EINVAL;
1967 }
89078d57
MP		 1968 if (!reference || !bdev_get_integrity(reference->bdev))
1969 return 0;
ac5e7113
AN		 1970 /*
1971 * All component devices are integrity capable and have matching
1972 * profiles, register the common profile for the md device.
1973 */
1974 if (blk_integrity_register(mddev->gendisk,
1975 bdev_get_integrity(reference->bdev)) != 0) {
1976 printk(KERN_ERR "md: failed to register integrity for %s\n",
1977 mdname(mddev));
1978 return -EINVAL;
1979 }
a91a2785
MP		 1980 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1981 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1982 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1983 mdname(mddev));
1984 return -EINVAL;
1985 }
ac5e7113
AN		 1986 return 0;
1987}
1988EXPORT_SYMBOL(md_integrity_register);
1989
1990/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 1991void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 1992{
3f9d99c1 1993 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1994 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1995
ac5e7113 1996 if (!bi_mddev) /* nothing to do */
3f9d99c1 1997 return;
ac5e7113 1998 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1999 return;
ac5e7113
AN		 2000 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2001 rdev->bdev->bd_disk) >= 0)
2002 return;
2003 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2004 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2005}
ac5e7113 2006EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2007
fd01b88c 2008static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
1da177e4 2009{
7dd5e7c3 2010 char b[BDEVNAME_SIZE];
f637b9f9 2011 struct kobject *ko;
1edf80d3 2012 char *s;
5e55e2f5 2013 int err;
1da177e4
LT		 2014
2015 if (rdev->mddev) {
2016 MD_BUG();
2017 return -EINVAL;
2018 }
11e2ede0
DW		 2019
2020 /* prevent duplicates */
2021 if (find_rdev(mddev, rdev->bdev->bd_dev))
2022 return -EEXIST;
2023
dd8ac336
AN		 2024 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2025 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2026 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2027 if (mddev->pers) {
2028 /* Cannot change size, so fail
2029 * If mddev->level <= 0, then we don't care
2030 * about aligning sizes (e.g. linear)
2031 */
2032 if (mddev->level > 0)
2033 return -ENOSPC;
2034 } else
dd8ac336 2035 mddev->dev_sectors = rdev->sectors;
2bf071bf 2036 }
1da177e4
LT		 2037
2038 /* Verify rdev->desc_nr is unique.
2039 * If it is -1, assign a free number, else
2040 * check number is not in use
2041 */
2042 if (rdev->desc_nr < 0) {
2043 int choice = 0;
2044 if (mddev->pers) choice = mddev->raid_disks;
2045 while (find_rdev_nr(mddev, choice))
2046 choice++;
2047 rdev->desc_nr = choice;
2048 } else {
2049 if (find_rdev_nr(mddev, rdev->desc_nr))
2050 return -EBUSY;
2051 }
de01dfad
N		 2052 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2053 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2054 mdname(mddev), mddev->max_disks);
2055 return -EBUSY;
2056 }
19133a42 2057 bdevname(rdev->bdev,b);
649316b2 2058 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2059 *s = '!';
649316b2 2060
1da177e4 2061 rdev->mddev = mddev;
19133a42 2062 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2063
b2d6db58 2064 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2065 goto fail;
86e6ffdd 2066
0762b8bd 2067 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2068 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2069 /* failure here is OK */;
2070 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2071
4b80991c 2072 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2073 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2074
2075 /* May as well allow recovery to be retried once */
5389042f 2076 mddev->recovery_disabled++;
3f9d99c1 2077
1da177e4 2078 return 0;
5e55e2f5
N		 2079
2080 fail:
2081 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2082 b, mdname(mddev));
2083 return err;
1da177e4
LT		 2084}
2085
177a99b2 2086static void md_delayed_delete(struct work_struct *ws)
5792a285 2087{
3cb03002 2088 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2089 kobject_del(&rdev->kobj);
177a99b2 2090 kobject_put(&rdev->kobj);
5792a285
N		 2091}
2092
3cb03002 2093static void unbind_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2094{
2095 char b[BDEVNAME_SIZE];
2096 if (!rdev->mddev) {
2097 MD_BUG();
2098 return;
2099 }
49731baa 2100 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2101 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2102 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2103 rdev->mddev = NULL;
86e6ffdd 2104 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2105 sysfs_put(rdev->sysfs_state);
2106 rdev->sysfs_state = NULL;
2230dfe4
N		 2107 kfree(rdev->badblocks.page);
2108 rdev->badblocks.count = 0;
2109 rdev->badblocks.page = NULL;
5792a285 2110 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2111 * writing to 'remove' to "dev/state". We also need
2112 * to delay it due to rcu usage.
5792a285 2113 */
4b80991c 2114 synchronize_rcu();
177a99b2
N		 2115 INIT_WORK(&rdev->del_work, md_delayed_delete);
2116 kobject_get(&rdev->kobj);
e804ac78 2117 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2118}
2119
2120/*
2121 * prevent the device from being mounted, repartitioned or
2122 * otherwise reused by a RAID array (or any other kernel
2123 * subsystem), by bd_claiming the device.
2124 */
3cb03002 2125static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2126{
2127 int err = 0;
2128 struct block_device *bdev;
2129 char b[BDEVNAME_SIZE];
2130
d4d77629 2131 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2132 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2133 if (IS_ERR(bdev)) {
2134 printk(KERN_ERR "md: could not open %s.\n",
2135 __bdevname(dev, b));
2136 return PTR_ERR(bdev);
2137 }
1da177e4
LT		 2138 rdev->bdev = bdev;
2139 return err;
2140}
2141
3cb03002 2142static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2143{
2144 struct block_device *bdev = rdev->bdev;
2145 rdev->bdev = NULL;
2146 if (!bdev)
2147 MD_BUG();
e525fd89 2148 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2149}
2150
2151void md_autodetect_dev(dev_t dev);
2152
3cb03002 2153static void export_rdev(struct md_rdev * rdev)
1da177e4
LT		 2154{
2155 char b[BDEVNAME_SIZE];
2156 printk(KERN_INFO "md: export_rdev(%s)\n",
2157 bdevname(rdev->bdev,b));
2158 if (rdev->mddev)
2159 MD_BUG();
2160 free_disk_sb(rdev);
1da177e4 2161#ifndef MODULE
d0fae18f
N		 2162 if (test_bit(AutoDetected, &rdev->flags))
2163 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2164#endif
2165 unlock_rdev(rdev);
86e6ffdd 2166 kobject_put(&rdev->kobj);
1da177e4
LT		 2167}
2168
3cb03002 2169static void kick_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2170{
2171 unbind_rdev_from_array(rdev);
2172 export_rdev(rdev);
2173}
2174
fd01b88c 2175static void export_array(struct mddev *mddev)
1da177e4 2176{
3cb03002 2177 struct md_rdev *rdev, *tmp;
1da177e4 2178
d089c6af 2179 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 2180 if (!rdev->mddev) {
2181 MD_BUG();
2182 continue;
2183 }
2184 kick_rdev_from_array(rdev);
2185 }
2186 if (!list_empty(&mddev->disks))
2187 MD_BUG();
2188 mddev->raid_disks = 0;
2189 mddev->major_version = 0;
2190}
2191
2192static void print_desc(mdp_disk_t *desc)
2193{
2194 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2195 desc->major,desc->minor,desc->raid_disk,desc->state);
2196}
2197
cd2ac932 2198static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2199{
2200 int i;
2201
2202 printk(KERN_INFO
2203 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2204 sb->major_version, sb->minor_version, sb->patch_version,
2205 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2206 sb->ctime);
2207 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2208 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2209 sb->md_minor, sb->layout, sb->chunk_size);
2210 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2211 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2212 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2213 sb->failed_disks, sb->spare_disks,
2214 sb->sb_csum, (unsigned long)sb->events_lo);
2215
2216 printk(KERN_INFO);
2217 for (i = 0; i < MD_SB_DISKS; i++) {
2218 mdp_disk_t *desc;
2219
2220 desc = sb->disks + i;
2221 if (desc->number || desc->major || desc->minor ||
2222 desc->raid_disk || (desc->state && (desc->state != 4))) {
2223 printk(" D %2d: ", i);
2224 print_desc(desc);
2225 }
2226 }
2227 printk(KERN_INFO "md: THIS: ");
2228 print_desc(&sb->this_disk);
cd2ac932 2229}
1da177e4 2230
cd2ac932
CR		 2231static void print_sb_1(struct mdp_superblock_1 *sb)
2232{
2233 __u8 *uuid;
2234
2235 uuid = sb->set_uuid;
ad361c98 2236 printk(KERN_INFO
7b75c2f8 2237 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2238 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2239 le32_to_cpu(sb->major_version),
2240 le32_to_cpu(sb->feature_map),
7b75c2f8 2241 uuid,
cd2ac932
CR		 2242 sb->set_name,
2243 (unsigned long long)le64_to_cpu(sb->ctime)
2244 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2245
2246 uuid = sb->device_uuid;
ad361c98
JP		 2247 printk(KERN_INFO
2248 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2249 " RO:%llu\n"
7b75c2f8 2250 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2251 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2252 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2253 le32_to_cpu(sb->level),
2254 (unsigned long long)le64_to_cpu(sb->size),
2255 le32_to_cpu(sb->raid_disks),
2256 le32_to_cpu(sb->layout),
2257 le32_to_cpu(sb->chunksize),
2258 (unsigned long long)le64_to_cpu(sb->data_offset),
2259 (unsigned long long)le64_to_cpu(sb->data_size),
2260 (unsigned long long)le64_to_cpu(sb->super_offset),
2261 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2262 le32_to_cpu(sb->dev_number),
7b75c2f8 2263 uuid,
cd2ac932
CR		 2264 sb->devflags,
2265 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2266 (unsigned long long)le64_to_cpu(sb->events),
2267 (unsigned long long)le64_to_cpu(sb->resync_offset),
2268 le32_to_cpu(sb->sb_csum),
2269 le32_to_cpu(sb->max_dev)
2270 );
1da177e4
LT		 2271}
2272
3cb03002 2273static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT		 2274{
2275 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2276 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2277 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2278 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2279 rdev->desc_nr);
1da177e4 2280 if (rdev->sb_loaded) {
cd2ac932
CR		 2281 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2282 switch (major_version) {
2283 case 0:
65a06f06 2284 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR		 2285 break;
2286 case 1:
65a06f06 2287 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR		 2288 break;
2289 }
1da177e4
LT		 2290 } else
2291 printk(KERN_INFO "md: no rdev superblock!\n");
2292}
2293
5e56341d 2294static void md_print_devices(void)
1da177e4 2295{
159ec1fc 2296 struct list_head *tmp;
3cb03002 2297 struct md_rdev *rdev;
fd01b88c 2298 struct mddev *mddev;
1da177e4
LT		 2299 char b[BDEVNAME_SIZE];
2300
2301 printk("\n");
2302 printk("md: **********************************\n");
2303 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2304 printk("md: **********************************\n");
29ac4aa3 2305 for_each_mddev(mddev, tmp) {
1da177e4 2306
32a7627c
N		 2307 if (mddev->bitmap)
2308 bitmap_print_sb(mddev->bitmap);
2309 else
2310 printk("%s: ", mdname(mddev));
159ec1fc 2311 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 2312 printk("<%s>", bdevname(rdev->bdev,b));
2313 printk("\n");
2314
159ec1fc 2315 list_for_each_entry(rdev, &mddev->disks, same_set)
cd2ac932 2316 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2317 }
2318 printk("md: **********************************\n");
2319 printk("\n");
2320}
2321
2322
fd01b88c 2323static void sync_sbs(struct mddev * mddev, int nospares)
1da177e4 2324{
42543769
N		 2325 /* Update each superblock (in-memory image), but
2326 * if we are allowed to, skip spares which already
2327 * have the right event counter, or have one earlier
2328 * (which would mean they aren't being marked as dirty
2329 * with the rest of the array)
2330 */
3cb03002 2331 struct md_rdev *rdev;
159ec1fc 2332 list_for_each_entry(rdev, &mddev->disks, same_set) {
42543769
N		 2333 if (rdev->sb_events == mddev->events ||
2334 (nospares &&
2335 rdev->raid_disk < 0 &&
42543769
N		 2336 rdev->sb_events+1 == mddev->events)) {
2337 /* Don't update this superblock */
2338 rdev->sb_loaded = 2;
2339 } else {
076f968b 2340 sync_super(mddev, rdev);
42543769
N		 2341 rdev->sb_loaded = 1;
2342 }
1da177e4
LT		 2343 }
2344}
2345
fd01b88c 2346static void md_update_sb(struct mddev * mddev, int force_change)
1da177e4 2347{
3cb03002 2348 struct md_rdev *rdev;
06d91a5f 2349 int sync_req;
42543769 2350 int nospares = 0;
2699b672 2351 int any_badblocks_changed = 0;
1da177e4 2352
1da177e4 2353repeat:
3a3a5ddb
N		 2354 /* First make sure individual recovery_offsets are correct */
2355 list_for_each_entry(rdev, &mddev->disks, same_set) {
2356 if (rdev->raid_disk >= 0 &&
2357 mddev->delta_disks >= 0 &&
2358 !test_bit(In_sync, &rdev->flags) &&
2359 mddev->curr_resync_completed > rdev->recovery_offset)
2360 rdev->recovery_offset = mddev->curr_resync_completed;
2361
2362 }
bd52b746 2363 if (!mddev->persistent) {
070dc6dd 2364 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2365 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2366 if (!mddev->external) {
d97a41dc 2367 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
de393cde
N		 2368 list_for_each_entry(rdev, &mddev->disks, same_set) {
2369 if (rdev->badblocks.changed) {
2370 md_ack_all_badblocks(&rdev->badblocks);
2371 md_error(mddev, rdev);
2372 }
2373 clear_bit(Blocked, &rdev->flags);
2374 clear_bit(BlockedBadBlocks, &rdev->flags);
2375 wake_up(&rdev->blocked_wait);
2376 }
2377 }
3a3a5ddb
N		 2378 wake_up(&mddev->sb_wait);
2379 return;
2380 }
2381
a9701a30 2382 spin_lock_irq(&mddev->write_lock);
84692195 2383
3a3a5ddb
N		 2384 mddev->utime = get_seconds();
2385
850b2b42
N		 2386 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2387 force_change = 1;
2388 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2389 /* just a clean<-> dirty transition, possibly leave spares alone,
2390 * though if events isn't the right even/odd, we will have to do
2391 * spares after all
2392 */
2393 nospares = 1;
2394 if (force_change)
2395 nospares = 0;
2396 if (mddev->degraded)
84692195
N		 2397 /* If the array is degraded, then skipping spares is both
2398 * dangerous and fairly pointless.
2399 * Dangerous because a device that was removed from the array
2400 * might have a event_count that still looks up-to-date,
2401 * so it can be re-added without a resync.
2402 * Pointless because if there are any spares to skip,
2403 * then a recovery will happen and soon that array won't
2404 * be degraded any more and the spare can go back to sleep then.
2405 */
850b2b42 2406 nospares = 0;
84692195 2407
06d91a5f 2408 sync_req = mddev->in_sync;
42543769
N		 2409
2410 /* If this is just a dirty<->clean transition, and the array is clean
2411 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2412 if (nospares
42543769 2413 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2414 && mddev->can_decrease_events
2415 && mddev->events != 1) {
42543769 2416 mddev->events--;
a8707c08
N		 2417 mddev->can_decrease_events = 0;
2418 } else {
42543769
N		 2419 /* otherwise we have to go forward and ... */
2420 mddev->events ++;
a8707c08 2421 mddev->can_decrease_events = nospares;
42543769 2422 }
1da177e4
LT		 2423
2424 if (!mddev->events) {
2425 /*
2426 * oops, this 64-bit counter should never wrap.
2427 * Either we are in around ~1 trillion A.C., assuming
2428 * 1 reboot per second, or we have a bug:
2429 */
2430 MD_BUG();
2431 mddev->events --;
2432 }
2699b672 2433
de393cde 2434 list_for_each_entry(rdev, &mddev->disks, same_set) {
2699b672
N		 2435 if (rdev->badblocks.changed)
2436 any_badblocks_changed++;
de393cde
N		 2437 if (test_bit(Faulty, &rdev->flags))
2438 set_bit(FaultRecorded, &rdev->flags);
2439 }
2699b672 2440
e691063a 2441 sync_sbs(mddev, nospares);
a9701a30 2442 spin_unlock_irq(&mddev->write_lock);
1da177e4 2443
36a4e1fe
N		 2444 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2445 mdname(mddev), mddev->in_sync);
1da177e4 2446
4ad13663 2447 bitmap_update_sb(mddev->bitmap);
159ec1fc 2448 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 2449 char b[BDEVNAME_SIZE];
36a4e1fe 2450
42543769
N		 2451 if (rdev->sb_loaded != 1)
2452 continue; /* no noise on spare devices */
1da177e4 2453
d70ed2e4
AW		 2454 if (!test_bit(Faulty, &rdev->flags) &&
2455 rdev->saved_raid_disk == -1) {
7bfa19f2 2456 md_super_write(mddev,rdev,
0f420358 2457 rdev->sb_start, rdev->sb_size,
7bfa19f2 2458 rdev->sb_page);
36a4e1fe
N		 2459 pr_debug("md: (write) %s's sb offset: %llu\n",
2460 bdevname(rdev->bdev, b),
2461 (unsigned long long)rdev->sb_start);
42543769 2462 rdev->sb_events = mddev->events;
2699b672
N		 2463 if (rdev->badblocks.size) {
2464 md_super_write(mddev, rdev,
2465 rdev->badblocks.sector,
2466 rdev->badblocks.size << 9,
2467 rdev->bb_page);
2468 rdev->badblocks.size = 0;
2469 }
7bfa19f2 2470
d70ed2e4 2471 } else if (test_bit(Faulty, &rdev->flags))
36a4e1fe
N		 2472 pr_debug("md: %s (skipping faulty)\n",
2473 bdevname(rdev->bdev, b));
d70ed2e4
AW		 2474 else
2475 pr_debug("(skipping incremental s/r ");
2476
7bfa19f2 2477 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2478 /* only need to write one superblock... */
2479 break;
2480 }
a9701a30 2481 md_super_wait(mddev);
850b2b42 2482 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2483
a9701a30 2484 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2485 if (mddev->in_sync != sync_req ||
2486 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2487 /* have to write it out again */
a9701a30 2488 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2489 goto repeat;
2490 }
850b2b42 2491 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2492 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2493 wake_up(&mddev->sb_wait);
acb180b0
N		 2494 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2495 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2496
de393cde
N		 2497 list_for_each_entry(rdev, &mddev->disks, same_set) {
2498 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2499 clear_bit(Blocked, &rdev->flags);
2500
2501 if (any_badblocks_changed)
2699b672 2502 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2503 clear_bit(BlockedBadBlocks, &rdev->flags);
2504 wake_up(&rdev->blocked_wait);
2505 }
1da177e4
LT		 2506}
2507
7f6ce769 2508/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2509 * We want to accept with case. For this we use cmd_match.
2510 */
2511static int cmd_match(const char *cmd, const char *str)
2512{
2513 /* See if cmd, written into a sysfs file, matches
2514 * str. They must either be the same, or cmd can
2515 * have a trailing newline
2516 */
2517 while (*cmd && *str && *cmd == *str) {
2518 cmd++;
2519 str++;
2520 }
2521 if (*cmd == '\n')
2522 cmd++;
2523 if (*str || *cmd)
2524 return 0;
2525 return 1;
2526}
2527
86e6ffdd
N		 2528struct rdev_sysfs_entry {
2529 struct attribute attr;
3cb03002
N		 2530 ssize_t (*show)(struct md_rdev *, char *);
2531 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2532};
2533
2534static ssize_t
3cb03002 2535state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2536{
2537 char *sep = "";
20a49ff6 2538 size_t len = 0;
86e6ffdd 2539
de393cde
N		 2540 if (test_bit(Faulty, &rdev->flags) ||
2541 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2542 len+= sprintf(page+len, "%sfaulty",sep);
2543 sep = ",";
2544 }
b2d444d7 2545 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2546 len += sprintf(page+len, "%sin_sync",sep);
2547 sep = ",";
2548 }
f655675b
N		 2549 if (test_bit(WriteMostly, &rdev->flags)) {
2550 len += sprintf(page+len, "%swrite_mostly",sep);
2551 sep = ",";
2552 }
de393cde 2553 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N		 2554 (rdev->badblocks.unacked_exist
2555 && !test_bit(Faulty, &rdev->flags))) {
6bfe0b49
DW		 2556 len += sprintf(page+len, "%sblocked", sep);
2557 sep = ",";
2558 }
b2d444d7
N		 2559 if (!test_bit(Faulty, &rdev->flags) &&
2560 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2561 len += sprintf(page+len, "%sspare", sep);
2562 sep = ",";
2563 }
d7a9d443
N		 2564 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2565 len += sprintf(page+len, "%swrite_error", sep);
2566 sep = ",";
2567 }
2d78f8c4
N		 2568 if (test_bit(WantReplacement, &rdev->flags)) {
2569 len += sprintf(page+len, "%swant_replacement", sep);
2570 sep = ",";
2571 }
2572 if (test_bit(Replacement, &rdev->flags)) {
2573 len += sprintf(page+len, "%sreplacement", sep);
2574 sep = ",";
2575 }
2576
86e6ffdd
N		 2577 return len+sprintf(page+len, "\n");
2578}
2579
45dc2de1 2580static ssize_t
3cb03002 2581state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2582{
2583 /* can write
de393cde 2584 * faulty - simulates an error
45dc2de1 2585 * remove - disconnects the device
f655675b
N		 2586 * writemostly - sets write_mostly
2587 * -writemostly - clears write_mostly
de393cde
N		 2588 * blocked - sets the Blocked flags
2589 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2590 * insync - sets Insync providing device isn't active
d7a9d443
N		 2591 * write_error - sets WriteErrorSeen
2592 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2593 */
2594 int err = -EINVAL;
2595 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2596 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2597 if (test_bit(Faulty, &rdev->flags))
2598 err = 0;
2599 else
2600 err = -EBUSY;
45dc2de1
N		 2601 } else if (cmd_match(buf, "remove")) {
2602 if (rdev->raid_disk >= 0)
2603 err = -EBUSY;
2604 else {
fd01b88c 2605 struct mddev *mddev = rdev->mddev;
45dc2de1 2606 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2607 if (mddev->pers)
2608 md_update_sb(mddev, 1);
45dc2de1
N		 2609 md_new_event(mddev);
2610 err = 0;
2611 }
f655675b
N		 2612 } else if (cmd_match(buf, "writemostly")) {
2613 set_bit(WriteMostly, &rdev->flags);
2614 err = 0;
2615 } else if (cmd_match(buf, "-writemostly")) {
2616 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2617 err = 0;
2618 } else if (cmd_match(buf, "blocked")) {
2619 set_bit(Blocked, &rdev->flags);
2620 err = 0;
2621 } else if (cmd_match(buf, "-blocked")) {
de393cde 2622 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2623 rdev->badblocks.unacked_exist) {
de393cde
N		 2624 /* metadata handler doesn't understand badblocks,
2625 * so we need to fail the device
2626 */
2627 md_error(rdev->mddev, rdev);
2628 }
6bfe0b49 2629 clear_bit(Blocked, &rdev->flags);
de393cde 2630 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2631 wake_up(&rdev->blocked_wait);
2632 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2633 md_wakeup_thread(rdev->mddev->thread);
2634
6d56e278
N		 2635 err = 0;
2636 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2637 set_bit(In_sync, &rdev->flags);
f655675b 2638 err = 0;
d7a9d443
N		 2639 } else if (cmd_match(buf, "write_error")) {
2640 set_bit(WriteErrorSeen, &rdev->flags);
2641 err = 0;
2642 } else if (cmd_match(buf, "-write_error")) {
2643 clear_bit(WriteErrorSeen, &rdev->flags);
2644 err = 0;
2d78f8c4
N		 2645 } else if (cmd_match(buf, "want_replacement")) {
2646 /* Any non-spare device that is not a replacement can
2647 * become want_replacement at any time, but we then need to
2648 * check if recovery is needed.
2649 */
2650 if (rdev->raid_disk >= 0 &&
2651 !test_bit(Replacement, &rdev->flags))
2652 set_bit(WantReplacement, &rdev->flags);
2653 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2654 md_wakeup_thread(rdev->mddev->thread);
2655 err = 0;
2656 } else if (cmd_match(buf, "-want_replacement")) {
2657 /* Clearing 'want_replacement' is always allowed.
2658 * Once replacements starts it is too late though.
2659 */
2660 err = 0;
2661 clear_bit(WantReplacement, &rdev->flags);
2662 } else if (cmd_match(buf, "replacement")) {
2663 /* Can only set a device as a replacement when array has not
2664 * yet been started. Once running, replacement is automatic
2665 * from spares, or by assigning 'slot'.
2666 */
2667 if (rdev->mddev->pers)
2668 err = -EBUSY;
2669 else {
2670 set_bit(Replacement, &rdev->flags);
2671 err = 0;
2672 }
2673 } else if (cmd_match(buf, "-replacement")) {
2674 /* Similarly, can only clear Replacement before start */
2675 if (rdev->mddev->pers)
2676 err = -EBUSY;
2677 else {
2678 clear_bit(Replacement, &rdev->flags);
2679 err = 0;
2680 }
45dc2de1 2681 }
00bcb4ac
N		 2682 if (!err)
2683 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2684 return err ? err : len;
2685}
80ca3a44
N		 2686static struct rdev_sysfs_entry rdev_state =
2687__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2688
4dbcdc75 2689static ssize_t
3cb03002 2690errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2691{
2692 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2693}
2694
2695static ssize_t
3cb03002 2696errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2697{
2698 char *e;
2699 unsigned long n = simple_strtoul(buf, &e, 10);
2700 if (*buf && (*e == 0 || *e == '\n')) {
2701 atomic_set(&rdev->corrected_errors, n);
2702 return len;
2703 }
2704 return -EINVAL;
2705}
2706static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2707__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2708
014236d2 2709static ssize_t
3cb03002 2710slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2711{
2712 if (rdev->raid_disk < 0)
2713 return sprintf(page, "none\n");
2714 else
2715 return sprintf(page, "%d\n", rdev->raid_disk);
2716}
2717
2718static ssize_t
3cb03002 2719slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2720{
2721 char *e;
c303da6d 2722 int err;
014236d2
N		 2723 int slot = simple_strtoul(buf, &e, 10);
2724 if (strncmp(buf, "none", 4)==0)
2725 slot = -1;
2726 else if (e==buf || (*e && *e!= '\n'))
2727 return -EINVAL;
6c2fce2e 2728 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2729 /* Setting 'slot' on an active array requires also
2730 * updating the 'rd%d' link, and communicating
2731 * with the personality with ->hot_*_disk.
2732 * For now we only support removing
2733 * failed/spare devices. This normally happens automatically,
2734 * but not when the metadata is externally managed.
2735 */
c303da6d
N		 2736 if (rdev->raid_disk == -1)
2737 return -EEXIST;
2738 /* personality does all needed checks */
01393f3d 2739 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d
N		 2740 return -EINVAL;
2741 err = rdev->mddev->pers->
b8321b68 2742 hot_remove_disk(rdev->mddev, rdev);
c303da6d
N		 2743 if (err)
2744 return err;
36fad858 2745 sysfs_unlink_rdev(rdev->mddev, rdev);
b7103107 2746 rdev->raid_disk = -1;
c303da6d
N		 2747 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2748 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2749 } else if (rdev->mddev->pers) {
6c2fce2e 2750 /* Activating a spare .. or possibly reactivating
6d56e278 2751 * if we ever get bitmaps working here.
6c2fce2e
NB		 2752 */
2753
2754 if (rdev->raid_disk != -1)
2755 return -EBUSY;
2756
c6751b2b
N		 2757 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2758 return -EBUSY;
2759
6c2fce2e
NB		 2760 if (rdev->mddev->pers->hot_add_disk == NULL)
2761 return -EINVAL;
2762
ba1b41b6
N		 2763 if (slot >= rdev->mddev->raid_disks &&
2764 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2765 return -ENOSPC;
2766
6c2fce2e
NB		 2767 rdev->raid_disk = slot;
2768 if (test_bit(In_sync, &rdev->flags))
2769 rdev->saved_raid_disk = slot;
2770 else
2771 rdev->saved_raid_disk = -1;
d30519fc 2772 clear_bit(In_sync, &rdev->flags);
6c2fce2e
NB		 2773 err = rdev->mddev->pers->
2774 hot_add_disk(rdev->mddev, rdev);
199050ea 2775 if (err) {
6c2fce2e 2776 rdev->raid_disk = -1;
6c2fce2e 2777 return err;
52664732 2778 } else
00bcb4ac 2779 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2780 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2781 /* failure here is OK */;
6c2fce2e 2782 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2783 } else {
ba1b41b6
N		 2784 if (slot >= rdev->mddev->raid_disks &&
2785 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2786 return -ENOSPC;
2787 rdev->raid_disk = slot;
2788 /* assume it is working */
c5d79adb
N		 2789 clear_bit(Faulty, &rdev->flags);
2790 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2791 set_bit(In_sync, &rdev->flags);
00bcb4ac 2792 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2793 }
014236d2
N		 2794 return len;
2795}
2796
2797
2798static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2799__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2800
93c8cad0 2801static ssize_t
3cb03002 2802offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2803{
6961ece4 2804 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2805}
2806
2807static ssize_t
3cb03002 2808offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0
N		 2809{
2810 char *e;
2811 unsigned long long offset = simple_strtoull(buf, &e, 10);
2812 if (e==buf || (*e && *e != '\n'))
2813 return -EINVAL;
8ed0a521 2814 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2815 return -EBUSY;
dd8ac336 2816 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2817 /* Must set offset before size, so overlap checks
2818 * can be sane */
2819 return -EBUSY;
93c8cad0
N		 2820 rdev->data_offset = offset;
2821 return len;
2822}
2823
2824static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2825__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2826
83303b61 2827static ssize_t
3cb03002 2828rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2829{
dd8ac336 2830 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2831}
2832
c5d79adb
N		 2833static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2834{
2835 /* check if two start/length pairs overlap */
2836 if (s1+l1 <= s2)
2837 return 0;
2838 if (s2+l2 <= s1)
2839 return 0;
2840 return 1;
2841}
2842
b522adcd
DW		 2843static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2844{
2845 unsigned long long blocks;
2846 sector_t new;
2847
2848 if (strict_strtoull(buf, 10, &blocks) < 0)
2849 return -EINVAL;
2850
2851 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2852 return -EINVAL; /* sector conversion overflow */
2853
2854 new = blocks * 2;
2855 if (new != blocks * 2)
2856 return -EINVAL; /* unsigned long long to sector_t overflow */
2857
2858 *sectors = new;
2859 return 0;
2860}
2861
83303b61 2862static ssize_t
3cb03002 2863rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2864{
fd01b88c 2865 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2866 sector_t oldsectors = rdev->sectors;
b522adcd 2867 sector_t sectors;
27c529bb 2868
b522adcd 2869 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2870 return -EINVAL;
0cd17fec 2871 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2872 if (my_mddev->persistent) {
dd8ac336
AN		 2873 sectors = super_types[my_mddev->major_version].
2874 rdev_size_change(rdev, sectors);
2875 if (!sectors)
0cd17fec 2876 return -EBUSY;
dd8ac336 2877 } else if (!sectors)
77304d2a 2878 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2879 rdev->data_offset;
0cd17fec 2880 }
dd8ac336 2881 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2882 return -EINVAL; /* component must fit device */
0cd17fec 2883
dd8ac336
AN		 2884 rdev->sectors = sectors;
2885 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2886 /* need to check that all other rdevs with the same ->bdev
2887 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2888 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2889 * we have to change it back, we will have the lock again.
2890 */
fd01b88c 2891 struct mddev *mddev;
c5d79adb 2892 int overlap = 0;
159ec1fc 2893 struct list_head *tmp;
c5d79adb 2894
27c529bb 2895 mddev_unlock(my_mddev);
29ac4aa3 2896 for_each_mddev(mddev, tmp) {
3cb03002 2897 struct md_rdev *rdev2;
c5d79adb
N		 2898
2899 mddev_lock(mddev);
159ec1fc 2900 list_for_each_entry(rdev2, &mddev->disks, same_set)
f21e9ff7
N		 2901 if (rdev->bdev == rdev2->bdev &&
2902 rdev != rdev2 &&
2903 overlaps(rdev->data_offset, rdev->sectors,
2904 rdev2->data_offset,
2905 rdev2->sectors)) {
c5d79adb
N		 2906 overlap = 1;
2907 break;
2908 }
2909 mddev_unlock(mddev);
2910 if (overlap) {
2911 mddev_put(mddev);
2912 break;
2913 }
2914 }
27c529bb 2915 mddev_lock(my_mddev);
c5d79adb
N		 2916 if (overlap) {
2917 /* Someone else could have slipped in a size
2918 * change here, but doing so is just silly.
dd8ac336 2919 * We put oldsectors back because we *know* it is
c5d79adb
N		 2920 * safe, and trust userspace not to race with
2921 * itself
2922 */
dd8ac336 2923 rdev->sectors = oldsectors;
c5d79adb
N		 2924 return -EBUSY;
2925 }
2926 }
83303b61
N		 2927 return len;
2928}
2929
2930static struct rdev_sysfs_entry rdev_size =
80ca3a44 2931__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2932
06e3c817 2933
3cb03002 2934static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2935{
2936 unsigned long long recovery_start = rdev->recovery_offset;
2937
2938 if (test_bit(In_sync, &rdev->flags) ||
2939 recovery_start == MaxSector)
2940 return sprintf(page, "none\n");
2941
2942 return sprintf(page, "%llu\n", recovery_start);
2943}
2944
3cb03002 2945static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2946{
2947 unsigned long long recovery_start;
2948
2949 if (cmd_match(buf, "none"))
2950 recovery_start = MaxSector;
2951 else if (strict_strtoull(buf, 10, &recovery_start))
2952 return -EINVAL;
2953
2954 if (rdev->mddev->pers &&
2955 rdev->raid_disk >= 0)
2956 return -EBUSY;
2957
2958 rdev->recovery_offset = recovery_start;
2959 if (recovery_start == MaxSector)
2960 set_bit(In_sync, &rdev->flags);
2961 else
2962 clear_bit(In_sync, &rdev->flags);
2963 return len;
2964}
2965
2966static struct rdev_sysfs_entry rdev_recovery_start =
2967__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2968
16c791a5
N		 2969
2970static ssize_t
2971badblocks_show(struct badblocks *bb, char *page, int unack);
2972static ssize_t
2973badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2974
3cb03002 2975static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2976{
2977 return badblocks_show(&rdev->badblocks, page, 0);
2978}
3cb03002 2979static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2980{
de393cde
N		 2981 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2982 /* Maybe that ack was all we needed */
2983 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2984 wake_up(&rdev->blocked_wait);
2985 return rv;
16c791a5
N		 2986}
2987static struct rdev_sysfs_entry rdev_bad_blocks =
2988__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2989
2990
3cb03002 2991static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2992{
2993 return badblocks_show(&rdev->badblocks, page, 1);
2994}
3cb03002 2995static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2996{
2997 return badblocks_store(&rdev->badblocks, page, len, 1);
2998}
2999static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3000__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3001
86e6ffdd
N		 3002static struct attribute *rdev_default_attrs[] = {
3003 &rdev_state.attr,
4dbcdc75 3004 &rdev_errors.attr,
014236d2 3005 &rdev_slot.attr,
93c8cad0 3006 &rdev_offset.attr,
83303b61 3007 &rdev_size.attr,
06e3c817 3008 &rdev_recovery_start.attr,
16c791a5
N		 3009 &rdev_bad_blocks.attr,
3010 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3011 NULL,
3012};
3013static ssize_t
3014rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3015{
3016 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3017 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 3018 struct mddev *mddev = rdev->mddev;
27c529bb 3019 ssize_t rv;
86e6ffdd
N		 3020
3021 if (!entry->show)
3022 return -EIO;
27c529bb
N		 3023
3024 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3025 if (!rv) {
3026 if (rdev->mddev == NULL)
3027 rv = -EBUSY;
3028 else
3029 rv = entry->show(rdev, page);
3030 mddev_unlock(mddev);
3031 }
3032 return rv;
86e6ffdd
N		 3033}
3034
3035static ssize_t
3036rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3037 const char *page, size_t length)
3038{
3039 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3040 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3041 ssize_t rv;
fd01b88c 3042 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3043
3044 if (!entry->store)
3045 return -EIO;
67463acb
N		 3046 if (!capable(CAP_SYS_ADMIN))
3047 return -EACCES;
27c529bb 3048 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3049 if (!rv) {
27c529bb
N		 3050 if (rdev->mddev == NULL)
3051 rv = -EBUSY;
3052 else
3053 rv = entry->store(rdev, page, length);
6a51830e 3054 mddev_unlock(mddev);
ca388059
N		 3055 }
3056 return rv;
86e6ffdd
N		 3057}
3058
3059static void rdev_free(struct kobject *ko)
3060{
3cb03002 3061 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3062 kfree(rdev);
3063}
52cf25d0 3064static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3065 .show = rdev_attr_show,
3066 .store = rdev_attr_store,
3067};
3068static struct kobj_type rdev_ktype = {
3069 .release = rdev_free,
3070 .sysfs_ops = &rdev_sysfs_ops,
3071 .default_attrs = rdev_default_attrs,
3072};
3073
3cb03002 3074int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3075{
3076 rdev->desc_nr = -1;
3077 rdev->saved_raid_disk = -1;
3078 rdev->raid_disk = -1;
3079 rdev->flags = 0;
3080 rdev->data_offset = 0;
3081 rdev->sb_events = 0;
3082 rdev->last_read_error.tv_sec = 0;
3083 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3084 rdev->sb_loaded = 0;
3085 rdev->bb_page = NULL;
e8bb9a83
N		 3086 atomic_set(&rdev->nr_pending, 0);
3087 atomic_set(&rdev->read_errors, 0);
3088 atomic_set(&rdev->corrected_errors, 0);
3089
3090 INIT_LIST_HEAD(&rdev->same_set);
3091 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3092
3093 /* Add space to store bad block list.
3094 * This reserves the space even on arrays where it cannot
3095 * be used - I wonder if that matters
3096 */
3097 rdev->badblocks.count = 0;
3098 rdev->badblocks.shift = 0;
3099 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3100 seqlock_init(&rdev->badblocks.lock);
3101 if (rdev->badblocks.page == NULL)
3102 return -ENOMEM;
3103
3104 return 0;
e8bb9a83
N		 3105}
3106EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3107/*
3108 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3109 *
3110 * mark the device faulty if:
3111 *
3112 * - the device is nonexistent (zero size)
3113 * - the device has no valid superblock
3114 *
3115 * a faulty rdev _never_ has rdev->sb set.
3116 */
3cb03002 3117static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3118{
3119 char b[BDEVNAME_SIZE];
3120 int err;
3cb03002 3121 struct md_rdev *rdev;
1da177e4
LT		 3122 sector_t size;
3123
9ffae0cf 3124 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3125 if (!rdev) {
3126 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3127 return ERR_PTR(-ENOMEM);
3128 }
1da177e4 3129
2230dfe4
N		 3130 err = md_rdev_init(rdev);
3131 if (err)
3132 goto abort_free;
3133 err = alloc_disk_sb(rdev);
3134 if (err)
1da177e4
LT		 3135 goto abort_free;
3136
c5d79adb 3137 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3138 if (err)
3139 goto abort_free;
3140
f9cb074b 3141 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3142
77304d2a 3143 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 3144 if (!size) {
3145 printk(KERN_WARNING
3146 "md: %s has zero or unknown size, marking faulty!\n",
3147 bdevname(rdev->bdev,b));
3148 err = -EINVAL;
3149 goto abort_free;
3150 }
3151
3152 if (super_format >= 0) {
3153 err = super_types[super_format].
3154 load_super(rdev, NULL, super_minor);
3155 if (err == -EINVAL) {
df968c4e
N		 3156 printk(KERN_WARNING
3157 "md: %s does not have a valid v%d.%d "
3158 "superblock, not importing!\n",
3159 bdevname(rdev->bdev,b),
3160 super_format, super_minor);
1da177e4
LT		 3161 goto abort_free;
3162 }
3163 if (err < 0) {
3164 printk(KERN_WARNING
3165 "md: could not read %s's sb, not importing!\n",
3166 bdevname(rdev->bdev,b));
3167 goto abort_free;
3168 }
3169 }
9f2f3830
N		 3170 if (super_format == -1)
3171 /* hot-add for 0.90, or non-persistent: so no badblocks */
3172 rdev->badblocks.shift = -1;
6bfe0b49 3173
1da177e4
LT		 3174 return rdev;
3175
3176abort_free:
2699b672
N		 3177 if (rdev->bdev)
3178 unlock_rdev(rdev);
3179 free_disk_sb(rdev);
2230dfe4 3180 kfree(rdev->badblocks.page);
1da177e4
LT		 3181 kfree(rdev);
3182 return ERR_PTR(err);
3183}
3184
3185/*
3186 * Check a full RAID array for plausibility
3187 */
3188
3189
fd01b88c 3190static void analyze_sbs(struct mddev * mddev)
1da177e4
LT		 3191{
3192 int i;
3cb03002 3193 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3194 char b[BDEVNAME_SIZE];
3195
3196 freshest = NULL;
d089c6af 3197 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 3198 switch (super_types[mddev->major_version].
3199 load_super(rdev, freshest, mddev->minor_version)) {
3200 case 1:
3201 freshest = rdev;
3202 break;
3203 case 0:
3204 break;
3205 default:
3206 printk( KERN_ERR \
3207 "md: fatal superblock inconsistency in %s"
3208 " -- removing from array\n",
3209 bdevname(rdev->bdev,b));
3210 kick_rdev_from_array(rdev);
3211 }
3212
3213
3214 super_types[mddev->major_version].
3215 validate_super(mddev, freshest);
3216
3217 i = 0;
d089c6af 3218 rdev_for_each(rdev, tmp, mddev) {
233fca36
N		 3219 if (mddev->max_disks &&
3220 (rdev->desc_nr >= mddev->max_disks ||
3221 i > mddev->max_disks)) {
de01dfad
N		 3222 printk(KERN_WARNING
3223 "md: %s: %s: only %d devices permitted\n",
3224 mdname(mddev), bdevname(rdev->bdev, b),
3225 mddev->max_disks);
3226 kick_rdev_from_array(rdev);
3227 continue;
3228 }
1da177e4
LT		 3229 if (rdev != freshest)
3230 if (super_types[mddev->major_version].
3231 validate_super(mddev, rdev)) {
3232 printk(KERN_WARNING "md: kicking non-fresh %s"
3233 " from array!\n",
3234 bdevname(rdev->bdev,b));
3235 kick_rdev_from_array(rdev);
3236 continue;
3237 }
3238 if (mddev->level == LEVEL_MULTIPATH) {
3239 rdev->desc_nr = i++;
3240 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3241 set_bit(In_sync, &rdev->flags);
5e5e3e78 3242 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3243 rdev->raid_disk = -1;
3244 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3245 }
3246 }
1da177e4
LT		 3247}
3248
72e02075
N		 3249/* Read a fixed-point number.
3250 * Numbers in sysfs attributes should be in "standard" units where
3251 * possible, so time should be in seconds.
3252 * However we internally use a a much smaller unit such as
3253 * milliseconds or jiffies.
3254 * This function takes a decimal number with a possible fractional
3255 * component, and produces an integer which is the result of
3256 * multiplying that number by 10^'scale'.
3257 * all without any floating-point arithmetic.
3258 */
3259int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3260{
3261 unsigned long result = 0;
3262 long decimals = -1;
3263 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3264 if (*cp == '.')
3265 decimals = 0;
3266 else if (decimals < scale) {
3267 unsigned int value;
3268 value = *cp - '0';
3269 result = result * 10 + value;
3270 if (decimals >= 0)
3271 decimals++;
3272 }
3273 cp++;
3274 }
3275 if (*cp == '\n')
3276 cp++;
3277 if (*cp)
3278 return -EINVAL;
3279 if (decimals < 0)
3280 decimals = 0;
3281 while (decimals < scale) {
3282 result *= 10;
3283 decimals ++;
3284 }
3285 *res = result;
3286 return 0;
3287}
3288
3289
19052c0e
N		 3290static void md_safemode_timeout(unsigned long data);
3291
16f17b39 3292static ssize_t
fd01b88c 3293safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3294{
3295 int msec = (mddev->safemode_delay*1000)/HZ;
3296 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3297}
3298static ssize_t
fd01b88c 3299safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3300{
16f17b39 3301 unsigned long msec;
97ce0a7f 3302
72e02075 3303 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3304 return -EINVAL;
16f17b39
N		 3305 if (msec == 0)
3306 mddev->safemode_delay = 0;
3307 else {
19052c0e 3308 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 3309 mddev->safemode_delay = (msec*HZ)/1000;
3310 if (mddev->safemode_delay == 0)
3311 mddev->safemode_delay = 1;
19052c0e
N		 3312 if (mddev->safemode_delay < old_delay)
3313 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 3314 }
3315 return len;
3316}
3317static struct md_sysfs_entry md_safe_delay =
80ca3a44 3318__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3319
eae1701f 3320static ssize_t
fd01b88c 3321level_show(struct mddev *mddev, char *page)
eae1701f 3322{
84fc4b56 3323 struct md_personality *p = mddev->pers;
d9d166c2 3324 if (p)
eae1701f 3325 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 3326 else if (mddev->clevel[0])
3327 return sprintf(page, "%s\n", mddev->clevel);
3328 else if (mddev->level != LEVEL_NONE)
3329 return sprintf(page, "%d\n", mddev->level);
3330 else
3331 return 0;
eae1701f
N		 3332}
3333
d9d166c2 3334static ssize_t
fd01b88c 3335level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3336{
f2859af6 3337 char clevel[16];
20a49ff6 3338 ssize_t rv = len;
84fc4b56 3339 struct md_personality *pers;
f2859af6 3340 long level;
245f46c2 3341 void *priv;
3cb03002 3342 struct md_rdev *rdev;
245f46c2
N		 3343
3344 if (mddev->pers == NULL) {
3345 if (len == 0)
3346 return 0;
3347 if (len >= sizeof(mddev->clevel))
3348 return -ENOSPC;
3349 strncpy(mddev->clevel, buf, len);
3350 if (mddev->clevel[len-1] == '\n')
3351 len--;
3352 mddev->clevel[len] = 0;
3353 mddev->level = LEVEL_NONE;
3354 return rv;
3355 }
3356
3357 /* request to change the personality. Need to ensure:
3358 * - array is not engaged in resync/recovery/reshape
3359 * - old personality can be suspended
3360 * - new personality will access other array.
3361 */
3362
bb4f1e9d
N		 3363 if (mddev->sync_thread ||
3364 mddev->reshape_position != MaxSector ||
3365 mddev->sysfs_active)
d9d166c2 3366 return -EBUSY;
245f46c2
N		 3367
3368 if (!mddev->pers->quiesce) {
3369 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3370 mdname(mddev), mddev->pers->name);
3371 return -EINVAL;
3372 }
3373
3374 /* Now find the new personality */
f2859af6 3375 if (len == 0 || len >= sizeof(clevel))
245f46c2 3376 return -EINVAL;
f2859af6
DW		 3377 strncpy(clevel, buf, len);
3378 if (clevel[len-1] == '\n')
d9d166c2 3379 len--;
f2859af6
DW		 3380 clevel[len] = 0;
3381 if (strict_strtol(clevel, 10, &level))
3382 level = LEVEL_NONE;
245f46c2 3383
f2859af6
DW		 3384 if (request_module("md-%s", clevel) != 0)
3385 request_module("md-level-%s", clevel);
245f46c2 3386 spin_lock(&pers_lock);
f2859af6 3387 pers = find_pers(level, clevel);
245f46c2
N		 3388 if (!pers || !try_module_get(pers->owner)) {
3389 spin_unlock(&pers_lock);
f2859af6 3390 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3391 return -EINVAL;
3392 }
3393 spin_unlock(&pers_lock);
3394
3395 if (pers == mddev->pers) {
3396 /* Nothing to do! */
3397 module_put(pers->owner);
3398 return rv;
3399 }
3400 if (!pers->takeover) {
3401 module_put(pers->owner);
3402 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3403 mdname(mddev), clevel);
245f46c2
N		 3404 return -EINVAL;
3405 }
3406
e93f68a1
N		 3407 list_for_each_entry(rdev, &mddev->disks, same_set)
3408 rdev->new_raid_disk = rdev->raid_disk;
3409
245f46c2
N		 3410 /* ->takeover must set new_* and/or delta_disks
3411 * if it succeeds, and may set them when it fails.
3412 */
3413 priv = pers->takeover(mddev);
3414 if (IS_ERR(priv)) {
3415 mddev->new_level = mddev->level;
3416 mddev->new_layout = mddev->layout;
664e7c41 3417 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3418 mddev->raid_disks -= mddev->delta_disks;
3419 mddev->delta_disks = 0;
3420 module_put(pers->owner);
3421 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3422 mdname(mddev), clevel);
245f46c2
N		 3423 return PTR_ERR(priv);
3424 }
3425
3426 /* Looks like we have a winner */
3427 mddev_suspend(mddev);
3428 mddev->pers->stop(mddev);
a64c876f
N		 3429
3430 if (mddev->pers->sync_request == NULL &&
3431 pers->sync_request != NULL) {
3432 /* need to add the md_redundancy_group */
3433 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3434 printk(KERN_WARNING
3435 "md: cannot register extra attributes for %s\n",
3436 mdname(mddev));
19fdb9ee 3437 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3438 }
3439 if (mddev->pers->sync_request != NULL &&
3440 pers->sync_request == NULL) {
3441 /* need to remove the md_redundancy_group */
3442 if (mddev->to_remove == NULL)
3443 mddev->to_remove = &md_redundancy_group;
3444 }
3445
54071b38
TM		 3446 if (mddev->pers->sync_request == NULL &&
3447 mddev->external) {
3448 /* We are converting from a no-redundancy array
3449 * to a redundancy array and metadata is managed
3450 * externally so we need to be sure that writes
3451 * won't block due to a need to transition
3452 * clean->dirty
3453 * until external management is started.
3454 */
3455 mddev->in_sync = 0;
3456 mddev->safemode_delay = 0;
3457 mddev->safemode = 0;
3458 }
3459
e93f68a1 3460 list_for_each_entry(rdev, &mddev->disks, same_set) {
e93f68a1
N		 3461 if (rdev->raid_disk < 0)
3462 continue;
bf2cb0da 3463 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3464 rdev->new_raid_disk = -1;
3465 if (rdev->new_raid_disk == rdev->raid_disk)
3466 continue;
36fad858 3467 sysfs_unlink_rdev(mddev, rdev);
e93f68a1
N		 3468 }
3469 list_for_each_entry(rdev, &mddev->disks, same_set) {
3470 if (rdev->raid_disk < 0)
3471 continue;
3472 if (rdev->new_raid_disk == rdev->raid_disk)
3473 continue;
3474 rdev->raid_disk = rdev->new_raid_disk;
3475 if (rdev->raid_disk < 0)
3a981b03 3476 clear_bit(In_sync, &rdev->flags);
e93f68a1 3477 else {
36fad858
NK		 3478 if (sysfs_link_rdev(mddev, rdev))
3479 printk(KERN_WARNING "md: cannot register rd%d"
3480 " for %s after level change\n",
3481 rdev->raid_disk, mdname(mddev));
3a981b03 3482 }
e93f68a1
N		 3483 }
3484
3485 module_put(mddev->pers->owner);
245f46c2
N		 3486 mddev->pers = pers;
3487 mddev->private = priv;
3488 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3489 mddev->level = mddev->new_level;
3490 mddev->layout = mddev->new_layout;
664e7c41 3491 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3492 mddev->delta_disks = 0;
fee68723 3493 mddev->degraded = 0;
9af204cf
TM		 3494 if (mddev->pers->sync_request == NULL) {
3495 /* this is now an array without redundancy, so
3496 * it must always be in_sync
3497 */
3498 mddev->in_sync = 1;
3499 del_timer_sync(&mddev->safemode_timer);
3500 }
245f46c2
N		 3501 pers->run(mddev);
3502 mddev_resume(mddev);
3503 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3504 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3505 md_wakeup_thread(mddev->thread);
5cac7861 3506 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3507 md_new_event(mddev);
d9d166c2
N		 3508 return rv;
3509}
3510
3511static struct md_sysfs_entry md_level =
80ca3a44 3512__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3513
d4dbd025
N		 3514
3515static ssize_t
fd01b88c 3516layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3517{
3518 /* just a number, not meaningful for all levels */
08a02ecd
N		 3519 if (mddev->reshape_position != MaxSector &&
3520 mddev->layout != mddev->new_layout)
3521 return sprintf(page, "%d (%d)\n",
3522 mddev->new_layout, mddev->layout);
d4dbd025
N		 3523 return sprintf(page, "%d\n", mddev->layout);
3524}
3525
3526static ssize_t
fd01b88c 3527layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3528{
3529 char *e;
3530 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3531
3532 if (!*buf || (*e && *e != '\n'))
3533 return -EINVAL;
3534
b3546035
N		 3535 if (mddev->pers) {
3536 int err;
50ac168a 3537 if (mddev->pers->check_reshape == NULL)
b3546035 3538 return -EBUSY;
597a711b 3539 mddev->new_layout = n;
50ac168a 3540 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3541 if (err) {
3542 mddev->new_layout = mddev->layout;
b3546035 3543 return err;
597a711b 3544 }
b3546035 3545 } else {
08a02ecd 3546 mddev->new_layout = n;
b3546035
N		 3547 if (mddev->reshape_position == MaxSector)
3548 mddev->layout = n;
3549 }
d4dbd025
N		 3550 return len;
3551}
3552static struct md_sysfs_entry md_layout =
80ca3a44 3553__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3554
3555
eae1701f 3556static ssize_t
fd01b88c 3557raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3558{
bb636547
N		 3559 if (mddev->raid_disks == 0)
3560 return 0;
08a02ecd
N		 3561 if (mddev->reshape_position != MaxSector &&
3562 mddev->delta_disks != 0)
3563 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3564 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3565 return sprintf(page, "%d\n", mddev->raid_disks);
3566}
3567
fd01b88c 3568static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3569
3570static ssize_t
fd01b88c 3571raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3572{
da943b99
N		 3573 char *e;
3574 int rv = 0;
3575 unsigned long n = simple_strtoul(buf, &e, 10);
3576
3577 if (!*buf || (*e && *e != '\n'))
3578 return -EINVAL;
3579
3580 if (mddev->pers)
3581 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3582 else if (mddev->reshape_position != MaxSector) {
3583 int olddisks = mddev->raid_disks - mddev->delta_disks;
3584 mddev->delta_disks = n - olddisks;
3585 mddev->raid_disks = n;
3586 } else
da943b99
N		 3587 mddev->raid_disks = n;
3588 return rv ? rv : len;
3589}
3590static struct md_sysfs_entry md_raid_disks =
80ca3a44 3591__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3592
3b34380a 3593static ssize_t
fd01b88c 3594chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3595{
08a02ecd 3596 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3597 mddev->chunk_sectors != mddev->new_chunk_sectors)
3598 return sprintf(page, "%d (%d)\n",
3599 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3600 mddev->chunk_sectors << 9);
3601 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3602}
3603
3604static ssize_t
fd01b88c 3605chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3606{
3b34380a
N		 3607 char *e;
3608 unsigned long n = simple_strtoul(buf, &e, 10);
3609
3b34380a
N		 3610 if (!*buf || (*e && *e != '\n'))
3611 return -EINVAL;
3612
b3546035
N		 3613 if (mddev->pers) {
3614 int err;
50ac168a 3615 if (mddev->pers->check_reshape == NULL)
b3546035 3616 return -EBUSY;
597a711b 3617 mddev->new_chunk_sectors = n >> 9;
50ac168a 3618 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3619 if (err) {
3620 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3621 return err;
597a711b 3622 }
b3546035 3623 } else {
664e7c41 3624 mddev->new_chunk_sectors = n >> 9;
b3546035 3625 if (mddev->reshape_position == MaxSector)
9d8f0363 3626 mddev->chunk_sectors = n >> 9;
b3546035 3627 }
3b34380a
N		 3628 return len;
3629}
3630static struct md_sysfs_entry md_chunk_size =
80ca3a44 3631__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3632
a94213b1 3633static ssize_t
fd01b88c 3634resync_start_show(struct mddev *mddev, char *page)
a94213b1 3635{
d1a7c503
N		 3636 if (mddev->recovery_cp == MaxSector)
3637 return sprintf(page, "none\n");
a94213b1
N		 3638 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3639}
3640
3641static ssize_t
fd01b88c 3642resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3643{
a94213b1
N		 3644 char *e;
3645 unsigned long long n = simple_strtoull(buf, &e, 10);
3646
b098636c 3647 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3648 return -EBUSY;
06e3c817
DW		 3649 if (cmd_match(buf, "none"))
3650 n = MaxSector;
3651 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3652 return -EINVAL;
3653
3654 mddev->recovery_cp = n;
3655 return len;
3656}
3657static struct md_sysfs_entry md_resync_start =
80ca3a44 3658__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3659
9e653b63
N		 3660/*
3661 * The array state can be:
3662 *
3663 * clear
3664 * No devices, no size, no level
3665 * Equivalent to STOP_ARRAY ioctl
3666 * inactive
3667 * May have some settings, but array is not active
3668 * all IO results in error
3669 * When written, doesn't tear down array, but just stops it
3670 * suspended (not supported yet)
3671 * All IO requests will block. The array can be reconfigured.
910d8cb3 3672 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3673 * readonly
3674 * no resync can happen. no superblocks get written.
3675 * write requests fail
3676 * read-auto
3677 * like readonly, but behaves like 'clean' on a write request.
3678 *
3679 * clean - no pending writes, but otherwise active.
3680 * When written to inactive array, starts without resync
3681 * If a write request arrives then
3682 * if metadata is known, mark 'dirty' and switch to 'active'.
3683 * if not known, block and switch to write-pending
3684 * If written to an active array that has pending writes, then fails.
3685 * active
3686 * fully active: IO and resync can be happening.
3687 * When written to inactive array, starts with resync
3688 *
3689 * write-pending
3690 * clean, but writes are blocked waiting for 'active' to be written.
3691 *
3692 * active-idle
3693 * like active, but no writes have been seen for a while (100msec).
3694 *
3695 */
3696enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3697 write_pending, active_idle, bad_word};
05381954 3698static char *array_states[] = {
9e653b63
N		 3699 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3700 "write-pending", "active-idle", NULL };
3701
3702static int match_word(const char *word, char **list)
3703{
3704 int n;
3705 for (n=0; list[n]; n++)
3706 if (cmd_match(word, list[n]))
3707 break;
3708 return n;
3709}
3710
3711static ssize_t
fd01b88c 3712array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3713{
3714 enum array_state st = inactive;
3715
3716 if (mddev->pers)
3717 switch(mddev->ro) {
3718 case 1:
3719 st = readonly;
3720 break;
3721 case 2:
3722 st = read_auto;
3723 break;
3724 case 0:
3725 if (mddev->in_sync)
3726 st = clean;
070dc6dd 3727 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3728 st = write_pending;
9e653b63
N		 3729 else if (mddev->safemode)
3730 st = active_idle;
3731 else
3732 st = active;
3733 }
3734 else {
3735 if (list_empty(&mddev->disks) &&
3736 mddev->raid_disks == 0 &&
58c0fed4 3737 mddev->dev_sectors == 0)
9e653b63
N		 3738 st = clear;
3739 else
3740 st = inactive;
3741 }
3742 return sprintf(page, "%s\n", array_states[st]);
3743}
3744
fd01b88c
N		 3745static int do_md_stop(struct mddev * mddev, int ro, int is_open);
3746static int md_set_readonly(struct mddev * mddev, int is_open);
3747static int do_md_run(struct mddev * mddev);
3748static int restart_array(struct mddev *mddev);
9e653b63
N		 3749
3750static ssize_t
fd01b88c 3751array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N		 3752{
3753 int err = -EINVAL;
3754 enum array_state st = match_word(buf, array_states);
3755 switch(st) {
3756 case bad_word:
3757 break;
3758 case clear:
3759 /* stopping an active array */
f2ea68cf 3760 if (atomic_read(&mddev->openers) > 0)
e691063a 3761 return -EBUSY;
df5b20cf 3762 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3763 break;
3764 case inactive:
3765 /* stopping an active array */
3766 if (mddev->pers) {
f2ea68cf 3767 if (atomic_read(&mddev->openers) > 0)
9e653b63 3768 return -EBUSY;
df5b20cf 3769 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3770 } else
3771 err = 0; /* already inactive */
9e653b63
N		 3772 break;
3773 case suspended:
3774 break; /* not supported yet */
3775 case readonly:
3776 if (mddev->pers)
a4bd82d0 3777 err = md_set_readonly(mddev, 0);
9e653b63
N		 3778 else {
3779 mddev->ro = 1;
648b629e 3780 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3781 err = do_md_run(mddev);
3782 }
3783 break;
3784 case read_auto:
9e653b63 3785 if (mddev->pers) {
80268ee9 3786 if (mddev->ro == 0)
a4bd82d0 3787 err = md_set_readonly(mddev, 0);
80268ee9 3788 else if (mddev->ro == 1)
648b629e
N		 3789 err = restart_array(mddev);
3790 if (err == 0) {
3791 mddev->ro = 2;
3792 set_disk_ro(mddev->gendisk, 0);
3793 }
9e653b63
N		 3794 } else {
3795 mddev->ro = 2;
3796 err = do_md_run(mddev);
3797 }
3798 break;
3799 case clean:
3800 if (mddev->pers) {
3801 restart_array(mddev);
3802 spin_lock_irq(&mddev->write_lock);
3803 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3804 if (mddev->in_sync == 0) {
3805 mddev->in_sync = 1;
31a59e34
N		 3806 if (mddev->safemode == 1)
3807 mddev->safemode = 0;
070dc6dd 3808 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3809 }
3810 err = 0;
3811 } else
3812 err = -EBUSY;
9e653b63 3813 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3814 } else
3815 err = -EINVAL;
9e653b63
N		 3816 break;
3817 case active:
3818 if (mddev->pers) {
3819 restart_array(mddev);
070dc6dd 3820 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3821 wake_up(&mddev->sb_wait);
3822 err = 0;
3823 } else {
3824 mddev->ro = 0;
648b629e 3825 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3826 err = do_md_run(mddev);
3827 }
3828 break;
3829 case write_pending:
3830 case active_idle:
3831 /* these cannot be set */
3832 break;
3833 }
3834 if (err)
3835 return err;
0fd62b86 3836 else {
1d23f178
N		 3837 if (mddev->hold_active == UNTIL_IOCTL)
3838 mddev->hold_active = 0;
00bcb4ac 3839 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3840 return len;
0fd62b86 3841 }
9e653b63 3842}
80ca3a44
N		 3843static struct md_sysfs_entry md_array_state =
3844__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3845
1e50915f 3846static ssize_t
fd01b88c 3847max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3848 return sprintf(page, "%d\n",
3849 atomic_read(&mddev->max_corr_read_errors));
3850}
3851
3852static ssize_t
fd01b88c 3853max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3854{
3855 char *e;
3856 unsigned long n = simple_strtoul(buf, &e, 10);
3857
3858 if (*buf && (*e == 0 || *e == '\n')) {
3859 atomic_set(&mddev->max_corr_read_errors, n);
3860 return len;
3861 }
3862 return -EINVAL;
3863}
3864
3865static struct md_sysfs_entry max_corr_read_errors =
3866__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3867 max_corrected_read_errors_store);
3868
6d7ff738 3869static ssize_t
fd01b88c 3870null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3871{
3872 return -EINVAL;
3873}
3874
3875static ssize_t
fd01b88c 3876new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3877{
3878 /* buf must be %d:%d\n? giving major and minor numbers */
3879 /* The new device is added to the array.
3880 * If the array has a persistent superblock, we read the
3881 * superblock to initialise info and check validity.
3882 * Otherwise, only checking done is that in bind_rdev_to_array,
3883 * which mainly checks size.
3884 */
3885 char *e;
3886 int major = simple_strtoul(buf, &e, 10);
3887 int minor;
3888 dev_t dev;
3cb03002 3889 struct md_rdev *rdev;
6d7ff738
N		 3890 int err;
3891
3892 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3893 return -EINVAL;
3894 minor = simple_strtoul(e+1, &e, 10);
3895 if (*e && *e != '\n')
3896 return -EINVAL;
3897 dev = MKDEV(major, minor);
3898 if (major != MAJOR(dev) ||
3899 minor != MINOR(dev))
3900 return -EOVERFLOW;
3901
3902
3903 if (mddev->persistent) {
3904 rdev = md_import_device(dev, mddev->major_version,
3905 mddev->minor_version);
3906 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 3907 struct md_rdev *rdev0
3908 = list_entry(mddev->disks.next,
3909 struct md_rdev, same_set);
6d7ff738
N		 3910 err = super_types[mddev->major_version]
3911 .load_super(rdev, rdev0, mddev->minor_version);
3912 if (err < 0)
3913 goto out;
3914 }
c5d79adb
N		 3915 } else if (mddev->external)
3916 rdev = md_import_device(dev, -2, -1);
3917 else
6d7ff738
N		 3918 rdev = md_import_device(dev, -1, -1);
3919
3920 if (IS_ERR(rdev))
3921 return PTR_ERR(rdev);
3922 err = bind_rdev_to_array(rdev, mddev);
3923 out:
3924 if (err)
3925 export_rdev(rdev);
3926 return err ? err : len;
3927}
3928
3929static struct md_sysfs_entry md_new_device =
80ca3a44 3930__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3931
9b1d1dac 3932static ssize_t
fd01b88c 3933bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 3934{
3935 char *end;
3936 unsigned long chunk, end_chunk;
3937
3938 if (!mddev->bitmap)
3939 goto out;
3940 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3941 while (*buf) {
3942 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3943 if (buf == end) break;
3944 if (*end == '-') { /* range */
3945 buf = end + 1;
3946 end_chunk = simple_strtoul(buf, &end, 0);
3947 if (buf == end) break;
3948 }
3949 if (*end && !isspace(*end)) break;
3950 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3951 buf = skip_spaces(end);
9b1d1dac
PC		 3952 }
3953 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3954out:
3955 return len;
3956}
3957
3958static struct md_sysfs_entry md_bitmap =
3959__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3960
a35b0d69 3961static ssize_t
fd01b88c 3962size_show(struct mddev *mddev, char *page)
a35b0d69 3963{
58c0fed4
AN		 3964 return sprintf(page, "%llu\n",
3965 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3966}
3967
fd01b88c 3968static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 3969
3970static ssize_t
fd01b88c 3971size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 3972{
3973 /* If array is inactive, we can reduce the component size, but
3974 * not increase it (except from 0).
3975 * If array is active, we can try an on-line resize
3976 */
b522adcd
DW		 3977 sector_t sectors;
3978 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3979
58c0fed4
AN		 3980 if (err < 0)
3981 return err;
a35b0d69 3982 if (mddev->pers) {
58c0fed4 3983 err = update_size(mddev, sectors);
850b2b42 3984 md_update_sb(mddev, 1);
a35b0d69 3985 } else {
58c0fed4
AN		 3986 if (mddev->dev_sectors == 0 ||
3987 mddev->dev_sectors > sectors)
3988 mddev->dev_sectors = sectors;
a35b0d69
N		 3989 else
3990 err = -ENOSPC;
3991 }
3992 return err ? err : len;
3993}
3994
3995static struct md_sysfs_entry md_size =
80ca3a44 3996__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3997
8bb93aac
N		 3998
3999/* Metdata version.
e691063a
N		 4000 * This is one of
4001 * 'none' for arrays with no metadata (good luck...)
4002 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4003 * or N.M for internally known formats
4004 */
4005static ssize_t
fd01b88c 4006metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4007{
4008 if (mddev->persistent)
4009 return sprintf(page, "%d.%d\n",
4010 mddev->major_version, mddev->minor_version);
e691063a
N		 4011 else if (mddev->external)
4012 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4013 else
4014 return sprintf(page, "none\n");
4015}
4016
4017static ssize_t
fd01b88c 4018metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4019{
4020 int major, minor;
4021 char *e;
ea43ddd8
N		 4022 /* Changing the details of 'external' metadata is
4023 * always permitted. Otherwise there must be
4024 * no devices attached to the array.
4025 */
4026 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4027 ;
4028 else if (!list_empty(&mddev->disks))
8bb93aac
N		 4029 return -EBUSY;
4030
4031 if (cmd_match(buf, "none")) {
4032 mddev->persistent = 0;
e691063a
N		 4033 mddev->external = 0;
4034 mddev->major_version = 0;
4035 mddev->minor_version = 90;
4036 return len;
4037 }
4038 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4039 size_t namelen = len-9;
e691063a
N		 4040 if (namelen >= sizeof(mddev->metadata_type))
4041 namelen = sizeof(mddev->metadata_type)-1;
4042 strncpy(mddev->metadata_type, buf+9, namelen);
4043 mddev->metadata_type[namelen] = 0;
4044 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4045 mddev->metadata_type[--namelen] = 0;
4046 mddev->persistent = 0;
4047 mddev->external = 1;
8bb93aac
N		 4048 mddev->major_version = 0;
4049 mddev->minor_version = 90;
4050 return len;
4051 }
4052 major = simple_strtoul(buf, &e, 10);
4053 if (e==buf || *e != '.')
4054 return -EINVAL;
4055 buf = e+1;
4056 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4057 if (e==buf || (*e && *e != '\n') )
8bb93aac 4058 return -EINVAL;
50511da3 4059 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 4060 return -ENOENT;
4061 mddev->major_version = major;
4062 mddev->minor_version = minor;
4063 mddev->persistent = 1;
e691063a 4064 mddev->external = 0;
8bb93aac
N		 4065 return len;
4066}
4067
4068static struct md_sysfs_entry md_metadata =
80ca3a44 4069__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4070
24dd469d 4071static ssize_t
fd01b88c 4072action_show(struct mddev *mddev, char *page)
24dd469d 4073{
7eec314d 4074 char *type = "idle";
b6a9ce68
N		 4075 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4076 type = "frozen";
4077 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4078 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 4079 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4080 type = "reshape";
4081 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 4082 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4083 type = "resync";
4084 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4085 type = "check";
4086 else
4087 type = "repair";
72a23c21 4088 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 4089 type = "recover";
4090 }
4091 return sprintf(page, "%s\n", type);
4092}
4093
fd01b88c 4094static void reap_sync_thread(struct mddev *mddev);
7ebc0be7 4095
24dd469d 4096static ssize_t
fd01b88c 4097action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4098{
7eec314d
N		 4099 if (!mddev->pers || !mddev->pers->sync_request)
4100 return -EINVAL;
4101
b6a9ce68
N		 4102 if (cmd_match(page, "frozen"))
4103 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4104 else
4105 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4106
4107 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 4108 if (mddev->sync_thread) {
4109 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4110 reap_sync_thread(mddev);
7eec314d 4111 }
03c902e1
N		 4112 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4113 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4114 return -EBUSY;
72a23c21
NB		 4115 else if (cmd_match(page, "resync"))
4116 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4117 else if (cmd_match(page, "recover")) {
4118 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4119 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4120 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4121 int err;
4122 if (mddev->pers->start_reshape == NULL)
4123 return -EINVAL;
4124 err = mddev->pers->start_reshape(mddev);
4125 if (err)
4126 return err;
a99ac971 4127 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4128 } else {
bce74dac 4129 if (cmd_match(page, "check"))
7eec314d 4130 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4131 else if (!cmd_match(page, "repair"))
7eec314d
N		 4132 return -EINVAL;
4133 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4134 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4135 }
03c902e1 4136 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4137 md_wakeup_thread(mddev->thread);
00bcb4ac 4138 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4139 return len;
4140}
4141
9d88883e 4142static ssize_t
fd01b88c 4143mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4144{
4145 return sprintf(page, "%llu\n",
4146 (unsigned long long) mddev->resync_mismatches);
4147}
4148
80ca3a44
N		 4149static struct md_sysfs_entry md_scan_mode =
4150__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 4151
96de1e66 4152
80ca3a44 4153static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4154
88202a0c 4155static ssize_t
fd01b88c 4156sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4157{
4158 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4159 mddev->sync_speed_min ? "local": "system");
4160}
4161
4162static ssize_t
fd01b88c 4163sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4164{
4165 int min;
4166 char *e;
4167 if (strncmp(buf, "system", 6)==0) {
4168 mddev->sync_speed_min = 0;
4169 return len;
4170 }
4171 min = simple_strtoul(buf, &e, 10);
4172 if (buf == e || (*e && *e != '\n') || min <= 0)
4173 return -EINVAL;
4174 mddev->sync_speed_min = min;
4175 return len;
4176}
4177
4178static struct md_sysfs_entry md_sync_min =
4179__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4180
4181static ssize_t
fd01b88c 4182sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4183{
4184 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4185 mddev->sync_speed_max ? "local": "system");
4186}
4187
4188static ssize_t
fd01b88c 4189sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4190{
4191 int max;
4192 char *e;
4193 if (strncmp(buf, "system", 6)==0) {
4194 mddev->sync_speed_max = 0;
4195 return len;
4196 }
4197 max = simple_strtoul(buf, &e, 10);
4198 if (buf == e || (*e && *e != '\n') || max <= 0)
4199 return -EINVAL;
4200 mddev->sync_speed_max = max;
4201 return len;
4202}
4203
4204static struct md_sysfs_entry md_sync_max =
4205__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4206
d7f3d291 4207static ssize_t
fd01b88c 4208degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4209{
4210 return sprintf(page, "%d\n", mddev->degraded);
4211}
4212static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4213
90b08710 4214static ssize_t
fd01b88c 4215sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4216{
4217 return sprintf(page, "%d\n", mddev->parallel_resync);
4218}
4219
4220static ssize_t
fd01b88c 4221sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4222{
4223 long n;
4224
4225 if (strict_strtol(buf, 10, &n))
4226 return -EINVAL;
4227
4228 if (n != 0 && n != 1)
4229 return -EINVAL;
4230
4231 mddev->parallel_resync = n;
4232
4233 if (mddev->sync_thread)
4234 wake_up(&resync_wait);
4235
4236 return len;
4237}
4238
4239/* force parallel resync, even with shared block devices */
4240static struct md_sysfs_entry md_sync_force_parallel =
4241__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4242 sync_force_parallel_show, sync_force_parallel_store);
4243
88202a0c 4244static ssize_t
fd01b88c 4245sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4246{
4247 unsigned long resync, dt, db;
d1a7c503
N		 4248 if (mddev->curr_resync == 0)
4249 return sprintf(page, "none\n");
9687a60c
AN		 4250 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4251 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4252 if (!dt) dt++;
9687a60c
AN		 4253 db = resync - mddev->resync_mark_cnt;
4254 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4255}
4256
80ca3a44 4257static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4258
4259static ssize_t
fd01b88c 4260sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4261{
13ae864b 4262 unsigned long long max_sectors, resync;
88202a0c 4263
acb180b0
N		 4264 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4265 return sprintf(page, "none\n");
4266
88202a0c 4267 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 4268 max_sectors = mddev->resync_max_sectors;
88202a0c 4269 else
58c0fed4 4270 max_sectors = mddev->dev_sectors;
88202a0c 4271
acb180b0 4272 resync = mddev->curr_resync_completed;
13ae864b 4273 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4274}
4275
80ca3a44 4276static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4277
5e96ee65 4278static ssize_t
fd01b88c 4279min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4280{
4281 return sprintf(page, "%llu\n",
4282 (unsigned long long)mddev->resync_min);
4283}
4284static ssize_t
fd01b88c 4285min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4286{
4287 unsigned long long min;
4288 if (strict_strtoull(buf, 10, &min))
4289 return -EINVAL;
4290 if (min > mddev->resync_max)
4291 return -EINVAL;
4292 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4293 return -EBUSY;
4294
4295 /* Must be a multiple of chunk_size */
9d8f0363 4296 if (mddev->chunk_sectors) {
2ac06c33 4297 sector_t temp = min;
9d8f0363 4298 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 4299 return -EINVAL;
4300 }
4301 mddev->resync_min = min;
4302
4303 return len;
4304}
4305
4306static struct md_sysfs_entry md_min_sync =
4307__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4308
c6207277 4309static ssize_t
fd01b88c 4310max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4311{
4312 if (mddev->resync_max == MaxSector)
4313 return sprintf(page, "max\n");
4314 else
4315 return sprintf(page, "%llu\n",
4316 (unsigned long long)mddev->resync_max);
4317}
4318static ssize_t
fd01b88c 4319max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N		 4320{
4321 if (strncmp(buf, "max", 3) == 0)
4322 mddev->resync_max = MaxSector;
4323 else {
5e96ee65
NB		 4324 unsigned long long max;
4325 if (strict_strtoull(buf, 10, &max))
4326 return -EINVAL;
4327 if (max < mddev->resync_min)
c6207277
N		 4328 return -EINVAL;
4329 if (max < mddev->resync_max &&
4d484a4a 4330 mddev->ro == 0 &&
c6207277
N		 4331 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4332 return -EBUSY;
4333
4334 /* Must be a multiple of chunk_size */
9d8f0363 4335 if (mddev->chunk_sectors) {
2ac06c33 4336 sector_t temp = max;
9d8f0363 4337 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4338 return -EINVAL;
4339 }
4340 mddev->resync_max = max;
4341 }
4342 wake_up(&mddev->recovery_wait);
4343 return len;
4344}
4345
4346static struct md_sysfs_entry md_max_sync =
4347__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4348
e464eafd 4349static ssize_t
fd01b88c 4350suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4351{
4352 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4353}
4354
4355static ssize_t
fd01b88c 4356suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4357{
4358 char *e;
4359 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4360 unsigned long long old = mddev->suspend_lo;
e464eafd 4361
b8d966ef
N		 4362 if (mddev->pers == NULL ||
4363 mddev->pers->quiesce == NULL)
e464eafd
N		 4364 return -EINVAL;
4365 if (buf == e || (*e && *e != '\n'))
4366 return -EINVAL;
23ddff37
N		 4367
4368 mddev->suspend_lo = new;
4369 if (new >= old)
4370 /* Shrinking suspended region */
e464eafd 4371 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4372 else {
4373 /* Expanding suspended region - need to wait */
4374 mddev->pers->quiesce(mddev, 1);
4375 mddev->pers->quiesce(mddev, 0);
4376 }
4377 return len;
e464eafd
N		 4378}
4379static struct md_sysfs_entry md_suspend_lo =
4380__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4381
4382
4383static ssize_t
fd01b88c 4384suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4385{
4386 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4387}
4388
4389static ssize_t
fd01b88c 4390suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4391{
4392 char *e;
4393 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4394 unsigned long long old = mddev->suspend_hi;
e464eafd 4395
b8d966ef
N		 4396 if (mddev->pers == NULL ||
4397 mddev->pers->quiesce == NULL)
e464eafd
N		 4398 return -EINVAL;
4399 if (buf == e || (*e && *e != '\n'))
4400 return -EINVAL;
23ddff37
N		 4401
4402 mddev->suspend_hi = new;
4403 if (new <= old)
4404 /* Shrinking suspended region */
4405 mddev->pers->quiesce(mddev, 2);
4406 else {
4407 /* Expanding suspended region - need to wait */
e464eafd
N		 4408 mddev->pers->quiesce(mddev, 1);
4409 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4410 }
4411 return len;
e464eafd
N		 4412}
4413static struct md_sysfs_entry md_suspend_hi =
4414__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4415
08a02ecd 4416static ssize_t
fd01b88c 4417reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4418{
4419 if (mddev->reshape_position != MaxSector)
4420 return sprintf(page, "%llu\n",
4421 (unsigned long long)mddev->reshape_position);
4422 strcpy(page, "none\n");
4423 return 5;
4424}
4425
4426static ssize_t
fd01b88c 4427reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd
N		 4428{
4429 char *e;
4430 unsigned long long new = simple_strtoull(buf, &e, 10);
4431 if (mddev->pers)
4432 return -EBUSY;
4433 if (buf == e || (*e && *e != '\n'))
4434 return -EINVAL;
4435 mddev->reshape_position = new;
4436 mddev->delta_disks = 0;
4437 mddev->new_level = mddev->level;
4438 mddev->new_layout = mddev->layout;
664e7c41 4439 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4440 return len;
4441}
4442
4443static struct md_sysfs_entry md_reshape_position =
4444__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4445 reshape_position_store);
4446
b522adcd 4447static ssize_t
fd01b88c 4448array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4449{
4450 if (mddev->external_size)
4451 return sprintf(page, "%llu\n",
4452 (unsigned long long)mddev->array_sectors/2);
4453 else
4454 return sprintf(page, "default\n");
4455}
4456
4457static ssize_t
fd01b88c 4458array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4459{
4460 sector_t sectors;
4461
4462 if (strncmp(buf, "default", 7) == 0) {
4463 if (mddev->pers)
4464 sectors = mddev->pers->size(mddev, 0, 0);
4465 else
4466 sectors = mddev->array_sectors;
4467
4468 mddev->external_size = 0;
4469 } else {
4470 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4471 return -EINVAL;
4472 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4473 return -E2BIG;
b522adcd
DW		 4474
4475 mddev->external_size = 1;
4476 }
4477
4478 mddev->array_sectors = sectors;
cbe6ef1d
N		 4479 if (mddev->pers) {
4480 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4481 revalidate_disk(mddev->gendisk);
cbe6ef1d 4482 }
b522adcd
DW		 4483 return len;
4484}
4485
4486static struct md_sysfs_entry md_array_size =
4487__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4488 array_size_store);
e464eafd 4489
eae1701f
N		 4490static struct attribute *md_default_attrs[] = {
4491 &md_level.attr,
d4dbd025 4492 &md_layout.attr,
eae1701f 4493 &md_raid_disks.attr,
3b34380a 4494 &md_chunk_size.attr,
a35b0d69 4495 &md_size.attr,
a94213b1 4496 &md_resync_start.attr,
8bb93aac 4497 &md_metadata.attr,
6d7ff738 4498 &md_new_device.attr,
16f17b39 4499 &md_safe_delay.attr,
9e653b63 4500 &md_array_state.attr,
08a02ecd 4501 &md_reshape_position.attr,
b522adcd 4502 &md_array_size.attr,
1e50915f 4503 &max_corr_read_errors.attr,
411036fa
N		 4504 NULL,
4505};
4506
4507static struct attribute *md_redundancy_attrs[] = {
24dd469d 4508 &md_scan_mode.attr,
9d88883e 4509 &md_mismatches.attr,
88202a0c
N		 4510 &md_sync_min.attr,
4511 &md_sync_max.attr,
4512 &md_sync_speed.attr,
90b08710 4513 &md_sync_force_parallel.attr,
88202a0c 4514 &md_sync_completed.attr,
5e96ee65 4515 &md_min_sync.attr,
c6207277 4516 &md_max_sync.attr,
e464eafd
N		 4517 &md_suspend_lo.attr,
4518 &md_suspend_hi.attr,
9b1d1dac 4519 &md_bitmap.attr,
d7f3d291 4520 &md_degraded.attr,
eae1701f
N		 4521 NULL,
4522};
411036fa
N		 4523static struct attribute_group md_redundancy_group = {
4524 .name = NULL,
4525 .attrs = md_redundancy_attrs,
4526};
4527
eae1701f
N		 4528
4529static ssize_t
4530md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4531{
4532 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4533 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4534 ssize_t rv;
eae1701f
N		 4535
4536 if (!entry->show)
4537 return -EIO;
af8a2434
N		 4538 spin_lock(&all_mddevs_lock);
4539 if (list_empty(&mddev->all_mddevs)) {
4540 spin_unlock(&all_mddevs_lock);
4541 return -EBUSY;
4542 }
4543 mddev_get(mddev);
4544 spin_unlock(&all_mddevs_lock);
4545
5dc5cf7d
IM		 4546 rv = mddev_lock(mddev);
4547 if (!rv) {
4548 rv = entry->show(mddev, page);
4549 mddev_unlock(mddev);
4550 }
af8a2434 4551 mddev_put(mddev);
96de1e66 4552 return rv;
eae1701f
N		 4553}
4554
4555static ssize_t
4556md_attr_store(struct kobject *kobj, struct attribute *attr,
4557 const char *page, size_t length)
4558{
4559 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4560 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4561 ssize_t rv;
eae1701f
N		 4562
4563 if (!entry->store)
4564 return -EIO;
67463acb
N		 4565 if (!capable(CAP_SYS_ADMIN))
4566 return -EACCES;
af8a2434
N		 4567 spin_lock(&all_mddevs_lock);
4568 if (list_empty(&mddev->all_mddevs)) {
4569 spin_unlock(&all_mddevs_lock);
4570 return -EBUSY;
4571 }
4572 mddev_get(mddev);
4573 spin_unlock(&all_mddevs_lock);
5dc5cf7d
IM		 4574 rv = mddev_lock(mddev);
4575 if (!rv) {
4576 rv = entry->store(mddev, page, length);
4577 mddev_unlock(mddev);
4578 }
af8a2434 4579 mddev_put(mddev);
96de1e66 4580 return rv;
eae1701f
N		 4581}
4582
4583static void md_free(struct kobject *ko)
4584{
fd01b88c 4585 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4586
4587 if (mddev->sysfs_state)
4588 sysfs_put(mddev->sysfs_state);
4589
4590 if (mddev->gendisk) {
4591 del_gendisk(mddev->gendisk);
4592 put_disk(mddev->gendisk);
4593 }
4594 if (mddev->queue)
4595 blk_cleanup_queue(mddev->queue);
4596
eae1701f
N		 4597 kfree(mddev);
4598}
4599
52cf25d0 4600static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4601 .show = md_attr_show,
4602 .store = md_attr_store,
4603};
4604static struct kobj_type md_ktype = {
4605 .release = md_free,
4606 .sysfs_ops = &md_sysfs_ops,
4607 .default_attrs = md_default_attrs,
4608};
4609
1da177e4
LT		 4610int mdp_major = 0;
4611
5fd3a17e
DW		 4612static void mddev_delayed_delete(struct work_struct *ws)
4613{
fd01b88c 4614 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4615
43a70507 4616 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4617 kobject_del(&mddev->kobj);
4618 kobject_put(&mddev->kobj);
4619}
4620
efeb53c0 4621static int md_alloc(dev_t dev, char *name)
1da177e4 4622{
48c9c27b 4623 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4624 struct mddev *mddev = mddev_find(dev);
1da177e4 4625 struct gendisk *disk;
efeb53c0
N		 4626 int partitioned;
4627 int shift;
4628 int unit;
3830c62f 4629 int error;
1da177e4
LT		 4630
4631 if (!mddev)
efeb53c0
N		 4632 return -ENODEV;
4633
4634 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4635 shift = partitioned ? MdpMinorShift : 0;
4636 unit = MINOR(mddev->unit) >> shift;
1da177e4 4637
e804ac78
TH		 4638 /* wait for any previous instance of this device to be
4639 * completely removed (mddev_delayed_delete).
d3374825 4640 */
e804ac78 4641 flush_workqueue(md_misc_wq);
d3374825 4642
48c9c27b 4643 mutex_lock(&disks_mutex);
0909dc44
N		 4644 error = -EEXIST;
4645 if (mddev->gendisk)
4646 goto abort;
efeb53c0
N		 4647
4648 if (name) {
4649 /* Need to ensure that 'name' is not a duplicate.
4650 */
fd01b88c 4651 struct mddev *mddev2;
efeb53c0
N		 4652 spin_lock(&all_mddevs_lock);
4653
4654 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4655 if (mddev2->gendisk &&
4656 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4657 spin_unlock(&all_mddevs_lock);
0909dc44 4658 goto abort;
efeb53c0
N		 4659 }
4660 spin_unlock(&all_mddevs_lock);
1da177e4 4661 }
8b765398 4662
0909dc44 4663 error = -ENOMEM;
8b765398 4664 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4665 if (!mddev->queue)
4666 goto abort;
409c57f3
N		 4667 mddev->queue->queuedata = mddev;
4668
409c57f3 4669 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4670
1da177e4
LT		 4671 disk = alloc_disk(1 << shift);
4672 if (!disk) {
8b765398
N		 4673 blk_cleanup_queue(mddev->queue);
4674 mddev->queue = NULL;
0909dc44 4675 goto abort;
1da177e4 4676 }
efeb53c0 4677 disk->major = MAJOR(mddev->unit);
1da177e4 4678 disk->first_minor = unit << shift;
efeb53c0
N		 4679 if (name)
4680 strcpy(disk->disk_name, name);
4681 else if (partitioned)
1da177e4 4682 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4683 else
1da177e4 4684 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4685 disk->fops = &md_fops;
4686 disk->private_data = mddev;
4687 disk->queue = mddev->queue;
b0140891 4688 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4689 /* Allow extended partitions. This makes the
d3374825 4690 * 'mdp' device redundant, but we can't really
92850bbd
N		 4691 * remove it now.
4692 */
4693 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4694 mddev->gendisk = disk;
b0140891
N		 4695 /* As soon as we call add_disk(), another thread could get
4696 * through to md_open, so make sure it doesn't get too far
4697 */
4698 mutex_lock(&mddev->open_mutex);
4699 add_disk(disk);
4700
ed9e1982
TH		 4701 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4702 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4703 if (error) {
4704 /* This isn't possible, but as kobject_init_and_add is marked
4705 * __must_check, we must do something with the result
4706 */
5e55e2f5
N		 4707 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4708 disk->disk_name);
0909dc44
N		 4709 error = 0;
4710 }
00bcb4ac
N		 4711 if (mddev->kobj.sd &&
4712 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4713 printk(KERN_DEBUG "pointless warning\n");
b0140891 4714 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4715 abort:
4716 mutex_unlock(&disks_mutex);
00bcb4ac 4717 if (!error && mddev->kobj.sd) {
3830c62f 4718 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4719 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4720 }
d3374825 4721 mddev_put(mddev);
0909dc44 4722 return error;
efeb53c0
N		 4723}
4724
4725static struct kobject *md_probe(dev_t dev, int *part, void *data)
4726{
4727 md_alloc(dev, NULL);
1da177e4
LT		 4728 return NULL;
4729}
4730
efeb53c0
N		 4731static int add_named_array(const char *val, struct kernel_param *kp)
4732{
4733 /* val must be "md_*" where * is not all digits.
4734 * We allocate an array with a large free minor number, and
4735 * set the name to val. val must not already be an active name.
4736 */
4737 int len = strlen(val);
4738 char buf[DISK_NAME_LEN];
4739
4740 while (len && val[len-1] == '\n')
4741 len--;
4742 if (len >= DISK_NAME_LEN)
4743 return -E2BIG;
4744 strlcpy(buf, val, len+1);
4745 if (strncmp(buf, "md_", 3) != 0)
4746 return -EINVAL;
4747 return md_alloc(0, buf);
4748}
4749
1da177e4
LT		 4750static void md_safemode_timeout(unsigned long data)
4751{
fd01b88c 4752 struct mddev *mddev = (struct mddev *) data;
1da177e4 4753
0fd62b86
NB		 4754 if (!atomic_read(&mddev->writes_pending)) {
4755 mddev->safemode = 1;
4756 if (mddev->external)
00bcb4ac 4757 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4758 }
1da177e4
LT		 4759 md_wakeup_thread(mddev->thread);
4760}
4761
6ff8d8ec 4762static int start_dirty_degraded;
1da177e4 4763
fd01b88c 4764int md_run(struct mddev *mddev)
1da177e4 4765{
2604b703 4766 int err;
3cb03002 4767 struct md_rdev *rdev;
84fc4b56 4768 struct md_personality *pers;
1da177e4 4769
a757e64c
N		 4770 if (list_empty(&mddev->disks))
4771 /* cannot run an array with no devices.. */
1da177e4 4772 return -EINVAL;
1da177e4
LT		 4773
4774 if (mddev->pers)
4775 return -EBUSY;
bb4f1e9d
N		 4776 /* Cannot run until previous stop completes properly */
4777 if (mddev->sysfs_active)
4778 return -EBUSY;
b6eb127d 4779
1da177e4
LT		 4780 /*
4781 * Analyze all RAID superblock(s)
4782 */
1ec4a939
N		 4783 if (!mddev->raid_disks) {
4784 if (!mddev->persistent)
4785 return -EINVAL;
a757e64c 4786 analyze_sbs(mddev);
1ec4a939 4787 }
1da177e4 4788
d9d166c2
N		 4789 if (mddev->level != LEVEL_NONE)
4790 request_module("md-level-%d", mddev->level);
4791 else if (mddev->clevel[0])
4792 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4793
4794 /*
4795 * Drop all container device buffers, from now on
4796 * the only valid external interface is through the md
4797 * device.
1da177e4 4798 */
159ec1fc 4799 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4800 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4801 continue;
4802 sync_blockdev(rdev->bdev);
f98393a6 4803 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4804
4805 /* perform some consistency tests on the device.
4806 * We don't want the data to overlap the metadata,
58c0fed4 4807 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4808 */
a6ff7e08
JB		 4809 if (rdev->meta_bdev) {
4810 /* Nothing to check */;
4811 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4812 if (mddev->dev_sectors &&
4813 rdev->data_offset + mddev->dev_sectors
0f420358 4814 > rdev->sb_start) {
f0d76d70
N		 4815 printk("md: %s: data overlaps metadata\n",
4816 mdname(mddev));
4817 return -EINVAL;
4818 }
4819 } else {
0f420358 4820 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4821 > rdev->data_offset) {
4822 printk("md: %s: metadata overlaps data\n",
4823 mdname(mddev));
4824 return -EINVAL;
4825 }
4826 }
00bcb4ac 4827 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4828 }
4829
a167f663 4830 if (mddev->bio_set == NULL)
a519b26d 4831 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
fd01b88c 4832 sizeof(struct mddev *));
a167f663 4833
1da177e4 4834 spin_lock(&pers_lock);
d9d166c2 4835 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4836 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4837 spin_unlock(&pers_lock);
d9d166c2
N		 4838 if (mddev->level != LEVEL_NONE)
4839 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4840 mddev->level);
4841 else
4842 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4843 mddev->clevel);
1da177e4
LT		 4844 return -EINVAL;
4845 }
2604b703 4846 mddev->pers = pers;
1da177e4 4847 spin_unlock(&pers_lock);
34817e8c
N		 4848 if (mddev->level != pers->level) {
4849 mddev->level = pers->level;
4850 mddev->new_level = pers->level;
4851 }
d9d166c2 4852 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4853
f6705578 4854 if (mddev->reshape_position != MaxSector &&
63c70c4f 4855 pers->start_reshape == NULL) {
f6705578
N		 4856 /* This personality cannot handle reshaping... */
4857 mddev->pers = NULL;
4858 module_put(pers->owner);
4859 return -EINVAL;
4860 }
4861
7dd5e7c3
N		 4862 if (pers->sync_request) {
4863 /* Warn if this is a potentially silly
4864 * configuration.
4865 */
4866 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 4867 struct md_rdev *rdev2;
7dd5e7c3 4868 int warned = 0;
159ec1fc
CR		 4869
4870 list_for_each_entry(rdev, &mddev->disks, same_set)
4871 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4872 if (rdev < rdev2 &&
4873 rdev->bdev->bd_contains ==
4874 rdev2->bdev->bd_contains) {
4875 printk(KERN_WARNING
4876 "%s: WARNING: %s appears to be"
4877 " on the same physical disk as"
4878 " %s.\n",
4879 mdname(mddev),
4880 bdevname(rdev->bdev,b),
4881 bdevname(rdev2->bdev,b2));
4882 warned = 1;
4883 }
4884 }
159ec1fc 4885
7dd5e7c3
N		 4886 if (warned)
4887 printk(KERN_WARNING
4888 "True protection against single-disk"
4889 " failure might be compromised.\n");
4890 }
4891
657390d2 4892 mddev->recovery = 0;
58c0fed4
AN		 4893 /* may be over-ridden by personality */
4894 mddev->resync_max_sectors = mddev->dev_sectors;
4895
6ff8d8ec 4896 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4897
0f9552b5 4898 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4899 mddev->ro = 2; /* read-only, but switch on first write */
4900
b15c2e57 4901 err = mddev->pers->run(mddev);
13e53df3
AN		 4902 if (err)
4903 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4904 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4905 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4906 " but 'external_size' not in effect?\n", __func__);
4907 printk(KERN_ERR
4908 "md: invalid array_size %llu > default size %llu\n",
4909 (unsigned long long)mddev->array_sectors / 2,
4910 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4911 err = -EINVAL;
4912 mddev->pers->stop(mddev);
4913 }
4914 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4915 err = bitmap_create(mddev);
4916 if (err) {
4917 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4918 mdname(mddev), err);
4919 mddev->pers->stop(mddev);
4920 }
4921 }
1da177e4 4922 if (err) {
1da177e4
LT		 4923 module_put(mddev->pers->owner);
4924 mddev->pers = NULL;
32a7627c
N		 4925 bitmap_destroy(mddev);
4926 return err;
1da177e4 4927 }
5e55e2f5 4928 if (mddev->pers->sync_request) {
00bcb4ac
N		 4929 if (mddev->kobj.sd &&
4930 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4931 printk(KERN_WARNING
4932 "md: cannot register extra attributes for %s\n",
4933 mdname(mddev));
00bcb4ac 4934 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4935 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4936 mddev->ro = 0;
4937
1da177e4 4938 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4939 atomic_set(&mddev->max_corr_read_errors,
4940 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4941 mddev->safemode = 0;
4942 mddev->safemode_timer.function = md_safemode_timeout;
4943 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4944 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4945 mddev->in_sync = 1;
0ca69886
N		 4946 smp_wmb();
4947 mddev->ready = 1;
159ec1fc 4948 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 4949 if (rdev->raid_disk >= 0)
4950 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 4951 /* failure here is OK */;
1da177e4
LT		 4952
4953 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4954
850b2b42
N		 4955 if (mddev->flags)
4956 md_update_sb(mddev, 0);
1da177e4 4957
d7603b7e 4958 md_new_event(mddev);
00bcb4ac
N		 4959 sysfs_notify_dirent_safe(mddev->sysfs_state);
4960 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4961 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4962 return 0;
4963}
390ee602 4964EXPORT_SYMBOL_GPL(md_run);
1da177e4 4965
fd01b88c 4966static int do_md_run(struct mddev *mddev)
fe60b014
N		 4967{
4968 int err;
4969
4970 err = md_run(mddev);
4971 if (err)
4972 goto out;
69e51b44
N		 4973 err = bitmap_load(mddev);
4974 if (err) {
4975 bitmap_destroy(mddev);
4976 goto out;
4977 }
0fd018af
JB		 4978
4979 md_wakeup_thread(mddev->thread);
4980 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4981
fe60b014
N		 4982 set_capacity(mddev->gendisk, mddev->array_sectors);
4983 revalidate_disk(mddev->gendisk);
f0b4f7e2 4984 mddev->changed = 1;
fe60b014
N		 4985 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4986out:
4987 return err;
4988}
4989
fd01b88c 4990static int restart_array(struct mddev *mddev)
1da177e4
LT		 4991{
4992 struct gendisk *disk = mddev->gendisk;
1da177e4 4993
80fab1d7 4994 /* Complain if it has no devices */
1da177e4 4995 if (list_empty(&mddev->disks))
80fab1d7
AN		 4996 return -ENXIO;
4997 if (!mddev->pers)
4998 return -EINVAL;
4999 if (!mddev->ro)
5000 return -EBUSY;
5001 mddev->safemode = 0;
5002 mddev->ro = 0;
5003 set_disk_ro(disk, 0);
5004 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5005 mdname(mddev));
5006 /* Kick recovery or resync if necessary */
5007 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5008 md_wakeup_thread(mddev->thread);
5009 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5010 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5011 return 0;
1da177e4
LT		 5012}
5013
acc55e22
N		 5014/* similar to deny_write_access, but accounts for our holding a reference
5015 * to the file ourselves */
5016static int deny_bitmap_write_access(struct file * file)
5017{
5018 struct inode *inode = file->f_mapping->host;
5019
5020 spin_lock(&inode->i_lock);
5021 if (atomic_read(&inode->i_writecount) > 1) {
5022 spin_unlock(&inode->i_lock);
5023 return -ETXTBSY;
5024 }
5025 atomic_set(&inode->i_writecount, -1);
5026 spin_unlock(&inode->i_lock);
5027
5028 return 0;
5029}
5030
43a70507 5031void restore_bitmap_write_access(struct file *file)
acc55e22
N		 5032{
5033 struct inode *inode = file->f_mapping->host;
5034
5035 spin_lock(&inode->i_lock);
5036 atomic_set(&inode->i_writecount, 1);
5037 spin_unlock(&inode->i_lock);
5038}
5039
fd01b88c 5040static void md_clean(struct mddev *mddev)
6177b472
N		 5041{
5042 mddev->array_sectors = 0;
5043 mddev->external_size = 0;
5044 mddev->dev_sectors = 0;
5045 mddev->raid_disks = 0;
5046 mddev->recovery_cp = 0;
5047 mddev->resync_min = 0;
5048 mddev->resync_max = MaxSector;
5049 mddev->reshape_position = MaxSector;
5050 mddev->external = 0;
5051 mddev->persistent = 0;
5052 mddev->level = LEVEL_NONE;
5053 mddev->clevel[0] = 0;
5054 mddev->flags = 0;
5055 mddev->ro = 0;
5056 mddev->metadata_type[0] = 0;
5057 mddev->chunk_sectors = 0;
5058 mddev->ctime = mddev->utime = 0;
5059 mddev->layout = 0;
5060 mddev->max_disks = 0;
5061 mddev->events = 0;
a8707c08 5062 mddev->can_decrease_events = 0;
6177b472
N		 5063 mddev->delta_disks = 0;
5064 mddev->new_level = LEVEL_NONE;
5065 mddev->new_layout = 0;
5066 mddev->new_chunk_sectors = 0;
5067 mddev->curr_resync = 0;
5068 mddev->resync_mismatches = 0;
5069 mddev->suspend_lo = mddev->suspend_hi = 0;
5070 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5071 mddev->recovery = 0;
5072 mddev->in_sync = 0;
f0b4f7e2 5073 mddev->changed = 0;
6177b472 5074 mddev->degraded = 0;
6177b472
N		 5075 mddev->safemode = 0;
5076 mddev->bitmap_info.offset = 0;
5077 mddev->bitmap_info.default_offset = 0;
5078 mddev->bitmap_info.chunksize = 0;
5079 mddev->bitmap_info.daemon_sleep = 0;
5080 mddev->bitmap_info.max_write_behind = 0;
5081}
5082
fd01b88c 5083static void __md_stop_writes(struct mddev *mddev)
a047e125
N		 5084{
5085 if (mddev->sync_thread) {
5086 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5087 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 5088 reap_sync_thread(mddev);
a047e125
N		 5089 }
5090
5091 del_timer_sync(&mddev->safemode_timer);
5092
5093 bitmap_flush(mddev);
5094 md_super_wait(mddev);
5095
5096 if (!mddev->in_sync || mddev->flags) {
5097 /* mark array as shutdown cleanly */
5098 mddev->in_sync = 1;
5099 md_update_sb(mddev, 1);
5100 }
5101}
defad61a 5102
fd01b88c 5103void md_stop_writes(struct mddev *mddev)
defad61a
N		 5104{
5105 mddev_lock(mddev);
5106 __md_stop_writes(mddev);
5107 mddev_unlock(mddev);
5108}
390ee602 5109EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5110
fd01b88c 5111void md_stop(struct mddev *mddev)
6177b472 5112{
0ca69886 5113 mddev->ready = 0;
6177b472
N		 5114 mddev->pers->stop(mddev);
5115 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5116 mddev->to_remove = &md_redundancy_group;
5117 module_put(mddev->pers->owner);
5118 mddev->pers = NULL;
cca9cf90 5119 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5120}
390ee602 5121EXPORT_SYMBOL_GPL(md_stop);
6177b472 5122
fd01b88c 5123static int md_set_readonly(struct mddev *mddev, int is_open)
a4bd82d0
N		 5124{
5125 int err = 0;
5126 mutex_lock(&mddev->open_mutex);
5127 if (atomic_read(&mddev->openers) > is_open) {
5128 printk("md: %s still in use.\n",mdname(mddev));
5129 err = -EBUSY;
5130 goto out;
5131 }
5132 if (mddev->pers) {
defad61a 5133 __md_stop_writes(mddev);
a4bd82d0
N		 5134
5135 err = -ENXIO;
5136 if (mddev->ro==1)
5137 goto out;
5138 mddev->ro = 1;
5139 set_disk_ro(mddev->gendisk, 1);
5140 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5141 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5142 err = 0;
5143 }
5144out:
5145 mutex_unlock(&mddev->open_mutex);
5146 return err;
5147}
5148
9e653b63
N		 5149/* mode:
5150 * 0 - completely stop and dis-assemble array
9e653b63
N		 5151 * 2 - stop but do not disassemble array
5152 */
fd01b88c 5153static int do_md_stop(struct mddev * mddev, int mode, int is_open)
1da177e4 5154{
1da177e4 5155 struct gendisk *disk = mddev->gendisk;
3cb03002 5156 struct md_rdev *rdev;
1da177e4 5157
c8c00a69 5158 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 5159 if (atomic_read(&mddev->openers) > is_open ||
5160 mddev->sysfs_active) {
df5b20cf 5161 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5162 mutex_unlock(&mddev->open_mutex);
5163 return -EBUSY;
5164 }
1da177e4 5165
6e17b027 5166 if (mddev->pers) {
a4bd82d0
N		 5167 if (mddev->ro)
5168 set_disk_ro(disk, 0);
409c57f3 5169
defad61a 5170 __md_stop_writes(mddev);
a4bd82d0
N		 5171 md_stop(mddev);
5172 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5173 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5174
a4bd82d0 5175 /* tell userspace to handle 'inactive' */
00bcb4ac 5176 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5177
a4bd82d0 5178 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 5179 if (rdev->raid_disk >= 0)
5180 sysfs_unlink_rdev(mddev, rdev);
c4647292 5181
a4bd82d0 5182 set_capacity(disk, 0);
6e17b027 5183 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5184 mddev->changed = 1;
a4bd82d0 5185 revalidate_disk(disk);
0d4ca600 5186
a4bd82d0
N		 5187 if (mddev->ro)
5188 mddev->ro = 0;
6e17b027
N		 5189 } else
5190 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5191 /*
5192 * Free resources if final stop
5193 */
9e653b63 5194 if (mode == 0) {
1da177e4
LT		 5195 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5196
978f946b 5197 bitmap_destroy(mddev);
c3d9714e
N		 5198 if (mddev->bitmap_info.file) {
5199 restore_bitmap_write_access(mddev->bitmap_info.file);
5200 fput(mddev->bitmap_info.file);
5201 mddev->bitmap_info.file = NULL;
978f946b 5202 }
c3d9714e 5203 mddev->bitmap_info.offset = 0;
978f946b 5204
1da177e4
LT		 5205 export_array(mddev);
5206
6177b472 5207 md_clean(mddev);
934d9c23 5208 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5209 if (mddev->hold_active == UNTIL_STOP)
5210 mddev->hold_active = 0;
a4bd82d0 5211 }
3f9d99c1 5212 blk_integrity_unregister(disk);
d7603b7e 5213 md_new_event(mddev);
00bcb4ac 5214 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5215 return 0;
1da177e4
LT		 5216}
5217
fdee8ae4 5218#ifndef MODULE
fd01b88c 5219static void autorun_array(struct mddev *mddev)
1da177e4 5220{
3cb03002 5221 struct md_rdev *rdev;
1da177e4
LT		 5222 int err;
5223
a757e64c 5224 if (list_empty(&mddev->disks))
1da177e4 5225 return;
1da177e4
LT		 5226
5227 printk(KERN_INFO "md: running: ");
5228
159ec1fc 5229 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 5230 char b[BDEVNAME_SIZE];
5231 printk("<%s>", bdevname(rdev->bdev,b));
5232 }
5233 printk("\n");
5234
d710e138 5235 err = do_md_run(mddev);
1da177e4
LT		 5236 if (err) {
5237 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 5238 do_md_stop(mddev, 0, 0);
1da177e4
LT		 5239 }
5240}
5241
5242/*
5243 * lets try to run arrays based on all disks that have arrived
5244 * until now. (those are in pending_raid_disks)
5245 *
5246 * the method: pick the first pending disk, collect all disks with
5247 * the same UUID, remove all from the pending list and put them into
5248 * the 'same_array' list. Then order this list based on superblock
5249 * update time (freshest comes first), kick out 'old' disks and
5250 * compare superblocks. If everything's fine then run it.
5251 *
5252 * If "unit" is allocated, then bump its reference count
5253 */
5254static void autorun_devices(int part)
5255{
3cb03002 5256 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5257 struct mddev *mddev;
1da177e4
LT		 5258 char b[BDEVNAME_SIZE];
5259
5260 printk(KERN_INFO "md: autorun ...\n");
5261 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5262 int unit;
1da177e4 5263 dev_t dev;
ad01c9e3 5264 LIST_HEAD(candidates);
1da177e4 5265 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5266 struct md_rdev, same_set);
1da177e4
LT		 5267
5268 printk(KERN_INFO "md: considering %s ...\n",
5269 bdevname(rdev0->bdev,b));
5270 INIT_LIST_HEAD(&candidates);
159ec1fc 5271 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5272 if (super_90_load(rdev, rdev0, 0) >= 0) {
5273 printk(KERN_INFO "md: adding %s ...\n",
5274 bdevname(rdev->bdev,b));
5275 list_move(&rdev->same_set, &candidates);
5276 }
5277 /*
5278 * now we have a set of devices, with all of them having
5279 * mostly sane superblocks. It's time to allocate the
5280 * mddev.
5281 */
e8703fe1
N		 5282 if (part) {
5283 dev = MKDEV(mdp_major,
5284 rdev0->preferred_minor << MdpMinorShift);
5285 unit = MINOR(dev) >> MdpMinorShift;
5286 } else {
5287 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5288 unit = MINOR(dev);
5289 }
5290 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5291 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5292 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5293 break;
5294 }
1da177e4
LT		 5295
5296 md_probe(dev, NULL, NULL);
5297 mddev = mddev_find(dev);
9bbbca3a
NB		 5298 if (!mddev || !mddev->gendisk) {
5299 if (mddev)
5300 mddev_put(mddev);
5301 printk(KERN_ERR
1da177e4
LT		 5302 "md: cannot allocate memory for md drive.\n");
5303 break;
5304 }
5305 if (mddev_lock(mddev))
5306 printk(KERN_WARNING "md: %s locked, cannot run\n",
5307 mdname(mddev));
5308 else if (mddev->raid_disks || mddev->major_version
5309 || !list_empty(&mddev->disks)) {
5310 printk(KERN_WARNING
5311 "md: %s already running, cannot run %s\n",
5312 mdname(mddev), bdevname(rdev0->bdev,b));
5313 mddev_unlock(mddev);
5314 } else {
5315 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5316 mddev->persistent = 1;
159ec1fc 5317 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5318 list_del_init(&rdev->same_set);
5319 if (bind_rdev_to_array(rdev, mddev))
5320 export_rdev(rdev);
5321 }
5322 autorun_array(mddev);
5323 mddev_unlock(mddev);
5324 }
5325 /* on success, candidates will be empty, on error
5326 * it won't...
5327 */
159ec1fc 5328 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5329 list_del_init(&rdev->same_set);
1da177e4 5330 export_rdev(rdev);
4b80991c 5331 }
1da177e4
LT		 5332 mddev_put(mddev);
5333 }
5334 printk(KERN_INFO "md: ... autorun DONE.\n");
5335}
fdee8ae4 5336#endif /* !MODULE */
1da177e4 5337
1da177e4
LT		 5338static int get_version(void __user * arg)
5339{
5340 mdu_version_t ver;
5341
5342 ver.major = MD_MAJOR_VERSION;
5343 ver.minor = MD_MINOR_VERSION;
5344 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5345
5346 if (copy_to_user(arg, &ver, sizeof(ver)))
5347 return -EFAULT;
5348
5349 return 0;
5350}
5351
fd01b88c 5352static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5353{
5354 mdu_array_info_t info;
a9f326eb 5355 int nr,working,insync,failed,spare;
3cb03002 5356 struct md_rdev *rdev;
1da177e4 5357
a9f326eb 5358 nr=working=insync=failed=spare=0;
159ec1fc 5359 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 5360 nr++;
b2d444d7 5361 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5362 failed++;
5363 else {
5364 working++;
b2d444d7 5365 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5366 insync++;
1da177e4
LT		 5367 else
5368 spare++;
5369 }
5370 }
5371
5372 info.major_version = mddev->major_version;
5373 info.minor_version = mddev->minor_version;
5374 info.patch_version = MD_PATCHLEVEL_VERSION;
5375 info.ctime = mddev->ctime;
5376 info.level = mddev->level;
58c0fed4
AN		 5377 info.size = mddev->dev_sectors / 2;
5378 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5379 info.size = -1;
1da177e4
LT		 5380 info.nr_disks = nr;
5381 info.raid_disks = mddev->raid_disks;
5382 info.md_minor = mddev->md_minor;
5383 info.not_persistent= !mddev->persistent;
5384
5385 info.utime = mddev->utime;
5386 info.state = 0;
5387 if (mddev->in_sync)
5388 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5389 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5390 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5391 info.active_disks = insync;
1da177e4
LT		 5392 info.working_disks = working;
5393 info.failed_disks = failed;
5394 info.spare_disks = spare;
5395
5396 info.layout = mddev->layout;
9d8f0363 5397 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5398
5399 if (copy_to_user(arg, &info, sizeof(info)))
5400 return -EFAULT;
5401
5402 return 0;
5403}
5404
fd01b88c 5405static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5406{
5407 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5408 char *ptr, *buf = NULL;
5409 int err = -ENOMEM;
5410
b5470dc5
DW		 5411 if (md_allow_write(mddev))
5412 file = kmalloc(sizeof(*file), GFP_NOIO);
5413 else
5414 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5415
32a7627c
N		 5416 if (!file)
5417 goto out;
5418
5419 /* bitmap disabled, zero the first byte and copy out */
5420 if (!mddev->bitmap || !mddev->bitmap->file) {
5421 file->pathname[0] = '\0';
5422 goto copy_out;
5423 }
5424
5425 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5426 if (!buf)
5427 goto out;
5428
6bcfd601
CH		 5429 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5430 if (IS_ERR(ptr))
32a7627c
N		 5431 goto out;
5432
5433 strcpy(file->pathname, ptr);
5434
5435copy_out:
5436 err = 0;
5437 if (copy_to_user(arg, file, sizeof(*file)))
5438 err = -EFAULT;
5439out:
5440 kfree(buf);
5441 kfree(file);
5442 return err;
5443}
5444
fd01b88c 5445static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5446{
5447 mdu_disk_info_t info;
3cb03002 5448 struct md_rdev *rdev;
1da177e4
LT		 5449
5450 if (copy_from_user(&info, arg, sizeof(info)))
5451 return -EFAULT;
5452
26ef379f 5453 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5454 if (rdev) {
5455 info.major = MAJOR(rdev->bdev->bd_dev);
5456 info.minor = MINOR(rdev->bdev->bd_dev);
5457 info.raid_disk = rdev->raid_disk;
5458 info.state = 0;
b2d444d7 5459 if (test_bit(Faulty, &rdev->flags))
1da177e4 5460 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5461 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5462 info.state |= (1<<MD_DISK_ACTIVE);
5463 info.state |= (1<<MD_DISK_SYNC);
5464 }
8ddf9efe
N		 5465 if (test_bit(WriteMostly, &rdev->flags))
5466 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5467 } else {
5468 info.major = info.minor = 0;
5469 info.raid_disk = -1;
5470 info.state = (1<<MD_DISK_REMOVED);
5471 }
5472
5473 if (copy_to_user(arg, &info, sizeof(info)))
5474 return -EFAULT;
5475
5476 return 0;
5477}
5478
fd01b88c 5479static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT		 5480{
5481 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5482 struct md_rdev *rdev;
1da177e4
LT		 5483 dev_t dev = MKDEV(info->major,info->minor);
5484
5485 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5486 return -EOVERFLOW;
5487
5488 if (!mddev->raid_disks) {
5489 int err;
5490 /* expecting a device which has a superblock */
5491 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5492 if (IS_ERR(rdev)) {
5493 printk(KERN_WARNING
5494 "md: md_import_device returned %ld\n",
5495 PTR_ERR(rdev));
5496 return PTR_ERR(rdev);
5497 }
5498 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5499 struct md_rdev *rdev0
5500 = list_entry(mddev->disks.next,
5501 struct md_rdev, same_set);
a9f326eb 5502 err = super_types[mddev->major_version]
1da177e4
LT		 5503 .load_super(rdev, rdev0, mddev->minor_version);
5504 if (err < 0) {
5505 printk(KERN_WARNING
5506 "md: %s has different UUID to %s\n",
5507 bdevname(rdev->bdev,b),
5508 bdevname(rdev0->bdev,b2));
5509 export_rdev(rdev);
5510 return -EINVAL;
5511 }
5512 }
5513 err = bind_rdev_to_array(rdev, mddev);
5514 if (err)
5515 export_rdev(rdev);
5516 return err;
5517 }
5518
5519 /*
5520 * add_new_disk can be used once the array is assembled
5521 * to add "hot spares". They must already have a superblock
5522 * written
5523 */
5524 if (mddev->pers) {
5525 int err;
5526 if (!mddev->pers->hot_add_disk) {
5527 printk(KERN_WARNING
5528 "%s: personality does not support diskops!\n",
5529 mdname(mddev));
5530 return -EINVAL;
5531 }
7b1e35f6
N		 5532 if (mddev->persistent)
5533 rdev = md_import_device(dev, mddev->major_version,
5534 mddev->minor_version);
5535 else
5536 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5537 if (IS_ERR(rdev)) {
5538 printk(KERN_WARNING
5539 "md: md_import_device returned %ld\n",
5540 PTR_ERR(rdev));
5541 return PTR_ERR(rdev);
5542 }
1a855a06 5543 /* set saved_raid_disk if appropriate */
41158c7e
N		 5544 if (!mddev->persistent) {
5545 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5546 info->raid_disk < mddev->raid_disks) {
41158c7e 5547 rdev->raid_disk = info->raid_disk;
bf572541
N		 5548 set_bit(In_sync, &rdev->flags);
5549 } else
41158c7e
N		 5550 rdev->raid_disk = -1;
5551 } else
5552 super_types[mddev->major_version].
5553 validate_super(mddev, rdev);
bedd86b7
N		 5554 if ((info->state & (1<<MD_DISK_SYNC)) &&
5555 (!test_bit(In_sync, &rdev->flags) ||
5556 rdev->raid_disk != info->raid_disk)) {
5557 /* This was a hot-add request, but events doesn't
5558 * match, so reject it.
5559 */
5560 export_rdev(rdev);
5561 return -EINVAL;
5562 }
5563
1a855a06
N		 5564 if (test_bit(In_sync, &rdev->flags))
5565 rdev->saved_raid_disk = rdev->raid_disk;
5566 else
5567 rdev->saved_raid_disk = -1;
41158c7e 5568
b2d444d7 5569 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5570 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5571 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5572 else
5573 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5574
1da177e4
LT		 5575 rdev->raid_disk = -1;
5576 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5577 if (!err && !mddev->pers->hot_remove_disk) {
5578 /* If there is hot_add_disk but no hot_remove_disk
5579 * then added disks for geometry changes,
5580 * and should be added immediately.
5581 */
5582 super_types[mddev->major_version].
5583 validate_super(mddev, rdev);
5584 err = mddev->pers->hot_add_disk(mddev, rdev);
5585 if (err)
5586 unbind_rdev_from_array(rdev);
5587 }
1da177e4
LT		 5588 if (err)
5589 export_rdev(rdev);
52664732 5590 else
00bcb4ac 5591 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5592
17571284 5593 md_update_sb(mddev, 1);
72a23c21
NB		 5594 if (mddev->degraded)
5595 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5596 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5597 if (!err)
5598 md_new_event(mddev);
005eca5e 5599 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5600 return err;
5601 }
5602
5603 /* otherwise, add_new_disk is only allowed
5604 * for major_version==0 superblocks
5605 */
5606 if (mddev->major_version != 0) {
5607 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5608 mdname(mddev));
5609 return -EINVAL;
5610 }
5611
5612 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5613 int err;
d710e138 5614 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5615 if (IS_ERR(rdev)) {
5616 printk(KERN_WARNING
5617 "md: error, md_import_device() returned %ld\n",
5618 PTR_ERR(rdev));
5619 return PTR_ERR(rdev);
5620 }
5621 rdev->desc_nr = info->number;
5622 if (info->raid_disk < mddev->raid_disks)
5623 rdev->raid_disk = info->raid_disk;
5624 else
5625 rdev->raid_disk = -1;
5626
1da177e4 5627 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5628 if (info->state & (1<<MD_DISK_SYNC))
5629 set_bit(In_sync, &rdev->flags);
1da177e4 5630
8ddf9efe
N		 5631 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5632 set_bit(WriteMostly, &rdev->flags);
5633
1da177e4
LT		 5634 if (!mddev->persistent) {
5635 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5636 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5637 } else
57b2caa3 5638 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5639 rdev->sectors = rdev->sb_start;
1da177e4 5640
2bf071bf
N		 5641 err = bind_rdev_to_array(rdev, mddev);
5642 if (err) {
5643 export_rdev(rdev);
5644 return err;
5645 }
1da177e4
LT		 5646 }
5647
5648 return 0;
5649}
5650
fd01b88c 5651static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5652{
5653 char b[BDEVNAME_SIZE];
3cb03002 5654 struct md_rdev *rdev;
1da177e4 5655
1da177e4
LT		 5656 rdev = find_rdev(mddev, dev);
5657 if (!rdev)
5658 return -ENXIO;
5659
5660 if (rdev->raid_disk >= 0)
5661 goto busy;
5662
5663 kick_rdev_from_array(rdev);
850b2b42 5664 md_update_sb(mddev, 1);
d7603b7e 5665 md_new_event(mddev);
1da177e4
LT		 5666
5667 return 0;
5668busy:
fdefa4d8 5669 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5670 bdevname(rdev->bdev,b), mdname(mddev));
5671 return -EBUSY;
5672}
5673
fd01b88c 5674static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5675{
5676 char b[BDEVNAME_SIZE];
5677 int err;
3cb03002 5678 struct md_rdev *rdev;
1da177e4
LT		 5679
5680 if (!mddev->pers)
5681 return -ENODEV;
5682
5683 if (mddev->major_version != 0) {
5684 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5685 " version-0 superblocks.\n",
5686 mdname(mddev));
5687 return -EINVAL;
5688 }
5689 if (!mddev->pers->hot_add_disk) {
5690 printk(KERN_WARNING
5691 "%s: personality does not support diskops!\n",
5692 mdname(mddev));
5693 return -EINVAL;
5694 }
5695
d710e138 5696 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5697 if (IS_ERR(rdev)) {
5698 printk(KERN_WARNING
5699 "md: error, md_import_device() returned %ld\n",
5700 PTR_ERR(rdev));
5701 return -EINVAL;
5702 }
5703
5704 if (mddev->persistent)
57b2caa3 5705 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5706 else
77304d2a 5707 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5708
8190e754 5709 rdev->sectors = rdev->sb_start;
1da177e4 5710
b2d444d7 5711 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5712 printk(KERN_WARNING
5713 "md: can not hot-add faulty %s disk to %s!\n",
5714 bdevname(rdev->bdev,b), mdname(mddev));
5715 err = -EINVAL;
5716 goto abort_export;
5717 }
b2d444d7 5718 clear_bit(In_sync, &rdev->flags);
1da177e4 5719 rdev->desc_nr = -1;
5842730d 5720 rdev->saved_raid_disk = -1;
2bf071bf
N		 5721 err = bind_rdev_to_array(rdev, mddev);
5722 if (err)
5723 goto abort_export;
1da177e4
LT		 5724
5725 /*
5726 * The rest should better be atomic, we can have disk failures
5727 * noticed in interrupt contexts ...
5728 */
5729
1da177e4
LT		 5730 rdev->raid_disk = -1;
5731
850b2b42 5732 md_update_sb(mddev, 1);
1da177e4
LT		 5733
5734 /*
5735 * Kick recovery, maybe this spare has to be added to the
5736 * array immediately.
5737 */
5738 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5739 md_wakeup_thread(mddev->thread);
d7603b7e 5740 md_new_event(mddev);
1da177e4
LT		 5741 return 0;
5742
1da177e4
LT		 5743abort_export:
5744 export_rdev(rdev);
5745 return err;
5746}
5747
fd01b88c 5748static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N		 5749{
5750 int err;
5751
36fa3063
N		 5752 if (mddev->pers) {
5753 if (!mddev->pers->quiesce)
5754 return -EBUSY;
5755 if (mddev->recovery || mddev->sync_thread)
5756 return -EBUSY;
5757 /* we should be able to change the bitmap.. */
5758 }
32a7627c 5759
32a7627c 5760
36fa3063
N		 5761 if (fd >= 0) {
5762 if (mddev->bitmap)
5763 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5764 mddev->bitmap_info.file = fget(fd);
32a7627c 5765
c3d9714e 5766 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5767 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5768 mdname(mddev));
5769 return -EBADF;
5770 }
5771
c3d9714e 5772 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5773 if (err) {
5774 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5775 mdname(mddev));
c3d9714e
N		 5776 fput(mddev->bitmap_info.file);
5777 mddev->bitmap_info.file = NULL;
36fa3063
N		 5778 return err;
5779 }
c3d9714e 5780 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5781 } else if (mddev->bitmap == NULL)
5782 return -ENOENT; /* cannot remove what isn't there */
5783 err = 0;
5784 if (mddev->pers) {
5785 mddev->pers->quiesce(mddev, 1);
69e51b44 5786 if (fd >= 0) {
36fa3063 5787 err = bitmap_create(mddev);
69e51b44
N		 5788 if (!err)
5789 err = bitmap_load(mddev);
5790 }
d7375ab3 5791 if (fd < 0 || err) {
36fa3063 5792 bitmap_destroy(mddev);
d7375ab3
N		 5793 fd = -1; /* make sure to put the file */
5794 }
36fa3063 5795 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5796 }
5797 if (fd < 0) {
c3d9714e
N		 5798 if (mddev->bitmap_info.file) {
5799 restore_bitmap_write_access(mddev->bitmap_info.file);
5800 fput(mddev->bitmap_info.file);
acc55e22 5801 }
c3d9714e 5802 mddev->bitmap_info.file = NULL;
36fa3063
N		 5803 }
5804
32a7627c
N		 5805 return err;
5806}
5807
1da177e4
LT		 5808/*
5809 * set_array_info is used two different ways
5810 * The original usage is when creating a new array.
5811 * In this usage, raid_disks is > 0 and it together with
5812 * level, size, not_persistent,layout,chunksize determine the
5813 * shape of the array.
5814 * This will always create an array with a type-0.90.0 superblock.
5815 * The newer usage is when assembling an array.
5816 * In this case raid_disks will be 0, and the major_version field is
5817 * use to determine which style super-blocks are to be found on the devices.
5818 * The minor and patch _version numbers are also kept incase the
5819 * super_block handler wishes to interpret them.
5820 */
fd01b88c 5821static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT		 5822{
5823
5824 if (info->raid_disks == 0) {
5825 /* just setting version number for superblock loading */
5826 if (info->major_version < 0 ||
50511da3 5827 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5828 super_types[info->major_version].name == NULL) {
5829 /* maybe try to auto-load a module? */
5830 printk(KERN_INFO
5831 "md: superblock version %d not known\n",
5832 info->major_version);
5833 return -EINVAL;
5834 }
5835 mddev->major_version = info->major_version;
5836 mddev->minor_version = info->minor_version;
5837 mddev->patch_version = info->patch_version;
3f9d7b0d 5838 mddev->persistent = !info->not_persistent;
cbd19983
N		 5839 /* ensure mddev_put doesn't delete this now that there
5840 * is some minimal configuration.
5841 */
5842 mddev->ctime = get_seconds();
1da177e4
LT		 5843 return 0;
5844 }
5845 mddev->major_version = MD_MAJOR_VERSION;
5846 mddev->minor_version = MD_MINOR_VERSION;
5847 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5848 mddev->ctime = get_seconds();
5849
5850 mddev->level = info->level;
17115e03 5851 mddev->clevel[0] = 0;
58c0fed4 5852 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5853 mddev->raid_disks = info->raid_disks;
5854 /* don't set md_minor, it is determined by which /dev/md* was
5855 * openned
5856 */
5857 if (info->state & (1<<MD_SB_CLEAN))
5858 mddev->recovery_cp = MaxSector;
5859 else
5860 mddev->recovery_cp = 0;
5861 mddev->persistent = ! info->not_persistent;
e691063a 5862 mddev->external = 0;
1da177e4
LT		 5863
5864 mddev->layout = info->layout;
9d8f0363 5865 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5866
5867 mddev->max_disks = MD_SB_DISKS;
5868
e691063a
N		 5869 if (mddev->persistent)
5870 mddev->flags = 0;
850b2b42 5871 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5872
c3d9714e
N		 5873 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5874 mddev->bitmap_info.offset = 0;
b2a2703c 5875
f6705578
N		 5876 mddev->reshape_position = MaxSector;
5877
1da177e4
LT		 5878 /*
5879 * Generate a 128 bit UUID
5880 */
5881 get_random_bytes(mddev->uuid, 16);
5882
f6705578 5883 mddev->new_level = mddev->level;
664e7c41 5884 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5885 mddev->new_layout = mddev->layout;
5886 mddev->delta_disks = 0;
5887
1da177e4
LT		 5888 return 0;
5889}
5890
fd01b88c 5891void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 5892{
b522adcd
DW		 5893 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5894
5895 if (mddev->external_size)
5896 return;
5897
1f403624
DW		 5898 mddev->array_sectors = array_sectors;
5899}
5900EXPORT_SYMBOL(md_set_array_sectors);
5901
fd01b88c 5902static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 5903{
3cb03002 5904 struct md_rdev *rdev;
a35b0d69 5905 int rv;
d71f9f88 5906 int fit = (num_sectors == 0);
a35b0d69
N		 5907
5908 if (mddev->pers->resize == NULL)
5909 return -EINVAL;
d71f9f88
AN		 5910 /* The "num_sectors" is the number of sectors of each device that
5911 * is used. This can only make sense for arrays with redundancy.
5912 * linear and raid0 always use whatever space is available. We can only
5913 * consider changing this number if no resync or reconstruction is
5914 * happening, and if the new size is acceptable. It must fit before the
0f420358 5915 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5916 * of each device. If num_sectors is zero, we find the largest size
5917 * that fits.
a35b0d69
N		 5918 */
5919 if (mddev->sync_thread)
5920 return -EBUSY;
dba034ee
N		 5921 if (mddev->bitmap)
5922 /* Sorry, cannot grow a bitmap yet, just remove it,
5923 * grow, and re-add.
5924 */
5925 return -EBUSY;
159ec1fc 5926 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5927 sector_t avail = rdev->sectors;
01ab5662 5928
d71f9f88
AN		 5929 if (fit && (num_sectors == 0 || num_sectors > avail))
5930 num_sectors = avail;
5931 if (avail < num_sectors)
a35b0d69
N		 5932 return -ENOSPC;
5933 }
d71f9f88 5934 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5935 if (!rv)
5936 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5937 return rv;
5938}
5939
fd01b88c 5940static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 5941{
5942 int rv;
5943 /* change the number of raid disks */
63c70c4f 5944 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5945 return -EINVAL;
5946 if (raid_disks <= 0 ||
233fca36 5947 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5948 return -EINVAL;
63c70c4f 5949 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5950 return -EBUSY;
63c70c4f
N		 5951 mddev->delta_disks = raid_disks - mddev->raid_disks;
5952
5953 rv = mddev->pers->check_reshape(mddev);
de171cb9
N		 5954 if (rv < 0)
5955 mddev->delta_disks = 0;
da943b99
N		 5956 return rv;
5957}
5958
5959
1da177e4
LT		 5960/*
5961 * update_array_info is used to change the configuration of an
5962 * on-line array.
5963 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5964 * fields in the info are checked against the array.
5965 * Any differences that cannot be handled will cause an error.
5966 * Normally, only one change can be managed at a time.
5967 */
fd01b88c 5968static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 5969{
5970 int rv = 0;
5971 int cnt = 0;
36fa3063
N		 5972 int state = 0;
5973
5974 /* calculate expected state,ignoring low bits */
c3d9714e 5975 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5976 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5977
5978 if (mddev->major_version != info->major_version ||
5979 mddev->minor_version != info->minor_version ||
5980/* mddev->patch_version != info->patch_version || */
5981 mddev->ctime != info->ctime ||
5982 mddev->level != info->level ||
5983/* mddev->layout != info->layout || */
5984 !mddev->persistent != info->not_persistent||
9d8f0363 5985 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5986 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5987 ((state^info->state) & 0xfffffe00)
5988 )
1da177e4
LT		 5989 return -EINVAL;
5990 /* Check there is only one change */
58c0fed4
AN		 5991 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5992 cnt++;
5993 if (mddev->raid_disks != info->raid_disks)
5994 cnt++;
5995 if (mddev->layout != info->layout)
5996 cnt++;
5997 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5998 cnt++;
5999 if (cnt == 0)
6000 return 0;
6001 if (cnt > 1)
6002 return -EINVAL;
1da177e4
LT		 6003
6004 if (mddev->layout != info->layout) {
6005 /* Change layout
6006 * we don't need to do anything at the md level, the
6007 * personality will take care of it all.
6008 */
50ac168a 6009 if (mddev->pers->check_reshape == NULL)
1da177e4 6010 return -EINVAL;
597a711b
N		 6011 else {
6012 mddev->new_layout = info->layout;
50ac168a 6013 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6014 if (rv)
6015 mddev->new_layout = mddev->layout;
6016 return rv;
6017 }
1da177e4 6018 }
58c0fed4 6019 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6020 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6021
da943b99
N		 6022 if (mddev->raid_disks != info->raid_disks)
6023 rv = update_raid_disks(mddev, info->raid_disks);
6024
36fa3063
N		 6025 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6026 if (mddev->pers->quiesce == NULL)
6027 return -EINVAL;
6028 if (mddev->recovery || mddev->sync_thread)
6029 return -EBUSY;
6030 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6031 /* add the bitmap */
6032 if (mddev->bitmap)
6033 return -EEXIST;
c3d9714e 6034 if (mddev->bitmap_info.default_offset == 0)
36fa3063 6035 return -EINVAL;
c3d9714e
N		 6036 mddev->bitmap_info.offset =
6037 mddev->bitmap_info.default_offset;
36fa3063
N		 6038 mddev->pers->quiesce(mddev, 1);
6039 rv = bitmap_create(mddev);
69e51b44
N		 6040 if (!rv)
6041 rv = bitmap_load(mddev);
36fa3063
N		 6042 if (rv)
6043 bitmap_destroy(mddev);
6044 mddev->pers->quiesce(mddev, 0);
6045 } else {
6046 /* remove the bitmap */
6047 if (!mddev->bitmap)
6048 return -ENOENT;
6049 if (mddev->bitmap->file)
6050 return -EINVAL;
6051 mddev->pers->quiesce(mddev, 1);
6052 bitmap_destroy(mddev);
6053 mddev->pers->quiesce(mddev, 0);
c3d9714e 6054 mddev->bitmap_info.offset = 0;
36fa3063
N		 6055 }
6056 }
850b2b42 6057 md_update_sb(mddev, 1);
1da177e4
LT		 6058 return rv;
6059}
6060
fd01b88c 6061static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6062{
3cb03002 6063 struct md_rdev *rdev;
1da177e4
LT		 6064
6065 if (mddev->pers == NULL)
6066 return -ENODEV;
6067
6068 rdev = find_rdev(mddev, dev);
6069 if (!rdev)
6070 return -ENODEV;
6071
6072 md_error(mddev, rdev);
5ef56c8f
N		 6073 if (!test_bit(Faulty, &rdev->flags))
6074 return -EBUSY;
1da177e4
LT		 6075 return 0;
6076}
6077
2f9618ce
AN		 6078/*
6079 * We have a problem here : there is no easy way to give a CHS
6080 * virtual geometry. We currently pretend that we have a 2 heads
6081 * 4 sectors (with a BIG number of cylinders...). This drives
6082 * dosfs just mad... ;-)
6083 */
a885c8c4
CH		 6084static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6085{
fd01b88c 6086 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6087
6088 geo->heads = 2;
6089 geo->sectors = 4;
49ce6cea 6090 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6091 return 0;
6092}
6093
a39907fa 6094static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6095 unsigned int cmd, unsigned long arg)
6096{
6097 int err = 0;
6098 void __user *argp = (void __user *)arg;
fd01b88c 6099 struct mddev *mddev = NULL;
e2218350 6100 int ro;
1da177e4 6101
506c9e44
N		 6102 switch (cmd) {
6103 case RAID_VERSION:
6104 case GET_ARRAY_INFO:
6105 case GET_DISK_INFO:
6106 break;
6107 default:
6108 if (!capable(CAP_SYS_ADMIN))
6109 return -EACCES;
6110 }
1da177e4
LT		 6111
6112 /*
6113 * Commands dealing with the RAID driver but not any
6114 * particular array:
6115 */
6116 switch (cmd)
6117 {
6118 case RAID_VERSION:
6119 err = get_version(argp);
6120 goto done;
6121
6122 case PRINT_RAID_DEBUG:
6123 err = 0;
6124 md_print_devices();
6125 goto done;
6126
6127#ifndef MODULE
6128 case RAID_AUTORUN:
6129 err = 0;
6130 autostart_arrays(arg);
6131 goto done;
6132#endif
6133 default:;
6134 }
6135
6136 /*
6137 * Commands creating/starting a new array:
6138 */
6139
a39907fa 6140 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6141
6142 if (!mddev) {
6143 BUG();
6144 goto abort;
6145 }
6146
1da177e4
LT		 6147 err = mddev_lock(mddev);
6148 if (err) {
6149 printk(KERN_INFO
6150 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6151 err, cmd);
6152 goto abort;
6153 }
6154
6155 switch (cmd)
6156 {
6157 case SET_ARRAY_INFO:
6158 {
6159 mdu_array_info_t info;
6160 if (!arg)
6161 memset(&info, 0, sizeof(info));
6162 else if (copy_from_user(&info, argp, sizeof(info))) {
6163 err = -EFAULT;
6164 goto abort_unlock;
6165 }
6166 if (mddev->pers) {
6167 err = update_array_info(mddev, &info);
6168 if (err) {
6169 printk(KERN_WARNING "md: couldn't update"
6170 " array info. %d\n", err);
6171 goto abort_unlock;
6172 }
6173 goto done_unlock;
6174 }
6175 if (!list_empty(&mddev->disks)) {
6176 printk(KERN_WARNING
6177 "md: array %s already has disks!\n",
6178 mdname(mddev));
6179 err = -EBUSY;
6180 goto abort_unlock;
6181 }
6182 if (mddev->raid_disks) {
6183 printk(KERN_WARNING
6184 "md: array %s already initialised!\n",
6185 mdname(mddev));
6186 err = -EBUSY;
6187 goto abort_unlock;
6188 }
6189 err = set_array_info(mddev, &info);
6190 if (err) {
6191 printk(KERN_WARNING "md: couldn't set"
6192 " array info. %d\n", err);
6193 goto abort_unlock;
6194 }
6195 }
6196 goto done_unlock;
6197
6198 default:;
6199 }
6200
6201 /*
6202 * Commands querying/configuring an existing array:
6203 */
32a7627c 6204 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6205 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6206 if ((!mddev->raid_disks && !mddev->external)
6207 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6208 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6209 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6210 err = -ENODEV;
6211 goto abort_unlock;
6212 }
6213
6214 /*
6215 * Commands even a read-only array can execute:
6216 */
6217 switch (cmd)
6218 {
6219 case GET_ARRAY_INFO:
6220 err = get_array_info(mddev, argp);
6221 goto done_unlock;
6222
32a7627c 6223 case GET_BITMAP_FILE:
87162a28 6224 err = get_bitmap_file(mddev, argp);
32a7627c
N		 6225 goto done_unlock;
6226
1da177e4
LT		 6227 case GET_DISK_INFO:
6228 err = get_disk_info(mddev, argp);
6229 goto done_unlock;
6230
6231 case RESTART_ARRAY_RW:
6232 err = restart_array(mddev);
6233 goto done_unlock;
6234
6235 case STOP_ARRAY:
d710e138 6236 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 6237 goto done_unlock;
6238
6239 case STOP_ARRAY_RO:
a4bd82d0 6240 err = md_set_readonly(mddev, 1);
1da177e4
LT		 6241 goto done_unlock;
6242
e2218350
DW		 6243 case BLKROSET:
6244 if (get_user(ro, (int __user *)(arg))) {
6245 err = -EFAULT;
6246 goto done_unlock;
6247 }
6248 err = -EINVAL;
6249
6250 /* if the bdev is going readonly the value of mddev->ro
6251 * does not matter, no writes are coming
6252 */
6253 if (ro)
6254 goto done_unlock;
6255
6256 /* are we are already prepared for writes? */
6257 if (mddev->ro != 1)
6258 goto done_unlock;
6259
6260 /* transitioning to readauto need only happen for
6261 * arrays that call md_write_start
6262 */
6263 if (mddev->pers) {
6264 err = restart_array(mddev);
6265 if (err == 0) {
6266 mddev->ro = 2;
6267 set_disk_ro(mddev->gendisk, 0);
6268 }
6269 }
6270 goto done_unlock;
1da177e4
LT		 6271 }
6272
6273 /*
6274 * The remaining ioctls are changing the state of the
f91de92e
N		 6275 * superblock, so we do not allow them on read-only arrays.
6276 * However non-MD ioctls (e.g. get-size) will still come through
6277 * here and hit the 'default' below, so only disallow
6278 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6279 */
bb57fc64 6280 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6281 if (mddev->ro == 2) {
6282 mddev->ro = 0;
00bcb4ac 6283 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 6284 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6285 md_wakeup_thread(mddev->thread);
f91de92e
N		 6286 } else {
6287 err = -EROFS;
6288 goto abort_unlock;
6289 }
1da177e4
LT		 6290 }
6291
6292 switch (cmd)
6293 {
6294 case ADD_NEW_DISK:
6295 {
6296 mdu_disk_info_t info;
6297 if (copy_from_user(&info, argp, sizeof(info)))
6298 err = -EFAULT;
6299 else
6300 err = add_new_disk(mddev, &info);
6301 goto done_unlock;
6302 }
6303
6304 case HOT_REMOVE_DISK:
6305 err = hot_remove_disk(mddev, new_decode_dev(arg));
6306 goto done_unlock;
6307
6308 case HOT_ADD_DISK:
6309 err = hot_add_disk(mddev, new_decode_dev(arg));
6310 goto done_unlock;
6311
6312 case SET_DISK_FAULTY:
6313 err = set_disk_faulty(mddev, new_decode_dev(arg));
6314 goto done_unlock;
6315
6316 case RUN_ARRAY:
d710e138 6317 err = do_md_run(mddev);
1da177e4
LT		 6318 goto done_unlock;
6319
32a7627c
N		 6320 case SET_BITMAP_FILE:
6321 err = set_bitmap_file(mddev, (int)arg);
6322 goto done_unlock;
6323
1da177e4 6324 default:
1da177e4
LT		 6325 err = -EINVAL;
6326 goto abort_unlock;
6327 }
6328
6329done_unlock:
6330abort_unlock:
d3374825
N		 6331 if (mddev->hold_active == UNTIL_IOCTL &&
6332 err != -EINVAL)
6333 mddev->hold_active = 0;
1da177e4
LT		 6334 mddev_unlock(mddev);
6335
6336 return err;
6337done:
6338 if (err)
6339 MD_BUG();
6340abort:
6341 return err;
6342}
aa98aa31
AB		 6343#ifdef CONFIG_COMPAT
6344static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6345 unsigned int cmd, unsigned long arg)
6346{
6347 switch (cmd) {
6348 case HOT_REMOVE_DISK:
6349 case HOT_ADD_DISK:
6350 case SET_DISK_FAULTY:
6351 case SET_BITMAP_FILE:
6352 /* These take in integer arg, do not convert */
6353 break;
6354 default:
6355 arg = (unsigned long)compat_ptr(arg);
6356 break;
6357 }
6358
6359 return md_ioctl(bdev, mode, cmd, arg);
6360}
6361#endif /* CONFIG_COMPAT */
1da177e4 6362
a39907fa 6363static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6364{
6365 /*
6366 * Succeed if we can lock the mddev, which confirms that
6367 * it isn't being stopped right now.
6368 */
fd01b88c 6369 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6370 int err;
6371
d3374825
N		 6372 if (mddev->gendisk != bdev->bd_disk) {
6373 /* we are racing with mddev_put which is discarding this
6374 * bd_disk.
6375 */
6376 mddev_put(mddev);
6377 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6378 flush_workqueue(md_misc_wq);
d3374825
N		 6379 /* Then retry the open from the top */
6380 return -ERESTARTSYS;
6381 }
6382 BUG_ON(mddev != bdev->bd_disk->private_data);
6383
c8c00a69 6384 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6385 goto out;
6386
6387 err = 0;
f2ea68cf 6388 atomic_inc(&mddev->openers);
c8c00a69 6389 mutex_unlock(&mddev->open_mutex);
1da177e4 6390
f0b4f7e2 6391 check_disk_change(bdev);
1da177e4
LT		 6392 out:
6393 return err;
6394}
6395
a39907fa 6396static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6397{
fd01b88c 6398 struct mddev *mddev = disk->private_data;
1da177e4 6399
52e5f9d1 6400 BUG_ON(!mddev);
f2ea68cf 6401 atomic_dec(&mddev->openers);
1da177e4
LT		 6402 mddev_put(mddev);
6403
6404 return 0;
6405}
f0b4f7e2
N		 6406
6407static int md_media_changed(struct gendisk *disk)
6408{
fd01b88c 6409 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6410
6411 return mddev->changed;
6412}
6413
6414static int md_revalidate(struct gendisk *disk)
6415{
fd01b88c 6416 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6417
6418 mddev->changed = 0;
6419 return 0;
6420}
83d5cde4 6421static const struct block_device_operations md_fops =
1da177e4
LT		 6422{
6423 .owner = THIS_MODULE,
a39907fa
AV		 6424 .open = md_open,
6425 .release = md_release,
b492b852 6426 .ioctl = md_ioctl,
aa98aa31
AB		 6427#ifdef CONFIG_COMPAT
6428 .compat_ioctl = md_compat_ioctl,
6429#endif
a885c8c4 6430 .getgeo = md_getgeo,
f0b4f7e2
N		 6431 .media_changed = md_media_changed,
6432 .revalidate_disk= md_revalidate,
1da177e4
LT		 6433};
6434
75c96f85 6435static int md_thread(void * arg)
1da177e4 6436{
2b8bf345 6437 struct md_thread *thread = arg;
1da177e4 6438
1da177e4
LT		 6439 /*
6440 * md_thread is a 'system-thread', it's priority should be very
6441 * high. We avoid resource deadlocks individually in each
6442 * raid personality. (RAID5 does preallocation) We also use RR and
6443 * the very same RT priority as kswapd, thus we will never get
6444 * into a priority inversion deadlock.
6445 *
6446 * we definitely have to have equal or higher priority than
6447 * bdflush, otherwise bdflush will deadlock if there are too
6448 * many dirty RAID5 blocks.
6449 */
1da177e4 6450
6985c43f 6451 allow_signal(SIGKILL);
a6fb0934 6452 while (!kthread_should_stop()) {
1da177e4 6453
93588e22
N		 6454 /* We need to wait INTERRUPTIBLE so that
6455 * we don't add to the load-average.
6456 * That means we need to be sure no signals are
6457 * pending
6458 */
6459 if (signal_pending(current))
6460 flush_signals(current);
6461
6462 wait_event_interruptible_timeout
6463 (thread->wqueue,
6464 test_bit(THREAD_WAKEUP, &thread->flags)
6465 || kthread_should_stop(),
6466 thread->timeout);
1da177e4 6467
6c987910
N		 6468 clear_bit(THREAD_WAKEUP, &thread->flags);
6469 if (!kthread_should_stop())
589a594b 6470 thread->run(thread->mddev);
1da177e4 6471 }
a6fb0934 6472
1da177e4
LT		 6473 return 0;
6474}
6475
2b8bf345 6476void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6477{
6478 if (thread) {
36a4e1fe 6479 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6480 set_bit(THREAD_WAKEUP, &thread->flags);
6481 wake_up(&thread->wqueue);
6482 }
6483}
6484
2b8bf345 6485struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
1da177e4
LT		 6486 const char *name)
6487{
2b8bf345 6488 struct md_thread *thread;
1da177e4 6489
2b8bf345 6490 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6491 if (!thread)
6492 return NULL;
6493
1da177e4
LT		 6494 init_waitqueue_head(&thread->wqueue);
6495
1da177e4
LT		 6496 thread->run = run;
6497 thread->mddev = mddev;
32a7627c 6498 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6499 thread->tsk = kthread_run(md_thread, thread,
6500 "%s_%s",
6501 mdname(thread->mddev),
6502 name ?: mddev->pers->name);
a6fb0934 6503 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6504 kfree(thread);
6505 return NULL;
6506 }
1da177e4
LT		 6507 return thread;
6508}
6509
2b8bf345 6510void md_unregister_thread(struct md_thread **threadp)
1da177e4 6511{
2b8bf345 6512 struct md_thread *thread = *threadp;
e0cf8f04
N		 6513 if (!thread)
6514 return;
36a4e1fe 6515 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6516 /* Locking ensures that mddev_unlock does not wake_up a
6517 * non-existent thread
6518 */
6519 spin_lock(&pers_lock);
6520 *threadp = NULL;
6521 spin_unlock(&pers_lock);
a6fb0934
N		 6522
6523 kthread_stop(thread->tsk);
1da177e4
LT		 6524 kfree(thread);
6525}
6526
fd01b88c 6527void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6528{
6529 if (!mddev) {
6530 MD_BUG();
6531 return;
6532 }
6533
b2d444d7 6534 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6535 return;
6bfe0b49 6536
de393cde 6537 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6538 return;
6539 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6540 if (mddev->degraded)
6541 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6542 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6543 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6544 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6545 md_wakeup_thread(mddev->thread);
768a418d 6546 if (mddev->event_work.func)
e804ac78 6547 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6548 md_new_event_inintr(mddev);
1da177e4
LT		 6549}
6550
6551/* seq_file implementation /proc/mdstat */
6552
6553static void status_unused(struct seq_file *seq)
6554{
6555 int i = 0;
3cb03002 6556 struct md_rdev *rdev;
1da177e4
LT		 6557
6558 seq_printf(seq, "unused devices: ");
6559
159ec1fc 6560 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6561 char b[BDEVNAME_SIZE];
6562 i++;
6563 seq_printf(seq, "%s ",
6564 bdevname(rdev->bdev,b));
6565 }
6566 if (!i)
6567 seq_printf(seq, "<none>");
6568
6569 seq_printf(seq, "\n");
6570}
6571
6572
fd01b88c 6573static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6574{
dd71cf6b
N		 6575 sector_t max_sectors, resync, res;
6576 unsigned long dt, db;
6577 sector_t rt;
4588b42e
N		 6578 int scale;
6579 unsigned int per_milli;
1da177e4 6580
dd71cf6b 6581 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6582
6583 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6584 max_sectors = mddev->resync_max_sectors;
1da177e4 6585 else
dd71cf6b 6586 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6587
6588 /*
6589 * Should not happen.
6590 */
dd71cf6b 6591 if (!max_sectors) {
1da177e4
LT		 6592 MD_BUG();
6593 return;
6594 }
4588b42e 6595 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6596 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6597 * u32, as those are the requirements for sector_div.
6598 * Thus 'scale' must be at least 10
6599 */
6600 scale = 10;
6601 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6602 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6603 scale++;
6604 }
6605 res = (resync>>scale)*1000;
dd71cf6b 6606 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6607
6608 per_milli = res;
1da177e4 6609 {
4588b42e 6610 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6611 seq_printf(seq, "[");
6612 for (i = 0; i < x; i++)
6613 seq_printf(seq, "=");
6614 seq_printf(seq, ">");
6615 for (i = 0; i < y; i++)
6616 seq_printf(seq, ".");
6617 seq_printf(seq, "] ");
6618 }
4588b42e 6619 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6620 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6621 "reshape" :
61df9d91
N		 6622 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6623 "check" :
6624 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6625 "resync" : "recovery"))),
6626 per_milli/10, per_milli % 10,
dd71cf6b
N		 6627 (unsigned long long) resync/2,
6628 (unsigned long long) max_sectors/2);
1da177e4
LT		 6629
6630 /*
1da177e4
LT		 6631 * dt: time from mark until now
6632 * db: blocks written from mark until now
6633 * rt: remaining time
dd71cf6b
N		 6634 *
6635 * rt is a sector_t, so could be 32bit or 64bit.
6636 * So we divide before multiply in case it is 32bit and close
6637 * to the limit.
25985edc 6638 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6639 * near the end of resync when the number of remaining sectors
6640 * is close to 'db'.
6641 * We then divide rt by 32 after multiplying by db to compensate.
6642 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6643 */
6644 dt = ((jiffies - mddev->resync_mark) / HZ);
6645 if (!dt) dt++;
ff4e8d9a
N		 6646 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6647 - mddev->resync_mark_cnt;
1da177e4 6648
dd71cf6b
N		 6649 rt = max_sectors - resync; /* number of remaining sectors */
6650 sector_div(rt, db/32+1);
6651 rt *= dt;
6652 rt >>= 5;
6653
6654 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6655 ((unsigned long)rt % 60)/6);
1da177e4 6656
ff4e8d9a 6657 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6658}
6659
6660static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6661{
6662 struct list_head *tmp;
6663 loff_t l = *pos;
fd01b88c 6664 struct mddev *mddev;
1da177e4
LT		 6665
6666 if (l >= 0x10000)
6667 return NULL;
6668 if (!l--)
6669 /* header */
6670 return (void*)1;
6671
6672 spin_lock(&all_mddevs_lock);
6673 list_for_each(tmp,&all_mddevs)
6674 if (!l--) {
fd01b88c 6675 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6676 mddev_get(mddev);
6677 spin_unlock(&all_mddevs_lock);
6678 return mddev;
6679 }
6680 spin_unlock(&all_mddevs_lock);
6681 if (!l--)
6682 return (void*)2;/* tail */
6683 return NULL;
6684}
6685
6686static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6687{
6688 struct list_head *tmp;
fd01b88c 6689 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6690
6691 ++*pos;
6692 if (v == (void*)2)
6693 return NULL;
6694
6695 spin_lock(&all_mddevs_lock);
6696 if (v == (void*)1)
6697 tmp = all_mddevs.next;
6698 else
6699 tmp = mddev->all_mddevs.next;
6700 if (tmp != &all_mddevs)
fd01b88c 6701 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 6702 else {
6703 next_mddev = (void*)2;
6704 *pos = 0x10000;
6705 }
6706 spin_unlock(&all_mddevs_lock);
6707
6708 if (v != (void*)1)
6709 mddev_put(mddev);
6710 return next_mddev;
6711
6712}
6713
6714static void md_seq_stop(struct seq_file *seq, void *v)
6715{
fd01b88c 6716 struct mddev *mddev = v;
1da177e4
LT		 6717
6718 if (mddev && v != (void*)1 && v != (void*)2)
6719 mddev_put(mddev);
6720}
6721
6722static int md_seq_show(struct seq_file *seq, void *v)
6723{
fd01b88c 6724 struct mddev *mddev = v;
dd8ac336 6725 sector_t sectors;
3cb03002 6726 struct md_rdev *rdev;
32a7627c 6727 struct bitmap *bitmap;
1da177e4
LT		 6728
6729 if (v == (void*)1) {
84fc4b56 6730 struct md_personality *pers;
1da177e4
LT		 6731 seq_printf(seq, "Personalities : ");
6732 spin_lock(&pers_lock);
2604b703
N		 6733 list_for_each_entry(pers, &pers_list, list)
6734 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6735
6736 spin_unlock(&pers_lock);
6737 seq_printf(seq, "\n");
f1514638 6738 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6739 return 0;
6740 }
6741 if (v == (void*)2) {
6742 status_unused(seq);
6743 return 0;
6744 }
6745
5dc5cf7d 6746 if (mddev_lock(mddev) < 0)
1da177e4 6747 return -EINTR;
5dc5cf7d 6748
1da177e4
LT		 6749 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6750 seq_printf(seq, "%s : %sactive", mdname(mddev),
6751 mddev->pers ? "" : "in");
6752 if (mddev->pers) {
f91de92e 6753 if (mddev->ro==1)
1da177e4 6754 seq_printf(seq, " (read-only)");
f91de92e 6755 if (mddev->ro==2)
52720ae7 6756 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6757 seq_printf(seq, " %s", mddev->pers->name);
6758 }
6759
dd8ac336 6760 sectors = 0;
159ec1fc 6761 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 6762 char b[BDEVNAME_SIZE];
6763 seq_printf(seq, " %s[%d]",
6764 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6765 if (test_bit(WriteMostly, &rdev->flags))
6766 seq_printf(seq, "(W)");
b2d444d7 6767 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6768 seq_printf(seq, "(F)");
6769 continue;
2d78f8c4
N		 6770 }
6771 if (rdev->raid_disk < 0)
b325a32e 6772 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 6773 if (test_bit(Replacement, &rdev->flags))
6774 seq_printf(seq, "(R)");
dd8ac336 6775 sectors += rdev->sectors;
1da177e4
LT		 6776 }
6777
6778 if (!list_empty(&mddev->disks)) {
6779 if (mddev->pers)
6780 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6781 (unsigned long long)
6782 mddev->array_sectors / 2);
1da177e4
LT		 6783 else
6784 seq_printf(seq, "\n %llu blocks",
dd8ac336 6785 (unsigned long long)sectors / 2);
1da177e4 6786 }
1cd6bf19
N		 6787 if (mddev->persistent) {
6788 if (mddev->major_version != 0 ||
6789 mddev->minor_version != 90) {
6790 seq_printf(seq," super %d.%d",
6791 mddev->major_version,
6792 mddev->minor_version);
6793 }
e691063a
N		 6794 } else if (mddev->external)
6795 seq_printf(seq, " super external:%s",
6796 mddev->metadata_type);
6797 else
1cd6bf19 6798 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6799
6800 if (mddev->pers) {
d710e138 6801 mddev->pers->status(seq, mddev);
1da177e4 6802 seq_printf(seq, "\n ");
8e1b39d6
N		 6803 if (mddev->pers->sync_request) {
6804 if (mddev->curr_resync > 2) {
d710e138 6805 status_resync(seq, mddev);
8e1b39d6
N		 6806 seq_printf(seq, "\n ");
6807 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6808 seq_printf(seq, "\tresync=DELAYED\n ");
6809 else if (mddev->recovery_cp < MaxSector)
6810 seq_printf(seq, "\tresync=PENDING\n ");
6811 }
32a7627c
N		 6812 } else
6813 seq_printf(seq, "\n ");
6814
6815 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6816 unsigned long chunk_kb;
6817 unsigned long flags;
32a7627c 6818 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6819 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6820 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6821 "%lu%s chunk",
6822 bitmap->pages - bitmap->missing_pages,
6823 bitmap->pages,
6824 (bitmap->pages - bitmap->missing_pages)
6825 << (PAGE_SHIFT - 10),
42a04b50 6826 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6827 chunk_kb ? "KB" : "B");
78d742d8
N		 6828 if (bitmap->file) {
6829 seq_printf(seq, ", file: ");
c32c2f63 6830 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6831 }
78d742d8 6832
32a7627c
N		 6833 seq_printf(seq, "\n");
6834 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6835 }
6836
6837 seq_printf(seq, "\n");
6838 }
6839 mddev_unlock(mddev);
6840
6841 return 0;
6842}
6843
110518bc 6844static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6845 .start = md_seq_start,
6846 .next = md_seq_next,
6847 .stop = md_seq_stop,
6848 .show = md_seq_show,
6849};
6850
6851static int md_seq_open(struct inode *inode, struct file *file)
6852{
f1514638 6853 struct seq_file *seq;
1da177e4
LT		 6854 int error;
6855
6856 error = seq_open(file, &md_seq_ops);
d7603b7e 6857 if (error)
f1514638
KS		 6858 return error;
6859
6860 seq = file->private_data;
6861 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6862 return error;
6863}
6864
d7603b7e
N		 6865static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6866{
f1514638 6867 struct seq_file *seq = filp->private_data;
d7603b7e
N		 6868 int mask;
6869
6870 poll_wait(filp, &md_event_waiters, wait);
6871
6872 /* always allow read */
6873 mask = POLLIN | POLLRDNORM;
6874
f1514638 6875 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 6876 mask |= POLLERR | POLLPRI;
6877 return mask;
6878}
6879
fa027c2a 6880static const struct file_operations md_seq_fops = {
e24650c2 6881 .owner = THIS_MODULE,
1da177e4
LT		 6882 .open = md_seq_open,
6883 .read = seq_read,
6884 .llseek = seq_lseek,
c3f94b40 6885 .release = seq_release_private,
d7603b7e 6886 .poll = mdstat_poll,
1da177e4
LT		 6887};
6888
84fc4b56 6889int register_md_personality(struct md_personality *p)
1da177e4 6890{
1da177e4 6891 spin_lock(&pers_lock);
2604b703
N		 6892 list_add_tail(&p->list, &pers_list);
6893 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6894 spin_unlock(&pers_lock);
6895 return 0;
6896}
6897
84fc4b56 6898int unregister_md_personality(struct md_personality *p)
1da177e4 6899{
2604b703 6900 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6901 spin_lock(&pers_lock);
2604b703 6902 list_del_init(&p->list);
1da177e4
LT		 6903 spin_unlock(&pers_lock);
6904 return 0;
6905}
6906
fd01b88c 6907static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 6908{
3cb03002 6909 struct md_rdev * rdev;
1da177e4 6910 int idle;
eea1bf38 6911 int curr_events;
1da177e4
LT		 6912
6913 idle = 1;
4b80991c
N		 6914 rcu_read_lock();
6915 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6916 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6917 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6918 (int)part_stat_read(&disk->part0, sectors[1]) -
6919 atomic_read(&disk->sync_io);
713f6ab1
N		 6920 /* sync IO will cause sync_io to increase before the disk_stats
6921 * as sync_io is counted when a request starts, and
6922 * disk_stats is counted when it completes.
6923 * So resync activity will cause curr_events to be smaller than
6924 * when there was no such activity.
6925 * non-sync IO will cause disk_stat to increase without
6926 * increasing sync_io so curr_events will (eventually)
6927 * be larger than it was before. Once it becomes
6928 * substantially larger, the test below will cause
6929 * the array to appear non-idle, and resync will slow
6930 * down.
6931 * If there is a lot of outstanding resync activity when
6932 * we set last_event to curr_events, then all that activity
6933 * completing might cause the array to appear non-idle
6934 * and resync will be slowed down even though there might
6935 * not have been non-resync activity. This will only
6936 * happen once though. 'last_events' will soon reflect
6937 * the state where there is little or no outstanding
6938 * resync requests, and further resync activity will
6939 * always make curr_events less than last_events.
c0e48521 6940 *
1da177e4 6941 */
eea1bf38 6942 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6943 rdev->last_events = curr_events;
6944 idle = 0;
6945 }
6946 }
4b80991c 6947 rcu_read_unlock();
1da177e4
LT		 6948 return idle;
6949}
6950
fd01b88c 6951void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 6952{
6953 /* another "blocks" (512byte) blocks have been synced */
6954 atomic_sub(blocks, &mddev->recovery_active);
6955 wake_up(&mddev->recovery_wait);
6956 if (!ok) {
dfc70645 6957 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6958 md_wakeup_thread(mddev->thread);
6959 // stop recovery, signal do_sync ....
6960 }
6961}
6962
6963
06d91a5f
N		 6964/* md_write_start(mddev, bi)
6965 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6966 * in superblock) before writing, schedule a superblock update
6967 * and wait for it to complete.
06d91a5f 6968 */
fd01b88c 6969void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 6970{
0fd62b86 6971 int did_change = 0;
06d91a5f 6972 if (bio_data_dir(bi) != WRITE)
3d310eb7 6973 return;
06d91a5f 6974
f91de92e
N		 6975 BUG_ON(mddev->ro == 1);
6976 if (mddev->ro == 2) {
6977 /* need to switch to read/write */
6978 mddev->ro = 0;
6979 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6980 md_wakeup_thread(mddev->thread);
25156198 6981 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6982 did_change = 1;
f91de92e 6983 }
06d91a5f 6984 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6985 if (mddev->safemode == 1)
6986 mddev->safemode = 0;
06d91a5f 6987 if (mddev->in_sync) {
a9701a30 6988 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6989 if (mddev->in_sync) {
6990 mddev->in_sync = 0;
850b2b42 6991 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6992 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6993 md_wakeup_thread(mddev->thread);
0fd62b86 6994 did_change = 1;
3d310eb7 6995 }
a9701a30 6996 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6997 }
0fd62b86 6998 if (did_change)
00bcb4ac 6999 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7000 wait_event(mddev->sb_wait,
09a44cc1 7001 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 7002}
7003
fd01b88c 7004void md_write_end(struct mddev *mddev)
1da177e4
LT		 7005{
7006 if (atomic_dec_and_test(&mddev->writes_pending)) {
7007 if (mddev->safemode == 2)
7008 md_wakeup_thread(mddev->thread);
16f17b39 7009 else if (mddev->safemode_delay)
1da177e4
LT		 7010 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7011 }
7012}
7013
2a2275d6
N		 7014/* md_allow_write(mddev)
7015 * Calling this ensures that the array is marked 'active' so that writes
7016 * may proceed without blocking. It is important to call this before
7017 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7018 * Must be called with mddev_lock held.
b5470dc5
DW		 7019 *
7020 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7021 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7022 */
fd01b88c 7023int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7024{
7025 if (!mddev->pers)
b5470dc5 7026 return 0;
2a2275d6 7027 if (mddev->ro)
b5470dc5 7028 return 0;
1a0fd497 7029 if (!mddev->pers->sync_request)
b5470dc5 7030 return 0;
2a2275d6
N		 7031
7032 spin_lock_irq(&mddev->write_lock);
7033 if (mddev->in_sync) {
7034 mddev->in_sync = 0;
7035 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7036 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7037 if (mddev->safemode_delay &&
7038 mddev->safemode == 0)
7039 mddev->safemode = 1;
7040 spin_unlock_irq(&mddev->write_lock);
7041 md_update_sb(mddev, 0);
00bcb4ac 7042 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 7043 } else
7044 spin_unlock_irq(&mddev->write_lock);
b5470dc5 7045
070dc6dd 7046 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7047 return -EAGAIN;
7048 else
7049 return 0;
2a2275d6
N		 7050}
7051EXPORT_SYMBOL_GPL(md_allow_write);
7052
1da177e4
LT		 7053#define SYNC_MARKS 10
7054#define SYNC_MARK_STEP (3*HZ)
fd01b88c 7055void md_do_sync(struct mddev *mddev)
1da177e4 7056{
fd01b88c 7057 struct mddev *mddev2;
1da177e4
LT		 7058 unsigned int currspeed = 0,
7059 window;
57afd89f 7060 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 7061 unsigned long mark[SYNC_MARKS];
7062 sector_t mark_cnt[SYNC_MARKS];
7063 int last_mark,m;
7064 struct list_head *tmp;
7065 sector_t last_check;
57afd89f 7066 int skipped = 0;
3cb03002 7067 struct md_rdev *rdev;
61df9d91 7068 char *desc;
1da177e4
LT		 7069
7070 /* just incase thread restarts... */
7071 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7072 return;
5fd6c1dc
N		 7073 if (mddev->ro) /* never try to sync a read-only array */
7074 return;
1da177e4 7075
61df9d91
N		 7076 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7077 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7078 desc = "data-check";
7079 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7080 desc = "requested-resync";
7081 else
7082 desc = "resync";
7083 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7084 desc = "reshape";
7085 else
7086 desc = "recovery";
7087
1da177e4
LT		 7088 /* we overload curr_resync somewhat here.
7089 * 0 == not engaged in resync at all
7090 * 2 == checking that there is no conflict with another sync
7091 * 1 == like 2, but have yielded to allow conflicting resync to
7092 * commense
7093 * other == active in resync - this many blocks
7094 *
7095 * Before starting a resync we must have set curr_resync to
7096 * 2, and then checked that every "conflicting" array has curr_resync
7097 * less than ours. When we find one that is the same or higher
7098 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7099 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7100 * This will mean we have to start checking from the beginning again.
7101 *
7102 */
7103
7104 do {
7105 mddev->curr_resync = 2;
7106
7107 try_again:
404e4b43 7108 if (kthread_should_stop())
6985c43f 7109 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7110
7111 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7112 goto skip;
29ac4aa3 7113 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7114 if (mddev2 == mddev)
7115 continue;
90b08710
BS		 7116 if (!mddev->parallel_resync
7117 && mddev2->curr_resync
7118 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7119 DEFINE_WAIT(wq);
7120 if (mddev < mddev2 && mddev->curr_resync == 2) {
7121 /* arbitrarily yield */
7122 mddev->curr_resync = 1;
7123 wake_up(&resync_wait);
7124 }
7125 if (mddev > mddev2 && mddev->curr_resync == 1)
7126 /* no need to wait here, we can wait the next
7127 * time 'round when curr_resync == 2
7128 */
7129 continue;
9744197c
N		 7130 /* We need to wait 'interruptible' so as not to
7131 * contribute to the load average, and not to
7132 * be caught by 'softlockup'
7133 */
7134 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7135 if (!kthread_should_stop() &&
8712e553 7136 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7137 printk(KERN_INFO "md: delaying %s of %s"
7138 " until %s has finished (they"
1da177e4 7139 " share one or more physical units)\n",
61df9d91 7140 desc, mdname(mddev), mdname(mddev2));
1da177e4 7141 mddev_put(mddev2);
9744197c
N		 7142 if (signal_pending(current))
7143 flush_signals(current);
1da177e4
LT		 7144 schedule();
7145 finish_wait(&resync_wait, &wq);
7146 goto try_again;
7147 }
7148 finish_wait(&resync_wait, &wq);
7149 }
7150 }
7151 } while (mddev->curr_resync < 2);
7152
5fd6c1dc 7153 j = 0;
9d88883e 7154 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7155 /* resync follows the size requested by the personality,
57afd89f 7156 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7157 */
7158 max_sectors = mddev->resync_max_sectors;
9d88883e 7159 mddev->resync_mismatches = 0;
5fd6c1dc 7160 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7161 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7162 j = mddev->resync_min;
7163 else if (!mddev->bitmap)
5fd6c1dc 7164 j = mddev->recovery_cp;
5e96ee65 7165
ccfcc3c1 7166 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 7167 max_sectors = mddev->dev_sectors;
5fd6c1dc 7168 else {
1da177e4 7169 /* recovery follows the physical size of devices */
58c0fed4 7170 max_sectors = mddev->dev_sectors;
5fd6c1dc 7171 j = MaxSector;
4e59ca7d
DW		 7172 rcu_read_lock();
7173 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 7174 if (rdev->raid_disk >= 0 &&
7175 !test_bit(Faulty, &rdev->flags) &&
7176 !test_bit(In_sync, &rdev->flags) &&
7177 rdev->recovery_offset < j)
7178 j = rdev->recovery_offset;
4e59ca7d 7179 rcu_read_unlock();
5fd6c1dc 7180 }
1da177e4 7181
61df9d91
N		 7182 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7183 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7184 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7185 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7186 "(but not more than %d KB/sec) for %s.\n",
7187 speed_max(mddev), desc);
1da177e4 7188
eea1bf38 7189 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7190
57afd89f 7191 io_sectors = 0;
1da177e4
LT		 7192 for (m = 0; m < SYNC_MARKS; m++) {
7193 mark[m] = jiffies;
57afd89f 7194 mark_cnt[m] = io_sectors;
1da177e4
LT		 7195 }
7196 last_mark = 0;
7197 mddev->resync_mark = mark[last_mark];
7198 mddev->resync_mark_cnt = mark_cnt[last_mark];
7199
7200 /*
7201 * Tune reconstruction:
7202 */
7203 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7204 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7205 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7206
7207 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7208 last_check = 0;
7209
7210 if (j>2) {
7211 printk(KERN_INFO
61df9d91
N		 7212 "md: resuming %s of %s from checkpoint.\n",
7213 desc, mdname(mddev));
1da177e4
LT		 7214 mddev->curr_resync = j;
7215 }
75d3da43 7216 mddev->curr_resync_completed = j;
1da177e4
LT		 7217
7218 while (j < max_sectors) {
57afd89f 7219 sector_t sectors;
1da177e4 7220
57afd89f 7221 skipped = 0;
97e4f42d 7222
7a91ee1f
N		 7223 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7224 ((mddev->curr_resync > mddev->curr_resync_completed &&
7225 (mddev->curr_resync - mddev->curr_resync_completed)
7226 > (max_sectors >> 4)) ||
7227 (j - mddev->curr_resync_completed)*2
7228 >= mddev->resync_max - mddev->curr_resync_completed
7229 )) {
97e4f42d 7230 /* time to update curr_resync_completed */
97e4f42d
N		 7231 wait_event(mddev->recovery_wait,
7232 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7233 mddev->curr_resync_completed = j;
070dc6dd 7234 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7235 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7236 }
acb180b0 7237
e62e58a5
N		 7238 while (j >= mddev->resync_max && !kthread_should_stop()) {
7239 /* As this condition is controlled by user-space,
7240 * we can block indefinitely, so use '_interruptible'
7241 * to avoid triggering warnings.
7242 */
7243 flush_signals(current); /* just in case */
7244 wait_event_interruptible(mddev->recovery_wait,
7245 mddev->resync_max > j
7246 || kthread_should_stop());
7247 }
acb180b0
N		 7248
7249 if (kthread_should_stop())
7250 goto interrupted;
7251
57afd89f 7252 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7253 currspeed < speed_min(mddev));
57afd89f 7254 if (sectors == 0) {
dfc70645 7255 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7256 goto out;
7257 }
57afd89f
N		 7258
7259 if (!skipped) { /* actual IO requested */
7260 io_sectors += sectors;
7261 atomic_add(sectors, &mddev->recovery_active);
7262 }
7263
e875ecea
N		 7264 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7265 break;
7266
1da177e4
LT		 7267 j += sectors;
7268 if (j>1) mddev->curr_resync = j;
ff4e8d9a 7269 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7270 if (last_check == 0)
e875ecea 7271 /* this is the earliest that rebuild will be
d7603b7e
N		 7272 * visible in /proc/mdstat
7273 */
7274 md_new_event(mddev);
57afd89f
N		 7275
7276 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7277 continue;
7278
57afd89f 7279 last_check = io_sectors;
1da177e4
LT		 7280 repeat:
7281 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7282 /* step marks */
7283 int next = (last_mark+1) % SYNC_MARKS;
7284
7285 mddev->resync_mark = mark[next];
7286 mddev->resync_mark_cnt = mark_cnt[next];
7287 mark[next] = jiffies;
57afd89f 7288 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7289 last_mark = next;
7290 }
7291
7292
c6207277
N		 7293 if (kthread_should_stop())
7294 goto interrupted;
7295
1da177e4
LT		 7296
7297 /*
7298 * this loop exits only if either when we are slower than
7299 * the 'hard' speed limit, or the system was IO-idle for
7300 * a jiffy.
7301 * the system might be non-idle CPU-wise, but we only care
7302 * about not overloading the IO subsystem. (things like an
7303 * e2fsck being done on the RAID array should execute fast)
7304 */
1da177e4
LT		 7305 cond_resched();
7306
57afd89f
N		 7307 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7308 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7309
88202a0c
N		 7310 if (currspeed > speed_min(mddev)) {
7311 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7312 !is_mddev_idle(mddev, 0)) {
c0e48521 7313 msleep(500);
1da177e4
LT		 7314 goto repeat;
7315 }
7316 }
7317 }
61df9d91 7318 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7319 /*
7320 * this also signals 'finished resyncing' to md_stop
7321 */
7322 out:
1da177e4
LT		 7323 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7324
7325 /* tell personality that we are finished */
57afd89f 7326 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7327
dfc70645 7328 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7329 mddev->curr_resync > 2) {
7330 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7331 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7332 if (mddev->curr_resync >= mddev->recovery_cp) {
7333 printk(KERN_INFO
61df9d91
N		 7334 "md: checkpointing %s of %s.\n",
7335 desc, mdname(mddev));
5fd6c1dc
N		 7336 mddev->recovery_cp = mddev->curr_resync;
7337 }
7338 } else
7339 mddev->recovery_cp = MaxSector;
7340 } else {
7341 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7342 mddev->curr_resync = MaxSector;
4e59ca7d
DW		 7343 rcu_read_lock();
7344 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc 7345 if (rdev->raid_disk >= 0 &&
70fffd0b 7346 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7347 !test_bit(Faulty, &rdev->flags) &&
7348 !test_bit(In_sync, &rdev->flags) &&
7349 rdev->recovery_offset < mddev->curr_resync)
7350 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7351 rcu_read_unlock();
5fd6c1dc 7352 }
1da177e4 7353 }
17571284 7354 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7355
1da177e4 7356 skip:
c07b70ad
N		 7357 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7358 /* We completed so min/max setting can be forgotten if used. */
7359 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7360 mddev->resync_min = 0;
7361 mddev->resync_max = MaxSector;
7362 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7363 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7364 mddev->curr_resync = 0;
7365 wake_up(&resync_wait);
7366 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7367 md_wakeup_thread(mddev->thread);
c6207277
N		 7368 return;
7369
7370 interrupted:
7371 /*
7372 * got a signal, exit.
7373 */
7374 printk(KERN_INFO
7375 "md: md_do_sync() got signal ... exiting\n");
7376 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7377 goto out;
7378
1da177e4 7379}
29269553 7380EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7381
fd01b88c 7382static int remove_and_add_spares(struct mddev *mddev)
b4c4c7b8 7383{
3cb03002 7384 struct md_rdev *rdev;
b4c4c7b8
N		 7385 int spares = 0;
7386
97e4f42d
N		 7387 mddev->curr_resync_completed = 0;
7388
159ec1fc 7389 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 7390 if (rdev->raid_disk >= 0 &&
6bfe0b49 7391 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7392 (test_bit(Faulty, &rdev->flags) ||
7393 ! test_bit(In_sync, &rdev->flags)) &&
7394 atomic_read(&rdev->nr_pending)==0) {
7395 if (mddev->pers->hot_remove_disk(
b8321b68 7396 mddev, rdev) == 0) {
36fad858 7397 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8
N		 7398 rdev->raid_disk = -1;
7399 }
7400 }
7401
5389042f 7402 if (mddev->degraded) {
159ec1fc 7403 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7404 if (rdev->raid_disk >= 0 &&
e5427135 7405 !test_bit(In_sync, &rdev->flags) &&
de393cde 7406 !test_bit(Faulty, &rdev->flags))
dfc70645 7407 spares++;
b4c4c7b8
N		 7408 if (rdev->raid_disk < 0
7409 && !test_bit(Faulty, &rdev->flags)) {
7410 rdev->recovery_offset = 0;
199050ea
NB		 7411 if (mddev->pers->
7412 hot_add_disk(mddev, rdev) == 0) {
36fad858 7413 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 7414 /* failure here is OK */;
b4c4c7b8
N		 7415 spares++;
7416 md_new_event(mddev);
93be75ff 7417 set_bit(MD_CHANGE_DEVS, &mddev->flags);
60fc1370 7418 }
b4c4c7b8 7419 }
dfc70645 7420 }
b4c4c7b8
N		 7421 }
7422 return spares;
7423}
7ebc0be7 7424
fd01b88c 7425static void reap_sync_thread(struct mddev *mddev)
7ebc0be7 7426{
3cb03002 7427 struct md_rdev *rdev;
7ebc0be7
N		 7428
7429 /* resync has finished, collect result */
01f96c0a 7430 md_unregister_thread(&mddev->sync_thread);
7ebc0be7
N		 7431 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7432 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7433 /* success...*/
7434 /* activate any spares */
7435 if (mddev->pers->spare_active(mddev))
7436 sysfs_notify(&mddev->kobj, NULL,
7437 "degraded");
7438 }
7439 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7440 mddev->pers->finish_reshape)
7441 mddev->pers->finish_reshape(mddev);
7ebc0be7 7442
d70ed2e4
AW		 7443 /* If array is no-longer degraded, then any saved_raid_disk
7444 * information must be scrapped. Also if any device is now
7445 * In_sync we must scrape the saved_raid_disk for that device
7446 * do the superblock for an incrementally recovered device
7447 * written out.
7ebc0be7 7448 */
d70ed2e4
AW		 7449 list_for_each_entry(rdev, &mddev->disks, same_set)
7450 if (!mddev->degraded ||
7451 test_bit(In_sync, &rdev->flags))
7ebc0be7
N		 7452 rdev->saved_raid_disk = -1;
7453
d70ed2e4 7454 md_update_sb(mddev, 1);
7ebc0be7
N		 7455 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7456 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7457 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7458 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7459 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7460 /* flag recovery needed just to double check */
7461 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7462 sysfs_notify_dirent_safe(mddev->sysfs_action);
7463 md_new_event(mddev);
768e587e
JB		 7464 if (mddev->event_work.func)
7465 queue_work(md_misc_wq, &mddev->event_work);
7ebc0be7
N		 7466}
7467
1da177e4
LT		 7468/*
7469 * This routine is regularly called by all per-raid-array threads to
7470 * deal with generic issues like resync and super-block update.
7471 * Raid personalities that don't have a thread (linear/raid0) do not
7472 * need this as they never do any recovery or update the superblock.
7473 *
7474 * It does not do any resync itself, but rather "forks" off other threads
7475 * to do that as needed.
7476 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7477 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7478 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7479 * and wakeups up this thread which will reap the thread and finish up.
7480 * This thread also removes any faulty devices (with nr_pending == 0).
7481 *
7482 * The overall approach is:
7483 * 1/ if the superblock needs updating, update it.
7484 * 2/ If a recovery thread is running, don't do anything else.
7485 * 3/ If recovery has finished, clean up, possibly marking spares active.
7486 * 4/ If there are any faulty devices, remove them.
7487 * 5/ If array is degraded, try to add spares devices
7488 * 6/ If array has spares or is not in-sync, start a resync thread.
7489 */
fd01b88c 7490void md_check_recovery(struct mddev *mddev)
1da177e4 7491{
68866e42
JB		 7492 if (mddev->suspended)
7493 return;
7494
5f40402d 7495 if (mddev->bitmap)
aa5cbd10 7496 bitmap_daemon_work(mddev);
1da177e4 7497
fca4d848 7498 if (signal_pending(current)) {
31a59e34 7499 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7500 printk(KERN_INFO "md: %s in immediate safe mode\n",
7501 mdname(mddev));
7502 mddev->safemode = 2;
7503 }
7504 flush_signals(current);
7505 }
7506
c89a8eee
N		 7507 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7508 return;
1da177e4 7509 if ( ! (
126925c0 7510 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7511 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7512 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7513 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7514 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7515 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7516 ))
7517 return;
fca4d848 7518
df5b89b3 7519 if (mddev_trylock(mddev)) {
b4c4c7b8 7520 int spares = 0;
fca4d848 7521
c89a8eee
N		 7522 if (mddev->ro) {
7523 /* Only thing we do on a ro array is remove
7524 * failed devices.
7525 */
3cb03002 7526 struct md_rdev *rdev;
a8c42c7f
N		 7527 list_for_each_entry(rdev, &mddev->disks, same_set)
7528 if (rdev->raid_disk >= 0 &&
7529 !test_bit(Blocked, &rdev->flags) &&
7530 test_bit(Faulty, &rdev->flags) &&
7531 atomic_read(&rdev->nr_pending)==0) {
7532 if (mddev->pers->hot_remove_disk(
b8321b68 7533 mddev, rdev) == 0) {
36fad858 7534 sysfs_unlink_rdev(mddev, rdev);
a8c42c7f
N		 7535 rdev->raid_disk = -1;
7536 }
7537 }
c89a8eee
N		 7538 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7539 goto unlock;
7540 }
7541
31a59e34 7542 if (!mddev->external) {
0fd62b86 7543 int did_change = 0;
31a59e34
N		 7544 spin_lock_irq(&mddev->write_lock);
7545 if (mddev->safemode &&
7546 !atomic_read(&mddev->writes_pending) &&
7547 !mddev->in_sync &&
7548 mddev->recovery_cp == MaxSector) {
7549 mddev->in_sync = 1;
0fd62b86 7550 did_change = 1;
070dc6dd 7551 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7552 }
7553 if (mddev->safemode == 1)
7554 mddev->safemode = 0;
7555 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7556 if (did_change)
00bcb4ac 7557 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7558 }
fca4d848 7559
850b2b42
N		 7560 if (mddev->flags)
7561 md_update_sb(mddev, 0);
06d91a5f 7562
1da177e4
LT		 7563 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7564 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7565 /* resync/recovery still happening */
7566 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7567 goto unlock;
7568 }
7569 if (mddev->sync_thread) {
7ebc0be7 7570 reap_sync_thread(mddev);
1da177e4
LT		 7571 goto unlock;
7572 }
72a23c21
NB		 7573 /* Set RUNNING before clearing NEEDED to avoid
7574 * any transients in the value of "sync_action".
7575 */
7576 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7577 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7578 /* Clear some bits that don't mean anything, but
7579 * might be left set
7580 */
24dd469d
N		 7581 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7582 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7583
5fd6c1dc
N		 7584 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7585 goto unlock;
1da177e4
LT		 7586 /* no recovery is running.
7587 * remove any failed drives, then
7588 * add spares if possible.
7589 * Spare are also removed and re-added, to allow
7590 * the personality to fail the re-add.
7591 */
1da177e4 7592
b4c4c7b8 7593 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7594 if (mddev->pers->check_reshape == NULL ||
7595 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7596 /* Cannot proceed */
7597 goto unlock;
7598 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7599 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7600 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7601 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7602 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7603 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7604 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7605 } else if (mddev->recovery_cp < MaxSector) {
7606 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7607 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7608 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7609 /* nothing to be done ... */
1da177e4 7610 goto unlock;
24dd469d 7611
1da177e4 7612 if (mddev->pers->sync_request) {
a654b9d8
N		 7613 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7614 /* We are adding a device or devices to an array
7615 * which has the bitmap stored on all devices.
7616 * So make sure all bitmap pages get written
7617 */
7618 bitmap_write_all(mddev->bitmap);
7619 }
1da177e4
LT		 7620 mddev->sync_thread = md_register_thread(md_do_sync,
7621 mddev,
0da3c619 7622 "resync");
1da177e4
LT		 7623 if (!mddev->sync_thread) {
7624 printk(KERN_ERR "%s: could not start resync"
7625 " thread...\n",
7626 mdname(mddev));
7627 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7628 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7629 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7630 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7631 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7632 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7633 } else
1da177e4 7634 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7635 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7636 md_new_event(mddev);
1da177e4
LT		 7637 }
7638 unlock:
72a23c21
NB		 7639 if (!mddev->sync_thread) {
7640 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7641 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7642 &mddev->recovery))
0c3573f1 7643 if (mddev->sysfs_action)
00bcb4ac 7644 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7645 }
1da177e4
LT		 7646 mddev_unlock(mddev);
7647 }
7648}
7649
fd01b88c 7650void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7651{
00bcb4ac 7652 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7653 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7654 !test_bit(Blocked, &rdev->flags) &&
7655 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7656 msecs_to_jiffies(5000));
7657 rdev_dec_pending(rdev, mddev);
7658}
7659EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7660
2230dfe4
N		 7661
7662/* Bad block management.
7663 * We can record which blocks on each device are 'bad' and so just
7664 * fail those blocks, or that stripe, rather than the whole device.
7665 * Entries in the bad-block table are 64bits wide. This comprises:
7666 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7667 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7668 * A 'shift' can be set so that larger blocks are tracked and
7669 * consequently larger devices can be covered.
7670 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7671 *
7672 * Locking of the bad-block table uses a seqlock so md_is_badblock
7673 * might need to retry if it is very unlucky.
7674 * We will sometimes want to check for bad blocks in a bi_end_io function,
7675 * so we use the write_seqlock_irq variant.
7676 *
7677 * When looking for a bad block we specify a range and want to
7678 * know if any block in the range is bad. So we binary-search
7679 * to the last range that starts at-or-before the given endpoint,
7680 * (or "before the sector after the target range")
7681 * then see if it ends after the given start.
7682 * We return
7683 * 0 if there are no known bad blocks in the range
7684 * 1 if there are known bad block which are all acknowledged
7685 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7686 * plus the start/length of the first bad section we overlap.
7687 */
7688int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7689 sector_t *first_bad, int *bad_sectors)
7690{
7691 int hi;
7692 int lo = 0;
7693 u64 *p = bb->page;
7694 int rv = 0;
7695 sector_t target = s + sectors;
7696 unsigned seq;
7697
7698 if (bb->shift > 0) {
7699 /* round the start down, and the end up */
7700 s >>= bb->shift;
7701 target += (1<<bb->shift) - 1;
7702 target >>= bb->shift;
7703 sectors = target - s;
7704 }
7705 /* 'target' is now the first block after the bad range */
7706
7707retry:
7708 seq = read_seqbegin(&bb->lock);
7709
7710 hi = bb->count;
7711
7712 /* Binary search between lo and hi for 'target'
7713 * i.e. for the last range that starts before 'target'
7714 */
7715 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7716 * are known not to be the last range before target.
7717 * VARIANT: hi-lo is the number of possible
7718 * ranges, and decreases until it reaches 1
7719 */
7720 while (hi - lo > 1) {
7721 int mid = (lo + hi) / 2;
7722 sector_t a = BB_OFFSET(p[mid]);
7723 if (a < target)
7724 /* This could still be the one, earlier ranges
7725 * could not. */
7726 lo = mid;
7727 else
7728 /* This and later ranges are definitely out. */
7729 hi = mid;
7730 }
7731 /* 'lo' might be the last that started before target, but 'hi' isn't */
7732 if (hi > lo) {
7733 /* need to check all range that end after 's' to see if
7734 * any are unacknowledged.
7735 */
7736 while (lo >= 0 &&
7737 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7738 if (BB_OFFSET(p[lo]) < target) {
7739 /* starts before the end, and finishes after
7740 * the start, so they must overlap
7741 */
7742 if (rv != -1 && BB_ACK(p[lo]))
7743 rv = 1;
7744 else
7745 rv = -1;
7746 *first_bad = BB_OFFSET(p[lo]);
7747 *bad_sectors = BB_LEN(p[lo]);
7748 }
7749 lo--;
7750 }
7751 }
7752
7753 if (read_seqretry(&bb->lock, seq))
7754 goto retry;
7755
7756 return rv;
7757}
7758EXPORT_SYMBOL_GPL(md_is_badblock);
7759
7760/*
7761 * Add a range of bad blocks to the table.
7762 * This might extend the table, or might contract it
7763 * if two adjacent ranges can be merged.
7764 * We binary-search to find the 'insertion' point, then
7765 * decide how best to handle it.
7766 */
7767static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7768 int acknowledged)
7769{
7770 u64 *p;
7771 int lo, hi;
7772 int rv = 1;
7773
7774 if (bb->shift < 0)
7775 /* badblocks are disabled */
7776 return 0;
7777
7778 if (bb->shift) {
7779 /* round the start down, and the end up */
7780 sector_t next = s + sectors;
7781 s >>= bb->shift;
7782 next += (1<<bb->shift) - 1;
7783 next >>= bb->shift;
7784 sectors = next - s;
7785 }
7786
7787 write_seqlock_irq(&bb->lock);
7788
7789 p = bb->page;
7790 lo = 0;
7791 hi = bb->count;
7792 /* Find the last range that starts at-or-before 's' */
7793 while (hi - lo > 1) {
7794 int mid = (lo + hi) / 2;
7795 sector_t a = BB_OFFSET(p[mid]);
7796 if (a <= s)
7797 lo = mid;
7798 else
7799 hi = mid;
7800 }
7801 if (hi > lo && BB_OFFSET(p[lo]) > s)
7802 hi = lo;
7803
7804 if (hi > lo) {
7805 /* we found a range that might merge with the start
7806 * of our new range
7807 */
7808 sector_t a = BB_OFFSET(p[lo]);
7809 sector_t e = a + BB_LEN(p[lo]);
7810 int ack = BB_ACK(p[lo]);
7811 if (e >= s) {
7812 /* Yes, we can merge with a previous range */
7813 if (s == a && s + sectors >= e)
7814 /* new range covers old */
7815 ack = acknowledged;
7816 else
7817 ack = ack && acknowledged;
7818
7819 if (e < s + sectors)
7820 e = s + sectors;
7821 if (e - a <= BB_MAX_LEN) {
7822 p[lo] = BB_MAKE(a, e-a, ack);
7823 s = e;
7824 } else {
7825 /* does not all fit in one range,
7826 * make p[lo] maximal
7827 */
7828 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7829 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7830 s = a + BB_MAX_LEN;
7831 }
7832 sectors = e - s;
7833 }
7834 }
7835 if (sectors && hi < bb->count) {
7836 /* 'hi' points to the first range that starts after 's'.
7837 * Maybe we can merge with the start of that range */
7838 sector_t a = BB_OFFSET(p[hi]);
7839 sector_t e = a + BB_LEN(p[hi]);
7840 int ack = BB_ACK(p[hi]);
7841 if (a <= s + sectors) {
7842 /* merging is possible */
7843 if (e <= s + sectors) {
7844 /* full overlap */
7845 e = s + sectors;
7846 ack = acknowledged;
7847 } else
7848 ack = ack && acknowledged;
7849
7850 a = s;
7851 if (e - a <= BB_MAX_LEN) {
7852 p[hi] = BB_MAKE(a, e-a, ack);
7853 s = e;
7854 } else {
7855 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7856 s = a + BB_MAX_LEN;
7857 }
7858 sectors = e - s;
7859 lo = hi;
7860 hi++;
7861 }
7862 }
7863 if (sectors == 0 && hi < bb->count) {
7864 /* we might be able to combine lo and hi */
7865 /* Note: 's' is at the end of 'lo' */
7866 sector_t a = BB_OFFSET(p[hi]);
7867 int lolen = BB_LEN(p[lo]);
7868 int hilen = BB_LEN(p[hi]);
7869 int newlen = lolen + hilen - (s - a);
7870 if (s >= a && newlen < BB_MAX_LEN) {
7871 /* yes, we can combine them */
7872 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7873 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7874 memmove(p + hi, p + hi + 1,
7875 (bb->count - hi - 1) * 8);
7876 bb->count--;
7877 }
7878 }
7879 while (sectors) {
7880 /* didn't merge (it all).
7881 * Need to add a range just before 'hi' */
7882 if (bb->count >= MD_MAX_BADBLOCKS) {
7883 /* No room for more */
7884 rv = 0;
7885 break;
7886 } else {
7887 int this_sectors = sectors;
7888 memmove(p + hi + 1, p + hi,
7889 (bb->count - hi) * 8);
7890 bb->count++;
7891
7892 if (this_sectors > BB_MAX_LEN)
7893 this_sectors = BB_MAX_LEN;
7894 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7895 sectors -= this_sectors;
7896 s += this_sectors;
7897 }
7898 }
7899
7900 bb->changed = 1;
de393cde
N		 7901 if (!acknowledged)
7902 bb->unacked_exist = 1;
2230dfe4
N		 7903 write_sequnlock_irq(&bb->lock);
7904
7905 return rv;
7906}
7907
3cb03002 7908int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
2230dfe4
N		 7909 int acknowledged)
7910{
7911 int rv = md_set_badblocks(&rdev->badblocks,
7912 s + rdev->data_offset, sectors, acknowledged);
7913 if (rv) {
7914 /* Make sure they get written out promptly */
8bd2f0a0 7915 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 7916 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7917 md_wakeup_thread(rdev->mddev->thread);
7918 }
7919 return rv;
7920}
7921EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7922
7923/*
7924 * Remove a range of bad blocks from the table.
7925 * This may involve extending the table if we spilt a region,
7926 * but it must not fail. So if the table becomes full, we just
7927 * drop the remove request.
7928 */
7929static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7930{
7931 u64 *p;
7932 int lo, hi;
7933 sector_t target = s + sectors;
7934 int rv = 0;
7935
7936 if (bb->shift > 0) {
7937 /* When clearing we round the start up and the end down.
7938 * This should not matter as the shift should align with
7939 * the block size and no rounding should ever be needed.
7940 * However it is better the think a block is bad when it
7941 * isn't than to think a block is not bad when it is.
7942 */
7943 s += (1<<bb->shift) - 1;
7944 s >>= bb->shift;
7945 target >>= bb->shift;
7946 sectors = target - s;
7947 }
7948
7949 write_seqlock_irq(&bb->lock);
7950
7951 p = bb->page;
7952 lo = 0;
7953 hi = bb->count;
7954 /* Find the last range that starts before 'target' */
7955 while (hi - lo > 1) {
7956 int mid = (lo + hi) / 2;
7957 sector_t a = BB_OFFSET(p[mid]);
7958 if (a < target)
7959 lo = mid;
7960 else
7961 hi = mid;
7962 }
7963 if (hi > lo) {
7964 /* p[lo] is the last range that could overlap the
7965 * current range. Earlier ranges could also overlap,
7966 * but only this one can overlap the end of the range.
7967 */
7968 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7969 /* Partial overlap, leave the tail of this range */
7970 int ack = BB_ACK(p[lo]);
7971 sector_t a = BB_OFFSET(p[lo]);
7972 sector_t end = a + BB_LEN(p[lo]);
7973
7974 if (a < s) {
7975 /* we need to split this range */
7976 if (bb->count >= MD_MAX_BADBLOCKS) {
7977 rv = 0;
7978 goto out;
7979 }
7980 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7981 bb->count++;
7982 p[lo] = BB_MAKE(a, s-a, ack);
7983 lo++;
7984 }
7985 p[lo] = BB_MAKE(target, end - target, ack);
7986 /* there is no longer an overlap */
7987 hi = lo;
7988 lo--;
7989 }
7990 while (lo >= 0 &&
7991 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7992 /* This range does overlap */
7993 if (BB_OFFSET(p[lo]) < s) {
7994 /* Keep the early parts of this range. */
7995 int ack = BB_ACK(p[lo]);
7996 sector_t start = BB_OFFSET(p[lo]);
7997 p[lo] = BB_MAKE(start, s - start, ack);
7998 /* now low doesn't overlap, so.. */
7999 break;
8000 }
8001 lo--;
8002 }
8003 /* 'lo' is strictly before, 'hi' is strictly after,
8004 * anything between needs to be discarded
8005 */
8006 if (hi - lo > 1) {
8007 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8008 bb->count -= (hi - lo - 1);
8009 }
8010 }
8011
8012 bb->changed = 1;
8013out:
8014 write_sequnlock_irq(&bb->lock);
8015 return rv;
8016}
8017
3cb03002 8018int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
2230dfe4
N		 8019{
8020 return md_clear_badblocks(&rdev->badblocks,
8021 s + rdev->data_offset,
8022 sectors);
8023}
8024EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8025
8026/*
8027 * Acknowledge all bad blocks in a list.
8028 * This only succeeds if ->changed is clear. It is used by
8029 * in-kernel metadata updates
8030 */
8031void md_ack_all_badblocks(struct badblocks *bb)
8032{
8033 if (bb->page == NULL || bb->changed)
8034 /* no point even trying */
8035 return;
8036 write_seqlock_irq(&bb->lock);
8037
8038 if (bb->changed == 0) {
8039 u64 *p = bb->page;
8040 int i;
8041 for (i = 0; i < bb->count ; i++) {
8042 if (!BB_ACK(p[i])) {
8043 sector_t start = BB_OFFSET(p[i]);
8044 int len = BB_LEN(p[i]);
8045 p[i] = BB_MAKE(start, len, 1);
8046 }
8047 }
de393cde 8048 bb->unacked_exist = 0;
2230dfe4
N		 8049 }
8050 write_sequnlock_irq(&bb->lock);
8051}
8052EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8053
16c791a5
N		 8054/* sysfs access to bad-blocks list.
8055 * We present two files.
8056 * 'bad-blocks' lists sector numbers and lengths of ranges that
8057 * are recorded as bad. The list is truncated to fit within
8058 * the one-page limit of sysfs.
8059 * Writing "sector length" to this file adds an acknowledged
8060 * bad block list.
8061 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8062 * been acknowledged. Writing to this file adds bad blocks
8063 * without acknowledging them. This is largely for testing.
8064 */
8065
8066static ssize_t
8067badblocks_show(struct badblocks *bb, char *page, int unack)
8068{
8069 size_t len;
8070 int i;
8071 u64 *p = bb->page;
8072 unsigned seq;
8073
8074 if (bb->shift < 0)
8075 return 0;
8076
8077retry:
8078 seq = read_seqbegin(&bb->lock);
8079
8080 len = 0;
8081 i = 0;
8082
8083 while (len < PAGE_SIZE && i < bb->count) {
8084 sector_t s = BB_OFFSET(p[i]);
8085 unsigned int length = BB_LEN(p[i]);
8086 int ack = BB_ACK(p[i]);
8087 i++;
8088
8089 if (unack && ack)
8090 continue;
8091
8092 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8093 (unsigned long long)s << bb->shift,
8094 length << bb->shift);
8095 }
de393cde
N		 8096 if (unack && len == 0)
8097 bb->unacked_exist = 0;
16c791a5
N		 8098
8099 if (read_seqretry(&bb->lock, seq))
8100 goto retry;
8101
8102 return len;
8103}
8104
8105#define DO_DEBUG 1
8106
8107static ssize_t
8108badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8109{
8110 unsigned long long sector;
8111 int length;
8112 char newline;
8113#ifdef DO_DEBUG
8114 /* Allow clearing via sysfs *only* for testing/debugging.
8115 * Normally only a successful write may clear a badblock
8116 */
8117 int clear = 0;
8118 if (page[0] == '-') {
8119 clear = 1;
8120 page++;
8121 }
8122#endif /* DO_DEBUG */
8123
8124 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8125 case 3:
8126 if (newline != '\n')
8127 return -EINVAL;
8128 case 2:
8129 if (length <= 0)
8130 return -EINVAL;
8131 break;
8132 default:
8133 return -EINVAL;
8134 }
8135
8136#ifdef DO_DEBUG
8137 if (clear) {
8138 md_clear_badblocks(bb, sector, length);
8139 return len;
8140 }
8141#endif /* DO_DEBUG */
8142 if (md_set_badblocks(bb, sector, length, !unack))
8143 return len;
8144 else
8145 return -ENOSPC;
8146}
8147
75c96f85
AB		 8148static int md_notify_reboot(struct notifier_block *this,
8149 unsigned long code, void *x)
1da177e4
LT		 8150{
8151 struct list_head *tmp;
fd01b88c 8152 struct mddev *mddev;
2dba6a91 8153 int need_delay = 0;
1da177e4
LT		 8154
8155 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
8156
8157 printk(KERN_INFO "md: stopping all md devices.\n");
8158
2dba6a91 8159 for_each_mddev(mddev, tmp) {
c71d4887 8160 if (mddev_trylock(mddev)) {
2b25000b
N		 8161 /* Force a switch to readonly even array
8162 * appears to still be in use. Hence
8163 * the '100'.
8164 */
a4bd82d0 8165 md_set_readonly(mddev, 100);
c71d4887
NB		 8166 mddev_unlock(mddev);
8167 }
2dba6a91
DB		 8168 need_delay = 1;
8169 }
1da177e4
LT		 8170 /*
8171 * certain more exotic SCSI devices are known to be
8172 * volatile wrt too early system reboots. While the
8173 * right place to handle this issue is the given
8174 * driver, we do want to have a safe RAID driver ...
8175 */
2dba6a91
DB		 8176 if (need_delay)
8177 mdelay(1000*1);
1da177e4
LT		 8178 }
8179 return NOTIFY_DONE;
8180}
8181
75c96f85 8182static struct notifier_block md_notifier = {
1da177e4
LT		 8183 .notifier_call = md_notify_reboot,
8184 .next = NULL,
8185 .priority = INT_MAX, /* before any real devices */
8186};
8187
8188static void md_geninit(void)
8189{
36a4e1fe 8190 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8191
c7705f34 8192 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8193}
8194
75c96f85 8195static int __init md_init(void)
1da177e4 8196{
e804ac78
TH		 8197 int ret = -ENOMEM;
8198
ada609ee 8199 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8200 if (!md_wq)
8201 goto err_wq;
8202
8203 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8204 if (!md_misc_wq)
8205 goto err_misc_wq;
8206
8207 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8208 goto err_md;
8209
8210 if ((ret = register_blkdev(0, "mdp")) < 0)
8211 goto err_mdp;
8212 mdp_major = ret;
8213
3dbd8c2e 8214 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8215 md_probe, NULL, NULL);
8216 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8217 md_probe, NULL, NULL);
8218
1da177e4 8219 register_reboot_notifier(&md_notifier);
0b4d4147 8220 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8221
8222 md_geninit();
d710e138 8223 return 0;
1da177e4 8224
e804ac78
TH		 8225err_mdp:
8226 unregister_blkdev(MD_MAJOR, "md");
8227err_md:
8228 destroy_workqueue(md_misc_wq);
8229err_misc_wq:
8230 destroy_workqueue(md_wq);
8231err_wq:
8232 return ret;
8233}
1da177e4
LT		 8234
8235#ifndef MODULE
8236
8237/*
8238 * Searches all registered partitions for autorun RAID arrays
8239 * at boot time.
8240 */
4d936ec1
ME		 8241
8242static LIST_HEAD(all_detected_devices);
8243struct detected_devices_node {
8244 struct list_head list;
8245 dev_t dev;
8246};
1da177e4
LT		 8247
8248void md_autodetect_dev(dev_t dev)
8249{
4d936ec1
ME		 8250 struct detected_devices_node *node_detected_dev;
8251
8252 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8253 if (node_detected_dev) {
8254 node_detected_dev->dev = dev;
8255 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8256 } else {
8257 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8258 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8259 }
1da177e4
LT		 8260}
8261
8262
8263static void autostart_arrays(int part)
8264{
3cb03002 8265 struct md_rdev *rdev;
4d936ec1
ME		 8266 struct detected_devices_node *node_detected_dev;
8267 dev_t dev;
8268 int i_scanned, i_passed;
1da177e4 8269
4d936ec1
ME		 8270 i_scanned = 0;
8271 i_passed = 0;
1da177e4 8272
4d936ec1 8273 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8274
4d936ec1
ME		 8275 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8276 i_scanned++;
8277 node_detected_dev = list_entry(all_detected_devices.next,
8278 struct detected_devices_node, list);
8279 list_del(&node_detected_dev->list);
8280 dev = node_detected_dev->dev;
8281 kfree(node_detected_dev);
df968c4e 8282 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8283 if (IS_ERR(rdev))
8284 continue;
8285
b2d444d7 8286 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8287 MD_BUG();
8288 continue;
8289 }
d0fae18f 8290 set_bit(AutoDetected, &rdev->flags);
1da177e4 8291 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8292 i_passed++;
1da177e4 8293 }
4d936ec1
ME		 8294
8295 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8296 i_scanned, i_passed);
1da177e4
LT		 8297
8298 autorun_devices(part);
8299}
8300
fdee8ae4 8301#endif /* !MODULE */
1da177e4
LT		 8302
8303static __exit void md_exit(void)
8304{
fd01b88c 8305 struct mddev *mddev;
1da177e4 8306 struct list_head *tmp;
8ab5e4c1 8307
3dbd8c2e 8308 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8309 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8310
3dbd8c2e 8311 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8312 unregister_blkdev(mdp_major, "mdp");
8313 unregister_reboot_notifier(&md_notifier);
8314 unregister_sysctl_table(raid_table_header);
8315 remove_proc_entry("mdstat", NULL);
29ac4aa3 8316 for_each_mddev(mddev, tmp) {
1da177e4 8317 export_array(mddev);
d3374825 8318 mddev->hold_active = 0;
1da177e4 8319 }
e804ac78
TH		 8320 destroy_workqueue(md_misc_wq);
8321 destroy_workqueue(md_wq);
1da177e4
LT		 8322}
8323
685784aa 8324subsys_initcall(md_init);
1da177e4
LT		 8325module_exit(md_exit)
8326
f91de92e
N		 8327static int get_ro(char *buffer, struct kernel_param *kp)
8328{
8329 return sprintf(buffer, "%d", start_readonly);
8330}
8331static int set_ro(const char *val, struct kernel_param *kp)
8332{
8333 char *e;
8334 int num = simple_strtoul(val, &e, 10);
8335 if (*val && (*e == '\0' || *e == '\n')) {
8336 start_readonly = num;
4dbcdc75 8337 return 0;
f91de92e
N		 8338 }
8339 return -EINVAL;
8340}
8341
80ca3a44
N		 8342module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8343module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8344
efeb53c0 8345module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8346
1da177e4
LT		 8347EXPORT_SYMBOL(register_md_personality);
8348EXPORT_SYMBOL(unregister_md_personality);
8349EXPORT_SYMBOL(md_error);
8350EXPORT_SYMBOL(md_done_sync);
8351EXPORT_SYMBOL(md_write_start);
8352EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8353EXPORT_SYMBOL(md_register_thread);
8354EXPORT_SYMBOL(md_unregister_thread);
8355EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 8356EXPORT_SYMBOL(md_check_recovery);
8357MODULE_LICENSE("GPL");
0efb9e61 8358MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8359MODULE_ALIAS("md");
72008652 8360MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)