]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | md.c : Multiple Devices driver for Linux | |
3 | Copyright (C) 1998, 1999, 2000 Ingo Molnar | |
4 | ||
5 | completely rewritten, based on the MD driver code from Marc Zyngier | |
6 | ||
7 | Changes: | |
8 | ||
9 | - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar | |
10 | - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com> | |
11 | - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net> | |
12 | - kerneld support by Boris Tobotras <boris@xtalk.msk.su> | |
13 | - kmod support by: Cyrus Durgin | |
14 | - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com> | |
15 | - Devfs support by Richard Gooch <rgooch@atnf.csiro.au> | |
16 | ||
17 | - lots of fixes and improvements to the RAID1/RAID5 and generic | |
18 | RAID code (such as request based resynchronization): | |
19 | ||
20 | Neil Brown <neilb@cse.unsw.edu.au>. | |
21 | ||
32a7627c N |
22 | - persistent bitmap code |
23 | Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. | |
24 | ||
1da177e4 LT |
25 | This program is free software; you can redistribute it and/or modify |
26 | it under the terms of the GNU General Public License as published by | |
27 | the Free Software Foundation; either version 2, or (at your option) | |
28 | any later version. | |
29 | ||
30 | You should have received a copy of the GNU General Public License | |
31 | (for example /usr/src/linux/COPYING); if not, write to the Free | |
32 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
33 | */ | |
34 | ||
a6fb0934 | 35 | #include <linux/kthread.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
1da177e4 | 37 | #include <linux/sysctl.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
1da177e4 | 39 | #include <linux/buffer_head.h> /* for invalidate_bdev */ |
d7603b7e | 40 | #include <linux/poll.h> |
16f17b39 | 41 | #include <linux/ctype.h> |
fb4d8c76 N |
42 | #include <linux/hdreg.h> |
43 | #include <linux/proc_fs.h> | |
44 | #include <linux/random.h> | |
45 | #include <linux/reboot.h> | |
32a7627c | 46 | #include <linux/file.h> |
25570727 | 47 | #include <linux/delay.h> |
bff61975 N |
48 | #include <linux/raid/md_p.h> |
49 | #include <linux/raid/md_u.h> | |
43b2e5d8 | 50 | #include "md.h" |
ef740c37 | 51 | #include "bitmap.h" |
1da177e4 LT |
52 | |
53 | #define DEBUG 0 | |
54 | #define dprintk(x...) ((void)(DEBUG && printk(x))) | |
55 | ||
56 | ||
57 | #ifndef MODULE | |
d710e138 | 58 | static void autostart_arrays(int part); |
1da177e4 LT |
59 | #endif |
60 | ||
2604b703 | 61 | static LIST_HEAD(pers_list); |
1da177e4 LT |
62 | static DEFINE_SPINLOCK(pers_lock); |
63 | ||
5e56341d AB |
64 | static void md_print_devices(void); |
65 | ||
90b08710 BS |
66 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
67 | ||
5e56341d AB |
68 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } |
69 | ||
1da177e4 LT |
70 | /* |
71 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
72 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
73 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 74 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
75 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
76 | * speed limit - in case reconstruction slows down your system despite | |
77 | * idle IO detection. | |
78 | * | |
79 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 80 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
81 | */ |
82 | ||
83 | static int sysctl_speed_limit_min = 1000; | |
84 | static int sysctl_speed_limit_max = 200000; | |
88202a0c N |
85 | static inline int speed_min(mddev_t *mddev) |
86 | { | |
87 | return mddev->sync_speed_min ? | |
88 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
89 | } | |
90 | ||
91 | static inline int speed_max(mddev_t *mddev) | |
92 | { | |
93 | return mddev->sync_speed_max ? | |
94 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
95 | } | |
1da177e4 LT |
96 | |
97 | static struct ctl_table_header *raid_table_header; | |
98 | ||
99 | static ctl_table raid_table[] = { | |
100 | { | |
101 | .ctl_name = DEV_RAID_SPEED_LIMIT_MIN, | |
102 | .procname = "speed_limit_min", | |
103 | .data = &sysctl_speed_limit_min, | |
104 | .maxlen = sizeof(int), | |
80ca3a44 | 105 | .mode = S_IRUGO|S_IWUSR, |
1da177e4 LT |
106 | .proc_handler = &proc_dointvec, |
107 | }, | |
108 | { | |
109 | .ctl_name = DEV_RAID_SPEED_LIMIT_MAX, | |
110 | .procname = "speed_limit_max", | |
111 | .data = &sysctl_speed_limit_max, | |
112 | .maxlen = sizeof(int), | |
80ca3a44 | 113 | .mode = S_IRUGO|S_IWUSR, |
1da177e4 LT |
114 | .proc_handler = &proc_dointvec, |
115 | }, | |
116 | { .ctl_name = 0 } | |
117 | }; | |
118 | ||
119 | static ctl_table raid_dir_table[] = { | |
120 | { | |
121 | .ctl_name = DEV_RAID, | |
122 | .procname = "raid", | |
123 | .maxlen = 0, | |
80ca3a44 | 124 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
125 | .child = raid_table, |
126 | }, | |
127 | { .ctl_name = 0 } | |
128 | }; | |
129 | ||
130 | static ctl_table raid_root_table[] = { | |
131 | { | |
132 | .ctl_name = CTL_DEV, | |
133 | .procname = "dev", | |
134 | .maxlen = 0, | |
135 | .mode = 0555, | |
136 | .child = raid_dir_table, | |
137 | }, | |
138 | { .ctl_name = 0 } | |
139 | }; | |
140 | ||
141 | static struct block_device_operations md_fops; | |
142 | ||
f91de92e N |
143 | static int start_readonly; |
144 | ||
d7603b7e N |
145 | /* |
146 | * We have a system wide 'event count' that is incremented | |
147 | * on any 'interesting' event, and readers of /proc/mdstat | |
148 | * can use 'poll' or 'select' to find out when the event | |
149 | * count increases. | |
150 | * | |
151 | * Events are: | |
152 | * start array, stop array, error, add device, remove device, | |
153 | * start build, activate spare | |
154 | */ | |
2989ddbd | 155 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 156 | static atomic_t md_event_count; |
29269553 | 157 | void md_new_event(mddev_t *mddev) |
d7603b7e N |
158 | { |
159 | atomic_inc(&md_event_count); | |
160 | wake_up(&md_event_waiters); | |
161 | } | |
29269553 | 162 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 163 | |
c331eb04 N |
164 | /* Alternate version that can be called from interrupts |
165 | * when calling sysfs_notify isn't needed. | |
166 | */ | |
05381954 | 167 | static void md_new_event_inintr(mddev_t *mddev) |
c331eb04 N |
168 | { |
169 | atomic_inc(&md_event_count); | |
170 | wake_up(&md_event_waiters); | |
171 | } | |
172 | ||
1da177e4 LT |
173 | /* |
174 | * Enables to iterate over all existing md arrays | |
175 | * all_mddevs_lock protects this list. | |
176 | */ | |
177 | static LIST_HEAD(all_mddevs); | |
178 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
179 | ||
180 | ||
181 | /* | |
182 | * iterates through all used mddevs in the system. | |
183 | * We take care to grab the all_mddevs_lock whenever navigating | |
184 | * the list, and to always hold a refcount when unlocked. | |
185 | * Any code which breaks out of this loop while own | |
186 | * a reference to the current mddev and must mddev_put it. | |
187 | */ | |
29ac4aa3 | 188 | #define for_each_mddev(mddev,tmp) \ |
1da177e4 LT |
189 | \ |
190 | for (({ spin_lock(&all_mddevs_lock); \ | |
191 | tmp = all_mddevs.next; \ | |
192 | mddev = NULL;}); \ | |
193 | ({ if (tmp != &all_mddevs) \ | |
194 | mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ | |
195 | spin_unlock(&all_mddevs_lock); \ | |
196 | if (mddev) mddev_put(mddev); \ | |
197 | mddev = list_entry(tmp, mddev_t, all_mddevs); \ | |
198 | tmp != &all_mddevs;}); \ | |
199 | ({ spin_lock(&all_mddevs_lock); \ | |
200 | tmp = tmp->next;}) \ | |
201 | ) | |
202 | ||
203 | ||
409c57f3 N |
204 | /* Rather than calling directly into the personality make_request function, |
205 | * IO requests come here first so that we can check if the device is | |
206 | * being suspended pending a reconfiguration. | |
207 | * We hold a refcount over the call to ->make_request. By the time that | |
208 | * call has finished, the bio has been linked into some internal structure | |
209 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
210 | */ | |
211 | static int md_make_request(struct request_queue *q, struct bio *bio) | |
1da177e4 | 212 | { |
409c57f3 N |
213 | mddev_t *mddev = q->queuedata; |
214 | int rv; | |
215 | if (mddev == NULL || mddev->pers == NULL) { | |
216 | bio_io_error(bio); | |
217 | return 0; | |
218 | } | |
219 | rcu_read_lock(); | |
220 | if (mddev->suspended) { | |
221 | DEFINE_WAIT(__wait); | |
222 | for (;;) { | |
223 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
224 | TASK_UNINTERRUPTIBLE); | |
225 | if (!mddev->suspended) | |
226 | break; | |
227 | rcu_read_unlock(); | |
228 | schedule(); | |
229 | rcu_read_lock(); | |
230 | } | |
231 | finish_wait(&mddev->sb_wait, &__wait); | |
232 | } | |
233 | atomic_inc(&mddev->active_io); | |
234 | rcu_read_unlock(); | |
235 | rv = mddev->pers->make_request(q, bio); | |
236 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) | |
237 | wake_up(&mddev->sb_wait); | |
238 | ||
239 | return rv; | |
240 | } | |
241 | ||
242 | static void mddev_suspend(mddev_t *mddev) | |
243 | { | |
244 | BUG_ON(mddev->suspended); | |
245 | mddev->suspended = 1; | |
246 | synchronize_rcu(); | |
247 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
248 | mddev->pers->quiesce(mddev, 1); | |
249 | md_unregister_thread(mddev->thread); | |
250 | mddev->thread = NULL; | |
251 | /* we now know that no code is executing in the personality module, | |
252 | * except possibly the tail end of a ->bi_end_io function, but that | |
253 | * is certain to complete before the module has a chance to get | |
254 | * unloaded | |
255 | */ | |
256 | } | |
257 | ||
258 | static void mddev_resume(mddev_t *mddev) | |
259 | { | |
260 | mddev->suspended = 0; | |
261 | wake_up(&mddev->sb_wait); | |
262 | mddev->pers->quiesce(mddev, 0); | |
1da177e4 LT |
263 | } |
264 | ||
409c57f3 | 265 | |
1da177e4 LT |
266 | static inline mddev_t *mddev_get(mddev_t *mddev) |
267 | { | |
268 | atomic_inc(&mddev->active); | |
269 | return mddev; | |
270 | } | |
271 | ||
5fd3a17e | 272 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 273 | |
1da177e4 LT |
274 | static void mddev_put(mddev_t *mddev) |
275 | { | |
276 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
277 | return; | |
d3374825 N |
278 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
279 | !mddev->hold_active) { | |
1da177e4 | 280 | list_del(&mddev->all_mddevs); |
d3374825 N |
281 | if (mddev->gendisk) { |
282 | /* we did a probe so need to clean up. | |
283 | * Call schedule_work inside the spinlock | |
284 | * so that flush_scheduled_work() after | |
285 | * mddev_find will succeed in waiting for the | |
286 | * work to be done. | |
287 | */ | |
288 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
289 | schedule_work(&mddev->del_work); | |
290 | } else | |
291 | kfree(mddev); | |
292 | } | |
293 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
294 | } |
295 | ||
296 | static mddev_t * mddev_find(dev_t unit) | |
297 | { | |
298 | mddev_t *mddev, *new = NULL; | |
299 | ||
300 | retry: | |
301 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
302 | |
303 | if (unit) { | |
304 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
305 | if (mddev->unit == unit) { | |
306 | mddev_get(mddev); | |
307 | spin_unlock(&all_mddevs_lock); | |
308 | kfree(new); | |
309 | return mddev; | |
310 | } | |
311 | ||
312 | if (new) { | |
313 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 314 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
315 | new->hold_active = UNTIL_IOCTL; |
316 | return new; | |
1da177e4 | 317 | } |
efeb53c0 N |
318 | } else if (new) { |
319 | /* find an unused unit number */ | |
320 | static int next_minor = 512; | |
321 | int start = next_minor; | |
322 | int is_free = 0; | |
323 | int dev = 0; | |
324 | while (!is_free) { | |
325 | dev = MKDEV(MD_MAJOR, next_minor); | |
326 | next_minor++; | |
327 | if (next_minor > MINORMASK) | |
328 | next_minor = 0; | |
329 | if (next_minor == start) { | |
330 | /* Oh dear, all in use. */ | |
331 | spin_unlock(&all_mddevs_lock); | |
332 | kfree(new); | |
333 | return NULL; | |
334 | } | |
335 | ||
336 | is_free = 1; | |
337 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
338 | if (mddev->unit == dev) { | |
339 | is_free = 0; | |
340 | break; | |
341 | } | |
342 | } | |
343 | new->unit = dev; | |
344 | new->md_minor = MINOR(dev); | |
345 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
346 | list_add(&new->all_mddevs, &all_mddevs); |
347 | spin_unlock(&all_mddevs_lock); | |
348 | return new; | |
349 | } | |
350 | spin_unlock(&all_mddevs_lock); | |
351 | ||
9ffae0cf | 352 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
353 | if (!new) |
354 | return NULL; | |
355 | ||
1da177e4 LT |
356 | new->unit = unit; |
357 | if (MAJOR(unit) == MD_MAJOR) | |
358 | new->md_minor = MINOR(unit); | |
359 | else | |
360 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
361 | ||
df5b89b3 | 362 | mutex_init(&new->reconfig_mutex); |
1da177e4 LT |
363 | INIT_LIST_HEAD(&new->disks); |
364 | INIT_LIST_HEAD(&new->all_mddevs); | |
365 | init_timer(&new->safemode_timer); | |
366 | atomic_set(&new->active, 1); | |
f2ea68cf | 367 | atomic_set(&new->openers, 0); |
409c57f3 | 368 | atomic_set(&new->active_io, 0); |
06d91a5f | 369 | spin_lock_init(&new->write_lock); |
3d310eb7 | 370 | init_waitqueue_head(&new->sb_wait); |
a6d8113a | 371 | init_waitqueue_head(&new->recovery_wait); |
08a02ecd | 372 | new->reshape_position = MaxSector; |
5e96ee65 | 373 | new->resync_min = 0; |
c6207277 | 374 | new->resync_max = MaxSector; |
d897dbf9 | 375 | new->level = LEVEL_NONE; |
1da177e4 | 376 | |
1da177e4 LT |
377 | goto retry; |
378 | } | |
379 | ||
380 | static inline int mddev_lock(mddev_t * mddev) | |
381 | { | |
df5b89b3 | 382 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
383 | } |
384 | ||
b522adcd DW |
385 | static inline int mddev_is_locked(mddev_t *mddev) |
386 | { | |
387 | return mutex_is_locked(&mddev->reconfig_mutex); | |
388 | } | |
389 | ||
1da177e4 LT |
390 | static inline int mddev_trylock(mddev_t * mddev) |
391 | { | |
df5b89b3 | 392 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
393 | } |
394 | ||
395 | static inline void mddev_unlock(mddev_t * mddev) | |
396 | { | |
df5b89b3 | 397 | mutex_unlock(&mddev->reconfig_mutex); |
1da177e4 | 398 | |
005eca5e | 399 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
400 | } |
401 | ||
2989ddbd | 402 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 403 | { |
159ec1fc | 404 | mdk_rdev_t *rdev; |
1da177e4 | 405 | |
159ec1fc | 406 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
407 | if (rdev->desc_nr == nr) |
408 | return rdev; | |
159ec1fc | 409 | |
1da177e4 LT |
410 | return NULL; |
411 | } | |
412 | ||
413 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
414 | { | |
1da177e4 LT |
415 | mdk_rdev_t *rdev; |
416 | ||
159ec1fc | 417 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
418 | if (rdev->bdev->bd_dev == dev) |
419 | return rdev; | |
159ec1fc | 420 | |
1da177e4 LT |
421 | return NULL; |
422 | } | |
423 | ||
d9d166c2 | 424 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
425 | { |
426 | struct mdk_personality *pers; | |
d9d166c2 N |
427 | list_for_each_entry(pers, &pers_list, list) { |
428 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 429 | return pers; |
d9d166c2 N |
430 | if (strcmp(pers->name, clevel)==0) |
431 | return pers; | |
432 | } | |
2604b703 N |
433 | return NULL; |
434 | } | |
435 | ||
b73df2d3 | 436 | /* return the offset of the super block in 512byte sectors */ |
77933d72 | 437 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 | 438 | { |
b73df2d3 AN |
439 | sector_t num_sectors = bdev->bd_inode->i_size / 512; |
440 | return MD_NEW_SIZE_SECTORS(num_sectors); | |
1da177e4 LT |
441 | } |
442 | ||
1da177e4 LT |
443 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
444 | { | |
445 | if (rdev->sb_page) | |
446 | MD_BUG(); | |
447 | ||
448 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
449 | if (!rdev->sb_page) { | |
450 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 451 | return -ENOMEM; |
1da177e4 LT |
452 | } |
453 | ||
454 | return 0; | |
455 | } | |
456 | ||
457 | static void free_disk_sb(mdk_rdev_t * rdev) | |
458 | { | |
459 | if (rdev->sb_page) { | |
2d1f3b5d | 460 | put_page(rdev->sb_page); |
1da177e4 LT |
461 | rdev->sb_loaded = 0; |
462 | rdev->sb_page = NULL; | |
0f420358 | 463 | rdev->sb_start = 0; |
dd8ac336 | 464 | rdev->sectors = 0; |
1da177e4 LT |
465 | } |
466 | } | |
467 | ||
468 | ||
6712ecf8 | 469 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
470 | { |
471 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 472 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 473 | |
3a0f5bbb N |
474 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
475 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
476 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
477 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 478 | md_error(mddev, rdev); |
3a0f5bbb | 479 | } |
7bfa19f2 | 480 | |
a9701a30 N |
481 | if (atomic_dec_and_test(&mddev->pending_writes)) |
482 | wake_up(&mddev->sb_wait); | |
f8b58edf | 483 | bio_put(bio); |
7bfa19f2 N |
484 | } |
485 | ||
6712ecf8 | 486 | static void super_written_barrier(struct bio *bio, int error) |
a9701a30 N |
487 | { |
488 | struct bio *bio2 = bio->bi_private; | |
489 | mdk_rdev_t *rdev = bio2->bi_private; | |
490 | mddev_t *mddev = rdev->mddev; | |
a9701a30 N |
491 | |
492 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
493 | error == -EOPNOTSUPP) { | |
494 | unsigned long flags; | |
495 | /* barriers don't appear to be supported :-( */ | |
496 | set_bit(BarriersNotsupp, &rdev->flags); | |
497 | mddev->barriers_work = 0; | |
498 | spin_lock_irqsave(&mddev->write_lock, flags); | |
499 | bio2->bi_next = mddev->biolist; | |
500 | mddev->biolist = bio2; | |
501 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
502 | wake_up(&mddev->sb_wait); | |
503 | bio_put(bio); | |
6712ecf8 N |
504 | } else { |
505 | bio_put(bio2); | |
506 | bio->bi_private = rdev; | |
507 | super_written(bio, error); | |
a9701a30 | 508 | } |
a9701a30 N |
509 | } |
510 | ||
7bfa19f2 N |
511 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
512 | sector_t sector, int size, struct page *page) | |
513 | { | |
514 | /* write first size bytes of page to sector of rdev | |
515 | * Increment mddev->pending_writes before returning | |
516 | * and decrement it on completion, waking up sb_wait | |
517 | * if zero is reached. | |
518 | * If an error occurred, call md_error | |
a9701a30 N |
519 | * |
520 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
521 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
522 | */ |
523 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
93dbb393 | 524 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG); |
7bfa19f2 N |
525 | |
526 | bio->bi_bdev = rdev->bdev; | |
527 | bio->bi_sector = sector; | |
528 | bio_add_page(bio, page, size, 0); | |
529 | bio->bi_private = rdev; | |
530 | bio->bi_end_io = super_written; | |
a9701a30 N |
531 | bio->bi_rw = rw; |
532 | ||
7bfa19f2 | 533 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
534 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
535 | struct bio *rbio; | |
536 | rw |= (1<<BIO_RW_BARRIER); | |
537 | rbio = bio_clone(bio, GFP_NOIO); | |
538 | rbio->bi_private = bio; | |
539 | rbio->bi_end_io = super_written_barrier; | |
540 | submit_bio(rw, rbio); | |
541 | } else | |
542 | submit_bio(rw, bio); | |
543 | } | |
544 | ||
545 | void md_super_wait(mddev_t *mddev) | |
546 | { | |
547 | /* wait for all superblock writes that were scheduled to complete. | |
548 | * if any had to be retried (due to BARRIER problems), retry them | |
549 | */ | |
550 | DEFINE_WAIT(wq); | |
551 | for(;;) { | |
552 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
553 | if (atomic_read(&mddev->pending_writes)==0) | |
554 | break; | |
555 | while (mddev->biolist) { | |
556 | struct bio *bio; | |
557 | spin_lock_irq(&mddev->write_lock); | |
558 | bio = mddev->biolist; | |
559 | mddev->biolist = bio->bi_next ; | |
560 | bio->bi_next = NULL; | |
561 | spin_unlock_irq(&mddev->write_lock); | |
562 | submit_bio(bio->bi_rw, bio); | |
563 | } | |
564 | schedule(); | |
565 | } | |
566 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
567 | } |
568 | ||
6712ecf8 | 569 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 570 | { |
1da177e4 | 571 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
572 | } |
573 | ||
a654b9d8 | 574 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
575 | struct page *page, int rw) |
576 | { | |
baaa2c51 | 577 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
578 | struct completion event; |
579 | int ret; | |
580 | ||
93dbb393 | 581 | rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG); |
1da177e4 LT |
582 | |
583 | bio->bi_bdev = bdev; | |
584 | bio->bi_sector = sector; | |
585 | bio_add_page(bio, page, size, 0); | |
586 | init_completion(&event); | |
587 | bio->bi_private = &event; | |
588 | bio->bi_end_io = bi_complete; | |
589 | submit_bio(rw, bio); | |
590 | wait_for_completion(&event); | |
591 | ||
592 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
593 | bio_put(bio); | |
594 | return ret; | |
595 | } | |
a8745db2 | 596 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 597 | |
0002b271 | 598 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
599 | { |
600 | char b[BDEVNAME_SIZE]; | |
601 | if (!rdev->sb_page) { | |
602 | MD_BUG(); | |
603 | return -EINVAL; | |
604 | } | |
605 | if (rdev->sb_loaded) | |
606 | return 0; | |
607 | ||
608 | ||
0f420358 | 609 | if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ)) |
1da177e4 LT |
610 | goto fail; |
611 | rdev->sb_loaded = 1; | |
612 | return 0; | |
613 | ||
614 | fail: | |
615 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
616 | bdevname(rdev->bdev,b)); | |
617 | return -EINVAL; | |
618 | } | |
619 | ||
620 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
621 | { | |
05710466 AN |
622 | return sb1->set_uuid0 == sb2->set_uuid0 && |
623 | sb1->set_uuid1 == sb2->set_uuid1 && | |
624 | sb1->set_uuid2 == sb2->set_uuid2 && | |
625 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
626 | } |
627 | ||
1da177e4 LT |
628 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
629 | { | |
630 | int ret; | |
631 | mdp_super_t *tmp1, *tmp2; | |
632 | ||
633 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
634 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
635 | ||
636 | if (!tmp1 || !tmp2) { | |
637 | ret = 0; | |
35020f1a | 638 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
639 | goto abort; |
640 | } | |
641 | ||
642 | *tmp1 = *sb1; | |
643 | *tmp2 = *sb2; | |
644 | ||
645 | /* | |
646 | * nr_disks is not constant | |
647 | */ | |
648 | tmp1->nr_disks = 0; | |
649 | tmp2->nr_disks = 0; | |
650 | ||
ce0c8e05 | 651 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 652 | abort: |
990a8baf JJ |
653 | kfree(tmp1); |
654 | kfree(tmp2); | |
1da177e4 LT |
655 | return ret; |
656 | } | |
657 | ||
4d167f09 N |
658 | |
659 | static u32 md_csum_fold(u32 csum) | |
660 | { | |
661 | csum = (csum & 0xffff) + (csum >> 16); | |
662 | return (csum & 0xffff) + (csum >> 16); | |
663 | } | |
664 | ||
1da177e4 LT |
665 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
666 | { | |
4d167f09 N |
667 | u64 newcsum = 0; |
668 | u32 *sb32 = (u32*)sb; | |
669 | int i; | |
1da177e4 LT |
670 | unsigned int disk_csum, csum; |
671 | ||
672 | disk_csum = sb->sb_csum; | |
673 | sb->sb_csum = 0; | |
4d167f09 N |
674 | |
675 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
676 | newcsum += sb32[i]; | |
677 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
678 | ||
679 | ||
680 | #ifdef CONFIG_ALPHA | |
681 | /* This used to use csum_partial, which was wrong for several | |
682 | * reasons including that different results are returned on | |
683 | * different architectures. It isn't critical that we get exactly | |
684 | * the same return value as before (we always csum_fold before | |
685 | * testing, and that removes any differences). However as we | |
686 | * know that csum_partial always returned a 16bit value on | |
687 | * alphas, do a fold to maximise conformity to previous behaviour. | |
688 | */ | |
689 | sb->sb_csum = md_csum_fold(disk_csum); | |
690 | #else | |
1da177e4 | 691 | sb->sb_csum = disk_csum; |
4d167f09 | 692 | #endif |
1da177e4 LT |
693 | return csum; |
694 | } | |
695 | ||
696 | ||
697 | /* | |
698 | * Handle superblock details. | |
699 | * We want to be able to handle multiple superblock formats | |
700 | * so we have a common interface to them all, and an array of | |
701 | * different handlers. | |
702 | * We rely on user-space to write the initial superblock, and support | |
703 | * reading and updating of superblocks. | |
704 | * Interface methods are: | |
705 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
706 | * loads and validates a superblock on dev. | |
707 | * if refdev != NULL, compare superblocks on both devices | |
708 | * Return: | |
709 | * 0 - dev has a superblock that is compatible with refdev | |
710 | * 1 - dev has a superblock that is compatible and newer than refdev | |
711 | * so dev should be used as the refdev in future | |
712 | * -EINVAL superblock incompatible or invalid | |
713 | * -othererror e.g. -EIO | |
714 | * | |
715 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
716 | * Verify that dev is acceptable into mddev. | |
717 | * The first time, mddev->raid_disks will be 0, and data from | |
718 | * dev should be merged in. Subsequent calls check that dev | |
719 | * is new enough. Return 0 or -EINVAL | |
720 | * | |
721 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
722 | * Update the superblock for rdev with data in mddev | |
723 | * This does not write to disc. | |
724 | * | |
725 | */ | |
726 | ||
727 | struct super_type { | |
0cd17fec CW |
728 | char *name; |
729 | struct module *owner; | |
730 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
731 | int minor_version); | |
732 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
733 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
734 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 735 | sector_t num_sectors); |
1da177e4 LT |
736 | }; |
737 | ||
738 | /* | |
739 | * load_super for 0.90.0 | |
740 | */ | |
741 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
742 | { | |
743 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
744 | mdp_super_t *sb; | |
745 | int ret; | |
1da177e4 LT |
746 | |
747 | /* | |
0f420358 | 748 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
749 | * it's at the end of the disk. |
750 | * | |
751 | * It also happens to be a multiple of 4Kb. | |
752 | */ | |
0f420358 | 753 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 754 | |
0002b271 | 755 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
756 | if (ret) return ret; |
757 | ||
758 | ret = -EINVAL; | |
759 | ||
760 | bdevname(rdev->bdev, b); | |
761 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
762 | ||
763 | if (sb->md_magic != MD_SB_MAGIC) { | |
764 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
765 | b); | |
766 | goto abort; | |
767 | } | |
768 | ||
769 | if (sb->major_version != 0 || | |
f6705578 N |
770 | sb->minor_version < 90 || |
771 | sb->minor_version > 91) { | |
1da177e4 LT |
772 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
773 | sb->major_version, sb->minor_version, | |
774 | b); | |
775 | goto abort; | |
776 | } | |
777 | ||
778 | if (sb->raid_disks <= 0) | |
779 | goto abort; | |
780 | ||
4d167f09 | 781 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
782 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
783 | b); | |
784 | goto abort; | |
785 | } | |
786 | ||
787 | rdev->preferred_minor = sb->md_minor; | |
788 | rdev->data_offset = 0; | |
0002b271 | 789 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 | 790 | |
e11e93fa N |
791 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) { |
792 | if (sb->level != 1 && sb->level != 4 | |
793 | && sb->level != 5 && sb->level != 6 | |
794 | && sb->level != 10) { | |
795 | /* FIXME use a better test */ | |
796 | printk(KERN_WARNING | |
797 | "md: bitmaps not supported for this level.\n"); | |
798 | goto abort; | |
799 | } | |
800 | } | |
801 | ||
1da177e4 LT |
802 | if (sb->level == LEVEL_MULTIPATH) |
803 | rdev->desc_nr = -1; | |
804 | else | |
805 | rdev->desc_nr = sb->this_disk.number; | |
806 | ||
9a7b2b0f | 807 | if (!refdev) { |
1da177e4 | 808 | ret = 1; |
9a7b2b0f | 809 | } else { |
1da177e4 LT |
810 | __u64 ev1, ev2; |
811 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
812 | if (!uuid_equal(refsb, sb)) { | |
813 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
814 | b, bdevname(refdev->bdev,b2)); | |
815 | goto abort; | |
816 | } | |
817 | if (!sb_equal(refsb, sb)) { | |
818 | printk(KERN_WARNING "md: %s has same UUID" | |
819 | " but different superblock to %s\n", | |
820 | b, bdevname(refdev->bdev, b2)); | |
821 | goto abort; | |
822 | } | |
823 | ev1 = md_event(sb); | |
824 | ev2 = md_event(refsb); | |
825 | if (ev1 > ev2) | |
826 | ret = 1; | |
827 | else | |
828 | ret = 0; | |
829 | } | |
8190e754 | 830 | rdev->sectors = rdev->sb_start; |
1da177e4 | 831 | |
dd8ac336 | 832 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
833 | /* "this cannot possibly happen" ... */ |
834 | ret = -EINVAL; | |
835 | ||
1da177e4 LT |
836 | abort: |
837 | return ret; | |
838 | } | |
839 | ||
840 | /* | |
841 | * validate_super for 0.90.0 | |
842 | */ | |
843 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
844 | { | |
845 | mdp_disk_t *desc; | |
846 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 847 | __u64 ev1 = md_event(sb); |
1da177e4 | 848 | |
41158c7e | 849 | rdev->raid_disk = -1; |
c5d79adb N |
850 | clear_bit(Faulty, &rdev->flags); |
851 | clear_bit(In_sync, &rdev->flags); | |
852 | clear_bit(WriteMostly, &rdev->flags); | |
853 | clear_bit(BarriersNotsupp, &rdev->flags); | |
854 | ||
1da177e4 LT |
855 | if (mddev->raid_disks == 0) { |
856 | mddev->major_version = 0; | |
857 | mddev->minor_version = sb->minor_version; | |
858 | mddev->patch_version = sb->patch_version; | |
e691063a | 859 | mddev->external = 0; |
9d8f0363 | 860 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
861 | mddev->ctime = sb->ctime; |
862 | mddev->utime = sb->utime; | |
863 | mddev->level = sb->level; | |
d9d166c2 | 864 | mddev->clevel[0] = 0; |
1da177e4 LT |
865 | mddev->layout = sb->layout; |
866 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 867 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 868 | mddev->events = ev1; |
9223214e | 869 | mddev->bitmap_offset = 0; |
36fa3063 | 870 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
1da177e4 | 871 | |
f6705578 N |
872 | if (mddev->minor_version >= 91) { |
873 | mddev->reshape_position = sb->reshape_position; | |
874 | mddev->delta_disks = sb->delta_disks; | |
875 | mddev->new_level = sb->new_level; | |
876 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 877 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
878 | } else { |
879 | mddev->reshape_position = MaxSector; | |
880 | mddev->delta_disks = 0; | |
881 | mddev->new_level = mddev->level; | |
882 | mddev->new_layout = mddev->layout; | |
664e7c41 | 883 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
884 | } |
885 | ||
1da177e4 LT |
886 | if (sb->state & (1<<MD_SB_CLEAN)) |
887 | mddev->recovery_cp = MaxSector; | |
888 | else { | |
889 | if (sb->events_hi == sb->cp_events_hi && | |
890 | sb->events_lo == sb->cp_events_lo) { | |
891 | mddev->recovery_cp = sb->recovery_cp; | |
892 | } else | |
893 | mddev->recovery_cp = 0; | |
894 | } | |
895 | ||
896 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
897 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
898 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
899 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
900 | ||
901 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
902 | |
903 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
e11e93fa | 904 | mddev->bitmap_file == NULL) |
36fa3063 | 905 | mddev->bitmap_offset = mddev->default_bitmap_offset; |
a654b9d8 | 906 | |
41158c7e N |
907 | } else if (mddev->pers == NULL) { |
908 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
909 | ++ev1; |
910 | if (ev1 < mddev->events) | |
911 | return -EINVAL; | |
41158c7e N |
912 | } else if (mddev->bitmap) { |
913 | /* if adding to array with a bitmap, then we can accept an | |
914 | * older device ... but not too old. | |
915 | */ | |
41158c7e N |
916 | if (ev1 < mddev->bitmap->events_cleared) |
917 | return 0; | |
07d84d10 N |
918 | } else { |
919 | if (ev1 < mddev->events) | |
920 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
921 | return 0; | |
922 | } | |
41158c7e | 923 | |
1da177e4 | 924 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
925 | desc = sb->disks + rdev->desc_nr; |
926 | ||
927 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 928 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
929 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
930 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 931 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
932 | rdev->raid_disk = desc->raid_disk; |
933 | } | |
8ddf9efe N |
934 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
935 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 936 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 937 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
938 | return 0; |
939 | } | |
940 | ||
941 | /* | |
942 | * sync_super for 0.90.0 | |
943 | */ | |
944 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
945 | { | |
946 | mdp_super_t *sb; | |
1da177e4 LT |
947 | mdk_rdev_t *rdev2; |
948 | int next_spare = mddev->raid_disks; | |
19133a42 | 949 | |
1da177e4 LT |
950 | |
951 | /* make rdev->sb match mddev data.. | |
952 | * | |
953 | * 1/ zero out disks | |
954 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
955 | * 3/ any empty disks < next_spare become removed | |
956 | * | |
957 | * disks[0] gets initialised to REMOVED because | |
958 | * we cannot be sure from other fields if it has | |
959 | * been initialised or not. | |
960 | */ | |
961 | int i; | |
962 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
963 | ||
61181565 N |
964 | rdev->sb_size = MD_SB_BYTES; |
965 | ||
1da177e4 LT |
966 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
967 | ||
968 | memset(sb, 0, sizeof(*sb)); | |
969 | ||
970 | sb->md_magic = MD_SB_MAGIC; | |
971 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
972 | sb->patch_version = mddev->patch_version; |
973 | sb->gvalid_words = 0; /* ignored */ | |
974 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
975 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
976 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
977 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
978 | ||
979 | sb->ctime = mddev->ctime; | |
980 | sb->level = mddev->level; | |
58c0fed4 | 981 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
982 | sb->raid_disks = mddev->raid_disks; |
983 | sb->md_minor = mddev->md_minor; | |
e691063a | 984 | sb->not_persistent = 0; |
1da177e4 LT |
985 | sb->utime = mddev->utime; |
986 | sb->state = 0; | |
987 | sb->events_hi = (mddev->events>>32); | |
988 | sb->events_lo = (u32)mddev->events; | |
989 | ||
f6705578 N |
990 | if (mddev->reshape_position == MaxSector) |
991 | sb->minor_version = 90; | |
992 | else { | |
993 | sb->minor_version = 91; | |
994 | sb->reshape_position = mddev->reshape_position; | |
995 | sb->new_level = mddev->new_level; | |
996 | sb->delta_disks = mddev->delta_disks; | |
997 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 998 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
999 | } |
1000 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1001 | if (mddev->in_sync) |
1002 | { | |
1003 | sb->recovery_cp = mddev->recovery_cp; | |
1004 | sb->cp_events_hi = (mddev->events>>32); | |
1005 | sb->cp_events_lo = (u32)mddev->events; | |
1006 | if (mddev->recovery_cp == MaxSector) | |
1007 | sb->state = (1<< MD_SB_CLEAN); | |
1008 | } else | |
1009 | sb->recovery_cp = 0; | |
1010 | ||
1011 | sb->layout = mddev->layout; | |
9d8f0363 | 1012 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1013 | |
a654b9d8 N |
1014 | if (mddev->bitmap && mddev->bitmap_file == NULL) |
1015 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); | |
1016 | ||
1da177e4 | 1017 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1018 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1019 | mdp_disk_t *d; |
86e6ffdd | 1020 | int desc_nr; |
b2d444d7 N |
1021 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
1022 | && !test_bit(Faulty, &rdev2->flags)) | |
86e6ffdd | 1023 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1024 | else |
86e6ffdd | 1025 | desc_nr = next_spare++; |
19133a42 | 1026 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1027 | d = &sb->disks[rdev2->desc_nr]; |
1028 | nr_disks++; | |
1029 | d->number = rdev2->desc_nr; | |
1030 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1031 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
b2d444d7 N |
1032 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
1033 | && !test_bit(Faulty, &rdev2->flags)) | |
1da177e4 LT |
1034 | d->raid_disk = rdev2->raid_disk; |
1035 | else | |
1036 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1037 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1038 | d->state = (1<<MD_DISK_FAULTY); |
1be7892f | 1039 | else if (test_bit(In_sync, &rdev2->flags)) { |
1da177e4 LT |
1040 | d->state = (1<<MD_DISK_ACTIVE); |
1041 | d->state |= (1<<MD_DISK_SYNC); | |
1042 | active++; | |
1043 | working++; | |
1044 | } else { | |
1045 | d->state = 0; | |
1046 | spare++; | |
1047 | working++; | |
1048 | } | |
8ddf9efe N |
1049 | if (test_bit(WriteMostly, &rdev2->flags)) |
1050 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1051 | } |
1da177e4 LT |
1052 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1053 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1054 | mdp_disk_t *d = &sb->disks[i]; | |
1055 | if (d->state == 0 && d->number == 0) { | |
1056 | d->number = i; | |
1057 | d->raid_disk = i; | |
1058 | d->state = (1<<MD_DISK_REMOVED); | |
1059 | d->state |= (1<<MD_DISK_FAULTY); | |
1060 | failed++; | |
1061 | } | |
1062 | } | |
1063 | sb->nr_disks = nr_disks; | |
1064 | sb->active_disks = active; | |
1065 | sb->working_disks = working; | |
1066 | sb->failed_disks = failed; | |
1067 | sb->spare_disks = spare; | |
1068 | ||
1069 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1070 | sb->sb_csum = calc_sb_csum(sb); | |
1071 | } | |
1072 | ||
0cd17fec CW |
1073 | /* |
1074 | * rdev_size_change for 0.90.0 | |
1075 | */ | |
1076 | static unsigned long long | |
15f4a5fd | 1077 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1078 | { |
58c0fed4 | 1079 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1080 | return 0; /* component must fit device */ |
0cd17fec CW |
1081 | if (rdev->mddev->bitmap_offset) |
1082 | return 0; /* can't move bitmap */ | |
0f420358 | 1083 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
15f4a5fd AN |
1084 | if (!num_sectors || num_sectors > rdev->sb_start) |
1085 | num_sectors = rdev->sb_start; | |
0f420358 | 1086 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1087 | rdev->sb_page); |
1088 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1089 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec CW |
1090 | } |
1091 | ||
1092 | ||
1da177e4 LT |
1093 | /* |
1094 | * version 1 superblock | |
1095 | */ | |
1096 | ||
1c05b4bc | 1097 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1098 | { |
1c05b4bc N |
1099 | __le32 disk_csum; |
1100 | u32 csum; | |
1da177e4 LT |
1101 | unsigned long long newcsum; |
1102 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1103 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1104 | int i; |
1105 | ||
1106 | disk_csum = sb->sb_csum; | |
1107 | sb->sb_csum = 0; | |
1108 | newcsum = 0; | |
1109 | for (i=0; size>=4; size -= 4 ) | |
1110 | newcsum += le32_to_cpu(*isuper++); | |
1111 | ||
1112 | if (size == 2) | |
1c05b4bc | 1113 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1114 | |
1115 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1116 | sb->sb_csum = disk_csum; | |
1117 | return cpu_to_le32(csum); | |
1118 | } | |
1119 | ||
1120 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1121 | { | |
1122 | struct mdp_superblock_1 *sb; | |
1123 | int ret; | |
0f420358 | 1124 | sector_t sb_start; |
1da177e4 | 1125 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1126 | int bmask; |
1da177e4 LT |
1127 | |
1128 | /* | |
0f420358 | 1129 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1130 | * It is always aligned to a 4K boundary and |
1131 | * depeding on minor_version, it can be: | |
1132 | * 0: At least 8K, but less than 12K, from end of device | |
1133 | * 1: At start of device | |
1134 | * 2: 4K from start of device. | |
1135 | */ | |
1136 | switch(minor_version) { | |
1137 | case 0: | |
0f420358 AN |
1138 | sb_start = rdev->bdev->bd_inode->i_size >> 9; |
1139 | sb_start -= 8*2; | |
1140 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1141 | break; |
1142 | case 1: | |
0f420358 | 1143 | sb_start = 0; |
1da177e4 LT |
1144 | break; |
1145 | case 2: | |
0f420358 | 1146 | sb_start = 8; |
1da177e4 LT |
1147 | break; |
1148 | default: | |
1149 | return -EINVAL; | |
1150 | } | |
0f420358 | 1151 | rdev->sb_start = sb_start; |
1da177e4 | 1152 | |
0002b271 N |
1153 | /* superblock is rarely larger than 1K, but it can be larger, |
1154 | * and it is safe to read 4k, so we do that | |
1155 | */ | |
1156 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1157 | if (ret) return ret; |
1158 | ||
1159 | ||
1160 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1161 | ||
1162 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1163 | sb->major_version != cpu_to_le32(1) || | |
1164 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1165 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1166 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1167 | return -EINVAL; |
1168 | ||
1169 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1170 | printk("md: invalid superblock checksum on %s\n", | |
1171 | bdevname(rdev->bdev,b)); | |
1172 | return -EINVAL; | |
1173 | } | |
1174 | if (le64_to_cpu(sb->data_size) < 10) { | |
1175 | printk("md: data_size too small on %s\n", | |
1176 | bdevname(rdev->bdev,b)); | |
1177 | return -EINVAL; | |
1178 | } | |
e11e93fa N |
1179 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) { |
1180 | if (sb->level != cpu_to_le32(1) && | |
1181 | sb->level != cpu_to_le32(4) && | |
1182 | sb->level != cpu_to_le32(5) && | |
1183 | sb->level != cpu_to_le32(6) && | |
1184 | sb->level != cpu_to_le32(10)) { | |
1185 | printk(KERN_WARNING | |
1186 | "md: bitmaps not supported for this level.\n"); | |
1187 | return -EINVAL; | |
1188 | } | |
1189 | } | |
1190 | ||
1da177e4 LT |
1191 | rdev->preferred_minor = 0xffff; |
1192 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1193 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1194 | |
0002b271 | 1195 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1196 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1197 | if (rdev->sb_size & bmask) |
a1801f85 N |
1198 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1199 | ||
1200 | if (minor_version | |
0f420358 | 1201 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1202 | return -EINVAL; |
0002b271 | 1203 | |
31b65a0d N |
1204 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1205 | rdev->desc_nr = -1; | |
1206 | else | |
1207 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1208 | ||
9a7b2b0f | 1209 | if (!refdev) { |
8ed75463 | 1210 | ret = 1; |
9a7b2b0f | 1211 | } else { |
1da177e4 LT |
1212 | __u64 ev1, ev2; |
1213 | struct mdp_superblock_1 *refsb = | |
1214 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1215 | ||
1216 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1217 | sb->level != refsb->level || | |
1218 | sb->layout != refsb->layout || | |
1219 | sb->chunksize != refsb->chunksize) { | |
1220 | printk(KERN_WARNING "md: %s has strangely different" | |
1221 | " superblock to %s\n", | |
1222 | bdevname(rdev->bdev,b), | |
1223 | bdevname(refdev->bdev,b2)); | |
1224 | return -EINVAL; | |
1225 | } | |
1226 | ev1 = le64_to_cpu(sb->events); | |
1227 | ev2 = le64_to_cpu(refsb->events); | |
1228 | ||
1229 | if (ev1 > ev2) | |
8ed75463 N |
1230 | ret = 1; |
1231 | else | |
1232 | ret = 0; | |
1da177e4 | 1233 | } |
a1801f85 | 1234 | if (minor_version) |
dd8ac336 AN |
1235 | rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) - |
1236 | le64_to_cpu(sb->data_offset); | |
1da177e4 | 1237 | else |
dd8ac336 AN |
1238 | rdev->sectors = rdev->sb_start; |
1239 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1240 | return -EINVAL; |
dd8ac336 | 1241 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1242 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1243 | return -EINVAL; |
8ed75463 | 1244 | return ret; |
1da177e4 LT |
1245 | } |
1246 | ||
1247 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1248 | { | |
1249 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1250 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1251 | |
41158c7e | 1252 | rdev->raid_disk = -1; |
c5d79adb N |
1253 | clear_bit(Faulty, &rdev->flags); |
1254 | clear_bit(In_sync, &rdev->flags); | |
1255 | clear_bit(WriteMostly, &rdev->flags); | |
1256 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1257 | ||
1da177e4 LT |
1258 | if (mddev->raid_disks == 0) { |
1259 | mddev->major_version = 1; | |
1260 | mddev->patch_version = 0; | |
e691063a | 1261 | mddev->external = 0; |
9d8f0363 | 1262 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1263 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1264 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1265 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1266 | mddev->clevel[0] = 0; |
1da177e4 LT |
1267 | mddev->layout = le32_to_cpu(sb->layout); |
1268 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1269 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1270 | mddev->events = ev1; |
9223214e | 1271 | mddev->bitmap_offset = 0; |
29fc7e3e | 1272 | mddev->default_bitmap_offset = 1024 >> 9; |
1da177e4 LT |
1273 | |
1274 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1275 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1276 | ||
1277 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1278 | |
71c0805c | 1279 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
e11e93fa | 1280 | mddev->bitmap_file == NULL ) |
a654b9d8 | 1281 | mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); |
e11e93fa | 1282 | |
f6705578 N |
1283 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1284 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1285 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1286 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1287 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1288 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1289 | } else { |
1290 | mddev->reshape_position = MaxSector; | |
1291 | mddev->delta_disks = 0; | |
1292 | mddev->new_level = mddev->level; | |
1293 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1294 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1295 | } |
1296 | ||
41158c7e N |
1297 | } else if (mddev->pers == NULL) { |
1298 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1299 | ++ev1; |
1300 | if (ev1 < mddev->events) | |
1301 | return -EINVAL; | |
41158c7e N |
1302 | } else if (mddev->bitmap) { |
1303 | /* If adding to array with a bitmap, then we can accept an | |
1304 | * older device, but not too old. | |
1305 | */ | |
41158c7e N |
1306 | if (ev1 < mddev->bitmap->events_cleared) |
1307 | return 0; | |
07d84d10 N |
1308 | } else { |
1309 | if (ev1 < mddev->events) | |
1310 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1311 | return 0; | |
1312 | } | |
1da177e4 LT |
1313 | if (mddev->level != LEVEL_MULTIPATH) { |
1314 | int role; | |
1da177e4 LT |
1315 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); |
1316 | switch(role) { | |
1317 | case 0xffff: /* spare */ | |
1da177e4 LT |
1318 | break; |
1319 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1320 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1321 | break; |
1322 | default: | |
5fd6c1dc N |
1323 | if ((le32_to_cpu(sb->feature_map) & |
1324 | MD_FEATURE_RECOVERY_OFFSET)) | |
1325 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1326 | else | |
1327 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1328 | rdev->raid_disk = role; |
1329 | break; | |
1330 | } | |
8ddf9efe N |
1331 | if (sb->devflags & WriteMostly1) |
1332 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1333 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1334 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1335 | |
1da177e4 LT |
1336 | return 0; |
1337 | } | |
1338 | ||
1339 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1340 | { | |
1341 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1342 | mdk_rdev_t *rdev2; |
1343 | int max_dev, i; | |
1344 | /* make rdev->sb match mddev and rdev data. */ | |
1345 | ||
1346 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1347 | ||
1348 | sb->feature_map = 0; | |
1349 | sb->pad0 = 0; | |
5fd6c1dc | 1350 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1351 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1352 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1353 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1354 | ||
1355 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1356 | sb->events = cpu_to_le64(mddev->events); | |
1357 | if (mddev->in_sync) | |
1358 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1359 | else | |
1360 | sb->resync_offset = cpu_to_le64(0); | |
1361 | ||
1c05b4bc | 1362 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1363 | |
f0ca340c | 1364 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1365 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1366 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1367 | sb->level = cpu_to_le32(mddev->level); |
1368 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1369 | |
a654b9d8 N |
1370 | if (mddev->bitmap && mddev->bitmap_file == NULL) { |
1371 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); | |
71c0805c | 1372 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1373 | } |
5fd6c1dc N |
1374 | |
1375 | if (rdev->raid_disk >= 0 && | |
97e4f42d N |
1376 | !test_bit(In_sync, &rdev->flags)) { |
1377 | if (mddev->curr_resync_completed > rdev->recovery_offset) | |
1378 | rdev->recovery_offset = mddev->curr_resync_completed; | |
1379 | if (rdev->recovery_offset > 0) { | |
1380 | sb->feature_map |= | |
1381 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1382 | sb->recovery_offset = | |
1383 | cpu_to_le64(rdev->recovery_offset); | |
1384 | } | |
5fd6c1dc N |
1385 | } |
1386 | ||
f6705578 N |
1387 | if (mddev->reshape_position != MaxSector) { |
1388 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1389 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1390 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1391 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1392 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1393 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1394 | } |
a654b9d8 | 1395 | |
1da177e4 | 1396 | max_dev = 0; |
159ec1fc | 1397 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1398 | if (rdev2->desc_nr+1 > max_dev) |
1399 | max_dev = rdev2->desc_nr+1; | |
a778b73f N |
1400 | |
1401 | if (max_dev > le32_to_cpu(sb->max_dev)) | |
1402 | sb->max_dev = cpu_to_le32(max_dev); | |
1da177e4 LT |
1403 | for (i=0; i<max_dev;i++) |
1404 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1405 | ||
159ec1fc | 1406 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1407 | i = rdev2->desc_nr; |
b2d444d7 | 1408 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1409 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1410 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1411 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
5fd6c1dc N |
1412 | else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0) |
1413 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); | |
1da177e4 LT |
1414 | else |
1415 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1416 | } | |
1417 | ||
1da177e4 LT |
1418 | sb->sb_csum = calc_sb_1_csum(sb); |
1419 | } | |
1420 | ||
0cd17fec | 1421 | static unsigned long long |
15f4a5fd | 1422 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1423 | { |
1424 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1425 | sector_t max_sectors; |
58c0fed4 | 1426 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1427 | return 0; /* component must fit device */ |
0f420358 | 1428 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1429 | /* minor versions 1 and 2; superblock before data */ |
15f4a5fd AN |
1430 | max_sectors = rdev->bdev->bd_inode->i_size >> 9; |
1431 | max_sectors -= rdev->data_offset; | |
1432 | if (!num_sectors || num_sectors > max_sectors) | |
1433 | num_sectors = max_sectors; | |
0cd17fec CW |
1434 | } else if (rdev->mddev->bitmap_offset) { |
1435 | /* minor version 0 with bitmap we can't move */ | |
1436 | return 0; | |
1437 | } else { | |
1438 | /* minor version 0; superblock after data */ | |
0f420358 AN |
1439 | sector_t sb_start; |
1440 | sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; | |
1441 | sb_start &= ~(sector_t)(4*2 - 1); | |
dd8ac336 | 1442 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1443 | if (!num_sectors || num_sectors > max_sectors) |
1444 | num_sectors = max_sectors; | |
0f420358 | 1445 | rdev->sb_start = sb_start; |
0cd17fec CW |
1446 | } |
1447 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1448 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1449 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1450 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1451 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1452 | rdev->sb_page); |
1453 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1454 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec | 1455 | } |
1da177e4 | 1456 | |
75c96f85 | 1457 | static struct super_type super_types[] = { |
1da177e4 LT |
1458 | [0] = { |
1459 | .name = "0.90.0", | |
1460 | .owner = THIS_MODULE, | |
0cd17fec CW |
1461 | .load_super = super_90_load, |
1462 | .validate_super = super_90_validate, | |
1463 | .sync_super = super_90_sync, | |
1464 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1465 | }, |
1466 | [1] = { | |
1467 | .name = "md-1", | |
1468 | .owner = THIS_MODULE, | |
0cd17fec CW |
1469 | .load_super = super_1_load, |
1470 | .validate_super = super_1_validate, | |
1471 | .sync_super = super_1_sync, | |
1472 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1473 | }, |
1474 | }; | |
1da177e4 LT |
1475 | |
1476 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1477 | { | |
7dd5e7c3 | 1478 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1479 | |
4b80991c N |
1480 | rcu_read_lock(); |
1481 | rdev_for_each_rcu(rdev, mddev1) | |
1482 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1483 | if (rdev->bdev->bd_contains == |
4b80991c N |
1484 | rdev2->bdev->bd_contains) { |
1485 | rcu_read_unlock(); | |
7dd5e7c3 | 1486 | return 1; |
4b80991c N |
1487 | } |
1488 | rcu_read_unlock(); | |
1da177e4 LT |
1489 | return 0; |
1490 | } | |
1491 | ||
1492 | static LIST_HEAD(pending_raid_disks); | |
1493 | ||
3f9d99c1 MP |
1494 | static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev) |
1495 | { | |
1496 | struct mdk_personality *pers = mddev->pers; | |
1497 | struct gendisk *disk = mddev->gendisk; | |
1498 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); | |
1499 | struct blk_integrity *bi_mddev = blk_get_integrity(disk); | |
1500 | ||
1501 | /* Data integrity passthrough not supported on RAID 4, 5 and 6 */ | |
1502 | if (pers && pers->level >= 4 && pers->level <= 6) | |
1503 | return; | |
1504 | ||
1505 | /* If rdev is integrity capable, register profile for mddev */ | |
1506 | if (!bi_mddev && bi_rdev) { | |
1507 | if (blk_integrity_register(disk, bi_rdev)) | |
1508 | printk(KERN_ERR "%s: %s Could not register integrity!\n", | |
1509 | __func__, disk->disk_name); | |
1510 | else | |
1511 | printk(KERN_NOTICE "Enabling data integrity on %s\n", | |
1512 | disk->disk_name); | |
1513 | return; | |
1514 | } | |
1515 | ||
1516 | /* Check that mddev and rdev have matching profiles */ | |
1517 | if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) { | |
1518 | printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__, | |
1519 | disk->disk_name, rdev->bdev->bd_disk->disk_name); | |
1520 | printk(KERN_NOTICE "Disabling data integrity on %s\n", | |
1521 | disk->disk_name); | |
1522 | blk_integrity_unregister(disk); | |
1523 | } | |
1524 | } | |
1525 | ||
1da177e4 LT |
1526 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1527 | { | |
7dd5e7c3 | 1528 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1529 | struct kobject *ko; |
1edf80d3 | 1530 | char *s; |
5e55e2f5 | 1531 | int err; |
1da177e4 LT |
1532 | |
1533 | if (rdev->mddev) { | |
1534 | MD_BUG(); | |
1535 | return -EINVAL; | |
1536 | } | |
11e2ede0 DW |
1537 | |
1538 | /* prevent duplicates */ | |
1539 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1540 | return -EEXIST; | |
1541 | ||
dd8ac336 AN |
1542 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1543 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1544 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1545 | if (mddev->pers) { |
1546 | /* Cannot change size, so fail | |
1547 | * If mddev->level <= 0, then we don't care | |
1548 | * about aligning sizes (e.g. linear) | |
1549 | */ | |
1550 | if (mddev->level > 0) | |
1551 | return -ENOSPC; | |
1552 | } else | |
dd8ac336 | 1553 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1554 | } |
1da177e4 LT |
1555 | |
1556 | /* Verify rdev->desc_nr is unique. | |
1557 | * If it is -1, assign a free number, else | |
1558 | * check number is not in use | |
1559 | */ | |
1560 | if (rdev->desc_nr < 0) { | |
1561 | int choice = 0; | |
1562 | if (mddev->pers) choice = mddev->raid_disks; | |
1563 | while (find_rdev_nr(mddev, choice)) | |
1564 | choice++; | |
1565 | rdev->desc_nr = choice; | |
1566 | } else { | |
1567 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1568 | return -EBUSY; | |
1569 | } | |
de01dfad N |
1570 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1571 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1572 | mdname(mddev), mddev->max_disks); | |
1573 | return -EBUSY; | |
1574 | } | |
19133a42 | 1575 | bdevname(rdev->bdev,b); |
649316b2 | 1576 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1577 | *s = '!'; |
649316b2 | 1578 | |
1da177e4 | 1579 | rdev->mddev = mddev; |
19133a42 | 1580 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1581 | |
b2d6db58 | 1582 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1583 | goto fail; |
86e6ffdd | 1584 | |
0762b8bd | 1585 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
5e55e2f5 N |
1586 | if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) { |
1587 | kobject_del(&rdev->kobj); | |
1588 | goto fail; | |
1589 | } | |
3c0ee63a N |
1590 | rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state"); |
1591 | ||
4b80991c | 1592 | list_add_rcu(&rdev->same_set, &mddev->disks); |
c5d79adb | 1593 | bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); |
4044ba58 N |
1594 | |
1595 | /* May as well allow recovery to be retried once */ | |
1596 | mddev->recovery_disabled = 0; | |
3f9d99c1 MP |
1597 | |
1598 | md_integrity_check(rdev, mddev); | |
1da177e4 | 1599 | return 0; |
5e55e2f5 N |
1600 | |
1601 | fail: | |
1602 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1603 | b, mdname(mddev)); | |
1604 | return err; | |
1da177e4 LT |
1605 | } |
1606 | ||
177a99b2 | 1607 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1608 | { |
1609 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1610 | kobject_del(&rdev->kobj); | |
177a99b2 | 1611 | kobject_put(&rdev->kobj); |
5792a285 N |
1612 | } |
1613 | ||
1da177e4 LT |
1614 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1615 | { | |
1616 | char b[BDEVNAME_SIZE]; | |
1617 | if (!rdev->mddev) { | |
1618 | MD_BUG(); | |
1619 | return; | |
1620 | } | |
5463c790 | 1621 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1622 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1623 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1624 | rdev->mddev = NULL; | |
86e6ffdd | 1625 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1626 | sysfs_put(rdev->sysfs_state); |
1627 | rdev->sysfs_state = NULL; | |
5792a285 | 1628 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1629 | * writing to 'remove' to "dev/state". We also need |
1630 | * to delay it due to rcu usage. | |
5792a285 | 1631 | */ |
4b80991c | 1632 | synchronize_rcu(); |
177a99b2 N |
1633 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1634 | kobject_get(&rdev->kobj); | |
5792a285 | 1635 | schedule_work(&rdev->del_work); |
1da177e4 LT |
1636 | } |
1637 | ||
1638 | /* | |
1639 | * prevent the device from being mounted, repartitioned or | |
1640 | * otherwise reused by a RAID array (or any other kernel | |
1641 | * subsystem), by bd_claiming the device. | |
1642 | */ | |
c5d79adb | 1643 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1644 | { |
1645 | int err = 0; | |
1646 | struct block_device *bdev; | |
1647 | char b[BDEVNAME_SIZE]; | |
1648 | ||
2e7b651d | 1649 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1650 | if (IS_ERR(bdev)) { |
1651 | printk(KERN_ERR "md: could not open %s.\n", | |
1652 | __bdevname(dev, b)); | |
1653 | return PTR_ERR(bdev); | |
1654 | } | |
c5d79adb | 1655 | err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); |
1da177e4 LT |
1656 | if (err) { |
1657 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1658 | bdevname(bdev, b)); | |
9a1c3542 | 1659 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1660 | return err; |
1661 | } | |
c5d79adb N |
1662 | if (!shared) |
1663 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1664 | rdev->bdev = bdev; |
1665 | return err; | |
1666 | } | |
1667 | ||
1668 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1669 | { | |
1670 | struct block_device *bdev = rdev->bdev; | |
1671 | rdev->bdev = NULL; | |
1672 | if (!bdev) | |
1673 | MD_BUG(); | |
1674 | bd_release(bdev); | |
9a1c3542 | 1675 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1676 | } |
1677 | ||
1678 | void md_autodetect_dev(dev_t dev); | |
1679 | ||
1680 | static void export_rdev(mdk_rdev_t * rdev) | |
1681 | { | |
1682 | char b[BDEVNAME_SIZE]; | |
1683 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1684 | bdevname(rdev->bdev,b)); | |
1685 | if (rdev->mddev) | |
1686 | MD_BUG(); | |
1687 | free_disk_sb(rdev); | |
1da177e4 | 1688 | #ifndef MODULE |
d0fae18f N |
1689 | if (test_bit(AutoDetected, &rdev->flags)) |
1690 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1691 | #endif |
1692 | unlock_rdev(rdev); | |
86e6ffdd | 1693 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1694 | } |
1695 | ||
1696 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1697 | { | |
1698 | unbind_rdev_from_array(rdev); | |
1699 | export_rdev(rdev); | |
1700 | } | |
1701 | ||
1702 | static void export_array(mddev_t *mddev) | |
1703 | { | |
159ec1fc | 1704 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 1705 | |
d089c6af | 1706 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1707 | if (!rdev->mddev) { |
1708 | MD_BUG(); | |
1709 | continue; | |
1710 | } | |
1711 | kick_rdev_from_array(rdev); | |
1712 | } | |
1713 | if (!list_empty(&mddev->disks)) | |
1714 | MD_BUG(); | |
1715 | mddev->raid_disks = 0; | |
1716 | mddev->major_version = 0; | |
1717 | } | |
1718 | ||
1719 | static void print_desc(mdp_disk_t *desc) | |
1720 | { | |
1721 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1722 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1723 | } | |
1724 | ||
cd2ac932 | 1725 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
1726 | { |
1727 | int i; | |
1728 | ||
1729 | printk(KERN_INFO | |
1730 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1731 | sb->major_version, sb->minor_version, sb->patch_version, | |
1732 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1733 | sb->ctime); | |
1734 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1735 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1736 | sb->md_minor, sb->layout, sb->chunk_size); | |
1737 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1738 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1739 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1740 | sb->failed_disks, sb->spare_disks, | |
1741 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1742 | ||
1743 | printk(KERN_INFO); | |
1744 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1745 | mdp_disk_t *desc; | |
1746 | ||
1747 | desc = sb->disks + i; | |
1748 | if (desc->number || desc->major || desc->minor || | |
1749 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1750 | printk(" D %2d: ", i); | |
1751 | print_desc(desc); | |
1752 | } | |
1753 | } | |
1754 | printk(KERN_INFO "md: THIS: "); | |
1755 | print_desc(&sb->this_disk); | |
cd2ac932 | 1756 | } |
1da177e4 | 1757 | |
cd2ac932 CR |
1758 | static void print_sb_1(struct mdp_superblock_1 *sb) |
1759 | { | |
1760 | __u8 *uuid; | |
1761 | ||
1762 | uuid = sb->set_uuid; | |
1763 | printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x" | |
1764 | ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n" | |
1765 | KERN_INFO "md: Name: \"%s\" CT:%llu\n", | |
1766 | le32_to_cpu(sb->major_version), | |
1767 | le32_to_cpu(sb->feature_map), | |
1768 | uuid[0], uuid[1], uuid[2], uuid[3], | |
1769 | uuid[4], uuid[5], uuid[6], uuid[7], | |
1770 | uuid[8], uuid[9], uuid[10], uuid[11], | |
1771 | uuid[12], uuid[13], uuid[14], uuid[15], | |
1772 | sb->set_name, | |
1773 | (unsigned long long)le64_to_cpu(sb->ctime) | |
1774 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
1775 | ||
1776 | uuid = sb->device_uuid; | |
1777 | printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
1778 | " RO:%llu\n" | |
1779 | KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x" | |
1780 | ":%02x%02x%02x%02x%02x%02x\n" | |
1781 | KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" | |
1782 | KERN_INFO "md: (MaxDev:%u) \n", | |
1783 | le32_to_cpu(sb->level), | |
1784 | (unsigned long long)le64_to_cpu(sb->size), | |
1785 | le32_to_cpu(sb->raid_disks), | |
1786 | le32_to_cpu(sb->layout), | |
1787 | le32_to_cpu(sb->chunksize), | |
1788 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
1789 | (unsigned long long)le64_to_cpu(sb->data_size), | |
1790 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
1791 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
1792 | le32_to_cpu(sb->dev_number), | |
1793 | uuid[0], uuid[1], uuid[2], uuid[3], | |
1794 | uuid[4], uuid[5], uuid[6], uuid[7], | |
1795 | uuid[8], uuid[9], uuid[10], uuid[11], | |
1796 | uuid[12], uuid[13], uuid[14], uuid[15], | |
1797 | sb->devflags, | |
1798 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
1799 | (unsigned long long)le64_to_cpu(sb->events), | |
1800 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
1801 | le32_to_cpu(sb->sb_csum), | |
1802 | le32_to_cpu(sb->max_dev) | |
1803 | ); | |
1da177e4 LT |
1804 | } |
1805 | ||
cd2ac932 | 1806 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
1807 | { |
1808 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
1809 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
1810 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
1811 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
1812 | rdev->desc_nr); | |
1da177e4 | 1813 | if (rdev->sb_loaded) { |
cd2ac932 CR |
1814 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
1815 | switch (major_version) { | |
1816 | case 0: | |
1817 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
1818 | break; | |
1819 | case 1: | |
1820 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
1821 | break; | |
1822 | } | |
1da177e4 LT |
1823 | } else |
1824 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
1825 | } | |
1826 | ||
5e56341d | 1827 | static void md_print_devices(void) |
1da177e4 | 1828 | { |
159ec1fc | 1829 | struct list_head *tmp; |
1da177e4 LT |
1830 | mdk_rdev_t *rdev; |
1831 | mddev_t *mddev; | |
1832 | char b[BDEVNAME_SIZE]; | |
1833 | ||
1834 | printk("\n"); | |
1835 | printk("md: **********************************\n"); | |
1836 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
1837 | printk("md: **********************************\n"); | |
29ac4aa3 | 1838 | for_each_mddev(mddev, tmp) { |
1da177e4 | 1839 | |
32a7627c N |
1840 | if (mddev->bitmap) |
1841 | bitmap_print_sb(mddev->bitmap); | |
1842 | else | |
1843 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 1844 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
1845 | printk("<%s>", bdevname(rdev->bdev,b)); |
1846 | printk("\n"); | |
1847 | ||
159ec1fc | 1848 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 1849 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
1850 | } |
1851 | printk("md: **********************************\n"); | |
1852 | printk("\n"); | |
1853 | } | |
1854 | ||
1855 | ||
42543769 | 1856 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 1857 | { |
42543769 N |
1858 | /* Update each superblock (in-memory image), but |
1859 | * if we are allowed to, skip spares which already | |
1860 | * have the right event counter, or have one earlier | |
1861 | * (which would mean they aren't being marked as dirty | |
1862 | * with the rest of the array) | |
1863 | */ | |
1da177e4 | 1864 | mdk_rdev_t *rdev; |
1da177e4 | 1865 | |
159ec1fc | 1866 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
1867 | if (rdev->sb_events == mddev->events || |
1868 | (nospares && | |
1869 | rdev->raid_disk < 0 && | |
1870 | (rdev->sb_events&1)==0 && | |
1871 | rdev->sb_events+1 == mddev->events)) { | |
1872 | /* Don't update this superblock */ | |
1873 | rdev->sb_loaded = 2; | |
1874 | } else { | |
1875 | super_types[mddev->major_version]. | |
1876 | sync_super(mddev, rdev); | |
1877 | rdev->sb_loaded = 1; | |
1878 | } | |
1da177e4 LT |
1879 | } |
1880 | } | |
1881 | ||
850b2b42 | 1882 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 1883 | { |
1da177e4 | 1884 | mdk_rdev_t *rdev; |
06d91a5f | 1885 | int sync_req; |
42543769 | 1886 | int nospares = 0; |
1da177e4 | 1887 | |
1b57f132 | 1888 | mddev->utime = get_seconds(); |
8377bc80 N |
1889 | if (mddev->external) |
1890 | return; | |
1da177e4 | 1891 | repeat: |
a9701a30 | 1892 | spin_lock_irq(&mddev->write_lock); |
84692195 | 1893 | |
850b2b42 N |
1894 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
1895 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) | |
1896 | force_change = 1; | |
1897 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
1898 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
1899 | * though if events isn't the right even/odd, we will have to do | |
1900 | * spares after all | |
1901 | */ | |
1902 | nospares = 1; | |
1903 | if (force_change) | |
1904 | nospares = 0; | |
1905 | if (mddev->degraded) | |
84692195 N |
1906 | /* If the array is degraded, then skipping spares is both |
1907 | * dangerous and fairly pointless. | |
1908 | * Dangerous because a device that was removed from the array | |
1909 | * might have a event_count that still looks up-to-date, | |
1910 | * so it can be re-added without a resync. | |
1911 | * Pointless because if there are any spares to skip, | |
1912 | * then a recovery will happen and soon that array won't | |
1913 | * be degraded any more and the spare can go back to sleep then. | |
1914 | */ | |
850b2b42 | 1915 | nospares = 0; |
84692195 | 1916 | |
06d91a5f | 1917 | sync_req = mddev->in_sync; |
42543769 N |
1918 | |
1919 | /* If this is just a dirty<->clean transition, and the array is clean | |
1920 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 1921 | if (nospares |
42543769 | 1922 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
1031be7a N |
1923 | && (mddev->events & 1) |
1924 | && mddev->events != 1) | |
42543769 N |
1925 | mddev->events--; |
1926 | else { | |
1927 | /* otherwise we have to go forward and ... */ | |
1928 | mddev->events ++; | |
1929 | if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ | |
1930 | /* .. if the array isn't clean, insist on an odd 'events' */ | |
1931 | if ((mddev->events&1)==0) { | |
1932 | mddev->events++; | |
1933 | nospares = 0; | |
1934 | } | |
1935 | } else { | |
1936 | /* otherwise insist on an even 'events' (for clean states) */ | |
1937 | if ((mddev->events&1)) { | |
1938 | mddev->events++; | |
1939 | nospares = 0; | |
1940 | } | |
1941 | } | |
1942 | } | |
1da177e4 LT |
1943 | |
1944 | if (!mddev->events) { | |
1945 | /* | |
1946 | * oops, this 64-bit counter should never wrap. | |
1947 | * Either we are in around ~1 trillion A.C., assuming | |
1948 | * 1 reboot per second, or we have a bug: | |
1949 | */ | |
1950 | MD_BUG(); | |
1951 | mddev->events --; | |
1952 | } | |
1da177e4 LT |
1953 | |
1954 | /* | |
1955 | * do not write anything to disk if using | |
1956 | * nonpersistent superblocks | |
1957 | */ | |
06d91a5f | 1958 | if (!mddev->persistent) { |
e691063a N |
1959 | if (!mddev->external) |
1960 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
1961 | ||
a9701a30 | 1962 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1963 | wake_up(&mddev->sb_wait); |
1da177e4 | 1964 | return; |
06d91a5f | 1965 | } |
e691063a | 1966 | sync_sbs(mddev, nospares); |
a9701a30 | 1967 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
1968 | |
1969 | dprintk(KERN_INFO | |
1970 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
1971 | mdname(mddev),mddev->in_sync); | |
1972 | ||
4ad13663 | 1973 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 1974 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
1975 | char b[BDEVNAME_SIZE]; |
1976 | dprintk(KERN_INFO "md: "); | |
42543769 N |
1977 | if (rdev->sb_loaded != 1) |
1978 | continue; /* no noise on spare devices */ | |
b2d444d7 | 1979 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
1980 | dprintk("(skipping faulty "); |
1981 | ||
1982 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 1983 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 1984 | md_super_write(mddev,rdev, |
0f420358 | 1985 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
1986 | rdev->sb_page); |
1987 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
1988 | bdevname(rdev->bdev,b), | |
0f420358 | 1989 | (unsigned long long)rdev->sb_start); |
42543769 | 1990 | rdev->sb_events = mddev->events; |
7bfa19f2 | 1991 | |
1da177e4 LT |
1992 | } else |
1993 | dprintk(")\n"); | |
7bfa19f2 | 1994 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
1995 | /* only need to write one superblock... */ |
1996 | break; | |
1997 | } | |
a9701a30 | 1998 | md_super_wait(mddev); |
850b2b42 | 1999 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2000 | |
a9701a30 | 2001 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2002 | if (mddev->in_sync != sync_req || |
2003 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2004 | /* have to write it out again */ |
a9701a30 | 2005 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2006 | goto repeat; |
2007 | } | |
850b2b42 | 2008 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2009 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2010 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2011 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2012 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2013 | |
1da177e4 LT |
2014 | } |
2015 | ||
7f6ce769 | 2016 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2017 | * We want to accept with case. For this we use cmd_match. |
2018 | */ | |
2019 | static int cmd_match(const char *cmd, const char *str) | |
2020 | { | |
2021 | /* See if cmd, written into a sysfs file, matches | |
2022 | * str. They must either be the same, or cmd can | |
2023 | * have a trailing newline | |
2024 | */ | |
2025 | while (*cmd && *str && *cmd == *str) { | |
2026 | cmd++; | |
2027 | str++; | |
2028 | } | |
2029 | if (*cmd == '\n') | |
2030 | cmd++; | |
2031 | if (*str || *cmd) | |
2032 | return 0; | |
2033 | return 1; | |
2034 | } | |
2035 | ||
86e6ffdd N |
2036 | struct rdev_sysfs_entry { |
2037 | struct attribute attr; | |
2038 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2039 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2040 | }; | |
2041 | ||
2042 | static ssize_t | |
96de1e66 | 2043 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2044 | { |
2045 | char *sep = ""; | |
20a49ff6 | 2046 | size_t len = 0; |
86e6ffdd | 2047 | |
b2d444d7 | 2048 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2049 | len+= sprintf(page+len, "%sfaulty",sep); |
2050 | sep = ","; | |
2051 | } | |
b2d444d7 | 2052 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2053 | len += sprintf(page+len, "%sin_sync",sep); |
2054 | sep = ","; | |
2055 | } | |
f655675b N |
2056 | if (test_bit(WriteMostly, &rdev->flags)) { |
2057 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2058 | sep = ","; | |
2059 | } | |
6bfe0b49 DW |
2060 | if (test_bit(Blocked, &rdev->flags)) { |
2061 | len += sprintf(page+len, "%sblocked", sep); | |
2062 | sep = ","; | |
2063 | } | |
b2d444d7 N |
2064 | if (!test_bit(Faulty, &rdev->flags) && |
2065 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2066 | len += sprintf(page+len, "%sspare", sep); |
2067 | sep = ","; | |
2068 | } | |
2069 | return len+sprintf(page+len, "\n"); | |
2070 | } | |
2071 | ||
45dc2de1 N |
2072 | static ssize_t |
2073 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2074 | { | |
2075 | /* can write | |
2076 | * faulty - simulates and error | |
2077 | * remove - disconnects the device | |
f655675b N |
2078 | * writemostly - sets write_mostly |
2079 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2080 | * blocked - sets the Blocked flag |
2081 | * -blocked - clears the Blocked flag | |
6d56e278 | 2082 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2083 | */ |
2084 | int err = -EINVAL; | |
2085 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2086 | md_error(rdev->mddev, rdev); | |
2087 | err = 0; | |
2088 | } else if (cmd_match(buf, "remove")) { | |
2089 | if (rdev->raid_disk >= 0) | |
2090 | err = -EBUSY; | |
2091 | else { | |
2092 | mddev_t *mddev = rdev->mddev; | |
2093 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2094 | if (mddev->pers) |
2095 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2096 | md_new_event(mddev); |
2097 | err = 0; | |
2098 | } | |
f655675b N |
2099 | } else if (cmd_match(buf, "writemostly")) { |
2100 | set_bit(WriteMostly, &rdev->flags); | |
2101 | err = 0; | |
2102 | } else if (cmd_match(buf, "-writemostly")) { | |
2103 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2104 | err = 0; |
2105 | } else if (cmd_match(buf, "blocked")) { | |
2106 | set_bit(Blocked, &rdev->flags); | |
2107 | err = 0; | |
2108 | } else if (cmd_match(buf, "-blocked")) { | |
2109 | clear_bit(Blocked, &rdev->flags); | |
2110 | wake_up(&rdev->blocked_wait); | |
2111 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2112 | md_wakeup_thread(rdev->mddev->thread); | |
2113 | ||
6d56e278 N |
2114 | err = 0; |
2115 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2116 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2117 | err = 0; |
45dc2de1 | 2118 | } |
3c0ee63a N |
2119 | if (!err && rdev->sysfs_state) |
2120 | sysfs_notify_dirent(rdev->sysfs_state); | |
45dc2de1 N |
2121 | return err ? err : len; |
2122 | } | |
80ca3a44 N |
2123 | static struct rdev_sysfs_entry rdev_state = |
2124 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2125 | |
4dbcdc75 N |
2126 | static ssize_t |
2127 | errors_show(mdk_rdev_t *rdev, char *page) | |
2128 | { | |
2129 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2130 | } | |
2131 | ||
2132 | static ssize_t | |
2133 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2134 | { | |
2135 | char *e; | |
2136 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2137 | if (*buf && (*e == 0 || *e == '\n')) { | |
2138 | atomic_set(&rdev->corrected_errors, n); | |
2139 | return len; | |
2140 | } | |
2141 | return -EINVAL; | |
2142 | } | |
2143 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2144 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2145 | |
014236d2 N |
2146 | static ssize_t |
2147 | slot_show(mdk_rdev_t *rdev, char *page) | |
2148 | { | |
2149 | if (rdev->raid_disk < 0) | |
2150 | return sprintf(page, "none\n"); | |
2151 | else | |
2152 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2153 | } | |
2154 | ||
2155 | static ssize_t | |
2156 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2157 | { | |
2158 | char *e; | |
c303da6d N |
2159 | int err; |
2160 | char nm[20]; | |
014236d2 N |
2161 | int slot = simple_strtoul(buf, &e, 10); |
2162 | if (strncmp(buf, "none", 4)==0) | |
2163 | slot = -1; | |
2164 | else if (e==buf || (*e && *e!= '\n')) | |
2165 | return -EINVAL; | |
6c2fce2e | 2166 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2167 | /* Setting 'slot' on an active array requires also |
2168 | * updating the 'rd%d' link, and communicating | |
2169 | * with the personality with ->hot_*_disk. | |
2170 | * For now we only support removing | |
2171 | * failed/spare devices. This normally happens automatically, | |
2172 | * but not when the metadata is externally managed. | |
2173 | */ | |
c303da6d N |
2174 | if (rdev->raid_disk == -1) |
2175 | return -EEXIST; | |
2176 | /* personality does all needed checks */ | |
2177 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2178 | return -EINVAL; | |
2179 | err = rdev->mddev->pers-> | |
2180 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2181 | if (err) | |
2182 | return err; | |
2183 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2184 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
2185 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2186 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2187 | } else if (rdev->mddev->pers) { |
2188 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2189 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2190 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2191 | */ |
2192 | ||
2193 | if (rdev->raid_disk != -1) | |
2194 | return -EBUSY; | |
2195 | ||
2196 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2197 | return -EINVAL; | |
2198 | ||
159ec1fc | 2199 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2200 | if (rdev2->raid_disk == slot) |
2201 | return -EEXIST; | |
2202 | ||
2203 | rdev->raid_disk = slot; | |
2204 | if (test_bit(In_sync, &rdev->flags)) | |
2205 | rdev->saved_raid_disk = slot; | |
2206 | else | |
2207 | rdev->saved_raid_disk = -1; | |
2208 | err = rdev->mddev->pers-> | |
2209 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2210 | if (err) { |
6c2fce2e | 2211 | rdev->raid_disk = -1; |
6c2fce2e | 2212 | return err; |
52664732 | 2213 | } else |
3c0ee63a | 2214 | sysfs_notify_dirent(rdev->sysfs_state); |
6c2fce2e NB |
2215 | sprintf(nm, "rd%d", rdev->raid_disk); |
2216 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
2217 | printk(KERN_WARNING | |
2218 | "md: cannot register " | |
2219 | "%s for %s\n", | |
2220 | nm, mdname(rdev->mddev)); | |
2221 | ||
2222 | /* don't wakeup anyone, leave that to userspace. */ | |
c303da6d N |
2223 | } else { |
2224 | if (slot >= rdev->mddev->raid_disks) | |
2225 | return -ENOSPC; | |
2226 | rdev->raid_disk = slot; | |
2227 | /* assume it is working */ | |
c5d79adb N |
2228 | clear_bit(Faulty, &rdev->flags); |
2229 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2230 | set_bit(In_sync, &rdev->flags); |
3c0ee63a | 2231 | sysfs_notify_dirent(rdev->sysfs_state); |
c303da6d | 2232 | } |
014236d2 N |
2233 | return len; |
2234 | } | |
2235 | ||
2236 | ||
2237 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2238 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2239 | |
93c8cad0 N |
2240 | static ssize_t |
2241 | offset_show(mdk_rdev_t *rdev, char *page) | |
2242 | { | |
6961ece4 | 2243 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2244 | } |
2245 | ||
2246 | static ssize_t | |
2247 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2248 | { | |
2249 | char *e; | |
2250 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2251 | if (e==buf || (*e && *e != '\n')) | |
2252 | return -EINVAL; | |
8ed0a521 | 2253 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2254 | return -EBUSY; |
dd8ac336 | 2255 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2256 | /* Must set offset before size, so overlap checks |
2257 | * can be sane */ | |
2258 | return -EBUSY; | |
93c8cad0 N |
2259 | rdev->data_offset = offset; |
2260 | return len; | |
2261 | } | |
2262 | ||
2263 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2264 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2265 | |
83303b61 N |
2266 | static ssize_t |
2267 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2268 | { | |
dd8ac336 | 2269 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2270 | } |
2271 | ||
c5d79adb N |
2272 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2273 | { | |
2274 | /* check if two start/length pairs overlap */ | |
2275 | if (s1+l1 <= s2) | |
2276 | return 0; | |
2277 | if (s2+l2 <= s1) | |
2278 | return 0; | |
2279 | return 1; | |
2280 | } | |
2281 | ||
b522adcd DW |
2282 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2283 | { | |
2284 | unsigned long long blocks; | |
2285 | sector_t new; | |
2286 | ||
2287 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2288 | return -EINVAL; | |
2289 | ||
2290 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2291 | return -EINVAL; /* sector conversion overflow */ | |
2292 | ||
2293 | new = blocks * 2; | |
2294 | if (new != blocks * 2) | |
2295 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2296 | ||
2297 | *sectors = new; | |
2298 | return 0; | |
2299 | } | |
2300 | ||
83303b61 N |
2301 | static ssize_t |
2302 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2303 | { | |
27c529bb | 2304 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2305 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2306 | sector_t sectors; |
27c529bb | 2307 | |
b522adcd | 2308 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2309 | return -EINVAL; |
0cd17fec | 2310 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2311 | if (my_mddev->persistent) { |
dd8ac336 AN |
2312 | sectors = super_types[my_mddev->major_version]. |
2313 | rdev_size_change(rdev, sectors); | |
2314 | if (!sectors) | |
0cd17fec | 2315 | return -EBUSY; |
dd8ac336 AN |
2316 | } else if (!sectors) |
2317 | sectors = (rdev->bdev->bd_inode->i_size >> 9) - | |
2318 | rdev->data_offset; | |
0cd17fec | 2319 | } |
dd8ac336 | 2320 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2321 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2322 | |
dd8ac336 AN |
2323 | rdev->sectors = sectors; |
2324 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2325 | /* need to check that all other rdevs with the same ->bdev |
2326 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2327 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2328 | * we have to change it back, we will have the lock again. |
2329 | */ | |
2330 | mddev_t *mddev; | |
2331 | int overlap = 0; | |
159ec1fc | 2332 | struct list_head *tmp; |
c5d79adb | 2333 | |
27c529bb | 2334 | mddev_unlock(my_mddev); |
29ac4aa3 | 2335 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2336 | mdk_rdev_t *rdev2; |
2337 | ||
2338 | mddev_lock(mddev); | |
159ec1fc | 2339 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
c5d79adb N |
2340 | if (test_bit(AllReserved, &rdev2->flags) || |
2341 | (rdev->bdev == rdev2->bdev && | |
2342 | rdev != rdev2 && | |
dd8ac336 | 2343 | overlaps(rdev->data_offset, rdev->sectors, |
d07bd3bc | 2344 | rdev2->data_offset, |
dd8ac336 | 2345 | rdev2->sectors))) { |
c5d79adb N |
2346 | overlap = 1; |
2347 | break; | |
2348 | } | |
2349 | mddev_unlock(mddev); | |
2350 | if (overlap) { | |
2351 | mddev_put(mddev); | |
2352 | break; | |
2353 | } | |
2354 | } | |
27c529bb | 2355 | mddev_lock(my_mddev); |
c5d79adb N |
2356 | if (overlap) { |
2357 | /* Someone else could have slipped in a size | |
2358 | * change here, but doing so is just silly. | |
dd8ac336 | 2359 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2360 | * safe, and trust userspace not to race with |
2361 | * itself | |
2362 | */ | |
dd8ac336 | 2363 | rdev->sectors = oldsectors; |
c5d79adb N |
2364 | return -EBUSY; |
2365 | } | |
2366 | } | |
83303b61 N |
2367 | return len; |
2368 | } | |
2369 | ||
2370 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2371 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2372 | |
86e6ffdd N |
2373 | static struct attribute *rdev_default_attrs[] = { |
2374 | &rdev_state.attr, | |
4dbcdc75 | 2375 | &rdev_errors.attr, |
014236d2 | 2376 | &rdev_slot.attr, |
93c8cad0 | 2377 | &rdev_offset.attr, |
83303b61 | 2378 | &rdev_size.attr, |
86e6ffdd N |
2379 | NULL, |
2380 | }; | |
2381 | static ssize_t | |
2382 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2383 | { | |
2384 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2385 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2386 | mddev_t *mddev = rdev->mddev; |
2387 | ssize_t rv; | |
86e6ffdd N |
2388 | |
2389 | if (!entry->show) | |
2390 | return -EIO; | |
27c529bb N |
2391 | |
2392 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2393 | if (!rv) { | |
2394 | if (rdev->mddev == NULL) | |
2395 | rv = -EBUSY; | |
2396 | else | |
2397 | rv = entry->show(rdev, page); | |
2398 | mddev_unlock(mddev); | |
2399 | } | |
2400 | return rv; | |
86e6ffdd N |
2401 | } |
2402 | ||
2403 | static ssize_t | |
2404 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2405 | const char *page, size_t length) | |
2406 | { | |
2407 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2408 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2409 | ssize_t rv; |
2410 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2411 | |
2412 | if (!entry->store) | |
2413 | return -EIO; | |
67463acb N |
2414 | if (!capable(CAP_SYS_ADMIN)) |
2415 | return -EACCES; | |
27c529bb | 2416 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2417 | if (!rv) { |
27c529bb N |
2418 | if (rdev->mddev == NULL) |
2419 | rv = -EBUSY; | |
2420 | else | |
2421 | rv = entry->store(rdev, page, length); | |
6a51830e | 2422 | mddev_unlock(mddev); |
ca388059 N |
2423 | } |
2424 | return rv; | |
86e6ffdd N |
2425 | } |
2426 | ||
2427 | static void rdev_free(struct kobject *ko) | |
2428 | { | |
2429 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2430 | kfree(rdev); | |
2431 | } | |
2432 | static struct sysfs_ops rdev_sysfs_ops = { | |
2433 | .show = rdev_attr_show, | |
2434 | .store = rdev_attr_store, | |
2435 | }; | |
2436 | static struct kobj_type rdev_ktype = { | |
2437 | .release = rdev_free, | |
2438 | .sysfs_ops = &rdev_sysfs_ops, | |
2439 | .default_attrs = rdev_default_attrs, | |
2440 | }; | |
2441 | ||
1da177e4 LT |
2442 | /* |
2443 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2444 | * | |
2445 | * mark the device faulty if: | |
2446 | * | |
2447 | * - the device is nonexistent (zero size) | |
2448 | * - the device has no valid superblock | |
2449 | * | |
2450 | * a faulty rdev _never_ has rdev->sb set. | |
2451 | */ | |
2452 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2453 | { | |
2454 | char b[BDEVNAME_SIZE]; | |
2455 | int err; | |
2456 | mdk_rdev_t *rdev; | |
2457 | sector_t size; | |
2458 | ||
9ffae0cf | 2459 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2460 | if (!rdev) { |
2461 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2462 | return ERR_PTR(-ENOMEM); | |
2463 | } | |
1da177e4 LT |
2464 | |
2465 | if ((err = alloc_disk_sb(rdev))) | |
2466 | goto abort_free; | |
2467 | ||
c5d79adb | 2468 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2469 | if (err) |
2470 | goto abort_free; | |
2471 | ||
f9cb074b | 2472 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2473 | |
1da177e4 | 2474 | rdev->desc_nr = -1; |
2b6e8459 | 2475 | rdev->saved_raid_disk = -1; |
3f9d7b0d | 2476 | rdev->raid_disk = -1; |
b2d444d7 | 2477 | rdev->flags = 0; |
1da177e4 | 2478 | rdev->data_offset = 0; |
42543769 | 2479 | rdev->sb_events = 0; |
1da177e4 | 2480 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 2481 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 2482 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
2483 | |
2484 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
2485 | if (!size) { | |
2486 | printk(KERN_WARNING | |
2487 | "md: %s has zero or unknown size, marking faulty!\n", | |
2488 | bdevname(rdev->bdev,b)); | |
2489 | err = -EINVAL; | |
2490 | goto abort_free; | |
2491 | } | |
2492 | ||
2493 | if (super_format >= 0) { | |
2494 | err = super_types[super_format]. | |
2495 | load_super(rdev, NULL, super_minor); | |
2496 | if (err == -EINVAL) { | |
df968c4e N |
2497 | printk(KERN_WARNING |
2498 | "md: %s does not have a valid v%d.%d " | |
2499 | "superblock, not importing!\n", | |
2500 | bdevname(rdev->bdev,b), | |
2501 | super_format, super_minor); | |
1da177e4 LT |
2502 | goto abort_free; |
2503 | } | |
2504 | if (err < 0) { | |
2505 | printk(KERN_WARNING | |
2506 | "md: could not read %s's sb, not importing!\n", | |
2507 | bdevname(rdev->bdev,b)); | |
2508 | goto abort_free; | |
2509 | } | |
2510 | } | |
6bfe0b49 | 2511 | |
1da177e4 | 2512 | INIT_LIST_HEAD(&rdev->same_set); |
6bfe0b49 | 2513 | init_waitqueue_head(&rdev->blocked_wait); |
1da177e4 LT |
2514 | |
2515 | return rdev; | |
2516 | ||
2517 | abort_free: | |
2518 | if (rdev->sb_page) { | |
2519 | if (rdev->bdev) | |
2520 | unlock_rdev(rdev); | |
2521 | free_disk_sb(rdev); | |
2522 | } | |
2523 | kfree(rdev); | |
2524 | return ERR_PTR(err); | |
2525 | } | |
2526 | ||
2527 | /* | |
2528 | * Check a full RAID array for plausibility | |
2529 | */ | |
2530 | ||
2531 | ||
a757e64c | 2532 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2533 | { |
2534 | int i; | |
159ec1fc | 2535 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2536 | char b[BDEVNAME_SIZE]; |
2537 | ||
2538 | freshest = NULL; | |
d089c6af | 2539 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2540 | switch (super_types[mddev->major_version]. |
2541 | load_super(rdev, freshest, mddev->minor_version)) { | |
2542 | case 1: | |
2543 | freshest = rdev; | |
2544 | break; | |
2545 | case 0: | |
2546 | break; | |
2547 | default: | |
2548 | printk( KERN_ERR \ | |
2549 | "md: fatal superblock inconsistency in %s" | |
2550 | " -- removing from array\n", | |
2551 | bdevname(rdev->bdev,b)); | |
2552 | kick_rdev_from_array(rdev); | |
2553 | } | |
2554 | ||
2555 | ||
2556 | super_types[mddev->major_version]. | |
2557 | validate_super(mddev, freshest); | |
2558 | ||
2559 | i = 0; | |
d089c6af | 2560 | rdev_for_each(rdev, tmp, mddev) { |
de01dfad N |
2561 | if (rdev->desc_nr >= mddev->max_disks || |
2562 | i > mddev->max_disks) { | |
2563 | printk(KERN_WARNING | |
2564 | "md: %s: %s: only %d devices permitted\n", | |
2565 | mdname(mddev), bdevname(rdev->bdev, b), | |
2566 | mddev->max_disks); | |
2567 | kick_rdev_from_array(rdev); | |
2568 | continue; | |
2569 | } | |
1da177e4 LT |
2570 | if (rdev != freshest) |
2571 | if (super_types[mddev->major_version]. | |
2572 | validate_super(mddev, rdev)) { | |
2573 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2574 | " from array!\n", | |
2575 | bdevname(rdev->bdev,b)); | |
2576 | kick_rdev_from_array(rdev); | |
2577 | continue; | |
2578 | } | |
2579 | if (mddev->level == LEVEL_MULTIPATH) { | |
2580 | rdev->desc_nr = i++; | |
2581 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2582 | set_bit(In_sync, &rdev->flags); |
a778b73f N |
2583 | } else if (rdev->raid_disk >= mddev->raid_disks) { |
2584 | rdev->raid_disk = -1; | |
2585 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2586 | } |
2587 | } | |
1da177e4 LT |
2588 | } |
2589 | ||
19052c0e N |
2590 | static void md_safemode_timeout(unsigned long data); |
2591 | ||
16f17b39 N |
2592 | static ssize_t |
2593 | safe_delay_show(mddev_t *mddev, char *page) | |
2594 | { | |
2595 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2596 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2597 | } | |
2598 | static ssize_t | |
2599 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2600 | { | |
2601 | int scale=1; | |
2602 | int dot=0; | |
2603 | int i; | |
2604 | unsigned long msec; | |
2605 | char buf[30]; | |
97ce0a7f | 2606 | |
16f17b39 N |
2607 | /* remove a period, and count digits after it */ |
2608 | if (len >= sizeof(buf)) | |
2609 | return -EINVAL; | |
97ce0a7f | 2610 | strlcpy(buf, cbuf, sizeof(buf)); |
16f17b39 N |
2611 | for (i=0; i<len; i++) { |
2612 | if (dot) { | |
2613 | if (isdigit(buf[i])) { | |
2614 | buf[i-1] = buf[i]; | |
2615 | scale *= 10; | |
2616 | } | |
2617 | buf[i] = 0; | |
2618 | } else if (buf[i] == '.') { | |
2619 | dot=1; | |
2620 | buf[i] = 0; | |
2621 | } | |
2622 | } | |
97ce0a7f | 2623 | if (strict_strtoul(buf, 10, &msec) < 0) |
16f17b39 N |
2624 | return -EINVAL; |
2625 | msec = (msec * 1000) / scale; | |
2626 | if (msec == 0) | |
2627 | mddev->safemode_delay = 0; | |
2628 | else { | |
19052c0e | 2629 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2630 | mddev->safemode_delay = (msec*HZ)/1000; |
2631 | if (mddev->safemode_delay == 0) | |
2632 | mddev->safemode_delay = 1; | |
19052c0e N |
2633 | if (mddev->safemode_delay < old_delay) |
2634 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2635 | } |
2636 | return len; | |
2637 | } | |
2638 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2639 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2640 | |
eae1701f | 2641 | static ssize_t |
96de1e66 | 2642 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2643 | { |
2604b703 | 2644 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2645 | if (p) |
eae1701f | 2646 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2647 | else if (mddev->clevel[0]) |
2648 | return sprintf(page, "%s\n", mddev->clevel); | |
2649 | else if (mddev->level != LEVEL_NONE) | |
2650 | return sprintf(page, "%d\n", mddev->level); | |
2651 | else | |
2652 | return 0; | |
eae1701f N |
2653 | } |
2654 | ||
d9d166c2 N |
2655 | static ssize_t |
2656 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2657 | { | |
245f46c2 | 2658 | char level[16]; |
20a49ff6 | 2659 | ssize_t rv = len; |
245f46c2 N |
2660 | struct mdk_personality *pers; |
2661 | void *priv; | |
2662 | ||
2663 | if (mddev->pers == NULL) { | |
2664 | if (len == 0) | |
2665 | return 0; | |
2666 | if (len >= sizeof(mddev->clevel)) | |
2667 | return -ENOSPC; | |
2668 | strncpy(mddev->clevel, buf, len); | |
2669 | if (mddev->clevel[len-1] == '\n') | |
2670 | len--; | |
2671 | mddev->clevel[len] = 0; | |
2672 | mddev->level = LEVEL_NONE; | |
2673 | return rv; | |
2674 | } | |
2675 | ||
2676 | /* request to change the personality. Need to ensure: | |
2677 | * - array is not engaged in resync/recovery/reshape | |
2678 | * - old personality can be suspended | |
2679 | * - new personality will access other array. | |
2680 | */ | |
2681 | ||
2682 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) | |
d9d166c2 | 2683 | return -EBUSY; |
245f46c2 N |
2684 | |
2685 | if (!mddev->pers->quiesce) { | |
2686 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
2687 | mdname(mddev), mddev->pers->name); | |
2688 | return -EINVAL; | |
2689 | } | |
2690 | ||
2691 | /* Now find the new personality */ | |
2692 | if (len == 0 || len >= sizeof(level)) | |
2693 | return -EINVAL; | |
2694 | strncpy(level, buf, len); | |
2695 | if (level[len-1] == '\n') | |
d9d166c2 | 2696 | len--; |
245f46c2 N |
2697 | level[len] = 0; |
2698 | ||
2699 | request_module("md-%s", level); | |
2700 | spin_lock(&pers_lock); | |
2701 | pers = find_pers(LEVEL_NONE, level); | |
2702 | if (!pers || !try_module_get(pers->owner)) { | |
2703 | spin_unlock(&pers_lock); | |
2704 | printk(KERN_WARNING "md: personality %s not loaded\n", level); | |
2705 | return -EINVAL; | |
2706 | } | |
2707 | spin_unlock(&pers_lock); | |
2708 | ||
2709 | if (pers == mddev->pers) { | |
2710 | /* Nothing to do! */ | |
2711 | module_put(pers->owner); | |
2712 | return rv; | |
2713 | } | |
2714 | if (!pers->takeover) { | |
2715 | module_put(pers->owner); | |
2716 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
2717 | mdname(mddev), level); | |
2718 | return -EINVAL; | |
2719 | } | |
2720 | ||
2721 | /* ->takeover must set new_* and/or delta_disks | |
2722 | * if it succeeds, and may set them when it fails. | |
2723 | */ | |
2724 | priv = pers->takeover(mddev); | |
2725 | if (IS_ERR(priv)) { | |
2726 | mddev->new_level = mddev->level; | |
2727 | mddev->new_layout = mddev->layout; | |
664e7c41 | 2728 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
2729 | mddev->raid_disks -= mddev->delta_disks; |
2730 | mddev->delta_disks = 0; | |
2731 | module_put(pers->owner); | |
2732 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
2733 | mdname(mddev), level); | |
2734 | return PTR_ERR(priv); | |
2735 | } | |
2736 | ||
2737 | /* Looks like we have a winner */ | |
2738 | mddev_suspend(mddev); | |
2739 | mddev->pers->stop(mddev); | |
2740 | module_put(mddev->pers->owner); | |
2741 | mddev->pers = pers; | |
2742 | mddev->private = priv; | |
2743 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
2744 | mddev->level = mddev->new_level; | |
2745 | mddev->layout = mddev->new_layout; | |
664e7c41 | 2746 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 N |
2747 | mddev->delta_disks = 0; |
2748 | pers->run(mddev); | |
2749 | mddev_resume(mddev); | |
2750 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2751 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2752 | md_wakeup_thread(mddev->thread); | |
d9d166c2 N |
2753 | return rv; |
2754 | } | |
2755 | ||
2756 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 2757 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 2758 | |
d4dbd025 N |
2759 | |
2760 | static ssize_t | |
2761 | layout_show(mddev_t *mddev, char *page) | |
2762 | { | |
2763 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
2764 | if (mddev->reshape_position != MaxSector && |
2765 | mddev->layout != mddev->new_layout) | |
2766 | return sprintf(page, "%d (%d)\n", | |
2767 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
2768 | return sprintf(page, "%d\n", mddev->layout); |
2769 | } | |
2770 | ||
2771 | static ssize_t | |
2772 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
2773 | { | |
2774 | char *e; | |
2775 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
2776 | |
2777 | if (!*buf || (*e && *e != '\n')) | |
2778 | return -EINVAL; | |
2779 | ||
b3546035 N |
2780 | if (mddev->pers) { |
2781 | int err; | |
50ac168a | 2782 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 2783 | return -EBUSY; |
597a711b | 2784 | mddev->new_layout = n; |
50ac168a | 2785 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
2786 | if (err) { |
2787 | mddev->new_layout = mddev->layout; | |
b3546035 | 2788 | return err; |
597a711b | 2789 | } |
b3546035 | 2790 | } else { |
08a02ecd | 2791 | mddev->new_layout = n; |
b3546035 N |
2792 | if (mddev->reshape_position == MaxSector) |
2793 | mddev->layout = n; | |
2794 | } | |
d4dbd025 N |
2795 | return len; |
2796 | } | |
2797 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 2798 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
2799 | |
2800 | ||
eae1701f | 2801 | static ssize_t |
96de1e66 | 2802 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 2803 | { |
bb636547 N |
2804 | if (mddev->raid_disks == 0) |
2805 | return 0; | |
08a02ecd N |
2806 | if (mddev->reshape_position != MaxSector && |
2807 | mddev->delta_disks != 0) | |
2808 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
2809 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
2810 | return sprintf(page, "%d\n", mddev->raid_disks); |
2811 | } | |
2812 | ||
da943b99 N |
2813 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
2814 | ||
2815 | static ssize_t | |
2816 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
2817 | { | |
da943b99 N |
2818 | char *e; |
2819 | int rv = 0; | |
2820 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2821 | ||
2822 | if (!*buf || (*e && *e != '\n')) | |
2823 | return -EINVAL; | |
2824 | ||
2825 | if (mddev->pers) | |
2826 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
2827 | else if (mddev->reshape_position != MaxSector) { |
2828 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
2829 | mddev->delta_disks = n - olddisks; | |
2830 | mddev->raid_disks = n; | |
2831 | } else | |
da943b99 N |
2832 | mddev->raid_disks = n; |
2833 | return rv ? rv : len; | |
2834 | } | |
2835 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 2836 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 2837 | |
3b34380a N |
2838 | static ssize_t |
2839 | chunk_size_show(mddev_t *mddev, char *page) | |
2840 | { | |
08a02ecd | 2841 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
2842 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
2843 | return sprintf(page, "%d (%d)\n", | |
2844 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
2845 | mddev->chunk_sectors << 9); |
2846 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
2847 | } |
2848 | ||
2849 | static ssize_t | |
2850 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
2851 | { | |
3b34380a N |
2852 | char *e; |
2853 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2854 | ||
3b34380a N |
2855 | if (!*buf || (*e && *e != '\n')) |
2856 | return -EINVAL; | |
2857 | ||
b3546035 N |
2858 | if (mddev->pers) { |
2859 | int err; | |
50ac168a | 2860 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 2861 | return -EBUSY; |
597a711b | 2862 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 2863 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
2864 | if (err) { |
2865 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 2866 | return err; |
597a711b | 2867 | } |
b3546035 | 2868 | } else { |
664e7c41 | 2869 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 2870 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 2871 | mddev->chunk_sectors = n >> 9; |
b3546035 | 2872 | } |
3b34380a N |
2873 | return len; |
2874 | } | |
2875 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 2876 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 2877 | |
a94213b1 N |
2878 | static ssize_t |
2879 | resync_start_show(mddev_t *mddev, char *page) | |
2880 | { | |
d1a7c503 N |
2881 | if (mddev->recovery_cp == MaxSector) |
2882 | return sprintf(page, "none\n"); | |
a94213b1 N |
2883 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
2884 | } | |
2885 | ||
2886 | static ssize_t | |
2887 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
2888 | { | |
a94213b1 N |
2889 | char *e; |
2890 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
2891 | ||
2892 | if (mddev->pers) | |
2893 | return -EBUSY; | |
2894 | if (!*buf || (*e && *e != '\n')) | |
2895 | return -EINVAL; | |
2896 | ||
2897 | mddev->recovery_cp = n; | |
2898 | return len; | |
2899 | } | |
2900 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 2901 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 2902 | |
9e653b63 N |
2903 | /* |
2904 | * The array state can be: | |
2905 | * | |
2906 | * clear | |
2907 | * No devices, no size, no level | |
2908 | * Equivalent to STOP_ARRAY ioctl | |
2909 | * inactive | |
2910 | * May have some settings, but array is not active | |
2911 | * all IO results in error | |
2912 | * When written, doesn't tear down array, but just stops it | |
2913 | * suspended (not supported yet) | |
2914 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 2915 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
2916 | * readonly |
2917 | * no resync can happen. no superblocks get written. | |
2918 | * write requests fail | |
2919 | * read-auto | |
2920 | * like readonly, but behaves like 'clean' on a write request. | |
2921 | * | |
2922 | * clean - no pending writes, but otherwise active. | |
2923 | * When written to inactive array, starts without resync | |
2924 | * If a write request arrives then | |
2925 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
2926 | * if not known, block and switch to write-pending | |
2927 | * If written to an active array that has pending writes, then fails. | |
2928 | * active | |
2929 | * fully active: IO and resync can be happening. | |
2930 | * When written to inactive array, starts with resync | |
2931 | * | |
2932 | * write-pending | |
2933 | * clean, but writes are blocked waiting for 'active' to be written. | |
2934 | * | |
2935 | * active-idle | |
2936 | * like active, but no writes have been seen for a while (100msec). | |
2937 | * | |
2938 | */ | |
2939 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
2940 | write_pending, active_idle, bad_word}; | |
05381954 | 2941 | static char *array_states[] = { |
9e653b63 N |
2942 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
2943 | "write-pending", "active-idle", NULL }; | |
2944 | ||
2945 | static int match_word(const char *word, char **list) | |
2946 | { | |
2947 | int n; | |
2948 | for (n=0; list[n]; n++) | |
2949 | if (cmd_match(word, list[n])) | |
2950 | break; | |
2951 | return n; | |
2952 | } | |
2953 | ||
2954 | static ssize_t | |
2955 | array_state_show(mddev_t *mddev, char *page) | |
2956 | { | |
2957 | enum array_state st = inactive; | |
2958 | ||
2959 | if (mddev->pers) | |
2960 | switch(mddev->ro) { | |
2961 | case 1: | |
2962 | st = readonly; | |
2963 | break; | |
2964 | case 2: | |
2965 | st = read_auto; | |
2966 | break; | |
2967 | case 0: | |
2968 | if (mddev->in_sync) | |
2969 | st = clean; | |
e691063a N |
2970 | else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) |
2971 | st = write_pending; | |
9e653b63 N |
2972 | else if (mddev->safemode) |
2973 | st = active_idle; | |
2974 | else | |
2975 | st = active; | |
2976 | } | |
2977 | else { | |
2978 | if (list_empty(&mddev->disks) && | |
2979 | mddev->raid_disks == 0 && | |
58c0fed4 | 2980 | mddev->dev_sectors == 0) |
9e653b63 N |
2981 | st = clear; |
2982 | else | |
2983 | st = inactive; | |
2984 | } | |
2985 | return sprintf(page, "%s\n", array_states[st]); | |
2986 | } | |
2987 | ||
df5b20cf | 2988 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
9e653b63 N |
2989 | static int do_md_run(mddev_t * mddev); |
2990 | static int restart_array(mddev_t *mddev); | |
2991 | ||
2992 | static ssize_t | |
2993 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
2994 | { | |
2995 | int err = -EINVAL; | |
2996 | enum array_state st = match_word(buf, array_states); | |
2997 | switch(st) { | |
2998 | case bad_word: | |
2999 | break; | |
3000 | case clear: | |
3001 | /* stopping an active array */ | |
f2ea68cf | 3002 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3003 | return -EBUSY; |
df5b20cf | 3004 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3005 | break; |
3006 | case inactive: | |
3007 | /* stopping an active array */ | |
3008 | if (mddev->pers) { | |
f2ea68cf | 3009 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3010 | return -EBUSY; |
df5b20cf | 3011 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3012 | } else |
3013 | err = 0; /* already inactive */ | |
9e653b63 N |
3014 | break; |
3015 | case suspended: | |
3016 | break; /* not supported yet */ | |
3017 | case readonly: | |
3018 | if (mddev->pers) | |
df5b20cf | 3019 | err = do_md_stop(mddev, 1, 0); |
9e653b63 N |
3020 | else { |
3021 | mddev->ro = 1; | |
648b629e | 3022 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3023 | err = do_md_run(mddev); |
3024 | } | |
3025 | break; | |
3026 | case read_auto: | |
9e653b63 | 3027 | if (mddev->pers) { |
80268ee9 | 3028 | if (mddev->ro == 0) |
df5b20cf | 3029 | err = do_md_stop(mddev, 1, 0); |
80268ee9 | 3030 | else if (mddev->ro == 1) |
648b629e N |
3031 | err = restart_array(mddev); |
3032 | if (err == 0) { | |
3033 | mddev->ro = 2; | |
3034 | set_disk_ro(mddev->gendisk, 0); | |
3035 | } | |
9e653b63 N |
3036 | } else { |
3037 | mddev->ro = 2; | |
3038 | err = do_md_run(mddev); | |
3039 | } | |
3040 | break; | |
3041 | case clean: | |
3042 | if (mddev->pers) { | |
3043 | restart_array(mddev); | |
3044 | spin_lock_irq(&mddev->write_lock); | |
3045 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3046 | if (mddev->in_sync == 0) { |
3047 | mddev->in_sync = 1; | |
31a59e34 N |
3048 | if (mddev->safemode == 1) |
3049 | mddev->safemode = 0; | |
e691063a N |
3050 | if (mddev->persistent) |
3051 | set_bit(MD_CHANGE_CLEAN, | |
3052 | &mddev->flags); | |
3053 | } | |
3054 | err = 0; | |
3055 | } else | |
3056 | err = -EBUSY; | |
9e653b63 | 3057 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3058 | } else |
3059 | err = -EINVAL; | |
9e653b63 N |
3060 | break; |
3061 | case active: | |
3062 | if (mddev->pers) { | |
3063 | restart_array(mddev); | |
e691063a N |
3064 | if (mddev->external) |
3065 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
9e653b63 N |
3066 | wake_up(&mddev->sb_wait); |
3067 | err = 0; | |
3068 | } else { | |
3069 | mddev->ro = 0; | |
648b629e | 3070 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3071 | err = do_md_run(mddev); |
3072 | } | |
3073 | break; | |
3074 | case write_pending: | |
3075 | case active_idle: | |
3076 | /* these cannot be set */ | |
3077 | break; | |
3078 | } | |
3079 | if (err) | |
3080 | return err; | |
0fd62b86 | 3081 | else { |
b62b7590 | 3082 | sysfs_notify_dirent(mddev->sysfs_state); |
9e653b63 | 3083 | return len; |
0fd62b86 | 3084 | } |
9e653b63 | 3085 | } |
80ca3a44 N |
3086 | static struct md_sysfs_entry md_array_state = |
3087 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3088 | |
6d7ff738 N |
3089 | static ssize_t |
3090 | null_show(mddev_t *mddev, char *page) | |
3091 | { | |
3092 | return -EINVAL; | |
3093 | } | |
3094 | ||
3095 | static ssize_t | |
3096 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3097 | { | |
3098 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3099 | /* The new device is added to the array. | |
3100 | * If the array has a persistent superblock, we read the | |
3101 | * superblock to initialise info and check validity. | |
3102 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3103 | * which mainly checks size. | |
3104 | */ | |
3105 | char *e; | |
3106 | int major = simple_strtoul(buf, &e, 10); | |
3107 | int minor; | |
3108 | dev_t dev; | |
3109 | mdk_rdev_t *rdev; | |
3110 | int err; | |
3111 | ||
3112 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3113 | return -EINVAL; | |
3114 | minor = simple_strtoul(e+1, &e, 10); | |
3115 | if (*e && *e != '\n') | |
3116 | return -EINVAL; | |
3117 | dev = MKDEV(major, minor); | |
3118 | if (major != MAJOR(dev) || | |
3119 | minor != MINOR(dev)) | |
3120 | return -EOVERFLOW; | |
3121 | ||
3122 | ||
3123 | if (mddev->persistent) { | |
3124 | rdev = md_import_device(dev, mddev->major_version, | |
3125 | mddev->minor_version); | |
3126 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3127 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3128 | mdk_rdev_t, same_set); | |
3129 | err = super_types[mddev->major_version] | |
3130 | .load_super(rdev, rdev0, mddev->minor_version); | |
3131 | if (err < 0) | |
3132 | goto out; | |
3133 | } | |
c5d79adb N |
3134 | } else if (mddev->external) |
3135 | rdev = md_import_device(dev, -2, -1); | |
3136 | else | |
6d7ff738 N |
3137 | rdev = md_import_device(dev, -1, -1); |
3138 | ||
3139 | if (IS_ERR(rdev)) | |
3140 | return PTR_ERR(rdev); | |
3141 | err = bind_rdev_to_array(rdev, mddev); | |
3142 | out: | |
3143 | if (err) | |
3144 | export_rdev(rdev); | |
3145 | return err ? err : len; | |
3146 | } | |
3147 | ||
3148 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3149 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3150 | |
9b1d1dac PC |
3151 | static ssize_t |
3152 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3153 | { | |
3154 | char *end; | |
3155 | unsigned long chunk, end_chunk; | |
3156 | ||
3157 | if (!mddev->bitmap) | |
3158 | goto out; | |
3159 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3160 | while (*buf) { | |
3161 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3162 | if (buf == end) break; | |
3163 | if (*end == '-') { /* range */ | |
3164 | buf = end + 1; | |
3165 | end_chunk = simple_strtoul(buf, &end, 0); | |
3166 | if (buf == end) break; | |
3167 | } | |
3168 | if (*end && !isspace(*end)) break; | |
3169 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
3170 | buf = end; | |
3171 | while (isspace(*buf)) buf++; | |
3172 | } | |
3173 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3174 | out: | |
3175 | return len; | |
3176 | } | |
3177 | ||
3178 | static struct md_sysfs_entry md_bitmap = | |
3179 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3180 | ||
a35b0d69 N |
3181 | static ssize_t |
3182 | size_show(mddev_t *mddev, char *page) | |
3183 | { | |
58c0fed4 AN |
3184 | return sprintf(page, "%llu\n", |
3185 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3186 | } |
3187 | ||
d71f9f88 | 3188 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3189 | |
3190 | static ssize_t | |
3191 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3192 | { | |
3193 | /* If array is inactive, we can reduce the component size, but | |
3194 | * not increase it (except from 0). | |
3195 | * If array is active, we can try an on-line resize | |
3196 | */ | |
b522adcd DW |
3197 | sector_t sectors; |
3198 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3199 | |
58c0fed4 AN |
3200 | if (err < 0) |
3201 | return err; | |
a35b0d69 | 3202 | if (mddev->pers) { |
58c0fed4 | 3203 | err = update_size(mddev, sectors); |
850b2b42 | 3204 | md_update_sb(mddev, 1); |
a35b0d69 | 3205 | } else { |
58c0fed4 AN |
3206 | if (mddev->dev_sectors == 0 || |
3207 | mddev->dev_sectors > sectors) | |
3208 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3209 | else |
3210 | err = -ENOSPC; | |
3211 | } | |
3212 | return err ? err : len; | |
3213 | } | |
3214 | ||
3215 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3216 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3217 | |
8bb93aac N |
3218 | |
3219 | /* Metdata version. | |
e691063a N |
3220 | * This is one of |
3221 | * 'none' for arrays with no metadata (good luck...) | |
3222 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3223 | * or N.M for internally known formats |
3224 | */ | |
3225 | static ssize_t | |
3226 | metadata_show(mddev_t *mddev, char *page) | |
3227 | { | |
3228 | if (mddev->persistent) | |
3229 | return sprintf(page, "%d.%d\n", | |
3230 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3231 | else if (mddev->external) |
3232 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3233 | else |
3234 | return sprintf(page, "none\n"); | |
3235 | } | |
3236 | ||
3237 | static ssize_t | |
3238 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3239 | { | |
3240 | int major, minor; | |
3241 | char *e; | |
ea43ddd8 N |
3242 | /* Changing the details of 'external' metadata is |
3243 | * always permitted. Otherwise there must be | |
3244 | * no devices attached to the array. | |
3245 | */ | |
3246 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3247 | ; | |
3248 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3249 | return -EBUSY; |
3250 | ||
3251 | if (cmd_match(buf, "none")) { | |
3252 | mddev->persistent = 0; | |
e691063a N |
3253 | mddev->external = 0; |
3254 | mddev->major_version = 0; | |
3255 | mddev->minor_version = 90; | |
3256 | return len; | |
3257 | } | |
3258 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3259 | size_t namelen = len-9; |
e691063a N |
3260 | if (namelen >= sizeof(mddev->metadata_type)) |
3261 | namelen = sizeof(mddev->metadata_type)-1; | |
3262 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3263 | mddev->metadata_type[namelen] = 0; | |
3264 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3265 | mddev->metadata_type[--namelen] = 0; | |
3266 | mddev->persistent = 0; | |
3267 | mddev->external = 1; | |
8bb93aac N |
3268 | mddev->major_version = 0; |
3269 | mddev->minor_version = 90; | |
3270 | return len; | |
3271 | } | |
3272 | major = simple_strtoul(buf, &e, 10); | |
3273 | if (e==buf || *e != '.') | |
3274 | return -EINVAL; | |
3275 | buf = e+1; | |
3276 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3277 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3278 | return -EINVAL; |
50511da3 | 3279 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3280 | return -ENOENT; |
3281 | mddev->major_version = major; | |
3282 | mddev->minor_version = minor; | |
3283 | mddev->persistent = 1; | |
e691063a | 3284 | mddev->external = 0; |
8bb93aac N |
3285 | return len; |
3286 | } | |
3287 | ||
3288 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3289 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3290 | |
24dd469d | 3291 | static ssize_t |
7eec314d | 3292 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3293 | { |
7eec314d | 3294 | char *type = "idle"; |
b6a9ce68 N |
3295 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3296 | type = "frozen"; | |
3297 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3298 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3299 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3300 | type = "reshape"; | |
3301 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3302 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3303 | type = "resync"; | |
3304 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3305 | type = "check"; | |
3306 | else | |
3307 | type = "repair"; | |
72a23c21 | 3308 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3309 | type = "recover"; |
3310 | } | |
3311 | return sprintf(page, "%s\n", type); | |
3312 | } | |
3313 | ||
3314 | static ssize_t | |
7eec314d | 3315 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3316 | { |
7eec314d N |
3317 | if (!mddev->pers || !mddev->pers->sync_request) |
3318 | return -EINVAL; | |
3319 | ||
b6a9ce68 N |
3320 | if (cmd_match(page, "frozen")) |
3321 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3322 | else | |
3323 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3324 | ||
3325 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3326 | if (mddev->sync_thread) { |
3327 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
3328 | md_unregister_thread(mddev->sync_thread); | |
3329 | mddev->sync_thread = NULL; | |
3330 | mddev->recovery = 0; | |
3331 | } | |
03c902e1 N |
3332 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3333 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3334 | return -EBUSY; |
72a23c21 NB |
3335 | else if (cmd_match(page, "resync")) |
3336 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3337 | else if (cmd_match(page, "recover")) { | |
3338 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3339 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3340 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3341 | int err; |
3342 | if (mddev->pers->start_reshape == NULL) | |
3343 | return -EINVAL; | |
3344 | err = mddev->pers->start_reshape(mddev); | |
3345 | if (err) | |
3346 | return err; | |
a99ac971 | 3347 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3348 | } else { |
bce74dac | 3349 | if (cmd_match(page, "check")) |
7eec314d | 3350 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3351 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3352 | return -EINVAL; |
3353 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3354 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3355 | } |
03c902e1 | 3356 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3357 | md_wakeup_thread(mddev->thread); |
0c3573f1 | 3358 | sysfs_notify_dirent(mddev->sysfs_action); |
24dd469d N |
3359 | return len; |
3360 | } | |
3361 | ||
9d88883e | 3362 | static ssize_t |
96de1e66 | 3363 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3364 | { |
3365 | return sprintf(page, "%llu\n", | |
3366 | (unsigned long long) mddev->resync_mismatches); | |
3367 | } | |
3368 | ||
80ca3a44 N |
3369 | static struct md_sysfs_entry md_scan_mode = |
3370 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3371 | |
96de1e66 | 3372 | |
80ca3a44 | 3373 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3374 | |
88202a0c N |
3375 | static ssize_t |
3376 | sync_min_show(mddev_t *mddev, char *page) | |
3377 | { | |
3378 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3379 | mddev->sync_speed_min ? "local": "system"); | |
3380 | } | |
3381 | ||
3382 | static ssize_t | |
3383 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3384 | { | |
3385 | int min; | |
3386 | char *e; | |
3387 | if (strncmp(buf, "system", 6)==0) { | |
3388 | mddev->sync_speed_min = 0; | |
3389 | return len; | |
3390 | } | |
3391 | min = simple_strtoul(buf, &e, 10); | |
3392 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3393 | return -EINVAL; | |
3394 | mddev->sync_speed_min = min; | |
3395 | return len; | |
3396 | } | |
3397 | ||
3398 | static struct md_sysfs_entry md_sync_min = | |
3399 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3400 | ||
3401 | static ssize_t | |
3402 | sync_max_show(mddev_t *mddev, char *page) | |
3403 | { | |
3404 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3405 | mddev->sync_speed_max ? "local": "system"); | |
3406 | } | |
3407 | ||
3408 | static ssize_t | |
3409 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3410 | { | |
3411 | int max; | |
3412 | char *e; | |
3413 | if (strncmp(buf, "system", 6)==0) { | |
3414 | mddev->sync_speed_max = 0; | |
3415 | return len; | |
3416 | } | |
3417 | max = simple_strtoul(buf, &e, 10); | |
3418 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3419 | return -EINVAL; | |
3420 | mddev->sync_speed_max = max; | |
3421 | return len; | |
3422 | } | |
3423 | ||
3424 | static struct md_sysfs_entry md_sync_max = | |
3425 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3426 | ||
d7f3d291 IP |
3427 | static ssize_t |
3428 | degraded_show(mddev_t *mddev, char *page) | |
3429 | { | |
3430 | return sprintf(page, "%d\n", mddev->degraded); | |
3431 | } | |
3432 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3433 | |
90b08710 BS |
3434 | static ssize_t |
3435 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3436 | { | |
3437 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3438 | } | |
3439 | ||
3440 | static ssize_t | |
3441 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3442 | { | |
3443 | long n; | |
3444 | ||
3445 | if (strict_strtol(buf, 10, &n)) | |
3446 | return -EINVAL; | |
3447 | ||
3448 | if (n != 0 && n != 1) | |
3449 | return -EINVAL; | |
3450 | ||
3451 | mddev->parallel_resync = n; | |
3452 | ||
3453 | if (mddev->sync_thread) | |
3454 | wake_up(&resync_wait); | |
3455 | ||
3456 | return len; | |
3457 | } | |
3458 | ||
3459 | /* force parallel resync, even with shared block devices */ | |
3460 | static struct md_sysfs_entry md_sync_force_parallel = | |
3461 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3462 | sync_force_parallel_show, sync_force_parallel_store); | |
3463 | ||
88202a0c N |
3464 | static ssize_t |
3465 | sync_speed_show(mddev_t *mddev, char *page) | |
3466 | { | |
3467 | unsigned long resync, dt, db; | |
d1a7c503 N |
3468 | if (mddev->curr_resync == 0) |
3469 | return sprintf(page, "none\n"); | |
9687a60c AN |
3470 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3471 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3472 | if (!dt) dt++; |
9687a60c AN |
3473 | db = resync - mddev->resync_mark_cnt; |
3474 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3475 | } |
3476 | ||
80ca3a44 | 3477 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3478 | |
3479 | static ssize_t | |
3480 | sync_completed_show(mddev_t *mddev, char *page) | |
3481 | { | |
58c0fed4 | 3482 | unsigned long max_sectors, resync; |
88202a0c | 3483 | |
acb180b0 N |
3484 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3485 | return sprintf(page, "none\n"); | |
3486 | ||
88202a0c | 3487 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3488 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3489 | else |
58c0fed4 | 3490 | max_sectors = mddev->dev_sectors; |
88202a0c | 3491 | |
acb180b0 | 3492 | resync = mddev->curr_resync_completed; |
58c0fed4 | 3493 | return sprintf(page, "%lu / %lu\n", resync, max_sectors); |
88202a0c N |
3494 | } |
3495 | ||
80ca3a44 | 3496 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3497 | |
5e96ee65 NB |
3498 | static ssize_t |
3499 | min_sync_show(mddev_t *mddev, char *page) | |
3500 | { | |
3501 | return sprintf(page, "%llu\n", | |
3502 | (unsigned long long)mddev->resync_min); | |
3503 | } | |
3504 | static ssize_t | |
3505 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3506 | { | |
3507 | unsigned long long min; | |
3508 | if (strict_strtoull(buf, 10, &min)) | |
3509 | return -EINVAL; | |
3510 | if (min > mddev->resync_max) | |
3511 | return -EINVAL; | |
3512 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3513 | return -EBUSY; | |
3514 | ||
3515 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3516 | if (mddev->chunk_sectors) { |
2ac06c33 | 3517 | sector_t temp = min; |
9d8f0363 | 3518 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3519 | return -EINVAL; |
3520 | } | |
3521 | mddev->resync_min = min; | |
3522 | ||
3523 | return len; | |
3524 | } | |
3525 | ||
3526 | static struct md_sysfs_entry md_min_sync = | |
3527 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3528 | ||
c6207277 N |
3529 | static ssize_t |
3530 | max_sync_show(mddev_t *mddev, char *page) | |
3531 | { | |
3532 | if (mddev->resync_max == MaxSector) | |
3533 | return sprintf(page, "max\n"); | |
3534 | else | |
3535 | return sprintf(page, "%llu\n", | |
3536 | (unsigned long long)mddev->resync_max); | |
3537 | } | |
3538 | static ssize_t | |
3539 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3540 | { | |
3541 | if (strncmp(buf, "max", 3) == 0) | |
3542 | mddev->resync_max = MaxSector; | |
3543 | else { | |
5e96ee65 NB |
3544 | unsigned long long max; |
3545 | if (strict_strtoull(buf, 10, &max)) | |
3546 | return -EINVAL; | |
3547 | if (max < mddev->resync_min) | |
c6207277 N |
3548 | return -EINVAL; |
3549 | if (max < mddev->resync_max && | |
3550 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3551 | return -EBUSY; | |
3552 | ||
3553 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3554 | if (mddev->chunk_sectors) { |
2ac06c33 | 3555 | sector_t temp = max; |
9d8f0363 | 3556 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
3557 | return -EINVAL; |
3558 | } | |
3559 | mddev->resync_max = max; | |
3560 | } | |
3561 | wake_up(&mddev->recovery_wait); | |
3562 | return len; | |
3563 | } | |
3564 | ||
3565 | static struct md_sysfs_entry md_max_sync = | |
3566 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
3567 | ||
e464eafd N |
3568 | static ssize_t |
3569 | suspend_lo_show(mddev_t *mddev, char *page) | |
3570 | { | |
3571 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
3572 | } | |
3573 | ||
3574 | static ssize_t | |
3575 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
3576 | { | |
3577 | char *e; | |
3578 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3579 | ||
3580 | if (mddev->pers->quiesce == NULL) | |
3581 | return -EINVAL; | |
3582 | if (buf == e || (*e && *e != '\n')) | |
3583 | return -EINVAL; | |
3584 | if (new >= mddev->suspend_hi || | |
3585 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
3586 | mddev->suspend_lo = new; | |
3587 | mddev->pers->quiesce(mddev, 2); | |
3588 | return len; | |
3589 | } else | |
3590 | return -EINVAL; | |
3591 | } | |
3592 | static struct md_sysfs_entry md_suspend_lo = | |
3593 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
3594 | ||
3595 | ||
3596 | static ssize_t | |
3597 | suspend_hi_show(mddev_t *mddev, char *page) | |
3598 | { | |
3599 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
3600 | } | |
3601 | ||
3602 | static ssize_t | |
3603 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
3604 | { | |
3605 | char *e; | |
3606 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3607 | ||
3608 | if (mddev->pers->quiesce == NULL) | |
3609 | return -EINVAL; | |
3610 | if (buf == e || (*e && *e != '\n')) | |
3611 | return -EINVAL; | |
3612 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
3613 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
3614 | mddev->suspend_hi = new; | |
3615 | mddev->pers->quiesce(mddev, 1); | |
3616 | mddev->pers->quiesce(mddev, 0); | |
3617 | return len; | |
3618 | } else | |
3619 | return -EINVAL; | |
3620 | } | |
3621 | static struct md_sysfs_entry md_suspend_hi = | |
3622 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
3623 | ||
08a02ecd N |
3624 | static ssize_t |
3625 | reshape_position_show(mddev_t *mddev, char *page) | |
3626 | { | |
3627 | if (mddev->reshape_position != MaxSector) | |
3628 | return sprintf(page, "%llu\n", | |
3629 | (unsigned long long)mddev->reshape_position); | |
3630 | strcpy(page, "none\n"); | |
3631 | return 5; | |
3632 | } | |
3633 | ||
3634 | static ssize_t | |
3635 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
3636 | { | |
3637 | char *e; | |
3638 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3639 | if (mddev->pers) | |
3640 | return -EBUSY; | |
3641 | if (buf == e || (*e && *e != '\n')) | |
3642 | return -EINVAL; | |
3643 | mddev->reshape_position = new; | |
3644 | mddev->delta_disks = 0; | |
3645 | mddev->new_level = mddev->level; | |
3646 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3647 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
3648 | return len; |
3649 | } | |
3650 | ||
3651 | static struct md_sysfs_entry md_reshape_position = | |
3652 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
3653 | reshape_position_store); | |
3654 | ||
b522adcd DW |
3655 | static ssize_t |
3656 | array_size_show(mddev_t *mddev, char *page) | |
3657 | { | |
3658 | if (mddev->external_size) | |
3659 | return sprintf(page, "%llu\n", | |
3660 | (unsigned long long)mddev->array_sectors/2); | |
3661 | else | |
3662 | return sprintf(page, "default\n"); | |
3663 | } | |
3664 | ||
3665 | static ssize_t | |
3666 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3667 | { | |
3668 | sector_t sectors; | |
3669 | ||
3670 | if (strncmp(buf, "default", 7) == 0) { | |
3671 | if (mddev->pers) | |
3672 | sectors = mddev->pers->size(mddev, 0, 0); | |
3673 | else | |
3674 | sectors = mddev->array_sectors; | |
3675 | ||
3676 | mddev->external_size = 0; | |
3677 | } else { | |
3678 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
3679 | return -EINVAL; | |
3680 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 3681 | return -E2BIG; |
b522adcd DW |
3682 | |
3683 | mddev->external_size = 1; | |
3684 | } | |
3685 | ||
3686 | mddev->array_sectors = sectors; | |
3687 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
3688 | if (mddev->pers) { | |
3689 | struct block_device *bdev = bdget_disk(mddev->gendisk, 0); | |
3690 | ||
3691 | if (bdev) { | |
3692 | mutex_lock(&bdev->bd_inode->i_mutex); | |
3693 | i_size_write(bdev->bd_inode, | |
3694 | (loff_t)mddev->array_sectors << 9); | |
3695 | mutex_unlock(&bdev->bd_inode->i_mutex); | |
3696 | bdput(bdev); | |
3697 | } | |
3698 | } | |
3699 | ||
3700 | return len; | |
3701 | } | |
3702 | ||
3703 | static struct md_sysfs_entry md_array_size = | |
3704 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
3705 | array_size_store); | |
e464eafd | 3706 | |
eae1701f N |
3707 | static struct attribute *md_default_attrs[] = { |
3708 | &md_level.attr, | |
d4dbd025 | 3709 | &md_layout.attr, |
eae1701f | 3710 | &md_raid_disks.attr, |
3b34380a | 3711 | &md_chunk_size.attr, |
a35b0d69 | 3712 | &md_size.attr, |
a94213b1 | 3713 | &md_resync_start.attr, |
8bb93aac | 3714 | &md_metadata.attr, |
6d7ff738 | 3715 | &md_new_device.attr, |
16f17b39 | 3716 | &md_safe_delay.attr, |
9e653b63 | 3717 | &md_array_state.attr, |
08a02ecd | 3718 | &md_reshape_position.attr, |
b522adcd | 3719 | &md_array_size.attr, |
411036fa N |
3720 | NULL, |
3721 | }; | |
3722 | ||
3723 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 3724 | &md_scan_mode.attr, |
9d88883e | 3725 | &md_mismatches.attr, |
88202a0c N |
3726 | &md_sync_min.attr, |
3727 | &md_sync_max.attr, | |
3728 | &md_sync_speed.attr, | |
90b08710 | 3729 | &md_sync_force_parallel.attr, |
88202a0c | 3730 | &md_sync_completed.attr, |
5e96ee65 | 3731 | &md_min_sync.attr, |
c6207277 | 3732 | &md_max_sync.attr, |
e464eafd N |
3733 | &md_suspend_lo.attr, |
3734 | &md_suspend_hi.attr, | |
9b1d1dac | 3735 | &md_bitmap.attr, |
d7f3d291 | 3736 | &md_degraded.attr, |
eae1701f N |
3737 | NULL, |
3738 | }; | |
411036fa N |
3739 | static struct attribute_group md_redundancy_group = { |
3740 | .name = NULL, | |
3741 | .attrs = md_redundancy_attrs, | |
3742 | }; | |
3743 | ||
eae1701f N |
3744 | |
3745 | static ssize_t | |
3746 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3747 | { | |
3748 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3749 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3750 | ssize_t rv; |
eae1701f N |
3751 | |
3752 | if (!entry->show) | |
3753 | return -EIO; | |
5dc5cf7d IM |
3754 | rv = mddev_lock(mddev); |
3755 | if (!rv) { | |
3756 | rv = entry->show(mddev, page); | |
3757 | mddev_unlock(mddev); | |
3758 | } | |
96de1e66 | 3759 | return rv; |
eae1701f N |
3760 | } |
3761 | ||
3762 | static ssize_t | |
3763 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
3764 | const char *page, size_t length) | |
3765 | { | |
3766 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3767 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3768 | ssize_t rv; |
eae1701f N |
3769 | |
3770 | if (!entry->store) | |
3771 | return -EIO; | |
67463acb N |
3772 | if (!capable(CAP_SYS_ADMIN)) |
3773 | return -EACCES; | |
5dc5cf7d | 3774 | rv = mddev_lock(mddev); |
d3374825 N |
3775 | if (mddev->hold_active == UNTIL_IOCTL) |
3776 | mddev->hold_active = 0; | |
5dc5cf7d IM |
3777 | if (!rv) { |
3778 | rv = entry->store(mddev, page, length); | |
3779 | mddev_unlock(mddev); | |
3780 | } | |
96de1e66 | 3781 | return rv; |
eae1701f N |
3782 | } |
3783 | ||
3784 | static void md_free(struct kobject *ko) | |
3785 | { | |
3786 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
3787 | |
3788 | if (mddev->sysfs_state) | |
3789 | sysfs_put(mddev->sysfs_state); | |
3790 | ||
3791 | if (mddev->gendisk) { | |
3792 | del_gendisk(mddev->gendisk); | |
3793 | put_disk(mddev->gendisk); | |
3794 | } | |
3795 | if (mddev->queue) | |
3796 | blk_cleanup_queue(mddev->queue); | |
3797 | ||
eae1701f N |
3798 | kfree(mddev); |
3799 | } | |
3800 | ||
3801 | static struct sysfs_ops md_sysfs_ops = { | |
3802 | .show = md_attr_show, | |
3803 | .store = md_attr_store, | |
3804 | }; | |
3805 | static struct kobj_type md_ktype = { | |
3806 | .release = md_free, | |
3807 | .sysfs_ops = &md_sysfs_ops, | |
3808 | .default_attrs = md_default_attrs, | |
3809 | }; | |
3810 | ||
1da177e4 LT |
3811 | int mdp_major = 0; |
3812 | ||
5fd3a17e DW |
3813 | static void mddev_delayed_delete(struct work_struct *ws) |
3814 | { | |
3815 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
3816 | ||
3817 | if (mddev->private == &md_redundancy_group) { | |
3818 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
3819 | if (mddev->sysfs_action) | |
3820 | sysfs_put(mddev->sysfs_action); | |
3821 | mddev->sysfs_action = NULL; | |
3822 | mddev->private = NULL; | |
3823 | } | |
3824 | kobject_del(&mddev->kobj); | |
3825 | kobject_put(&mddev->kobj); | |
3826 | } | |
3827 | ||
efeb53c0 | 3828 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 3829 | { |
48c9c27b | 3830 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
3831 | mddev_t *mddev = mddev_find(dev); |
3832 | struct gendisk *disk; | |
efeb53c0 N |
3833 | int partitioned; |
3834 | int shift; | |
3835 | int unit; | |
3830c62f | 3836 | int error; |
1da177e4 LT |
3837 | |
3838 | if (!mddev) | |
efeb53c0 N |
3839 | return -ENODEV; |
3840 | ||
3841 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
3842 | shift = partitioned ? MdpMinorShift : 0; | |
3843 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 3844 | |
d3374825 N |
3845 | /* wait for any previous instance if this device |
3846 | * to be completed removed (mddev_delayed_delete). | |
3847 | */ | |
3848 | flush_scheduled_work(); | |
3849 | ||
48c9c27b | 3850 | mutex_lock(&disks_mutex); |
1da177e4 | 3851 | if (mddev->gendisk) { |
48c9c27b | 3852 | mutex_unlock(&disks_mutex); |
1da177e4 | 3853 | mddev_put(mddev); |
efeb53c0 N |
3854 | return -EEXIST; |
3855 | } | |
3856 | ||
3857 | if (name) { | |
3858 | /* Need to ensure that 'name' is not a duplicate. | |
3859 | */ | |
3860 | mddev_t *mddev2; | |
3861 | spin_lock(&all_mddevs_lock); | |
3862 | ||
3863 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
3864 | if (mddev2->gendisk && | |
3865 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
3866 | spin_unlock(&all_mddevs_lock); | |
3867 | return -EEXIST; | |
3868 | } | |
3869 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 3870 | } |
8b765398 N |
3871 | |
3872 | mddev->queue = blk_alloc_queue(GFP_KERNEL); | |
3873 | if (!mddev->queue) { | |
3874 | mutex_unlock(&disks_mutex); | |
3875 | mddev_put(mddev); | |
efeb53c0 | 3876 | return -ENOMEM; |
8b765398 | 3877 | } |
409c57f3 N |
3878 | mddev->queue->queuedata = mddev; |
3879 | ||
8b765398 N |
3880 | /* Can be unlocked because the queue is new: no concurrency */ |
3881 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); | |
3882 | ||
409c57f3 | 3883 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 3884 | |
1da177e4 LT |
3885 | disk = alloc_disk(1 << shift); |
3886 | if (!disk) { | |
48c9c27b | 3887 | mutex_unlock(&disks_mutex); |
8b765398 N |
3888 | blk_cleanup_queue(mddev->queue); |
3889 | mddev->queue = NULL; | |
1da177e4 | 3890 | mddev_put(mddev); |
efeb53c0 | 3891 | return -ENOMEM; |
1da177e4 | 3892 | } |
efeb53c0 | 3893 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 3894 | disk->first_minor = unit << shift; |
efeb53c0 N |
3895 | if (name) |
3896 | strcpy(disk->disk_name, name); | |
3897 | else if (partitioned) | |
1da177e4 | 3898 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 3899 | else |
1da177e4 | 3900 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
3901 | disk->fops = &md_fops; |
3902 | disk->private_data = mddev; | |
3903 | disk->queue = mddev->queue; | |
92850bbd | 3904 | /* Allow extended partitions. This makes the |
d3374825 | 3905 | * 'mdp' device redundant, but we can't really |
92850bbd N |
3906 | * remove it now. |
3907 | */ | |
3908 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 LT |
3909 | add_disk(disk); |
3910 | mddev->gendisk = disk; | |
ed9e1982 TH |
3911 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
3912 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
f48ed538 | 3913 | mutex_unlock(&disks_mutex); |
3830c62f | 3914 | if (error) |
5e55e2f5 N |
3915 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
3916 | disk->disk_name); | |
b62b7590 | 3917 | else { |
3830c62f | 3918 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
b62b7590 N |
3919 | mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state"); |
3920 | } | |
d3374825 | 3921 | mddev_put(mddev); |
efeb53c0 N |
3922 | return 0; |
3923 | } | |
3924 | ||
3925 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
3926 | { | |
3927 | md_alloc(dev, NULL); | |
1da177e4 LT |
3928 | return NULL; |
3929 | } | |
3930 | ||
efeb53c0 N |
3931 | static int add_named_array(const char *val, struct kernel_param *kp) |
3932 | { | |
3933 | /* val must be "md_*" where * is not all digits. | |
3934 | * We allocate an array with a large free minor number, and | |
3935 | * set the name to val. val must not already be an active name. | |
3936 | */ | |
3937 | int len = strlen(val); | |
3938 | char buf[DISK_NAME_LEN]; | |
3939 | ||
3940 | while (len && val[len-1] == '\n') | |
3941 | len--; | |
3942 | if (len >= DISK_NAME_LEN) | |
3943 | return -E2BIG; | |
3944 | strlcpy(buf, val, len+1); | |
3945 | if (strncmp(buf, "md_", 3) != 0) | |
3946 | return -EINVAL; | |
3947 | return md_alloc(0, buf); | |
3948 | } | |
3949 | ||
1da177e4 LT |
3950 | static void md_safemode_timeout(unsigned long data) |
3951 | { | |
3952 | mddev_t *mddev = (mddev_t *) data; | |
3953 | ||
0fd62b86 NB |
3954 | if (!atomic_read(&mddev->writes_pending)) { |
3955 | mddev->safemode = 1; | |
3956 | if (mddev->external) | |
b62b7590 | 3957 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 | 3958 | } |
1da177e4 LT |
3959 | md_wakeup_thread(mddev->thread); |
3960 | } | |
3961 | ||
6ff8d8ec | 3962 | static int start_dirty_degraded; |
1da177e4 LT |
3963 | |
3964 | static int do_md_run(mddev_t * mddev) | |
3965 | { | |
2604b703 | 3966 | int err; |
1da177e4 LT |
3967 | mdk_rdev_t *rdev; |
3968 | struct gendisk *disk; | |
2604b703 | 3969 | struct mdk_personality *pers; |
1da177e4 | 3970 | |
a757e64c N |
3971 | if (list_empty(&mddev->disks)) |
3972 | /* cannot run an array with no devices.. */ | |
1da177e4 | 3973 | return -EINVAL; |
1da177e4 LT |
3974 | |
3975 | if (mddev->pers) | |
3976 | return -EBUSY; | |
3977 | ||
3978 | /* | |
3979 | * Analyze all RAID superblock(s) | |
3980 | */ | |
1ec4a939 N |
3981 | if (!mddev->raid_disks) { |
3982 | if (!mddev->persistent) | |
3983 | return -EINVAL; | |
a757e64c | 3984 | analyze_sbs(mddev); |
1ec4a939 | 3985 | } |
1da177e4 | 3986 | |
d9d166c2 N |
3987 | if (mddev->level != LEVEL_NONE) |
3988 | request_module("md-level-%d", mddev->level); | |
3989 | else if (mddev->clevel[0]) | |
3990 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
3991 | |
3992 | /* | |
3993 | * Drop all container device buffers, from now on | |
3994 | * the only valid external interface is through the md | |
3995 | * device. | |
1da177e4 | 3996 | */ |
159ec1fc | 3997 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 3998 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
3999 | continue; |
4000 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4001 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4002 | |
4003 | /* perform some consistency tests on the device. | |
4004 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4005 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4006 | */ |
0f420358 | 4007 | if (rdev->data_offset < rdev->sb_start) { |
58c0fed4 AN |
4008 | if (mddev->dev_sectors && |
4009 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4010 | > rdev->sb_start) { |
f0d76d70 N |
4011 | printk("md: %s: data overlaps metadata\n", |
4012 | mdname(mddev)); | |
4013 | return -EINVAL; | |
4014 | } | |
4015 | } else { | |
0f420358 | 4016 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4017 | > rdev->data_offset) { |
4018 | printk("md: %s: metadata overlaps data\n", | |
4019 | mdname(mddev)); | |
4020 | return -EINVAL; | |
4021 | } | |
4022 | } | |
3c0ee63a | 4023 | sysfs_notify_dirent(rdev->sysfs_state); |
1da177e4 LT |
4024 | } |
4025 | ||
4026 | md_probe(mddev->unit, NULL, NULL); | |
4027 | disk = mddev->gendisk; | |
4028 | if (!disk) | |
4029 | return -ENOMEM; | |
4030 | ||
4031 | spin_lock(&pers_lock); | |
d9d166c2 | 4032 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4033 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4034 | spin_unlock(&pers_lock); |
d9d166c2 N |
4035 | if (mddev->level != LEVEL_NONE) |
4036 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4037 | mddev->level); | |
4038 | else | |
4039 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4040 | mddev->clevel); | |
1da177e4 LT |
4041 | return -EINVAL; |
4042 | } | |
2604b703 | 4043 | mddev->pers = pers; |
1da177e4 | 4044 | spin_unlock(&pers_lock); |
34817e8c N |
4045 | if (mddev->level != pers->level) { |
4046 | mddev->level = pers->level; | |
4047 | mddev->new_level = pers->level; | |
4048 | } | |
d9d166c2 | 4049 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4050 | |
3f9d99c1 MP |
4051 | if (pers->level >= 4 && pers->level <= 6) |
4052 | /* Cannot support integrity (yet) */ | |
4053 | blk_integrity_unregister(mddev->gendisk); | |
4054 | ||
f6705578 | 4055 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4056 | pers->start_reshape == NULL) { |
f6705578 N |
4057 | /* This personality cannot handle reshaping... */ |
4058 | mddev->pers = NULL; | |
4059 | module_put(pers->owner); | |
4060 | return -EINVAL; | |
4061 | } | |
4062 | ||
7dd5e7c3 N |
4063 | if (pers->sync_request) { |
4064 | /* Warn if this is a potentially silly | |
4065 | * configuration. | |
4066 | */ | |
4067 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4068 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4069 | int warned = 0; |
159ec1fc CR |
4070 | |
4071 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4072 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4073 | if (rdev < rdev2 && |
4074 | rdev->bdev->bd_contains == | |
4075 | rdev2->bdev->bd_contains) { | |
4076 | printk(KERN_WARNING | |
4077 | "%s: WARNING: %s appears to be" | |
4078 | " on the same physical disk as" | |
4079 | " %s.\n", | |
4080 | mdname(mddev), | |
4081 | bdevname(rdev->bdev,b), | |
4082 | bdevname(rdev2->bdev,b2)); | |
4083 | warned = 1; | |
4084 | } | |
4085 | } | |
159ec1fc | 4086 | |
7dd5e7c3 N |
4087 | if (warned) |
4088 | printk(KERN_WARNING | |
4089 | "True protection against single-disk" | |
4090 | " failure might be compromised.\n"); | |
4091 | } | |
4092 | ||
657390d2 | 4093 | mddev->recovery = 0; |
58c0fed4 AN |
4094 | /* may be over-ridden by personality */ |
4095 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4096 | ||
a9701a30 | 4097 | mddev->barriers_work = 1; |
6ff8d8ec | 4098 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4099 | |
f91de92e N |
4100 | if (start_readonly) |
4101 | mddev->ro = 2; /* read-only, but switch on first write */ | |
4102 | ||
b15c2e57 | 4103 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4104 | if (err) |
4105 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4106 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4107 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4108 | " but 'external_size' not in effect?\n", __func__); | |
4109 | printk(KERN_ERR | |
4110 | "md: invalid array_size %llu > default size %llu\n", | |
4111 | (unsigned long long)mddev->array_sectors / 2, | |
4112 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4113 | err = -EINVAL; | |
4114 | mddev->pers->stop(mddev); | |
4115 | } | |
4116 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4117 | err = bitmap_create(mddev); |
4118 | if (err) { | |
4119 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4120 | mdname(mddev), err); | |
4121 | mddev->pers->stop(mddev); | |
4122 | } | |
4123 | } | |
1da177e4 | 4124 | if (err) { |
1da177e4 LT |
4125 | module_put(mddev->pers->owner); |
4126 | mddev->pers = NULL; | |
32a7627c N |
4127 | bitmap_destroy(mddev); |
4128 | return err; | |
1da177e4 | 4129 | } |
5e55e2f5 N |
4130 | if (mddev->pers->sync_request) { |
4131 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
4132 | printk(KERN_WARNING | |
4133 | "md: cannot register extra attributes for %s\n", | |
4134 | mdname(mddev)); | |
0c3573f1 | 4135 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4136 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4137 | mddev->ro = 0; |
4138 | ||
1da177e4 LT |
4139 | atomic_set(&mddev->writes_pending,0); |
4140 | mddev->safemode = 0; | |
4141 | mddev->safemode_timer.function = md_safemode_timeout; | |
4142 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4143 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4144 | mddev->in_sync = 1; |
86e6ffdd | 4145 | |
159ec1fc | 4146 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4147 | if (rdev->raid_disk >= 0) { |
4148 | char nm[20]; | |
4149 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
4150 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
4151 | printk("md: cannot register %s for %s\n", | |
4152 | nm, mdname(mddev)); | |
86e6ffdd | 4153 | } |
1da177e4 LT |
4154 | |
4155 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4156 | ||
850b2b42 N |
4157 | if (mddev->flags) |
4158 | md_update_sb(mddev, 0); | |
1da177e4 | 4159 | |
f233ea5c | 4160 | set_capacity(disk, mddev->array_sectors); |
1da177e4 | 4161 | |
5fd6c1dc N |
4162 | /* If there is a partially-recovered drive we need to |
4163 | * start recovery here. If we leave it to md_check_recovery, | |
4164 | * it will remove the drives and not do the right thing | |
4165 | */ | |
0b8c9de0 | 4166 | if (mddev->degraded && !mddev->sync_thread) { |
5fd6c1dc | 4167 | int spares = 0; |
159ec1fc | 4168 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
4169 | if (rdev->raid_disk >= 0 && |
4170 | !test_bit(In_sync, &rdev->flags) && | |
4171 | !test_bit(Faulty, &rdev->flags)) | |
4172 | /* complete an interrupted recovery */ | |
4173 | spares++; | |
4174 | if (spares && mddev->pers->sync_request) { | |
4175 | mddev->recovery = 0; | |
4176 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4177 | mddev->sync_thread = md_register_thread(md_do_sync, | |
4178 | mddev, | |
4179 | "%s_resync"); | |
4180 | if (!mddev->sync_thread) { | |
4181 | printk(KERN_ERR "%s: could not start resync" | |
4182 | " thread...\n", | |
4183 | mdname(mddev)); | |
4184 | /* leave the spares where they are, it shouldn't hurt */ | |
4185 | mddev->recovery = 0; | |
0b8c9de0 | 4186 | } |
5fd6c1dc N |
4187 | } |
4188 | } | |
0b8c9de0 N |
4189 | md_wakeup_thread(mddev->thread); |
4190 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4191 | |
44ce6294 | 4192 | mddev->changed = 1; |
d7603b7e | 4193 | md_new_event(mddev); |
b62b7590 | 4194 | sysfs_notify_dirent(mddev->sysfs_state); |
0c3573f1 N |
4195 | if (mddev->sysfs_action) |
4196 | sysfs_notify_dirent(mddev->sysfs_action); | |
a99ac971 | 4197 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
ed9e1982 | 4198 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
1da177e4 LT |
4199 | return 0; |
4200 | } | |
4201 | ||
4202 | static int restart_array(mddev_t *mddev) | |
4203 | { | |
4204 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4205 | |
80fab1d7 | 4206 | /* Complain if it has no devices */ |
1da177e4 | 4207 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4208 | return -ENXIO; |
4209 | if (!mddev->pers) | |
4210 | return -EINVAL; | |
4211 | if (!mddev->ro) | |
4212 | return -EBUSY; | |
4213 | mddev->safemode = 0; | |
4214 | mddev->ro = 0; | |
4215 | set_disk_ro(disk, 0); | |
4216 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4217 | mdname(mddev)); | |
4218 | /* Kick recovery or resync if necessary */ | |
4219 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4220 | md_wakeup_thread(mddev->thread); | |
4221 | md_wakeup_thread(mddev->sync_thread); | |
b62b7590 | 4222 | sysfs_notify_dirent(mddev->sysfs_state); |
80fab1d7 | 4223 | return 0; |
1da177e4 LT |
4224 | } |
4225 | ||
acc55e22 N |
4226 | /* similar to deny_write_access, but accounts for our holding a reference |
4227 | * to the file ourselves */ | |
4228 | static int deny_bitmap_write_access(struct file * file) | |
4229 | { | |
4230 | struct inode *inode = file->f_mapping->host; | |
4231 | ||
4232 | spin_lock(&inode->i_lock); | |
4233 | if (atomic_read(&inode->i_writecount) > 1) { | |
4234 | spin_unlock(&inode->i_lock); | |
4235 | return -ETXTBSY; | |
4236 | } | |
4237 | atomic_set(&inode->i_writecount, -1); | |
4238 | spin_unlock(&inode->i_lock); | |
4239 | ||
4240 | return 0; | |
4241 | } | |
4242 | ||
4243 | static void restore_bitmap_write_access(struct file *file) | |
4244 | { | |
4245 | struct inode *inode = file->f_mapping->host; | |
4246 | ||
4247 | spin_lock(&inode->i_lock); | |
4248 | atomic_set(&inode->i_writecount, 1); | |
4249 | spin_unlock(&inode->i_lock); | |
4250 | } | |
4251 | ||
9e653b63 N |
4252 | /* mode: |
4253 | * 0 - completely stop and dis-assemble array | |
4254 | * 1 - switch to readonly | |
4255 | * 2 - stop but do not disassemble array | |
4256 | */ | |
df5b20cf | 4257 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 LT |
4258 | { |
4259 | int err = 0; | |
4260 | struct gendisk *disk = mddev->gendisk; | |
c4647292 | 4261 | mdk_rdev_t *rdev; |
1da177e4 | 4262 | |
f2ea68cf | 4263 | if (atomic_read(&mddev->openers) > is_open) { |
df5b20cf NB |
4264 | printk("md: %s still in use.\n",mdname(mddev)); |
4265 | return -EBUSY; | |
4266 | } | |
4267 | ||
1da177e4 | 4268 | if (mddev->pers) { |
1da177e4 LT |
4269 | |
4270 | if (mddev->sync_thread) { | |
5fd6c1dc | 4271 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 LT |
4272 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4273 | md_unregister_thread(mddev->sync_thread); | |
4274 | mddev->sync_thread = NULL; | |
4275 | } | |
4276 | ||
4277 | del_timer_sync(&mddev->safemode_timer); | |
4278 | ||
9e653b63 N |
4279 | switch(mode) { |
4280 | case 1: /* readonly */ | |
1da177e4 | 4281 | err = -ENXIO; |
f91de92e | 4282 | if (mddev->ro==1) |
1da177e4 LT |
4283 | goto out; |
4284 | mddev->ro = 1; | |
9e653b63 N |
4285 | break; |
4286 | case 0: /* disassemble */ | |
4287 | case 2: /* stop */ | |
6b8b3e8a | 4288 | bitmap_flush(mddev); |
a9701a30 | 4289 | md_super_wait(mddev); |
1da177e4 LT |
4290 | if (mddev->ro) |
4291 | set_disk_ro(disk, 0); | |
409c57f3 | 4292 | |
1da177e4 | 4293 | mddev->pers->stop(mddev); |
d1b5380c N |
4294 | mddev->queue->merge_bvec_fn = NULL; |
4295 | mddev->queue->unplug_fn = NULL; | |
041ae52e | 4296 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4297 | module_put(mddev->pers->owner); |
5fd3a17e DW |
4298 | if (mddev->pers->sync_request) |
4299 | mddev->private = &md_redundancy_group; | |
1da177e4 | 4300 | mddev->pers = NULL; |
4f54b0e9 | 4301 | /* tell userspace to handle 'inactive' */ |
b62b7590 | 4302 | sysfs_notify_dirent(mddev->sysfs_state); |
0d4ca600 | 4303 | |
c4647292 N |
4304 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4305 | if (rdev->raid_disk >= 0) { | |
4306 | char nm[20]; | |
4307 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4308 | sysfs_remove_link(&mddev->kobj, nm); | |
4309 | } | |
4310 | ||
0d4ca600 | 4311 | set_capacity(disk, 0); |
44ce6294 | 4312 | mddev->changed = 1; |
0d4ca600 | 4313 | |
1da177e4 LT |
4314 | if (mddev->ro) |
4315 | mddev->ro = 0; | |
4316 | } | |
850b2b42 | 4317 | if (!mddev->in_sync || mddev->flags) { |
1da177e4 LT |
4318 | /* mark array as shutdown cleanly */ |
4319 | mddev->in_sync = 1; | |
850b2b42 | 4320 | md_update_sb(mddev, 1); |
1da177e4 | 4321 | } |
9e653b63 | 4322 | if (mode == 1) |
1da177e4 | 4323 | set_disk_ro(disk, 1); |
5fd6c1dc | 4324 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 | 4325 | } |
32a7627c | 4326 | |
1da177e4 LT |
4327 | /* |
4328 | * Free resources if final stop | |
4329 | */ | |
9e653b63 | 4330 | if (mode == 0) { |
0d4ca600 | 4331 | |
1da177e4 LT |
4332 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4333 | ||
978f946b N |
4334 | bitmap_destroy(mddev); |
4335 | if (mddev->bitmap_file) { | |
acc55e22 | 4336 | restore_bitmap_write_access(mddev->bitmap_file); |
978f946b N |
4337 | fput(mddev->bitmap_file); |
4338 | mddev->bitmap_file = NULL; | |
4339 | } | |
4340 | mddev->bitmap_offset = 0; | |
4341 | ||
177a99b2 | 4342 | /* make sure all md_delayed_delete calls have finished */ |
5792a285 N |
4343 | flush_scheduled_work(); |
4344 | ||
1da177e4 LT |
4345 | export_array(mddev); |
4346 | ||
f233ea5c | 4347 | mddev->array_sectors = 0; |
b522adcd | 4348 | mddev->external_size = 0; |
58c0fed4 | 4349 | mddev->dev_sectors = 0; |
9e653b63 | 4350 | mddev->raid_disks = 0; |
a94213b1 | 4351 | mddev->recovery_cp = 0; |
5e96ee65 | 4352 | mddev->resync_min = 0; |
c6207277 | 4353 | mddev->resync_max = MaxSector; |
08a02ecd | 4354 | mddev->reshape_position = MaxSector; |
e691063a | 4355 | mddev->external = 0; |
1ec4a939 | 4356 | mddev->persistent = 0; |
d897dbf9 N |
4357 | mddev->level = LEVEL_NONE; |
4358 | mddev->clevel[0] = 0; | |
4359 | mddev->flags = 0; | |
4360 | mddev->ro = 0; | |
4361 | mddev->metadata_type[0] = 0; | |
9d8f0363 | 4362 | mddev->chunk_sectors = 0; |
d897dbf9 N |
4363 | mddev->ctime = mddev->utime = 0; |
4364 | mddev->layout = 0; | |
4365 | mddev->max_disks = 0; | |
4366 | mddev->events = 0; | |
4367 | mddev->delta_disks = 0; | |
4368 | mddev->new_level = LEVEL_NONE; | |
4369 | mddev->new_layout = 0; | |
664e7c41 | 4370 | mddev->new_chunk_sectors = 0; |
d897dbf9 N |
4371 | mddev->curr_resync = 0; |
4372 | mddev->resync_mismatches = 0; | |
4373 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4374 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4375 | mddev->recovery = 0; | |
4376 | mddev->in_sync = 0; | |
4377 | mddev->changed = 0; | |
4378 | mddev->degraded = 0; | |
4379 | mddev->barriers_work = 0; | |
4380 | mddev->safemode = 0; | |
934d9c23 | 4381 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4382 | if (mddev->hold_active == UNTIL_STOP) |
4383 | mddev->hold_active = 0; | |
9e653b63 | 4384 | |
a8a55c38 | 4385 | } else if (mddev->pers) |
1da177e4 LT |
4386 | printk(KERN_INFO "md: %s switched to read-only mode.\n", |
4387 | mdname(mddev)); | |
4388 | err = 0; | |
3f9d99c1 | 4389 | blk_integrity_unregister(disk); |
d7603b7e | 4390 | md_new_event(mddev); |
b62b7590 | 4391 | sysfs_notify_dirent(mddev->sysfs_state); |
1da177e4 LT |
4392 | out: |
4393 | return err; | |
4394 | } | |
4395 | ||
fdee8ae4 | 4396 | #ifndef MODULE |
1da177e4 LT |
4397 | static void autorun_array(mddev_t *mddev) |
4398 | { | |
4399 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4400 | int err; |
4401 | ||
a757e64c | 4402 | if (list_empty(&mddev->disks)) |
1da177e4 | 4403 | return; |
1da177e4 LT |
4404 | |
4405 | printk(KERN_INFO "md: running: "); | |
4406 | ||
159ec1fc | 4407 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4408 | char b[BDEVNAME_SIZE]; |
4409 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4410 | } | |
4411 | printk("\n"); | |
4412 | ||
d710e138 | 4413 | err = do_md_run(mddev); |
1da177e4 LT |
4414 | if (err) { |
4415 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4416 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4417 | } |
4418 | } | |
4419 | ||
4420 | /* | |
4421 | * lets try to run arrays based on all disks that have arrived | |
4422 | * until now. (those are in pending_raid_disks) | |
4423 | * | |
4424 | * the method: pick the first pending disk, collect all disks with | |
4425 | * the same UUID, remove all from the pending list and put them into | |
4426 | * the 'same_array' list. Then order this list based on superblock | |
4427 | * update time (freshest comes first), kick out 'old' disks and | |
4428 | * compare superblocks. If everything's fine then run it. | |
4429 | * | |
4430 | * If "unit" is allocated, then bump its reference count | |
4431 | */ | |
4432 | static void autorun_devices(int part) | |
4433 | { | |
159ec1fc | 4434 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4435 | mddev_t *mddev; |
4436 | char b[BDEVNAME_SIZE]; | |
4437 | ||
4438 | printk(KERN_INFO "md: autorun ...\n"); | |
4439 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4440 | int unit; |
1da177e4 | 4441 | dev_t dev; |
ad01c9e3 | 4442 | LIST_HEAD(candidates); |
1da177e4 LT |
4443 | rdev0 = list_entry(pending_raid_disks.next, |
4444 | mdk_rdev_t, same_set); | |
4445 | ||
4446 | printk(KERN_INFO "md: considering %s ...\n", | |
4447 | bdevname(rdev0->bdev,b)); | |
4448 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4449 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4450 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4451 | printk(KERN_INFO "md: adding %s ...\n", | |
4452 | bdevname(rdev->bdev,b)); | |
4453 | list_move(&rdev->same_set, &candidates); | |
4454 | } | |
4455 | /* | |
4456 | * now we have a set of devices, with all of them having | |
4457 | * mostly sane superblocks. It's time to allocate the | |
4458 | * mddev. | |
4459 | */ | |
e8703fe1 N |
4460 | if (part) { |
4461 | dev = MKDEV(mdp_major, | |
4462 | rdev0->preferred_minor << MdpMinorShift); | |
4463 | unit = MINOR(dev) >> MdpMinorShift; | |
4464 | } else { | |
4465 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4466 | unit = MINOR(dev); | |
4467 | } | |
4468 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4469 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4470 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4471 | break; | |
4472 | } | |
1da177e4 LT |
4473 | |
4474 | md_probe(dev, NULL, NULL); | |
4475 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4476 | if (!mddev || !mddev->gendisk) { |
4477 | if (mddev) | |
4478 | mddev_put(mddev); | |
4479 | printk(KERN_ERR | |
1da177e4 LT |
4480 | "md: cannot allocate memory for md drive.\n"); |
4481 | break; | |
4482 | } | |
4483 | if (mddev_lock(mddev)) | |
4484 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4485 | mdname(mddev)); | |
4486 | else if (mddev->raid_disks || mddev->major_version | |
4487 | || !list_empty(&mddev->disks)) { | |
4488 | printk(KERN_WARNING | |
4489 | "md: %s already running, cannot run %s\n", | |
4490 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4491 | mddev_unlock(mddev); | |
4492 | } else { | |
4493 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4494 | mddev->persistent = 1; |
159ec1fc | 4495 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4496 | list_del_init(&rdev->same_set); |
4497 | if (bind_rdev_to_array(rdev, mddev)) | |
4498 | export_rdev(rdev); | |
4499 | } | |
4500 | autorun_array(mddev); | |
4501 | mddev_unlock(mddev); | |
4502 | } | |
4503 | /* on success, candidates will be empty, on error | |
4504 | * it won't... | |
4505 | */ | |
159ec1fc | 4506 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4507 | list_del_init(&rdev->same_set); |
1da177e4 | 4508 | export_rdev(rdev); |
4b80991c | 4509 | } |
1da177e4 LT |
4510 | mddev_put(mddev); |
4511 | } | |
4512 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4513 | } | |
fdee8ae4 | 4514 | #endif /* !MODULE */ |
1da177e4 | 4515 | |
1da177e4 LT |
4516 | static int get_version(void __user * arg) |
4517 | { | |
4518 | mdu_version_t ver; | |
4519 | ||
4520 | ver.major = MD_MAJOR_VERSION; | |
4521 | ver.minor = MD_MINOR_VERSION; | |
4522 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4523 | ||
4524 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4525 | return -EFAULT; | |
4526 | ||
4527 | return 0; | |
4528 | } | |
4529 | ||
4530 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4531 | { | |
4532 | mdu_array_info_t info; | |
4533 | int nr,working,active,failed,spare; | |
4534 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4535 | |
4536 | nr=working=active=failed=spare=0; | |
159ec1fc | 4537 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4538 | nr++; |
b2d444d7 | 4539 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4540 | failed++; |
4541 | else { | |
4542 | working++; | |
b2d444d7 | 4543 | if (test_bit(In_sync, &rdev->flags)) |
1da177e4 LT |
4544 | active++; |
4545 | else | |
4546 | spare++; | |
4547 | } | |
4548 | } | |
4549 | ||
4550 | info.major_version = mddev->major_version; | |
4551 | info.minor_version = mddev->minor_version; | |
4552 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
4553 | info.ctime = mddev->ctime; | |
4554 | info.level = mddev->level; | |
58c0fed4 AN |
4555 | info.size = mddev->dev_sectors / 2; |
4556 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 4557 | info.size = -1; |
1da177e4 LT |
4558 | info.nr_disks = nr; |
4559 | info.raid_disks = mddev->raid_disks; | |
4560 | info.md_minor = mddev->md_minor; | |
4561 | info.not_persistent= !mddev->persistent; | |
4562 | ||
4563 | info.utime = mddev->utime; | |
4564 | info.state = 0; | |
4565 | if (mddev->in_sync) | |
4566 | info.state = (1<<MD_SB_CLEAN); | |
36fa3063 N |
4567 | if (mddev->bitmap && mddev->bitmap_offset) |
4568 | info.state = (1<<MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
4569 | info.active_disks = active; |
4570 | info.working_disks = working; | |
4571 | info.failed_disks = failed; | |
4572 | info.spare_disks = spare; | |
4573 | ||
4574 | info.layout = mddev->layout; | |
9d8f0363 | 4575 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
4576 | |
4577 | if (copy_to_user(arg, &info, sizeof(info))) | |
4578 | return -EFAULT; | |
4579 | ||
4580 | return 0; | |
4581 | } | |
4582 | ||
87162a28 | 4583 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
4584 | { |
4585 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
4586 | char *ptr, *buf = NULL; | |
4587 | int err = -ENOMEM; | |
4588 | ||
b5470dc5 DW |
4589 | if (md_allow_write(mddev)) |
4590 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
4591 | else | |
4592 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 4593 | |
32a7627c N |
4594 | if (!file) |
4595 | goto out; | |
4596 | ||
4597 | /* bitmap disabled, zero the first byte and copy out */ | |
4598 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
4599 | file->pathname[0] = '\0'; | |
4600 | goto copy_out; | |
4601 | } | |
4602 | ||
4603 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
4604 | if (!buf) | |
4605 | goto out; | |
4606 | ||
6bcfd601 CH |
4607 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
4608 | if (IS_ERR(ptr)) | |
32a7627c N |
4609 | goto out; |
4610 | ||
4611 | strcpy(file->pathname, ptr); | |
4612 | ||
4613 | copy_out: | |
4614 | err = 0; | |
4615 | if (copy_to_user(arg, file, sizeof(*file))) | |
4616 | err = -EFAULT; | |
4617 | out: | |
4618 | kfree(buf); | |
4619 | kfree(file); | |
4620 | return err; | |
4621 | } | |
4622 | ||
1da177e4 LT |
4623 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
4624 | { | |
4625 | mdu_disk_info_t info; | |
1da177e4 LT |
4626 | mdk_rdev_t *rdev; |
4627 | ||
4628 | if (copy_from_user(&info, arg, sizeof(info))) | |
4629 | return -EFAULT; | |
4630 | ||
26ef379f | 4631 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
4632 | if (rdev) { |
4633 | info.major = MAJOR(rdev->bdev->bd_dev); | |
4634 | info.minor = MINOR(rdev->bdev->bd_dev); | |
4635 | info.raid_disk = rdev->raid_disk; | |
4636 | info.state = 0; | |
b2d444d7 | 4637 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4638 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 4639 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4640 | info.state |= (1<<MD_DISK_ACTIVE); |
4641 | info.state |= (1<<MD_DISK_SYNC); | |
4642 | } | |
8ddf9efe N |
4643 | if (test_bit(WriteMostly, &rdev->flags)) |
4644 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
4645 | } else { |
4646 | info.major = info.minor = 0; | |
4647 | info.raid_disk = -1; | |
4648 | info.state = (1<<MD_DISK_REMOVED); | |
4649 | } | |
4650 | ||
4651 | if (copy_to_user(arg, &info, sizeof(info))) | |
4652 | return -EFAULT; | |
4653 | ||
4654 | return 0; | |
4655 | } | |
4656 | ||
4657 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
4658 | { | |
4659 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4660 | mdk_rdev_t *rdev; | |
4661 | dev_t dev = MKDEV(info->major,info->minor); | |
4662 | ||
4663 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
4664 | return -EOVERFLOW; | |
4665 | ||
4666 | if (!mddev->raid_disks) { | |
4667 | int err; | |
4668 | /* expecting a device which has a superblock */ | |
4669 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
4670 | if (IS_ERR(rdev)) { | |
4671 | printk(KERN_WARNING | |
4672 | "md: md_import_device returned %ld\n", | |
4673 | PTR_ERR(rdev)); | |
4674 | return PTR_ERR(rdev); | |
4675 | } | |
4676 | if (!list_empty(&mddev->disks)) { | |
4677 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
4678 | mdk_rdev_t, same_set); | |
4679 | int err = super_types[mddev->major_version] | |
4680 | .load_super(rdev, rdev0, mddev->minor_version); | |
4681 | if (err < 0) { | |
4682 | printk(KERN_WARNING | |
4683 | "md: %s has different UUID to %s\n", | |
4684 | bdevname(rdev->bdev,b), | |
4685 | bdevname(rdev0->bdev,b2)); | |
4686 | export_rdev(rdev); | |
4687 | return -EINVAL; | |
4688 | } | |
4689 | } | |
4690 | err = bind_rdev_to_array(rdev, mddev); | |
4691 | if (err) | |
4692 | export_rdev(rdev); | |
4693 | return err; | |
4694 | } | |
4695 | ||
4696 | /* | |
4697 | * add_new_disk can be used once the array is assembled | |
4698 | * to add "hot spares". They must already have a superblock | |
4699 | * written | |
4700 | */ | |
4701 | if (mddev->pers) { | |
4702 | int err; | |
4703 | if (!mddev->pers->hot_add_disk) { | |
4704 | printk(KERN_WARNING | |
4705 | "%s: personality does not support diskops!\n", | |
4706 | mdname(mddev)); | |
4707 | return -EINVAL; | |
4708 | } | |
7b1e35f6 N |
4709 | if (mddev->persistent) |
4710 | rdev = md_import_device(dev, mddev->major_version, | |
4711 | mddev->minor_version); | |
4712 | else | |
4713 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
4714 | if (IS_ERR(rdev)) { |
4715 | printk(KERN_WARNING | |
4716 | "md: md_import_device returned %ld\n", | |
4717 | PTR_ERR(rdev)); | |
4718 | return PTR_ERR(rdev); | |
4719 | } | |
41158c7e N |
4720 | /* set save_raid_disk if appropriate */ |
4721 | if (!mddev->persistent) { | |
4722 | if (info->state & (1<<MD_DISK_SYNC) && | |
4723 | info->raid_disk < mddev->raid_disks) | |
4724 | rdev->raid_disk = info->raid_disk; | |
4725 | else | |
4726 | rdev->raid_disk = -1; | |
4727 | } else | |
4728 | super_types[mddev->major_version]. | |
4729 | validate_super(mddev, rdev); | |
4730 | rdev->saved_raid_disk = rdev->raid_disk; | |
4731 | ||
b2d444d7 | 4732 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
4733 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4734 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
4735 | else |
4736 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 4737 | |
1da177e4 LT |
4738 | rdev->raid_disk = -1; |
4739 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
4740 | if (!err && !mddev->pers->hot_remove_disk) { |
4741 | /* If there is hot_add_disk but no hot_remove_disk | |
4742 | * then added disks for geometry changes, | |
4743 | * and should be added immediately. | |
4744 | */ | |
4745 | super_types[mddev->major_version]. | |
4746 | validate_super(mddev, rdev); | |
4747 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
4748 | if (err) | |
4749 | unbind_rdev_from_array(rdev); | |
4750 | } | |
1da177e4 LT |
4751 | if (err) |
4752 | export_rdev(rdev); | |
52664732 | 4753 | else |
3c0ee63a | 4754 | sysfs_notify_dirent(rdev->sysfs_state); |
c361777f | 4755 | |
17571284 | 4756 | md_update_sb(mddev, 1); |
72a23c21 NB |
4757 | if (mddev->degraded) |
4758 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 4759 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 4760 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4761 | return err; |
4762 | } | |
4763 | ||
4764 | /* otherwise, add_new_disk is only allowed | |
4765 | * for major_version==0 superblocks | |
4766 | */ | |
4767 | if (mddev->major_version != 0) { | |
4768 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
4769 | mdname(mddev)); | |
4770 | return -EINVAL; | |
4771 | } | |
4772 | ||
4773 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
4774 | int err; | |
d710e138 | 4775 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
4776 | if (IS_ERR(rdev)) { |
4777 | printk(KERN_WARNING | |
4778 | "md: error, md_import_device() returned %ld\n", | |
4779 | PTR_ERR(rdev)); | |
4780 | return PTR_ERR(rdev); | |
4781 | } | |
4782 | rdev->desc_nr = info->number; | |
4783 | if (info->raid_disk < mddev->raid_disks) | |
4784 | rdev->raid_disk = info->raid_disk; | |
4785 | else | |
4786 | rdev->raid_disk = -1; | |
4787 | ||
1da177e4 | 4788 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
4789 | if (info->state & (1<<MD_DISK_SYNC)) |
4790 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 4791 | |
8ddf9efe N |
4792 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4793 | set_bit(WriteMostly, &rdev->flags); | |
4794 | ||
1da177e4 LT |
4795 | if (!mddev->persistent) { |
4796 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
0f420358 | 4797 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 4798 | } else |
0f420358 | 4799 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
8190e754 | 4800 | rdev->sectors = rdev->sb_start; |
1da177e4 | 4801 | |
2bf071bf N |
4802 | err = bind_rdev_to_array(rdev, mddev); |
4803 | if (err) { | |
4804 | export_rdev(rdev); | |
4805 | return err; | |
4806 | } | |
1da177e4 LT |
4807 | } |
4808 | ||
4809 | return 0; | |
4810 | } | |
4811 | ||
4812 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
4813 | { | |
4814 | char b[BDEVNAME_SIZE]; | |
4815 | mdk_rdev_t *rdev; | |
4816 | ||
1da177e4 LT |
4817 | rdev = find_rdev(mddev, dev); |
4818 | if (!rdev) | |
4819 | return -ENXIO; | |
4820 | ||
4821 | if (rdev->raid_disk >= 0) | |
4822 | goto busy; | |
4823 | ||
4824 | kick_rdev_from_array(rdev); | |
850b2b42 | 4825 | md_update_sb(mddev, 1); |
d7603b7e | 4826 | md_new_event(mddev); |
1da177e4 LT |
4827 | |
4828 | return 0; | |
4829 | busy: | |
fdefa4d8 | 4830 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
4831 | bdevname(rdev->bdev,b), mdname(mddev)); |
4832 | return -EBUSY; | |
4833 | } | |
4834 | ||
4835 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
4836 | { | |
4837 | char b[BDEVNAME_SIZE]; | |
4838 | int err; | |
1da177e4 LT |
4839 | mdk_rdev_t *rdev; |
4840 | ||
4841 | if (!mddev->pers) | |
4842 | return -ENODEV; | |
4843 | ||
4844 | if (mddev->major_version != 0) { | |
4845 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
4846 | " version-0 superblocks.\n", | |
4847 | mdname(mddev)); | |
4848 | return -EINVAL; | |
4849 | } | |
4850 | if (!mddev->pers->hot_add_disk) { | |
4851 | printk(KERN_WARNING | |
4852 | "%s: personality does not support diskops!\n", | |
4853 | mdname(mddev)); | |
4854 | return -EINVAL; | |
4855 | } | |
4856 | ||
d710e138 | 4857 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
4858 | if (IS_ERR(rdev)) { |
4859 | printk(KERN_WARNING | |
4860 | "md: error, md_import_device() returned %ld\n", | |
4861 | PTR_ERR(rdev)); | |
4862 | return -EINVAL; | |
4863 | } | |
4864 | ||
4865 | if (mddev->persistent) | |
0f420358 | 4866 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 4867 | else |
0f420358 | 4868 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 4869 | |
8190e754 | 4870 | rdev->sectors = rdev->sb_start; |
1da177e4 | 4871 | |
b2d444d7 | 4872 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
4873 | printk(KERN_WARNING |
4874 | "md: can not hot-add faulty %s disk to %s!\n", | |
4875 | bdevname(rdev->bdev,b), mdname(mddev)); | |
4876 | err = -EINVAL; | |
4877 | goto abort_export; | |
4878 | } | |
b2d444d7 | 4879 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4880 | rdev->desc_nr = -1; |
5842730d | 4881 | rdev->saved_raid_disk = -1; |
2bf071bf N |
4882 | err = bind_rdev_to_array(rdev, mddev); |
4883 | if (err) | |
4884 | goto abort_export; | |
1da177e4 LT |
4885 | |
4886 | /* | |
4887 | * The rest should better be atomic, we can have disk failures | |
4888 | * noticed in interrupt contexts ... | |
4889 | */ | |
4890 | ||
1da177e4 LT |
4891 | rdev->raid_disk = -1; |
4892 | ||
850b2b42 | 4893 | md_update_sb(mddev, 1); |
1da177e4 LT |
4894 | |
4895 | /* | |
4896 | * Kick recovery, maybe this spare has to be added to the | |
4897 | * array immediately. | |
4898 | */ | |
4899 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4900 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 4901 | md_new_event(mddev); |
1da177e4 LT |
4902 | return 0; |
4903 | ||
1da177e4 LT |
4904 | abort_export: |
4905 | export_rdev(rdev); | |
4906 | return err; | |
4907 | } | |
4908 | ||
32a7627c N |
4909 | static int set_bitmap_file(mddev_t *mddev, int fd) |
4910 | { | |
4911 | int err; | |
4912 | ||
36fa3063 N |
4913 | if (mddev->pers) { |
4914 | if (!mddev->pers->quiesce) | |
4915 | return -EBUSY; | |
4916 | if (mddev->recovery || mddev->sync_thread) | |
4917 | return -EBUSY; | |
4918 | /* we should be able to change the bitmap.. */ | |
4919 | } | |
32a7627c | 4920 | |
32a7627c | 4921 | |
36fa3063 N |
4922 | if (fd >= 0) { |
4923 | if (mddev->bitmap) | |
4924 | return -EEXIST; /* cannot add when bitmap is present */ | |
4925 | mddev->bitmap_file = fget(fd); | |
32a7627c | 4926 | |
36fa3063 N |
4927 | if (mddev->bitmap_file == NULL) { |
4928 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", | |
4929 | mdname(mddev)); | |
4930 | return -EBADF; | |
4931 | } | |
4932 | ||
4933 | err = deny_bitmap_write_access(mddev->bitmap_file); | |
4934 | if (err) { | |
4935 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
4936 | mdname(mddev)); | |
4937 | fput(mddev->bitmap_file); | |
4938 | mddev->bitmap_file = NULL; | |
4939 | return err; | |
4940 | } | |
a654b9d8 | 4941 | mddev->bitmap_offset = 0; /* file overrides offset */ |
36fa3063 N |
4942 | } else if (mddev->bitmap == NULL) |
4943 | return -ENOENT; /* cannot remove what isn't there */ | |
4944 | err = 0; | |
4945 | if (mddev->pers) { | |
4946 | mddev->pers->quiesce(mddev, 1); | |
4947 | if (fd >= 0) | |
4948 | err = bitmap_create(mddev); | |
d7375ab3 | 4949 | if (fd < 0 || err) { |
36fa3063 | 4950 | bitmap_destroy(mddev); |
d7375ab3 N |
4951 | fd = -1; /* make sure to put the file */ |
4952 | } | |
36fa3063 | 4953 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
4954 | } |
4955 | if (fd < 0) { | |
acc55e22 N |
4956 | if (mddev->bitmap_file) { |
4957 | restore_bitmap_write_access(mddev->bitmap_file); | |
36fa3063 | 4958 | fput(mddev->bitmap_file); |
acc55e22 | 4959 | } |
36fa3063 N |
4960 | mddev->bitmap_file = NULL; |
4961 | } | |
4962 | ||
32a7627c N |
4963 | return err; |
4964 | } | |
4965 | ||
1da177e4 LT |
4966 | /* |
4967 | * set_array_info is used two different ways | |
4968 | * The original usage is when creating a new array. | |
4969 | * In this usage, raid_disks is > 0 and it together with | |
4970 | * level, size, not_persistent,layout,chunksize determine the | |
4971 | * shape of the array. | |
4972 | * This will always create an array with a type-0.90.0 superblock. | |
4973 | * The newer usage is when assembling an array. | |
4974 | * In this case raid_disks will be 0, and the major_version field is | |
4975 | * use to determine which style super-blocks are to be found on the devices. | |
4976 | * The minor and patch _version numbers are also kept incase the | |
4977 | * super_block handler wishes to interpret them. | |
4978 | */ | |
4979 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
4980 | { | |
4981 | ||
4982 | if (info->raid_disks == 0) { | |
4983 | /* just setting version number for superblock loading */ | |
4984 | if (info->major_version < 0 || | |
50511da3 | 4985 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
4986 | super_types[info->major_version].name == NULL) { |
4987 | /* maybe try to auto-load a module? */ | |
4988 | printk(KERN_INFO | |
4989 | "md: superblock version %d not known\n", | |
4990 | info->major_version); | |
4991 | return -EINVAL; | |
4992 | } | |
4993 | mddev->major_version = info->major_version; | |
4994 | mddev->minor_version = info->minor_version; | |
4995 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 4996 | mddev->persistent = !info->not_persistent; |
1da177e4 LT |
4997 | return 0; |
4998 | } | |
4999 | mddev->major_version = MD_MAJOR_VERSION; | |
5000 | mddev->minor_version = MD_MINOR_VERSION; | |
5001 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5002 | mddev->ctime = get_seconds(); | |
5003 | ||
5004 | mddev->level = info->level; | |
17115e03 | 5005 | mddev->clevel[0] = 0; |
58c0fed4 | 5006 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5007 | mddev->raid_disks = info->raid_disks; |
5008 | /* don't set md_minor, it is determined by which /dev/md* was | |
5009 | * openned | |
5010 | */ | |
5011 | if (info->state & (1<<MD_SB_CLEAN)) | |
5012 | mddev->recovery_cp = MaxSector; | |
5013 | else | |
5014 | mddev->recovery_cp = 0; | |
5015 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5016 | mddev->external = 0; |
1da177e4 LT |
5017 | |
5018 | mddev->layout = info->layout; | |
9d8f0363 | 5019 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5020 | |
5021 | mddev->max_disks = MD_SB_DISKS; | |
5022 | ||
e691063a N |
5023 | if (mddev->persistent) |
5024 | mddev->flags = 0; | |
850b2b42 | 5025 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5026 | |
b2a2703c N |
5027 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
5028 | mddev->bitmap_offset = 0; | |
5029 | ||
f6705578 N |
5030 | mddev->reshape_position = MaxSector; |
5031 | ||
1da177e4 LT |
5032 | /* |
5033 | * Generate a 128 bit UUID | |
5034 | */ | |
5035 | get_random_bytes(mddev->uuid, 16); | |
5036 | ||
f6705578 | 5037 | mddev->new_level = mddev->level; |
664e7c41 | 5038 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5039 | mddev->new_layout = mddev->layout; |
5040 | mddev->delta_disks = 0; | |
5041 | ||
1da177e4 LT |
5042 | return 0; |
5043 | } | |
5044 | ||
1f403624 DW |
5045 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5046 | { | |
b522adcd DW |
5047 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5048 | ||
5049 | if (mddev->external_size) | |
5050 | return; | |
5051 | ||
1f403624 DW |
5052 | mddev->array_sectors = array_sectors; |
5053 | } | |
5054 | EXPORT_SYMBOL(md_set_array_sectors); | |
5055 | ||
d71f9f88 | 5056 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5057 | { |
159ec1fc | 5058 | mdk_rdev_t *rdev; |
a35b0d69 | 5059 | int rv; |
d71f9f88 | 5060 | int fit = (num_sectors == 0); |
a35b0d69 N |
5061 | |
5062 | if (mddev->pers->resize == NULL) | |
5063 | return -EINVAL; | |
d71f9f88 AN |
5064 | /* The "num_sectors" is the number of sectors of each device that |
5065 | * is used. This can only make sense for arrays with redundancy. | |
5066 | * linear and raid0 always use whatever space is available. We can only | |
5067 | * consider changing this number if no resync or reconstruction is | |
5068 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5069 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5070 | * of each device. If num_sectors is zero, we find the largest size |
5071 | * that fits. | |
5072 | ||
a35b0d69 N |
5073 | */ |
5074 | if (mddev->sync_thread) | |
5075 | return -EBUSY; | |
dba034ee N |
5076 | if (mddev->bitmap) |
5077 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5078 | * grow, and re-add. | |
5079 | */ | |
5080 | return -EBUSY; | |
159ec1fc | 5081 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5082 | sector_t avail = rdev->sectors; |
01ab5662 | 5083 | |
d71f9f88 AN |
5084 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5085 | num_sectors = avail; | |
5086 | if (avail < num_sectors) | |
a35b0d69 N |
5087 | return -ENOSPC; |
5088 | } | |
d71f9f88 | 5089 | rv = mddev->pers->resize(mddev, num_sectors); |
a35b0d69 N |
5090 | if (!rv) { |
5091 | struct block_device *bdev; | |
5092 | ||
5093 | bdev = bdget_disk(mddev->gendisk, 0); | |
5094 | if (bdev) { | |
1b1dcc1b | 5095 | mutex_lock(&bdev->bd_inode->i_mutex); |
f233ea5c AN |
5096 | i_size_write(bdev->bd_inode, |
5097 | (loff_t)mddev->array_sectors << 9); | |
1b1dcc1b | 5098 | mutex_unlock(&bdev->bd_inode->i_mutex); |
a35b0d69 N |
5099 | bdput(bdev); |
5100 | } | |
5101 | } | |
5102 | return rv; | |
5103 | } | |
5104 | ||
da943b99 N |
5105 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5106 | { | |
5107 | int rv; | |
5108 | /* change the number of raid disks */ | |
63c70c4f | 5109 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5110 | return -EINVAL; |
5111 | if (raid_disks <= 0 || | |
5112 | raid_disks >= mddev->max_disks) | |
5113 | return -EINVAL; | |
63c70c4f | 5114 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5115 | return -EBUSY; |
63c70c4f N |
5116 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5117 | ||
5118 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
5119 | return rv; |
5120 | } | |
5121 | ||
5122 | ||
1da177e4 LT |
5123 | /* |
5124 | * update_array_info is used to change the configuration of an | |
5125 | * on-line array. | |
5126 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5127 | * fields in the info are checked against the array. | |
5128 | * Any differences that cannot be handled will cause an error. | |
5129 | * Normally, only one change can be managed at a time. | |
5130 | */ | |
5131 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5132 | { | |
5133 | int rv = 0; | |
5134 | int cnt = 0; | |
36fa3063 N |
5135 | int state = 0; |
5136 | ||
5137 | /* calculate expected state,ignoring low bits */ | |
5138 | if (mddev->bitmap && mddev->bitmap_offset) | |
5139 | state |= (1 << MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
5140 | |
5141 | if (mddev->major_version != info->major_version || | |
5142 | mddev->minor_version != info->minor_version || | |
5143 | /* mddev->patch_version != info->patch_version || */ | |
5144 | mddev->ctime != info->ctime || | |
5145 | mddev->level != info->level || | |
5146 | /* mddev->layout != info->layout || */ | |
5147 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5148 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5149 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5150 | ((state^info->state) & 0xfffffe00) | |
5151 | ) | |
1da177e4 LT |
5152 | return -EINVAL; |
5153 | /* Check there is only one change */ | |
58c0fed4 AN |
5154 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5155 | cnt++; | |
5156 | if (mddev->raid_disks != info->raid_disks) | |
5157 | cnt++; | |
5158 | if (mddev->layout != info->layout) | |
5159 | cnt++; | |
5160 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5161 | cnt++; | |
5162 | if (cnt == 0) | |
5163 | return 0; | |
5164 | if (cnt > 1) | |
5165 | return -EINVAL; | |
1da177e4 LT |
5166 | |
5167 | if (mddev->layout != info->layout) { | |
5168 | /* Change layout | |
5169 | * we don't need to do anything at the md level, the | |
5170 | * personality will take care of it all. | |
5171 | */ | |
50ac168a | 5172 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5173 | return -EINVAL; |
597a711b N |
5174 | else { |
5175 | mddev->new_layout = info->layout; | |
50ac168a | 5176 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5177 | if (rv) |
5178 | mddev->new_layout = mddev->layout; | |
5179 | return rv; | |
5180 | } | |
1da177e4 | 5181 | } |
58c0fed4 | 5182 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5183 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5184 | |
da943b99 N |
5185 | if (mddev->raid_disks != info->raid_disks) |
5186 | rv = update_raid_disks(mddev, info->raid_disks); | |
5187 | ||
36fa3063 N |
5188 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5189 | if (mddev->pers->quiesce == NULL) | |
5190 | return -EINVAL; | |
5191 | if (mddev->recovery || mddev->sync_thread) | |
5192 | return -EBUSY; | |
5193 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5194 | /* add the bitmap */ | |
5195 | if (mddev->bitmap) | |
5196 | return -EEXIST; | |
5197 | if (mddev->default_bitmap_offset == 0) | |
5198 | return -EINVAL; | |
5199 | mddev->bitmap_offset = mddev->default_bitmap_offset; | |
5200 | mddev->pers->quiesce(mddev, 1); | |
5201 | rv = bitmap_create(mddev); | |
5202 | if (rv) | |
5203 | bitmap_destroy(mddev); | |
5204 | mddev->pers->quiesce(mddev, 0); | |
5205 | } else { | |
5206 | /* remove the bitmap */ | |
5207 | if (!mddev->bitmap) | |
5208 | return -ENOENT; | |
5209 | if (mddev->bitmap->file) | |
5210 | return -EINVAL; | |
5211 | mddev->pers->quiesce(mddev, 1); | |
5212 | bitmap_destroy(mddev); | |
5213 | mddev->pers->quiesce(mddev, 0); | |
5214 | mddev->bitmap_offset = 0; | |
5215 | } | |
5216 | } | |
850b2b42 | 5217 | md_update_sb(mddev, 1); |
1da177e4 LT |
5218 | return rv; |
5219 | } | |
5220 | ||
5221 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5222 | { | |
5223 | mdk_rdev_t *rdev; | |
5224 | ||
5225 | if (mddev->pers == NULL) | |
5226 | return -ENODEV; | |
5227 | ||
5228 | rdev = find_rdev(mddev, dev); | |
5229 | if (!rdev) | |
5230 | return -ENODEV; | |
5231 | ||
5232 | md_error(mddev, rdev); | |
5233 | return 0; | |
5234 | } | |
5235 | ||
2f9618ce AN |
5236 | /* |
5237 | * We have a problem here : there is no easy way to give a CHS | |
5238 | * virtual geometry. We currently pretend that we have a 2 heads | |
5239 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5240 | * dosfs just mad... ;-) | |
5241 | */ | |
a885c8c4 CH |
5242 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5243 | { | |
5244 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5245 | ||
5246 | geo->heads = 2; | |
5247 | geo->sectors = 4; | |
5248 | geo->cylinders = get_capacity(mddev->gendisk) / 8; | |
5249 | return 0; | |
5250 | } | |
5251 | ||
a39907fa | 5252 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5253 | unsigned int cmd, unsigned long arg) |
5254 | { | |
5255 | int err = 0; | |
5256 | void __user *argp = (void __user *)arg; | |
1da177e4 LT |
5257 | mddev_t *mddev = NULL; |
5258 | ||
5259 | if (!capable(CAP_SYS_ADMIN)) | |
5260 | return -EACCES; | |
5261 | ||
5262 | /* | |
5263 | * Commands dealing with the RAID driver but not any | |
5264 | * particular array: | |
5265 | */ | |
5266 | switch (cmd) | |
5267 | { | |
5268 | case RAID_VERSION: | |
5269 | err = get_version(argp); | |
5270 | goto done; | |
5271 | ||
5272 | case PRINT_RAID_DEBUG: | |
5273 | err = 0; | |
5274 | md_print_devices(); | |
5275 | goto done; | |
5276 | ||
5277 | #ifndef MODULE | |
5278 | case RAID_AUTORUN: | |
5279 | err = 0; | |
5280 | autostart_arrays(arg); | |
5281 | goto done; | |
5282 | #endif | |
5283 | default:; | |
5284 | } | |
5285 | ||
5286 | /* | |
5287 | * Commands creating/starting a new array: | |
5288 | */ | |
5289 | ||
a39907fa | 5290 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5291 | |
5292 | if (!mddev) { | |
5293 | BUG(); | |
5294 | goto abort; | |
5295 | } | |
5296 | ||
1da177e4 LT |
5297 | err = mddev_lock(mddev); |
5298 | if (err) { | |
5299 | printk(KERN_INFO | |
5300 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5301 | err, cmd); | |
5302 | goto abort; | |
5303 | } | |
5304 | ||
5305 | switch (cmd) | |
5306 | { | |
5307 | case SET_ARRAY_INFO: | |
5308 | { | |
5309 | mdu_array_info_t info; | |
5310 | if (!arg) | |
5311 | memset(&info, 0, sizeof(info)); | |
5312 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5313 | err = -EFAULT; | |
5314 | goto abort_unlock; | |
5315 | } | |
5316 | if (mddev->pers) { | |
5317 | err = update_array_info(mddev, &info); | |
5318 | if (err) { | |
5319 | printk(KERN_WARNING "md: couldn't update" | |
5320 | " array info. %d\n", err); | |
5321 | goto abort_unlock; | |
5322 | } | |
5323 | goto done_unlock; | |
5324 | } | |
5325 | if (!list_empty(&mddev->disks)) { | |
5326 | printk(KERN_WARNING | |
5327 | "md: array %s already has disks!\n", | |
5328 | mdname(mddev)); | |
5329 | err = -EBUSY; | |
5330 | goto abort_unlock; | |
5331 | } | |
5332 | if (mddev->raid_disks) { | |
5333 | printk(KERN_WARNING | |
5334 | "md: array %s already initialised!\n", | |
5335 | mdname(mddev)); | |
5336 | err = -EBUSY; | |
5337 | goto abort_unlock; | |
5338 | } | |
5339 | err = set_array_info(mddev, &info); | |
5340 | if (err) { | |
5341 | printk(KERN_WARNING "md: couldn't set" | |
5342 | " array info. %d\n", err); | |
5343 | goto abort_unlock; | |
5344 | } | |
5345 | } | |
5346 | goto done_unlock; | |
5347 | ||
5348 | default:; | |
5349 | } | |
5350 | ||
5351 | /* | |
5352 | * Commands querying/configuring an existing array: | |
5353 | */ | |
32a7627c | 5354 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5355 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5356 | if ((!mddev->raid_disks && !mddev->external) |
5357 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5358 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5359 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5360 | err = -ENODEV; |
5361 | goto abort_unlock; | |
5362 | } | |
5363 | ||
5364 | /* | |
5365 | * Commands even a read-only array can execute: | |
5366 | */ | |
5367 | switch (cmd) | |
5368 | { | |
5369 | case GET_ARRAY_INFO: | |
5370 | err = get_array_info(mddev, argp); | |
5371 | goto done_unlock; | |
5372 | ||
32a7627c | 5373 | case GET_BITMAP_FILE: |
87162a28 | 5374 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5375 | goto done_unlock; |
5376 | ||
1da177e4 LT |
5377 | case GET_DISK_INFO: |
5378 | err = get_disk_info(mddev, argp); | |
5379 | goto done_unlock; | |
5380 | ||
5381 | case RESTART_ARRAY_RW: | |
5382 | err = restart_array(mddev); | |
5383 | goto done_unlock; | |
5384 | ||
5385 | case STOP_ARRAY: | |
d710e138 | 5386 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5387 | goto done_unlock; |
5388 | ||
5389 | case STOP_ARRAY_RO: | |
d710e138 | 5390 | err = do_md_stop(mddev, 1, 1); |
1da177e4 LT |
5391 | goto done_unlock; |
5392 | ||
1da177e4 LT |
5393 | } |
5394 | ||
5395 | /* | |
5396 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5397 | * superblock, so we do not allow them on read-only arrays. |
5398 | * However non-MD ioctls (e.g. get-size) will still come through | |
5399 | * here and hit the 'default' below, so only disallow | |
5400 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5401 | */ |
bb57fc64 | 5402 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5403 | if (mddev->ro == 2) { |
5404 | mddev->ro = 0; | |
b62b7590 | 5405 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 NB |
5406 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5407 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5408 | } else { |
5409 | err = -EROFS; | |
5410 | goto abort_unlock; | |
5411 | } | |
1da177e4 LT |
5412 | } |
5413 | ||
5414 | switch (cmd) | |
5415 | { | |
5416 | case ADD_NEW_DISK: | |
5417 | { | |
5418 | mdu_disk_info_t info; | |
5419 | if (copy_from_user(&info, argp, sizeof(info))) | |
5420 | err = -EFAULT; | |
5421 | else | |
5422 | err = add_new_disk(mddev, &info); | |
5423 | goto done_unlock; | |
5424 | } | |
5425 | ||
5426 | case HOT_REMOVE_DISK: | |
5427 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5428 | goto done_unlock; | |
5429 | ||
5430 | case HOT_ADD_DISK: | |
5431 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5432 | goto done_unlock; | |
5433 | ||
5434 | case SET_DISK_FAULTY: | |
5435 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5436 | goto done_unlock; | |
5437 | ||
5438 | case RUN_ARRAY: | |
d710e138 | 5439 | err = do_md_run(mddev); |
1da177e4 LT |
5440 | goto done_unlock; |
5441 | ||
32a7627c N |
5442 | case SET_BITMAP_FILE: |
5443 | err = set_bitmap_file(mddev, (int)arg); | |
5444 | goto done_unlock; | |
5445 | ||
1da177e4 | 5446 | default: |
1da177e4 LT |
5447 | err = -EINVAL; |
5448 | goto abort_unlock; | |
5449 | } | |
5450 | ||
5451 | done_unlock: | |
5452 | abort_unlock: | |
d3374825 N |
5453 | if (mddev->hold_active == UNTIL_IOCTL && |
5454 | err != -EINVAL) | |
5455 | mddev->hold_active = 0; | |
1da177e4 LT |
5456 | mddev_unlock(mddev); |
5457 | ||
5458 | return err; | |
5459 | done: | |
5460 | if (err) | |
5461 | MD_BUG(); | |
5462 | abort: | |
5463 | return err; | |
5464 | } | |
5465 | ||
a39907fa | 5466 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
5467 | { |
5468 | /* | |
5469 | * Succeed if we can lock the mddev, which confirms that | |
5470 | * it isn't being stopped right now. | |
5471 | */ | |
d3374825 | 5472 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
5473 | int err; |
5474 | ||
d3374825 N |
5475 | if (mddev->gendisk != bdev->bd_disk) { |
5476 | /* we are racing with mddev_put which is discarding this | |
5477 | * bd_disk. | |
5478 | */ | |
5479 | mddev_put(mddev); | |
5480 | /* Wait until bdev->bd_disk is definitely gone */ | |
5481 | flush_scheduled_work(); | |
5482 | /* Then retry the open from the top */ | |
5483 | return -ERESTARTSYS; | |
5484 | } | |
5485 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
5486 | ||
d63a5a74 | 5487 | if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1))) |
1da177e4 LT |
5488 | goto out; |
5489 | ||
5490 | err = 0; | |
f2ea68cf | 5491 | atomic_inc(&mddev->openers); |
1da177e4 LT |
5492 | mddev_unlock(mddev); |
5493 | ||
a39907fa | 5494 | check_disk_change(bdev); |
1da177e4 LT |
5495 | out: |
5496 | return err; | |
5497 | } | |
5498 | ||
a39907fa | 5499 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 5500 | { |
a39907fa | 5501 | mddev_t *mddev = disk->private_data; |
1da177e4 | 5502 | |
52e5f9d1 | 5503 | BUG_ON(!mddev); |
f2ea68cf | 5504 | atomic_dec(&mddev->openers); |
1da177e4 LT |
5505 | mddev_put(mddev); |
5506 | ||
5507 | return 0; | |
5508 | } | |
5509 | ||
44ce6294 LT |
5510 | static int md_media_changed(struct gendisk *disk) |
5511 | { | |
5512 | mddev_t *mddev = disk->private_data; | |
5513 | ||
5514 | return mddev->changed; | |
5515 | } | |
5516 | ||
5517 | static int md_revalidate(struct gendisk *disk) | |
5518 | { | |
5519 | mddev_t *mddev = disk->private_data; | |
5520 | ||
5521 | mddev->changed = 0; | |
5522 | return 0; | |
5523 | } | |
1da177e4 LT |
5524 | static struct block_device_operations md_fops = |
5525 | { | |
5526 | .owner = THIS_MODULE, | |
a39907fa AV |
5527 | .open = md_open, |
5528 | .release = md_release, | |
b492b852 | 5529 | .ioctl = md_ioctl, |
a885c8c4 | 5530 | .getgeo = md_getgeo, |
44ce6294 LT |
5531 | .media_changed = md_media_changed, |
5532 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
5533 | }; |
5534 | ||
75c96f85 | 5535 | static int md_thread(void * arg) |
1da177e4 LT |
5536 | { |
5537 | mdk_thread_t *thread = arg; | |
5538 | ||
1da177e4 LT |
5539 | /* |
5540 | * md_thread is a 'system-thread', it's priority should be very | |
5541 | * high. We avoid resource deadlocks individually in each | |
5542 | * raid personality. (RAID5 does preallocation) We also use RR and | |
5543 | * the very same RT priority as kswapd, thus we will never get | |
5544 | * into a priority inversion deadlock. | |
5545 | * | |
5546 | * we definitely have to have equal or higher priority than | |
5547 | * bdflush, otherwise bdflush will deadlock if there are too | |
5548 | * many dirty RAID5 blocks. | |
5549 | */ | |
1da177e4 | 5550 | |
6985c43f | 5551 | allow_signal(SIGKILL); |
a6fb0934 | 5552 | while (!kthread_should_stop()) { |
1da177e4 | 5553 | |
93588e22 N |
5554 | /* We need to wait INTERRUPTIBLE so that |
5555 | * we don't add to the load-average. | |
5556 | * That means we need to be sure no signals are | |
5557 | * pending | |
5558 | */ | |
5559 | if (signal_pending(current)) | |
5560 | flush_signals(current); | |
5561 | ||
5562 | wait_event_interruptible_timeout | |
5563 | (thread->wqueue, | |
5564 | test_bit(THREAD_WAKEUP, &thread->flags) | |
5565 | || kthread_should_stop(), | |
5566 | thread->timeout); | |
1da177e4 LT |
5567 | |
5568 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
5569 | ||
787453c2 | 5570 | thread->run(thread->mddev); |
1da177e4 | 5571 | } |
a6fb0934 | 5572 | |
1da177e4 LT |
5573 | return 0; |
5574 | } | |
5575 | ||
5576 | void md_wakeup_thread(mdk_thread_t *thread) | |
5577 | { | |
5578 | if (thread) { | |
5579 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
5580 | set_bit(THREAD_WAKEUP, &thread->flags); | |
5581 | wake_up(&thread->wqueue); | |
5582 | } | |
5583 | } | |
5584 | ||
5585 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
5586 | const char *name) | |
5587 | { | |
5588 | mdk_thread_t *thread; | |
1da177e4 | 5589 | |
9ffae0cf | 5590 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
5591 | if (!thread) |
5592 | return NULL; | |
5593 | ||
1da177e4 LT |
5594 | init_waitqueue_head(&thread->wqueue); |
5595 | ||
1da177e4 LT |
5596 | thread->run = run; |
5597 | thread->mddev = mddev; | |
32a7627c | 5598 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
6985c43f | 5599 | thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev)); |
a6fb0934 | 5600 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
5601 | kfree(thread); |
5602 | return NULL; | |
5603 | } | |
1da177e4 LT |
5604 | return thread; |
5605 | } | |
5606 | ||
1da177e4 LT |
5607 | void md_unregister_thread(mdk_thread_t *thread) |
5608 | { | |
e0cf8f04 N |
5609 | if (!thread) |
5610 | return; | |
ba25f9dc | 5611 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
5612 | |
5613 | kthread_stop(thread->tsk); | |
1da177e4 LT |
5614 | kfree(thread); |
5615 | } | |
5616 | ||
5617 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
5618 | { | |
5619 | if (!mddev) { | |
5620 | MD_BUG(); | |
5621 | return; | |
5622 | } | |
5623 | ||
b2d444d7 | 5624 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5625 | return; |
6bfe0b49 DW |
5626 | |
5627 | if (mddev->external) | |
5628 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 5629 | /* |
1da177e4 LT |
5630 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
5631 | mdname(mddev), | |
5632 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
5633 | __builtin_return_address(0),__builtin_return_address(1), | |
5634 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 5635 | */ |
d0a0a5ee AM |
5636 | if (!mddev->pers) |
5637 | return; | |
1da177e4 LT |
5638 | if (!mddev->pers->error_handler) |
5639 | return; | |
5640 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
5641 | if (mddev->degraded) |
5642 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
52664732 | 5643 | set_bit(StateChanged, &rdev->flags); |
1da177e4 LT |
5644 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
5645 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5646 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 5647 | md_new_event_inintr(mddev); |
1da177e4 LT |
5648 | } |
5649 | ||
5650 | /* seq_file implementation /proc/mdstat */ | |
5651 | ||
5652 | static void status_unused(struct seq_file *seq) | |
5653 | { | |
5654 | int i = 0; | |
5655 | mdk_rdev_t *rdev; | |
1da177e4 LT |
5656 | |
5657 | seq_printf(seq, "unused devices: "); | |
5658 | ||
159ec1fc | 5659 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
5660 | char b[BDEVNAME_SIZE]; |
5661 | i++; | |
5662 | seq_printf(seq, "%s ", | |
5663 | bdevname(rdev->bdev,b)); | |
5664 | } | |
5665 | if (!i) | |
5666 | seq_printf(seq, "<none>"); | |
5667 | ||
5668 | seq_printf(seq, "\n"); | |
5669 | } | |
5670 | ||
5671 | ||
5672 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
5673 | { | |
dd71cf6b N |
5674 | sector_t max_sectors, resync, res; |
5675 | unsigned long dt, db; | |
5676 | sector_t rt; | |
4588b42e N |
5677 | int scale; |
5678 | unsigned int per_milli; | |
1da177e4 | 5679 | |
dd71cf6b | 5680 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
5681 | |
5682 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 5683 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 5684 | else |
dd71cf6b | 5685 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
5686 | |
5687 | /* | |
5688 | * Should not happen. | |
5689 | */ | |
dd71cf6b | 5690 | if (!max_sectors) { |
1da177e4 LT |
5691 | MD_BUG(); |
5692 | return; | |
5693 | } | |
4588b42e | 5694 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 5695 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
5696 | * u32, as those are the requirements for sector_div. |
5697 | * Thus 'scale' must be at least 10 | |
5698 | */ | |
5699 | scale = 10; | |
5700 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 5701 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
5702 | scale++; |
5703 | } | |
5704 | res = (resync>>scale)*1000; | |
dd71cf6b | 5705 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
5706 | |
5707 | per_milli = res; | |
1da177e4 | 5708 | { |
4588b42e | 5709 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
5710 | seq_printf(seq, "["); |
5711 | for (i = 0; i < x; i++) | |
5712 | seq_printf(seq, "="); | |
5713 | seq_printf(seq, ">"); | |
5714 | for (i = 0; i < y; i++) | |
5715 | seq_printf(seq, "."); | |
5716 | seq_printf(seq, "] "); | |
5717 | } | |
4588b42e | 5718 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
5719 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
5720 | "reshape" : | |
61df9d91 N |
5721 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
5722 | "check" : | |
5723 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
5724 | "resync" : "recovery"))), | |
5725 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
5726 | (unsigned long long) resync/2, |
5727 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
5728 | |
5729 | /* | |
1da177e4 LT |
5730 | * dt: time from mark until now |
5731 | * db: blocks written from mark until now | |
5732 | * rt: remaining time | |
dd71cf6b N |
5733 | * |
5734 | * rt is a sector_t, so could be 32bit or 64bit. | |
5735 | * So we divide before multiply in case it is 32bit and close | |
5736 | * to the limit. | |
5737 | * We scale the divisor (db) by 32 to avoid loosing precision | |
5738 | * near the end of resync when the number of remaining sectors | |
5739 | * is close to 'db'. | |
5740 | * We then divide rt by 32 after multiplying by db to compensate. | |
5741 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
5742 | */ |
5743 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
5744 | if (!dt) dt++; | |
ff4e8d9a N |
5745 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
5746 | - mddev->resync_mark_cnt; | |
1da177e4 | 5747 | |
dd71cf6b N |
5748 | rt = max_sectors - resync; /* number of remaining sectors */ |
5749 | sector_div(rt, db/32+1); | |
5750 | rt *= dt; | |
5751 | rt >>= 5; | |
5752 | ||
5753 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
5754 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 5755 | |
ff4e8d9a | 5756 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
5757 | } |
5758 | ||
5759 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
5760 | { | |
5761 | struct list_head *tmp; | |
5762 | loff_t l = *pos; | |
5763 | mddev_t *mddev; | |
5764 | ||
5765 | if (l >= 0x10000) | |
5766 | return NULL; | |
5767 | if (!l--) | |
5768 | /* header */ | |
5769 | return (void*)1; | |
5770 | ||
5771 | spin_lock(&all_mddevs_lock); | |
5772 | list_for_each(tmp,&all_mddevs) | |
5773 | if (!l--) { | |
5774 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
5775 | mddev_get(mddev); | |
5776 | spin_unlock(&all_mddevs_lock); | |
5777 | return mddev; | |
5778 | } | |
5779 | spin_unlock(&all_mddevs_lock); | |
5780 | if (!l--) | |
5781 | return (void*)2;/* tail */ | |
5782 | return NULL; | |
5783 | } | |
5784 | ||
5785 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
5786 | { | |
5787 | struct list_head *tmp; | |
5788 | mddev_t *next_mddev, *mddev = v; | |
5789 | ||
5790 | ++*pos; | |
5791 | if (v == (void*)2) | |
5792 | return NULL; | |
5793 | ||
5794 | spin_lock(&all_mddevs_lock); | |
5795 | if (v == (void*)1) | |
5796 | tmp = all_mddevs.next; | |
5797 | else | |
5798 | tmp = mddev->all_mddevs.next; | |
5799 | if (tmp != &all_mddevs) | |
5800 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
5801 | else { | |
5802 | next_mddev = (void*)2; | |
5803 | *pos = 0x10000; | |
5804 | } | |
5805 | spin_unlock(&all_mddevs_lock); | |
5806 | ||
5807 | if (v != (void*)1) | |
5808 | mddev_put(mddev); | |
5809 | return next_mddev; | |
5810 | ||
5811 | } | |
5812 | ||
5813 | static void md_seq_stop(struct seq_file *seq, void *v) | |
5814 | { | |
5815 | mddev_t *mddev = v; | |
5816 | ||
5817 | if (mddev && v != (void*)1 && v != (void*)2) | |
5818 | mddev_put(mddev); | |
5819 | } | |
5820 | ||
d7603b7e N |
5821 | struct mdstat_info { |
5822 | int event; | |
5823 | }; | |
5824 | ||
1da177e4 LT |
5825 | static int md_seq_show(struct seq_file *seq, void *v) |
5826 | { | |
5827 | mddev_t *mddev = v; | |
dd8ac336 | 5828 | sector_t sectors; |
1da177e4 | 5829 | mdk_rdev_t *rdev; |
d7603b7e | 5830 | struct mdstat_info *mi = seq->private; |
32a7627c | 5831 | struct bitmap *bitmap; |
1da177e4 LT |
5832 | |
5833 | if (v == (void*)1) { | |
2604b703 | 5834 | struct mdk_personality *pers; |
1da177e4 LT |
5835 | seq_printf(seq, "Personalities : "); |
5836 | spin_lock(&pers_lock); | |
2604b703 N |
5837 | list_for_each_entry(pers, &pers_list, list) |
5838 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
5839 | |
5840 | spin_unlock(&pers_lock); | |
5841 | seq_printf(seq, "\n"); | |
d7603b7e | 5842 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
5843 | return 0; |
5844 | } | |
5845 | if (v == (void*)2) { | |
5846 | status_unused(seq); | |
5847 | return 0; | |
5848 | } | |
5849 | ||
5dc5cf7d | 5850 | if (mddev_lock(mddev) < 0) |
1da177e4 | 5851 | return -EINTR; |
5dc5cf7d | 5852 | |
1da177e4 LT |
5853 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
5854 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
5855 | mddev->pers ? "" : "in"); | |
5856 | if (mddev->pers) { | |
f91de92e | 5857 | if (mddev->ro==1) |
1da177e4 | 5858 | seq_printf(seq, " (read-only)"); |
f91de92e | 5859 | if (mddev->ro==2) |
52720ae7 | 5860 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
5861 | seq_printf(seq, " %s", mddev->pers->name); |
5862 | } | |
5863 | ||
dd8ac336 | 5864 | sectors = 0; |
159ec1fc | 5865 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
5866 | char b[BDEVNAME_SIZE]; |
5867 | seq_printf(seq, " %s[%d]", | |
5868 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
5869 | if (test_bit(WriteMostly, &rdev->flags)) |
5870 | seq_printf(seq, "(W)"); | |
b2d444d7 | 5871 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5872 | seq_printf(seq, "(F)"); |
5873 | continue; | |
b325a32e N |
5874 | } else if (rdev->raid_disk < 0) |
5875 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 5876 | sectors += rdev->sectors; |
1da177e4 LT |
5877 | } |
5878 | ||
5879 | if (!list_empty(&mddev->disks)) { | |
5880 | if (mddev->pers) | |
5881 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
5882 | (unsigned long long) |
5883 | mddev->array_sectors / 2); | |
1da177e4 LT |
5884 | else |
5885 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 5886 | (unsigned long long)sectors / 2); |
1da177e4 | 5887 | } |
1cd6bf19 N |
5888 | if (mddev->persistent) { |
5889 | if (mddev->major_version != 0 || | |
5890 | mddev->minor_version != 90) { | |
5891 | seq_printf(seq," super %d.%d", | |
5892 | mddev->major_version, | |
5893 | mddev->minor_version); | |
5894 | } | |
e691063a N |
5895 | } else if (mddev->external) |
5896 | seq_printf(seq, " super external:%s", | |
5897 | mddev->metadata_type); | |
5898 | else | |
1cd6bf19 | 5899 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
5900 | |
5901 | if (mddev->pers) { | |
d710e138 | 5902 | mddev->pers->status(seq, mddev); |
1da177e4 | 5903 | seq_printf(seq, "\n "); |
8e1b39d6 N |
5904 | if (mddev->pers->sync_request) { |
5905 | if (mddev->curr_resync > 2) { | |
d710e138 | 5906 | status_resync(seq, mddev); |
8e1b39d6 N |
5907 | seq_printf(seq, "\n "); |
5908 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
5909 | seq_printf(seq, "\tresync=DELAYED\n "); | |
5910 | else if (mddev->recovery_cp < MaxSector) | |
5911 | seq_printf(seq, "\tresync=PENDING\n "); | |
5912 | } | |
32a7627c N |
5913 | } else |
5914 | seq_printf(seq, "\n "); | |
5915 | ||
5916 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
5917 | unsigned long chunk_kb; |
5918 | unsigned long flags; | |
32a7627c N |
5919 | spin_lock_irqsave(&bitmap->lock, flags); |
5920 | chunk_kb = bitmap->chunksize >> 10; | |
5921 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
5922 | "%lu%s chunk", | |
5923 | bitmap->pages - bitmap->missing_pages, | |
5924 | bitmap->pages, | |
5925 | (bitmap->pages - bitmap->missing_pages) | |
5926 | << (PAGE_SHIFT - 10), | |
5927 | chunk_kb ? chunk_kb : bitmap->chunksize, | |
5928 | chunk_kb ? "KB" : "B"); | |
78d742d8 N |
5929 | if (bitmap->file) { |
5930 | seq_printf(seq, ", file: "); | |
c32c2f63 | 5931 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 5932 | } |
78d742d8 | 5933 | |
32a7627c N |
5934 | seq_printf(seq, "\n"); |
5935 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
5936 | } |
5937 | ||
5938 | seq_printf(seq, "\n"); | |
5939 | } | |
5940 | mddev_unlock(mddev); | |
5941 | ||
5942 | return 0; | |
5943 | } | |
5944 | ||
110518bc | 5945 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
5946 | .start = md_seq_start, |
5947 | .next = md_seq_next, | |
5948 | .stop = md_seq_stop, | |
5949 | .show = md_seq_show, | |
5950 | }; | |
5951 | ||
5952 | static int md_seq_open(struct inode *inode, struct file *file) | |
5953 | { | |
5954 | int error; | |
d7603b7e N |
5955 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
5956 | if (mi == NULL) | |
5957 | return -ENOMEM; | |
1da177e4 LT |
5958 | |
5959 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
5960 | if (error) |
5961 | kfree(mi); | |
5962 | else { | |
5963 | struct seq_file *p = file->private_data; | |
5964 | p->private = mi; | |
5965 | mi->event = atomic_read(&md_event_count); | |
5966 | } | |
1da177e4 LT |
5967 | return error; |
5968 | } | |
5969 | ||
d7603b7e N |
5970 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
5971 | { | |
5972 | struct seq_file *m = filp->private_data; | |
5973 | struct mdstat_info *mi = m->private; | |
5974 | int mask; | |
5975 | ||
5976 | poll_wait(filp, &md_event_waiters, wait); | |
5977 | ||
5978 | /* always allow read */ | |
5979 | mask = POLLIN | POLLRDNORM; | |
5980 | ||
5981 | if (mi->event != atomic_read(&md_event_count)) | |
5982 | mask |= POLLERR | POLLPRI; | |
5983 | return mask; | |
5984 | } | |
5985 | ||
fa027c2a | 5986 | static const struct file_operations md_seq_fops = { |
e24650c2 | 5987 | .owner = THIS_MODULE, |
1da177e4 LT |
5988 | .open = md_seq_open, |
5989 | .read = seq_read, | |
5990 | .llseek = seq_lseek, | |
c3f94b40 | 5991 | .release = seq_release_private, |
d7603b7e | 5992 | .poll = mdstat_poll, |
1da177e4 LT |
5993 | }; |
5994 | ||
2604b703 | 5995 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 5996 | { |
1da177e4 | 5997 | spin_lock(&pers_lock); |
2604b703 N |
5998 | list_add_tail(&p->list, &pers_list); |
5999 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6000 | spin_unlock(&pers_lock); |
6001 | return 0; | |
6002 | } | |
6003 | ||
2604b703 | 6004 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6005 | { |
2604b703 | 6006 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6007 | spin_lock(&pers_lock); |
2604b703 | 6008 | list_del_init(&p->list); |
1da177e4 LT |
6009 | spin_unlock(&pers_lock); |
6010 | return 0; | |
6011 | } | |
6012 | ||
eea1bf38 | 6013 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6014 | { |
6015 | mdk_rdev_t * rdev; | |
1da177e4 | 6016 | int idle; |
eea1bf38 | 6017 | int curr_events; |
1da177e4 LT |
6018 | |
6019 | idle = 1; | |
4b80991c N |
6020 | rcu_read_lock(); |
6021 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6022 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6023 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6024 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6025 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6026 | /* sync IO will cause sync_io to increase before the disk_stats |
6027 | * as sync_io is counted when a request starts, and | |
6028 | * disk_stats is counted when it completes. | |
6029 | * So resync activity will cause curr_events to be smaller than | |
6030 | * when there was no such activity. | |
6031 | * non-sync IO will cause disk_stat to increase without | |
6032 | * increasing sync_io so curr_events will (eventually) | |
6033 | * be larger than it was before. Once it becomes | |
6034 | * substantially larger, the test below will cause | |
6035 | * the array to appear non-idle, and resync will slow | |
6036 | * down. | |
6037 | * If there is a lot of outstanding resync activity when | |
6038 | * we set last_event to curr_events, then all that activity | |
6039 | * completing might cause the array to appear non-idle | |
6040 | * and resync will be slowed down even though there might | |
6041 | * not have been non-resync activity. This will only | |
6042 | * happen once though. 'last_events' will soon reflect | |
6043 | * the state where there is little or no outstanding | |
6044 | * resync requests, and further resync activity will | |
6045 | * always make curr_events less than last_events. | |
c0e48521 | 6046 | * |
1da177e4 | 6047 | */ |
eea1bf38 | 6048 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6049 | rdev->last_events = curr_events; |
6050 | idle = 0; | |
6051 | } | |
6052 | } | |
4b80991c | 6053 | rcu_read_unlock(); |
1da177e4 LT |
6054 | return idle; |
6055 | } | |
6056 | ||
6057 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6058 | { | |
6059 | /* another "blocks" (512byte) blocks have been synced */ | |
6060 | atomic_sub(blocks, &mddev->recovery_active); | |
6061 | wake_up(&mddev->recovery_wait); | |
6062 | if (!ok) { | |
dfc70645 | 6063 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6064 | md_wakeup_thread(mddev->thread); |
6065 | // stop recovery, signal do_sync .... | |
6066 | } | |
6067 | } | |
6068 | ||
6069 | ||
06d91a5f N |
6070 | /* md_write_start(mddev, bi) |
6071 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6072 | * in superblock) before writing, schedule a superblock update |
6073 | * and wait for it to complete. | |
06d91a5f | 6074 | */ |
3d310eb7 | 6075 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6076 | { |
0fd62b86 | 6077 | int did_change = 0; |
06d91a5f | 6078 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6079 | return; |
06d91a5f | 6080 | |
f91de92e N |
6081 | BUG_ON(mddev->ro == 1); |
6082 | if (mddev->ro == 2) { | |
6083 | /* need to switch to read/write */ | |
6084 | mddev->ro = 0; | |
6085 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6086 | md_wakeup_thread(mddev->thread); | |
25156198 | 6087 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6088 | did_change = 1; |
f91de92e | 6089 | } |
06d91a5f | 6090 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6091 | if (mddev->safemode == 1) |
6092 | mddev->safemode = 0; | |
06d91a5f | 6093 | if (mddev->in_sync) { |
a9701a30 | 6094 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6095 | if (mddev->in_sync) { |
6096 | mddev->in_sync = 0; | |
850b2b42 | 6097 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3d310eb7 | 6098 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6099 | did_change = 1; |
3d310eb7 | 6100 | } |
a9701a30 | 6101 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6102 | } |
0fd62b86 | 6103 | if (did_change) |
b62b7590 | 6104 | sysfs_notify_dirent(mddev->sysfs_state); |
09a44cc1 N |
6105 | wait_event(mddev->sb_wait, |
6106 | !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && | |
6107 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1da177e4 LT |
6108 | } |
6109 | ||
6110 | void md_write_end(mddev_t *mddev) | |
6111 | { | |
6112 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6113 | if (mddev->safemode == 2) | |
6114 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6115 | else if (mddev->safemode_delay) |
1da177e4 LT |
6116 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6117 | } | |
6118 | } | |
6119 | ||
2a2275d6 N |
6120 | /* md_allow_write(mddev) |
6121 | * Calling this ensures that the array is marked 'active' so that writes | |
6122 | * may proceed without blocking. It is important to call this before | |
6123 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6124 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6125 | * |
6126 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6127 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6128 | */ |
b5470dc5 | 6129 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6130 | { |
6131 | if (!mddev->pers) | |
b5470dc5 | 6132 | return 0; |
2a2275d6 | 6133 | if (mddev->ro) |
b5470dc5 | 6134 | return 0; |
1a0fd497 | 6135 | if (!mddev->pers->sync_request) |
b5470dc5 | 6136 | return 0; |
2a2275d6 N |
6137 | |
6138 | spin_lock_irq(&mddev->write_lock); | |
6139 | if (mddev->in_sync) { | |
6140 | mddev->in_sync = 0; | |
6141 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6142 | if (mddev->safemode_delay && | |
6143 | mddev->safemode == 0) | |
6144 | mddev->safemode = 1; | |
6145 | spin_unlock_irq(&mddev->write_lock); | |
6146 | md_update_sb(mddev, 0); | |
b62b7590 | 6147 | sysfs_notify_dirent(mddev->sysfs_state); |
2a2275d6 N |
6148 | } else |
6149 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 DW |
6150 | |
6151 | if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
6152 | return -EAGAIN; | |
6153 | else | |
6154 | return 0; | |
2a2275d6 N |
6155 | } |
6156 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6157 | ||
1da177e4 LT |
6158 | #define SYNC_MARKS 10 |
6159 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6160 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6161 | { |
6162 | mddev_t *mddev2; | |
6163 | unsigned int currspeed = 0, | |
6164 | window; | |
57afd89f | 6165 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6166 | unsigned long mark[SYNC_MARKS]; |
6167 | sector_t mark_cnt[SYNC_MARKS]; | |
6168 | int last_mark,m; | |
6169 | struct list_head *tmp; | |
6170 | sector_t last_check; | |
57afd89f | 6171 | int skipped = 0; |
5fd6c1dc | 6172 | mdk_rdev_t *rdev; |
61df9d91 | 6173 | char *desc; |
1da177e4 LT |
6174 | |
6175 | /* just incase thread restarts... */ | |
6176 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6177 | return; | |
5fd6c1dc N |
6178 | if (mddev->ro) /* never try to sync a read-only array */ |
6179 | return; | |
1da177e4 | 6180 | |
61df9d91 N |
6181 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6182 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6183 | desc = "data-check"; | |
6184 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6185 | desc = "requested-resync"; | |
6186 | else | |
6187 | desc = "resync"; | |
6188 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6189 | desc = "reshape"; | |
6190 | else | |
6191 | desc = "recovery"; | |
6192 | ||
1da177e4 LT |
6193 | /* we overload curr_resync somewhat here. |
6194 | * 0 == not engaged in resync at all | |
6195 | * 2 == checking that there is no conflict with another sync | |
6196 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6197 | * commense | |
6198 | * other == active in resync - this many blocks | |
6199 | * | |
6200 | * Before starting a resync we must have set curr_resync to | |
6201 | * 2, and then checked that every "conflicting" array has curr_resync | |
6202 | * less than ours. When we find one that is the same or higher | |
6203 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6204 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6205 | * This will mean we have to start checking from the beginning again. | |
6206 | * | |
6207 | */ | |
6208 | ||
6209 | do { | |
6210 | mddev->curr_resync = 2; | |
6211 | ||
6212 | try_again: | |
787453c2 | 6213 | if (kthread_should_stop()) { |
6985c43f | 6214 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6215 | goto skip; |
6216 | } | |
29ac4aa3 | 6217 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6218 | if (mddev2 == mddev) |
6219 | continue; | |
90b08710 BS |
6220 | if (!mddev->parallel_resync |
6221 | && mddev2->curr_resync | |
6222 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6223 | DEFINE_WAIT(wq); |
6224 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6225 | /* arbitrarily yield */ | |
6226 | mddev->curr_resync = 1; | |
6227 | wake_up(&resync_wait); | |
6228 | } | |
6229 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6230 | /* no need to wait here, we can wait the next | |
6231 | * time 'round when curr_resync == 2 | |
6232 | */ | |
6233 | continue; | |
9744197c N |
6234 | /* We need to wait 'interruptible' so as not to |
6235 | * contribute to the load average, and not to | |
6236 | * be caught by 'softlockup' | |
6237 | */ | |
6238 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6239 | if (!kthread_should_stop() && |
8712e553 | 6240 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6241 | printk(KERN_INFO "md: delaying %s of %s" |
6242 | " until %s has finished (they" | |
1da177e4 | 6243 | " share one or more physical units)\n", |
61df9d91 | 6244 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6245 | mddev_put(mddev2); |
9744197c N |
6246 | if (signal_pending(current)) |
6247 | flush_signals(current); | |
1da177e4 LT |
6248 | schedule(); |
6249 | finish_wait(&resync_wait, &wq); | |
6250 | goto try_again; | |
6251 | } | |
6252 | finish_wait(&resync_wait, &wq); | |
6253 | } | |
6254 | } | |
6255 | } while (mddev->curr_resync < 2); | |
6256 | ||
5fd6c1dc | 6257 | j = 0; |
9d88883e | 6258 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6259 | /* resync follows the size requested by the personality, |
57afd89f | 6260 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6261 | */ |
6262 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6263 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6264 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6265 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6266 | j = mddev->resync_min; | |
6267 | else if (!mddev->bitmap) | |
5fd6c1dc | 6268 | j = mddev->recovery_cp; |
5e96ee65 | 6269 | |
ccfcc3c1 | 6270 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6271 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6272 | else { |
1da177e4 | 6273 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6274 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6275 | j = MaxSector; |
159ec1fc | 6276 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
6277 | if (rdev->raid_disk >= 0 && |
6278 | !test_bit(Faulty, &rdev->flags) && | |
6279 | !test_bit(In_sync, &rdev->flags) && | |
6280 | rdev->recovery_offset < j) | |
6281 | j = rdev->recovery_offset; | |
6282 | } | |
1da177e4 | 6283 | |
61df9d91 N |
6284 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6285 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6286 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6287 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6288 | "(but not more than %d KB/sec) for %s.\n", |
6289 | speed_max(mddev), desc); | |
1da177e4 | 6290 | |
eea1bf38 | 6291 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6292 | |
57afd89f | 6293 | io_sectors = 0; |
1da177e4 LT |
6294 | for (m = 0; m < SYNC_MARKS; m++) { |
6295 | mark[m] = jiffies; | |
57afd89f | 6296 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6297 | } |
6298 | last_mark = 0; | |
6299 | mddev->resync_mark = mark[last_mark]; | |
6300 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6301 | ||
6302 | /* | |
6303 | * Tune reconstruction: | |
6304 | */ | |
6305 | window = 32*(PAGE_SIZE/512); | |
6306 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6307 | window/2,(unsigned long long) max_sectors/2); | |
6308 | ||
6309 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6310 | last_check = 0; |
6311 | ||
6312 | if (j>2) { | |
6313 | printk(KERN_INFO | |
61df9d91 N |
6314 | "md: resuming %s of %s from checkpoint.\n", |
6315 | desc, mdname(mddev)); | |
1da177e4 LT |
6316 | mddev->curr_resync = j; |
6317 | } | |
6318 | ||
6319 | while (j < max_sectors) { | |
57afd89f | 6320 | sector_t sectors; |
1da177e4 | 6321 | |
57afd89f | 6322 | skipped = 0; |
97e4f42d | 6323 | |
7a91ee1f N |
6324 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6325 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6326 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6327 | > (max_sectors >> 4)) || | |
6328 | (j - mddev->curr_resync_completed)*2 | |
6329 | >= mddev->resync_max - mddev->curr_resync_completed | |
6330 | )) { | |
97e4f42d N |
6331 | /* time to update curr_resync_completed */ |
6332 | blk_unplug(mddev->queue); | |
6333 | wait_event(mddev->recovery_wait, | |
6334 | atomic_read(&mddev->recovery_active) == 0); | |
6335 | mddev->curr_resync_completed = | |
6336 | mddev->curr_resync; | |
6337 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
acb180b0 | 6338 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6339 | } |
acb180b0 N |
6340 | |
6341 | if (j >= mddev->resync_max) | |
6342 | wait_event(mddev->recovery_wait, | |
6343 | mddev->resync_max > j | |
6344 | || kthread_should_stop()); | |
6345 | ||
6346 | if (kthread_should_stop()) | |
6347 | goto interrupted; | |
6348 | ||
57afd89f | 6349 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6350 | currspeed < speed_min(mddev)); |
57afd89f | 6351 | if (sectors == 0) { |
dfc70645 | 6352 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6353 | goto out; |
6354 | } | |
57afd89f N |
6355 | |
6356 | if (!skipped) { /* actual IO requested */ | |
6357 | io_sectors += sectors; | |
6358 | atomic_add(sectors, &mddev->recovery_active); | |
6359 | } | |
6360 | ||
1da177e4 LT |
6361 | j += sectors; |
6362 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6363 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6364 | if (last_check == 0) |
6365 | /* this is the earliers that rebuilt will be | |
6366 | * visible in /proc/mdstat | |
6367 | */ | |
6368 | md_new_event(mddev); | |
57afd89f N |
6369 | |
6370 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6371 | continue; |
6372 | ||
57afd89f | 6373 | last_check = io_sectors; |
1da177e4 | 6374 | |
dfc70645 | 6375 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6376 | break; |
6377 | ||
6378 | repeat: | |
6379 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6380 | /* step marks */ | |
6381 | int next = (last_mark+1) % SYNC_MARKS; | |
6382 | ||
6383 | mddev->resync_mark = mark[next]; | |
6384 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6385 | mark[next] = jiffies; | |
57afd89f | 6386 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6387 | last_mark = next; |
6388 | } | |
6389 | ||
6390 | ||
c6207277 N |
6391 | if (kthread_should_stop()) |
6392 | goto interrupted; | |
6393 | ||
1da177e4 LT |
6394 | |
6395 | /* | |
6396 | * this loop exits only if either when we are slower than | |
6397 | * the 'hard' speed limit, or the system was IO-idle for | |
6398 | * a jiffy. | |
6399 | * the system might be non-idle CPU-wise, but we only care | |
6400 | * about not overloading the IO subsystem. (things like an | |
6401 | * e2fsck being done on the RAID array should execute fast) | |
6402 | */ | |
2ad8b1ef | 6403 | blk_unplug(mddev->queue); |
1da177e4 LT |
6404 | cond_resched(); |
6405 | ||
57afd89f N |
6406 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6407 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6408 | |
88202a0c N |
6409 | if (currspeed > speed_min(mddev)) { |
6410 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6411 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6412 | msleep(500); |
1da177e4 LT |
6413 | goto repeat; |
6414 | } | |
6415 | } | |
6416 | } | |
61df9d91 | 6417 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6418 | /* |
6419 | * this also signals 'finished resyncing' to md_stop | |
6420 | */ | |
6421 | out: | |
2ad8b1ef | 6422 | blk_unplug(mddev->queue); |
1da177e4 LT |
6423 | |
6424 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
6425 | ||
6426 | /* tell personality that we are finished */ | |
57afd89f | 6427 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6428 | |
dfc70645 | 6429 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6430 | mddev->curr_resync > 2) { |
6431 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6432 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6433 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6434 | printk(KERN_INFO | |
61df9d91 N |
6435 | "md: checkpointing %s of %s.\n", |
6436 | desc, mdname(mddev)); | |
5fd6c1dc N |
6437 | mddev->recovery_cp = mddev->curr_resync; |
6438 | } | |
6439 | } else | |
6440 | mddev->recovery_cp = MaxSector; | |
6441 | } else { | |
6442 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6443 | mddev->curr_resync = MaxSector; | |
159ec1fc | 6444 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
6445 | if (rdev->raid_disk >= 0 && |
6446 | !test_bit(Faulty, &rdev->flags) && | |
6447 | !test_bit(In_sync, &rdev->flags) && | |
6448 | rdev->recovery_offset < mddev->curr_resync) | |
6449 | rdev->recovery_offset = mddev->curr_resync; | |
5fd6c1dc | 6450 | } |
1da177e4 | 6451 | } |
17571284 | 6452 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6453 | |
1da177e4 LT |
6454 | skip: |
6455 | mddev->curr_resync = 0; | |
acb180b0 | 6456 | mddev->curr_resync_completed = 0; |
5e96ee65 | 6457 | mddev->resync_min = 0; |
c6207277 N |
6458 | mddev->resync_max = MaxSector; |
6459 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
1da177e4 LT |
6460 | wake_up(&resync_wait); |
6461 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
6462 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
6463 | return; |
6464 | ||
6465 | interrupted: | |
6466 | /* | |
6467 | * got a signal, exit. | |
6468 | */ | |
6469 | printk(KERN_INFO | |
6470 | "md: md_do_sync() got signal ... exiting\n"); | |
6471 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6472 | goto out; | |
6473 | ||
1da177e4 | 6474 | } |
29269553 | 6475 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
6476 | |
6477 | ||
b4c4c7b8 N |
6478 | static int remove_and_add_spares(mddev_t *mddev) |
6479 | { | |
6480 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
6481 | int spares = 0; |
6482 | ||
97e4f42d N |
6483 | mddev->curr_resync_completed = 0; |
6484 | ||
159ec1fc | 6485 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 6486 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 6487 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
6488 | (test_bit(Faulty, &rdev->flags) || |
6489 | ! test_bit(In_sync, &rdev->flags)) && | |
6490 | atomic_read(&rdev->nr_pending)==0) { | |
6491 | if (mddev->pers->hot_remove_disk( | |
6492 | mddev, rdev->raid_disk)==0) { | |
6493 | char nm[20]; | |
6494 | sprintf(nm,"rd%d", rdev->raid_disk); | |
6495 | sysfs_remove_link(&mddev->kobj, nm); | |
6496 | rdev->raid_disk = -1; | |
6497 | } | |
6498 | } | |
6499 | ||
4044ba58 | 6500 | if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { |
159ec1fc | 6501 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 6502 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
6503 | !test_bit(In_sync, &rdev->flags) && |
6504 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 6505 | spares++; |
b4c4c7b8 N |
6506 | if (rdev->raid_disk < 0 |
6507 | && !test_bit(Faulty, &rdev->flags)) { | |
6508 | rdev->recovery_offset = 0; | |
199050ea NB |
6509 | if (mddev->pers-> |
6510 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
6511 | char nm[20]; |
6512 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
6513 | if (sysfs_create_link(&mddev->kobj, |
6514 | &rdev->kobj, nm)) | |
6515 | printk(KERN_WARNING | |
6516 | "md: cannot register " | |
6517 | "%s for %s\n", | |
6518 | nm, mdname(mddev)); | |
b4c4c7b8 N |
6519 | spares++; |
6520 | md_new_event(mddev); | |
6521 | } else | |
6522 | break; | |
6523 | } | |
dfc70645 | 6524 | } |
b4c4c7b8 N |
6525 | } |
6526 | return spares; | |
6527 | } | |
1da177e4 LT |
6528 | /* |
6529 | * This routine is regularly called by all per-raid-array threads to | |
6530 | * deal with generic issues like resync and super-block update. | |
6531 | * Raid personalities that don't have a thread (linear/raid0) do not | |
6532 | * need this as they never do any recovery or update the superblock. | |
6533 | * | |
6534 | * It does not do any resync itself, but rather "forks" off other threads | |
6535 | * to do that as needed. | |
6536 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
6537 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 6538 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
6539 | * and wakeups up this thread which will reap the thread and finish up. |
6540 | * This thread also removes any faulty devices (with nr_pending == 0). | |
6541 | * | |
6542 | * The overall approach is: | |
6543 | * 1/ if the superblock needs updating, update it. | |
6544 | * 2/ If a recovery thread is running, don't do anything else. | |
6545 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
6546 | * 4/ If there are any faulty devices, remove them. | |
6547 | * 5/ If array is degraded, try to add spares devices | |
6548 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
6549 | */ | |
6550 | void md_check_recovery(mddev_t *mddev) | |
6551 | { | |
6552 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6553 | |
6554 | ||
5f40402d N |
6555 | if (mddev->bitmap) |
6556 | bitmap_daemon_work(mddev->bitmap); | |
1da177e4 LT |
6557 | |
6558 | if (mddev->ro) | |
6559 | return; | |
fca4d848 N |
6560 | |
6561 | if (signal_pending(current)) { | |
31a59e34 | 6562 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
6563 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
6564 | mdname(mddev)); | |
6565 | mddev->safemode = 2; | |
6566 | } | |
6567 | flush_signals(current); | |
6568 | } | |
6569 | ||
c89a8eee N |
6570 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
6571 | return; | |
1da177e4 | 6572 | if ( ! ( |
e691063a | 6573 | (mddev->flags && !mddev->external) || |
1da177e4 | 6574 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 6575 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 6576 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
6577 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
6578 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
6579 | )) |
6580 | return; | |
fca4d848 | 6581 | |
df5b89b3 | 6582 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 6583 | int spares = 0; |
fca4d848 | 6584 | |
c89a8eee N |
6585 | if (mddev->ro) { |
6586 | /* Only thing we do on a ro array is remove | |
6587 | * failed devices. | |
6588 | */ | |
6589 | remove_and_add_spares(mddev); | |
6590 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6591 | goto unlock; | |
6592 | } | |
6593 | ||
31a59e34 | 6594 | if (!mddev->external) { |
0fd62b86 | 6595 | int did_change = 0; |
31a59e34 N |
6596 | spin_lock_irq(&mddev->write_lock); |
6597 | if (mddev->safemode && | |
6598 | !atomic_read(&mddev->writes_pending) && | |
6599 | !mddev->in_sync && | |
6600 | mddev->recovery_cp == MaxSector) { | |
6601 | mddev->in_sync = 1; | |
0fd62b86 | 6602 | did_change = 1; |
31a59e34 N |
6603 | if (mddev->persistent) |
6604 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6605 | } | |
6606 | if (mddev->safemode == 1) | |
6607 | mddev->safemode = 0; | |
6608 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 6609 | if (did_change) |
b62b7590 | 6610 | sysfs_notify_dirent(mddev->sysfs_state); |
fca4d848 | 6611 | } |
fca4d848 | 6612 | |
850b2b42 N |
6613 | if (mddev->flags) |
6614 | md_update_sb(mddev, 0); | |
06d91a5f | 6615 | |
159ec1fc | 6616 | list_for_each_entry(rdev, &mddev->disks, same_set) |
52664732 | 6617 | if (test_and_clear_bit(StateChanged, &rdev->flags)) |
3c0ee63a | 6618 | sysfs_notify_dirent(rdev->sysfs_state); |
52664732 | 6619 | |
06d91a5f | 6620 | |
1da177e4 LT |
6621 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
6622 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
6623 | /* resync/recovery still happening */ | |
6624 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6625 | goto unlock; | |
6626 | } | |
6627 | if (mddev->sync_thread) { | |
6628 | /* resync has finished, collect result */ | |
6629 | md_unregister_thread(mddev->sync_thread); | |
6630 | mddev->sync_thread = NULL; | |
56ac36d7 DW |
6631 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
6632 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1da177e4 LT |
6633 | /* success...*/ |
6634 | /* activate any spares */ | |
a99ac971 NB |
6635 | if (mddev->pers->spare_active(mddev)) |
6636 | sysfs_notify(&mddev->kobj, NULL, | |
6637 | "degraded"); | |
1da177e4 | 6638 | } |
cea9c228 N |
6639 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6640 | mddev->pers->finish_reshape) | |
6641 | mddev->pers->finish_reshape(mddev); | |
850b2b42 | 6642 | md_update_sb(mddev, 1); |
41158c7e N |
6643 | |
6644 | /* if array is no-longer degraded, then any saved_raid_disk | |
6645 | * information must be scrapped | |
6646 | */ | |
6647 | if (!mddev->degraded) | |
159ec1fc | 6648 | list_for_each_entry(rdev, &mddev->disks, same_set) |
41158c7e N |
6649 | rdev->saved_raid_disk = -1; |
6650 | ||
1da177e4 LT |
6651 | mddev->recovery = 0; |
6652 | /* flag recovery needed just to double check */ | |
6653 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
0c3573f1 | 6654 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 6655 | md_new_event(mddev); |
1da177e4 LT |
6656 | goto unlock; |
6657 | } | |
72a23c21 NB |
6658 | /* Set RUNNING before clearing NEEDED to avoid |
6659 | * any transients in the value of "sync_action". | |
6660 | */ | |
6661 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6662 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
6663 | /* Clear some bits that don't mean anything, but |
6664 | * might be left set | |
6665 | */ | |
24dd469d N |
6666 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6667 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 6668 | |
5fd6c1dc N |
6669 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6670 | goto unlock; | |
1da177e4 LT |
6671 | /* no recovery is running. |
6672 | * remove any failed drives, then | |
6673 | * add spares if possible. | |
6674 | * Spare are also removed and re-added, to allow | |
6675 | * the personality to fail the re-add. | |
6676 | */ | |
1da177e4 | 6677 | |
b4c4c7b8 | 6678 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
6679 | if (mddev->pers->check_reshape == NULL || |
6680 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
6681 | /* Cannot proceed */ |
6682 | goto unlock; | |
6683 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 6684 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 6685 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
6686 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
6687 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 6688 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 6689 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6690 | } else if (mddev->recovery_cp < MaxSector) { |
6691 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 6692 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6693 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
6694 | /* nothing to be done ... */ | |
1da177e4 | 6695 | goto unlock; |
24dd469d | 6696 | |
1da177e4 | 6697 | if (mddev->pers->sync_request) { |
a654b9d8 N |
6698 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
6699 | /* We are adding a device or devices to an array | |
6700 | * which has the bitmap stored on all devices. | |
6701 | * So make sure all bitmap pages get written | |
6702 | */ | |
6703 | bitmap_write_all(mddev->bitmap); | |
6704 | } | |
1da177e4 LT |
6705 | mddev->sync_thread = md_register_thread(md_do_sync, |
6706 | mddev, | |
6707 | "%s_resync"); | |
6708 | if (!mddev->sync_thread) { | |
6709 | printk(KERN_ERR "%s: could not start resync" | |
6710 | " thread...\n", | |
6711 | mdname(mddev)); | |
6712 | /* leave the spares where they are, it shouldn't hurt */ | |
6713 | mddev->recovery = 0; | |
d7603b7e | 6714 | } else |
1da177e4 | 6715 | md_wakeup_thread(mddev->sync_thread); |
0c3573f1 | 6716 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 6717 | md_new_event(mddev); |
1da177e4 LT |
6718 | } |
6719 | unlock: | |
72a23c21 NB |
6720 | if (!mddev->sync_thread) { |
6721 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6722 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
6723 | &mddev->recovery)) | |
0c3573f1 N |
6724 | if (mddev->sysfs_action) |
6725 | sysfs_notify_dirent(mddev->sysfs_action); | |
72a23c21 | 6726 | } |
1da177e4 LT |
6727 | mddev_unlock(mddev); |
6728 | } | |
6729 | } | |
6730 | ||
6bfe0b49 DW |
6731 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
6732 | { | |
3c0ee63a | 6733 | sysfs_notify_dirent(rdev->sysfs_state); |
6bfe0b49 DW |
6734 | wait_event_timeout(rdev->blocked_wait, |
6735 | !test_bit(Blocked, &rdev->flags), | |
6736 | msecs_to_jiffies(5000)); | |
6737 | rdev_dec_pending(rdev, mddev); | |
6738 | } | |
6739 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
6740 | ||
75c96f85 AB |
6741 | static int md_notify_reboot(struct notifier_block *this, |
6742 | unsigned long code, void *x) | |
1da177e4 LT |
6743 | { |
6744 | struct list_head *tmp; | |
6745 | mddev_t *mddev; | |
6746 | ||
6747 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
6748 | ||
6749 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
6750 | ||
29ac4aa3 | 6751 | for_each_mddev(mddev, tmp) |
c71d4887 | 6752 | if (mddev_trylock(mddev)) { |
2b25000b N |
6753 | /* Force a switch to readonly even array |
6754 | * appears to still be in use. Hence | |
6755 | * the '100'. | |
6756 | */ | |
d710e138 | 6757 | do_md_stop(mddev, 1, 100); |
c71d4887 NB |
6758 | mddev_unlock(mddev); |
6759 | } | |
1da177e4 LT |
6760 | /* |
6761 | * certain more exotic SCSI devices are known to be | |
6762 | * volatile wrt too early system reboots. While the | |
6763 | * right place to handle this issue is the given | |
6764 | * driver, we do want to have a safe RAID driver ... | |
6765 | */ | |
6766 | mdelay(1000*1); | |
6767 | } | |
6768 | return NOTIFY_DONE; | |
6769 | } | |
6770 | ||
75c96f85 | 6771 | static struct notifier_block md_notifier = { |
1da177e4 LT |
6772 | .notifier_call = md_notify_reboot, |
6773 | .next = NULL, | |
6774 | .priority = INT_MAX, /* before any real devices */ | |
6775 | }; | |
6776 | ||
6777 | static void md_geninit(void) | |
6778 | { | |
1da177e4 LT |
6779 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
6780 | ||
c7705f34 | 6781 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
6782 | } |
6783 | ||
75c96f85 | 6784 | static int __init md_init(void) |
1da177e4 | 6785 | { |
3dbd8c2e | 6786 | if (register_blkdev(MD_MAJOR, "md")) |
1da177e4 LT |
6787 | return -1; |
6788 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
3dbd8c2e | 6789 | unregister_blkdev(MD_MAJOR, "md"); |
1da177e4 LT |
6790 | return -1; |
6791 | } | |
3dbd8c2e | 6792 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
6793 | md_probe, NULL, NULL); |
6794 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
6795 | md_probe, NULL, NULL); |
6796 | ||
1da177e4 | 6797 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 6798 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
6799 | |
6800 | md_geninit(); | |
d710e138 | 6801 | return 0; |
1da177e4 LT |
6802 | } |
6803 | ||
6804 | ||
6805 | #ifndef MODULE | |
6806 | ||
6807 | /* | |
6808 | * Searches all registered partitions for autorun RAID arrays | |
6809 | * at boot time. | |
6810 | */ | |
4d936ec1 ME |
6811 | |
6812 | static LIST_HEAD(all_detected_devices); | |
6813 | struct detected_devices_node { | |
6814 | struct list_head list; | |
6815 | dev_t dev; | |
6816 | }; | |
1da177e4 LT |
6817 | |
6818 | void md_autodetect_dev(dev_t dev) | |
6819 | { | |
4d936ec1 ME |
6820 | struct detected_devices_node *node_detected_dev; |
6821 | ||
6822 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
6823 | if (node_detected_dev) { | |
6824 | node_detected_dev->dev = dev; | |
6825 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
6826 | } else { | |
6827 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
6828 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
6829 | } | |
1da177e4 LT |
6830 | } |
6831 | ||
6832 | ||
6833 | static void autostart_arrays(int part) | |
6834 | { | |
6835 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
6836 | struct detected_devices_node *node_detected_dev; |
6837 | dev_t dev; | |
6838 | int i_scanned, i_passed; | |
1da177e4 | 6839 | |
4d936ec1 ME |
6840 | i_scanned = 0; |
6841 | i_passed = 0; | |
1da177e4 | 6842 | |
4d936ec1 | 6843 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 6844 | |
4d936ec1 ME |
6845 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
6846 | i_scanned++; | |
6847 | node_detected_dev = list_entry(all_detected_devices.next, | |
6848 | struct detected_devices_node, list); | |
6849 | list_del(&node_detected_dev->list); | |
6850 | dev = node_detected_dev->dev; | |
6851 | kfree(node_detected_dev); | |
df968c4e | 6852 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
6853 | if (IS_ERR(rdev)) |
6854 | continue; | |
6855 | ||
b2d444d7 | 6856 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6857 | MD_BUG(); |
6858 | continue; | |
6859 | } | |
d0fae18f | 6860 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 6861 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 6862 | i_passed++; |
1da177e4 | 6863 | } |
4d936ec1 ME |
6864 | |
6865 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
6866 | i_scanned, i_passed); | |
1da177e4 LT |
6867 | |
6868 | autorun_devices(part); | |
6869 | } | |
6870 | ||
fdee8ae4 | 6871 | #endif /* !MODULE */ |
1da177e4 LT |
6872 | |
6873 | static __exit void md_exit(void) | |
6874 | { | |
6875 | mddev_t *mddev; | |
6876 | struct list_head *tmp; | |
8ab5e4c1 | 6877 | |
3dbd8c2e | 6878 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 6879 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 6880 | |
3dbd8c2e | 6881 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
6882 | unregister_blkdev(mdp_major, "mdp"); |
6883 | unregister_reboot_notifier(&md_notifier); | |
6884 | unregister_sysctl_table(raid_table_header); | |
6885 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 6886 | for_each_mddev(mddev, tmp) { |
1da177e4 | 6887 | export_array(mddev); |
d3374825 | 6888 | mddev->hold_active = 0; |
1da177e4 LT |
6889 | } |
6890 | } | |
6891 | ||
685784aa | 6892 | subsys_initcall(md_init); |
1da177e4 LT |
6893 | module_exit(md_exit) |
6894 | ||
f91de92e N |
6895 | static int get_ro(char *buffer, struct kernel_param *kp) |
6896 | { | |
6897 | return sprintf(buffer, "%d", start_readonly); | |
6898 | } | |
6899 | static int set_ro(const char *val, struct kernel_param *kp) | |
6900 | { | |
6901 | char *e; | |
6902 | int num = simple_strtoul(val, &e, 10); | |
6903 | if (*val && (*e == '\0' || *e == '\n')) { | |
6904 | start_readonly = num; | |
4dbcdc75 | 6905 | return 0; |
f91de92e N |
6906 | } |
6907 | return -EINVAL; | |
6908 | } | |
6909 | ||
80ca3a44 N |
6910 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
6911 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 6912 | |
efeb53c0 | 6913 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 6914 | |
1da177e4 LT |
6915 | EXPORT_SYMBOL(register_md_personality); |
6916 | EXPORT_SYMBOL(unregister_md_personality); | |
6917 | EXPORT_SYMBOL(md_error); | |
6918 | EXPORT_SYMBOL(md_done_sync); | |
6919 | EXPORT_SYMBOL(md_write_start); | |
6920 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
6921 | EXPORT_SYMBOL(md_register_thread); |
6922 | EXPORT_SYMBOL(md_unregister_thread); | |
6923 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
6924 | EXPORT_SYMBOL(md_check_recovery); |
6925 | MODULE_LICENSE("GPL"); | |
aa1595e9 | 6926 | MODULE_ALIAS("md"); |
72008652 | 6927 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |