]>
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 | ||
83d5cde4 | 141 | static const struct block_device_operations md_fops; |
1da177e4 | 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 | ||
3fa841d7 N |
265 | int mddev_congested(mddev_t *mddev, int bits) |
266 | { | |
267 | return mddev->suspended; | |
268 | } | |
269 | EXPORT_SYMBOL(mddev_congested); | |
270 | ||
409c57f3 | 271 | |
1da177e4 LT |
272 | static inline mddev_t *mddev_get(mddev_t *mddev) |
273 | { | |
274 | atomic_inc(&mddev->active); | |
275 | return mddev; | |
276 | } | |
277 | ||
5fd3a17e | 278 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 279 | |
1da177e4 LT |
280 | static void mddev_put(mddev_t *mddev) |
281 | { | |
282 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
283 | return; | |
d3374825 N |
284 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
285 | !mddev->hold_active) { | |
1da177e4 | 286 | list_del(&mddev->all_mddevs); |
d3374825 N |
287 | if (mddev->gendisk) { |
288 | /* we did a probe so need to clean up. | |
289 | * Call schedule_work inside the spinlock | |
290 | * so that flush_scheduled_work() after | |
291 | * mddev_find will succeed in waiting for the | |
292 | * work to be done. | |
293 | */ | |
294 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
295 | schedule_work(&mddev->del_work); | |
296 | } else | |
297 | kfree(mddev); | |
298 | } | |
299 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
300 | } |
301 | ||
302 | static mddev_t * mddev_find(dev_t unit) | |
303 | { | |
304 | mddev_t *mddev, *new = NULL; | |
305 | ||
306 | retry: | |
307 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
308 | |
309 | if (unit) { | |
310 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
311 | if (mddev->unit == unit) { | |
312 | mddev_get(mddev); | |
313 | spin_unlock(&all_mddevs_lock); | |
314 | kfree(new); | |
315 | return mddev; | |
316 | } | |
317 | ||
318 | if (new) { | |
319 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 320 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
321 | new->hold_active = UNTIL_IOCTL; |
322 | return new; | |
1da177e4 | 323 | } |
efeb53c0 N |
324 | } else if (new) { |
325 | /* find an unused unit number */ | |
326 | static int next_minor = 512; | |
327 | int start = next_minor; | |
328 | int is_free = 0; | |
329 | int dev = 0; | |
330 | while (!is_free) { | |
331 | dev = MKDEV(MD_MAJOR, next_minor); | |
332 | next_minor++; | |
333 | if (next_minor > MINORMASK) | |
334 | next_minor = 0; | |
335 | if (next_minor == start) { | |
336 | /* Oh dear, all in use. */ | |
337 | spin_unlock(&all_mddevs_lock); | |
338 | kfree(new); | |
339 | return NULL; | |
340 | } | |
341 | ||
342 | is_free = 1; | |
343 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
344 | if (mddev->unit == dev) { | |
345 | is_free = 0; | |
346 | break; | |
347 | } | |
348 | } | |
349 | new->unit = dev; | |
350 | new->md_minor = MINOR(dev); | |
351 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
352 | list_add(&new->all_mddevs, &all_mddevs); |
353 | spin_unlock(&all_mddevs_lock); | |
354 | return new; | |
355 | } | |
356 | spin_unlock(&all_mddevs_lock); | |
357 | ||
9ffae0cf | 358 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
359 | if (!new) |
360 | return NULL; | |
361 | ||
1da177e4 LT |
362 | new->unit = unit; |
363 | if (MAJOR(unit) == MD_MAJOR) | |
364 | new->md_minor = MINOR(unit); | |
365 | else | |
366 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
367 | ||
c8c00a69 | 368 | mutex_init(&new->open_mutex); |
df5b89b3 | 369 | mutex_init(&new->reconfig_mutex); |
1da177e4 LT |
370 | INIT_LIST_HEAD(&new->disks); |
371 | INIT_LIST_HEAD(&new->all_mddevs); | |
372 | init_timer(&new->safemode_timer); | |
373 | atomic_set(&new->active, 1); | |
f2ea68cf | 374 | atomic_set(&new->openers, 0); |
409c57f3 | 375 | atomic_set(&new->active_io, 0); |
06d91a5f | 376 | spin_lock_init(&new->write_lock); |
3d310eb7 | 377 | init_waitqueue_head(&new->sb_wait); |
a6d8113a | 378 | init_waitqueue_head(&new->recovery_wait); |
08a02ecd | 379 | new->reshape_position = MaxSector; |
5e96ee65 | 380 | new->resync_min = 0; |
c6207277 | 381 | new->resync_max = MaxSector; |
d897dbf9 | 382 | new->level = LEVEL_NONE; |
1da177e4 | 383 | |
1da177e4 LT |
384 | goto retry; |
385 | } | |
386 | ||
387 | static inline int mddev_lock(mddev_t * mddev) | |
388 | { | |
df5b89b3 | 389 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
390 | } |
391 | ||
b522adcd DW |
392 | static inline int mddev_is_locked(mddev_t *mddev) |
393 | { | |
394 | return mutex_is_locked(&mddev->reconfig_mutex); | |
395 | } | |
396 | ||
1da177e4 LT |
397 | static inline int mddev_trylock(mddev_t * mddev) |
398 | { | |
df5b89b3 | 399 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
400 | } |
401 | ||
402 | static inline void mddev_unlock(mddev_t * mddev) | |
403 | { | |
df5b89b3 | 404 | mutex_unlock(&mddev->reconfig_mutex); |
1da177e4 | 405 | |
005eca5e | 406 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
407 | } |
408 | ||
2989ddbd | 409 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 410 | { |
159ec1fc | 411 | mdk_rdev_t *rdev; |
1da177e4 | 412 | |
159ec1fc | 413 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
414 | if (rdev->desc_nr == nr) |
415 | return rdev; | |
159ec1fc | 416 | |
1da177e4 LT |
417 | return NULL; |
418 | } | |
419 | ||
420 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
421 | { | |
1da177e4 LT |
422 | mdk_rdev_t *rdev; |
423 | ||
159ec1fc | 424 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
425 | if (rdev->bdev->bd_dev == dev) |
426 | return rdev; | |
159ec1fc | 427 | |
1da177e4 LT |
428 | return NULL; |
429 | } | |
430 | ||
d9d166c2 | 431 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
432 | { |
433 | struct mdk_personality *pers; | |
d9d166c2 N |
434 | list_for_each_entry(pers, &pers_list, list) { |
435 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 436 | return pers; |
d9d166c2 N |
437 | if (strcmp(pers->name, clevel)==0) |
438 | return pers; | |
439 | } | |
2604b703 N |
440 | return NULL; |
441 | } | |
442 | ||
b73df2d3 | 443 | /* return the offset of the super block in 512byte sectors */ |
77933d72 | 444 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 | 445 | { |
b73df2d3 AN |
446 | sector_t num_sectors = bdev->bd_inode->i_size / 512; |
447 | return MD_NEW_SIZE_SECTORS(num_sectors); | |
1da177e4 LT |
448 | } |
449 | ||
1da177e4 LT |
450 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
451 | { | |
452 | if (rdev->sb_page) | |
453 | MD_BUG(); | |
454 | ||
455 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
456 | if (!rdev->sb_page) { | |
457 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 458 | return -ENOMEM; |
1da177e4 LT |
459 | } |
460 | ||
461 | return 0; | |
462 | } | |
463 | ||
464 | static void free_disk_sb(mdk_rdev_t * rdev) | |
465 | { | |
466 | if (rdev->sb_page) { | |
2d1f3b5d | 467 | put_page(rdev->sb_page); |
1da177e4 LT |
468 | rdev->sb_loaded = 0; |
469 | rdev->sb_page = NULL; | |
0f420358 | 470 | rdev->sb_start = 0; |
dd8ac336 | 471 | rdev->sectors = 0; |
1da177e4 LT |
472 | } |
473 | } | |
474 | ||
475 | ||
6712ecf8 | 476 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
477 | { |
478 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 479 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 480 | |
3a0f5bbb N |
481 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
482 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
483 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
484 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 485 | md_error(mddev, rdev); |
3a0f5bbb | 486 | } |
7bfa19f2 | 487 | |
a9701a30 N |
488 | if (atomic_dec_and_test(&mddev->pending_writes)) |
489 | wake_up(&mddev->sb_wait); | |
f8b58edf | 490 | bio_put(bio); |
7bfa19f2 N |
491 | } |
492 | ||
6712ecf8 | 493 | static void super_written_barrier(struct bio *bio, int error) |
a9701a30 N |
494 | { |
495 | struct bio *bio2 = bio->bi_private; | |
496 | mdk_rdev_t *rdev = bio2->bi_private; | |
497 | mddev_t *mddev = rdev->mddev; | |
a9701a30 N |
498 | |
499 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
500 | error == -EOPNOTSUPP) { | |
501 | unsigned long flags; | |
502 | /* barriers don't appear to be supported :-( */ | |
503 | set_bit(BarriersNotsupp, &rdev->flags); | |
504 | mddev->barriers_work = 0; | |
505 | spin_lock_irqsave(&mddev->write_lock, flags); | |
506 | bio2->bi_next = mddev->biolist; | |
507 | mddev->biolist = bio2; | |
508 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
509 | wake_up(&mddev->sb_wait); | |
510 | bio_put(bio); | |
6712ecf8 N |
511 | } else { |
512 | bio_put(bio2); | |
513 | bio->bi_private = rdev; | |
514 | super_written(bio, error); | |
a9701a30 | 515 | } |
a9701a30 N |
516 | } |
517 | ||
7bfa19f2 N |
518 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
519 | sector_t sector, int size, struct page *page) | |
520 | { | |
521 | /* write first size bytes of page to sector of rdev | |
522 | * Increment mddev->pending_writes before returning | |
523 | * and decrement it on completion, waking up sb_wait | |
524 | * if zero is reached. | |
525 | * If an error occurred, call md_error | |
a9701a30 N |
526 | * |
527 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
528 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
529 | */ |
530 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
93dbb393 | 531 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG); |
7bfa19f2 N |
532 | |
533 | bio->bi_bdev = rdev->bdev; | |
534 | bio->bi_sector = sector; | |
535 | bio_add_page(bio, page, size, 0); | |
536 | bio->bi_private = rdev; | |
537 | bio->bi_end_io = super_written; | |
a9701a30 N |
538 | bio->bi_rw = rw; |
539 | ||
7bfa19f2 | 540 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
541 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
542 | struct bio *rbio; | |
543 | rw |= (1<<BIO_RW_BARRIER); | |
544 | rbio = bio_clone(bio, GFP_NOIO); | |
545 | rbio->bi_private = bio; | |
546 | rbio->bi_end_io = super_written_barrier; | |
547 | submit_bio(rw, rbio); | |
548 | } else | |
549 | submit_bio(rw, bio); | |
550 | } | |
551 | ||
552 | void md_super_wait(mddev_t *mddev) | |
553 | { | |
554 | /* wait for all superblock writes that were scheduled to complete. | |
555 | * if any had to be retried (due to BARRIER problems), retry them | |
556 | */ | |
557 | DEFINE_WAIT(wq); | |
558 | for(;;) { | |
559 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
560 | if (atomic_read(&mddev->pending_writes)==0) | |
561 | break; | |
562 | while (mddev->biolist) { | |
563 | struct bio *bio; | |
564 | spin_lock_irq(&mddev->write_lock); | |
565 | bio = mddev->biolist; | |
566 | mddev->biolist = bio->bi_next ; | |
567 | bio->bi_next = NULL; | |
568 | spin_unlock_irq(&mddev->write_lock); | |
569 | submit_bio(bio->bi_rw, bio); | |
570 | } | |
571 | schedule(); | |
572 | } | |
573 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
574 | } |
575 | ||
6712ecf8 | 576 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 577 | { |
1da177e4 | 578 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
579 | } |
580 | ||
a654b9d8 | 581 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
582 | struct page *page, int rw) |
583 | { | |
baaa2c51 | 584 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
585 | struct completion event; |
586 | int ret; | |
587 | ||
93dbb393 | 588 | rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG); |
1da177e4 LT |
589 | |
590 | bio->bi_bdev = bdev; | |
591 | bio->bi_sector = sector; | |
592 | bio_add_page(bio, page, size, 0); | |
593 | init_completion(&event); | |
594 | bio->bi_private = &event; | |
595 | bio->bi_end_io = bi_complete; | |
596 | submit_bio(rw, bio); | |
597 | wait_for_completion(&event); | |
598 | ||
599 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
600 | bio_put(bio); | |
601 | return ret; | |
602 | } | |
a8745db2 | 603 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 604 | |
0002b271 | 605 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
606 | { |
607 | char b[BDEVNAME_SIZE]; | |
608 | if (!rdev->sb_page) { | |
609 | MD_BUG(); | |
610 | return -EINVAL; | |
611 | } | |
612 | if (rdev->sb_loaded) | |
613 | return 0; | |
614 | ||
615 | ||
0f420358 | 616 | if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ)) |
1da177e4 LT |
617 | goto fail; |
618 | rdev->sb_loaded = 1; | |
619 | return 0; | |
620 | ||
621 | fail: | |
622 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
623 | bdevname(rdev->bdev,b)); | |
624 | return -EINVAL; | |
625 | } | |
626 | ||
627 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
628 | { | |
05710466 AN |
629 | return sb1->set_uuid0 == sb2->set_uuid0 && |
630 | sb1->set_uuid1 == sb2->set_uuid1 && | |
631 | sb1->set_uuid2 == sb2->set_uuid2 && | |
632 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
633 | } |
634 | ||
1da177e4 LT |
635 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
636 | { | |
637 | int ret; | |
638 | mdp_super_t *tmp1, *tmp2; | |
639 | ||
640 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
641 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
642 | ||
643 | if (!tmp1 || !tmp2) { | |
644 | ret = 0; | |
35020f1a | 645 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
646 | goto abort; |
647 | } | |
648 | ||
649 | *tmp1 = *sb1; | |
650 | *tmp2 = *sb2; | |
651 | ||
652 | /* | |
653 | * nr_disks is not constant | |
654 | */ | |
655 | tmp1->nr_disks = 0; | |
656 | tmp2->nr_disks = 0; | |
657 | ||
ce0c8e05 | 658 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 659 | abort: |
990a8baf JJ |
660 | kfree(tmp1); |
661 | kfree(tmp2); | |
1da177e4 LT |
662 | return ret; |
663 | } | |
664 | ||
4d167f09 N |
665 | |
666 | static u32 md_csum_fold(u32 csum) | |
667 | { | |
668 | csum = (csum & 0xffff) + (csum >> 16); | |
669 | return (csum & 0xffff) + (csum >> 16); | |
670 | } | |
671 | ||
1da177e4 LT |
672 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
673 | { | |
4d167f09 N |
674 | u64 newcsum = 0; |
675 | u32 *sb32 = (u32*)sb; | |
676 | int i; | |
1da177e4 LT |
677 | unsigned int disk_csum, csum; |
678 | ||
679 | disk_csum = sb->sb_csum; | |
680 | sb->sb_csum = 0; | |
4d167f09 N |
681 | |
682 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
683 | newcsum += sb32[i]; | |
684 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
685 | ||
686 | ||
687 | #ifdef CONFIG_ALPHA | |
688 | /* This used to use csum_partial, which was wrong for several | |
689 | * reasons including that different results are returned on | |
690 | * different architectures. It isn't critical that we get exactly | |
691 | * the same return value as before (we always csum_fold before | |
692 | * testing, and that removes any differences). However as we | |
693 | * know that csum_partial always returned a 16bit value on | |
694 | * alphas, do a fold to maximise conformity to previous behaviour. | |
695 | */ | |
696 | sb->sb_csum = md_csum_fold(disk_csum); | |
697 | #else | |
1da177e4 | 698 | sb->sb_csum = disk_csum; |
4d167f09 | 699 | #endif |
1da177e4 LT |
700 | return csum; |
701 | } | |
702 | ||
703 | ||
704 | /* | |
705 | * Handle superblock details. | |
706 | * We want to be able to handle multiple superblock formats | |
707 | * so we have a common interface to them all, and an array of | |
708 | * different handlers. | |
709 | * We rely on user-space to write the initial superblock, and support | |
710 | * reading and updating of superblocks. | |
711 | * Interface methods are: | |
712 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
713 | * loads and validates a superblock on dev. | |
714 | * if refdev != NULL, compare superblocks on both devices | |
715 | * Return: | |
716 | * 0 - dev has a superblock that is compatible with refdev | |
717 | * 1 - dev has a superblock that is compatible and newer than refdev | |
718 | * so dev should be used as the refdev in future | |
719 | * -EINVAL superblock incompatible or invalid | |
720 | * -othererror e.g. -EIO | |
721 | * | |
722 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
723 | * Verify that dev is acceptable into mddev. | |
724 | * The first time, mddev->raid_disks will be 0, and data from | |
725 | * dev should be merged in. Subsequent calls check that dev | |
726 | * is new enough. Return 0 or -EINVAL | |
727 | * | |
728 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
729 | * Update the superblock for rdev with data in mddev | |
730 | * This does not write to disc. | |
731 | * | |
732 | */ | |
733 | ||
734 | struct super_type { | |
0cd17fec CW |
735 | char *name; |
736 | struct module *owner; | |
737 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
738 | int minor_version); | |
739 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
740 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
741 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 742 | sector_t num_sectors); |
1da177e4 LT |
743 | }; |
744 | ||
0894cc30 AN |
745 | /* |
746 | * Check that the given mddev has no bitmap. | |
747 | * | |
748 | * This function is called from the run method of all personalities that do not | |
749 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
750 | * has a bitmap. Otherwise, it returns 0. | |
751 | * | |
752 | */ | |
753 | int md_check_no_bitmap(mddev_t *mddev) | |
754 | { | |
755 | if (!mddev->bitmap_file && !mddev->bitmap_offset) | |
756 | return 0; | |
757 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
758 | mdname(mddev), mddev->pers->name); | |
759 | return 1; | |
760 | } | |
761 | EXPORT_SYMBOL(md_check_no_bitmap); | |
762 | ||
1da177e4 LT |
763 | /* |
764 | * load_super for 0.90.0 | |
765 | */ | |
766 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
767 | { | |
768 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
769 | mdp_super_t *sb; | |
770 | int ret; | |
1da177e4 LT |
771 | |
772 | /* | |
0f420358 | 773 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
774 | * it's at the end of the disk. |
775 | * | |
776 | * It also happens to be a multiple of 4Kb. | |
777 | */ | |
0f420358 | 778 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 779 | |
0002b271 | 780 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
781 | if (ret) return ret; |
782 | ||
783 | ret = -EINVAL; | |
784 | ||
785 | bdevname(rdev->bdev, b); | |
786 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
787 | ||
788 | if (sb->md_magic != MD_SB_MAGIC) { | |
789 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
790 | b); | |
791 | goto abort; | |
792 | } | |
793 | ||
794 | if (sb->major_version != 0 || | |
f6705578 N |
795 | sb->minor_version < 90 || |
796 | sb->minor_version > 91) { | |
1da177e4 LT |
797 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
798 | sb->major_version, sb->minor_version, | |
799 | b); | |
800 | goto abort; | |
801 | } | |
802 | ||
803 | if (sb->raid_disks <= 0) | |
804 | goto abort; | |
805 | ||
4d167f09 | 806 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
807 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
808 | b); | |
809 | goto abort; | |
810 | } | |
811 | ||
812 | rdev->preferred_minor = sb->md_minor; | |
813 | rdev->data_offset = 0; | |
0002b271 | 814 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
815 | |
816 | if (sb->level == LEVEL_MULTIPATH) | |
817 | rdev->desc_nr = -1; | |
818 | else | |
819 | rdev->desc_nr = sb->this_disk.number; | |
820 | ||
9a7b2b0f | 821 | if (!refdev) { |
1da177e4 | 822 | ret = 1; |
9a7b2b0f | 823 | } else { |
1da177e4 LT |
824 | __u64 ev1, ev2; |
825 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
826 | if (!uuid_equal(refsb, sb)) { | |
827 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
828 | b, bdevname(refdev->bdev,b2)); | |
829 | goto abort; | |
830 | } | |
831 | if (!sb_equal(refsb, sb)) { | |
832 | printk(KERN_WARNING "md: %s has same UUID" | |
833 | " but different superblock to %s\n", | |
834 | b, bdevname(refdev->bdev, b2)); | |
835 | goto abort; | |
836 | } | |
837 | ev1 = md_event(sb); | |
838 | ev2 = md_event(refsb); | |
839 | if (ev1 > ev2) | |
840 | ret = 1; | |
841 | else | |
842 | ret = 0; | |
843 | } | |
8190e754 | 844 | rdev->sectors = rdev->sb_start; |
1da177e4 | 845 | |
dd8ac336 | 846 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
847 | /* "this cannot possibly happen" ... */ |
848 | ret = -EINVAL; | |
849 | ||
1da177e4 LT |
850 | abort: |
851 | return ret; | |
852 | } | |
853 | ||
854 | /* | |
855 | * validate_super for 0.90.0 | |
856 | */ | |
857 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
858 | { | |
859 | mdp_disk_t *desc; | |
860 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 861 | __u64 ev1 = md_event(sb); |
1da177e4 | 862 | |
41158c7e | 863 | rdev->raid_disk = -1; |
c5d79adb N |
864 | clear_bit(Faulty, &rdev->flags); |
865 | clear_bit(In_sync, &rdev->flags); | |
866 | clear_bit(WriteMostly, &rdev->flags); | |
867 | clear_bit(BarriersNotsupp, &rdev->flags); | |
868 | ||
1da177e4 LT |
869 | if (mddev->raid_disks == 0) { |
870 | mddev->major_version = 0; | |
871 | mddev->minor_version = sb->minor_version; | |
872 | mddev->patch_version = sb->patch_version; | |
e691063a | 873 | mddev->external = 0; |
9d8f0363 | 874 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
875 | mddev->ctime = sb->ctime; |
876 | mddev->utime = sb->utime; | |
877 | mddev->level = sb->level; | |
d9d166c2 | 878 | mddev->clevel[0] = 0; |
1da177e4 LT |
879 | mddev->layout = sb->layout; |
880 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 881 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 882 | mddev->events = ev1; |
9223214e | 883 | mddev->bitmap_offset = 0; |
36fa3063 | 884 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
1da177e4 | 885 | |
f6705578 N |
886 | if (mddev->minor_version >= 91) { |
887 | mddev->reshape_position = sb->reshape_position; | |
888 | mddev->delta_disks = sb->delta_disks; | |
889 | mddev->new_level = sb->new_level; | |
890 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 891 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
892 | } else { |
893 | mddev->reshape_position = MaxSector; | |
894 | mddev->delta_disks = 0; | |
895 | mddev->new_level = mddev->level; | |
896 | mddev->new_layout = mddev->layout; | |
664e7c41 | 897 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
898 | } |
899 | ||
1da177e4 LT |
900 | if (sb->state & (1<<MD_SB_CLEAN)) |
901 | mddev->recovery_cp = MaxSector; | |
902 | else { | |
903 | if (sb->events_hi == sb->cp_events_hi && | |
904 | sb->events_lo == sb->cp_events_lo) { | |
905 | mddev->recovery_cp = sb->recovery_cp; | |
906 | } else | |
907 | mddev->recovery_cp = 0; | |
908 | } | |
909 | ||
910 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
911 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
912 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
913 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
914 | ||
915 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
916 | |
917 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
e11e93fa | 918 | mddev->bitmap_file == NULL) |
36fa3063 | 919 | mddev->bitmap_offset = mddev->default_bitmap_offset; |
a654b9d8 | 920 | |
41158c7e N |
921 | } else if (mddev->pers == NULL) { |
922 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
923 | ++ev1; |
924 | if (ev1 < mddev->events) | |
925 | return -EINVAL; | |
41158c7e N |
926 | } else if (mddev->bitmap) { |
927 | /* if adding to array with a bitmap, then we can accept an | |
928 | * older device ... but not too old. | |
929 | */ | |
41158c7e N |
930 | if (ev1 < mddev->bitmap->events_cleared) |
931 | return 0; | |
07d84d10 N |
932 | } else { |
933 | if (ev1 < mddev->events) | |
934 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
935 | return 0; | |
936 | } | |
41158c7e | 937 | |
1da177e4 | 938 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
939 | desc = sb->disks + rdev->desc_nr; |
940 | ||
941 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 942 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
943 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
944 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 945 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
946 | rdev->raid_disk = desc->raid_disk; |
947 | } | |
8ddf9efe N |
948 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
949 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 950 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 951 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
952 | return 0; |
953 | } | |
954 | ||
955 | /* | |
956 | * sync_super for 0.90.0 | |
957 | */ | |
958 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
959 | { | |
960 | mdp_super_t *sb; | |
1da177e4 LT |
961 | mdk_rdev_t *rdev2; |
962 | int next_spare = mddev->raid_disks; | |
19133a42 | 963 | |
1da177e4 LT |
964 | |
965 | /* make rdev->sb match mddev data.. | |
966 | * | |
967 | * 1/ zero out disks | |
968 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
969 | * 3/ any empty disks < next_spare become removed | |
970 | * | |
971 | * disks[0] gets initialised to REMOVED because | |
972 | * we cannot be sure from other fields if it has | |
973 | * been initialised or not. | |
974 | */ | |
975 | int i; | |
976 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
977 | ||
61181565 N |
978 | rdev->sb_size = MD_SB_BYTES; |
979 | ||
1da177e4 LT |
980 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
981 | ||
982 | memset(sb, 0, sizeof(*sb)); | |
983 | ||
984 | sb->md_magic = MD_SB_MAGIC; | |
985 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
986 | sb->patch_version = mddev->patch_version; |
987 | sb->gvalid_words = 0; /* ignored */ | |
988 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
989 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
990 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
991 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
992 | ||
993 | sb->ctime = mddev->ctime; | |
994 | sb->level = mddev->level; | |
58c0fed4 | 995 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
996 | sb->raid_disks = mddev->raid_disks; |
997 | sb->md_minor = mddev->md_minor; | |
e691063a | 998 | sb->not_persistent = 0; |
1da177e4 LT |
999 | sb->utime = mddev->utime; |
1000 | sb->state = 0; | |
1001 | sb->events_hi = (mddev->events>>32); | |
1002 | sb->events_lo = (u32)mddev->events; | |
1003 | ||
f6705578 N |
1004 | if (mddev->reshape_position == MaxSector) |
1005 | sb->minor_version = 90; | |
1006 | else { | |
1007 | sb->minor_version = 91; | |
1008 | sb->reshape_position = mddev->reshape_position; | |
1009 | sb->new_level = mddev->new_level; | |
1010 | sb->delta_disks = mddev->delta_disks; | |
1011 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1012 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1013 | } |
1014 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1015 | if (mddev->in_sync) |
1016 | { | |
1017 | sb->recovery_cp = mddev->recovery_cp; | |
1018 | sb->cp_events_hi = (mddev->events>>32); | |
1019 | sb->cp_events_lo = (u32)mddev->events; | |
1020 | if (mddev->recovery_cp == MaxSector) | |
1021 | sb->state = (1<< MD_SB_CLEAN); | |
1022 | } else | |
1023 | sb->recovery_cp = 0; | |
1024 | ||
1025 | sb->layout = mddev->layout; | |
9d8f0363 | 1026 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1027 | |
a654b9d8 N |
1028 | if (mddev->bitmap && mddev->bitmap_file == NULL) |
1029 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); | |
1030 | ||
1da177e4 | 1031 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1032 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1033 | mdp_disk_t *d; |
86e6ffdd | 1034 | int desc_nr; |
b2d444d7 N |
1035 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
1036 | && !test_bit(Faulty, &rdev2->flags)) | |
86e6ffdd | 1037 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1038 | else |
86e6ffdd | 1039 | desc_nr = next_spare++; |
19133a42 | 1040 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1041 | d = &sb->disks[rdev2->desc_nr]; |
1042 | nr_disks++; | |
1043 | d->number = rdev2->desc_nr; | |
1044 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1045 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
b2d444d7 N |
1046 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
1047 | && !test_bit(Faulty, &rdev2->flags)) | |
1da177e4 LT |
1048 | d->raid_disk = rdev2->raid_disk; |
1049 | else | |
1050 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1051 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1052 | d->state = (1<<MD_DISK_FAULTY); |
1be7892f | 1053 | else if (test_bit(In_sync, &rdev2->flags)) { |
1da177e4 LT |
1054 | d->state = (1<<MD_DISK_ACTIVE); |
1055 | d->state |= (1<<MD_DISK_SYNC); | |
1056 | active++; | |
1057 | working++; | |
1058 | } else { | |
1059 | d->state = 0; | |
1060 | spare++; | |
1061 | working++; | |
1062 | } | |
8ddf9efe N |
1063 | if (test_bit(WriteMostly, &rdev2->flags)) |
1064 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1065 | } |
1da177e4 LT |
1066 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1067 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1068 | mdp_disk_t *d = &sb->disks[i]; | |
1069 | if (d->state == 0 && d->number == 0) { | |
1070 | d->number = i; | |
1071 | d->raid_disk = i; | |
1072 | d->state = (1<<MD_DISK_REMOVED); | |
1073 | d->state |= (1<<MD_DISK_FAULTY); | |
1074 | failed++; | |
1075 | } | |
1076 | } | |
1077 | sb->nr_disks = nr_disks; | |
1078 | sb->active_disks = active; | |
1079 | sb->working_disks = working; | |
1080 | sb->failed_disks = failed; | |
1081 | sb->spare_disks = spare; | |
1082 | ||
1083 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1084 | sb->sb_csum = calc_sb_csum(sb); | |
1085 | } | |
1086 | ||
0cd17fec CW |
1087 | /* |
1088 | * rdev_size_change for 0.90.0 | |
1089 | */ | |
1090 | static unsigned long long | |
15f4a5fd | 1091 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1092 | { |
58c0fed4 | 1093 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1094 | return 0; /* component must fit device */ |
0cd17fec CW |
1095 | if (rdev->mddev->bitmap_offset) |
1096 | return 0; /* can't move bitmap */ | |
0f420358 | 1097 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
15f4a5fd AN |
1098 | if (!num_sectors || num_sectors > rdev->sb_start) |
1099 | num_sectors = rdev->sb_start; | |
0f420358 | 1100 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1101 | rdev->sb_page); |
1102 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1103 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec CW |
1104 | } |
1105 | ||
1106 | ||
1da177e4 LT |
1107 | /* |
1108 | * version 1 superblock | |
1109 | */ | |
1110 | ||
1c05b4bc | 1111 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1112 | { |
1c05b4bc N |
1113 | __le32 disk_csum; |
1114 | u32 csum; | |
1da177e4 LT |
1115 | unsigned long long newcsum; |
1116 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1117 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1118 | int i; |
1119 | ||
1120 | disk_csum = sb->sb_csum; | |
1121 | sb->sb_csum = 0; | |
1122 | newcsum = 0; | |
1123 | for (i=0; size>=4; size -= 4 ) | |
1124 | newcsum += le32_to_cpu(*isuper++); | |
1125 | ||
1126 | if (size == 2) | |
1c05b4bc | 1127 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1128 | |
1129 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1130 | sb->sb_csum = disk_csum; | |
1131 | return cpu_to_le32(csum); | |
1132 | } | |
1133 | ||
1134 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1135 | { | |
1136 | struct mdp_superblock_1 *sb; | |
1137 | int ret; | |
0f420358 | 1138 | sector_t sb_start; |
1da177e4 | 1139 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1140 | int bmask; |
1da177e4 LT |
1141 | |
1142 | /* | |
0f420358 | 1143 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1144 | * It is always aligned to a 4K boundary and |
1145 | * depeding on minor_version, it can be: | |
1146 | * 0: At least 8K, but less than 12K, from end of device | |
1147 | * 1: At start of device | |
1148 | * 2: 4K from start of device. | |
1149 | */ | |
1150 | switch(minor_version) { | |
1151 | case 0: | |
0f420358 AN |
1152 | sb_start = rdev->bdev->bd_inode->i_size >> 9; |
1153 | sb_start -= 8*2; | |
1154 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1155 | break; |
1156 | case 1: | |
0f420358 | 1157 | sb_start = 0; |
1da177e4 LT |
1158 | break; |
1159 | case 2: | |
0f420358 | 1160 | sb_start = 8; |
1da177e4 LT |
1161 | break; |
1162 | default: | |
1163 | return -EINVAL; | |
1164 | } | |
0f420358 | 1165 | rdev->sb_start = sb_start; |
1da177e4 | 1166 | |
0002b271 N |
1167 | /* superblock is rarely larger than 1K, but it can be larger, |
1168 | * and it is safe to read 4k, so we do that | |
1169 | */ | |
1170 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1171 | if (ret) return ret; |
1172 | ||
1173 | ||
1174 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1175 | ||
1176 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1177 | sb->major_version != cpu_to_le32(1) || | |
1178 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1179 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1180 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1181 | return -EINVAL; |
1182 | ||
1183 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1184 | printk("md: invalid superblock checksum on %s\n", | |
1185 | bdevname(rdev->bdev,b)); | |
1186 | return -EINVAL; | |
1187 | } | |
1188 | if (le64_to_cpu(sb->data_size) < 10) { | |
1189 | printk("md: data_size too small on %s\n", | |
1190 | bdevname(rdev->bdev,b)); | |
1191 | return -EINVAL; | |
1192 | } | |
e11e93fa | 1193 | |
1da177e4 LT |
1194 | rdev->preferred_minor = 0xffff; |
1195 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1196 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1197 | |
0002b271 | 1198 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1199 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1200 | if (rdev->sb_size & bmask) |
a1801f85 N |
1201 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1202 | ||
1203 | if (minor_version | |
0f420358 | 1204 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1205 | return -EINVAL; |
0002b271 | 1206 | |
31b65a0d N |
1207 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1208 | rdev->desc_nr = -1; | |
1209 | else | |
1210 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1211 | ||
9a7b2b0f | 1212 | if (!refdev) { |
8ed75463 | 1213 | ret = 1; |
9a7b2b0f | 1214 | } else { |
1da177e4 LT |
1215 | __u64 ev1, ev2; |
1216 | struct mdp_superblock_1 *refsb = | |
1217 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1218 | ||
1219 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1220 | sb->level != refsb->level || | |
1221 | sb->layout != refsb->layout || | |
1222 | sb->chunksize != refsb->chunksize) { | |
1223 | printk(KERN_WARNING "md: %s has strangely different" | |
1224 | " superblock to %s\n", | |
1225 | bdevname(rdev->bdev,b), | |
1226 | bdevname(refdev->bdev,b2)); | |
1227 | return -EINVAL; | |
1228 | } | |
1229 | ev1 = le64_to_cpu(sb->events); | |
1230 | ev2 = le64_to_cpu(refsb->events); | |
1231 | ||
1232 | if (ev1 > ev2) | |
8ed75463 N |
1233 | ret = 1; |
1234 | else | |
1235 | ret = 0; | |
1da177e4 | 1236 | } |
a1801f85 | 1237 | if (minor_version) |
dd8ac336 AN |
1238 | rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) - |
1239 | le64_to_cpu(sb->data_offset); | |
1da177e4 | 1240 | else |
dd8ac336 AN |
1241 | rdev->sectors = rdev->sb_start; |
1242 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1243 | return -EINVAL; |
dd8ac336 | 1244 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1245 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1246 | return -EINVAL; |
8ed75463 | 1247 | return ret; |
1da177e4 LT |
1248 | } |
1249 | ||
1250 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1251 | { | |
1252 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1253 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1254 | |
41158c7e | 1255 | rdev->raid_disk = -1; |
c5d79adb N |
1256 | clear_bit(Faulty, &rdev->flags); |
1257 | clear_bit(In_sync, &rdev->flags); | |
1258 | clear_bit(WriteMostly, &rdev->flags); | |
1259 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1260 | ||
1da177e4 LT |
1261 | if (mddev->raid_disks == 0) { |
1262 | mddev->major_version = 1; | |
1263 | mddev->patch_version = 0; | |
e691063a | 1264 | mddev->external = 0; |
9d8f0363 | 1265 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1266 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1267 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1268 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1269 | mddev->clevel[0] = 0; |
1da177e4 LT |
1270 | mddev->layout = le32_to_cpu(sb->layout); |
1271 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1272 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1273 | mddev->events = ev1; |
9223214e | 1274 | mddev->bitmap_offset = 0; |
29fc7e3e | 1275 | mddev->default_bitmap_offset = 1024 >> 9; |
1da177e4 LT |
1276 | |
1277 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1278 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1279 | ||
1280 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1281 | |
71c0805c | 1282 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
e11e93fa | 1283 | mddev->bitmap_file == NULL ) |
a654b9d8 | 1284 | mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); |
e11e93fa | 1285 | |
f6705578 N |
1286 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1287 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1288 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1289 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1290 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1291 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1292 | } else { |
1293 | mddev->reshape_position = MaxSector; | |
1294 | mddev->delta_disks = 0; | |
1295 | mddev->new_level = mddev->level; | |
1296 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1297 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1298 | } |
1299 | ||
41158c7e N |
1300 | } else if (mddev->pers == NULL) { |
1301 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1302 | ++ev1; |
1303 | if (ev1 < mddev->events) | |
1304 | return -EINVAL; | |
41158c7e N |
1305 | } else if (mddev->bitmap) { |
1306 | /* If adding to array with a bitmap, then we can accept an | |
1307 | * older device, but not too old. | |
1308 | */ | |
41158c7e N |
1309 | if (ev1 < mddev->bitmap->events_cleared) |
1310 | return 0; | |
07d84d10 N |
1311 | } else { |
1312 | if (ev1 < mddev->events) | |
1313 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1314 | return 0; | |
1315 | } | |
1da177e4 LT |
1316 | if (mddev->level != LEVEL_MULTIPATH) { |
1317 | int role; | |
3673f305 N |
1318 | if (rdev->desc_nr < 0 || |
1319 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1320 | role = 0xffff; | |
1321 | rdev->desc_nr = -1; | |
1322 | } else | |
1323 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1324 | switch(role) { |
1325 | case 0xffff: /* spare */ | |
1da177e4 LT |
1326 | break; |
1327 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1328 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1329 | break; |
1330 | default: | |
5fd6c1dc N |
1331 | if ((le32_to_cpu(sb->feature_map) & |
1332 | MD_FEATURE_RECOVERY_OFFSET)) | |
1333 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1334 | else | |
1335 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1336 | rdev->raid_disk = role; |
1337 | break; | |
1338 | } | |
8ddf9efe N |
1339 | if (sb->devflags & WriteMostly1) |
1340 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1341 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1342 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1343 | |
1da177e4 LT |
1344 | return 0; |
1345 | } | |
1346 | ||
1347 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1348 | { | |
1349 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1350 | mdk_rdev_t *rdev2; |
1351 | int max_dev, i; | |
1352 | /* make rdev->sb match mddev and rdev data. */ | |
1353 | ||
1354 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1355 | ||
1356 | sb->feature_map = 0; | |
1357 | sb->pad0 = 0; | |
5fd6c1dc | 1358 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1359 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1360 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1361 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1362 | ||
1363 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1364 | sb->events = cpu_to_le64(mddev->events); | |
1365 | if (mddev->in_sync) | |
1366 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1367 | else | |
1368 | sb->resync_offset = cpu_to_le64(0); | |
1369 | ||
1c05b4bc | 1370 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1371 | |
f0ca340c | 1372 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1373 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1374 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1375 | sb->level = cpu_to_le32(mddev->level); |
1376 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1377 | |
a654b9d8 N |
1378 | if (mddev->bitmap && mddev->bitmap_file == NULL) { |
1379 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); | |
71c0805c | 1380 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1381 | } |
5fd6c1dc N |
1382 | |
1383 | if (rdev->raid_disk >= 0 && | |
97e4f42d N |
1384 | !test_bit(In_sync, &rdev->flags)) { |
1385 | if (mddev->curr_resync_completed > rdev->recovery_offset) | |
1386 | rdev->recovery_offset = mddev->curr_resync_completed; | |
1387 | if (rdev->recovery_offset > 0) { | |
1388 | sb->feature_map |= | |
1389 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1390 | sb->recovery_offset = | |
1391 | cpu_to_le64(rdev->recovery_offset); | |
1392 | } | |
5fd6c1dc N |
1393 | } |
1394 | ||
f6705578 N |
1395 | if (mddev->reshape_position != MaxSector) { |
1396 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1397 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1398 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1399 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1400 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1401 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1402 | } |
a654b9d8 | 1403 | |
1da177e4 | 1404 | max_dev = 0; |
159ec1fc | 1405 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1406 | if (rdev2->desc_nr+1 > max_dev) |
1407 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1408 | |
70471daf N |
1409 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1410 | int bmask; | |
a778b73f | 1411 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1412 | rdev->sb_size = max_dev * 2 + 256; |
1413 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1414 | if (rdev->sb_size & bmask) | |
1415 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
1416 | } | |
1da177e4 LT |
1417 | for (i=0; i<max_dev;i++) |
1418 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1419 | ||
159ec1fc | 1420 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1421 | i = rdev2->desc_nr; |
b2d444d7 | 1422 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1423 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1424 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1425 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
5fd6c1dc N |
1426 | else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0) |
1427 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); | |
1da177e4 LT |
1428 | else |
1429 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1430 | } | |
1431 | ||
1da177e4 LT |
1432 | sb->sb_csum = calc_sb_1_csum(sb); |
1433 | } | |
1434 | ||
0cd17fec | 1435 | static unsigned long long |
15f4a5fd | 1436 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1437 | { |
1438 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1439 | sector_t max_sectors; |
58c0fed4 | 1440 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1441 | return 0; /* component must fit device */ |
0f420358 | 1442 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1443 | /* minor versions 1 and 2; superblock before data */ |
15f4a5fd AN |
1444 | max_sectors = rdev->bdev->bd_inode->i_size >> 9; |
1445 | max_sectors -= rdev->data_offset; | |
1446 | if (!num_sectors || num_sectors > max_sectors) | |
1447 | num_sectors = max_sectors; | |
0cd17fec CW |
1448 | } else if (rdev->mddev->bitmap_offset) { |
1449 | /* minor version 0 with bitmap we can't move */ | |
1450 | return 0; | |
1451 | } else { | |
1452 | /* minor version 0; superblock after data */ | |
0f420358 AN |
1453 | sector_t sb_start; |
1454 | sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; | |
1455 | sb_start &= ~(sector_t)(4*2 - 1); | |
dd8ac336 | 1456 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1457 | if (!num_sectors || num_sectors > max_sectors) |
1458 | num_sectors = max_sectors; | |
0f420358 | 1459 | rdev->sb_start = sb_start; |
0cd17fec CW |
1460 | } |
1461 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1462 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1463 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1464 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1465 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1466 | rdev->sb_page); |
1467 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1468 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec | 1469 | } |
1da177e4 | 1470 | |
75c96f85 | 1471 | static struct super_type super_types[] = { |
1da177e4 LT |
1472 | [0] = { |
1473 | .name = "0.90.0", | |
1474 | .owner = THIS_MODULE, | |
0cd17fec CW |
1475 | .load_super = super_90_load, |
1476 | .validate_super = super_90_validate, | |
1477 | .sync_super = super_90_sync, | |
1478 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1479 | }, |
1480 | [1] = { | |
1481 | .name = "md-1", | |
1482 | .owner = THIS_MODULE, | |
0cd17fec CW |
1483 | .load_super = super_1_load, |
1484 | .validate_super = super_1_validate, | |
1485 | .sync_super = super_1_sync, | |
1486 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1487 | }, |
1488 | }; | |
1da177e4 LT |
1489 | |
1490 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1491 | { | |
7dd5e7c3 | 1492 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1493 | |
4b80991c N |
1494 | rcu_read_lock(); |
1495 | rdev_for_each_rcu(rdev, mddev1) | |
1496 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1497 | if (rdev->bdev->bd_contains == |
4b80991c N |
1498 | rdev2->bdev->bd_contains) { |
1499 | rcu_read_unlock(); | |
7dd5e7c3 | 1500 | return 1; |
4b80991c N |
1501 | } |
1502 | rcu_read_unlock(); | |
1da177e4 LT |
1503 | return 0; |
1504 | } | |
1505 | ||
1506 | static LIST_HEAD(pending_raid_disks); | |
1507 | ||
ac5e7113 AN |
1508 | /* |
1509 | * Try to register data integrity profile for an mddev | |
1510 | * | |
1511 | * This is called when an array is started and after a disk has been kicked | |
1512 | * from the array. It only succeeds if all working and active component devices | |
1513 | * are integrity capable with matching profiles. | |
1514 | */ | |
1515 | int md_integrity_register(mddev_t *mddev) | |
1516 | { | |
1517 | mdk_rdev_t *rdev, *reference = NULL; | |
1518 | ||
1519 | if (list_empty(&mddev->disks)) | |
1520 | return 0; /* nothing to do */ | |
1521 | if (blk_get_integrity(mddev->gendisk)) | |
1522 | return 0; /* already registered */ | |
1523 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
1524 | /* skip spares and non-functional disks */ | |
1525 | if (test_bit(Faulty, &rdev->flags)) | |
1526 | continue; | |
1527 | if (rdev->raid_disk < 0) | |
1528 | continue; | |
1529 | /* | |
1530 | * If at least one rdev is not integrity capable, we can not | |
1531 | * enable data integrity for the md device. | |
1532 | */ | |
1533 | if (!bdev_get_integrity(rdev->bdev)) | |
1534 | return -EINVAL; | |
1535 | if (!reference) { | |
1536 | /* Use the first rdev as the reference */ | |
1537 | reference = rdev; | |
1538 | continue; | |
1539 | } | |
1540 | /* does this rdev's profile match the reference profile? */ | |
1541 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1542 | rdev->bdev->bd_disk) < 0) | |
1543 | return -EINVAL; | |
1544 | } | |
1545 | /* | |
1546 | * All component devices are integrity capable and have matching | |
1547 | * profiles, register the common profile for the md device. | |
1548 | */ | |
1549 | if (blk_integrity_register(mddev->gendisk, | |
1550 | bdev_get_integrity(reference->bdev)) != 0) { | |
1551 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1552 | mdname(mddev)); | |
1553 | return -EINVAL; | |
1554 | } | |
1555 | printk(KERN_NOTICE "md: data integrity on %s enabled\n", | |
1556 | mdname(mddev)); | |
1557 | return 0; | |
1558 | } | |
1559 | EXPORT_SYMBOL(md_integrity_register); | |
1560 | ||
1561 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
1562 | void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) | |
3f9d99c1 | 1563 | { |
3f9d99c1 | 1564 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); |
ac5e7113 | 1565 | struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); |
3f9d99c1 | 1566 | |
ac5e7113 | 1567 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1568 | return; |
ac5e7113 | 1569 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1570 | return; |
ac5e7113 AN |
1571 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1572 | rdev->bdev->bd_disk) >= 0) | |
1573 | return; | |
1574 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1575 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1576 | } |
ac5e7113 | 1577 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1578 | |
1da177e4 LT |
1579 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1580 | { | |
7dd5e7c3 | 1581 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1582 | struct kobject *ko; |
1edf80d3 | 1583 | char *s; |
5e55e2f5 | 1584 | int err; |
1da177e4 LT |
1585 | |
1586 | if (rdev->mddev) { | |
1587 | MD_BUG(); | |
1588 | return -EINVAL; | |
1589 | } | |
11e2ede0 DW |
1590 | |
1591 | /* prevent duplicates */ | |
1592 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1593 | return -EEXIST; | |
1594 | ||
dd8ac336 AN |
1595 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1596 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1597 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1598 | if (mddev->pers) { |
1599 | /* Cannot change size, so fail | |
1600 | * If mddev->level <= 0, then we don't care | |
1601 | * about aligning sizes (e.g. linear) | |
1602 | */ | |
1603 | if (mddev->level > 0) | |
1604 | return -ENOSPC; | |
1605 | } else | |
dd8ac336 | 1606 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1607 | } |
1da177e4 LT |
1608 | |
1609 | /* Verify rdev->desc_nr is unique. | |
1610 | * If it is -1, assign a free number, else | |
1611 | * check number is not in use | |
1612 | */ | |
1613 | if (rdev->desc_nr < 0) { | |
1614 | int choice = 0; | |
1615 | if (mddev->pers) choice = mddev->raid_disks; | |
1616 | while (find_rdev_nr(mddev, choice)) | |
1617 | choice++; | |
1618 | rdev->desc_nr = choice; | |
1619 | } else { | |
1620 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1621 | return -EBUSY; | |
1622 | } | |
de01dfad N |
1623 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1624 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1625 | mdname(mddev), mddev->max_disks); | |
1626 | return -EBUSY; | |
1627 | } | |
19133a42 | 1628 | bdevname(rdev->bdev,b); |
649316b2 | 1629 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1630 | *s = '!'; |
649316b2 | 1631 | |
1da177e4 | 1632 | rdev->mddev = mddev; |
19133a42 | 1633 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1634 | |
b2d6db58 | 1635 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1636 | goto fail; |
86e6ffdd | 1637 | |
0762b8bd | 1638 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
5e55e2f5 N |
1639 | if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) { |
1640 | kobject_del(&rdev->kobj); | |
1641 | goto fail; | |
1642 | } | |
3c0ee63a N |
1643 | rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state"); |
1644 | ||
4b80991c | 1645 | list_add_rcu(&rdev->same_set, &mddev->disks); |
c5d79adb | 1646 | bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); |
4044ba58 N |
1647 | |
1648 | /* May as well allow recovery to be retried once */ | |
1649 | mddev->recovery_disabled = 0; | |
3f9d99c1 | 1650 | |
1da177e4 | 1651 | return 0; |
5e55e2f5 N |
1652 | |
1653 | fail: | |
1654 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1655 | b, mdname(mddev)); | |
1656 | return err; | |
1da177e4 LT |
1657 | } |
1658 | ||
177a99b2 | 1659 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1660 | { |
1661 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1662 | kobject_del(&rdev->kobj); | |
177a99b2 | 1663 | kobject_put(&rdev->kobj); |
5792a285 N |
1664 | } |
1665 | ||
1da177e4 LT |
1666 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1667 | { | |
1668 | char b[BDEVNAME_SIZE]; | |
1669 | if (!rdev->mddev) { | |
1670 | MD_BUG(); | |
1671 | return; | |
1672 | } | |
5463c790 | 1673 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1674 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1675 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1676 | rdev->mddev = NULL; | |
86e6ffdd | 1677 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1678 | sysfs_put(rdev->sysfs_state); |
1679 | rdev->sysfs_state = NULL; | |
5792a285 | 1680 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1681 | * writing to 'remove' to "dev/state". We also need |
1682 | * to delay it due to rcu usage. | |
5792a285 | 1683 | */ |
4b80991c | 1684 | synchronize_rcu(); |
177a99b2 N |
1685 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1686 | kobject_get(&rdev->kobj); | |
5792a285 | 1687 | schedule_work(&rdev->del_work); |
1da177e4 LT |
1688 | } |
1689 | ||
1690 | /* | |
1691 | * prevent the device from being mounted, repartitioned or | |
1692 | * otherwise reused by a RAID array (or any other kernel | |
1693 | * subsystem), by bd_claiming the device. | |
1694 | */ | |
c5d79adb | 1695 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1696 | { |
1697 | int err = 0; | |
1698 | struct block_device *bdev; | |
1699 | char b[BDEVNAME_SIZE]; | |
1700 | ||
2e7b651d | 1701 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1702 | if (IS_ERR(bdev)) { |
1703 | printk(KERN_ERR "md: could not open %s.\n", | |
1704 | __bdevname(dev, b)); | |
1705 | return PTR_ERR(bdev); | |
1706 | } | |
c5d79adb | 1707 | err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); |
1da177e4 LT |
1708 | if (err) { |
1709 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1710 | bdevname(bdev, b)); | |
9a1c3542 | 1711 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1712 | return err; |
1713 | } | |
c5d79adb N |
1714 | if (!shared) |
1715 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1716 | rdev->bdev = bdev; |
1717 | return err; | |
1718 | } | |
1719 | ||
1720 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1721 | { | |
1722 | struct block_device *bdev = rdev->bdev; | |
1723 | rdev->bdev = NULL; | |
1724 | if (!bdev) | |
1725 | MD_BUG(); | |
1726 | bd_release(bdev); | |
9a1c3542 | 1727 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1728 | } |
1729 | ||
1730 | void md_autodetect_dev(dev_t dev); | |
1731 | ||
1732 | static void export_rdev(mdk_rdev_t * rdev) | |
1733 | { | |
1734 | char b[BDEVNAME_SIZE]; | |
1735 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1736 | bdevname(rdev->bdev,b)); | |
1737 | if (rdev->mddev) | |
1738 | MD_BUG(); | |
1739 | free_disk_sb(rdev); | |
1da177e4 | 1740 | #ifndef MODULE |
d0fae18f N |
1741 | if (test_bit(AutoDetected, &rdev->flags)) |
1742 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1743 | #endif |
1744 | unlock_rdev(rdev); | |
86e6ffdd | 1745 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1746 | } |
1747 | ||
1748 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1749 | { | |
1750 | unbind_rdev_from_array(rdev); | |
1751 | export_rdev(rdev); | |
1752 | } | |
1753 | ||
1754 | static void export_array(mddev_t *mddev) | |
1755 | { | |
159ec1fc | 1756 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 1757 | |
d089c6af | 1758 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1759 | if (!rdev->mddev) { |
1760 | MD_BUG(); | |
1761 | continue; | |
1762 | } | |
1763 | kick_rdev_from_array(rdev); | |
1764 | } | |
1765 | if (!list_empty(&mddev->disks)) | |
1766 | MD_BUG(); | |
1767 | mddev->raid_disks = 0; | |
1768 | mddev->major_version = 0; | |
1769 | } | |
1770 | ||
1771 | static void print_desc(mdp_disk_t *desc) | |
1772 | { | |
1773 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1774 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1775 | } | |
1776 | ||
cd2ac932 | 1777 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
1778 | { |
1779 | int i; | |
1780 | ||
1781 | printk(KERN_INFO | |
1782 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1783 | sb->major_version, sb->minor_version, sb->patch_version, | |
1784 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1785 | sb->ctime); | |
1786 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1787 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1788 | sb->md_minor, sb->layout, sb->chunk_size); | |
1789 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1790 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1791 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1792 | sb->failed_disks, sb->spare_disks, | |
1793 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1794 | ||
1795 | printk(KERN_INFO); | |
1796 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1797 | mdp_disk_t *desc; | |
1798 | ||
1799 | desc = sb->disks + i; | |
1800 | if (desc->number || desc->major || desc->minor || | |
1801 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1802 | printk(" D %2d: ", i); | |
1803 | print_desc(desc); | |
1804 | } | |
1805 | } | |
1806 | printk(KERN_INFO "md: THIS: "); | |
1807 | print_desc(&sb->this_disk); | |
cd2ac932 | 1808 | } |
1da177e4 | 1809 | |
cd2ac932 CR |
1810 | static void print_sb_1(struct mdp_superblock_1 *sb) |
1811 | { | |
1812 | __u8 *uuid; | |
1813 | ||
1814 | uuid = sb->set_uuid; | |
ad361c98 JP |
1815 | printk(KERN_INFO |
1816 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x" | |
1817 | ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n" | |
1818 | "md: Name: \"%s\" CT:%llu\n", | |
cd2ac932 CR |
1819 | le32_to_cpu(sb->major_version), |
1820 | le32_to_cpu(sb->feature_map), | |
1821 | uuid[0], uuid[1], uuid[2], uuid[3], | |
1822 | uuid[4], uuid[5], uuid[6], uuid[7], | |
1823 | uuid[8], uuid[9], uuid[10], uuid[11], | |
1824 | uuid[12], uuid[13], uuid[14], uuid[15], | |
1825 | sb->set_name, | |
1826 | (unsigned long long)le64_to_cpu(sb->ctime) | |
1827 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
1828 | ||
1829 | uuid = sb->device_uuid; | |
ad361c98 JP |
1830 | printk(KERN_INFO |
1831 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 1832 | " RO:%llu\n" |
ad361c98 JP |
1833 | "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x" |
1834 | ":%02x%02x%02x%02x%02x%02x\n" | |
1835 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" | |
1836 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
1837 | le32_to_cpu(sb->level), |
1838 | (unsigned long long)le64_to_cpu(sb->size), | |
1839 | le32_to_cpu(sb->raid_disks), | |
1840 | le32_to_cpu(sb->layout), | |
1841 | le32_to_cpu(sb->chunksize), | |
1842 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
1843 | (unsigned long long)le64_to_cpu(sb->data_size), | |
1844 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
1845 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
1846 | le32_to_cpu(sb->dev_number), | |
1847 | uuid[0], uuid[1], uuid[2], uuid[3], | |
1848 | uuid[4], uuid[5], uuid[6], uuid[7], | |
1849 | uuid[8], uuid[9], uuid[10], uuid[11], | |
1850 | uuid[12], uuid[13], uuid[14], uuid[15], | |
1851 | sb->devflags, | |
1852 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
1853 | (unsigned long long)le64_to_cpu(sb->events), | |
1854 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
1855 | le32_to_cpu(sb->sb_csum), | |
1856 | le32_to_cpu(sb->max_dev) | |
1857 | ); | |
1da177e4 LT |
1858 | } |
1859 | ||
cd2ac932 | 1860 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
1861 | { |
1862 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
1863 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
1864 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
1865 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
1866 | rdev->desc_nr); | |
1da177e4 | 1867 | if (rdev->sb_loaded) { |
cd2ac932 CR |
1868 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
1869 | switch (major_version) { | |
1870 | case 0: | |
1871 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
1872 | break; | |
1873 | case 1: | |
1874 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
1875 | break; | |
1876 | } | |
1da177e4 LT |
1877 | } else |
1878 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
1879 | } | |
1880 | ||
5e56341d | 1881 | static void md_print_devices(void) |
1da177e4 | 1882 | { |
159ec1fc | 1883 | struct list_head *tmp; |
1da177e4 LT |
1884 | mdk_rdev_t *rdev; |
1885 | mddev_t *mddev; | |
1886 | char b[BDEVNAME_SIZE]; | |
1887 | ||
1888 | printk("\n"); | |
1889 | printk("md: **********************************\n"); | |
1890 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
1891 | printk("md: **********************************\n"); | |
29ac4aa3 | 1892 | for_each_mddev(mddev, tmp) { |
1da177e4 | 1893 | |
32a7627c N |
1894 | if (mddev->bitmap) |
1895 | bitmap_print_sb(mddev->bitmap); | |
1896 | else | |
1897 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 1898 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
1899 | printk("<%s>", bdevname(rdev->bdev,b)); |
1900 | printk("\n"); | |
1901 | ||
159ec1fc | 1902 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 1903 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
1904 | } |
1905 | printk("md: **********************************\n"); | |
1906 | printk("\n"); | |
1907 | } | |
1908 | ||
1909 | ||
42543769 | 1910 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 1911 | { |
42543769 N |
1912 | /* Update each superblock (in-memory image), but |
1913 | * if we are allowed to, skip spares which already | |
1914 | * have the right event counter, or have one earlier | |
1915 | * (which would mean they aren't being marked as dirty | |
1916 | * with the rest of the array) | |
1917 | */ | |
1da177e4 | 1918 | mdk_rdev_t *rdev; |
1da177e4 | 1919 | |
159ec1fc | 1920 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
1921 | if (rdev->sb_events == mddev->events || |
1922 | (nospares && | |
1923 | rdev->raid_disk < 0 && | |
1924 | (rdev->sb_events&1)==0 && | |
1925 | rdev->sb_events+1 == mddev->events)) { | |
1926 | /* Don't update this superblock */ | |
1927 | rdev->sb_loaded = 2; | |
1928 | } else { | |
1929 | super_types[mddev->major_version]. | |
1930 | sync_super(mddev, rdev); | |
1931 | rdev->sb_loaded = 1; | |
1932 | } | |
1da177e4 LT |
1933 | } |
1934 | } | |
1935 | ||
850b2b42 | 1936 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 1937 | { |
1da177e4 | 1938 | mdk_rdev_t *rdev; |
06d91a5f | 1939 | int sync_req; |
42543769 | 1940 | int nospares = 0; |
1da177e4 | 1941 | |
1b57f132 | 1942 | mddev->utime = get_seconds(); |
8377bc80 N |
1943 | if (mddev->external) |
1944 | return; | |
1da177e4 | 1945 | repeat: |
a9701a30 | 1946 | spin_lock_irq(&mddev->write_lock); |
84692195 | 1947 | |
850b2b42 N |
1948 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
1949 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) | |
1950 | force_change = 1; | |
1951 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
1952 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
1953 | * though if events isn't the right even/odd, we will have to do | |
1954 | * spares after all | |
1955 | */ | |
1956 | nospares = 1; | |
1957 | if (force_change) | |
1958 | nospares = 0; | |
1959 | if (mddev->degraded) | |
84692195 N |
1960 | /* If the array is degraded, then skipping spares is both |
1961 | * dangerous and fairly pointless. | |
1962 | * Dangerous because a device that was removed from the array | |
1963 | * might have a event_count that still looks up-to-date, | |
1964 | * so it can be re-added without a resync. | |
1965 | * Pointless because if there are any spares to skip, | |
1966 | * then a recovery will happen and soon that array won't | |
1967 | * be degraded any more and the spare can go back to sleep then. | |
1968 | */ | |
850b2b42 | 1969 | nospares = 0; |
84692195 | 1970 | |
06d91a5f | 1971 | sync_req = mddev->in_sync; |
42543769 N |
1972 | |
1973 | /* If this is just a dirty<->clean transition, and the array is clean | |
1974 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 1975 | if (nospares |
42543769 | 1976 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
1031be7a N |
1977 | && (mddev->events & 1) |
1978 | && mddev->events != 1) | |
42543769 N |
1979 | mddev->events--; |
1980 | else { | |
1981 | /* otherwise we have to go forward and ... */ | |
1982 | mddev->events ++; | |
1983 | if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ | |
51d5668c N |
1984 | /* .. if the array isn't clean, an 'even' event must also go |
1985 | * to spares. */ | |
1986 | if ((mddev->events&1)==0) | |
42543769 | 1987 | nospares = 0; |
42543769 | 1988 | } else { |
51d5668c N |
1989 | /* otherwise an 'odd' event must go to spares */ |
1990 | if ((mddev->events&1)) | |
42543769 | 1991 | nospares = 0; |
42543769 N |
1992 | } |
1993 | } | |
1da177e4 LT |
1994 | |
1995 | if (!mddev->events) { | |
1996 | /* | |
1997 | * oops, this 64-bit counter should never wrap. | |
1998 | * Either we are in around ~1 trillion A.C., assuming | |
1999 | * 1 reboot per second, or we have a bug: | |
2000 | */ | |
2001 | MD_BUG(); | |
2002 | mddev->events --; | |
2003 | } | |
1da177e4 LT |
2004 | |
2005 | /* | |
2006 | * do not write anything to disk if using | |
2007 | * nonpersistent superblocks | |
2008 | */ | |
06d91a5f | 2009 | if (!mddev->persistent) { |
e691063a N |
2010 | if (!mddev->external) |
2011 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
2012 | ||
a9701a30 | 2013 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2014 | wake_up(&mddev->sb_wait); |
1da177e4 | 2015 | return; |
06d91a5f | 2016 | } |
e691063a | 2017 | sync_sbs(mddev, nospares); |
a9701a30 | 2018 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2019 | |
2020 | dprintk(KERN_INFO | |
2021 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2022 | mdname(mddev),mddev->in_sync); | |
2023 | ||
4ad13663 | 2024 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2025 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2026 | char b[BDEVNAME_SIZE]; |
2027 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2028 | if (rdev->sb_loaded != 1) |
2029 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2030 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2031 | dprintk("(skipping faulty "); |
2032 | ||
2033 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2034 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2035 | md_super_write(mddev,rdev, |
0f420358 | 2036 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2037 | rdev->sb_page); |
2038 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2039 | bdevname(rdev->bdev,b), | |
0f420358 | 2040 | (unsigned long long)rdev->sb_start); |
42543769 | 2041 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2042 | |
1da177e4 LT |
2043 | } else |
2044 | dprintk(")\n"); | |
7bfa19f2 | 2045 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2046 | /* only need to write one superblock... */ |
2047 | break; | |
2048 | } | |
a9701a30 | 2049 | md_super_wait(mddev); |
850b2b42 | 2050 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2051 | |
a9701a30 | 2052 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2053 | if (mddev->in_sync != sync_req || |
2054 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2055 | /* have to write it out again */ |
a9701a30 | 2056 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2057 | goto repeat; |
2058 | } | |
850b2b42 | 2059 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2060 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2061 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2062 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2063 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2064 | |
1da177e4 LT |
2065 | } |
2066 | ||
7f6ce769 | 2067 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2068 | * We want to accept with case. For this we use cmd_match. |
2069 | */ | |
2070 | static int cmd_match(const char *cmd, const char *str) | |
2071 | { | |
2072 | /* See if cmd, written into a sysfs file, matches | |
2073 | * str. They must either be the same, or cmd can | |
2074 | * have a trailing newline | |
2075 | */ | |
2076 | while (*cmd && *str && *cmd == *str) { | |
2077 | cmd++; | |
2078 | str++; | |
2079 | } | |
2080 | if (*cmd == '\n') | |
2081 | cmd++; | |
2082 | if (*str || *cmd) | |
2083 | return 0; | |
2084 | return 1; | |
2085 | } | |
2086 | ||
86e6ffdd N |
2087 | struct rdev_sysfs_entry { |
2088 | struct attribute attr; | |
2089 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2090 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2091 | }; | |
2092 | ||
2093 | static ssize_t | |
96de1e66 | 2094 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2095 | { |
2096 | char *sep = ""; | |
20a49ff6 | 2097 | size_t len = 0; |
86e6ffdd | 2098 | |
b2d444d7 | 2099 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2100 | len+= sprintf(page+len, "%sfaulty",sep); |
2101 | sep = ","; | |
2102 | } | |
b2d444d7 | 2103 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2104 | len += sprintf(page+len, "%sin_sync",sep); |
2105 | sep = ","; | |
2106 | } | |
f655675b N |
2107 | if (test_bit(WriteMostly, &rdev->flags)) { |
2108 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2109 | sep = ","; | |
2110 | } | |
6bfe0b49 DW |
2111 | if (test_bit(Blocked, &rdev->flags)) { |
2112 | len += sprintf(page+len, "%sblocked", sep); | |
2113 | sep = ","; | |
2114 | } | |
b2d444d7 N |
2115 | if (!test_bit(Faulty, &rdev->flags) && |
2116 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2117 | len += sprintf(page+len, "%sspare", sep); |
2118 | sep = ","; | |
2119 | } | |
2120 | return len+sprintf(page+len, "\n"); | |
2121 | } | |
2122 | ||
45dc2de1 N |
2123 | static ssize_t |
2124 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2125 | { | |
2126 | /* can write | |
2127 | * faulty - simulates and error | |
2128 | * remove - disconnects the device | |
f655675b N |
2129 | * writemostly - sets write_mostly |
2130 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2131 | * blocked - sets the Blocked flag |
2132 | * -blocked - clears the Blocked flag | |
6d56e278 | 2133 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2134 | */ |
2135 | int err = -EINVAL; | |
2136 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2137 | md_error(rdev->mddev, rdev); | |
2138 | err = 0; | |
2139 | } else if (cmd_match(buf, "remove")) { | |
2140 | if (rdev->raid_disk >= 0) | |
2141 | err = -EBUSY; | |
2142 | else { | |
2143 | mddev_t *mddev = rdev->mddev; | |
2144 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2145 | if (mddev->pers) |
2146 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2147 | md_new_event(mddev); |
2148 | err = 0; | |
2149 | } | |
f655675b N |
2150 | } else if (cmd_match(buf, "writemostly")) { |
2151 | set_bit(WriteMostly, &rdev->flags); | |
2152 | err = 0; | |
2153 | } else if (cmd_match(buf, "-writemostly")) { | |
2154 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2155 | err = 0; |
2156 | } else if (cmd_match(buf, "blocked")) { | |
2157 | set_bit(Blocked, &rdev->flags); | |
2158 | err = 0; | |
2159 | } else if (cmd_match(buf, "-blocked")) { | |
2160 | clear_bit(Blocked, &rdev->flags); | |
2161 | wake_up(&rdev->blocked_wait); | |
2162 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2163 | md_wakeup_thread(rdev->mddev->thread); | |
2164 | ||
6d56e278 N |
2165 | err = 0; |
2166 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2167 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2168 | err = 0; |
45dc2de1 | 2169 | } |
3c0ee63a N |
2170 | if (!err && rdev->sysfs_state) |
2171 | sysfs_notify_dirent(rdev->sysfs_state); | |
45dc2de1 N |
2172 | return err ? err : len; |
2173 | } | |
80ca3a44 N |
2174 | static struct rdev_sysfs_entry rdev_state = |
2175 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2176 | |
4dbcdc75 N |
2177 | static ssize_t |
2178 | errors_show(mdk_rdev_t *rdev, char *page) | |
2179 | { | |
2180 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2181 | } | |
2182 | ||
2183 | static ssize_t | |
2184 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2185 | { | |
2186 | char *e; | |
2187 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2188 | if (*buf && (*e == 0 || *e == '\n')) { | |
2189 | atomic_set(&rdev->corrected_errors, n); | |
2190 | return len; | |
2191 | } | |
2192 | return -EINVAL; | |
2193 | } | |
2194 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2195 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2196 | |
014236d2 N |
2197 | static ssize_t |
2198 | slot_show(mdk_rdev_t *rdev, char *page) | |
2199 | { | |
2200 | if (rdev->raid_disk < 0) | |
2201 | return sprintf(page, "none\n"); | |
2202 | else | |
2203 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2204 | } | |
2205 | ||
2206 | static ssize_t | |
2207 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2208 | { | |
2209 | char *e; | |
c303da6d N |
2210 | int err; |
2211 | char nm[20]; | |
014236d2 N |
2212 | int slot = simple_strtoul(buf, &e, 10); |
2213 | if (strncmp(buf, "none", 4)==0) | |
2214 | slot = -1; | |
2215 | else if (e==buf || (*e && *e!= '\n')) | |
2216 | return -EINVAL; | |
6c2fce2e | 2217 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2218 | /* Setting 'slot' on an active array requires also |
2219 | * updating the 'rd%d' link, and communicating | |
2220 | * with the personality with ->hot_*_disk. | |
2221 | * For now we only support removing | |
2222 | * failed/spare devices. This normally happens automatically, | |
2223 | * but not when the metadata is externally managed. | |
2224 | */ | |
c303da6d N |
2225 | if (rdev->raid_disk == -1) |
2226 | return -EEXIST; | |
2227 | /* personality does all needed checks */ | |
2228 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2229 | return -EINVAL; | |
2230 | err = rdev->mddev->pers-> | |
2231 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2232 | if (err) | |
2233 | return err; | |
2234 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2235 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
2236 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2237 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2238 | } else if (rdev->mddev->pers) { |
2239 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2240 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2241 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2242 | */ |
2243 | ||
2244 | if (rdev->raid_disk != -1) | |
2245 | return -EBUSY; | |
2246 | ||
2247 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2248 | return -EINVAL; | |
2249 | ||
159ec1fc | 2250 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2251 | if (rdev2->raid_disk == slot) |
2252 | return -EEXIST; | |
2253 | ||
2254 | rdev->raid_disk = slot; | |
2255 | if (test_bit(In_sync, &rdev->flags)) | |
2256 | rdev->saved_raid_disk = slot; | |
2257 | else | |
2258 | rdev->saved_raid_disk = -1; | |
2259 | err = rdev->mddev->pers-> | |
2260 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2261 | if (err) { |
6c2fce2e | 2262 | rdev->raid_disk = -1; |
6c2fce2e | 2263 | return err; |
52664732 | 2264 | } else |
3c0ee63a | 2265 | sysfs_notify_dirent(rdev->sysfs_state); |
6c2fce2e NB |
2266 | sprintf(nm, "rd%d", rdev->raid_disk); |
2267 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
2268 | printk(KERN_WARNING | |
2269 | "md: cannot register " | |
2270 | "%s for %s\n", | |
2271 | nm, mdname(rdev->mddev)); | |
2272 | ||
2273 | /* don't wakeup anyone, leave that to userspace. */ | |
c303da6d N |
2274 | } else { |
2275 | if (slot >= rdev->mddev->raid_disks) | |
2276 | return -ENOSPC; | |
2277 | rdev->raid_disk = slot; | |
2278 | /* assume it is working */ | |
c5d79adb N |
2279 | clear_bit(Faulty, &rdev->flags); |
2280 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2281 | set_bit(In_sync, &rdev->flags); |
3c0ee63a | 2282 | sysfs_notify_dirent(rdev->sysfs_state); |
c303da6d | 2283 | } |
014236d2 N |
2284 | return len; |
2285 | } | |
2286 | ||
2287 | ||
2288 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2289 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2290 | |
93c8cad0 N |
2291 | static ssize_t |
2292 | offset_show(mdk_rdev_t *rdev, char *page) | |
2293 | { | |
6961ece4 | 2294 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2295 | } |
2296 | ||
2297 | static ssize_t | |
2298 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2299 | { | |
2300 | char *e; | |
2301 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2302 | if (e==buf || (*e && *e != '\n')) | |
2303 | return -EINVAL; | |
8ed0a521 | 2304 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2305 | return -EBUSY; |
dd8ac336 | 2306 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2307 | /* Must set offset before size, so overlap checks |
2308 | * can be sane */ | |
2309 | return -EBUSY; | |
93c8cad0 N |
2310 | rdev->data_offset = offset; |
2311 | return len; | |
2312 | } | |
2313 | ||
2314 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2315 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2316 | |
83303b61 N |
2317 | static ssize_t |
2318 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2319 | { | |
dd8ac336 | 2320 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2321 | } |
2322 | ||
c5d79adb N |
2323 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2324 | { | |
2325 | /* check if two start/length pairs overlap */ | |
2326 | if (s1+l1 <= s2) | |
2327 | return 0; | |
2328 | if (s2+l2 <= s1) | |
2329 | return 0; | |
2330 | return 1; | |
2331 | } | |
2332 | ||
b522adcd DW |
2333 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2334 | { | |
2335 | unsigned long long blocks; | |
2336 | sector_t new; | |
2337 | ||
2338 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2339 | return -EINVAL; | |
2340 | ||
2341 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2342 | return -EINVAL; /* sector conversion overflow */ | |
2343 | ||
2344 | new = blocks * 2; | |
2345 | if (new != blocks * 2) | |
2346 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2347 | ||
2348 | *sectors = new; | |
2349 | return 0; | |
2350 | } | |
2351 | ||
83303b61 N |
2352 | static ssize_t |
2353 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2354 | { | |
27c529bb | 2355 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2356 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2357 | sector_t sectors; |
27c529bb | 2358 | |
b522adcd | 2359 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2360 | return -EINVAL; |
0cd17fec | 2361 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2362 | if (my_mddev->persistent) { |
dd8ac336 AN |
2363 | sectors = super_types[my_mddev->major_version]. |
2364 | rdev_size_change(rdev, sectors); | |
2365 | if (!sectors) | |
0cd17fec | 2366 | return -EBUSY; |
dd8ac336 AN |
2367 | } else if (!sectors) |
2368 | sectors = (rdev->bdev->bd_inode->i_size >> 9) - | |
2369 | rdev->data_offset; | |
0cd17fec | 2370 | } |
dd8ac336 | 2371 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2372 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2373 | |
dd8ac336 AN |
2374 | rdev->sectors = sectors; |
2375 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2376 | /* need to check that all other rdevs with the same ->bdev |
2377 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2378 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2379 | * we have to change it back, we will have the lock again. |
2380 | */ | |
2381 | mddev_t *mddev; | |
2382 | int overlap = 0; | |
159ec1fc | 2383 | struct list_head *tmp; |
c5d79adb | 2384 | |
27c529bb | 2385 | mddev_unlock(my_mddev); |
29ac4aa3 | 2386 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2387 | mdk_rdev_t *rdev2; |
2388 | ||
2389 | mddev_lock(mddev); | |
159ec1fc | 2390 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
c5d79adb N |
2391 | if (test_bit(AllReserved, &rdev2->flags) || |
2392 | (rdev->bdev == rdev2->bdev && | |
2393 | rdev != rdev2 && | |
dd8ac336 | 2394 | overlaps(rdev->data_offset, rdev->sectors, |
d07bd3bc | 2395 | rdev2->data_offset, |
dd8ac336 | 2396 | rdev2->sectors))) { |
c5d79adb N |
2397 | overlap = 1; |
2398 | break; | |
2399 | } | |
2400 | mddev_unlock(mddev); | |
2401 | if (overlap) { | |
2402 | mddev_put(mddev); | |
2403 | break; | |
2404 | } | |
2405 | } | |
27c529bb | 2406 | mddev_lock(my_mddev); |
c5d79adb N |
2407 | if (overlap) { |
2408 | /* Someone else could have slipped in a size | |
2409 | * change here, but doing so is just silly. | |
dd8ac336 | 2410 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2411 | * safe, and trust userspace not to race with |
2412 | * itself | |
2413 | */ | |
dd8ac336 | 2414 | rdev->sectors = oldsectors; |
c5d79adb N |
2415 | return -EBUSY; |
2416 | } | |
2417 | } | |
83303b61 N |
2418 | return len; |
2419 | } | |
2420 | ||
2421 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2422 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2423 | |
86e6ffdd N |
2424 | static struct attribute *rdev_default_attrs[] = { |
2425 | &rdev_state.attr, | |
4dbcdc75 | 2426 | &rdev_errors.attr, |
014236d2 | 2427 | &rdev_slot.attr, |
93c8cad0 | 2428 | &rdev_offset.attr, |
83303b61 | 2429 | &rdev_size.attr, |
86e6ffdd N |
2430 | NULL, |
2431 | }; | |
2432 | static ssize_t | |
2433 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2434 | { | |
2435 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2436 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2437 | mddev_t *mddev = rdev->mddev; |
2438 | ssize_t rv; | |
86e6ffdd N |
2439 | |
2440 | if (!entry->show) | |
2441 | return -EIO; | |
27c529bb N |
2442 | |
2443 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2444 | if (!rv) { | |
2445 | if (rdev->mddev == NULL) | |
2446 | rv = -EBUSY; | |
2447 | else | |
2448 | rv = entry->show(rdev, page); | |
2449 | mddev_unlock(mddev); | |
2450 | } | |
2451 | return rv; | |
86e6ffdd N |
2452 | } |
2453 | ||
2454 | static ssize_t | |
2455 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2456 | const char *page, size_t length) | |
2457 | { | |
2458 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2459 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2460 | ssize_t rv; |
2461 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2462 | |
2463 | if (!entry->store) | |
2464 | return -EIO; | |
67463acb N |
2465 | if (!capable(CAP_SYS_ADMIN)) |
2466 | return -EACCES; | |
27c529bb | 2467 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2468 | if (!rv) { |
27c529bb N |
2469 | if (rdev->mddev == NULL) |
2470 | rv = -EBUSY; | |
2471 | else | |
2472 | rv = entry->store(rdev, page, length); | |
6a51830e | 2473 | mddev_unlock(mddev); |
ca388059 N |
2474 | } |
2475 | return rv; | |
86e6ffdd N |
2476 | } |
2477 | ||
2478 | static void rdev_free(struct kobject *ko) | |
2479 | { | |
2480 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2481 | kfree(rdev); | |
2482 | } | |
2483 | static struct sysfs_ops rdev_sysfs_ops = { | |
2484 | .show = rdev_attr_show, | |
2485 | .store = rdev_attr_store, | |
2486 | }; | |
2487 | static struct kobj_type rdev_ktype = { | |
2488 | .release = rdev_free, | |
2489 | .sysfs_ops = &rdev_sysfs_ops, | |
2490 | .default_attrs = rdev_default_attrs, | |
2491 | }; | |
2492 | ||
1da177e4 LT |
2493 | /* |
2494 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2495 | * | |
2496 | * mark the device faulty if: | |
2497 | * | |
2498 | * - the device is nonexistent (zero size) | |
2499 | * - the device has no valid superblock | |
2500 | * | |
2501 | * a faulty rdev _never_ has rdev->sb set. | |
2502 | */ | |
2503 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2504 | { | |
2505 | char b[BDEVNAME_SIZE]; | |
2506 | int err; | |
2507 | mdk_rdev_t *rdev; | |
2508 | sector_t size; | |
2509 | ||
9ffae0cf | 2510 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2511 | if (!rdev) { |
2512 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2513 | return ERR_PTR(-ENOMEM); | |
2514 | } | |
1da177e4 LT |
2515 | |
2516 | if ((err = alloc_disk_sb(rdev))) | |
2517 | goto abort_free; | |
2518 | ||
c5d79adb | 2519 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2520 | if (err) |
2521 | goto abort_free; | |
2522 | ||
f9cb074b | 2523 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2524 | |
1da177e4 | 2525 | rdev->desc_nr = -1; |
2b6e8459 | 2526 | rdev->saved_raid_disk = -1; |
3f9d7b0d | 2527 | rdev->raid_disk = -1; |
b2d444d7 | 2528 | rdev->flags = 0; |
1da177e4 | 2529 | rdev->data_offset = 0; |
42543769 | 2530 | rdev->sb_events = 0; |
1da177e4 | 2531 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 2532 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 2533 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
2534 | |
2535 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
2536 | if (!size) { | |
2537 | printk(KERN_WARNING | |
2538 | "md: %s has zero or unknown size, marking faulty!\n", | |
2539 | bdevname(rdev->bdev,b)); | |
2540 | err = -EINVAL; | |
2541 | goto abort_free; | |
2542 | } | |
2543 | ||
2544 | if (super_format >= 0) { | |
2545 | err = super_types[super_format]. | |
2546 | load_super(rdev, NULL, super_minor); | |
2547 | if (err == -EINVAL) { | |
df968c4e N |
2548 | printk(KERN_WARNING |
2549 | "md: %s does not have a valid v%d.%d " | |
2550 | "superblock, not importing!\n", | |
2551 | bdevname(rdev->bdev,b), | |
2552 | super_format, super_minor); | |
1da177e4 LT |
2553 | goto abort_free; |
2554 | } | |
2555 | if (err < 0) { | |
2556 | printk(KERN_WARNING | |
2557 | "md: could not read %s's sb, not importing!\n", | |
2558 | bdevname(rdev->bdev,b)); | |
2559 | goto abort_free; | |
2560 | } | |
2561 | } | |
6bfe0b49 | 2562 | |
1da177e4 | 2563 | INIT_LIST_HEAD(&rdev->same_set); |
6bfe0b49 | 2564 | init_waitqueue_head(&rdev->blocked_wait); |
1da177e4 LT |
2565 | |
2566 | return rdev; | |
2567 | ||
2568 | abort_free: | |
2569 | if (rdev->sb_page) { | |
2570 | if (rdev->bdev) | |
2571 | unlock_rdev(rdev); | |
2572 | free_disk_sb(rdev); | |
2573 | } | |
2574 | kfree(rdev); | |
2575 | return ERR_PTR(err); | |
2576 | } | |
2577 | ||
2578 | /* | |
2579 | * Check a full RAID array for plausibility | |
2580 | */ | |
2581 | ||
2582 | ||
a757e64c | 2583 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2584 | { |
2585 | int i; | |
159ec1fc | 2586 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2587 | char b[BDEVNAME_SIZE]; |
2588 | ||
2589 | freshest = NULL; | |
d089c6af | 2590 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2591 | switch (super_types[mddev->major_version]. |
2592 | load_super(rdev, freshest, mddev->minor_version)) { | |
2593 | case 1: | |
2594 | freshest = rdev; | |
2595 | break; | |
2596 | case 0: | |
2597 | break; | |
2598 | default: | |
2599 | printk( KERN_ERR \ | |
2600 | "md: fatal superblock inconsistency in %s" | |
2601 | " -- removing from array\n", | |
2602 | bdevname(rdev->bdev,b)); | |
2603 | kick_rdev_from_array(rdev); | |
2604 | } | |
2605 | ||
2606 | ||
2607 | super_types[mddev->major_version]. | |
2608 | validate_super(mddev, freshest); | |
2609 | ||
2610 | i = 0; | |
d089c6af | 2611 | rdev_for_each(rdev, tmp, mddev) { |
de01dfad N |
2612 | if (rdev->desc_nr >= mddev->max_disks || |
2613 | i > mddev->max_disks) { | |
2614 | printk(KERN_WARNING | |
2615 | "md: %s: %s: only %d devices permitted\n", | |
2616 | mdname(mddev), bdevname(rdev->bdev, b), | |
2617 | mddev->max_disks); | |
2618 | kick_rdev_from_array(rdev); | |
2619 | continue; | |
2620 | } | |
1da177e4 LT |
2621 | if (rdev != freshest) |
2622 | if (super_types[mddev->major_version]. | |
2623 | validate_super(mddev, rdev)) { | |
2624 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2625 | " from array!\n", | |
2626 | bdevname(rdev->bdev,b)); | |
2627 | kick_rdev_from_array(rdev); | |
2628 | continue; | |
2629 | } | |
2630 | if (mddev->level == LEVEL_MULTIPATH) { | |
2631 | rdev->desc_nr = i++; | |
2632 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2633 | set_bit(In_sync, &rdev->flags); |
a778b73f N |
2634 | } else if (rdev->raid_disk >= mddev->raid_disks) { |
2635 | rdev->raid_disk = -1; | |
2636 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2637 | } |
2638 | } | |
1da177e4 LT |
2639 | } |
2640 | ||
19052c0e N |
2641 | static void md_safemode_timeout(unsigned long data); |
2642 | ||
16f17b39 N |
2643 | static ssize_t |
2644 | safe_delay_show(mddev_t *mddev, char *page) | |
2645 | { | |
2646 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2647 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2648 | } | |
2649 | static ssize_t | |
2650 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2651 | { | |
2652 | int scale=1; | |
2653 | int dot=0; | |
2654 | int i; | |
2655 | unsigned long msec; | |
2656 | char buf[30]; | |
97ce0a7f | 2657 | |
16f17b39 N |
2658 | /* remove a period, and count digits after it */ |
2659 | if (len >= sizeof(buf)) | |
2660 | return -EINVAL; | |
97ce0a7f | 2661 | strlcpy(buf, cbuf, sizeof(buf)); |
16f17b39 N |
2662 | for (i=0; i<len; i++) { |
2663 | if (dot) { | |
2664 | if (isdigit(buf[i])) { | |
2665 | buf[i-1] = buf[i]; | |
2666 | scale *= 10; | |
2667 | } | |
2668 | buf[i] = 0; | |
2669 | } else if (buf[i] == '.') { | |
2670 | dot=1; | |
2671 | buf[i] = 0; | |
2672 | } | |
2673 | } | |
97ce0a7f | 2674 | if (strict_strtoul(buf, 10, &msec) < 0) |
16f17b39 N |
2675 | return -EINVAL; |
2676 | msec = (msec * 1000) / scale; | |
2677 | if (msec == 0) | |
2678 | mddev->safemode_delay = 0; | |
2679 | else { | |
19052c0e | 2680 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2681 | mddev->safemode_delay = (msec*HZ)/1000; |
2682 | if (mddev->safemode_delay == 0) | |
2683 | mddev->safemode_delay = 1; | |
19052c0e N |
2684 | if (mddev->safemode_delay < old_delay) |
2685 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2686 | } |
2687 | return len; | |
2688 | } | |
2689 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2690 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2691 | |
eae1701f | 2692 | static ssize_t |
96de1e66 | 2693 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2694 | { |
2604b703 | 2695 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2696 | if (p) |
eae1701f | 2697 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2698 | else if (mddev->clevel[0]) |
2699 | return sprintf(page, "%s\n", mddev->clevel); | |
2700 | else if (mddev->level != LEVEL_NONE) | |
2701 | return sprintf(page, "%d\n", mddev->level); | |
2702 | else | |
2703 | return 0; | |
eae1701f N |
2704 | } |
2705 | ||
d9d166c2 N |
2706 | static ssize_t |
2707 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2708 | { | |
245f46c2 | 2709 | char level[16]; |
20a49ff6 | 2710 | ssize_t rv = len; |
245f46c2 N |
2711 | struct mdk_personality *pers; |
2712 | void *priv; | |
3a981b03 | 2713 | mdk_rdev_t *rdev; |
245f46c2 N |
2714 | |
2715 | if (mddev->pers == NULL) { | |
2716 | if (len == 0) | |
2717 | return 0; | |
2718 | if (len >= sizeof(mddev->clevel)) | |
2719 | return -ENOSPC; | |
2720 | strncpy(mddev->clevel, buf, len); | |
2721 | if (mddev->clevel[len-1] == '\n') | |
2722 | len--; | |
2723 | mddev->clevel[len] = 0; | |
2724 | mddev->level = LEVEL_NONE; | |
2725 | return rv; | |
2726 | } | |
2727 | ||
2728 | /* request to change the personality. Need to ensure: | |
2729 | * - array is not engaged in resync/recovery/reshape | |
2730 | * - old personality can be suspended | |
2731 | * - new personality will access other array. | |
2732 | */ | |
2733 | ||
2734 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) | |
d9d166c2 | 2735 | return -EBUSY; |
245f46c2 N |
2736 | |
2737 | if (!mddev->pers->quiesce) { | |
2738 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
2739 | mdname(mddev), mddev->pers->name); | |
2740 | return -EINVAL; | |
2741 | } | |
2742 | ||
2743 | /* Now find the new personality */ | |
2744 | if (len == 0 || len >= sizeof(level)) | |
2745 | return -EINVAL; | |
2746 | strncpy(level, buf, len); | |
2747 | if (level[len-1] == '\n') | |
d9d166c2 | 2748 | len--; |
245f46c2 N |
2749 | level[len] = 0; |
2750 | ||
2751 | request_module("md-%s", level); | |
2752 | spin_lock(&pers_lock); | |
2753 | pers = find_pers(LEVEL_NONE, level); | |
2754 | if (!pers || !try_module_get(pers->owner)) { | |
2755 | spin_unlock(&pers_lock); | |
2756 | printk(KERN_WARNING "md: personality %s not loaded\n", level); | |
2757 | return -EINVAL; | |
2758 | } | |
2759 | spin_unlock(&pers_lock); | |
2760 | ||
2761 | if (pers == mddev->pers) { | |
2762 | /* Nothing to do! */ | |
2763 | module_put(pers->owner); | |
2764 | return rv; | |
2765 | } | |
2766 | if (!pers->takeover) { | |
2767 | module_put(pers->owner); | |
2768 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
2769 | mdname(mddev), level); | |
2770 | return -EINVAL; | |
2771 | } | |
2772 | ||
2773 | /* ->takeover must set new_* and/or delta_disks | |
2774 | * if it succeeds, and may set them when it fails. | |
2775 | */ | |
2776 | priv = pers->takeover(mddev); | |
2777 | if (IS_ERR(priv)) { | |
2778 | mddev->new_level = mddev->level; | |
2779 | mddev->new_layout = mddev->layout; | |
664e7c41 | 2780 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
2781 | mddev->raid_disks -= mddev->delta_disks; |
2782 | mddev->delta_disks = 0; | |
2783 | module_put(pers->owner); | |
2784 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
2785 | mdname(mddev), level); | |
2786 | return PTR_ERR(priv); | |
2787 | } | |
2788 | ||
2789 | /* Looks like we have a winner */ | |
2790 | mddev_suspend(mddev); | |
2791 | mddev->pers->stop(mddev); | |
2792 | module_put(mddev->pers->owner); | |
3a981b03 N |
2793 | /* Invalidate devices that are now superfluous */ |
2794 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
2795 | if (rdev->raid_disk >= mddev->raid_disks) { | |
2796 | rdev->raid_disk = -1; | |
2797 | clear_bit(In_sync, &rdev->flags); | |
2798 | } | |
245f46c2 N |
2799 | mddev->pers = pers; |
2800 | mddev->private = priv; | |
2801 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
2802 | mddev->level = mddev->new_level; | |
2803 | mddev->layout = mddev->new_layout; | |
664e7c41 | 2804 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 N |
2805 | mddev->delta_disks = 0; |
2806 | pers->run(mddev); | |
2807 | mddev_resume(mddev); | |
2808 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2809 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2810 | md_wakeup_thread(mddev->thread); | |
d9d166c2 N |
2811 | return rv; |
2812 | } | |
2813 | ||
2814 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 2815 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 2816 | |
d4dbd025 N |
2817 | |
2818 | static ssize_t | |
2819 | layout_show(mddev_t *mddev, char *page) | |
2820 | { | |
2821 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
2822 | if (mddev->reshape_position != MaxSector && |
2823 | mddev->layout != mddev->new_layout) | |
2824 | return sprintf(page, "%d (%d)\n", | |
2825 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
2826 | return sprintf(page, "%d\n", mddev->layout); |
2827 | } | |
2828 | ||
2829 | static ssize_t | |
2830 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
2831 | { | |
2832 | char *e; | |
2833 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
2834 | |
2835 | if (!*buf || (*e && *e != '\n')) | |
2836 | return -EINVAL; | |
2837 | ||
b3546035 N |
2838 | if (mddev->pers) { |
2839 | int err; | |
50ac168a | 2840 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 2841 | return -EBUSY; |
597a711b | 2842 | mddev->new_layout = n; |
50ac168a | 2843 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
2844 | if (err) { |
2845 | mddev->new_layout = mddev->layout; | |
b3546035 | 2846 | return err; |
597a711b | 2847 | } |
b3546035 | 2848 | } else { |
08a02ecd | 2849 | mddev->new_layout = n; |
b3546035 N |
2850 | if (mddev->reshape_position == MaxSector) |
2851 | mddev->layout = n; | |
2852 | } | |
d4dbd025 N |
2853 | return len; |
2854 | } | |
2855 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 2856 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
2857 | |
2858 | ||
eae1701f | 2859 | static ssize_t |
96de1e66 | 2860 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 2861 | { |
bb636547 N |
2862 | if (mddev->raid_disks == 0) |
2863 | return 0; | |
08a02ecd N |
2864 | if (mddev->reshape_position != MaxSector && |
2865 | mddev->delta_disks != 0) | |
2866 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
2867 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
2868 | return sprintf(page, "%d\n", mddev->raid_disks); |
2869 | } | |
2870 | ||
da943b99 N |
2871 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
2872 | ||
2873 | static ssize_t | |
2874 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
2875 | { | |
da943b99 N |
2876 | char *e; |
2877 | int rv = 0; | |
2878 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2879 | ||
2880 | if (!*buf || (*e && *e != '\n')) | |
2881 | return -EINVAL; | |
2882 | ||
2883 | if (mddev->pers) | |
2884 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
2885 | else if (mddev->reshape_position != MaxSector) { |
2886 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
2887 | mddev->delta_disks = n - olddisks; | |
2888 | mddev->raid_disks = n; | |
2889 | } else | |
da943b99 N |
2890 | mddev->raid_disks = n; |
2891 | return rv ? rv : len; | |
2892 | } | |
2893 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 2894 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 2895 | |
3b34380a N |
2896 | static ssize_t |
2897 | chunk_size_show(mddev_t *mddev, char *page) | |
2898 | { | |
08a02ecd | 2899 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
2900 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
2901 | return sprintf(page, "%d (%d)\n", | |
2902 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
2903 | mddev->chunk_sectors << 9); |
2904 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
2905 | } |
2906 | ||
2907 | static ssize_t | |
2908 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
2909 | { | |
3b34380a N |
2910 | char *e; |
2911 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2912 | ||
3b34380a N |
2913 | if (!*buf || (*e && *e != '\n')) |
2914 | return -EINVAL; | |
2915 | ||
b3546035 N |
2916 | if (mddev->pers) { |
2917 | int err; | |
50ac168a | 2918 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 2919 | return -EBUSY; |
597a711b | 2920 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 2921 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
2922 | if (err) { |
2923 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 2924 | return err; |
597a711b | 2925 | } |
b3546035 | 2926 | } else { |
664e7c41 | 2927 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 2928 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 2929 | mddev->chunk_sectors = n >> 9; |
b3546035 | 2930 | } |
3b34380a N |
2931 | return len; |
2932 | } | |
2933 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 2934 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 2935 | |
a94213b1 N |
2936 | static ssize_t |
2937 | resync_start_show(mddev_t *mddev, char *page) | |
2938 | { | |
d1a7c503 N |
2939 | if (mddev->recovery_cp == MaxSector) |
2940 | return sprintf(page, "none\n"); | |
a94213b1 N |
2941 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
2942 | } | |
2943 | ||
2944 | static ssize_t | |
2945 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
2946 | { | |
a94213b1 N |
2947 | char *e; |
2948 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
2949 | ||
2950 | if (mddev->pers) | |
2951 | return -EBUSY; | |
2952 | if (!*buf || (*e && *e != '\n')) | |
2953 | return -EINVAL; | |
2954 | ||
2955 | mddev->recovery_cp = n; | |
2956 | return len; | |
2957 | } | |
2958 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 2959 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 2960 | |
9e653b63 N |
2961 | /* |
2962 | * The array state can be: | |
2963 | * | |
2964 | * clear | |
2965 | * No devices, no size, no level | |
2966 | * Equivalent to STOP_ARRAY ioctl | |
2967 | * inactive | |
2968 | * May have some settings, but array is not active | |
2969 | * all IO results in error | |
2970 | * When written, doesn't tear down array, but just stops it | |
2971 | * suspended (not supported yet) | |
2972 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 2973 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
2974 | * readonly |
2975 | * no resync can happen. no superblocks get written. | |
2976 | * write requests fail | |
2977 | * read-auto | |
2978 | * like readonly, but behaves like 'clean' on a write request. | |
2979 | * | |
2980 | * clean - no pending writes, but otherwise active. | |
2981 | * When written to inactive array, starts without resync | |
2982 | * If a write request arrives then | |
2983 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
2984 | * if not known, block and switch to write-pending | |
2985 | * If written to an active array that has pending writes, then fails. | |
2986 | * active | |
2987 | * fully active: IO and resync can be happening. | |
2988 | * When written to inactive array, starts with resync | |
2989 | * | |
2990 | * write-pending | |
2991 | * clean, but writes are blocked waiting for 'active' to be written. | |
2992 | * | |
2993 | * active-idle | |
2994 | * like active, but no writes have been seen for a while (100msec). | |
2995 | * | |
2996 | */ | |
2997 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
2998 | write_pending, active_idle, bad_word}; | |
05381954 | 2999 | static char *array_states[] = { |
9e653b63 N |
3000 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3001 | "write-pending", "active-idle", NULL }; | |
3002 | ||
3003 | static int match_word(const char *word, char **list) | |
3004 | { | |
3005 | int n; | |
3006 | for (n=0; list[n]; n++) | |
3007 | if (cmd_match(word, list[n])) | |
3008 | break; | |
3009 | return n; | |
3010 | } | |
3011 | ||
3012 | static ssize_t | |
3013 | array_state_show(mddev_t *mddev, char *page) | |
3014 | { | |
3015 | enum array_state st = inactive; | |
3016 | ||
3017 | if (mddev->pers) | |
3018 | switch(mddev->ro) { | |
3019 | case 1: | |
3020 | st = readonly; | |
3021 | break; | |
3022 | case 2: | |
3023 | st = read_auto; | |
3024 | break; | |
3025 | case 0: | |
3026 | if (mddev->in_sync) | |
3027 | st = clean; | |
e691063a N |
3028 | else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) |
3029 | st = write_pending; | |
9e653b63 N |
3030 | else if (mddev->safemode) |
3031 | st = active_idle; | |
3032 | else | |
3033 | st = active; | |
3034 | } | |
3035 | else { | |
3036 | if (list_empty(&mddev->disks) && | |
3037 | mddev->raid_disks == 0 && | |
58c0fed4 | 3038 | mddev->dev_sectors == 0) |
9e653b63 N |
3039 | st = clear; |
3040 | else | |
3041 | st = inactive; | |
3042 | } | |
3043 | return sprintf(page, "%s\n", array_states[st]); | |
3044 | } | |
3045 | ||
df5b20cf | 3046 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
9e653b63 N |
3047 | static int do_md_run(mddev_t * mddev); |
3048 | static int restart_array(mddev_t *mddev); | |
3049 | ||
3050 | static ssize_t | |
3051 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3052 | { | |
3053 | int err = -EINVAL; | |
3054 | enum array_state st = match_word(buf, array_states); | |
3055 | switch(st) { | |
3056 | case bad_word: | |
3057 | break; | |
3058 | case clear: | |
3059 | /* stopping an active array */ | |
f2ea68cf | 3060 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3061 | return -EBUSY; |
df5b20cf | 3062 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3063 | break; |
3064 | case inactive: | |
3065 | /* stopping an active array */ | |
3066 | if (mddev->pers) { | |
f2ea68cf | 3067 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3068 | return -EBUSY; |
df5b20cf | 3069 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3070 | } else |
3071 | err = 0; /* already inactive */ | |
9e653b63 N |
3072 | break; |
3073 | case suspended: | |
3074 | break; /* not supported yet */ | |
3075 | case readonly: | |
3076 | if (mddev->pers) | |
df5b20cf | 3077 | err = do_md_stop(mddev, 1, 0); |
9e653b63 N |
3078 | else { |
3079 | mddev->ro = 1; | |
648b629e | 3080 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3081 | err = do_md_run(mddev); |
3082 | } | |
3083 | break; | |
3084 | case read_auto: | |
9e653b63 | 3085 | if (mddev->pers) { |
80268ee9 | 3086 | if (mddev->ro == 0) |
df5b20cf | 3087 | err = do_md_stop(mddev, 1, 0); |
80268ee9 | 3088 | else if (mddev->ro == 1) |
648b629e N |
3089 | err = restart_array(mddev); |
3090 | if (err == 0) { | |
3091 | mddev->ro = 2; | |
3092 | set_disk_ro(mddev->gendisk, 0); | |
3093 | } | |
9e653b63 N |
3094 | } else { |
3095 | mddev->ro = 2; | |
3096 | err = do_md_run(mddev); | |
3097 | } | |
3098 | break; | |
3099 | case clean: | |
3100 | if (mddev->pers) { | |
3101 | restart_array(mddev); | |
3102 | spin_lock_irq(&mddev->write_lock); | |
3103 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3104 | if (mddev->in_sync == 0) { |
3105 | mddev->in_sync = 1; | |
31a59e34 N |
3106 | if (mddev->safemode == 1) |
3107 | mddev->safemode = 0; | |
e691063a N |
3108 | if (mddev->persistent) |
3109 | set_bit(MD_CHANGE_CLEAN, | |
3110 | &mddev->flags); | |
3111 | } | |
3112 | err = 0; | |
3113 | } else | |
3114 | err = -EBUSY; | |
9e653b63 | 3115 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3116 | } else |
3117 | err = -EINVAL; | |
9e653b63 N |
3118 | break; |
3119 | case active: | |
3120 | if (mddev->pers) { | |
3121 | restart_array(mddev); | |
e691063a N |
3122 | if (mddev->external) |
3123 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
9e653b63 N |
3124 | wake_up(&mddev->sb_wait); |
3125 | err = 0; | |
3126 | } else { | |
3127 | mddev->ro = 0; | |
648b629e | 3128 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3129 | err = do_md_run(mddev); |
3130 | } | |
3131 | break; | |
3132 | case write_pending: | |
3133 | case active_idle: | |
3134 | /* these cannot be set */ | |
3135 | break; | |
3136 | } | |
3137 | if (err) | |
3138 | return err; | |
0fd62b86 | 3139 | else { |
b62b7590 | 3140 | sysfs_notify_dirent(mddev->sysfs_state); |
9e653b63 | 3141 | return len; |
0fd62b86 | 3142 | } |
9e653b63 | 3143 | } |
80ca3a44 N |
3144 | static struct md_sysfs_entry md_array_state = |
3145 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3146 | |
6d7ff738 N |
3147 | static ssize_t |
3148 | null_show(mddev_t *mddev, char *page) | |
3149 | { | |
3150 | return -EINVAL; | |
3151 | } | |
3152 | ||
3153 | static ssize_t | |
3154 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3155 | { | |
3156 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3157 | /* The new device is added to the array. | |
3158 | * If the array has a persistent superblock, we read the | |
3159 | * superblock to initialise info and check validity. | |
3160 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3161 | * which mainly checks size. | |
3162 | */ | |
3163 | char *e; | |
3164 | int major = simple_strtoul(buf, &e, 10); | |
3165 | int minor; | |
3166 | dev_t dev; | |
3167 | mdk_rdev_t *rdev; | |
3168 | int err; | |
3169 | ||
3170 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3171 | return -EINVAL; | |
3172 | minor = simple_strtoul(e+1, &e, 10); | |
3173 | if (*e && *e != '\n') | |
3174 | return -EINVAL; | |
3175 | dev = MKDEV(major, minor); | |
3176 | if (major != MAJOR(dev) || | |
3177 | minor != MINOR(dev)) | |
3178 | return -EOVERFLOW; | |
3179 | ||
3180 | ||
3181 | if (mddev->persistent) { | |
3182 | rdev = md_import_device(dev, mddev->major_version, | |
3183 | mddev->minor_version); | |
3184 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3185 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3186 | mdk_rdev_t, same_set); | |
3187 | err = super_types[mddev->major_version] | |
3188 | .load_super(rdev, rdev0, mddev->minor_version); | |
3189 | if (err < 0) | |
3190 | goto out; | |
3191 | } | |
c5d79adb N |
3192 | } else if (mddev->external) |
3193 | rdev = md_import_device(dev, -2, -1); | |
3194 | else | |
6d7ff738 N |
3195 | rdev = md_import_device(dev, -1, -1); |
3196 | ||
3197 | if (IS_ERR(rdev)) | |
3198 | return PTR_ERR(rdev); | |
3199 | err = bind_rdev_to_array(rdev, mddev); | |
3200 | out: | |
3201 | if (err) | |
3202 | export_rdev(rdev); | |
3203 | return err ? err : len; | |
3204 | } | |
3205 | ||
3206 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3207 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3208 | |
9b1d1dac PC |
3209 | static ssize_t |
3210 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3211 | { | |
3212 | char *end; | |
3213 | unsigned long chunk, end_chunk; | |
3214 | ||
3215 | if (!mddev->bitmap) | |
3216 | goto out; | |
3217 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3218 | while (*buf) { | |
3219 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3220 | if (buf == end) break; | |
3221 | if (*end == '-') { /* range */ | |
3222 | buf = end + 1; | |
3223 | end_chunk = simple_strtoul(buf, &end, 0); | |
3224 | if (buf == end) break; | |
3225 | } | |
3226 | if (*end && !isspace(*end)) break; | |
3227 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
3228 | buf = end; | |
3229 | while (isspace(*buf)) buf++; | |
3230 | } | |
3231 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3232 | out: | |
3233 | return len; | |
3234 | } | |
3235 | ||
3236 | static struct md_sysfs_entry md_bitmap = | |
3237 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3238 | ||
a35b0d69 N |
3239 | static ssize_t |
3240 | size_show(mddev_t *mddev, char *page) | |
3241 | { | |
58c0fed4 AN |
3242 | return sprintf(page, "%llu\n", |
3243 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3244 | } |
3245 | ||
d71f9f88 | 3246 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3247 | |
3248 | static ssize_t | |
3249 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3250 | { | |
3251 | /* If array is inactive, we can reduce the component size, but | |
3252 | * not increase it (except from 0). | |
3253 | * If array is active, we can try an on-line resize | |
3254 | */ | |
b522adcd DW |
3255 | sector_t sectors; |
3256 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3257 | |
58c0fed4 AN |
3258 | if (err < 0) |
3259 | return err; | |
a35b0d69 | 3260 | if (mddev->pers) { |
58c0fed4 | 3261 | err = update_size(mddev, sectors); |
850b2b42 | 3262 | md_update_sb(mddev, 1); |
a35b0d69 | 3263 | } else { |
58c0fed4 AN |
3264 | if (mddev->dev_sectors == 0 || |
3265 | mddev->dev_sectors > sectors) | |
3266 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3267 | else |
3268 | err = -ENOSPC; | |
3269 | } | |
3270 | return err ? err : len; | |
3271 | } | |
3272 | ||
3273 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3274 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3275 | |
8bb93aac N |
3276 | |
3277 | /* Metdata version. | |
e691063a N |
3278 | * This is one of |
3279 | * 'none' for arrays with no metadata (good luck...) | |
3280 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3281 | * or N.M for internally known formats |
3282 | */ | |
3283 | static ssize_t | |
3284 | metadata_show(mddev_t *mddev, char *page) | |
3285 | { | |
3286 | if (mddev->persistent) | |
3287 | return sprintf(page, "%d.%d\n", | |
3288 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3289 | else if (mddev->external) |
3290 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3291 | else |
3292 | return sprintf(page, "none\n"); | |
3293 | } | |
3294 | ||
3295 | static ssize_t | |
3296 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3297 | { | |
3298 | int major, minor; | |
3299 | char *e; | |
ea43ddd8 N |
3300 | /* Changing the details of 'external' metadata is |
3301 | * always permitted. Otherwise there must be | |
3302 | * no devices attached to the array. | |
3303 | */ | |
3304 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3305 | ; | |
3306 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3307 | return -EBUSY; |
3308 | ||
3309 | if (cmd_match(buf, "none")) { | |
3310 | mddev->persistent = 0; | |
e691063a N |
3311 | mddev->external = 0; |
3312 | mddev->major_version = 0; | |
3313 | mddev->minor_version = 90; | |
3314 | return len; | |
3315 | } | |
3316 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3317 | size_t namelen = len-9; |
e691063a N |
3318 | if (namelen >= sizeof(mddev->metadata_type)) |
3319 | namelen = sizeof(mddev->metadata_type)-1; | |
3320 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3321 | mddev->metadata_type[namelen] = 0; | |
3322 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3323 | mddev->metadata_type[--namelen] = 0; | |
3324 | mddev->persistent = 0; | |
3325 | mddev->external = 1; | |
8bb93aac N |
3326 | mddev->major_version = 0; |
3327 | mddev->minor_version = 90; | |
3328 | return len; | |
3329 | } | |
3330 | major = simple_strtoul(buf, &e, 10); | |
3331 | if (e==buf || *e != '.') | |
3332 | return -EINVAL; | |
3333 | buf = e+1; | |
3334 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3335 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3336 | return -EINVAL; |
50511da3 | 3337 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3338 | return -ENOENT; |
3339 | mddev->major_version = major; | |
3340 | mddev->minor_version = minor; | |
3341 | mddev->persistent = 1; | |
e691063a | 3342 | mddev->external = 0; |
8bb93aac N |
3343 | return len; |
3344 | } | |
3345 | ||
3346 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3347 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3348 | |
24dd469d | 3349 | static ssize_t |
7eec314d | 3350 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3351 | { |
7eec314d | 3352 | char *type = "idle"; |
b6a9ce68 N |
3353 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3354 | type = "frozen"; | |
3355 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3356 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3357 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3358 | type = "reshape"; | |
3359 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3360 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3361 | type = "resync"; | |
3362 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3363 | type = "check"; | |
3364 | else | |
3365 | type = "repair"; | |
72a23c21 | 3366 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3367 | type = "recover"; |
3368 | } | |
3369 | return sprintf(page, "%s\n", type); | |
3370 | } | |
3371 | ||
3372 | static ssize_t | |
7eec314d | 3373 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3374 | { |
7eec314d N |
3375 | if (!mddev->pers || !mddev->pers->sync_request) |
3376 | return -EINVAL; | |
3377 | ||
b6a9ce68 N |
3378 | if (cmd_match(page, "frozen")) |
3379 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3380 | else | |
3381 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3382 | ||
3383 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3384 | if (mddev->sync_thread) { |
3385 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
3386 | md_unregister_thread(mddev->sync_thread); | |
3387 | mddev->sync_thread = NULL; | |
3388 | mddev->recovery = 0; | |
3389 | } | |
03c902e1 N |
3390 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3391 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3392 | return -EBUSY; |
72a23c21 NB |
3393 | else if (cmd_match(page, "resync")) |
3394 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3395 | else if (cmd_match(page, "recover")) { | |
3396 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3397 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3398 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3399 | int err; |
3400 | if (mddev->pers->start_reshape == NULL) | |
3401 | return -EINVAL; | |
3402 | err = mddev->pers->start_reshape(mddev); | |
3403 | if (err) | |
3404 | return err; | |
a99ac971 | 3405 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3406 | } else { |
bce74dac | 3407 | if (cmd_match(page, "check")) |
7eec314d | 3408 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3409 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3410 | return -EINVAL; |
3411 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3412 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3413 | } |
03c902e1 | 3414 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3415 | md_wakeup_thread(mddev->thread); |
0c3573f1 | 3416 | sysfs_notify_dirent(mddev->sysfs_action); |
24dd469d N |
3417 | return len; |
3418 | } | |
3419 | ||
9d88883e | 3420 | static ssize_t |
96de1e66 | 3421 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3422 | { |
3423 | return sprintf(page, "%llu\n", | |
3424 | (unsigned long long) mddev->resync_mismatches); | |
3425 | } | |
3426 | ||
80ca3a44 N |
3427 | static struct md_sysfs_entry md_scan_mode = |
3428 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3429 | |
96de1e66 | 3430 | |
80ca3a44 | 3431 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3432 | |
88202a0c N |
3433 | static ssize_t |
3434 | sync_min_show(mddev_t *mddev, char *page) | |
3435 | { | |
3436 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3437 | mddev->sync_speed_min ? "local": "system"); | |
3438 | } | |
3439 | ||
3440 | static ssize_t | |
3441 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3442 | { | |
3443 | int min; | |
3444 | char *e; | |
3445 | if (strncmp(buf, "system", 6)==0) { | |
3446 | mddev->sync_speed_min = 0; | |
3447 | return len; | |
3448 | } | |
3449 | min = simple_strtoul(buf, &e, 10); | |
3450 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3451 | return -EINVAL; | |
3452 | mddev->sync_speed_min = min; | |
3453 | return len; | |
3454 | } | |
3455 | ||
3456 | static struct md_sysfs_entry md_sync_min = | |
3457 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3458 | ||
3459 | static ssize_t | |
3460 | sync_max_show(mddev_t *mddev, char *page) | |
3461 | { | |
3462 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3463 | mddev->sync_speed_max ? "local": "system"); | |
3464 | } | |
3465 | ||
3466 | static ssize_t | |
3467 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3468 | { | |
3469 | int max; | |
3470 | char *e; | |
3471 | if (strncmp(buf, "system", 6)==0) { | |
3472 | mddev->sync_speed_max = 0; | |
3473 | return len; | |
3474 | } | |
3475 | max = simple_strtoul(buf, &e, 10); | |
3476 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3477 | return -EINVAL; | |
3478 | mddev->sync_speed_max = max; | |
3479 | return len; | |
3480 | } | |
3481 | ||
3482 | static struct md_sysfs_entry md_sync_max = | |
3483 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3484 | ||
d7f3d291 IP |
3485 | static ssize_t |
3486 | degraded_show(mddev_t *mddev, char *page) | |
3487 | { | |
3488 | return sprintf(page, "%d\n", mddev->degraded); | |
3489 | } | |
3490 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3491 | |
90b08710 BS |
3492 | static ssize_t |
3493 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3494 | { | |
3495 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3496 | } | |
3497 | ||
3498 | static ssize_t | |
3499 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3500 | { | |
3501 | long n; | |
3502 | ||
3503 | if (strict_strtol(buf, 10, &n)) | |
3504 | return -EINVAL; | |
3505 | ||
3506 | if (n != 0 && n != 1) | |
3507 | return -EINVAL; | |
3508 | ||
3509 | mddev->parallel_resync = n; | |
3510 | ||
3511 | if (mddev->sync_thread) | |
3512 | wake_up(&resync_wait); | |
3513 | ||
3514 | return len; | |
3515 | } | |
3516 | ||
3517 | /* force parallel resync, even with shared block devices */ | |
3518 | static struct md_sysfs_entry md_sync_force_parallel = | |
3519 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3520 | sync_force_parallel_show, sync_force_parallel_store); | |
3521 | ||
88202a0c N |
3522 | static ssize_t |
3523 | sync_speed_show(mddev_t *mddev, char *page) | |
3524 | { | |
3525 | unsigned long resync, dt, db; | |
d1a7c503 N |
3526 | if (mddev->curr_resync == 0) |
3527 | return sprintf(page, "none\n"); | |
9687a60c AN |
3528 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3529 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3530 | if (!dt) dt++; |
9687a60c AN |
3531 | db = resync - mddev->resync_mark_cnt; |
3532 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3533 | } |
3534 | ||
80ca3a44 | 3535 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3536 | |
3537 | static ssize_t | |
3538 | sync_completed_show(mddev_t *mddev, char *page) | |
3539 | { | |
58c0fed4 | 3540 | unsigned long max_sectors, resync; |
88202a0c | 3541 | |
acb180b0 N |
3542 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3543 | return sprintf(page, "none\n"); | |
3544 | ||
88202a0c | 3545 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3546 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3547 | else |
58c0fed4 | 3548 | max_sectors = mddev->dev_sectors; |
88202a0c | 3549 | |
acb180b0 | 3550 | resync = mddev->curr_resync_completed; |
58c0fed4 | 3551 | return sprintf(page, "%lu / %lu\n", resync, max_sectors); |
88202a0c N |
3552 | } |
3553 | ||
80ca3a44 | 3554 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3555 | |
5e96ee65 NB |
3556 | static ssize_t |
3557 | min_sync_show(mddev_t *mddev, char *page) | |
3558 | { | |
3559 | return sprintf(page, "%llu\n", | |
3560 | (unsigned long long)mddev->resync_min); | |
3561 | } | |
3562 | static ssize_t | |
3563 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3564 | { | |
3565 | unsigned long long min; | |
3566 | if (strict_strtoull(buf, 10, &min)) | |
3567 | return -EINVAL; | |
3568 | if (min > mddev->resync_max) | |
3569 | return -EINVAL; | |
3570 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3571 | return -EBUSY; | |
3572 | ||
3573 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3574 | if (mddev->chunk_sectors) { |
2ac06c33 | 3575 | sector_t temp = min; |
9d8f0363 | 3576 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3577 | return -EINVAL; |
3578 | } | |
3579 | mddev->resync_min = min; | |
3580 | ||
3581 | return len; | |
3582 | } | |
3583 | ||
3584 | static struct md_sysfs_entry md_min_sync = | |
3585 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3586 | ||
c6207277 N |
3587 | static ssize_t |
3588 | max_sync_show(mddev_t *mddev, char *page) | |
3589 | { | |
3590 | if (mddev->resync_max == MaxSector) | |
3591 | return sprintf(page, "max\n"); | |
3592 | else | |
3593 | return sprintf(page, "%llu\n", | |
3594 | (unsigned long long)mddev->resync_max); | |
3595 | } | |
3596 | static ssize_t | |
3597 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3598 | { | |
3599 | if (strncmp(buf, "max", 3) == 0) | |
3600 | mddev->resync_max = MaxSector; | |
3601 | else { | |
5e96ee65 NB |
3602 | unsigned long long max; |
3603 | if (strict_strtoull(buf, 10, &max)) | |
3604 | return -EINVAL; | |
3605 | if (max < mddev->resync_min) | |
c6207277 N |
3606 | return -EINVAL; |
3607 | if (max < mddev->resync_max && | |
4d484a4a | 3608 | mddev->ro == 0 && |
c6207277 N |
3609 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3610 | return -EBUSY; | |
3611 | ||
3612 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3613 | if (mddev->chunk_sectors) { |
2ac06c33 | 3614 | sector_t temp = max; |
9d8f0363 | 3615 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
3616 | return -EINVAL; |
3617 | } | |
3618 | mddev->resync_max = max; | |
3619 | } | |
3620 | wake_up(&mddev->recovery_wait); | |
3621 | return len; | |
3622 | } | |
3623 | ||
3624 | static struct md_sysfs_entry md_max_sync = | |
3625 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
3626 | ||
e464eafd N |
3627 | static ssize_t |
3628 | suspend_lo_show(mddev_t *mddev, char *page) | |
3629 | { | |
3630 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
3631 | } | |
3632 | ||
3633 | static ssize_t | |
3634 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
3635 | { | |
3636 | char *e; | |
3637 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3638 | ||
b8d966ef N |
3639 | if (mddev->pers == NULL || |
3640 | mddev->pers->quiesce == NULL) | |
e464eafd N |
3641 | return -EINVAL; |
3642 | if (buf == e || (*e && *e != '\n')) | |
3643 | return -EINVAL; | |
3644 | if (new >= mddev->suspend_hi || | |
3645 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
3646 | mddev->suspend_lo = new; | |
3647 | mddev->pers->quiesce(mddev, 2); | |
3648 | return len; | |
3649 | } else | |
3650 | return -EINVAL; | |
3651 | } | |
3652 | static struct md_sysfs_entry md_suspend_lo = | |
3653 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
3654 | ||
3655 | ||
3656 | static ssize_t | |
3657 | suspend_hi_show(mddev_t *mddev, char *page) | |
3658 | { | |
3659 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
3660 | } | |
3661 | ||
3662 | static ssize_t | |
3663 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
3664 | { | |
3665 | char *e; | |
3666 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3667 | ||
b8d966ef N |
3668 | if (mddev->pers == NULL || |
3669 | mddev->pers->quiesce == NULL) | |
e464eafd N |
3670 | return -EINVAL; |
3671 | if (buf == e || (*e && *e != '\n')) | |
3672 | return -EINVAL; | |
3673 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
3674 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
3675 | mddev->suspend_hi = new; | |
3676 | mddev->pers->quiesce(mddev, 1); | |
3677 | mddev->pers->quiesce(mddev, 0); | |
3678 | return len; | |
3679 | } else | |
3680 | return -EINVAL; | |
3681 | } | |
3682 | static struct md_sysfs_entry md_suspend_hi = | |
3683 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
3684 | ||
08a02ecd N |
3685 | static ssize_t |
3686 | reshape_position_show(mddev_t *mddev, char *page) | |
3687 | { | |
3688 | if (mddev->reshape_position != MaxSector) | |
3689 | return sprintf(page, "%llu\n", | |
3690 | (unsigned long long)mddev->reshape_position); | |
3691 | strcpy(page, "none\n"); | |
3692 | return 5; | |
3693 | } | |
3694 | ||
3695 | static ssize_t | |
3696 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
3697 | { | |
3698 | char *e; | |
3699 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3700 | if (mddev->pers) | |
3701 | return -EBUSY; | |
3702 | if (buf == e || (*e && *e != '\n')) | |
3703 | return -EINVAL; | |
3704 | mddev->reshape_position = new; | |
3705 | mddev->delta_disks = 0; | |
3706 | mddev->new_level = mddev->level; | |
3707 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3708 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
3709 | return len; |
3710 | } | |
3711 | ||
3712 | static struct md_sysfs_entry md_reshape_position = | |
3713 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
3714 | reshape_position_store); | |
3715 | ||
b522adcd DW |
3716 | static ssize_t |
3717 | array_size_show(mddev_t *mddev, char *page) | |
3718 | { | |
3719 | if (mddev->external_size) | |
3720 | return sprintf(page, "%llu\n", | |
3721 | (unsigned long long)mddev->array_sectors/2); | |
3722 | else | |
3723 | return sprintf(page, "default\n"); | |
3724 | } | |
3725 | ||
3726 | static ssize_t | |
3727 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3728 | { | |
3729 | sector_t sectors; | |
3730 | ||
3731 | if (strncmp(buf, "default", 7) == 0) { | |
3732 | if (mddev->pers) | |
3733 | sectors = mddev->pers->size(mddev, 0, 0); | |
3734 | else | |
3735 | sectors = mddev->array_sectors; | |
3736 | ||
3737 | mddev->external_size = 0; | |
3738 | } else { | |
3739 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
3740 | return -EINVAL; | |
3741 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 3742 | return -E2BIG; |
b522adcd DW |
3743 | |
3744 | mddev->external_size = 1; | |
3745 | } | |
3746 | ||
3747 | mddev->array_sectors = sectors; | |
3748 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e N |
3749 | if (mddev->pers) |
3750 | revalidate_disk(mddev->gendisk); | |
b522adcd DW |
3751 | |
3752 | return len; | |
3753 | } | |
3754 | ||
3755 | static struct md_sysfs_entry md_array_size = | |
3756 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
3757 | array_size_store); | |
e464eafd | 3758 | |
eae1701f N |
3759 | static struct attribute *md_default_attrs[] = { |
3760 | &md_level.attr, | |
d4dbd025 | 3761 | &md_layout.attr, |
eae1701f | 3762 | &md_raid_disks.attr, |
3b34380a | 3763 | &md_chunk_size.attr, |
a35b0d69 | 3764 | &md_size.attr, |
a94213b1 | 3765 | &md_resync_start.attr, |
8bb93aac | 3766 | &md_metadata.attr, |
6d7ff738 | 3767 | &md_new_device.attr, |
16f17b39 | 3768 | &md_safe_delay.attr, |
9e653b63 | 3769 | &md_array_state.attr, |
08a02ecd | 3770 | &md_reshape_position.attr, |
b522adcd | 3771 | &md_array_size.attr, |
411036fa N |
3772 | NULL, |
3773 | }; | |
3774 | ||
3775 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 3776 | &md_scan_mode.attr, |
9d88883e | 3777 | &md_mismatches.attr, |
88202a0c N |
3778 | &md_sync_min.attr, |
3779 | &md_sync_max.attr, | |
3780 | &md_sync_speed.attr, | |
90b08710 | 3781 | &md_sync_force_parallel.attr, |
88202a0c | 3782 | &md_sync_completed.attr, |
5e96ee65 | 3783 | &md_min_sync.attr, |
c6207277 | 3784 | &md_max_sync.attr, |
e464eafd N |
3785 | &md_suspend_lo.attr, |
3786 | &md_suspend_hi.attr, | |
9b1d1dac | 3787 | &md_bitmap.attr, |
d7f3d291 | 3788 | &md_degraded.attr, |
eae1701f N |
3789 | NULL, |
3790 | }; | |
411036fa N |
3791 | static struct attribute_group md_redundancy_group = { |
3792 | .name = NULL, | |
3793 | .attrs = md_redundancy_attrs, | |
3794 | }; | |
3795 | ||
eae1701f N |
3796 | |
3797 | static ssize_t | |
3798 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3799 | { | |
3800 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3801 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3802 | ssize_t rv; |
eae1701f N |
3803 | |
3804 | if (!entry->show) | |
3805 | return -EIO; | |
5dc5cf7d IM |
3806 | rv = mddev_lock(mddev); |
3807 | if (!rv) { | |
3808 | rv = entry->show(mddev, page); | |
3809 | mddev_unlock(mddev); | |
3810 | } | |
96de1e66 | 3811 | return rv; |
eae1701f N |
3812 | } |
3813 | ||
3814 | static ssize_t | |
3815 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
3816 | const char *page, size_t length) | |
3817 | { | |
3818 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3819 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3820 | ssize_t rv; |
eae1701f N |
3821 | |
3822 | if (!entry->store) | |
3823 | return -EIO; | |
67463acb N |
3824 | if (!capable(CAP_SYS_ADMIN)) |
3825 | return -EACCES; | |
5dc5cf7d | 3826 | rv = mddev_lock(mddev); |
d3374825 N |
3827 | if (mddev->hold_active == UNTIL_IOCTL) |
3828 | mddev->hold_active = 0; | |
5dc5cf7d IM |
3829 | if (!rv) { |
3830 | rv = entry->store(mddev, page, length); | |
3831 | mddev_unlock(mddev); | |
3832 | } | |
96de1e66 | 3833 | return rv; |
eae1701f N |
3834 | } |
3835 | ||
3836 | static void md_free(struct kobject *ko) | |
3837 | { | |
3838 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
3839 | |
3840 | if (mddev->sysfs_state) | |
3841 | sysfs_put(mddev->sysfs_state); | |
3842 | ||
3843 | if (mddev->gendisk) { | |
3844 | del_gendisk(mddev->gendisk); | |
3845 | put_disk(mddev->gendisk); | |
3846 | } | |
3847 | if (mddev->queue) | |
3848 | blk_cleanup_queue(mddev->queue); | |
3849 | ||
eae1701f N |
3850 | kfree(mddev); |
3851 | } | |
3852 | ||
3853 | static struct sysfs_ops md_sysfs_ops = { | |
3854 | .show = md_attr_show, | |
3855 | .store = md_attr_store, | |
3856 | }; | |
3857 | static struct kobj_type md_ktype = { | |
3858 | .release = md_free, | |
3859 | .sysfs_ops = &md_sysfs_ops, | |
3860 | .default_attrs = md_default_attrs, | |
3861 | }; | |
3862 | ||
1da177e4 LT |
3863 | int mdp_major = 0; |
3864 | ||
5fd3a17e DW |
3865 | static void mddev_delayed_delete(struct work_struct *ws) |
3866 | { | |
3867 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
3868 | ||
3869 | if (mddev->private == &md_redundancy_group) { | |
3870 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
3871 | if (mddev->sysfs_action) | |
3872 | sysfs_put(mddev->sysfs_action); | |
3873 | mddev->sysfs_action = NULL; | |
3874 | mddev->private = NULL; | |
3875 | } | |
3876 | kobject_del(&mddev->kobj); | |
3877 | kobject_put(&mddev->kobj); | |
3878 | } | |
3879 | ||
efeb53c0 | 3880 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 3881 | { |
48c9c27b | 3882 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
3883 | mddev_t *mddev = mddev_find(dev); |
3884 | struct gendisk *disk; | |
efeb53c0 N |
3885 | int partitioned; |
3886 | int shift; | |
3887 | int unit; | |
3830c62f | 3888 | int error; |
1da177e4 LT |
3889 | |
3890 | if (!mddev) | |
efeb53c0 N |
3891 | return -ENODEV; |
3892 | ||
3893 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
3894 | shift = partitioned ? MdpMinorShift : 0; | |
3895 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 3896 | |
d3374825 N |
3897 | /* wait for any previous instance if this device |
3898 | * to be completed removed (mddev_delayed_delete). | |
3899 | */ | |
3900 | flush_scheduled_work(); | |
3901 | ||
48c9c27b | 3902 | mutex_lock(&disks_mutex); |
0909dc44 N |
3903 | error = -EEXIST; |
3904 | if (mddev->gendisk) | |
3905 | goto abort; | |
efeb53c0 N |
3906 | |
3907 | if (name) { | |
3908 | /* Need to ensure that 'name' is not a duplicate. | |
3909 | */ | |
3910 | mddev_t *mddev2; | |
3911 | spin_lock(&all_mddevs_lock); | |
3912 | ||
3913 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
3914 | if (mddev2->gendisk && | |
3915 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
3916 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 3917 | goto abort; |
efeb53c0 N |
3918 | } |
3919 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 3920 | } |
8b765398 | 3921 | |
0909dc44 | 3922 | error = -ENOMEM; |
8b765398 | 3923 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
3924 | if (!mddev->queue) |
3925 | goto abort; | |
409c57f3 N |
3926 | mddev->queue->queuedata = mddev; |
3927 | ||
8b765398 N |
3928 | /* Can be unlocked because the queue is new: no concurrency */ |
3929 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); | |
3930 | ||
409c57f3 | 3931 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 3932 | |
1da177e4 LT |
3933 | disk = alloc_disk(1 << shift); |
3934 | if (!disk) { | |
8b765398 N |
3935 | blk_cleanup_queue(mddev->queue); |
3936 | mddev->queue = NULL; | |
0909dc44 | 3937 | goto abort; |
1da177e4 | 3938 | } |
efeb53c0 | 3939 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 3940 | disk->first_minor = unit << shift; |
efeb53c0 N |
3941 | if (name) |
3942 | strcpy(disk->disk_name, name); | |
3943 | else if (partitioned) | |
1da177e4 | 3944 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 3945 | else |
1da177e4 | 3946 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
3947 | disk->fops = &md_fops; |
3948 | disk->private_data = mddev; | |
3949 | disk->queue = mddev->queue; | |
92850bbd | 3950 | /* Allow extended partitions. This makes the |
d3374825 | 3951 | * 'mdp' device redundant, but we can't really |
92850bbd N |
3952 | * remove it now. |
3953 | */ | |
3954 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 LT |
3955 | add_disk(disk); |
3956 | mddev->gendisk = disk; | |
ed9e1982 TH |
3957 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
3958 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
3959 | if (error) { |
3960 | /* This isn't possible, but as kobject_init_and_add is marked | |
3961 | * __must_check, we must do something with the result | |
3962 | */ | |
5e55e2f5 N |
3963 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
3964 | disk->disk_name); | |
0909dc44 N |
3965 | error = 0; |
3966 | } | |
3967 | abort: | |
3968 | mutex_unlock(&disks_mutex); | |
3969 | if (!error) { | |
3830c62f | 3970 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
b62b7590 N |
3971 | mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state"); |
3972 | } | |
d3374825 | 3973 | mddev_put(mddev); |
0909dc44 | 3974 | return error; |
efeb53c0 N |
3975 | } |
3976 | ||
3977 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
3978 | { | |
3979 | md_alloc(dev, NULL); | |
1da177e4 LT |
3980 | return NULL; |
3981 | } | |
3982 | ||
efeb53c0 N |
3983 | static int add_named_array(const char *val, struct kernel_param *kp) |
3984 | { | |
3985 | /* val must be "md_*" where * is not all digits. | |
3986 | * We allocate an array with a large free minor number, and | |
3987 | * set the name to val. val must not already be an active name. | |
3988 | */ | |
3989 | int len = strlen(val); | |
3990 | char buf[DISK_NAME_LEN]; | |
3991 | ||
3992 | while (len && val[len-1] == '\n') | |
3993 | len--; | |
3994 | if (len >= DISK_NAME_LEN) | |
3995 | return -E2BIG; | |
3996 | strlcpy(buf, val, len+1); | |
3997 | if (strncmp(buf, "md_", 3) != 0) | |
3998 | return -EINVAL; | |
3999 | return md_alloc(0, buf); | |
4000 | } | |
4001 | ||
1da177e4 LT |
4002 | static void md_safemode_timeout(unsigned long data) |
4003 | { | |
4004 | mddev_t *mddev = (mddev_t *) data; | |
4005 | ||
0fd62b86 NB |
4006 | if (!atomic_read(&mddev->writes_pending)) { |
4007 | mddev->safemode = 1; | |
4008 | if (mddev->external) | |
b62b7590 | 4009 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 | 4010 | } |
1da177e4 LT |
4011 | md_wakeup_thread(mddev->thread); |
4012 | } | |
4013 | ||
6ff8d8ec | 4014 | static int start_dirty_degraded; |
1da177e4 LT |
4015 | |
4016 | static int do_md_run(mddev_t * mddev) | |
4017 | { | |
2604b703 | 4018 | int err; |
1da177e4 LT |
4019 | mdk_rdev_t *rdev; |
4020 | struct gendisk *disk; | |
2604b703 | 4021 | struct mdk_personality *pers; |
1da177e4 | 4022 | |
a757e64c N |
4023 | if (list_empty(&mddev->disks)) |
4024 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4025 | return -EINVAL; |
1da177e4 LT |
4026 | |
4027 | if (mddev->pers) | |
4028 | return -EBUSY; | |
4029 | ||
4030 | /* | |
4031 | * Analyze all RAID superblock(s) | |
4032 | */ | |
1ec4a939 N |
4033 | if (!mddev->raid_disks) { |
4034 | if (!mddev->persistent) | |
4035 | return -EINVAL; | |
a757e64c | 4036 | analyze_sbs(mddev); |
1ec4a939 | 4037 | } |
1da177e4 | 4038 | |
d9d166c2 N |
4039 | if (mddev->level != LEVEL_NONE) |
4040 | request_module("md-level-%d", mddev->level); | |
4041 | else if (mddev->clevel[0]) | |
4042 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4043 | |
4044 | /* | |
4045 | * Drop all container device buffers, from now on | |
4046 | * the only valid external interface is through the md | |
4047 | * device. | |
1da177e4 | 4048 | */ |
159ec1fc | 4049 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 4050 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4051 | continue; |
4052 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4053 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4054 | |
4055 | /* perform some consistency tests on the device. | |
4056 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4057 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4058 | */ |
0f420358 | 4059 | if (rdev->data_offset < rdev->sb_start) { |
58c0fed4 AN |
4060 | if (mddev->dev_sectors && |
4061 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4062 | > rdev->sb_start) { |
f0d76d70 N |
4063 | printk("md: %s: data overlaps metadata\n", |
4064 | mdname(mddev)); | |
4065 | return -EINVAL; | |
4066 | } | |
4067 | } else { | |
0f420358 | 4068 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4069 | > rdev->data_offset) { |
4070 | printk("md: %s: metadata overlaps data\n", | |
4071 | mdname(mddev)); | |
4072 | return -EINVAL; | |
4073 | } | |
4074 | } | |
3c0ee63a | 4075 | sysfs_notify_dirent(rdev->sysfs_state); |
1da177e4 LT |
4076 | } |
4077 | ||
4078 | md_probe(mddev->unit, NULL, NULL); | |
4079 | disk = mddev->gendisk; | |
4080 | if (!disk) | |
4081 | return -ENOMEM; | |
4082 | ||
4083 | spin_lock(&pers_lock); | |
d9d166c2 | 4084 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4085 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4086 | spin_unlock(&pers_lock); |
d9d166c2 N |
4087 | if (mddev->level != LEVEL_NONE) |
4088 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4089 | mddev->level); | |
4090 | else | |
4091 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4092 | mddev->clevel); | |
1da177e4 LT |
4093 | return -EINVAL; |
4094 | } | |
2604b703 | 4095 | mddev->pers = pers; |
1da177e4 | 4096 | spin_unlock(&pers_lock); |
34817e8c N |
4097 | if (mddev->level != pers->level) { |
4098 | mddev->level = pers->level; | |
4099 | mddev->new_level = pers->level; | |
4100 | } | |
d9d166c2 | 4101 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4102 | |
f6705578 | 4103 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4104 | pers->start_reshape == NULL) { |
f6705578 N |
4105 | /* This personality cannot handle reshaping... */ |
4106 | mddev->pers = NULL; | |
4107 | module_put(pers->owner); | |
4108 | return -EINVAL; | |
4109 | } | |
4110 | ||
7dd5e7c3 N |
4111 | if (pers->sync_request) { |
4112 | /* Warn if this is a potentially silly | |
4113 | * configuration. | |
4114 | */ | |
4115 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4116 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4117 | int warned = 0; |
159ec1fc CR |
4118 | |
4119 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4120 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4121 | if (rdev < rdev2 && |
4122 | rdev->bdev->bd_contains == | |
4123 | rdev2->bdev->bd_contains) { | |
4124 | printk(KERN_WARNING | |
4125 | "%s: WARNING: %s appears to be" | |
4126 | " on the same physical disk as" | |
4127 | " %s.\n", | |
4128 | mdname(mddev), | |
4129 | bdevname(rdev->bdev,b), | |
4130 | bdevname(rdev2->bdev,b2)); | |
4131 | warned = 1; | |
4132 | } | |
4133 | } | |
159ec1fc | 4134 | |
7dd5e7c3 N |
4135 | if (warned) |
4136 | printk(KERN_WARNING | |
4137 | "True protection against single-disk" | |
4138 | " failure might be compromised.\n"); | |
4139 | } | |
4140 | ||
657390d2 | 4141 | mddev->recovery = 0; |
58c0fed4 AN |
4142 | /* may be over-ridden by personality */ |
4143 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4144 | ||
a9701a30 | 4145 | mddev->barriers_work = 1; |
6ff8d8ec | 4146 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4147 | |
f91de92e N |
4148 | if (start_readonly) |
4149 | mddev->ro = 2; /* read-only, but switch on first write */ | |
4150 | ||
b15c2e57 | 4151 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4152 | if (err) |
4153 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4154 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4155 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4156 | " but 'external_size' not in effect?\n", __func__); | |
4157 | printk(KERN_ERR | |
4158 | "md: invalid array_size %llu > default size %llu\n", | |
4159 | (unsigned long long)mddev->array_sectors / 2, | |
4160 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4161 | err = -EINVAL; | |
4162 | mddev->pers->stop(mddev); | |
4163 | } | |
4164 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4165 | err = bitmap_create(mddev); |
4166 | if (err) { | |
4167 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4168 | mdname(mddev), err); | |
4169 | mddev->pers->stop(mddev); | |
4170 | } | |
4171 | } | |
1da177e4 | 4172 | if (err) { |
1da177e4 LT |
4173 | module_put(mddev->pers->owner); |
4174 | mddev->pers = NULL; | |
32a7627c N |
4175 | bitmap_destroy(mddev); |
4176 | return err; | |
1da177e4 | 4177 | } |
5e55e2f5 N |
4178 | if (mddev->pers->sync_request) { |
4179 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
4180 | printk(KERN_WARNING | |
4181 | "md: cannot register extra attributes for %s\n", | |
4182 | mdname(mddev)); | |
0c3573f1 | 4183 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4184 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4185 | mddev->ro = 0; |
4186 | ||
1da177e4 LT |
4187 | atomic_set(&mddev->writes_pending,0); |
4188 | mddev->safemode = 0; | |
4189 | mddev->safemode_timer.function = md_safemode_timeout; | |
4190 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4191 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4192 | mddev->in_sync = 1; |
86e6ffdd | 4193 | |
159ec1fc | 4194 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4195 | if (rdev->raid_disk >= 0) { |
4196 | char nm[20]; | |
4197 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
4198 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
4199 | printk("md: cannot register %s for %s\n", | |
4200 | nm, mdname(mddev)); | |
86e6ffdd | 4201 | } |
1da177e4 LT |
4202 | |
4203 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4204 | ||
850b2b42 N |
4205 | if (mddev->flags) |
4206 | md_update_sb(mddev, 0); | |
1da177e4 | 4207 | |
f233ea5c | 4208 | set_capacity(disk, mddev->array_sectors); |
1da177e4 | 4209 | |
5fd6c1dc N |
4210 | /* If there is a partially-recovered drive we need to |
4211 | * start recovery here. If we leave it to md_check_recovery, | |
4212 | * it will remove the drives and not do the right thing | |
4213 | */ | |
0b8c9de0 | 4214 | if (mddev->degraded && !mddev->sync_thread) { |
5fd6c1dc | 4215 | int spares = 0; |
159ec1fc | 4216 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
4217 | if (rdev->raid_disk >= 0 && |
4218 | !test_bit(In_sync, &rdev->flags) && | |
4219 | !test_bit(Faulty, &rdev->flags)) | |
4220 | /* complete an interrupted recovery */ | |
4221 | spares++; | |
4222 | if (spares && mddev->pers->sync_request) { | |
4223 | mddev->recovery = 0; | |
4224 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4225 | mddev->sync_thread = md_register_thread(md_do_sync, | |
4226 | mddev, | |
0da3c619 | 4227 | "resync"); |
5fd6c1dc N |
4228 | if (!mddev->sync_thread) { |
4229 | printk(KERN_ERR "%s: could not start resync" | |
4230 | " thread...\n", | |
4231 | mdname(mddev)); | |
4232 | /* leave the spares where they are, it shouldn't hurt */ | |
4233 | mddev->recovery = 0; | |
0b8c9de0 | 4234 | } |
5fd6c1dc N |
4235 | } |
4236 | } | |
0b8c9de0 N |
4237 | md_wakeup_thread(mddev->thread); |
4238 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4239 | |
449aad3e | 4240 | revalidate_disk(mddev->gendisk); |
44ce6294 | 4241 | mddev->changed = 1; |
d7603b7e | 4242 | md_new_event(mddev); |
b62b7590 | 4243 | sysfs_notify_dirent(mddev->sysfs_state); |
0c3573f1 N |
4244 | if (mddev->sysfs_action) |
4245 | sysfs_notify_dirent(mddev->sysfs_action); | |
a99ac971 | 4246 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
ed9e1982 | 4247 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
1da177e4 LT |
4248 | return 0; |
4249 | } | |
4250 | ||
4251 | static int restart_array(mddev_t *mddev) | |
4252 | { | |
4253 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4254 | |
80fab1d7 | 4255 | /* Complain if it has no devices */ |
1da177e4 | 4256 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4257 | return -ENXIO; |
4258 | if (!mddev->pers) | |
4259 | return -EINVAL; | |
4260 | if (!mddev->ro) | |
4261 | return -EBUSY; | |
4262 | mddev->safemode = 0; | |
4263 | mddev->ro = 0; | |
4264 | set_disk_ro(disk, 0); | |
4265 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4266 | mdname(mddev)); | |
4267 | /* Kick recovery or resync if necessary */ | |
4268 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4269 | md_wakeup_thread(mddev->thread); | |
4270 | md_wakeup_thread(mddev->sync_thread); | |
b62b7590 | 4271 | sysfs_notify_dirent(mddev->sysfs_state); |
80fab1d7 | 4272 | return 0; |
1da177e4 LT |
4273 | } |
4274 | ||
acc55e22 N |
4275 | /* similar to deny_write_access, but accounts for our holding a reference |
4276 | * to the file ourselves */ | |
4277 | static int deny_bitmap_write_access(struct file * file) | |
4278 | { | |
4279 | struct inode *inode = file->f_mapping->host; | |
4280 | ||
4281 | spin_lock(&inode->i_lock); | |
4282 | if (atomic_read(&inode->i_writecount) > 1) { | |
4283 | spin_unlock(&inode->i_lock); | |
4284 | return -ETXTBSY; | |
4285 | } | |
4286 | atomic_set(&inode->i_writecount, -1); | |
4287 | spin_unlock(&inode->i_lock); | |
4288 | ||
4289 | return 0; | |
4290 | } | |
4291 | ||
4292 | static void restore_bitmap_write_access(struct file *file) | |
4293 | { | |
4294 | struct inode *inode = file->f_mapping->host; | |
4295 | ||
4296 | spin_lock(&inode->i_lock); | |
4297 | atomic_set(&inode->i_writecount, 1); | |
4298 | spin_unlock(&inode->i_lock); | |
4299 | } | |
4300 | ||
9e653b63 N |
4301 | /* mode: |
4302 | * 0 - completely stop and dis-assemble array | |
4303 | * 1 - switch to readonly | |
4304 | * 2 - stop but do not disassemble array | |
4305 | */ | |
df5b20cf | 4306 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 LT |
4307 | { |
4308 | int err = 0; | |
4309 | struct gendisk *disk = mddev->gendisk; | |
c4647292 | 4310 | mdk_rdev_t *rdev; |
1da177e4 | 4311 | |
c8c00a69 | 4312 | mutex_lock(&mddev->open_mutex); |
f2ea68cf | 4313 | if (atomic_read(&mddev->openers) > is_open) { |
df5b20cf | 4314 | printk("md: %s still in use.\n",mdname(mddev)); |
c8c00a69 N |
4315 | err = -EBUSY; |
4316 | } else if (mddev->pers) { | |
1da177e4 LT |
4317 | |
4318 | if (mddev->sync_thread) { | |
5fd6c1dc | 4319 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 LT |
4320 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4321 | md_unregister_thread(mddev->sync_thread); | |
4322 | mddev->sync_thread = NULL; | |
4323 | } | |
4324 | ||
4325 | del_timer_sync(&mddev->safemode_timer); | |
4326 | ||
9e653b63 N |
4327 | switch(mode) { |
4328 | case 1: /* readonly */ | |
1da177e4 | 4329 | err = -ENXIO; |
f91de92e | 4330 | if (mddev->ro==1) |
1da177e4 LT |
4331 | goto out; |
4332 | mddev->ro = 1; | |
9e653b63 N |
4333 | break; |
4334 | case 0: /* disassemble */ | |
4335 | case 2: /* stop */ | |
6b8b3e8a | 4336 | bitmap_flush(mddev); |
a9701a30 | 4337 | md_super_wait(mddev); |
1da177e4 LT |
4338 | if (mddev->ro) |
4339 | set_disk_ro(disk, 0); | |
409c57f3 | 4340 | |
1da177e4 | 4341 | mddev->pers->stop(mddev); |
d1b5380c N |
4342 | mddev->queue->merge_bvec_fn = NULL; |
4343 | mddev->queue->unplug_fn = NULL; | |
041ae52e | 4344 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4345 | module_put(mddev->pers->owner); |
5fd3a17e DW |
4346 | if (mddev->pers->sync_request) |
4347 | mddev->private = &md_redundancy_group; | |
1da177e4 | 4348 | mddev->pers = NULL; |
4f54b0e9 | 4349 | /* tell userspace to handle 'inactive' */ |
b62b7590 | 4350 | sysfs_notify_dirent(mddev->sysfs_state); |
0d4ca600 | 4351 | |
c4647292 N |
4352 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4353 | if (rdev->raid_disk >= 0) { | |
4354 | char nm[20]; | |
4355 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4356 | sysfs_remove_link(&mddev->kobj, nm); | |
4357 | } | |
4358 | ||
0d4ca600 | 4359 | set_capacity(disk, 0); |
44ce6294 | 4360 | mddev->changed = 1; |
0d4ca600 | 4361 | |
1da177e4 LT |
4362 | if (mddev->ro) |
4363 | mddev->ro = 0; | |
4364 | } | |
850b2b42 | 4365 | if (!mddev->in_sync || mddev->flags) { |
1da177e4 LT |
4366 | /* mark array as shutdown cleanly */ |
4367 | mddev->in_sync = 1; | |
850b2b42 | 4368 | md_update_sb(mddev, 1); |
1da177e4 | 4369 | } |
9e653b63 | 4370 | if (mode == 1) |
1da177e4 | 4371 | set_disk_ro(disk, 1); |
5fd6c1dc | 4372 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
80ffb3cc | 4373 | err = 0; |
1da177e4 | 4374 | } |
c8c00a69 N |
4375 | out: |
4376 | mutex_unlock(&mddev->open_mutex); | |
4377 | if (err) | |
4378 | return err; | |
1da177e4 LT |
4379 | /* |
4380 | * Free resources if final stop | |
4381 | */ | |
9e653b63 | 4382 | if (mode == 0) { |
0d4ca600 | 4383 | |
1da177e4 LT |
4384 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4385 | ||
978f946b N |
4386 | bitmap_destroy(mddev); |
4387 | if (mddev->bitmap_file) { | |
acc55e22 | 4388 | restore_bitmap_write_access(mddev->bitmap_file); |
978f946b N |
4389 | fput(mddev->bitmap_file); |
4390 | mddev->bitmap_file = NULL; | |
4391 | } | |
4392 | mddev->bitmap_offset = 0; | |
4393 | ||
177a99b2 | 4394 | /* make sure all md_delayed_delete calls have finished */ |
5792a285 N |
4395 | flush_scheduled_work(); |
4396 | ||
1da177e4 LT |
4397 | export_array(mddev); |
4398 | ||
f233ea5c | 4399 | mddev->array_sectors = 0; |
b522adcd | 4400 | mddev->external_size = 0; |
58c0fed4 | 4401 | mddev->dev_sectors = 0; |
9e653b63 | 4402 | mddev->raid_disks = 0; |
a94213b1 | 4403 | mddev->recovery_cp = 0; |
5e96ee65 | 4404 | mddev->resync_min = 0; |
c6207277 | 4405 | mddev->resync_max = MaxSector; |
08a02ecd | 4406 | mddev->reshape_position = MaxSector; |
e691063a | 4407 | mddev->external = 0; |
1ec4a939 | 4408 | mddev->persistent = 0; |
d897dbf9 N |
4409 | mddev->level = LEVEL_NONE; |
4410 | mddev->clevel[0] = 0; | |
4411 | mddev->flags = 0; | |
4412 | mddev->ro = 0; | |
4413 | mddev->metadata_type[0] = 0; | |
9d8f0363 | 4414 | mddev->chunk_sectors = 0; |
d897dbf9 N |
4415 | mddev->ctime = mddev->utime = 0; |
4416 | mddev->layout = 0; | |
4417 | mddev->max_disks = 0; | |
4418 | mddev->events = 0; | |
4419 | mddev->delta_disks = 0; | |
4420 | mddev->new_level = LEVEL_NONE; | |
4421 | mddev->new_layout = 0; | |
664e7c41 | 4422 | mddev->new_chunk_sectors = 0; |
d897dbf9 N |
4423 | mddev->curr_resync = 0; |
4424 | mddev->resync_mismatches = 0; | |
4425 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4426 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4427 | mddev->recovery = 0; | |
4428 | mddev->in_sync = 0; | |
4429 | mddev->changed = 0; | |
4430 | mddev->degraded = 0; | |
4431 | mddev->barriers_work = 0; | |
4432 | mddev->safemode = 0; | |
934d9c23 | 4433 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4434 | if (mddev->hold_active == UNTIL_STOP) |
4435 | mddev->hold_active = 0; | |
9e653b63 | 4436 | |
a8a55c38 | 4437 | } else if (mddev->pers) |
1da177e4 LT |
4438 | printk(KERN_INFO "md: %s switched to read-only mode.\n", |
4439 | mdname(mddev)); | |
4440 | err = 0; | |
3f9d99c1 | 4441 | blk_integrity_unregister(disk); |
d7603b7e | 4442 | md_new_event(mddev); |
b62b7590 | 4443 | sysfs_notify_dirent(mddev->sysfs_state); |
1da177e4 LT |
4444 | return err; |
4445 | } | |
4446 | ||
fdee8ae4 | 4447 | #ifndef MODULE |
1da177e4 LT |
4448 | static void autorun_array(mddev_t *mddev) |
4449 | { | |
4450 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4451 | int err; |
4452 | ||
a757e64c | 4453 | if (list_empty(&mddev->disks)) |
1da177e4 | 4454 | return; |
1da177e4 LT |
4455 | |
4456 | printk(KERN_INFO "md: running: "); | |
4457 | ||
159ec1fc | 4458 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4459 | char b[BDEVNAME_SIZE]; |
4460 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4461 | } | |
4462 | printk("\n"); | |
4463 | ||
d710e138 | 4464 | err = do_md_run(mddev); |
1da177e4 LT |
4465 | if (err) { |
4466 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4467 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4468 | } |
4469 | } | |
4470 | ||
4471 | /* | |
4472 | * lets try to run arrays based on all disks that have arrived | |
4473 | * until now. (those are in pending_raid_disks) | |
4474 | * | |
4475 | * the method: pick the first pending disk, collect all disks with | |
4476 | * the same UUID, remove all from the pending list and put them into | |
4477 | * the 'same_array' list. Then order this list based on superblock | |
4478 | * update time (freshest comes first), kick out 'old' disks and | |
4479 | * compare superblocks. If everything's fine then run it. | |
4480 | * | |
4481 | * If "unit" is allocated, then bump its reference count | |
4482 | */ | |
4483 | static void autorun_devices(int part) | |
4484 | { | |
159ec1fc | 4485 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4486 | mddev_t *mddev; |
4487 | char b[BDEVNAME_SIZE]; | |
4488 | ||
4489 | printk(KERN_INFO "md: autorun ...\n"); | |
4490 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4491 | int unit; |
1da177e4 | 4492 | dev_t dev; |
ad01c9e3 | 4493 | LIST_HEAD(candidates); |
1da177e4 LT |
4494 | rdev0 = list_entry(pending_raid_disks.next, |
4495 | mdk_rdev_t, same_set); | |
4496 | ||
4497 | printk(KERN_INFO "md: considering %s ...\n", | |
4498 | bdevname(rdev0->bdev,b)); | |
4499 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4500 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4501 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4502 | printk(KERN_INFO "md: adding %s ...\n", | |
4503 | bdevname(rdev->bdev,b)); | |
4504 | list_move(&rdev->same_set, &candidates); | |
4505 | } | |
4506 | /* | |
4507 | * now we have a set of devices, with all of them having | |
4508 | * mostly sane superblocks. It's time to allocate the | |
4509 | * mddev. | |
4510 | */ | |
e8703fe1 N |
4511 | if (part) { |
4512 | dev = MKDEV(mdp_major, | |
4513 | rdev0->preferred_minor << MdpMinorShift); | |
4514 | unit = MINOR(dev) >> MdpMinorShift; | |
4515 | } else { | |
4516 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4517 | unit = MINOR(dev); | |
4518 | } | |
4519 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4520 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4521 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4522 | break; | |
4523 | } | |
1da177e4 LT |
4524 | |
4525 | md_probe(dev, NULL, NULL); | |
4526 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4527 | if (!mddev || !mddev->gendisk) { |
4528 | if (mddev) | |
4529 | mddev_put(mddev); | |
4530 | printk(KERN_ERR | |
1da177e4 LT |
4531 | "md: cannot allocate memory for md drive.\n"); |
4532 | break; | |
4533 | } | |
4534 | if (mddev_lock(mddev)) | |
4535 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4536 | mdname(mddev)); | |
4537 | else if (mddev->raid_disks || mddev->major_version | |
4538 | || !list_empty(&mddev->disks)) { | |
4539 | printk(KERN_WARNING | |
4540 | "md: %s already running, cannot run %s\n", | |
4541 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4542 | mddev_unlock(mddev); | |
4543 | } else { | |
4544 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4545 | mddev->persistent = 1; |
159ec1fc | 4546 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4547 | list_del_init(&rdev->same_set); |
4548 | if (bind_rdev_to_array(rdev, mddev)) | |
4549 | export_rdev(rdev); | |
4550 | } | |
4551 | autorun_array(mddev); | |
4552 | mddev_unlock(mddev); | |
4553 | } | |
4554 | /* on success, candidates will be empty, on error | |
4555 | * it won't... | |
4556 | */ | |
159ec1fc | 4557 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4558 | list_del_init(&rdev->same_set); |
1da177e4 | 4559 | export_rdev(rdev); |
4b80991c | 4560 | } |
1da177e4 LT |
4561 | mddev_put(mddev); |
4562 | } | |
4563 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4564 | } | |
fdee8ae4 | 4565 | #endif /* !MODULE */ |
1da177e4 | 4566 | |
1da177e4 LT |
4567 | static int get_version(void __user * arg) |
4568 | { | |
4569 | mdu_version_t ver; | |
4570 | ||
4571 | ver.major = MD_MAJOR_VERSION; | |
4572 | ver.minor = MD_MINOR_VERSION; | |
4573 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4574 | ||
4575 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4576 | return -EFAULT; | |
4577 | ||
4578 | return 0; | |
4579 | } | |
4580 | ||
4581 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4582 | { | |
4583 | mdu_array_info_t info; | |
a9f326eb | 4584 | int nr,working,insync,failed,spare; |
1da177e4 | 4585 | mdk_rdev_t *rdev; |
1da177e4 | 4586 | |
a9f326eb | 4587 | nr=working=insync=failed=spare=0; |
159ec1fc | 4588 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4589 | nr++; |
b2d444d7 | 4590 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4591 | failed++; |
4592 | else { | |
4593 | working++; | |
b2d444d7 | 4594 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 4595 | insync++; |
1da177e4 LT |
4596 | else |
4597 | spare++; | |
4598 | } | |
4599 | } | |
4600 | ||
4601 | info.major_version = mddev->major_version; | |
4602 | info.minor_version = mddev->minor_version; | |
4603 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
4604 | info.ctime = mddev->ctime; | |
4605 | info.level = mddev->level; | |
58c0fed4 AN |
4606 | info.size = mddev->dev_sectors / 2; |
4607 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 4608 | info.size = -1; |
1da177e4 LT |
4609 | info.nr_disks = nr; |
4610 | info.raid_disks = mddev->raid_disks; | |
4611 | info.md_minor = mddev->md_minor; | |
4612 | info.not_persistent= !mddev->persistent; | |
4613 | ||
4614 | info.utime = mddev->utime; | |
4615 | info.state = 0; | |
4616 | if (mddev->in_sync) | |
4617 | info.state = (1<<MD_SB_CLEAN); | |
36fa3063 N |
4618 | if (mddev->bitmap && mddev->bitmap_offset) |
4619 | info.state = (1<<MD_SB_BITMAP_PRESENT); | |
a9f326eb | 4620 | info.active_disks = insync; |
1da177e4 LT |
4621 | info.working_disks = working; |
4622 | info.failed_disks = failed; | |
4623 | info.spare_disks = spare; | |
4624 | ||
4625 | info.layout = mddev->layout; | |
9d8f0363 | 4626 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
4627 | |
4628 | if (copy_to_user(arg, &info, sizeof(info))) | |
4629 | return -EFAULT; | |
4630 | ||
4631 | return 0; | |
4632 | } | |
4633 | ||
87162a28 | 4634 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
4635 | { |
4636 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
4637 | char *ptr, *buf = NULL; | |
4638 | int err = -ENOMEM; | |
4639 | ||
b5470dc5 DW |
4640 | if (md_allow_write(mddev)) |
4641 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
4642 | else | |
4643 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 4644 | |
32a7627c N |
4645 | if (!file) |
4646 | goto out; | |
4647 | ||
4648 | /* bitmap disabled, zero the first byte and copy out */ | |
4649 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
4650 | file->pathname[0] = '\0'; | |
4651 | goto copy_out; | |
4652 | } | |
4653 | ||
4654 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
4655 | if (!buf) | |
4656 | goto out; | |
4657 | ||
6bcfd601 CH |
4658 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
4659 | if (IS_ERR(ptr)) | |
32a7627c N |
4660 | goto out; |
4661 | ||
4662 | strcpy(file->pathname, ptr); | |
4663 | ||
4664 | copy_out: | |
4665 | err = 0; | |
4666 | if (copy_to_user(arg, file, sizeof(*file))) | |
4667 | err = -EFAULT; | |
4668 | out: | |
4669 | kfree(buf); | |
4670 | kfree(file); | |
4671 | return err; | |
4672 | } | |
4673 | ||
1da177e4 LT |
4674 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
4675 | { | |
4676 | mdu_disk_info_t info; | |
1da177e4 LT |
4677 | mdk_rdev_t *rdev; |
4678 | ||
4679 | if (copy_from_user(&info, arg, sizeof(info))) | |
4680 | return -EFAULT; | |
4681 | ||
26ef379f | 4682 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
4683 | if (rdev) { |
4684 | info.major = MAJOR(rdev->bdev->bd_dev); | |
4685 | info.minor = MINOR(rdev->bdev->bd_dev); | |
4686 | info.raid_disk = rdev->raid_disk; | |
4687 | info.state = 0; | |
b2d444d7 | 4688 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4689 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 4690 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4691 | info.state |= (1<<MD_DISK_ACTIVE); |
4692 | info.state |= (1<<MD_DISK_SYNC); | |
4693 | } | |
8ddf9efe N |
4694 | if (test_bit(WriteMostly, &rdev->flags)) |
4695 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
4696 | } else { |
4697 | info.major = info.minor = 0; | |
4698 | info.raid_disk = -1; | |
4699 | info.state = (1<<MD_DISK_REMOVED); | |
4700 | } | |
4701 | ||
4702 | if (copy_to_user(arg, &info, sizeof(info))) | |
4703 | return -EFAULT; | |
4704 | ||
4705 | return 0; | |
4706 | } | |
4707 | ||
4708 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
4709 | { | |
4710 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4711 | mdk_rdev_t *rdev; | |
4712 | dev_t dev = MKDEV(info->major,info->minor); | |
4713 | ||
4714 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
4715 | return -EOVERFLOW; | |
4716 | ||
4717 | if (!mddev->raid_disks) { | |
4718 | int err; | |
4719 | /* expecting a device which has a superblock */ | |
4720 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
4721 | if (IS_ERR(rdev)) { | |
4722 | printk(KERN_WARNING | |
4723 | "md: md_import_device returned %ld\n", | |
4724 | PTR_ERR(rdev)); | |
4725 | return PTR_ERR(rdev); | |
4726 | } | |
4727 | if (!list_empty(&mddev->disks)) { | |
4728 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
4729 | mdk_rdev_t, same_set); | |
a9f326eb | 4730 | err = super_types[mddev->major_version] |
1da177e4 LT |
4731 | .load_super(rdev, rdev0, mddev->minor_version); |
4732 | if (err < 0) { | |
4733 | printk(KERN_WARNING | |
4734 | "md: %s has different UUID to %s\n", | |
4735 | bdevname(rdev->bdev,b), | |
4736 | bdevname(rdev0->bdev,b2)); | |
4737 | export_rdev(rdev); | |
4738 | return -EINVAL; | |
4739 | } | |
4740 | } | |
4741 | err = bind_rdev_to_array(rdev, mddev); | |
4742 | if (err) | |
4743 | export_rdev(rdev); | |
4744 | return err; | |
4745 | } | |
4746 | ||
4747 | /* | |
4748 | * add_new_disk can be used once the array is assembled | |
4749 | * to add "hot spares". They must already have a superblock | |
4750 | * written | |
4751 | */ | |
4752 | if (mddev->pers) { | |
4753 | int err; | |
4754 | if (!mddev->pers->hot_add_disk) { | |
4755 | printk(KERN_WARNING | |
4756 | "%s: personality does not support diskops!\n", | |
4757 | mdname(mddev)); | |
4758 | return -EINVAL; | |
4759 | } | |
7b1e35f6 N |
4760 | if (mddev->persistent) |
4761 | rdev = md_import_device(dev, mddev->major_version, | |
4762 | mddev->minor_version); | |
4763 | else | |
4764 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
4765 | if (IS_ERR(rdev)) { |
4766 | printk(KERN_WARNING | |
4767 | "md: md_import_device returned %ld\n", | |
4768 | PTR_ERR(rdev)); | |
4769 | return PTR_ERR(rdev); | |
4770 | } | |
41158c7e N |
4771 | /* set save_raid_disk if appropriate */ |
4772 | if (!mddev->persistent) { | |
4773 | if (info->state & (1<<MD_DISK_SYNC) && | |
4774 | info->raid_disk < mddev->raid_disks) | |
4775 | rdev->raid_disk = info->raid_disk; | |
4776 | else | |
4777 | rdev->raid_disk = -1; | |
4778 | } else | |
4779 | super_types[mddev->major_version]. | |
4780 | validate_super(mddev, rdev); | |
4781 | rdev->saved_raid_disk = rdev->raid_disk; | |
4782 | ||
b2d444d7 | 4783 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
4784 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4785 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
4786 | else |
4787 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 4788 | |
1da177e4 LT |
4789 | rdev->raid_disk = -1; |
4790 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
4791 | if (!err && !mddev->pers->hot_remove_disk) { |
4792 | /* If there is hot_add_disk but no hot_remove_disk | |
4793 | * then added disks for geometry changes, | |
4794 | * and should be added immediately. | |
4795 | */ | |
4796 | super_types[mddev->major_version]. | |
4797 | validate_super(mddev, rdev); | |
4798 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
4799 | if (err) | |
4800 | unbind_rdev_from_array(rdev); | |
4801 | } | |
1da177e4 LT |
4802 | if (err) |
4803 | export_rdev(rdev); | |
52664732 | 4804 | else |
3c0ee63a | 4805 | sysfs_notify_dirent(rdev->sysfs_state); |
c361777f | 4806 | |
17571284 | 4807 | md_update_sb(mddev, 1); |
72a23c21 NB |
4808 | if (mddev->degraded) |
4809 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 4810 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 4811 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4812 | return err; |
4813 | } | |
4814 | ||
4815 | /* otherwise, add_new_disk is only allowed | |
4816 | * for major_version==0 superblocks | |
4817 | */ | |
4818 | if (mddev->major_version != 0) { | |
4819 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
4820 | mdname(mddev)); | |
4821 | return -EINVAL; | |
4822 | } | |
4823 | ||
4824 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
4825 | int err; | |
d710e138 | 4826 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
4827 | if (IS_ERR(rdev)) { |
4828 | printk(KERN_WARNING | |
4829 | "md: error, md_import_device() returned %ld\n", | |
4830 | PTR_ERR(rdev)); | |
4831 | return PTR_ERR(rdev); | |
4832 | } | |
4833 | rdev->desc_nr = info->number; | |
4834 | if (info->raid_disk < mddev->raid_disks) | |
4835 | rdev->raid_disk = info->raid_disk; | |
4836 | else | |
4837 | rdev->raid_disk = -1; | |
4838 | ||
1da177e4 | 4839 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
4840 | if (info->state & (1<<MD_DISK_SYNC)) |
4841 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 4842 | |
8ddf9efe N |
4843 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4844 | set_bit(WriteMostly, &rdev->flags); | |
4845 | ||
1da177e4 LT |
4846 | if (!mddev->persistent) { |
4847 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
0f420358 | 4848 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 4849 | } else |
0f420358 | 4850 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
8190e754 | 4851 | rdev->sectors = rdev->sb_start; |
1da177e4 | 4852 | |
2bf071bf N |
4853 | err = bind_rdev_to_array(rdev, mddev); |
4854 | if (err) { | |
4855 | export_rdev(rdev); | |
4856 | return err; | |
4857 | } | |
1da177e4 LT |
4858 | } |
4859 | ||
4860 | return 0; | |
4861 | } | |
4862 | ||
4863 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
4864 | { | |
4865 | char b[BDEVNAME_SIZE]; | |
4866 | mdk_rdev_t *rdev; | |
4867 | ||
1da177e4 LT |
4868 | rdev = find_rdev(mddev, dev); |
4869 | if (!rdev) | |
4870 | return -ENXIO; | |
4871 | ||
4872 | if (rdev->raid_disk >= 0) | |
4873 | goto busy; | |
4874 | ||
4875 | kick_rdev_from_array(rdev); | |
850b2b42 | 4876 | md_update_sb(mddev, 1); |
d7603b7e | 4877 | md_new_event(mddev); |
1da177e4 LT |
4878 | |
4879 | return 0; | |
4880 | busy: | |
fdefa4d8 | 4881 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
4882 | bdevname(rdev->bdev,b), mdname(mddev)); |
4883 | return -EBUSY; | |
4884 | } | |
4885 | ||
4886 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
4887 | { | |
4888 | char b[BDEVNAME_SIZE]; | |
4889 | int err; | |
1da177e4 LT |
4890 | mdk_rdev_t *rdev; |
4891 | ||
4892 | if (!mddev->pers) | |
4893 | return -ENODEV; | |
4894 | ||
4895 | if (mddev->major_version != 0) { | |
4896 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
4897 | " version-0 superblocks.\n", | |
4898 | mdname(mddev)); | |
4899 | return -EINVAL; | |
4900 | } | |
4901 | if (!mddev->pers->hot_add_disk) { | |
4902 | printk(KERN_WARNING | |
4903 | "%s: personality does not support diskops!\n", | |
4904 | mdname(mddev)); | |
4905 | return -EINVAL; | |
4906 | } | |
4907 | ||
d710e138 | 4908 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
4909 | if (IS_ERR(rdev)) { |
4910 | printk(KERN_WARNING | |
4911 | "md: error, md_import_device() returned %ld\n", | |
4912 | PTR_ERR(rdev)); | |
4913 | return -EINVAL; | |
4914 | } | |
4915 | ||
4916 | if (mddev->persistent) | |
0f420358 | 4917 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 4918 | else |
0f420358 | 4919 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 4920 | |
8190e754 | 4921 | rdev->sectors = rdev->sb_start; |
1da177e4 | 4922 | |
b2d444d7 | 4923 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
4924 | printk(KERN_WARNING |
4925 | "md: can not hot-add faulty %s disk to %s!\n", | |
4926 | bdevname(rdev->bdev,b), mdname(mddev)); | |
4927 | err = -EINVAL; | |
4928 | goto abort_export; | |
4929 | } | |
b2d444d7 | 4930 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4931 | rdev->desc_nr = -1; |
5842730d | 4932 | rdev->saved_raid_disk = -1; |
2bf071bf N |
4933 | err = bind_rdev_to_array(rdev, mddev); |
4934 | if (err) | |
4935 | goto abort_export; | |
1da177e4 LT |
4936 | |
4937 | /* | |
4938 | * The rest should better be atomic, we can have disk failures | |
4939 | * noticed in interrupt contexts ... | |
4940 | */ | |
4941 | ||
1da177e4 LT |
4942 | rdev->raid_disk = -1; |
4943 | ||
850b2b42 | 4944 | md_update_sb(mddev, 1); |
1da177e4 LT |
4945 | |
4946 | /* | |
4947 | * Kick recovery, maybe this spare has to be added to the | |
4948 | * array immediately. | |
4949 | */ | |
4950 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4951 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 4952 | md_new_event(mddev); |
1da177e4 LT |
4953 | return 0; |
4954 | ||
1da177e4 LT |
4955 | abort_export: |
4956 | export_rdev(rdev); | |
4957 | return err; | |
4958 | } | |
4959 | ||
32a7627c N |
4960 | static int set_bitmap_file(mddev_t *mddev, int fd) |
4961 | { | |
4962 | int err; | |
4963 | ||
36fa3063 N |
4964 | if (mddev->pers) { |
4965 | if (!mddev->pers->quiesce) | |
4966 | return -EBUSY; | |
4967 | if (mddev->recovery || mddev->sync_thread) | |
4968 | return -EBUSY; | |
4969 | /* we should be able to change the bitmap.. */ | |
4970 | } | |
32a7627c | 4971 | |
32a7627c | 4972 | |
36fa3063 N |
4973 | if (fd >= 0) { |
4974 | if (mddev->bitmap) | |
4975 | return -EEXIST; /* cannot add when bitmap is present */ | |
4976 | mddev->bitmap_file = fget(fd); | |
32a7627c | 4977 | |
36fa3063 N |
4978 | if (mddev->bitmap_file == NULL) { |
4979 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", | |
4980 | mdname(mddev)); | |
4981 | return -EBADF; | |
4982 | } | |
4983 | ||
4984 | err = deny_bitmap_write_access(mddev->bitmap_file); | |
4985 | if (err) { | |
4986 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
4987 | mdname(mddev)); | |
4988 | fput(mddev->bitmap_file); | |
4989 | mddev->bitmap_file = NULL; | |
4990 | return err; | |
4991 | } | |
a654b9d8 | 4992 | mddev->bitmap_offset = 0; /* file overrides offset */ |
36fa3063 N |
4993 | } else if (mddev->bitmap == NULL) |
4994 | return -ENOENT; /* cannot remove what isn't there */ | |
4995 | err = 0; | |
4996 | if (mddev->pers) { | |
4997 | mddev->pers->quiesce(mddev, 1); | |
4998 | if (fd >= 0) | |
4999 | err = bitmap_create(mddev); | |
d7375ab3 | 5000 | if (fd < 0 || err) { |
36fa3063 | 5001 | bitmap_destroy(mddev); |
d7375ab3 N |
5002 | fd = -1; /* make sure to put the file */ |
5003 | } | |
36fa3063 | 5004 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5005 | } |
5006 | if (fd < 0) { | |
acc55e22 N |
5007 | if (mddev->bitmap_file) { |
5008 | restore_bitmap_write_access(mddev->bitmap_file); | |
36fa3063 | 5009 | fput(mddev->bitmap_file); |
acc55e22 | 5010 | } |
36fa3063 N |
5011 | mddev->bitmap_file = NULL; |
5012 | } | |
5013 | ||
32a7627c N |
5014 | return err; |
5015 | } | |
5016 | ||
1da177e4 LT |
5017 | /* |
5018 | * set_array_info is used two different ways | |
5019 | * The original usage is when creating a new array. | |
5020 | * In this usage, raid_disks is > 0 and it together with | |
5021 | * level, size, not_persistent,layout,chunksize determine the | |
5022 | * shape of the array. | |
5023 | * This will always create an array with a type-0.90.0 superblock. | |
5024 | * The newer usage is when assembling an array. | |
5025 | * In this case raid_disks will be 0, and the major_version field is | |
5026 | * use to determine which style super-blocks are to be found on the devices. | |
5027 | * The minor and patch _version numbers are also kept incase the | |
5028 | * super_block handler wishes to interpret them. | |
5029 | */ | |
5030 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5031 | { | |
5032 | ||
5033 | if (info->raid_disks == 0) { | |
5034 | /* just setting version number for superblock loading */ | |
5035 | if (info->major_version < 0 || | |
50511da3 | 5036 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5037 | super_types[info->major_version].name == NULL) { |
5038 | /* maybe try to auto-load a module? */ | |
5039 | printk(KERN_INFO | |
5040 | "md: superblock version %d not known\n", | |
5041 | info->major_version); | |
5042 | return -EINVAL; | |
5043 | } | |
5044 | mddev->major_version = info->major_version; | |
5045 | mddev->minor_version = info->minor_version; | |
5046 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5047 | mddev->persistent = !info->not_persistent; |
1da177e4 LT |
5048 | return 0; |
5049 | } | |
5050 | mddev->major_version = MD_MAJOR_VERSION; | |
5051 | mddev->minor_version = MD_MINOR_VERSION; | |
5052 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5053 | mddev->ctime = get_seconds(); | |
5054 | ||
5055 | mddev->level = info->level; | |
17115e03 | 5056 | mddev->clevel[0] = 0; |
58c0fed4 | 5057 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5058 | mddev->raid_disks = info->raid_disks; |
5059 | /* don't set md_minor, it is determined by which /dev/md* was | |
5060 | * openned | |
5061 | */ | |
5062 | if (info->state & (1<<MD_SB_CLEAN)) | |
5063 | mddev->recovery_cp = MaxSector; | |
5064 | else | |
5065 | mddev->recovery_cp = 0; | |
5066 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5067 | mddev->external = 0; |
1da177e4 LT |
5068 | |
5069 | mddev->layout = info->layout; | |
9d8f0363 | 5070 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5071 | |
5072 | mddev->max_disks = MD_SB_DISKS; | |
5073 | ||
e691063a N |
5074 | if (mddev->persistent) |
5075 | mddev->flags = 0; | |
850b2b42 | 5076 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5077 | |
b2a2703c N |
5078 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
5079 | mddev->bitmap_offset = 0; | |
5080 | ||
f6705578 N |
5081 | mddev->reshape_position = MaxSector; |
5082 | ||
1da177e4 LT |
5083 | /* |
5084 | * Generate a 128 bit UUID | |
5085 | */ | |
5086 | get_random_bytes(mddev->uuid, 16); | |
5087 | ||
f6705578 | 5088 | mddev->new_level = mddev->level; |
664e7c41 | 5089 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5090 | mddev->new_layout = mddev->layout; |
5091 | mddev->delta_disks = 0; | |
5092 | ||
1da177e4 LT |
5093 | return 0; |
5094 | } | |
5095 | ||
1f403624 DW |
5096 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5097 | { | |
b522adcd DW |
5098 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5099 | ||
5100 | if (mddev->external_size) | |
5101 | return; | |
5102 | ||
1f403624 DW |
5103 | mddev->array_sectors = array_sectors; |
5104 | } | |
5105 | EXPORT_SYMBOL(md_set_array_sectors); | |
5106 | ||
d71f9f88 | 5107 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5108 | { |
159ec1fc | 5109 | mdk_rdev_t *rdev; |
a35b0d69 | 5110 | int rv; |
d71f9f88 | 5111 | int fit = (num_sectors == 0); |
a35b0d69 N |
5112 | |
5113 | if (mddev->pers->resize == NULL) | |
5114 | return -EINVAL; | |
d71f9f88 AN |
5115 | /* The "num_sectors" is the number of sectors of each device that |
5116 | * is used. This can only make sense for arrays with redundancy. | |
5117 | * linear and raid0 always use whatever space is available. We can only | |
5118 | * consider changing this number if no resync or reconstruction is | |
5119 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5120 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5121 | * of each device. If num_sectors is zero, we find the largest size |
5122 | * that fits. | |
5123 | ||
a35b0d69 N |
5124 | */ |
5125 | if (mddev->sync_thread) | |
5126 | return -EBUSY; | |
dba034ee N |
5127 | if (mddev->bitmap) |
5128 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5129 | * grow, and re-add. | |
5130 | */ | |
5131 | return -EBUSY; | |
159ec1fc | 5132 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5133 | sector_t avail = rdev->sectors; |
01ab5662 | 5134 | |
d71f9f88 AN |
5135 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5136 | num_sectors = avail; | |
5137 | if (avail < num_sectors) | |
a35b0d69 N |
5138 | return -ENOSPC; |
5139 | } | |
d71f9f88 | 5140 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5141 | if (!rv) |
5142 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5143 | return rv; |
5144 | } | |
5145 | ||
da943b99 N |
5146 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5147 | { | |
5148 | int rv; | |
5149 | /* change the number of raid disks */ | |
63c70c4f | 5150 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5151 | return -EINVAL; |
5152 | if (raid_disks <= 0 || | |
5153 | raid_disks >= mddev->max_disks) | |
5154 | return -EINVAL; | |
63c70c4f | 5155 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5156 | return -EBUSY; |
63c70c4f N |
5157 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5158 | ||
5159 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
5160 | return rv; |
5161 | } | |
5162 | ||
5163 | ||
1da177e4 LT |
5164 | /* |
5165 | * update_array_info is used to change the configuration of an | |
5166 | * on-line array. | |
5167 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5168 | * fields in the info are checked against the array. | |
5169 | * Any differences that cannot be handled will cause an error. | |
5170 | * Normally, only one change can be managed at a time. | |
5171 | */ | |
5172 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5173 | { | |
5174 | int rv = 0; | |
5175 | int cnt = 0; | |
36fa3063 N |
5176 | int state = 0; |
5177 | ||
5178 | /* calculate expected state,ignoring low bits */ | |
5179 | if (mddev->bitmap && mddev->bitmap_offset) | |
5180 | state |= (1 << MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
5181 | |
5182 | if (mddev->major_version != info->major_version || | |
5183 | mddev->minor_version != info->minor_version || | |
5184 | /* mddev->patch_version != info->patch_version || */ | |
5185 | mddev->ctime != info->ctime || | |
5186 | mddev->level != info->level || | |
5187 | /* mddev->layout != info->layout || */ | |
5188 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5189 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5190 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5191 | ((state^info->state) & 0xfffffe00) | |
5192 | ) | |
1da177e4 LT |
5193 | return -EINVAL; |
5194 | /* Check there is only one change */ | |
58c0fed4 AN |
5195 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5196 | cnt++; | |
5197 | if (mddev->raid_disks != info->raid_disks) | |
5198 | cnt++; | |
5199 | if (mddev->layout != info->layout) | |
5200 | cnt++; | |
5201 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5202 | cnt++; | |
5203 | if (cnt == 0) | |
5204 | return 0; | |
5205 | if (cnt > 1) | |
5206 | return -EINVAL; | |
1da177e4 LT |
5207 | |
5208 | if (mddev->layout != info->layout) { | |
5209 | /* Change layout | |
5210 | * we don't need to do anything at the md level, the | |
5211 | * personality will take care of it all. | |
5212 | */ | |
50ac168a | 5213 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5214 | return -EINVAL; |
597a711b N |
5215 | else { |
5216 | mddev->new_layout = info->layout; | |
50ac168a | 5217 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5218 | if (rv) |
5219 | mddev->new_layout = mddev->layout; | |
5220 | return rv; | |
5221 | } | |
1da177e4 | 5222 | } |
58c0fed4 | 5223 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5224 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5225 | |
da943b99 N |
5226 | if (mddev->raid_disks != info->raid_disks) |
5227 | rv = update_raid_disks(mddev, info->raid_disks); | |
5228 | ||
36fa3063 N |
5229 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5230 | if (mddev->pers->quiesce == NULL) | |
5231 | return -EINVAL; | |
5232 | if (mddev->recovery || mddev->sync_thread) | |
5233 | return -EBUSY; | |
5234 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5235 | /* add the bitmap */ | |
5236 | if (mddev->bitmap) | |
5237 | return -EEXIST; | |
5238 | if (mddev->default_bitmap_offset == 0) | |
5239 | return -EINVAL; | |
5240 | mddev->bitmap_offset = mddev->default_bitmap_offset; | |
5241 | mddev->pers->quiesce(mddev, 1); | |
5242 | rv = bitmap_create(mddev); | |
5243 | if (rv) | |
5244 | bitmap_destroy(mddev); | |
5245 | mddev->pers->quiesce(mddev, 0); | |
5246 | } else { | |
5247 | /* remove the bitmap */ | |
5248 | if (!mddev->bitmap) | |
5249 | return -ENOENT; | |
5250 | if (mddev->bitmap->file) | |
5251 | return -EINVAL; | |
5252 | mddev->pers->quiesce(mddev, 1); | |
5253 | bitmap_destroy(mddev); | |
5254 | mddev->pers->quiesce(mddev, 0); | |
5255 | mddev->bitmap_offset = 0; | |
5256 | } | |
5257 | } | |
850b2b42 | 5258 | md_update_sb(mddev, 1); |
1da177e4 LT |
5259 | return rv; |
5260 | } | |
5261 | ||
5262 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5263 | { | |
5264 | mdk_rdev_t *rdev; | |
5265 | ||
5266 | if (mddev->pers == NULL) | |
5267 | return -ENODEV; | |
5268 | ||
5269 | rdev = find_rdev(mddev, dev); | |
5270 | if (!rdev) | |
5271 | return -ENODEV; | |
5272 | ||
5273 | md_error(mddev, rdev); | |
5274 | return 0; | |
5275 | } | |
5276 | ||
2f9618ce AN |
5277 | /* |
5278 | * We have a problem here : there is no easy way to give a CHS | |
5279 | * virtual geometry. We currently pretend that we have a 2 heads | |
5280 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5281 | * dosfs just mad... ;-) | |
5282 | */ | |
a885c8c4 CH |
5283 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5284 | { | |
5285 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5286 | ||
5287 | geo->heads = 2; | |
5288 | geo->sectors = 4; | |
5289 | geo->cylinders = get_capacity(mddev->gendisk) / 8; | |
5290 | return 0; | |
5291 | } | |
5292 | ||
a39907fa | 5293 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5294 | unsigned int cmd, unsigned long arg) |
5295 | { | |
5296 | int err = 0; | |
5297 | void __user *argp = (void __user *)arg; | |
1da177e4 LT |
5298 | mddev_t *mddev = NULL; |
5299 | ||
5300 | if (!capable(CAP_SYS_ADMIN)) | |
5301 | return -EACCES; | |
5302 | ||
5303 | /* | |
5304 | * Commands dealing with the RAID driver but not any | |
5305 | * particular array: | |
5306 | */ | |
5307 | switch (cmd) | |
5308 | { | |
5309 | case RAID_VERSION: | |
5310 | err = get_version(argp); | |
5311 | goto done; | |
5312 | ||
5313 | case PRINT_RAID_DEBUG: | |
5314 | err = 0; | |
5315 | md_print_devices(); | |
5316 | goto done; | |
5317 | ||
5318 | #ifndef MODULE | |
5319 | case RAID_AUTORUN: | |
5320 | err = 0; | |
5321 | autostart_arrays(arg); | |
5322 | goto done; | |
5323 | #endif | |
5324 | default:; | |
5325 | } | |
5326 | ||
5327 | /* | |
5328 | * Commands creating/starting a new array: | |
5329 | */ | |
5330 | ||
a39907fa | 5331 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5332 | |
5333 | if (!mddev) { | |
5334 | BUG(); | |
5335 | goto abort; | |
5336 | } | |
5337 | ||
1da177e4 LT |
5338 | err = mddev_lock(mddev); |
5339 | if (err) { | |
5340 | printk(KERN_INFO | |
5341 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5342 | err, cmd); | |
5343 | goto abort; | |
5344 | } | |
5345 | ||
5346 | switch (cmd) | |
5347 | { | |
5348 | case SET_ARRAY_INFO: | |
5349 | { | |
5350 | mdu_array_info_t info; | |
5351 | if (!arg) | |
5352 | memset(&info, 0, sizeof(info)); | |
5353 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5354 | err = -EFAULT; | |
5355 | goto abort_unlock; | |
5356 | } | |
5357 | if (mddev->pers) { | |
5358 | err = update_array_info(mddev, &info); | |
5359 | if (err) { | |
5360 | printk(KERN_WARNING "md: couldn't update" | |
5361 | " array info. %d\n", err); | |
5362 | goto abort_unlock; | |
5363 | } | |
5364 | goto done_unlock; | |
5365 | } | |
5366 | if (!list_empty(&mddev->disks)) { | |
5367 | printk(KERN_WARNING | |
5368 | "md: array %s already has disks!\n", | |
5369 | mdname(mddev)); | |
5370 | err = -EBUSY; | |
5371 | goto abort_unlock; | |
5372 | } | |
5373 | if (mddev->raid_disks) { | |
5374 | printk(KERN_WARNING | |
5375 | "md: array %s already initialised!\n", | |
5376 | mdname(mddev)); | |
5377 | err = -EBUSY; | |
5378 | goto abort_unlock; | |
5379 | } | |
5380 | err = set_array_info(mddev, &info); | |
5381 | if (err) { | |
5382 | printk(KERN_WARNING "md: couldn't set" | |
5383 | " array info. %d\n", err); | |
5384 | goto abort_unlock; | |
5385 | } | |
5386 | } | |
5387 | goto done_unlock; | |
5388 | ||
5389 | default:; | |
5390 | } | |
5391 | ||
5392 | /* | |
5393 | * Commands querying/configuring an existing array: | |
5394 | */ | |
32a7627c | 5395 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5396 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5397 | if ((!mddev->raid_disks && !mddev->external) |
5398 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5399 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5400 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5401 | err = -ENODEV; |
5402 | goto abort_unlock; | |
5403 | } | |
5404 | ||
5405 | /* | |
5406 | * Commands even a read-only array can execute: | |
5407 | */ | |
5408 | switch (cmd) | |
5409 | { | |
5410 | case GET_ARRAY_INFO: | |
5411 | err = get_array_info(mddev, argp); | |
5412 | goto done_unlock; | |
5413 | ||
32a7627c | 5414 | case GET_BITMAP_FILE: |
87162a28 | 5415 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5416 | goto done_unlock; |
5417 | ||
1da177e4 LT |
5418 | case GET_DISK_INFO: |
5419 | err = get_disk_info(mddev, argp); | |
5420 | goto done_unlock; | |
5421 | ||
5422 | case RESTART_ARRAY_RW: | |
5423 | err = restart_array(mddev); | |
5424 | goto done_unlock; | |
5425 | ||
5426 | case STOP_ARRAY: | |
d710e138 | 5427 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5428 | goto done_unlock; |
5429 | ||
5430 | case STOP_ARRAY_RO: | |
d710e138 | 5431 | err = do_md_stop(mddev, 1, 1); |
1da177e4 LT |
5432 | goto done_unlock; |
5433 | ||
1da177e4 LT |
5434 | } |
5435 | ||
5436 | /* | |
5437 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5438 | * superblock, so we do not allow them on read-only arrays. |
5439 | * However non-MD ioctls (e.g. get-size) will still come through | |
5440 | * here and hit the 'default' below, so only disallow | |
5441 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5442 | */ |
bb57fc64 | 5443 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5444 | if (mddev->ro == 2) { |
5445 | mddev->ro = 0; | |
b62b7590 | 5446 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 NB |
5447 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5448 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5449 | } else { |
5450 | err = -EROFS; | |
5451 | goto abort_unlock; | |
5452 | } | |
1da177e4 LT |
5453 | } |
5454 | ||
5455 | switch (cmd) | |
5456 | { | |
5457 | case ADD_NEW_DISK: | |
5458 | { | |
5459 | mdu_disk_info_t info; | |
5460 | if (copy_from_user(&info, argp, sizeof(info))) | |
5461 | err = -EFAULT; | |
5462 | else | |
5463 | err = add_new_disk(mddev, &info); | |
5464 | goto done_unlock; | |
5465 | } | |
5466 | ||
5467 | case HOT_REMOVE_DISK: | |
5468 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5469 | goto done_unlock; | |
5470 | ||
5471 | case HOT_ADD_DISK: | |
5472 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5473 | goto done_unlock; | |
5474 | ||
5475 | case SET_DISK_FAULTY: | |
5476 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5477 | goto done_unlock; | |
5478 | ||
5479 | case RUN_ARRAY: | |
d710e138 | 5480 | err = do_md_run(mddev); |
1da177e4 LT |
5481 | goto done_unlock; |
5482 | ||
32a7627c N |
5483 | case SET_BITMAP_FILE: |
5484 | err = set_bitmap_file(mddev, (int)arg); | |
5485 | goto done_unlock; | |
5486 | ||
1da177e4 | 5487 | default: |
1da177e4 LT |
5488 | err = -EINVAL; |
5489 | goto abort_unlock; | |
5490 | } | |
5491 | ||
5492 | done_unlock: | |
5493 | abort_unlock: | |
d3374825 N |
5494 | if (mddev->hold_active == UNTIL_IOCTL && |
5495 | err != -EINVAL) | |
5496 | mddev->hold_active = 0; | |
1da177e4 LT |
5497 | mddev_unlock(mddev); |
5498 | ||
5499 | return err; | |
5500 | done: | |
5501 | if (err) | |
5502 | MD_BUG(); | |
5503 | abort: | |
5504 | return err; | |
5505 | } | |
5506 | ||
a39907fa | 5507 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
5508 | { |
5509 | /* | |
5510 | * Succeed if we can lock the mddev, which confirms that | |
5511 | * it isn't being stopped right now. | |
5512 | */ | |
d3374825 | 5513 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
5514 | int err; |
5515 | ||
d3374825 N |
5516 | if (mddev->gendisk != bdev->bd_disk) { |
5517 | /* we are racing with mddev_put which is discarding this | |
5518 | * bd_disk. | |
5519 | */ | |
5520 | mddev_put(mddev); | |
5521 | /* Wait until bdev->bd_disk is definitely gone */ | |
5522 | flush_scheduled_work(); | |
5523 | /* Then retry the open from the top */ | |
5524 | return -ERESTARTSYS; | |
5525 | } | |
5526 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
5527 | ||
c8c00a69 | 5528 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
5529 | goto out; |
5530 | ||
5531 | err = 0; | |
f2ea68cf | 5532 | atomic_inc(&mddev->openers); |
c8c00a69 | 5533 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 5534 | |
a39907fa | 5535 | check_disk_change(bdev); |
1da177e4 LT |
5536 | out: |
5537 | return err; | |
5538 | } | |
5539 | ||
a39907fa | 5540 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 5541 | { |
a39907fa | 5542 | mddev_t *mddev = disk->private_data; |
1da177e4 | 5543 | |
52e5f9d1 | 5544 | BUG_ON(!mddev); |
f2ea68cf | 5545 | atomic_dec(&mddev->openers); |
1da177e4 LT |
5546 | mddev_put(mddev); |
5547 | ||
5548 | return 0; | |
5549 | } | |
5550 | ||
44ce6294 LT |
5551 | static int md_media_changed(struct gendisk *disk) |
5552 | { | |
5553 | mddev_t *mddev = disk->private_data; | |
5554 | ||
5555 | return mddev->changed; | |
5556 | } | |
5557 | ||
5558 | static int md_revalidate(struct gendisk *disk) | |
5559 | { | |
5560 | mddev_t *mddev = disk->private_data; | |
5561 | ||
5562 | mddev->changed = 0; | |
5563 | return 0; | |
5564 | } | |
83d5cde4 | 5565 | static const struct block_device_operations md_fops = |
1da177e4 LT |
5566 | { |
5567 | .owner = THIS_MODULE, | |
a39907fa AV |
5568 | .open = md_open, |
5569 | .release = md_release, | |
b492b852 | 5570 | .ioctl = md_ioctl, |
a885c8c4 | 5571 | .getgeo = md_getgeo, |
44ce6294 LT |
5572 | .media_changed = md_media_changed, |
5573 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
5574 | }; |
5575 | ||
75c96f85 | 5576 | static int md_thread(void * arg) |
1da177e4 LT |
5577 | { |
5578 | mdk_thread_t *thread = arg; | |
5579 | ||
1da177e4 LT |
5580 | /* |
5581 | * md_thread is a 'system-thread', it's priority should be very | |
5582 | * high. We avoid resource deadlocks individually in each | |
5583 | * raid personality. (RAID5 does preallocation) We also use RR and | |
5584 | * the very same RT priority as kswapd, thus we will never get | |
5585 | * into a priority inversion deadlock. | |
5586 | * | |
5587 | * we definitely have to have equal or higher priority than | |
5588 | * bdflush, otherwise bdflush will deadlock if there are too | |
5589 | * many dirty RAID5 blocks. | |
5590 | */ | |
1da177e4 | 5591 | |
6985c43f | 5592 | allow_signal(SIGKILL); |
a6fb0934 | 5593 | while (!kthread_should_stop()) { |
1da177e4 | 5594 | |
93588e22 N |
5595 | /* We need to wait INTERRUPTIBLE so that |
5596 | * we don't add to the load-average. | |
5597 | * That means we need to be sure no signals are | |
5598 | * pending | |
5599 | */ | |
5600 | if (signal_pending(current)) | |
5601 | flush_signals(current); | |
5602 | ||
5603 | wait_event_interruptible_timeout | |
5604 | (thread->wqueue, | |
5605 | test_bit(THREAD_WAKEUP, &thread->flags) | |
5606 | || kthread_should_stop(), | |
5607 | thread->timeout); | |
1da177e4 LT |
5608 | |
5609 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
5610 | ||
787453c2 | 5611 | thread->run(thread->mddev); |
1da177e4 | 5612 | } |
a6fb0934 | 5613 | |
1da177e4 LT |
5614 | return 0; |
5615 | } | |
5616 | ||
5617 | void md_wakeup_thread(mdk_thread_t *thread) | |
5618 | { | |
5619 | if (thread) { | |
5620 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
5621 | set_bit(THREAD_WAKEUP, &thread->flags); | |
5622 | wake_up(&thread->wqueue); | |
5623 | } | |
5624 | } | |
5625 | ||
5626 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
5627 | const char *name) | |
5628 | { | |
5629 | mdk_thread_t *thread; | |
1da177e4 | 5630 | |
9ffae0cf | 5631 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
5632 | if (!thread) |
5633 | return NULL; | |
5634 | ||
1da177e4 LT |
5635 | init_waitqueue_head(&thread->wqueue); |
5636 | ||
1da177e4 LT |
5637 | thread->run = run; |
5638 | thread->mddev = mddev; | |
32a7627c | 5639 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
5640 | thread->tsk = kthread_run(md_thread, thread, |
5641 | "%s_%s", | |
5642 | mdname(thread->mddev), | |
5643 | name ?: mddev->pers->name); | |
a6fb0934 | 5644 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
5645 | kfree(thread); |
5646 | return NULL; | |
5647 | } | |
1da177e4 LT |
5648 | return thread; |
5649 | } | |
5650 | ||
1da177e4 LT |
5651 | void md_unregister_thread(mdk_thread_t *thread) |
5652 | { | |
e0cf8f04 N |
5653 | if (!thread) |
5654 | return; | |
ba25f9dc | 5655 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
5656 | |
5657 | kthread_stop(thread->tsk); | |
1da177e4 LT |
5658 | kfree(thread); |
5659 | } | |
5660 | ||
5661 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
5662 | { | |
5663 | if (!mddev) { | |
5664 | MD_BUG(); | |
5665 | return; | |
5666 | } | |
5667 | ||
b2d444d7 | 5668 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5669 | return; |
6bfe0b49 DW |
5670 | |
5671 | if (mddev->external) | |
5672 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 5673 | /* |
1da177e4 LT |
5674 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
5675 | mdname(mddev), | |
5676 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
5677 | __builtin_return_address(0),__builtin_return_address(1), | |
5678 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 5679 | */ |
d0a0a5ee AM |
5680 | if (!mddev->pers) |
5681 | return; | |
1da177e4 LT |
5682 | if (!mddev->pers->error_handler) |
5683 | return; | |
5684 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
5685 | if (mddev->degraded) |
5686 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
52664732 | 5687 | set_bit(StateChanged, &rdev->flags); |
1da177e4 LT |
5688 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
5689 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5690 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 5691 | md_new_event_inintr(mddev); |
1da177e4 LT |
5692 | } |
5693 | ||
5694 | /* seq_file implementation /proc/mdstat */ | |
5695 | ||
5696 | static void status_unused(struct seq_file *seq) | |
5697 | { | |
5698 | int i = 0; | |
5699 | mdk_rdev_t *rdev; | |
1da177e4 LT |
5700 | |
5701 | seq_printf(seq, "unused devices: "); | |
5702 | ||
159ec1fc | 5703 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
5704 | char b[BDEVNAME_SIZE]; |
5705 | i++; | |
5706 | seq_printf(seq, "%s ", | |
5707 | bdevname(rdev->bdev,b)); | |
5708 | } | |
5709 | if (!i) | |
5710 | seq_printf(seq, "<none>"); | |
5711 | ||
5712 | seq_printf(seq, "\n"); | |
5713 | } | |
5714 | ||
5715 | ||
5716 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
5717 | { | |
dd71cf6b N |
5718 | sector_t max_sectors, resync, res; |
5719 | unsigned long dt, db; | |
5720 | sector_t rt; | |
4588b42e N |
5721 | int scale; |
5722 | unsigned int per_milli; | |
1da177e4 | 5723 | |
dd71cf6b | 5724 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
5725 | |
5726 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 5727 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 5728 | else |
dd71cf6b | 5729 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
5730 | |
5731 | /* | |
5732 | * Should not happen. | |
5733 | */ | |
dd71cf6b | 5734 | if (!max_sectors) { |
1da177e4 LT |
5735 | MD_BUG(); |
5736 | return; | |
5737 | } | |
4588b42e | 5738 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 5739 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
5740 | * u32, as those are the requirements for sector_div. |
5741 | * Thus 'scale' must be at least 10 | |
5742 | */ | |
5743 | scale = 10; | |
5744 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 5745 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
5746 | scale++; |
5747 | } | |
5748 | res = (resync>>scale)*1000; | |
dd71cf6b | 5749 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
5750 | |
5751 | per_milli = res; | |
1da177e4 | 5752 | { |
4588b42e | 5753 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
5754 | seq_printf(seq, "["); |
5755 | for (i = 0; i < x; i++) | |
5756 | seq_printf(seq, "="); | |
5757 | seq_printf(seq, ">"); | |
5758 | for (i = 0; i < y; i++) | |
5759 | seq_printf(seq, "."); | |
5760 | seq_printf(seq, "] "); | |
5761 | } | |
4588b42e | 5762 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
5763 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
5764 | "reshape" : | |
61df9d91 N |
5765 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
5766 | "check" : | |
5767 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
5768 | "resync" : "recovery"))), | |
5769 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
5770 | (unsigned long long) resync/2, |
5771 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
5772 | |
5773 | /* | |
1da177e4 LT |
5774 | * dt: time from mark until now |
5775 | * db: blocks written from mark until now | |
5776 | * rt: remaining time | |
dd71cf6b N |
5777 | * |
5778 | * rt is a sector_t, so could be 32bit or 64bit. | |
5779 | * So we divide before multiply in case it is 32bit and close | |
5780 | * to the limit. | |
5781 | * We scale the divisor (db) by 32 to avoid loosing precision | |
5782 | * near the end of resync when the number of remaining sectors | |
5783 | * is close to 'db'. | |
5784 | * We then divide rt by 32 after multiplying by db to compensate. | |
5785 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
5786 | */ |
5787 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
5788 | if (!dt) dt++; | |
ff4e8d9a N |
5789 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
5790 | - mddev->resync_mark_cnt; | |
1da177e4 | 5791 | |
dd71cf6b N |
5792 | rt = max_sectors - resync; /* number of remaining sectors */ |
5793 | sector_div(rt, db/32+1); | |
5794 | rt *= dt; | |
5795 | rt >>= 5; | |
5796 | ||
5797 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
5798 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 5799 | |
ff4e8d9a | 5800 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
5801 | } |
5802 | ||
5803 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
5804 | { | |
5805 | struct list_head *tmp; | |
5806 | loff_t l = *pos; | |
5807 | mddev_t *mddev; | |
5808 | ||
5809 | if (l >= 0x10000) | |
5810 | return NULL; | |
5811 | if (!l--) | |
5812 | /* header */ | |
5813 | return (void*)1; | |
5814 | ||
5815 | spin_lock(&all_mddevs_lock); | |
5816 | list_for_each(tmp,&all_mddevs) | |
5817 | if (!l--) { | |
5818 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
5819 | mddev_get(mddev); | |
5820 | spin_unlock(&all_mddevs_lock); | |
5821 | return mddev; | |
5822 | } | |
5823 | spin_unlock(&all_mddevs_lock); | |
5824 | if (!l--) | |
5825 | return (void*)2;/* tail */ | |
5826 | return NULL; | |
5827 | } | |
5828 | ||
5829 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
5830 | { | |
5831 | struct list_head *tmp; | |
5832 | mddev_t *next_mddev, *mddev = v; | |
5833 | ||
5834 | ++*pos; | |
5835 | if (v == (void*)2) | |
5836 | return NULL; | |
5837 | ||
5838 | spin_lock(&all_mddevs_lock); | |
5839 | if (v == (void*)1) | |
5840 | tmp = all_mddevs.next; | |
5841 | else | |
5842 | tmp = mddev->all_mddevs.next; | |
5843 | if (tmp != &all_mddevs) | |
5844 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
5845 | else { | |
5846 | next_mddev = (void*)2; | |
5847 | *pos = 0x10000; | |
5848 | } | |
5849 | spin_unlock(&all_mddevs_lock); | |
5850 | ||
5851 | if (v != (void*)1) | |
5852 | mddev_put(mddev); | |
5853 | return next_mddev; | |
5854 | ||
5855 | } | |
5856 | ||
5857 | static void md_seq_stop(struct seq_file *seq, void *v) | |
5858 | { | |
5859 | mddev_t *mddev = v; | |
5860 | ||
5861 | if (mddev && v != (void*)1 && v != (void*)2) | |
5862 | mddev_put(mddev); | |
5863 | } | |
5864 | ||
d7603b7e N |
5865 | struct mdstat_info { |
5866 | int event; | |
5867 | }; | |
5868 | ||
1da177e4 LT |
5869 | static int md_seq_show(struct seq_file *seq, void *v) |
5870 | { | |
5871 | mddev_t *mddev = v; | |
dd8ac336 | 5872 | sector_t sectors; |
1da177e4 | 5873 | mdk_rdev_t *rdev; |
d7603b7e | 5874 | struct mdstat_info *mi = seq->private; |
32a7627c | 5875 | struct bitmap *bitmap; |
1da177e4 LT |
5876 | |
5877 | if (v == (void*)1) { | |
2604b703 | 5878 | struct mdk_personality *pers; |
1da177e4 LT |
5879 | seq_printf(seq, "Personalities : "); |
5880 | spin_lock(&pers_lock); | |
2604b703 N |
5881 | list_for_each_entry(pers, &pers_list, list) |
5882 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
5883 | |
5884 | spin_unlock(&pers_lock); | |
5885 | seq_printf(seq, "\n"); | |
d7603b7e | 5886 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
5887 | return 0; |
5888 | } | |
5889 | if (v == (void*)2) { | |
5890 | status_unused(seq); | |
5891 | return 0; | |
5892 | } | |
5893 | ||
5dc5cf7d | 5894 | if (mddev_lock(mddev) < 0) |
1da177e4 | 5895 | return -EINTR; |
5dc5cf7d | 5896 | |
1da177e4 LT |
5897 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
5898 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
5899 | mddev->pers ? "" : "in"); | |
5900 | if (mddev->pers) { | |
f91de92e | 5901 | if (mddev->ro==1) |
1da177e4 | 5902 | seq_printf(seq, " (read-only)"); |
f91de92e | 5903 | if (mddev->ro==2) |
52720ae7 | 5904 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
5905 | seq_printf(seq, " %s", mddev->pers->name); |
5906 | } | |
5907 | ||
dd8ac336 | 5908 | sectors = 0; |
159ec1fc | 5909 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
5910 | char b[BDEVNAME_SIZE]; |
5911 | seq_printf(seq, " %s[%d]", | |
5912 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
5913 | if (test_bit(WriteMostly, &rdev->flags)) |
5914 | seq_printf(seq, "(W)"); | |
b2d444d7 | 5915 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5916 | seq_printf(seq, "(F)"); |
5917 | continue; | |
b325a32e N |
5918 | } else if (rdev->raid_disk < 0) |
5919 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 5920 | sectors += rdev->sectors; |
1da177e4 LT |
5921 | } |
5922 | ||
5923 | if (!list_empty(&mddev->disks)) { | |
5924 | if (mddev->pers) | |
5925 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
5926 | (unsigned long long) |
5927 | mddev->array_sectors / 2); | |
1da177e4 LT |
5928 | else |
5929 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 5930 | (unsigned long long)sectors / 2); |
1da177e4 | 5931 | } |
1cd6bf19 N |
5932 | if (mddev->persistent) { |
5933 | if (mddev->major_version != 0 || | |
5934 | mddev->minor_version != 90) { | |
5935 | seq_printf(seq," super %d.%d", | |
5936 | mddev->major_version, | |
5937 | mddev->minor_version); | |
5938 | } | |
e691063a N |
5939 | } else if (mddev->external) |
5940 | seq_printf(seq, " super external:%s", | |
5941 | mddev->metadata_type); | |
5942 | else | |
1cd6bf19 | 5943 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
5944 | |
5945 | if (mddev->pers) { | |
d710e138 | 5946 | mddev->pers->status(seq, mddev); |
1da177e4 | 5947 | seq_printf(seq, "\n "); |
8e1b39d6 N |
5948 | if (mddev->pers->sync_request) { |
5949 | if (mddev->curr_resync > 2) { | |
d710e138 | 5950 | status_resync(seq, mddev); |
8e1b39d6 N |
5951 | seq_printf(seq, "\n "); |
5952 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
5953 | seq_printf(seq, "\tresync=DELAYED\n "); | |
5954 | else if (mddev->recovery_cp < MaxSector) | |
5955 | seq_printf(seq, "\tresync=PENDING\n "); | |
5956 | } | |
32a7627c N |
5957 | } else |
5958 | seq_printf(seq, "\n "); | |
5959 | ||
5960 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
5961 | unsigned long chunk_kb; |
5962 | unsigned long flags; | |
32a7627c N |
5963 | spin_lock_irqsave(&bitmap->lock, flags); |
5964 | chunk_kb = bitmap->chunksize >> 10; | |
5965 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
5966 | "%lu%s chunk", | |
5967 | bitmap->pages - bitmap->missing_pages, | |
5968 | bitmap->pages, | |
5969 | (bitmap->pages - bitmap->missing_pages) | |
5970 | << (PAGE_SHIFT - 10), | |
5971 | chunk_kb ? chunk_kb : bitmap->chunksize, | |
5972 | chunk_kb ? "KB" : "B"); | |
78d742d8 N |
5973 | if (bitmap->file) { |
5974 | seq_printf(seq, ", file: "); | |
c32c2f63 | 5975 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 5976 | } |
78d742d8 | 5977 | |
32a7627c N |
5978 | seq_printf(seq, "\n"); |
5979 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
5980 | } |
5981 | ||
5982 | seq_printf(seq, "\n"); | |
5983 | } | |
5984 | mddev_unlock(mddev); | |
5985 | ||
5986 | return 0; | |
5987 | } | |
5988 | ||
110518bc | 5989 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
5990 | .start = md_seq_start, |
5991 | .next = md_seq_next, | |
5992 | .stop = md_seq_stop, | |
5993 | .show = md_seq_show, | |
5994 | }; | |
5995 | ||
5996 | static int md_seq_open(struct inode *inode, struct file *file) | |
5997 | { | |
5998 | int error; | |
d7603b7e N |
5999 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6000 | if (mi == NULL) | |
6001 | return -ENOMEM; | |
1da177e4 LT |
6002 | |
6003 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6004 | if (error) |
6005 | kfree(mi); | |
6006 | else { | |
6007 | struct seq_file *p = file->private_data; | |
6008 | p->private = mi; | |
6009 | mi->event = atomic_read(&md_event_count); | |
6010 | } | |
1da177e4 LT |
6011 | return error; |
6012 | } | |
6013 | ||
d7603b7e N |
6014 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6015 | { | |
6016 | struct seq_file *m = filp->private_data; | |
6017 | struct mdstat_info *mi = m->private; | |
6018 | int mask; | |
6019 | ||
6020 | poll_wait(filp, &md_event_waiters, wait); | |
6021 | ||
6022 | /* always allow read */ | |
6023 | mask = POLLIN | POLLRDNORM; | |
6024 | ||
6025 | if (mi->event != atomic_read(&md_event_count)) | |
6026 | mask |= POLLERR | POLLPRI; | |
6027 | return mask; | |
6028 | } | |
6029 | ||
fa027c2a | 6030 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6031 | .owner = THIS_MODULE, |
1da177e4 LT |
6032 | .open = md_seq_open, |
6033 | .read = seq_read, | |
6034 | .llseek = seq_lseek, | |
c3f94b40 | 6035 | .release = seq_release_private, |
d7603b7e | 6036 | .poll = mdstat_poll, |
1da177e4 LT |
6037 | }; |
6038 | ||
2604b703 | 6039 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6040 | { |
1da177e4 | 6041 | spin_lock(&pers_lock); |
2604b703 N |
6042 | list_add_tail(&p->list, &pers_list); |
6043 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6044 | spin_unlock(&pers_lock); |
6045 | return 0; | |
6046 | } | |
6047 | ||
2604b703 | 6048 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6049 | { |
2604b703 | 6050 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6051 | spin_lock(&pers_lock); |
2604b703 | 6052 | list_del_init(&p->list); |
1da177e4 LT |
6053 | spin_unlock(&pers_lock); |
6054 | return 0; | |
6055 | } | |
6056 | ||
eea1bf38 | 6057 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6058 | { |
6059 | mdk_rdev_t * rdev; | |
1da177e4 | 6060 | int idle; |
eea1bf38 | 6061 | int curr_events; |
1da177e4 LT |
6062 | |
6063 | idle = 1; | |
4b80991c N |
6064 | rcu_read_lock(); |
6065 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6066 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6067 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6068 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6069 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6070 | /* sync IO will cause sync_io to increase before the disk_stats |
6071 | * as sync_io is counted when a request starts, and | |
6072 | * disk_stats is counted when it completes. | |
6073 | * So resync activity will cause curr_events to be smaller than | |
6074 | * when there was no such activity. | |
6075 | * non-sync IO will cause disk_stat to increase without | |
6076 | * increasing sync_io so curr_events will (eventually) | |
6077 | * be larger than it was before. Once it becomes | |
6078 | * substantially larger, the test below will cause | |
6079 | * the array to appear non-idle, and resync will slow | |
6080 | * down. | |
6081 | * If there is a lot of outstanding resync activity when | |
6082 | * we set last_event to curr_events, then all that activity | |
6083 | * completing might cause the array to appear non-idle | |
6084 | * and resync will be slowed down even though there might | |
6085 | * not have been non-resync activity. This will only | |
6086 | * happen once though. 'last_events' will soon reflect | |
6087 | * the state where there is little or no outstanding | |
6088 | * resync requests, and further resync activity will | |
6089 | * always make curr_events less than last_events. | |
c0e48521 | 6090 | * |
1da177e4 | 6091 | */ |
eea1bf38 | 6092 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6093 | rdev->last_events = curr_events; |
6094 | idle = 0; | |
6095 | } | |
6096 | } | |
4b80991c | 6097 | rcu_read_unlock(); |
1da177e4 LT |
6098 | return idle; |
6099 | } | |
6100 | ||
6101 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6102 | { | |
6103 | /* another "blocks" (512byte) blocks have been synced */ | |
6104 | atomic_sub(blocks, &mddev->recovery_active); | |
6105 | wake_up(&mddev->recovery_wait); | |
6106 | if (!ok) { | |
dfc70645 | 6107 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6108 | md_wakeup_thread(mddev->thread); |
6109 | // stop recovery, signal do_sync .... | |
6110 | } | |
6111 | } | |
6112 | ||
6113 | ||
06d91a5f N |
6114 | /* md_write_start(mddev, bi) |
6115 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6116 | * in superblock) before writing, schedule a superblock update |
6117 | * and wait for it to complete. | |
06d91a5f | 6118 | */ |
3d310eb7 | 6119 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6120 | { |
0fd62b86 | 6121 | int did_change = 0; |
06d91a5f | 6122 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6123 | return; |
06d91a5f | 6124 | |
f91de92e N |
6125 | BUG_ON(mddev->ro == 1); |
6126 | if (mddev->ro == 2) { | |
6127 | /* need to switch to read/write */ | |
6128 | mddev->ro = 0; | |
6129 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6130 | md_wakeup_thread(mddev->thread); | |
25156198 | 6131 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6132 | did_change = 1; |
f91de92e | 6133 | } |
06d91a5f | 6134 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6135 | if (mddev->safemode == 1) |
6136 | mddev->safemode = 0; | |
06d91a5f | 6137 | if (mddev->in_sync) { |
a9701a30 | 6138 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6139 | if (mddev->in_sync) { |
6140 | mddev->in_sync = 0; | |
850b2b42 | 6141 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3d310eb7 | 6142 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6143 | did_change = 1; |
3d310eb7 | 6144 | } |
a9701a30 | 6145 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6146 | } |
0fd62b86 | 6147 | if (did_change) |
b62b7590 | 6148 | sysfs_notify_dirent(mddev->sysfs_state); |
09a44cc1 N |
6149 | wait_event(mddev->sb_wait, |
6150 | !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && | |
6151 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1da177e4 LT |
6152 | } |
6153 | ||
6154 | void md_write_end(mddev_t *mddev) | |
6155 | { | |
6156 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6157 | if (mddev->safemode == 2) | |
6158 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6159 | else if (mddev->safemode_delay) |
1da177e4 LT |
6160 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6161 | } | |
6162 | } | |
6163 | ||
2a2275d6 N |
6164 | /* md_allow_write(mddev) |
6165 | * Calling this ensures that the array is marked 'active' so that writes | |
6166 | * may proceed without blocking. It is important to call this before | |
6167 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6168 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6169 | * |
6170 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6171 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6172 | */ |
b5470dc5 | 6173 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6174 | { |
6175 | if (!mddev->pers) | |
b5470dc5 | 6176 | return 0; |
2a2275d6 | 6177 | if (mddev->ro) |
b5470dc5 | 6178 | return 0; |
1a0fd497 | 6179 | if (!mddev->pers->sync_request) |
b5470dc5 | 6180 | return 0; |
2a2275d6 N |
6181 | |
6182 | spin_lock_irq(&mddev->write_lock); | |
6183 | if (mddev->in_sync) { | |
6184 | mddev->in_sync = 0; | |
6185 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6186 | if (mddev->safemode_delay && | |
6187 | mddev->safemode == 0) | |
6188 | mddev->safemode = 1; | |
6189 | spin_unlock_irq(&mddev->write_lock); | |
6190 | md_update_sb(mddev, 0); | |
b62b7590 | 6191 | sysfs_notify_dirent(mddev->sysfs_state); |
2a2275d6 N |
6192 | } else |
6193 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 DW |
6194 | |
6195 | if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
6196 | return -EAGAIN; | |
6197 | else | |
6198 | return 0; | |
2a2275d6 N |
6199 | } |
6200 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6201 | ||
1da177e4 LT |
6202 | #define SYNC_MARKS 10 |
6203 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6204 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6205 | { |
6206 | mddev_t *mddev2; | |
6207 | unsigned int currspeed = 0, | |
6208 | window; | |
57afd89f | 6209 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6210 | unsigned long mark[SYNC_MARKS]; |
6211 | sector_t mark_cnt[SYNC_MARKS]; | |
6212 | int last_mark,m; | |
6213 | struct list_head *tmp; | |
6214 | sector_t last_check; | |
57afd89f | 6215 | int skipped = 0; |
5fd6c1dc | 6216 | mdk_rdev_t *rdev; |
61df9d91 | 6217 | char *desc; |
1da177e4 LT |
6218 | |
6219 | /* just incase thread restarts... */ | |
6220 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6221 | return; | |
5fd6c1dc N |
6222 | if (mddev->ro) /* never try to sync a read-only array */ |
6223 | return; | |
1da177e4 | 6224 | |
61df9d91 N |
6225 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6226 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6227 | desc = "data-check"; | |
6228 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6229 | desc = "requested-resync"; | |
6230 | else | |
6231 | desc = "resync"; | |
6232 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6233 | desc = "reshape"; | |
6234 | else | |
6235 | desc = "recovery"; | |
6236 | ||
1da177e4 LT |
6237 | /* we overload curr_resync somewhat here. |
6238 | * 0 == not engaged in resync at all | |
6239 | * 2 == checking that there is no conflict with another sync | |
6240 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6241 | * commense | |
6242 | * other == active in resync - this many blocks | |
6243 | * | |
6244 | * Before starting a resync we must have set curr_resync to | |
6245 | * 2, and then checked that every "conflicting" array has curr_resync | |
6246 | * less than ours. When we find one that is the same or higher | |
6247 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6248 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6249 | * This will mean we have to start checking from the beginning again. | |
6250 | * | |
6251 | */ | |
6252 | ||
6253 | do { | |
6254 | mddev->curr_resync = 2; | |
6255 | ||
6256 | try_again: | |
787453c2 | 6257 | if (kthread_should_stop()) { |
6985c43f | 6258 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6259 | goto skip; |
6260 | } | |
29ac4aa3 | 6261 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6262 | if (mddev2 == mddev) |
6263 | continue; | |
90b08710 BS |
6264 | if (!mddev->parallel_resync |
6265 | && mddev2->curr_resync | |
6266 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6267 | DEFINE_WAIT(wq); |
6268 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6269 | /* arbitrarily yield */ | |
6270 | mddev->curr_resync = 1; | |
6271 | wake_up(&resync_wait); | |
6272 | } | |
6273 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6274 | /* no need to wait here, we can wait the next | |
6275 | * time 'round when curr_resync == 2 | |
6276 | */ | |
6277 | continue; | |
9744197c N |
6278 | /* We need to wait 'interruptible' so as not to |
6279 | * contribute to the load average, and not to | |
6280 | * be caught by 'softlockup' | |
6281 | */ | |
6282 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6283 | if (!kthread_should_stop() && |
8712e553 | 6284 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6285 | printk(KERN_INFO "md: delaying %s of %s" |
6286 | " until %s has finished (they" | |
1da177e4 | 6287 | " share one or more physical units)\n", |
61df9d91 | 6288 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6289 | mddev_put(mddev2); |
9744197c N |
6290 | if (signal_pending(current)) |
6291 | flush_signals(current); | |
1da177e4 LT |
6292 | schedule(); |
6293 | finish_wait(&resync_wait, &wq); | |
6294 | goto try_again; | |
6295 | } | |
6296 | finish_wait(&resync_wait, &wq); | |
6297 | } | |
6298 | } | |
6299 | } while (mddev->curr_resync < 2); | |
6300 | ||
5fd6c1dc | 6301 | j = 0; |
9d88883e | 6302 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6303 | /* resync follows the size requested by the personality, |
57afd89f | 6304 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6305 | */ |
6306 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6307 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6308 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6309 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6310 | j = mddev->resync_min; | |
6311 | else if (!mddev->bitmap) | |
5fd6c1dc | 6312 | j = mddev->recovery_cp; |
5e96ee65 | 6313 | |
ccfcc3c1 | 6314 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6315 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6316 | else { |
1da177e4 | 6317 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6318 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6319 | j = MaxSector; |
159ec1fc | 6320 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
6321 | if (rdev->raid_disk >= 0 && |
6322 | !test_bit(Faulty, &rdev->flags) && | |
6323 | !test_bit(In_sync, &rdev->flags) && | |
6324 | rdev->recovery_offset < j) | |
6325 | j = rdev->recovery_offset; | |
6326 | } | |
1da177e4 | 6327 | |
61df9d91 N |
6328 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6329 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6330 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6331 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6332 | "(but not more than %d KB/sec) for %s.\n", |
6333 | speed_max(mddev), desc); | |
1da177e4 | 6334 | |
eea1bf38 | 6335 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6336 | |
57afd89f | 6337 | io_sectors = 0; |
1da177e4 LT |
6338 | for (m = 0; m < SYNC_MARKS; m++) { |
6339 | mark[m] = jiffies; | |
57afd89f | 6340 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6341 | } |
6342 | last_mark = 0; | |
6343 | mddev->resync_mark = mark[last_mark]; | |
6344 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6345 | ||
6346 | /* | |
6347 | * Tune reconstruction: | |
6348 | */ | |
6349 | window = 32*(PAGE_SIZE/512); | |
6350 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6351 | window/2,(unsigned long long) max_sectors/2); | |
6352 | ||
6353 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6354 | last_check = 0; |
6355 | ||
6356 | if (j>2) { | |
6357 | printk(KERN_INFO | |
61df9d91 N |
6358 | "md: resuming %s of %s from checkpoint.\n", |
6359 | desc, mdname(mddev)); | |
1da177e4 LT |
6360 | mddev->curr_resync = j; |
6361 | } | |
6362 | ||
6363 | while (j < max_sectors) { | |
57afd89f | 6364 | sector_t sectors; |
1da177e4 | 6365 | |
57afd89f | 6366 | skipped = 0; |
97e4f42d | 6367 | |
7a91ee1f N |
6368 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6369 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6370 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6371 | > (max_sectors >> 4)) || | |
6372 | (j - mddev->curr_resync_completed)*2 | |
6373 | >= mddev->resync_max - mddev->curr_resync_completed | |
6374 | )) { | |
97e4f42d N |
6375 | /* time to update curr_resync_completed */ |
6376 | blk_unplug(mddev->queue); | |
6377 | wait_event(mddev->recovery_wait, | |
6378 | atomic_read(&mddev->recovery_active) == 0); | |
6379 | mddev->curr_resync_completed = | |
6380 | mddev->curr_resync; | |
6381 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
acb180b0 | 6382 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6383 | } |
acb180b0 | 6384 | |
e62e58a5 N |
6385 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6386 | /* As this condition is controlled by user-space, | |
6387 | * we can block indefinitely, so use '_interruptible' | |
6388 | * to avoid triggering warnings. | |
6389 | */ | |
6390 | flush_signals(current); /* just in case */ | |
6391 | wait_event_interruptible(mddev->recovery_wait, | |
6392 | mddev->resync_max > j | |
6393 | || kthread_should_stop()); | |
6394 | } | |
acb180b0 N |
6395 | |
6396 | if (kthread_should_stop()) | |
6397 | goto interrupted; | |
6398 | ||
57afd89f | 6399 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6400 | currspeed < speed_min(mddev)); |
57afd89f | 6401 | if (sectors == 0) { |
dfc70645 | 6402 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6403 | goto out; |
6404 | } | |
57afd89f N |
6405 | |
6406 | if (!skipped) { /* actual IO requested */ | |
6407 | io_sectors += sectors; | |
6408 | atomic_add(sectors, &mddev->recovery_active); | |
6409 | } | |
6410 | ||
1da177e4 LT |
6411 | j += sectors; |
6412 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6413 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6414 | if (last_check == 0) |
6415 | /* this is the earliers that rebuilt will be | |
6416 | * visible in /proc/mdstat | |
6417 | */ | |
6418 | md_new_event(mddev); | |
57afd89f N |
6419 | |
6420 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6421 | continue; |
6422 | ||
57afd89f | 6423 | last_check = io_sectors; |
1da177e4 | 6424 | |
dfc70645 | 6425 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6426 | break; |
6427 | ||
6428 | repeat: | |
6429 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6430 | /* step marks */ | |
6431 | int next = (last_mark+1) % SYNC_MARKS; | |
6432 | ||
6433 | mddev->resync_mark = mark[next]; | |
6434 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6435 | mark[next] = jiffies; | |
57afd89f | 6436 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6437 | last_mark = next; |
6438 | } | |
6439 | ||
6440 | ||
c6207277 N |
6441 | if (kthread_should_stop()) |
6442 | goto interrupted; | |
6443 | ||
1da177e4 LT |
6444 | |
6445 | /* | |
6446 | * this loop exits only if either when we are slower than | |
6447 | * the 'hard' speed limit, or the system was IO-idle for | |
6448 | * a jiffy. | |
6449 | * the system might be non-idle CPU-wise, but we only care | |
6450 | * about not overloading the IO subsystem. (things like an | |
6451 | * e2fsck being done on the RAID array should execute fast) | |
6452 | */ | |
2ad8b1ef | 6453 | blk_unplug(mddev->queue); |
1da177e4 LT |
6454 | cond_resched(); |
6455 | ||
57afd89f N |
6456 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6457 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6458 | |
88202a0c N |
6459 | if (currspeed > speed_min(mddev)) { |
6460 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6461 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6462 | msleep(500); |
1da177e4 LT |
6463 | goto repeat; |
6464 | } | |
6465 | } | |
6466 | } | |
61df9d91 | 6467 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6468 | /* |
6469 | * this also signals 'finished resyncing' to md_stop | |
6470 | */ | |
6471 | out: | |
2ad8b1ef | 6472 | blk_unplug(mddev->queue); |
1da177e4 LT |
6473 | |
6474 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
6475 | ||
6476 | /* tell personality that we are finished */ | |
57afd89f | 6477 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6478 | |
dfc70645 | 6479 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6480 | mddev->curr_resync > 2) { |
6481 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6482 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6483 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6484 | printk(KERN_INFO | |
61df9d91 N |
6485 | "md: checkpointing %s of %s.\n", |
6486 | desc, mdname(mddev)); | |
5fd6c1dc N |
6487 | mddev->recovery_cp = mddev->curr_resync; |
6488 | } | |
6489 | } else | |
6490 | mddev->recovery_cp = MaxSector; | |
6491 | } else { | |
6492 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6493 | mddev->curr_resync = MaxSector; | |
159ec1fc | 6494 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
6495 | if (rdev->raid_disk >= 0 && |
6496 | !test_bit(Faulty, &rdev->flags) && | |
6497 | !test_bit(In_sync, &rdev->flags) && | |
6498 | rdev->recovery_offset < mddev->curr_resync) | |
6499 | rdev->recovery_offset = mddev->curr_resync; | |
5fd6c1dc | 6500 | } |
1da177e4 | 6501 | } |
17571284 | 6502 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6503 | |
1da177e4 LT |
6504 | skip: |
6505 | mddev->curr_resync = 0; | |
acb180b0 | 6506 | mddev->curr_resync_completed = 0; |
5e96ee65 | 6507 | mddev->resync_min = 0; |
c6207277 N |
6508 | mddev->resync_max = MaxSector; |
6509 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
1da177e4 LT |
6510 | wake_up(&resync_wait); |
6511 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
6512 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
6513 | return; |
6514 | ||
6515 | interrupted: | |
6516 | /* | |
6517 | * got a signal, exit. | |
6518 | */ | |
6519 | printk(KERN_INFO | |
6520 | "md: md_do_sync() got signal ... exiting\n"); | |
6521 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6522 | goto out; | |
6523 | ||
1da177e4 | 6524 | } |
29269553 | 6525 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
6526 | |
6527 | ||
b4c4c7b8 N |
6528 | static int remove_and_add_spares(mddev_t *mddev) |
6529 | { | |
6530 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
6531 | int spares = 0; |
6532 | ||
97e4f42d N |
6533 | mddev->curr_resync_completed = 0; |
6534 | ||
159ec1fc | 6535 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 6536 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 6537 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
6538 | (test_bit(Faulty, &rdev->flags) || |
6539 | ! test_bit(In_sync, &rdev->flags)) && | |
6540 | atomic_read(&rdev->nr_pending)==0) { | |
6541 | if (mddev->pers->hot_remove_disk( | |
6542 | mddev, rdev->raid_disk)==0) { | |
6543 | char nm[20]; | |
6544 | sprintf(nm,"rd%d", rdev->raid_disk); | |
6545 | sysfs_remove_link(&mddev->kobj, nm); | |
6546 | rdev->raid_disk = -1; | |
6547 | } | |
6548 | } | |
6549 | ||
4044ba58 | 6550 | if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { |
159ec1fc | 6551 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 6552 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
6553 | !test_bit(In_sync, &rdev->flags) && |
6554 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 6555 | spares++; |
b4c4c7b8 N |
6556 | if (rdev->raid_disk < 0 |
6557 | && !test_bit(Faulty, &rdev->flags)) { | |
6558 | rdev->recovery_offset = 0; | |
199050ea NB |
6559 | if (mddev->pers-> |
6560 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
6561 | char nm[20]; |
6562 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
6563 | if (sysfs_create_link(&mddev->kobj, |
6564 | &rdev->kobj, nm)) | |
6565 | printk(KERN_WARNING | |
6566 | "md: cannot register " | |
6567 | "%s for %s\n", | |
6568 | nm, mdname(mddev)); | |
b4c4c7b8 N |
6569 | spares++; |
6570 | md_new_event(mddev); | |
6571 | } else | |
6572 | break; | |
6573 | } | |
dfc70645 | 6574 | } |
b4c4c7b8 N |
6575 | } |
6576 | return spares; | |
6577 | } | |
1da177e4 LT |
6578 | /* |
6579 | * This routine is regularly called by all per-raid-array threads to | |
6580 | * deal with generic issues like resync and super-block update. | |
6581 | * Raid personalities that don't have a thread (linear/raid0) do not | |
6582 | * need this as they never do any recovery or update the superblock. | |
6583 | * | |
6584 | * It does not do any resync itself, but rather "forks" off other threads | |
6585 | * to do that as needed. | |
6586 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
6587 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 6588 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
6589 | * and wakeups up this thread which will reap the thread and finish up. |
6590 | * This thread also removes any faulty devices (with nr_pending == 0). | |
6591 | * | |
6592 | * The overall approach is: | |
6593 | * 1/ if the superblock needs updating, update it. | |
6594 | * 2/ If a recovery thread is running, don't do anything else. | |
6595 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
6596 | * 4/ If there are any faulty devices, remove them. | |
6597 | * 5/ If array is degraded, try to add spares devices | |
6598 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
6599 | */ | |
6600 | void md_check_recovery(mddev_t *mddev) | |
6601 | { | |
6602 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6603 | |
6604 | ||
5f40402d N |
6605 | if (mddev->bitmap) |
6606 | bitmap_daemon_work(mddev->bitmap); | |
1da177e4 LT |
6607 | |
6608 | if (mddev->ro) | |
6609 | return; | |
fca4d848 N |
6610 | |
6611 | if (signal_pending(current)) { | |
31a59e34 | 6612 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
6613 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
6614 | mdname(mddev)); | |
6615 | mddev->safemode = 2; | |
6616 | } | |
6617 | flush_signals(current); | |
6618 | } | |
6619 | ||
c89a8eee N |
6620 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
6621 | return; | |
1da177e4 | 6622 | if ( ! ( |
e691063a | 6623 | (mddev->flags && !mddev->external) || |
1da177e4 | 6624 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 6625 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 6626 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
6627 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
6628 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
6629 | )) |
6630 | return; | |
fca4d848 | 6631 | |
df5b89b3 | 6632 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 6633 | int spares = 0; |
fca4d848 | 6634 | |
c89a8eee N |
6635 | if (mddev->ro) { |
6636 | /* Only thing we do on a ro array is remove | |
6637 | * failed devices. | |
6638 | */ | |
6639 | remove_and_add_spares(mddev); | |
6640 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6641 | goto unlock; | |
6642 | } | |
6643 | ||
31a59e34 | 6644 | if (!mddev->external) { |
0fd62b86 | 6645 | int did_change = 0; |
31a59e34 N |
6646 | spin_lock_irq(&mddev->write_lock); |
6647 | if (mddev->safemode && | |
6648 | !atomic_read(&mddev->writes_pending) && | |
6649 | !mddev->in_sync && | |
6650 | mddev->recovery_cp == MaxSector) { | |
6651 | mddev->in_sync = 1; | |
0fd62b86 | 6652 | did_change = 1; |
31a59e34 N |
6653 | if (mddev->persistent) |
6654 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6655 | } | |
6656 | if (mddev->safemode == 1) | |
6657 | mddev->safemode = 0; | |
6658 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 6659 | if (did_change) |
b62b7590 | 6660 | sysfs_notify_dirent(mddev->sysfs_state); |
fca4d848 | 6661 | } |
fca4d848 | 6662 | |
850b2b42 N |
6663 | if (mddev->flags) |
6664 | md_update_sb(mddev, 0); | |
06d91a5f | 6665 | |
159ec1fc | 6666 | list_for_each_entry(rdev, &mddev->disks, same_set) |
52664732 | 6667 | if (test_and_clear_bit(StateChanged, &rdev->flags)) |
3c0ee63a | 6668 | sysfs_notify_dirent(rdev->sysfs_state); |
52664732 | 6669 | |
06d91a5f | 6670 | |
1da177e4 LT |
6671 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
6672 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
6673 | /* resync/recovery still happening */ | |
6674 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6675 | goto unlock; | |
6676 | } | |
6677 | if (mddev->sync_thread) { | |
6678 | /* resync has finished, collect result */ | |
6679 | md_unregister_thread(mddev->sync_thread); | |
6680 | mddev->sync_thread = NULL; | |
56ac36d7 DW |
6681 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
6682 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1da177e4 LT |
6683 | /* success...*/ |
6684 | /* activate any spares */ | |
a99ac971 NB |
6685 | if (mddev->pers->spare_active(mddev)) |
6686 | sysfs_notify(&mddev->kobj, NULL, | |
6687 | "degraded"); | |
1da177e4 | 6688 | } |
cea9c228 N |
6689 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6690 | mddev->pers->finish_reshape) | |
6691 | mddev->pers->finish_reshape(mddev); | |
850b2b42 | 6692 | md_update_sb(mddev, 1); |
41158c7e N |
6693 | |
6694 | /* if array is no-longer degraded, then any saved_raid_disk | |
6695 | * information must be scrapped | |
6696 | */ | |
6697 | if (!mddev->degraded) | |
159ec1fc | 6698 | list_for_each_entry(rdev, &mddev->disks, same_set) |
41158c7e N |
6699 | rdev->saved_raid_disk = -1; |
6700 | ||
1da177e4 LT |
6701 | mddev->recovery = 0; |
6702 | /* flag recovery needed just to double check */ | |
6703 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
0c3573f1 | 6704 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 6705 | md_new_event(mddev); |
1da177e4 LT |
6706 | goto unlock; |
6707 | } | |
72a23c21 NB |
6708 | /* Set RUNNING before clearing NEEDED to avoid |
6709 | * any transients in the value of "sync_action". | |
6710 | */ | |
6711 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6712 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
6713 | /* Clear some bits that don't mean anything, but |
6714 | * might be left set | |
6715 | */ | |
24dd469d N |
6716 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6717 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 6718 | |
5fd6c1dc N |
6719 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6720 | goto unlock; | |
1da177e4 LT |
6721 | /* no recovery is running. |
6722 | * remove any failed drives, then | |
6723 | * add spares if possible. | |
6724 | * Spare are also removed and re-added, to allow | |
6725 | * the personality to fail the re-add. | |
6726 | */ | |
1da177e4 | 6727 | |
b4c4c7b8 | 6728 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
6729 | if (mddev->pers->check_reshape == NULL || |
6730 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
6731 | /* Cannot proceed */ |
6732 | goto unlock; | |
6733 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 6734 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 6735 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
6736 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
6737 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 6738 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 6739 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6740 | } else if (mddev->recovery_cp < MaxSector) { |
6741 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 6742 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6743 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
6744 | /* nothing to be done ... */ | |
1da177e4 | 6745 | goto unlock; |
24dd469d | 6746 | |
1da177e4 | 6747 | if (mddev->pers->sync_request) { |
a654b9d8 N |
6748 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
6749 | /* We are adding a device or devices to an array | |
6750 | * which has the bitmap stored on all devices. | |
6751 | * So make sure all bitmap pages get written | |
6752 | */ | |
6753 | bitmap_write_all(mddev->bitmap); | |
6754 | } | |
1da177e4 LT |
6755 | mddev->sync_thread = md_register_thread(md_do_sync, |
6756 | mddev, | |
0da3c619 | 6757 | "resync"); |
1da177e4 LT |
6758 | if (!mddev->sync_thread) { |
6759 | printk(KERN_ERR "%s: could not start resync" | |
6760 | " thread...\n", | |
6761 | mdname(mddev)); | |
6762 | /* leave the spares where they are, it shouldn't hurt */ | |
6763 | mddev->recovery = 0; | |
d7603b7e | 6764 | } else |
1da177e4 | 6765 | md_wakeup_thread(mddev->sync_thread); |
0c3573f1 | 6766 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 6767 | md_new_event(mddev); |
1da177e4 LT |
6768 | } |
6769 | unlock: | |
72a23c21 NB |
6770 | if (!mddev->sync_thread) { |
6771 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6772 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
6773 | &mddev->recovery)) | |
0c3573f1 N |
6774 | if (mddev->sysfs_action) |
6775 | sysfs_notify_dirent(mddev->sysfs_action); | |
72a23c21 | 6776 | } |
1da177e4 LT |
6777 | mddev_unlock(mddev); |
6778 | } | |
6779 | } | |
6780 | ||
6bfe0b49 DW |
6781 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
6782 | { | |
3c0ee63a | 6783 | sysfs_notify_dirent(rdev->sysfs_state); |
6bfe0b49 DW |
6784 | wait_event_timeout(rdev->blocked_wait, |
6785 | !test_bit(Blocked, &rdev->flags), | |
6786 | msecs_to_jiffies(5000)); | |
6787 | rdev_dec_pending(rdev, mddev); | |
6788 | } | |
6789 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
6790 | ||
75c96f85 AB |
6791 | static int md_notify_reboot(struct notifier_block *this, |
6792 | unsigned long code, void *x) | |
1da177e4 LT |
6793 | { |
6794 | struct list_head *tmp; | |
6795 | mddev_t *mddev; | |
6796 | ||
6797 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
6798 | ||
6799 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
6800 | ||
29ac4aa3 | 6801 | for_each_mddev(mddev, tmp) |
c71d4887 | 6802 | if (mddev_trylock(mddev)) { |
2b25000b N |
6803 | /* Force a switch to readonly even array |
6804 | * appears to still be in use. Hence | |
6805 | * the '100'. | |
6806 | */ | |
d710e138 | 6807 | do_md_stop(mddev, 1, 100); |
c71d4887 NB |
6808 | mddev_unlock(mddev); |
6809 | } | |
1da177e4 LT |
6810 | /* |
6811 | * certain more exotic SCSI devices are known to be | |
6812 | * volatile wrt too early system reboots. While the | |
6813 | * right place to handle this issue is the given | |
6814 | * driver, we do want to have a safe RAID driver ... | |
6815 | */ | |
6816 | mdelay(1000*1); | |
6817 | } | |
6818 | return NOTIFY_DONE; | |
6819 | } | |
6820 | ||
75c96f85 | 6821 | static struct notifier_block md_notifier = { |
1da177e4 LT |
6822 | .notifier_call = md_notify_reboot, |
6823 | .next = NULL, | |
6824 | .priority = INT_MAX, /* before any real devices */ | |
6825 | }; | |
6826 | ||
6827 | static void md_geninit(void) | |
6828 | { | |
1da177e4 LT |
6829 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
6830 | ||
c7705f34 | 6831 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
6832 | } |
6833 | ||
75c96f85 | 6834 | static int __init md_init(void) |
1da177e4 | 6835 | { |
3dbd8c2e | 6836 | if (register_blkdev(MD_MAJOR, "md")) |
1da177e4 LT |
6837 | return -1; |
6838 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
3dbd8c2e | 6839 | unregister_blkdev(MD_MAJOR, "md"); |
1da177e4 LT |
6840 | return -1; |
6841 | } | |
3dbd8c2e | 6842 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
6843 | md_probe, NULL, NULL); |
6844 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
6845 | md_probe, NULL, NULL); |
6846 | ||
1da177e4 | 6847 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 6848 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
6849 | |
6850 | md_geninit(); | |
d710e138 | 6851 | return 0; |
1da177e4 LT |
6852 | } |
6853 | ||
6854 | ||
6855 | #ifndef MODULE | |
6856 | ||
6857 | /* | |
6858 | * Searches all registered partitions for autorun RAID arrays | |
6859 | * at boot time. | |
6860 | */ | |
4d936ec1 ME |
6861 | |
6862 | static LIST_HEAD(all_detected_devices); | |
6863 | struct detected_devices_node { | |
6864 | struct list_head list; | |
6865 | dev_t dev; | |
6866 | }; | |
1da177e4 LT |
6867 | |
6868 | void md_autodetect_dev(dev_t dev) | |
6869 | { | |
4d936ec1 ME |
6870 | struct detected_devices_node *node_detected_dev; |
6871 | ||
6872 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
6873 | if (node_detected_dev) { | |
6874 | node_detected_dev->dev = dev; | |
6875 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
6876 | } else { | |
6877 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
6878 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
6879 | } | |
1da177e4 LT |
6880 | } |
6881 | ||
6882 | ||
6883 | static void autostart_arrays(int part) | |
6884 | { | |
6885 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
6886 | struct detected_devices_node *node_detected_dev; |
6887 | dev_t dev; | |
6888 | int i_scanned, i_passed; | |
1da177e4 | 6889 | |
4d936ec1 ME |
6890 | i_scanned = 0; |
6891 | i_passed = 0; | |
1da177e4 | 6892 | |
4d936ec1 | 6893 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 6894 | |
4d936ec1 ME |
6895 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
6896 | i_scanned++; | |
6897 | node_detected_dev = list_entry(all_detected_devices.next, | |
6898 | struct detected_devices_node, list); | |
6899 | list_del(&node_detected_dev->list); | |
6900 | dev = node_detected_dev->dev; | |
6901 | kfree(node_detected_dev); | |
df968c4e | 6902 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
6903 | if (IS_ERR(rdev)) |
6904 | continue; | |
6905 | ||
b2d444d7 | 6906 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6907 | MD_BUG(); |
6908 | continue; | |
6909 | } | |
d0fae18f | 6910 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 6911 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 6912 | i_passed++; |
1da177e4 | 6913 | } |
4d936ec1 ME |
6914 | |
6915 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
6916 | i_scanned, i_passed); | |
1da177e4 LT |
6917 | |
6918 | autorun_devices(part); | |
6919 | } | |
6920 | ||
fdee8ae4 | 6921 | #endif /* !MODULE */ |
1da177e4 LT |
6922 | |
6923 | static __exit void md_exit(void) | |
6924 | { | |
6925 | mddev_t *mddev; | |
6926 | struct list_head *tmp; | |
8ab5e4c1 | 6927 | |
3dbd8c2e | 6928 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 6929 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 6930 | |
3dbd8c2e | 6931 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
6932 | unregister_blkdev(mdp_major, "mdp"); |
6933 | unregister_reboot_notifier(&md_notifier); | |
6934 | unregister_sysctl_table(raid_table_header); | |
6935 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 6936 | for_each_mddev(mddev, tmp) { |
1da177e4 | 6937 | export_array(mddev); |
d3374825 | 6938 | mddev->hold_active = 0; |
1da177e4 LT |
6939 | } |
6940 | } | |
6941 | ||
685784aa | 6942 | subsys_initcall(md_init); |
1da177e4 LT |
6943 | module_exit(md_exit) |
6944 | ||
f91de92e N |
6945 | static int get_ro(char *buffer, struct kernel_param *kp) |
6946 | { | |
6947 | return sprintf(buffer, "%d", start_readonly); | |
6948 | } | |
6949 | static int set_ro(const char *val, struct kernel_param *kp) | |
6950 | { | |
6951 | char *e; | |
6952 | int num = simple_strtoul(val, &e, 10); | |
6953 | if (*val && (*e == '\0' || *e == '\n')) { | |
6954 | start_readonly = num; | |
4dbcdc75 | 6955 | return 0; |
f91de92e N |
6956 | } |
6957 | return -EINVAL; | |
6958 | } | |
6959 | ||
80ca3a44 N |
6960 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
6961 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 6962 | |
efeb53c0 | 6963 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 6964 | |
1da177e4 LT |
6965 | EXPORT_SYMBOL(register_md_personality); |
6966 | EXPORT_SYMBOL(unregister_md_personality); | |
6967 | EXPORT_SYMBOL(md_error); | |
6968 | EXPORT_SYMBOL(md_done_sync); | |
6969 | EXPORT_SYMBOL(md_write_start); | |
6970 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
6971 | EXPORT_SYMBOL(md_register_thread); |
6972 | EXPORT_SYMBOL(md_unregister_thread); | |
6973 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
6974 | EXPORT_SYMBOL(md_check_recovery); |
6975 | MODULE_LICENSE("GPL"); | |
aa1595e9 | 6976 | MODULE_ALIAS("md"); |
72008652 | 6977 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |