]>
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 | { | |
1da177e4 LT |
101 | .procname = "speed_limit_min", |
102 | .data = &sysctl_speed_limit_min, | |
103 | .maxlen = sizeof(int), | |
80ca3a44 | 104 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 105 | .proc_handler = proc_dointvec, |
1da177e4 LT |
106 | }, |
107 | { | |
1da177e4 LT |
108 | .procname = "speed_limit_max", |
109 | .data = &sysctl_speed_limit_max, | |
110 | .maxlen = sizeof(int), | |
80ca3a44 | 111 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 112 | .proc_handler = proc_dointvec, |
1da177e4 | 113 | }, |
894d2491 | 114 | { } |
1da177e4 LT |
115 | }; |
116 | ||
117 | static ctl_table raid_dir_table[] = { | |
118 | { | |
1da177e4 LT |
119 | .procname = "raid", |
120 | .maxlen = 0, | |
80ca3a44 | 121 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
122 | .child = raid_table, |
123 | }, | |
894d2491 | 124 | { } |
1da177e4 LT |
125 | }; |
126 | ||
127 | static ctl_table raid_root_table[] = { | |
128 | { | |
1da177e4 LT |
129 | .procname = "dev", |
130 | .maxlen = 0, | |
131 | .mode = 0555, | |
132 | .child = raid_dir_table, | |
133 | }, | |
894d2491 | 134 | { } |
1da177e4 LT |
135 | }; |
136 | ||
83d5cde4 | 137 | static const struct block_device_operations md_fops; |
1da177e4 | 138 | |
f91de92e N |
139 | static int start_readonly; |
140 | ||
d7603b7e N |
141 | /* |
142 | * We have a system wide 'event count' that is incremented | |
143 | * on any 'interesting' event, and readers of /proc/mdstat | |
144 | * can use 'poll' or 'select' to find out when the event | |
145 | * count increases. | |
146 | * | |
147 | * Events are: | |
148 | * start array, stop array, error, add device, remove device, | |
149 | * start build, activate spare | |
150 | */ | |
2989ddbd | 151 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 152 | static atomic_t md_event_count; |
29269553 | 153 | void md_new_event(mddev_t *mddev) |
d7603b7e N |
154 | { |
155 | atomic_inc(&md_event_count); | |
156 | wake_up(&md_event_waiters); | |
157 | } | |
29269553 | 158 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 159 | |
c331eb04 N |
160 | /* Alternate version that can be called from interrupts |
161 | * when calling sysfs_notify isn't needed. | |
162 | */ | |
05381954 | 163 | static void md_new_event_inintr(mddev_t *mddev) |
c331eb04 N |
164 | { |
165 | atomic_inc(&md_event_count); | |
166 | wake_up(&md_event_waiters); | |
167 | } | |
168 | ||
1da177e4 LT |
169 | /* |
170 | * Enables to iterate over all existing md arrays | |
171 | * all_mddevs_lock protects this list. | |
172 | */ | |
173 | static LIST_HEAD(all_mddevs); | |
174 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
175 | ||
176 | ||
177 | /* | |
178 | * iterates through all used mddevs in the system. | |
179 | * We take care to grab the all_mddevs_lock whenever navigating | |
180 | * the list, and to always hold a refcount when unlocked. | |
181 | * Any code which breaks out of this loop while own | |
182 | * a reference to the current mddev and must mddev_put it. | |
183 | */ | |
29ac4aa3 | 184 | #define for_each_mddev(mddev,tmp) \ |
1da177e4 LT |
185 | \ |
186 | for (({ spin_lock(&all_mddevs_lock); \ | |
187 | tmp = all_mddevs.next; \ | |
188 | mddev = NULL;}); \ | |
189 | ({ if (tmp != &all_mddevs) \ | |
190 | mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ | |
191 | spin_unlock(&all_mddevs_lock); \ | |
192 | if (mddev) mddev_put(mddev); \ | |
193 | mddev = list_entry(tmp, mddev_t, all_mddevs); \ | |
194 | tmp != &all_mddevs;}); \ | |
195 | ({ spin_lock(&all_mddevs_lock); \ | |
196 | tmp = tmp->next;}) \ | |
197 | ) | |
198 | ||
199 | ||
409c57f3 N |
200 | /* Rather than calling directly into the personality make_request function, |
201 | * IO requests come here first so that we can check if the device is | |
202 | * being suspended pending a reconfiguration. | |
203 | * We hold a refcount over the call to ->make_request. By the time that | |
204 | * call has finished, the bio has been linked into some internal structure | |
205 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
206 | */ | |
207 | static int md_make_request(struct request_queue *q, struct bio *bio) | |
1da177e4 | 208 | { |
409c57f3 N |
209 | mddev_t *mddev = q->queuedata; |
210 | int rv; | |
211 | if (mddev == NULL || mddev->pers == NULL) { | |
212 | bio_io_error(bio); | |
213 | return 0; | |
214 | } | |
215 | rcu_read_lock(); | |
216 | if (mddev->suspended) { | |
217 | DEFINE_WAIT(__wait); | |
218 | for (;;) { | |
219 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
220 | TASK_UNINTERRUPTIBLE); | |
221 | if (!mddev->suspended) | |
222 | break; | |
223 | rcu_read_unlock(); | |
224 | schedule(); | |
225 | rcu_read_lock(); | |
226 | } | |
227 | finish_wait(&mddev->sb_wait, &__wait); | |
228 | } | |
229 | atomic_inc(&mddev->active_io); | |
230 | rcu_read_unlock(); | |
231 | rv = mddev->pers->make_request(q, bio); | |
232 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) | |
233 | wake_up(&mddev->sb_wait); | |
234 | ||
235 | return rv; | |
236 | } | |
237 | ||
238 | static void mddev_suspend(mddev_t *mddev) | |
239 | { | |
240 | BUG_ON(mddev->suspended); | |
241 | mddev->suspended = 1; | |
242 | synchronize_rcu(); | |
243 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
244 | mddev->pers->quiesce(mddev, 1); | |
245 | md_unregister_thread(mddev->thread); | |
246 | mddev->thread = NULL; | |
247 | /* we now know that no code is executing in the personality module, | |
248 | * except possibly the tail end of a ->bi_end_io function, but that | |
249 | * is certain to complete before the module has a chance to get | |
250 | * unloaded | |
251 | */ | |
252 | } | |
253 | ||
254 | static void mddev_resume(mddev_t *mddev) | |
255 | { | |
256 | mddev->suspended = 0; | |
257 | wake_up(&mddev->sb_wait); | |
258 | mddev->pers->quiesce(mddev, 0); | |
1da177e4 LT |
259 | } |
260 | ||
3fa841d7 N |
261 | int mddev_congested(mddev_t *mddev, int bits) |
262 | { | |
263 | return mddev->suspended; | |
264 | } | |
265 | EXPORT_SYMBOL(mddev_congested); | |
266 | ||
409c57f3 | 267 | |
1da177e4 LT |
268 | static inline mddev_t *mddev_get(mddev_t *mddev) |
269 | { | |
270 | atomic_inc(&mddev->active); | |
271 | return mddev; | |
272 | } | |
273 | ||
5fd3a17e | 274 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 275 | |
1da177e4 LT |
276 | static void mddev_put(mddev_t *mddev) |
277 | { | |
278 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
279 | return; | |
d3374825 N |
280 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
281 | !mddev->hold_active) { | |
1da177e4 | 282 | list_del(&mddev->all_mddevs); |
d3374825 N |
283 | if (mddev->gendisk) { |
284 | /* we did a probe so need to clean up. | |
285 | * Call schedule_work inside the spinlock | |
286 | * so that flush_scheduled_work() after | |
287 | * mddev_find will succeed in waiting for the | |
288 | * work to be done. | |
289 | */ | |
290 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
291 | schedule_work(&mddev->del_work); | |
292 | } else | |
293 | kfree(mddev); | |
294 | } | |
295 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
296 | } |
297 | ||
298 | static mddev_t * mddev_find(dev_t unit) | |
299 | { | |
300 | mddev_t *mddev, *new = NULL; | |
301 | ||
302 | retry: | |
303 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
304 | |
305 | if (unit) { | |
306 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
307 | if (mddev->unit == unit) { | |
308 | mddev_get(mddev); | |
309 | spin_unlock(&all_mddevs_lock); | |
310 | kfree(new); | |
311 | return mddev; | |
312 | } | |
313 | ||
314 | if (new) { | |
315 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 316 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
317 | new->hold_active = UNTIL_IOCTL; |
318 | return new; | |
1da177e4 | 319 | } |
efeb53c0 N |
320 | } else if (new) { |
321 | /* find an unused unit number */ | |
322 | static int next_minor = 512; | |
323 | int start = next_minor; | |
324 | int is_free = 0; | |
325 | int dev = 0; | |
326 | while (!is_free) { | |
327 | dev = MKDEV(MD_MAJOR, next_minor); | |
328 | next_minor++; | |
329 | if (next_minor > MINORMASK) | |
330 | next_minor = 0; | |
331 | if (next_minor == start) { | |
332 | /* Oh dear, all in use. */ | |
333 | spin_unlock(&all_mddevs_lock); | |
334 | kfree(new); | |
335 | return NULL; | |
336 | } | |
337 | ||
338 | is_free = 1; | |
339 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
340 | if (mddev->unit == dev) { | |
341 | is_free = 0; | |
342 | break; | |
343 | } | |
344 | } | |
345 | new->unit = dev; | |
346 | new->md_minor = MINOR(dev); | |
347 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
348 | list_add(&new->all_mddevs, &all_mddevs); |
349 | spin_unlock(&all_mddevs_lock); | |
350 | return new; | |
351 | } | |
352 | spin_unlock(&all_mddevs_lock); | |
353 | ||
9ffae0cf | 354 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
355 | if (!new) |
356 | return NULL; | |
357 | ||
1da177e4 LT |
358 | new->unit = unit; |
359 | if (MAJOR(unit) == MD_MAJOR) | |
360 | new->md_minor = MINOR(unit); | |
361 | else | |
362 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
363 | ||
c8c00a69 | 364 | mutex_init(&new->open_mutex); |
df5b89b3 | 365 | mutex_init(&new->reconfig_mutex); |
aa5cbd10 | 366 | mutex_init(&new->bitmap_mutex); |
1da177e4 LT |
367 | INIT_LIST_HEAD(&new->disks); |
368 | INIT_LIST_HEAD(&new->all_mddevs); | |
369 | init_timer(&new->safemode_timer); | |
370 | atomic_set(&new->active, 1); | |
f2ea68cf | 371 | atomic_set(&new->openers, 0); |
409c57f3 | 372 | atomic_set(&new->active_io, 0); |
06d91a5f | 373 | spin_lock_init(&new->write_lock); |
3d310eb7 | 374 | init_waitqueue_head(&new->sb_wait); |
a6d8113a | 375 | init_waitqueue_head(&new->recovery_wait); |
08a02ecd | 376 | new->reshape_position = MaxSector; |
5e96ee65 | 377 | new->resync_min = 0; |
c6207277 | 378 | new->resync_max = MaxSector; |
d897dbf9 | 379 | new->level = LEVEL_NONE; |
1da177e4 | 380 | |
1da177e4 LT |
381 | goto retry; |
382 | } | |
383 | ||
384 | static inline int mddev_lock(mddev_t * mddev) | |
385 | { | |
df5b89b3 | 386 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
387 | } |
388 | ||
b522adcd DW |
389 | static inline int mddev_is_locked(mddev_t *mddev) |
390 | { | |
391 | return mutex_is_locked(&mddev->reconfig_mutex); | |
392 | } | |
393 | ||
1da177e4 LT |
394 | static inline int mddev_trylock(mddev_t * mddev) |
395 | { | |
df5b89b3 | 396 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
397 | } |
398 | ||
399 | static inline void mddev_unlock(mddev_t * mddev) | |
400 | { | |
df5b89b3 | 401 | mutex_unlock(&mddev->reconfig_mutex); |
1da177e4 | 402 | |
005eca5e | 403 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
404 | } |
405 | ||
2989ddbd | 406 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 407 | { |
159ec1fc | 408 | mdk_rdev_t *rdev; |
1da177e4 | 409 | |
159ec1fc | 410 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
411 | if (rdev->desc_nr == nr) |
412 | return rdev; | |
159ec1fc | 413 | |
1da177e4 LT |
414 | return NULL; |
415 | } | |
416 | ||
417 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
418 | { | |
1da177e4 LT |
419 | mdk_rdev_t *rdev; |
420 | ||
159ec1fc | 421 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
422 | if (rdev->bdev->bd_dev == dev) |
423 | return rdev; | |
159ec1fc | 424 | |
1da177e4 LT |
425 | return NULL; |
426 | } | |
427 | ||
d9d166c2 | 428 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
429 | { |
430 | struct mdk_personality *pers; | |
d9d166c2 N |
431 | list_for_each_entry(pers, &pers_list, list) { |
432 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 433 | return pers; |
d9d166c2 N |
434 | if (strcmp(pers->name, clevel)==0) |
435 | return pers; | |
436 | } | |
2604b703 N |
437 | return NULL; |
438 | } | |
439 | ||
b73df2d3 | 440 | /* return the offset of the super block in 512byte sectors */ |
77933d72 | 441 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 | 442 | { |
b73df2d3 AN |
443 | sector_t num_sectors = bdev->bd_inode->i_size / 512; |
444 | return MD_NEW_SIZE_SECTORS(num_sectors); | |
1da177e4 LT |
445 | } |
446 | ||
1da177e4 LT |
447 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
448 | { | |
449 | if (rdev->sb_page) | |
450 | MD_BUG(); | |
451 | ||
452 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
453 | if (!rdev->sb_page) { | |
454 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 455 | return -ENOMEM; |
1da177e4 LT |
456 | } |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
461 | static void free_disk_sb(mdk_rdev_t * rdev) | |
462 | { | |
463 | if (rdev->sb_page) { | |
2d1f3b5d | 464 | put_page(rdev->sb_page); |
1da177e4 LT |
465 | rdev->sb_loaded = 0; |
466 | rdev->sb_page = NULL; | |
0f420358 | 467 | rdev->sb_start = 0; |
dd8ac336 | 468 | rdev->sectors = 0; |
1da177e4 LT |
469 | } |
470 | } | |
471 | ||
472 | ||
6712ecf8 | 473 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
474 | { |
475 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 476 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 477 | |
3a0f5bbb N |
478 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
479 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
480 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
481 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 482 | md_error(mddev, rdev); |
3a0f5bbb | 483 | } |
7bfa19f2 | 484 | |
a9701a30 N |
485 | if (atomic_dec_and_test(&mddev->pending_writes)) |
486 | wake_up(&mddev->sb_wait); | |
f8b58edf | 487 | bio_put(bio); |
7bfa19f2 N |
488 | } |
489 | ||
6712ecf8 | 490 | static void super_written_barrier(struct bio *bio, int error) |
a9701a30 N |
491 | { |
492 | struct bio *bio2 = bio->bi_private; | |
493 | mdk_rdev_t *rdev = bio2->bi_private; | |
494 | mddev_t *mddev = rdev->mddev; | |
a9701a30 N |
495 | |
496 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
497 | error == -EOPNOTSUPP) { | |
498 | unsigned long flags; | |
499 | /* barriers don't appear to be supported :-( */ | |
500 | set_bit(BarriersNotsupp, &rdev->flags); | |
501 | mddev->barriers_work = 0; | |
502 | spin_lock_irqsave(&mddev->write_lock, flags); | |
503 | bio2->bi_next = mddev->biolist; | |
504 | mddev->biolist = bio2; | |
505 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
506 | wake_up(&mddev->sb_wait); | |
507 | bio_put(bio); | |
6712ecf8 N |
508 | } else { |
509 | bio_put(bio2); | |
510 | bio->bi_private = rdev; | |
511 | super_written(bio, error); | |
a9701a30 | 512 | } |
a9701a30 N |
513 | } |
514 | ||
7bfa19f2 N |
515 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
516 | sector_t sector, int size, struct page *page) | |
517 | { | |
518 | /* write first size bytes of page to sector of rdev | |
519 | * Increment mddev->pending_writes before returning | |
520 | * and decrement it on completion, waking up sb_wait | |
521 | * if zero is reached. | |
522 | * If an error occurred, call md_error | |
a9701a30 N |
523 | * |
524 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
525 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
526 | */ |
527 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
93dbb393 | 528 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG); |
7bfa19f2 N |
529 | |
530 | bio->bi_bdev = rdev->bdev; | |
531 | bio->bi_sector = sector; | |
532 | bio_add_page(bio, page, size, 0); | |
533 | bio->bi_private = rdev; | |
534 | bio->bi_end_io = super_written; | |
a9701a30 N |
535 | bio->bi_rw = rw; |
536 | ||
7bfa19f2 | 537 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
538 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
539 | struct bio *rbio; | |
540 | rw |= (1<<BIO_RW_BARRIER); | |
541 | rbio = bio_clone(bio, GFP_NOIO); | |
542 | rbio->bi_private = bio; | |
543 | rbio->bi_end_io = super_written_barrier; | |
544 | submit_bio(rw, rbio); | |
545 | } else | |
546 | submit_bio(rw, bio); | |
547 | } | |
548 | ||
549 | void md_super_wait(mddev_t *mddev) | |
550 | { | |
551 | /* wait for all superblock writes that were scheduled to complete. | |
552 | * if any had to be retried (due to BARRIER problems), retry them | |
553 | */ | |
554 | DEFINE_WAIT(wq); | |
555 | for(;;) { | |
556 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
557 | if (atomic_read(&mddev->pending_writes)==0) | |
558 | break; | |
559 | while (mddev->biolist) { | |
560 | struct bio *bio; | |
561 | spin_lock_irq(&mddev->write_lock); | |
562 | bio = mddev->biolist; | |
563 | mddev->biolist = bio->bi_next ; | |
564 | bio->bi_next = NULL; | |
565 | spin_unlock_irq(&mddev->write_lock); | |
566 | submit_bio(bio->bi_rw, bio); | |
567 | } | |
568 | schedule(); | |
569 | } | |
570 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
571 | } |
572 | ||
6712ecf8 | 573 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 574 | { |
1da177e4 | 575 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
576 | } |
577 | ||
a654b9d8 | 578 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
579 | struct page *page, int rw) |
580 | { | |
baaa2c51 | 581 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
582 | struct completion event; |
583 | int ret; | |
584 | ||
93dbb393 | 585 | rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG); |
1da177e4 LT |
586 | |
587 | bio->bi_bdev = bdev; | |
588 | bio->bi_sector = sector; | |
589 | bio_add_page(bio, page, size, 0); | |
590 | init_completion(&event); | |
591 | bio->bi_private = &event; | |
592 | bio->bi_end_io = bi_complete; | |
593 | submit_bio(rw, bio); | |
594 | wait_for_completion(&event); | |
595 | ||
596 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
597 | bio_put(bio); | |
598 | return ret; | |
599 | } | |
a8745db2 | 600 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 601 | |
0002b271 | 602 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
603 | { |
604 | char b[BDEVNAME_SIZE]; | |
605 | if (!rdev->sb_page) { | |
606 | MD_BUG(); | |
607 | return -EINVAL; | |
608 | } | |
609 | if (rdev->sb_loaded) | |
610 | return 0; | |
611 | ||
612 | ||
0f420358 | 613 | if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ)) |
1da177e4 LT |
614 | goto fail; |
615 | rdev->sb_loaded = 1; | |
616 | return 0; | |
617 | ||
618 | fail: | |
619 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
620 | bdevname(rdev->bdev,b)); | |
621 | return -EINVAL; | |
622 | } | |
623 | ||
624 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
625 | { | |
05710466 AN |
626 | return sb1->set_uuid0 == sb2->set_uuid0 && |
627 | sb1->set_uuid1 == sb2->set_uuid1 && | |
628 | sb1->set_uuid2 == sb2->set_uuid2 && | |
629 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
630 | } |
631 | ||
1da177e4 LT |
632 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
633 | { | |
634 | int ret; | |
635 | mdp_super_t *tmp1, *tmp2; | |
636 | ||
637 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
638 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
639 | ||
640 | if (!tmp1 || !tmp2) { | |
641 | ret = 0; | |
35020f1a | 642 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
643 | goto abort; |
644 | } | |
645 | ||
646 | *tmp1 = *sb1; | |
647 | *tmp2 = *sb2; | |
648 | ||
649 | /* | |
650 | * nr_disks is not constant | |
651 | */ | |
652 | tmp1->nr_disks = 0; | |
653 | tmp2->nr_disks = 0; | |
654 | ||
ce0c8e05 | 655 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 656 | abort: |
990a8baf JJ |
657 | kfree(tmp1); |
658 | kfree(tmp2); | |
1da177e4 LT |
659 | return ret; |
660 | } | |
661 | ||
4d167f09 N |
662 | |
663 | static u32 md_csum_fold(u32 csum) | |
664 | { | |
665 | csum = (csum & 0xffff) + (csum >> 16); | |
666 | return (csum & 0xffff) + (csum >> 16); | |
667 | } | |
668 | ||
1da177e4 LT |
669 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
670 | { | |
4d167f09 N |
671 | u64 newcsum = 0; |
672 | u32 *sb32 = (u32*)sb; | |
673 | int i; | |
1da177e4 LT |
674 | unsigned int disk_csum, csum; |
675 | ||
676 | disk_csum = sb->sb_csum; | |
677 | sb->sb_csum = 0; | |
4d167f09 N |
678 | |
679 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
680 | newcsum += sb32[i]; | |
681 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
682 | ||
683 | ||
684 | #ifdef CONFIG_ALPHA | |
685 | /* This used to use csum_partial, which was wrong for several | |
686 | * reasons including that different results are returned on | |
687 | * different architectures. It isn't critical that we get exactly | |
688 | * the same return value as before (we always csum_fold before | |
689 | * testing, and that removes any differences). However as we | |
690 | * know that csum_partial always returned a 16bit value on | |
691 | * alphas, do a fold to maximise conformity to previous behaviour. | |
692 | */ | |
693 | sb->sb_csum = md_csum_fold(disk_csum); | |
694 | #else | |
1da177e4 | 695 | sb->sb_csum = disk_csum; |
4d167f09 | 696 | #endif |
1da177e4 LT |
697 | return csum; |
698 | } | |
699 | ||
700 | ||
701 | /* | |
702 | * Handle superblock details. | |
703 | * We want to be able to handle multiple superblock formats | |
704 | * so we have a common interface to them all, and an array of | |
705 | * different handlers. | |
706 | * We rely on user-space to write the initial superblock, and support | |
707 | * reading and updating of superblocks. | |
708 | * Interface methods are: | |
709 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
710 | * loads and validates a superblock on dev. | |
711 | * if refdev != NULL, compare superblocks on both devices | |
712 | * Return: | |
713 | * 0 - dev has a superblock that is compatible with refdev | |
714 | * 1 - dev has a superblock that is compatible and newer than refdev | |
715 | * so dev should be used as the refdev in future | |
716 | * -EINVAL superblock incompatible or invalid | |
717 | * -othererror e.g. -EIO | |
718 | * | |
719 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
720 | * Verify that dev is acceptable into mddev. | |
721 | * The first time, mddev->raid_disks will be 0, and data from | |
722 | * dev should be merged in. Subsequent calls check that dev | |
723 | * is new enough. Return 0 or -EINVAL | |
724 | * | |
725 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
726 | * Update the superblock for rdev with data in mddev | |
727 | * This does not write to disc. | |
728 | * | |
729 | */ | |
730 | ||
731 | struct super_type { | |
0cd17fec CW |
732 | char *name; |
733 | struct module *owner; | |
734 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
735 | int minor_version); | |
736 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
737 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
738 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 739 | sector_t num_sectors); |
1da177e4 LT |
740 | }; |
741 | ||
0894cc30 AN |
742 | /* |
743 | * Check that the given mddev has no bitmap. | |
744 | * | |
745 | * This function is called from the run method of all personalities that do not | |
746 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
747 | * has a bitmap. Otherwise, it returns 0. | |
748 | * | |
749 | */ | |
750 | int md_check_no_bitmap(mddev_t *mddev) | |
751 | { | |
752 | if (!mddev->bitmap_file && !mddev->bitmap_offset) | |
753 | return 0; | |
754 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
755 | mdname(mddev), mddev->pers->name); | |
756 | return 1; | |
757 | } | |
758 | EXPORT_SYMBOL(md_check_no_bitmap); | |
759 | ||
1da177e4 LT |
760 | /* |
761 | * load_super for 0.90.0 | |
762 | */ | |
763 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
764 | { | |
765 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
766 | mdp_super_t *sb; | |
767 | int ret; | |
1da177e4 LT |
768 | |
769 | /* | |
0f420358 | 770 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
771 | * it's at the end of the disk. |
772 | * | |
773 | * It also happens to be a multiple of 4Kb. | |
774 | */ | |
0f420358 | 775 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 776 | |
0002b271 | 777 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
778 | if (ret) return ret; |
779 | ||
780 | ret = -EINVAL; | |
781 | ||
782 | bdevname(rdev->bdev, b); | |
783 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
784 | ||
785 | if (sb->md_magic != MD_SB_MAGIC) { | |
786 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
787 | b); | |
788 | goto abort; | |
789 | } | |
790 | ||
791 | if (sb->major_version != 0 || | |
f6705578 N |
792 | sb->minor_version < 90 || |
793 | sb->minor_version > 91) { | |
1da177e4 LT |
794 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
795 | sb->major_version, sb->minor_version, | |
796 | b); | |
797 | goto abort; | |
798 | } | |
799 | ||
800 | if (sb->raid_disks <= 0) | |
801 | goto abort; | |
802 | ||
4d167f09 | 803 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
804 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
805 | b); | |
806 | goto abort; | |
807 | } | |
808 | ||
809 | rdev->preferred_minor = sb->md_minor; | |
810 | rdev->data_offset = 0; | |
0002b271 | 811 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
812 | |
813 | if (sb->level == LEVEL_MULTIPATH) | |
814 | rdev->desc_nr = -1; | |
815 | else | |
816 | rdev->desc_nr = sb->this_disk.number; | |
817 | ||
9a7b2b0f | 818 | if (!refdev) { |
1da177e4 | 819 | ret = 1; |
9a7b2b0f | 820 | } else { |
1da177e4 LT |
821 | __u64 ev1, ev2; |
822 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
823 | if (!uuid_equal(refsb, sb)) { | |
824 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
825 | b, bdevname(refdev->bdev,b2)); | |
826 | goto abort; | |
827 | } | |
828 | if (!sb_equal(refsb, sb)) { | |
829 | printk(KERN_WARNING "md: %s has same UUID" | |
830 | " but different superblock to %s\n", | |
831 | b, bdevname(refdev->bdev, b2)); | |
832 | goto abort; | |
833 | } | |
834 | ev1 = md_event(sb); | |
835 | ev2 = md_event(refsb); | |
836 | if (ev1 > ev2) | |
837 | ret = 1; | |
838 | else | |
839 | ret = 0; | |
840 | } | |
8190e754 | 841 | rdev->sectors = rdev->sb_start; |
1da177e4 | 842 | |
dd8ac336 | 843 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
844 | /* "this cannot possibly happen" ... */ |
845 | ret = -EINVAL; | |
846 | ||
1da177e4 LT |
847 | abort: |
848 | return ret; | |
849 | } | |
850 | ||
851 | /* | |
852 | * validate_super for 0.90.0 | |
853 | */ | |
854 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
855 | { | |
856 | mdp_disk_t *desc; | |
857 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 858 | __u64 ev1 = md_event(sb); |
1da177e4 | 859 | |
41158c7e | 860 | rdev->raid_disk = -1; |
c5d79adb N |
861 | clear_bit(Faulty, &rdev->flags); |
862 | clear_bit(In_sync, &rdev->flags); | |
863 | clear_bit(WriteMostly, &rdev->flags); | |
864 | clear_bit(BarriersNotsupp, &rdev->flags); | |
865 | ||
1da177e4 LT |
866 | if (mddev->raid_disks == 0) { |
867 | mddev->major_version = 0; | |
868 | mddev->minor_version = sb->minor_version; | |
869 | mddev->patch_version = sb->patch_version; | |
e691063a | 870 | mddev->external = 0; |
9d8f0363 | 871 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
872 | mddev->ctime = sb->ctime; |
873 | mddev->utime = sb->utime; | |
874 | mddev->level = sb->level; | |
d9d166c2 | 875 | mddev->clevel[0] = 0; |
1da177e4 LT |
876 | mddev->layout = sb->layout; |
877 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 878 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 879 | mddev->events = ev1; |
9223214e | 880 | mddev->bitmap_offset = 0; |
36fa3063 | 881 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
1da177e4 | 882 | |
f6705578 N |
883 | if (mddev->minor_version >= 91) { |
884 | mddev->reshape_position = sb->reshape_position; | |
885 | mddev->delta_disks = sb->delta_disks; | |
886 | mddev->new_level = sb->new_level; | |
887 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 888 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
889 | } else { |
890 | mddev->reshape_position = MaxSector; | |
891 | mddev->delta_disks = 0; | |
892 | mddev->new_level = mddev->level; | |
893 | mddev->new_layout = mddev->layout; | |
664e7c41 | 894 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
895 | } |
896 | ||
1da177e4 LT |
897 | if (sb->state & (1<<MD_SB_CLEAN)) |
898 | mddev->recovery_cp = MaxSector; | |
899 | else { | |
900 | if (sb->events_hi == sb->cp_events_hi && | |
901 | sb->events_lo == sb->cp_events_lo) { | |
902 | mddev->recovery_cp = sb->recovery_cp; | |
903 | } else | |
904 | mddev->recovery_cp = 0; | |
905 | } | |
906 | ||
907 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
908 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
909 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
910 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
911 | ||
912 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
913 | |
914 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
e11e93fa | 915 | mddev->bitmap_file == NULL) |
36fa3063 | 916 | mddev->bitmap_offset = mddev->default_bitmap_offset; |
a654b9d8 | 917 | |
41158c7e N |
918 | } else if (mddev->pers == NULL) { |
919 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
920 | ++ev1; |
921 | if (ev1 < mddev->events) | |
922 | return -EINVAL; | |
41158c7e N |
923 | } else if (mddev->bitmap) { |
924 | /* if adding to array with a bitmap, then we can accept an | |
925 | * older device ... but not too old. | |
926 | */ | |
41158c7e N |
927 | if (ev1 < mddev->bitmap->events_cleared) |
928 | return 0; | |
07d84d10 N |
929 | } else { |
930 | if (ev1 < mddev->events) | |
931 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
932 | return 0; | |
933 | } | |
41158c7e | 934 | |
1da177e4 | 935 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
936 | desc = sb->disks + rdev->desc_nr; |
937 | ||
938 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 939 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
940 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
941 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 942 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 943 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
944 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
945 | /* active but not in sync implies recovery up to | |
946 | * reshape position. We don't know exactly where | |
947 | * that is, so set to zero for now */ | |
948 | if (mddev->minor_version >= 91) { | |
949 | rdev->recovery_offset = 0; | |
950 | rdev->raid_disk = desc->raid_disk; | |
951 | } | |
1da177e4 | 952 | } |
8ddf9efe N |
953 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
954 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 955 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 956 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
957 | return 0; |
958 | } | |
959 | ||
960 | /* | |
961 | * sync_super for 0.90.0 | |
962 | */ | |
963 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
964 | { | |
965 | mdp_super_t *sb; | |
1da177e4 LT |
966 | mdk_rdev_t *rdev2; |
967 | int next_spare = mddev->raid_disks; | |
19133a42 | 968 | |
1da177e4 LT |
969 | |
970 | /* make rdev->sb match mddev data.. | |
971 | * | |
972 | * 1/ zero out disks | |
973 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
974 | * 3/ any empty disks < next_spare become removed | |
975 | * | |
976 | * disks[0] gets initialised to REMOVED because | |
977 | * we cannot be sure from other fields if it has | |
978 | * been initialised or not. | |
979 | */ | |
980 | int i; | |
981 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
982 | ||
61181565 N |
983 | rdev->sb_size = MD_SB_BYTES; |
984 | ||
1da177e4 LT |
985 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
986 | ||
987 | memset(sb, 0, sizeof(*sb)); | |
988 | ||
989 | sb->md_magic = MD_SB_MAGIC; | |
990 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
991 | sb->patch_version = mddev->patch_version; |
992 | sb->gvalid_words = 0; /* ignored */ | |
993 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
994 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
995 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
996 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
997 | ||
998 | sb->ctime = mddev->ctime; | |
999 | sb->level = mddev->level; | |
58c0fed4 | 1000 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1001 | sb->raid_disks = mddev->raid_disks; |
1002 | sb->md_minor = mddev->md_minor; | |
e691063a | 1003 | sb->not_persistent = 0; |
1da177e4 LT |
1004 | sb->utime = mddev->utime; |
1005 | sb->state = 0; | |
1006 | sb->events_hi = (mddev->events>>32); | |
1007 | sb->events_lo = (u32)mddev->events; | |
1008 | ||
f6705578 N |
1009 | if (mddev->reshape_position == MaxSector) |
1010 | sb->minor_version = 90; | |
1011 | else { | |
1012 | sb->minor_version = 91; | |
1013 | sb->reshape_position = mddev->reshape_position; | |
1014 | sb->new_level = mddev->new_level; | |
1015 | sb->delta_disks = mddev->delta_disks; | |
1016 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1017 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1018 | } |
1019 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1020 | if (mddev->in_sync) |
1021 | { | |
1022 | sb->recovery_cp = mddev->recovery_cp; | |
1023 | sb->cp_events_hi = (mddev->events>>32); | |
1024 | sb->cp_events_lo = (u32)mddev->events; | |
1025 | if (mddev->recovery_cp == MaxSector) | |
1026 | sb->state = (1<< MD_SB_CLEAN); | |
1027 | } else | |
1028 | sb->recovery_cp = 0; | |
1029 | ||
1030 | sb->layout = mddev->layout; | |
9d8f0363 | 1031 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1032 | |
a654b9d8 N |
1033 | if (mddev->bitmap && mddev->bitmap_file == NULL) |
1034 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); | |
1035 | ||
1da177e4 | 1036 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1037 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1038 | mdp_disk_t *d; |
86e6ffdd | 1039 | int desc_nr; |
0261cd9f N |
1040 | int is_active = test_bit(In_sync, &rdev2->flags); |
1041 | ||
1042 | if (rdev2->raid_disk >= 0 && | |
1043 | sb->minor_version >= 91) | |
1044 | /* we have nowhere to store the recovery_offset, | |
1045 | * but if it is not below the reshape_position, | |
1046 | * we can piggy-back on that. | |
1047 | */ | |
1048 | is_active = 1; | |
1049 | if (rdev2->raid_disk < 0 || | |
1050 | test_bit(Faulty, &rdev2->flags)) | |
1051 | is_active = 0; | |
1052 | if (is_active) | |
86e6ffdd | 1053 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1054 | else |
86e6ffdd | 1055 | desc_nr = next_spare++; |
19133a42 | 1056 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1057 | d = &sb->disks[rdev2->desc_nr]; |
1058 | nr_disks++; | |
1059 | d->number = rdev2->desc_nr; | |
1060 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1061 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1062 | if (is_active) |
1da177e4 LT |
1063 | d->raid_disk = rdev2->raid_disk; |
1064 | else | |
1065 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1066 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1067 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1068 | else if (is_active) { |
1da177e4 | 1069 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1070 | if (test_bit(In_sync, &rdev2->flags)) |
1071 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1072 | active++; |
1073 | working++; | |
1074 | } else { | |
1075 | d->state = 0; | |
1076 | spare++; | |
1077 | working++; | |
1078 | } | |
8ddf9efe N |
1079 | if (test_bit(WriteMostly, &rdev2->flags)) |
1080 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1081 | } |
1da177e4 LT |
1082 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1083 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1084 | mdp_disk_t *d = &sb->disks[i]; | |
1085 | if (d->state == 0 && d->number == 0) { | |
1086 | d->number = i; | |
1087 | d->raid_disk = i; | |
1088 | d->state = (1<<MD_DISK_REMOVED); | |
1089 | d->state |= (1<<MD_DISK_FAULTY); | |
1090 | failed++; | |
1091 | } | |
1092 | } | |
1093 | sb->nr_disks = nr_disks; | |
1094 | sb->active_disks = active; | |
1095 | sb->working_disks = working; | |
1096 | sb->failed_disks = failed; | |
1097 | sb->spare_disks = spare; | |
1098 | ||
1099 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1100 | sb->sb_csum = calc_sb_csum(sb); | |
1101 | } | |
1102 | ||
0cd17fec CW |
1103 | /* |
1104 | * rdev_size_change for 0.90.0 | |
1105 | */ | |
1106 | static unsigned long long | |
15f4a5fd | 1107 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1108 | { |
58c0fed4 | 1109 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1110 | return 0; /* component must fit device */ |
0cd17fec CW |
1111 | if (rdev->mddev->bitmap_offset) |
1112 | return 0; /* can't move bitmap */ | |
0f420358 | 1113 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
15f4a5fd AN |
1114 | if (!num_sectors || num_sectors > rdev->sb_start) |
1115 | num_sectors = rdev->sb_start; | |
0f420358 | 1116 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1117 | rdev->sb_page); |
1118 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1119 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec CW |
1120 | } |
1121 | ||
1122 | ||
1da177e4 LT |
1123 | /* |
1124 | * version 1 superblock | |
1125 | */ | |
1126 | ||
1c05b4bc | 1127 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1128 | { |
1c05b4bc N |
1129 | __le32 disk_csum; |
1130 | u32 csum; | |
1da177e4 LT |
1131 | unsigned long long newcsum; |
1132 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1133 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1134 | int i; |
1135 | ||
1136 | disk_csum = sb->sb_csum; | |
1137 | sb->sb_csum = 0; | |
1138 | newcsum = 0; | |
1139 | for (i=0; size>=4; size -= 4 ) | |
1140 | newcsum += le32_to_cpu(*isuper++); | |
1141 | ||
1142 | if (size == 2) | |
1c05b4bc | 1143 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1144 | |
1145 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1146 | sb->sb_csum = disk_csum; | |
1147 | return cpu_to_le32(csum); | |
1148 | } | |
1149 | ||
1150 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1151 | { | |
1152 | struct mdp_superblock_1 *sb; | |
1153 | int ret; | |
0f420358 | 1154 | sector_t sb_start; |
1da177e4 | 1155 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1156 | int bmask; |
1da177e4 LT |
1157 | |
1158 | /* | |
0f420358 | 1159 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1160 | * It is always aligned to a 4K boundary and |
1161 | * depeding on minor_version, it can be: | |
1162 | * 0: At least 8K, but less than 12K, from end of device | |
1163 | * 1: At start of device | |
1164 | * 2: 4K from start of device. | |
1165 | */ | |
1166 | switch(minor_version) { | |
1167 | case 0: | |
0f420358 AN |
1168 | sb_start = rdev->bdev->bd_inode->i_size >> 9; |
1169 | sb_start -= 8*2; | |
1170 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1171 | break; |
1172 | case 1: | |
0f420358 | 1173 | sb_start = 0; |
1da177e4 LT |
1174 | break; |
1175 | case 2: | |
0f420358 | 1176 | sb_start = 8; |
1da177e4 LT |
1177 | break; |
1178 | default: | |
1179 | return -EINVAL; | |
1180 | } | |
0f420358 | 1181 | rdev->sb_start = sb_start; |
1da177e4 | 1182 | |
0002b271 N |
1183 | /* superblock is rarely larger than 1K, but it can be larger, |
1184 | * and it is safe to read 4k, so we do that | |
1185 | */ | |
1186 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1187 | if (ret) return ret; |
1188 | ||
1189 | ||
1190 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1191 | ||
1192 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1193 | sb->major_version != cpu_to_le32(1) || | |
1194 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1195 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1196 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1197 | return -EINVAL; |
1198 | ||
1199 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1200 | printk("md: invalid superblock checksum on %s\n", | |
1201 | bdevname(rdev->bdev,b)); | |
1202 | return -EINVAL; | |
1203 | } | |
1204 | if (le64_to_cpu(sb->data_size) < 10) { | |
1205 | printk("md: data_size too small on %s\n", | |
1206 | bdevname(rdev->bdev,b)); | |
1207 | return -EINVAL; | |
1208 | } | |
e11e93fa | 1209 | |
1da177e4 LT |
1210 | rdev->preferred_minor = 0xffff; |
1211 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1212 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1213 | |
0002b271 | 1214 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1215 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1216 | if (rdev->sb_size & bmask) |
a1801f85 N |
1217 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1218 | ||
1219 | if (minor_version | |
0f420358 | 1220 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1221 | return -EINVAL; |
0002b271 | 1222 | |
31b65a0d N |
1223 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1224 | rdev->desc_nr = -1; | |
1225 | else | |
1226 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1227 | ||
9a7b2b0f | 1228 | if (!refdev) { |
8ed75463 | 1229 | ret = 1; |
9a7b2b0f | 1230 | } else { |
1da177e4 LT |
1231 | __u64 ev1, ev2; |
1232 | struct mdp_superblock_1 *refsb = | |
1233 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1234 | ||
1235 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1236 | sb->level != refsb->level || | |
1237 | sb->layout != refsb->layout || | |
1238 | sb->chunksize != refsb->chunksize) { | |
1239 | printk(KERN_WARNING "md: %s has strangely different" | |
1240 | " superblock to %s\n", | |
1241 | bdevname(rdev->bdev,b), | |
1242 | bdevname(refdev->bdev,b2)); | |
1243 | return -EINVAL; | |
1244 | } | |
1245 | ev1 = le64_to_cpu(sb->events); | |
1246 | ev2 = le64_to_cpu(refsb->events); | |
1247 | ||
1248 | if (ev1 > ev2) | |
8ed75463 N |
1249 | ret = 1; |
1250 | else | |
1251 | ret = 0; | |
1da177e4 | 1252 | } |
a1801f85 | 1253 | if (minor_version) |
dd8ac336 AN |
1254 | rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) - |
1255 | le64_to_cpu(sb->data_offset); | |
1da177e4 | 1256 | else |
dd8ac336 AN |
1257 | rdev->sectors = rdev->sb_start; |
1258 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1259 | return -EINVAL; |
dd8ac336 | 1260 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1261 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1262 | return -EINVAL; |
8ed75463 | 1263 | return ret; |
1da177e4 LT |
1264 | } |
1265 | ||
1266 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1267 | { | |
1268 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1269 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1270 | |
41158c7e | 1271 | rdev->raid_disk = -1; |
c5d79adb N |
1272 | clear_bit(Faulty, &rdev->flags); |
1273 | clear_bit(In_sync, &rdev->flags); | |
1274 | clear_bit(WriteMostly, &rdev->flags); | |
1275 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1276 | ||
1da177e4 LT |
1277 | if (mddev->raid_disks == 0) { |
1278 | mddev->major_version = 1; | |
1279 | mddev->patch_version = 0; | |
e691063a | 1280 | mddev->external = 0; |
9d8f0363 | 1281 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1282 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1283 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1284 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1285 | mddev->clevel[0] = 0; |
1da177e4 LT |
1286 | mddev->layout = le32_to_cpu(sb->layout); |
1287 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1288 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1289 | mddev->events = ev1; |
9223214e | 1290 | mddev->bitmap_offset = 0; |
29fc7e3e | 1291 | mddev->default_bitmap_offset = 1024 >> 9; |
1da177e4 LT |
1292 | |
1293 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1294 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1295 | ||
1296 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1297 | |
71c0805c | 1298 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
e11e93fa | 1299 | mddev->bitmap_file == NULL ) |
a654b9d8 | 1300 | mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); |
e11e93fa | 1301 | |
f6705578 N |
1302 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1303 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1304 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1305 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1306 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1307 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1308 | } else { |
1309 | mddev->reshape_position = MaxSector; | |
1310 | mddev->delta_disks = 0; | |
1311 | mddev->new_level = mddev->level; | |
1312 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1313 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1314 | } |
1315 | ||
41158c7e N |
1316 | } else if (mddev->pers == NULL) { |
1317 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1318 | ++ev1; |
1319 | if (ev1 < mddev->events) | |
1320 | return -EINVAL; | |
41158c7e N |
1321 | } else if (mddev->bitmap) { |
1322 | /* If adding to array with a bitmap, then we can accept an | |
1323 | * older device, but not too old. | |
1324 | */ | |
41158c7e N |
1325 | if (ev1 < mddev->bitmap->events_cleared) |
1326 | return 0; | |
07d84d10 N |
1327 | } else { |
1328 | if (ev1 < mddev->events) | |
1329 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1330 | return 0; | |
1331 | } | |
1da177e4 LT |
1332 | if (mddev->level != LEVEL_MULTIPATH) { |
1333 | int role; | |
3673f305 N |
1334 | if (rdev->desc_nr < 0 || |
1335 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1336 | role = 0xffff; | |
1337 | rdev->desc_nr = -1; | |
1338 | } else | |
1339 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1340 | switch(role) { |
1341 | case 0xffff: /* spare */ | |
1da177e4 LT |
1342 | break; |
1343 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1344 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1345 | break; |
1346 | default: | |
5fd6c1dc N |
1347 | if ((le32_to_cpu(sb->feature_map) & |
1348 | MD_FEATURE_RECOVERY_OFFSET)) | |
1349 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1350 | else | |
1351 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1352 | rdev->raid_disk = role; |
1353 | break; | |
1354 | } | |
8ddf9efe N |
1355 | if (sb->devflags & WriteMostly1) |
1356 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1357 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1358 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1359 | |
1da177e4 LT |
1360 | return 0; |
1361 | } | |
1362 | ||
1363 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1364 | { | |
1365 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1366 | mdk_rdev_t *rdev2; |
1367 | int max_dev, i; | |
1368 | /* make rdev->sb match mddev and rdev data. */ | |
1369 | ||
1370 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1371 | ||
1372 | sb->feature_map = 0; | |
1373 | sb->pad0 = 0; | |
5fd6c1dc | 1374 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1375 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1376 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1377 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1378 | ||
1379 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1380 | sb->events = cpu_to_le64(mddev->events); | |
1381 | if (mddev->in_sync) | |
1382 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1383 | else | |
1384 | sb->resync_offset = cpu_to_le64(0); | |
1385 | ||
1c05b4bc | 1386 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1387 | |
f0ca340c | 1388 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1389 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1390 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1391 | sb->level = cpu_to_le32(mddev->level); |
1392 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1393 | |
a654b9d8 N |
1394 | if (mddev->bitmap && mddev->bitmap_file == NULL) { |
1395 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); | |
71c0805c | 1396 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1397 | } |
5fd6c1dc N |
1398 | |
1399 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1400 | !test_bit(In_sync, &rdev->flags)) { |
97e4f42d N |
1401 | if (rdev->recovery_offset > 0) { |
1402 | sb->feature_map |= | |
1403 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1404 | sb->recovery_offset = | |
1405 | cpu_to_le64(rdev->recovery_offset); | |
1406 | } | |
5fd6c1dc N |
1407 | } |
1408 | ||
f6705578 N |
1409 | if (mddev->reshape_position != MaxSector) { |
1410 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1411 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1412 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1413 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1414 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1415 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1416 | } |
a654b9d8 | 1417 | |
1da177e4 | 1418 | max_dev = 0; |
159ec1fc | 1419 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1420 | if (rdev2->desc_nr+1 > max_dev) |
1421 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1422 | |
70471daf N |
1423 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1424 | int bmask; | |
a778b73f | 1425 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1426 | rdev->sb_size = max_dev * 2 + 256; |
1427 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1428 | if (rdev->sb_size & bmask) | |
1429 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
1430 | } | |
1da177e4 LT |
1431 | for (i=0; i<max_dev;i++) |
1432 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1433 | ||
159ec1fc | 1434 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1435 | i = rdev2->desc_nr; |
b2d444d7 | 1436 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1437 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1438 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1439 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
5fd6c1dc N |
1440 | else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0) |
1441 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); | |
1da177e4 LT |
1442 | else |
1443 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1444 | } | |
1445 | ||
1da177e4 LT |
1446 | sb->sb_csum = calc_sb_1_csum(sb); |
1447 | } | |
1448 | ||
0cd17fec | 1449 | static unsigned long long |
15f4a5fd | 1450 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1451 | { |
1452 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1453 | sector_t max_sectors; |
58c0fed4 | 1454 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1455 | return 0; /* component must fit device */ |
0f420358 | 1456 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1457 | /* minor versions 1 and 2; superblock before data */ |
15f4a5fd AN |
1458 | max_sectors = rdev->bdev->bd_inode->i_size >> 9; |
1459 | max_sectors -= rdev->data_offset; | |
1460 | if (!num_sectors || num_sectors > max_sectors) | |
1461 | num_sectors = max_sectors; | |
0cd17fec CW |
1462 | } else if (rdev->mddev->bitmap_offset) { |
1463 | /* minor version 0 with bitmap we can't move */ | |
1464 | return 0; | |
1465 | } else { | |
1466 | /* minor version 0; superblock after data */ | |
0f420358 AN |
1467 | sector_t sb_start; |
1468 | sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; | |
1469 | sb_start &= ~(sector_t)(4*2 - 1); | |
dd8ac336 | 1470 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1471 | if (!num_sectors || num_sectors > max_sectors) |
1472 | num_sectors = max_sectors; | |
0f420358 | 1473 | rdev->sb_start = sb_start; |
0cd17fec CW |
1474 | } |
1475 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1476 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1477 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1478 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1479 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1480 | rdev->sb_page); |
1481 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1482 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec | 1483 | } |
1da177e4 | 1484 | |
75c96f85 | 1485 | static struct super_type super_types[] = { |
1da177e4 LT |
1486 | [0] = { |
1487 | .name = "0.90.0", | |
1488 | .owner = THIS_MODULE, | |
0cd17fec CW |
1489 | .load_super = super_90_load, |
1490 | .validate_super = super_90_validate, | |
1491 | .sync_super = super_90_sync, | |
1492 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1493 | }, |
1494 | [1] = { | |
1495 | .name = "md-1", | |
1496 | .owner = THIS_MODULE, | |
0cd17fec CW |
1497 | .load_super = super_1_load, |
1498 | .validate_super = super_1_validate, | |
1499 | .sync_super = super_1_sync, | |
1500 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1501 | }, |
1502 | }; | |
1da177e4 LT |
1503 | |
1504 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1505 | { | |
7dd5e7c3 | 1506 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1507 | |
4b80991c N |
1508 | rcu_read_lock(); |
1509 | rdev_for_each_rcu(rdev, mddev1) | |
1510 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1511 | if (rdev->bdev->bd_contains == |
4b80991c N |
1512 | rdev2->bdev->bd_contains) { |
1513 | rcu_read_unlock(); | |
7dd5e7c3 | 1514 | return 1; |
4b80991c N |
1515 | } |
1516 | rcu_read_unlock(); | |
1da177e4 LT |
1517 | return 0; |
1518 | } | |
1519 | ||
1520 | static LIST_HEAD(pending_raid_disks); | |
1521 | ||
ac5e7113 AN |
1522 | /* |
1523 | * Try to register data integrity profile for an mddev | |
1524 | * | |
1525 | * This is called when an array is started and after a disk has been kicked | |
1526 | * from the array. It only succeeds if all working and active component devices | |
1527 | * are integrity capable with matching profiles. | |
1528 | */ | |
1529 | int md_integrity_register(mddev_t *mddev) | |
1530 | { | |
1531 | mdk_rdev_t *rdev, *reference = NULL; | |
1532 | ||
1533 | if (list_empty(&mddev->disks)) | |
1534 | return 0; /* nothing to do */ | |
1535 | if (blk_get_integrity(mddev->gendisk)) | |
1536 | return 0; /* already registered */ | |
1537 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
1538 | /* skip spares and non-functional disks */ | |
1539 | if (test_bit(Faulty, &rdev->flags)) | |
1540 | continue; | |
1541 | if (rdev->raid_disk < 0) | |
1542 | continue; | |
1543 | /* | |
1544 | * If at least one rdev is not integrity capable, we can not | |
1545 | * enable data integrity for the md device. | |
1546 | */ | |
1547 | if (!bdev_get_integrity(rdev->bdev)) | |
1548 | return -EINVAL; | |
1549 | if (!reference) { | |
1550 | /* Use the first rdev as the reference */ | |
1551 | reference = rdev; | |
1552 | continue; | |
1553 | } | |
1554 | /* does this rdev's profile match the reference profile? */ | |
1555 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1556 | rdev->bdev->bd_disk) < 0) | |
1557 | return -EINVAL; | |
1558 | } | |
1559 | /* | |
1560 | * All component devices are integrity capable and have matching | |
1561 | * profiles, register the common profile for the md device. | |
1562 | */ | |
1563 | if (blk_integrity_register(mddev->gendisk, | |
1564 | bdev_get_integrity(reference->bdev)) != 0) { | |
1565 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1566 | mdname(mddev)); | |
1567 | return -EINVAL; | |
1568 | } | |
1569 | printk(KERN_NOTICE "md: data integrity on %s enabled\n", | |
1570 | mdname(mddev)); | |
1571 | return 0; | |
1572 | } | |
1573 | EXPORT_SYMBOL(md_integrity_register); | |
1574 | ||
1575 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
1576 | void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) | |
3f9d99c1 | 1577 | { |
3f9d99c1 | 1578 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); |
ac5e7113 | 1579 | struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); |
3f9d99c1 | 1580 | |
ac5e7113 | 1581 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1582 | return; |
ac5e7113 | 1583 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1584 | return; |
ac5e7113 AN |
1585 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1586 | rdev->bdev->bd_disk) >= 0) | |
1587 | return; | |
1588 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1589 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1590 | } |
ac5e7113 | 1591 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1592 | |
1da177e4 LT |
1593 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1594 | { | |
7dd5e7c3 | 1595 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1596 | struct kobject *ko; |
1edf80d3 | 1597 | char *s; |
5e55e2f5 | 1598 | int err; |
1da177e4 LT |
1599 | |
1600 | if (rdev->mddev) { | |
1601 | MD_BUG(); | |
1602 | return -EINVAL; | |
1603 | } | |
11e2ede0 DW |
1604 | |
1605 | /* prevent duplicates */ | |
1606 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1607 | return -EEXIST; | |
1608 | ||
dd8ac336 AN |
1609 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1610 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1611 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1612 | if (mddev->pers) { |
1613 | /* Cannot change size, so fail | |
1614 | * If mddev->level <= 0, then we don't care | |
1615 | * about aligning sizes (e.g. linear) | |
1616 | */ | |
1617 | if (mddev->level > 0) | |
1618 | return -ENOSPC; | |
1619 | } else | |
dd8ac336 | 1620 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1621 | } |
1da177e4 LT |
1622 | |
1623 | /* Verify rdev->desc_nr is unique. | |
1624 | * If it is -1, assign a free number, else | |
1625 | * check number is not in use | |
1626 | */ | |
1627 | if (rdev->desc_nr < 0) { | |
1628 | int choice = 0; | |
1629 | if (mddev->pers) choice = mddev->raid_disks; | |
1630 | while (find_rdev_nr(mddev, choice)) | |
1631 | choice++; | |
1632 | rdev->desc_nr = choice; | |
1633 | } else { | |
1634 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1635 | return -EBUSY; | |
1636 | } | |
de01dfad N |
1637 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1638 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1639 | mdname(mddev), mddev->max_disks); | |
1640 | return -EBUSY; | |
1641 | } | |
19133a42 | 1642 | bdevname(rdev->bdev,b); |
649316b2 | 1643 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1644 | *s = '!'; |
649316b2 | 1645 | |
1da177e4 | 1646 | rdev->mddev = mddev; |
19133a42 | 1647 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1648 | |
b2d6db58 | 1649 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1650 | goto fail; |
86e6ffdd | 1651 | |
0762b8bd | 1652 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
5e55e2f5 N |
1653 | if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) { |
1654 | kobject_del(&rdev->kobj); | |
1655 | goto fail; | |
1656 | } | |
3c0ee63a N |
1657 | rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state"); |
1658 | ||
4b80991c | 1659 | list_add_rcu(&rdev->same_set, &mddev->disks); |
c5d79adb | 1660 | bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); |
4044ba58 N |
1661 | |
1662 | /* May as well allow recovery to be retried once */ | |
1663 | mddev->recovery_disabled = 0; | |
3f9d99c1 | 1664 | |
1da177e4 | 1665 | return 0; |
5e55e2f5 N |
1666 | |
1667 | fail: | |
1668 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1669 | b, mdname(mddev)); | |
1670 | return err; | |
1da177e4 LT |
1671 | } |
1672 | ||
177a99b2 | 1673 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1674 | { |
1675 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1676 | kobject_del(&rdev->kobj); | |
177a99b2 | 1677 | kobject_put(&rdev->kobj); |
5792a285 N |
1678 | } |
1679 | ||
1da177e4 LT |
1680 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1681 | { | |
1682 | char b[BDEVNAME_SIZE]; | |
1683 | if (!rdev->mddev) { | |
1684 | MD_BUG(); | |
1685 | return; | |
1686 | } | |
5463c790 | 1687 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1688 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1689 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1690 | rdev->mddev = NULL; | |
86e6ffdd | 1691 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1692 | sysfs_put(rdev->sysfs_state); |
1693 | rdev->sysfs_state = NULL; | |
5792a285 | 1694 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1695 | * writing to 'remove' to "dev/state". We also need |
1696 | * to delay it due to rcu usage. | |
5792a285 | 1697 | */ |
4b80991c | 1698 | synchronize_rcu(); |
177a99b2 N |
1699 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1700 | kobject_get(&rdev->kobj); | |
5792a285 | 1701 | schedule_work(&rdev->del_work); |
1da177e4 LT |
1702 | } |
1703 | ||
1704 | /* | |
1705 | * prevent the device from being mounted, repartitioned or | |
1706 | * otherwise reused by a RAID array (or any other kernel | |
1707 | * subsystem), by bd_claiming the device. | |
1708 | */ | |
c5d79adb | 1709 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1710 | { |
1711 | int err = 0; | |
1712 | struct block_device *bdev; | |
1713 | char b[BDEVNAME_SIZE]; | |
1714 | ||
2e7b651d | 1715 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1716 | if (IS_ERR(bdev)) { |
1717 | printk(KERN_ERR "md: could not open %s.\n", | |
1718 | __bdevname(dev, b)); | |
1719 | return PTR_ERR(bdev); | |
1720 | } | |
c5d79adb | 1721 | err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); |
1da177e4 LT |
1722 | if (err) { |
1723 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1724 | bdevname(bdev, b)); | |
9a1c3542 | 1725 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1726 | return err; |
1727 | } | |
c5d79adb N |
1728 | if (!shared) |
1729 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1730 | rdev->bdev = bdev; |
1731 | return err; | |
1732 | } | |
1733 | ||
1734 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1735 | { | |
1736 | struct block_device *bdev = rdev->bdev; | |
1737 | rdev->bdev = NULL; | |
1738 | if (!bdev) | |
1739 | MD_BUG(); | |
1740 | bd_release(bdev); | |
9a1c3542 | 1741 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1742 | } |
1743 | ||
1744 | void md_autodetect_dev(dev_t dev); | |
1745 | ||
1746 | static void export_rdev(mdk_rdev_t * rdev) | |
1747 | { | |
1748 | char b[BDEVNAME_SIZE]; | |
1749 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1750 | bdevname(rdev->bdev,b)); | |
1751 | if (rdev->mddev) | |
1752 | MD_BUG(); | |
1753 | free_disk_sb(rdev); | |
1da177e4 | 1754 | #ifndef MODULE |
d0fae18f N |
1755 | if (test_bit(AutoDetected, &rdev->flags)) |
1756 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1757 | #endif |
1758 | unlock_rdev(rdev); | |
86e6ffdd | 1759 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1760 | } |
1761 | ||
1762 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1763 | { | |
1764 | unbind_rdev_from_array(rdev); | |
1765 | export_rdev(rdev); | |
1766 | } | |
1767 | ||
1768 | static void export_array(mddev_t *mddev) | |
1769 | { | |
159ec1fc | 1770 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 1771 | |
d089c6af | 1772 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1773 | if (!rdev->mddev) { |
1774 | MD_BUG(); | |
1775 | continue; | |
1776 | } | |
1777 | kick_rdev_from_array(rdev); | |
1778 | } | |
1779 | if (!list_empty(&mddev->disks)) | |
1780 | MD_BUG(); | |
1781 | mddev->raid_disks = 0; | |
1782 | mddev->major_version = 0; | |
1783 | } | |
1784 | ||
1785 | static void print_desc(mdp_disk_t *desc) | |
1786 | { | |
1787 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1788 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1789 | } | |
1790 | ||
cd2ac932 | 1791 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
1792 | { |
1793 | int i; | |
1794 | ||
1795 | printk(KERN_INFO | |
1796 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1797 | sb->major_version, sb->minor_version, sb->patch_version, | |
1798 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1799 | sb->ctime); | |
1800 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1801 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1802 | sb->md_minor, sb->layout, sb->chunk_size); | |
1803 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1804 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1805 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1806 | sb->failed_disks, sb->spare_disks, | |
1807 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1808 | ||
1809 | printk(KERN_INFO); | |
1810 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1811 | mdp_disk_t *desc; | |
1812 | ||
1813 | desc = sb->disks + i; | |
1814 | if (desc->number || desc->major || desc->minor || | |
1815 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1816 | printk(" D %2d: ", i); | |
1817 | print_desc(desc); | |
1818 | } | |
1819 | } | |
1820 | printk(KERN_INFO "md: THIS: "); | |
1821 | print_desc(&sb->this_disk); | |
cd2ac932 | 1822 | } |
1da177e4 | 1823 | |
cd2ac932 CR |
1824 | static void print_sb_1(struct mdp_superblock_1 *sb) |
1825 | { | |
1826 | __u8 *uuid; | |
1827 | ||
1828 | uuid = sb->set_uuid; | |
ad361c98 JP |
1829 | printk(KERN_INFO |
1830 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x" | |
1831 | ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n" | |
1832 | "md: Name: \"%s\" CT:%llu\n", | |
cd2ac932 CR |
1833 | le32_to_cpu(sb->major_version), |
1834 | le32_to_cpu(sb->feature_map), | |
1835 | uuid[0], uuid[1], uuid[2], uuid[3], | |
1836 | uuid[4], uuid[5], uuid[6], uuid[7], | |
1837 | uuid[8], uuid[9], uuid[10], uuid[11], | |
1838 | uuid[12], uuid[13], uuid[14], uuid[15], | |
1839 | sb->set_name, | |
1840 | (unsigned long long)le64_to_cpu(sb->ctime) | |
1841 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
1842 | ||
1843 | uuid = sb->device_uuid; | |
ad361c98 JP |
1844 | printk(KERN_INFO |
1845 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 1846 | " RO:%llu\n" |
ad361c98 JP |
1847 | "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x" |
1848 | ":%02x%02x%02x%02x%02x%02x\n" | |
1849 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" | |
1850 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
1851 | le32_to_cpu(sb->level), |
1852 | (unsigned long long)le64_to_cpu(sb->size), | |
1853 | le32_to_cpu(sb->raid_disks), | |
1854 | le32_to_cpu(sb->layout), | |
1855 | le32_to_cpu(sb->chunksize), | |
1856 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
1857 | (unsigned long long)le64_to_cpu(sb->data_size), | |
1858 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
1859 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
1860 | le32_to_cpu(sb->dev_number), | |
1861 | uuid[0], uuid[1], uuid[2], uuid[3], | |
1862 | uuid[4], uuid[5], uuid[6], uuid[7], | |
1863 | uuid[8], uuid[9], uuid[10], uuid[11], | |
1864 | uuid[12], uuid[13], uuid[14], uuid[15], | |
1865 | sb->devflags, | |
1866 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
1867 | (unsigned long long)le64_to_cpu(sb->events), | |
1868 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
1869 | le32_to_cpu(sb->sb_csum), | |
1870 | le32_to_cpu(sb->max_dev) | |
1871 | ); | |
1da177e4 LT |
1872 | } |
1873 | ||
cd2ac932 | 1874 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
1875 | { |
1876 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
1877 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
1878 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
1879 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
1880 | rdev->desc_nr); | |
1da177e4 | 1881 | if (rdev->sb_loaded) { |
cd2ac932 CR |
1882 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
1883 | switch (major_version) { | |
1884 | case 0: | |
1885 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
1886 | break; | |
1887 | case 1: | |
1888 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
1889 | break; | |
1890 | } | |
1da177e4 LT |
1891 | } else |
1892 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
1893 | } | |
1894 | ||
5e56341d | 1895 | static void md_print_devices(void) |
1da177e4 | 1896 | { |
159ec1fc | 1897 | struct list_head *tmp; |
1da177e4 LT |
1898 | mdk_rdev_t *rdev; |
1899 | mddev_t *mddev; | |
1900 | char b[BDEVNAME_SIZE]; | |
1901 | ||
1902 | printk("\n"); | |
1903 | printk("md: **********************************\n"); | |
1904 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
1905 | printk("md: **********************************\n"); | |
29ac4aa3 | 1906 | for_each_mddev(mddev, tmp) { |
1da177e4 | 1907 | |
32a7627c N |
1908 | if (mddev->bitmap) |
1909 | bitmap_print_sb(mddev->bitmap); | |
1910 | else | |
1911 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 1912 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
1913 | printk("<%s>", bdevname(rdev->bdev,b)); |
1914 | printk("\n"); | |
1915 | ||
159ec1fc | 1916 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 1917 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
1918 | } |
1919 | printk("md: **********************************\n"); | |
1920 | printk("\n"); | |
1921 | } | |
1922 | ||
1923 | ||
42543769 | 1924 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 1925 | { |
42543769 N |
1926 | /* Update each superblock (in-memory image), but |
1927 | * if we are allowed to, skip spares which already | |
1928 | * have the right event counter, or have one earlier | |
1929 | * (which would mean they aren't being marked as dirty | |
1930 | * with the rest of the array) | |
1931 | */ | |
1da177e4 | 1932 | mdk_rdev_t *rdev; |
1da177e4 | 1933 | |
5e865106 N |
1934 | /* First make sure individual recovery_offsets are correct */ |
1935 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
1936 | if (rdev->raid_disk >= 0 && | |
1937 | !test_bit(In_sync, &rdev->flags) && | |
1938 | mddev->curr_resync_completed > rdev->recovery_offset) | |
1939 | rdev->recovery_offset = mddev->curr_resync_completed; | |
1940 | ||
1941 | } | |
159ec1fc | 1942 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
1943 | if (rdev->sb_events == mddev->events || |
1944 | (nospares && | |
1945 | rdev->raid_disk < 0 && | |
1946 | (rdev->sb_events&1)==0 && | |
1947 | rdev->sb_events+1 == mddev->events)) { | |
1948 | /* Don't update this superblock */ | |
1949 | rdev->sb_loaded = 2; | |
1950 | } else { | |
1951 | super_types[mddev->major_version]. | |
1952 | sync_super(mddev, rdev); | |
1953 | rdev->sb_loaded = 1; | |
1954 | } | |
1da177e4 LT |
1955 | } |
1956 | } | |
1957 | ||
850b2b42 | 1958 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 1959 | { |
1da177e4 | 1960 | mdk_rdev_t *rdev; |
06d91a5f | 1961 | int sync_req; |
42543769 | 1962 | int nospares = 0; |
1da177e4 | 1963 | |
1b57f132 | 1964 | mddev->utime = get_seconds(); |
8377bc80 N |
1965 | if (mddev->external) |
1966 | return; | |
1da177e4 | 1967 | repeat: |
a9701a30 | 1968 | spin_lock_irq(&mddev->write_lock); |
84692195 | 1969 | |
850b2b42 N |
1970 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
1971 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) | |
1972 | force_change = 1; | |
1973 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
1974 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
1975 | * though if events isn't the right even/odd, we will have to do | |
1976 | * spares after all | |
1977 | */ | |
1978 | nospares = 1; | |
1979 | if (force_change) | |
1980 | nospares = 0; | |
1981 | if (mddev->degraded) | |
84692195 N |
1982 | /* If the array is degraded, then skipping spares is both |
1983 | * dangerous and fairly pointless. | |
1984 | * Dangerous because a device that was removed from the array | |
1985 | * might have a event_count that still looks up-to-date, | |
1986 | * so it can be re-added without a resync. | |
1987 | * Pointless because if there are any spares to skip, | |
1988 | * then a recovery will happen and soon that array won't | |
1989 | * be degraded any more and the spare can go back to sleep then. | |
1990 | */ | |
850b2b42 | 1991 | nospares = 0; |
84692195 | 1992 | |
06d91a5f | 1993 | sync_req = mddev->in_sync; |
42543769 N |
1994 | |
1995 | /* If this is just a dirty<->clean transition, and the array is clean | |
1996 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 1997 | if (nospares |
42543769 | 1998 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
1031be7a N |
1999 | && (mddev->events & 1) |
2000 | && mddev->events != 1) | |
42543769 N |
2001 | mddev->events--; |
2002 | else { | |
2003 | /* otherwise we have to go forward and ... */ | |
2004 | mddev->events ++; | |
2005 | if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ | |
51d5668c N |
2006 | /* .. if the array isn't clean, an 'even' event must also go |
2007 | * to spares. */ | |
2008 | if ((mddev->events&1)==0) | |
42543769 | 2009 | nospares = 0; |
42543769 | 2010 | } else { |
51d5668c N |
2011 | /* otherwise an 'odd' event must go to spares */ |
2012 | if ((mddev->events&1)) | |
42543769 | 2013 | nospares = 0; |
42543769 N |
2014 | } |
2015 | } | |
1da177e4 LT |
2016 | |
2017 | if (!mddev->events) { | |
2018 | /* | |
2019 | * oops, this 64-bit counter should never wrap. | |
2020 | * Either we are in around ~1 trillion A.C., assuming | |
2021 | * 1 reboot per second, or we have a bug: | |
2022 | */ | |
2023 | MD_BUG(); | |
2024 | mddev->events --; | |
2025 | } | |
1da177e4 LT |
2026 | |
2027 | /* | |
2028 | * do not write anything to disk if using | |
2029 | * nonpersistent superblocks | |
2030 | */ | |
06d91a5f | 2031 | if (!mddev->persistent) { |
e691063a N |
2032 | if (!mddev->external) |
2033 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
2034 | ||
a9701a30 | 2035 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2036 | wake_up(&mddev->sb_wait); |
1da177e4 | 2037 | return; |
06d91a5f | 2038 | } |
e691063a | 2039 | sync_sbs(mddev, nospares); |
a9701a30 | 2040 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2041 | |
2042 | dprintk(KERN_INFO | |
2043 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2044 | mdname(mddev),mddev->in_sync); | |
2045 | ||
4ad13663 | 2046 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2047 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2048 | char b[BDEVNAME_SIZE]; |
2049 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2050 | if (rdev->sb_loaded != 1) |
2051 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2052 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2053 | dprintk("(skipping faulty "); |
2054 | ||
2055 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2056 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2057 | md_super_write(mddev,rdev, |
0f420358 | 2058 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2059 | rdev->sb_page); |
2060 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2061 | bdevname(rdev->bdev,b), | |
0f420358 | 2062 | (unsigned long long)rdev->sb_start); |
42543769 | 2063 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2064 | |
1da177e4 LT |
2065 | } else |
2066 | dprintk(")\n"); | |
7bfa19f2 | 2067 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2068 | /* only need to write one superblock... */ |
2069 | break; | |
2070 | } | |
a9701a30 | 2071 | md_super_wait(mddev); |
850b2b42 | 2072 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2073 | |
a9701a30 | 2074 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2075 | if (mddev->in_sync != sync_req || |
2076 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2077 | /* have to write it out again */ |
a9701a30 | 2078 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2079 | goto repeat; |
2080 | } | |
850b2b42 | 2081 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2082 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2083 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2084 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2085 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2086 | |
1da177e4 LT |
2087 | } |
2088 | ||
7f6ce769 | 2089 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2090 | * We want to accept with case. For this we use cmd_match. |
2091 | */ | |
2092 | static int cmd_match(const char *cmd, const char *str) | |
2093 | { | |
2094 | /* See if cmd, written into a sysfs file, matches | |
2095 | * str. They must either be the same, or cmd can | |
2096 | * have a trailing newline | |
2097 | */ | |
2098 | while (*cmd && *str && *cmd == *str) { | |
2099 | cmd++; | |
2100 | str++; | |
2101 | } | |
2102 | if (*cmd == '\n') | |
2103 | cmd++; | |
2104 | if (*str || *cmd) | |
2105 | return 0; | |
2106 | return 1; | |
2107 | } | |
2108 | ||
86e6ffdd N |
2109 | struct rdev_sysfs_entry { |
2110 | struct attribute attr; | |
2111 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2112 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2113 | }; | |
2114 | ||
2115 | static ssize_t | |
96de1e66 | 2116 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2117 | { |
2118 | char *sep = ""; | |
20a49ff6 | 2119 | size_t len = 0; |
86e6ffdd | 2120 | |
b2d444d7 | 2121 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2122 | len+= sprintf(page+len, "%sfaulty",sep); |
2123 | sep = ","; | |
2124 | } | |
b2d444d7 | 2125 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2126 | len += sprintf(page+len, "%sin_sync",sep); |
2127 | sep = ","; | |
2128 | } | |
f655675b N |
2129 | if (test_bit(WriteMostly, &rdev->flags)) { |
2130 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2131 | sep = ","; | |
2132 | } | |
6bfe0b49 DW |
2133 | if (test_bit(Blocked, &rdev->flags)) { |
2134 | len += sprintf(page+len, "%sblocked", sep); | |
2135 | sep = ","; | |
2136 | } | |
b2d444d7 N |
2137 | if (!test_bit(Faulty, &rdev->flags) && |
2138 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2139 | len += sprintf(page+len, "%sspare", sep); |
2140 | sep = ","; | |
2141 | } | |
2142 | return len+sprintf(page+len, "\n"); | |
2143 | } | |
2144 | ||
45dc2de1 N |
2145 | static ssize_t |
2146 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2147 | { | |
2148 | /* can write | |
2149 | * faulty - simulates and error | |
2150 | * remove - disconnects the device | |
f655675b N |
2151 | * writemostly - sets write_mostly |
2152 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2153 | * blocked - sets the Blocked flag |
2154 | * -blocked - clears the Blocked flag | |
6d56e278 | 2155 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2156 | */ |
2157 | int err = -EINVAL; | |
2158 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2159 | md_error(rdev->mddev, rdev); | |
2160 | err = 0; | |
2161 | } else if (cmd_match(buf, "remove")) { | |
2162 | if (rdev->raid_disk >= 0) | |
2163 | err = -EBUSY; | |
2164 | else { | |
2165 | mddev_t *mddev = rdev->mddev; | |
2166 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2167 | if (mddev->pers) |
2168 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2169 | md_new_event(mddev); |
2170 | err = 0; | |
2171 | } | |
f655675b N |
2172 | } else if (cmd_match(buf, "writemostly")) { |
2173 | set_bit(WriteMostly, &rdev->flags); | |
2174 | err = 0; | |
2175 | } else if (cmd_match(buf, "-writemostly")) { | |
2176 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2177 | err = 0; |
2178 | } else if (cmd_match(buf, "blocked")) { | |
2179 | set_bit(Blocked, &rdev->flags); | |
2180 | err = 0; | |
2181 | } else if (cmd_match(buf, "-blocked")) { | |
2182 | clear_bit(Blocked, &rdev->flags); | |
2183 | wake_up(&rdev->blocked_wait); | |
2184 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2185 | md_wakeup_thread(rdev->mddev->thread); | |
2186 | ||
6d56e278 N |
2187 | err = 0; |
2188 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2189 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2190 | err = 0; |
45dc2de1 | 2191 | } |
3c0ee63a N |
2192 | if (!err && rdev->sysfs_state) |
2193 | sysfs_notify_dirent(rdev->sysfs_state); | |
45dc2de1 N |
2194 | return err ? err : len; |
2195 | } | |
80ca3a44 N |
2196 | static struct rdev_sysfs_entry rdev_state = |
2197 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2198 | |
4dbcdc75 N |
2199 | static ssize_t |
2200 | errors_show(mdk_rdev_t *rdev, char *page) | |
2201 | { | |
2202 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2203 | } | |
2204 | ||
2205 | static ssize_t | |
2206 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2207 | { | |
2208 | char *e; | |
2209 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2210 | if (*buf && (*e == 0 || *e == '\n')) { | |
2211 | atomic_set(&rdev->corrected_errors, n); | |
2212 | return len; | |
2213 | } | |
2214 | return -EINVAL; | |
2215 | } | |
2216 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2217 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2218 | |
014236d2 N |
2219 | static ssize_t |
2220 | slot_show(mdk_rdev_t *rdev, char *page) | |
2221 | { | |
2222 | if (rdev->raid_disk < 0) | |
2223 | return sprintf(page, "none\n"); | |
2224 | else | |
2225 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2226 | } | |
2227 | ||
2228 | static ssize_t | |
2229 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2230 | { | |
2231 | char *e; | |
c303da6d N |
2232 | int err; |
2233 | char nm[20]; | |
014236d2 N |
2234 | int slot = simple_strtoul(buf, &e, 10); |
2235 | if (strncmp(buf, "none", 4)==0) | |
2236 | slot = -1; | |
2237 | else if (e==buf || (*e && *e!= '\n')) | |
2238 | return -EINVAL; | |
6c2fce2e | 2239 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2240 | /* Setting 'slot' on an active array requires also |
2241 | * updating the 'rd%d' link, and communicating | |
2242 | * with the personality with ->hot_*_disk. | |
2243 | * For now we only support removing | |
2244 | * failed/spare devices. This normally happens automatically, | |
2245 | * but not when the metadata is externally managed. | |
2246 | */ | |
c303da6d N |
2247 | if (rdev->raid_disk == -1) |
2248 | return -EEXIST; | |
2249 | /* personality does all needed checks */ | |
2250 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2251 | return -EINVAL; | |
2252 | err = rdev->mddev->pers-> | |
2253 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2254 | if (err) | |
2255 | return err; | |
2256 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2257 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
2258 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2259 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2260 | } else if (rdev->mddev->pers) { |
2261 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2262 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2263 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2264 | */ |
2265 | ||
2266 | if (rdev->raid_disk != -1) | |
2267 | return -EBUSY; | |
2268 | ||
2269 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2270 | return -EINVAL; | |
2271 | ||
159ec1fc | 2272 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2273 | if (rdev2->raid_disk == slot) |
2274 | return -EEXIST; | |
2275 | ||
2276 | rdev->raid_disk = slot; | |
2277 | if (test_bit(In_sync, &rdev->flags)) | |
2278 | rdev->saved_raid_disk = slot; | |
2279 | else | |
2280 | rdev->saved_raid_disk = -1; | |
2281 | err = rdev->mddev->pers-> | |
2282 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2283 | if (err) { |
6c2fce2e | 2284 | rdev->raid_disk = -1; |
6c2fce2e | 2285 | return err; |
52664732 | 2286 | } else |
3c0ee63a | 2287 | sysfs_notify_dirent(rdev->sysfs_state); |
6c2fce2e NB |
2288 | sprintf(nm, "rd%d", rdev->raid_disk); |
2289 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
2290 | printk(KERN_WARNING | |
2291 | "md: cannot register " | |
2292 | "%s for %s\n", | |
2293 | nm, mdname(rdev->mddev)); | |
2294 | ||
2295 | /* don't wakeup anyone, leave that to userspace. */ | |
c303da6d N |
2296 | } else { |
2297 | if (slot >= rdev->mddev->raid_disks) | |
2298 | return -ENOSPC; | |
2299 | rdev->raid_disk = slot; | |
2300 | /* assume it is working */ | |
c5d79adb N |
2301 | clear_bit(Faulty, &rdev->flags); |
2302 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2303 | set_bit(In_sync, &rdev->flags); |
3c0ee63a | 2304 | sysfs_notify_dirent(rdev->sysfs_state); |
c303da6d | 2305 | } |
014236d2 N |
2306 | return len; |
2307 | } | |
2308 | ||
2309 | ||
2310 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2311 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2312 | |
93c8cad0 N |
2313 | static ssize_t |
2314 | offset_show(mdk_rdev_t *rdev, char *page) | |
2315 | { | |
6961ece4 | 2316 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2317 | } |
2318 | ||
2319 | static ssize_t | |
2320 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2321 | { | |
2322 | char *e; | |
2323 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2324 | if (e==buf || (*e && *e != '\n')) | |
2325 | return -EINVAL; | |
8ed0a521 | 2326 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2327 | return -EBUSY; |
dd8ac336 | 2328 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2329 | /* Must set offset before size, so overlap checks |
2330 | * can be sane */ | |
2331 | return -EBUSY; | |
93c8cad0 N |
2332 | rdev->data_offset = offset; |
2333 | return len; | |
2334 | } | |
2335 | ||
2336 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2337 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2338 | |
83303b61 N |
2339 | static ssize_t |
2340 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2341 | { | |
dd8ac336 | 2342 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2343 | } |
2344 | ||
c5d79adb N |
2345 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2346 | { | |
2347 | /* check if two start/length pairs overlap */ | |
2348 | if (s1+l1 <= s2) | |
2349 | return 0; | |
2350 | if (s2+l2 <= s1) | |
2351 | return 0; | |
2352 | return 1; | |
2353 | } | |
2354 | ||
b522adcd DW |
2355 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2356 | { | |
2357 | unsigned long long blocks; | |
2358 | sector_t new; | |
2359 | ||
2360 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2361 | return -EINVAL; | |
2362 | ||
2363 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2364 | return -EINVAL; /* sector conversion overflow */ | |
2365 | ||
2366 | new = blocks * 2; | |
2367 | if (new != blocks * 2) | |
2368 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2369 | ||
2370 | *sectors = new; | |
2371 | return 0; | |
2372 | } | |
2373 | ||
83303b61 N |
2374 | static ssize_t |
2375 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2376 | { | |
27c529bb | 2377 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2378 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2379 | sector_t sectors; |
27c529bb | 2380 | |
b522adcd | 2381 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2382 | return -EINVAL; |
0cd17fec | 2383 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2384 | if (my_mddev->persistent) { |
dd8ac336 AN |
2385 | sectors = super_types[my_mddev->major_version]. |
2386 | rdev_size_change(rdev, sectors); | |
2387 | if (!sectors) | |
0cd17fec | 2388 | return -EBUSY; |
dd8ac336 AN |
2389 | } else if (!sectors) |
2390 | sectors = (rdev->bdev->bd_inode->i_size >> 9) - | |
2391 | rdev->data_offset; | |
0cd17fec | 2392 | } |
dd8ac336 | 2393 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2394 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2395 | |
dd8ac336 AN |
2396 | rdev->sectors = sectors; |
2397 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2398 | /* need to check that all other rdevs with the same ->bdev |
2399 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2400 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2401 | * we have to change it back, we will have the lock again. |
2402 | */ | |
2403 | mddev_t *mddev; | |
2404 | int overlap = 0; | |
159ec1fc | 2405 | struct list_head *tmp; |
c5d79adb | 2406 | |
27c529bb | 2407 | mddev_unlock(my_mddev); |
29ac4aa3 | 2408 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2409 | mdk_rdev_t *rdev2; |
2410 | ||
2411 | mddev_lock(mddev); | |
159ec1fc | 2412 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
c5d79adb N |
2413 | if (test_bit(AllReserved, &rdev2->flags) || |
2414 | (rdev->bdev == rdev2->bdev && | |
2415 | rdev != rdev2 && | |
dd8ac336 | 2416 | overlaps(rdev->data_offset, rdev->sectors, |
d07bd3bc | 2417 | rdev2->data_offset, |
dd8ac336 | 2418 | rdev2->sectors))) { |
c5d79adb N |
2419 | overlap = 1; |
2420 | break; | |
2421 | } | |
2422 | mddev_unlock(mddev); | |
2423 | if (overlap) { | |
2424 | mddev_put(mddev); | |
2425 | break; | |
2426 | } | |
2427 | } | |
27c529bb | 2428 | mddev_lock(my_mddev); |
c5d79adb N |
2429 | if (overlap) { |
2430 | /* Someone else could have slipped in a size | |
2431 | * change here, but doing so is just silly. | |
dd8ac336 | 2432 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2433 | * safe, and trust userspace not to race with |
2434 | * itself | |
2435 | */ | |
dd8ac336 | 2436 | rdev->sectors = oldsectors; |
c5d79adb N |
2437 | return -EBUSY; |
2438 | } | |
2439 | } | |
83303b61 N |
2440 | return len; |
2441 | } | |
2442 | ||
2443 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2444 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2445 | |
86e6ffdd N |
2446 | static struct attribute *rdev_default_attrs[] = { |
2447 | &rdev_state.attr, | |
4dbcdc75 | 2448 | &rdev_errors.attr, |
014236d2 | 2449 | &rdev_slot.attr, |
93c8cad0 | 2450 | &rdev_offset.attr, |
83303b61 | 2451 | &rdev_size.attr, |
86e6ffdd N |
2452 | NULL, |
2453 | }; | |
2454 | static ssize_t | |
2455 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2456 | { | |
2457 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2458 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2459 | mddev_t *mddev = rdev->mddev; |
2460 | ssize_t rv; | |
86e6ffdd N |
2461 | |
2462 | if (!entry->show) | |
2463 | return -EIO; | |
27c529bb N |
2464 | |
2465 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2466 | if (!rv) { | |
2467 | if (rdev->mddev == NULL) | |
2468 | rv = -EBUSY; | |
2469 | else | |
2470 | rv = entry->show(rdev, page); | |
2471 | mddev_unlock(mddev); | |
2472 | } | |
2473 | return rv; | |
86e6ffdd N |
2474 | } |
2475 | ||
2476 | static ssize_t | |
2477 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2478 | const char *page, size_t length) | |
2479 | { | |
2480 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2481 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2482 | ssize_t rv; |
2483 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2484 | |
2485 | if (!entry->store) | |
2486 | return -EIO; | |
67463acb N |
2487 | if (!capable(CAP_SYS_ADMIN)) |
2488 | return -EACCES; | |
27c529bb | 2489 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2490 | if (!rv) { |
27c529bb N |
2491 | if (rdev->mddev == NULL) |
2492 | rv = -EBUSY; | |
2493 | else | |
2494 | rv = entry->store(rdev, page, length); | |
6a51830e | 2495 | mddev_unlock(mddev); |
ca388059 N |
2496 | } |
2497 | return rv; | |
86e6ffdd N |
2498 | } |
2499 | ||
2500 | static void rdev_free(struct kobject *ko) | |
2501 | { | |
2502 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2503 | kfree(rdev); | |
2504 | } | |
2505 | static struct sysfs_ops rdev_sysfs_ops = { | |
2506 | .show = rdev_attr_show, | |
2507 | .store = rdev_attr_store, | |
2508 | }; | |
2509 | static struct kobj_type rdev_ktype = { | |
2510 | .release = rdev_free, | |
2511 | .sysfs_ops = &rdev_sysfs_ops, | |
2512 | .default_attrs = rdev_default_attrs, | |
2513 | }; | |
2514 | ||
1da177e4 LT |
2515 | /* |
2516 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2517 | * | |
2518 | * mark the device faulty if: | |
2519 | * | |
2520 | * - the device is nonexistent (zero size) | |
2521 | * - the device has no valid superblock | |
2522 | * | |
2523 | * a faulty rdev _never_ has rdev->sb set. | |
2524 | */ | |
2525 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2526 | { | |
2527 | char b[BDEVNAME_SIZE]; | |
2528 | int err; | |
2529 | mdk_rdev_t *rdev; | |
2530 | sector_t size; | |
2531 | ||
9ffae0cf | 2532 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2533 | if (!rdev) { |
2534 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2535 | return ERR_PTR(-ENOMEM); | |
2536 | } | |
1da177e4 LT |
2537 | |
2538 | if ((err = alloc_disk_sb(rdev))) | |
2539 | goto abort_free; | |
2540 | ||
c5d79adb | 2541 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2542 | if (err) |
2543 | goto abort_free; | |
2544 | ||
f9cb074b | 2545 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2546 | |
1da177e4 | 2547 | rdev->desc_nr = -1; |
2b6e8459 | 2548 | rdev->saved_raid_disk = -1; |
3f9d7b0d | 2549 | rdev->raid_disk = -1; |
b2d444d7 | 2550 | rdev->flags = 0; |
1da177e4 | 2551 | rdev->data_offset = 0; |
42543769 | 2552 | rdev->sb_events = 0; |
1da177e4 | 2553 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 2554 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 2555 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
2556 | |
2557 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
2558 | if (!size) { | |
2559 | printk(KERN_WARNING | |
2560 | "md: %s has zero or unknown size, marking faulty!\n", | |
2561 | bdevname(rdev->bdev,b)); | |
2562 | err = -EINVAL; | |
2563 | goto abort_free; | |
2564 | } | |
2565 | ||
2566 | if (super_format >= 0) { | |
2567 | err = super_types[super_format]. | |
2568 | load_super(rdev, NULL, super_minor); | |
2569 | if (err == -EINVAL) { | |
df968c4e N |
2570 | printk(KERN_WARNING |
2571 | "md: %s does not have a valid v%d.%d " | |
2572 | "superblock, not importing!\n", | |
2573 | bdevname(rdev->bdev,b), | |
2574 | super_format, super_minor); | |
1da177e4 LT |
2575 | goto abort_free; |
2576 | } | |
2577 | if (err < 0) { | |
2578 | printk(KERN_WARNING | |
2579 | "md: could not read %s's sb, not importing!\n", | |
2580 | bdevname(rdev->bdev,b)); | |
2581 | goto abort_free; | |
2582 | } | |
2583 | } | |
6bfe0b49 | 2584 | |
1da177e4 | 2585 | INIT_LIST_HEAD(&rdev->same_set); |
6bfe0b49 | 2586 | init_waitqueue_head(&rdev->blocked_wait); |
1da177e4 LT |
2587 | |
2588 | return rdev; | |
2589 | ||
2590 | abort_free: | |
2591 | if (rdev->sb_page) { | |
2592 | if (rdev->bdev) | |
2593 | unlock_rdev(rdev); | |
2594 | free_disk_sb(rdev); | |
2595 | } | |
2596 | kfree(rdev); | |
2597 | return ERR_PTR(err); | |
2598 | } | |
2599 | ||
2600 | /* | |
2601 | * Check a full RAID array for plausibility | |
2602 | */ | |
2603 | ||
2604 | ||
a757e64c | 2605 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2606 | { |
2607 | int i; | |
159ec1fc | 2608 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2609 | char b[BDEVNAME_SIZE]; |
2610 | ||
2611 | freshest = NULL; | |
d089c6af | 2612 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2613 | switch (super_types[mddev->major_version]. |
2614 | load_super(rdev, freshest, mddev->minor_version)) { | |
2615 | case 1: | |
2616 | freshest = rdev; | |
2617 | break; | |
2618 | case 0: | |
2619 | break; | |
2620 | default: | |
2621 | printk( KERN_ERR \ | |
2622 | "md: fatal superblock inconsistency in %s" | |
2623 | " -- removing from array\n", | |
2624 | bdevname(rdev->bdev,b)); | |
2625 | kick_rdev_from_array(rdev); | |
2626 | } | |
2627 | ||
2628 | ||
2629 | super_types[mddev->major_version]. | |
2630 | validate_super(mddev, freshest); | |
2631 | ||
2632 | i = 0; | |
d089c6af | 2633 | rdev_for_each(rdev, tmp, mddev) { |
de01dfad N |
2634 | if (rdev->desc_nr >= mddev->max_disks || |
2635 | i > mddev->max_disks) { | |
2636 | printk(KERN_WARNING | |
2637 | "md: %s: %s: only %d devices permitted\n", | |
2638 | mdname(mddev), bdevname(rdev->bdev, b), | |
2639 | mddev->max_disks); | |
2640 | kick_rdev_from_array(rdev); | |
2641 | continue; | |
2642 | } | |
1da177e4 LT |
2643 | if (rdev != freshest) |
2644 | if (super_types[mddev->major_version]. | |
2645 | validate_super(mddev, rdev)) { | |
2646 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2647 | " from array!\n", | |
2648 | bdevname(rdev->bdev,b)); | |
2649 | kick_rdev_from_array(rdev); | |
2650 | continue; | |
2651 | } | |
2652 | if (mddev->level == LEVEL_MULTIPATH) { | |
2653 | rdev->desc_nr = i++; | |
2654 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2655 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 2656 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
2657 | rdev->raid_disk = -1; |
2658 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2659 | } |
2660 | } | |
1da177e4 LT |
2661 | } |
2662 | ||
19052c0e N |
2663 | static void md_safemode_timeout(unsigned long data); |
2664 | ||
16f17b39 N |
2665 | static ssize_t |
2666 | safe_delay_show(mddev_t *mddev, char *page) | |
2667 | { | |
2668 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2669 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2670 | } | |
2671 | static ssize_t | |
2672 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2673 | { | |
2674 | int scale=1; | |
2675 | int dot=0; | |
2676 | int i; | |
2677 | unsigned long msec; | |
2678 | char buf[30]; | |
97ce0a7f | 2679 | |
16f17b39 N |
2680 | /* remove a period, and count digits after it */ |
2681 | if (len >= sizeof(buf)) | |
2682 | return -EINVAL; | |
97ce0a7f | 2683 | strlcpy(buf, cbuf, sizeof(buf)); |
16f17b39 N |
2684 | for (i=0; i<len; i++) { |
2685 | if (dot) { | |
2686 | if (isdigit(buf[i])) { | |
2687 | buf[i-1] = buf[i]; | |
2688 | scale *= 10; | |
2689 | } | |
2690 | buf[i] = 0; | |
2691 | } else if (buf[i] == '.') { | |
2692 | dot=1; | |
2693 | buf[i] = 0; | |
2694 | } | |
2695 | } | |
97ce0a7f | 2696 | if (strict_strtoul(buf, 10, &msec) < 0) |
16f17b39 N |
2697 | return -EINVAL; |
2698 | msec = (msec * 1000) / scale; | |
2699 | if (msec == 0) | |
2700 | mddev->safemode_delay = 0; | |
2701 | else { | |
19052c0e | 2702 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2703 | mddev->safemode_delay = (msec*HZ)/1000; |
2704 | if (mddev->safemode_delay == 0) | |
2705 | mddev->safemode_delay = 1; | |
19052c0e N |
2706 | if (mddev->safemode_delay < old_delay) |
2707 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2708 | } |
2709 | return len; | |
2710 | } | |
2711 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2712 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2713 | |
eae1701f | 2714 | static ssize_t |
96de1e66 | 2715 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2716 | { |
2604b703 | 2717 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2718 | if (p) |
eae1701f | 2719 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2720 | else if (mddev->clevel[0]) |
2721 | return sprintf(page, "%s\n", mddev->clevel); | |
2722 | else if (mddev->level != LEVEL_NONE) | |
2723 | return sprintf(page, "%d\n", mddev->level); | |
2724 | else | |
2725 | return 0; | |
eae1701f N |
2726 | } |
2727 | ||
d9d166c2 N |
2728 | static ssize_t |
2729 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2730 | { | |
245f46c2 | 2731 | char level[16]; |
20a49ff6 | 2732 | ssize_t rv = len; |
245f46c2 N |
2733 | struct mdk_personality *pers; |
2734 | void *priv; | |
3a981b03 | 2735 | mdk_rdev_t *rdev; |
245f46c2 N |
2736 | |
2737 | if (mddev->pers == NULL) { | |
2738 | if (len == 0) | |
2739 | return 0; | |
2740 | if (len >= sizeof(mddev->clevel)) | |
2741 | return -ENOSPC; | |
2742 | strncpy(mddev->clevel, buf, len); | |
2743 | if (mddev->clevel[len-1] == '\n') | |
2744 | len--; | |
2745 | mddev->clevel[len] = 0; | |
2746 | mddev->level = LEVEL_NONE; | |
2747 | return rv; | |
2748 | } | |
2749 | ||
2750 | /* request to change the personality. Need to ensure: | |
2751 | * - array is not engaged in resync/recovery/reshape | |
2752 | * - old personality can be suspended | |
2753 | * - new personality will access other array. | |
2754 | */ | |
2755 | ||
2756 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) | |
d9d166c2 | 2757 | return -EBUSY; |
245f46c2 N |
2758 | |
2759 | if (!mddev->pers->quiesce) { | |
2760 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
2761 | mdname(mddev), mddev->pers->name); | |
2762 | return -EINVAL; | |
2763 | } | |
2764 | ||
2765 | /* Now find the new personality */ | |
2766 | if (len == 0 || len >= sizeof(level)) | |
2767 | return -EINVAL; | |
2768 | strncpy(level, buf, len); | |
2769 | if (level[len-1] == '\n') | |
d9d166c2 | 2770 | len--; |
245f46c2 N |
2771 | level[len] = 0; |
2772 | ||
2773 | request_module("md-%s", level); | |
2774 | spin_lock(&pers_lock); | |
2775 | pers = find_pers(LEVEL_NONE, level); | |
2776 | if (!pers || !try_module_get(pers->owner)) { | |
2777 | spin_unlock(&pers_lock); | |
2778 | printk(KERN_WARNING "md: personality %s not loaded\n", level); | |
2779 | return -EINVAL; | |
2780 | } | |
2781 | spin_unlock(&pers_lock); | |
2782 | ||
2783 | if (pers == mddev->pers) { | |
2784 | /* Nothing to do! */ | |
2785 | module_put(pers->owner); | |
2786 | return rv; | |
2787 | } | |
2788 | if (!pers->takeover) { | |
2789 | module_put(pers->owner); | |
2790 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
2791 | mdname(mddev), level); | |
2792 | return -EINVAL; | |
2793 | } | |
2794 | ||
2795 | /* ->takeover must set new_* and/or delta_disks | |
2796 | * if it succeeds, and may set them when it fails. | |
2797 | */ | |
2798 | priv = pers->takeover(mddev); | |
2799 | if (IS_ERR(priv)) { | |
2800 | mddev->new_level = mddev->level; | |
2801 | mddev->new_layout = mddev->layout; | |
664e7c41 | 2802 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
2803 | mddev->raid_disks -= mddev->delta_disks; |
2804 | mddev->delta_disks = 0; | |
2805 | module_put(pers->owner); | |
2806 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
2807 | mdname(mddev), level); | |
2808 | return PTR_ERR(priv); | |
2809 | } | |
2810 | ||
2811 | /* Looks like we have a winner */ | |
2812 | mddev_suspend(mddev); | |
2813 | mddev->pers->stop(mddev); | |
2814 | module_put(mddev->pers->owner); | |
3a981b03 N |
2815 | /* Invalidate devices that are now superfluous */ |
2816 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
2817 | if (rdev->raid_disk >= mddev->raid_disks) { | |
2818 | rdev->raid_disk = -1; | |
2819 | clear_bit(In_sync, &rdev->flags); | |
2820 | } | |
245f46c2 N |
2821 | mddev->pers = pers; |
2822 | mddev->private = priv; | |
2823 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
2824 | mddev->level = mddev->new_level; | |
2825 | mddev->layout = mddev->new_layout; | |
664e7c41 | 2826 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 N |
2827 | mddev->delta_disks = 0; |
2828 | pers->run(mddev); | |
2829 | mddev_resume(mddev); | |
2830 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2831 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2832 | md_wakeup_thread(mddev->thread); | |
d9d166c2 N |
2833 | return rv; |
2834 | } | |
2835 | ||
2836 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 2837 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 2838 | |
d4dbd025 N |
2839 | |
2840 | static ssize_t | |
2841 | layout_show(mddev_t *mddev, char *page) | |
2842 | { | |
2843 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
2844 | if (mddev->reshape_position != MaxSector && |
2845 | mddev->layout != mddev->new_layout) | |
2846 | return sprintf(page, "%d (%d)\n", | |
2847 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
2848 | return sprintf(page, "%d\n", mddev->layout); |
2849 | } | |
2850 | ||
2851 | static ssize_t | |
2852 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
2853 | { | |
2854 | char *e; | |
2855 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
2856 | |
2857 | if (!*buf || (*e && *e != '\n')) | |
2858 | return -EINVAL; | |
2859 | ||
b3546035 N |
2860 | if (mddev->pers) { |
2861 | int err; | |
50ac168a | 2862 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 2863 | return -EBUSY; |
597a711b | 2864 | mddev->new_layout = n; |
50ac168a | 2865 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
2866 | if (err) { |
2867 | mddev->new_layout = mddev->layout; | |
b3546035 | 2868 | return err; |
597a711b | 2869 | } |
b3546035 | 2870 | } else { |
08a02ecd | 2871 | mddev->new_layout = n; |
b3546035 N |
2872 | if (mddev->reshape_position == MaxSector) |
2873 | mddev->layout = n; | |
2874 | } | |
d4dbd025 N |
2875 | return len; |
2876 | } | |
2877 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 2878 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
2879 | |
2880 | ||
eae1701f | 2881 | static ssize_t |
96de1e66 | 2882 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 2883 | { |
bb636547 N |
2884 | if (mddev->raid_disks == 0) |
2885 | return 0; | |
08a02ecd N |
2886 | if (mddev->reshape_position != MaxSector && |
2887 | mddev->delta_disks != 0) | |
2888 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
2889 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
2890 | return sprintf(page, "%d\n", mddev->raid_disks); |
2891 | } | |
2892 | ||
da943b99 N |
2893 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
2894 | ||
2895 | static ssize_t | |
2896 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
2897 | { | |
da943b99 N |
2898 | char *e; |
2899 | int rv = 0; | |
2900 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2901 | ||
2902 | if (!*buf || (*e && *e != '\n')) | |
2903 | return -EINVAL; | |
2904 | ||
2905 | if (mddev->pers) | |
2906 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
2907 | else if (mddev->reshape_position != MaxSector) { |
2908 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
2909 | mddev->delta_disks = n - olddisks; | |
2910 | mddev->raid_disks = n; | |
2911 | } else | |
da943b99 N |
2912 | mddev->raid_disks = n; |
2913 | return rv ? rv : len; | |
2914 | } | |
2915 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 2916 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 2917 | |
3b34380a N |
2918 | static ssize_t |
2919 | chunk_size_show(mddev_t *mddev, char *page) | |
2920 | { | |
08a02ecd | 2921 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
2922 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
2923 | return sprintf(page, "%d (%d)\n", | |
2924 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
2925 | mddev->chunk_sectors << 9); |
2926 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
2927 | } |
2928 | ||
2929 | static ssize_t | |
2930 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
2931 | { | |
3b34380a N |
2932 | char *e; |
2933 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2934 | ||
3b34380a N |
2935 | if (!*buf || (*e && *e != '\n')) |
2936 | return -EINVAL; | |
2937 | ||
b3546035 N |
2938 | if (mddev->pers) { |
2939 | int err; | |
50ac168a | 2940 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 2941 | return -EBUSY; |
597a711b | 2942 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 2943 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
2944 | if (err) { |
2945 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 2946 | return err; |
597a711b | 2947 | } |
b3546035 | 2948 | } else { |
664e7c41 | 2949 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 2950 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 2951 | mddev->chunk_sectors = n >> 9; |
b3546035 | 2952 | } |
3b34380a N |
2953 | return len; |
2954 | } | |
2955 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 2956 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 2957 | |
a94213b1 N |
2958 | static ssize_t |
2959 | resync_start_show(mddev_t *mddev, char *page) | |
2960 | { | |
d1a7c503 N |
2961 | if (mddev->recovery_cp == MaxSector) |
2962 | return sprintf(page, "none\n"); | |
a94213b1 N |
2963 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
2964 | } | |
2965 | ||
2966 | static ssize_t | |
2967 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
2968 | { | |
a94213b1 N |
2969 | char *e; |
2970 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
2971 | ||
2972 | if (mddev->pers) | |
2973 | return -EBUSY; | |
2974 | if (!*buf || (*e && *e != '\n')) | |
2975 | return -EINVAL; | |
2976 | ||
2977 | mddev->recovery_cp = n; | |
2978 | return len; | |
2979 | } | |
2980 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 2981 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 2982 | |
9e653b63 N |
2983 | /* |
2984 | * The array state can be: | |
2985 | * | |
2986 | * clear | |
2987 | * No devices, no size, no level | |
2988 | * Equivalent to STOP_ARRAY ioctl | |
2989 | * inactive | |
2990 | * May have some settings, but array is not active | |
2991 | * all IO results in error | |
2992 | * When written, doesn't tear down array, but just stops it | |
2993 | * suspended (not supported yet) | |
2994 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 2995 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
2996 | * readonly |
2997 | * no resync can happen. no superblocks get written. | |
2998 | * write requests fail | |
2999 | * read-auto | |
3000 | * like readonly, but behaves like 'clean' on a write request. | |
3001 | * | |
3002 | * clean - no pending writes, but otherwise active. | |
3003 | * When written to inactive array, starts without resync | |
3004 | * If a write request arrives then | |
3005 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3006 | * if not known, block and switch to write-pending | |
3007 | * If written to an active array that has pending writes, then fails. | |
3008 | * active | |
3009 | * fully active: IO and resync can be happening. | |
3010 | * When written to inactive array, starts with resync | |
3011 | * | |
3012 | * write-pending | |
3013 | * clean, but writes are blocked waiting for 'active' to be written. | |
3014 | * | |
3015 | * active-idle | |
3016 | * like active, but no writes have been seen for a while (100msec). | |
3017 | * | |
3018 | */ | |
3019 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3020 | write_pending, active_idle, bad_word}; | |
05381954 | 3021 | static char *array_states[] = { |
9e653b63 N |
3022 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3023 | "write-pending", "active-idle", NULL }; | |
3024 | ||
3025 | static int match_word(const char *word, char **list) | |
3026 | { | |
3027 | int n; | |
3028 | for (n=0; list[n]; n++) | |
3029 | if (cmd_match(word, list[n])) | |
3030 | break; | |
3031 | return n; | |
3032 | } | |
3033 | ||
3034 | static ssize_t | |
3035 | array_state_show(mddev_t *mddev, char *page) | |
3036 | { | |
3037 | enum array_state st = inactive; | |
3038 | ||
3039 | if (mddev->pers) | |
3040 | switch(mddev->ro) { | |
3041 | case 1: | |
3042 | st = readonly; | |
3043 | break; | |
3044 | case 2: | |
3045 | st = read_auto; | |
3046 | break; | |
3047 | case 0: | |
3048 | if (mddev->in_sync) | |
3049 | st = clean; | |
e691063a N |
3050 | else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) |
3051 | st = write_pending; | |
9e653b63 N |
3052 | else if (mddev->safemode) |
3053 | st = active_idle; | |
3054 | else | |
3055 | st = active; | |
3056 | } | |
3057 | else { | |
3058 | if (list_empty(&mddev->disks) && | |
3059 | mddev->raid_disks == 0 && | |
58c0fed4 | 3060 | mddev->dev_sectors == 0) |
9e653b63 N |
3061 | st = clear; |
3062 | else | |
3063 | st = inactive; | |
3064 | } | |
3065 | return sprintf(page, "%s\n", array_states[st]); | |
3066 | } | |
3067 | ||
df5b20cf | 3068 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
9e653b63 N |
3069 | static int do_md_run(mddev_t * mddev); |
3070 | static int restart_array(mddev_t *mddev); | |
3071 | ||
3072 | static ssize_t | |
3073 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3074 | { | |
3075 | int err = -EINVAL; | |
3076 | enum array_state st = match_word(buf, array_states); | |
3077 | switch(st) { | |
3078 | case bad_word: | |
3079 | break; | |
3080 | case clear: | |
3081 | /* stopping an active array */ | |
f2ea68cf | 3082 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3083 | return -EBUSY; |
df5b20cf | 3084 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3085 | break; |
3086 | case inactive: | |
3087 | /* stopping an active array */ | |
3088 | if (mddev->pers) { | |
f2ea68cf | 3089 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3090 | return -EBUSY; |
df5b20cf | 3091 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3092 | } else |
3093 | err = 0; /* already inactive */ | |
9e653b63 N |
3094 | break; |
3095 | case suspended: | |
3096 | break; /* not supported yet */ | |
3097 | case readonly: | |
3098 | if (mddev->pers) | |
df5b20cf | 3099 | err = do_md_stop(mddev, 1, 0); |
9e653b63 N |
3100 | else { |
3101 | mddev->ro = 1; | |
648b629e | 3102 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3103 | err = do_md_run(mddev); |
3104 | } | |
3105 | break; | |
3106 | case read_auto: | |
9e653b63 | 3107 | if (mddev->pers) { |
80268ee9 | 3108 | if (mddev->ro == 0) |
df5b20cf | 3109 | err = do_md_stop(mddev, 1, 0); |
80268ee9 | 3110 | else if (mddev->ro == 1) |
648b629e N |
3111 | err = restart_array(mddev); |
3112 | if (err == 0) { | |
3113 | mddev->ro = 2; | |
3114 | set_disk_ro(mddev->gendisk, 0); | |
3115 | } | |
9e653b63 N |
3116 | } else { |
3117 | mddev->ro = 2; | |
3118 | err = do_md_run(mddev); | |
3119 | } | |
3120 | break; | |
3121 | case clean: | |
3122 | if (mddev->pers) { | |
3123 | restart_array(mddev); | |
3124 | spin_lock_irq(&mddev->write_lock); | |
3125 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3126 | if (mddev->in_sync == 0) { |
3127 | mddev->in_sync = 1; | |
31a59e34 N |
3128 | if (mddev->safemode == 1) |
3129 | mddev->safemode = 0; | |
e691063a N |
3130 | if (mddev->persistent) |
3131 | set_bit(MD_CHANGE_CLEAN, | |
3132 | &mddev->flags); | |
3133 | } | |
3134 | err = 0; | |
3135 | } else | |
3136 | err = -EBUSY; | |
9e653b63 | 3137 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3138 | } else |
3139 | err = -EINVAL; | |
9e653b63 N |
3140 | break; |
3141 | case active: | |
3142 | if (mddev->pers) { | |
3143 | restart_array(mddev); | |
e691063a N |
3144 | if (mddev->external) |
3145 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
9e653b63 N |
3146 | wake_up(&mddev->sb_wait); |
3147 | err = 0; | |
3148 | } else { | |
3149 | mddev->ro = 0; | |
648b629e | 3150 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3151 | err = do_md_run(mddev); |
3152 | } | |
3153 | break; | |
3154 | case write_pending: | |
3155 | case active_idle: | |
3156 | /* these cannot be set */ | |
3157 | break; | |
3158 | } | |
3159 | if (err) | |
3160 | return err; | |
0fd62b86 | 3161 | else { |
b62b7590 | 3162 | sysfs_notify_dirent(mddev->sysfs_state); |
9e653b63 | 3163 | return len; |
0fd62b86 | 3164 | } |
9e653b63 | 3165 | } |
80ca3a44 N |
3166 | static struct md_sysfs_entry md_array_state = |
3167 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3168 | |
6d7ff738 N |
3169 | static ssize_t |
3170 | null_show(mddev_t *mddev, char *page) | |
3171 | { | |
3172 | return -EINVAL; | |
3173 | } | |
3174 | ||
3175 | static ssize_t | |
3176 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3177 | { | |
3178 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3179 | /* The new device is added to the array. | |
3180 | * If the array has a persistent superblock, we read the | |
3181 | * superblock to initialise info and check validity. | |
3182 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3183 | * which mainly checks size. | |
3184 | */ | |
3185 | char *e; | |
3186 | int major = simple_strtoul(buf, &e, 10); | |
3187 | int minor; | |
3188 | dev_t dev; | |
3189 | mdk_rdev_t *rdev; | |
3190 | int err; | |
3191 | ||
3192 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3193 | return -EINVAL; | |
3194 | minor = simple_strtoul(e+1, &e, 10); | |
3195 | if (*e && *e != '\n') | |
3196 | return -EINVAL; | |
3197 | dev = MKDEV(major, minor); | |
3198 | if (major != MAJOR(dev) || | |
3199 | minor != MINOR(dev)) | |
3200 | return -EOVERFLOW; | |
3201 | ||
3202 | ||
3203 | if (mddev->persistent) { | |
3204 | rdev = md_import_device(dev, mddev->major_version, | |
3205 | mddev->minor_version); | |
3206 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3207 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3208 | mdk_rdev_t, same_set); | |
3209 | err = super_types[mddev->major_version] | |
3210 | .load_super(rdev, rdev0, mddev->minor_version); | |
3211 | if (err < 0) | |
3212 | goto out; | |
3213 | } | |
c5d79adb N |
3214 | } else if (mddev->external) |
3215 | rdev = md_import_device(dev, -2, -1); | |
3216 | else | |
6d7ff738 N |
3217 | rdev = md_import_device(dev, -1, -1); |
3218 | ||
3219 | if (IS_ERR(rdev)) | |
3220 | return PTR_ERR(rdev); | |
3221 | err = bind_rdev_to_array(rdev, mddev); | |
3222 | out: | |
3223 | if (err) | |
3224 | export_rdev(rdev); | |
3225 | return err ? err : len; | |
3226 | } | |
3227 | ||
3228 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3229 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3230 | |
9b1d1dac PC |
3231 | static ssize_t |
3232 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3233 | { | |
3234 | char *end; | |
3235 | unsigned long chunk, end_chunk; | |
3236 | ||
3237 | if (!mddev->bitmap) | |
3238 | goto out; | |
3239 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3240 | while (*buf) { | |
3241 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3242 | if (buf == end) break; | |
3243 | if (*end == '-') { /* range */ | |
3244 | buf = end + 1; | |
3245 | end_chunk = simple_strtoul(buf, &end, 0); | |
3246 | if (buf == end) break; | |
3247 | } | |
3248 | if (*end && !isspace(*end)) break; | |
3249 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
3250 | buf = end; | |
3251 | while (isspace(*buf)) buf++; | |
3252 | } | |
3253 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3254 | out: | |
3255 | return len; | |
3256 | } | |
3257 | ||
3258 | static struct md_sysfs_entry md_bitmap = | |
3259 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3260 | ||
a35b0d69 N |
3261 | static ssize_t |
3262 | size_show(mddev_t *mddev, char *page) | |
3263 | { | |
58c0fed4 AN |
3264 | return sprintf(page, "%llu\n", |
3265 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3266 | } |
3267 | ||
d71f9f88 | 3268 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3269 | |
3270 | static ssize_t | |
3271 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3272 | { | |
3273 | /* If array is inactive, we can reduce the component size, but | |
3274 | * not increase it (except from 0). | |
3275 | * If array is active, we can try an on-line resize | |
3276 | */ | |
b522adcd DW |
3277 | sector_t sectors; |
3278 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3279 | |
58c0fed4 AN |
3280 | if (err < 0) |
3281 | return err; | |
a35b0d69 | 3282 | if (mddev->pers) { |
58c0fed4 | 3283 | err = update_size(mddev, sectors); |
850b2b42 | 3284 | md_update_sb(mddev, 1); |
a35b0d69 | 3285 | } else { |
58c0fed4 AN |
3286 | if (mddev->dev_sectors == 0 || |
3287 | mddev->dev_sectors > sectors) | |
3288 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3289 | else |
3290 | err = -ENOSPC; | |
3291 | } | |
3292 | return err ? err : len; | |
3293 | } | |
3294 | ||
3295 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3296 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3297 | |
8bb93aac N |
3298 | |
3299 | /* Metdata version. | |
e691063a N |
3300 | * This is one of |
3301 | * 'none' for arrays with no metadata (good luck...) | |
3302 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3303 | * or N.M for internally known formats |
3304 | */ | |
3305 | static ssize_t | |
3306 | metadata_show(mddev_t *mddev, char *page) | |
3307 | { | |
3308 | if (mddev->persistent) | |
3309 | return sprintf(page, "%d.%d\n", | |
3310 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3311 | else if (mddev->external) |
3312 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3313 | else |
3314 | return sprintf(page, "none\n"); | |
3315 | } | |
3316 | ||
3317 | static ssize_t | |
3318 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3319 | { | |
3320 | int major, minor; | |
3321 | char *e; | |
ea43ddd8 N |
3322 | /* Changing the details of 'external' metadata is |
3323 | * always permitted. Otherwise there must be | |
3324 | * no devices attached to the array. | |
3325 | */ | |
3326 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3327 | ; | |
3328 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3329 | return -EBUSY; |
3330 | ||
3331 | if (cmd_match(buf, "none")) { | |
3332 | mddev->persistent = 0; | |
e691063a N |
3333 | mddev->external = 0; |
3334 | mddev->major_version = 0; | |
3335 | mddev->minor_version = 90; | |
3336 | return len; | |
3337 | } | |
3338 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3339 | size_t namelen = len-9; |
e691063a N |
3340 | if (namelen >= sizeof(mddev->metadata_type)) |
3341 | namelen = sizeof(mddev->metadata_type)-1; | |
3342 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3343 | mddev->metadata_type[namelen] = 0; | |
3344 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3345 | mddev->metadata_type[--namelen] = 0; | |
3346 | mddev->persistent = 0; | |
3347 | mddev->external = 1; | |
8bb93aac N |
3348 | mddev->major_version = 0; |
3349 | mddev->minor_version = 90; | |
3350 | return len; | |
3351 | } | |
3352 | major = simple_strtoul(buf, &e, 10); | |
3353 | if (e==buf || *e != '.') | |
3354 | return -EINVAL; | |
3355 | buf = e+1; | |
3356 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3357 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3358 | return -EINVAL; |
50511da3 | 3359 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3360 | return -ENOENT; |
3361 | mddev->major_version = major; | |
3362 | mddev->minor_version = minor; | |
3363 | mddev->persistent = 1; | |
e691063a | 3364 | mddev->external = 0; |
8bb93aac N |
3365 | return len; |
3366 | } | |
3367 | ||
3368 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3369 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3370 | |
24dd469d | 3371 | static ssize_t |
7eec314d | 3372 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3373 | { |
7eec314d | 3374 | char *type = "idle"; |
b6a9ce68 N |
3375 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3376 | type = "frozen"; | |
3377 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3378 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3379 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3380 | type = "reshape"; | |
3381 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3382 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3383 | type = "resync"; | |
3384 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3385 | type = "check"; | |
3386 | else | |
3387 | type = "repair"; | |
72a23c21 | 3388 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3389 | type = "recover"; |
3390 | } | |
3391 | return sprintf(page, "%s\n", type); | |
3392 | } | |
3393 | ||
3394 | static ssize_t | |
7eec314d | 3395 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3396 | { |
7eec314d N |
3397 | if (!mddev->pers || !mddev->pers->sync_request) |
3398 | return -EINVAL; | |
3399 | ||
b6a9ce68 N |
3400 | if (cmd_match(page, "frozen")) |
3401 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3402 | else | |
3403 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3404 | ||
3405 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3406 | if (mddev->sync_thread) { |
3407 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
3408 | md_unregister_thread(mddev->sync_thread); | |
3409 | mddev->sync_thread = NULL; | |
3410 | mddev->recovery = 0; | |
3411 | } | |
03c902e1 N |
3412 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3413 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3414 | return -EBUSY; |
72a23c21 NB |
3415 | else if (cmd_match(page, "resync")) |
3416 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3417 | else if (cmd_match(page, "recover")) { | |
3418 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3419 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3420 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3421 | int err; |
3422 | if (mddev->pers->start_reshape == NULL) | |
3423 | return -EINVAL; | |
3424 | err = mddev->pers->start_reshape(mddev); | |
3425 | if (err) | |
3426 | return err; | |
a99ac971 | 3427 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3428 | } else { |
bce74dac | 3429 | if (cmd_match(page, "check")) |
7eec314d | 3430 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3431 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3432 | return -EINVAL; |
3433 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3434 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3435 | } |
03c902e1 | 3436 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3437 | md_wakeup_thread(mddev->thread); |
0c3573f1 | 3438 | sysfs_notify_dirent(mddev->sysfs_action); |
24dd469d N |
3439 | return len; |
3440 | } | |
3441 | ||
9d88883e | 3442 | static ssize_t |
96de1e66 | 3443 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3444 | { |
3445 | return sprintf(page, "%llu\n", | |
3446 | (unsigned long long) mddev->resync_mismatches); | |
3447 | } | |
3448 | ||
80ca3a44 N |
3449 | static struct md_sysfs_entry md_scan_mode = |
3450 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3451 | |
96de1e66 | 3452 | |
80ca3a44 | 3453 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3454 | |
88202a0c N |
3455 | static ssize_t |
3456 | sync_min_show(mddev_t *mddev, char *page) | |
3457 | { | |
3458 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3459 | mddev->sync_speed_min ? "local": "system"); | |
3460 | } | |
3461 | ||
3462 | static ssize_t | |
3463 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3464 | { | |
3465 | int min; | |
3466 | char *e; | |
3467 | if (strncmp(buf, "system", 6)==0) { | |
3468 | mddev->sync_speed_min = 0; | |
3469 | return len; | |
3470 | } | |
3471 | min = simple_strtoul(buf, &e, 10); | |
3472 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3473 | return -EINVAL; | |
3474 | mddev->sync_speed_min = min; | |
3475 | return len; | |
3476 | } | |
3477 | ||
3478 | static struct md_sysfs_entry md_sync_min = | |
3479 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3480 | ||
3481 | static ssize_t | |
3482 | sync_max_show(mddev_t *mddev, char *page) | |
3483 | { | |
3484 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3485 | mddev->sync_speed_max ? "local": "system"); | |
3486 | } | |
3487 | ||
3488 | static ssize_t | |
3489 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3490 | { | |
3491 | int max; | |
3492 | char *e; | |
3493 | if (strncmp(buf, "system", 6)==0) { | |
3494 | mddev->sync_speed_max = 0; | |
3495 | return len; | |
3496 | } | |
3497 | max = simple_strtoul(buf, &e, 10); | |
3498 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3499 | return -EINVAL; | |
3500 | mddev->sync_speed_max = max; | |
3501 | return len; | |
3502 | } | |
3503 | ||
3504 | static struct md_sysfs_entry md_sync_max = | |
3505 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3506 | ||
d7f3d291 IP |
3507 | static ssize_t |
3508 | degraded_show(mddev_t *mddev, char *page) | |
3509 | { | |
3510 | return sprintf(page, "%d\n", mddev->degraded); | |
3511 | } | |
3512 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3513 | |
90b08710 BS |
3514 | static ssize_t |
3515 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3516 | { | |
3517 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3518 | } | |
3519 | ||
3520 | static ssize_t | |
3521 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3522 | { | |
3523 | long n; | |
3524 | ||
3525 | if (strict_strtol(buf, 10, &n)) | |
3526 | return -EINVAL; | |
3527 | ||
3528 | if (n != 0 && n != 1) | |
3529 | return -EINVAL; | |
3530 | ||
3531 | mddev->parallel_resync = n; | |
3532 | ||
3533 | if (mddev->sync_thread) | |
3534 | wake_up(&resync_wait); | |
3535 | ||
3536 | return len; | |
3537 | } | |
3538 | ||
3539 | /* force parallel resync, even with shared block devices */ | |
3540 | static struct md_sysfs_entry md_sync_force_parallel = | |
3541 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3542 | sync_force_parallel_show, sync_force_parallel_store); | |
3543 | ||
88202a0c N |
3544 | static ssize_t |
3545 | sync_speed_show(mddev_t *mddev, char *page) | |
3546 | { | |
3547 | unsigned long resync, dt, db; | |
d1a7c503 N |
3548 | if (mddev->curr_resync == 0) |
3549 | return sprintf(page, "none\n"); | |
9687a60c AN |
3550 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3551 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3552 | if (!dt) dt++; |
9687a60c AN |
3553 | db = resync - mddev->resync_mark_cnt; |
3554 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3555 | } |
3556 | ||
80ca3a44 | 3557 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3558 | |
3559 | static ssize_t | |
3560 | sync_completed_show(mddev_t *mddev, char *page) | |
3561 | { | |
58c0fed4 | 3562 | unsigned long max_sectors, resync; |
88202a0c | 3563 | |
acb180b0 N |
3564 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3565 | return sprintf(page, "none\n"); | |
3566 | ||
88202a0c | 3567 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3568 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3569 | else |
58c0fed4 | 3570 | max_sectors = mddev->dev_sectors; |
88202a0c | 3571 | |
acb180b0 | 3572 | resync = mddev->curr_resync_completed; |
58c0fed4 | 3573 | return sprintf(page, "%lu / %lu\n", resync, max_sectors); |
88202a0c N |
3574 | } |
3575 | ||
80ca3a44 | 3576 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3577 | |
5e96ee65 NB |
3578 | static ssize_t |
3579 | min_sync_show(mddev_t *mddev, char *page) | |
3580 | { | |
3581 | return sprintf(page, "%llu\n", | |
3582 | (unsigned long long)mddev->resync_min); | |
3583 | } | |
3584 | static ssize_t | |
3585 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3586 | { | |
3587 | unsigned long long min; | |
3588 | if (strict_strtoull(buf, 10, &min)) | |
3589 | return -EINVAL; | |
3590 | if (min > mddev->resync_max) | |
3591 | return -EINVAL; | |
3592 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3593 | return -EBUSY; | |
3594 | ||
3595 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3596 | if (mddev->chunk_sectors) { |
2ac06c33 | 3597 | sector_t temp = min; |
9d8f0363 | 3598 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3599 | return -EINVAL; |
3600 | } | |
3601 | mddev->resync_min = min; | |
3602 | ||
3603 | return len; | |
3604 | } | |
3605 | ||
3606 | static struct md_sysfs_entry md_min_sync = | |
3607 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3608 | ||
c6207277 N |
3609 | static ssize_t |
3610 | max_sync_show(mddev_t *mddev, char *page) | |
3611 | { | |
3612 | if (mddev->resync_max == MaxSector) | |
3613 | return sprintf(page, "max\n"); | |
3614 | else | |
3615 | return sprintf(page, "%llu\n", | |
3616 | (unsigned long long)mddev->resync_max); | |
3617 | } | |
3618 | static ssize_t | |
3619 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3620 | { | |
3621 | if (strncmp(buf, "max", 3) == 0) | |
3622 | mddev->resync_max = MaxSector; | |
3623 | else { | |
5e96ee65 NB |
3624 | unsigned long long max; |
3625 | if (strict_strtoull(buf, 10, &max)) | |
3626 | return -EINVAL; | |
3627 | if (max < mddev->resync_min) | |
c6207277 N |
3628 | return -EINVAL; |
3629 | if (max < mddev->resync_max && | |
4d484a4a | 3630 | mddev->ro == 0 && |
c6207277 N |
3631 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3632 | return -EBUSY; | |
3633 | ||
3634 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3635 | if (mddev->chunk_sectors) { |
2ac06c33 | 3636 | sector_t temp = max; |
9d8f0363 | 3637 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
3638 | return -EINVAL; |
3639 | } | |
3640 | mddev->resync_max = max; | |
3641 | } | |
3642 | wake_up(&mddev->recovery_wait); | |
3643 | return len; | |
3644 | } | |
3645 | ||
3646 | static struct md_sysfs_entry md_max_sync = | |
3647 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
3648 | ||
e464eafd N |
3649 | static ssize_t |
3650 | suspend_lo_show(mddev_t *mddev, char *page) | |
3651 | { | |
3652 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
3653 | } | |
3654 | ||
3655 | static ssize_t | |
3656 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
3657 | { | |
3658 | char *e; | |
3659 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3660 | ||
b8d966ef N |
3661 | if (mddev->pers == NULL || |
3662 | mddev->pers->quiesce == NULL) | |
e464eafd N |
3663 | return -EINVAL; |
3664 | if (buf == e || (*e && *e != '\n')) | |
3665 | return -EINVAL; | |
3666 | if (new >= mddev->suspend_hi || | |
3667 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
3668 | mddev->suspend_lo = new; | |
3669 | mddev->pers->quiesce(mddev, 2); | |
3670 | return len; | |
3671 | } else | |
3672 | return -EINVAL; | |
3673 | } | |
3674 | static struct md_sysfs_entry md_suspend_lo = | |
3675 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
3676 | ||
3677 | ||
3678 | static ssize_t | |
3679 | suspend_hi_show(mddev_t *mddev, char *page) | |
3680 | { | |
3681 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
3682 | } | |
3683 | ||
3684 | static ssize_t | |
3685 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
3686 | { | |
3687 | char *e; | |
3688 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3689 | ||
b8d966ef N |
3690 | if (mddev->pers == NULL || |
3691 | mddev->pers->quiesce == NULL) | |
e464eafd N |
3692 | return -EINVAL; |
3693 | if (buf == e || (*e && *e != '\n')) | |
3694 | return -EINVAL; | |
3695 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
3696 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
3697 | mddev->suspend_hi = new; | |
3698 | mddev->pers->quiesce(mddev, 1); | |
3699 | mddev->pers->quiesce(mddev, 0); | |
3700 | return len; | |
3701 | } else | |
3702 | return -EINVAL; | |
3703 | } | |
3704 | static struct md_sysfs_entry md_suspend_hi = | |
3705 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
3706 | ||
08a02ecd N |
3707 | static ssize_t |
3708 | reshape_position_show(mddev_t *mddev, char *page) | |
3709 | { | |
3710 | if (mddev->reshape_position != MaxSector) | |
3711 | return sprintf(page, "%llu\n", | |
3712 | (unsigned long long)mddev->reshape_position); | |
3713 | strcpy(page, "none\n"); | |
3714 | return 5; | |
3715 | } | |
3716 | ||
3717 | static ssize_t | |
3718 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
3719 | { | |
3720 | char *e; | |
3721 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3722 | if (mddev->pers) | |
3723 | return -EBUSY; | |
3724 | if (buf == e || (*e && *e != '\n')) | |
3725 | return -EINVAL; | |
3726 | mddev->reshape_position = new; | |
3727 | mddev->delta_disks = 0; | |
3728 | mddev->new_level = mddev->level; | |
3729 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3730 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
3731 | return len; |
3732 | } | |
3733 | ||
3734 | static struct md_sysfs_entry md_reshape_position = | |
3735 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
3736 | reshape_position_store); | |
3737 | ||
b522adcd DW |
3738 | static ssize_t |
3739 | array_size_show(mddev_t *mddev, char *page) | |
3740 | { | |
3741 | if (mddev->external_size) | |
3742 | return sprintf(page, "%llu\n", | |
3743 | (unsigned long long)mddev->array_sectors/2); | |
3744 | else | |
3745 | return sprintf(page, "default\n"); | |
3746 | } | |
3747 | ||
3748 | static ssize_t | |
3749 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3750 | { | |
3751 | sector_t sectors; | |
3752 | ||
3753 | if (strncmp(buf, "default", 7) == 0) { | |
3754 | if (mddev->pers) | |
3755 | sectors = mddev->pers->size(mddev, 0, 0); | |
3756 | else | |
3757 | sectors = mddev->array_sectors; | |
3758 | ||
3759 | mddev->external_size = 0; | |
3760 | } else { | |
3761 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
3762 | return -EINVAL; | |
3763 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 3764 | return -E2BIG; |
b522adcd DW |
3765 | |
3766 | mddev->external_size = 1; | |
3767 | } | |
3768 | ||
3769 | mddev->array_sectors = sectors; | |
3770 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e N |
3771 | if (mddev->pers) |
3772 | revalidate_disk(mddev->gendisk); | |
b522adcd DW |
3773 | |
3774 | return len; | |
3775 | } | |
3776 | ||
3777 | static struct md_sysfs_entry md_array_size = | |
3778 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
3779 | array_size_store); | |
e464eafd | 3780 | |
eae1701f N |
3781 | static struct attribute *md_default_attrs[] = { |
3782 | &md_level.attr, | |
d4dbd025 | 3783 | &md_layout.attr, |
eae1701f | 3784 | &md_raid_disks.attr, |
3b34380a | 3785 | &md_chunk_size.attr, |
a35b0d69 | 3786 | &md_size.attr, |
a94213b1 | 3787 | &md_resync_start.attr, |
8bb93aac | 3788 | &md_metadata.attr, |
6d7ff738 | 3789 | &md_new_device.attr, |
16f17b39 | 3790 | &md_safe_delay.attr, |
9e653b63 | 3791 | &md_array_state.attr, |
08a02ecd | 3792 | &md_reshape_position.attr, |
b522adcd | 3793 | &md_array_size.attr, |
411036fa N |
3794 | NULL, |
3795 | }; | |
3796 | ||
3797 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 3798 | &md_scan_mode.attr, |
9d88883e | 3799 | &md_mismatches.attr, |
88202a0c N |
3800 | &md_sync_min.attr, |
3801 | &md_sync_max.attr, | |
3802 | &md_sync_speed.attr, | |
90b08710 | 3803 | &md_sync_force_parallel.attr, |
88202a0c | 3804 | &md_sync_completed.attr, |
5e96ee65 | 3805 | &md_min_sync.attr, |
c6207277 | 3806 | &md_max_sync.attr, |
e464eafd N |
3807 | &md_suspend_lo.attr, |
3808 | &md_suspend_hi.attr, | |
9b1d1dac | 3809 | &md_bitmap.attr, |
d7f3d291 | 3810 | &md_degraded.attr, |
eae1701f N |
3811 | NULL, |
3812 | }; | |
411036fa N |
3813 | static struct attribute_group md_redundancy_group = { |
3814 | .name = NULL, | |
3815 | .attrs = md_redundancy_attrs, | |
3816 | }; | |
3817 | ||
eae1701f N |
3818 | |
3819 | static ssize_t | |
3820 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3821 | { | |
3822 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3823 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3824 | ssize_t rv; |
eae1701f N |
3825 | |
3826 | if (!entry->show) | |
3827 | return -EIO; | |
5dc5cf7d IM |
3828 | rv = mddev_lock(mddev); |
3829 | if (!rv) { | |
3830 | rv = entry->show(mddev, page); | |
3831 | mddev_unlock(mddev); | |
3832 | } | |
96de1e66 | 3833 | return rv; |
eae1701f N |
3834 | } |
3835 | ||
3836 | static ssize_t | |
3837 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
3838 | const char *page, size_t length) | |
3839 | { | |
3840 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3841 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3842 | ssize_t rv; |
eae1701f N |
3843 | |
3844 | if (!entry->store) | |
3845 | return -EIO; | |
67463acb N |
3846 | if (!capable(CAP_SYS_ADMIN)) |
3847 | return -EACCES; | |
5dc5cf7d | 3848 | rv = mddev_lock(mddev); |
d3374825 N |
3849 | if (mddev->hold_active == UNTIL_IOCTL) |
3850 | mddev->hold_active = 0; | |
5dc5cf7d IM |
3851 | if (!rv) { |
3852 | rv = entry->store(mddev, page, length); | |
3853 | mddev_unlock(mddev); | |
3854 | } | |
96de1e66 | 3855 | return rv; |
eae1701f N |
3856 | } |
3857 | ||
3858 | static void md_free(struct kobject *ko) | |
3859 | { | |
3860 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
3861 | |
3862 | if (mddev->sysfs_state) | |
3863 | sysfs_put(mddev->sysfs_state); | |
3864 | ||
3865 | if (mddev->gendisk) { | |
3866 | del_gendisk(mddev->gendisk); | |
3867 | put_disk(mddev->gendisk); | |
3868 | } | |
3869 | if (mddev->queue) | |
3870 | blk_cleanup_queue(mddev->queue); | |
3871 | ||
eae1701f N |
3872 | kfree(mddev); |
3873 | } | |
3874 | ||
3875 | static struct sysfs_ops md_sysfs_ops = { | |
3876 | .show = md_attr_show, | |
3877 | .store = md_attr_store, | |
3878 | }; | |
3879 | static struct kobj_type md_ktype = { | |
3880 | .release = md_free, | |
3881 | .sysfs_ops = &md_sysfs_ops, | |
3882 | .default_attrs = md_default_attrs, | |
3883 | }; | |
3884 | ||
1da177e4 LT |
3885 | int mdp_major = 0; |
3886 | ||
5fd3a17e DW |
3887 | static void mddev_delayed_delete(struct work_struct *ws) |
3888 | { | |
3889 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
3890 | ||
3891 | if (mddev->private == &md_redundancy_group) { | |
3892 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
3893 | if (mddev->sysfs_action) | |
3894 | sysfs_put(mddev->sysfs_action); | |
3895 | mddev->sysfs_action = NULL; | |
3896 | mddev->private = NULL; | |
3897 | } | |
3898 | kobject_del(&mddev->kobj); | |
3899 | kobject_put(&mddev->kobj); | |
3900 | } | |
3901 | ||
efeb53c0 | 3902 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 3903 | { |
48c9c27b | 3904 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
3905 | mddev_t *mddev = mddev_find(dev); |
3906 | struct gendisk *disk; | |
efeb53c0 N |
3907 | int partitioned; |
3908 | int shift; | |
3909 | int unit; | |
3830c62f | 3910 | int error; |
1da177e4 LT |
3911 | |
3912 | if (!mddev) | |
efeb53c0 N |
3913 | return -ENODEV; |
3914 | ||
3915 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
3916 | shift = partitioned ? MdpMinorShift : 0; | |
3917 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 3918 | |
d3374825 N |
3919 | /* wait for any previous instance if this device |
3920 | * to be completed removed (mddev_delayed_delete). | |
3921 | */ | |
3922 | flush_scheduled_work(); | |
3923 | ||
48c9c27b | 3924 | mutex_lock(&disks_mutex); |
0909dc44 N |
3925 | error = -EEXIST; |
3926 | if (mddev->gendisk) | |
3927 | goto abort; | |
efeb53c0 N |
3928 | |
3929 | if (name) { | |
3930 | /* Need to ensure that 'name' is not a duplicate. | |
3931 | */ | |
3932 | mddev_t *mddev2; | |
3933 | spin_lock(&all_mddevs_lock); | |
3934 | ||
3935 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
3936 | if (mddev2->gendisk && | |
3937 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
3938 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 3939 | goto abort; |
efeb53c0 N |
3940 | } |
3941 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 3942 | } |
8b765398 | 3943 | |
0909dc44 | 3944 | error = -ENOMEM; |
8b765398 | 3945 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
3946 | if (!mddev->queue) |
3947 | goto abort; | |
409c57f3 N |
3948 | mddev->queue->queuedata = mddev; |
3949 | ||
8b765398 N |
3950 | /* Can be unlocked because the queue is new: no concurrency */ |
3951 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); | |
3952 | ||
409c57f3 | 3953 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 3954 | |
1da177e4 LT |
3955 | disk = alloc_disk(1 << shift); |
3956 | if (!disk) { | |
8b765398 N |
3957 | blk_cleanup_queue(mddev->queue); |
3958 | mddev->queue = NULL; | |
0909dc44 | 3959 | goto abort; |
1da177e4 | 3960 | } |
efeb53c0 | 3961 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 3962 | disk->first_minor = unit << shift; |
efeb53c0 N |
3963 | if (name) |
3964 | strcpy(disk->disk_name, name); | |
3965 | else if (partitioned) | |
1da177e4 | 3966 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 3967 | else |
1da177e4 | 3968 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
3969 | disk->fops = &md_fops; |
3970 | disk->private_data = mddev; | |
3971 | disk->queue = mddev->queue; | |
92850bbd | 3972 | /* Allow extended partitions. This makes the |
d3374825 | 3973 | * 'mdp' device redundant, but we can't really |
92850bbd N |
3974 | * remove it now. |
3975 | */ | |
3976 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 LT |
3977 | add_disk(disk); |
3978 | mddev->gendisk = disk; | |
ed9e1982 TH |
3979 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
3980 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
3981 | if (error) { |
3982 | /* This isn't possible, but as kobject_init_and_add is marked | |
3983 | * __must_check, we must do something with the result | |
3984 | */ | |
5e55e2f5 N |
3985 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
3986 | disk->disk_name); | |
0909dc44 N |
3987 | error = 0; |
3988 | } | |
3989 | abort: | |
3990 | mutex_unlock(&disks_mutex); | |
3991 | if (!error) { | |
3830c62f | 3992 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
b62b7590 N |
3993 | mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state"); |
3994 | } | |
d3374825 | 3995 | mddev_put(mddev); |
0909dc44 | 3996 | return error; |
efeb53c0 N |
3997 | } |
3998 | ||
3999 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4000 | { | |
4001 | md_alloc(dev, NULL); | |
1da177e4 LT |
4002 | return NULL; |
4003 | } | |
4004 | ||
efeb53c0 N |
4005 | static int add_named_array(const char *val, struct kernel_param *kp) |
4006 | { | |
4007 | /* val must be "md_*" where * is not all digits. | |
4008 | * We allocate an array with a large free minor number, and | |
4009 | * set the name to val. val must not already be an active name. | |
4010 | */ | |
4011 | int len = strlen(val); | |
4012 | char buf[DISK_NAME_LEN]; | |
4013 | ||
4014 | while (len && val[len-1] == '\n') | |
4015 | len--; | |
4016 | if (len >= DISK_NAME_LEN) | |
4017 | return -E2BIG; | |
4018 | strlcpy(buf, val, len+1); | |
4019 | if (strncmp(buf, "md_", 3) != 0) | |
4020 | return -EINVAL; | |
4021 | return md_alloc(0, buf); | |
4022 | } | |
4023 | ||
1da177e4 LT |
4024 | static void md_safemode_timeout(unsigned long data) |
4025 | { | |
4026 | mddev_t *mddev = (mddev_t *) data; | |
4027 | ||
0fd62b86 NB |
4028 | if (!atomic_read(&mddev->writes_pending)) { |
4029 | mddev->safemode = 1; | |
4030 | if (mddev->external) | |
b62b7590 | 4031 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 | 4032 | } |
1da177e4 LT |
4033 | md_wakeup_thread(mddev->thread); |
4034 | } | |
4035 | ||
6ff8d8ec | 4036 | static int start_dirty_degraded; |
1da177e4 LT |
4037 | |
4038 | static int do_md_run(mddev_t * mddev) | |
4039 | { | |
2604b703 | 4040 | int err; |
1da177e4 LT |
4041 | mdk_rdev_t *rdev; |
4042 | struct gendisk *disk; | |
2604b703 | 4043 | struct mdk_personality *pers; |
1da177e4 | 4044 | |
a757e64c N |
4045 | if (list_empty(&mddev->disks)) |
4046 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4047 | return -EINVAL; |
1da177e4 LT |
4048 | |
4049 | if (mddev->pers) | |
4050 | return -EBUSY; | |
4051 | ||
4052 | /* | |
4053 | * Analyze all RAID superblock(s) | |
4054 | */ | |
1ec4a939 N |
4055 | if (!mddev->raid_disks) { |
4056 | if (!mddev->persistent) | |
4057 | return -EINVAL; | |
a757e64c | 4058 | analyze_sbs(mddev); |
1ec4a939 | 4059 | } |
1da177e4 | 4060 | |
d9d166c2 N |
4061 | if (mddev->level != LEVEL_NONE) |
4062 | request_module("md-level-%d", mddev->level); | |
4063 | else if (mddev->clevel[0]) | |
4064 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4065 | |
4066 | /* | |
4067 | * Drop all container device buffers, from now on | |
4068 | * the only valid external interface is through the md | |
4069 | * device. | |
1da177e4 | 4070 | */ |
159ec1fc | 4071 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 4072 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4073 | continue; |
4074 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4075 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4076 | |
4077 | /* perform some consistency tests on the device. | |
4078 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4079 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4080 | */ |
0f420358 | 4081 | if (rdev->data_offset < rdev->sb_start) { |
58c0fed4 AN |
4082 | if (mddev->dev_sectors && |
4083 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4084 | > rdev->sb_start) { |
f0d76d70 N |
4085 | printk("md: %s: data overlaps metadata\n", |
4086 | mdname(mddev)); | |
4087 | return -EINVAL; | |
4088 | } | |
4089 | } else { | |
0f420358 | 4090 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4091 | > rdev->data_offset) { |
4092 | printk("md: %s: metadata overlaps data\n", | |
4093 | mdname(mddev)); | |
4094 | return -EINVAL; | |
4095 | } | |
4096 | } | |
3c0ee63a | 4097 | sysfs_notify_dirent(rdev->sysfs_state); |
1da177e4 LT |
4098 | } |
4099 | ||
4100 | md_probe(mddev->unit, NULL, NULL); | |
4101 | disk = mddev->gendisk; | |
4102 | if (!disk) | |
4103 | return -ENOMEM; | |
4104 | ||
4105 | spin_lock(&pers_lock); | |
d9d166c2 | 4106 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4107 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4108 | spin_unlock(&pers_lock); |
d9d166c2 N |
4109 | if (mddev->level != LEVEL_NONE) |
4110 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4111 | mddev->level); | |
4112 | else | |
4113 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4114 | mddev->clevel); | |
1da177e4 LT |
4115 | return -EINVAL; |
4116 | } | |
2604b703 | 4117 | mddev->pers = pers; |
1da177e4 | 4118 | spin_unlock(&pers_lock); |
34817e8c N |
4119 | if (mddev->level != pers->level) { |
4120 | mddev->level = pers->level; | |
4121 | mddev->new_level = pers->level; | |
4122 | } | |
d9d166c2 | 4123 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4124 | |
f6705578 | 4125 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4126 | pers->start_reshape == NULL) { |
f6705578 N |
4127 | /* This personality cannot handle reshaping... */ |
4128 | mddev->pers = NULL; | |
4129 | module_put(pers->owner); | |
4130 | return -EINVAL; | |
4131 | } | |
4132 | ||
7dd5e7c3 N |
4133 | if (pers->sync_request) { |
4134 | /* Warn if this is a potentially silly | |
4135 | * configuration. | |
4136 | */ | |
4137 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4138 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4139 | int warned = 0; |
159ec1fc CR |
4140 | |
4141 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4142 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4143 | if (rdev < rdev2 && |
4144 | rdev->bdev->bd_contains == | |
4145 | rdev2->bdev->bd_contains) { | |
4146 | printk(KERN_WARNING | |
4147 | "%s: WARNING: %s appears to be" | |
4148 | " on the same physical disk as" | |
4149 | " %s.\n", | |
4150 | mdname(mddev), | |
4151 | bdevname(rdev->bdev,b), | |
4152 | bdevname(rdev2->bdev,b2)); | |
4153 | warned = 1; | |
4154 | } | |
4155 | } | |
159ec1fc | 4156 | |
7dd5e7c3 N |
4157 | if (warned) |
4158 | printk(KERN_WARNING | |
4159 | "True protection against single-disk" | |
4160 | " failure might be compromised.\n"); | |
4161 | } | |
4162 | ||
657390d2 | 4163 | mddev->recovery = 0; |
58c0fed4 AN |
4164 | /* may be over-ridden by personality */ |
4165 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4166 | ||
a9701a30 | 4167 | mddev->barriers_work = 1; |
6ff8d8ec | 4168 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4169 | |
f91de92e N |
4170 | if (start_readonly) |
4171 | mddev->ro = 2; /* read-only, but switch on first write */ | |
4172 | ||
b15c2e57 | 4173 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4174 | if (err) |
4175 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4176 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4177 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4178 | " but 'external_size' not in effect?\n", __func__); | |
4179 | printk(KERN_ERR | |
4180 | "md: invalid array_size %llu > default size %llu\n", | |
4181 | (unsigned long long)mddev->array_sectors / 2, | |
4182 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4183 | err = -EINVAL; | |
4184 | mddev->pers->stop(mddev); | |
4185 | } | |
4186 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4187 | err = bitmap_create(mddev); |
4188 | if (err) { | |
4189 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4190 | mdname(mddev), err); | |
4191 | mddev->pers->stop(mddev); | |
4192 | } | |
4193 | } | |
1da177e4 | 4194 | if (err) { |
1da177e4 LT |
4195 | module_put(mddev->pers->owner); |
4196 | mddev->pers = NULL; | |
32a7627c N |
4197 | bitmap_destroy(mddev); |
4198 | return err; | |
1da177e4 | 4199 | } |
5e55e2f5 N |
4200 | if (mddev->pers->sync_request) { |
4201 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
4202 | printk(KERN_WARNING | |
4203 | "md: cannot register extra attributes for %s\n", | |
4204 | mdname(mddev)); | |
0c3573f1 | 4205 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4206 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4207 | mddev->ro = 0; |
4208 | ||
1da177e4 LT |
4209 | atomic_set(&mddev->writes_pending,0); |
4210 | mddev->safemode = 0; | |
4211 | mddev->safemode_timer.function = md_safemode_timeout; | |
4212 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4213 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4214 | mddev->in_sync = 1; |
86e6ffdd | 4215 | |
159ec1fc | 4216 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4217 | if (rdev->raid_disk >= 0) { |
4218 | char nm[20]; | |
4219 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
4220 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
4221 | printk("md: cannot register %s for %s\n", | |
4222 | nm, mdname(mddev)); | |
86e6ffdd | 4223 | } |
1da177e4 LT |
4224 | |
4225 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4226 | ||
850b2b42 N |
4227 | if (mddev->flags) |
4228 | md_update_sb(mddev, 0); | |
1da177e4 | 4229 | |
f233ea5c | 4230 | set_capacity(disk, mddev->array_sectors); |
1da177e4 | 4231 | |
5fd6c1dc N |
4232 | /* If there is a partially-recovered drive we need to |
4233 | * start recovery here. If we leave it to md_check_recovery, | |
4234 | * it will remove the drives and not do the right thing | |
4235 | */ | |
0b8c9de0 | 4236 | if (mddev->degraded && !mddev->sync_thread) { |
5fd6c1dc | 4237 | int spares = 0; |
159ec1fc | 4238 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
4239 | if (rdev->raid_disk >= 0 && |
4240 | !test_bit(In_sync, &rdev->flags) && | |
4241 | !test_bit(Faulty, &rdev->flags)) | |
4242 | /* complete an interrupted recovery */ | |
4243 | spares++; | |
4244 | if (spares && mddev->pers->sync_request) { | |
4245 | mddev->recovery = 0; | |
4246 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4247 | mddev->sync_thread = md_register_thread(md_do_sync, | |
4248 | mddev, | |
0da3c619 | 4249 | "resync"); |
5fd6c1dc N |
4250 | if (!mddev->sync_thread) { |
4251 | printk(KERN_ERR "%s: could not start resync" | |
4252 | " thread...\n", | |
4253 | mdname(mddev)); | |
4254 | /* leave the spares where they are, it shouldn't hurt */ | |
4255 | mddev->recovery = 0; | |
0b8c9de0 | 4256 | } |
5fd6c1dc N |
4257 | } |
4258 | } | |
0b8c9de0 N |
4259 | md_wakeup_thread(mddev->thread); |
4260 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4261 | |
449aad3e | 4262 | revalidate_disk(mddev->gendisk); |
44ce6294 | 4263 | mddev->changed = 1; |
d7603b7e | 4264 | md_new_event(mddev); |
b62b7590 | 4265 | sysfs_notify_dirent(mddev->sysfs_state); |
0c3573f1 N |
4266 | if (mddev->sysfs_action) |
4267 | sysfs_notify_dirent(mddev->sysfs_action); | |
a99ac971 | 4268 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
ed9e1982 | 4269 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
1da177e4 LT |
4270 | return 0; |
4271 | } | |
4272 | ||
4273 | static int restart_array(mddev_t *mddev) | |
4274 | { | |
4275 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4276 | |
80fab1d7 | 4277 | /* Complain if it has no devices */ |
1da177e4 | 4278 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4279 | return -ENXIO; |
4280 | if (!mddev->pers) | |
4281 | return -EINVAL; | |
4282 | if (!mddev->ro) | |
4283 | return -EBUSY; | |
4284 | mddev->safemode = 0; | |
4285 | mddev->ro = 0; | |
4286 | set_disk_ro(disk, 0); | |
4287 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4288 | mdname(mddev)); | |
4289 | /* Kick recovery or resync if necessary */ | |
4290 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4291 | md_wakeup_thread(mddev->thread); | |
4292 | md_wakeup_thread(mddev->sync_thread); | |
b62b7590 | 4293 | sysfs_notify_dirent(mddev->sysfs_state); |
80fab1d7 | 4294 | return 0; |
1da177e4 LT |
4295 | } |
4296 | ||
acc55e22 N |
4297 | /* similar to deny_write_access, but accounts for our holding a reference |
4298 | * to the file ourselves */ | |
4299 | static int deny_bitmap_write_access(struct file * file) | |
4300 | { | |
4301 | struct inode *inode = file->f_mapping->host; | |
4302 | ||
4303 | spin_lock(&inode->i_lock); | |
4304 | if (atomic_read(&inode->i_writecount) > 1) { | |
4305 | spin_unlock(&inode->i_lock); | |
4306 | return -ETXTBSY; | |
4307 | } | |
4308 | atomic_set(&inode->i_writecount, -1); | |
4309 | spin_unlock(&inode->i_lock); | |
4310 | ||
4311 | return 0; | |
4312 | } | |
4313 | ||
4314 | static void restore_bitmap_write_access(struct file *file) | |
4315 | { | |
4316 | struct inode *inode = file->f_mapping->host; | |
4317 | ||
4318 | spin_lock(&inode->i_lock); | |
4319 | atomic_set(&inode->i_writecount, 1); | |
4320 | spin_unlock(&inode->i_lock); | |
4321 | } | |
4322 | ||
9e653b63 N |
4323 | /* mode: |
4324 | * 0 - completely stop and dis-assemble array | |
4325 | * 1 - switch to readonly | |
4326 | * 2 - stop but do not disassemble array | |
4327 | */ | |
df5b20cf | 4328 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 LT |
4329 | { |
4330 | int err = 0; | |
4331 | struct gendisk *disk = mddev->gendisk; | |
c4647292 | 4332 | mdk_rdev_t *rdev; |
1da177e4 | 4333 | |
c8c00a69 | 4334 | mutex_lock(&mddev->open_mutex); |
f2ea68cf | 4335 | if (atomic_read(&mddev->openers) > is_open) { |
df5b20cf | 4336 | printk("md: %s still in use.\n",mdname(mddev)); |
c8c00a69 N |
4337 | err = -EBUSY; |
4338 | } else if (mddev->pers) { | |
1da177e4 LT |
4339 | |
4340 | if (mddev->sync_thread) { | |
5fd6c1dc | 4341 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 LT |
4342 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4343 | md_unregister_thread(mddev->sync_thread); | |
4344 | mddev->sync_thread = NULL; | |
4345 | } | |
4346 | ||
4347 | del_timer_sync(&mddev->safemode_timer); | |
4348 | ||
9e653b63 N |
4349 | switch(mode) { |
4350 | case 1: /* readonly */ | |
1da177e4 | 4351 | err = -ENXIO; |
f91de92e | 4352 | if (mddev->ro==1) |
1da177e4 LT |
4353 | goto out; |
4354 | mddev->ro = 1; | |
9e653b63 N |
4355 | break; |
4356 | case 0: /* disassemble */ | |
4357 | case 2: /* stop */ | |
6b8b3e8a | 4358 | bitmap_flush(mddev); |
a9701a30 | 4359 | md_super_wait(mddev); |
1da177e4 LT |
4360 | if (mddev->ro) |
4361 | set_disk_ro(disk, 0); | |
409c57f3 | 4362 | |
1da177e4 | 4363 | mddev->pers->stop(mddev); |
d1b5380c N |
4364 | mddev->queue->merge_bvec_fn = NULL; |
4365 | mddev->queue->unplug_fn = NULL; | |
041ae52e | 4366 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4367 | module_put(mddev->pers->owner); |
5fd3a17e DW |
4368 | if (mddev->pers->sync_request) |
4369 | mddev->private = &md_redundancy_group; | |
1da177e4 | 4370 | mddev->pers = NULL; |
4f54b0e9 | 4371 | /* tell userspace to handle 'inactive' */ |
b62b7590 | 4372 | sysfs_notify_dirent(mddev->sysfs_state); |
0d4ca600 | 4373 | |
c4647292 N |
4374 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4375 | if (rdev->raid_disk >= 0) { | |
4376 | char nm[20]; | |
4377 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4378 | sysfs_remove_link(&mddev->kobj, nm); | |
4379 | } | |
4380 | ||
0d4ca600 | 4381 | set_capacity(disk, 0); |
44ce6294 | 4382 | mddev->changed = 1; |
0d4ca600 | 4383 | |
1da177e4 LT |
4384 | if (mddev->ro) |
4385 | mddev->ro = 0; | |
4386 | } | |
850b2b42 | 4387 | if (!mddev->in_sync || mddev->flags) { |
1da177e4 LT |
4388 | /* mark array as shutdown cleanly */ |
4389 | mddev->in_sync = 1; | |
850b2b42 | 4390 | md_update_sb(mddev, 1); |
1da177e4 | 4391 | } |
9e653b63 | 4392 | if (mode == 1) |
1da177e4 | 4393 | set_disk_ro(disk, 1); |
5fd6c1dc | 4394 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
80ffb3cc | 4395 | err = 0; |
1da177e4 | 4396 | } |
c8c00a69 N |
4397 | out: |
4398 | mutex_unlock(&mddev->open_mutex); | |
4399 | if (err) | |
4400 | return err; | |
1da177e4 LT |
4401 | /* |
4402 | * Free resources if final stop | |
4403 | */ | |
9e653b63 | 4404 | if (mode == 0) { |
0d4ca600 | 4405 | |
1da177e4 LT |
4406 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4407 | ||
978f946b N |
4408 | bitmap_destroy(mddev); |
4409 | if (mddev->bitmap_file) { | |
acc55e22 | 4410 | restore_bitmap_write_access(mddev->bitmap_file); |
978f946b N |
4411 | fput(mddev->bitmap_file); |
4412 | mddev->bitmap_file = NULL; | |
4413 | } | |
4414 | mddev->bitmap_offset = 0; | |
4415 | ||
177a99b2 | 4416 | /* make sure all md_delayed_delete calls have finished */ |
5792a285 N |
4417 | flush_scheduled_work(); |
4418 | ||
1da177e4 LT |
4419 | export_array(mddev); |
4420 | ||
f233ea5c | 4421 | mddev->array_sectors = 0; |
b522adcd | 4422 | mddev->external_size = 0; |
58c0fed4 | 4423 | mddev->dev_sectors = 0; |
9e653b63 | 4424 | mddev->raid_disks = 0; |
a94213b1 | 4425 | mddev->recovery_cp = 0; |
5e96ee65 | 4426 | mddev->resync_min = 0; |
c6207277 | 4427 | mddev->resync_max = MaxSector; |
08a02ecd | 4428 | mddev->reshape_position = MaxSector; |
e691063a | 4429 | mddev->external = 0; |
1ec4a939 | 4430 | mddev->persistent = 0; |
d897dbf9 N |
4431 | mddev->level = LEVEL_NONE; |
4432 | mddev->clevel[0] = 0; | |
4433 | mddev->flags = 0; | |
4434 | mddev->ro = 0; | |
4435 | mddev->metadata_type[0] = 0; | |
9d8f0363 | 4436 | mddev->chunk_sectors = 0; |
d897dbf9 N |
4437 | mddev->ctime = mddev->utime = 0; |
4438 | mddev->layout = 0; | |
4439 | mddev->max_disks = 0; | |
4440 | mddev->events = 0; | |
4441 | mddev->delta_disks = 0; | |
4442 | mddev->new_level = LEVEL_NONE; | |
4443 | mddev->new_layout = 0; | |
664e7c41 | 4444 | mddev->new_chunk_sectors = 0; |
d897dbf9 N |
4445 | mddev->curr_resync = 0; |
4446 | mddev->resync_mismatches = 0; | |
4447 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4448 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4449 | mddev->recovery = 0; | |
4450 | mddev->in_sync = 0; | |
4451 | mddev->changed = 0; | |
4452 | mddev->degraded = 0; | |
4453 | mddev->barriers_work = 0; | |
4454 | mddev->safemode = 0; | |
934d9c23 | 4455 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4456 | if (mddev->hold_active == UNTIL_STOP) |
4457 | mddev->hold_active = 0; | |
9e653b63 | 4458 | |
a8a55c38 | 4459 | } else if (mddev->pers) |
1da177e4 LT |
4460 | printk(KERN_INFO "md: %s switched to read-only mode.\n", |
4461 | mdname(mddev)); | |
4462 | err = 0; | |
3f9d99c1 | 4463 | blk_integrity_unregister(disk); |
d7603b7e | 4464 | md_new_event(mddev); |
b62b7590 | 4465 | sysfs_notify_dirent(mddev->sysfs_state); |
1da177e4 LT |
4466 | return err; |
4467 | } | |
4468 | ||
fdee8ae4 | 4469 | #ifndef MODULE |
1da177e4 LT |
4470 | static void autorun_array(mddev_t *mddev) |
4471 | { | |
4472 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4473 | int err; |
4474 | ||
a757e64c | 4475 | if (list_empty(&mddev->disks)) |
1da177e4 | 4476 | return; |
1da177e4 LT |
4477 | |
4478 | printk(KERN_INFO "md: running: "); | |
4479 | ||
159ec1fc | 4480 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4481 | char b[BDEVNAME_SIZE]; |
4482 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4483 | } | |
4484 | printk("\n"); | |
4485 | ||
d710e138 | 4486 | err = do_md_run(mddev); |
1da177e4 LT |
4487 | if (err) { |
4488 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4489 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4490 | } |
4491 | } | |
4492 | ||
4493 | /* | |
4494 | * lets try to run arrays based on all disks that have arrived | |
4495 | * until now. (those are in pending_raid_disks) | |
4496 | * | |
4497 | * the method: pick the first pending disk, collect all disks with | |
4498 | * the same UUID, remove all from the pending list and put them into | |
4499 | * the 'same_array' list. Then order this list based on superblock | |
4500 | * update time (freshest comes first), kick out 'old' disks and | |
4501 | * compare superblocks. If everything's fine then run it. | |
4502 | * | |
4503 | * If "unit" is allocated, then bump its reference count | |
4504 | */ | |
4505 | static void autorun_devices(int part) | |
4506 | { | |
159ec1fc | 4507 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4508 | mddev_t *mddev; |
4509 | char b[BDEVNAME_SIZE]; | |
4510 | ||
4511 | printk(KERN_INFO "md: autorun ...\n"); | |
4512 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4513 | int unit; |
1da177e4 | 4514 | dev_t dev; |
ad01c9e3 | 4515 | LIST_HEAD(candidates); |
1da177e4 LT |
4516 | rdev0 = list_entry(pending_raid_disks.next, |
4517 | mdk_rdev_t, same_set); | |
4518 | ||
4519 | printk(KERN_INFO "md: considering %s ...\n", | |
4520 | bdevname(rdev0->bdev,b)); | |
4521 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4522 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4523 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4524 | printk(KERN_INFO "md: adding %s ...\n", | |
4525 | bdevname(rdev->bdev,b)); | |
4526 | list_move(&rdev->same_set, &candidates); | |
4527 | } | |
4528 | /* | |
4529 | * now we have a set of devices, with all of them having | |
4530 | * mostly sane superblocks. It's time to allocate the | |
4531 | * mddev. | |
4532 | */ | |
e8703fe1 N |
4533 | if (part) { |
4534 | dev = MKDEV(mdp_major, | |
4535 | rdev0->preferred_minor << MdpMinorShift); | |
4536 | unit = MINOR(dev) >> MdpMinorShift; | |
4537 | } else { | |
4538 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4539 | unit = MINOR(dev); | |
4540 | } | |
4541 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4542 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4543 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4544 | break; | |
4545 | } | |
1da177e4 LT |
4546 | |
4547 | md_probe(dev, NULL, NULL); | |
4548 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4549 | if (!mddev || !mddev->gendisk) { |
4550 | if (mddev) | |
4551 | mddev_put(mddev); | |
4552 | printk(KERN_ERR | |
1da177e4 LT |
4553 | "md: cannot allocate memory for md drive.\n"); |
4554 | break; | |
4555 | } | |
4556 | if (mddev_lock(mddev)) | |
4557 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4558 | mdname(mddev)); | |
4559 | else if (mddev->raid_disks || mddev->major_version | |
4560 | || !list_empty(&mddev->disks)) { | |
4561 | printk(KERN_WARNING | |
4562 | "md: %s already running, cannot run %s\n", | |
4563 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4564 | mddev_unlock(mddev); | |
4565 | } else { | |
4566 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4567 | mddev->persistent = 1; |
159ec1fc | 4568 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4569 | list_del_init(&rdev->same_set); |
4570 | if (bind_rdev_to_array(rdev, mddev)) | |
4571 | export_rdev(rdev); | |
4572 | } | |
4573 | autorun_array(mddev); | |
4574 | mddev_unlock(mddev); | |
4575 | } | |
4576 | /* on success, candidates will be empty, on error | |
4577 | * it won't... | |
4578 | */ | |
159ec1fc | 4579 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4580 | list_del_init(&rdev->same_set); |
1da177e4 | 4581 | export_rdev(rdev); |
4b80991c | 4582 | } |
1da177e4 LT |
4583 | mddev_put(mddev); |
4584 | } | |
4585 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4586 | } | |
fdee8ae4 | 4587 | #endif /* !MODULE */ |
1da177e4 | 4588 | |
1da177e4 LT |
4589 | static int get_version(void __user * arg) |
4590 | { | |
4591 | mdu_version_t ver; | |
4592 | ||
4593 | ver.major = MD_MAJOR_VERSION; | |
4594 | ver.minor = MD_MINOR_VERSION; | |
4595 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4596 | ||
4597 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4598 | return -EFAULT; | |
4599 | ||
4600 | return 0; | |
4601 | } | |
4602 | ||
4603 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4604 | { | |
4605 | mdu_array_info_t info; | |
a9f326eb | 4606 | int nr,working,insync,failed,spare; |
1da177e4 | 4607 | mdk_rdev_t *rdev; |
1da177e4 | 4608 | |
a9f326eb | 4609 | nr=working=insync=failed=spare=0; |
159ec1fc | 4610 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4611 | nr++; |
b2d444d7 | 4612 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4613 | failed++; |
4614 | else { | |
4615 | working++; | |
b2d444d7 | 4616 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 4617 | insync++; |
1da177e4 LT |
4618 | else |
4619 | spare++; | |
4620 | } | |
4621 | } | |
4622 | ||
4623 | info.major_version = mddev->major_version; | |
4624 | info.minor_version = mddev->minor_version; | |
4625 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
4626 | info.ctime = mddev->ctime; | |
4627 | info.level = mddev->level; | |
58c0fed4 AN |
4628 | info.size = mddev->dev_sectors / 2; |
4629 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 4630 | info.size = -1; |
1da177e4 LT |
4631 | info.nr_disks = nr; |
4632 | info.raid_disks = mddev->raid_disks; | |
4633 | info.md_minor = mddev->md_minor; | |
4634 | info.not_persistent= !mddev->persistent; | |
4635 | ||
4636 | info.utime = mddev->utime; | |
4637 | info.state = 0; | |
4638 | if (mddev->in_sync) | |
4639 | info.state = (1<<MD_SB_CLEAN); | |
36fa3063 N |
4640 | if (mddev->bitmap && mddev->bitmap_offset) |
4641 | info.state = (1<<MD_SB_BITMAP_PRESENT); | |
a9f326eb | 4642 | info.active_disks = insync; |
1da177e4 LT |
4643 | info.working_disks = working; |
4644 | info.failed_disks = failed; | |
4645 | info.spare_disks = spare; | |
4646 | ||
4647 | info.layout = mddev->layout; | |
9d8f0363 | 4648 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
4649 | |
4650 | if (copy_to_user(arg, &info, sizeof(info))) | |
4651 | return -EFAULT; | |
4652 | ||
4653 | return 0; | |
4654 | } | |
4655 | ||
87162a28 | 4656 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
4657 | { |
4658 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
4659 | char *ptr, *buf = NULL; | |
4660 | int err = -ENOMEM; | |
4661 | ||
b5470dc5 DW |
4662 | if (md_allow_write(mddev)) |
4663 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
4664 | else | |
4665 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 4666 | |
32a7627c N |
4667 | if (!file) |
4668 | goto out; | |
4669 | ||
4670 | /* bitmap disabled, zero the first byte and copy out */ | |
4671 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
4672 | file->pathname[0] = '\0'; | |
4673 | goto copy_out; | |
4674 | } | |
4675 | ||
4676 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
4677 | if (!buf) | |
4678 | goto out; | |
4679 | ||
6bcfd601 CH |
4680 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
4681 | if (IS_ERR(ptr)) | |
32a7627c N |
4682 | goto out; |
4683 | ||
4684 | strcpy(file->pathname, ptr); | |
4685 | ||
4686 | copy_out: | |
4687 | err = 0; | |
4688 | if (copy_to_user(arg, file, sizeof(*file))) | |
4689 | err = -EFAULT; | |
4690 | out: | |
4691 | kfree(buf); | |
4692 | kfree(file); | |
4693 | return err; | |
4694 | } | |
4695 | ||
1da177e4 LT |
4696 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
4697 | { | |
4698 | mdu_disk_info_t info; | |
1da177e4 LT |
4699 | mdk_rdev_t *rdev; |
4700 | ||
4701 | if (copy_from_user(&info, arg, sizeof(info))) | |
4702 | return -EFAULT; | |
4703 | ||
26ef379f | 4704 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
4705 | if (rdev) { |
4706 | info.major = MAJOR(rdev->bdev->bd_dev); | |
4707 | info.minor = MINOR(rdev->bdev->bd_dev); | |
4708 | info.raid_disk = rdev->raid_disk; | |
4709 | info.state = 0; | |
b2d444d7 | 4710 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4711 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 4712 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4713 | info.state |= (1<<MD_DISK_ACTIVE); |
4714 | info.state |= (1<<MD_DISK_SYNC); | |
4715 | } | |
8ddf9efe N |
4716 | if (test_bit(WriteMostly, &rdev->flags)) |
4717 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
4718 | } else { |
4719 | info.major = info.minor = 0; | |
4720 | info.raid_disk = -1; | |
4721 | info.state = (1<<MD_DISK_REMOVED); | |
4722 | } | |
4723 | ||
4724 | if (copy_to_user(arg, &info, sizeof(info))) | |
4725 | return -EFAULT; | |
4726 | ||
4727 | return 0; | |
4728 | } | |
4729 | ||
4730 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
4731 | { | |
4732 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4733 | mdk_rdev_t *rdev; | |
4734 | dev_t dev = MKDEV(info->major,info->minor); | |
4735 | ||
4736 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
4737 | return -EOVERFLOW; | |
4738 | ||
4739 | if (!mddev->raid_disks) { | |
4740 | int err; | |
4741 | /* expecting a device which has a superblock */ | |
4742 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
4743 | if (IS_ERR(rdev)) { | |
4744 | printk(KERN_WARNING | |
4745 | "md: md_import_device returned %ld\n", | |
4746 | PTR_ERR(rdev)); | |
4747 | return PTR_ERR(rdev); | |
4748 | } | |
4749 | if (!list_empty(&mddev->disks)) { | |
4750 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
4751 | mdk_rdev_t, same_set); | |
a9f326eb | 4752 | err = super_types[mddev->major_version] |
1da177e4 LT |
4753 | .load_super(rdev, rdev0, mddev->minor_version); |
4754 | if (err < 0) { | |
4755 | printk(KERN_WARNING | |
4756 | "md: %s has different UUID to %s\n", | |
4757 | bdevname(rdev->bdev,b), | |
4758 | bdevname(rdev0->bdev,b2)); | |
4759 | export_rdev(rdev); | |
4760 | return -EINVAL; | |
4761 | } | |
4762 | } | |
4763 | err = bind_rdev_to_array(rdev, mddev); | |
4764 | if (err) | |
4765 | export_rdev(rdev); | |
4766 | return err; | |
4767 | } | |
4768 | ||
4769 | /* | |
4770 | * add_new_disk can be used once the array is assembled | |
4771 | * to add "hot spares". They must already have a superblock | |
4772 | * written | |
4773 | */ | |
4774 | if (mddev->pers) { | |
4775 | int err; | |
4776 | if (!mddev->pers->hot_add_disk) { | |
4777 | printk(KERN_WARNING | |
4778 | "%s: personality does not support diskops!\n", | |
4779 | mdname(mddev)); | |
4780 | return -EINVAL; | |
4781 | } | |
7b1e35f6 N |
4782 | if (mddev->persistent) |
4783 | rdev = md_import_device(dev, mddev->major_version, | |
4784 | mddev->minor_version); | |
4785 | else | |
4786 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
4787 | if (IS_ERR(rdev)) { |
4788 | printk(KERN_WARNING | |
4789 | "md: md_import_device returned %ld\n", | |
4790 | PTR_ERR(rdev)); | |
4791 | return PTR_ERR(rdev); | |
4792 | } | |
41158c7e N |
4793 | /* set save_raid_disk if appropriate */ |
4794 | if (!mddev->persistent) { | |
4795 | if (info->state & (1<<MD_DISK_SYNC) && | |
4796 | info->raid_disk < mddev->raid_disks) | |
4797 | rdev->raid_disk = info->raid_disk; | |
4798 | else | |
4799 | rdev->raid_disk = -1; | |
4800 | } else | |
4801 | super_types[mddev->major_version]. | |
4802 | validate_super(mddev, rdev); | |
4803 | rdev->saved_raid_disk = rdev->raid_disk; | |
4804 | ||
b2d444d7 | 4805 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
4806 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4807 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
4808 | else |
4809 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 4810 | |
1da177e4 LT |
4811 | rdev->raid_disk = -1; |
4812 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
4813 | if (!err && !mddev->pers->hot_remove_disk) { |
4814 | /* If there is hot_add_disk but no hot_remove_disk | |
4815 | * then added disks for geometry changes, | |
4816 | * and should be added immediately. | |
4817 | */ | |
4818 | super_types[mddev->major_version]. | |
4819 | validate_super(mddev, rdev); | |
4820 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
4821 | if (err) | |
4822 | unbind_rdev_from_array(rdev); | |
4823 | } | |
1da177e4 LT |
4824 | if (err) |
4825 | export_rdev(rdev); | |
52664732 | 4826 | else |
3c0ee63a | 4827 | sysfs_notify_dirent(rdev->sysfs_state); |
c361777f | 4828 | |
17571284 | 4829 | md_update_sb(mddev, 1); |
72a23c21 NB |
4830 | if (mddev->degraded) |
4831 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 4832 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 4833 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4834 | return err; |
4835 | } | |
4836 | ||
4837 | /* otherwise, add_new_disk is only allowed | |
4838 | * for major_version==0 superblocks | |
4839 | */ | |
4840 | if (mddev->major_version != 0) { | |
4841 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
4842 | mdname(mddev)); | |
4843 | return -EINVAL; | |
4844 | } | |
4845 | ||
4846 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
4847 | int err; | |
d710e138 | 4848 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
4849 | if (IS_ERR(rdev)) { |
4850 | printk(KERN_WARNING | |
4851 | "md: error, md_import_device() returned %ld\n", | |
4852 | PTR_ERR(rdev)); | |
4853 | return PTR_ERR(rdev); | |
4854 | } | |
4855 | rdev->desc_nr = info->number; | |
4856 | if (info->raid_disk < mddev->raid_disks) | |
4857 | rdev->raid_disk = info->raid_disk; | |
4858 | else | |
4859 | rdev->raid_disk = -1; | |
4860 | ||
1da177e4 | 4861 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
4862 | if (info->state & (1<<MD_DISK_SYNC)) |
4863 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 4864 | |
8ddf9efe N |
4865 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4866 | set_bit(WriteMostly, &rdev->flags); | |
4867 | ||
1da177e4 LT |
4868 | if (!mddev->persistent) { |
4869 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
0f420358 | 4870 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 4871 | } else |
0f420358 | 4872 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
8190e754 | 4873 | rdev->sectors = rdev->sb_start; |
1da177e4 | 4874 | |
2bf071bf N |
4875 | err = bind_rdev_to_array(rdev, mddev); |
4876 | if (err) { | |
4877 | export_rdev(rdev); | |
4878 | return err; | |
4879 | } | |
1da177e4 LT |
4880 | } |
4881 | ||
4882 | return 0; | |
4883 | } | |
4884 | ||
4885 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
4886 | { | |
4887 | char b[BDEVNAME_SIZE]; | |
4888 | mdk_rdev_t *rdev; | |
4889 | ||
1da177e4 LT |
4890 | rdev = find_rdev(mddev, dev); |
4891 | if (!rdev) | |
4892 | return -ENXIO; | |
4893 | ||
4894 | if (rdev->raid_disk >= 0) | |
4895 | goto busy; | |
4896 | ||
4897 | kick_rdev_from_array(rdev); | |
850b2b42 | 4898 | md_update_sb(mddev, 1); |
d7603b7e | 4899 | md_new_event(mddev); |
1da177e4 LT |
4900 | |
4901 | return 0; | |
4902 | busy: | |
fdefa4d8 | 4903 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
4904 | bdevname(rdev->bdev,b), mdname(mddev)); |
4905 | return -EBUSY; | |
4906 | } | |
4907 | ||
4908 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
4909 | { | |
4910 | char b[BDEVNAME_SIZE]; | |
4911 | int err; | |
1da177e4 LT |
4912 | mdk_rdev_t *rdev; |
4913 | ||
4914 | if (!mddev->pers) | |
4915 | return -ENODEV; | |
4916 | ||
4917 | if (mddev->major_version != 0) { | |
4918 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
4919 | " version-0 superblocks.\n", | |
4920 | mdname(mddev)); | |
4921 | return -EINVAL; | |
4922 | } | |
4923 | if (!mddev->pers->hot_add_disk) { | |
4924 | printk(KERN_WARNING | |
4925 | "%s: personality does not support diskops!\n", | |
4926 | mdname(mddev)); | |
4927 | return -EINVAL; | |
4928 | } | |
4929 | ||
d710e138 | 4930 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
4931 | if (IS_ERR(rdev)) { |
4932 | printk(KERN_WARNING | |
4933 | "md: error, md_import_device() returned %ld\n", | |
4934 | PTR_ERR(rdev)); | |
4935 | return -EINVAL; | |
4936 | } | |
4937 | ||
4938 | if (mddev->persistent) | |
0f420358 | 4939 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 4940 | else |
0f420358 | 4941 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 4942 | |
8190e754 | 4943 | rdev->sectors = rdev->sb_start; |
1da177e4 | 4944 | |
b2d444d7 | 4945 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
4946 | printk(KERN_WARNING |
4947 | "md: can not hot-add faulty %s disk to %s!\n", | |
4948 | bdevname(rdev->bdev,b), mdname(mddev)); | |
4949 | err = -EINVAL; | |
4950 | goto abort_export; | |
4951 | } | |
b2d444d7 | 4952 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4953 | rdev->desc_nr = -1; |
5842730d | 4954 | rdev->saved_raid_disk = -1; |
2bf071bf N |
4955 | err = bind_rdev_to_array(rdev, mddev); |
4956 | if (err) | |
4957 | goto abort_export; | |
1da177e4 LT |
4958 | |
4959 | /* | |
4960 | * The rest should better be atomic, we can have disk failures | |
4961 | * noticed in interrupt contexts ... | |
4962 | */ | |
4963 | ||
1da177e4 LT |
4964 | rdev->raid_disk = -1; |
4965 | ||
850b2b42 | 4966 | md_update_sb(mddev, 1); |
1da177e4 LT |
4967 | |
4968 | /* | |
4969 | * Kick recovery, maybe this spare has to be added to the | |
4970 | * array immediately. | |
4971 | */ | |
4972 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4973 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 4974 | md_new_event(mddev); |
1da177e4 LT |
4975 | return 0; |
4976 | ||
1da177e4 LT |
4977 | abort_export: |
4978 | export_rdev(rdev); | |
4979 | return err; | |
4980 | } | |
4981 | ||
32a7627c N |
4982 | static int set_bitmap_file(mddev_t *mddev, int fd) |
4983 | { | |
4984 | int err; | |
4985 | ||
36fa3063 N |
4986 | if (mddev->pers) { |
4987 | if (!mddev->pers->quiesce) | |
4988 | return -EBUSY; | |
4989 | if (mddev->recovery || mddev->sync_thread) | |
4990 | return -EBUSY; | |
4991 | /* we should be able to change the bitmap.. */ | |
4992 | } | |
32a7627c | 4993 | |
32a7627c | 4994 | |
36fa3063 N |
4995 | if (fd >= 0) { |
4996 | if (mddev->bitmap) | |
4997 | return -EEXIST; /* cannot add when bitmap is present */ | |
4998 | mddev->bitmap_file = fget(fd); | |
32a7627c | 4999 | |
36fa3063 N |
5000 | if (mddev->bitmap_file == NULL) { |
5001 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", | |
5002 | mdname(mddev)); | |
5003 | return -EBADF; | |
5004 | } | |
5005 | ||
5006 | err = deny_bitmap_write_access(mddev->bitmap_file); | |
5007 | if (err) { | |
5008 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5009 | mdname(mddev)); | |
5010 | fput(mddev->bitmap_file); | |
5011 | mddev->bitmap_file = NULL; | |
5012 | return err; | |
5013 | } | |
a654b9d8 | 5014 | mddev->bitmap_offset = 0; /* file overrides offset */ |
36fa3063 N |
5015 | } else if (mddev->bitmap == NULL) |
5016 | return -ENOENT; /* cannot remove what isn't there */ | |
5017 | err = 0; | |
5018 | if (mddev->pers) { | |
5019 | mddev->pers->quiesce(mddev, 1); | |
5020 | if (fd >= 0) | |
5021 | err = bitmap_create(mddev); | |
d7375ab3 | 5022 | if (fd < 0 || err) { |
36fa3063 | 5023 | bitmap_destroy(mddev); |
d7375ab3 N |
5024 | fd = -1; /* make sure to put the file */ |
5025 | } | |
36fa3063 | 5026 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5027 | } |
5028 | if (fd < 0) { | |
acc55e22 N |
5029 | if (mddev->bitmap_file) { |
5030 | restore_bitmap_write_access(mddev->bitmap_file); | |
36fa3063 | 5031 | fput(mddev->bitmap_file); |
acc55e22 | 5032 | } |
36fa3063 N |
5033 | mddev->bitmap_file = NULL; |
5034 | } | |
5035 | ||
32a7627c N |
5036 | return err; |
5037 | } | |
5038 | ||
1da177e4 LT |
5039 | /* |
5040 | * set_array_info is used two different ways | |
5041 | * The original usage is when creating a new array. | |
5042 | * In this usage, raid_disks is > 0 and it together with | |
5043 | * level, size, not_persistent,layout,chunksize determine the | |
5044 | * shape of the array. | |
5045 | * This will always create an array with a type-0.90.0 superblock. | |
5046 | * The newer usage is when assembling an array. | |
5047 | * In this case raid_disks will be 0, and the major_version field is | |
5048 | * use to determine which style super-blocks are to be found on the devices. | |
5049 | * The minor and patch _version numbers are also kept incase the | |
5050 | * super_block handler wishes to interpret them. | |
5051 | */ | |
5052 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5053 | { | |
5054 | ||
5055 | if (info->raid_disks == 0) { | |
5056 | /* just setting version number for superblock loading */ | |
5057 | if (info->major_version < 0 || | |
50511da3 | 5058 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5059 | super_types[info->major_version].name == NULL) { |
5060 | /* maybe try to auto-load a module? */ | |
5061 | printk(KERN_INFO | |
5062 | "md: superblock version %d not known\n", | |
5063 | info->major_version); | |
5064 | return -EINVAL; | |
5065 | } | |
5066 | mddev->major_version = info->major_version; | |
5067 | mddev->minor_version = info->minor_version; | |
5068 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5069 | mddev->persistent = !info->not_persistent; |
1da177e4 LT |
5070 | return 0; |
5071 | } | |
5072 | mddev->major_version = MD_MAJOR_VERSION; | |
5073 | mddev->minor_version = MD_MINOR_VERSION; | |
5074 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5075 | mddev->ctime = get_seconds(); | |
5076 | ||
5077 | mddev->level = info->level; | |
17115e03 | 5078 | mddev->clevel[0] = 0; |
58c0fed4 | 5079 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5080 | mddev->raid_disks = info->raid_disks; |
5081 | /* don't set md_minor, it is determined by which /dev/md* was | |
5082 | * openned | |
5083 | */ | |
5084 | if (info->state & (1<<MD_SB_CLEAN)) | |
5085 | mddev->recovery_cp = MaxSector; | |
5086 | else | |
5087 | mddev->recovery_cp = 0; | |
5088 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5089 | mddev->external = 0; |
1da177e4 LT |
5090 | |
5091 | mddev->layout = info->layout; | |
9d8f0363 | 5092 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5093 | |
5094 | mddev->max_disks = MD_SB_DISKS; | |
5095 | ||
e691063a N |
5096 | if (mddev->persistent) |
5097 | mddev->flags = 0; | |
850b2b42 | 5098 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5099 | |
b2a2703c N |
5100 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
5101 | mddev->bitmap_offset = 0; | |
5102 | ||
f6705578 N |
5103 | mddev->reshape_position = MaxSector; |
5104 | ||
1da177e4 LT |
5105 | /* |
5106 | * Generate a 128 bit UUID | |
5107 | */ | |
5108 | get_random_bytes(mddev->uuid, 16); | |
5109 | ||
f6705578 | 5110 | mddev->new_level = mddev->level; |
664e7c41 | 5111 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5112 | mddev->new_layout = mddev->layout; |
5113 | mddev->delta_disks = 0; | |
5114 | ||
1da177e4 LT |
5115 | return 0; |
5116 | } | |
5117 | ||
1f403624 DW |
5118 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5119 | { | |
b522adcd DW |
5120 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5121 | ||
5122 | if (mddev->external_size) | |
5123 | return; | |
5124 | ||
1f403624 DW |
5125 | mddev->array_sectors = array_sectors; |
5126 | } | |
5127 | EXPORT_SYMBOL(md_set_array_sectors); | |
5128 | ||
d71f9f88 | 5129 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5130 | { |
159ec1fc | 5131 | mdk_rdev_t *rdev; |
a35b0d69 | 5132 | int rv; |
d71f9f88 | 5133 | int fit = (num_sectors == 0); |
a35b0d69 N |
5134 | |
5135 | if (mddev->pers->resize == NULL) | |
5136 | return -EINVAL; | |
d71f9f88 AN |
5137 | /* The "num_sectors" is the number of sectors of each device that |
5138 | * is used. This can only make sense for arrays with redundancy. | |
5139 | * linear and raid0 always use whatever space is available. We can only | |
5140 | * consider changing this number if no resync or reconstruction is | |
5141 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5142 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5143 | * of each device. If num_sectors is zero, we find the largest size |
5144 | * that fits. | |
5145 | ||
a35b0d69 N |
5146 | */ |
5147 | if (mddev->sync_thread) | |
5148 | return -EBUSY; | |
dba034ee N |
5149 | if (mddev->bitmap) |
5150 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5151 | * grow, and re-add. | |
5152 | */ | |
5153 | return -EBUSY; | |
159ec1fc | 5154 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5155 | sector_t avail = rdev->sectors; |
01ab5662 | 5156 | |
d71f9f88 AN |
5157 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5158 | num_sectors = avail; | |
5159 | if (avail < num_sectors) | |
a35b0d69 N |
5160 | return -ENOSPC; |
5161 | } | |
d71f9f88 | 5162 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5163 | if (!rv) |
5164 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5165 | return rv; |
5166 | } | |
5167 | ||
da943b99 N |
5168 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5169 | { | |
5170 | int rv; | |
5171 | /* change the number of raid disks */ | |
63c70c4f | 5172 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5173 | return -EINVAL; |
5174 | if (raid_disks <= 0 || | |
5175 | raid_disks >= mddev->max_disks) | |
5176 | return -EINVAL; | |
63c70c4f | 5177 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5178 | return -EBUSY; |
63c70c4f N |
5179 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5180 | ||
5181 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
5182 | return rv; |
5183 | } | |
5184 | ||
5185 | ||
1da177e4 LT |
5186 | /* |
5187 | * update_array_info is used to change the configuration of an | |
5188 | * on-line array. | |
5189 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5190 | * fields in the info are checked against the array. | |
5191 | * Any differences that cannot be handled will cause an error. | |
5192 | * Normally, only one change can be managed at a time. | |
5193 | */ | |
5194 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5195 | { | |
5196 | int rv = 0; | |
5197 | int cnt = 0; | |
36fa3063 N |
5198 | int state = 0; |
5199 | ||
5200 | /* calculate expected state,ignoring low bits */ | |
5201 | if (mddev->bitmap && mddev->bitmap_offset) | |
5202 | state |= (1 << MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
5203 | |
5204 | if (mddev->major_version != info->major_version || | |
5205 | mddev->minor_version != info->minor_version || | |
5206 | /* mddev->patch_version != info->patch_version || */ | |
5207 | mddev->ctime != info->ctime || | |
5208 | mddev->level != info->level || | |
5209 | /* mddev->layout != info->layout || */ | |
5210 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5211 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5212 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5213 | ((state^info->state) & 0xfffffe00) | |
5214 | ) | |
1da177e4 LT |
5215 | return -EINVAL; |
5216 | /* Check there is only one change */ | |
58c0fed4 AN |
5217 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5218 | cnt++; | |
5219 | if (mddev->raid_disks != info->raid_disks) | |
5220 | cnt++; | |
5221 | if (mddev->layout != info->layout) | |
5222 | cnt++; | |
5223 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5224 | cnt++; | |
5225 | if (cnt == 0) | |
5226 | return 0; | |
5227 | if (cnt > 1) | |
5228 | return -EINVAL; | |
1da177e4 LT |
5229 | |
5230 | if (mddev->layout != info->layout) { | |
5231 | /* Change layout | |
5232 | * we don't need to do anything at the md level, the | |
5233 | * personality will take care of it all. | |
5234 | */ | |
50ac168a | 5235 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5236 | return -EINVAL; |
597a711b N |
5237 | else { |
5238 | mddev->new_layout = info->layout; | |
50ac168a | 5239 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5240 | if (rv) |
5241 | mddev->new_layout = mddev->layout; | |
5242 | return rv; | |
5243 | } | |
1da177e4 | 5244 | } |
58c0fed4 | 5245 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5246 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5247 | |
da943b99 N |
5248 | if (mddev->raid_disks != info->raid_disks) |
5249 | rv = update_raid_disks(mddev, info->raid_disks); | |
5250 | ||
36fa3063 N |
5251 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5252 | if (mddev->pers->quiesce == NULL) | |
5253 | return -EINVAL; | |
5254 | if (mddev->recovery || mddev->sync_thread) | |
5255 | return -EBUSY; | |
5256 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5257 | /* add the bitmap */ | |
5258 | if (mddev->bitmap) | |
5259 | return -EEXIST; | |
5260 | if (mddev->default_bitmap_offset == 0) | |
5261 | return -EINVAL; | |
5262 | mddev->bitmap_offset = mddev->default_bitmap_offset; | |
5263 | mddev->pers->quiesce(mddev, 1); | |
5264 | rv = bitmap_create(mddev); | |
5265 | if (rv) | |
5266 | bitmap_destroy(mddev); | |
5267 | mddev->pers->quiesce(mddev, 0); | |
5268 | } else { | |
5269 | /* remove the bitmap */ | |
5270 | if (!mddev->bitmap) | |
5271 | return -ENOENT; | |
5272 | if (mddev->bitmap->file) | |
5273 | return -EINVAL; | |
5274 | mddev->pers->quiesce(mddev, 1); | |
5275 | bitmap_destroy(mddev); | |
5276 | mddev->pers->quiesce(mddev, 0); | |
5277 | mddev->bitmap_offset = 0; | |
5278 | } | |
5279 | } | |
850b2b42 | 5280 | md_update_sb(mddev, 1); |
1da177e4 LT |
5281 | return rv; |
5282 | } | |
5283 | ||
5284 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5285 | { | |
5286 | mdk_rdev_t *rdev; | |
5287 | ||
5288 | if (mddev->pers == NULL) | |
5289 | return -ENODEV; | |
5290 | ||
5291 | rdev = find_rdev(mddev, dev); | |
5292 | if (!rdev) | |
5293 | return -ENODEV; | |
5294 | ||
5295 | md_error(mddev, rdev); | |
5296 | return 0; | |
5297 | } | |
5298 | ||
2f9618ce AN |
5299 | /* |
5300 | * We have a problem here : there is no easy way to give a CHS | |
5301 | * virtual geometry. We currently pretend that we have a 2 heads | |
5302 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5303 | * dosfs just mad... ;-) | |
5304 | */ | |
a885c8c4 CH |
5305 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5306 | { | |
5307 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5308 | ||
5309 | geo->heads = 2; | |
5310 | geo->sectors = 4; | |
5311 | geo->cylinders = get_capacity(mddev->gendisk) / 8; | |
5312 | return 0; | |
5313 | } | |
5314 | ||
a39907fa | 5315 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5316 | unsigned int cmd, unsigned long arg) |
5317 | { | |
5318 | int err = 0; | |
5319 | void __user *argp = (void __user *)arg; | |
1da177e4 LT |
5320 | mddev_t *mddev = NULL; |
5321 | ||
5322 | if (!capable(CAP_SYS_ADMIN)) | |
5323 | return -EACCES; | |
5324 | ||
5325 | /* | |
5326 | * Commands dealing with the RAID driver but not any | |
5327 | * particular array: | |
5328 | */ | |
5329 | switch (cmd) | |
5330 | { | |
5331 | case RAID_VERSION: | |
5332 | err = get_version(argp); | |
5333 | goto done; | |
5334 | ||
5335 | case PRINT_RAID_DEBUG: | |
5336 | err = 0; | |
5337 | md_print_devices(); | |
5338 | goto done; | |
5339 | ||
5340 | #ifndef MODULE | |
5341 | case RAID_AUTORUN: | |
5342 | err = 0; | |
5343 | autostart_arrays(arg); | |
5344 | goto done; | |
5345 | #endif | |
5346 | default:; | |
5347 | } | |
5348 | ||
5349 | /* | |
5350 | * Commands creating/starting a new array: | |
5351 | */ | |
5352 | ||
a39907fa | 5353 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5354 | |
5355 | if (!mddev) { | |
5356 | BUG(); | |
5357 | goto abort; | |
5358 | } | |
5359 | ||
1da177e4 LT |
5360 | err = mddev_lock(mddev); |
5361 | if (err) { | |
5362 | printk(KERN_INFO | |
5363 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5364 | err, cmd); | |
5365 | goto abort; | |
5366 | } | |
5367 | ||
5368 | switch (cmd) | |
5369 | { | |
5370 | case SET_ARRAY_INFO: | |
5371 | { | |
5372 | mdu_array_info_t info; | |
5373 | if (!arg) | |
5374 | memset(&info, 0, sizeof(info)); | |
5375 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5376 | err = -EFAULT; | |
5377 | goto abort_unlock; | |
5378 | } | |
5379 | if (mddev->pers) { | |
5380 | err = update_array_info(mddev, &info); | |
5381 | if (err) { | |
5382 | printk(KERN_WARNING "md: couldn't update" | |
5383 | " array info. %d\n", err); | |
5384 | goto abort_unlock; | |
5385 | } | |
5386 | goto done_unlock; | |
5387 | } | |
5388 | if (!list_empty(&mddev->disks)) { | |
5389 | printk(KERN_WARNING | |
5390 | "md: array %s already has disks!\n", | |
5391 | mdname(mddev)); | |
5392 | err = -EBUSY; | |
5393 | goto abort_unlock; | |
5394 | } | |
5395 | if (mddev->raid_disks) { | |
5396 | printk(KERN_WARNING | |
5397 | "md: array %s already initialised!\n", | |
5398 | mdname(mddev)); | |
5399 | err = -EBUSY; | |
5400 | goto abort_unlock; | |
5401 | } | |
5402 | err = set_array_info(mddev, &info); | |
5403 | if (err) { | |
5404 | printk(KERN_WARNING "md: couldn't set" | |
5405 | " array info. %d\n", err); | |
5406 | goto abort_unlock; | |
5407 | } | |
5408 | } | |
5409 | goto done_unlock; | |
5410 | ||
5411 | default:; | |
5412 | } | |
5413 | ||
5414 | /* | |
5415 | * Commands querying/configuring an existing array: | |
5416 | */ | |
32a7627c | 5417 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5418 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5419 | if ((!mddev->raid_disks && !mddev->external) |
5420 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5421 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5422 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5423 | err = -ENODEV; |
5424 | goto abort_unlock; | |
5425 | } | |
5426 | ||
5427 | /* | |
5428 | * Commands even a read-only array can execute: | |
5429 | */ | |
5430 | switch (cmd) | |
5431 | { | |
5432 | case GET_ARRAY_INFO: | |
5433 | err = get_array_info(mddev, argp); | |
5434 | goto done_unlock; | |
5435 | ||
32a7627c | 5436 | case GET_BITMAP_FILE: |
87162a28 | 5437 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5438 | goto done_unlock; |
5439 | ||
1da177e4 LT |
5440 | case GET_DISK_INFO: |
5441 | err = get_disk_info(mddev, argp); | |
5442 | goto done_unlock; | |
5443 | ||
5444 | case RESTART_ARRAY_RW: | |
5445 | err = restart_array(mddev); | |
5446 | goto done_unlock; | |
5447 | ||
5448 | case STOP_ARRAY: | |
d710e138 | 5449 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5450 | goto done_unlock; |
5451 | ||
5452 | case STOP_ARRAY_RO: | |
d710e138 | 5453 | err = do_md_stop(mddev, 1, 1); |
1da177e4 LT |
5454 | goto done_unlock; |
5455 | ||
1da177e4 LT |
5456 | } |
5457 | ||
5458 | /* | |
5459 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5460 | * superblock, so we do not allow them on read-only arrays. |
5461 | * However non-MD ioctls (e.g. get-size) will still come through | |
5462 | * here and hit the 'default' below, so only disallow | |
5463 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5464 | */ |
bb57fc64 | 5465 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5466 | if (mddev->ro == 2) { |
5467 | mddev->ro = 0; | |
b62b7590 | 5468 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 NB |
5469 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5470 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5471 | } else { |
5472 | err = -EROFS; | |
5473 | goto abort_unlock; | |
5474 | } | |
1da177e4 LT |
5475 | } |
5476 | ||
5477 | switch (cmd) | |
5478 | { | |
5479 | case ADD_NEW_DISK: | |
5480 | { | |
5481 | mdu_disk_info_t info; | |
5482 | if (copy_from_user(&info, argp, sizeof(info))) | |
5483 | err = -EFAULT; | |
5484 | else | |
5485 | err = add_new_disk(mddev, &info); | |
5486 | goto done_unlock; | |
5487 | } | |
5488 | ||
5489 | case HOT_REMOVE_DISK: | |
5490 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5491 | goto done_unlock; | |
5492 | ||
5493 | case HOT_ADD_DISK: | |
5494 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5495 | goto done_unlock; | |
5496 | ||
5497 | case SET_DISK_FAULTY: | |
5498 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5499 | goto done_unlock; | |
5500 | ||
5501 | case RUN_ARRAY: | |
d710e138 | 5502 | err = do_md_run(mddev); |
1da177e4 LT |
5503 | goto done_unlock; |
5504 | ||
32a7627c N |
5505 | case SET_BITMAP_FILE: |
5506 | err = set_bitmap_file(mddev, (int)arg); | |
5507 | goto done_unlock; | |
5508 | ||
1da177e4 | 5509 | default: |
1da177e4 LT |
5510 | err = -EINVAL; |
5511 | goto abort_unlock; | |
5512 | } | |
5513 | ||
5514 | done_unlock: | |
5515 | abort_unlock: | |
d3374825 N |
5516 | if (mddev->hold_active == UNTIL_IOCTL && |
5517 | err != -EINVAL) | |
5518 | mddev->hold_active = 0; | |
1da177e4 LT |
5519 | mddev_unlock(mddev); |
5520 | ||
5521 | return err; | |
5522 | done: | |
5523 | if (err) | |
5524 | MD_BUG(); | |
5525 | abort: | |
5526 | return err; | |
5527 | } | |
5528 | ||
a39907fa | 5529 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
5530 | { |
5531 | /* | |
5532 | * Succeed if we can lock the mddev, which confirms that | |
5533 | * it isn't being stopped right now. | |
5534 | */ | |
d3374825 | 5535 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
5536 | int err; |
5537 | ||
d3374825 N |
5538 | if (mddev->gendisk != bdev->bd_disk) { |
5539 | /* we are racing with mddev_put which is discarding this | |
5540 | * bd_disk. | |
5541 | */ | |
5542 | mddev_put(mddev); | |
5543 | /* Wait until bdev->bd_disk is definitely gone */ | |
5544 | flush_scheduled_work(); | |
5545 | /* Then retry the open from the top */ | |
5546 | return -ERESTARTSYS; | |
5547 | } | |
5548 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
5549 | ||
c8c00a69 | 5550 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
5551 | goto out; |
5552 | ||
5553 | err = 0; | |
f2ea68cf | 5554 | atomic_inc(&mddev->openers); |
c8c00a69 | 5555 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 5556 | |
a39907fa | 5557 | check_disk_change(bdev); |
1da177e4 LT |
5558 | out: |
5559 | return err; | |
5560 | } | |
5561 | ||
a39907fa | 5562 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 5563 | { |
a39907fa | 5564 | mddev_t *mddev = disk->private_data; |
1da177e4 | 5565 | |
52e5f9d1 | 5566 | BUG_ON(!mddev); |
f2ea68cf | 5567 | atomic_dec(&mddev->openers); |
1da177e4 LT |
5568 | mddev_put(mddev); |
5569 | ||
5570 | return 0; | |
5571 | } | |
5572 | ||
44ce6294 LT |
5573 | static int md_media_changed(struct gendisk *disk) |
5574 | { | |
5575 | mddev_t *mddev = disk->private_data; | |
5576 | ||
5577 | return mddev->changed; | |
5578 | } | |
5579 | ||
5580 | static int md_revalidate(struct gendisk *disk) | |
5581 | { | |
5582 | mddev_t *mddev = disk->private_data; | |
5583 | ||
5584 | mddev->changed = 0; | |
5585 | return 0; | |
5586 | } | |
83d5cde4 | 5587 | static const struct block_device_operations md_fops = |
1da177e4 LT |
5588 | { |
5589 | .owner = THIS_MODULE, | |
a39907fa AV |
5590 | .open = md_open, |
5591 | .release = md_release, | |
b492b852 | 5592 | .ioctl = md_ioctl, |
a885c8c4 | 5593 | .getgeo = md_getgeo, |
44ce6294 LT |
5594 | .media_changed = md_media_changed, |
5595 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
5596 | }; |
5597 | ||
75c96f85 | 5598 | static int md_thread(void * arg) |
1da177e4 LT |
5599 | { |
5600 | mdk_thread_t *thread = arg; | |
5601 | ||
1da177e4 LT |
5602 | /* |
5603 | * md_thread is a 'system-thread', it's priority should be very | |
5604 | * high. We avoid resource deadlocks individually in each | |
5605 | * raid personality. (RAID5 does preallocation) We also use RR and | |
5606 | * the very same RT priority as kswapd, thus we will never get | |
5607 | * into a priority inversion deadlock. | |
5608 | * | |
5609 | * we definitely have to have equal or higher priority than | |
5610 | * bdflush, otherwise bdflush will deadlock if there are too | |
5611 | * many dirty RAID5 blocks. | |
5612 | */ | |
1da177e4 | 5613 | |
6985c43f | 5614 | allow_signal(SIGKILL); |
a6fb0934 | 5615 | while (!kthread_should_stop()) { |
1da177e4 | 5616 | |
93588e22 N |
5617 | /* We need to wait INTERRUPTIBLE so that |
5618 | * we don't add to the load-average. | |
5619 | * That means we need to be sure no signals are | |
5620 | * pending | |
5621 | */ | |
5622 | if (signal_pending(current)) | |
5623 | flush_signals(current); | |
5624 | ||
5625 | wait_event_interruptible_timeout | |
5626 | (thread->wqueue, | |
5627 | test_bit(THREAD_WAKEUP, &thread->flags) | |
5628 | || kthread_should_stop(), | |
5629 | thread->timeout); | |
1da177e4 LT |
5630 | |
5631 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
5632 | ||
787453c2 | 5633 | thread->run(thread->mddev); |
1da177e4 | 5634 | } |
a6fb0934 | 5635 | |
1da177e4 LT |
5636 | return 0; |
5637 | } | |
5638 | ||
5639 | void md_wakeup_thread(mdk_thread_t *thread) | |
5640 | { | |
5641 | if (thread) { | |
5642 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
5643 | set_bit(THREAD_WAKEUP, &thread->flags); | |
5644 | wake_up(&thread->wqueue); | |
5645 | } | |
5646 | } | |
5647 | ||
5648 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
5649 | const char *name) | |
5650 | { | |
5651 | mdk_thread_t *thread; | |
1da177e4 | 5652 | |
9ffae0cf | 5653 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
5654 | if (!thread) |
5655 | return NULL; | |
5656 | ||
1da177e4 LT |
5657 | init_waitqueue_head(&thread->wqueue); |
5658 | ||
1da177e4 LT |
5659 | thread->run = run; |
5660 | thread->mddev = mddev; | |
32a7627c | 5661 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
5662 | thread->tsk = kthread_run(md_thread, thread, |
5663 | "%s_%s", | |
5664 | mdname(thread->mddev), | |
5665 | name ?: mddev->pers->name); | |
a6fb0934 | 5666 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
5667 | kfree(thread); |
5668 | return NULL; | |
5669 | } | |
1da177e4 LT |
5670 | return thread; |
5671 | } | |
5672 | ||
1da177e4 LT |
5673 | void md_unregister_thread(mdk_thread_t *thread) |
5674 | { | |
e0cf8f04 N |
5675 | if (!thread) |
5676 | return; | |
ba25f9dc | 5677 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
5678 | |
5679 | kthread_stop(thread->tsk); | |
1da177e4 LT |
5680 | kfree(thread); |
5681 | } | |
5682 | ||
5683 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
5684 | { | |
5685 | if (!mddev) { | |
5686 | MD_BUG(); | |
5687 | return; | |
5688 | } | |
5689 | ||
b2d444d7 | 5690 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5691 | return; |
6bfe0b49 DW |
5692 | |
5693 | if (mddev->external) | |
5694 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 5695 | /* |
1da177e4 LT |
5696 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
5697 | mdname(mddev), | |
5698 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
5699 | __builtin_return_address(0),__builtin_return_address(1), | |
5700 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 5701 | */ |
d0a0a5ee AM |
5702 | if (!mddev->pers) |
5703 | return; | |
1da177e4 LT |
5704 | if (!mddev->pers->error_handler) |
5705 | return; | |
5706 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
5707 | if (mddev->degraded) |
5708 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
52664732 | 5709 | set_bit(StateChanged, &rdev->flags); |
1da177e4 LT |
5710 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
5711 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5712 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 5713 | md_new_event_inintr(mddev); |
1da177e4 LT |
5714 | } |
5715 | ||
5716 | /* seq_file implementation /proc/mdstat */ | |
5717 | ||
5718 | static void status_unused(struct seq_file *seq) | |
5719 | { | |
5720 | int i = 0; | |
5721 | mdk_rdev_t *rdev; | |
1da177e4 LT |
5722 | |
5723 | seq_printf(seq, "unused devices: "); | |
5724 | ||
159ec1fc | 5725 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
5726 | char b[BDEVNAME_SIZE]; |
5727 | i++; | |
5728 | seq_printf(seq, "%s ", | |
5729 | bdevname(rdev->bdev,b)); | |
5730 | } | |
5731 | if (!i) | |
5732 | seq_printf(seq, "<none>"); | |
5733 | ||
5734 | seq_printf(seq, "\n"); | |
5735 | } | |
5736 | ||
5737 | ||
5738 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
5739 | { | |
dd71cf6b N |
5740 | sector_t max_sectors, resync, res; |
5741 | unsigned long dt, db; | |
5742 | sector_t rt; | |
4588b42e N |
5743 | int scale; |
5744 | unsigned int per_milli; | |
1da177e4 | 5745 | |
dd71cf6b | 5746 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
5747 | |
5748 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 5749 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 5750 | else |
dd71cf6b | 5751 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
5752 | |
5753 | /* | |
5754 | * Should not happen. | |
5755 | */ | |
dd71cf6b | 5756 | if (!max_sectors) { |
1da177e4 LT |
5757 | MD_BUG(); |
5758 | return; | |
5759 | } | |
4588b42e | 5760 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 5761 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
5762 | * u32, as those are the requirements for sector_div. |
5763 | * Thus 'scale' must be at least 10 | |
5764 | */ | |
5765 | scale = 10; | |
5766 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 5767 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
5768 | scale++; |
5769 | } | |
5770 | res = (resync>>scale)*1000; | |
dd71cf6b | 5771 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
5772 | |
5773 | per_milli = res; | |
1da177e4 | 5774 | { |
4588b42e | 5775 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
5776 | seq_printf(seq, "["); |
5777 | for (i = 0; i < x; i++) | |
5778 | seq_printf(seq, "="); | |
5779 | seq_printf(seq, ">"); | |
5780 | for (i = 0; i < y; i++) | |
5781 | seq_printf(seq, "."); | |
5782 | seq_printf(seq, "] "); | |
5783 | } | |
4588b42e | 5784 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
5785 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
5786 | "reshape" : | |
61df9d91 N |
5787 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
5788 | "check" : | |
5789 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
5790 | "resync" : "recovery"))), | |
5791 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
5792 | (unsigned long long) resync/2, |
5793 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
5794 | |
5795 | /* | |
1da177e4 LT |
5796 | * dt: time from mark until now |
5797 | * db: blocks written from mark until now | |
5798 | * rt: remaining time | |
dd71cf6b N |
5799 | * |
5800 | * rt is a sector_t, so could be 32bit or 64bit. | |
5801 | * So we divide before multiply in case it is 32bit and close | |
5802 | * to the limit. | |
5803 | * We scale the divisor (db) by 32 to avoid loosing precision | |
5804 | * near the end of resync when the number of remaining sectors | |
5805 | * is close to 'db'. | |
5806 | * We then divide rt by 32 after multiplying by db to compensate. | |
5807 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
5808 | */ |
5809 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
5810 | if (!dt) dt++; | |
ff4e8d9a N |
5811 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
5812 | - mddev->resync_mark_cnt; | |
1da177e4 | 5813 | |
dd71cf6b N |
5814 | rt = max_sectors - resync; /* number of remaining sectors */ |
5815 | sector_div(rt, db/32+1); | |
5816 | rt *= dt; | |
5817 | rt >>= 5; | |
5818 | ||
5819 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
5820 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 5821 | |
ff4e8d9a | 5822 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
5823 | } |
5824 | ||
5825 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
5826 | { | |
5827 | struct list_head *tmp; | |
5828 | loff_t l = *pos; | |
5829 | mddev_t *mddev; | |
5830 | ||
5831 | if (l >= 0x10000) | |
5832 | return NULL; | |
5833 | if (!l--) | |
5834 | /* header */ | |
5835 | return (void*)1; | |
5836 | ||
5837 | spin_lock(&all_mddevs_lock); | |
5838 | list_for_each(tmp,&all_mddevs) | |
5839 | if (!l--) { | |
5840 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
5841 | mddev_get(mddev); | |
5842 | spin_unlock(&all_mddevs_lock); | |
5843 | return mddev; | |
5844 | } | |
5845 | spin_unlock(&all_mddevs_lock); | |
5846 | if (!l--) | |
5847 | return (void*)2;/* tail */ | |
5848 | return NULL; | |
5849 | } | |
5850 | ||
5851 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
5852 | { | |
5853 | struct list_head *tmp; | |
5854 | mddev_t *next_mddev, *mddev = v; | |
5855 | ||
5856 | ++*pos; | |
5857 | if (v == (void*)2) | |
5858 | return NULL; | |
5859 | ||
5860 | spin_lock(&all_mddevs_lock); | |
5861 | if (v == (void*)1) | |
5862 | tmp = all_mddevs.next; | |
5863 | else | |
5864 | tmp = mddev->all_mddevs.next; | |
5865 | if (tmp != &all_mddevs) | |
5866 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
5867 | else { | |
5868 | next_mddev = (void*)2; | |
5869 | *pos = 0x10000; | |
5870 | } | |
5871 | spin_unlock(&all_mddevs_lock); | |
5872 | ||
5873 | if (v != (void*)1) | |
5874 | mddev_put(mddev); | |
5875 | return next_mddev; | |
5876 | ||
5877 | } | |
5878 | ||
5879 | static void md_seq_stop(struct seq_file *seq, void *v) | |
5880 | { | |
5881 | mddev_t *mddev = v; | |
5882 | ||
5883 | if (mddev && v != (void*)1 && v != (void*)2) | |
5884 | mddev_put(mddev); | |
5885 | } | |
5886 | ||
d7603b7e N |
5887 | struct mdstat_info { |
5888 | int event; | |
5889 | }; | |
5890 | ||
1da177e4 LT |
5891 | static int md_seq_show(struct seq_file *seq, void *v) |
5892 | { | |
5893 | mddev_t *mddev = v; | |
dd8ac336 | 5894 | sector_t sectors; |
1da177e4 | 5895 | mdk_rdev_t *rdev; |
d7603b7e | 5896 | struct mdstat_info *mi = seq->private; |
32a7627c | 5897 | struct bitmap *bitmap; |
1da177e4 LT |
5898 | |
5899 | if (v == (void*)1) { | |
2604b703 | 5900 | struct mdk_personality *pers; |
1da177e4 LT |
5901 | seq_printf(seq, "Personalities : "); |
5902 | spin_lock(&pers_lock); | |
2604b703 N |
5903 | list_for_each_entry(pers, &pers_list, list) |
5904 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
5905 | |
5906 | spin_unlock(&pers_lock); | |
5907 | seq_printf(seq, "\n"); | |
d7603b7e | 5908 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
5909 | return 0; |
5910 | } | |
5911 | if (v == (void*)2) { | |
5912 | status_unused(seq); | |
5913 | return 0; | |
5914 | } | |
5915 | ||
5dc5cf7d | 5916 | if (mddev_lock(mddev) < 0) |
1da177e4 | 5917 | return -EINTR; |
5dc5cf7d | 5918 | |
1da177e4 LT |
5919 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
5920 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
5921 | mddev->pers ? "" : "in"); | |
5922 | if (mddev->pers) { | |
f91de92e | 5923 | if (mddev->ro==1) |
1da177e4 | 5924 | seq_printf(seq, " (read-only)"); |
f91de92e | 5925 | if (mddev->ro==2) |
52720ae7 | 5926 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
5927 | seq_printf(seq, " %s", mddev->pers->name); |
5928 | } | |
5929 | ||
dd8ac336 | 5930 | sectors = 0; |
159ec1fc | 5931 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
5932 | char b[BDEVNAME_SIZE]; |
5933 | seq_printf(seq, " %s[%d]", | |
5934 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
5935 | if (test_bit(WriteMostly, &rdev->flags)) |
5936 | seq_printf(seq, "(W)"); | |
b2d444d7 | 5937 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5938 | seq_printf(seq, "(F)"); |
5939 | continue; | |
b325a32e N |
5940 | } else if (rdev->raid_disk < 0) |
5941 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 5942 | sectors += rdev->sectors; |
1da177e4 LT |
5943 | } |
5944 | ||
5945 | if (!list_empty(&mddev->disks)) { | |
5946 | if (mddev->pers) | |
5947 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
5948 | (unsigned long long) |
5949 | mddev->array_sectors / 2); | |
1da177e4 LT |
5950 | else |
5951 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 5952 | (unsigned long long)sectors / 2); |
1da177e4 | 5953 | } |
1cd6bf19 N |
5954 | if (mddev->persistent) { |
5955 | if (mddev->major_version != 0 || | |
5956 | mddev->minor_version != 90) { | |
5957 | seq_printf(seq," super %d.%d", | |
5958 | mddev->major_version, | |
5959 | mddev->minor_version); | |
5960 | } | |
e691063a N |
5961 | } else if (mddev->external) |
5962 | seq_printf(seq, " super external:%s", | |
5963 | mddev->metadata_type); | |
5964 | else | |
1cd6bf19 | 5965 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
5966 | |
5967 | if (mddev->pers) { | |
d710e138 | 5968 | mddev->pers->status(seq, mddev); |
1da177e4 | 5969 | seq_printf(seq, "\n "); |
8e1b39d6 N |
5970 | if (mddev->pers->sync_request) { |
5971 | if (mddev->curr_resync > 2) { | |
d710e138 | 5972 | status_resync(seq, mddev); |
8e1b39d6 N |
5973 | seq_printf(seq, "\n "); |
5974 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
5975 | seq_printf(seq, "\tresync=DELAYED\n "); | |
5976 | else if (mddev->recovery_cp < MaxSector) | |
5977 | seq_printf(seq, "\tresync=PENDING\n "); | |
5978 | } | |
32a7627c N |
5979 | } else |
5980 | seq_printf(seq, "\n "); | |
5981 | ||
5982 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
5983 | unsigned long chunk_kb; |
5984 | unsigned long flags; | |
32a7627c N |
5985 | spin_lock_irqsave(&bitmap->lock, flags); |
5986 | chunk_kb = bitmap->chunksize >> 10; | |
5987 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
5988 | "%lu%s chunk", | |
5989 | bitmap->pages - bitmap->missing_pages, | |
5990 | bitmap->pages, | |
5991 | (bitmap->pages - bitmap->missing_pages) | |
5992 | << (PAGE_SHIFT - 10), | |
5993 | chunk_kb ? chunk_kb : bitmap->chunksize, | |
5994 | chunk_kb ? "KB" : "B"); | |
78d742d8 N |
5995 | if (bitmap->file) { |
5996 | seq_printf(seq, ", file: "); | |
c32c2f63 | 5997 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 5998 | } |
78d742d8 | 5999 | |
32a7627c N |
6000 | seq_printf(seq, "\n"); |
6001 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
6002 | } |
6003 | ||
6004 | seq_printf(seq, "\n"); | |
6005 | } | |
6006 | mddev_unlock(mddev); | |
6007 | ||
6008 | return 0; | |
6009 | } | |
6010 | ||
110518bc | 6011 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6012 | .start = md_seq_start, |
6013 | .next = md_seq_next, | |
6014 | .stop = md_seq_stop, | |
6015 | .show = md_seq_show, | |
6016 | }; | |
6017 | ||
6018 | static int md_seq_open(struct inode *inode, struct file *file) | |
6019 | { | |
6020 | int error; | |
d7603b7e N |
6021 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6022 | if (mi == NULL) | |
6023 | return -ENOMEM; | |
1da177e4 LT |
6024 | |
6025 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6026 | if (error) |
6027 | kfree(mi); | |
6028 | else { | |
6029 | struct seq_file *p = file->private_data; | |
6030 | p->private = mi; | |
6031 | mi->event = atomic_read(&md_event_count); | |
6032 | } | |
1da177e4 LT |
6033 | return error; |
6034 | } | |
6035 | ||
d7603b7e N |
6036 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6037 | { | |
6038 | struct seq_file *m = filp->private_data; | |
6039 | struct mdstat_info *mi = m->private; | |
6040 | int mask; | |
6041 | ||
6042 | poll_wait(filp, &md_event_waiters, wait); | |
6043 | ||
6044 | /* always allow read */ | |
6045 | mask = POLLIN | POLLRDNORM; | |
6046 | ||
6047 | if (mi->event != atomic_read(&md_event_count)) | |
6048 | mask |= POLLERR | POLLPRI; | |
6049 | return mask; | |
6050 | } | |
6051 | ||
fa027c2a | 6052 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6053 | .owner = THIS_MODULE, |
1da177e4 LT |
6054 | .open = md_seq_open, |
6055 | .read = seq_read, | |
6056 | .llseek = seq_lseek, | |
c3f94b40 | 6057 | .release = seq_release_private, |
d7603b7e | 6058 | .poll = mdstat_poll, |
1da177e4 LT |
6059 | }; |
6060 | ||
2604b703 | 6061 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6062 | { |
1da177e4 | 6063 | spin_lock(&pers_lock); |
2604b703 N |
6064 | list_add_tail(&p->list, &pers_list); |
6065 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6066 | spin_unlock(&pers_lock); |
6067 | return 0; | |
6068 | } | |
6069 | ||
2604b703 | 6070 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6071 | { |
2604b703 | 6072 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6073 | spin_lock(&pers_lock); |
2604b703 | 6074 | list_del_init(&p->list); |
1da177e4 LT |
6075 | spin_unlock(&pers_lock); |
6076 | return 0; | |
6077 | } | |
6078 | ||
eea1bf38 | 6079 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6080 | { |
6081 | mdk_rdev_t * rdev; | |
1da177e4 | 6082 | int idle; |
eea1bf38 | 6083 | int curr_events; |
1da177e4 LT |
6084 | |
6085 | idle = 1; | |
4b80991c N |
6086 | rcu_read_lock(); |
6087 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6088 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6089 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6090 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6091 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6092 | /* sync IO will cause sync_io to increase before the disk_stats |
6093 | * as sync_io is counted when a request starts, and | |
6094 | * disk_stats is counted when it completes. | |
6095 | * So resync activity will cause curr_events to be smaller than | |
6096 | * when there was no such activity. | |
6097 | * non-sync IO will cause disk_stat to increase without | |
6098 | * increasing sync_io so curr_events will (eventually) | |
6099 | * be larger than it was before. Once it becomes | |
6100 | * substantially larger, the test below will cause | |
6101 | * the array to appear non-idle, and resync will slow | |
6102 | * down. | |
6103 | * If there is a lot of outstanding resync activity when | |
6104 | * we set last_event to curr_events, then all that activity | |
6105 | * completing might cause the array to appear non-idle | |
6106 | * and resync will be slowed down even though there might | |
6107 | * not have been non-resync activity. This will only | |
6108 | * happen once though. 'last_events' will soon reflect | |
6109 | * the state where there is little or no outstanding | |
6110 | * resync requests, and further resync activity will | |
6111 | * always make curr_events less than last_events. | |
c0e48521 | 6112 | * |
1da177e4 | 6113 | */ |
eea1bf38 | 6114 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6115 | rdev->last_events = curr_events; |
6116 | idle = 0; | |
6117 | } | |
6118 | } | |
4b80991c | 6119 | rcu_read_unlock(); |
1da177e4 LT |
6120 | return idle; |
6121 | } | |
6122 | ||
6123 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6124 | { | |
6125 | /* another "blocks" (512byte) blocks have been synced */ | |
6126 | atomic_sub(blocks, &mddev->recovery_active); | |
6127 | wake_up(&mddev->recovery_wait); | |
6128 | if (!ok) { | |
dfc70645 | 6129 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6130 | md_wakeup_thread(mddev->thread); |
6131 | // stop recovery, signal do_sync .... | |
6132 | } | |
6133 | } | |
6134 | ||
6135 | ||
06d91a5f N |
6136 | /* md_write_start(mddev, bi) |
6137 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6138 | * in superblock) before writing, schedule a superblock update |
6139 | * and wait for it to complete. | |
06d91a5f | 6140 | */ |
3d310eb7 | 6141 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6142 | { |
0fd62b86 | 6143 | int did_change = 0; |
06d91a5f | 6144 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6145 | return; |
06d91a5f | 6146 | |
f91de92e N |
6147 | BUG_ON(mddev->ro == 1); |
6148 | if (mddev->ro == 2) { | |
6149 | /* need to switch to read/write */ | |
6150 | mddev->ro = 0; | |
6151 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6152 | md_wakeup_thread(mddev->thread); | |
25156198 | 6153 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6154 | did_change = 1; |
f91de92e | 6155 | } |
06d91a5f | 6156 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6157 | if (mddev->safemode == 1) |
6158 | mddev->safemode = 0; | |
06d91a5f | 6159 | if (mddev->in_sync) { |
a9701a30 | 6160 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6161 | if (mddev->in_sync) { |
6162 | mddev->in_sync = 0; | |
850b2b42 | 6163 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3d310eb7 | 6164 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6165 | did_change = 1; |
3d310eb7 | 6166 | } |
a9701a30 | 6167 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6168 | } |
0fd62b86 | 6169 | if (did_change) |
b62b7590 | 6170 | sysfs_notify_dirent(mddev->sysfs_state); |
09a44cc1 N |
6171 | wait_event(mddev->sb_wait, |
6172 | !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && | |
6173 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1da177e4 LT |
6174 | } |
6175 | ||
6176 | void md_write_end(mddev_t *mddev) | |
6177 | { | |
6178 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6179 | if (mddev->safemode == 2) | |
6180 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6181 | else if (mddev->safemode_delay) |
1da177e4 LT |
6182 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6183 | } | |
6184 | } | |
6185 | ||
2a2275d6 N |
6186 | /* md_allow_write(mddev) |
6187 | * Calling this ensures that the array is marked 'active' so that writes | |
6188 | * may proceed without blocking. It is important to call this before | |
6189 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6190 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6191 | * |
6192 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6193 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6194 | */ |
b5470dc5 | 6195 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6196 | { |
6197 | if (!mddev->pers) | |
b5470dc5 | 6198 | return 0; |
2a2275d6 | 6199 | if (mddev->ro) |
b5470dc5 | 6200 | return 0; |
1a0fd497 | 6201 | if (!mddev->pers->sync_request) |
b5470dc5 | 6202 | return 0; |
2a2275d6 N |
6203 | |
6204 | spin_lock_irq(&mddev->write_lock); | |
6205 | if (mddev->in_sync) { | |
6206 | mddev->in_sync = 0; | |
6207 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6208 | if (mddev->safemode_delay && | |
6209 | mddev->safemode == 0) | |
6210 | mddev->safemode = 1; | |
6211 | spin_unlock_irq(&mddev->write_lock); | |
6212 | md_update_sb(mddev, 0); | |
b62b7590 | 6213 | sysfs_notify_dirent(mddev->sysfs_state); |
2a2275d6 N |
6214 | } else |
6215 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 DW |
6216 | |
6217 | if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
6218 | return -EAGAIN; | |
6219 | else | |
6220 | return 0; | |
2a2275d6 N |
6221 | } |
6222 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6223 | ||
1da177e4 LT |
6224 | #define SYNC_MARKS 10 |
6225 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6226 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6227 | { |
6228 | mddev_t *mddev2; | |
6229 | unsigned int currspeed = 0, | |
6230 | window; | |
57afd89f | 6231 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6232 | unsigned long mark[SYNC_MARKS]; |
6233 | sector_t mark_cnt[SYNC_MARKS]; | |
6234 | int last_mark,m; | |
6235 | struct list_head *tmp; | |
6236 | sector_t last_check; | |
57afd89f | 6237 | int skipped = 0; |
5fd6c1dc | 6238 | mdk_rdev_t *rdev; |
61df9d91 | 6239 | char *desc; |
1da177e4 LT |
6240 | |
6241 | /* just incase thread restarts... */ | |
6242 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6243 | return; | |
5fd6c1dc N |
6244 | if (mddev->ro) /* never try to sync a read-only array */ |
6245 | return; | |
1da177e4 | 6246 | |
61df9d91 N |
6247 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6248 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6249 | desc = "data-check"; | |
6250 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6251 | desc = "requested-resync"; | |
6252 | else | |
6253 | desc = "resync"; | |
6254 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6255 | desc = "reshape"; | |
6256 | else | |
6257 | desc = "recovery"; | |
6258 | ||
1da177e4 LT |
6259 | /* we overload curr_resync somewhat here. |
6260 | * 0 == not engaged in resync at all | |
6261 | * 2 == checking that there is no conflict with another sync | |
6262 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6263 | * commense | |
6264 | * other == active in resync - this many blocks | |
6265 | * | |
6266 | * Before starting a resync we must have set curr_resync to | |
6267 | * 2, and then checked that every "conflicting" array has curr_resync | |
6268 | * less than ours. When we find one that is the same or higher | |
6269 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6270 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6271 | * This will mean we have to start checking from the beginning again. | |
6272 | * | |
6273 | */ | |
6274 | ||
6275 | do { | |
6276 | mddev->curr_resync = 2; | |
6277 | ||
6278 | try_again: | |
787453c2 | 6279 | if (kthread_should_stop()) { |
6985c43f | 6280 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6281 | goto skip; |
6282 | } | |
29ac4aa3 | 6283 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6284 | if (mddev2 == mddev) |
6285 | continue; | |
90b08710 BS |
6286 | if (!mddev->parallel_resync |
6287 | && mddev2->curr_resync | |
6288 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6289 | DEFINE_WAIT(wq); |
6290 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6291 | /* arbitrarily yield */ | |
6292 | mddev->curr_resync = 1; | |
6293 | wake_up(&resync_wait); | |
6294 | } | |
6295 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6296 | /* no need to wait here, we can wait the next | |
6297 | * time 'round when curr_resync == 2 | |
6298 | */ | |
6299 | continue; | |
9744197c N |
6300 | /* We need to wait 'interruptible' so as not to |
6301 | * contribute to the load average, and not to | |
6302 | * be caught by 'softlockup' | |
6303 | */ | |
6304 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6305 | if (!kthread_should_stop() && |
8712e553 | 6306 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6307 | printk(KERN_INFO "md: delaying %s of %s" |
6308 | " until %s has finished (they" | |
1da177e4 | 6309 | " share one or more physical units)\n", |
61df9d91 | 6310 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6311 | mddev_put(mddev2); |
9744197c N |
6312 | if (signal_pending(current)) |
6313 | flush_signals(current); | |
1da177e4 LT |
6314 | schedule(); |
6315 | finish_wait(&resync_wait, &wq); | |
6316 | goto try_again; | |
6317 | } | |
6318 | finish_wait(&resync_wait, &wq); | |
6319 | } | |
6320 | } | |
6321 | } while (mddev->curr_resync < 2); | |
6322 | ||
5fd6c1dc | 6323 | j = 0; |
9d88883e | 6324 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6325 | /* resync follows the size requested by the personality, |
57afd89f | 6326 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6327 | */ |
6328 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6329 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6330 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6331 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6332 | j = mddev->resync_min; | |
6333 | else if (!mddev->bitmap) | |
5fd6c1dc | 6334 | j = mddev->recovery_cp; |
5e96ee65 | 6335 | |
ccfcc3c1 | 6336 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6337 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6338 | else { |
1da177e4 | 6339 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6340 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6341 | j = MaxSector; |
159ec1fc | 6342 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
6343 | if (rdev->raid_disk >= 0 && |
6344 | !test_bit(Faulty, &rdev->flags) && | |
6345 | !test_bit(In_sync, &rdev->flags) && | |
6346 | rdev->recovery_offset < j) | |
6347 | j = rdev->recovery_offset; | |
6348 | } | |
1da177e4 | 6349 | |
61df9d91 N |
6350 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6351 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6352 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6353 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6354 | "(but not more than %d KB/sec) for %s.\n", |
6355 | speed_max(mddev), desc); | |
1da177e4 | 6356 | |
eea1bf38 | 6357 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6358 | |
57afd89f | 6359 | io_sectors = 0; |
1da177e4 LT |
6360 | for (m = 0; m < SYNC_MARKS; m++) { |
6361 | mark[m] = jiffies; | |
57afd89f | 6362 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6363 | } |
6364 | last_mark = 0; | |
6365 | mddev->resync_mark = mark[last_mark]; | |
6366 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6367 | ||
6368 | /* | |
6369 | * Tune reconstruction: | |
6370 | */ | |
6371 | window = 32*(PAGE_SIZE/512); | |
6372 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6373 | window/2,(unsigned long long) max_sectors/2); | |
6374 | ||
6375 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6376 | last_check = 0; |
6377 | ||
6378 | if (j>2) { | |
6379 | printk(KERN_INFO | |
61df9d91 N |
6380 | "md: resuming %s of %s from checkpoint.\n", |
6381 | desc, mdname(mddev)); | |
1da177e4 LT |
6382 | mddev->curr_resync = j; |
6383 | } | |
6384 | ||
6385 | while (j < max_sectors) { | |
57afd89f | 6386 | sector_t sectors; |
1da177e4 | 6387 | |
57afd89f | 6388 | skipped = 0; |
97e4f42d | 6389 | |
7a91ee1f N |
6390 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6391 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6392 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6393 | > (max_sectors >> 4)) || | |
6394 | (j - mddev->curr_resync_completed)*2 | |
6395 | >= mddev->resync_max - mddev->curr_resync_completed | |
6396 | )) { | |
97e4f42d N |
6397 | /* time to update curr_resync_completed */ |
6398 | blk_unplug(mddev->queue); | |
6399 | wait_event(mddev->recovery_wait, | |
6400 | atomic_read(&mddev->recovery_active) == 0); | |
6401 | mddev->curr_resync_completed = | |
6402 | mddev->curr_resync; | |
6403 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
acb180b0 | 6404 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6405 | } |
acb180b0 | 6406 | |
e62e58a5 N |
6407 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6408 | /* As this condition is controlled by user-space, | |
6409 | * we can block indefinitely, so use '_interruptible' | |
6410 | * to avoid triggering warnings. | |
6411 | */ | |
6412 | flush_signals(current); /* just in case */ | |
6413 | wait_event_interruptible(mddev->recovery_wait, | |
6414 | mddev->resync_max > j | |
6415 | || kthread_should_stop()); | |
6416 | } | |
acb180b0 N |
6417 | |
6418 | if (kthread_should_stop()) | |
6419 | goto interrupted; | |
6420 | ||
57afd89f | 6421 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6422 | currspeed < speed_min(mddev)); |
57afd89f | 6423 | if (sectors == 0) { |
dfc70645 | 6424 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6425 | goto out; |
6426 | } | |
57afd89f N |
6427 | |
6428 | if (!skipped) { /* actual IO requested */ | |
6429 | io_sectors += sectors; | |
6430 | atomic_add(sectors, &mddev->recovery_active); | |
6431 | } | |
6432 | ||
1da177e4 LT |
6433 | j += sectors; |
6434 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6435 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6436 | if (last_check == 0) |
6437 | /* this is the earliers that rebuilt will be | |
6438 | * visible in /proc/mdstat | |
6439 | */ | |
6440 | md_new_event(mddev); | |
57afd89f N |
6441 | |
6442 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6443 | continue; |
6444 | ||
57afd89f | 6445 | last_check = io_sectors; |
1da177e4 | 6446 | |
dfc70645 | 6447 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6448 | break; |
6449 | ||
6450 | repeat: | |
6451 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6452 | /* step marks */ | |
6453 | int next = (last_mark+1) % SYNC_MARKS; | |
6454 | ||
6455 | mddev->resync_mark = mark[next]; | |
6456 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6457 | mark[next] = jiffies; | |
57afd89f | 6458 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6459 | last_mark = next; |
6460 | } | |
6461 | ||
6462 | ||
c6207277 N |
6463 | if (kthread_should_stop()) |
6464 | goto interrupted; | |
6465 | ||
1da177e4 LT |
6466 | |
6467 | /* | |
6468 | * this loop exits only if either when we are slower than | |
6469 | * the 'hard' speed limit, or the system was IO-idle for | |
6470 | * a jiffy. | |
6471 | * the system might be non-idle CPU-wise, but we only care | |
6472 | * about not overloading the IO subsystem. (things like an | |
6473 | * e2fsck being done on the RAID array should execute fast) | |
6474 | */ | |
2ad8b1ef | 6475 | blk_unplug(mddev->queue); |
1da177e4 LT |
6476 | cond_resched(); |
6477 | ||
57afd89f N |
6478 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6479 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6480 | |
88202a0c N |
6481 | if (currspeed > speed_min(mddev)) { |
6482 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6483 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6484 | msleep(500); |
1da177e4 LT |
6485 | goto repeat; |
6486 | } | |
6487 | } | |
6488 | } | |
61df9d91 | 6489 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6490 | /* |
6491 | * this also signals 'finished resyncing' to md_stop | |
6492 | */ | |
6493 | out: | |
2ad8b1ef | 6494 | blk_unplug(mddev->queue); |
1da177e4 LT |
6495 | |
6496 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
6497 | ||
6498 | /* tell personality that we are finished */ | |
57afd89f | 6499 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6500 | |
dfc70645 | 6501 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6502 | mddev->curr_resync > 2) { |
6503 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6504 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6505 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6506 | printk(KERN_INFO | |
61df9d91 N |
6507 | "md: checkpointing %s of %s.\n", |
6508 | desc, mdname(mddev)); | |
5fd6c1dc N |
6509 | mddev->recovery_cp = mddev->curr_resync; |
6510 | } | |
6511 | } else | |
6512 | mddev->recovery_cp = MaxSector; | |
6513 | } else { | |
6514 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6515 | mddev->curr_resync = MaxSector; | |
159ec1fc | 6516 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5fd6c1dc N |
6517 | if (rdev->raid_disk >= 0 && |
6518 | !test_bit(Faulty, &rdev->flags) && | |
6519 | !test_bit(In_sync, &rdev->flags) && | |
6520 | rdev->recovery_offset < mddev->curr_resync) | |
6521 | rdev->recovery_offset = mddev->curr_resync; | |
5fd6c1dc | 6522 | } |
1da177e4 | 6523 | } |
17571284 | 6524 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6525 | |
1da177e4 LT |
6526 | skip: |
6527 | mddev->curr_resync = 0; | |
acb180b0 | 6528 | mddev->curr_resync_completed = 0; |
24395a85 N |
6529 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
6530 | /* We completed so max setting can be forgotten. */ | |
6531 | mddev->resync_max = MaxSector; | |
c6207277 | 6532 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
1da177e4 LT |
6533 | wake_up(&resync_wait); |
6534 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
6535 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
6536 | return; |
6537 | ||
6538 | interrupted: | |
6539 | /* | |
6540 | * got a signal, exit. | |
6541 | */ | |
6542 | printk(KERN_INFO | |
6543 | "md: md_do_sync() got signal ... exiting\n"); | |
6544 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6545 | goto out; | |
6546 | ||
1da177e4 | 6547 | } |
29269553 | 6548 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
6549 | |
6550 | ||
b4c4c7b8 N |
6551 | static int remove_and_add_spares(mddev_t *mddev) |
6552 | { | |
6553 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
6554 | int spares = 0; |
6555 | ||
97e4f42d N |
6556 | mddev->curr_resync_completed = 0; |
6557 | ||
159ec1fc | 6558 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 6559 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 6560 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
6561 | (test_bit(Faulty, &rdev->flags) || |
6562 | ! test_bit(In_sync, &rdev->flags)) && | |
6563 | atomic_read(&rdev->nr_pending)==0) { | |
6564 | if (mddev->pers->hot_remove_disk( | |
6565 | mddev, rdev->raid_disk)==0) { | |
6566 | char nm[20]; | |
6567 | sprintf(nm,"rd%d", rdev->raid_disk); | |
6568 | sysfs_remove_link(&mddev->kobj, nm); | |
6569 | rdev->raid_disk = -1; | |
6570 | } | |
6571 | } | |
6572 | ||
4044ba58 | 6573 | if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { |
159ec1fc | 6574 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 6575 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
6576 | !test_bit(In_sync, &rdev->flags) && |
6577 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 6578 | spares++; |
b4c4c7b8 N |
6579 | if (rdev->raid_disk < 0 |
6580 | && !test_bit(Faulty, &rdev->flags)) { | |
6581 | rdev->recovery_offset = 0; | |
199050ea NB |
6582 | if (mddev->pers-> |
6583 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
6584 | char nm[20]; |
6585 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
6586 | if (sysfs_create_link(&mddev->kobj, |
6587 | &rdev->kobj, nm)) | |
6588 | printk(KERN_WARNING | |
6589 | "md: cannot register " | |
6590 | "%s for %s\n", | |
6591 | nm, mdname(mddev)); | |
b4c4c7b8 N |
6592 | spares++; |
6593 | md_new_event(mddev); | |
6594 | } else | |
6595 | break; | |
6596 | } | |
dfc70645 | 6597 | } |
b4c4c7b8 N |
6598 | } |
6599 | return spares; | |
6600 | } | |
1da177e4 LT |
6601 | /* |
6602 | * This routine is regularly called by all per-raid-array threads to | |
6603 | * deal with generic issues like resync and super-block update. | |
6604 | * Raid personalities that don't have a thread (linear/raid0) do not | |
6605 | * need this as they never do any recovery or update the superblock. | |
6606 | * | |
6607 | * It does not do any resync itself, but rather "forks" off other threads | |
6608 | * to do that as needed. | |
6609 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
6610 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 6611 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
6612 | * and wakeups up this thread which will reap the thread and finish up. |
6613 | * This thread also removes any faulty devices (with nr_pending == 0). | |
6614 | * | |
6615 | * The overall approach is: | |
6616 | * 1/ if the superblock needs updating, update it. | |
6617 | * 2/ If a recovery thread is running, don't do anything else. | |
6618 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
6619 | * 4/ If there are any faulty devices, remove them. | |
6620 | * 5/ If array is degraded, try to add spares devices | |
6621 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
6622 | */ | |
6623 | void md_check_recovery(mddev_t *mddev) | |
6624 | { | |
6625 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6626 | |
6627 | ||
5f40402d | 6628 | if (mddev->bitmap) |
aa5cbd10 | 6629 | bitmap_daemon_work(mddev); |
1da177e4 LT |
6630 | |
6631 | if (mddev->ro) | |
6632 | return; | |
fca4d848 N |
6633 | |
6634 | if (signal_pending(current)) { | |
31a59e34 | 6635 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
6636 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
6637 | mdname(mddev)); | |
6638 | mddev->safemode = 2; | |
6639 | } | |
6640 | flush_signals(current); | |
6641 | } | |
6642 | ||
c89a8eee N |
6643 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
6644 | return; | |
1da177e4 | 6645 | if ( ! ( |
e691063a | 6646 | (mddev->flags && !mddev->external) || |
1da177e4 | 6647 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 6648 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 6649 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
6650 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
6651 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
6652 | )) |
6653 | return; | |
fca4d848 | 6654 | |
df5b89b3 | 6655 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 6656 | int spares = 0; |
fca4d848 | 6657 | |
c89a8eee N |
6658 | if (mddev->ro) { |
6659 | /* Only thing we do on a ro array is remove | |
6660 | * failed devices. | |
6661 | */ | |
6662 | remove_and_add_spares(mddev); | |
6663 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6664 | goto unlock; | |
6665 | } | |
6666 | ||
31a59e34 | 6667 | if (!mddev->external) { |
0fd62b86 | 6668 | int did_change = 0; |
31a59e34 N |
6669 | spin_lock_irq(&mddev->write_lock); |
6670 | if (mddev->safemode && | |
6671 | !atomic_read(&mddev->writes_pending) && | |
6672 | !mddev->in_sync && | |
6673 | mddev->recovery_cp == MaxSector) { | |
6674 | mddev->in_sync = 1; | |
0fd62b86 | 6675 | did_change = 1; |
31a59e34 N |
6676 | if (mddev->persistent) |
6677 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6678 | } | |
6679 | if (mddev->safemode == 1) | |
6680 | mddev->safemode = 0; | |
6681 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 6682 | if (did_change) |
b62b7590 | 6683 | sysfs_notify_dirent(mddev->sysfs_state); |
fca4d848 | 6684 | } |
fca4d848 | 6685 | |
850b2b42 N |
6686 | if (mddev->flags) |
6687 | md_update_sb(mddev, 0); | |
06d91a5f | 6688 | |
159ec1fc | 6689 | list_for_each_entry(rdev, &mddev->disks, same_set) |
52664732 | 6690 | if (test_and_clear_bit(StateChanged, &rdev->flags)) |
3c0ee63a | 6691 | sysfs_notify_dirent(rdev->sysfs_state); |
52664732 | 6692 | |
06d91a5f | 6693 | |
1da177e4 LT |
6694 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
6695 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
6696 | /* resync/recovery still happening */ | |
6697 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6698 | goto unlock; | |
6699 | } | |
6700 | if (mddev->sync_thread) { | |
6701 | /* resync has finished, collect result */ | |
6702 | md_unregister_thread(mddev->sync_thread); | |
6703 | mddev->sync_thread = NULL; | |
56ac36d7 DW |
6704 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
6705 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1da177e4 LT |
6706 | /* success...*/ |
6707 | /* activate any spares */ | |
a99ac971 NB |
6708 | if (mddev->pers->spare_active(mddev)) |
6709 | sysfs_notify(&mddev->kobj, NULL, | |
6710 | "degraded"); | |
1da177e4 | 6711 | } |
cea9c228 N |
6712 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6713 | mddev->pers->finish_reshape) | |
6714 | mddev->pers->finish_reshape(mddev); | |
850b2b42 | 6715 | md_update_sb(mddev, 1); |
41158c7e N |
6716 | |
6717 | /* if array is no-longer degraded, then any saved_raid_disk | |
6718 | * information must be scrapped | |
6719 | */ | |
6720 | if (!mddev->degraded) | |
159ec1fc | 6721 | list_for_each_entry(rdev, &mddev->disks, same_set) |
41158c7e N |
6722 | rdev->saved_raid_disk = -1; |
6723 | ||
1da177e4 LT |
6724 | mddev->recovery = 0; |
6725 | /* flag recovery needed just to double check */ | |
6726 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
0c3573f1 | 6727 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 6728 | md_new_event(mddev); |
1da177e4 LT |
6729 | goto unlock; |
6730 | } | |
72a23c21 NB |
6731 | /* Set RUNNING before clearing NEEDED to avoid |
6732 | * any transients in the value of "sync_action". | |
6733 | */ | |
6734 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6735 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
6736 | /* Clear some bits that don't mean anything, but |
6737 | * might be left set | |
6738 | */ | |
24dd469d N |
6739 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6740 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 6741 | |
5fd6c1dc N |
6742 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6743 | goto unlock; | |
1da177e4 LT |
6744 | /* no recovery is running. |
6745 | * remove any failed drives, then | |
6746 | * add spares if possible. | |
6747 | * Spare are also removed and re-added, to allow | |
6748 | * the personality to fail the re-add. | |
6749 | */ | |
1da177e4 | 6750 | |
b4c4c7b8 | 6751 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
6752 | if (mddev->pers->check_reshape == NULL || |
6753 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
6754 | /* Cannot proceed */ |
6755 | goto unlock; | |
6756 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 6757 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 6758 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
6759 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
6760 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 6761 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 6762 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6763 | } else if (mddev->recovery_cp < MaxSector) { |
6764 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 6765 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6766 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
6767 | /* nothing to be done ... */ | |
1da177e4 | 6768 | goto unlock; |
24dd469d | 6769 | |
1da177e4 | 6770 | if (mddev->pers->sync_request) { |
a654b9d8 N |
6771 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
6772 | /* We are adding a device or devices to an array | |
6773 | * which has the bitmap stored on all devices. | |
6774 | * So make sure all bitmap pages get written | |
6775 | */ | |
6776 | bitmap_write_all(mddev->bitmap); | |
6777 | } | |
1da177e4 LT |
6778 | mddev->sync_thread = md_register_thread(md_do_sync, |
6779 | mddev, | |
0da3c619 | 6780 | "resync"); |
1da177e4 LT |
6781 | if (!mddev->sync_thread) { |
6782 | printk(KERN_ERR "%s: could not start resync" | |
6783 | " thread...\n", | |
6784 | mdname(mddev)); | |
6785 | /* leave the spares where they are, it shouldn't hurt */ | |
6786 | mddev->recovery = 0; | |
d7603b7e | 6787 | } else |
1da177e4 | 6788 | md_wakeup_thread(mddev->sync_thread); |
0c3573f1 | 6789 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 6790 | md_new_event(mddev); |
1da177e4 LT |
6791 | } |
6792 | unlock: | |
72a23c21 NB |
6793 | if (!mddev->sync_thread) { |
6794 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6795 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
6796 | &mddev->recovery)) | |
0c3573f1 N |
6797 | if (mddev->sysfs_action) |
6798 | sysfs_notify_dirent(mddev->sysfs_action); | |
72a23c21 | 6799 | } |
1da177e4 LT |
6800 | mddev_unlock(mddev); |
6801 | } | |
6802 | } | |
6803 | ||
6bfe0b49 DW |
6804 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
6805 | { | |
3c0ee63a | 6806 | sysfs_notify_dirent(rdev->sysfs_state); |
6bfe0b49 DW |
6807 | wait_event_timeout(rdev->blocked_wait, |
6808 | !test_bit(Blocked, &rdev->flags), | |
6809 | msecs_to_jiffies(5000)); | |
6810 | rdev_dec_pending(rdev, mddev); | |
6811 | } | |
6812 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
6813 | ||
75c96f85 AB |
6814 | static int md_notify_reboot(struct notifier_block *this, |
6815 | unsigned long code, void *x) | |
1da177e4 LT |
6816 | { |
6817 | struct list_head *tmp; | |
6818 | mddev_t *mddev; | |
6819 | ||
6820 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
6821 | ||
6822 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
6823 | ||
29ac4aa3 | 6824 | for_each_mddev(mddev, tmp) |
c71d4887 | 6825 | if (mddev_trylock(mddev)) { |
2b25000b N |
6826 | /* Force a switch to readonly even array |
6827 | * appears to still be in use. Hence | |
6828 | * the '100'. | |
6829 | */ | |
d710e138 | 6830 | do_md_stop(mddev, 1, 100); |
c71d4887 NB |
6831 | mddev_unlock(mddev); |
6832 | } | |
1da177e4 LT |
6833 | /* |
6834 | * certain more exotic SCSI devices are known to be | |
6835 | * volatile wrt too early system reboots. While the | |
6836 | * right place to handle this issue is the given | |
6837 | * driver, we do want to have a safe RAID driver ... | |
6838 | */ | |
6839 | mdelay(1000*1); | |
6840 | } | |
6841 | return NOTIFY_DONE; | |
6842 | } | |
6843 | ||
75c96f85 | 6844 | static struct notifier_block md_notifier = { |
1da177e4 LT |
6845 | .notifier_call = md_notify_reboot, |
6846 | .next = NULL, | |
6847 | .priority = INT_MAX, /* before any real devices */ | |
6848 | }; | |
6849 | ||
6850 | static void md_geninit(void) | |
6851 | { | |
1da177e4 LT |
6852 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
6853 | ||
c7705f34 | 6854 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
6855 | } |
6856 | ||
75c96f85 | 6857 | static int __init md_init(void) |
1da177e4 | 6858 | { |
3dbd8c2e | 6859 | if (register_blkdev(MD_MAJOR, "md")) |
1da177e4 LT |
6860 | return -1; |
6861 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
3dbd8c2e | 6862 | unregister_blkdev(MD_MAJOR, "md"); |
1da177e4 LT |
6863 | return -1; |
6864 | } | |
3dbd8c2e | 6865 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
6866 | md_probe, NULL, NULL); |
6867 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
6868 | md_probe, NULL, NULL); |
6869 | ||
1da177e4 | 6870 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 6871 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
6872 | |
6873 | md_geninit(); | |
d710e138 | 6874 | return 0; |
1da177e4 LT |
6875 | } |
6876 | ||
6877 | ||
6878 | #ifndef MODULE | |
6879 | ||
6880 | /* | |
6881 | * Searches all registered partitions for autorun RAID arrays | |
6882 | * at boot time. | |
6883 | */ | |
4d936ec1 ME |
6884 | |
6885 | static LIST_HEAD(all_detected_devices); | |
6886 | struct detected_devices_node { | |
6887 | struct list_head list; | |
6888 | dev_t dev; | |
6889 | }; | |
1da177e4 LT |
6890 | |
6891 | void md_autodetect_dev(dev_t dev) | |
6892 | { | |
4d936ec1 ME |
6893 | struct detected_devices_node *node_detected_dev; |
6894 | ||
6895 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
6896 | if (node_detected_dev) { | |
6897 | node_detected_dev->dev = dev; | |
6898 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
6899 | } else { | |
6900 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
6901 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
6902 | } | |
1da177e4 LT |
6903 | } |
6904 | ||
6905 | ||
6906 | static void autostart_arrays(int part) | |
6907 | { | |
6908 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
6909 | struct detected_devices_node *node_detected_dev; |
6910 | dev_t dev; | |
6911 | int i_scanned, i_passed; | |
1da177e4 | 6912 | |
4d936ec1 ME |
6913 | i_scanned = 0; |
6914 | i_passed = 0; | |
1da177e4 | 6915 | |
4d936ec1 | 6916 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 6917 | |
4d936ec1 ME |
6918 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
6919 | i_scanned++; | |
6920 | node_detected_dev = list_entry(all_detected_devices.next, | |
6921 | struct detected_devices_node, list); | |
6922 | list_del(&node_detected_dev->list); | |
6923 | dev = node_detected_dev->dev; | |
6924 | kfree(node_detected_dev); | |
df968c4e | 6925 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
6926 | if (IS_ERR(rdev)) |
6927 | continue; | |
6928 | ||
b2d444d7 | 6929 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6930 | MD_BUG(); |
6931 | continue; | |
6932 | } | |
d0fae18f | 6933 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 6934 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 6935 | i_passed++; |
1da177e4 | 6936 | } |
4d936ec1 ME |
6937 | |
6938 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
6939 | i_scanned, i_passed); | |
1da177e4 LT |
6940 | |
6941 | autorun_devices(part); | |
6942 | } | |
6943 | ||
fdee8ae4 | 6944 | #endif /* !MODULE */ |
1da177e4 LT |
6945 | |
6946 | static __exit void md_exit(void) | |
6947 | { | |
6948 | mddev_t *mddev; | |
6949 | struct list_head *tmp; | |
8ab5e4c1 | 6950 | |
3dbd8c2e | 6951 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 6952 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 6953 | |
3dbd8c2e | 6954 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
6955 | unregister_blkdev(mdp_major, "mdp"); |
6956 | unregister_reboot_notifier(&md_notifier); | |
6957 | unregister_sysctl_table(raid_table_header); | |
6958 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 6959 | for_each_mddev(mddev, tmp) { |
1da177e4 | 6960 | export_array(mddev); |
d3374825 | 6961 | mddev->hold_active = 0; |
1da177e4 LT |
6962 | } |
6963 | } | |
6964 | ||
685784aa | 6965 | subsys_initcall(md_init); |
1da177e4 LT |
6966 | module_exit(md_exit) |
6967 | ||
f91de92e N |
6968 | static int get_ro(char *buffer, struct kernel_param *kp) |
6969 | { | |
6970 | return sprintf(buffer, "%d", start_readonly); | |
6971 | } | |
6972 | static int set_ro(const char *val, struct kernel_param *kp) | |
6973 | { | |
6974 | char *e; | |
6975 | int num = simple_strtoul(val, &e, 10); | |
6976 | if (*val && (*e == '\0' || *e == '\n')) { | |
6977 | start_readonly = num; | |
4dbcdc75 | 6978 | return 0; |
f91de92e N |
6979 | } |
6980 | return -EINVAL; | |
6981 | } | |
6982 | ||
80ca3a44 N |
6983 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
6984 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 6985 | |
efeb53c0 | 6986 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 6987 | |
1da177e4 LT |
6988 | EXPORT_SYMBOL(register_md_personality); |
6989 | EXPORT_SYMBOL(unregister_md_personality); | |
6990 | EXPORT_SYMBOL(md_error); | |
6991 | EXPORT_SYMBOL(md_done_sync); | |
6992 | EXPORT_SYMBOL(md_write_start); | |
6993 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
6994 | EXPORT_SYMBOL(md_register_thread); |
6995 | EXPORT_SYMBOL(md_unregister_thread); | |
6996 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
6997 | EXPORT_SYMBOL(md_check_recovery); |
6998 | MODULE_LICENSE("GPL"); | |
aa1595e9 | 6999 | MODULE_ALIAS("md"); |
72008652 | 7000 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |