blob: 7cf512a34ccfc1ecd84c5264aff7d44184a14d95 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
NeilBrown32a76272005-06-21 17:17:14 -070022 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 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
35#include <linux/module.h>
Ahmed S. Darwish50511da2007-05-09 02:35:34 -070036#include <linux/kernel.h>
NeilBrowna6fb0932005-09-09 16:23:56 -070037#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/linkage.h>
39#include <linux/raid/md.h>
NeilBrown32a76272005-06-21 17:17:14 -070040#include <linux/raid/bitmap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/sysctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/buffer_head.h> /* for invalidate_bdev */
NeilBrownd7603b72006-01-06 00:20:30 -080043#include <linux/poll.h>
Arjan van de Ven48c9c272006-03-27 01:18:20 -080044#include <linux/mutex.h>
NeilBrown16f17b32006-06-26 00:27:37 -070045#include <linux/ctype.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080046#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#include <linux/init.h>
49
NeilBrown32a76272005-06-21 17:17:14 -070050#include <linux/file.h>
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#ifdef CONFIG_KMOD
53#include <linux/kmod.h>
54#endif
55
56#include <asm/unaligned.h>
57
58#define MAJOR_NR MD_MAJOR
59#define MD_DRIVER
60
61/* 63 partitions with the alternate major number (mdp) */
62#define MdpMinorShift 6
63
64#define DEBUG 0
65#define dprintk(x...) ((void)(DEBUG && printk(x)))
66
67
68#ifndef MODULE
69static void autostart_arrays (int part);
70#endif
71
NeilBrown2604b702006-01-06 00:20:36 -080072static LIST_HEAD(pers_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073static DEFINE_SPINLOCK(pers_lock);
74
Adrian Bunk5e563412006-06-26 00:27:42 -070075static void md_print_devices(void);
76
Bernd Schubert90b08712008-05-23 13:04:38 -070077static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
78
Adrian Bunk5e563412006-06-26 00:27:42 -070079#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/*
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
Adrian Bunk338cec32005-09-10 00:26:54 -070085 * the RAID driver will use the maximum available bandwidth if the IO
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
88 * idle IO detection.
89 *
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
NeilBrown88202a02006-01-06 00:21:36 -080091 * or /sys/block/mdX/md/sync_speed_{min,max}
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 */
93
94static int sysctl_speed_limit_min = 1000;
95static int sysctl_speed_limit_max = 200000;
NeilBrown88202a02006-01-06 00:21:36 -080096static inline int speed_min(mddev_t *mddev)
97{
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
100}
101
102static inline int speed_max(mddev_t *mddev)
103{
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
106}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108static struct ctl_table_header *raid_table_header;
109
110static ctl_table raid_table[] = {
111 {
112 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
113 .procname = "speed_limit_min",
114 .data = &sysctl_speed_limit_min,
115 .maxlen = sizeof(int),
NeilBrown80ca3a42006-07-10 04:44:18 -0700116 .mode = S_IRUGO|S_IWUSR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 .proc_handler = &proc_dointvec,
118 },
119 {
120 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
NeilBrown80ca3a42006-07-10 04:44:18 -0700124 .mode = S_IRUGO|S_IWUSR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 .proc_handler = &proc_dointvec,
126 },
127 { .ctl_name = 0 }
128};
129
130static ctl_table raid_dir_table[] = {
131 {
132 .ctl_name = DEV_RAID,
133 .procname = "raid",
134 .maxlen = 0,
NeilBrown80ca3a42006-07-10 04:44:18 -0700135 .mode = S_IRUGO|S_IXUGO,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 .child = raid_table,
137 },
138 { .ctl_name = 0 }
139};
140
141static ctl_table raid_root_table[] = {
142 {
143 .ctl_name = CTL_DEV,
144 .procname = "dev",
145 .maxlen = 0,
146 .mode = 0555,
147 .child = raid_dir_table,
148 },
149 { .ctl_name = 0 }
150};
151
152static struct block_device_operations md_fops;
153
NeilBrownf91de922005-11-08 21:39:36 -0800154static int start_readonly;
155
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156/*
NeilBrownd7603b72006-01-06 00:20:30 -0800157 * We have a system wide 'event count' that is incremented
158 * on any 'interesting' event, and readers of /proc/mdstat
159 * can use 'poll' or 'select' to find out when the event
160 * count increases.
161 *
162 * Events are:
163 * start array, stop array, error, add device, remove device,
164 * start build, activate spare
165 */
NeilBrown2989ddb2006-01-06 00:20:43 -0800166static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
NeilBrownd7603b72006-01-06 00:20:30 -0800167static atomic_t md_event_count;
NeilBrown29269552006-03-27 01:18:10 -0800168void md_new_event(mddev_t *mddev)
NeilBrownd7603b72006-01-06 00:20:30 -0800169{
170 atomic_inc(&md_event_count);
171 wake_up(&md_event_waiters);
NeilBrown4508a7a2006-03-20 17:53:53 +1100172 sysfs_notify(&mddev->kobj, NULL, "sync_action");
NeilBrownd7603b72006-01-06 00:20:30 -0800173}
NeilBrown29269552006-03-27 01:18:10 -0800174EXPORT_SYMBOL_GPL(md_new_event);
NeilBrownd7603b72006-01-06 00:20:30 -0800175
NeilBrownc331eb02006-05-30 21:27:13 -0700176/* Alternate version that can be called from interrupts
177 * when calling sysfs_notify isn't needed.
178 */
Adrian Bunk05381952006-06-26 00:28:01 -0700179static void md_new_event_inintr(mddev_t *mddev)
NeilBrownc331eb02006-05-30 21:27:13 -0700180{
181 atomic_inc(&md_event_count);
182 wake_up(&md_event_waiters);
183}
184
NeilBrownd7603b72006-01-06 00:20:30 -0800185/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 * Enables to iterate over all existing md arrays
187 * all_mddevs_lock protects this list.
188 */
189static LIST_HEAD(all_mddevs);
190static DEFINE_SPINLOCK(all_mddevs_lock);
191
192
193/*
194 * iterates through all used mddevs in the system.
195 * We take care to grab the all_mddevs_lock whenever navigating
196 * the list, and to always hold a refcount when unlocked.
197 * Any code which breaks out of this loop while own
198 * a reference to the current mddev and must mddev_put it.
199 */
NeilBrown29ac4aa2008-02-06 01:39:58 -0800200#define for_each_mddev(mddev,tmp) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 \
202 for (({ spin_lock(&all_mddevs_lock); \
203 tmp = all_mddevs.next; \
204 mddev = NULL;}); \
205 ({ if (tmp != &all_mddevs) \
206 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
207 spin_unlock(&all_mddevs_lock); \
208 if (mddev) mddev_put(mddev); \
209 mddev = list_entry(tmp, mddev_t, all_mddevs); \
210 tmp != &all_mddevs;}); \
211 ({ spin_lock(&all_mddevs_lock); \
212 tmp = tmp->next;}) \
213 )
214
215
Jens Axboe165125e2007-07-24 09:28:11 +0200216static int md_fail_request (struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
NeilBrown6712ecf2007-09-27 12:47:43 +0200218 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 return 0;
220}
221
222static inline mddev_t *mddev_get(mddev_t *mddev)
223{
224 atomic_inc(&mddev->active);
225 return mddev;
226}
227
228static void mddev_put(mddev_t *mddev)
229{
230 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
231 return;
232 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
233 list_del(&mddev->all_mddevs);
NeilBrown926ce2d2006-03-31 02:32:02 -0800234 spin_unlock(&all_mddevs_lock);
Al Viro1312f402006-03-12 11:02:03 -0500235 blk_cleanup_queue(mddev->queue);
Greg Kroah-Hartmanc10997f2007-12-20 08:13:05 -0800236 kobject_put(&mddev->kobj);
NeilBrown926ce2d2006-03-31 02:32:02 -0800237 } else
238 spin_unlock(&all_mddevs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
241static mddev_t * mddev_find(dev_t unit)
242{
243 mddev_t *mddev, *new = NULL;
244
245 retry:
246 spin_lock(&all_mddevs_lock);
247 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
248 if (mddev->unit == unit) {
249 mddev_get(mddev);
250 spin_unlock(&all_mddevs_lock);
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700251 kfree(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 return mddev;
253 }
254
255 if (new) {
256 list_add(&new->all_mddevs, &all_mddevs);
257 spin_unlock(&all_mddevs_lock);
258 return new;
259 }
260 spin_unlock(&all_mddevs_lock);
261
NeilBrown9ffae0c2006-01-06 00:20:32 -0800262 new = kzalloc(sizeof(*new), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 if (!new)
264 return NULL;
265
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 new->unit = unit;
267 if (MAJOR(unit) == MD_MAJOR)
268 new->md_minor = MINOR(unit);
269 else
270 new->md_minor = MINOR(unit) >> MdpMinorShift;
271
NeilBrowndf5b89b2006-03-27 01:18:20 -0800272 mutex_init(&new->reconfig_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 INIT_LIST_HEAD(&new->disks);
274 INIT_LIST_HEAD(&new->all_mddevs);
275 init_timer(&new->safemode_timer);
276 atomic_set(&new->active, 1);
NeilBrown06d91a52005-06-21 17:17:12 -0700277 spin_lock_init(&new->write_lock);
NeilBrown3d310eb2005-06-21 17:17:26 -0700278 init_waitqueue_head(&new->sb_wait);
Dan Williamsa6d81132008-06-05 22:45:53 -0700279 init_waitqueue_head(&new->recovery_wait);
NeilBrown08a02ec2007-05-09 02:35:38 -0700280 new->reshape_position = MaxSector;
NeilBrownc6207272008-02-06 01:39:52 -0800281 new->resync_max = MaxSector;
NeilBrownd897dbf2008-04-30 00:52:29 -0700282 new->level = LEVEL_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284 new->queue = blk_alloc_queue(GFP_KERNEL);
285 if (!new->queue) {
286 kfree(new);
287 return NULL;
288 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200289 /* Can be unlocked because the queue is new: no concurrency */
290 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
292 blk_queue_make_request(new->queue, md_fail_request);
293
294 goto retry;
295}
296
297static inline int mddev_lock(mddev_t * mddev)
298{
NeilBrowndf5b89b2006-03-27 01:18:20 -0800299 return mutex_lock_interruptible(&mddev->reconfig_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302static inline int mddev_trylock(mddev_t * mddev)
303{
NeilBrowndf5b89b2006-03-27 01:18:20 -0800304 return mutex_trylock(&mddev->reconfig_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305}
306
307static inline void mddev_unlock(mddev_t * mddev)
308{
NeilBrowndf5b89b2006-03-27 01:18:20 -0800309 mutex_unlock(&mddev->reconfig_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
NeilBrown005eca52005-08-22 13:11:08 -0700311 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312}
313
NeilBrown2989ddb2006-01-06 00:20:43 -0800314static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
316 mdk_rdev_t * rdev;
317 struct list_head *tmp;
318
NeilBrownd089c6a2008-02-06 01:39:59 -0800319 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 if (rdev->desc_nr == nr)
321 return rdev;
322 }
323 return NULL;
324}
325
326static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
327{
328 struct list_head *tmp;
329 mdk_rdev_t *rdev;
330
NeilBrownd089c6a2008-02-06 01:39:59 -0800331 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 if (rdev->bdev->bd_dev == dev)
333 return rdev;
334 }
335 return NULL;
336}
337
NeilBrownd9d166c2006-01-06 00:20:51 -0800338static struct mdk_personality *find_pers(int level, char *clevel)
NeilBrown2604b702006-01-06 00:20:36 -0800339{
340 struct mdk_personality *pers;
NeilBrownd9d166c2006-01-06 00:20:51 -0800341 list_for_each_entry(pers, &pers_list, list) {
342 if (level != LEVEL_NONE && pers->level == level)
NeilBrown2604b702006-01-06 00:20:36 -0800343 return pers;
NeilBrownd9d166c2006-01-06 00:20:51 -0800344 if (strcmp(pers->name, clevel)==0)
345 return pers;
346 }
NeilBrown2604b702006-01-06 00:20:36 -0800347 return NULL;
348}
349
Jesper Juhl77933d72005-07-27 11:46:09 -0700350static inline sector_t calc_dev_sboffset(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
352 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
353 return MD_NEW_SIZE_BLOCKS(size);
354}
355
356static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
357{
358 sector_t size;
359
360 size = rdev->sb_offset;
361
362 if (chunk_size)
363 size &= ~((sector_t)chunk_size/1024 - 1);
364 return size;
365}
366
367static int alloc_disk_sb(mdk_rdev_t * rdev)
368{
369 if (rdev->sb_page)
370 MD_BUG();
371
372 rdev->sb_page = alloc_page(GFP_KERNEL);
373 if (!rdev->sb_page) {
374 printk(KERN_ALERT "md: out of memory.\n");
375 return -EINVAL;
376 }
377
378 return 0;
379}
380
381static void free_disk_sb(mdk_rdev_t * rdev)
382{
383 if (rdev->sb_page) {
NeilBrown2d1f3b52006-01-06 00:20:31 -0800384 put_page(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 rdev->sb_loaded = 0;
386 rdev->sb_page = NULL;
387 rdev->sb_offset = 0;
388 rdev->size = 0;
389 }
390}
391
392
NeilBrown6712ecf2007-09-27 12:47:43 +0200393static void super_written(struct bio *bio, int error)
NeilBrown7bfa19f2005-06-21 17:17:28 -0700394{
395 mdk_rdev_t *rdev = bio->bi_private;
NeilBrowna9701a32005-11-08 21:39:34 -0800396 mddev_t *mddev = rdev->mddev;
NeilBrown7bfa19f2005-06-21 17:17:28 -0700397
NeilBrown3a0f5bb2006-10-03 01:16:03 -0700398 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
399 printk("md: super_written gets error=%d, uptodate=%d\n",
400 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
401 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
NeilBrowna9701a32005-11-08 21:39:34 -0800402 md_error(mddev, rdev);
NeilBrown3a0f5bb2006-10-03 01:16:03 -0700403 }
NeilBrown7bfa19f2005-06-21 17:17:28 -0700404
NeilBrowna9701a32005-11-08 21:39:34 -0800405 if (atomic_dec_and_test(&mddev->pending_writes))
406 wake_up(&mddev->sb_wait);
Neil Brownf8b58ed2005-06-27 22:29:34 -0700407 bio_put(bio);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700408}
409
NeilBrown6712ecf2007-09-27 12:47:43 +0200410static void super_written_barrier(struct bio *bio, int error)
NeilBrowna9701a32005-11-08 21:39:34 -0800411{
412 struct bio *bio2 = bio->bi_private;
413 mdk_rdev_t *rdev = bio2->bi_private;
414 mddev_t *mddev = rdev->mddev;
NeilBrowna9701a32005-11-08 21:39:34 -0800415
416 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
417 error == -EOPNOTSUPP) {
418 unsigned long flags;
419 /* barriers don't appear to be supported :-( */
420 set_bit(BarriersNotsupp, &rdev->flags);
421 mddev->barriers_work = 0;
422 spin_lock_irqsave(&mddev->write_lock, flags);
423 bio2->bi_next = mddev->biolist;
424 mddev->biolist = bio2;
425 spin_unlock_irqrestore(&mddev->write_lock, flags);
426 wake_up(&mddev->sb_wait);
427 bio_put(bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200428 } else {
429 bio_put(bio2);
430 bio->bi_private = rdev;
431 super_written(bio, error);
NeilBrowna9701a32005-11-08 21:39:34 -0800432 }
NeilBrowna9701a32005-11-08 21:39:34 -0800433}
434
NeilBrown7bfa19f2005-06-21 17:17:28 -0700435void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
436 sector_t sector, int size, struct page *page)
437{
438 /* write first size bytes of page to sector of rdev
439 * Increment mddev->pending_writes before returning
440 * and decrement it on completion, waking up sb_wait
441 * if zero is reached.
442 * If an error occurred, call md_error
NeilBrowna9701a32005-11-08 21:39:34 -0800443 *
444 * As we might need to resubmit the request if BIO_RW_BARRIER
445 * causes ENOTSUPP, we allocate a spare bio...
NeilBrown7bfa19f2005-06-21 17:17:28 -0700446 */
447 struct bio *bio = bio_alloc(GFP_NOIO, 1);
NeilBrowna9701a32005-11-08 21:39:34 -0800448 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700449
450 bio->bi_bdev = rdev->bdev;
451 bio->bi_sector = sector;
452 bio_add_page(bio, page, size, 0);
453 bio->bi_private = rdev;
454 bio->bi_end_io = super_written;
NeilBrowna9701a32005-11-08 21:39:34 -0800455 bio->bi_rw = rw;
456
NeilBrown7bfa19f2005-06-21 17:17:28 -0700457 atomic_inc(&mddev->pending_writes);
NeilBrowna9701a32005-11-08 21:39:34 -0800458 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
459 struct bio *rbio;
460 rw |= (1<<BIO_RW_BARRIER);
461 rbio = bio_clone(bio, GFP_NOIO);
462 rbio->bi_private = bio;
463 rbio->bi_end_io = super_written_barrier;
464 submit_bio(rw, rbio);
465 } else
466 submit_bio(rw, bio);
467}
468
469void md_super_wait(mddev_t *mddev)
470{
471 /* wait for all superblock writes that were scheduled to complete.
472 * if any had to be retried (due to BARRIER problems), retry them
473 */
474 DEFINE_WAIT(wq);
475 for(;;) {
476 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
477 if (atomic_read(&mddev->pending_writes)==0)
478 break;
479 while (mddev->biolist) {
480 struct bio *bio;
481 spin_lock_irq(&mddev->write_lock);
482 bio = mddev->biolist;
483 mddev->biolist = bio->bi_next ;
484 bio->bi_next = NULL;
485 spin_unlock_irq(&mddev->write_lock);
486 submit_bio(bio->bi_rw, bio);
487 }
488 schedule();
489 }
490 finish_wait(&mddev->sb_wait, &wq);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700491}
492
NeilBrown6712ecf2007-09-27 12:47:43 +0200493static void bi_complete(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 complete((struct completion*)bio->bi_private);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496}
497
NeilBrowna654b9d82005-06-21 17:17:27 -0700498int sync_page_io(struct block_device *bdev, sector_t sector, int size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 struct page *page, int rw)
500{
Neil Brownbaaa2c52005-04-16 15:23:54 -0700501 struct bio *bio = bio_alloc(GFP_NOIO, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 struct completion event;
503 int ret;
504
505 rw |= (1 << BIO_RW_SYNC);
506
507 bio->bi_bdev = bdev;
508 bio->bi_sector = sector;
509 bio_add_page(bio, page, size, 0);
510 init_completion(&event);
511 bio->bi_private = &event;
512 bio->bi_end_io = bi_complete;
513 submit_bio(rw, bio);
514 wait_for_completion(&event);
515
516 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
517 bio_put(bio);
518 return ret;
519}
NeilBrowna8745db2006-01-06 00:20:34 -0800520EXPORT_SYMBOL_GPL(sync_page_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
NeilBrown0002b272005-09-09 16:23:53 -0700522static int read_disk_sb(mdk_rdev_t * rdev, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
524 char b[BDEVNAME_SIZE];
525 if (!rdev->sb_page) {
526 MD_BUG();
527 return -EINVAL;
528 }
529 if (rdev->sb_loaded)
530 return 0;
531
532
NeilBrown0002b272005-09-09 16:23:53 -0700533 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 goto fail;
535 rdev->sb_loaded = 1;
536 return 0;
537
538fail:
539 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
540 bdevname(rdev->bdev,b));
541 return -EINVAL;
542}
543
544static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
545{
546 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
547 (sb1->set_uuid1 == sb2->set_uuid1) &&
548 (sb1->set_uuid2 == sb2->set_uuid2) &&
549 (sb1->set_uuid3 == sb2->set_uuid3))
550
551 return 1;
552
553 return 0;
554}
555
556
557static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
558{
559 int ret;
560 mdp_super_t *tmp1, *tmp2;
561
562 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
563 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
564
565 if (!tmp1 || !tmp2) {
566 ret = 0;
567 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
568 goto abort;
569 }
570
571 *tmp1 = *sb1;
572 *tmp2 = *sb2;
573
574 /*
575 * nr_disks is not constant
576 */
577 tmp1->nr_disks = 0;
578 tmp2->nr_disks = 0;
579
580 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
581 ret = 0;
582 else
583 ret = 1;
584
585abort:
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700586 kfree(tmp1);
587 kfree(tmp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 return ret;
589}
590
NeilBrown4d167f02007-05-09 02:35:37 -0700591
592static u32 md_csum_fold(u32 csum)
593{
594 csum = (csum & 0xffff) + (csum >> 16);
595 return (csum & 0xffff) + (csum >> 16);
596}
597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598static unsigned int calc_sb_csum(mdp_super_t * sb)
599{
NeilBrown4d167f02007-05-09 02:35:37 -0700600 u64 newcsum = 0;
601 u32 *sb32 = (u32*)sb;
602 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 unsigned int disk_csum, csum;
604
605 disk_csum = sb->sb_csum;
606 sb->sb_csum = 0;
NeilBrown4d167f02007-05-09 02:35:37 -0700607
608 for (i = 0; i < MD_SB_BYTES/4 ; i++)
609 newcsum += sb32[i];
610 csum = (newcsum & 0xffffffff) + (newcsum>>32);
611
612
613#ifdef CONFIG_ALPHA
614 /* This used to use csum_partial, which was wrong for several
615 * reasons including that different results are returned on
616 * different architectures. It isn't critical that we get exactly
617 * the same return value as before (we always csum_fold before
618 * testing, and that removes any differences). However as we
619 * know that csum_partial always returned a 16bit value on
620 * alphas, do a fold to maximise conformity to previous behaviour.
621 */
622 sb->sb_csum = md_csum_fold(disk_csum);
623#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 sb->sb_csum = disk_csum;
NeilBrown4d167f02007-05-09 02:35:37 -0700625#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 return csum;
627}
628
629
630/*
631 * Handle superblock details.
632 * We want to be able to handle multiple superblock formats
633 * so we have a common interface to them all, and an array of
634 * different handlers.
635 * We rely on user-space to write the initial superblock, and support
636 * reading and updating of superblocks.
637 * Interface methods are:
638 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
639 * loads and validates a superblock on dev.
640 * if refdev != NULL, compare superblocks on both devices
641 * Return:
642 * 0 - dev has a superblock that is compatible with refdev
643 * 1 - dev has a superblock that is compatible and newer than refdev
644 * so dev should be used as the refdev in future
645 * -EINVAL superblock incompatible or invalid
646 * -othererror e.g. -EIO
647 *
648 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
649 * Verify that dev is acceptable into mddev.
650 * The first time, mddev->raid_disks will be 0, and data from
651 * dev should be merged in. Subsequent calls check that dev
652 * is new enough. Return 0 or -EINVAL
653 *
654 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
655 * Update the superblock for rdev with data in mddev
656 * This does not write to disc.
657 *
658 */
659
660struct super_type {
661 char *name;
662 struct module *owner;
663 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
664 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
665 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
666};
667
668/*
669 * load_super for 0.90.0
670 */
671static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
672{
673 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
674 mdp_super_t *sb;
675 int ret;
676 sector_t sb_offset;
677
678 /*
679 * Calculate the position of the superblock,
680 * it's at the end of the disk.
681 *
682 * It also happens to be a multiple of 4Kb.
683 */
684 sb_offset = calc_dev_sboffset(rdev->bdev);
685 rdev->sb_offset = sb_offset;
686
NeilBrown0002b272005-09-09 16:23:53 -0700687 ret = read_disk_sb(rdev, MD_SB_BYTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (ret) return ret;
689
690 ret = -EINVAL;
691
692 bdevname(rdev->bdev, b);
693 sb = (mdp_super_t*)page_address(rdev->sb_page);
694
695 if (sb->md_magic != MD_SB_MAGIC) {
696 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
697 b);
698 goto abort;
699 }
700
701 if (sb->major_version != 0 ||
NeilBrownf6705572006-03-27 01:18:11 -0800702 sb->minor_version < 90 ||
703 sb->minor_version > 91) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
705 sb->major_version, sb->minor_version,
706 b);
707 goto abort;
708 }
709
710 if (sb->raid_disks <= 0)
711 goto abort;
712
NeilBrown4d167f02007-05-09 02:35:37 -0700713 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
715 b);
716 goto abort;
717 }
718
719 rdev->preferred_minor = sb->md_minor;
720 rdev->data_offset = 0;
NeilBrown0002b272005-09-09 16:23:53 -0700721 rdev->sb_size = MD_SB_BYTES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
NeilBrowne11e93f2007-05-09 02:35:36 -0700723 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
724 if (sb->level != 1 && sb->level != 4
725 && sb->level != 5 && sb->level != 6
726 && sb->level != 10) {
727 /* FIXME use a better test */
728 printk(KERN_WARNING
729 "md: bitmaps not supported for this level.\n");
730 goto abort;
731 }
732 }
733
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 if (sb->level == LEVEL_MULTIPATH)
735 rdev->desc_nr = -1;
736 else
737 rdev->desc_nr = sb->this_disk.number;
738
Harvey Harrison9a7b2b02008-04-28 02:15:49 -0700739 if (!refdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 ret = 1;
Harvey Harrison9a7b2b02008-04-28 02:15:49 -0700741 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 __u64 ev1, ev2;
743 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
744 if (!uuid_equal(refsb, sb)) {
745 printk(KERN_WARNING "md: %s has different UUID to %s\n",
746 b, bdevname(refdev->bdev,b2));
747 goto abort;
748 }
749 if (!sb_equal(refsb, sb)) {
750 printk(KERN_WARNING "md: %s has same UUID"
751 " but different superblock to %s\n",
752 b, bdevname(refdev->bdev, b2));
753 goto abort;
754 }
755 ev1 = md_event(sb);
756 ev2 = md_event(refsb);
757 if (ev1 > ev2)
758 ret = 1;
759 else
760 ret = 0;
761 }
762 rdev->size = calc_dev_size(rdev, sb->chunk_size);
763
NeilBrown2bf071b2006-01-06 00:20:55 -0800764 if (rdev->size < sb->size && sb->level > 1)
765 /* "this cannot possibly happen" ... */
766 ret = -EINVAL;
767
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 abort:
769 return ret;
770}
771
772/*
773 * validate_super for 0.90.0
774 */
775static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
776{
777 mdp_disk_t *desc;
778 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
NeilBrown07d84d102006-06-26 00:27:56 -0700779 __u64 ev1 = md_event(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
NeilBrown41158c72005-06-21 17:17:25 -0700781 rdev->raid_disk = -1;
NeilBrownc5d79ad2008-02-06 01:39:54 -0800782 clear_bit(Faulty, &rdev->flags);
783 clear_bit(In_sync, &rdev->flags);
784 clear_bit(WriteMostly, &rdev->flags);
785 clear_bit(BarriersNotsupp, &rdev->flags);
786
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 if (mddev->raid_disks == 0) {
788 mddev->major_version = 0;
789 mddev->minor_version = sb->minor_version;
790 mddev->patch_version = sb->patch_version;
NeilBrowne6910632008-02-06 01:39:51 -0800791 mddev->external = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 mddev->chunk_size = sb->chunk_size;
793 mddev->ctime = sb->ctime;
794 mddev->utime = sb->utime;
795 mddev->level = sb->level;
NeilBrownd9d166c2006-01-06 00:20:51 -0800796 mddev->clevel[0] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 mddev->layout = sb->layout;
798 mddev->raid_disks = sb->raid_disks;
799 mddev->size = sb->size;
NeilBrown07d84d102006-06-26 00:27:56 -0700800 mddev->events = ev1;
NeilBrown92232142005-08-18 11:24:16 -0700801 mddev->bitmap_offset = 0;
NeilBrown36fa3062005-09-09 16:23:45 -0700802 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
NeilBrownf6705572006-03-27 01:18:11 -0800804 if (mddev->minor_version >= 91) {
805 mddev->reshape_position = sb->reshape_position;
806 mddev->delta_disks = sb->delta_disks;
807 mddev->new_level = sb->new_level;
808 mddev->new_layout = sb->new_layout;
809 mddev->new_chunk = sb->new_chunk;
810 } else {
811 mddev->reshape_position = MaxSector;
812 mddev->delta_disks = 0;
813 mddev->new_level = mddev->level;
814 mddev->new_layout = mddev->layout;
815 mddev->new_chunk = mddev->chunk_size;
816 }
817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (sb->state & (1<<MD_SB_CLEAN))
819 mddev->recovery_cp = MaxSector;
820 else {
821 if (sb->events_hi == sb->cp_events_hi &&
822 sb->events_lo == sb->cp_events_lo) {
823 mddev->recovery_cp = sb->recovery_cp;
824 } else
825 mddev->recovery_cp = 0;
826 }
827
828 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
829 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
830 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
831 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
832
833 mddev->max_disks = MD_SB_DISKS;
NeilBrowna654b9d82005-06-21 17:17:27 -0700834
835 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
NeilBrowne11e93f2007-05-09 02:35:36 -0700836 mddev->bitmap_file == NULL)
NeilBrown36fa3062005-09-09 16:23:45 -0700837 mddev->bitmap_offset = mddev->default_bitmap_offset;
NeilBrowna654b9d82005-06-21 17:17:27 -0700838
NeilBrown41158c72005-06-21 17:17:25 -0700839 } else if (mddev->pers == NULL) {
840 /* Insist on good event counter while assembling */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 ++ev1;
842 if (ev1 < mddev->events)
843 return -EINVAL;
NeilBrown41158c72005-06-21 17:17:25 -0700844 } else if (mddev->bitmap) {
845 /* if adding to array with a bitmap, then we can accept an
846 * older device ... but not too old.
847 */
NeilBrown41158c72005-06-21 17:17:25 -0700848 if (ev1 < mddev->bitmap->events_cleared)
849 return 0;
NeilBrown07d84d102006-06-26 00:27:56 -0700850 } else {
851 if (ev1 < mddev->events)
852 /* just a hot-add of a new device, leave raid_disk at -1 */
853 return 0;
854 }
NeilBrown41158c72005-06-21 17:17:25 -0700855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (mddev->level != LEVEL_MULTIPATH) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 desc = sb->disks + rdev->desc_nr;
858
859 if (desc->state & (1<<MD_DISK_FAULTY))
NeilBrownb2d444d2005-11-08 21:39:31 -0800860 set_bit(Faulty, &rdev->flags);
NeilBrown7c7546c2006-06-26 00:27:41 -0700861 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
862 desc->raid_disk < mddev->raid_disks */) {
NeilBrownb2d444d2005-11-08 21:39:31 -0800863 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 rdev->raid_disk = desc->raid_disk;
865 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700866 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
867 set_bit(WriteMostly, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -0700868 } else /* MULTIPATH are always insync */
NeilBrownb2d444d2005-11-08 21:39:31 -0800869 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 return 0;
871}
872
873/*
874 * sync_super for 0.90.0
875 */
876static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
877{
878 mdp_super_t *sb;
879 struct list_head *tmp;
880 mdk_rdev_t *rdev2;
881 int next_spare = mddev->raid_disks;
NeilBrown19133a42005-11-08 21:39:35 -0800882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 /* make rdev->sb match mddev data..
885 *
886 * 1/ zero out disks
887 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
888 * 3/ any empty disks < next_spare become removed
889 *
890 * disks[0] gets initialised to REMOVED because
891 * we cannot be sure from other fields if it has
892 * been initialised or not.
893 */
894 int i;
895 int active=0, working=0,failed=0,spare=0,nr_disks=0;
896
NeilBrown61181562005-09-09 16:24:02 -0700897 rdev->sb_size = MD_SB_BYTES;
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 sb = (mdp_super_t*)page_address(rdev->sb_page);
900
901 memset(sb, 0, sizeof(*sb));
902
903 sb->md_magic = MD_SB_MAGIC;
904 sb->major_version = mddev->major_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 sb->patch_version = mddev->patch_version;
906 sb->gvalid_words = 0; /* ignored */
907 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
908 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
909 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
910 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
911
912 sb->ctime = mddev->ctime;
913 sb->level = mddev->level;
914 sb->size = mddev->size;
915 sb->raid_disks = mddev->raid_disks;
916 sb->md_minor = mddev->md_minor;
NeilBrowne6910632008-02-06 01:39:51 -0800917 sb->not_persistent = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 sb->utime = mddev->utime;
919 sb->state = 0;
920 sb->events_hi = (mddev->events>>32);
921 sb->events_lo = (u32)mddev->events;
922
NeilBrownf6705572006-03-27 01:18:11 -0800923 if (mddev->reshape_position == MaxSector)
924 sb->minor_version = 90;
925 else {
926 sb->minor_version = 91;
927 sb->reshape_position = mddev->reshape_position;
928 sb->new_level = mddev->new_level;
929 sb->delta_disks = mddev->delta_disks;
930 sb->new_layout = mddev->new_layout;
931 sb->new_chunk = mddev->new_chunk;
932 }
933 mddev->minor_version = sb->minor_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 if (mddev->in_sync)
935 {
936 sb->recovery_cp = mddev->recovery_cp;
937 sb->cp_events_hi = (mddev->events>>32);
938 sb->cp_events_lo = (u32)mddev->events;
939 if (mddev->recovery_cp == MaxSector)
940 sb->state = (1<< MD_SB_CLEAN);
941 } else
942 sb->recovery_cp = 0;
943
944 sb->layout = mddev->layout;
945 sb->chunk_size = mddev->chunk_size;
946
NeilBrowna654b9d82005-06-21 17:17:27 -0700947 if (mddev->bitmap && mddev->bitmap_file == NULL)
948 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 sb->disks[0].state = (1<<MD_DISK_REMOVED);
NeilBrownd089c6a2008-02-06 01:39:59 -0800951 rdev_for_each(rdev2, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 mdp_disk_t *d;
NeilBrown86e6ffd2005-11-08 21:39:24 -0800953 int desc_nr;
NeilBrownb2d444d2005-11-08 21:39:31 -0800954 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
955 && !test_bit(Faulty, &rdev2->flags))
NeilBrown86e6ffd2005-11-08 21:39:24 -0800956 desc_nr = rdev2->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 else
NeilBrown86e6ffd2005-11-08 21:39:24 -0800958 desc_nr = next_spare++;
NeilBrown19133a42005-11-08 21:39:35 -0800959 rdev2->desc_nr = desc_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 d = &sb->disks[rdev2->desc_nr];
961 nr_disks++;
962 d->number = rdev2->desc_nr;
963 d->major = MAJOR(rdev2->bdev->bd_dev);
964 d->minor = MINOR(rdev2->bdev->bd_dev);
NeilBrownb2d444d2005-11-08 21:39:31 -0800965 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
966 && !test_bit(Faulty, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 d->raid_disk = rdev2->raid_disk;
968 else
969 d->raid_disk = rdev2->desc_nr; /* compatibility */
NeilBrown1be78922006-03-27 01:18:03 -0800970 if (test_bit(Faulty, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 d->state = (1<<MD_DISK_FAULTY);
NeilBrown1be78922006-03-27 01:18:03 -0800972 else if (test_bit(In_sync, &rdev2->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 d->state = (1<<MD_DISK_ACTIVE);
974 d->state |= (1<<MD_DISK_SYNC);
975 active++;
976 working++;
977 } else {
978 d->state = 0;
979 spare++;
980 working++;
981 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700982 if (test_bit(WriteMostly, &rdev2->flags))
983 d->state |= (1<<MD_DISK_WRITEMOSTLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 /* now set the "removed" and "faulty" bits on any missing devices */
986 for (i=0 ; i < mddev->raid_disks ; i++) {
987 mdp_disk_t *d = &sb->disks[i];
988 if (d->state == 0 && d->number == 0) {
989 d->number = i;
990 d->raid_disk = i;
991 d->state = (1<<MD_DISK_REMOVED);
992 d->state |= (1<<MD_DISK_FAULTY);
993 failed++;
994 }
995 }
996 sb->nr_disks = nr_disks;
997 sb->active_disks = active;
998 sb->working_disks = working;
999 sb->failed_disks = failed;
1000 sb->spare_disks = spare;
1001
1002 sb->this_disk = sb->disks[rdev->desc_nr];
1003 sb->sb_csum = calc_sb_csum(sb);
1004}
1005
1006/*
1007 * version 1 superblock
1008 */
1009
NeilBrown1c05b4b2006-10-21 10:24:08 -07001010static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011{
NeilBrown1c05b4b2006-10-21 10:24:08 -07001012 __le32 disk_csum;
1013 u32 csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 unsigned long long newcsum;
1015 int size = 256 + le32_to_cpu(sb->max_dev)*2;
NeilBrown1c05b4b2006-10-21 10:24:08 -07001016 __le32 *isuper = (__le32*)sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 int i;
1018
1019 disk_csum = sb->sb_csum;
1020 sb->sb_csum = 0;
1021 newcsum = 0;
1022 for (i=0; size>=4; size -= 4 )
1023 newcsum += le32_to_cpu(*isuper++);
1024
1025 if (size == 2)
NeilBrown1c05b4b2006-10-21 10:24:08 -07001026 newcsum += le16_to_cpu(*(__le16*) isuper);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
1028 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1029 sb->sb_csum = disk_csum;
1030 return cpu_to_le32(csum);
1031}
1032
1033static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1034{
1035 struct mdp_superblock_1 *sb;
1036 int ret;
1037 sector_t sb_offset;
1038 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
NeilBrown0002b272005-09-09 16:23:53 -07001039 int bmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 /*
1042 * Calculate the position of the superblock.
1043 * It is always aligned to a 4K boundary and
1044 * depeding on minor_version, it can be:
1045 * 0: At least 8K, but less than 12K, from end of device
1046 * 1: At start of device
1047 * 2: 4K from start of device.
1048 */
1049 switch(minor_version) {
1050 case 0:
1051 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1052 sb_offset -= 8*2;
NeilBrown39730962005-06-21 17:17:28 -07001053 sb_offset &= ~(sector_t)(4*2-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 /* convert from sectors to K */
1055 sb_offset /= 2;
1056 break;
1057 case 1:
1058 sb_offset = 0;
1059 break;
1060 case 2:
1061 sb_offset = 4;
1062 break;
1063 default:
1064 return -EINVAL;
1065 }
1066 rdev->sb_offset = sb_offset;
1067
NeilBrown0002b272005-09-09 16:23:53 -07001068 /* superblock is rarely larger than 1K, but it can be larger,
1069 * and it is safe to read 4k, so we do that
1070 */
1071 ret = read_disk_sb(rdev, 4096);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (ret) return ret;
1073
1074
1075 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1076
1077 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1078 sb->major_version != cpu_to_le32(1) ||
1079 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1080 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
NeilBrown71c08052005-09-09 16:23:51 -07001081 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 return -EINVAL;
1083
1084 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1085 printk("md: invalid superblock checksum on %s\n",
1086 bdevname(rdev->bdev,b));
1087 return -EINVAL;
1088 }
1089 if (le64_to_cpu(sb->data_size) < 10) {
1090 printk("md: data_size too small on %s\n",
1091 bdevname(rdev->bdev,b));
1092 return -EINVAL;
1093 }
NeilBrowne11e93f2007-05-09 02:35:36 -07001094 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1095 if (sb->level != cpu_to_le32(1) &&
1096 sb->level != cpu_to_le32(4) &&
1097 sb->level != cpu_to_le32(5) &&
1098 sb->level != cpu_to_le32(6) &&
1099 sb->level != cpu_to_le32(10)) {
1100 printk(KERN_WARNING
1101 "md: bitmaps not supported for this level.\n");
1102 return -EINVAL;
1103 }
1104 }
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 rdev->preferred_minor = 0xffff;
1107 rdev->data_offset = le64_to_cpu(sb->data_offset);
NeilBrown4dbcdc72006-01-06 00:20:52 -08001108 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
NeilBrown0002b272005-09-09 16:23:53 -07001110 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
NeilBrown720a3dc2005-09-09 16:23:59 -07001111 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
NeilBrown0002b272005-09-09 16:23:53 -07001112 if (rdev->sb_size & bmask)
NeilBrowna1801f82008-03-04 14:29:31 -08001113 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1114
1115 if (minor_version
1116 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1117 return -EINVAL;
NeilBrown0002b272005-09-09 16:23:53 -07001118
NeilBrown31b65a02006-07-10 04:44:14 -07001119 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1120 rdev->desc_nr = -1;
1121 else
1122 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1123
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001124 if (!refdev) {
NeilBrown8ed75462006-02-03 03:03:41 -08001125 ret = 1;
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001126 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 __u64 ev1, ev2;
1128 struct mdp_superblock_1 *refsb =
1129 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1130
1131 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1132 sb->level != refsb->level ||
1133 sb->layout != refsb->layout ||
1134 sb->chunksize != refsb->chunksize) {
1135 printk(KERN_WARNING "md: %s has strangely different"
1136 " superblock to %s\n",
1137 bdevname(rdev->bdev,b),
1138 bdevname(refdev->bdev,b2));
1139 return -EINVAL;
1140 }
1141 ev1 = le64_to_cpu(sb->events);
1142 ev2 = le64_to_cpu(refsb->events);
1143
1144 if (ev1 > ev2)
NeilBrown8ed75462006-02-03 03:03:41 -08001145 ret = 1;
1146 else
1147 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 }
NeilBrowna1801f82008-03-04 14:29:31 -08001149 if (minor_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1151 else
1152 rdev->size = rdev->sb_offset;
1153 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1154 return -EINVAL;
1155 rdev->size = le64_to_cpu(sb->data_size)/2;
1156 if (le32_to_cpu(sb->chunksize))
1157 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
NeilBrown2bf071b2006-01-06 00:20:55 -08001158
NeilBrown1c05b4b2006-10-21 10:24:08 -07001159 if (le64_to_cpu(sb->size) > rdev->size*2)
NeilBrown2bf071b2006-01-06 00:20:55 -08001160 return -EINVAL;
NeilBrown8ed75462006-02-03 03:03:41 -08001161 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162}
1163
1164static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1165{
1166 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
NeilBrown07d84d102006-06-26 00:27:56 -07001167 __u64 ev1 = le64_to_cpu(sb->events);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
NeilBrown41158c72005-06-21 17:17:25 -07001169 rdev->raid_disk = -1;
NeilBrownc5d79ad2008-02-06 01:39:54 -08001170 clear_bit(Faulty, &rdev->flags);
1171 clear_bit(In_sync, &rdev->flags);
1172 clear_bit(WriteMostly, &rdev->flags);
1173 clear_bit(BarriersNotsupp, &rdev->flags);
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (mddev->raid_disks == 0) {
1176 mddev->major_version = 1;
1177 mddev->patch_version = 0;
NeilBrowne6910632008-02-06 01:39:51 -08001178 mddev->external = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1180 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1181 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1182 mddev->level = le32_to_cpu(sb->level);
NeilBrownd9d166c2006-01-06 00:20:51 -08001183 mddev->clevel[0] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 mddev->layout = le32_to_cpu(sb->layout);
1185 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1186 mddev->size = le64_to_cpu(sb->size)/2;
NeilBrown07d84d102006-06-26 00:27:56 -07001187 mddev->events = ev1;
NeilBrown92232142005-08-18 11:24:16 -07001188 mddev->bitmap_offset = 0;
NeilBrown29fc7e32006-02-03 03:03:41 -08001189 mddev->default_bitmap_offset = 1024 >> 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1192 memcpy(mddev->uuid, sb->set_uuid, 16);
1193
1194 mddev->max_disks = (4096-256)/2;
NeilBrowna654b9d82005-06-21 17:17:27 -07001195
NeilBrown71c08052005-09-09 16:23:51 -07001196 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
NeilBrowne11e93f2007-05-09 02:35:36 -07001197 mddev->bitmap_file == NULL )
NeilBrowna654b9d82005-06-21 17:17:27 -07001198 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
NeilBrowne11e93f2007-05-09 02:35:36 -07001199
NeilBrownf6705572006-03-27 01:18:11 -08001200 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1201 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1202 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1203 mddev->new_level = le32_to_cpu(sb->new_level);
1204 mddev->new_layout = le32_to_cpu(sb->new_layout);
1205 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1206 } else {
1207 mddev->reshape_position = MaxSector;
1208 mddev->delta_disks = 0;
1209 mddev->new_level = mddev->level;
1210 mddev->new_layout = mddev->layout;
1211 mddev->new_chunk = mddev->chunk_size;
1212 }
1213
NeilBrown41158c72005-06-21 17:17:25 -07001214 } else if (mddev->pers == NULL) {
1215 /* Insist of good event counter while assembling */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 ++ev1;
1217 if (ev1 < mddev->events)
1218 return -EINVAL;
NeilBrown41158c72005-06-21 17:17:25 -07001219 } else if (mddev->bitmap) {
1220 /* If adding to array with a bitmap, then we can accept an
1221 * older device, but not too old.
1222 */
NeilBrown41158c72005-06-21 17:17:25 -07001223 if (ev1 < mddev->bitmap->events_cleared)
1224 return 0;
NeilBrown07d84d102006-06-26 00:27:56 -07001225 } else {
1226 if (ev1 < mddev->events)
1227 /* just a hot-add of a new device, leave raid_disk at -1 */
1228 return 0;
1229 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 if (mddev->level != LEVEL_MULTIPATH) {
1231 int role;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1233 switch(role) {
1234 case 0xffff: /* spare */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 break;
1236 case 0xfffe: /* faulty */
NeilBrownb2d444d2005-11-08 21:39:31 -08001237 set_bit(Faulty, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 break;
1239 default:
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001240 if ((le32_to_cpu(sb->feature_map) &
1241 MD_FEATURE_RECOVERY_OFFSET))
1242 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1243 else
1244 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 rdev->raid_disk = role;
1246 break;
1247 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07001248 if (sb->devflags & WriteMostly1)
1249 set_bit(WriteMostly, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -07001250 } else /* MULTIPATH are always insync */
NeilBrownb2d444d2005-11-08 21:39:31 -08001251 set_bit(In_sync, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -07001252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 return 0;
1254}
1255
1256static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1257{
1258 struct mdp_superblock_1 *sb;
1259 struct list_head *tmp;
1260 mdk_rdev_t *rdev2;
1261 int max_dev, i;
1262 /* make rdev->sb match mddev and rdev data. */
1263
1264 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1265
1266 sb->feature_map = 0;
1267 sb->pad0 = 0;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001268 sb->recovery_offset = cpu_to_le64(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 memset(sb->pad1, 0, sizeof(sb->pad1));
1270 memset(sb->pad2, 0, sizeof(sb->pad2));
1271 memset(sb->pad3, 0, sizeof(sb->pad3));
1272
1273 sb->utime = cpu_to_le64((__u64)mddev->utime);
1274 sb->events = cpu_to_le64(mddev->events);
1275 if (mddev->in_sync)
1276 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1277 else
1278 sb->resync_offset = cpu_to_le64(0);
1279
NeilBrown1c05b4b2006-10-21 10:24:08 -07001280 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
NeilBrown4dbcdc72006-01-06 00:20:52 -08001281
NeilBrownf0ca3402006-02-02 14:28:04 -08001282 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
NeilBrown29fc7e32006-02-03 03:03:41 -08001283 sb->size = cpu_to_le64(mddev->size<<1);
NeilBrownf0ca3402006-02-02 14:28:04 -08001284
NeilBrowna654b9d82005-06-21 17:17:27 -07001285 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1286 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
NeilBrown71c08052005-09-09 16:23:51 -07001287 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
NeilBrowna654b9d82005-06-21 17:17:27 -07001288 }
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001289
1290 if (rdev->raid_disk >= 0 &&
1291 !test_bit(In_sync, &rdev->flags) &&
1292 rdev->recovery_offset > 0) {
1293 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1294 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1295 }
1296
NeilBrownf6705572006-03-27 01:18:11 -08001297 if (mddev->reshape_position != MaxSector) {
1298 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1299 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1300 sb->new_layout = cpu_to_le32(mddev->new_layout);
1301 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1302 sb->new_level = cpu_to_le32(mddev->new_level);
1303 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1304 }
NeilBrowna654b9d82005-06-21 17:17:27 -07001305
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 max_dev = 0;
NeilBrownd089c6a2008-02-06 01:39:59 -08001307 rdev_for_each(rdev2, tmp, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 if (rdev2->desc_nr+1 > max_dev)
1309 max_dev = rdev2->desc_nr+1;
NeilBrowna778b732007-05-23 13:58:10 -07001310
1311 if (max_dev > le32_to_cpu(sb->max_dev))
1312 sb->max_dev = cpu_to_le32(max_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 for (i=0; i<max_dev;i++)
1314 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1315
NeilBrownd089c6a2008-02-06 01:39:59 -08001316 rdev_for_each(rdev2, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 i = rdev2->desc_nr;
NeilBrownb2d444d2005-11-08 21:39:31 -08001318 if (test_bit(Faulty, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 sb->dev_roles[i] = cpu_to_le16(0xfffe);
NeilBrownb2d444d2005-11-08 21:39:31 -08001320 else if (test_bit(In_sync, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001322 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1323 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 else
1325 sb->dev_roles[i] = cpu_to_le16(0xffff);
1326 }
1327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 sb->sb_csum = calc_sb_1_csum(sb);
1329}
1330
1331
Adrian Bunk75c96f82005-05-05 16:16:09 -07001332static struct super_type super_types[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 [0] = {
1334 .name = "0.90.0",
1335 .owner = THIS_MODULE,
1336 .load_super = super_90_load,
1337 .validate_super = super_90_validate,
1338 .sync_super = super_90_sync,
1339 },
1340 [1] = {
1341 .name = "md-1",
1342 .owner = THIS_MODULE,
1343 .load_super = super_1_load,
1344 .validate_super = super_1_validate,
1345 .sync_super = super_1_sync,
1346 },
1347};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
1349static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1350{
NeilBrown7dd5e7c32007-02-28 20:11:35 -08001351 struct list_head *tmp, *tmp2;
1352 mdk_rdev_t *rdev, *rdev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
NeilBrownd089c6a2008-02-06 01:39:59 -08001354 rdev_for_each(rdev, tmp, mddev1)
1355 rdev_for_each(rdev2, tmp2, mddev2)
NeilBrown7dd5e7c32007-02-28 20:11:35 -08001356 if (rdev->bdev->bd_contains ==
1357 rdev2->bdev->bd_contains)
1358 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360 return 0;
1361}
1362
1363static LIST_HEAD(pending_raid_disks);
1364
1365static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1366{
NeilBrown7dd5e7c32007-02-28 20:11:35 -08001367 char b[BDEVNAME_SIZE];
NeilBrownf637b9f2005-11-08 21:39:37 -08001368 struct kobject *ko;
Neil Brown1edf80d2006-01-12 01:05:23 -08001369 char *s;
NeilBrown5e55e2f2007-03-26 21:32:14 -08001370 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
1372 if (rdev->mddev) {
1373 MD_BUG();
1374 return -EINVAL;
1375 }
Dan Williams11e2ede2008-04-30 00:52:32 -07001376
1377 /* prevent duplicates */
1378 if (find_rdev(mddev, rdev->bdev->bd_dev))
1379 return -EEXIST;
1380
NeilBrown2bf071b2006-01-06 00:20:55 -08001381 /* make sure rdev->size exceeds mddev->size */
1382 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
NeilBrowna778b732007-05-23 13:58:10 -07001383 if (mddev->pers) {
1384 /* Cannot change size, so fail
1385 * If mddev->level <= 0, then we don't care
1386 * about aligning sizes (e.g. linear)
1387 */
1388 if (mddev->level > 0)
1389 return -ENOSPC;
1390 } else
NeilBrown2bf071b2006-01-06 00:20:55 -08001391 mddev->size = rdev->size;
1392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
1394 /* Verify rdev->desc_nr is unique.
1395 * If it is -1, assign a free number, else
1396 * check number is not in use
1397 */
1398 if (rdev->desc_nr < 0) {
1399 int choice = 0;
1400 if (mddev->pers) choice = mddev->raid_disks;
1401 while (find_rdev_nr(mddev, choice))
1402 choice++;
1403 rdev->desc_nr = choice;
1404 } else {
1405 if (find_rdev_nr(mddev, rdev->desc_nr))
1406 return -EBUSY;
1407 }
NeilBrown19133a42005-11-08 21:39:35 -08001408 bdevname(rdev->bdev,b);
Greg Kroah-Hartman649316b2007-12-17 23:05:35 -07001409 while ( (s=strchr(b, '/')) != NULL)
Neil Brown1edf80d2006-01-12 01:05:23 -08001410 *s = '!';
Greg Kroah-Hartman649316b2007-12-17 23:05:35 -07001411
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 rdev->mddev = mddev;
NeilBrown19133a42005-11-08 21:39:35 -08001413 printk(KERN_INFO "md: bind<%s>\n", b);
NeilBrown86e6ffd2005-11-08 21:39:24 -08001414
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -07001415 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
NeilBrown5e55e2f2007-03-26 21:32:14 -08001416 goto fail;
NeilBrown86e6ffd2005-11-08 21:39:24 -08001417
NeilBrownf637b9f2005-11-08 21:39:37 -08001418 if (rdev->bdev->bd_part)
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001419 ko = &rdev->bdev->bd_part->dev.kobj;
NeilBrownf637b9f2005-11-08 21:39:37 -08001420 else
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001421 ko = &rdev->bdev->bd_disk->dev.kobj;
NeilBrown5e55e2f2007-03-26 21:32:14 -08001422 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1423 kobject_del(&rdev->kobj);
1424 goto fail;
1425 }
1426 list_add(&rdev->same_set, &mddev->disks);
NeilBrownc5d79ad2008-02-06 01:39:54 -08001427 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 return 0;
NeilBrown5e55e2f2007-03-26 21:32:14 -08001429
1430 fail:
1431 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1432 b, mdname(mddev));
1433 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434}
1435
NeilBrown177a99b2008-02-06 01:39:56 -08001436static void md_delayed_delete(struct work_struct *ws)
NeilBrown5792a282007-04-04 19:08:18 -07001437{
1438 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1439 kobject_del(&rdev->kobj);
NeilBrown177a99b2008-02-06 01:39:56 -08001440 kobject_put(&rdev->kobj);
NeilBrown5792a282007-04-04 19:08:18 -07001441}
1442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1444{
1445 char b[BDEVNAME_SIZE];
1446 if (!rdev->mddev) {
1447 MD_BUG();
1448 return;
1449 }
Jun'ichi Nomura5463c792006-03-27 01:17:58 -08001450 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 list_del_init(&rdev->same_set);
1452 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1453 rdev->mddev = NULL;
NeilBrown86e6ffd2005-11-08 21:39:24 -08001454 sysfs_remove_link(&rdev->kobj, "block");
NeilBrown5792a282007-04-04 19:08:18 -07001455
1456 /* We need to delay this, otherwise we can deadlock when
1457 * writing to 'remove' to "dev/state"
1458 */
NeilBrown177a99b2008-02-06 01:39:56 -08001459 INIT_WORK(&rdev->del_work, md_delayed_delete);
1460 kobject_get(&rdev->kobj);
NeilBrown5792a282007-04-04 19:08:18 -07001461 schedule_work(&rdev->del_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462}
1463
1464/*
1465 * prevent the device from being mounted, repartitioned or
1466 * otherwise reused by a RAID array (or any other kernel
1467 * subsystem), by bd_claiming the device.
1468 */
NeilBrownc5d79ad2008-02-06 01:39:54 -08001469static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470{
1471 int err = 0;
1472 struct block_device *bdev;
1473 char b[BDEVNAME_SIZE];
1474
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001475 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (IS_ERR(bdev)) {
1477 printk(KERN_ERR "md: could not open %s.\n",
1478 __bdevname(dev, b));
1479 return PTR_ERR(bdev);
1480 }
NeilBrownc5d79ad2008-02-06 01:39:54 -08001481 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 if (err) {
1483 printk(KERN_ERR "md: could not bd_claim %s.\n",
1484 bdevname(bdev, b));
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001485 blkdev_put(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return err;
1487 }
NeilBrownc5d79ad2008-02-06 01:39:54 -08001488 if (!shared)
1489 set_bit(AllReserved, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 rdev->bdev = bdev;
1491 return err;
1492}
1493
1494static void unlock_rdev(mdk_rdev_t *rdev)
1495{
1496 struct block_device *bdev = rdev->bdev;
1497 rdev->bdev = NULL;
1498 if (!bdev)
1499 MD_BUG();
1500 bd_release(bdev);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001501 blkdev_put(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502}
1503
1504void md_autodetect_dev(dev_t dev);
1505
1506static void export_rdev(mdk_rdev_t * rdev)
1507{
1508 char b[BDEVNAME_SIZE];
1509 printk(KERN_INFO "md: export_rdev(%s)\n",
1510 bdevname(rdev->bdev,b));
1511 if (rdev->mddev)
1512 MD_BUG();
1513 free_disk_sb(rdev);
1514 list_del_init(&rdev->same_set);
1515#ifndef MODULE
NeilBrownd0fae182008-03-04 14:29:31 -08001516 if (test_bit(AutoDetected, &rdev->flags))
1517 md_autodetect_dev(rdev->bdev->bd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518#endif
1519 unlock_rdev(rdev);
NeilBrown86e6ffd2005-11-08 21:39:24 -08001520 kobject_put(&rdev->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521}
1522
1523static void kick_rdev_from_array(mdk_rdev_t * rdev)
1524{
1525 unbind_rdev_from_array(rdev);
1526 export_rdev(rdev);
1527}
1528
1529static void export_array(mddev_t *mddev)
1530{
1531 struct list_head *tmp;
1532 mdk_rdev_t *rdev;
1533
NeilBrownd089c6a2008-02-06 01:39:59 -08001534 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 if (!rdev->mddev) {
1536 MD_BUG();
1537 continue;
1538 }
1539 kick_rdev_from_array(rdev);
1540 }
1541 if (!list_empty(&mddev->disks))
1542 MD_BUG();
1543 mddev->raid_disks = 0;
1544 mddev->major_version = 0;
1545}
1546
1547static void print_desc(mdp_disk_t *desc)
1548{
1549 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1550 desc->major,desc->minor,desc->raid_disk,desc->state);
1551}
1552
1553static void print_sb(mdp_super_t *sb)
1554{
1555 int i;
1556
1557 printk(KERN_INFO
1558 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1559 sb->major_version, sb->minor_version, sb->patch_version,
1560 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1561 sb->ctime);
1562 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1563 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1564 sb->md_minor, sb->layout, sb->chunk_size);
1565 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1566 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1567 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1568 sb->failed_disks, sb->spare_disks,
1569 sb->sb_csum, (unsigned long)sb->events_lo);
1570
1571 printk(KERN_INFO);
1572 for (i = 0; i < MD_SB_DISKS; i++) {
1573 mdp_disk_t *desc;
1574
1575 desc = sb->disks + i;
1576 if (desc->number || desc->major || desc->minor ||
1577 desc->raid_disk || (desc->state && (desc->state != 4))) {
1578 printk(" D %2d: ", i);
1579 print_desc(desc);
1580 }
1581 }
1582 printk(KERN_INFO "md: THIS: ");
1583 print_desc(&sb->this_disk);
1584
1585}
1586
1587static void print_rdev(mdk_rdev_t *rdev)
1588{
1589 char b[BDEVNAME_SIZE];
1590 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1591 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
NeilBrownb2d444d2005-11-08 21:39:31 -08001592 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1593 rdev->desc_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (rdev->sb_loaded) {
1595 printk(KERN_INFO "md: rdev superblock:\n");
1596 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1597 } else
1598 printk(KERN_INFO "md: no rdev superblock!\n");
1599}
1600
Adrian Bunk5e563412006-06-26 00:27:42 -07001601static void md_print_devices(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602{
1603 struct list_head *tmp, *tmp2;
1604 mdk_rdev_t *rdev;
1605 mddev_t *mddev;
1606 char b[BDEVNAME_SIZE];
1607
1608 printk("\n");
1609 printk("md: **********************************\n");
1610 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1611 printk("md: **********************************\n");
NeilBrown29ac4aa2008-02-06 01:39:58 -08001612 for_each_mddev(mddev, tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
NeilBrown32a76272005-06-21 17:17:14 -07001614 if (mddev->bitmap)
1615 bitmap_print_sb(mddev->bitmap);
1616 else
1617 printk("%s: ", mdname(mddev));
NeilBrownd089c6a2008-02-06 01:39:59 -08001618 rdev_for_each(rdev, tmp2, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 printk("<%s>", bdevname(rdev->bdev,b));
1620 printk("\n");
1621
NeilBrownd089c6a2008-02-06 01:39:59 -08001622 rdev_for_each(rdev, tmp2, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 print_rdev(rdev);
1624 }
1625 printk("md: **********************************\n");
1626 printk("\n");
1627}
1628
1629
NeilBrown42543762006-06-26 00:27:57 -07001630static void sync_sbs(mddev_t * mddev, int nospares)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631{
NeilBrown42543762006-06-26 00:27:57 -07001632 /* Update each superblock (in-memory image), but
1633 * if we are allowed to, skip spares which already
1634 * have the right event counter, or have one earlier
1635 * (which would mean they aren't being marked as dirty
1636 * with the rest of the array)
1637 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 mdk_rdev_t *rdev;
1639 struct list_head *tmp;
1640
NeilBrownd089c6a2008-02-06 01:39:59 -08001641 rdev_for_each(rdev, tmp, mddev) {
NeilBrown42543762006-06-26 00:27:57 -07001642 if (rdev->sb_events == mddev->events ||
1643 (nospares &&
1644 rdev->raid_disk < 0 &&
1645 (rdev->sb_events&1)==0 &&
1646 rdev->sb_events+1 == mddev->events)) {
1647 /* Don't update this superblock */
1648 rdev->sb_loaded = 2;
1649 } else {
1650 super_types[mddev->major_version].
1651 sync_super(mddev, rdev);
1652 rdev->sb_loaded = 1;
1653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 }
1655}
1656
NeilBrown850b2b42006-10-03 01:15:46 -07001657static void md_update_sb(mddev_t * mddev, int force_change)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 struct list_head *tmp;
1660 mdk_rdev_t *rdev;
NeilBrown06d91a52005-06-21 17:17:12 -07001661 int sync_req;
NeilBrown42543762006-06-26 00:27:57 -07001662 int nospares = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
NeilBrown8377bc82008-04-30 00:52:28 -07001664 if (mddev->external)
1665 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666repeat:
NeilBrowna9701a32005-11-08 21:39:34 -08001667 spin_lock_irq(&mddev->write_lock);
NeilBrown84692192006-08-27 01:23:49 -07001668
NeilBrown850b2b42006-10-03 01:15:46 -07001669 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1670 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1671 force_change = 1;
1672 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1673 /* just a clean<-> dirty transition, possibly leave spares alone,
1674 * though if events isn't the right even/odd, we will have to do
1675 * spares after all
1676 */
1677 nospares = 1;
1678 if (force_change)
1679 nospares = 0;
1680 if (mddev->degraded)
NeilBrown84692192006-08-27 01:23:49 -07001681 /* If the array is degraded, then skipping spares is both
1682 * dangerous and fairly pointless.
1683 * Dangerous because a device that was removed from the array
1684 * might have a event_count that still looks up-to-date,
1685 * so it can be re-added without a resync.
1686 * Pointless because if there are any spares to skip,
1687 * then a recovery will happen and soon that array won't
1688 * be degraded any more and the spare can go back to sleep then.
1689 */
NeilBrown850b2b42006-10-03 01:15:46 -07001690 nospares = 0;
NeilBrown84692192006-08-27 01:23:49 -07001691
NeilBrown06d91a52005-06-21 17:17:12 -07001692 sync_req = mddev->in_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 mddev->utime = get_seconds();
NeilBrown42543762006-06-26 00:27:57 -07001694
1695 /* If this is just a dirty<->clean transition, and the array is clean
1696 * and 'events' is odd, we can roll back to the previous clean state */
NeilBrown850b2b42006-10-03 01:15:46 -07001697 if (nospares
NeilBrown42543762006-06-26 00:27:57 -07001698 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
NeilBrown1031be72007-01-26 00:57:02 -08001699 && (mddev->events & 1)
1700 && mddev->events != 1)
NeilBrown42543762006-06-26 00:27:57 -07001701 mddev->events--;
1702 else {
1703 /* otherwise we have to go forward and ... */
1704 mddev->events ++;
1705 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1706 /* .. if the array isn't clean, insist on an odd 'events' */
1707 if ((mddev->events&1)==0) {
1708 mddev->events++;
1709 nospares = 0;
1710 }
1711 } else {
1712 /* otherwise insist on an even 'events' (for clean states) */
1713 if ((mddev->events&1)) {
1714 mddev->events++;
1715 nospares = 0;
1716 }
1717 }
1718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
1720 if (!mddev->events) {
1721 /*
1722 * oops, this 64-bit counter should never wrap.
1723 * Either we are in around ~1 trillion A.C., assuming
1724 * 1 reboot per second, or we have a bug:
1725 */
1726 MD_BUG();
1727 mddev->events --;
1728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
1730 /*
1731 * do not write anything to disk if using
1732 * nonpersistent superblocks
1733 */
NeilBrown06d91a52005-06-21 17:17:12 -07001734 if (!mddev->persistent) {
NeilBrowne6910632008-02-06 01:39:51 -08001735 if (!mddev->external)
1736 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1737
NeilBrowna9701a32005-11-08 21:39:34 -08001738 spin_unlock_irq(&mddev->write_lock);
NeilBrown3d310eb2005-06-21 17:17:26 -07001739 wake_up(&mddev->sb_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 return;
NeilBrown06d91a52005-06-21 17:17:12 -07001741 }
NeilBrowne6910632008-02-06 01:39:51 -08001742 sync_sbs(mddev, nospares);
NeilBrowna9701a32005-11-08 21:39:34 -08001743 spin_unlock_irq(&mddev->write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
1745 dprintk(KERN_INFO
1746 "md: updating %s RAID superblock on device (in sync %d)\n",
1747 mdname(mddev),mddev->in_sync);
1748
NeilBrown4ad13662007-07-17 04:06:13 -07001749 bitmap_update_sb(mddev->bitmap);
NeilBrownd089c6a2008-02-06 01:39:59 -08001750 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 char b[BDEVNAME_SIZE];
1752 dprintk(KERN_INFO "md: ");
NeilBrown42543762006-06-26 00:27:57 -07001753 if (rdev->sb_loaded != 1)
1754 continue; /* no noise on spare devices */
NeilBrownb2d444d2005-11-08 21:39:31 -08001755 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 dprintk("(skipping faulty ");
1757
1758 dprintk("%s ", bdevname(rdev->bdev,b));
NeilBrownb2d444d2005-11-08 21:39:31 -08001759 if (!test_bit(Faulty, &rdev->flags)) {
NeilBrown7bfa19f2005-06-21 17:17:28 -07001760 md_super_write(mddev,rdev,
NeilBrown0002b272005-09-09 16:23:53 -07001761 rdev->sb_offset<<1, rdev->sb_size,
NeilBrown7bfa19f2005-06-21 17:17:28 -07001762 rdev->sb_page);
1763 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1764 bdevname(rdev->bdev,b),
1765 (unsigned long long)rdev->sb_offset);
NeilBrown42543762006-06-26 00:27:57 -07001766 rdev->sb_events = mddev->events;
NeilBrown7bfa19f2005-06-21 17:17:28 -07001767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 } else
1769 dprintk(")\n");
NeilBrown7bfa19f2005-06-21 17:17:28 -07001770 if (mddev->level == LEVEL_MULTIPATH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 /* only need to write one superblock... */
1772 break;
1773 }
NeilBrowna9701a32005-11-08 21:39:34 -08001774 md_super_wait(mddev);
NeilBrown850b2b42006-10-03 01:15:46 -07001775 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
NeilBrown7bfa19f2005-06-21 17:17:28 -07001776
NeilBrowna9701a32005-11-08 21:39:34 -08001777 spin_lock_irq(&mddev->write_lock);
NeilBrown850b2b42006-10-03 01:15:46 -07001778 if (mddev->in_sync != sync_req ||
1779 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
NeilBrown06d91a52005-06-21 17:17:12 -07001780 /* have to write it out again */
NeilBrowna9701a32005-11-08 21:39:34 -08001781 spin_unlock_irq(&mddev->write_lock);
NeilBrown06d91a52005-06-21 17:17:12 -07001782 goto repeat;
1783 }
NeilBrown850b2b42006-10-03 01:15:46 -07001784 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrowna9701a32005-11-08 21:39:34 -08001785 spin_unlock_irq(&mddev->write_lock);
NeilBrown3d310eb2005-06-21 17:17:26 -07001786 wake_up(&mddev->sb_wait);
NeilBrown06d91a52005-06-21 17:17:12 -07001787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788}
1789
NeilBrownbce74da2006-01-06 00:20:41 -08001790/* words written to sysfs files may, or my not, be \n terminated.
1791 * We want to accept with case. For this we use cmd_match.
1792 */
1793static int cmd_match(const char *cmd, const char *str)
1794{
1795 /* See if cmd, written into a sysfs file, matches
1796 * str. They must either be the same, or cmd can
1797 * have a trailing newline
1798 */
1799 while (*cmd && *str && *cmd == *str) {
1800 cmd++;
1801 str++;
1802 }
1803 if (*cmd == '\n')
1804 cmd++;
1805 if (*str || *cmd)
1806 return 0;
1807 return 1;
1808}
1809
NeilBrown86e6ffd2005-11-08 21:39:24 -08001810struct rdev_sysfs_entry {
1811 struct attribute attr;
1812 ssize_t (*show)(mdk_rdev_t *, char *);
1813 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1814};
1815
1816static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08001817state_show(mdk_rdev_t *rdev, char *page)
NeilBrown86e6ffd2005-11-08 21:39:24 -08001818{
1819 char *sep = "";
NeilBrown20a49ff2008-02-06 01:39:57 -08001820 size_t len = 0;
NeilBrown86e6ffd2005-11-08 21:39:24 -08001821
NeilBrownb2d444d2005-11-08 21:39:31 -08001822 if (test_bit(Faulty, &rdev->flags)) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08001823 len+= sprintf(page+len, "%sfaulty",sep);
1824 sep = ",";
1825 }
NeilBrownb2d444d2005-11-08 21:39:31 -08001826 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08001827 len += sprintf(page+len, "%sin_sync",sep);
1828 sep = ",";
1829 }
NeilBrownf6556752006-06-26 00:28:01 -07001830 if (test_bit(WriteMostly, &rdev->flags)) {
1831 len += sprintf(page+len, "%swrite_mostly",sep);
1832 sep = ",";
1833 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07001834 if (test_bit(Blocked, &rdev->flags)) {
1835 len += sprintf(page+len, "%sblocked", sep);
1836 sep = ",";
1837 }
NeilBrownb2d444d2005-11-08 21:39:31 -08001838 if (!test_bit(Faulty, &rdev->flags) &&
1839 !test_bit(In_sync, &rdev->flags)) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08001840 len += sprintf(page+len, "%sspare", sep);
1841 sep = ",";
1842 }
1843 return len+sprintf(page+len, "\n");
1844}
1845
NeilBrown45dc2de2006-06-26 00:27:58 -07001846static ssize_t
1847state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1848{
1849 /* can write
1850 * faulty - simulates and error
1851 * remove - disconnects the device
NeilBrownf6556752006-06-26 00:28:01 -07001852 * writemostly - sets write_mostly
1853 * -writemostly - clears write_mostly
Dan Williams6bfe0b42008-04-30 00:52:32 -07001854 * blocked - sets the Blocked flag
1855 * -blocked - clears the Blocked flag
NeilBrown45dc2de2006-06-26 00:27:58 -07001856 */
1857 int err = -EINVAL;
1858 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1859 md_error(rdev->mddev, rdev);
1860 err = 0;
1861 } else if (cmd_match(buf, "remove")) {
1862 if (rdev->raid_disk >= 0)
1863 err = -EBUSY;
1864 else {
1865 mddev_t *mddev = rdev->mddev;
1866 kick_rdev_from_array(rdev);
NeilBrown3f9d7b02006-12-22 01:11:41 -08001867 if (mddev->pers)
1868 md_update_sb(mddev, 1);
NeilBrown45dc2de2006-06-26 00:27:58 -07001869 md_new_event(mddev);
1870 err = 0;
1871 }
NeilBrownf6556752006-06-26 00:28:01 -07001872 } else if (cmd_match(buf, "writemostly")) {
1873 set_bit(WriteMostly, &rdev->flags);
1874 err = 0;
1875 } else if (cmd_match(buf, "-writemostly")) {
1876 clear_bit(WriteMostly, &rdev->flags);
1877 err = 0;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001878 } else if (cmd_match(buf, "blocked")) {
1879 set_bit(Blocked, &rdev->flags);
1880 err = 0;
1881 } else if (cmd_match(buf, "-blocked")) {
1882 clear_bit(Blocked, &rdev->flags);
1883 wake_up(&rdev->blocked_wait);
1884 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1885 md_wakeup_thread(rdev->mddev->thread);
1886
1887 err = 0;
NeilBrown45dc2de2006-06-26 00:27:58 -07001888 }
1889 return err ? err : len;
1890}
NeilBrown80ca3a42006-07-10 04:44:18 -07001891static struct rdev_sysfs_entry rdev_state =
1892__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
NeilBrown86e6ffd2005-11-08 21:39:24 -08001893
1894static ssize_t
NeilBrown4dbcdc72006-01-06 00:20:52 -08001895errors_show(mdk_rdev_t *rdev, char *page)
1896{
1897 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1898}
1899
1900static ssize_t
1901errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1902{
1903 char *e;
1904 unsigned long n = simple_strtoul(buf, &e, 10);
1905 if (*buf && (*e == 0 || *e == '\n')) {
1906 atomic_set(&rdev->corrected_errors, n);
1907 return len;
1908 }
1909 return -EINVAL;
1910}
1911static struct rdev_sysfs_entry rdev_errors =
NeilBrown80ca3a42006-07-10 04:44:18 -07001912__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
NeilBrown4dbcdc72006-01-06 00:20:52 -08001913
NeilBrown014236d2006-01-06 00:20:55 -08001914static ssize_t
1915slot_show(mdk_rdev_t *rdev, char *page)
1916{
1917 if (rdev->raid_disk < 0)
1918 return sprintf(page, "none\n");
1919 else
1920 return sprintf(page, "%d\n", rdev->raid_disk);
1921}
1922
1923static ssize_t
1924slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1925{
1926 char *e;
NeilBrownc303da62008-02-06 01:39:51 -08001927 int err;
1928 char nm[20];
NeilBrown014236d2006-01-06 00:20:55 -08001929 int slot = simple_strtoul(buf, &e, 10);
1930 if (strncmp(buf, "none", 4)==0)
1931 slot = -1;
1932 else if (e==buf || (*e && *e!= '\n'))
1933 return -EINVAL;
NeilBrownc303da62008-02-06 01:39:51 -08001934 if (rdev->mddev->pers) {
1935 /* Setting 'slot' on an active array requires also
1936 * updating the 'rd%d' link, and communicating
1937 * with the personality with ->hot_*_disk.
1938 * For now we only support removing
1939 * failed/spare devices. This normally happens automatically,
1940 * but not when the metadata is externally managed.
1941 */
1942 if (slot != -1)
1943 return -EBUSY;
1944 if (rdev->raid_disk == -1)
1945 return -EEXIST;
1946 /* personality does all needed checks */
1947 if (rdev->mddev->pers->hot_add_disk == NULL)
1948 return -EINVAL;
1949 err = rdev->mddev->pers->
1950 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1951 if (err)
1952 return err;
1953 sprintf(nm, "rd%d", rdev->raid_disk);
1954 sysfs_remove_link(&rdev->mddev->kobj, nm);
1955 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1956 md_wakeup_thread(rdev->mddev->thread);
1957 } else {
1958 if (slot >= rdev->mddev->raid_disks)
1959 return -ENOSPC;
1960 rdev->raid_disk = slot;
1961 /* assume it is working */
NeilBrownc5d79ad2008-02-06 01:39:54 -08001962 clear_bit(Faulty, &rdev->flags);
1963 clear_bit(WriteMostly, &rdev->flags);
NeilBrownc303da62008-02-06 01:39:51 -08001964 set_bit(In_sync, &rdev->flags);
1965 }
NeilBrown014236d2006-01-06 00:20:55 -08001966 return len;
1967}
1968
1969
1970static struct rdev_sysfs_entry rdev_slot =
NeilBrown80ca3a42006-07-10 04:44:18 -07001971__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
NeilBrown014236d2006-01-06 00:20:55 -08001972
NeilBrown93c8cad2006-01-06 00:20:56 -08001973static ssize_t
1974offset_show(mdk_rdev_t *rdev, char *page)
1975{
Andrew Morton6961ece2006-01-06 00:20:59 -08001976 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
NeilBrown93c8cad2006-01-06 00:20:56 -08001977}
1978
1979static ssize_t
1980offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1981{
1982 char *e;
1983 unsigned long long offset = simple_strtoull(buf, &e, 10);
1984 if (e==buf || (*e && *e != '\n'))
1985 return -EINVAL;
1986 if (rdev->mddev->pers)
1987 return -EBUSY;
NeilBrownc5d79ad2008-02-06 01:39:54 -08001988 if (rdev->size && rdev->mddev->external)
1989 /* Must set offset before size, so overlap checks
1990 * can be sane */
1991 return -EBUSY;
NeilBrown93c8cad2006-01-06 00:20:56 -08001992 rdev->data_offset = offset;
1993 return len;
1994}
1995
1996static struct rdev_sysfs_entry rdev_offset =
NeilBrown80ca3a42006-07-10 04:44:18 -07001997__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
NeilBrown93c8cad2006-01-06 00:20:56 -08001998
NeilBrown83303b62006-01-06 00:21:06 -08001999static ssize_t
2000rdev_size_show(mdk_rdev_t *rdev, char *page)
2001{
2002 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2003}
2004
NeilBrownc5d79ad2008-02-06 01:39:54 -08002005static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2006{
2007 /* check if two start/length pairs overlap */
2008 if (s1+l1 <= s2)
2009 return 0;
2010 if (s2+l2 <= s1)
2011 return 0;
2012 return 1;
2013}
2014
NeilBrown83303b62006-01-06 00:21:06 -08002015static ssize_t
2016rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2017{
2018 char *e;
2019 unsigned long long size = simple_strtoull(buf, &e, 10);
NeilBrownc5d79ad2008-02-06 01:39:54 -08002020 unsigned long long oldsize = rdev->size;
NeilBrown27c529b2008-03-04 14:29:33 -08002021 mddev_t *my_mddev = rdev->mddev;
2022
NeilBrown83303b62006-01-06 00:21:06 -08002023 if (e==buf || (*e && *e != '\n'))
2024 return -EINVAL;
NeilBrown27c529b2008-03-04 14:29:33 -08002025 if (my_mddev->pers)
NeilBrown83303b62006-01-06 00:21:06 -08002026 return -EBUSY;
2027 rdev->size = size;
NeilBrownc5d79ad2008-02-06 01:39:54 -08002028 if (size > oldsize && rdev->mddev->external) {
2029 /* need to check that all other rdevs with the same ->bdev
2030 * do not overlap. We need to unlock the mddev to avoid
2031 * a deadlock. We have already changed rdev->size, and if
2032 * we have to change it back, we will have the lock again.
2033 */
2034 mddev_t *mddev;
2035 int overlap = 0;
2036 struct list_head *tmp, *tmp2;
2037
NeilBrown27c529b2008-03-04 14:29:33 -08002038 mddev_unlock(my_mddev);
NeilBrown29ac4aa2008-02-06 01:39:58 -08002039 for_each_mddev(mddev, tmp) {
NeilBrownc5d79ad2008-02-06 01:39:54 -08002040 mdk_rdev_t *rdev2;
2041
2042 mddev_lock(mddev);
NeilBrownd089c6a2008-02-06 01:39:59 -08002043 rdev_for_each(rdev2, tmp2, mddev)
NeilBrownc5d79ad2008-02-06 01:39:54 -08002044 if (test_bit(AllReserved, &rdev2->flags) ||
2045 (rdev->bdev == rdev2->bdev &&
2046 rdev != rdev2 &&
2047 overlaps(rdev->data_offset, rdev->size,
2048 rdev2->data_offset, rdev2->size))) {
2049 overlap = 1;
2050 break;
2051 }
2052 mddev_unlock(mddev);
2053 if (overlap) {
2054 mddev_put(mddev);
2055 break;
2056 }
2057 }
NeilBrown27c529b2008-03-04 14:29:33 -08002058 mddev_lock(my_mddev);
NeilBrownc5d79ad2008-02-06 01:39:54 -08002059 if (overlap) {
2060 /* Someone else could have slipped in a size
2061 * change here, but doing so is just silly.
2062 * We put oldsize back because we *know* it is
2063 * safe, and trust userspace not to race with
2064 * itself
2065 */
2066 rdev->size = oldsize;
2067 return -EBUSY;
2068 }
2069 }
NeilBrown27c529b2008-03-04 14:29:33 -08002070 if (size < my_mddev->size || my_mddev->size == 0)
2071 my_mddev->size = size;
NeilBrown83303b62006-01-06 00:21:06 -08002072 return len;
2073}
2074
2075static struct rdev_sysfs_entry rdev_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07002076__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
NeilBrown83303b62006-01-06 00:21:06 -08002077
NeilBrown86e6ffd2005-11-08 21:39:24 -08002078static struct attribute *rdev_default_attrs[] = {
2079 &rdev_state.attr,
NeilBrown4dbcdc72006-01-06 00:20:52 -08002080 &rdev_errors.attr,
NeilBrown014236d2006-01-06 00:20:55 -08002081 &rdev_slot.attr,
NeilBrown93c8cad2006-01-06 00:20:56 -08002082 &rdev_offset.attr,
NeilBrown83303b62006-01-06 00:21:06 -08002083 &rdev_size.attr,
NeilBrown86e6ffd2005-11-08 21:39:24 -08002084 NULL,
2085};
2086static ssize_t
2087rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2088{
2089 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2090 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
NeilBrown27c529b2008-03-04 14:29:33 -08002091 mddev_t *mddev = rdev->mddev;
2092 ssize_t rv;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002093
2094 if (!entry->show)
2095 return -EIO;
NeilBrown27c529b2008-03-04 14:29:33 -08002096
2097 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2098 if (!rv) {
2099 if (rdev->mddev == NULL)
2100 rv = -EBUSY;
2101 else
2102 rv = entry->show(rdev, page);
2103 mddev_unlock(mddev);
2104 }
2105 return rv;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002106}
2107
2108static ssize_t
2109rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2110 const char *page, size_t length)
2111{
2112 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2113 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
NeilBrown27c529b2008-03-04 14:29:33 -08002114 ssize_t rv;
2115 mddev_t *mddev = rdev->mddev;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002116
2117 if (!entry->store)
2118 return -EIO;
NeilBrown67463ac2006-07-10 04:44:19 -07002119 if (!capable(CAP_SYS_ADMIN))
2120 return -EACCES;
NeilBrown27c529b2008-03-04 14:29:33 -08002121 rv = mddev ? mddev_lock(mddev): -EBUSY;
NeilBrownca388052008-02-06 01:39:55 -08002122 if (!rv) {
NeilBrown27c529b2008-03-04 14:29:33 -08002123 if (rdev->mddev == NULL)
2124 rv = -EBUSY;
2125 else
2126 rv = entry->store(rdev, page, length);
Dan Williams6a518302008-04-30 00:52:28 -07002127 mddev_unlock(mddev);
NeilBrownca388052008-02-06 01:39:55 -08002128 }
2129 return rv;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002130}
2131
2132static void rdev_free(struct kobject *ko)
2133{
2134 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2135 kfree(rdev);
2136}
2137static struct sysfs_ops rdev_sysfs_ops = {
2138 .show = rdev_attr_show,
2139 .store = rdev_attr_store,
2140};
2141static struct kobj_type rdev_ktype = {
2142 .release = rdev_free,
2143 .sysfs_ops = &rdev_sysfs_ops,
2144 .default_attrs = rdev_default_attrs,
2145};
2146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147/*
2148 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2149 *
2150 * mark the device faulty if:
2151 *
2152 * - the device is nonexistent (zero size)
2153 * - the device has no valid superblock
2154 *
2155 * a faulty rdev _never_ has rdev->sb set.
2156 */
2157static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2158{
2159 char b[BDEVNAME_SIZE];
2160 int err;
2161 mdk_rdev_t *rdev;
2162 sector_t size;
2163
NeilBrown9ffae0c2006-01-06 00:20:32 -08002164 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 if (!rdev) {
2166 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2167 return ERR_PTR(-ENOMEM);
2168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169
2170 if ((err = alloc_disk_sb(rdev)))
2171 goto abort_free;
2172
NeilBrownc5d79ad2008-02-06 01:39:54 -08002173 err = lock_rdev(rdev, newdev, super_format == -2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 if (err)
2175 goto abort_free;
2176
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -07002177 kobject_init(&rdev->kobj, &rdev_ktype);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002178
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 rdev->desc_nr = -1;
NeilBrown2b6e8452006-10-28 10:38:30 -07002180 rdev->saved_raid_disk = -1;
NeilBrown3f9d7b02006-12-22 01:11:41 -08002181 rdev->raid_disk = -1;
NeilBrownb2d444d2005-11-08 21:39:31 -08002182 rdev->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 rdev->data_offset = 0;
NeilBrown42543762006-06-26 00:27:57 -07002184 rdev->sb_events = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 atomic_set(&rdev->nr_pending, 0);
NeilBrownba22dcb2005-11-08 21:39:31 -08002186 atomic_set(&rdev->read_errors, 0);
NeilBrown4dbcdc72006-01-06 00:20:52 -08002187 atomic_set(&rdev->corrected_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188
2189 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2190 if (!size) {
2191 printk(KERN_WARNING
2192 "md: %s has zero or unknown size, marking faulty!\n",
2193 bdevname(rdev->bdev,b));
2194 err = -EINVAL;
2195 goto abort_free;
2196 }
2197
2198 if (super_format >= 0) {
2199 err = super_types[super_format].
2200 load_super(rdev, NULL, super_minor);
2201 if (err == -EINVAL) {
NeilBrowndf968c42007-07-17 04:06:11 -07002202 printk(KERN_WARNING
2203 "md: %s does not have a valid v%d.%d "
2204 "superblock, not importing!\n",
2205 bdevname(rdev->bdev,b),
2206 super_format, super_minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 goto abort_free;
2208 }
2209 if (err < 0) {
2210 printk(KERN_WARNING
2211 "md: could not read %s's sb, not importing!\n",
2212 bdevname(rdev->bdev,b));
2213 goto abort_free;
2214 }
2215 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07002216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 INIT_LIST_HEAD(&rdev->same_set);
Dan Williams6bfe0b42008-04-30 00:52:32 -07002218 init_waitqueue_head(&rdev->blocked_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219
2220 return rdev;
2221
2222abort_free:
2223 if (rdev->sb_page) {
2224 if (rdev->bdev)
2225 unlock_rdev(rdev);
2226 free_disk_sb(rdev);
2227 }
2228 kfree(rdev);
2229 return ERR_PTR(err);
2230}
2231
2232/*
2233 * Check a full RAID array for plausibility
2234 */
2235
2236
NeilBrowna757e642005-04-16 15:26:42 -07002237static void analyze_sbs(mddev_t * mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238{
2239 int i;
2240 struct list_head *tmp;
2241 mdk_rdev_t *rdev, *freshest;
2242 char b[BDEVNAME_SIZE];
2243
2244 freshest = NULL;
NeilBrownd089c6a2008-02-06 01:39:59 -08002245 rdev_for_each(rdev, tmp, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 switch (super_types[mddev->major_version].
2247 load_super(rdev, freshest, mddev->minor_version)) {
2248 case 1:
2249 freshest = rdev;
2250 break;
2251 case 0:
2252 break;
2253 default:
2254 printk( KERN_ERR \
2255 "md: fatal superblock inconsistency in %s"
2256 " -- removing from array\n",
2257 bdevname(rdev->bdev,b));
2258 kick_rdev_from_array(rdev);
2259 }
2260
2261
2262 super_types[mddev->major_version].
2263 validate_super(mddev, freshest);
2264
2265 i = 0;
NeilBrownd089c6a2008-02-06 01:39:59 -08002266 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 if (rdev != freshest)
2268 if (super_types[mddev->major_version].
2269 validate_super(mddev, rdev)) {
2270 printk(KERN_WARNING "md: kicking non-fresh %s"
2271 " from array!\n",
2272 bdevname(rdev->bdev,b));
2273 kick_rdev_from_array(rdev);
2274 continue;
2275 }
2276 if (mddev->level == LEVEL_MULTIPATH) {
2277 rdev->desc_nr = i++;
2278 rdev->raid_disk = rdev->desc_nr;
NeilBrownb2d444d2005-11-08 21:39:31 -08002279 set_bit(In_sync, &rdev->flags);
NeilBrowna778b732007-05-23 13:58:10 -07002280 } else if (rdev->raid_disk >= mddev->raid_disks) {
2281 rdev->raid_disk = -1;
2282 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 }
2284 }
2285
2286
2287
2288 if (mddev->recovery_cp != MaxSector &&
2289 mddev->level >= 1)
2290 printk(KERN_ERR "md: %s: raid array is not clean"
2291 " -- starting background reconstruction\n",
2292 mdname(mddev));
2293
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294}
2295
NeilBrowneae17012005-11-08 21:39:23 -08002296static ssize_t
NeilBrown16f17b32006-06-26 00:27:37 -07002297safe_delay_show(mddev_t *mddev, char *page)
2298{
2299 int msec = (mddev->safemode_delay*1000)/HZ;
2300 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2301}
2302static ssize_t
2303safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2304{
2305 int scale=1;
2306 int dot=0;
2307 int i;
2308 unsigned long msec;
2309 char buf[30];
2310 char *e;
2311 /* remove a period, and count digits after it */
2312 if (len >= sizeof(buf))
2313 return -EINVAL;
2314 strlcpy(buf, cbuf, len);
2315 buf[len] = 0;
2316 for (i=0; i<len; i++) {
2317 if (dot) {
2318 if (isdigit(buf[i])) {
2319 buf[i-1] = buf[i];
2320 scale *= 10;
2321 }
2322 buf[i] = 0;
2323 } else if (buf[i] == '.') {
2324 dot=1;
2325 buf[i] = 0;
2326 }
2327 }
2328 msec = simple_strtoul(buf, &e, 10);
2329 if (e == buf || (*e && *e != '\n'))
2330 return -EINVAL;
2331 msec = (msec * 1000) / scale;
2332 if (msec == 0)
2333 mddev->safemode_delay = 0;
2334 else {
2335 mddev->safemode_delay = (msec*HZ)/1000;
2336 if (mddev->safemode_delay == 0)
2337 mddev->safemode_delay = 1;
2338 }
2339 return len;
2340}
2341static struct md_sysfs_entry md_safe_delay =
NeilBrown80ca3a42006-07-10 04:44:18 -07002342__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
NeilBrown16f17b32006-06-26 00:27:37 -07002343
2344static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08002345level_show(mddev_t *mddev, char *page)
NeilBrowneae17012005-11-08 21:39:23 -08002346{
NeilBrown2604b702006-01-06 00:20:36 -08002347 struct mdk_personality *p = mddev->pers;
NeilBrownd9d166c2006-01-06 00:20:51 -08002348 if (p)
NeilBrowneae17012005-11-08 21:39:23 -08002349 return sprintf(page, "%s\n", p->name);
NeilBrownd9d166c2006-01-06 00:20:51 -08002350 else if (mddev->clevel[0])
2351 return sprintf(page, "%s\n", mddev->clevel);
2352 else if (mddev->level != LEVEL_NONE)
2353 return sprintf(page, "%d\n", mddev->level);
2354 else
2355 return 0;
NeilBrowneae17012005-11-08 21:39:23 -08002356}
2357
NeilBrownd9d166c2006-01-06 00:20:51 -08002358static ssize_t
2359level_store(mddev_t *mddev, const char *buf, size_t len)
2360{
NeilBrown20a49ff2008-02-06 01:39:57 -08002361 ssize_t rv = len;
NeilBrownd9d166c2006-01-06 00:20:51 -08002362 if (mddev->pers)
2363 return -EBUSY;
2364 if (len == 0)
2365 return 0;
2366 if (len >= sizeof(mddev->clevel))
2367 return -ENOSPC;
2368 strncpy(mddev->clevel, buf, len);
2369 if (mddev->clevel[len-1] == '\n')
2370 len--;
2371 mddev->clevel[len] = 0;
2372 mddev->level = LEVEL_NONE;
2373 return rv;
2374}
2375
2376static struct md_sysfs_entry md_level =
NeilBrown80ca3a42006-07-10 04:44:18 -07002377__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
NeilBrowneae17012005-11-08 21:39:23 -08002378
NeilBrownd4dbd022006-06-26 00:27:59 -07002379
2380static ssize_t
2381layout_show(mddev_t *mddev, char *page)
2382{
2383 /* just a number, not meaningful for all levels */
NeilBrown08a02ec2007-05-09 02:35:38 -07002384 if (mddev->reshape_position != MaxSector &&
2385 mddev->layout != mddev->new_layout)
2386 return sprintf(page, "%d (%d)\n",
2387 mddev->new_layout, mddev->layout);
NeilBrownd4dbd022006-06-26 00:27:59 -07002388 return sprintf(page, "%d\n", mddev->layout);
2389}
2390
2391static ssize_t
2392layout_store(mddev_t *mddev, const char *buf, size_t len)
2393{
2394 char *e;
2395 unsigned long n = simple_strtoul(buf, &e, 10);
NeilBrownd4dbd022006-06-26 00:27:59 -07002396
2397 if (!*buf || (*e && *e != '\n'))
2398 return -EINVAL;
2399
NeilBrown08a02ec2007-05-09 02:35:38 -07002400 if (mddev->pers)
2401 return -EBUSY;
2402 if (mddev->reshape_position != MaxSector)
2403 mddev->new_layout = n;
2404 else
2405 mddev->layout = n;
NeilBrownd4dbd022006-06-26 00:27:59 -07002406 return len;
2407}
2408static struct md_sysfs_entry md_layout =
NeilBrown80ca3a42006-07-10 04:44:18 -07002409__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
NeilBrownd4dbd022006-06-26 00:27:59 -07002410
2411
NeilBrowneae17012005-11-08 21:39:23 -08002412static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08002413raid_disks_show(mddev_t *mddev, char *page)
NeilBrowneae17012005-11-08 21:39:23 -08002414{
NeilBrownbb636542005-11-08 21:39:45 -08002415 if (mddev->raid_disks == 0)
2416 return 0;
NeilBrown08a02ec2007-05-09 02:35:38 -07002417 if (mddev->reshape_position != MaxSector &&
2418 mddev->delta_disks != 0)
2419 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2420 mddev->raid_disks - mddev->delta_disks);
NeilBrowneae17012005-11-08 21:39:23 -08002421 return sprintf(page, "%d\n", mddev->raid_disks);
2422}
2423
NeilBrownda943b992006-01-06 00:20:54 -08002424static int update_raid_disks(mddev_t *mddev, int raid_disks);
2425
2426static ssize_t
2427raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2428{
NeilBrownda943b992006-01-06 00:20:54 -08002429 char *e;
2430 int rv = 0;
2431 unsigned long n = simple_strtoul(buf, &e, 10);
2432
2433 if (!*buf || (*e && *e != '\n'))
2434 return -EINVAL;
2435
2436 if (mddev->pers)
2437 rv = update_raid_disks(mddev, n);
NeilBrown08a02ec2007-05-09 02:35:38 -07002438 else if (mddev->reshape_position != MaxSector) {
2439 int olddisks = mddev->raid_disks - mddev->delta_disks;
2440 mddev->delta_disks = n - olddisks;
2441 mddev->raid_disks = n;
2442 } else
NeilBrownda943b992006-01-06 00:20:54 -08002443 mddev->raid_disks = n;
2444 return rv ? rv : len;
2445}
2446static struct md_sysfs_entry md_raid_disks =
NeilBrown80ca3a42006-07-10 04:44:18 -07002447__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
NeilBrowneae17012005-11-08 21:39:23 -08002448
NeilBrown24dd4692005-11-08 21:39:26 -08002449static ssize_t
NeilBrown3b343802006-01-06 00:20:47 -08002450chunk_size_show(mddev_t *mddev, char *page)
2451{
NeilBrown08a02ec2007-05-09 02:35:38 -07002452 if (mddev->reshape_position != MaxSector &&
2453 mddev->chunk_size != mddev->new_chunk)
2454 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2455 mddev->chunk_size);
NeilBrown3b343802006-01-06 00:20:47 -08002456 return sprintf(page, "%d\n", mddev->chunk_size);
2457}
2458
2459static ssize_t
2460chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2461{
2462 /* can only set chunk_size if array is not yet active */
2463 char *e;
2464 unsigned long n = simple_strtoul(buf, &e, 10);
2465
NeilBrown3b343802006-01-06 00:20:47 -08002466 if (!*buf || (*e && *e != '\n'))
2467 return -EINVAL;
2468
NeilBrown08a02ec2007-05-09 02:35:38 -07002469 if (mddev->pers)
2470 return -EBUSY;
2471 else if (mddev->reshape_position != MaxSector)
2472 mddev->new_chunk = n;
2473 else
2474 mddev->chunk_size = n;
NeilBrown3b343802006-01-06 00:20:47 -08002475 return len;
2476}
2477static struct md_sysfs_entry md_chunk_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07002478__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
NeilBrown3b343802006-01-06 00:20:47 -08002479
NeilBrowna94213b2006-06-26 00:28:00 -07002480static ssize_t
2481resync_start_show(mddev_t *mddev, char *page)
2482{
2483 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2484}
2485
2486static ssize_t
2487resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2488{
NeilBrowna94213b2006-06-26 00:28:00 -07002489 char *e;
2490 unsigned long long n = simple_strtoull(buf, &e, 10);
2491
2492 if (mddev->pers)
2493 return -EBUSY;
2494 if (!*buf || (*e && *e != '\n'))
2495 return -EINVAL;
2496
2497 mddev->recovery_cp = n;
2498 return len;
2499}
2500static struct md_sysfs_entry md_resync_start =
NeilBrown80ca3a42006-07-10 04:44:18 -07002501__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
NeilBrowna94213b2006-06-26 00:28:00 -07002502
NeilBrown9e653b62006-06-26 00:27:58 -07002503/*
2504 * The array state can be:
2505 *
2506 * clear
2507 * No devices, no size, no level
2508 * Equivalent to STOP_ARRAY ioctl
2509 * inactive
2510 * May have some settings, but array is not active
2511 * all IO results in error
2512 * When written, doesn't tear down array, but just stops it
2513 * suspended (not supported yet)
2514 * All IO requests will block. The array can be reconfigured.
2515 * Writing this, if accepted, will block until array is quiessent
2516 * readonly
2517 * no resync can happen. no superblocks get written.
2518 * write requests fail
2519 * read-auto
2520 * like readonly, but behaves like 'clean' on a write request.
2521 *
2522 * clean - no pending writes, but otherwise active.
2523 * When written to inactive array, starts without resync
2524 * If a write request arrives then
2525 * if metadata is known, mark 'dirty' and switch to 'active'.
2526 * if not known, block and switch to write-pending
2527 * If written to an active array that has pending writes, then fails.
2528 * active
2529 * fully active: IO and resync can be happening.
2530 * When written to inactive array, starts with resync
2531 *
2532 * write-pending
2533 * clean, but writes are blocked waiting for 'active' to be written.
2534 *
2535 * active-idle
2536 * like active, but no writes have been seen for a while (100msec).
2537 *
2538 */
2539enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2540 write_pending, active_idle, bad_word};
Adrian Bunk05381952006-06-26 00:28:01 -07002541static char *array_states[] = {
NeilBrown9e653b62006-06-26 00:27:58 -07002542 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2543 "write-pending", "active-idle", NULL };
2544
2545static int match_word(const char *word, char **list)
2546{
2547 int n;
2548 for (n=0; list[n]; n++)
2549 if (cmd_match(word, list[n]))
2550 break;
2551 return n;
2552}
2553
2554static ssize_t
2555array_state_show(mddev_t *mddev, char *page)
2556{
2557 enum array_state st = inactive;
2558
2559 if (mddev->pers)
2560 switch(mddev->ro) {
2561 case 1:
2562 st = readonly;
2563 break;
2564 case 2:
2565 st = read_auto;
2566 break;
2567 case 0:
2568 if (mddev->in_sync)
2569 st = clean;
NeilBrowne6910632008-02-06 01:39:51 -08002570 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2571 st = write_pending;
NeilBrown9e653b62006-06-26 00:27:58 -07002572 else if (mddev->safemode)
2573 st = active_idle;
2574 else
2575 st = active;
2576 }
2577 else {
2578 if (list_empty(&mddev->disks) &&
2579 mddev->raid_disks == 0 &&
2580 mddev->size == 0)
2581 st = clear;
2582 else
2583 st = inactive;
2584 }
2585 return sprintf(page, "%s\n", array_states[st]);
2586}
2587
2588static int do_md_stop(mddev_t * mddev, int ro);
2589static int do_md_run(mddev_t * mddev);
2590static int restart_array(mddev_t *mddev);
2591
2592static ssize_t
2593array_state_store(mddev_t *mddev, const char *buf, size_t len)
2594{
2595 int err = -EINVAL;
2596 enum array_state st = match_word(buf, array_states);
2597 switch(st) {
2598 case bad_word:
2599 break;
2600 case clear:
2601 /* stopping an active array */
NeilBrowne6910632008-02-06 01:39:51 -08002602 if (atomic_read(&mddev->active) > 1)
2603 return -EBUSY;
2604 err = do_md_stop(mddev, 0);
NeilBrown9e653b62006-06-26 00:27:58 -07002605 break;
2606 case inactive:
2607 /* stopping an active array */
2608 if (mddev->pers) {
2609 if (atomic_read(&mddev->active) > 1)
2610 return -EBUSY;
2611 err = do_md_stop(mddev, 2);
NeilBrowne6910632008-02-06 01:39:51 -08002612 } else
2613 err = 0; /* already inactive */
NeilBrown9e653b62006-06-26 00:27:58 -07002614 break;
2615 case suspended:
2616 break; /* not supported yet */
2617 case readonly:
2618 if (mddev->pers)
2619 err = do_md_stop(mddev, 1);
2620 else {
2621 mddev->ro = 1;
NeilBrown648b6292008-04-30 00:52:30 -07002622 set_disk_ro(mddev->gendisk, 1);
NeilBrown9e653b62006-06-26 00:27:58 -07002623 err = do_md_run(mddev);
2624 }
2625 break;
2626 case read_auto:
NeilBrown9e653b62006-06-26 00:27:58 -07002627 if (mddev->pers) {
NeilBrown648b6292008-04-30 00:52:30 -07002628 if (mddev->ro != 1)
2629 err = do_md_stop(mddev, 1);
2630 else
2631 err = restart_array(mddev);
2632 if (err == 0) {
2633 mddev->ro = 2;
2634 set_disk_ro(mddev->gendisk, 0);
2635 }
NeilBrown9e653b62006-06-26 00:27:58 -07002636 } else {
2637 mddev->ro = 2;
2638 err = do_md_run(mddev);
2639 }
2640 break;
2641 case clean:
2642 if (mddev->pers) {
2643 restart_array(mddev);
2644 spin_lock_irq(&mddev->write_lock);
2645 if (atomic_read(&mddev->writes_pending) == 0) {
NeilBrowne6910632008-02-06 01:39:51 -08002646 if (mddev->in_sync == 0) {
2647 mddev->in_sync = 1;
NeilBrown31a59e32008-04-30 00:52:30 -07002648 if (mddev->safemode == 1)
2649 mddev->safemode = 0;
NeilBrowne6910632008-02-06 01:39:51 -08002650 if (mddev->persistent)
2651 set_bit(MD_CHANGE_CLEAN,
2652 &mddev->flags);
2653 }
2654 err = 0;
2655 } else
2656 err = -EBUSY;
NeilBrown9e653b62006-06-26 00:27:58 -07002657 spin_unlock_irq(&mddev->write_lock);
2658 } else {
2659 mddev->ro = 0;
2660 mddev->recovery_cp = MaxSector;
2661 err = do_md_run(mddev);
2662 }
2663 break;
2664 case active:
2665 if (mddev->pers) {
2666 restart_array(mddev);
NeilBrowne6910632008-02-06 01:39:51 -08002667 if (mddev->external)
2668 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown9e653b62006-06-26 00:27:58 -07002669 wake_up(&mddev->sb_wait);
2670 err = 0;
2671 } else {
2672 mddev->ro = 0;
NeilBrown648b6292008-04-30 00:52:30 -07002673 set_disk_ro(mddev->gendisk, 0);
NeilBrown9e653b62006-06-26 00:27:58 -07002674 err = do_md_run(mddev);
2675 }
2676 break;
2677 case write_pending:
2678 case active_idle:
2679 /* these cannot be set */
2680 break;
2681 }
2682 if (err)
2683 return err;
2684 else
2685 return len;
2686}
NeilBrown80ca3a42006-07-10 04:44:18 -07002687static struct md_sysfs_entry md_array_state =
2688__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
NeilBrown9e653b62006-06-26 00:27:58 -07002689
NeilBrown6d7ff732006-01-06 00:21:16 -08002690static ssize_t
2691null_show(mddev_t *mddev, char *page)
2692{
2693 return -EINVAL;
2694}
2695
2696static ssize_t
2697new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2698{
2699 /* buf must be %d:%d\n? giving major and minor numbers */
2700 /* The new device is added to the array.
2701 * If the array has a persistent superblock, we read the
2702 * superblock to initialise info and check validity.
2703 * Otherwise, only checking done is that in bind_rdev_to_array,
2704 * which mainly checks size.
2705 */
2706 char *e;
2707 int major = simple_strtoul(buf, &e, 10);
2708 int minor;
2709 dev_t dev;
2710 mdk_rdev_t *rdev;
2711 int err;
2712
2713 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2714 return -EINVAL;
2715 minor = simple_strtoul(e+1, &e, 10);
2716 if (*e && *e != '\n')
2717 return -EINVAL;
2718 dev = MKDEV(major, minor);
2719 if (major != MAJOR(dev) ||
2720 minor != MINOR(dev))
2721 return -EOVERFLOW;
2722
2723
2724 if (mddev->persistent) {
2725 rdev = md_import_device(dev, mddev->major_version,
2726 mddev->minor_version);
2727 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2728 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2729 mdk_rdev_t, same_set);
2730 err = super_types[mddev->major_version]
2731 .load_super(rdev, rdev0, mddev->minor_version);
2732 if (err < 0)
2733 goto out;
2734 }
NeilBrownc5d79ad2008-02-06 01:39:54 -08002735 } else if (mddev->external)
2736 rdev = md_import_device(dev, -2, -1);
2737 else
NeilBrown6d7ff732006-01-06 00:21:16 -08002738 rdev = md_import_device(dev, -1, -1);
2739
2740 if (IS_ERR(rdev))
2741 return PTR_ERR(rdev);
2742 err = bind_rdev_to_array(rdev, mddev);
2743 out:
2744 if (err)
2745 export_rdev(rdev);
2746 return err ? err : len;
2747}
2748
2749static struct md_sysfs_entry md_new_device =
NeilBrown80ca3a42006-07-10 04:44:18 -07002750__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
NeilBrown3b343802006-01-06 00:20:47 -08002751
2752static ssize_t
Paul Clements9b1d1da2006-10-03 01:15:49 -07002753bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2754{
2755 char *end;
2756 unsigned long chunk, end_chunk;
2757
2758 if (!mddev->bitmap)
2759 goto out;
2760 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2761 while (*buf) {
2762 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2763 if (buf == end) break;
2764 if (*end == '-') { /* range */
2765 buf = end + 1;
2766 end_chunk = simple_strtoul(buf, &end, 0);
2767 if (buf == end) break;
2768 }
2769 if (*end && !isspace(*end)) break;
2770 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2771 buf = end;
2772 while (isspace(*buf)) buf++;
2773 }
2774 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2775out:
2776 return len;
2777}
2778
2779static struct md_sysfs_entry md_bitmap =
2780__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2781
2782static ssize_t
NeilBrowna35b0d62006-01-06 00:20:49 -08002783size_show(mddev_t *mddev, char *page)
2784{
2785 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2786}
2787
2788static int update_size(mddev_t *mddev, unsigned long size);
2789
2790static ssize_t
2791size_store(mddev_t *mddev, const char *buf, size_t len)
2792{
2793 /* If array is inactive, we can reduce the component size, but
2794 * not increase it (except from 0).
2795 * If array is active, we can try an on-line resize
2796 */
2797 char *e;
2798 int err = 0;
2799 unsigned long long size = simple_strtoull(buf, &e, 10);
2800 if (!*buf || *buf == '\n' ||
2801 (*e && *e != '\n'))
2802 return -EINVAL;
2803
2804 if (mddev->pers) {
2805 err = update_size(mddev, size);
NeilBrown850b2b42006-10-03 01:15:46 -07002806 md_update_sb(mddev, 1);
NeilBrowna35b0d62006-01-06 00:20:49 -08002807 } else {
2808 if (mddev->size == 0 ||
2809 mddev->size > size)
2810 mddev->size = size;
2811 else
2812 err = -ENOSPC;
2813 }
2814 return err ? err : len;
2815}
2816
2817static struct md_sysfs_entry md_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07002818__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
NeilBrowna35b0d62006-01-06 00:20:49 -08002819
NeilBrown8bb93aa2006-01-06 00:20:50 -08002820
2821/* Metdata version.
NeilBrowne6910632008-02-06 01:39:51 -08002822 * This is one of
2823 * 'none' for arrays with no metadata (good luck...)
2824 * 'external' for arrays with externally managed metadata,
NeilBrown8bb93aa2006-01-06 00:20:50 -08002825 * or N.M for internally known formats
2826 */
2827static ssize_t
2828metadata_show(mddev_t *mddev, char *page)
2829{
2830 if (mddev->persistent)
2831 return sprintf(page, "%d.%d\n",
2832 mddev->major_version, mddev->minor_version);
NeilBrowne6910632008-02-06 01:39:51 -08002833 else if (mddev->external)
2834 return sprintf(page, "external:%s\n", mddev->metadata_type);
NeilBrown8bb93aa2006-01-06 00:20:50 -08002835 else
2836 return sprintf(page, "none\n");
2837}
2838
2839static ssize_t
2840metadata_store(mddev_t *mddev, const char *buf, size_t len)
2841{
2842 int major, minor;
2843 char *e;
2844 if (!list_empty(&mddev->disks))
2845 return -EBUSY;
2846
2847 if (cmd_match(buf, "none")) {
2848 mddev->persistent = 0;
NeilBrowne6910632008-02-06 01:39:51 -08002849 mddev->external = 0;
2850 mddev->major_version = 0;
2851 mddev->minor_version = 90;
2852 return len;
2853 }
2854 if (strncmp(buf, "external:", 9) == 0) {
NeilBrown20a49ff2008-02-06 01:39:57 -08002855 size_t namelen = len-9;
NeilBrowne6910632008-02-06 01:39:51 -08002856 if (namelen >= sizeof(mddev->metadata_type))
2857 namelen = sizeof(mddev->metadata_type)-1;
2858 strncpy(mddev->metadata_type, buf+9, namelen);
2859 mddev->metadata_type[namelen] = 0;
2860 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2861 mddev->metadata_type[--namelen] = 0;
2862 mddev->persistent = 0;
2863 mddev->external = 1;
NeilBrown8bb93aa2006-01-06 00:20:50 -08002864 mddev->major_version = 0;
2865 mddev->minor_version = 90;
2866 return len;
2867 }
2868 major = simple_strtoul(buf, &e, 10);
2869 if (e==buf || *e != '.')
2870 return -EINVAL;
2871 buf = e+1;
2872 minor = simple_strtoul(buf, &e, 10);
NeilBrown3f9d7b02006-12-22 01:11:41 -08002873 if (e==buf || (*e && *e != '\n') )
NeilBrown8bb93aa2006-01-06 00:20:50 -08002874 return -EINVAL;
Ahmed S. Darwish50511da2007-05-09 02:35:34 -07002875 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
NeilBrown8bb93aa2006-01-06 00:20:50 -08002876 return -ENOENT;
2877 mddev->major_version = major;
2878 mddev->minor_version = minor;
2879 mddev->persistent = 1;
NeilBrowne6910632008-02-06 01:39:51 -08002880 mddev->external = 0;
NeilBrown8bb93aa2006-01-06 00:20:50 -08002881 return len;
2882}
2883
2884static struct md_sysfs_entry md_metadata =
NeilBrown80ca3a42006-07-10 04:44:18 -07002885__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
NeilBrown8bb93aa2006-01-06 00:20:50 -08002886
NeilBrowna35b0d62006-01-06 00:20:49 -08002887static ssize_t
NeilBrown7eec3142005-11-08 21:39:44 -08002888action_show(mddev_t *mddev, char *page)
NeilBrown24dd4692005-11-08 21:39:26 -08002889{
NeilBrown7eec3142005-11-08 21:39:44 -08002890 char *type = "idle";
NeilBrown31399d92005-11-08 21:39:29 -08002891 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
NeilBrown2b12ab62007-10-16 23:30:53 -07002892 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08002893 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2894 type = "reshape";
2895 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
NeilBrown24dd4692005-11-08 21:39:26 -08002896 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2897 type = "resync";
2898 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2899 type = "check";
2900 else
2901 type = "repair";
2902 } else
2903 type = "recover";
2904 }
2905 return sprintf(page, "%s\n", type);
2906}
2907
2908static ssize_t
NeilBrown7eec3142005-11-08 21:39:44 -08002909action_store(mddev_t *mddev, const char *page, size_t len)
NeilBrown24dd4692005-11-08 21:39:26 -08002910{
NeilBrown7eec3142005-11-08 21:39:44 -08002911 if (!mddev->pers || !mddev->pers->sync_request)
2912 return -EINVAL;
2913
NeilBrownbce74da2006-01-06 00:20:41 -08002914 if (cmd_match(page, "idle")) {
NeilBrown7eec3142005-11-08 21:39:44 -08002915 if (mddev->sync_thread) {
2916 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2917 md_unregister_thread(mddev->sync_thread);
2918 mddev->sync_thread = NULL;
2919 mddev->recovery = 0;
2920 }
NeilBrown03c902e2006-01-06 00:20:46 -08002921 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2922 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
NeilBrown24dd4692005-11-08 21:39:26 -08002923 return -EBUSY;
NeilBrown03c902e2006-01-06 00:20:46 -08002924 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
NeilBrown7eec3142005-11-08 21:39:44 -08002925 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrown16484bf2006-03-27 01:18:13 -08002926 else if (cmd_match(page, "reshape")) {
2927 int err;
2928 if (mddev->pers->start_reshape == NULL)
2929 return -EINVAL;
2930 err = mddev->pers->start_reshape(mddev);
2931 if (err)
2932 return err;
2933 } else {
NeilBrownbce74da2006-01-06 00:20:41 -08002934 if (cmd_match(page, "check"))
NeilBrown7eec3142005-11-08 21:39:44 -08002935 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
NeilBrown2adc7d42006-05-20 14:59:57 -07002936 else if (!cmd_match(page, "repair"))
NeilBrown7eec3142005-11-08 21:39:44 -08002937 return -EINVAL;
2938 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2939 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
NeilBrown7eec3142005-11-08 21:39:44 -08002940 }
NeilBrown03c902e2006-01-06 00:20:46 -08002941 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08002942 md_wakeup_thread(mddev->thread);
2943 return len;
2944}
2945
NeilBrown9d888832005-11-08 21:39:26 -08002946static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08002947mismatch_cnt_show(mddev_t *mddev, char *page)
NeilBrown9d888832005-11-08 21:39:26 -08002948{
2949 return sprintf(page, "%llu\n",
2950 (unsigned long long) mddev->resync_mismatches);
2951}
2952
NeilBrown80ca3a42006-07-10 04:44:18 -07002953static struct md_sysfs_entry md_scan_mode =
2954__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
NeilBrown24dd4692005-11-08 21:39:26 -08002955
NeilBrown96de1e62005-11-08 21:39:39 -08002956
NeilBrown80ca3a42006-07-10 04:44:18 -07002957static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
NeilBrown9d888832005-11-08 21:39:26 -08002958
NeilBrown88202a02006-01-06 00:21:36 -08002959static ssize_t
2960sync_min_show(mddev_t *mddev, char *page)
2961{
2962 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2963 mddev->sync_speed_min ? "local": "system");
2964}
2965
2966static ssize_t
2967sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2968{
2969 int min;
2970 char *e;
2971 if (strncmp(buf, "system", 6)==0) {
2972 mddev->sync_speed_min = 0;
2973 return len;
2974 }
2975 min = simple_strtoul(buf, &e, 10);
2976 if (buf == e || (*e && *e != '\n') || min <= 0)
2977 return -EINVAL;
2978 mddev->sync_speed_min = min;
2979 return len;
2980}
2981
2982static struct md_sysfs_entry md_sync_min =
2983__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2984
2985static ssize_t
2986sync_max_show(mddev_t *mddev, char *page)
2987{
2988 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2989 mddev->sync_speed_max ? "local": "system");
2990}
2991
2992static ssize_t
2993sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2994{
2995 int max;
2996 char *e;
2997 if (strncmp(buf, "system", 6)==0) {
2998 mddev->sync_speed_max = 0;
2999 return len;
3000 }
3001 max = simple_strtoul(buf, &e, 10);
3002 if (buf == e || (*e && *e != '\n') || max <= 0)
3003 return -EINVAL;
3004 mddev->sync_speed_max = max;
3005 return len;
3006}
3007
3008static struct md_sysfs_entry md_sync_max =
3009__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3010
Iustin Popd7f3d292007-10-16 23:30:54 -07003011static ssize_t
3012degraded_show(mddev_t *mddev, char *page)
3013{
3014 return sprintf(page, "%d\n", mddev->degraded);
3015}
3016static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
NeilBrown88202a02006-01-06 00:21:36 -08003017
3018static ssize_t
Bernd Schubert90b08712008-05-23 13:04:38 -07003019sync_force_parallel_show(mddev_t *mddev, char *page)
3020{
3021 return sprintf(page, "%d\n", mddev->parallel_resync);
3022}
3023
3024static ssize_t
3025sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3026{
3027 long n;
3028
3029 if (strict_strtol(buf, 10, &n))
3030 return -EINVAL;
3031
3032 if (n != 0 && n != 1)
3033 return -EINVAL;
3034
3035 mddev->parallel_resync = n;
3036
3037 if (mddev->sync_thread)
3038 wake_up(&resync_wait);
3039
3040 return len;
3041}
3042
3043/* force parallel resync, even with shared block devices */
3044static struct md_sysfs_entry md_sync_force_parallel =
3045__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3046 sync_force_parallel_show, sync_force_parallel_store);
3047
3048static ssize_t
NeilBrown88202a02006-01-06 00:21:36 -08003049sync_speed_show(mddev_t *mddev, char *page)
3050{
3051 unsigned long resync, dt, db;
NeilBrownff4e8d92006-07-10 04:44:16 -07003052 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
NeilBrown88202a02006-01-06 00:21:36 -08003053 dt = ((jiffies - mddev->resync_mark) / HZ);
3054 if (!dt) dt++;
3055 db = resync - (mddev->resync_mark_cnt);
3056 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
3057}
3058
NeilBrown80ca3a42006-07-10 04:44:18 -07003059static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
NeilBrown88202a02006-01-06 00:21:36 -08003060
3061static ssize_t
3062sync_completed_show(mddev_t *mddev, char *page)
3063{
3064 unsigned long max_blocks, resync;
3065
3066 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3067 max_blocks = mddev->resync_max_sectors;
3068 else
3069 max_blocks = mddev->size << 1;
3070
3071 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3072 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3073}
3074
NeilBrown80ca3a42006-07-10 04:44:18 -07003075static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
NeilBrown88202a02006-01-06 00:21:36 -08003076
NeilBrowne464eaf2006-03-27 01:18:14 -08003077static ssize_t
NeilBrownc6207272008-02-06 01:39:52 -08003078max_sync_show(mddev_t *mddev, char *page)
3079{
3080 if (mddev->resync_max == MaxSector)
3081 return sprintf(page, "max\n");
3082 else
3083 return sprintf(page, "%llu\n",
3084 (unsigned long long)mddev->resync_max);
3085}
3086static ssize_t
3087max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3088{
3089 if (strncmp(buf, "max", 3) == 0)
3090 mddev->resync_max = MaxSector;
3091 else {
3092 char *ep;
3093 unsigned long long max = simple_strtoull(buf, &ep, 10);
3094 if (ep == buf || (*ep != 0 && *ep != '\n'))
3095 return -EINVAL;
3096 if (max < mddev->resync_max &&
3097 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3098 return -EBUSY;
3099
3100 /* Must be a multiple of chunk_size */
3101 if (mddev->chunk_size) {
3102 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3103 return -EINVAL;
3104 }
3105 mddev->resync_max = max;
3106 }
3107 wake_up(&mddev->recovery_wait);
3108 return len;
3109}
3110
3111static struct md_sysfs_entry md_max_sync =
3112__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3113
3114static ssize_t
NeilBrowne464eaf2006-03-27 01:18:14 -08003115suspend_lo_show(mddev_t *mddev, char *page)
3116{
3117 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3118}
3119
3120static ssize_t
3121suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3122{
3123 char *e;
3124 unsigned long long new = simple_strtoull(buf, &e, 10);
3125
3126 if (mddev->pers->quiesce == NULL)
3127 return -EINVAL;
3128 if (buf == e || (*e && *e != '\n'))
3129 return -EINVAL;
3130 if (new >= mddev->suspend_hi ||
3131 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3132 mddev->suspend_lo = new;
3133 mddev->pers->quiesce(mddev, 2);
3134 return len;
3135 } else
3136 return -EINVAL;
3137}
3138static struct md_sysfs_entry md_suspend_lo =
3139__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3140
3141
3142static ssize_t
3143suspend_hi_show(mddev_t *mddev, char *page)
3144{
3145 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3146}
3147
3148static ssize_t
3149suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3150{
3151 char *e;
3152 unsigned long long new = simple_strtoull(buf, &e, 10);
3153
3154 if (mddev->pers->quiesce == NULL)
3155 return -EINVAL;
3156 if (buf == e || (*e && *e != '\n'))
3157 return -EINVAL;
3158 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3159 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3160 mddev->suspend_hi = new;
3161 mddev->pers->quiesce(mddev, 1);
3162 mddev->pers->quiesce(mddev, 0);
3163 return len;
3164 } else
3165 return -EINVAL;
3166}
3167static struct md_sysfs_entry md_suspend_hi =
3168__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3169
NeilBrown08a02ec2007-05-09 02:35:38 -07003170static ssize_t
3171reshape_position_show(mddev_t *mddev, char *page)
3172{
3173 if (mddev->reshape_position != MaxSector)
3174 return sprintf(page, "%llu\n",
3175 (unsigned long long)mddev->reshape_position);
3176 strcpy(page, "none\n");
3177 return 5;
3178}
3179
3180static ssize_t
3181reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3182{
3183 char *e;
3184 unsigned long long new = simple_strtoull(buf, &e, 10);
3185 if (mddev->pers)
3186 return -EBUSY;
3187 if (buf == e || (*e && *e != '\n'))
3188 return -EINVAL;
3189 mddev->reshape_position = new;
3190 mddev->delta_disks = 0;
3191 mddev->new_level = mddev->level;
3192 mddev->new_layout = mddev->layout;
3193 mddev->new_chunk = mddev->chunk_size;
3194 return len;
3195}
3196
3197static struct md_sysfs_entry md_reshape_position =
3198__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3199 reshape_position_store);
3200
NeilBrowne464eaf2006-03-27 01:18:14 -08003201
NeilBrowneae17012005-11-08 21:39:23 -08003202static struct attribute *md_default_attrs[] = {
3203 &md_level.attr,
NeilBrownd4dbd022006-06-26 00:27:59 -07003204 &md_layout.attr,
NeilBrowneae17012005-11-08 21:39:23 -08003205 &md_raid_disks.attr,
NeilBrown3b343802006-01-06 00:20:47 -08003206 &md_chunk_size.attr,
NeilBrowna35b0d62006-01-06 00:20:49 -08003207 &md_size.attr,
NeilBrowna94213b2006-06-26 00:28:00 -07003208 &md_resync_start.attr,
NeilBrown8bb93aa2006-01-06 00:20:50 -08003209 &md_metadata.attr,
NeilBrown6d7ff732006-01-06 00:21:16 -08003210 &md_new_device.attr,
NeilBrown16f17b32006-06-26 00:27:37 -07003211 &md_safe_delay.attr,
NeilBrown9e653b62006-06-26 00:27:58 -07003212 &md_array_state.attr,
NeilBrown08a02ec2007-05-09 02:35:38 -07003213 &md_reshape_position.attr,
NeilBrown411036f2005-11-08 21:39:40 -08003214 NULL,
3215};
3216
3217static struct attribute *md_redundancy_attrs[] = {
NeilBrown24dd4692005-11-08 21:39:26 -08003218 &md_scan_mode.attr,
NeilBrown9d888832005-11-08 21:39:26 -08003219 &md_mismatches.attr,
NeilBrown88202a02006-01-06 00:21:36 -08003220 &md_sync_min.attr,
3221 &md_sync_max.attr,
3222 &md_sync_speed.attr,
Bernd Schubert90b08712008-05-23 13:04:38 -07003223 &md_sync_force_parallel.attr,
NeilBrown88202a02006-01-06 00:21:36 -08003224 &md_sync_completed.attr,
NeilBrownc6207272008-02-06 01:39:52 -08003225 &md_max_sync.attr,
NeilBrowne464eaf2006-03-27 01:18:14 -08003226 &md_suspend_lo.attr,
3227 &md_suspend_hi.attr,
Paul Clements9b1d1da2006-10-03 01:15:49 -07003228 &md_bitmap.attr,
Iustin Popd7f3d292007-10-16 23:30:54 -07003229 &md_degraded.attr,
NeilBrowneae17012005-11-08 21:39:23 -08003230 NULL,
3231};
NeilBrown411036f2005-11-08 21:39:40 -08003232static struct attribute_group md_redundancy_group = {
3233 .name = NULL,
3234 .attrs = md_redundancy_attrs,
3235};
3236
NeilBrowneae17012005-11-08 21:39:23 -08003237
3238static ssize_t
3239md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3240{
3241 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3242 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
NeilBrown96de1e62005-11-08 21:39:39 -08003243 ssize_t rv;
NeilBrowneae17012005-11-08 21:39:23 -08003244
3245 if (!entry->show)
3246 return -EIO;
Ingo Molnar5dc5cf72006-04-20 02:43:23 -07003247 rv = mddev_lock(mddev);
3248 if (!rv) {
3249 rv = entry->show(mddev, page);
3250 mddev_unlock(mddev);
3251 }
NeilBrown96de1e62005-11-08 21:39:39 -08003252 return rv;
NeilBrowneae17012005-11-08 21:39:23 -08003253}
3254
3255static ssize_t
3256md_attr_store(struct kobject *kobj, struct attribute *attr,
3257 const char *page, size_t length)
3258{
3259 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3260 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
NeilBrown96de1e62005-11-08 21:39:39 -08003261 ssize_t rv;
NeilBrowneae17012005-11-08 21:39:23 -08003262
3263 if (!entry->store)
3264 return -EIO;
NeilBrown67463ac2006-07-10 04:44:19 -07003265 if (!capable(CAP_SYS_ADMIN))
3266 return -EACCES;
Ingo Molnar5dc5cf72006-04-20 02:43:23 -07003267 rv = mddev_lock(mddev);
3268 if (!rv) {
3269 rv = entry->store(mddev, page, length);
3270 mddev_unlock(mddev);
3271 }
NeilBrown96de1e62005-11-08 21:39:39 -08003272 return rv;
NeilBrowneae17012005-11-08 21:39:23 -08003273}
3274
3275static void md_free(struct kobject *ko)
3276{
3277 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3278 kfree(mddev);
3279}
3280
3281static struct sysfs_ops md_sysfs_ops = {
3282 .show = md_attr_show,
3283 .store = md_attr_store,
3284};
3285static struct kobj_type md_ktype = {
3286 .release = md_free,
3287 .sysfs_ops = &md_sysfs_ops,
3288 .default_attrs = md_default_attrs,
3289};
3290
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291int mdp_major = 0;
3292
3293static struct kobject *md_probe(dev_t dev, int *part, void *data)
3294{
Arjan van de Ven48c9c272006-03-27 01:18:20 -08003295 static DEFINE_MUTEX(disks_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296 mddev_t *mddev = mddev_find(dev);
3297 struct gendisk *disk;
3298 int partitioned = (MAJOR(dev) != MD_MAJOR);
3299 int shift = partitioned ? MdpMinorShift : 0;
3300 int unit = MINOR(dev) >> shift;
Greg Kroah-Hartman3830c622007-12-17 15:54:39 -04003301 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302
3303 if (!mddev)
3304 return NULL;
3305
Arjan van de Ven48c9c272006-03-27 01:18:20 -08003306 mutex_lock(&disks_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307 if (mddev->gendisk) {
Arjan van de Ven48c9c272006-03-27 01:18:20 -08003308 mutex_unlock(&disks_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 mddev_put(mddev);
3310 return NULL;
3311 }
3312 disk = alloc_disk(1 << shift);
3313 if (!disk) {
Arjan van de Ven48c9c272006-03-27 01:18:20 -08003314 mutex_unlock(&disks_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 mddev_put(mddev);
3316 return NULL;
3317 }
3318 disk->major = MAJOR(dev);
3319 disk->first_minor = unit << shift;
Greg Kroah-Hartmance7b0f42005-06-20 21:15:16 -07003320 if (partitioned)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321 sprintf(disk->disk_name, "md_d%d", unit);
Greg Kroah-Hartmance7b0f42005-06-20 21:15:16 -07003322 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 sprintf(disk->disk_name, "md%d", unit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 disk->fops = &md_fops;
3325 disk->private_data = mddev;
3326 disk->queue = mddev->queue;
3327 add_disk(disk);
3328 mddev->gendisk = disk;
Arjan van de Ven48c9c272006-03-27 01:18:20 -08003329 mutex_unlock(&disks_mutex);
Kay Sieversedfaa7c2007-05-21 22:08:01 +02003330 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
Greg Kroah-Hartman3830c622007-12-17 15:54:39 -04003331 "%s", "md");
3332 if (error)
NeilBrown5e55e2f2007-03-26 21:32:14 -08003333 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3334 disk->disk_name);
Greg Kroah-Hartman3830c622007-12-17 15:54:39 -04003335 else
3336 kobject_uevent(&mddev->kobj, KOBJ_ADD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 return NULL;
3338}
3339
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340static void md_safemode_timeout(unsigned long data)
3341{
3342 mddev_t *mddev = (mddev_t *) data;
3343
3344 mddev->safemode = 1;
3345 md_wakeup_thread(mddev->thread);
3346}
3347
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08003348static int start_dirty_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349
3350static int do_md_run(mddev_t * mddev)
3351{
NeilBrown2604b702006-01-06 00:20:36 -08003352 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353 int chunk_size;
3354 struct list_head *tmp;
3355 mdk_rdev_t *rdev;
3356 struct gendisk *disk;
NeilBrown2604b702006-01-06 00:20:36 -08003357 struct mdk_personality *pers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 char b[BDEVNAME_SIZE];
3359
NeilBrowna757e642005-04-16 15:26:42 -07003360 if (list_empty(&mddev->disks))
3361 /* cannot run an array with no devices.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363
3364 if (mddev->pers)
3365 return -EBUSY;
3366
3367 /*
3368 * Analyze all RAID superblock(s)
3369 */
NeilBrown1ec4a932008-02-06 01:39:53 -08003370 if (!mddev->raid_disks) {
3371 if (!mddev->persistent)
3372 return -EINVAL;
NeilBrowna757e642005-04-16 15:26:42 -07003373 analyze_sbs(mddev);
NeilBrown1ec4a932008-02-06 01:39:53 -08003374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375
3376 chunk_size = mddev->chunk_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377
NeilBrown2604b702006-01-06 00:20:36 -08003378 if (chunk_size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 if (chunk_size > MAX_CHUNK_SIZE) {
3380 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3381 chunk_size, MAX_CHUNK_SIZE);
3382 return -EINVAL;
3383 }
3384 /*
3385 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3386 */
3387 if ( (1 << ffz(~chunk_size)) != chunk_size) {
NeilBrowna757e642005-04-16 15:26:42 -07003388 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389 return -EINVAL;
3390 }
3391 if (chunk_size < PAGE_SIZE) {
3392 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3393 chunk_size, PAGE_SIZE);
3394 return -EINVAL;
3395 }
3396
3397 /* devices must have minimum size of one chunk */
NeilBrownd089c6a2008-02-06 01:39:59 -08003398 rdev_for_each(rdev, tmp, mddev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08003399 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 continue;
3401 if (rdev->size < chunk_size / 1024) {
3402 printk(KERN_WARNING
3403 "md: Dev %s smaller than chunk_size:"
3404 " %lluk < %dk\n",
3405 bdevname(rdev->bdev,b),
3406 (unsigned long long)rdev->size,
3407 chunk_size / 1024);
3408 return -EINVAL;
3409 }
3410 }
3411 }
3412
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413#ifdef CONFIG_KMOD
NeilBrownd9d166c2006-01-06 00:20:51 -08003414 if (mddev->level != LEVEL_NONE)
3415 request_module("md-level-%d", mddev->level);
3416 else if (mddev->clevel[0])
3417 request_module("md-%s", mddev->clevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418#endif
3419
3420 /*
3421 * Drop all container device buffers, from now on
3422 * the only valid external interface is through the md
3423 * device.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 */
NeilBrownd089c6a2008-02-06 01:39:59 -08003425 rdev_for_each(rdev, tmp, mddev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08003426 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 continue;
3428 sync_blockdev(rdev->bdev);
Peter Zijlstraf98393a2007-05-06 14:49:54 -07003429 invalidate_bdev(rdev->bdev);
NeilBrownf0d76d72007-07-17 04:06:12 -07003430
3431 /* perform some consistency tests on the device.
3432 * We don't want the data to overlap the metadata,
3433 * Internal Bitmap issues has handled elsewhere.
3434 */
3435 if (rdev->data_offset < rdev->sb_offset) {
3436 if (mddev->size &&
3437 rdev->data_offset + mddev->size*2
3438 > rdev->sb_offset*2) {
3439 printk("md: %s: data overlaps metadata\n",
3440 mdname(mddev));
3441 return -EINVAL;
3442 }
3443 } else {
3444 if (rdev->sb_offset*2 + rdev->sb_size/512
3445 > rdev->data_offset) {
3446 printk("md: %s: metadata overlaps data\n",
3447 mdname(mddev));
3448 return -EINVAL;
3449 }
3450 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451 }
3452
3453 md_probe(mddev->unit, NULL, NULL);
3454 disk = mddev->gendisk;
3455 if (!disk)
3456 return -ENOMEM;
3457
3458 spin_lock(&pers_lock);
NeilBrownd9d166c2006-01-06 00:20:51 -08003459 pers = find_pers(mddev->level, mddev->clevel);
NeilBrown2604b702006-01-06 00:20:36 -08003460 if (!pers || !try_module_get(pers->owner)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 spin_unlock(&pers_lock);
NeilBrownd9d166c2006-01-06 00:20:51 -08003462 if (mddev->level != LEVEL_NONE)
3463 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3464 mddev->level);
3465 else
3466 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3467 mddev->clevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003468 return -EINVAL;
3469 }
NeilBrown2604b702006-01-06 00:20:36 -08003470 mddev->pers = pers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 spin_unlock(&pers_lock);
NeilBrownd9d166c2006-01-06 00:20:51 -08003472 mddev->level = pers->level;
3473 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474
NeilBrownf6705572006-03-27 01:18:11 -08003475 if (mddev->reshape_position != MaxSector &&
NeilBrown63c70c42006-03-27 01:18:13 -08003476 pers->start_reshape == NULL) {
NeilBrownf6705572006-03-27 01:18:11 -08003477 /* This personality cannot handle reshaping... */
3478 mddev->pers = NULL;
3479 module_put(pers->owner);
3480 return -EINVAL;
3481 }
3482
NeilBrown7dd5e7c32007-02-28 20:11:35 -08003483 if (pers->sync_request) {
3484 /* Warn if this is a potentially silly
3485 * configuration.
3486 */
3487 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3488 mdk_rdev_t *rdev2;
3489 struct list_head *tmp2;
3490 int warned = 0;
NeilBrownd089c6a2008-02-06 01:39:59 -08003491 rdev_for_each(rdev, tmp, mddev) {
3492 rdev_for_each(rdev2, tmp2, mddev) {
NeilBrown7dd5e7c32007-02-28 20:11:35 -08003493 if (rdev < rdev2 &&
3494 rdev->bdev->bd_contains ==
3495 rdev2->bdev->bd_contains) {
3496 printk(KERN_WARNING
3497 "%s: WARNING: %s appears to be"
3498 " on the same physical disk as"
3499 " %s.\n",
3500 mdname(mddev),
3501 bdevname(rdev->bdev,b),
3502 bdevname(rdev2->bdev,b2));
3503 warned = 1;
3504 }
3505 }
3506 }
3507 if (warned)
3508 printk(KERN_WARNING
3509 "True protection against single-disk"
3510 " failure might be compromised.\n");
3511 }
3512
NeilBrown657390d2005-08-26 18:34:16 -07003513 mddev->recovery = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
NeilBrowna9701a32005-11-08 21:39:34 -08003515 mddev->barriers_work = 1;
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08003516 mddev->ok_start_degraded = start_dirty_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517
NeilBrownf91de922005-11-08 21:39:36 -08003518 if (start_readonly)
3519 mddev->ro = 2; /* read-only, but switch on first write */
3520
NeilBrownb15c2e52006-01-06 00:20:16 -08003521 err = mddev->pers->run(mddev);
3522 if (!err && mddev->pers->sync_request) {
3523 err = bitmap_create(mddev);
3524 if (err) {
3525 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3526 mdname(mddev), err);
3527 mddev->pers->stop(mddev);
3528 }
3529 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530 if (err) {
3531 printk(KERN_ERR "md: pers->run() failed ...\n");
3532 module_put(mddev->pers->owner);
3533 mddev->pers = NULL;
NeilBrown32a76272005-06-21 17:17:14 -07003534 bitmap_destroy(mddev);
3535 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536 }
NeilBrown5e55e2f2007-03-26 21:32:14 -08003537 if (mddev->pers->sync_request) {
3538 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3539 printk(KERN_WARNING
3540 "md: cannot register extra attributes for %s\n",
3541 mdname(mddev));
3542 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
NeilBrownfd9d49c2005-11-08 21:39:42 -08003543 mddev->ro = 0;
3544
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 atomic_set(&mddev->writes_pending,0);
3546 mddev->safemode = 0;
3547 mddev->safemode_timer.function = md_safemode_timeout;
3548 mddev->safemode_timer.data = (unsigned long) mddev;
NeilBrown16f17b32006-06-26 00:27:37 -07003549 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 mddev->in_sync = 1;
NeilBrown86e6ffd2005-11-08 21:39:24 -08003551
NeilBrownd089c6a2008-02-06 01:39:59 -08003552 rdev_for_each(rdev, tmp, mddev)
NeilBrown86e6ffd2005-11-08 21:39:24 -08003553 if (rdev->raid_disk >= 0) {
3554 char nm[20];
3555 sprintf(nm, "rd%d", rdev->raid_disk);
NeilBrown5e55e2f2007-03-26 21:32:14 -08003556 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3557 printk("md: cannot register %s for %s\n",
3558 nm, mdname(mddev));
NeilBrown86e6ffd2005-11-08 21:39:24 -08003559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
3561 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3562
NeilBrown850b2b42006-10-03 01:15:46 -07003563 if (mddev->flags)
3564 md_update_sb(mddev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565
3566 set_capacity(disk, mddev->array_size<<1);
3567
3568 /* If we call blk_queue_make_request here, it will
3569 * re-initialise max_sectors etc which may have been
3570 * refined inside -> run. So just set the bits we need to set.
3571 * Most initialisation happended when we called
3572 * blk_queue_make_request(..., md_fail_request)
3573 * earlier.
3574 */
3575 mddev->queue->queuedata = mddev;
3576 mddev->queue->make_request_fn = mddev->pers->make_request;
3577
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003578 /* If there is a partially-recovered drive we need to
3579 * start recovery here. If we leave it to md_check_recovery,
3580 * it will remove the drives and not do the right thing
3581 */
NeilBrown0b8c9de2006-07-10 04:44:15 -07003582 if (mddev->degraded && !mddev->sync_thread) {
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003583 struct list_head *rtmp;
3584 int spares = 0;
NeilBrownd089c6a2008-02-06 01:39:59 -08003585 rdev_for_each(rdev, rtmp, mddev)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003586 if (rdev->raid_disk >= 0 &&
3587 !test_bit(In_sync, &rdev->flags) &&
3588 !test_bit(Faulty, &rdev->flags))
3589 /* complete an interrupted recovery */
3590 spares++;
3591 if (spares && mddev->pers->sync_request) {
3592 mddev->recovery = 0;
3593 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3594 mddev->sync_thread = md_register_thread(md_do_sync,
3595 mddev,
3596 "%s_resync");
3597 if (!mddev->sync_thread) {
3598 printk(KERN_ERR "%s: could not start resync"
3599 " thread...\n",
3600 mdname(mddev));
3601 /* leave the spares where they are, it shouldn't hurt */
3602 mddev->recovery = 0;
NeilBrown0b8c9de2006-07-10 04:44:15 -07003603 }
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003604 }
3605 }
NeilBrown0b8c9de2006-07-10 04:44:15 -07003606 md_wakeup_thread(mddev->thread);
3607 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003608
Linus Torvalds44ce62942007-05-09 18:51:36 -07003609 mddev->changed = 1;
NeilBrownd7603b72006-01-06 00:20:30 -08003610 md_new_event(mddev);
Kay Sieversedfaa7c2007-05-21 22:08:01 +02003611 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 return 0;
3613}
3614
3615static int restart_array(mddev_t *mddev)
3616{
3617 struct gendisk *disk = mddev->gendisk;
3618 int err;
3619
3620 /*
3621 * Complain if it has no devices
3622 */
3623 err = -ENXIO;
3624 if (list_empty(&mddev->disks))
3625 goto out;
3626
3627 if (mddev->pers) {
3628 err = -EBUSY;
3629 if (!mddev->ro)
3630 goto out;
3631
3632 mddev->safemode = 0;
3633 mddev->ro = 0;
3634 set_disk_ro(disk, 0);
3635
3636 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3637 mdname(mddev));
3638 /*
3639 * Kick recovery or resync if necessary
3640 */
3641 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3642 md_wakeup_thread(mddev->thread);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003643 md_wakeup_thread(mddev->sync_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644 err = 0;
NeilBrown9e653b62006-06-26 00:27:58 -07003645 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647
3648out:
3649 return err;
3650}
3651
NeilBrownacc55e22006-06-26 00:27:47 -07003652/* similar to deny_write_access, but accounts for our holding a reference
3653 * to the file ourselves */
3654static int deny_bitmap_write_access(struct file * file)
3655{
3656 struct inode *inode = file->f_mapping->host;
3657
3658 spin_lock(&inode->i_lock);
3659 if (atomic_read(&inode->i_writecount) > 1) {
3660 spin_unlock(&inode->i_lock);
3661 return -ETXTBSY;
3662 }
3663 atomic_set(&inode->i_writecount, -1);
3664 spin_unlock(&inode->i_lock);
3665
3666 return 0;
3667}
3668
3669static void restore_bitmap_write_access(struct file *file)
3670{
3671 struct inode *inode = file->f_mapping->host;
3672
3673 spin_lock(&inode->i_lock);
3674 atomic_set(&inode->i_writecount, 1);
3675 spin_unlock(&inode->i_lock);
3676}
3677
NeilBrown9e653b62006-06-26 00:27:58 -07003678/* mode:
3679 * 0 - completely stop and dis-assemble array
3680 * 1 - switch to readonly
3681 * 2 - stop but do not disassemble array
3682 */
3683static int do_md_stop(mddev_t * mddev, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684{
3685 int err = 0;
3686 struct gendisk *disk = mddev->gendisk;
3687
3688 if (mddev->pers) {
3689 if (atomic_read(&mddev->active)>2) {
3690 printk("md: %s still in use.\n",mdname(mddev));
3691 return -EBUSY;
3692 }
3693
3694 if (mddev->sync_thread) {
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003695 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3697 md_unregister_thread(mddev->sync_thread);
3698 mddev->sync_thread = NULL;
3699 }
3700
3701 del_timer_sync(&mddev->safemode_timer);
3702
3703 invalidate_partition(disk, 0);
3704
NeilBrown9e653b62006-06-26 00:27:58 -07003705 switch(mode) {
3706 case 1: /* readonly */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 err = -ENXIO;
NeilBrownf91de922005-11-08 21:39:36 -08003708 if (mddev->ro==1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709 goto out;
3710 mddev->ro = 1;
NeilBrown9e653b62006-06-26 00:27:58 -07003711 break;
3712 case 0: /* disassemble */
3713 case 2: /* stop */
NeilBrown6b8b3e82005-08-04 12:53:35 -07003714 bitmap_flush(mddev);
NeilBrowna9701a32005-11-08 21:39:34 -08003715 md_super_wait(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 if (mddev->ro)
3717 set_disk_ro(disk, 0);
3718 blk_queue_make_request(mddev->queue, md_fail_request);
3719 mddev->pers->stop(mddev);
NeilBrownd1b53802007-02-28 20:11:42 -08003720 mddev->queue->merge_bvec_fn = NULL;
3721 mddev->queue->unplug_fn = NULL;
NeilBrown041ae522007-03-26 21:32:14 -08003722 mddev->queue->backing_dev_info.congested_fn = NULL;
NeilBrown411036f2005-11-08 21:39:40 -08003723 if (mddev->pers->sync_request)
3724 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3725
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726 module_put(mddev->pers->owner);
3727 mddev->pers = NULL;
Dan Williams4f54b0e2008-05-23 13:04:37 -07003728 /* tell userspace to handle 'inactive' */
3729 sysfs_notify(&mddev->kobj, NULL, "array_state");
NeilBrown0d4ca602006-12-10 02:20:44 -08003730
3731 set_capacity(disk, 0);
Linus Torvalds44ce62942007-05-09 18:51:36 -07003732 mddev->changed = 1;
NeilBrown0d4ca602006-12-10 02:20:44 -08003733
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 if (mddev->ro)
3735 mddev->ro = 0;
3736 }
NeilBrown850b2b42006-10-03 01:15:46 -07003737 if (!mddev->in_sync || mddev->flags) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 /* mark array as shutdown cleanly */
3739 mddev->in_sync = 1;
NeilBrown850b2b42006-10-03 01:15:46 -07003740 md_update_sb(mddev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 }
NeilBrown9e653b62006-06-26 00:27:58 -07003742 if (mode == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 set_disk_ro(disk, 1);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003744 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745 }
NeilBrown32a76272005-06-21 17:17:14 -07003746
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 /*
3748 * Free resources if final stop
3749 */
NeilBrown9e653b62006-06-26 00:27:58 -07003750 if (mode == 0) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08003751 mdk_rdev_t *rdev;
3752 struct list_head *tmp;
NeilBrown0d4ca602006-12-10 02:20:44 -08003753
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3755
NeilBrown978f9462006-02-02 14:28:05 -08003756 bitmap_destroy(mddev);
3757 if (mddev->bitmap_file) {
NeilBrownacc55e22006-06-26 00:27:47 -07003758 restore_bitmap_write_access(mddev->bitmap_file);
NeilBrown978f9462006-02-02 14:28:05 -08003759 fput(mddev->bitmap_file);
3760 mddev->bitmap_file = NULL;
3761 }
3762 mddev->bitmap_offset = 0;
3763
NeilBrownd089c6a2008-02-06 01:39:59 -08003764 rdev_for_each(rdev, tmp, mddev)
NeilBrown86e6ffd2005-11-08 21:39:24 -08003765 if (rdev->raid_disk >= 0) {
3766 char nm[20];
3767 sprintf(nm, "rd%d", rdev->raid_disk);
3768 sysfs_remove_link(&mddev->kobj, nm);
3769 }
3770
NeilBrown177a99b2008-02-06 01:39:56 -08003771 /* make sure all md_delayed_delete calls have finished */
NeilBrown5792a282007-04-04 19:08:18 -07003772 flush_scheduled_work();
3773
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774 export_array(mddev);
3775
3776 mddev->array_size = 0;
NeilBrown9e653b62006-06-26 00:27:58 -07003777 mddev->size = 0;
3778 mddev->raid_disks = 0;
NeilBrowna94213b2006-06-26 00:28:00 -07003779 mddev->recovery_cp = 0;
NeilBrownc6207272008-02-06 01:39:52 -08003780 mddev->resync_max = MaxSector;
NeilBrown08a02ec2007-05-09 02:35:38 -07003781 mddev->reshape_position = MaxSector;
NeilBrowne6910632008-02-06 01:39:51 -08003782 mddev->external = 0;
NeilBrown1ec4a932008-02-06 01:39:53 -08003783 mddev->persistent = 0;
NeilBrownd897dbf2008-04-30 00:52:29 -07003784 mddev->level = LEVEL_NONE;
3785 mddev->clevel[0] = 0;
3786 mddev->flags = 0;
3787 mddev->ro = 0;
3788 mddev->metadata_type[0] = 0;
3789 mddev->chunk_size = 0;
3790 mddev->ctime = mddev->utime = 0;
3791 mddev->layout = 0;
3792 mddev->max_disks = 0;
3793 mddev->events = 0;
3794 mddev->delta_disks = 0;
3795 mddev->new_level = LEVEL_NONE;
3796 mddev->new_layout = 0;
3797 mddev->new_chunk = 0;
3798 mddev->curr_resync = 0;
3799 mddev->resync_mismatches = 0;
3800 mddev->suspend_lo = mddev->suspend_hi = 0;
3801 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3802 mddev->recovery = 0;
3803 mddev->in_sync = 0;
3804 mddev->changed = 0;
3805 mddev->degraded = 0;
3806 mddev->barriers_work = 0;
3807 mddev->safemode = 0;
NeilBrown9e653b62006-06-26 00:27:58 -07003808
NeilBrowna8a55c32006-06-26 00:27:39 -07003809 } else if (mddev->pers)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3811 mdname(mddev));
3812 err = 0;
NeilBrownd7603b72006-01-06 00:20:30 -08003813 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814out:
3815 return err;
3816}
3817
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08003818#ifndef MODULE
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819static void autorun_array(mddev_t *mddev)
3820{
3821 mdk_rdev_t *rdev;
3822 struct list_head *tmp;
3823 int err;
3824
NeilBrowna757e642005-04-16 15:26:42 -07003825 if (list_empty(&mddev->disks))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827
3828 printk(KERN_INFO "md: running: ");
3829
NeilBrownd089c6a2008-02-06 01:39:59 -08003830 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831 char b[BDEVNAME_SIZE];
3832 printk("<%s>", bdevname(rdev->bdev,b));
3833 }
3834 printk("\n");
3835
3836 err = do_md_run (mddev);
3837 if (err) {
3838 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3839 do_md_stop (mddev, 0);
3840 }
3841}
3842
3843/*
3844 * lets try to run arrays based on all disks that have arrived
3845 * until now. (those are in pending_raid_disks)
3846 *
3847 * the method: pick the first pending disk, collect all disks with
3848 * the same UUID, remove all from the pending list and put them into
3849 * the 'same_array' list. Then order this list based on superblock
3850 * update time (freshest comes first), kick out 'old' disks and
3851 * compare superblocks. If everything's fine then run it.
3852 *
3853 * If "unit" is allocated, then bump its reference count
3854 */
3855static void autorun_devices(int part)
3856{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857 struct list_head *tmp;
3858 mdk_rdev_t *rdev0, *rdev;
3859 mddev_t *mddev;
3860 char b[BDEVNAME_SIZE];
3861
3862 printk(KERN_INFO "md: autorun ...\n");
3863 while (!list_empty(&pending_raid_disks)) {
NeilBrowne8703fe2006-10-03 01:15:59 -07003864 int unit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 dev_t dev;
NeilBrownad01c9e2006-03-27 01:18:07 -08003866 LIST_HEAD(candidates);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867 rdev0 = list_entry(pending_raid_disks.next,
3868 mdk_rdev_t, same_set);
3869
3870 printk(KERN_INFO "md: considering %s ...\n",
3871 bdevname(rdev0->bdev,b));
3872 INIT_LIST_HEAD(&candidates);
NeilBrown73c34432008-02-06 01:39:59 -08003873 rdev_for_each_list(rdev, tmp, pending_raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 if (super_90_load(rdev, rdev0, 0) >= 0) {
3875 printk(KERN_INFO "md: adding %s ...\n",
3876 bdevname(rdev->bdev,b));
3877 list_move(&rdev->same_set, &candidates);
3878 }
3879 /*
3880 * now we have a set of devices, with all of them having
3881 * mostly sane superblocks. It's time to allocate the
3882 * mddev.
3883 */
NeilBrowne8703fe2006-10-03 01:15:59 -07003884 if (part) {
3885 dev = MKDEV(mdp_major,
3886 rdev0->preferred_minor << MdpMinorShift);
3887 unit = MINOR(dev) >> MdpMinorShift;
3888 } else {
3889 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3890 unit = MINOR(dev);
3891 }
3892 if (rdev0->preferred_minor != unit) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3894 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3895 break;
3896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003897
3898 md_probe(dev, NULL, NULL);
3899 mddev = mddev_find(dev);
3900 if (!mddev) {
3901 printk(KERN_ERR
3902 "md: cannot allocate memory for md drive.\n");
3903 break;
3904 }
3905 if (mddev_lock(mddev))
3906 printk(KERN_WARNING "md: %s locked, cannot run\n",
3907 mdname(mddev));
3908 else if (mddev->raid_disks || mddev->major_version
3909 || !list_empty(&mddev->disks)) {
3910 printk(KERN_WARNING
3911 "md: %s already running, cannot run %s\n",
3912 mdname(mddev), bdevname(rdev0->bdev,b));
3913 mddev_unlock(mddev);
3914 } else {
3915 printk(KERN_INFO "md: created %s\n", mdname(mddev));
NeilBrown1ec4a932008-02-06 01:39:53 -08003916 mddev->persistent = 1;
NeilBrown73c34432008-02-06 01:39:59 -08003917 rdev_for_each_list(rdev, tmp, candidates) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918 list_del_init(&rdev->same_set);
3919 if (bind_rdev_to_array(rdev, mddev))
3920 export_rdev(rdev);
3921 }
3922 autorun_array(mddev);
3923 mddev_unlock(mddev);
3924 }
3925 /* on success, candidates will be empty, on error
3926 * it won't...
3927 */
NeilBrown73c34432008-02-06 01:39:59 -08003928 rdev_for_each_list(rdev, tmp, candidates)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003929 export_rdev(rdev);
3930 mddev_put(mddev);
3931 }
3932 printk(KERN_INFO "md: ... autorun DONE.\n");
3933}
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08003934#endif /* !MODULE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936static int get_version(void __user * arg)
3937{
3938 mdu_version_t ver;
3939
3940 ver.major = MD_MAJOR_VERSION;
3941 ver.minor = MD_MINOR_VERSION;
3942 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3943
3944 if (copy_to_user(arg, &ver, sizeof(ver)))
3945 return -EFAULT;
3946
3947 return 0;
3948}
3949
3950static int get_array_info(mddev_t * mddev, void __user * arg)
3951{
3952 mdu_array_info_t info;
3953 int nr,working,active,failed,spare;
3954 mdk_rdev_t *rdev;
3955 struct list_head *tmp;
3956
3957 nr=working=active=failed=spare=0;
NeilBrownd089c6a2008-02-06 01:39:59 -08003958 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 nr++;
NeilBrownb2d444d2005-11-08 21:39:31 -08003960 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961 failed++;
3962 else {
3963 working++;
NeilBrownb2d444d2005-11-08 21:39:31 -08003964 if (test_bit(In_sync, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 active++;
3966 else
3967 spare++;
3968 }
3969 }
3970
3971 info.major_version = mddev->major_version;
3972 info.minor_version = mddev->minor_version;
3973 info.patch_version = MD_PATCHLEVEL_VERSION;
3974 info.ctime = mddev->ctime;
3975 info.level = mddev->level;
3976 info.size = mddev->size;
NeilBrown284ae7c2006-02-03 03:03:40 -08003977 if (info.size != mddev->size) /* overflow */
3978 info.size = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 info.nr_disks = nr;
3980 info.raid_disks = mddev->raid_disks;
3981 info.md_minor = mddev->md_minor;
3982 info.not_persistent= !mddev->persistent;
3983
3984 info.utime = mddev->utime;
3985 info.state = 0;
3986 if (mddev->in_sync)
3987 info.state = (1<<MD_SB_CLEAN);
NeilBrown36fa3062005-09-09 16:23:45 -07003988 if (mddev->bitmap && mddev->bitmap_offset)
3989 info.state = (1<<MD_SB_BITMAP_PRESENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990 info.active_disks = active;
3991 info.working_disks = working;
3992 info.failed_disks = failed;
3993 info.spare_disks = spare;
3994
3995 info.layout = mddev->layout;
3996 info.chunk_size = mddev->chunk_size;
3997
3998 if (copy_to_user(arg, &info, sizeof(info)))
3999 return -EFAULT;
4000
4001 return 0;
4002}
4003
viro@ZenIV.linux.org.uk87162a22005-09-09 20:36:43 +01004004static int get_bitmap_file(mddev_t * mddev, void __user * arg)
NeilBrown32a76272005-06-21 17:17:14 -07004005{
4006 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4007 char *ptr, *buf = NULL;
4008 int err = -ENOMEM;
4009
NeilBrown2a2275d2007-01-26 00:57:11 -08004010 md_allow_write(mddev);
4011
NeilBrown32a76272005-06-21 17:17:14 -07004012 file = kmalloc(sizeof(*file), GFP_KERNEL);
4013 if (!file)
4014 goto out;
4015
4016 /* bitmap disabled, zero the first byte and copy out */
4017 if (!mddev->bitmap || !mddev->bitmap->file) {
4018 file->pathname[0] = '\0';
4019 goto copy_out;
4020 }
4021
4022 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4023 if (!buf)
4024 goto out;
4025
Christoph Hellwig6bcfd602008-05-23 13:04:34 -07004026 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4027 if (IS_ERR(ptr))
NeilBrown32a76272005-06-21 17:17:14 -07004028 goto out;
4029
4030 strcpy(file->pathname, ptr);
4031
4032copy_out:
4033 err = 0;
4034 if (copy_to_user(arg, file, sizeof(*file)))
4035 err = -EFAULT;
4036out:
4037 kfree(buf);
4038 kfree(file);
4039 return err;
4040}
4041
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042static int get_disk_info(mddev_t * mddev, void __user * arg)
4043{
4044 mdu_disk_info_t info;
4045 unsigned int nr;
4046 mdk_rdev_t *rdev;
4047
4048 if (copy_from_user(&info, arg, sizeof(info)))
4049 return -EFAULT;
4050
4051 nr = info.number;
4052
4053 rdev = find_rdev_nr(mddev, nr);
4054 if (rdev) {
4055 info.major = MAJOR(rdev->bdev->bd_dev);
4056 info.minor = MINOR(rdev->bdev->bd_dev);
4057 info.raid_disk = rdev->raid_disk;
4058 info.state = 0;
NeilBrownb2d444d2005-11-08 21:39:31 -08004059 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060 info.state |= (1<<MD_DISK_FAULTY);
NeilBrownb2d444d2005-11-08 21:39:31 -08004061 else if (test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062 info.state |= (1<<MD_DISK_ACTIVE);
4063 info.state |= (1<<MD_DISK_SYNC);
4064 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07004065 if (test_bit(WriteMostly, &rdev->flags))
4066 info.state |= (1<<MD_DISK_WRITEMOSTLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067 } else {
4068 info.major = info.minor = 0;
4069 info.raid_disk = -1;
4070 info.state = (1<<MD_DISK_REMOVED);
4071 }
4072
4073 if (copy_to_user(arg, &info, sizeof(info)))
4074 return -EFAULT;
4075
4076 return 0;
4077}
4078
4079static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4080{
4081 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4082 mdk_rdev_t *rdev;
4083 dev_t dev = MKDEV(info->major,info->minor);
4084
4085 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4086 return -EOVERFLOW;
4087
4088 if (!mddev->raid_disks) {
4089 int err;
4090 /* expecting a device which has a superblock */
4091 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4092 if (IS_ERR(rdev)) {
4093 printk(KERN_WARNING
4094 "md: md_import_device returned %ld\n",
4095 PTR_ERR(rdev));
4096 return PTR_ERR(rdev);
4097 }
4098 if (!list_empty(&mddev->disks)) {
4099 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4100 mdk_rdev_t, same_set);
4101 int err = super_types[mddev->major_version]
4102 .load_super(rdev, rdev0, mddev->minor_version);
4103 if (err < 0) {
4104 printk(KERN_WARNING
4105 "md: %s has different UUID to %s\n",
4106 bdevname(rdev->bdev,b),
4107 bdevname(rdev0->bdev,b2));
4108 export_rdev(rdev);
4109 return -EINVAL;
4110 }
4111 }
4112 err = bind_rdev_to_array(rdev, mddev);
4113 if (err)
4114 export_rdev(rdev);
4115 return err;
4116 }
4117
4118 /*
4119 * add_new_disk can be used once the array is assembled
4120 * to add "hot spares". They must already have a superblock
4121 * written
4122 */
4123 if (mddev->pers) {
4124 int err;
4125 if (!mddev->pers->hot_add_disk) {
4126 printk(KERN_WARNING
4127 "%s: personality does not support diskops!\n",
4128 mdname(mddev));
4129 return -EINVAL;
4130 }
NeilBrown7b1e35f2005-09-09 16:23:50 -07004131 if (mddev->persistent)
4132 rdev = md_import_device(dev, mddev->major_version,
4133 mddev->minor_version);
4134 else
4135 rdev = md_import_device(dev, -1, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136 if (IS_ERR(rdev)) {
4137 printk(KERN_WARNING
4138 "md: md_import_device returned %ld\n",
4139 PTR_ERR(rdev));
4140 return PTR_ERR(rdev);
4141 }
NeilBrown41158c72005-06-21 17:17:25 -07004142 /* set save_raid_disk if appropriate */
4143 if (!mddev->persistent) {
4144 if (info->state & (1<<MD_DISK_SYNC) &&
4145 info->raid_disk < mddev->raid_disks)
4146 rdev->raid_disk = info->raid_disk;
4147 else
4148 rdev->raid_disk = -1;
4149 } else
4150 super_types[mddev->major_version].
4151 validate_super(mddev, rdev);
4152 rdev->saved_raid_disk = rdev->raid_disk;
4153
NeilBrownb2d444d2005-11-08 21:39:31 -08004154 clear_bit(In_sync, &rdev->flags); /* just to be sure */
NeilBrown8ddf9ef2005-09-09 16:23:45 -07004155 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4156 set_bit(WriteMostly, &rdev->flags);
4157
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 rdev->raid_disk = -1;
4159 err = bind_rdev_to_array(rdev, mddev);
NeilBrown7c7546c2006-06-26 00:27:41 -07004160 if (!err && !mddev->pers->hot_remove_disk) {
4161 /* If there is hot_add_disk but no hot_remove_disk
4162 * then added disks for geometry changes,
4163 * and should be added immediately.
4164 */
4165 super_types[mddev->major_version].
4166 validate_super(mddev, rdev);
4167 err = mddev->pers->hot_add_disk(mddev, rdev);
4168 if (err)
4169 unbind_rdev_from_array(rdev);
4170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171 if (err)
4172 export_rdev(rdev);
NeilBrownc3617772005-06-21 17:17:10 -07004173
NeilBrown17571282006-12-10 02:20:52 -08004174 md_update_sb(mddev, 1);
NeilBrownc3617772005-06-21 17:17:10 -07004175 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrown005eca52005-08-22 13:11:08 -07004176 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177 return err;
4178 }
4179
4180 /* otherwise, add_new_disk is only allowed
4181 * for major_version==0 superblocks
4182 */
4183 if (mddev->major_version != 0) {
4184 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4185 mdname(mddev));
4186 return -EINVAL;
4187 }
4188
4189 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4190 int err;
4191 rdev = md_import_device (dev, -1, 0);
4192 if (IS_ERR(rdev)) {
4193 printk(KERN_WARNING
4194 "md: error, md_import_device() returned %ld\n",
4195 PTR_ERR(rdev));
4196 return PTR_ERR(rdev);
4197 }
4198 rdev->desc_nr = info->number;
4199 if (info->raid_disk < mddev->raid_disks)
4200 rdev->raid_disk = info->raid_disk;
4201 else
4202 rdev->raid_disk = -1;
4203
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204 if (rdev->raid_disk < mddev->raid_disks)
NeilBrownb2d444d2005-11-08 21:39:31 -08004205 if (info->state & (1<<MD_DISK_SYNC))
4206 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004207
NeilBrown8ddf9ef2005-09-09 16:23:45 -07004208 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4209 set_bit(WriteMostly, &rdev->flags);
4210
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211 if (!mddev->persistent) {
4212 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4213 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4214 } else
4215 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4216 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4217
NeilBrown2bf071b2006-01-06 00:20:55 -08004218 err = bind_rdev_to_array(rdev, mddev);
4219 if (err) {
4220 export_rdev(rdev);
4221 return err;
4222 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004223 }
4224
4225 return 0;
4226}
4227
4228static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4229{
4230 char b[BDEVNAME_SIZE];
4231 mdk_rdev_t *rdev;
4232
4233 if (!mddev->pers)
4234 return -ENODEV;
4235
4236 rdev = find_rdev(mddev, dev);
4237 if (!rdev)
4238 return -ENXIO;
4239
4240 if (rdev->raid_disk >= 0)
4241 goto busy;
4242
4243 kick_rdev_from_array(rdev);
NeilBrown850b2b42006-10-03 01:15:46 -07004244 md_update_sb(mddev, 1);
NeilBrownd7603b72006-01-06 00:20:30 -08004245 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246
4247 return 0;
4248busy:
Nick Andrewfdefa4d2008-04-21 22:42:58 +00004249 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004250 bdevname(rdev->bdev,b), mdname(mddev));
4251 return -EBUSY;
4252}
4253
4254static int hot_add_disk(mddev_t * mddev, dev_t dev)
4255{
4256 char b[BDEVNAME_SIZE];
4257 int err;
4258 unsigned int size;
4259 mdk_rdev_t *rdev;
4260
4261 if (!mddev->pers)
4262 return -ENODEV;
4263
4264 if (mddev->major_version != 0) {
4265 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4266 " version-0 superblocks.\n",
4267 mdname(mddev));
4268 return -EINVAL;
4269 }
4270 if (!mddev->pers->hot_add_disk) {
4271 printk(KERN_WARNING
4272 "%s: personality does not support diskops!\n",
4273 mdname(mddev));
4274 return -EINVAL;
4275 }
4276
4277 rdev = md_import_device (dev, -1, 0);
4278 if (IS_ERR(rdev)) {
4279 printk(KERN_WARNING
4280 "md: error, md_import_device() returned %ld\n",
4281 PTR_ERR(rdev));
4282 return -EINVAL;
4283 }
4284
4285 if (mddev->persistent)
4286 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4287 else
4288 rdev->sb_offset =
4289 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4290
4291 size = calc_dev_size(rdev, mddev->chunk_size);
4292 rdev->size = size;
4293
NeilBrownb2d444d2005-11-08 21:39:31 -08004294 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295 printk(KERN_WARNING
4296 "md: can not hot-add faulty %s disk to %s!\n",
4297 bdevname(rdev->bdev,b), mdname(mddev));
4298 err = -EINVAL;
4299 goto abort_export;
4300 }
NeilBrownb2d444d2005-11-08 21:39:31 -08004301 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004302 rdev->desc_nr = -1;
NeilBrown58427302006-10-06 00:44:04 -07004303 rdev->saved_raid_disk = -1;
NeilBrown2bf071b2006-01-06 00:20:55 -08004304 err = bind_rdev_to_array(rdev, mddev);
4305 if (err)
4306 goto abort_export;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004307
4308 /*
4309 * The rest should better be atomic, we can have disk failures
4310 * noticed in interrupt contexts ...
4311 */
4312
4313 if (rdev->desc_nr == mddev->max_disks) {
4314 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4315 mdname(mddev));
4316 err = -EBUSY;
4317 goto abort_unbind_export;
4318 }
4319
4320 rdev->raid_disk = -1;
4321
NeilBrown850b2b42006-10-03 01:15:46 -07004322 md_update_sb(mddev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004323
4324 /*
4325 * Kick recovery, maybe this spare has to be added to the
4326 * array immediately.
4327 */
4328 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4329 md_wakeup_thread(mddev->thread);
NeilBrownd7603b72006-01-06 00:20:30 -08004330 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331 return 0;
4332
4333abort_unbind_export:
4334 unbind_rdev_from_array(rdev);
4335
4336abort_export:
4337 export_rdev(rdev);
4338 return err;
4339}
4340
NeilBrown32a76272005-06-21 17:17:14 -07004341static int set_bitmap_file(mddev_t *mddev, int fd)
4342{
4343 int err;
4344
NeilBrown36fa3062005-09-09 16:23:45 -07004345 if (mddev->pers) {
4346 if (!mddev->pers->quiesce)
4347 return -EBUSY;
4348 if (mddev->recovery || mddev->sync_thread)
4349 return -EBUSY;
4350 /* we should be able to change the bitmap.. */
NeilBrown32a76272005-06-21 17:17:14 -07004351 }
4352
NeilBrown36fa3062005-09-09 16:23:45 -07004353
4354 if (fd >= 0) {
4355 if (mddev->bitmap)
4356 return -EEXIST; /* cannot add when bitmap is present */
4357 mddev->bitmap_file = fget(fd);
4358
4359 if (mddev->bitmap_file == NULL) {
4360 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4361 mdname(mddev));
4362 return -EBADF;
4363 }
4364
4365 err = deny_bitmap_write_access(mddev->bitmap_file);
4366 if (err) {
4367 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4368 mdname(mddev));
4369 fput(mddev->bitmap_file);
4370 mddev->bitmap_file = NULL;
4371 return err;
4372 }
NeilBrowna654b9d82005-06-21 17:17:27 -07004373 mddev->bitmap_offset = 0; /* file overrides offset */
NeilBrown36fa3062005-09-09 16:23:45 -07004374 } else if (mddev->bitmap == NULL)
4375 return -ENOENT; /* cannot remove what isn't there */
4376 err = 0;
4377 if (mddev->pers) {
4378 mddev->pers->quiesce(mddev, 1);
4379 if (fd >= 0)
4380 err = bitmap_create(mddev);
NeilBrownd7375ab2006-06-26 00:27:43 -07004381 if (fd < 0 || err) {
NeilBrown36fa3062005-09-09 16:23:45 -07004382 bitmap_destroy(mddev);
NeilBrownd7375ab2006-06-26 00:27:43 -07004383 fd = -1; /* make sure to put the file */
4384 }
NeilBrown36fa3062005-09-09 16:23:45 -07004385 mddev->pers->quiesce(mddev, 0);
NeilBrownd7375ab2006-06-26 00:27:43 -07004386 }
4387 if (fd < 0) {
NeilBrownacc55e22006-06-26 00:27:47 -07004388 if (mddev->bitmap_file) {
4389 restore_bitmap_write_access(mddev->bitmap_file);
NeilBrown36fa3062005-09-09 16:23:45 -07004390 fput(mddev->bitmap_file);
NeilBrownacc55e22006-06-26 00:27:47 -07004391 }
NeilBrown36fa3062005-09-09 16:23:45 -07004392 mddev->bitmap_file = NULL;
4393 }
4394
NeilBrown32a76272005-06-21 17:17:14 -07004395 return err;
4396}
4397
Linus Torvalds1da177e2005-04-16 15:20:36 -07004398/*
4399 * set_array_info is used two different ways
4400 * The original usage is when creating a new array.
4401 * In this usage, raid_disks is > 0 and it together with
4402 * level, size, not_persistent,layout,chunksize determine the
4403 * shape of the array.
4404 * This will always create an array with a type-0.90.0 superblock.
4405 * The newer usage is when assembling an array.
4406 * In this case raid_disks will be 0, and the major_version field is
4407 * use to determine which style super-blocks are to be found on the devices.
4408 * The minor and patch _version numbers are also kept incase the
4409 * super_block handler wishes to interpret them.
4410 */
4411static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4412{
4413
4414 if (info->raid_disks == 0) {
4415 /* just setting version number for superblock loading */
4416 if (info->major_version < 0 ||
Ahmed S. Darwish50511da2007-05-09 02:35:34 -07004417 info->major_version >= ARRAY_SIZE(super_types) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07004418 super_types[info->major_version].name == NULL) {
4419 /* maybe try to auto-load a module? */
4420 printk(KERN_INFO
4421 "md: superblock version %d not known\n",
4422 info->major_version);
4423 return -EINVAL;
4424 }
4425 mddev->major_version = info->major_version;
4426 mddev->minor_version = info->minor_version;
4427 mddev->patch_version = info->patch_version;
NeilBrown3f9d7b02006-12-22 01:11:41 -08004428 mddev->persistent = !info->not_persistent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429 return 0;
4430 }
4431 mddev->major_version = MD_MAJOR_VERSION;
4432 mddev->minor_version = MD_MINOR_VERSION;
4433 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4434 mddev->ctime = get_seconds();
4435
4436 mddev->level = info->level;
NeilBrown17115e02006-01-16 22:14:57 -08004437 mddev->clevel[0] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004438 mddev->size = info->size;
4439 mddev->raid_disks = info->raid_disks;
4440 /* don't set md_minor, it is determined by which /dev/md* was
4441 * openned
4442 */
4443 if (info->state & (1<<MD_SB_CLEAN))
4444 mddev->recovery_cp = MaxSector;
4445 else
4446 mddev->recovery_cp = 0;
4447 mddev->persistent = ! info->not_persistent;
NeilBrowne6910632008-02-06 01:39:51 -08004448 mddev->external = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449
4450 mddev->layout = info->layout;
4451 mddev->chunk_size = info->chunk_size;
4452
4453 mddev->max_disks = MD_SB_DISKS;
4454
NeilBrowne6910632008-02-06 01:39:51 -08004455 if (mddev->persistent)
4456 mddev->flags = 0;
NeilBrown850b2b42006-10-03 01:15:46 -07004457 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004458
NeilBrownb2a27032005-11-28 13:44:12 -08004459 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4460 mddev->bitmap_offset = 0;
4461
NeilBrownf6705572006-03-27 01:18:11 -08004462 mddev->reshape_position = MaxSector;
4463
Linus Torvalds1da177e2005-04-16 15:20:36 -07004464 /*
4465 * Generate a 128 bit UUID
4466 */
4467 get_random_bytes(mddev->uuid, 16);
4468
NeilBrownf6705572006-03-27 01:18:11 -08004469 mddev->new_level = mddev->level;
4470 mddev->new_chunk = mddev->chunk_size;
4471 mddev->new_layout = mddev->layout;
4472 mddev->delta_disks = 0;
4473
Linus Torvalds1da177e2005-04-16 15:20:36 -07004474 return 0;
4475}
4476
NeilBrowna35b0d62006-01-06 00:20:49 -08004477static int update_size(mddev_t *mddev, unsigned long size)
4478{
4479 mdk_rdev_t * rdev;
4480 int rv;
4481 struct list_head *tmp;
NeilBrown8ddeeae2006-03-27 01:18:18 -08004482 int fit = (size == 0);
NeilBrowna35b0d62006-01-06 00:20:49 -08004483
4484 if (mddev->pers->resize == NULL)
4485 return -EINVAL;
4486 /* The "size" is the amount of each device that is used.
4487 * This can only make sense for arrays with redundancy.
4488 * linear and raid0 always use whatever space is available
4489 * We can only consider changing the size if no resync
4490 * or reconstruction is happening, and if the new size
4491 * is acceptable. It must fit before the sb_offset or,
4492 * if that is <data_offset, it must fit before the
4493 * size of each device.
4494 * If size is zero, we find the largest size that fits.
4495 */
4496 if (mddev->sync_thread)
4497 return -EBUSY;
NeilBrownd089c6a2008-02-06 01:39:59 -08004498 rdev_for_each(rdev, tmp, mddev) {
NeilBrowna35b0d62006-01-06 00:20:49 -08004499 sector_t avail;
NeilBrown01ab5662006-10-28 10:38:30 -07004500 avail = rdev->size * 2;
4501
NeilBrowna35b0d62006-01-06 00:20:49 -08004502 if (fit && (size == 0 || size > avail/2))
4503 size = avail/2;
4504 if (avail < ((sector_t)size << 1))
4505 return -ENOSPC;
4506 }
4507 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4508 if (!rv) {
4509 struct block_device *bdev;
4510
4511 bdev = bdget_disk(mddev->gendisk, 0);
4512 if (bdev) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004513 mutex_lock(&bdev->bd_inode->i_mutex);
NeilBrown6d893322006-02-02 14:28:03 -08004514 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08004515 mutex_unlock(&bdev->bd_inode->i_mutex);
NeilBrowna35b0d62006-01-06 00:20:49 -08004516 bdput(bdev);
4517 }
4518 }
4519 return rv;
4520}
4521
NeilBrownda943b992006-01-06 00:20:54 -08004522static int update_raid_disks(mddev_t *mddev, int raid_disks)
4523{
4524 int rv;
4525 /* change the number of raid disks */
NeilBrown63c70c42006-03-27 01:18:13 -08004526 if (mddev->pers->check_reshape == NULL)
NeilBrownda943b992006-01-06 00:20:54 -08004527 return -EINVAL;
4528 if (raid_disks <= 0 ||
4529 raid_disks >= mddev->max_disks)
4530 return -EINVAL;
NeilBrown63c70c42006-03-27 01:18:13 -08004531 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
NeilBrownda943b992006-01-06 00:20:54 -08004532 return -EBUSY;
NeilBrown63c70c42006-03-27 01:18:13 -08004533 mddev->delta_disks = raid_disks - mddev->raid_disks;
4534
4535 rv = mddev->pers->check_reshape(mddev);
NeilBrownda943b992006-01-06 00:20:54 -08004536 return rv;
4537}
4538
4539
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540/*
4541 * update_array_info is used to change the configuration of an
4542 * on-line array.
4543 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4544 * fields in the info are checked against the array.
4545 * Any differences that cannot be handled will cause an error.
4546 * Normally, only one change can be managed at a time.
4547 */
4548static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4549{
4550 int rv = 0;
4551 int cnt = 0;
NeilBrown36fa3062005-09-09 16:23:45 -07004552 int state = 0;
4553
4554 /* calculate expected state,ignoring low bits */
4555 if (mddev->bitmap && mddev->bitmap_offset)
4556 state |= (1 << MD_SB_BITMAP_PRESENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004557
4558 if (mddev->major_version != info->major_version ||
4559 mddev->minor_version != info->minor_version ||
4560/* mddev->patch_version != info->patch_version || */
4561 mddev->ctime != info->ctime ||
4562 mddev->level != info->level ||
4563/* mddev->layout != info->layout || */
4564 !mddev->persistent != info->not_persistent||
NeilBrown36fa3062005-09-09 16:23:45 -07004565 mddev->chunk_size != info->chunk_size ||
4566 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4567 ((state^info->state) & 0xfffffe00)
4568 )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004569 return -EINVAL;
4570 /* Check there is only one change */
NeilBrown284ae7c2006-02-03 03:03:40 -08004571 if (info->size >= 0 && mddev->size != info->size) cnt++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004572 if (mddev->raid_disks != info->raid_disks) cnt++;
4573 if (mddev->layout != info->layout) cnt++;
NeilBrown36fa3062005-09-09 16:23:45 -07004574 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004575 if (cnt == 0) return 0;
4576 if (cnt > 1) return -EINVAL;
4577
4578 if (mddev->layout != info->layout) {
4579 /* Change layout
4580 * we don't need to do anything at the md level, the
4581 * personality will take care of it all.
4582 */
4583 if (mddev->pers->reconfig == NULL)
4584 return -EINVAL;
4585 else
4586 return mddev->pers->reconfig(mddev, info->layout, -1);
4587 }
NeilBrown284ae7c2006-02-03 03:03:40 -08004588 if (info->size >= 0 && mddev->size != info->size)
NeilBrowna35b0d62006-01-06 00:20:49 -08004589 rv = update_size(mddev, info->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004590
NeilBrownda943b992006-01-06 00:20:54 -08004591 if (mddev->raid_disks != info->raid_disks)
4592 rv = update_raid_disks(mddev, info->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004593
NeilBrown36fa3062005-09-09 16:23:45 -07004594 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4595 if (mddev->pers->quiesce == NULL)
4596 return -EINVAL;
4597 if (mddev->recovery || mddev->sync_thread)
4598 return -EBUSY;
4599 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4600 /* add the bitmap */
4601 if (mddev->bitmap)
4602 return -EEXIST;
4603 if (mddev->default_bitmap_offset == 0)
4604 return -EINVAL;
4605 mddev->bitmap_offset = mddev->default_bitmap_offset;
4606 mddev->pers->quiesce(mddev, 1);
4607 rv = bitmap_create(mddev);
4608 if (rv)
4609 bitmap_destroy(mddev);
4610 mddev->pers->quiesce(mddev, 0);
4611 } else {
4612 /* remove the bitmap */
4613 if (!mddev->bitmap)
4614 return -ENOENT;
4615 if (mddev->bitmap->file)
4616 return -EINVAL;
4617 mddev->pers->quiesce(mddev, 1);
4618 bitmap_destroy(mddev);
4619 mddev->pers->quiesce(mddev, 0);
4620 mddev->bitmap_offset = 0;
4621 }
4622 }
NeilBrown850b2b42006-10-03 01:15:46 -07004623 md_update_sb(mddev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004624 return rv;
4625}
4626
4627static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4628{
4629 mdk_rdev_t *rdev;
4630
4631 if (mddev->pers == NULL)
4632 return -ENODEV;
4633
4634 rdev = find_rdev(mddev, dev);
4635 if (!rdev)
4636 return -ENODEV;
4637
4638 md_error(mddev, rdev);
4639 return 0;
4640}
4641
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08004642static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4643{
4644 mddev_t *mddev = bdev->bd_disk->private_data;
4645
4646 geo->heads = 2;
4647 geo->sectors = 4;
4648 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4649 return 0;
4650}
4651
Linus Torvalds1da177e2005-04-16 15:20:36 -07004652static int md_ioctl(struct inode *inode, struct file *file,
4653 unsigned int cmd, unsigned long arg)
4654{
4655 int err = 0;
4656 void __user *argp = (void __user *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004657 mddev_t *mddev = NULL;
4658
4659 if (!capable(CAP_SYS_ADMIN))
4660 return -EACCES;
4661
4662 /*
4663 * Commands dealing with the RAID driver but not any
4664 * particular array:
4665 */
4666 switch (cmd)
4667 {
4668 case RAID_VERSION:
4669 err = get_version(argp);
4670 goto done;
4671
4672 case PRINT_RAID_DEBUG:
4673 err = 0;
4674 md_print_devices();
4675 goto done;
4676
4677#ifndef MODULE
4678 case RAID_AUTORUN:
4679 err = 0;
4680 autostart_arrays(arg);
4681 goto done;
4682#endif
4683 default:;
4684 }
4685
4686 /*
4687 * Commands creating/starting a new array:
4688 */
4689
4690 mddev = inode->i_bdev->bd_disk->private_data;
4691
4692 if (!mddev) {
4693 BUG();
4694 goto abort;
4695 }
4696
Linus Torvalds1da177e2005-04-16 15:20:36 -07004697 err = mddev_lock(mddev);
4698 if (err) {
4699 printk(KERN_INFO
4700 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4701 err, cmd);
4702 goto abort;
4703 }
4704
4705 switch (cmd)
4706 {
4707 case SET_ARRAY_INFO:
4708 {
4709 mdu_array_info_t info;
4710 if (!arg)
4711 memset(&info, 0, sizeof(info));
4712 else if (copy_from_user(&info, argp, sizeof(info))) {
4713 err = -EFAULT;
4714 goto abort_unlock;
4715 }
4716 if (mddev->pers) {
4717 err = update_array_info(mddev, &info);
4718 if (err) {
4719 printk(KERN_WARNING "md: couldn't update"
4720 " array info. %d\n", err);
4721 goto abort_unlock;
4722 }
4723 goto done_unlock;
4724 }
4725 if (!list_empty(&mddev->disks)) {
4726 printk(KERN_WARNING
4727 "md: array %s already has disks!\n",
4728 mdname(mddev));
4729 err = -EBUSY;
4730 goto abort_unlock;
4731 }
4732 if (mddev->raid_disks) {
4733 printk(KERN_WARNING
4734 "md: array %s already initialised!\n",
4735 mdname(mddev));
4736 err = -EBUSY;
4737 goto abort_unlock;
4738 }
4739 err = set_array_info(mddev, &info);
4740 if (err) {
4741 printk(KERN_WARNING "md: couldn't set"
4742 " array info. %d\n", err);
4743 goto abort_unlock;
4744 }
4745 }
4746 goto done_unlock;
4747
4748 default:;
4749 }
4750
4751 /*
4752 * Commands querying/configuring an existing array:
4753 */
NeilBrown32a76272005-06-21 17:17:14 -07004754 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
NeilBrown3f9d7b02006-12-22 01:11:41 -08004755 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
NeilBrowna17184a2008-02-06 01:39:55 -08004756 if ((!mddev->raid_disks && !mddev->external)
4757 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4758 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4759 && cmd != GET_BITMAP_FILE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004760 err = -ENODEV;
4761 goto abort_unlock;
4762 }
4763
4764 /*
4765 * Commands even a read-only array can execute:
4766 */
4767 switch (cmd)
4768 {
4769 case GET_ARRAY_INFO:
4770 err = get_array_info(mddev, argp);
4771 goto done_unlock;
4772
NeilBrown32a76272005-06-21 17:17:14 -07004773 case GET_BITMAP_FILE:
viro@ZenIV.linux.org.uk87162a22005-09-09 20:36:43 +01004774 err = get_bitmap_file(mddev, argp);
NeilBrown32a76272005-06-21 17:17:14 -07004775 goto done_unlock;
4776
Linus Torvalds1da177e2005-04-16 15:20:36 -07004777 case GET_DISK_INFO:
4778 err = get_disk_info(mddev, argp);
4779 goto done_unlock;
4780
4781 case RESTART_ARRAY_RW:
4782 err = restart_array(mddev);
4783 goto done_unlock;
4784
4785 case STOP_ARRAY:
4786 err = do_md_stop (mddev, 0);
4787 goto done_unlock;
4788
4789 case STOP_ARRAY_RO:
4790 err = do_md_stop (mddev, 1);
4791 goto done_unlock;
4792
4793 /*
4794 * We have a problem here : there is no easy way to give a CHS
4795 * virtual geometry. We currently pretend that we have a 2 heads
4796 * 4 sectors (with a BIG number of cylinders...). This drives
4797 * dosfs just mad... ;-)
4798 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004799 }
4800
4801 /*
4802 * The remaining ioctls are changing the state of the
NeilBrownf91de922005-11-08 21:39:36 -08004803 * superblock, so we do not allow them on read-only arrays.
4804 * However non-MD ioctls (e.g. get-size) will still come through
4805 * here and hit the 'default' below, so only disallow
4806 * 'md' ioctls, and switch to rw mode if started auto-readonly.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004807 */
NeilBrownf91de922005-11-08 21:39:36 -08004808 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4809 mddev->ro && mddev->pers) {
4810 if (mddev->ro == 2) {
4811 mddev->ro = 0;
4812 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4813 md_wakeup_thread(mddev->thread);
4814
4815 } else {
4816 err = -EROFS;
4817 goto abort_unlock;
4818 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004819 }
4820
4821 switch (cmd)
4822 {
4823 case ADD_NEW_DISK:
4824 {
4825 mdu_disk_info_t info;
4826 if (copy_from_user(&info, argp, sizeof(info)))
4827 err = -EFAULT;
4828 else
4829 err = add_new_disk(mddev, &info);
4830 goto done_unlock;
4831 }
4832
4833 case HOT_REMOVE_DISK:
4834 err = hot_remove_disk(mddev, new_decode_dev(arg));
4835 goto done_unlock;
4836
4837 case HOT_ADD_DISK:
4838 err = hot_add_disk(mddev, new_decode_dev(arg));
4839 goto done_unlock;
4840
4841 case SET_DISK_FAULTY:
4842 err = set_disk_faulty(mddev, new_decode_dev(arg));
4843 goto done_unlock;
4844
4845 case RUN_ARRAY:
4846 err = do_md_run (mddev);
4847 goto done_unlock;
4848
NeilBrown32a76272005-06-21 17:17:14 -07004849 case SET_BITMAP_FILE:
4850 err = set_bitmap_file(mddev, (int)arg);
4851 goto done_unlock;
4852
Linus Torvalds1da177e2005-04-16 15:20:36 -07004853 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854 err = -EINVAL;
4855 goto abort_unlock;
4856 }
4857
4858done_unlock:
4859abort_unlock:
4860 mddev_unlock(mddev);
4861
4862 return err;
4863done:
4864 if (err)
4865 MD_BUG();
4866abort:
4867 return err;
4868}
4869
4870static int md_open(struct inode *inode, struct file *file)
4871{
4872 /*
4873 * Succeed if we can lock the mddev, which confirms that
4874 * it isn't being stopped right now.
4875 */
4876 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4877 int err;
4878
NeilBrownd63a5a72006-12-08 02:36:17 -08004879 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004880 goto out;
4881
4882 err = 0;
4883 mddev_get(mddev);
4884 mddev_unlock(mddev);
4885
4886 check_disk_change(inode->i_bdev);
4887 out:
4888 return err;
4889}
4890
4891static int md_release(struct inode *inode, struct file * file)
4892{
4893 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4894
Eric Sesterhenn52e5f9d2006-10-03 23:33:23 +02004895 BUG_ON(!mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004896 mddev_put(mddev);
4897
4898 return 0;
4899}
4900
Linus Torvalds44ce62942007-05-09 18:51:36 -07004901static int md_media_changed(struct gendisk *disk)
4902{
4903 mddev_t *mddev = disk->private_data;
4904
4905 return mddev->changed;
4906}
4907
4908static int md_revalidate(struct gendisk *disk)
4909{
4910 mddev_t *mddev = disk->private_data;
4911
4912 mddev->changed = 0;
4913 return 0;
4914}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004915static struct block_device_operations md_fops =
4916{
4917 .owner = THIS_MODULE,
4918 .open = md_open,
4919 .release = md_release,
4920 .ioctl = md_ioctl,
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08004921 .getgeo = md_getgeo,
Linus Torvalds44ce62942007-05-09 18:51:36 -07004922 .media_changed = md_media_changed,
4923 .revalidate_disk= md_revalidate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004924};
4925
Adrian Bunk75c96f82005-05-05 16:16:09 -07004926static int md_thread(void * arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004927{
4928 mdk_thread_t *thread = arg;
4929
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930 /*
4931 * md_thread is a 'system-thread', it's priority should be very
4932 * high. We avoid resource deadlocks individually in each
4933 * raid personality. (RAID5 does preallocation) We also use RR and
4934 * the very same RT priority as kswapd, thus we will never get
4935 * into a priority inversion deadlock.
4936 *
4937 * we definitely have to have equal or higher priority than
4938 * bdflush, otherwise bdflush will deadlock if there are too
4939 * many dirty RAID5 blocks.
4940 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004941
NeilBrown6985c432005-10-19 21:23:47 -07004942 allow_signal(SIGKILL);
NeilBrowna6fb0932005-09-09 16:23:56 -07004943 while (!kthread_should_stop()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004944
NeilBrown93588e22005-11-15 00:09:12 -08004945 /* We need to wait INTERRUPTIBLE so that
4946 * we don't add to the load-average.
4947 * That means we need to be sure no signals are
4948 * pending
4949 */
4950 if (signal_pending(current))
4951 flush_signals(current);
4952
4953 wait_event_interruptible_timeout
4954 (thread->wqueue,
4955 test_bit(THREAD_WAKEUP, &thread->flags)
4956 || kthread_should_stop(),
4957 thread->timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004958
4959 clear_bit(THREAD_WAKEUP, &thread->flags);
4960
NeilBrown787453c2005-11-08 21:39:43 -08004961 thread->run(thread->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004962 }
NeilBrowna6fb0932005-09-09 16:23:56 -07004963
Linus Torvalds1da177e2005-04-16 15:20:36 -07004964 return 0;
4965}
4966
4967void md_wakeup_thread(mdk_thread_t *thread)
4968{
4969 if (thread) {
4970 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4971 set_bit(THREAD_WAKEUP, &thread->flags);
4972 wake_up(&thread->wqueue);
4973 }
4974}
4975
4976mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4977 const char *name)
4978{
4979 mdk_thread_t *thread;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004980
NeilBrown9ffae0c2006-01-06 00:20:32 -08004981 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004982 if (!thread)
4983 return NULL;
4984
Linus Torvalds1da177e2005-04-16 15:20:36 -07004985 init_waitqueue_head(&thread->wqueue);
4986
Linus Torvalds1da177e2005-04-16 15:20:36 -07004987 thread->run = run;
4988 thread->mddev = mddev;
NeilBrown32a76272005-06-21 17:17:14 -07004989 thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrown6985c432005-10-19 21:23:47 -07004990 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
NeilBrowna6fb0932005-09-09 16:23:56 -07004991 if (IS_ERR(thread->tsk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004992 kfree(thread);
4993 return NULL;
4994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004995 return thread;
4996}
4997
Linus Torvalds1da177e2005-04-16 15:20:36 -07004998void md_unregister_thread(mdk_thread_t *thread)
4999{
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07005000 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
NeilBrowna6fb0932005-09-09 16:23:56 -07005001
5002 kthread_stop(thread->tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005003 kfree(thread);
5004}
5005
5006void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5007{
5008 if (!mddev) {
5009 MD_BUG();
5010 return;
5011 }
5012
NeilBrownb2d444d2005-11-08 21:39:31 -08005013 if (!rdev || test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005014 return;
Dan Williams6bfe0b42008-04-30 00:52:32 -07005015
5016 if (mddev->external)
5017 set_bit(Blocked, &rdev->flags);
NeilBrown32a76272005-06-21 17:17:14 -07005018/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005019 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5020 mdname(mddev),
5021 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5022 __builtin_return_address(0),__builtin_return_address(1),
5023 __builtin_return_address(2),__builtin_return_address(3));
NeilBrown32a76272005-06-21 17:17:14 -07005024*/
Andrew Mortond0a0a5e2006-07-10 04:44:20 -07005025 if (!mddev->pers)
5026 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005027 if (!mddev->pers->error_handler)
5028 return;
5029 mddev->pers->error_handler(mddev,rdev);
5030 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5031 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5032 md_wakeup_thread(mddev->thread);
NeilBrownc331eb02006-05-30 21:27:13 -07005033 md_new_event_inintr(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005034}
5035
5036/* seq_file implementation /proc/mdstat */
5037
5038static void status_unused(struct seq_file *seq)
5039{
5040 int i = 0;
5041 mdk_rdev_t *rdev;
5042 struct list_head *tmp;
5043
5044 seq_printf(seq, "unused devices: ");
5045
NeilBrown73c34432008-02-06 01:39:59 -08005046 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005047 char b[BDEVNAME_SIZE];
5048 i++;
5049 seq_printf(seq, "%s ",
5050 bdevname(rdev->bdev,b));
5051 }
5052 if (!i)
5053 seq_printf(seq, "<none>");
5054
5055 seq_printf(seq, "\n");
5056}
5057
5058
5059static void status_resync(struct seq_file *seq, mddev_t * mddev)
5060{
NeilBrown4588b422006-03-27 01:18:04 -08005061 sector_t max_blocks, resync, res;
5062 unsigned long dt, db, rt;
5063 int scale;
5064 unsigned int per_milli;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005065
5066 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5067
5068 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5069 max_blocks = mddev->resync_max_sectors >> 1;
5070 else
5071 max_blocks = mddev->size;
5072
5073 /*
5074 * Should not happen.
5075 */
5076 if (!max_blocks) {
5077 MD_BUG();
5078 return;
5079 }
NeilBrown4588b422006-03-27 01:18:04 -08005080 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5081 * in a sector_t, and (max_blocks>>scale) will fit in a
5082 * u32, as those are the requirements for sector_div.
5083 * Thus 'scale' must be at least 10
5084 */
5085 scale = 10;
5086 if (sizeof(sector_t) > sizeof(unsigned long)) {
5087 while ( max_blocks/2 > (1ULL<<(scale+32)))
5088 scale++;
5089 }
5090 res = (resync>>scale)*1000;
5091 sector_div(res, (u32)((max_blocks>>scale)+1));
5092
5093 per_milli = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005094 {
NeilBrown4588b422006-03-27 01:18:04 -08005095 int i, x = per_milli/50, y = 20-x;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005096 seq_printf(seq, "[");
5097 for (i = 0; i < x; i++)
5098 seq_printf(seq, "=");
5099 seq_printf(seq, ">");
5100 for (i = 0; i < y; i++)
5101 seq_printf(seq, ".");
5102 seq_printf(seq, "] ");
5103 }
NeilBrown4588b422006-03-27 01:18:04 -08005104 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
NeilBrownccfcc3c2006-03-27 01:18:09 -08005105 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5106 "reshape" :
NeilBrown61df9d92006-10-03 01:15:57 -07005107 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5108 "check" :
5109 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5110 "resync" : "recovery"))),
5111 per_milli/10, per_milli % 10,
NeilBrown4588b422006-03-27 01:18:04 -08005112 (unsigned long long) resync,
5113 (unsigned long long) max_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005114
5115 /*
5116 * We do not want to overflow, so the order of operands and
5117 * the * 100 / 100 trick are important. We do a +1 to be
5118 * safe against division by zero. We only estimate anyway.
5119 *
5120 * dt: time from mark until now
5121 * db: blocks written from mark until now
5122 * rt: remaining time
5123 */
5124 dt = ((jiffies - mddev->resync_mark) / HZ);
5125 if (!dt) dt++;
NeilBrownff4e8d92006-07-10 04:44:16 -07005126 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5127 - mddev->resync_mark_cnt;
5128 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005129
5130 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5131
NeilBrownff4e8d92006-07-10 04:44:16 -07005132 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005133}
5134
5135static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5136{
5137 struct list_head *tmp;
5138 loff_t l = *pos;
5139 mddev_t *mddev;
5140
5141 if (l >= 0x10000)
5142 return NULL;
5143 if (!l--)
5144 /* header */
5145 return (void*)1;
5146
5147 spin_lock(&all_mddevs_lock);
5148 list_for_each(tmp,&all_mddevs)
5149 if (!l--) {
5150 mddev = list_entry(tmp, mddev_t, all_mddevs);
5151 mddev_get(mddev);
5152 spin_unlock(&all_mddevs_lock);
5153 return mddev;
5154 }
5155 spin_unlock(&all_mddevs_lock);
5156 if (!l--)
5157 return (void*)2;/* tail */
5158 return NULL;
5159}
5160
5161static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5162{
5163 struct list_head *tmp;
5164 mddev_t *next_mddev, *mddev = v;
5165
5166 ++*pos;
5167 if (v == (void*)2)
5168 return NULL;
5169
5170 spin_lock(&all_mddevs_lock);
5171 if (v == (void*)1)
5172 tmp = all_mddevs.next;
5173 else
5174 tmp = mddev->all_mddevs.next;
5175 if (tmp != &all_mddevs)
5176 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5177 else {
5178 next_mddev = (void*)2;
5179 *pos = 0x10000;
5180 }
5181 spin_unlock(&all_mddevs_lock);
5182
5183 if (v != (void*)1)
5184 mddev_put(mddev);
5185 return next_mddev;
5186
5187}
5188
5189static void md_seq_stop(struct seq_file *seq, void *v)
5190{
5191 mddev_t *mddev = v;
5192
5193 if (mddev && v != (void*)1 && v != (void*)2)
5194 mddev_put(mddev);
5195}
5196
NeilBrownd7603b72006-01-06 00:20:30 -08005197struct mdstat_info {
5198 int event;
5199};
5200
Linus Torvalds1da177e2005-04-16 15:20:36 -07005201static int md_seq_show(struct seq_file *seq, void *v)
5202{
5203 mddev_t *mddev = v;
5204 sector_t size;
5205 struct list_head *tmp2;
5206 mdk_rdev_t *rdev;
NeilBrownd7603b72006-01-06 00:20:30 -08005207 struct mdstat_info *mi = seq->private;
NeilBrown32a76272005-06-21 17:17:14 -07005208 struct bitmap *bitmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005209
5210 if (v == (void*)1) {
NeilBrown2604b702006-01-06 00:20:36 -08005211 struct mdk_personality *pers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005212 seq_printf(seq, "Personalities : ");
5213 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08005214 list_for_each_entry(pers, &pers_list, list)
5215 seq_printf(seq, "[%s] ", pers->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005216
5217 spin_unlock(&pers_lock);
5218 seq_printf(seq, "\n");
NeilBrownd7603b72006-01-06 00:20:30 -08005219 mi->event = atomic_read(&md_event_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005220 return 0;
5221 }
5222 if (v == (void*)2) {
5223 status_unused(seq);
5224 return 0;
5225 }
5226
Ingo Molnar5dc5cf72006-04-20 02:43:23 -07005227 if (mddev_lock(mddev) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005228 return -EINTR;
Ingo Molnar5dc5cf72006-04-20 02:43:23 -07005229
Linus Torvalds1da177e2005-04-16 15:20:36 -07005230 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5231 seq_printf(seq, "%s : %sactive", mdname(mddev),
5232 mddev->pers ? "" : "in");
5233 if (mddev->pers) {
NeilBrownf91de922005-11-08 21:39:36 -08005234 if (mddev->ro==1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005235 seq_printf(seq, " (read-only)");
NeilBrownf91de922005-11-08 21:39:36 -08005236 if (mddev->ro==2)
NeilBrown52720ae2008-03-10 11:43:47 -07005237 seq_printf(seq, " (auto-read-only)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005238 seq_printf(seq, " %s", mddev->pers->name);
5239 }
5240
5241 size = 0;
NeilBrownd089c6a2008-02-06 01:39:59 -08005242 rdev_for_each(rdev, tmp2, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005243 char b[BDEVNAME_SIZE];
5244 seq_printf(seq, " %s[%d]",
5245 bdevname(rdev->bdev,b), rdev->desc_nr);
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005246 if (test_bit(WriteMostly, &rdev->flags))
5247 seq_printf(seq, "(W)");
NeilBrownb2d444d2005-11-08 21:39:31 -08005248 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005249 seq_printf(seq, "(F)");
5250 continue;
NeilBrownb325a322005-09-09 16:24:00 -07005251 } else if (rdev->raid_disk < 0)
5252 seq_printf(seq, "(S)"); /* spare */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005253 size += rdev->size;
5254 }
5255
5256 if (!list_empty(&mddev->disks)) {
5257 if (mddev->pers)
5258 seq_printf(seq, "\n %llu blocks",
5259 (unsigned long long)mddev->array_size);
5260 else
5261 seq_printf(seq, "\n %llu blocks",
5262 (unsigned long long)size);
5263 }
NeilBrown1cd6bf12005-09-09 16:24:00 -07005264 if (mddev->persistent) {
5265 if (mddev->major_version != 0 ||
5266 mddev->minor_version != 90) {
5267 seq_printf(seq," super %d.%d",
5268 mddev->major_version,
5269 mddev->minor_version);
5270 }
NeilBrowne6910632008-02-06 01:39:51 -08005271 } else if (mddev->external)
5272 seq_printf(seq, " super external:%s",
5273 mddev->metadata_type);
5274 else
NeilBrown1cd6bf12005-09-09 16:24:00 -07005275 seq_printf(seq, " super non-persistent");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005276
5277 if (mddev->pers) {
5278 mddev->pers->status (seq, mddev);
5279 seq_printf(seq, "\n ");
NeilBrown8e1b39d2005-11-08 21:39:41 -08005280 if (mddev->pers->sync_request) {
5281 if (mddev->curr_resync > 2) {
5282 status_resync (seq, mddev);
5283 seq_printf(seq, "\n ");
5284 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5285 seq_printf(seq, "\tresync=DELAYED\n ");
5286 else if (mddev->recovery_cp < MaxSector)
5287 seq_printf(seq, "\tresync=PENDING\n ");
5288 }
NeilBrown32a76272005-06-21 17:17:14 -07005289 } else
5290 seq_printf(seq, "\n ");
5291
5292 if ((bitmap = mddev->bitmap)) {
NeilBrown32a76272005-06-21 17:17:14 -07005293 unsigned long chunk_kb;
5294 unsigned long flags;
NeilBrown32a76272005-06-21 17:17:14 -07005295 spin_lock_irqsave(&bitmap->lock, flags);
5296 chunk_kb = bitmap->chunksize >> 10;
5297 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5298 "%lu%s chunk",
5299 bitmap->pages - bitmap->missing_pages,
5300 bitmap->pages,
5301 (bitmap->pages - bitmap->missing_pages)
5302 << (PAGE_SHIFT - 10),
5303 chunk_kb ? chunk_kb : bitmap->chunksize,
5304 chunk_kb ? "KB" : "B");
NeilBrown78d742d2005-06-21 17:17:15 -07005305 if (bitmap->file) {
5306 seq_printf(seq, ", file: ");
Jan Blunckc32c2f62008-02-14 19:38:43 -08005307 seq_path(seq, &bitmap->file->f_path, " \t\n");
NeilBrown32a76272005-06-21 17:17:14 -07005308 }
NeilBrown78d742d2005-06-21 17:17:15 -07005309
NeilBrown32a76272005-06-21 17:17:14 -07005310 seq_printf(seq, "\n");
5311 spin_unlock_irqrestore(&bitmap->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005312 }
5313
5314 seq_printf(seq, "\n");
5315 }
5316 mddev_unlock(mddev);
5317
5318 return 0;
5319}
5320
5321static struct seq_operations md_seq_ops = {
5322 .start = md_seq_start,
5323 .next = md_seq_next,
5324 .stop = md_seq_stop,
5325 .show = md_seq_show,
5326};
5327
5328static int md_seq_open(struct inode *inode, struct file *file)
5329{
5330 int error;
NeilBrownd7603b72006-01-06 00:20:30 -08005331 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5332 if (mi == NULL)
5333 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005334
5335 error = seq_open(file, &md_seq_ops);
NeilBrownd7603b72006-01-06 00:20:30 -08005336 if (error)
5337 kfree(mi);
5338 else {
5339 struct seq_file *p = file->private_data;
5340 p->private = mi;
5341 mi->event = atomic_read(&md_event_count);
5342 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005343 return error;
5344}
5345
NeilBrownd7603b72006-01-06 00:20:30 -08005346static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5347{
5348 struct seq_file *m = filp->private_data;
5349 struct mdstat_info *mi = m->private;
5350 int mask;
5351
5352 poll_wait(filp, &md_event_waiters, wait);
5353
5354 /* always allow read */
5355 mask = POLLIN | POLLRDNORM;
5356
5357 if (mi->event != atomic_read(&md_event_count))
5358 mask |= POLLERR | POLLPRI;
5359 return mask;
5360}
5361
Arjan van de Venfa027c22007-02-12 00:55:33 -08005362static const struct file_operations md_seq_fops = {
Akinobu Mitae24650c2006-10-17 00:09:38 -07005363 .owner = THIS_MODULE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005364 .open = md_seq_open,
5365 .read = seq_read,
5366 .llseek = seq_lseek,
Martin Peschkec3f94b402007-05-09 02:35:35 -07005367 .release = seq_release_private,
NeilBrownd7603b72006-01-06 00:20:30 -08005368 .poll = mdstat_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005369};
5370
NeilBrown2604b702006-01-06 00:20:36 -08005371int register_md_personality(struct mdk_personality *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005372{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005373 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08005374 list_add_tail(&p->list, &pers_list);
5375 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005376 spin_unlock(&pers_lock);
5377 return 0;
5378}
5379
NeilBrown2604b702006-01-06 00:20:36 -08005380int unregister_md_personality(struct mdk_personality *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005381{
NeilBrown2604b702006-01-06 00:20:36 -08005382 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005383 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08005384 list_del_init(&p->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005385 spin_unlock(&pers_lock);
5386 return 0;
5387}
5388
5389static int is_mddev_idle(mddev_t *mddev)
5390{
5391 mdk_rdev_t * rdev;
5392 struct list_head *tmp;
5393 int idle;
NeilBrown713f6ab2007-07-17 04:06:12 -07005394 long curr_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005395
5396 idle = 1;
NeilBrownd089c6a2008-02-06 01:39:59 -08005397 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005398 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
Jens Axboea3623572005-11-01 09:26:16 +01005399 curr_events = disk_stat_read(disk, sectors[0]) +
5400 disk_stat_read(disk, sectors[1]) -
Linus Torvalds1da177e2005-04-16 15:20:36 -07005401 atomic_read(&disk->sync_io);
NeilBrown713f6ab2007-07-17 04:06:12 -07005402 /* sync IO will cause sync_io to increase before the disk_stats
5403 * as sync_io is counted when a request starts, and
5404 * disk_stats is counted when it completes.
5405 * So resync activity will cause curr_events to be smaller than
5406 * when there was no such activity.
5407 * non-sync IO will cause disk_stat to increase without
5408 * increasing sync_io so curr_events will (eventually)
5409 * be larger than it was before. Once it becomes
5410 * substantially larger, the test below will cause
5411 * the array to appear non-idle, and resync will slow
5412 * down.
5413 * If there is a lot of outstanding resync activity when
5414 * we set last_event to curr_events, then all that activity
5415 * completing might cause the array to appear non-idle
5416 * and resync will be slowed down even though there might
5417 * not have been non-resync activity. This will only
5418 * happen once though. 'last_events' will soon reflect
5419 * the state where there is little or no outstanding
5420 * resync requests, and further resync activity will
5421 * always make curr_events less than last_events.
NeilBrownc0e48522005-11-18 01:11:01 -08005422 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07005423 */
NeilBrown713f6ab2007-07-17 04:06:12 -07005424 if (curr_events - rdev->last_events > 4096) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005425 rdev->last_events = curr_events;
5426 idle = 0;
5427 }
5428 }
5429 return idle;
5430}
5431
5432void md_done_sync(mddev_t *mddev, int blocks, int ok)
5433{
5434 /* another "blocks" (512byte) blocks have been synced */
5435 atomic_sub(blocks, &mddev->recovery_active);
5436 wake_up(&mddev->recovery_wait);
5437 if (!ok) {
NeilBrowndfc70642008-05-23 13:04:39 -07005438 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005439 md_wakeup_thread(mddev->thread);
5440 // stop recovery, signal do_sync ....
5441 }
5442}
5443
5444
NeilBrown06d91a52005-06-21 17:17:12 -07005445/* md_write_start(mddev, bi)
5446 * If we need to update some array metadata (e.g. 'active' flag
NeilBrown3d310eb2005-06-21 17:17:26 -07005447 * in superblock) before writing, schedule a superblock update
5448 * and wait for it to complete.
NeilBrown06d91a52005-06-21 17:17:12 -07005449 */
NeilBrown3d310eb2005-06-21 17:17:26 -07005450void md_write_start(mddev_t *mddev, struct bio *bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005451{
NeilBrown06d91a52005-06-21 17:17:12 -07005452 if (bio_data_dir(bi) != WRITE)
NeilBrown3d310eb2005-06-21 17:17:26 -07005453 return;
NeilBrown06d91a52005-06-21 17:17:12 -07005454
NeilBrownf91de922005-11-08 21:39:36 -08005455 BUG_ON(mddev->ro == 1);
5456 if (mddev->ro == 2) {
5457 /* need to switch to read/write */
5458 mddev->ro = 0;
5459 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5460 md_wakeup_thread(mddev->thread);
NeilBrown25156192008-03-04 14:29:32 -08005461 md_wakeup_thread(mddev->sync_thread);
NeilBrownf91de922005-11-08 21:39:36 -08005462 }
NeilBrown06d91a52005-06-21 17:17:12 -07005463 atomic_inc(&mddev->writes_pending);
NeilBrown31a59e32008-04-30 00:52:30 -07005464 if (mddev->safemode == 1)
5465 mddev->safemode = 0;
NeilBrown06d91a52005-06-21 17:17:12 -07005466 if (mddev->in_sync) {
NeilBrowna9701a32005-11-08 21:39:34 -08005467 spin_lock_irq(&mddev->write_lock);
NeilBrown3d310eb2005-06-21 17:17:26 -07005468 if (mddev->in_sync) {
5469 mddev->in_sync = 0;
NeilBrown850b2b42006-10-03 01:15:46 -07005470 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown3d310eb2005-06-21 17:17:26 -07005471 md_wakeup_thread(mddev->thread);
5472 }
NeilBrowna9701a32005-11-08 21:39:34 -08005473 spin_unlock_irq(&mddev->write_lock);
NeilBrown09a44cc2008-05-23 13:04:36 -07005474 sysfs_notify(&mddev->kobj, NULL, "array_state");
NeilBrown06d91a52005-06-21 17:17:12 -07005475 }
NeilBrown09a44cc2008-05-23 13:04:36 -07005476 wait_event(mddev->sb_wait,
5477 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5478 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005479}
5480
5481void md_write_end(mddev_t *mddev)
5482{
5483 if (atomic_dec_and_test(&mddev->writes_pending)) {
5484 if (mddev->safemode == 2)
5485 md_wakeup_thread(mddev->thread);
NeilBrown16f17b32006-06-26 00:27:37 -07005486 else if (mddev->safemode_delay)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005487 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5488 }
5489}
5490
NeilBrown2a2275d2007-01-26 00:57:11 -08005491/* md_allow_write(mddev)
5492 * Calling this ensures that the array is marked 'active' so that writes
5493 * may proceed without blocking. It is important to call this before
5494 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5495 * Must be called with mddev_lock held.
5496 */
5497void md_allow_write(mddev_t *mddev)
5498{
5499 if (!mddev->pers)
5500 return;
5501 if (mddev->ro)
5502 return;
5503
5504 spin_lock_irq(&mddev->write_lock);
5505 if (mddev->in_sync) {
5506 mddev->in_sync = 0;
5507 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5508 if (mddev->safemode_delay &&
5509 mddev->safemode == 0)
5510 mddev->safemode = 1;
5511 spin_unlock_irq(&mddev->write_lock);
5512 md_update_sb(mddev, 0);
NeilBrown09a44cc2008-05-23 13:04:36 -07005513
5514 sysfs_notify(&mddev->kobj, NULL, "array_state");
5515 /* wait for the dirty state to be recorded in the metadata */
5516 wait_event(mddev->sb_wait,
5517 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5518 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
NeilBrown2a2275d2007-01-26 00:57:11 -08005519 } else
5520 spin_unlock_irq(&mddev->write_lock);
5521}
5522EXPORT_SYMBOL_GPL(md_allow_write);
5523
Linus Torvalds1da177e2005-04-16 15:20:36 -07005524#define SYNC_MARKS 10
5525#define SYNC_MARK_STEP (3*HZ)
NeilBrown29269552006-03-27 01:18:10 -08005526void md_do_sync(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005527{
5528 mddev_t *mddev2;
5529 unsigned int currspeed = 0,
5530 window;
NeilBrown57afd892005-06-21 17:17:13 -07005531 sector_t max_sectors,j, io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005532 unsigned long mark[SYNC_MARKS];
5533 sector_t mark_cnt[SYNC_MARKS];
5534 int last_mark,m;
5535 struct list_head *tmp;
5536 sector_t last_check;
NeilBrown57afd892005-06-21 17:17:13 -07005537 int skipped = 0;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005538 struct list_head *rtmp;
5539 mdk_rdev_t *rdev;
NeilBrown61df9d92006-10-03 01:15:57 -07005540 char *desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005541
5542 /* just incase thread restarts... */
5543 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5544 return;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005545 if (mddev->ro) /* never try to sync a read-only array */
5546 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005547
NeilBrown61df9d92006-10-03 01:15:57 -07005548 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5549 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5550 desc = "data-check";
5551 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5552 desc = "requested-resync";
5553 else
5554 desc = "resync";
5555 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5556 desc = "reshape";
5557 else
5558 desc = "recovery";
5559
Linus Torvalds1da177e2005-04-16 15:20:36 -07005560 /* we overload curr_resync somewhat here.
5561 * 0 == not engaged in resync at all
5562 * 2 == checking that there is no conflict with another sync
5563 * 1 == like 2, but have yielded to allow conflicting resync to
5564 * commense
5565 * other == active in resync - this many blocks
5566 *
5567 * Before starting a resync we must have set curr_resync to
5568 * 2, and then checked that every "conflicting" array has curr_resync
5569 * less than ours. When we find one that is the same or higher
5570 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5571 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5572 * This will mean we have to start checking from the beginning again.
5573 *
5574 */
5575
5576 do {
5577 mddev->curr_resync = 2;
5578
5579 try_again:
NeilBrown787453c2005-11-08 21:39:43 -08005580 if (kthread_should_stop()) {
NeilBrown6985c432005-10-19 21:23:47 -07005581 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005582 goto skip;
5583 }
NeilBrown29ac4aa2008-02-06 01:39:58 -08005584 for_each_mddev(mddev2, tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005585 if (mddev2 == mddev)
5586 continue;
Bernd Schubert90b08712008-05-23 13:04:38 -07005587 if (!mddev->parallel_resync
5588 && mddev2->curr_resync
5589 && match_mddev_units(mddev, mddev2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005590 DEFINE_WAIT(wq);
5591 if (mddev < mddev2 && mddev->curr_resync == 2) {
5592 /* arbitrarily yield */
5593 mddev->curr_resync = 1;
5594 wake_up(&resync_wait);
5595 }
5596 if (mddev > mddev2 && mddev->curr_resync == 1)
5597 /* no need to wait here, we can wait the next
5598 * time 'round when curr_resync == 2
5599 */
5600 continue;
NeilBrown787453c2005-11-08 21:39:43 -08005601 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5602 if (!kthread_should_stop() &&
NeilBrown8712e552005-10-26 01:58:58 -07005603 mddev2->curr_resync >= mddev->curr_resync) {
NeilBrown61df9d92006-10-03 01:15:57 -07005604 printk(KERN_INFO "md: delaying %s of %s"
5605 " until %s has finished (they"
Linus Torvalds1da177e2005-04-16 15:20:36 -07005606 " share one or more physical units)\n",
NeilBrown61df9d92006-10-03 01:15:57 -07005607 desc, mdname(mddev), mdname(mddev2));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005608 mddev_put(mddev2);
5609 schedule();
5610 finish_wait(&resync_wait, &wq);
5611 goto try_again;
5612 }
5613 finish_wait(&resync_wait, &wq);
5614 }
5615 }
5616 } while (mddev->curr_resync < 2);
5617
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005618 j = 0;
NeilBrown9d888832005-11-08 21:39:26 -08005619 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005620 /* resync follows the size requested by the personality,
NeilBrown57afd892005-06-21 17:17:13 -07005621 * which defaults to physical size, but can be virtual size
Linus Torvalds1da177e2005-04-16 15:20:36 -07005622 */
5623 max_sectors = mddev->resync_max_sectors;
NeilBrown9d888832005-11-08 21:39:26 -08005624 mddev->resync_mismatches = 0;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005625 /* we don't use the checkpoint if there's a bitmap */
5626 if (!mddev->bitmap &&
5627 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5628 j = mddev->recovery_cp;
NeilBrownccfcc3c2006-03-27 01:18:09 -08005629 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5630 max_sectors = mddev->size << 1;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005631 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005632 /* recovery follows the physical size of devices */
5633 max_sectors = mddev->size << 1;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005634 j = MaxSector;
NeilBrownd089c6a2008-02-06 01:39:59 -08005635 rdev_for_each(rdev, rtmp, mddev)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005636 if (rdev->raid_disk >= 0 &&
5637 !test_bit(Faulty, &rdev->flags) &&
5638 !test_bit(In_sync, &rdev->flags) &&
5639 rdev->recovery_offset < j)
5640 j = rdev->recovery_offset;
5641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005642
NeilBrown61df9d92006-10-03 01:15:57 -07005643 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5644 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5645 " %d KB/sec/disk.\n", speed_min(mddev));
Adrian Bunk338cec32005-09-10 00:26:54 -07005646 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
NeilBrown61df9d92006-10-03 01:15:57 -07005647 "(but not more than %d KB/sec) for %s.\n",
5648 speed_max(mddev), desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005649
5650 is_mddev_idle(mddev); /* this also initializes IO event counters */
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005651
NeilBrown57afd892005-06-21 17:17:13 -07005652 io_sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005653 for (m = 0; m < SYNC_MARKS; m++) {
5654 mark[m] = jiffies;
NeilBrown57afd892005-06-21 17:17:13 -07005655 mark_cnt[m] = io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005656 }
5657 last_mark = 0;
5658 mddev->resync_mark = mark[last_mark];
5659 mddev->resync_mark_cnt = mark_cnt[last_mark];
5660
5661 /*
5662 * Tune reconstruction:
5663 */
5664 window = 32*(PAGE_SIZE/512);
5665 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5666 window/2,(unsigned long long) max_sectors/2);
5667
5668 atomic_set(&mddev->recovery_active, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005669 last_check = 0;
5670
5671 if (j>2) {
5672 printk(KERN_INFO
NeilBrown61df9d92006-10-03 01:15:57 -07005673 "md: resuming %s of %s from checkpoint.\n",
5674 desc, mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005675 mddev->curr_resync = j;
5676 }
5677
5678 while (j < max_sectors) {
NeilBrown57afd892005-06-21 17:17:13 -07005679 sector_t sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005680
NeilBrown57afd892005-06-21 17:17:13 -07005681 skipped = 0;
NeilBrownc6207272008-02-06 01:39:52 -08005682 if (j >= mddev->resync_max) {
5683 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5684 wait_event(mddev->recovery_wait,
5685 mddev->resync_max > j
5686 || kthread_should_stop());
5687 }
5688 if (kthread_should_stop())
5689 goto interrupted;
NeilBrown57afd892005-06-21 17:17:13 -07005690 sectors = mddev->pers->sync_request(mddev, j, &skipped,
NeilBrownc6207272008-02-06 01:39:52 -08005691 currspeed < speed_min(mddev));
NeilBrown57afd892005-06-21 17:17:13 -07005692 if (sectors == 0) {
NeilBrowndfc70642008-05-23 13:04:39 -07005693 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005694 goto out;
5695 }
NeilBrown57afd892005-06-21 17:17:13 -07005696
5697 if (!skipped) { /* actual IO requested */
5698 io_sectors += sectors;
5699 atomic_add(sectors, &mddev->recovery_active);
5700 }
5701
Linus Torvalds1da177e2005-04-16 15:20:36 -07005702 j += sectors;
5703 if (j>1) mddev->curr_resync = j;
NeilBrownff4e8d92006-07-10 04:44:16 -07005704 mddev->curr_mark_cnt = io_sectors;
NeilBrownd7603b72006-01-06 00:20:30 -08005705 if (last_check == 0)
5706 /* this is the earliers that rebuilt will be
5707 * visible in /proc/mdstat
5708 */
5709 md_new_event(mddev);
NeilBrown57afd892005-06-21 17:17:13 -07005710
5711 if (last_check + window > io_sectors || j == max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005712 continue;
5713
NeilBrown57afd892005-06-21 17:17:13 -07005714 last_check = io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005715
NeilBrowndfc70642008-05-23 13:04:39 -07005716 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005717 break;
5718
5719 repeat:
5720 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5721 /* step marks */
5722 int next = (last_mark+1) % SYNC_MARKS;
5723
5724 mddev->resync_mark = mark[next];
5725 mddev->resync_mark_cnt = mark_cnt[next];
5726 mark[next] = jiffies;
NeilBrown57afd892005-06-21 17:17:13 -07005727 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005728 last_mark = next;
5729 }
5730
5731
NeilBrownc6207272008-02-06 01:39:52 -08005732 if (kthread_should_stop())
5733 goto interrupted;
5734
Linus Torvalds1da177e2005-04-16 15:20:36 -07005735
5736 /*
5737 * this loop exits only if either when we are slower than
5738 * the 'hard' speed limit, or the system was IO-idle for
5739 * a jiffy.
5740 * the system might be non-idle CPU-wise, but we only care
5741 * about not overloading the IO subsystem. (things like an
5742 * e2fsck being done on the RAID array should execute fast)
5743 */
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -05005744 blk_unplug(mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005745 cond_resched();
5746
NeilBrown57afd892005-06-21 17:17:13 -07005747 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5748 /((jiffies-mddev->resync_mark)/HZ +1) +1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005749
NeilBrown88202a02006-01-06 00:21:36 -08005750 if (currspeed > speed_min(mddev)) {
5751 if ((currspeed > speed_max(mddev)) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07005752 !is_mddev_idle(mddev)) {
NeilBrownc0e48522005-11-18 01:11:01 -08005753 msleep(500);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005754 goto repeat;
5755 }
5756 }
5757 }
NeilBrown61df9d92006-10-03 01:15:57 -07005758 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005759 /*
5760 * this also signals 'finished resyncing' to md_stop
5761 */
5762 out:
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -05005763 blk_unplug(mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005764
5765 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5766
5767 /* tell personality that we are finished */
NeilBrown57afd892005-06-21 17:17:13 -07005768 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005769
NeilBrowndfc70642008-05-23 13:04:39 -07005770 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005771 mddev->curr_resync > 2) {
5772 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5773 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5774 if (mddev->curr_resync >= mddev->recovery_cp) {
5775 printk(KERN_INFO
NeilBrown61df9d92006-10-03 01:15:57 -07005776 "md: checkpointing %s of %s.\n",
5777 desc, mdname(mddev));
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005778 mddev->recovery_cp = mddev->curr_resync;
5779 }
5780 } else
5781 mddev->recovery_cp = MaxSector;
5782 } else {
5783 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5784 mddev->curr_resync = MaxSector;
NeilBrownd089c6a2008-02-06 01:39:59 -08005785 rdev_for_each(rdev, rtmp, mddev)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005786 if (rdev->raid_disk >= 0 &&
5787 !test_bit(Faulty, &rdev->flags) &&
5788 !test_bit(In_sync, &rdev->flags) &&
5789 rdev->recovery_offset < mddev->curr_resync)
5790 rdev->recovery_offset = mddev->curr_resync;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005791 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005792 }
NeilBrown17571282006-12-10 02:20:52 -08005793 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005794
Linus Torvalds1da177e2005-04-16 15:20:36 -07005795 skip:
5796 mddev->curr_resync = 0;
NeilBrownc6207272008-02-06 01:39:52 -08005797 mddev->resync_max = MaxSector;
5798 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005799 wake_up(&resync_wait);
5800 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5801 md_wakeup_thread(mddev->thread);
NeilBrownc6207272008-02-06 01:39:52 -08005802 return;
5803
5804 interrupted:
5805 /*
5806 * got a signal, exit.
5807 */
5808 printk(KERN_INFO
5809 "md: md_do_sync() got signal ... exiting\n");
5810 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5811 goto out;
5812
Linus Torvalds1da177e2005-04-16 15:20:36 -07005813}
NeilBrown29269552006-03-27 01:18:10 -08005814EXPORT_SYMBOL_GPL(md_do_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005815
5816
NeilBrownb4c4c7b2007-02-28 20:11:48 -08005817static int remove_and_add_spares(mddev_t *mddev)
5818{
5819 mdk_rdev_t *rdev;
5820 struct list_head *rtmp;
5821 int spares = 0;
5822
NeilBrownd089c6a2008-02-06 01:39:59 -08005823 rdev_for_each(rdev, rtmp, mddev)
NeilBrownb4c4c7b2007-02-28 20:11:48 -08005824 if (rdev->raid_disk >= 0 &&
Dan Williams6bfe0b42008-04-30 00:52:32 -07005825 !test_bit(Blocked, &rdev->flags) &&
NeilBrownb4c4c7b2007-02-28 20:11:48 -08005826 (test_bit(Faulty, &rdev->flags) ||
5827 ! test_bit(In_sync, &rdev->flags)) &&
5828 atomic_read(&rdev->nr_pending)==0) {
5829 if (mddev->pers->hot_remove_disk(
5830 mddev, rdev->raid_disk)==0) {
5831 char nm[20];
5832 sprintf(nm,"rd%d", rdev->raid_disk);
5833 sysfs_remove_link(&mddev->kobj, nm);
5834 rdev->raid_disk = -1;
5835 }
5836 }
5837
5838 if (mddev->degraded) {
NeilBrowndfc70642008-05-23 13:04:39 -07005839 rdev_for_each(rdev, rtmp, mddev) {
5840 if (rdev->raid_disk >= 0 &&
5841 !test_bit(In_sync, &rdev->flags))
5842 spares++;
NeilBrownb4c4c7b2007-02-28 20:11:48 -08005843 if (rdev->raid_disk < 0
5844 && !test_bit(Faulty, &rdev->flags)) {
5845 rdev->recovery_offset = 0;
5846 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5847 char nm[20];
5848 sprintf(nm, "rd%d", rdev->raid_disk);
NeilBrown5e55e2f2007-03-26 21:32:14 -08005849 if (sysfs_create_link(&mddev->kobj,
5850 &rdev->kobj, nm))
5851 printk(KERN_WARNING
5852 "md: cannot register "
5853 "%s for %s\n",
5854 nm, mdname(mddev));
NeilBrownb4c4c7b2007-02-28 20:11:48 -08005855 spares++;
5856 md_new_event(mddev);
5857 } else
5858 break;
5859 }
NeilBrowndfc70642008-05-23 13:04:39 -07005860 }
NeilBrownb4c4c7b2007-02-28 20:11:48 -08005861 }
5862 return spares;
5863}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005864/*
5865 * This routine is regularly called by all per-raid-array threads to
5866 * deal with generic issues like resync and super-block update.
5867 * Raid personalities that don't have a thread (linear/raid0) do not
5868 * need this as they never do any recovery or update the superblock.
5869 *
5870 * It does not do any resync itself, but rather "forks" off other threads
5871 * to do that as needed.
5872 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5873 * "->recovery" and create a thread at ->sync_thread.
NeilBrowndfc70642008-05-23 13:04:39 -07005874 * When the thread finishes it sets MD_RECOVERY_DONE
Linus Torvalds1da177e2005-04-16 15:20:36 -07005875 * and wakeups up this thread which will reap the thread and finish up.
5876 * This thread also removes any faulty devices (with nr_pending == 0).
5877 *
5878 * The overall approach is:
5879 * 1/ if the superblock needs updating, update it.
5880 * 2/ If a recovery thread is running, don't do anything else.
5881 * 3/ If recovery has finished, clean up, possibly marking spares active.
5882 * 4/ If there are any faulty devices, remove them.
5883 * 5/ If array is degraded, try to add spares devices
5884 * 6/ If array has spares or is not in-sync, start a resync thread.
5885 */
5886void md_check_recovery(mddev_t *mddev)
5887{
5888 mdk_rdev_t *rdev;
5889 struct list_head *rtmp;
5890
5891
NeilBrown5f404022005-06-21 17:17:16 -07005892 if (mddev->bitmap)
5893 bitmap_daemon_work(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005894
5895 if (mddev->ro)
5896 return;
NeilBrownfca4d842005-06-21 17:17:11 -07005897
5898 if (signal_pending(current)) {
NeilBrown31a59e32008-04-30 00:52:30 -07005899 if (mddev->pers->sync_request && !mddev->external) {
NeilBrownfca4d842005-06-21 17:17:11 -07005900 printk(KERN_INFO "md: %s in immediate safe mode\n",
5901 mdname(mddev));
5902 mddev->safemode = 2;
5903 }
5904 flush_signals(current);
5905 }
5906
Linus Torvalds1da177e2005-04-16 15:20:36 -07005907 if ( ! (
NeilBrowne6910632008-02-06 01:39:51 -08005908 (mddev->flags && !mddev->external) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07005909 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
NeilBrownfca4d842005-06-21 17:17:11 -07005910 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
NeilBrown31a59e32008-04-30 00:52:30 -07005911 (mddev->external == 0 && mddev->safemode == 1) ||
NeilBrownfca4d842005-06-21 17:17:11 -07005912 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5913 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005914 ))
5915 return;
NeilBrownfca4d842005-06-21 17:17:11 -07005916
NeilBrowndf5b89b2006-03-27 01:18:20 -08005917 if (mddev_trylock(mddev)) {
NeilBrownb4c4c7b2007-02-28 20:11:48 -08005918 int spares = 0;
NeilBrownfca4d842005-06-21 17:17:11 -07005919
NeilBrown31a59e32008-04-30 00:52:30 -07005920 if (!mddev->external) {
5921 spin_lock_irq(&mddev->write_lock);
5922 if (mddev->safemode &&
5923 !atomic_read(&mddev->writes_pending) &&
5924 !mddev->in_sync &&
5925 mddev->recovery_cp == MaxSector) {
5926 mddev->in_sync = 1;
5927 if (mddev->persistent)
5928 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5929 }
5930 if (mddev->safemode == 1)
5931 mddev->safemode = 0;
5932 spin_unlock_irq(&mddev->write_lock);
NeilBrownfca4d842005-06-21 17:17:11 -07005933 }
NeilBrownfca4d842005-06-21 17:17:11 -07005934
NeilBrown850b2b42006-10-03 01:15:46 -07005935 if (mddev->flags)
5936 md_update_sb(mddev, 0);
NeilBrown06d91a52005-06-21 17:17:12 -07005937
NeilBrown06d91a52005-06-21 17:17:12 -07005938
Linus Torvalds1da177e2005-04-16 15:20:36 -07005939 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5940 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5941 /* resync/recovery still happening */
5942 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5943 goto unlock;
5944 }
5945 if (mddev->sync_thread) {
5946 /* resync has finished, collect result */
5947 md_unregister_thread(mddev->sync_thread);
5948 mddev->sync_thread = NULL;
NeilBrowndfc70642008-05-23 13:04:39 -07005949 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005950 /* success...*/
5951 /* activate any spares */
5952 mddev->pers->spare_active(mddev);
5953 }
NeilBrown850b2b42006-10-03 01:15:46 -07005954 md_update_sb(mddev, 1);
NeilBrown41158c72005-06-21 17:17:25 -07005955
5956 /* if array is no-longer degraded, then any saved_raid_disk
5957 * information must be scrapped
5958 */
5959 if (!mddev->degraded)
NeilBrownd089c6a2008-02-06 01:39:59 -08005960 rdev_for_each(rdev, rtmp, mddev)
NeilBrown41158c72005-06-21 17:17:25 -07005961 rdev->saved_raid_disk = -1;
5962
Linus Torvalds1da177e2005-04-16 15:20:36 -07005963 mddev->recovery = 0;
5964 /* flag recovery needed just to double check */
5965 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrownd7603b72006-01-06 00:20:30 -08005966 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005967 goto unlock;
5968 }
NeilBrown24dd4692005-11-08 21:39:26 -08005969 /* Clear some bits that don't mean anything, but
5970 * might be left set
5971 */
5972 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08005973 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5974 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005975
NeilBrown5fd6c1d2006-06-26 00:27:40 -07005976 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5977 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005978 /* no recovery is running.
5979 * remove any failed drives, then
5980 * add spares if possible.
5981 * Spare are also removed and re-added, to allow
5982 * the personality to fail the re-add.
5983 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005984
NeilBrownb4c4c7b2007-02-28 20:11:48 -08005985 if (mddev->reshape_position != MaxSector) {
5986 if (mddev->pers->check_reshape(mddev) != 0)
5987 /* Cannot proceed */
5988 goto unlock;
5989 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5990 } else if ((spares = remove_and_add_spares(mddev))) {
NeilBrown24dd4692005-11-08 21:39:26 -08005991 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5992 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5993 } else if (mddev->recovery_cp < MaxSector) {
5994 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5995 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5996 /* nothing to be done ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005997 goto unlock;
NeilBrown24dd4692005-11-08 21:39:26 -08005998
Linus Torvalds1da177e2005-04-16 15:20:36 -07005999 if (mddev->pers->sync_request) {
6000 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
NeilBrowna654b9d82005-06-21 17:17:27 -07006001 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6002 /* We are adding a device or devices to an array
6003 * which has the bitmap stored on all devices.
6004 * So make sure all bitmap pages get written
6005 */
6006 bitmap_write_all(mddev->bitmap);
6007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006008 mddev->sync_thread = md_register_thread(md_do_sync,
6009 mddev,
6010 "%s_resync");
6011 if (!mddev->sync_thread) {
6012 printk(KERN_ERR "%s: could not start resync"
6013 " thread...\n",
6014 mdname(mddev));
6015 /* leave the spares where they are, it shouldn't hurt */
6016 mddev->recovery = 0;
NeilBrownd7603b72006-01-06 00:20:30 -08006017 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07006018 md_wakeup_thread(mddev->sync_thread);
NeilBrownd7603b72006-01-06 00:20:30 -08006019 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006020 }
6021 unlock:
6022 mddev_unlock(mddev);
6023 }
6024}
6025
Dan Williams6bfe0b42008-04-30 00:52:32 -07006026void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6027{
6028 sysfs_notify(&rdev->kobj, NULL, "state");
6029 wait_event_timeout(rdev->blocked_wait,
6030 !test_bit(Blocked, &rdev->flags),
6031 msecs_to_jiffies(5000));
6032 rdev_dec_pending(rdev, mddev);
6033}
6034EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6035
Adrian Bunk75c96f82005-05-05 16:16:09 -07006036static int md_notify_reboot(struct notifier_block *this,
6037 unsigned long code, void *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006038{
6039 struct list_head *tmp;
6040 mddev_t *mddev;
6041
6042 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6043
6044 printk(KERN_INFO "md: stopping all md devices.\n");
6045
NeilBrown29ac4aa2008-02-06 01:39:58 -08006046 for_each_mddev(mddev, tmp)
Neil Brownc71d4882006-05-26 10:39:25 +10006047 if (mddev_trylock(mddev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006048 do_md_stop (mddev, 1);
Neil Brownc71d4882006-05-26 10:39:25 +10006049 mddev_unlock(mddev);
6050 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006051 /*
6052 * certain more exotic SCSI devices are known to be
6053 * volatile wrt too early system reboots. While the
6054 * right place to handle this issue is the given
6055 * driver, we do want to have a safe RAID driver ...
6056 */
6057 mdelay(1000*1);
6058 }
6059 return NOTIFY_DONE;
6060}
6061
Adrian Bunk75c96f82005-05-05 16:16:09 -07006062static struct notifier_block md_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006063 .notifier_call = md_notify_reboot,
6064 .next = NULL,
6065 .priority = INT_MAX, /* before any real devices */
6066};
6067
6068static void md_geninit(void)
6069{
Linus Torvalds1da177e2005-04-16 15:20:36 -07006070 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6071
Denis V. Lunevc7705f32008-04-29 01:02:35 -07006072 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006073}
6074
Adrian Bunk75c96f82005-05-05 16:16:09 -07006075static int __init md_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006076{
Linus Torvalds1da177e2005-04-16 15:20:36 -07006077 if (register_blkdev(MAJOR_NR, "md"))
6078 return -1;
6079 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6080 unregister_blkdev(MAJOR_NR, "md");
6081 return -1;
6082 }
NeilBrowne8703fe2006-10-03 01:15:59 -07006083 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6084 md_probe, NULL, NULL);
6085 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006086 md_probe, NULL, NULL);
6087
Linus Torvalds1da177e2005-04-16 15:20:36 -07006088 register_reboot_notifier(&md_notifier);
Eric W. Biederman0b4d4142007-02-14 00:34:09 -08006089 raid_table_header = register_sysctl_table(raid_root_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006090
6091 md_geninit();
6092 return (0);
6093}
6094
6095
6096#ifndef MODULE
6097
6098/*
6099 * Searches all registered partitions for autorun RAID arrays
6100 * at boot time.
6101 */
Michael J. Evans4d936ec2007-10-16 23:30:52 -07006102
6103static LIST_HEAD(all_detected_devices);
6104struct detected_devices_node {
6105 struct list_head list;
6106 dev_t dev;
6107};
Linus Torvalds1da177e2005-04-16 15:20:36 -07006108
6109void md_autodetect_dev(dev_t dev)
6110{
Michael J. Evans4d936ec2007-10-16 23:30:52 -07006111 struct detected_devices_node *node_detected_dev;
6112
6113 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6114 if (node_detected_dev) {
6115 node_detected_dev->dev = dev;
6116 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6117 } else {
6118 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6119 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6120 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006121}
6122
6123
6124static void autostart_arrays(int part)
6125{
6126 mdk_rdev_t *rdev;
Michael J. Evans4d936ec2007-10-16 23:30:52 -07006127 struct detected_devices_node *node_detected_dev;
6128 dev_t dev;
6129 int i_scanned, i_passed;
6130
6131 i_scanned = 0;
6132 i_passed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006133
6134 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6135
Michael J. Evans4d936ec2007-10-16 23:30:52 -07006136 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6137 i_scanned++;
6138 node_detected_dev = list_entry(all_detected_devices.next,
6139 struct detected_devices_node, list);
6140 list_del(&node_detected_dev->list);
6141 dev = node_detected_dev->dev;
6142 kfree(node_detected_dev);
NeilBrowndf968c42007-07-17 04:06:11 -07006143 rdev = md_import_device(dev,0, 90);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006144 if (IS_ERR(rdev))
6145 continue;
6146
NeilBrownb2d444d2005-11-08 21:39:31 -08006147 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006148 MD_BUG();
6149 continue;
6150 }
NeilBrownd0fae182008-03-04 14:29:31 -08006151 set_bit(AutoDetected, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006152 list_add(&rdev->same_set, &pending_raid_disks);
Michael J. Evans4d936ec2007-10-16 23:30:52 -07006153 i_passed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006154 }
Michael J. Evans4d936ec2007-10-16 23:30:52 -07006155
6156 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6157 i_scanned, i_passed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006158
6159 autorun_devices(part);
6160}
6161
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08006162#endif /* !MODULE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006163
6164static __exit void md_exit(void)
6165{
6166 mddev_t *mddev;
6167 struct list_head *tmp;
Greg Kroah-Hartman8ab5e4c2005-06-20 21:15:16 -07006168
NeilBrowne8703fe2006-10-03 01:15:59 -07006169 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6170 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006171
6172 unregister_blkdev(MAJOR_NR,"md");
6173 unregister_blkdev(mdp_major, "mdp");
6174 unregister_reboot_notifier(&md_notifier);
6175 unregister_sysctl_table(raid_table_header);
6176 remove_proc_entry("mdstat", NULL);
NeilBrown29ac4aa2008-02-06 01:39:58 -08006177 for_each_mddev(mddev, tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006178 struct gendisk *disk = mddev->gendisk;
6179 if (!disk)
6180 continue;
6181 export_array(mddev);
6182 del_gendisk(disk);
6183 put_disk(disk);
6184 mddev->gendisk = NULL;
6185 mddev_put(mddev);
6186 }
6187}
6188
Dan Williams685784a2007-07-09 11:56:42 -07006189subsys_initcall(md_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006190module_exit(md_exit)
6191
NeilBrownf91de922005-11-08 21:39:36 -08006192static int get_ro(char *buffer, struct kernel_param *kp)
6193{
6194 return sprintf(buffer, "%d", start_readonly);
6195}
6196static int set_ro(const char *val, struct kernel_param *kp)
6197{
6198 char *e;
6199 int num = simple_strtoul(val, &e, 10);
6200 if (*val && (*e == '\0' || *e == '\n')) {
6201 start_readonly = num;
NeilBrown4dbcdc72006-01-06 00:20:52 -08006202 return 0;
NeilBrownf91de922005-11-08 21:39:36 -08006203 }
6204 return -EINVAL;
6205}
6206
NeilBrown80ca3a42006-07-10 04:44:18 -07006207module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6208module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006209
NeilBrownf91de922005-11-08 21:39:36 -08006210
Linus Torvalds1da177e2005-04-16 15:20:36 -07006211EXPORT_SYMBOL(register_md_personality);
6212EXPORT_SYMBOL(unregister_md_personality);
6213EXPORT_SYMBOL(md_error);
6214EXPORT_SYMBOL(md_done_sync);
6215EXPORT_SYMBOL(md_write_start);
6216EXPORT_SYMBOL(md_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006217EXPORT_SYMBOL(md_register_thread);
6218EXPORT_SYMBOL(md_unregister_thread);
6219EXPORT_SYMBOL(md_wakeup_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006220EXPORT_SYMBOL(md_check_recovery);
6221MODULE_LICENSE("GPL");
NeilBrownaa1595e2005-08-04 12:53:32 -07006222MODULE_ALIAS("md");
NeilBrown72008652005-08-26 18:34:15 -07006223MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);