]>
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> |
e7d2860b | 42 | #include <linux/string.h> |
fb4d8c76 N |
43 | #include <linux/hdreg.h> |
44 | #include <linux/proc_fs.h> | |
45 | #include <linux/random.h> | |
46 | #include <linux/reboot.h> | |
32a7627c | 47 | #include <linux/file.h> |
aa98aa31 | 48 | #include <linux/compat.h> |
25570727 | 49 | #include <linux/delay.h> |
bff61975 N |
50 | #include <linux/raid/md_p.h> |
51 | #include <linux/raid/md_u.h> | |
43b2e5d8 | 52 | #include "md.h" |
ef740c37 | 53 | #include "bitmap.h" |
1da177e4 LT |
54 | |
55 | #define DEBUG 0 | |
56 | #define dprintk(x...) ((void)(DEBUG && printk(x))) | |
57 | ||
58 | ||
59 | #ifndef MODULE | |
d710e138 | 60 | static void autostart_arrays(int part); |
1da177e4 LT |
61 | #endif |
62 | ||
2604b703 | 63 | static LIST_HEAD(pers_list); |
1da177e4 LT |
64 | static DEFINE_SPINLOCK(pers_lock); |
65 | ||
5e56341d AB |
66 | static void md_print_devices(void); |
67 | ||
90b08710 BS |
68 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
69 | ||
5e56341d AB |
70 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } |
71 | ||
1e50915f RB |
72 | /* |
73 | * Default number of read corrections we'll attempt on an rdev | |
74 | * before ejecting it from the array. We divide the read error | |
75 | * count by 2 for every hour elapsed between read errors. | |
76 | */ | |
77 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
78 | /* |
79 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
80 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
81 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 82 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
83 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
84 | * speed limit - in case reconstruction slows down your system despite | |
85 | * idle IO detection. | |
86 | * | |
87 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 88 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
89 | */ |
90 | ||
91 | static int sysctl_speed_limit_min = 1000; | |
92 | static int sysctl_speed_limit_max = 200000; | |
88202a0c N |
93 | static inline int speed_min(mddev_t *mddev) |
94 | { | |
95 | return mddev->sync_speed_min ? | |
96 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
97 | } | |
98 | ||
99 | static inline int speed_max(mddev_t *mddev) | |
100 | { | |
101 | return mddev->sync_speed_max ? | |
102 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
103 | } | |
1da177e4 LT |
104 | |
105 | static struct ctl_table_header *raid_table_header; | |
106 | ||
107 | static ctl_table raid_table[] = { | |
108 | { | |
1da177e4 LT |
109 | .procname = "speed_limit_min", |
110 | .data = &sysctl_speed_limit_min, | |
111 | .maxlen = sizeof(int), | |
80ca3a44 | 112 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 113 | .proc_handler = proc_dointvec, |
1da177e4 LT |
114 | }, |
115 | { | |
1da177e4 LT |
116 | .procname = "speed_limit_max", |
117 | .data = &sysctl_speed_limit_max, | |
118 | .maxlen = sizeof(int), | |
80ca3a44 | 119 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 120 | .proc_handler = proc_dointvec, |
1da177e4 | 121 | }, |
894d2491 | 122 | { } |
1da177e4 LT |
123 | }; |
124 | ||
125 | static ctl_table raid_dir_table[] = { | |
126 | { | |
1da177e4 LT |
127 | .procname = "raid", |
128 | .maxlen = 0, | |
80ca3a44 | 129 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
130 | .child = raid_table, |
131 | }, | |
894d2491 | 132 | { } |
1da177e4 LT |
133 | }; |
134 | ||
135 | static ctl_table raid_root_table[] = { | |
136 | { | |
1da177e4 LT |
137 | .procname = "dev", |
138 | .maxlen = 0, | |
139 | .mode = 0555, | |
140 | .child = raid_dir_table, | |
141 | }, | |
894d2491 | 142 | { } |
1da177e4 LT |
143 | }; |
144 | ||
83d5cde4 | 145 | static const struct block_device_operations md_fops; |
1da177e4 | 146 | |
f91de92e N |
147 | static int start_readonly; |
148 | ||
d7603b7e N |
149 | /* |
150 | * We have a system wide 'event count' that is incremented | |
151 | * on any 'interesting' event, and readers of /proc/mdstat | |
152 | * can use 'poll' or 'select' to find out when the event | |
153 | * count increases. | |
154 | * | |
155 | * Events are: | |
156 | * start array, stop array, error, add device, remove device, | |
157 | * start build, activate spare | |
158 | */ | |
2989ddbd | 159 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 160 | static atomic_t md_event_count; |
29269553 | 161 | void md_new_event(mddev_t *mddev) |
d7603b7e N |
162 | { |
163 | atomic_inc(&md_event_count); | |
164 | wake_up(&md_event_waiters); | |
165 | } | |
29269553 | 166 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 167 | |
c331eb04 N |
168 | /* Alternate version that can be called from interrupts |
169 | * when calling sysfs_notify isn't needed. | |
170 | */ | |
05381954 | 171 | static void md_new_event_inintr(mddev_t *mddev) |
c331eb04 N |
172 | { |
173 | atomic_inc(&md_event_count); | |
174 | wake_up(&md_event_waiters); | |
175 | } | |
176 | ||
1da177e4 LT |
177 | /* |
178 | * Enables to iterate over all existing md arrays | |
179 | * all_mddevs_lock protects this list. | |
180 | */ | |
181 | static LIST_HEAD(all_mddevs); | |
182 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
183 | ||
184 | ||
185 | /* | |
186 | * iterates through all used mddevs in the system. | |
187 | * We take care to grab the all_mddevs_lock whenever navigating | |
188 | * the list, and to always hold a refcount when unlocked. | |
189 | * Any code which breaks out of this loop while own | |
190 | * a reference to the current mddev and must mddev_put it. | |
191 | */ | |
29ac4aa3 | 192 | #define for_each_mddev(mddev,tmp) \ |
1da177e4 LT |
193 | \ |
194 | for (({ spin_lock(&all_mddevs_lock); \ | |
195 | tmp = all_mddevs.next; \ | |
196 | mddev = NULL;}); \ | |
197 | ({ if (tmp != &all_mddevs) \ | |
198 | mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ | |
199 | spin_unlock(&all_mddevs_lock); \ | |
200 | if (mddev) mddev_put(mddev); \ | |
201 | mddev = list_entry(tmp, mddev_t, all_mddevs); \ | |
202 | tmp != &all_mddevs;}); \ | |
203 | ({ spin_lock(&all_mddevs_lock); \ | |
204 | tmp = tmp->next;}) \ | |
205 | ) | |
206 | ||
207 | ||
409c57f3 N |
208 | /* Rather than calling directly into the personality make_request function, |
209 | * IO requests come here first so that we can check if the device is | |
210 | * being suspended pending a reconfiguration. | |
211 | * We hold a refcount over the call to ->make_request. By the time that | |
212 | * call has finished, the bio has been linked into some internal structure | |
213 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
214 | */ | |
215 | static int md_make_request(struct request_queue *q, struct bio *bio) | |
1da177e4 | 216 | { |
49077326 | 217 | const int rw = bio_data_dir(bio); |
409c57f3 N |
218 | mddev_t *mddev = q->queuedata; |
219 | int rv; | |
49077326 N |
220 | int cpu; |
221 | ||
409c57f3 N |
222 | if (mddev == NULL || mddev->pers == NULL) { |
223 | bio_io_error(bio); | |
224 | return 0; | |
225 | } | |
226 | rcu_read_lock(); | |
a2826aa9 | 227 | if (mddev->suspended || mddev->barrier) { |
409c57f3 N |
228 | DEFINE_WAIT(__wait); |
229 | for (;;) { | |
230 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
231 | TASK_UNINTERRUPTIBLE); | |
a2826aa9 | 232 | if (!mddev->suspended && !mddev->barrier) |
409c57f3 N |
233 | break; |
234 | rcu_read_unlock(); | |
235 | schedule(); | |
236 | rcu_read_lock(); | |
237 | } | |
238 | finish_wait(&mddev->sb_wait, &__wait); | |
239 | } | |
240 | atomic_inc(&mddev->active_io); | |
241 | rcu_read_unlock(); | |
49077326 | 242 | |
21a52c6d | 243 | rv = mddev->pers->make_request(mddev, bio); |
49077326 N |
244 | |
245 | cpu = part_stat_lock(); | |
246 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
247 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
248 | bio_sectors(bio)); | |
249 | part_stat_unlock(); | |
250 | ||
409c57f3 N |
251 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
252 | wake_up(&mddev->sb_wait); | |
253 | ||
254 | return rv; | |
255 | } | |
256 | ||
9e35b99c N |
257 | /* mddev_suspend makes sure no new requests are submitted |
258 | * to the device, and that any requests that have been submitted | |
259 | * are completely handled. | |
260 | * Once ->stop is called and completes, the module will be completely | |
261 | * unused. | |
262 | */ | |
409c57f3 N |
263 | static void mddev_suspend(mddev_t *mddev) |
264 | { | |
265 | BUG_ON(mddev->suspended); | |
266 | mddev->suspended = 1; | |
267 | synchronize_rcu(); | |
268 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
269 | mddev->pers->quiesce(mddev, 1); | |
409c57f3 N |
270 | } |
271 | ||
272 | static void mddev_resume(mddev_t *mddev) | |
273 | { | |
274 | mddev->suspended = 0; | |
275 | wake_up(&mddev->sb_wait); | |
276 | mddev->pers->quiesce(mddev, 0); | |
1da177e4 LT |
277 | } |
278 | ||
3fa841d7 N |
279 | int mddev_congested(mddev_t *mddev, int bits) |
280 | { | |
a2826aa9 N |
281 | if (mddev->barrier) |
282 | return 1; | |
3fa841d7 N |
283 | return mddev->suspended; |
284 | } | |
285 | EXPORT_SYMBOL(mddev_congested); | |
286 | ||
a2826aa9 N |
287 | /* |
288 | * Generic barrier handling for md | |
289 | */ | |
290 | ||
291 | #define POST_REQUEST_BARRIER ((void*)1) | |
292 | ||
293 | static void md_end_barrier(struct bio *bio, int err) | |
294 | { | |
295 | mdk_rdev_t *rdev = bio->bi_private; | |
296 | mddev_t *mddev = rdev->mddev; | |
297 | if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER) | |
298 | set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags); | |
299 | ||
300 | rdev_dec_pending(rdev, mddev); | |
301 | ||
302 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
303 | if (mddev->barrier == POST_REQUEST_BARRIER) { | |
304 | /* This was a post-request barrier */ | |
305 | mddev->barrier = NULL; | |
306 | wake_up(&mddev->sb_wait); | |
307 | } else | |
308 | /* The pre-request barrier has finished */ | |
309 | schedule_work(&mddev->barrier_work); | |
310 | } | |
311 | bio_put(bio); | |
312 | } | |
313 | ||
314 | static void submit_barriers(mddev_t *mddev) | |
315 | { | |
316 | mdk_rdev_t *rdev; | |
317 | ||
318 | rcu_read_lock(); | |
319 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
320 | if (rdev->raid_disk >= 0 && | |
321 | !test_bit(Faulty, &rdev->flags)) { | |
322 | /* Take two references, one is dropped | |
323 | * when request finishes, one after | |
324 | * we reclaim rcu_read_lock | |
325 | */ | |
326 | struct bio *bi; | |
327 | atomic_inc(&rdev->nr_pending); | |
328 | atomic_inc(&rdev->nr_pending); | |
329 | rcu_read_unlock(); | |
330 | bi = bio_alloc(GFP_KERNEL, 0); | |
331 | bi->bi_end_io = md_end_barrier; | |
332 | bi->bi_private = rdev; | |
333 | bi->bi_bdev = rdev->bdev; | |
334 | atomic_inc(&mddev->flush_pending); | |
335 | submit_bio(WRITE_BARRIER, bi); | |
336 | rcu_read_lock(); | |
337 | rdev_dec_pending(rdev, mddev); | |
338 | } | |
339 | rcu_read_unlock(); | |
340 | } | |
341 | ||
342 | static void md_submit_barrier(struct work_struct *ws) | |
343 | { | |
344 | mddev_t *mddev = container_of(ws, mddev_t, barrier_work); | |
345 | struct bio *bio = mddev->barrier; | |
346 | ||
347 | atomic_set(&mddev->flush_pending, 1); | |
348 | ||
349 | if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags)) | |
350 | bio_endio(bio, -EOPNOTSUPP); | |
351 | else if (bio->bi_size == 0) | |
352 | /* an empty barrier - all done */ | |
353 | bio_endio(bio, 0); | |
354 | else { | |
355 | bio->bi_rw &= ~(1<<BIO_RW_BARRIER); | |
21a52c6d | 356 | if (mddev->pers->make_request(mddev, bio)) |
a2826aa9 N |
357 | generic_make_request(bio); |
358 | mddev->barrier = POST_REQUEST_BARRIER; | |
359 | submit_barriers(mddev); | |
360 | } | |
361 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
362 | mddev->barrier = NULL; | |
363 | wake_up(&mddev->sb_wait); | |
364 | } | |
365 | } | |
366 | ||
367 | void md_barrier_request(mddev_t *mddev, struct bio *bio) | |
368 | { | |
369 | spin_lock_irq(&mddev->write_lock); | |
370 | wait_event_lock_irq(mddev->sb_wait, | |
371 | !mddev->barrier, | |
372 | mddev->write_lock, /*nothing*/); | |
373 | mddev->barrier = bio; | |
374 | spin_unlock_irq(&mddev->write_lock); | |
375 | ||
376 | atomic_set(&mddev->flush_pending, 1); | |
377 | INIT_WORK(&mddev->barrier_work, md_submit_barrier); | |
378 | ||
379 | submit_barriers(mddev); | |
380 | ||
381 | if (atomic_dec_and_test(&mddev->flush_pending)) | |
382 | schedule_work(&mddev->barrier_work); | |
383 | } | |
384 | EXPORT_SYMBOL(md_barrier_request); | |
409c57f3 | 385 | |
1da177e4 LT |
386 | static inline mddev_t *mddev_get(mddev_t *mddev) |
387 | { | |
388 | atomic_inc(&mddev->active); | |
389 | return mddev; | |
390 | } | |
391 | ||
5fd3a17e | 392 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 393 | |
1da177e4 LT |
394 | static void mddev_put(mddev_t *mddev) |
395 | { | |
396 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
397 | return; | |
d3374825 | 398 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
399 | mddev->ctime == 0 && !mddev->hold_active) { |
400 | /* Array is not configured at all, and not held active, | |
401 | * so destroy it */ | |
1da177e4 | 402 | list_del(&mddev->all_mddevs); |
d3374825 N |
403 | if (mddev->gendisk) { |
404 | /* we did a probe so need to clean up. | |
405 | * Call schedule_work inside the spinlock | |
406 | * so that flush_scheduled_work() after | |
407 | * mddev_find will succeed in waiting for the | |
408 | * work to be done. | |
409 | */ | |
410 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
411 | schedule_work(&mddev->del_work); | |
412 | } else | |
413 | kfree(mddev); | |
414 | } | |
415 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
416 | } |
417 | ||
fafd7fb0 N |
418 | static void mddev_init(mddev_t *mddev) |
419 | { | |
420 | mutex_init(&mddev->open_mutex); | |
421 | mutex_init(&mddev->reconfig_mutex); | |
422 | mutex_init(&mddev->bitmap_info.mutex); | |
423 | INIT_LIST_HEAD(&mddev->disks); | |
424 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
425 | init_timer(&mddev->safemode_timer); | |
426 | atomic_set(&mddev->active, 1); | |
427 | atomic_set(&mddev->openers, 0); | |
428 | atomic_set(&mddev->active_io, 0); | |
429 | spin_lock_init(&mddev->write_lock); | |
430 | atomic_set(&mddev->flush_pending, 0); | |
431 | init_waitqueue_head(&mddev->sb_wait); | |
432 | init_waitqueue_head(&mddev->recovery_wait); | |
433 | mddev->reshape_position = MaxSector; | |
434 | mddev->resync_min = 0; | |
435 | mddev->resync_max = MaxSector; | |
436 | mddev->level = LEVEL_NONE; | |
437 | } | |
438 | ||
1da177e4 LT |
439 | static mddev_t * mddev_find(dev_t unit) |
440 | { | |
441 | mddev_t *mddev, *new = NULL; | |
442 | ||
443 | retry: | |
444 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
445 | |
446 | if (unit) { | |
447 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
448 | if (mddev->unit == unit) { | |
449 | mddev_get(mddev); | |
450 | spin_unlock(&all_mddevs_lock); | |
451 | kfree(new); | |
452 | return mddev; | |
453 | } | |
454 | ||
455 | if (new) { | |
456 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 457 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
458 | new->hold_active = UNTIL_IOCTL; |
459 | return new; | |
1da177e4 | 460 | } |
efeb53c0 N |
461 | } else if (new) { |
462 | /* find an unused unit number */ | |
463 | static int next_minor = 512; | |
464 | int start = next_minor; | |
465 | int is_free = 0; | |
466 | int dev = 0; | |
467 | while (!is_free) { | |
468 | dev = MKDEV(MD_MAJOR, next_minor); | |
469 | next_minor++; | |
470 | if (next_minor > MINORMASK) | |
471 | next_minor = 0; | |
472 | if (next_minor == start) { | |
473 | /* Oh dear, all in use. */ | |
474 | spin_unlock(&all_mddevs_lock); | |
475 | kfree(new); | |
476 | return NULL; | |
477 | } | |
478 | ||
479 | is_free = 1; | |
480 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
481 | if (mddev->unit == dev) { | |
482 | is_free = 0; | |
483 | break; | |
484 | } | |
485 | } | |
486 | new->unit = dev; | |
487 | new->md_minor = MINOR(dev); | |
488 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
489 | list_add(&new->all_mddevs, &all_mddevs); |
490 | spin_unlock(&all_mddevs_lock); | |
491 | return new; | |
492 | } | |
493 | spin_unlock(&all_mddevs_lock); | |
494 | ||
9ffae0cf | 495 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
496 | if (!new) |
497 | return NULL; | |
498 | ||
1da177e4 LT |
499 | new->unit = unit; |
500 | if (MAJOR(unit) == MD_MAJOR) | |
501 | new->md_minor = MINOR(unit); | |
502 | else | |
503 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
504 | ||
fafd7fb0 | 505 | mddev_init(new); |
1da177e4 | 506 | |
1da177e4 LT |
507 | goto retry; |
508 | } | |
509 | ||
510 | static inline int mddev_lock(mddev_t * mddev) | |
511 | { | |
df5b89b3 | 512 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
513 | } |
514 | ||
b522adcd DW |
515 | static inline int mddev_is_locked(mddev_t *mddev) |
516 | { | |
517 | return mutex_is_locked(&mddev->reconfig_mutex); | |
518 | } | |
519 | ||
1da177e4 LT |
520 | static inline int mddev_trylock(mddev_t * mddev) |
521 | { | |
df5b89b3 | 522 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
523 | } |
524 | ||
b6eb127d N |
525 | static struct attribute_group md_redundancy_group; |
526 | ||
a64c876f | 527 | static void mddev_unlock(mddev_t * mddev) |
1da177e4 | 528 | { |
a64c876f | 529 | if (mddev->to_remove) { |
b6eb127d N |
530 | /* These cannot be removed under reconfig_mutex as |
531 | * an access to the files will try to take reconfig_mutex | |
532 | * while holding the file unremovable, which leads to | |
533 | * a deadlock. | |
534 | * So hold open_mutex instead - we are allowed to take | |
535 | * it while holding reconfig_mutex, and md_run can | |
536 | * use it to wait for the remove to complete. | |
537 | */ | |
a64c876f N |
538 | struct attribute_group *to_remove = mddev->to_remove; |
539 | mddev->to_remove = NULL; | |
b6eb127d N |
540 | mutex_lock(&mddev->open_mutex); |
541 | mutex_unlock(&mddev->reconfig_mutex); | |
542 | ||
a64c876f N |
543 | if (to_remove != &md_redundancy_group) |
544 | sysfs_remove_group(&mddev->kobj, to_remove); | |
545 | if (mddev->pers == NULL || | |
546 | mddev->pers->sync_request == NULL) { | |
547 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
548 | if (mddev->sysfs_action) | |
549 | sysfs_put(mddev->sysfs_action); | |
550 | mddev->sysfs_action = NULL; | |
551 | } | |
b6eb127d N |
552 | mutex_unlock(&mddev->open_mutex); |
553 | } else | |
554 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 555 | |
005eca5e | 556 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
557 | } |
558 | ||
2989ddbd | 559 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 560 | { |
159ec1fc | 561 | mdk_rdev_t *rdev; |
1da177e4 | 562 | |
159ec1fc | 563 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
564 | if (rdev->desc_nr == nr) |
565 | return rdev; | |
159ec1fc | 566 | |
1da177e4 LT |
567 | return NULL; |
568 | } | |
569 | ||
570 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
571 | { | |
1da177e4 LT |
572 | mdk_rdev_t *rdev; |
573 | ||
159ec1fc | 574 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
575 | if (rdev->bdev->bd_dev == dev) |
576 | return rdev; | |
159ec1fc | 577 | |
1da177e4 LT |
578 | return NULL; |
579 | } | |
580 | ||
d9d166c2 | 581 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
582 | { |
583 | struct mdk_personality *pers; | |
d9d166c2 N |
584 | list_for_each_entry(pers, &pers_list, list) { |
585 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 586 | return pers; |
d9d166c2 N |
587 | if (strcmp(pers->name, clevel)==0) |
588 | return pers; | |
589 | } | |
2604b703 N |
590 | return NULL; |
591 | } | |
592 | ||
b73df2d3 | 593 | /* return the offset of the super block in 512byte sectors */ |
77933d72 | 594 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 | 595 | { |
b73df2d3 AN |
596 | sector_t num_sectors = bdev->bd_inode->i_size / 512; |
597 | return MD_NEW_SIZE_SECTORS(num_sectors); | |
1da177e4 LT |
598 | } |
599 | ||
1da177e4 LT |
600 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
601 | { | |
602 | if (rdev->sb_page) | |
603 | MD_BUG(); | |
604 | ||
605 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
606 | if (!rdev->sb_page) { | |
607 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 608 | return -ENOMEM; |
1da177e4 LT |
609 | } |
610 | ||
611 | return 0; | |
612 | } | |
613 | ||
614 | static void free_disk_sb(mdk_rdev_t * rdev) | |
615 | { | |
616 | if (rdev->sb_page) { | |
2d1f3b5d | 617 | put_page(rdev->sb_page); |
1da177e4 LT |
618 | rdev->sb_loaded = 0; |
619 | rdev->sb_page = NULL; | |
0f420358 | 620 | rdev->sb_start = 0; |
dd8ac336 | 621 | rdev->sectors = 0; |
1da177e4 LT |
622 | } |
623 | } | |
624 | ||
625 | ||
6712ecf8 | 626 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
627 | { |
628 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 629 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 630 | |
3a0f5bbb N |
631 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
632 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
633 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
634 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 635 | md_error(mddev, rdev); |
3a0f5bbb | 636 | } |
7bfa19f2 | 637 | |
a9701a30 N |
638 | if (atomic_dec_and_test(&mddev->pending_writes)) |
639 | wake_up(&mddev->sb_wait); | |
f8b58edf | 640 | bio_put(bio); |
7bfa19f2 N |
641 | } |
642 | ||
6712ecf8 | 643 | static void super_written_barrier(struct bio *bio, int error) |
a9701a30 N |
644 | { |
645 | struct bio *bio2 = bio->bi_private; | |
646 | mdk_rdev_t *rdev = bio2->bi_private; | |
647 | mddev_t *mddev = rdev->mddev; | |
a9701a30 N |
648 | |
649 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
650 | error == -EOPNOTSUPP) { | |
651 | unsigned long flags; | |
652 | /* barriers don't appear to be supported :-( */ | |
653 | set_bit(BarriersNotsupp, &rdev->flags); | |
654 | mddev->barriers_work = 0; | |
655 | spin_lock_irqsave(&mddev->write_lock, flags); | |
656 | bio2->bi_next = mddev->biolist; | |
657 | mddev->biolist = bio2; | |
658 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
659 | wake_up(&mddev->sb_wait); | |
660 | bio_put(bio); | |
6712ecf8 N |
661 | } else { |
662 | bio_put(bio2); | |
663 | bio->bi_private = rdev; | |
664 | super_written(bio, error); | |
a9701a30 | 665 | } |
a9701a30 N |
666 | } |
667 | ||
7bfa19f2 N |
668 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
669 | sector_t sector, int size, struct page *page) | |
670 | { | |
671 | /* write first size bytes of page to sector of rdev | |
672 | * Increment mddev->pending_writes before returning | |
673 | * and decrement it on completion, waking up sb_wait | |
674 | * if zero is reached. | |
675 | * If an error occurred, call md_error | |
a9701a30 N |
676 | * |
677 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
678 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
679 | */ |
680 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
93dbb393 | 681 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG); |
7bfa19f2 N |
682 | |
683 | bio->bi_bdev = rdev->bdev; | |
684 | bio->bi_sector = sector; | |
685 | bio_add_page(bio, page, size, 0); | |
686 | bio->bi_private = rdev; | |
687 | bio->bi_end_io = super_written; | |
a9701a30 N |
688 | bio->bi_rw = rw; |
689 | ||
7bfa19f2 | 690 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
691 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
692 | struct bio *rbio; | |
693 | rw |= (1<<BIO_RW_BARRIER); | |
694 | rbio = bio_clone(bio, GFP_NOIO); | |
695 | rbio->bi_private = bio; | |
696 | rbio->bi_end_io = super_written_barrier; | |
697 | submit_bio(rw, rbio); | |
698 | } else | |
699 | submit_bio(rw, bio); | |
700 | } | |
701 | ||
702 | void md_super_wait(mddev_t *mddev) | |
703 | { | |
704 | /* wait for all superblock writes that were scheduled to complete. | |
705 | * if any had to be retried (due to BARRIER problems), retry them | |
706 | */ | |
707 | DEFINE_WAIT(wq); | |
708 | for(;;) { | |
709 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
710 | if (atomic_read(&mddev->pending_writes)==0) | |
711 | break; | |
712 | while (mddev->biolist) { | |
713 | struct bio *bio; | |
714 | spin_lock_irq(&mddev->write_lock); | |
715 | bio = mddev->biolist; | |
716 | mddev->biolist = bio->bi_next ; | |
717 | bio->bi_next = NULL; | |
718 | spin_unlock_irq(&mddev->write_lock); | |
719 | submit_bio(bio->bi_rw, bio); | |
720 | } | |
721 | schedule(); | |
722 | } | |
723 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
724 | } |
725 | ||
6712ecf8 | 726 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 727 | { |
1da177e4 | 728 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
729 | } |
730 | ||
a654b9d8 | 731 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
732 | struct page *page, int rw) |
733 | { | |
baaa2c51 | 734 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
735 | struct completion event; |
736 | int ret; | |
737 | ||
93dbb393 | 738 | rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG); |
1da177e4 LT |
739 | |
740 | bio->bi_bdev = bdev; | |
741 | bio->bi_sector = sector; | |
742 | bio_add_page(bio, page, size, 0); | |
743 | init_completion(&event); | |
744 | bio->bi_private = &event; | |
745 | bio->bi_end_io = bi_complete; | |
746 | submit_bio(rw, bio); | |
747 | wait_for_completion(&event); | |
748 | ||
749 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
750 | bio_put(bio); | |
751 | return ret; | |
752 | } | |
a8745db2 | 753 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 754 | |
0002b271 | 755 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
756 | { |
757 | char b[BDEVNAME_SIZE]; | |
758 | if (!rdev->sb_page) { | |
759 | MD_BUG(); | |
760 | return -EINVAL; | |
761 | } | |
762 | if (rdev->sb_loaded) | |
763 | return 0; | |
764 | ||
765 | ||
0f420358 | 766 | if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ)) |
1da177e4 LT |
767 | goto fail; |
768 | rdev->sb_loaded = 1; | |
769 | return 0; | |
770 | ||
771 | fail: | |
772 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
773 | bdevname(rdev->bdev,b)); | |
774 | return -EINVAL; | |
775 | } | |
776 | ||
777 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
778 | { | |
05710466 AN |
779 | return sb1->set_uuid0 == sb2->set_uuid0 && |
780 | sb1->set_uuid1 == sb2->set_uuid1 && | |
781 | sb1->set_uuid2 == sb2->set_uuid2 && | |
782 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
783 | } |
784 | ||
1da177e4 LT |
785 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
786 | { | |
787 | int ret; | |
788 | mdp_super_t *tmp1, *tmp2; | |
789 | ||
790 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
791 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
792 | ||
793 | if (!tmp1 || !tmp2) { | |
794 | ret = 0; | |
35020f1a | 795 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
796 | goto abort; |
797 | } | |
798 | ||
799 | *tmp1 = *sb1; | |
800 | *tmp2 = *sb2; | |
801 | ||
802 | /* | |
803 | * nr_disks is not constant | |
804 | */ | |
805 | tmp1->nr_disks = 0; | |
806 | tmp2->nr_disks = 0; | |
807 | ||
ce0c8e05 | 808 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 809 | abort: |
990a8baf JJ |
810 | kfree(tmp1); |
811 | kfree(tmp2); | |
1da177e4 LT |
812 | return ret; |
813 | } | |
814 | ||
4d167f09 N |
815 | |
816 | static u32 md_csum_fold(u32 csum) | |
817 | { | |
818 | csum = (csum & 0xffff) + (csum >> 16); | |
819 | return (csum & 0xffff) + (csum >> 16); | |
820 | } | |
821 | ||
1da177e4 LT |
822 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
823 | { | |
4d167f09 N |
824 | u64 newcsum = 0; |
825 | u32 *sb32 = (u32*)sb; | |
826 | int i; | |
1da177e4 LT |
827 | unsigned int disk_csum, csum; |
828 | ||
829 | disk_csum = sb->sb_csum; | |
830 | sb->sb_csum = 0; | |
4d167f09 N |
831 | |
832 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
833 | newcsum += sb32[i]; | |
834 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
835 | ||
836 | ||
837 | #ifdef CONFIG_ALPHA | |
838 | /* This used to use csum_partial, which was wrong for several | |
839 | * reasons including that different results are returned on | |
840 | * different architectures. It isn't critical that we get exactly | |
841 | * the same return value as before (we always csum_fold before | |
842 | * testing, and that removes any differences). However as we | |
843 | * know that csum_partial always returned a 16bit value on | |
844 | * alphas, do a fold to maximise conformity to previous behaviour. | |
845 | */ | |
846 | sb->sb_csum = md_csum_fold(disk_csum); | |
847 | #else | |
1da177e4 | 848 | sb->sb_csum = disk_csum; |
4d167f09 | 849 | #endif |
1da177e4 LT |
850 | return csum; |
851 | } | |
852 | ||
853 | ||
854 | /* | |
855 | * Handle superblock details. | |
856 | * We want to be able to handle multiple superblock formats | |
857 | * so we have a common interface to them all, and an array of | |
858 | * different handlers. | |
859 | * We rely on user-space to write the initial superblock, and support | |
860 | * reading and updating of superblocks. | |
861 | * Interface methods are: | |
862 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
863 | * loads and validates a superblock on dev. | |
864 | * if refdev != NULL, compare superblocks on both devices | |
865 | * Return: | |
866 | * 0 - dev has a superblock that is compatible with refdev | |
867 | * 1 - dev has a superblock that is compatible and newer than refdev | |
868 | * so dev should be used as the refdev in future | |
869 | * -EINVAL superblock incompatible or invalid | |
870 | * -othererror e.g. -EIO | |
871 | * | |
872 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
873 | * Verify that dev is acceptable into mddev. | |
874 | * The first time, mddev->raid_disks will be 0, and data from | |
875 | * dev should be merged in. Subsequent calls check that dev | |
876 | * is new enough. Return 0 or -EINVAL | |
877 | * | |
878 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
879 | * Update the superblock for rdev with data in mddev | |
880 | * This does not write to disc. | |
881 | * | |
882 | */ | |
883 | ||
884 | struct super_type { | |
0cd17fec CW |
885 | char *name; |
886 | struct module *owner; | |
887 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
888 | int minor_version); | |
889 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
890 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
891 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 892 | sector_t num_sectors); |
1da177e4 LT |
893 | }; |
894 | ||
0894cc30 AN |
895 | /* |
896 | * Check that the given mddev has no bitmap. | |
897 | * | |
898 | * This function is called from the run method of all personalities that do not | |
899 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
900 | * has a bitmap. Otherwise, it returns 0. | |
901 | * | |
902 | */ | |
903 | int md_check_no_bitmap(mddev_t *mddev) | |
904 | { | |
c3d9714e | 905 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
906 | return 0; |
907 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
908 | mdname(mddev), mddev->pers->name); | |
909 | return 1; | |
910 | } | |
911 | EXPORT_SYMBOL(md_check_no_bitmap); | |
912 | ||
1da177e4 LT |
913 | /* |
914 | * load_super for 0.90.0 | |
915 | */ | |
916 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
917 | { | |
918 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
919 | mdp_super_t *sb; | |
920 | int ret; | |
1da177e4 LT |
921 | |
922 | /* | |
0f420358 | 923 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
924 | * it's at the end of the disk. |
925 | * | |
926 | * It also happens to be a multiple of 4Kb. | |
927 | */ | |
0f420358 | 928 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 929 | |
0002b271 | 930 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
931 | if (ret) return ret; |
932 | ||
933 | ret = -EINVAL; | |
934 | ||
935 | bdevname(rdev->bdev, b); | |
936 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
937 | ||
938 | if (sb->md_magic != MD_SB_MAGIC) { | |
939 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
940 | b); | |
941 | goto abort; | |
942 | } | |
943 | ||
944 | if (sb->major_version != 0 || | |
f6705578 N |
945 | sb->minor_version < 90 || |
946 | sb->minor_version > 91) { | |
1da177e4 LT |
947 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
948 | sb->major_version, sb->minor_version, | |
949 | b); | |
950 | goto abort; | |
951 | } | |
952 | ||
953 | if (sb->raid_disks <= 0) | |
954 | goto abort; | |
955 | ||
4d167f09 | 956 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
957 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
958 | b); | |
959 | goto abort; | |
960 | } | |
961 | ||
962 | rdev->preferred_minor = sb->md_minor; | |
963 | rdev->data_offset = 0; | |
0002b271 | 964 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
965 | |
966 | if (sb->level == LEVEL_MULTIPATH) | |
967 | rdev->desc_nr = -1; | |
968 | else | |
969 | rdev->desc_nr = sb->this_disk.number; | |
970 | ||
9a7b2b0f | 971 | if (!refdev) { |
1da177e4 | 972 | ret = 1; |
9a7b2b0f | 973 | } else { |
1da177e4 LT |
974 | __u64 ev1, ev2; |
975 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
976 | if (!uuid_equal(refsb, sb)) { | |
977 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
978 | b, bdevname(refdev->bdev,b2)); | |
979 | goto abort; | |
980 | } | |
981 | if (!sb_equal(refsb, sb)) { | |
982 | printk(KERN_WARNING "md: %s has same UUID" | |
983 | " but different superblock to %s\n", | |
984 | b, bdevname(refdev->bdev, b2)); | |
985 | goto abort; | |
986 | } | |
987 | ev1 = md_event(sb); | |
988 | ev2 = md_event(refsb); | |
989 | if (ev1 > ev2) | |
990 | ret = 1; | |
991 | else | |
992 | ret = 0; | |
993 | } | |
8190e754 | 994 | rdev->sectors = rdev->sb_start; |
1da177e4 | 995 | |
dd8ac336 | 996 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
997 | /* "this cannot possibly happen" ... */ |
998 | ret = -EINVAL; | |
999 | ||
1da177e4 LT |
1000 | abort: |
1001 | return ret; | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * validate_super for 0.90.0 | |
1006 | */ | |
1007 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1008 | { | |
1009 | mdp_disk_t *desc; | |
1010 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 1011 | __u64 ev1 = md_event(sb); |
1da177e4 | 1012 | |
41158c7e | 1013 | rdev->raid_disk = -1; |
c5d79adb N |
1014 | clear_bit(Faulty, &rdev->flags); |
1015 | clear_bit(In_sync, &rdev->flags); | |
1016 | clear_bit(WriteMostly, &rdev->flags); | |
1017 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1018 | ||
1da177e4 LT |
1019 | if (mddev->raid_disks == 0) { |
1020 | mddev->major_version = 0; | |
1021 | mddev->minor_version = sb->minor_version; | |
1022 | mddev->patch_version = sb->patch_version; | |
e691063a | 1023 | mddev->external = 0; |
9d8f0363 | 1024 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1025 | mddev->ctime = sb->ctime; |
1026 | mddev->utime = sb->utime; | |
1027 | mddev->level = sb->level; | |
d9d166c2 | 1028 | mddev->clevel[0] = 0; |
1da177e4 LT |
1029 | mddev->layout = sb->layout; |
1030 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 1031 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 1032 | mddev->events = ev1; |
c3d9714e N |
1033 | mddev->bitmap_info.offset = 0; |
1034 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; | |
1da177e4 | 1035 | |
f6705578 N |
1036 | if (mddev->minor_version >= 91) { |
1037 | mddev->reshape_position = sb->reshape_position; | |
1038 | mddev->delta_disks = sb->delta_disks; | |
1039 | mddev->new_level = sb->new_level; | |
1040 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1041 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
1042 | } else { |
1043 | mddev->reshape_position = MaxSector; | |
1044 | mddev->delta_disks = 0; | |
1045 | mddev->new_level = mddev->level; | |
1046 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1047 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1048 | } |
1049 | ||
1da177e4 LT |
1050 | if (sb->state & (1<<MD_SB_CLEAN)) |
1051 | mddev->recovery_cp = MaxSector; | |
1052 | else { | |
1053 | if (sb->events_hi == sb->cp_events_hi && | |
1054 | sb->events_lo == sb->cp_events_lo) { | |
1055 | mddev->recovery_cp = sb->recovery_cp; | |
1056 | } else | |
1057 | mddev->recovery_cp = 0; | |
1058 | } | |
1059 | ||
1060 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1061 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1062 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1063 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1064 | ||
1065 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1066 | |
1067 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
c3d9714e N |
1068 | mddev->bitmap_info.file == NULL) |
1069 | mddev->bitmap_info.offset = | |
1070 | mddev->bitmap_info.default_offset; | |
a654b9d8 | 1071 | |
41158c7e N |
1072 | } else if (mddev->pers == NULL) { |
1073 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
1074 | ++ev1; |
1075 | if (ev1 < mddev->events) | |
1076 | return -EINVAL; | |
41158c7e N |
1077 | } else if (mddev->bitmap) { |
1078 | /* if adding to array with a bitmap, then we can accept an | |
1079 | * older device ... but not too old. | |
1080 | */ | |
41158c7e N |
1081 | if (ev1 < mddev->bitmap->events_cleared) |
1082 | return 0; | |
07d84d10 N |
1083 | } else { |
1084 | if (ev1 < mddev->events) | |
1085 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1086 | return 0; | |
1087 | } | |
41158c7e | 1088 | |
1da177e4 | 1089 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1090 | desc = sb->disks + rdev->desc_nr; |
1091 | ||
1092 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1093 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1094 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1095 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1096 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1097 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
1098 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1099 | /* active but not in sync implies recovery up to | |
1100 | * reshape position. We don't know exactly where | |
1101 | * that is, so set to zero for now */ | |
1102 | if (mddev->minor_version >= 91) { | |
1103 | rdev->recovery_offset = 0; | |
1104 | rdev->raid_disk = desc->raid_disk; | |
1105 | } | |
1da177e4 | 1106 | } |
8ddf9efe N |
1107 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1108 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1109 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1110 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1111 | return 0; |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * sync_super for 0.90.0 | |
1116 | */ | |
1117 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1118 | { | |
1119 | mdp_super_t *sb; | |
1da177e4 LT |
1120 | mdk_rdev_t *rdev2; |
1121 | int next_spare = mddev->raid_disks; | |
19133a42 | 1122 | |
1da177e4 LT |
1123 | |
1124 | /* make rdev->sb match mddev data.. | |
1125 | * | |
1126 | * 1/ zero out disks | |
1127 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1128 | * 3/ any empty disks < next_spare become removed | |
1129 | * | |
1130 | * disks[0] gets initialised to REMOVED because | |
1131 | * we cannot be sure from other fields if it has | |
1132 | * been initialised or not. | |
1133 | */ | |
1134 | int i; | |
1135 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1136 | ||
61181565 N |
1137 | rdev->sb_size = MD_SB_BYTES; |
1138 | ||
1da177e4 LT |
1139 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
1140 | ||
1141 | memset(sb, 0, sizeof(*sb)); | |
1142 | ||
1143 | sb->md_magic = MD_SB_MAGIC; | |
1144 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1145 | sb->patch_version = mddev->patch_version; |
1146 | sb->gvalid_words = 0; /* ignored */ | |
1147 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1148 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1149 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1150 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1151 | ||
1152 | sb->ctime = mddev->ctime; | |
1153 | sb->level = mddev->level; | |
58c0fed4 | 1154 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1155 | sb->raid_disks = mddev->raid_disks; |
1156 | sb->md_minor = mddev->md_minor; | |
e691063a | 1157 | sb->not_persistent = 0; |
1da177e4 LT |
1158 | sb->utime = mddev->utime; |
1159 | sb->state = 0; | |
1160 | sb->events_hi = (mddev->events>>32); | |
1161 | sb->events_lo = (u32)mddev->events; | |
1162 | ||
f6705578 N |
1163 | if (mddev->reshape_position == MaxSector) |
1164 | sb->minor_version = 90; | |
1165 | else { | |
1166 | sb->minor_version = 91; | |
1167 | sb->reshape_position = mddev->reshape_position; | |
1168 | sb->new_level = mddev->new_level; | |
1169 | sb->delta_disks = mddev->delta_disks; | |
1170 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1171 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1172 | } |
1173 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1174 | if (mddev->in_sync) |
1175 | { | |
1176 | sb->recovery_cp = mddev->recovery_cp; | |
1177 | sb->cp_events_hi = (mddev->events>>32); | |
1178 | sb->cp_events_lo = (u32)mddev->events; | |
1179 | if (mddev->recovery_cp == MaxSector) | |
1180 | sb->state = (1<< MD_SB_CLEAN); | |
1181 | } else | |
1182 | sb->recovery_cp = 0; | |
1183 | ||
1184 | sb->layout = mddev->layout; | |
9d8f0363 | 1185 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1186 | |
c3d9714e | 1187 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1188 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1189 | ||
1da177e4 | 1190 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1191 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1192 | mdp_disk_t *d; |
86e6ffdd | 1193 | int desc_nr; |
0261cd9f N |
1194 | int is_active = test_bit(In_sync, &rdev2->flags); |
1195 | ||
1196 | if (rdev2->raid_disk >= 0 && | |
1197 | sb->minor_version >= 91) | |
1198 | /* we have nowhere to store the recovery_offset, | |
1199 | * but if it is not below the reshape_position, | |
1200 | * we can piggy-back on that. | |
1201 | */ | |
1202 | is_active = 1; | |
1203 | if (rdev2->raid_disk < 0 || | |
1204 | test_bit(Faulty, &rdev2->flags)) | |
1205 | is_active = 0; | |
1206 | if (is_active) | |
86e6ffdd | 1207 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1208 | else |
86e6ffdd | 1209 | desc_nr = next_spare++; |
19133a42 | 1210 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1211 | d = &sb->disks[rdev2->desc_nr]; |
1212 | nr_disks++; | |
1213 | d->number = rdev2->desc_nr; | |
1214 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1215 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1216 | if (is_active) |
1da177e4 LT |
1217 | d->raid_disk = rdev2->raid_disk; |
1218 | else | |
1219 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1220 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1221 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1222 | else if (is_active) { |
1da177e4 | 1223 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1224 | if (test_bit(In_sync, &rdev2->flags)) |
1225 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1226 | active++; |
1227 | working++; | |
1228 | } else { | |
1229 | d->state = 0; | |
1230 | spare++; | |
1231 | working++; | |
1232 | } | |
8ddf9efe N |
1233 | if (test_bit(WriteMostly, &rdev2->flags)) |
1234 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1235 | } |
1da177e4 LT |
1236 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1237 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1238 | mdp_disk_t *d = &sb->disks[i]; | |
1239 | if (d->state == 0 && d->number == 0) { | |
1240 | d->number = i; | |
1241 | d->raid_disk = i; | |
1242 | d->state = (1<<MD_DISK_REMOVED); | |
1243 | d->state |= (1<<MD_DISK_FAULTY); | |
1244 | failed++; | |
1245 | } | |
1246 | } | |
1247 | sb->nr_disks = nr_disks; | |
1248 | sb->active_disks = active; | |
1249 | sb->working_disks = working; | |
1250 | sb->failed_disks = failed; | |
1251 | sb->spare_disks = spare; | |
1252 | ||
1253 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1254 | sb->sb_csum = calc_sb_csum(sb); | |
1255 | } | |
1256 | ||
0cd17fec CW |
1257 | /* |
1258 | * rdev_size_change for 0.90.0 | |
1259 | */ | |
1260 | static unsigned long long | |
15f4a5fd | 1261 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1262 | { |
58c0fed4 | 1263 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1264 | return 0; /* component must fit device */ |
c3d9714e | 1265 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1266 | return 0; /* can't move bitmap */ |
0f420358 | 1267 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
15f4a5fd AN |
1268 | if (!num_sectors || num_sectors > rdev->sb_start) |
1269 | num_sectors = rdev->sb_start; | |
0f420358 | 1270 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1271 | rdev->sb_page); |
1272 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1273 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec CW |
1274 | } |
1275 | ||
1276 | ||
1da177e4 LT |
1277 | /* |
1278 | * version 1 superblock | |
1279 | */ | |
1280 | ||
1c05b4bc | 1281 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1282 | { |
1c05b4bc N |
1283 | __le32 disk_csum; |
1284 | u32 csum; | |
1da177e4 LT |
1285 | unsigned long long newcsum; |
1286 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1287 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1288 | int i; |
1289 | ||
1290 | disk_csum = sb->sb_csum; | |
1291 | sb->sb_csum = 0; | |
1292 | newcsum = 0; | |
1293 | for (i=0; size>=4; size -= 4 ) | |
1294 | newcsum += le32_to_cpu(*isuper++); | |
1295 | ||
1296 | if (size == 2) | |
1c05b4bc | 1297 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1298 | |
1299 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1300 | sb->sb_csum = disk_csum; | |
1301 | return cpu_to_le32(csum); | |
1302 | } | |
1303 | ||
1304 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1305 | { | |
1306 | struct mdp_superblock_1 *sb; | |
1307 | int ret; | |
0f420358 | 1308 | sector_t sb_start; |
1da177e4 | 1309 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1310 | int bmask; |
1da177e4 LT |
1311 | |
1312 | /* | |
0f420358 | 1313 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1314 | * It is always aligned to a 4K boundary and |
1315 | * depeding on minor_version, it can be: | |
1316 | * 0: At least 8K, but less than 12K, from end of device | |
1317 | * 1: At start of device | |
1318 | * 2: 4K from start of device. | |
1319 | */ | |
1320 | switch(minor_version) { | |
1321 | case 0: | |
0f420358 AN |
1322 | sb_start = rdev->bdev->bd_inode->i_size >> 9; |
1323 | sb_start -= 8*2; | |
1324 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1325 | break; |
1326 | case 1: | |
0f420358 | 1327 | sb_start = 0; |
1da177e4 LT |
1328 | break; |
1329 | case 2: | |
0f420358 | 1330 | sb_start = 8; |
1da177e4 LT |
1331 | break; |
1332 | default: | |
1333 | return -EINVAL; | |
1334 | } | |
0f420358 | 1335 | rdev->sb_start = sb_start; |
1da177e4 | 1336 | |
0002b271 N |
1337 | /* superblock is rarely larger than 1K, but it can be larger, |
1338 | * and it is safe to read 4k, so we do that | |
1339 | */ | |
1340 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1341 | if (ret) return ret; |
1342 | ||
1343 | ||
1344 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1345 | ||
1346 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1347 | sb->major_version != cpu_to_le32(1) || | |
1348 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1349 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1350 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1351 | return -EINVAL; |
1352 | ||
1353 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1354 | printk("md: invalid superblock checksum on %s\n", | |
1355 | bdevname(rdev->bdev,b)); | |
1356 | return -EINVAL; | |
1357 | } | |
1358 | if (le64_to_cpu(sb->data_size) < 10) { | |
1359 | printk("md: data_size too small on %s\n", | |
1360 | bdevname(rdev->bdev,b)); | |
1361 | return -EINVAL; | |
1362 | } | |
e11e93fa | 1363 | |
1da177e4 LT |
1364 | rdev->preferred_minor = 0xffff; |
1365 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1366 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1367 | |
0002b271 | 1368 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1369 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1370 | if (rdev->sb_size & bmask) |
a1801f85 N |
1371 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1372 | ||
1373 | if (minor_version | |
0f420358 | 1374 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1375 | return -EINVAL; |
0002b271 | 1376 | |
31b65a0d N |
1377 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1378 | rdev->desc_nr = -1; | |
1379 | else | |
1380 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1381 | ||
9a7b2b0f | 1382 | if (!refdev) { |
8ed75463 | 1383 | ret = 1; |
9a7b2b0f | 1384 | } else { |
1da177e4 LT |
1385 | __u64 ev1, ev2; |
1386 | struct mdp_superblock_1 *refsb = | |
1387 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1388 | ||
1389 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1390 | sb->level != refsb->level || | |
1391 | sb->layout != refsb->layout || | |
1392 | sb->chunksize != refsb->chunksize) { | |
1393 | printk(KERN_WARNING "md: %s has strangely different" | |
1394 | " superblock to %s\n", | |
1395 | bdevname(rdev->bdev,b), | |
1396 | bdevname(refdev->bdev,b2)); | |
1397 | return -EINVAL; | |
1398 | } | |
1399 | ev1 = le64_to_cpu(sb->events); | |
1400 | ev2 = le64_to_cpu(refsb->events); | |
1401 | ||
1402 | if (ev1 > ev2) | |
8ed75463 N |
1403 | ret = 1; |
1404 | else | |
1405 | ret = 0; | |
1da177e4 | 1406 | } |
a1801f85 | 1407 | if (minor_version) |
dd8ac336 AN |
1408 | rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) - |
1409 | le64_to_cpu(sb->data_offset); | |
1da177e4 | 1410 | else |
dd8ac336 AN |
1411 | rdev->sectors = rdev->sb_start; |
1412 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1413 | return -EINVAL; |
dd8ac336 | 1414 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1415 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1416 | return -EINVAL; |
8ed75463 | 1417 | return ret; |
1da177e4 LT |
1418 | } |
1419 | ||
1420 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1421 | { | |
1422 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1423 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1424 | |
41158c7e | 1425 | rdev->raid_disk = -1; |
c5d79adb N |
1426 | clear_bit(Faulty, &rdev->flags); |
1427 | clear_bit(In_sync, &rdev->flags); | |
1428 | clear_bit(WriteMostly, &rdev->flags); | |
1429 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1430 | ||
1da177e4 LT |
1431 | if (mddev->raid_disks == 0) { |
1432 | mddev->major_version = 1; | |
1433 | mddev->patch_version = 0; | |
e691063a | 1434 | mddev->external = 0; |
9d8f0363 | 1435 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1436 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1437 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1438 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1439 | mddev->clevel[0] = 0; |
1da177e4 LT |
1440 | mddev->layout = le32_to_cpu(sb->layout); |
1441 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1442 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1443 | mddev->events = ev1; |
c3d9714e N |
1444 | mddev->bitmap_info.offset = 0; |
1445 | mddev->bitmap_info.default_offset = 1024 >> 9; | |
1da177e4 LT |
1446 | |
1447 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1448 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1449 | ||
1450 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1451 | |
71c0805c | 1452 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
c3d9714e N |
1453 | mddev->bitmap_info.file == NULL ) |
1454 | mddev->bitmap_info.offset = | |
1455 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
e11e93fa | 1456 | |
f6705578 N |
1457 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1458 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1459 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1460 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1461 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1462 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1463 | } else { |
1464 | mddev->reshape_position = MaxSector; | |
1465 | mddev->delta_disks = 0; | |
1466 | mddev->new_level = mddev->level; | |
1467 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1468 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1469 | } |
1470 | ||
41158c7e N |
1471 | } else if (mddev->pers == NULL) { |
1472 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1473 | ++ev1; |
1474 | if (ev1 < mddev->events) | |
1475 | return -EINVAL; | |
41158c7e N |
1476 | } else if (mddev->bitmap) { |
1477 | /* If adding to array with a bitmap, then we can accept an | |
1478 | * older device, but not too old. | |
1479 | */ | |
41158c7e N |
1480 | if (ev1 < mddev->bitmap->events_cleared) |
1481 | return 0; | |
07d84d10 N |
1482 | } else { |
1483 | if (ev1 < mddev->events) | |
1484 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1485 | return 0; | |
1486 | } | |
1da177e4 LT |
1487 | if (mddev->level != LEVEL_MULTIPATH) { |
1488 | int role; | |
3673f305 N |
1489 | if (rdev->desc_nr < 0 || |
1490 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1491 | role = 0xffff; | |
1492 | rdev->desc_nr = -1; | |
1493 | } else | |
1494 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1495 | switch(role) { |
1496 | case 0xffff: /* spare */ | |
1da177e4 LT |
1497 | break; |
1498 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1499 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1500 | break; |
1501 | default: | |
5fd6c1dc N |
1502 | if ((le32_to_cpu(sb->feature_map) & |
1503 | MD_FEATURE_RECOVERY_OFFSET)) | |
1504 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1505 | else | |
1506 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1507 | rdev->raid_disk = role; |
1508 | break; | |
1509 | } | |
8ddf9efe N |
1510 | if (sb->devflags & WriteMostly1) |
1511 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1512 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1513 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1514 | |
1da177e4 LT |
1515 | return 0; |
1516 | } | |
1517 | ||
1518 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1519 | { | |
1520 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1521 | mdk_rdev_t *rdev2; |
1522 | int max_dev, i; | |
1523 | /* make rdev->sb match mddev and rdev data. */ | |
1524 | ||
1525 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1526 | ||
1527 | sb->feature_map = 0; | |
1528 | sb->pad0 = 0; | |
5fd6c1dc | 1529 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1530 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1531 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1532 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1533 | ||
1534 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1535 | sb->events = cpu_to_le64(mddev->events); | |
1536 | if (mddev->in_sync) | |
1537 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1538 | else | |
1539 | sb->resync_offset = cpu_to_le64(0); | |
1540 | ||
1c05b4bc | 1541 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1542 | |
f0ca340c | 1543 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1544 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1545 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1546 | sb->level = cpu_to_le32(mddev->level); |
1547 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1548 | |
c3d9714e N |
1549 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1550 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1551 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1552 | } |
5fd6c1dc N |
1553 | |
1554 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1555 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1556 | sb->feature_map |= |
1557 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1558 | sb->recovery_offset = | |
1559 | cpu_to_le64(rdev->recovery_offset); | |
5fd6c1dc N |
1560 | } |
1561 | ||
f6705578 N |
1562 | if (mddev->reshape_position != MaxSector) { |
1563 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1564 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1565 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1566 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1567 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1568 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1569 | } |
a654b9d8 | 1570 | |
1da177e4 | 1571 | max_dev = 0; |
159ec1fc | 1572 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1573 | if (rdev2->desc_nr+1 > max_dev) |
1574 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1575 | |
70471daf N |
1576 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1577 | int bmask; | |
a778b73f | 1578 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1579 | rdev->sb_size = max_dev * 2 + 256; |
1580 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1581 | if (rdev->sb_size & bmask) | |
1582 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
1583 | } | |
1da177e4 LT |
1584 | for (i=0; i<max_dev;i++) |
1585 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1586 | ||
159ec1fc | 1587 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1588 | i = rdev2->desc_nr; |
b2d444d7 | 1589 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1590 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1591 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1592 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1593 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1594 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1595 | else |
1596 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1597 | } | |
1598 | ||
1da177e4 LT |
1599 | sb->sb_csum = calc_sb_1_csum(sb); |
1600 | } | |
1601 | ||
0cd17fec | 1602 | static unsigned long long |
15f4a5fd | 1603 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1604 | { |
1605 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1606 | sector_t max_sectors; |
58c0fed4 | 1607 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1608 | return 0; /* component must fit device */ |
0f420358 | 1609 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1610 | /* minor versions 1 and 2; superblock before data */ |
15f4a5fd AN |
1611 | max_sectors = rdev->bdev->bd_inode->i_size >> 9; |
1612 | max_sectors -= rdev->data_offset; | |
1613 | if (!num_sectors || num_sectors > max_sectors) | |
1614 | num_sectors = max_sectors; | |
c3d9714e | 1615 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1616 | /* minor version 0 with bitmap we can't move */ |
1617 | return 0; | |
1618 | } else { | |
1619 | /* minor version 0; superblock after data */ | |
0f420358 AN |
1620 | sector_t sb_start; |
1621 | sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; | |
1622 | sb_start &= ~(sector_t)(4*2 - 1); | |
dd8ac336 | 1623 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1624 | if (!num_sectors || num_sectors > max_sectors) |
1625 | num_sectors = max_sectors; | |
0f420358 | 1626 | rdev->sb_start = sb_start; |
0cd17fec CW |
1627 | } |
1628 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1629 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1630 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1631 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1632 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1633 | rdev->sb_page); |
1634 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1635 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec | 1636 | } |
1da177e4 | 1637 | |
75c96f85 | 1638 | static struct super_type super_types[] = { |
1da177e4 LT |
1639 | [0] = { |
1640 | .name = "0.90.0", | |
1641 | .owner = THIS_MODULE, | |
0cd17fec CW |
1642 | .load_super = super_90_load, |
1643 | .validate_super = super_90_validate, | |
1644 | .sync_super = super_90_sync, | |
1645 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1646 | }, |
1647 | [1] = { | |
1648 | .name = "md-1", | |
1649 | .owner = THIS_MODULE, | |
0cd17fec CW |
1650 | .load_super = super_1_load, |
1651 | .validate_super = super_1_validate, | |
1652 | .sync_super = super_1_sync, | |
1653 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1654 | }, |
1655 | }; | |
1da177e4 LT |
1656 | |
1657 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1658 | { | |
7dd5e7c3 | 1659 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1660 | |
4b80991c N |
1661 | rcu_read_lock(); |
1662 | rdev_for_each_rcu(rdev, mddev1) | |
1663 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1664 | if (rdev->bdev->bd_contains == |
4b80991c N |
1665 | rdev2->bdev->bd_contains) { |
1666 | rcu_read_unlock(); | |
7dd5e7c3 | 1667 | return 1; |
4b80991c N |
1668 | } |
1669 | rcu_read_unlock(); | |
1da177e4 LT |
1670 | return 0; |
1671 | } | |
1672 | ||
1673 | static LIST_HEAD(pending_raid_disks); | |
1674 | ||
ac5e7113 AN |
1675 | /* |
1676 | * Try to register data integrity profile for an mddev | |
1677 | * | |
1678 | * This is called when an array is started and after a disk has been kicked | |
1679 | * from the array. It only succeeds if all working and active component devices | |
1680 | * are integrity capable with matching profiles. | |
1681 | */ | |
1682 | int md_integrity_register(mddev_t *mddev) | |
1683 | { | |
1684 | mdk_rdev_t *rdev, *reference = NULL; | |
1685 | ||
1686 | if (list_empty(&mddev->disks)) | |
1687 | return 0; /* nothing to do */ | |
1688 | if (blk_get_integrity(mddev->gendisk)) | |
1689 | return 0; /* already registered */ | |
1690 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
1691 | /* skip spares and non-functional disks */ | |
1692 | if (test_bit(Faulty, &rdev->flags)) | |
1693 | continue; | |
1694 | if (rdev->raid_disk < 0) | |
1695 | continue; | |
1696 | /* | |
1697 | * If at least one rdev is not integrity capable, we can not | |
1698 | * enable data integrity for the md device. | |
1699 | */ | |
1700 | if (!bdev_get_integrity(rdev->bdev)) | |
1701 | return -EINVAL; | |
1702 | if (!reference) { | |
1703 | /* Use the first rdev as the reference */ | |
1704 | reference = rdev; | |
1705 | continue; | |
1706 | } | |
1707 | /* does this rdev's profile match the reference profile? */ | |
1708 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1709 | rdev->bdev->bd_disk) < 0) | |
1710 | return -EINVAL; | |
1711 | } | |
1712 | /* | |
1713 | * All component devices are integrity capable and have matching | |
1714 | * profiles, register the common profile for the md device. | |
1715 | */ | |
1716 | if (blk_integrity_register(mddev->gendisk, | |
1717 | bdev_get_integrity(reference->bdev)) != 0) { | |
1718 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1719 | mdname(mddev)); | |
1720 | return -EINVAL; | |
1721 | } | |
1722 | printk(KERN_NOTICE "md: data integrity on %s enabled\n", | |
1723 | mdname(mddev)); | |
1724 | return 0; | |
1725 | } | |
1726 | EXPORT_SYMBOL(md_integrity_register); | |
1727 | ||
1728 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
1729 | void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) | |
3f9d99c1 | 1730 | { |
3f9d99c1 | 1731 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); |
ac5e7113 | 1732 | struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); |
3f9d99c1 | 1733 | |
ac5e7113 | 1734 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1735 | return; |
ac5e7113 | 1736 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1737 | return; |
ac5e7113 AN |
1738 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1739 | rdev->bdev->bd_disk) >= 0) | |
1740 | return; | |
1741 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1742 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1743 | } |
ac5e7113 | 1744 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1745 | |
1da177e4 LT |
1746 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1747 | { | |
7dd5e7c3 | 1748 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1749 | struct kobject *ko; |
1edf80d3 | 1750 | char *s; |
5e55e2f5 | 1751 | int err; |
1da177e4 LT |
1752 | |
1753 | if (rdev->mddev) { | |
1754 | MD_BUG(); | |
1755 | return -EINVAL; | |
1756 | } | |
11e2ede0 DW |
1757 | |
1758 | /* prevent duplicates */ | |
1759 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1760 | return -EEXIST; | |
1761 | ||
dd8ac336 AN |
1762 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1763 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1764 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1765 | if (mddev->pers) { |
1766 | /* Cannot change size, so fail | |
1767 | * If mddev->level <= 0, then we don't care | |
1768 | * about aligning sizes (e.g. linear) | |
1769 | */ | |
1770 | if (mddev->level > 0) | |
1771 | return -ENOSPC; | |
1772 | } else | |
dd8ac336 | 1773 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1774 | } |
1da177e4 LT |
1775 | |
1776 | /* Verify rdev->desc_nr is unique. | |
1777 | * If it is -1, assign a free number, else | |
1778 | * check number is not in use | |
1779 | */ | |
1780 | if (rdev->desc_nr < 0) { | |
1781 | int choice = 0; | |
1782 | if (mddev->pers) choice = mddev->raid_disks; | |
1783 | while (find_rdev_nr(mddev, choice)) | |
1784 | choice++; | |
1785 | rdev->desc_nr = choice; | |
1786 | } else { | |
1787 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1788 | return -EBUSY; | |
1789 | } | |
de01dfad N |
1790 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1791 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1792 | mdname(mddev), mddev->max_disks); | |
1793 | return -EBUSY; | |
1794 | } | |
19133a42 | 1795 | bdevname(rdev->bdev,b); |
649316b2 | 1796 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1797 | *s = '!'; |
649316b2 | 1798 | |
1da177e4 | 1799 | rdev->mddev = mddev; |
19133a42 | 1800 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1801 | |
b2d6db58 | 1802 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1803 | goto fail; |
86e6ffdd | 1804 | |
0762b8bd | 1805 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
5e55e2f5 N |
1806 | if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) { |
1807 | kobject_del(&rdev->kobj); | |
1808 | goto fail; | |
1809 | } | |
3c0ee63a N |
1810 | rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state"); |
1811 | ||
4b80991c | 1812 | list_add_rcu(&rdev->same_set, &mddev->disks); |
c5d79adb | 1813 | bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); |
4044ba58 N |
1814 | |
1815 | /* May as well allow recovery to be retried once */ | |
1816 | mddev->recovery_disabled = 0; | |
3f9d99c1 | 1817 | |
1da177e4 | 1818 | return 0; |
5e55e2f5 N |
1819 | |
1820 | fail: | |
1821 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1822 | b, mdname(mddev)); | |
1823 | return err; | |
1da177e4 LT |
1824 | } |
1825 | ||
177a99b2 | 1826 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1827 | { |
1828 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1829 | kobject_del(&rdev->kobj); | |
177a99b2 | 1830 | kobject_put(&rdev->kobj); |
5792a285 N |
1831 | } |
1832 | ||
1da177e4 LT |
1833 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1834 | { | |
1835 | char b[BDEVNAME_SIZE]; | |
1836 | if (!rdev->mddev) { | |
1837 | MD_BUG(); | |
1838 | return; | |
1839 | } | |
5463c790 | 1840 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1841 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1842 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1843 | rdev->mddev = NULL; | |
86e6ffdd | 1844 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1845 | sysfs_put(rdev->sysfs_state); |
1846 | rdev->sysfs_state = NULL; | |
5792a285 | 1847 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1848 | * writing to 'remove' to "dev/state". We also need |
1849 | * to delay it due to rcu usage. | |
5792a285 | 1850 | */ |
4b80991c | 1851 | synchronize_rcu(); |
177a99b2 N |
1852 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1853 | kobject_get(&rdev->kobj); | |
5792a285 | 1854 | schedule_work(&rdev->del_work); |
1da177e4 LT |
1855 | } |
1856 | ||
1857 | /* | |
1858 | * prevent the device from being mounted, repartitioned or | |
1859 | * otherwise reused by a RAID array (or any other kernel | |
1860 | * subsystem), by bd_claiming the device. | |
1861 | */ | |
c5d79adb | 1862 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1863 | { |
1864 | int err = 0; | |
1865 | struct block_device *bdev; | |
1866 | char b[BDEVNAME_SIZE]; | |
1867 | ||
2e7b651d | 1868 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1869 | if (IS_ERR(bdev)) { |
1870 | printk(KERN_ERR "md: could not open %s.\n", | |
1871 | __bdevname(dev, b)); | |
1872 | return PTR_ERR(bdev); | |
1873 | } | |
c5d79adb | 1874 | err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); |
1da177e4 LT |
1875 | if (err) { |
1876 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1877 | bdevname(bdev, b)); | |
9a1c3542 | 1878 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1879 | return err; |
1880 | } | |
c5d79adb N |
1881 | if (!shared) |
1882 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1883 | rdev->bdev = bdev; |
1884 | return err; | |
1885 | } | |
1886 | ||
1887 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1888 | { | |
1889 | struct block_device *bdev = rdev->bdev; | |
1890 | rdev->bdev = NULL; | |
1891 | if (!bdev) | |
1892 | MD_BUG(); | |
1893 | bd_release(bdev); | |
9a1c3542 | 1894 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1895 | } |
1896 | ||
1897 | void md_autodetect_dev(dev_t dev); | |
1898 | ||
1899 | static void export_rdev(mdk_rdev_t * rdev) | |
1900 | { | |
1901 | char b[BDEVNAME_SIZE]; | |
1902 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1903 | bdevname(rdev->bdev,b)); | |
1904 | if (rdev->mddev) | |
1905 | MD_BUG(); | |
1906 | free_disk_sb(rdev); | |
1da177e4 | 1907 | #ifndef MODULE |
d0fae18f N |
1908 | if (test_bit(AutoDetected, &rdev->flags)) |
1909 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1910 | #endif |
1911 | unlock_rdev(rdev); | |
86e6ffdd | 1912 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1913 | } |
1914 | ||
1915 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1916 | { | |
1917 | unbind_rdev_from_array(rdev); | |
1918 | export_rdev(rdev); | |
1919 | } | |
1920 | ||
1921 | static void export_array(mddev_t *mddev) | |
1922 | { | |
159ec1fc | 1923 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 1924 | |
d089c6af | 1925 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1926 | if (!rdev->mddev) { |
1927 | MD_BUG(); | |
1928 | continue; | |
1929 | } | |
1930 | kick_rdev_from_array(rdev); | |
1931 | } | |
1932 | if (!list_empty(&mddev->disks)) | |
1933 | MD_BUG(); | |
1934 | mddev->raid_disks = 0; | |
1935 | mddev->major_version = 0; | |
1936 | } | |
1937 | ||
1938 | static void print_desc(mdp_disk_t *desc) | |
1939 | { | |
1940 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1941 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1942 | } | |
1943 | ||
cd2ac932 | 1944 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
1945 | { |
1946 | int i; | |
1947 | ||
1948 | printk(KERN_INFO | |
1949 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1950 | sb->major_version, sb->minor_version, sb->patch_version, | |
1951 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1952 | sb->ctime); | |
1953 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1954 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1955 | sb->md_minor, sb->layout, sb->chunk_size); | |
1956 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1957 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1958 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1959 | sb->failed_disks, sb->spare_disks, | |
1960 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1961 | ||
1962 | printk(KERN_INFO); | |
1963 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1964 | mdp_disk_t *desc; | |
1965 | ||
1966 | desc = sb->disks + i; | |
1967 | if (desc->number || desc->major || desc->minor || | |
1968 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1969 | printk(" D %2d: ", i); | |
1970 | print_desc(desc); | |
1971 | } | |
1972 | } | |
1973 | printk(KERN_INFO "md: THIS: "); | |
1974 | print_desc(&sb->this_disk); | |
cd2ac932 | 1975 | } |
1da177e4 | 1976 | |
cd2ac932 CR |
1977 | static void print_sb_1(struct mdp_superblock_1 *sb) |
1978 | { | |
1979 | __u8 *uuid; | |
1980 | ||
1981 | uuid = sb->set_uuid; | |
ad361c98 | 1982 | printk(KERN_INFO |
7b75c2f8 | 1983 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 1984 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
1985 | le32_to_cpu(sb->major_version), |
1986 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 1987 | uuid, |
cd2ac932 CR |
1988 | sb->set_name, |
1989 | (unsigned long long)le64_to_cpu(sb->ctime) | |
1990 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
1991 | ||
1992 | uuid = sb->device_uuid; | |
ad361c98 JP |
1993 | printk(KERN_INFO |
1994 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 1995 | " RO:%llu\n" |
7b75c2f8 | 1996 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
1997 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
1998 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
1999 | le32_to_cpu(sb->level), |
2000 | (unsigned long long)le64_to_cpu(sb->size), | |
2001 | le32_to_cpu(sb->raid_disks), | |
2002 | le32_to_cpu(sb->layout), | |
2003 | le32_to_cpu(sb->chunksize), | |
2004 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2005 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2006 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2007 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2008 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2009 | uuid, |
cd2ac932 CR |
2010 | sb->devflags, |
2011 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2012 | (unsigned long long)le64_to_cpu(sb->events), | |
2013 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2014 | le32_to_cpu(sb->sb_csum), | |
2015 | le32_to_cpu(sb->max_dev) | |
2016 | ); | |
1da177e4 LT |
2017 | } |
2018 | ||
cd2ac932 | 2019 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
2020 | { |
2021 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2022 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2023 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2024 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2025 | rdev->desc_nr); | |
1da177e4 | 2026 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2027 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2028 | switch (major_version) { | |
2029 | case 0: | |
2030 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
2031 | break; | |
2032 | case 1: | |
2033 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
2034 | break; | |
2035 | } | |
1da177e4 LT |
2036 | } else |
2037 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2038 | } | |
2039 | ||
5e56341d | 2040 | static void md_print_devices(void) |
1da177e4 | 2041 | { |
159ec1fc | 2042 | struct list_head *tmp; |
1da177e4 LT |
2043 | mdk_rdev_t *rdev; |
2044 | mddev_t *mddev; | |
2045 | char b[BDEVNAME_SIZE]; | |
2046 | ||
2047 | printk("\n"); | |
2048 | printk("md: **********************************\n"); | |
2049 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2050 | printk("md: **********************************\n"); | |
29ac4aa3 | 2051 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2052 | |
32a7627c N |
2053 | if (mddev->bitmap) |
2054 | bitmap_print_sb(mddev->bitmap); | |
2055 | else | |
2056 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 2057 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
2058 | printk("<%s>", bdevname(rdev->bdev,b)); |
2059 | printk("\n"); | |
2060 | ||
159ec1fc | 2061 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 2062 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2063 | } |
2064 | printk("md: **********************************\n"); | |
2065 | printk("\n"); | |
2066 | } | |
2067 | ||
2068 | ||
42543769 | 2069 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 2070 | { |
42543769 N |
2071 | /* Update each superblock (in-memory image), but |
2072 | * if we are allowed to, skip spares which already | |
2073 | * have the right event counter, or have one earlier | |
2074 | * (which would mean they aren't being marked as dirty | |
2075 | * with the rest of the array) | |
2076 | */ | |
1da177e4 | 2077 | mdk_rdev_t *rdev; |
1da177e4 | 2078 | |
5e865106 N |
2079 | /* First make sure individual recovery_offsets are correct */ |
2080 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
2081 | if (rdev->raid_disk >= 0 && | |
2082 | !test_bit(In_sync, &rdev->flags) && | |
2083 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2084 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2085 | ||
2086 | } | |
159ec1fc | 2087 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
2088 | if (rdev->sb_events == mddev->events || |
2089 | (nospares && | |
2090 | rdev->raid_disk < 0 && | |
42543769 N |
2091 | rdev->sb_events+1 == mddev->events)) { |
2092 | /* Don't update this superblock */ | |
2093 | rdev->sb_loaded = 2; | |
2094 | } else { | |
2095 | super_types[mddev->major_version]. | |
2096 | sync_super(mddev, rdev); | |
2097 | rdev->sb_loaded = 1; | |
2098 | } | |
1da177e4 LT |
2099 | } |
2100 | } | |
2101 | ||
850b2b42 | 2102 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 2103 | { |
1da177e4 | 2104 | mdk_rdev_t *rdev; |
06d91a5f | 2105 | int sync_req; |
42543769 | 2106 | int nospares = 0; |
1da177e4 | 2107 | |
1b57f132 | 2108 | mddev->utime = get_seconds(); |
8377bc80 N |
2109 | if (mddev->external) |
2110 | return; | |
1da177e4 | 2111 | repeat: |
a9701a30 | 2112 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2113 | |
850b2b42 N |
2114 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2115 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) | |
2116 | force_change = 1; | |
2117 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2118 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2119 | * though if events isn't the right even/odd, we will have to do | |
2120 | * spares after all | |
2121 | */ | |
2122 | nospares = 1; | |
2123 | if (force_change) | |
2124 | nospares = 0; | |
2125 | if (mddev->degraded) | |
84692195 N |
2126 | /* If the array is degraded, then skipping spares is both |
2127 | * dangerous and fairly pointless. | |
2128 | * Dangerous because a device that was removed from the array | |
2129 | * might have a event_count that still looks up-to-date, | |
2130 | * so it can be re-added without a resync. | |
2131 | * Pointless because if there are any spares to skip, | |
2132 | * then a recovery will happen and soon that array won't | |
2133 | * be degraded any more and the spare can go back to sleep then. | |
2134 | */ | |
850b2b42 | 2135 | nospares = 0; |
84692195 | 2136 | |
06d91a5f | 2137 | sync_req = mddev->in_sync; |
42543769 N |
2138 | |
2139 | /* If this is just a dirty<->clean transition, and the array is clean | |
2140 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2141 | if (nospares |
42543769 | 2142 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2143 | && mddev->can_decrease_events |
2144 | && mddev->events != 1) { | |
42543769 | 2145 | mddev->events--; |
a8707c08 N |
2146 | mddev->can_decrease_events = 0; |
2147 | } else { | |
42543769 N |
2148 | /* otherwise we have to go forward and ... */ |
2149 | mddev->events ++; | |
a8707c08 | 2150 | mddev->can_decrease_events = nospares; |
42543769 | 2151 | } |
1da177e4 LT |
2152 | |
2153 | if (!mddev->events) { | |
2154 | /* | |
2155 | * oops, this 64-bit counter should never wrap. | |
2156 | * Either we are in around ~1 trillion A.C., assuming | |
2157 | * 1 reboot per second, or we have a bug: | |
2158 | */ | |
2159 | MD_BUG(); | |
2160 | mddev->events --; | |
2161 | } | |
1da177e4 LT |
2162 | |
2163 | /* | |
2164 | * do not write anything to disk if using | |
2165 | * nonpersistent superblocks | |
2166 | */ | |
06d91a5f | 2167 | if (!mddev->persistent) { |
e691063a N |
2168 | if (!mddev->external) |
2169 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
2170 | ||
a9701a30 | 2171 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2172 | wake_up(&mddev->sb_wait); |
1da177e4 | 2173 | return; |
06d91a5f | 2174 | } |
e691063a | 2175 | sync_sbs(mddev, nospares); |
a9701a30 | 2176 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2177 | |
2178 | dprintk(KERN_INFO | |
2179 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2180 | mdname(mddev),mddev->in_sync); | |
2181 | ||
4ad13663 | 2182 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2183 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2184 | char b[BDEVNAME_SIZE]; |
2185 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2186 | if (rdev->sb_loaded != 1) |
2187 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2188 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2189 | dprintk("(skipping faulty "); |
2190 | ||
2191 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2192 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2193 | md_super_write(mddev,rdev, |
0f420358 | 2194 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2195 | rdev->sb_page); |
2196 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2197 | bdevname(rdev->bdev,b), | |
0f420358 | 2198 | (unsigned long long)rdev->sb_start); |
42543769 | 2199 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2200 | |
1da177e4 LT |
2201 | } else |
2202 | dprintk(")\n"); | |
7bfa19f2 | 2203 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2204 | /* only need to write one superblock... */ |
2205 | break; | |
2206 | } | |
a9701a30 | 2207 | md_super_wait(mddev); |
850b2b42 | 2208 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2209 | |
a9701a30 | 2210 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2211 | if (mddev->in_sync != sync_req || |
2212 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2213 | /* have to write it out again */ |
a9701a30 | 2214 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2215 | goto repeat; |
2216 | } | |
850b2b42 | 2217 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2218 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2219 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2220 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2221 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2222 | |
1da177e4 LT |
2223 | } |
2224 | ||
7f6ce769 | 2225 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2226 | * We want to accept with case. For this we use cmd_match. |
2227 | */ | |
2228 | static int cmd_match(const char *cmd, const char *str) | |
2229 | { | |
2230 | /* See if cmd, written into a sysfs file, matches | |
2231 | * str. They must either be the same, or cmd can | |
2232 | * have a trailing newline | |
2233 | */ | |
2234 | while (*cmd && *str && *cmd == *str) { | |
2235 | cmd++; | |
2236 | str++; | |
2237 | } | |
2238 | if (*cmd == '\n') | |
2239 | cmd++; | |
2240 | if (*str || *cmd) | |
2241 | return 0; | |
2242 | return 1; | |
2243 | } | |
2244 | ||
86e6ffdd N |
2245 | struct rdev_sysfs_entry { |
2246 | struct attribute attr; | |
2247 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2248 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2249 | }; | |
2250 | ||
2251 | static ssize_t | |
96de1e66 | 2252 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2253 | { |
2254 | char *sep = ""; | |
20a49ff6 | 2255 | size_t len = 0; |
86e6ffdd | 2256 | |
b2d444d7 | 2257 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2258 | len+= sprintf(page+len, "%sfaulty",sep); |
2259 | sep = ","; | |
2260 | } | |
b2d444d7 | 2261 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2262 | len += sprintf(page+len, "%sin_sync",sep); |
2263 | sep = ","; | |
2264 | } | |
f655675b N |
2265 | if (test_bit(WriteMostly, &rdev->flags)) { |
2266 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2267 | sep = ","; | |
2268 | } | |
6bfe0b49 DW |
2269 | if (test_bit(Blocked, &rdev->flags)) { |
2270 | len += sprintf(page+len, "%sblocked", sep); | |
2271 | sep = ","; | |
2272 | } | |
b2d444d7 N |
2273 | if (!test_bit(Faulty, &rdev->flags) && |
2274 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2275 | len += sprintf(page+len, "%sspare", sep); |
2276 | sep = ","; | |
2277 | } | |
2278 | return len+sprintf(page+len, "\n"); | |
2279 | } | |
2280 | ||
45dc2de1 N |
2281 | static ssize_t |
2282 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2283 | { | |
2284 | /* can write | |
2285 | * faulty - simulates and error | |
2286 | * remove - disconnects the device | |
f655675b N |
2287 | * writemostly - sets write_mostly |
2288 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2289 | * blocked - sets the Blocked flag |
2290 | * -blocked - clears the Blocked flag | |
6d56e278 | 2291 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2292 | */ |
2293 | int err = -EINVAL; | |
2294 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2295 | md_error(rdev->mddev, rdev); | |
2296 | err = 0; | |
2297 | } else if (cmd_match(buf, "remove")) { | |
2298 | if (rdev->raid_disk >= 0) | |
2299 | err = -EBUSY; | |
2300 | else { | |
2301 | mddev_t *mddev = rdev->mddev; | |
2302 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2303 | if (mddev->pers) |
2304 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2305 | md_new_event(mddev); |
2306 | err = 0; | |
2307 | } | |
f655675b N |
2308 | } else if (cmd_match(buf, "writemostly")) { |
2309 | set_bit(WriteMostly, &rdev->flags); | |
2310 | err = 0; | |
2311 | } else if (cmd_match(buf, "-writemostly")) { | |
2312 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2313 | err = 0; |
2314 | } else if (cmd_match(buf, "blocked")) { | |
2315 | set_bit(Blocked, &rdev->flags); | |
2316 | err = 0; | |
2317 | } else if (cmd_match(buf, "-blocked")) { | |
2318 | clear_bit(Blocked, &rdev->flags); | |
2319 | wake_up(&rdev->blocked_wait); | |
2320 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2321 | md_wakeup_thread(rdev->mddev->thread); | |
2322 | ||
6d56e278 N |
2323 | err = 0; |
2324 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2325 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2326 | err = 0; |
45dc2de1 | 2327 | } |
3c0ee63a N |
2328 | if (!err && rdev->sysfs_state) |
2329 | sysfs_notify_dirent(rdev->sysfs_state); | |
45dc2de1 N |
2330 | return err ? err : len; |
2331 | } | |
80ca3a44 N |
2332 | static struct rdev_sysfs_entry rdev_state = |
2333 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2334 | |
4dbcdc75 N |
2335 | static ssize_t |
2336 | errors_show(mdk_rdev_t *rdev, char *page) | |
2337 | { | |
2338 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2339 | } | |
2340 | ||
2341 | static ssize_t | |
2342 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2343 | { | |
2344 | char *e; | |
2345 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2346 | if (*buf && (*e == 0 || *e == '\n')) { | |
2347 | atomic_set(&rdev->corrected_errors, n); | |
2348 | return len; | |
2349 | } | |
2350 | return -EINVAL; | |
2351 | } | |
2352 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2353 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2354 | |
014236d2 N |
2355 | static ssize_t |
2356 | slot_show(mdk_rdev_t *rdev, char *page) | |
2357 | { | |
2358 | if (rdev->raid_disk < 0) | |
2359 | return sprintf(page, "none\n"); | |
2360 | else | |
2361 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2362 | } | |
2363 | ||
2364 | static ssize_t | |
2365 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2366 | { | |
2367 | char *e; | |
c303da6d N |
2368 | int err; |
2369 | char nm[20]; | |
014236d2 N |
2370 | int slot = simple_strtoul(buf, &e, 10); |
2371 | if (strncmp(buf, "none", 4)==0) | |
2372 | slot = -1; | |
2373 | else if (e==buf || (*e && *e!= '\n')) | |
2374 | return -EINVAL; | |
6c2fce2e | 2375 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2376 | /* Setting 'slot' on an active array requires also |
2377 | * updating the 'rd%d' link, and communicating | |
2378 | * with the personality with ->hot_*_disk. | |
2379 | * For now we only support removing | |
2380 | * failed/spare devices. This normally happens automatically, | |
2381 | * but not when the metadata is externally managed. | |
2382 | */ | |
c303da6d N |
2383 | if (rdev->raid_disk == -1) |
2384 | return -EEXIST; | |
2385 | /* personality does all needed checks */ | |
2386 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2387 | return -EINVAL; | |
2388 | err = rdev->mddev->pers-> | |
2389 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2390 | if (err) | |
2391 | return err; | |
2392 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2393 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
b7103107 | 2394 | rdev->raid_disk = -1; |
c303da6d N |
2395 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2396 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2397 | } else if (rdev->mddev->pers) { |
2398 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2399 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2400 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2401 | */ |
2402 | ||
2403 | if (rdev->raid_disk != -1) | |
2404 | return -EBUSY; | |
2405 | ||
2406 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2407 | return -EINVAL; | |
2408 | ||
159ec1fc | 2409 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2410 | if (rdev2->raid_disk == slot) |
2411 | return -EEXIST; | |
2412 | ||
2413 | rdev->raid_disk = slot; | |
2414 | if (test_bit(In_sync, &rdev->flags)) | |
2415 | rdev->saved_raid_disk = slot; | |
2416 | else | |
2417 | rdev->saved_raid_disk = -1; | |
2418 | err = rdev->mddev->pers-> | |
2419 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2420 | if (err) { |
6c2fce2e | 2421 | rdev->raid_disk = -1; |
6c2fce2e | 2422 | return err; |
52664732 | 2423 | } else |
3c0ee63a | 2424 | sysfs_notify_dirent(rdev->sysfs_state); |
6c2fce2e NB |
2425 | sprintf(nm, "rd%d", rdev->raid_disk); |
2426 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
2427 | printk(KERN_WARNING | |
2428 | "md: cannot register " | |
2429 | "%s for %s\n", | |
2430 | nm, mdname(rdev->mddev)); | |
2431 | ||
2432 | /* don't wakeup anyone, leave that to userspace. */ | |
c303da6d N |
2433 | } else { |
2434 | if (slot >= rdev->mddev->raid_disks) | |
2435 | return -ENOSPC; | |
2436 | rdev->raid_disk = slot; | |
2437 | /* assume it is working */ | |
c5d79adb N |
2438 | clear_bit(Faulty, &rdev->flags); |
2439 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2440 | set_bit(In_sync, &rdev->flags); |
3c0ee63a | 2441 | sysfs_notify_dirent(rdev->sysfs_state); |
c303da6d | 2442 | } |
014236d2 N |
2443 | return len; |
2444 | } | |
2445 | ||
2446 | ||
2447 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2448 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2449 | |
93c8cad0 N |
2450 | static ssize_t |
2451 | offset_show(mdk_rdev_t *rdev, char *page) | |
2452 | { | |
6961ece4 | 2453 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2454 | } |
2455 | ||
2456 | static ssize_t | |
2457 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2458 | { | |
2459 | char *e; | |
2460 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2461 | if (e==buf || (*e && *e != '\n')) | |
2462 | return -EINVAL; | |
8ed0a521 | 2463 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2464 | return -EBUSY; |
dd8ac336 | 2465 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2466 | /* Must set offset before size, so overlap checks |
2467 | * can be sane */ | |
2468 | return -EBUSY; | |
93c8cad0 N |
2469 | rdev->data_offset = offset; |
2470 | return len; | |
2471 | } | |
2472 | ||
2473 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2474 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2475 | |
83303b61 N |
2476 | static ssize_t |
2477 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2478 | { | |
dd8ac336 | 2479 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2480 | } |
2481 | ||
c5d79adb N |
2482 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2483 | { | |
2484 | /* check if two start/length pairs overlap */ | |
2485 | if (s1+l1 <= s2) | |
2486 | return 0; | |
2487 | if (s2+l2 <= s1) | |
2488 | return 0; | |
2489 | return 1; | |
2490 | } | |
2491 | ||
b522adcd DW |
2492 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2493 | { | |
2494 | unsigned long long blocks; | |
2495 | sector_t new; | |
2496 | ||
2497 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2498 | return -EINVAL; | |
2499 | ||
2500 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2501 | return -EINVAL; /* sector conversion overflow */ | |
2502 | ||
2503 | new = blocks * 2; | |
2504 | if (new != blocks * 2) | |
2505 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2506 | ||
2507 | *sectors = new; | |
2508 | return 0; | |
2509 | } | |
2510 | ||
83303b61 N |
2511 | static ssize_t |
2512 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2513 | { | |
27c529bb | 2514 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2515 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2516 | sector_t sectors; |
27c529bb | 2517 | |
b522adcd | 2518 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2519 | return -EINVAL; |
0cd17fec | 2520 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2521 | if (my_mddev->persistent) { |
dd8ac336 AN |
2522 | sectors = super_types[my_mddev->major_version]. |
2523 | rdev_size_change(rdev, sectors); | |
2524 | if (!sectors) | |
0cd17fec | 2525 | return -EBUSY; |
dd8ac336 AN |
2526 | } else if (!sectors) |
2527 | sectors = (rdev->bdev->bd_inode->i_size >> 9) - | |
2528 | rdev->data_offset; | |
0cd17fec | 2529 | } |
dd8ac336 | 2530 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2531 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2532 | |
dd8ac336 AN |
2533 | rdev->sectors = sectors; |
2534 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2535 | /* need to check that all other rdevs with the same ->bdev |
2536 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2537 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2538 | * we have to change it back, we will have the lock again. |
2539 | */ | |
2540 | mddev_t *mddev; | |
2541 | int overlap = 0; | |
159ec1fc | 2542 | struct list_head *tmp; |
c5d79adb | 2543 | |
27c529bb | 2544 | mddev_unlock(my_mddev); |
29ac4aa3 | 2545 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2546 | mdk_rdev_t *rdev2; |
2547 | ||
2548 | mddev_lock(mddev); | |
159ec1fc | 2549 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
c5d79adb N |
2550 | if (test_bit(AllReserved, &rdev2->flags) || |
2551 | (rdev->bdev == rdev2->bdev && | |
2552 | rdev != rdev2 && | |
dd8ac336 | 2553 | overlaps(rdev->data_offset, rdev->sectors, |
d07bd3bc | 2554 | rdev2->data_offset, |
dd8ac336 | 2555 | rdev2->sectors))) { |
c5d79adb N |
2556 | overlap = 1; |
2557 | break; | |
2558 | } | |
2559 | mddev_unlock(mddev); | |
2560 | if (overlap) { | |
2561 | mddev_put(mddev); | |
2562 | break; | |
2563 | } | |
2564 | } | |
27c529bb | 2565 | mddev_lock(my_mddev); |
c5d79adb N |
2566 | if (overlap) { |
2567 | /* Someone else could have slipped in a size | |
2568 | * change here, but doing so is just silly. | |
dd8ac336 | 2569 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2570 | * safe, and trust userspace not to race with |
2571 | * itself | |
2572 | */ | |
dd8ac336 | 2573 | rdev->sectors = oldsectors; |
c5d79adb N |
2574 | return -EBUSY; |
2575 | } | |
2576 | } | |
83303b61 N |
2577 | return len; |
2578 | } | |
2579 | ||
2580 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2581 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2582 | |
06e3c817 DW |
2583 | |
2584 | static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) | |
2585 | { | |
2586 | unsigned long long recovery_start = rdev->recovery_offset; | |
2587 | ||
2588 | if (test_bit(In_sync, &rdev->flags) || | |
2589 | recovery_start == MaxSector) | |
2590 | return sprintf(page, "none\n"); | |
2591 | ||
2592 | return sprintf(page, "%llu\n", recovery_start); | |
2593 | } | |
2594 | ||
2595 | static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2596 | { | |
2597 | unsigned long long recovery_start; | |
2598 | ||
2599 | if (cmd_match(buf, "none")) | |
2600 | recovery_start = MaxSector; | |
2601 | else if (strict_strtoull(buf, 10, &recovery_start)) | |
2602 | return -EINVAL; | |
2603 | ||
2604 | if (rdev->mddev->pers && | |
2605 | rdev->raid_disk >= 0) | |
2606 | return -EBUSY; | |
2607 | ||
2608 | rdev->recovery_offset = recovery_start; | |
2609 | if (recovery_start == MaxSector) | |
2610 | set_bit(In_sync, &rdev->flags); | |
2611 | else | |
2612 | clear_bit(In_sync, &rdev->flags); | |
2613 | return len; | |
2614 | } | |
2615 | ||
2616 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2617 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2618 | ||
86e6ffdd N |
2619 | static struct attribute *rdev_default_attrs[] = { |
2620 | &rdev_state.attr, | |
4dbcdc75 | 2621 | &rdev_errors.attr, |
014236d2 | 2622 | &rdev_slot.attr, |
93c8cad0 | 2623 | &rdev_offset.attr, |
83303b61 | 2624 | &rdev_size.attr, |
06e3c817 | 2625 | &rdev_recovery_start.attr, |
86e6ffdd N |
2626 | NULL, |
2627 | }; | |
2628 | static ssize_t | |
2629 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2630 | { | |
2631 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2632 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2633 | mddev_t *mddev = rdev->mddev; |
2634 | ssize_t rv; | |
86e6ffdd N |
2635 | |
2636 | if (!entry->show) | |
2637 | return -EIO; | |
27c529bb N |
2638 | |
2639 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2640 | if (!rv) { | |
2641 | if (rdev->mddev == NULL) | |
2642 | rv = -EBUSY; | |
2643 | else | |
2644 | rv = entry->show(rdev, page); | |
2645 | mddev_unlock(mddev); | |
2646 | } | |
2647 | return rv; | |
86e6ffdd N |
2648 | } |
2649 | ||
2650 | static ssize_t | |
2651 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2652 | const char *page, size_t length) | |
2653 | { | |
2654 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2655 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2656 | ssize_t rv; |
2657 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2658 | |
2659 | if (!entry->store) | |
2660 | return -EIO; | |
67463acb N |
2661 | if (!capable(CAP_SYS_ADMIN)) |
2662 | return -EACCES; | |
27c529bb | 2663 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2664 | if (!rv) { |
27c529bb N |
2665 | if (rdev->mddev == NULL) |
2666 | rv = -EBUSY; | |
2667 | else | |
2668 | rv = entry->store(rdev, page, length); | |
6a51830e | 2669 | mddev_unlock(mddev); |
ca388059 N |
2670 | } |
2671 | return rv; | |
86e6ffdd N |
2672 | } |
2673 | ||
2674 | static void rdev_free(struct kobject *ko) | |
2675 | { | |
2676 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2677 | kfree(rdev); | |
2678 | } | |
2679 | static struct sysfs_ops rdev_sysfs_ops = { | |
2680 | .show = rdev_attr_show, | |
2681 | .store = rdev_attr_store, | |
2682 | }; | |
2683 | static struct kobj_type rdev_ktype = { | |
2684 | .release = rdev_free, | |
2685 | .sysfs_ops = &rdev_sysfs_ops, | |
2686 | .default_attrs = rdev_default_attrs, | |
2687 | }; | |
2688 | ||
1da177e4 LT |
2689 | /* |
2690 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2691 | * | |
2692 | * mark the device faulty if: | |
2693 | * | |
2694 | * - the device is nonexistent (zero size) | |
2695 | * - the device has no valid superblock | |
2696 | * | |
2697 | * a faulty rdev _never_ has rdev->sb set. | |
2698 | */ | |
2699 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2700 | { | |
2701 | char b[BDEVNAME_SIZE]; | |
2702 | int err; | |
2703 | mdk_rdev_t *rdev; | |
2704 | sector_t size; | |
2705 | ||
9ffae0cf | 2706 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2707 | if (!rdev) { |
2708 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2709 | return ERR_PTR(-ENOMEM); | |
2710 | } | |
1da177e4 LT |
2711 | |
2712 | if ((err = alloc_disk_sb(rdev))) | |
2713 | goto abort_free; | |
2714 | ||
c5d79adb | 2715 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2716 | if (err) |
2717 | goto abort_free; | |
2718 | ||
f9cb074b | 2719 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2720 | |
1da177e4 | 2721 | rdev->desc_nr = -1; |
2b6e8459 | 2722 | rdev->saved_raid_disk = -1; |
3f9d7b0d | 2723 | rdev->raid_disk = -1; |
b2d444d7 | 2724 | rdev->flags = 0; |
1da177e4 | 2725 | rdev->data_offset = 0; |
42543769 | 2726 | rdev->sb_events = 0; |
1e50915f RB |
2727 | rdev->last_read_error.tv_sec = 0; |
2728 | rdev->last_read_error.tv_nsec = 0; | |
1da177e4 | 2729 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 2730 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 2731 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
2732 | |
2733 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
2734 | if (!size) { | |
2735 | printk(KERN_WARNING | |
2736 | "md: %s has zero or unknown size, marking faulty!\n", | |
2737 | bdevname(rdev->bdev,b)); | |
2738 | err = -EINVAL; | |
2739 | goto abort_free; | |
2740 | } | |
2741 | ||
2742 | if (super_format >= 0) { | |
2743 | err = super_types[super_format]. | |
2744 | load_super(rdev, NULL, super_minor); | |
2745 | if (err == -EINVAL) { | |
df968c4e N |
2746 | printk(KERN_WARNING |
2747 | "md: %s does not have a valid v%d.%d " | |
2748 | "superblock, not importing!\n", | |
2749 | bdevname(rdev->bdev,b), | |
2750 | super_format, super_minor); | |
1da177e4 LT |
2751 | goto abort_free; |
2752 | } | |
2753 | if (err < 0) { | |
2754 | printk(KERN_WARNING | |
2755 | "md: could not read %s's sb, not importing!\n", | |
2756 | bdevname(rdev->bdev,b)); | |
2757 | goto abort_free; | |
2758 | } | |
2759 | } | |
6bfe0b49 | 2760 | |
1da177e4 | 2761 | INIT_LIST_HEAD(&rdev->same_set); |
6bfe0b49 | 2762 | init_waitqueue_head(&rdev->blocked_wait); |
1da177e4 LT |
2763 | |
2764 | return rdev; | |
2765 | ||
2766 | abort_free: | |
2767 | if (rdev->sb_page) { | |
2768 | if (rdev->bdev) | |
2769 | unlock_rdev(rdev); | |
2770 | free_disk_sb(rdev); | |
2771 | } | |
2772 | kfree(rdev); | |
2773 | return ERR_PTR(err); | |
2774 | } | |
2775 | ||
2776 | /* | |
2777 | * Check a full RAID array for plausibility | |
2778 | */ | |
2779 | ||
2780 | ||
a757e64c | 2781 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2782 | { |
2783 | int i; | |
159ec1fc | 2784 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2785 | char b[BDEVNAME_SIZE]; |
2786 | ||
2787 | freshest = NULL; | |
d089c6af | 2788 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2789 | switch (super_types[mddev->major_version]. |
2790 | load_super(rdev, freshest, mddev->minor_version)) { | |
2791 | case 1: | |
2792 | freshest = rdev; | |
2793 | break; | |
2794 | case 0: | |
2795 | break; | |
2796 | default: | |
2797 | printk( KERN_ERR \ | |
2798 | "md: fatal superblock inconsistency in %s" | |
2799 | " -- removing from array\n", | |
2800 | bdevname(rdev->bdev,b)); | |
2801 | kick_rdev_from_array(rdev); | |
2802 | } | |
2803 | ||
2804 | ||
2805 | super_types[mddev->major_version]. | |
2806 | validate_super(mddev, freshest); | |
2807 | ||
2808 | i = 0; | |
d089c6af | 2809 | rdev_for_each(rdev, tmp, mddev) { |
233fca36 N |
2810 | if (mddev->max_disks && |
2811 | (rdev->desc_nr >= mddev->max_disks || | |
2812 | i > mddev->max_disks)) { | |
de01dfad N |
2813 | printk(KERN_WARNING |
2814 | "md: %s: %s: only %d devices permitted\n", | |
2815 | mdname(mddev), bdevname(rdev->bdev, b), | |
2816 | mddev->max_disks); | |
2817 | kick_rdev_from_array(rdev); | |
2818 | continue; | |
2819 | } | |
1da177e4 LT |
2820 | if (rdev != freshest) |
2821 | if (super_types[mddev->major_version]. | |
2822 | validate_super(mddev, rdev)) { | |
2823 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2824 | " from array!\n", | |
2825 | bdevname(rdev->bdev,b)); | |
2826 | kick_rdev_from_array(rdev); | |
2827 | continue; | |
2828 | } | |
2829 | if (mddev->level == LEVEL_MULTIPATH) { | |
2830 | rdev->desc_nr = i++; | |
2831 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2832 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 2833 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
2834 | rdev->raid_disk = -1; |
2835 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2836 | } |
2837 | } | |
1da177e4 LT |
2838 | } |
2839 | ||
72e02075 N |
2840 | /* Read a fixed-point number. |
2841 | * Numbers in sysfs attributes should be in "standard" units where | |
2842 | * possible, so time should be in seconds. | |
2843 | * However we internally use a a much smaller unit such as | |
2844 | * milliseconds or jiffies. | |
2845 | * This function takes a decimal number with a possible fractional | |
2846 | * component, and produces an integer which is the result of | |
2847 | * multiplying that number by 10^'scale'. | |
2848 | * all without any floating-point arithmetic. | |
2849 | */ | |
2850 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
2851 | { | |
2852 | unsigned long result = 0; | |
2853 | long decimals = -1; | |
2854 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
2855 | if (*cp == '.') | |
2856 | decimals = 0; | |
2857 | else if (decimals < scale) { | |
2858 | unsigned int value; | |
2859 | value = *cp - '0'; | |
2860 | result = result * 10 + value; | |
2861 | if (decimals >= 0) | |
2862 | decimals++; | |
2863 | } | |
2864 | cp++; | |
2865 | } | |
2866 | if (*cp == '\n') | |
2867 | cp++; | |
2868 | if (*cp) | |
2869 | return -EINVAL; | |
2870 | if (decimals < 0) | |
2871 | decimals = 0; | |
2872 | while (decimals < scale) { | |
2873 | result *= 10; | |
2874 | decimals ++; | |
2875 | } | |
2876 | *res = result; | |
2877 | return 0; | |
2878 | } | |
2879 | ||
2880 | ||
19052c0e N |
2881 | static void md_safemode_timeout(unsigned long data); |
2882 | ||
16f17b39 N |
2883 | static ssize_t |
2884 | safe_delay_show(mddev_t *mddev, char *page) | |
2885 | { | |
2886 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2887 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2888 | } | |
2889 | static ssize_t | |
2890 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2891 | { | |
16f17b39 | 2892 | unsigned long msec; |
97ce0a7f | 2893 | |
72e02075 | 2894 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 2895 | return -EINVAL; |
16f17b39 N |
2896 | if (msec == 0) |
2897 | mddev->safemode_delay = 0; | |
2898 | else { | |
19052c0e | 2899 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2900 | mddev->safemode_delay = (msec*HZ)/1000; |
2901 | if (mddev->safemode_delay == 0) | |
2902 | mddev->safemode_delay = 1; | |
19052c0e N |
2903 | if (mddev->safemode_delay < old_delay) |
2904 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2905 | } |
2906 | return len; | |
2907 | } | |
2908 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2909 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2910 | |
eae1701f | 2911 | static ssize_t |
96de1e66 | 2912 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2913 | { |
2604b703 | 2914 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2915 | if (p) |
eae1701f | 2916 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2917 | else if (mddev->clevel[0]) |
2918 | return sprintf(page, "%s\n", mddev->clevel); | |
2919 | else if (mddev->level != LEVEL_NONE) | |
2920 | return sprintf(page, "%d\n", mddev->level); | |
2921 | else | |
2922 | return 0; | |
eae1701f N |
2923 | } |
2924 | ||
d9d166c2 N |
2925 | static ssize_t |
2926 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2927 | { | |
f2859af6 | 2928 | char clevel[16]; |
20a49ff6 | 2929 | ssize_t rv = len; |
245f46c2 | 2930 | struct mdk_personality *pers; |
f2859af6 | 2931 | long level; |
245f46c2 | 2932 | void *priv; |
3a981b03 | 2933 | mdk_rdev_t *rdev; |
245f46c2 N |
2934 | |
2935 | if (mddev->pers == NULL) { | |
2936 | if (len == 0) | |
2937 | return 0; | |
2938 | if (len >= sizeof(mddev->clevel)) | |
2939 | return -ENOSPC; | |
2940 | strncpy(mddev->clevel, buf, len); | |
2941 | if (mddev->clevel[len-1] == '\n') | |
2942 | len--; | |
2943 | mddev->clevel[len] = 0; | |
2944 | mddev->level = LEVEL_NONE; | |
2945 | return rv; | |
2946 | } | |
2947 | ||
2948 | /* request to change the personality. Need to ensure: | |
2949 | * - array is not engaged in resync/recovery/reshape | |
2950 | * - old personality can be suspended | |
2951 | * - new personality will access other array. | |
2952 | */ | |
2953 | ||
2954 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) | |
d9d166c2 | 2955 | return -EBUSY; |
245f46c2 N |
2956 | |
2957 | if (!mddev->pers->quiesce) { | |
2958 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
2959 | mdname(mddev), mddev->pers->name); | |
2960 | return -EINVAL; | |
2961 | } | |
2962 | ||
2963 | /* Now find the new personality */ | |
f2859af6 | 2964 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 2965 | return -EINVAL; |
f2859af6 DW |
2966 | strncpy(clevel, buf, len); |
2967 | if (clevel[len-1] == '\n') | |
d9d166c2 | 2968 | len--; |
f2859af6 DW |
2969 | clevel[len] = 0; |
2970 | if (strict_strtol(clevel, 10, &level)) | |
2971 | level = LEVEL_NONE; | |
245f46c2 | 2972 | |
f2859af6 DW |
2973 | if (request_module("md-%s", clevel) != 0) |
2974 | request_module("md-level-%s", clevel); | |
245f46c2 | 2975 | spin_lock(&pers_lock); |
f2859af6 | 2976 | pers = find_pers(level, clevel); |
245f46c2 N |
2977 | if (!pers || !try_module_get(pers->owner)) { |
2978 | spin_unlock(&pers_lock); | |
f2859af6 | 2979 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
2980 | return -EINVAL; |
2981 | } | |
2982 | spin_unlock(&pers_lock); | |
2983 | ||
2984 | if (pers == mddev->pers) { | |
2985 | /* Nothing to do! */ | |
2986 | module_put(pers->owner); | |
2987 | return rv; | |
2988 | } | |
2989 | if (!pers->takeover) { | |
2990 | module_put(pers->owner); | |
2991 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 2992 | mdname(mddev), clevel); |
245f46c2 N |
2993 | return -EINVAL; |
2994 | } | |
2995 | ||
2996 | /* ->takeover must set new_* and/or delta_disks | |
2997 | * if it succeeds, and may set them when it fails. | |
2998 | */ | |
2999 | priv = pers->takeover(mddev); | |
3000 | if (IS_ERR(priv)) { | |
3001 | mddev->new_level = mddev->level; | |
3002 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3003 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3004 | mddev->raid_disks -= mddev->delta_disks; |
3005 | mddev->delta_disks = 0; | |
3006 | module_put(pers->owner); | |
3007 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3008 | mdname(mddev), clevel); |
245f46c2 N |
3009 | return PTR_ERR(priv); |
3010 | } | |
3011 | ||
3012 | /* Looks like we have a winner */ | |
3013 | mddev_suspend(mddev); | |
3014 | mddev->pers->stop(mddev); | |
a64c876f N |
3015 | |
3016 | if (mddev->pers->sync_request == NULL && | |
3017 | pers->sync_request != NULL) { | |
3018 | /* need to add the md_redundancy_group */ | |
3019 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3020 | printk(KERN_WARNING | |
3021 | "md: cannot register extra attributes for %s\n", | |
3022 | mdname(mddev)); | |
3023 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); | |
3024 | } | |
3025 | if (mddev->pers->sync_request != NULL && | |
3026 | pers->sync_request == NULL) { | |
3027 | /* need to remove the md_redundancy_group */ | |
3028 | if (mddev->to_remove == NULL) | |
3029 | mddev->to_remove = &md_redundancy_group; | |
3030 | } | |
3031 | ||
54071b38 TM |
3032 | if (mddev->pers->sync_request == NULL && |
3033 | mddev->external) { | |
3034 | /* We are converting from a no-redundancy array | |
3035 | * to a redundancy array and metadata is managed | |
3036 | * externally so we need to be sure that writes | |
3037 | * won't block due to a need to transition | |
3038 | * clean->dirty | |
3039 | * until external management is started. | |
3040 | */ | |
3041 | mddev->in_sync = 0; | |
3042 | mddev->safemode_delay = 0; | |
3043 | mddev->safemode = 0; | |
3044 | } | |
3045 | ||
245f46c2 | 3046 | module_put(mddev->pers->owner); |
3a981b03 N |
3047 | /* Invalidate devices that are now superfluous */ |
3048 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
3049 | if (rdev->raid_disk >= mddev->raid_disks) { | |
3050 | rdev->raid_disk = -1; | |
3051 | clear_bit(In_sync, &rdev->flags); | |
3052 | } | |
245f46c2 N |
3053 | mddev->pers = pers; |
3054 | mddev->private = priv; | |
3055 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3056 | mddev->level = mddev->new_level; | |
3057 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3058 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3059 | mddev->delta_disks = 0; |
9af204cf TM |
3060 | if (mddev->pers->sync_request == NULL) { |
3061 | /* this is now an array without redundancy, so | |
3062 | * it must always be in_sync | |
3063 | */ | |
3064 | mddev->in_sync = 1; | |
3065 | del_timer_sync(&mddev->safemode_timer); | |
3066 | } | |
245f46c2 N |
3067 | pers->run(mddev); |
3068 | mddev_resume(mddev); | |
3069 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3070 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3071 | md_wakeup_thread(mddev->thread); | |
5cac7861 | 3072 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3073 | md_new_event(mddev); |
d9d166c2 N |
3074 | return rv; |
3075 | } | |
3076 | ||
3077 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3078 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3079 | |
d4dbd025 N |
3080 | |
3081 | static ssize_t | |
3082 | layout_show(mddev_t *mddev, char *page) | |
3083 | { | |
3084 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3085 | if (mddev->reshape_position != MaxSector && |
3086 | mddev->layout != mddev->new_layout) | |
3087 | return sprintf(page, "%d (%d)\n", | |
3088 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3089 | return sprintf(page, "%d\n", mddev->layout); |
3090 | } | |
3091 | ||
3092 | static ssize_t | |
3093 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
3094 | { | |
3095 | char *e; | |
3096 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3097 | |
3098 | if (!*buf || (*e && *e != '\n')) | |
3099 | return -EINVAL; | |
3100 | ||
b3546035 N |
3101 | if (mddev->pers) { |
3102 | int err; | |
50ac168a | 3103 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3104 | return -EBUSY; |
597a711b | 3105 | mddev->new_layout = n; |
50ac168a | 3106 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3107 | if (err) { |
3108 | mddev->new_layout = mddev->layout; | |
b3546035 | 3109 | return err; |
597a711b | 3110 | } |
b3546035 | 3111 | } else { |
08a02ecd | 3112 | mddev->new_layout = n; |
b3546035 N |
3113 | if (mddev->reshape_position == MaxSector) |
3114 | mddev->layout = n; | |
3115 | } | |
d4dbd025 N |
3116 | return len; |
3117 | } | |
3118 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3119 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3120 | |
3121 | ||
eae1701f | 3122 | static ssize_t |
96de1e66 | 3123 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 3124 | { |
bb636547 N |
3125 | if (mddev->raid_disks == 0) |
3126 | return 0; | |
08a02ecd N |
3127 | if (mddev->reshape_position != MaxSector && |
3128 | mddev->delta_disks != 0) | |
3129 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3130 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3131 | return sprintf(page, "%d\n", mddev->raid_disks); |
3132 | } | |
3133 | ||
da943b99 N |
3134 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
3135 | ||
3136 | static ssize_t | |
3137 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
3138 | { | |
da943b99 N |
3139 | char *e; |
3140 | int rv = 0; | |
3141 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3142 | ||
3143 | if (!*buf || (*e && *e != '\n')) | |
3144 | return -EINVAL; | |
3145 | ||
3146 | if (mddev->pers) | |
3147 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
3148 | else if (mddev->reshape_position != MaxSector) { |
3149 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
3150 | mddev->delta_disks = n - olddisks; | |
3151 | mddev->raid_disks = n; | |
3152 | } else | |
da943b99 N |
3153 | mddev->raid_disks = n; |
3154 | return rv ? rv : len; | |
3155 | } | |
3156 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3157 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3158 | |
3b34380a N |
3159 | static ssize_t |
3160 | chunk_size_show(mddev_t *mddev, char *page) | |
3161 | { | |
08a02ecd | 3162 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3163 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3164 | return sprintf(page, "%d (%d)\n", | |
3165 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3166 | mddev->chunk_sectors << 9); |
3167 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3168 | } |
3169 | ||
3170 | static ssize_t | |
3171 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3172 | { | |
3b34380a N |
3173 | char *e; |
3174 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3175 | ||
3b34380a N |
3176 | if (!*buf || (*e && *e != '\n')) |
3177 | return -EINVAL; | |
3178 | ||
b3546035 N |
3179 | if (mddev->pers) { |
3180 | int err; | |
50ac168a | 3181 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3182 | return -EBUSY; |
597a711b | 3183 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3184 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3185 | if (err) { |
3186 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3187 | return err; |
597a711b | 3188 | } |
b3546035 | 3189 | } else { |
664e7c41 | 3190 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3191 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3192 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3193 | } |
3b34380a N |
3194 | return len; |
3195 | } | |
3196 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3197 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3198 | |
a94213b1 N |
3199 | static ssize_t |
3200 | resync_start_show(mddev_t *mddev, char *page) | |
3201 | { | |
d1a7c503 N |
3202 | if (mddev->recovery_cp == MaxSector) |
3203 | return sprintf(page, "none\n"); | |
a94213b1 N |
3204 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3205 | } | |
3206 | ||
3207 | static ssize_t | |
3208 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
3209 | { | |
a94213b1 N |
3210 | char *e; |
3211 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3212 | ||
3213 | if (mddev->pers) | |
3214 | return -EBUSY; | |
06e3c817 DW |
3215 | if (cmd_match(buf, "none")) |
3216 | n = MaxSector; | |
3217 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3218 | return -EINVAL; |
3219 | ||
3220 | mddev->recovery_cp = n; | |
3221 | return len; | |
3222 | } | |
3223 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3224 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3225 | |
9e653b63 N |
3226 | /* |
3227 | * The array state can be: | |
3228 | * | |
3229 | * clear | |
3230 | * No devices, no size, no level | |
3231 | * Equivalent to STOP_ARRAY ioctl | |
3232 | * inactive | |
3233 | * May have some settings, but array is not active | |
3234 | * all IO results in error | |
3235 | * When written, doesn't tear down array, but just stops it | |
3236 | * suspended (not supported yet) | |
3237 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3238 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3239 | * readonly |
3240 | * no resync can happen. no superblocks get written. | |
3241 | * write requests fail | |
3242 | * read-auto | |
3243 | * like readonly, but behaves like 'clean' on a write request. | |
3244 | * | |
3245 | * clean - no pending writes, but otherwise active. | |
3246 | * When written to inactive array, starts without resync | |
3247 | * If a write request arrives then | |
3248 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3249 | * if not known, block and switch to write-pending | |
3250 | * If written to an active array that has pending writes, then fails. | |
3251 | * active | |
3252 | * fully active: IO and resync can be happening. | |
3253 | * When written to inactive array, starts with resync | |
3254 | * | |
3255 | * write-pending | |
3256 | * clean, but writes are blocked waiting for 'active' to be written. | |
3257 | * | |
3258 | * active-idle | |
3259 | * like active, but no writes have been seen for a while (100msec). | |
3260 | * | |
3261 | */ | |
3262 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3263 | write_pending, active_idle, bad_word}; | |
05381954 | 3264 | static char *array_states[] = { |
9e653b63 N |
3265 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3266 | "write-pending", "active-idle", NULL }; | |
3267 | ||
3268 | static int match_word(const char *word, char **list) | |
3269 | { | |
3270 | int n; | |
3271 | for (n=0; list[n]; n++) | |
3272 | if (cmd_match(word, list[n])) | |
3273 | break; | |
3274 | return n; | |
3275 | } | |
3276 | ||
3277 | static ssize_t | |
3278 | array_state_show(mddev_t *mddev, char *page) | |
3279 | { | |
3280 | enum array_state st = inactive; | |
3281 | ||
3282 | if (mddev->pers) | |
3283 | switch(mddev->ro) { | |
3284 | case 1: | |
3285 | st = readonly; | |
3286 | break; | |
3287 | case 2: | |
3288 | st = read_auto; | |
3289 | break; | |
3290 | case 0: | |
3291 | if (mddev->in_sync) | |
3292 | st = clean; | |
e691063a N |
3293 | else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) |
3294 | st = write_pending; | |
9e653b63 N |
3295 | else if (mddev->safemode) |
3296 | st = active_idle; | |
3297 | else | |
3298 | st = active; | |
3299 | } | |
3300 | else { | |
3301 | if (list_empty(&mddev->disks) && | |
3302 | mddev->raid_disks == 0 && | |
58c0fed4 | 3303 | mddev->dev_sectors == 0) |
9e653b63 N |
3304 | st = clear; |
3305 | else | |
3306 | st = inactive; | |
3307 | } | |
3308 | return sprintf(page, "%s\n", array_states[st]); | |
3309 | } | |
3310 | ||
df5b20cf | 3311 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
a4bd82d0 | 3312 | static int md_set_readonly(mddev_t * mddev, int is_open); |
9e653b63 N |
3313 | static int do_md_run(mddev_t * mddev); |
3314 | static int restart_array(mddev_t *mddev); | |
3315 | ||
3316 | static ssize_t | |
3317 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3318 | { | |
3319 | int err = -EINVAL; | |
3320 | enum array_state st = match_word(buf, array_states); | |
3321 | switch(st) { | |
3322 | case bad_word: | |
3323 | break; | |
3324 | case clear: | |
3325 | /* stopping an active array */ | |
f2ea68cf | 3326 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3327 | return -EBUSY; |
df5b20cf | 3328 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3329 | break; |
3330 | case inactive: | |
3331 | /* stopping an active array */ | |
3332 | if (mddev->pers) { | |
f2ea68cf | 3333 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3334 | return -EBUSY; |
df5b20cf | 3335 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3336 | } else |
3337 | err = 0; /* already inactive */ | |
9e653b63 N |
3338 | break; |
3339 | case suspended: | |
3340 | break; /* not supported yet */ | |
3341 | case readonly: | |
3342 | if (mddev->pers) | |
a4bd82d0 | 3343 | err = md_set_readonly(mddev, 0); |
9e653b63 N |
3344 | else { |
3345 | mddev->ro = 1; | |
648b629e | 3346 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3347 | err = do_md_run(mddev); |
3348 | } | |
3349 | break; | |
3350 | case read_auto: | |
9e653b63 | 3351 | if (mddev->pers) { |
80268ee9 | 3352 | if (mddev->ro == 0) |
a4bd82d0 | 3353 | err = md_set_readonly(mddev, 0); |
80268ee9 | 3354 | else if (mddev->ro == 1) |
648b629e N |
3355 | err = restart_array(mddev); |
3356 | if (err == 0) { | |
3357 | mddev->ro = 2; | |
3358 | set_disk_ro(mddev->gendisk, 0); | |
3359 | } | |
9e653b63 N |
3360 | } else { |
3361 | mddev->ro = 2; | |
3362 | err = do_md_run(mddev); | |
3363 | } | |
3364 | break; | |
3365 | case clean: | |
3366 | if (mddev->pers) { | |
3367 | restart_array(mddev); | |
3368 | spin_lock_irq(&mddev->write_lock); | |
3369 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3370 | if (mddev->in_sync == 0) { |
3371 | mddev->in_sync = 1; | |
31a59e34 N |
3372 | if (mddev->safemode == 1) |
3373 | mddev->safemode = 0; | |
e691063a N |
3374 | if (mddev->persistent) |
3375 | set_bit(MD_CHANGE_CLEAN, | |
3376 | &mddev->flags); | |
3377 | } | |
3378 | err = 0; | |
3379 | } else | |
3380 | err = -EBUSY; | |
9e653b63 | 3381 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3382 | } else |
3383 | err = -EINVAL; | |
9e653b63 N |
3384 | break; |
3385 | case active: | |
3386 | if (mddev->pers) { | |
3387 | restart_array(mddev); | |
e691063a N |
3388 | if (mddev->external) |
3389 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
9e653b63 N |
3390 | wake_up(&mddev->sb_wait); |
3391 | err = 0; | |
3392 | } else { | |
3393 | mddev->ro = 0; | |
648b629e | 3394 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3395 | err = do_md_run(mddev); |
3396 | } | |
3397 | break; | |
3398 | case write_pending: | |
3399 | case active_idle: | |
3400 | /* these cannot be set */ | |
3401 | break; | |
3402 | } | |
3403 | if (err) | |
3404 | return err; | |
0fd62b86 | 3405 | else { |
b62b7590 | 3406 | sysfs_notify_dirent(mddev->sysfs_state); |
9e653b63 | 3407 | return len; |
0fd62b86 | 3408 | } |
9e653b63 | 3409 | } |
80ca3a44 N |
3410 | static struct md_sysfs_entry md_array_state = |
3411 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3412 | |
1e50915f RB |
3413 | static ssize_t |
3414 | max_corrected_read_errors_show(mddev_t *mddev, char *page) { | |
3415 | return sprintf(page, "%d\n", | |
3416 | atomic_read(&mddev->max_corr_read_errors)); | |
3417 | } | |
3418 | ||
3419 | static ssize_t | |
3420 | max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) | |
3421 | { | |
3422 | char *e; | |
3423 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3424 | ||
3425 | if (*buf && (*e == 0 || *e == '\n')) { | |
3426 | atomic_set(&mddev->max_corr_read_errors, n); | |
3427 | return len; | |
3428 | } | |
3429 | return -EINVAL; | |
3430 | } | |
3431 | ||
3432 | static struct md_sysfs_entry max_corr_read_errors = | |
3433 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3434 | max_corrected_read_errors_store); | |
3435 | ||
6d7ff738 N |
3436 | static ssize_t |
3437 | null_show(mddev_t *mddev, char *page) | |
3438 | { | |
3439 | return -EINVAL; | |
3440 | } | |
3441 | ||
3442 | static ssize_t | |
3443 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3444 | { | |
3445 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3446 | /* The new device is added to the array. | |
3447 | * If the array has a persistent superblock, we read the | |
3448 | * superblock to initialise info and check validity. | |
3449 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3450 | * which mainly checks size. | |
3451 | */ | |
3452 | char *e; | |
3453 | int major = simple_strtoul(buf, &e, 10); | |
3454 | int minor; | |
3455 | dev_t dev; | |
3456 | mdk_rdev_t *rdev; | |
3457 | int err; | |
3458 | ||
3459 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3460 | return -EINVAL; | |
3461 | minor = simple_strtoul(e+1, &e, 10); | |
3462 | if (*e && *e != '\n') | |
3463 | return -EINVAL; | |
3464 | dev = MKDEV(major, minor); | |
3465 | if (major != MAJOR(dev) || | |
3466 | minor != MINOR(dev)) | |
3467 | return -EOVERFLOW; | |
3468 | ||
3469 | ||
3470 | if (mddev->persistent) { | |
3471 | rdev = md_import_device(dev, mddev->major_version, | |
3472 | mddev->minor_version); | |
3473 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3474 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3475 | mdk_rdev_t, same_set); | |
3476 | err = super_types[mddev->major_version] | |
3477 | .load_super(rdev, rdev0, mddev->minor_version); | |
3478 | if (err < 0) | |
3479 | goto out; | |
3480 | } | |
c5d79adb N |
3481 | } else if (mddev->external) |
3482 | rdev = md_import_device(dev, -2, -1); | |
3483 | else | |
6d7ff738 N |
3484 | rdev = md_import_device(dev, -1, -1); |
3485 | ||
3486 | if (IS_ERR(rdev)) | |
3487 | return PTR_ERR(rdev); | |
3488 | err = bind_rdev_to_array(rdev, mddev); | |
3489 | out: | |
3490 | if (err) | |
3491 | export_rdev(rdev); | |
3492 | return err ? err : len; | |
3493 | } | |
3494 | ||
3495 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3496 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3497 | |
9b1d1dac PC |
3498 | static ssize_t |
3499 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3500 | { | |
3501 | char *end; | |
3502 | unsigned long chunk, end_chunk; | |
3503 | ||
3504 | if (!mddev->bitmap) | |
3505 | goto out; | |
3506 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3507 | while (*buf) { | |
3508 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3509 | if (buf == end) break; | |
3510 | if (*end == '-') { /* range */ | |
3511 | buf = end + 1; | |
3512 | end_chunk = simple_strtoul(buf, &end, 0); | |
3513 | if (buf == end) break; | |
3514 | } | |
3515 | if (*end && !isspace(*end)) break; | |
3516 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3517 | buf = skip_spaces(end); |
9b1d1dac PC |
3518 | } |
3519 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3520 | out: | |
3521 | return len; | |
3522 | } | |
3523 | ||
3524 | static struct md_sysfs_entry md_bitmap = | |
3525 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3526 | ||
a35b0d69 N |
3527 | static ssize_t |
3528 | size_show(mddev_t *mddev, char *page) | |
3529 | { | |
58c0fed4 AN |
3530 | return sprintf(page, "%llu\n", |
3531 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3532 | } |
3533 | ||
d71f9f88 | 3534 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3535 | |
3536 | static ssize_t | |
3537 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3538 | { | |
3539 | /* If array is inactive, we can reduce the component size, but | |
3540 | * not increase it (except from 0). | |
3541 | * If array is active, we can try an on-line resize | |
3542 | */ | |
b522adcd DW |
3543 | sector_t sectors; |
3544 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3545 | |
58c0fed4 AN |
3546 | if (err < 0) |
3547 | return err; | |
a35b0d69 | 3548 | if (mddev->pers) { |
58c0fed4 | 3549 | err = update_size(mddev, sectors); |
850b2b42 | 3550 | md_update_sb(mddev, 1); |
a35b0d69 | 3551 | } else { |
58c0fed4 AN |
3552 | if (mddev->dev_sectors == 0 || |
3553 | mddev->dev_sectors > sectors) | |
3554 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3555 | else |
3556 | err = -ENOSPC; | |
3557 | } | |
3558 | return err ? err : len; | |
3559 | } | |
3560 | ||
3561 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3562 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3563 | |
8bb93aac N |
3564 | |
3565 | /* Metdata version. | |
e691063a N |
3566 | * This is one of |
3567 | * 'none' for arrays with no metadata (good luck...) | |
3568 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3569 | * or N.M for internally known formats |
3570 | */ | |
3571 | static ssize_t | |
3572 | metadata_show(mddev_t *mddev, char *page) | |
3573 | { | |
3574 | if (mddev->persistent) | |
3575 | return sprintf(page, "%d.%d\n", | |
3576 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3577 | else if (mddev->external) |
3578 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3579 | else |
3580 | return sprintf(page, "none\n"); | |
3581 | } | |
3582 | ||
3583 | static ssize_t | |
3584 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3585 | { | |
3586 | int major, minor; | |
3587 | char *e; | |
ea43ddd8 N |
3588 | /* Changing the details of 'external' metadata is |
3589 | * always permitted. Otherwise there must be | |
3590 | * no devices attached to the array. | |
3591 | */ | |
3592 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3593 | ; | |
3594 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3595 | return -EBUSY; |
3596 | ||
3597 | if (cmd_match(buf, "none")) { | |
3598 | mddev->persistent = 0; | |
e691063a N |
3599 | mddev->external = 0; |
3600 | mddev->major_version = 0; | |
3601 | mddev->minor_version = 90; | |
3602 | return len; | |
3603 | } | |
3604 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3605 | size_t namelen = len-9; |
e691063a N |
3606 | if (namelen >= sizeof(mddev->metadata_type)) |
3607 | namelen = sizeof(mddev->metadata_type)-1; | |
3608 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3609 | mddev->metadata_type[namelen] = 0; | |
3610 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3611 | mddev->metadata_type[--namelen] = 0; | |
3612 | mddev->persistent = 0; | |
3613 | mddev->external = 1; | |
8bb93aac N |
3614 | mddev->major_version = 0; |
3615 | mddev->minor_version = 90; | |
3616 | return len; | |
3617 | } | |
3618 | major = simple_strtoul(buf, &e, 10); | |
3619 | if (e==buf || *e != '.') | |
3620 | return -EINVAL; | |
3621 | buf = e+1; | |
3622 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3623 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3624 | return -EINVAL; |
50511da3 | 3625 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3626 | return -ENOENT; |
3627 | mddev->major_version = major; | |
3628 | mddev->minor_version = minor; | |
3629 | mddev->persistent = 1; | |
e691063a | 3630 | mddev->external = 0; |
8bb93aac N |
3631 | return len; |
3632 | } | |
3633 | ||
3634 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3635 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3636 | |
24dd469d | 3637 | static ssize_t |
7eec314d | 3638 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3639 | { |
7eec314d | 3640 | char *type = "idle"; |
b6a9ce68 N |
3641 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3642 | type = "frozen"; | |
3643 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3644 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3645 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3646 | type = "reshape"; | |
3647 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3648 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3649 | type = "resync"; | |
3650 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3651 | type = "check"; | |
3652 | else | |
3653 | type = "repair"; | |
72a23c21 | 3654 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3655 | type = "recover"; |
3656 | } | |
3657 | return sprintf(page, "%s\n", type); | |
3658 | } | |
3659 | ||
3660 | static ssize_t | |
7eec314d | 3661 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3662 | { |
7eec314d N |
3663 | if (!mddev->pers || !mddev->pers->sync_request) |
3664 | return -EINVAL; | |
3665 | ||
b6a9ce68 N |
3666 | if (cmd_match(page, "frozen")) |
3667 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3668 | else | |
3669 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3670 | ||
3671 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3672 | if (mddev->sync_thread) { |
3673 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
3674 | md_unregister_thread(mddev->sync_thread); | |
3675 | mddev->sync_thread = NULL; | |
3676 | mddev->recovery = 0; | |
3677 | } | |
03c902e1 N |
3678 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3679 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3680 | return -EBUSY; |
72a23c21 NB |
3681 | else if (cmd_match(page, "resync")) |
3682 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3683 | else if (cmd_match(page, "recover")) { | |
3684 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3685 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3686 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3687 | int err; |
3688 | if (mddev->pers->start_reshape == NULL) | |
3689 | return -EINVAL; | |
3690 | err = mddev->pers->start_reshape(mddev); | |
3691 | if (err) | |
3692 | return err; | |
a99ac971 | 3693 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3694 | } else { |
bce74dac | 3695 | if (cmd_match(page, "check")) |
7eec314d | 3696 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3697 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3698 | return -EINVAL; |
3699 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3700 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3701 | } |
03c902e1 | 3702 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3703 | md_wakeup_thread(mddev->thread); |
0c3573f1 | 3704 | sysfs_notify_dirent(mddev->sysfs_action); |
24dd469d N |
3705 | return len; |
3706 | } | |
3707 | ||
9d88883e | 3708 | static ssize_t |
96de1e66 | 3709 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3710 | { |
3711 | return sprintf(page, "%llu\n", | |
3712 | (unsigned long long) mddev->resync_mismatches); | |
3713 | } | |
3714 | ||
80ca3a44 N |
3715 | static struct md_sysfs_entry md_scan_mode = |
3716 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3717 | |
96de1e66 | 3718 | |
80ca3a44 | 3719 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3720 | |
88202a0c N |
3721 | static ssize_t |
3722 | sync_min_show(mddev_t *mddev, char *page) | |
3723 | { | |
3724 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3725 | mddev->sync_speed_min ? "local": "system"); | |
3726 | } | |
3727 | ||
3728 | static ssize_t | |
3729 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3730 | { | |
3731 | int min; | |
3732 | char *e; | |
3733 | if (strncmp(buf, "system", 6)==0) { | |
3734 | mddev->sync_speed_min = 0; | |
3735 | return len; | |
3736 | } | |
3737 | min = simple_strtoul(buf, &e, 10); | |
3738 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3739 | return -EINVAL; | |
3740 | mddev->sync_speed_min = min; | |
3741 | return len; | |
3742 | } | |
3743 | ||
3744 | static struct md_sysfs_entry md_sync_min = | |
3745 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3746 | ||
3747 | static ssize_t | |
3748 | sync_max_show(mddev_t *mddev, char *page) | |
3749 | { | |
3750 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3751 | mddev->sync_speed_max ? "local": "system"); | |
3752 | } | |
3753 | ||
3754 | static ssize_t | |
3755 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3756 | { | |
3757 | int max; | |
3758 | char *e; | |
3759 | if (strncmp(buf, "system", 6)==0) { | |
3760 | mddev->sync_speed_max = 0; | |
3761 | return len; | |
3762 | } | |
3763 | max = simple_strtoul(buf, &e, 10); | |
3764 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3765 | return -EINVAL; | |
3766 | mddev->sync_speed_max = max; | |
3767 | return len; | |
3768 | } | |
3769 | ||
3770 | static struct md_sysfs_entry md_sync_max = | |
3771 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3772 | ||
d7f3d291 IP |
3773 | static ssize_t |
3774 | degraded_show(mddev_t *mddev, char *page) | |
3775 | { | |
3776 | return sprintf(page, "%d\n", mddev->degraded); | |
3777 | } | |
3778 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3779 | |
90b08710 BS |
3780 | static ssize_t |
3781 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3782 | { | |
3783 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3784 | } | |
3785 | ||
3786 | static ssize_t | |
3787 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3788 | { | |
3789 | long n; | |
3790 | ||
3791 | if (strict_strtol(buf, 10, &n)) | |
3792 | return -EINVAL; | |
3793 | ||
3794 | if (n != 0 && n != 1) | |
3795 | return -EINVAL; | |
3796 | ||
3797 | mddev->parallel_resync = n; | |
3798 | ||
3799 | if (mddev->sync_thread) | |
3800 | wake_up(&resync_wait); | |
3801 | ||
3802 | return len; | |
3803 | } | |
3804 | ||
3805 | /* force parallel resync, even with shared block devices */ | |
3806 | static struct md_sysfs_entry md_sync_force_parallel = | |
3807 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3808 | sync_force_parallel_show, sync_force_parallel_store); | |
3809 | ||
88202a0c N |
3810 | static ssize_t |
3811 | sync_speed_show(mddev_t *mddev, char *page) | |
3812 | { | |
3813 | unsigned long resync, dt, db; | |
d1a7c503 N |
3814 | if (mddev->curr_resync == 0) |
3815 | return sprintf(page, "none\n"); | |
9687a60c AN |
3816 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3817 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3818 | if (!dt) dt++; |
9687a60c AN |
3819 | db = resync - mddev->resync_mark_cnt; |
3820 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3821 | } |
3822 | ||
80ca3a44 | 3823 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3824 | |
3825 | static ssize_t | |
3826 | sync_completed_show(mddev_t *mddev, char *page) | |
3827 | { | |
58c0fed4 | 3828 | unsigned long max_sectors, resync; |
88202a0c | 3829 | |
acb180b0 N |
3830 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3831 | return sprintf(page, "none\n"); | |
3832 | ||
88202a0c | 3833 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3834 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3835 | else |
58c0fed4 | 3836 | max_sectors = mddev->dev_sectors; |
88202a0c | 3837 | |
acb180b0 | 3838 | resync = mddev->curr_resync_completed; |
58c0fed4 | 3839 | return sprintf(page, "%lu / %lu\n", resync, max_sectors); |
88202a0c N |
3840 | } |
3841 | ||
80ca3a44 | 3842 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3843 | |
5e96ee65 NB |
3844 | static ssize_t |
3845 | min_sync_show(mddev_t *mddev, char *page) | |
3846 | { | |
3847 | return sprintf(page, "%llu\n", | |
3848 | (unsigned long long)mddev->resync_min); | |
3849 | } | |
3850 | static ssize_t | |
3851 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3852 | { | |
3853 | unsigned long long min; | |
3854 | if (strict_strtoull(buf, 10, &min)) | |
3855 | return -EINVAL; | |
3856 | if (min > mddev->resync_max) | |
3857 | return -EINVAL; | |
3858 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3859 | return -EBUSY; | |
3860 | ||
3861 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3862 | if (mddev->chunk_sectors) { |
2ac06c33 | 3863 | sector_t temp = min; |
9d8f0363 | 3864 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3865 | return -EINVAL; |
3866 | } | |
3867 | mddev->resync_min = min; | |
3868 | ||
3869 | return len; | |
3870 | } | |
3871 | ||
3872 | static struct md_sysfs_entry md_min_sync = | |
3873 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3874 | ||
c6207277 N |
3875 | static ssize_t |
3876 | max_sync_show(mddev_t *mddev, char *page) | |
3877 | { | |
3878 | if (mddev->resync_max == MaxSector) | |
3879 | return sprintf(page, "max\n"); | |
3880 | else | |
3881 | return sprintf(page, "%llu\n", | |
3882 | (unsigned long long)mddev->resync_max); | |
3883 | } | |
3884 | static ssize_t | |
3885 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3886 | { | |
3887 | if (strncmp(buf, "max", 3) == 0) | |
3888 | mddev->resync_max = MaxSector; | |
3889 | else { | |
5e96ee65 NB |
3890 | unsigned long long max; |
3891 | if (strict_strtoull(buf, 10, &max)) | |
3892 | return -EINVAL; | |
3893 | if (max < mddev->resync_min) | |
c6207277 N |
3894 | return -EINVAL; |
3895 | if (max < mddev->resync_max && | |
4d484a4a | 3896 | mddev->ro == 0 && |
c6207277 N |
3897 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3898 | return -EBUSY; | |
3899 | ||
3900 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3901 | if (mddev->chunk_sectors) { |
2ac06c33 | 3902 | sector_t temp = max; |
9d8f0363 | 3903 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
3904 | return -EINVAL; |
3905 | } | |
3906 | mddev->resync_max = max; | |
3907 | } | |
3908 | wake_up(&mddev->recovery_wait); | |
3909 | return len; | |
3910 | } | |
3911 | ||
3912 | static struct md_sysfs_entry md_max_sync = | |
3913 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
3914 | ||
e464eafd N |
3915 | static ssize_t |
3916 | suspend_lo_show(mddev_t *mddev, char *page) | |
3917 | { | |
3918 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
3919 | } | |
3920 | ||
3921 | static ssize_t | |
3922 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
3923 | { | |
3924 | char *e; | |
3925 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3926 | ||
b8d966ef N |
3927 | if (mddev->pers == NULL || |
3928 | mddev->pers->quiesce == NULL) | |
e464eafd N |
3929 | return -EINVAL; |
3930 | if (buf == e || (*e && *e != '\n')) | |
3931 | return -EINVAL; | |
3932 | if (new >= mddev->suspend_hi || | |
3933 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
3934 | mddev->suspend_lo = new; | |
3935 | mddev->pers->quiesce(mddev, 2); | |
3936 | return len; | |
3937 | } else | |
3938 | return -EINVAL; | |
3939 | } | |
3940 | static struct md_sysfs_entry md_suspend_lo = | |
3941 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
3942 | ||
3943 | ||
3944 | static ssize_t | |
3945 | suspend_hi_show(mddev_t *mddev, char *page) | |
3946 | { | |
3947 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
3948 | } | |
3949 | ||
3950 | static ssize_t | |
3951 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
3952 | { | |
3953 | char *e; | |
3954 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3955 | ||
b8d966ef N |
3956 | if (mddev->pers == NULL || |
3957 | mddev->pers->quiesce == NULL) | |
e464eafd N |
3958 | return -EINVAL; |
3959 | if (buf == e || (*e && *e != '\n')) | |
3960 | return -EINVAL; | |
3961 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
3962 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
3963 | mddev->suspend_hi = new; | |
3964 | mddev->pers->quiesce(mddev, 1); | |
3965 | mddev->pers->quiesce(mddev, 0); | |
3966 | return len; | |
3967 | } else | |
3968 | return -EINVAL; | |
3969 | } | |
3970 | static struct md_sysfs_entry md_suspend_hi = | |
3971 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
3972 | ||
08a02ecd N |
3973 | static ssize_t |
3974 | reshape_position_show(mddev_t *mddev, char *page) | |
3975 | { | |
3976 | if (mddev->reshape_position != MaxSector) | |
3977 | return sprintf(page, "%llu\n", | |
3978 | (unsigned long long)mddev->reshape_position); | |
3979 | strcpy(page, "none\n"); | |
3980 | return 5; | |
3981 | } | |
3982 | ||
3983 | static ssize_t | |
3984 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
3985 | { | |
3986 | char *e; | |
3987 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3988 | if (mddev->pers) | |
3989 | return -EBUSY; | |
3990 | if (buf == e || (*e && *e != '\n')) | |
3991 | return -EINVAL; | |
3992 | mddev->reshape_position = new; | |
3993 | mddev->delta_disks = 0; | |
3994 | mddev->new_level = mddev->level; | |
3995 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3996 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
3997 | return len; |
3998 | } | |
3999 | ||
4000 | static struct md_sysfs_entry md_reshape_position = | |
4001 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4002 | reshape_position_store); | |
4003 | ||
b522adcd DW |
4004 | static ssize_t |
4005 | array_size_show(mddev_t *mddev, char *page) | |
4006 | { | |
4007 | if (mddev->external_size) | |
4008 | return sprintf(page, "%llu\n", | |
4009 | (unsigned long long)mddev->array_sectors/2); | |
4010 | else | |
4011 | return sprintf(page, "default\n"); | |
4012 | } | |
4013 | ||
4014 | static ssize_t | |
4015 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
4016 | { | |
4017 | sector_t sectors; | |
4018 | ||
4019 | if (strncmp(buf, "default", 7) == 0) { | |
4020 | if (mddev->pers) | |
4021 | sectors = mddev->pers->size(mddev, 0, 0); | |
4022 | else | |
4023 | sectors = mddev->array_sectors; | |
4024 | ||
4025 | mddev->external_size = 0; | |
4026 | } else { | |
4027 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4028 | return -EINVAL; | |
4029 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4030 | return -E2BIG; |
b522adcd DW |
4031 | |
4032 | mddev->external_size = 1; | |
4033 | } | |
4034 | ||
4035 | mddev->array_sectors = sectors; | |
4036 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e N |
4037 | if (mddev->pers) |
4038 | revalidate_disk(mddev->gendisk); | |
b522adcd DW |
4039 | |
4040 | return len; | |
4041 | } | |
4042 | ||
4043 | static struct md_sysfs_entry md_array_size = | |
4044 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4045 | array_size_store); | |
e464eafd | 4046 | |
eae1701f N |
4047 | static struct attribute *md_default_attrs[] = { |
4048 | &md_level.attr, | |
d4dbd025 | 4049 | &md_layout.attr, |
eae1701f | 4050 | &md_raid_disks.attr, |
3b34380a | 4051 | &md_chunk_size.attr, |
a35b0d69 | 4052 | &md_size.attr, |
a94213b1 | 4053 | &md_resync_start.attr, |
8bb93aac | 4054 | &md_metadata.attr, |
6d7ff738 | 4055 | &md_new_device.attr, |
16f17b39 | 4056 | &md_safe_delay.attr, |
9e653b63 | 4057 | &md_array_state.attr, |
08a02ecd | 4058 | &md_reshape_position.attr, |
b522adcd | 4059 | &md_array_size.attr, |
1e50915f | 4060 | &max_corr_read_errors.attr, |
411036fa N |
4061 | NULL, |
4062 | }; | |
4063 | ||
4064 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4065 | &md_scan_mode.attr, |
9d88883e | 4066 | &md_mismatches.attr, |
88202a0c N |
4067 | &md_sync_min.attr, |
4068 | &md_sync_max.attr, | |
4069 | &md_sync_speed.attr, | |
90b08710 | 4070 | &md_sync_force_parallel.attr, |
88202a0c | 4071 | &md_sync_completed.attr, |
5e96ee65 | 4072 | &md_min_sync.attr, |
c6207277 | 4073 | &md_max_sync.attr, |
e464eafd N |
4074 | &md_suspend_lo.attr, |
4075 | &md_suspend_hi.attr, | |
9b1d1dac | 4076 | &md_bitmap.attr, |
d7f3d291 | 4077 | &md_degraded.attr, |
eae1701f N |
4078 | NULL, |
4079 | }; | |
411036fa N |
4080 | static struct attribute_group md_redundancy_group = { |
4081 | .name = NULL, | |
4082 | .attrs = md_redundancy_attrs, | |
4083 | }; | |
4084 | ||
eae1701f N |
4085 | |
4086 | static ssize_t | |
4087 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4088 | { | |
4089 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4090 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4091 | ssize_t rv; |
eae1701f N |
4092 | |
4093 | if (!entry->show) | |
4094 | return -EIO; | |
5dc5cf7d IM |
4095 | rv = mddev_lock(mddev); |
4096 | if (!rv) { | |
4097 | rv = entry->show(mddev, page); | |
4098 | mddev_unlock(mddev); | |
4099 | } | |
96de1e66 | 4100 | return rv; |
eae1701f N |
4101 | } |
4102 | ||
4103 | static ssize_t | |
4104 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4105 | const char *page, size_t length) | |
4106 | { | |
4107 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4108 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4109 | ssize_t rv; |
eae1701f N |
4110 | |
4111 | if (!entry->store) | |
4112 | return -EIO; | |
67463acb N |
4113 | if (!capable(CAP_SYS_ADMIN)) |
4114 | return -EACCES; | |
5dc5cf7d | 4115 | rv = mddev_lock(mddev); |
d3374825 N |
4116 | if (mddev->hold_active == UNTIL_IOCTL) |
4117 | mddev->hold_active = 0; | |
5dc5cf7d IM |
4118 | if (!rv) { |
4119 | rv = entry->store(mddev, page, length); | |
4120 | mddev_unlock(mddev); | |
4121 | } | |
96de1e66 | 4122 | return rv; |
eae1701f N |
4123 | } |
4124 | ||
4125 | static void md_free(struct kobject *ko) | |
4126 | { | |
4127 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
4128 | |
4129 | if (mddev->sysfs_state) | |
4130 | sysfs_put(mddev->sysfs_state); | |
4131 | ||
4132 | if (mddev->gendisk) { | |
4133 | del_gendisk(mddev->gendisk); | |
4134 | put_disk(mddev->gendisk); | |
4135 | } | |
4136 | if (mddev->queue) | |
4137 | blk_cleanup_queue(mddev->queue); | |
4138 | ||
eae1701f N |
4139 | kfree(mddev); |
4140 | } | |
4141 | ||
4142 | static struct sysfs_ops md_sysfs_ops = { | |
4143 | .show = md_attr_show, | |
4144 | .store = md_attr_store, | |
4145 | }; | |
4146 | static struct kobj_type md_ktype = { | |
4147 | .release = md_free, | |
4148 | .sysfs_ops = &md_sysfs_ops, | |
4149 | .default_attrs = md_default_attrs, | |
4150 | }; | |
4151 | ||
1da177e4 LT |
4152 | int mdp_major = 0; |
4153 | ||
5fd3a17e DW |
4154 | static void mddev_delayed_delete(struct work_struct *ws) |
4155 | { | |
4156 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
4157 | ||
43a70507 | 4158 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4159 | kobject_del(&mddev->kobj); |
4160 | kobject_put(&mddev->kobj); | |
4161 | } | |
4162 | ||
efeb53c0 | 4163 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4164 | { |
48c9c27b | 4165 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
4166 | mddev_t *mddev = mddev_find(dev); |
4167 | struct gendisk *disk; | |
efeb53c0 N |
4168 | int partitioned; |
4169 | int shift; | |
4170 | int unit; | |
3830c62f | 4171 | int error; |
1da177e4 LT |
4172 | |
4173 | if (!mddev) | |
efeb53c0 N |
4174 | return -ENODEV; |
4175 | ||
4176 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4177 | shift = partitioned ? MdpMinorShift : 0; | |
4178 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4179 | |
d3374825 N |
4180 | /* wait for any previous instance if this device |
4181 | * to be completed removed (mddev_delayed_delete). | |
4182 | */ | |
4183 | flush_scheduled_work(); | |
4184 | ||
48c9c27b | 4185 | mutex_lock(&disks_mutex); |
0909dc44 N |
4186 | error = -EEXIST; |
4187 | if (mddev->gendisk) | |
4188 | goto abort; | |
efeb53c0 N |
4189 | |
4190 | if (name) { | |
4191 | /* Need to ensure that 'name' is not a duplicate. | |
4192 | */ | |
4193 | mddev_t *mddev2; | |
4194 | spin_lock(&all_mddevs_lock); | |
4195 | ||
4196 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4197 | if (mddev2->gendisk && | |
4198 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4199 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4200 | goto abort; |
efeb53c0 N |
4201 | } |
4202 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4203 | } |
8b765398 | 4204 | |
0909dc44 | 4205 | error = -ENOMEM; |
8b765398 | 4206 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4207 | if (!mddev->queue) |
4208 | goto abort; | |
409c57f3 N |
4209 | mddev->queue->queuedata = mddev; |
4210 | ||
8b765398 N |
4211 | /* Can be unlocked because the queue is new: no concurrency */ |
4212 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); | |
4213 | ||
409c57f3 | 4214 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 4215 | |
1da177e4 LT |
4216 | disk = alloc_disk(1 << shift); |
4217 | if (!disk) { | |
8b765398 N |
4218 | blk_cleanup_queue(mddev->queue); |
4219 | mddev->queue = NULL; | |
0909dc44 | 4220 | goto abort; |
1da177e4 | 4221 | } |
efeb53c0 | 4222 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4223 | disk->first_minor = unit << shift; |
efeb53c0 N |
4224 | if (name) |
4225 | strcpy(disk->disk_name, name); | |
4226 | else if (partitioned) | |
1da177e4 | 4227 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4228 | else |
1da177e4 | 4229 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4230 | disk->fops = &md_fops; |
4231 | disk->private_data = mddev; | |
4232 | disk->queue = mddev->queue; | |
92850bbd | 4233 | /* Allow extended partitions. This makes the |
d3374825 | 4234 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4235 | * remove it now. |
4236 | */ | |
4237 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 LT |
4238 | add_disk(disk); |
4239 | mddev->gendisk = disk; | |
ed9e1982 TH |
4240 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4241 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4242 | if (error) { |
4243 | /* This isn't possible, but as kobject_init_and_add is marked | |
4244 | * __must_check, we must do something with the result | |
4245 | */ | |
5e55e2f5 N |
4246 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4247 | disk->disk_name); | |
0909dc44 N |
4248 | error = 0; |
4249 | } | |
43a70507 N |
4250 | if (sysfs_create_group(&mddev->kobj, &md_bitmap_group)) |
4251 | printk(KERN_DEBUG "pointless warning\n"); | |
0909dc44 N |
4252 | abort: |
4253 | mutex_unlock(&disks_mutex); | |
4254 | if (!error) { | |
3830c62f | 4255 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
b62b7590 N |
4256 | mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state"); |
4257 | } | |
d3374825 | 4258 | mddev_put(mddev); |
0909dc44 | 4259 | return error; |
efeb53c0 N |
4260 | } |
4261 | ||
4262 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4263 | { | |
4264 | md_alloc(dev, NULL); | |
1da177e4 LT |
4265 | return NULL; |
4266 | } | |
4267 | ||
efeb53c0 N |
4268 | static int add_named_array(const char *val, struct kernel_param *kp) |
4269 | { | |
4270 | /* val must be "md_*" where * is not all digits. | |
4271 | * We allocate an array with a large free minor number, and | |
4272 | * set the name to val. val must not already be an active name. | |
4273 | */ | |
4274 | int len = strlen(val); | |
4275 | char buf[DISK_NAME_LEN]; | |
4276 | ||
4277 | while (len && val[len-1] == '\n') | |
4278 | len--; | |
4279 | if (len >= DISK_NAME_LEN) | |
4280 | return -E2BIG; | |
4281 | strlcpy(buf, val, len+1); | |
4282 | if (strncmp(buf, "md_", 3) != 0) | |
4283 | return -EINVAL; | |
4284 | return md_alloc(0, buf); | |
4285 | } | |
4286 | ||
1da177e4 LT |
4287 | static void md_safemode_timeout(unsigned long data) |
4288 | { | |
4289 | mddev_t *mddev = (mddev_t *) data; | |
4290 | ||
0fd62b86 NB |
4291 | if (!atomic_read(&mddev->writes_pending)) { |
4292 | mddev->safemode = 1; | |
4293 | if (mddev->external) | |
b62b7590 | 4294 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 | 4295 | } |
1da177e4 LT |
4296 | md_wakeup_thread(mddev->thread); |
4297 | } | |
4298 | ||
6ff8d8ec | 4299 | static int start_dirty_degraded; |
1da177e4 | 4300 | |
fe60b014 | 4301 | static int md_run(mddev_t *mddev) |
1da177e4 | 4302 | { |
2604b703 | 4303 | int err; |
1da177e4 | 4304 | mdk_rdev_t *rdev; |
2604b703 | 4305 | struct mdk_personality *pers; |
1da177e4 | 4306 | |
a757e64c N |
4307 | if (list_empty(&mddev->disks)) |
4308 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4309 | return -EINVAL; |
1da177e4 LT |
4310 | |
4311 | if (mddev->pers) | |
4312 | return -EBUSY; | |
4313 | ||
b6eb127d N |
4314 | /* These two calls synchronise us with the |
4315 | * sysfs_remove_group calls in mddev_unlock, | |
4316 | * so they must have completed. | |
4317 | */ | |
4318 | mutex_lock(&mddev->open_mutex); | |
4319 | mutex_unlock(&mddev->open_mutex); | |
4320 | ||
1da177e4 LT |
4321 | /* |
4322 | * Analyze all RAID superblock(s) | |
4323 | */ | |
1ec4a939 N |
4324 | if (!mddev->raid_disks) { |
4325 | if (!mddev->persistent) | |
4326 | return -EINVAL; | |
a757e64c | 4327 | analyze_sbs(mddev); |
1ec4a939 | 4328 | } |
1da177e4 | 4329 | |
d9d166c2 N |
4330 | if (mddev->level != LEVEL_NONE) |
4331 | request_module("md-level-%d", mddev->level); | |
4332 | else if (mddev->clevel[0]) | |
4333 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4334 | |
4335 | /* | |
4336 | * Drop all container device buffers, from now on | |
4337 | * the only valid external interface is through the md | |
4338 | * device. | |
1da177e4 | 4339 | */ |
159ec1fc | 4340 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 4341 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4342 | continue; |
4343 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4344 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4345 | |
4346 | /* perform some consistency tests on the device. | |
4347 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4348 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4349 | */ |
0f420358 | 4350 | if (rdev->data_offset < rdev->sb_start) { |
58c0fed4 AN |
4351 | if (mddev->dev_sectors && |
4352 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4353 | > rdev->sb_start) { |
f0d76d70 N |
4354 | printk("md: %s: data overlaps metadata\n", |
4355 | mdname(mddev)); | |
4356 | return -EINVAL; | |
4357 | } | |
4358 | } else { | |
0f420358 | 4359 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4360 | > rdev->data_offset) { |
4361 | printk("md: %s: metadata overlaps data\n", | |
4362 | mdname(mddev)); | |
4363 | return -EINVAL; | |
4364 | } | |
4365 | } | |
3c0ee63a | 4366 | sysfs_notify_dirent(rdev->sysfs_state); |
1da177e4 LT |
4367 | } |
4368 | ||
1da177e4 | 4369 | spin_lock(&pers_lock); |
d9d166c2 | 4370 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4371 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4372 | spin_unlock(&pers_lock); |
d9d166c2 N |
4373 | if (mddev->level != LEVEL_NONE) |
4374 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4375 | mddev->level); | |
4376 | else | |
4377 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4378 | mddev->clevel); | |
1da177e4 LT |
4379 | return -EINVAL; |
4380 | } | |
2604b703 | 4381 | mddev->pers = pers; |
1da177e4 | 4382 | spin_unlock(&pers_lock); |
34817e8c N |
4383 | if (mddev->level != pers->level) { |
4384 | mddev->level = pers->level; | |
4385 | mddev->new_level = pers->level; | |
4386 | } | |
d9d166c2 | 4387 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4388 | |
f6705578 | 4389 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4390 | pers->start_reshape == NULL) { |
f6705578 N |
4391 | /* This personality cannot handle reshaping... */ |
4392 | mddev->pers = NULL; | |
4393 | module_put(pers->owner); | |
4394 | return -EINVAL; | |
4395 | } | |
4396 | ||
7dd5e7c3 N |
4397 | if (pers->sync_request) { |
4398 | /* Warn if this is a potentially silly | |
4399 | * configuration. | |
4400 | */ | |
4401 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4402 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4403 | int warned = 0; |
159ec1fc CR |
4404 | |
4405 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4406 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4407 | if (rdev < rdev2 && |
4408 | rdev->bdev->bd_contains == | |
4409 | rdev2->bdev->bd_contains) { | |
4410 | printk(KERN_WARNING | |
4411 | "%s: WARNING: %s appears to be" | |
4412 | " on the same physical disk as" | |
4413 | " %s.\n", | |
4414 | mdname(mddev), | |
4415 | bdevname(rdev->bdev,b), | |
4416 | bdevname(rdev2->bdev,b2)); | |
4417 | warned = 1; | |
4418 | } | |
4419 | } | |
159ec1fc | 4420 | |
7dd5e7c3 N |
4421 | if (warned) |
4422 | printk(KERN_WARNING | |
4423 | "True protection against single-disk" | |
4424 | " failure might be compromised.\n"); | |
4425 | } | |
4426 | ||
657390d2 | 4427 | mddev->recovery = 0; |
58c0fed4 AN |
4428 | /* may be over-ridden by personality */ |
4429 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4430 | ||
a9701a30 | 4431 | mddev->barriers_work = 1; |
6ff8d8ec | 4432 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4433 | |
0f9552b5 | 4434 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
4435 | mddev->ro = 2; /* read-only, but switch on first write */ |
4436 | ||
b15c2e57 | 4437 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4438 | if (err) |
4439 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4440 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4441 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4442 | " but 'external_size' not in effect?\n", __func__); | |
4443 | printk(KERN_ERR | |
4444 | "md: invalid array_size %llu > default size %llu\n", | |
4445 | (unsigned long long)mddev->array_sectors / 2, | |
4446 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4447 | err = -EINVAL; | |
4448 | mddev->pers->stop(mddev); | |
4449 | } | |
4450 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4451 | err = bitmap_create(mddev); |
4452 | if (err) { | |
4453 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4454 | mdname(mddev), err); | |
4455 | mddev->pers->stop(mddev); | |
4456 | } | |
4457 | } | |
1da177e4 | 4458 | if (err) { |
1da177e4 LT |
4459 | module_put(mddev->pers->owner); |
4460 | mddev->pers = NULL; | |
32a7627c N |
4461 | bitmap_destroy(mddev); |
4462 | return err; | |
1da177e4 | 4463 | } |
5e55e2f5 N |
4464 | if (mddev->pers->sync_request) { |
4465 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
4466 | printk(KERN_WARNING | |
4467 | "md: cannot register extra attributes for %s\n", | |
4468 | mdname(mddev)); | |
0c3573f1 | 4469 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4470 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4471 | mddev->ro = 0; |
4472 | ||
1da177e4 | 4473 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
4474 | atomic_set(&mddev->max_corr_read_errors, |
4475 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
4476 | mddev->safemode = 0; |
4477 | mddev->safemode_timer.function = md_safemode_timeout; | |
4478 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4479 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4480 | mddev->in_sync = 1; |
86e6ffdd | 4481 | |
159ec1fc | 4482 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4483 | if (rdev->raid_disk >= 0) { |
4484 | char nm[20]; | |
4485 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
4486 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
4487 | printk("md: cannot register %s for %s\n", | |
4488 | nm, mdname(mddev)); | |
86e6ffdd | 4489 | } |
1da177e4 LT |
4490 | |
4491 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4492 | ||
850b2b42 N |
4493 | if (mddev->flags) |
4494 | md_update_sb(mddev, 0); | |
1da177e4 | 4495 | |
0b8c9de0 N |
4496 | md_wakeup_thread(mddev->thread); |
4497 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4498 | |
d7603b7e | 4499 | md_new_event(mddev); |
b62b7590 | 4500 | sysfs_notify_dirent(mddev->sysfs_state); |
0c3573f1 N |
4501 | if (mddev->sysfs_action) |
4502 | sysfs_notify_dirent(mddev->sysfs_action); | |
a99ac971 | 4503 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
4504 | return 0; |
4505 | } | |
4506 | ||
fe60b014 N |
4507 | static int do_md_run(mddev_t *mddev) |
4508 | { | |
4509 | int err; | |
4510 | ||
4511 | err = md_run(mddev); | |
4512 | if (err) | |
4513 | goto out; | |
4514 | ||
4515 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4516 | revalidate_disk(mddev->gendisk); | |
4517 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); | |
4518 | out: | |
4519 | return err; | |
4520 | } | |
4521 | ||
1da177e4 LT |
4522 | static int restart_array(mddev_t *mddev) |
4523 | { | |
4524 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4525 | |
80fab1d7 | 4526 | /* Complain if it has no devices */ |
1da177e4 | 4527 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4528 | return -ENXIO; |
4529 | if (!mddev->pers) | |
4530 | return -EINVAL; | |
4531 | if (!mddev->ro) | |
4532 | return -EBUSY; | |
4533 | mddev->safemode = 0; | |
4534 | mddev->ro = 0; | |
4535 | set_disk_ro(disk, 0); | |
4536 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4537 | mdname(mddev)); | |
4538 | /* Kick recovery or resync if necessary */ | |
4539 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4540 | md_wakeup_thread(mddev->thread); | |
4541 | md_wakeup_thread(mddev->sync_thread); | |
b62b7590 | 4542 | sysfs_notify_dirent(mddev->sysfs_state); |
80fab1d7 | 4543 | return 0; |
1da177e4 LT |
4544 | } |
4545 | ||
acc55e22 N |
4546 | /* similar to deny_write_access, but accounts for our holding a reference |
4547 | * to the file ourselves */ | |
4548 | static int deny_bitmap_write_access(struct file * file) | |
4549 | { | |
4550 | struct inode *inode = file->f_mapping->host; | |
4551 | ||
4552 | spin_lock(&inode->i_lock); | |
4553 | if (atomic_read(&inode->i_writecount) > 1) { | |
4554 | spin_unlock(&inode->i_lock); | |
4555 | return -ETXTBSY; | |
4556 | } | |
4557 | atomic_set(&inode->i_writecount, -1); | |
4558 | spin_unlock(&inode->i_lock); | |
4559 | ||
4560 | return 0; | |
4561 | } | |
4562 | ||
43a70507 | 4563 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
4564 | { |
4565 | struct inode *inode = file->f_mapping->host; | |
4566 | ||
4567 | spin_lock(&inode->i_lock); | |
4568 | atomic_set(&inode->i_writecount, 1); | |
4569 | spin_unlock(&inode->i_lock); | |
4570 | } | |
4571 | ||
6177b472 N |
4572 | static void md_clean(mddev_t *mddev) |
4573 | { | |
4574 | mddev->array_sectors = 0; | |
4575 | mddev->external_size = 0; | |
4576 | mddev->dev_sectors = 0; | |
4577 | mddev->raid_disks = 0; | |
4578 | mddev->recovery_cp = 0; | |
4579 | mddev->resync_min = 0; | |
4580 | mddev->resync_max = MaxSector; | |
4581 | mddev->reshape_position = MaxSector; | |
4582 | mddev->external = 0; | |
4583 | mddev->persistent = 0; | |
4584 | mddev->level = LEVEL_NONE; | |
4585 | mddev->clevel[0] = 0; | |
4586 | mddev->flags = 0; | |
4587 | mddev->ro = 0; | |
4588 | mddev->metadata_type[0] = 0; | |
4589 | mddev->chunk_sectors = 0; | |
4590 | mddev->ctime = mddev->utime = 0; | |
4591 | mddev->layout = 0; | |
4592 | mddev->max_disks = 0; | |
4593 | mddev->events = 0; | |
a8707c08 | 4594 | mddev->can_decrease_events = 0; |
6177b472 N |
4595 | mddev->delta_disks = 0; |
4596 | mddev->new_level = LEVEL_NONE; | |
4597 | mddev->new_layout = 0; | |
4598 | mddev->new_chunk_sectors = 0; | |
4599 | mddev->curr_resync = 0; | |
4600 | mddev->resync_mismatches = 0; | |
4601 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4602 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4603 | mddev->recovery = 0; | |
4604 | mddev->in_sync = 0; | |
4605 | mddev->degraded = 0; | |
4606 | mddev->barriers_work = 0; | |
4607 | mddev->safemode = 0; | |
4608 | mddev->bitmap_info.offset = 0; | |
4609 | mddev->bitmap_info.default_offset = 0; | |
4610 | mddev->bitmap_info.chunksize = 0; | |
4611 | mddev->bitmap_info.daemon_sleep = 0; | |
4612 | mddev->bitmap_info.max_write_behind = 0; | |
4613 | } | |
4614 | ||
a047e125 N |
4615 | static void md_stop_writes(mddev_t *mddev) |
4616 | { | |
4617 | if (mddev->sync_thread) { | |
4618 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4619 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
4620 | md_unregister_thread(mddev->sync_thread); | |
4621 | mddev->sync_thread = NULL; | |
4622 | } | |
4623 | ||
4624 | del_timer_sync(&mddev->safemode_timer); | |
4625 | ||
4626 | bitmap_flush(mddev); | |
4627 | md_super_wait(mddev); | |
4628 | ||
4629 | if (!mddev->in_sync || mddev->flags) { | |
4630 | /* mark array as shutdown cleanly */ | |
4631 | mddev->in_sync = 1; | |
4632 | md_update_sb(mddev, 1); | |
4633 | } | |
4634 | } | |
4635 | ||
6177b472 N |
4636 | static void md_stop(mddev_t *mddev) |
4637 | { | |
cca9cf90 N |
4638 | md_stop_writes(mddev); |
4639 | ||
6177b472 N |
4640 | mddev->pers->stop(mddev); |
4641 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
4642 | mddev->to_remove = &md_redundancy_group; | |
4643 | module_put(mddev->pers->owner); | |
4644 | mddev->pers = NULL; | |
cca9cf90 | 4645 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 N |
4646 | } |
4647 | ||
a4bd82d0 N |
4648 | static int md_set_readonly(mddev_t *mddev, int is_open) |
4649 | { | |
4650 | int err = 0; | |
4651 | mutex_lock(&mddev->open_mutex); | |
4652 | if (atomic_read(&mddev->openers) > is_open) { | |
4653 | printk("md: %s still in use.\n",mdname(mddev)); | |
4654 | err = -EBUSY; | |
4655 | goto out; | |
4656 | } | |
4657 | if (mddev->pers) { | |
4658 | md_stop_writes(mddev); | |
4659 | ||
4660 | err = -ENXIO; | |
4661 | if (mddev->ro==1) | |
4662 | goto out; | |
4663 | mddev->ro = 1; | |
4664 | set_disk_ro(mddev->gendisk, 1); | |
4665 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4666 | sysfs_notify_dirent(mddev->sysfs_state); | |
4667 | err = 0; | |
4668 | } | |
4669 | out: | |
4670 | mutex_unlock(&mddev->open_mutex); | |
4671 | return err; | |
4672 | } | |
4673 | ||
9e653b63 N |
4674 | /* mode: |
4675 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
4676 | * 2 - stop but do not disassemble array |
4677 | */ | |
df5b20cf | 4678 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 LT |
4679 | { |
4680 | int err = 0; | |
4681 | struct gendisk *disk = mddev->gendisk; | |
c4647292 | 4682 | mdk_rdev_t *rdev; |
1da177e4 | 4683 | |
c8c00a69 | 4684 | mutex_lock(&mddev->open_mutex); |
f2ea68cf | 4685 | if (atomic_read(&mddev->openers) > is_open) { |
df5b20cf | 4686 | printk("md: %s still in use.\n",mdname(mddev)); |
c8c00a69 N |
4687 | err = -EBUSY; |
4688 | } else if (mddev->pers) { | |
1da177e4 | 4689 | |
a4bd82d0 N |
4690 | if (mddev->ro) |
4691 | set_disk_ro(disk, 0); | |
409c57f3 | 4692 | |
a4bd82d0 N |
4693 | md_stop(mddev); |
4694 | mddev->queue->merge_bvec_fn = NULL; | |
4695 | mddev->queue->unplug_fn = NULL; | |
4696 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
6177b472 | 4697 | |
a4bd82d0 N |
4698 | /* tell userspace to handle 'inactive' */ |
4699 | sysfs_notify_dirent(mddev->sysfs_state); | |
0d4ca600 | 4700 | |
a4bd82d0 N |
4701 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4702 | if (rdev->raid_disk >= 0) { | |
4703 | char nm[20]; | |
4704 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4705 | sysfs_remove_link(&mddev->kobj, nm); | |
4706 | } | |
c4647292 | 4707 | |
a4bd82d0 N |
4708 | set_capacity(disk, 0); |
4709 | revalidate_disk(disk); | |
0d4ca600 | 4710 | |
a4bd82d0 N |
4711 | if (mddev->ro) |
4712 | mddev->ro = 0; | |
4713 | ||
80ffb3cc | 4714 | err = 0; |
1da177e4 | 4715 | } |
c8c00a69 N |
4716 | mutex_unlock(&mddev->open_mutex); |
4717 | if (err) | |
4718 | return err; | |
1da177e4 LT |
4719 | /* |
4720 | * Free resources if final stop | |
4721 | */ | |
9e653b63 | 4722 | if (mode == 0) { |
0d4ca600 | 4723 | |
1da177e4 LT |
4724 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4725 | ||
978f946b | 4726 | bitmap_destroy(mddev); |
c3d9714e N |
4727 | if (mddev->bitmap_info.file) { |
4728 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
4729 | fput(mddev->bitmap_info.file); | |
4730 | mddev->bitmap_info.file = NULL; | |
978f946b | 4731 | } |
c3d9714e | 4732 | mddev->bitmap_info.offset = 0; |
978f946b | 4733 | |
1da177e4 LT |
4734 | export_array(mddev); |
4735 | ||
6177b472 | 4736 | md_clean(mddev); |
934d9c23 | 4737 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4738 | if (mddev->hold_active == UNTIL_STOP) |
4739 | mddev->hold_active = 0; | |
9e653b63 | 4740 | |
a4bd82d0 | 4741 | } |
1da177e4 | 4742 | err = 0; |
3f9d99c1 | 4743 | blk_integrity_unregister(disk); |
d7603b7e | 4744 | md_new_event(mddev); |
b62b7590 | 4745 | sysfs_notify_dirent(mddev->sysfs_state); |
1da177e4 LT |
4746 | return err; |
4747 | } | |
4748 | ||
fdee8ae4 | 4749 | #ifndef MODULE |
1da177e4 LT |
4750 | static void autorun_array(mddev_t *mddev) |
4751 | { | |
4752 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4753 | int err; |
4754 | ||
a757e64c | 4755 | if (list_empty(&mddev->disks)) |
1da177e4 | 4756 | return; |
1da177e4 LT |
4757 | |
4758 | printk(KERN_INFO "md: running: "); | |
4759 | ||
159ec1fc | 4760 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4761 | char b[BDEVNAME_SIZE]; |
4762 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4763 | } | |
4764 | printk("\n"); | |
4765 | ||
d710e138 | 4766 | err = do_md_run(mddev); |
1da177e4 LT |
4767 | if (err) { |
4768 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4769 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4770 | } |
4771 | } | |
4772 | ||
4773 | /* | |
4774 | * lets try to run arrays based on all disks that have arrived | |
4775 | * until now. (those are in pending_raid_disks) | |
4776 | * | |
4777 | * the method: pick the first pending disk, collect all disks with | |
4778 | * the same UUID, remove all from the pending list and put them into | |
4779 | * the 'same_array' list. Then order this list based on superblock | |
4780 | * update time (freshest comes first), kick out 'old' disks and | |
4781 | * compare superblocks. If everything's fine then run it. | |
4782 | * | |
4783 | * If "unit" is allocated, then bump its reference count | |
4784 | */ | |
4785 | static void autorun_devices(int part) | |
4786 | { | |
159ec1fc | 4787 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4788 | mddev_t *mddev; |
4789 | char b[BDEVNAME_SIZE]; | |
4790 | ||
4791 | printk(KERN_INFO "md: autorun ...\n"); | |
4792 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4793 | int unit; |
1da177e4 | 4794 | dev_t dev; |
ad01c9e3 | 4795 | LIST_HEAD(candidates); |
1da177e4 LT |
4796 | rdev0 = list_entry(pending_raid_disks.next, |
4797 | mdk_rdev_t, same_set); | |
4798 | ||
4799 | printk(KERN_INFO "md: considering %s ...\n", | |
4800 | bdevname(rdev0->bdev,b)); | |
4801 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4802 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4803 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4804 | printk(KERN_INFO "md: adding %s ...\n", | |
4805 | bdevname(rdev->bdev,b)); | |
4806 | list_move(&rdev->same_set, &candidates); | |
4807 | } | |
4808 | /* | |
4809 | * now we have a set of devices, with all of them having | |
4810 | * mostly sane superblocks. It's time to allocate the | |
4811 | * mddev. | |
4812 | */ | |
e8703fe1 N |
4813 | if (part) { |
4814 | dev = MKDEV(mdp_major, | |
4815 | rdev0->preferred_minor << MdpMinorShift); | |
4816 | unit = MINOR(dev) >> MdpMinorShift; | |
4817 | } else { | |
4818 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4819 | unit = MINOR(dev); | |
4820 | } | |
4821 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4822 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4823 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4824 | break; | |
4825 | } | |
1da177e4 LT |
4826 | |
4827 | md_probe(dev, NULL, NULL); | |
4828 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4829 | if (!mddev || !mddev->gendisk) { |
4830 | if (mddev) | |
4831 | mddev_put(mddev); | |
4832 | printk(KERN_ERR | |
1da177e4 LT |
4833 | "md: cannot allocate memory for md drive.\n"); |
4834 | break; | |
4835 | } | |
4836 | if (mddev_lock(mddev)) | |
4837 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4838 | mdname(mddev)); | |
4839 | else if (mddev->raid_disks || mddev->major_version | |
4840 | || !list_empty(&mddev->disks)) { | |
4841 | printk(KERN_WARNING | |
4842 | "md: %s already running, cannot run %s\n", | |
4843 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4844 | mddev_unlock(mddev); | |
4845 | } else { | |
4846 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4847 | mddev->persistent = 1; |
159ec1fc | 4848 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4849 | list_del_init(&rdev->same_set); |
4850 | if (bind_rdev_to_array(rdev, mddev)) | |
4851 | export_rdev(rdev); | |
4852 | } | |
4853 | autorun_array(mddev); | |
4854 | mddev_unlock(mddev); | |
4855 | } | |
4856 | /* on success, candidates will be empty, on error | |
4857 | * it won't... | |
4858 | */ | |
159ec1fc | 4859 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4860 | list_del_init(&rdev->same_set); |
1da177e4 | 4861 | export_rdev(rdev); |
4b80991c | 4862 | } |
1da177e4 LT |
4863 | mddev_put(mddev); |
4864 | } | |
4865 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4866 | } | |
fdee8ae4 | 4867 | #endif /* !MODULE */ |
1da177e4 | 4868 | |
1da177e4 LT |
4869 | static int get_version(void __user * arg) |
4870 | { | |
4871 | mdu_version_t ver; | |
4872 | ||
4873 | ver.major = MD_MAJOR_VERSION; | |
4874 | ver.minor = MD_MINOR_VERSION; | |
4875 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4876 | ||
4877 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4878 | return -EFAULT; | |
4879 | ||
4880 | return 0; | |
4881 | } | |
4882 | ||
4883 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4884 | { | |
4885 | mdu_array_info_t info; | |
a9f326eb | 4886 | int nr,working,insync,failed,spare; |
1da177e4 | 4887 | mdk_rdev_t *rdev; |
1da177e4 | 4888 | |
a9f326eb | 4889 | nr=working=insync=failed=spare=0; |
159ec1fc | 4890 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4891 | nr++; |
b2d444d7 | 4892 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4893 | failed++; |
4894 | else { | |
4895 | working++; | |
b2d444d7 | 4896 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 4897 | insync++; |
1da177e4 LT |
4898 | else |
4899 | spare++; | |
4900 | } | |
4901 | } | |
4902 | ||
4903 | info.major_version = mddev->major_version; | |
4904 | info.minor_version = mddev->minor_version; | |
4905 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
4906 | info.ctime = mddev->ctime; | |
4907 | info.level = mddev->level; | |
58c0fed4 AN |
4908 | info.size = mddev->dev_sectors / 2; |
4909 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 4910 | info.size = -1; |
1da177e4 LT |
4911 | info.nr_disks = nr; |
4912 | info.raid_disks = mddev->raid_disks; | |
4913 | info.md_minor = mddev->md_minor; | |
4914 | info.not_persistent= !mddev->persistent; | |
4915 | ||
4916 | info.utime = mddev->utime; | |
4917 | info.state = 0; | |
4918 | if (mddev->in_sync) | |
4919 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 4920 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 4921 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 4922 | info.active_disks = insync; |
1da177e4 LT |
4923 | info.working_disks = working; |
4924 | info.failed_disks = failed; | |
4925 | info.spare_disks = spare; | |
4926 | ||
4927 | info.layout = mddev->layout; | |
9d8f0363 | 4928 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
4929 | |
4930 | if (copy_to_user(arg, &info, sizeof(info))) | |
4931 | return -EFAULT; | |
4932 | ||
4933 | return 0; | |
4934 | } | |
4935 | ||
87162a28 | 4936 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
4937 | { |
4938 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
4939 | char *ptr, *buf = NULL; | |
4940 | int err = -ENOMEM; | |
4941 | ||
b5470dc5 DW |
4942 | if (md_allow_write(mddev)) |
4943 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
4944 | else | |
4945 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 4946 | |
32a7627c N |
4947 | if (!file) |
4948 | goto out; | |
4949 | ||
4950 | /* bitmap disabled, zero the first byte and copy out */ | |
4951 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
4952 | file->pathname[0] = '\0'; | |
4953 | goto copy_out; | |
4954 | } | |
4955 | ||
4956 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
4957 | if (!buf) | |
4958 | goto out; | |
4959 | ||
6bcfd601 CH |
4960 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
4961 | if (IS_ERR(ptr)) | |
32a7627c N |
4962 | goto out; |
4963 | ||
4964 | strcpy(file->pathname, ptr); | |
4965 | ||
4966 | copy_out: | |
4967 | err = 0; | |
4968 | if (copy_to_user(arg, file, sizeof(*file))) | |
4969 | err = -EFAULT; | |
4970 | out: | |
4971 | kfree(buf); | |
4972 | kfree(file); | |
4973 | return err; | |
4974 | } | |
4975 | ||
1da177e4 LT |
4976 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
4977 | { | |
4978 | mdu_disk_info_t info; | |
1da177e4 LT |
4979 | mdk_rdev_t *rdev; |
4980 | ||
4981 | if (copy_from_user(&info, arg, sizeof(info))) | |
4982 | return -EFAULT; | |
4983 | ||
26ef379f | 4984 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
4985 | if (rdev) { |
4986 | info.major = MAJOR(rdev->bdev->bd_dev); | |
4987 | info.minor = MINOR(rdev->bdev->bd_dev); | |
4988 | info.raid_disk = rdev->raid_disk; | |
4989 | info.state = 0; | |
b2d444d7 | 4990 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4991 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 4992 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4993 | info.state |= (1<<MD_DISK_ACTIVE); |
4994 | info.state |= (1<<MD_DISK_SYNC); | |
4995 | } | |
8ddf9efe N |
4996 | if (test_bit(WriteMostly, &rdev->flags)) |
4997 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
4998 | } else { |
4999 | info.major = info.minor = 0; | |
5000 | info.raid_disk = -1; | |
5001 | info.state = (1<<MD_DISK_REMOVED); | |
5002 | } | |
5003 | ||
5004 | if (copy_to_user(arg, &info, sizeof(info))) | |
5005 | return -EFAULT; | |
5006 | ||
5007 | return 0; | |
5008 | } | |
5009 | ||
5010 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
5011 | { | |
5012 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
5013 | mdk_rdev_t *rdev; | |
5014 | dev_t dev = MKDEV(info->major,info->minor); | |
5015 | ||
5016 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5017 | return -EOVERFLOW; | |
5018 | ||
5019 | if (!mddev->raid_disks) { | |
5020 | int err; | |
5021 | /* expecting a device which has a superblock */ | |
5022 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5023 | if (IS_ERR(rdev)) { | |
5024 | printk(KERN_WARNING | |
5025 | "md: md_import_device returned %ld\n", | |
5026 | PTR_ERR(rdev)); | |
5027 | return PTR_ERR(rdev); | |
5028 | } | |
5029 | if (!list_empty(&mddev->disks)) { | |
5030 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
5031 | mdk_rdev_t, same_set); | |
a9f326eb | 5032 | err = super_types[mddev->major_version] |
1da177e4 LT |
5033 | .load_super(rdev, rdev0, mddev->minor_version); |
5034 | if (err < 0) { | |
5035 | printk(KERN_WARNING | |
5036 | "md: %s has different UUID to %s\n", | |
5037 | bdevname(rdev->bdev,b), | |
5038 | bdevname(rdev0->bdev,b2)); | |
5039 | export_rdev(rdev); | |
5040 | return -EINVAL; | |
5041 | } | |
5042 | } | |
5043 | err = bind_rdev_to_array(rdev, mddev); | |
5044 | if (err) | |
5045 | export_rdev(rdev); | |
5046 | return err; | |
5047 | } | |
5048 | ||
5049 | /* | |
5050 | * add_new_disk can be used once the array is assembled | |
5051 | * to add "hot spares". They must already have a superblock | |
5052 | * written | |
5053 | */ | |
5054 | if (mddev->pers) { | |
5055 | int err; | |
5056 | if (!mddev->pers->hot_add_disk) { | |
5057 | printk(KERN_WARNING | |
5058 | "%s: personality does not support diskops!\n", | |
5059 | mdname(mddev)); | |
5060 | return -EINVAL; | |
5061 | } | |
7b1e35f6 N |
5062 | if (mddev->persistent) |
5063 | rdev = md_import_device(dev, mddev->major_version, | |
5064 | mddev->minor_version); | |
5065 | else | |
5066 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5067 | if (IS_ERR(rdev)) { |
5068 | printk(KERN_WARNING | |
5069 | "md: md_import_device returned %ld\n", | |
5070 | PTR_ERR(rdev)); | |
5071 | return PTR_ERR(rdev); | |
5072 | } | |
41158c7e N |
5073 | /* set save_raid_disk if appropriate */ |
5074 | if (!mddev->persistent) { | |
5075 | if (info->state & (1<<MD_DISK_SYNC) && | |
5076 | info->raid_disk < mddev->raid_disks) | |
5077 | rdev->raid_disk = info->raid_disk; | |
5078 | else | |
5079 | rdev->raid_disk = -1; | |
5080 | } else | |
5081 | super_types[mddev->major_version]. | |
5082 | validate_super(mddev, rdev); | |
5083 | rdev->saved_raid_disk = rdev->raid_disk; | |
5084 | ||
b2d444d7 | 5085 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5086 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5087 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5088 | else |
5089 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5090 | |
1da177e4 LT |
5091 | rdev->raid_disk = -1; |
5092 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5093 | if (!err && !mddev->pers->hot_remove_disk) { |
5094 | /* If there is hot_add_disk but no hot_remove_disk | |
5095 | * then added disks for geometry changes, | |
5096 | * and should be added immediately. | |
5097 | */ | |
5098 | super_types[mddev->major_version]. | |
5099 | validate_super(mddev, rdev); | |
5100 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5101 | if (err) | |
5102 | unbind_rdev_from_array(rdev); | |
5103 | } | |
1da177e4 LT |
5104 | if (err) |
5105 | export_rdev(rdev); | |
52664732 | 5106 | else |
3c0ee63a | 5107 | sysfs_notify_dirent(rdev->sysfs_state); |
c361777f | 5108 | |
17571284 | 5109 | md_update_sb(mddev, 1); |
72a23c21 NB |
5110 | if (mddev->degraded) |
5111 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5112 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 5113 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5114 | return err; |
5115 | } | |
5116 | ||
5117 | /* otherwise, add_new_disk is only allowed | |
5118 | * for major_version==0 superblocks | |
5119 | */ | |
5120 | if (mddev->major_version != 0) { | |
5121 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5122 | mdname(mddev)); | |
5123 | return -EINVAL; | |
5124 | } | |
5125 | ||
5126 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5127 | int err; | |
d710e138 | 5128 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5129 | if (IS_ERR(rdev)) { |
5130 | printk(KERN_WARNING | |
5131 | "md: error, md_import_device() returned %ld\n", | |
5132 | PTR_ERR(rdev)); | |
5133 | return PTR_ERR(rdev); | |
5134 | } | |
5135 | rdev->desc_nr = info->number; | |
5136 | if (info->raid_disk < mddev->raid_disks) | |
5137 | rdev->raid_disk = info->raid_disk; | |
5138 | else | |
5139 | rdev->raid_disk = -1; | |
5140 | ||
1da177e4 | 5141 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5142 | if (info->state & (1<<MD_DISK_SYNC)) |
5143 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5144 | |
8ddf9efe N |
5145 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5146 | set_bit(WriteMostly, &rdev->flags); | |
5147 | ||
1da177e4 LT |
5148 | if (!mddev->persistent) { |
5149 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
0f420358 | 5150 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 5151 | } else |
0f420358 | 5152 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
8190e754 | 5153 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5154 | |
2bf071bf N |
5155 | err = bind_rdev_to_array(rdev, mddev); |
5156 | if (err) { | |
5157 | export_rdev(rdev); | |
5158 | return err; | |
5159 | } | |
1da177e4 LT |
5160 | } |
5161 | ||
5162 | return 0; | |
5163 | } | |
5164 | ||
5165 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
5166 | { | |
5167 | char b[BDEVNAME_SIZE]; | |
5168 | mdk_rdev_t *rdev; | |
5169 | ||
1da177e4 LT |
5170 | rdev = find_rdev(mddev, dev); |
5171 | if (!rdev) | |
5172 | return -ENXIO; | |
5173 | ||
5174 | if (rdev->raid_disk >= 0) | |
5175 | goto busy; | |
5176 | ||
5177 | kick_rdev_from_array(rdev); | |
850b2b42 | 5178 | md_update_sb(mddev, 1); |
d7603b7e | 5179 | md_new_event(mddev); |
1da177e4 LT |
5180 | |
5181 | return 0; | |
5182 | busy: | |
fdefa4d8 | 5183 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5184 | bdevname(rdev->bdev,b), mdname(mddev)); |
5185 | return -EBUSY; | |
5186 | } | |
5187 | ||
5188 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
5189 | { | |
5190 | char b[BDEVNAME_SIZE]; | |
5191 | int err; | |
1da177e4 LT |
5192 | mdk_rdev_t *rdev; |
5193 | ||
5194 | if (!mddev->pers) | |
5195 | return -ENODEV; | |
5196 | ||
5197 | if (mddev->major_version != 0) { | |
5198 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5199 | " version-0 superblocks.\n", | |
5200 | mdname(mddev)); | |
5201 | return -EINVAL; | |
5202 | } | |
5203 | if (!mddev->pers->hot_add_disk) { | |
5204 | printk(KERN_WARNING | |
5205 | "%s: personality does not support diskops!\n", | |
5206 | mdname(mddev)); | |
5207 | return -EINVAL; | |
5208 | } | |
5209 | ||
d710e138 | 5210 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5211 | if (IS_ERR(rdev)) { |
5212 | printk(KERN_WARNING | |
5213 | "md: error, md_import_device() returned %ld\n", | |
5214 | PTR_ERR(rdev)); | |
5215 | return -EINVAL; | |
5216 | } | |
5217 | ||
5218 | if (mddev->persistent) | |
0f420358 | 5219 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 5220 | else |
0f420358 | 5221 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 5222 | |
8190e754 | 5223 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5224 | |
b2d444d7 | 5225 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5226 | printk(KERN_WARNING |
5227 | "md: can not hot-add faulty %s disk to %s!\n", | |
5228 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5229 | err = -EINVAL; | |
5230 | goto abort_export; | |
5231 | } | |
b2d444d7 | 5232 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5233 | rdev->desc_nr = -1; |
5842730d | 5234 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5235 | err = bind_rdev_to_array(rdev, mddev); |
5236 | if (err) | |
5237 | goto abort_export; | |
1da177e4 LT |
5238 | |
5239 | /* | |
5240 | * The rest should better be atomic, we can have disk failures | |
5241 | * noticed in interrupt contexts ... | |
5242 | */ | |
5243 | ||
1da177e4 LT |
5244 | rdev->raid_disk = -1; |
5245 | ||
850b2b42 | 5246 | md_update_sb(mddev, 1); |
1da177e4 LT |
5247 | |
5248 | /* | |
5249 | * Kick recovery, maybe this spare has to be added to the | |
5250 | * array immediately. | |
5251 | */ | |
5252 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5253 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5254 | md_new_event(mddev); |
1da177e4 LT |
5255 | return 0; |
5256 | ||
1da177e4 LT |
5257 | abort_export: |
5258 | export_rdev(rdev); | |
5259 | return err; | |
5260 | } | |
5261 | ||
32a7627c N |
5262 | static int set_bitmap_file(mddev_t *mddev, int fd) |
5263 | { | |
5264 | int err; | |
5265 | ||
36fa3063 N |
5266 | if (mddev->pers) { |
5267 | if (!mddev->pers->quiesce) | |
5268 | return -EBUSY; | |
5269 | if (mddev->recovery || mddev->sync_thread) | |
5270 | return -EBUSY; | |
5271 | /* we should be able to change the bitmap.. */ | |
5272 | } | |
32a7627c | 5273 | |
32a7627c | 5274 | |
36fa3063 N |
5275 | if (fd >= 0) { |
5276 | if (mddev->bitmap) | |
5277 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5278 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5279 | |
c3d9714e | 5280 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5281 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5282 | mdname(mddev)); | |
5283 | return -EBADF; | |
5284 | } | |
5285 | ||
c3d9714e | 5286 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
5287 | if (err) { |
5288 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5289 | mdname(mddev)); | |
c3d9714e N |
5290 | fput(mddev->bitmap_info.file); |
5291 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5292 | return err; |
5293 | } | |
c3d9714e | 5294 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5295 | } else if (mddev->bitmap == NULL) |
5296 | return -ENOENT; /* cannot remove what isn't there */ | |
5297 | err = 0; | |
5298 | if (mddev->pers) { | |
5299 | mddev->pers->quiesce(mddev, 1); | |
5300 | if (fd >= 0) | |
5301 | err = bitmap_create(mddev); | |
d7375ab3 | 5302 | if (fd < 0 || err) { |
36fa3063 | 5303 | bitmap_destroy(mddev); |
d7375ab3 N |
5304 | fd = -1; /* make sure to put the file */ |
5305 | } | |
36fa3063 | 5306 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5307 | } |
5308 | if (fd < 0) { | |
c3d9714e N |
5309 | if (mddev->bitmap_info.file) { |
5310 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5311 | fput(mddev->bitmap_info.file); | |
acc55e22 | 5312 | } |
c3d9714e | 5313 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
5314 | } |
5315 | ||
32a7627c N |
5316 | return err; |
5317 | } | |
5318 | ||
1da177e4 LT |
5319 | /* |
5320 | * set_array_info is used two different ways | |
5321 | * The original usage is when creating a new array. | |
5322 | * In this usage, raid_disks is > 0 and it together with | |
5323 | * level, size, not_persistent,layout,chunksize determine the | |
5324 | * shape of the array. | |
5325 | * This will always create an array with a type-0.90.0 superblock. | |
5326 | * The newer usage is when assembling an array. | |
5327 | * In this case raid_disks will be 0, and the major_version field is | |
5328 | * use to determine which style super-blocks are to be found on the devices. | |
5329 | * The minor and patch _version numbers are also kept incase the | |
5330 | * super_block handler wishes to interpret them. | |
5331 | */ | |
5332 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5333 | { | |
5334 | ||
5335 | if (info->raid_disks == 0) { | |
5336 | /* just setting version number for superblock loading */ | |
5337 | if (info->major_version < 0 || | |
50511da3 | 5338 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5339 | super_types[info->major_version].name == NULL) { |
5340 | /* maybe try to auto-load a module? */ | |
5341 | printk(KERN_INFO | |
5342 | "md: superblock version %d not known\n", | |
5343 | info->major_version); | |
5344 | return -EINVAL; | |
5345 | } | |
5346 | mddev->major_version = info->major_version; | |
5347 | mddev->minor_version = info->minor_version; | |
5348 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5349 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
5350 | /* ensure mddev_put doesn't delete this now that there |
5351 | * is some minimal configuration. | |
5352 | */ | |
5353 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
5354 | return 0; |
5355 | } | |
5356 | mddev->major_version = MD_MAJOR_VERSION; | |
5357 | mddev->minor_version = MD_MINOR_VERSION; | |
5358 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5359 | mddev->ctime = get_seconds(); | |
5360 | ||
5361 | mddev->level = info->level; | |
17115e03 | 5362 | mddev->clevel[0] = 0; |
58c0fed4 | 5363 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5364 | mddev->raid_disks = info->raid_disks; |
5365 | /* don't set md_minor, it is determined by which /dev/md* was | |
5366 | * openned | |
5367 | */ | |
5368 | if (info->state & (1<<MD_SB_CLEAN)) | |
5369 | mddev->recovery_cp = MaxSector; | |
5370 | else | |
5371 | mddev->recovery_cp = 0; | |
5372 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5373 | mddev->external = 0; |
1da177e4 LT |
5374 | |
5375 | mddev->layout = info->layout; | |
9d8f0363 | 5376 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5377 | |
5378 | mddev->max_disks = MD_SB_DISKS; | |
5379 | ||
e691063a N |
5380 | if (mddev->persistent) |
5381 | mddev->flags = 0; | |
850b2b42 | 5382 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5383 | |
c3d9714e N |
5384 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
5385 | mddev->bitmap_info.offset = 0; | |
b2a2703c | 5386 | |
f6705578 N |
5387 | mddev->reshape_position = MaxSector; |
5388 | ||
1da177e4 LT |
5389 | /* |
5390 | * Generate a 128 bit UUID | |
5391 | */ | |
5392 | get_random_bytes(mddev->uuid, 16); | |
5393 | ||
f6705578 | 5394 | mddev->new_level = mddev->level; |
664e7c41 | 5395 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5396 | mddev->new_layout = mddev->layout; |
5397 | mddev->delta_disks = 0; | |
5398 | ||
1da177e4 LT |
5399 | return 0; |
5400 | } | |
5401 | ||
1f403624 DW |
5402 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5403 | { | |
b522adcd DW |
5404 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5405 | ||
5406 | if (mddev->external_size) | |
5407 | return; | |
5408 | ||
1f403624 DW |
5409 | mddev->array_sectors = array_sectors; |
5410 | } | |
5411 | EXPORT_SYMBOL(md_set_array_sectors); | |
5412 | ||
d71f9f88 | 5413 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5414 | { |
159ec1fc | 5415 | mdk_rdev_t *rdev; |
a35b0d69 | 5416 | int rv; |
d71f9f88 | 5417 | int fit = (num_sectors == 0); |
a35b0d69 N |
5418 | |
5419 | if (mddev->pers->resize == NULL) | |
5420 | return -EINVAL; | |
d71f9f88 AN |
5421 | /* The "num_sectors" is the number of sectors of each device that |
5422 | * is used. This can only make sense for arrays with redundancy. | |
5423 | * linear and raid0 always use whatever space is available. We can only | |
5424 | * consider changing this number if no resync or reconstruction is | |
5425 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5426 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5427 | * of each device. If num_sectors is zero, we find the largest size |
5428 | * that fits. | |
5429 | ||
a35b0d69 N |
5430 | */ |
5431 | if (mddev->sync_thread) | |
5432 | return -EBUSY; | |
dba034ee N |
5433 | if (mddev->bitmap) |
5434 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5435 | * grow, and re-add. | |
5436 | */ | |
5437 | return -EBUSY; | |
159ec1fc | 5438 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5439 | sector_t avail = rdev->sectors; |
01ab5662 | 5440 | |
d71f9f88 AN |
5441 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5442 | num_sectors = avail; | |
5443 | if (avail < num_sectors) | |
a35b0d69 N |
5444 | return -ENOSPC; |
5445 | } | |
d71f9f88 | 5446 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5447 | if (!rv) |
5448 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5449 | return rv; |
5450 | } | |
5451 | ||
da943b99 N |
5452 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5453 | { | |
5454 | int rv; | |
5455 | /* change the number of raid disks */ | |
63c70c4f | 5456 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5457 | return -EINVAL; |
5458 | if (raid_disks <= 0 || | |
233fca36 | 5459 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 5460 | return -EINVAL; |
63c70c4f | 5461 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5462 | return -EBUSY; |
63c70c4f N |
5463 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5464 | ||
5465 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
5466 | return rv; |
5467 | } | |
5468 | ||
5469 | ||
1da177e4 LT |
5470 | /* |
5471 | * update_array_info is used to change the configuration of an | |
5472 | * on-line array. | |
5473 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5474 | * fields in the info are checked against the array. | |
5475 | * Any differences that cannot be handled will cause an error. | |
5476 | * Normally, only one change can be managed at a time. | |
5477 | */ | |
5478 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5479 | { | |
5480 | int rv = 0; | |
5481 | int cnt = 0; | |
36fa3063 N |
5482 | int state = 0; |
5483 | ||
5484 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 5485 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5486 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
5487 | |
5488 | if (mddev->major_version != info->major_version || | |
5489 | mddev->minor_version != info->minor_version || | |
5490 | /* mddev->patch_version != info->patch_version || */ | |
5491 | mddev->ctime != info->ctime || | |
5492 | mddev->level != info->level || | |
5493 | /* mddev->layout != info->layout || */ | |
5494 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5495 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5496 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5497 | ((state^info->state) & 0xfffffe00) | |
5498 | ) | |
1da177e4 LT |
5499 | return -EINVAL; |
5500 | /* Check there is only one change */ | |
58c0fed4 AN |
5501 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5502 | cnt++; | |
5503 | if (mddev->raid_disks != info->raid_disks) | |
5504 | cnt++; | |
5505 | if (mddev->layout != info->layout) | |
5506 | cnt++; | |
5507 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5508 | cnt++; | |
5509 | if (cnt == 0) | |
5510 | return 0; | |
5511 | if (cnt > 1) | |
5512 | return -EINVAL; | |
1da177e4 LT |
5513 | |
5514 | if (mddev->layout != info->layout) { | |
5515 | /* Change layout | |
5516 | * we don't need to do anything at the md level, the | |
5517 | * personality will take care of it all. | |
5518 | */ | |
50ac168a | 5519 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5520 | return -EINVAL; |
597a711b N |
5521 | else { |
5522 | mddev->new_layout = info->layout; | |
50ac168a | 5523 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5524 | if (rv) |
5525 | mddev->new_layout = mddev->layout; | |
5526 | return rv; | |
5527 | } | |
1da177e4 | 5528 | } |
58c0fed4 | 5529 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5530 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5531 | |
da943b99 N |
5532 | if (mddev->raid_disks != info->raid_disks) |
5533 | rv = update_raid_disks(mddev, info->raid_disks); | |
5534 | ||
36fa3063 N |
5535 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5536 | if (mddev->pers->quiesce == NULL) | |
5537 | return -EINVAL; | |
5538 | if (mddev->recovery || mddev->sync_thread) | |
5539 | return -EBUSY; | |
5540 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5541 | /* add the bitmap */ | |
5542 | if (mddev->bitmap) | |
5543 | return -EEXIST; | |
c3d9714e | 5544 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 5545 | return -EINVAL; |
c3d9714e N |
5546 | mddev->bitmap_info.offset = |
5547 | mddev->bitmap_info.default_offset; | |
36fa3063 N |
5548 | mddev->pers->quiesce(mddev, 1); |
5549 | rv = bitmap_create(mddev); | |
5550 | if (rv) | |
5551 | bitmap_destroy(mddev); | |
5552 | mddev->pers->quiesce(mddev, 0); | |
5553 | } else { | |
5554 | /* remove the bitmap */ | |
5555 | if (!mddev->bitmap) | |
5556 | return -ENOENT; | |
5557 | if (mddev->bitmap->file) | |
5558 | return -EINVAL; | |
5559 | mddev->pers->quiesce(mddev, 1); | |
5560 | bitmap_destroy(mddev); | |
5561 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 5562 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
5563 | } |
5564 | } | |
850b2b42 | 5565 | md_update_sb(mddev, 1); |
1da177e4 LT |
5566 | return rv; |
5567 | } | |
5568 | ||
5569 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5570 | { | |
5571 | mdk_rdev_t *rdev; | |
5572 | ||
5573 | if (mddev->pers == NULL) | |
5574 | return -ENODEV; | |
5575 | ||
5576 | rdev = find_rdev(mddev, dev); | |
5577 | if (!rdev) | |
5578 | return -ENODEV; | |
5579 | ||
5580 | md_error(mddev, rdev); | |
5581 | return 0; | |
5582 | } | |
5583 | ||
2f9618ce AN |
5584 | /* |
5585 | * We have a problem here : there is no easy way to give a CHS | |
5586 | * virtual geometry. We currently pretend that we have a 2 heads | |
5587 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5588 | * dosfs just mad... ;-) | |
5589 | */ | |
a885c8c4 CH |
5590 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5591 | { | |
5592 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5593 | ||
5594 | geo->heads = 2; | |
5595 | geo->sectors = 4; | |
49ce6cea | 5596 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
5597 | return 0; |
5598 | } | |
5599 | ||
a39907fa | 5600 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5601 | unsigned int cmd, unsigned long arg) |
5602 | { | |
5603 | int err = 0; | |
5604 | void __user *argp = (void __user *)arg; | |
1da177e4 | 5605 | mddev_t *mddev = NULL; |
e2218350 | 5606 | int ro; |
1da177e4 LT |
5607 | |
5608 | if (!capable(CAP_SYS_ADMIN)) | |
5609 | return -EACCES; | |
5610 | ||
5611 | /* | |
5612 | * Commands dealing with the RAID driver but not any | |
5613 | * particular array: | |
5614 | */ | |
5615 | switch (cmd) | |
5616 | { | |
5617 | case RAID_VERSION: | |
5618 | err = get_version(argp); | |
5619 | goto done; | |
5620 | ||
5621 | case PRINT_RAID_DEBUG: | |
5622 | err = 0; | |
5623 | md_print_devices(); | |
5624 | goto done; | |
5625 | ||
5626 | #ifndef MODULE | |
5627 | case RAID_AUTORUN: | |
5628 | err = 0; | |
5629 | autostart_arrays(arg); | |
5630 | goto done; | |
5631 | #endif | |
5632 | default:; | |
5633 | } | |
5634 | ||
5635 | /* | |
5636 | * Commands creating/starting a new array: | |
5637 | */ | |
5638 | ||
a39907fa | 5639 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5640 | |
5641 | if (!mddev) { | |
5642 | BUG(); | |
5643 | goto abort; | |
5644 | } | |
5645 | ||
1da177e4 LT |
5646 | err = mddev_lock(mddev); |
5647 | if (err) { | |
5648 | printk(KERN_INFO | |
5649 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5650 | err, cmd); | |
5651 | goto abort; | |
5652 | } | |
5653 | ||
5654 | switch (cmd) | |
5655 | { | |
5656 | case SET_ARRAY_INFO: | |
5657 | { | |
5658 | mdu_array_info_t info; | |
5659 | if (!arg) | |
5660 | memset(&info, 0, sizeof(info)); | |
5661 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5662 | err = -EFAULT; | |
5663 | goto abort_unlock; | |
5664 | } | |
5665 | if (mddev->pers) { | |
5666 | err = update_array_info(mddev, &info); | |
5667 | if (err) { | |
5668 | printk(KERN_WARNING "md: couldn't update" | |
5669 | " array info. %d\n", err); | |
5670 | goto abort_unlock; | |
5671 | } | |
5672 | goto done_unlock; | |
5673 | } | |
5674 | if (!list_empty(&mddev->disks)) { | |
5675 | printk(KERN_WARNING | |
5676 | "md: array %s already has disks!\n", | |
5677 | mdname(mddev)); | |
5678 | err = -EBUSY; | |
5679 | goto abort_unlock; | |
5680 | } | |
5681 | if (mddev->raid_disks) { | |
5682 | printk(KERN_WARNING | |
5683 | "md: array %s already initialised!\n", | |
5684 | mdname(mddev)); | |
5685 | err = -EBUSY; | |
5686 | goto abort_unlock; | |
5687 | } | |
5688 | err = set_array_info(mddev, &info); | |
5689 | if (err) { | |
5690 | printk(KERN_WARNING "md: couldn't set" | |
5691 | " array info. %d\n", err); | |
5692 | goto abort_unlock; | |
5693 | } | |
5694 | } | |
5695 | goto done_unlock; | |
5696 | ||
5697 | default:; | |
5698 | } | |
5699 | ||
5700 | /* | |
5701 | * Commands querying/configuring an existing array: | |
5702 | */ | |
32a7627c | 5703 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5704 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5705 | if ((!mddev->raid_disks && !mddev->external) |
5706 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5707 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5708 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5709 | err = -ENODEV; |
5710 | goto abort_unlock; | |
5711 | } | |
5712 | ||
5713 | /* | |
5714 | * Commands even a read-only array can execute: | |
5715 | */ | |
5716 | switch (cmd) | |
5717 | { | |
5718 | case GET_ARRAY_INFO: | |
5719 | err = get_array_info(mddev, argp); | |
5720 | goto done_unlock; | |
5721 | ||
32a7627c | 5722 | case GET_BITMAP_FILE: |
87162a28 | 5723 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5724 | goto done_unlock; |
5725 | ||
1da177e4 LT |
5726 | case GET_DISK_INFO: |
5727 | err = get_disk_info(mddev, argp); | |
5728 | goto done_unlock; | |
5729 | ||
5730 | case RESTART_ARRAY_RW: | |
5731 | err = restart_array(mddev); | |
5732 | goto done_unlock; | |
5733 | ||
5734 | case STOP_ARRAY: | |
d710e138 | 5735 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5736 | goto done_unlock; |
5737 | ||
5738 | case STOP_ARRAY_RO: | |
a4bd82d0 | 5739 | err = md_set_readonly(mddev, 1); |
1da177e4 LT |
5740 | goto done_unlock; |
5741 | ||
e2218350 DW |
5742 | case BLKROSET: |
5743 | if (get_user(ro, (int __user *)(arg))) { | |
5744 | err = -EFAULT; | |
5745 | goto done_unlock; | |
5746 | } | |
5747 | err = -EINVAL; | |
5748 | ||
5749 | /* if the bdev is going readonly the value of mddev->ro | |
5750 | * does not matter, no writes are coming | |
5751 | */ | |
5752 | if (ro) | |
5753 | goto done_unlock; | |
5754 | ||
5755 | /* are we are already prepared for writes? */ | |
5756 | if (mddev->ro != 1) | |
5757 | goto done_unlock; | |
5758 | ||
5759 | /* transitioning to readauto need only happen for | |
5760 | * arrays that call md_write_start | |
5761 | */ | |
5762 | if (mddev->pers) { | |
5763 | err = restart_array(mddev); | |
5764 | if (err == 0) { | |
5765 | mddev->ro = 2; | |
5766 | set_disk_ro(mddev->gendisk, 0); | |
5767 | } | |
5768 | } | |
5769 | goto done_unlock; | |
1da177e4 LT |
5770 | } |
5771 | ||
5772 | /* | |
5773 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5774 | * superblock, so we do not allow them on read-only arrays. |
5775 | * However non-MD ioctls (e.g. get-size) will still come through | |
5776 | * here and hit the 'default' below, so only disallow | |
5777 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5778 | */ |
bb57fc64 | 5779 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5780 | if (mddev->ro == 2) { |
5781 | mddev->ro = 0; | |
b62b7590 | 5782 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 NB |
5783 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5784 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5785 | } else { |
5786 | err = -EROFS; | |
5787 | goto abort_unlock; | |
5788 | } | |
1da177e4 LT |
5789 | } |
5790 | ||
5791 | switch (cmd) | |
5792 | { | |
5793 | case ADD_NEW_DISK: | |
5794 | { | |
5795 | mdu_disk_info_t info; | |
5796 | if (copy_from_user(&info, argp, sizeof(info))) | |
5797 | err = -EFAULT; | |
5798 | else | |
5799 | err = add_new_disk(mddev, &info); | |
5800 | goto done_unlock; | |
5801 | } | |
5802 | ||
5803 | case HOT_REMOVE_DISK: | |
5804 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5805 | goto done_unlock; | |
5806 | ||
5807 | case HOT_ADD_DISK: | |
5808 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5809 | goto done_unlock; | |
5810 | ||
5811 | case SET_DISK_FAULTY: | |
5812 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5813 | goto done_unlock; | |
5814 | ||
5815 | case RUN_ARRAY: | |
d710e138 | 5816 | err = do_md_run(mddev); |
1da177e4 LT |
5817 | goto done_unlock; |
5818 | ||
32a7627c N |
5819 | case SET_BITMAP_FILE: |
5820 | err = set_bitmap_file(mddev, (int)arg); | |
5821 | goto done_unlock; | |
5822 | ||
1da177e4 | 5823 | default: |
1da177e4 LT |
5824 | err = -EINVAL; |
5825 | goto abort_unlock; | |
5826 | } | |
5827 | ||
5828 | done_unlock: | |
5829 | abort_unlock: | |
d3374825 N |
5830 | if (mddev->hold_active == UNTIL_IOCTL && |
5831 | err != -EINVAL) | |
5832 | mddev->hold_active = 0; | |
1da177e4 LT |
5833 | mddev_unlock(mddev); |
5834 | ||
5835 | return err; | |
5836 | done: | |
5837 | if (err) | |
5838 | MD_BUG(); | |
5839 | abort: | |
5840 | return err; | |
5841 | } | |
aa98aa31 AB |
5842 | #ifdef CONFIG_COMPAT |
5843 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
5844 | unsigned int cmd, unsigned long arg) | |
5845 | { | |
5846 | switch (cmd) { | |
5847 | case HOT_REMOVE_DISK: | |
5848 | case HOT_ADD_DISK: | |
5849 | case SET_DISK_FAULTY: | |
5850 | case SET_BITMAP_FILE: | |
5851 | /* These take in integer arg, do not convert */ | |
5852 | break; | |
5853 | default: | |
5854 | arg = (unsigned long)compat_ptr(arg); | |
5855 | break; | |
5856 | } | |
5857 | ||
5858 | return md_ioctl(bdev, mode, cmd, arg); | |
5859 | } | |
5860 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 5861 | |
a39907fa | 5862 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
5863 | { |
5864 | /* | |
5865 | * Succeed if we can lock the mddev, which confirms that | |
5866 | * it isn't being stopped right now. | |
5867 | */ | |
d3374825 | 5868 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
5869 | int err; |
5870 | ||
d3374825 N |
5871 | if (mddev->gendisk != bdev->bd_disk) { |
5872 | /* we are racing with mddev_put which is discarding this | |
5873 | * bd_disk. | |
5874 | */ | |
5875 | mddev_put(mddev); | |
5876 | /* Wait until bdev->bd_disk is definitely gone */ | |
5877 | flush_scheduled_work(); | |
5878 | /* Then retry the open from the top */ | |
5879 | return -ERESTARTSYS; | |
5880 | } | |
5881 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
5882 | ||
c8c00a69 | 5883 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
5884 | goto out; |
5885 | ||
5886 | err = 0; | |
f2ea68cf | 5887 | atomic_inc(&mddev->openers); |
c8c00a69 | 5888 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 5889 | |
1da177e4 LT |
5890 | out: |
5891 | return err; | |
5892 | } | |
5893 | ||
a39907fa | 5894 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 5895 | { |
a39907fa | 5896 | mddev_t *mddev = disk->private_data; |
1da177e4 | 5897 | |
52e5f9d1 | 5898 | BUG_ON(!mddev); |
f2ea68cf | 5899 | atomic_dec(&mddev->openers); |
1da177e4 LT |
5900 | mddev_put(mddev); |
5901 | ||
5902 | return 0; | |
5903 | } | |
83d5cde4 | 5904 | static const struct block_device_operations md_fops = |
1da177e4 LT |
5905 | { |
5906 | .owner = THIS_MODULE, | |
a39907fa AV |
5907 | .open = md_open, |
5908 | .release = md_release, | |
b492b852 | 5909 | .ioctl = md_ioctl, |
aa98aa31 AB |
5910 | #ifdef CONFIG_COMPAT |
5911 | .compat_ioctl = md_compat_ioctl, | |
5912 | #endif | |
a885c8c4 | 5913 | .getgeo = md_getgeo, |
1da177e4 LT |
5914 | }; |
5915 | ||
75c96f85 | 5916 | static int md_thread(void * arg) |
1da177e4 LT |
5917 | { |
5918 | mdk_thread_t *thread = arg; | |
5919 | ||
1da177e4 LT |
5920 | /* |
5921 | * md_thread is a 'system-thread', it's priority should be very | |
5922 | * high. We avoid resource deadlocks individually in each | |
5923 | * raid personality. (RAID5 does preallocation) We also use RR and | |
5924 | * the very same RT priority as kswapd, thus we will never get | |
5925 | * into a priority inversion deadlock. | |
5926 | * | |
5927 | * we definitely have to have equal or higher priority than | |
5928 | * bdflush, otherwise bdflush will deadlock if there are too | |
5929 | * many dirty RAID5 blocks. | |
5930 | */ | |
1da177e4 | 5931 | |
6985c43f | 5932 | allow_signal(SIGKILL); |
a6fb0934 | 5933 | while (!kthread_should_stop()) { |
1da177e4 | 5934 | |
93588e22 N |
5935 | /* We need to wait INTERRUPTIBLE so that |
5936 | * we don't add to the load-average. | |
5937 | * That means we need to be sure no signals are | |
5938 | * pending | |
5939 | */ | |
5940 | if (signal_pending(current)) | |
5941 | flush_signals(current); | |
5942 | ||
5943 | wait_event_interruptible_timeout | |
5944 | (thread->wqueue, | |
5945 | test_bit(THREAD_WAKEUP, &thread->flags) | |
5946 | || kthread_should_stop(), | |
5947 | thread->timeout); | |
1da177e4 LT |
5948 | |
5949 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
5950 | ||
787453c2 | 5951 | thread->run(thread->mddev); |
1da177e4 | 5952 | } |
a6fb0934 | 5953 | |
1da177e4 LT |
5954 | return 0; |
5955 | } | |
5956 | ||
5957 | void md_wakeup_thread(mdk_thread_t *thread) | |
5958 | { | |
5959 | if (thread) { | |
5960 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
5961 | set_bit(THREAD_WAKEUP, &thread->flags); | |
5962 | wake_up(&thread->wqueue); | |
5963 | } | |
5964 | } | |
5965 | ||
5966 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
5967 | const char *name) | |
5968 | { | |
5969 | mdk_thread_t *thread; | |
1da177e4 | 5970 | |
9ffae0cf | 5971 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
5972 | if (!thread) |
5973 | return NULL; | |
5974 | ||
1da177e4 LT |
5975 | init_waitqueue_head(&thread->wqueue); |
5976 | ||
1da177e4 LT |
5977 | thread->run = run; |
5978 | thread->mddev = mddev; | |
32a7627c | 5979 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
5980 | thread->tsk = kthread_run(md_thread, thread, |
5981 | "%s_%s", | |
5982 | mdname(thread->mddev), | |
5983 | name ?: mddev->pers->name); | |
a6fb0934 | 5984 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
5985 | kfree(thread); |
5986 | return NULL; | |
5987 | } | |
1da177e4 LT |
5988 | return thread; |
5989 | } | |
5990 | ||
1da177e4 LT |
5991 | void md_unregister_thread(mdk_thread_t *thread) |
5992 | { | |
e0cf8f04 N |
5993 | if (!thread) |
5994 | return; | |
ba25f9dc | 5995 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
5996 | |
5997 | kthread_stop(thread->tsk); | |
1da177e4 LT |
5998 | kfree(thread); |
5999 | } | |
6000 | ||
6001 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
6002 | { | |
6003 | if (!mddev) { | |
6004 | MD_BUG(); | |
6005 | return; | |
6006 | } | |
6007 | ||
b2d444d7 | 6008 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6009 | return; |
6bfe0b49 DW |
6010 | |
6011 | if (mddev->external) | |
6012 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 6013 | /* |
1da177e4 LT |
6014 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
6015 | mdname(mddev), | |
6016 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
6017 | __builtin_return_address(0),__builtin_return_address(1), | |
6018 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 6019 | */ |
d0a0a5ee AM |
6020 | if (!mddev->pers) |
6021 | return; | |
1da177e4 LT |
6022 | if (!mddev->pers->error_handler) |
6023 | return; | |
6024 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6025 | if (mddev->degraded) |
6026 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c0cc75f8 | 6027 | sysfs_notify_dirent(rdev->sysfs_state); |
1da177e4 LT |
6028 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6029 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6030 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 6031 | md_new_event_inintr(mddev); |
1da177e4 LT |
6032 | } |
6033 | ||
6034 | /* seq_file implementation /proc/mdstat */ | |
6035 | ||
6036 | static void status_unused(struct seq_file *seq) | |
6037 | { | |
6038 | int i = 0; | |
6039 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6040 | |
6041 | seq_printf(seq, "unused devices: "); | |
6042 | ||
159ec1fc | 6043 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6044 | char b[BDEVNAME_SIZE]; |
6045 | i++; | |
6046 | seq_printf(seq, "%s ", | |
6047 | bdevname(rdev->bdev,b)); | |
6048 | } | |
6049 | if (!i) | |
6050 | seq_printf(seq, "<none>"); | |
6051 | ||
6052 | seq_printf(seq, "\n"); | |
6053 | } | |
6054 | ||
6055 | ||
6056 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
6057 | { | |
dd71cf6b N |
6058 | sector_t max_sectors, resync, res; |
6059 | unsigned long dt, db; | |
6060 | sector_t rt; | |
4588b42e N |
6061 | int scale; |
6062 | unsigned int per_milli; | |
1da177e4 | 6063 | |
dd71cf6b | 6064 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6065 | |
6066 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 6067 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6068 | else |
dd71cf6b | 6069 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6070 | |
6071 | /* | |
6072 | * Should not happen. | |
6073 | */ | |
dd71cf6b | 6074 | if (!max_sectors) { |
1da177e4 LT |
6075 | MD_BUG(); |
6076 | return; | |
6077 | } | |
4588b42e | 6078 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6079 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6080 | * u32, as those are the requirements for sector_div. |
6081 | * Thus 'scale' must be at least 10 | |
6082 | */ | |
6083 | scale = 10; | |
6084 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6085 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6086 | scale++; |
6087 | } | |
6088 | res = (resync>>scale)*1000; | |
dd71cf6b | 6089 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6090 | |
6091 | per_milli = res; | |
1da177e4 | 6092 | { |
4588b42e | 6093 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6094 | seq_printf(seq, "["); |
6095 | for (i = 0; i < x; i++) | |
6096 | seq_printf(seq, "="); | |
6097 | seq_printf(seq, ">"); | |
6098 | for (i = 0; i < y; i++) | |
6099 | seq_printf(seq, "."); | |
6100 | seq_printf(seq, "] "); | |
6101 | } | |
4588b42e | 6102 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6103 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6104 | "reshape" : | |
61df9d91 N |
6105 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6106 | "check" : | |
6107 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6108 | "resync" : "recovery"))), | |
6109 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6110 | (unsigned long long) resync/2, |
6111 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6112 | |
6113 | /* | |
1da177e4 LT |
6114 | * dt: time from mark until now |
6115 | * db: blocks written from mark until now | |
6116 | * rt: remaining time | |
dd71cf6b N |
6117 | * |
6118 | * rt is a sector_t, so could be 32bit or 64bit. | |
6119 | * So we divide before multiply in case it is 32bit and close | |
6120 | * to the limit. | |
6121 | * We scale the divisor (db) by 32 to avoid loosing precision | |
6122 | * near the end of resync when the number of remaining sectors | |
6123 | * is close to 'db'. | |
6124 | * We then divide rt by 32 after multiplying by db to compensate. | |
6125 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6126 | */ |
6127 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6128 | if (!dt) dt++; | |
ff4e8d9a N |
6129 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6130 | - mddev->resync_mark_cnt; | |
1da177e4 | 6131 | |
dd71cf6b N |
6132 | rt = max_sectors - resync; /* number of remaining sectors */ |
6133 | sector_div(rt, db/32+1); | |
6134 | rt *= dt; | |
6135 | rt >>= 5; | |
6136 | ||
6137 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6138 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6139 | |
ff4e8d9a | 6140 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6141 | } |
6142 | ||
6143 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6144 | { | |
6145 | struct list_head *tmp; | |
6146 | loff_t l = *pos; | |
6147 | mddev_t *mddev; | |
6148 | ||
6149 | if (l >= 0x10000) | |
6150 | return NULL; | |
6151 | if (!l--) | |
6152 | /* header */ | |
6153 | return (void*)1; | |
6154 | ||
6155 | spin_lock(&all_mddevs_lock); | |
6156 | list_for_each(tmp,&all_mddevs) | |
6157 | if (!l--) { | |
6158 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
6159 | mddev_get(mddev); | |
6160 | spin_unlock(&all_mddevs_lock); | |
6161 | return mddev; | |
6162 | } | |
6163 | spin_unlock(&all_mddevs_lock); | |
6164 | if (!l--) | |
6165 | return (void*)2;/* tail */ | |
6166 | return NULL; | |
6167 | } | |
6168 | ||
6169 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6170 | { | |
6171 | struct list_head *tmp; | |
6172 | mddev_t *next_mddev, *mddev = v; | |
6173 | ||
6174 | ++*pos; | |
6175 | if (v == (void*)2) | |
6176 | return NULL; | |
6177 | ||
6178 | spin_lock(&all_mddevs_lock); | |
6179 | if (v == (void*)1) | |
6180 | tmp = all_mddevs.next; | |
6181 | else | |
6182 | tmp = mddev->all_mddevs.next; | |
6183 | if (tmp != &all_mddevs) | |
6184 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
6185 | else { | |
6186 | next_mddev = (void*)2; | |
6187 | *pos = 0x10000; | |
6188 | } | |
6189 | spin_unlock(&all_mddevs_lock); | |
6190 | ||
6191 | if (v != (void*)1) | |
6192 | mddev_put(mddev); | |
6193 | return next_mddev; | |
6194 | ||
6195 | } | |
6196 | ||
6197 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6198 | { | |
6199 | mddev_t *mddev = v; | |
6200 | ||
6201 | if (mddev && v != (void*)1 && v != (void*)2) | |
6202 | mddev_put(mddev); | |
6203 | } | |
6204 | ||
d7603b7e N |
6205 | struct mdstat_info { |
6206 | int event; | |
6207 | }; | |
6208 | ||
1da177e4 LT |
6209 | static int md_seq_show(struct seq_file *seq, void *v) |
6210 | { | |
6211 | mddev_t *mddev = v; | |
dd8ac336 | 6212 | sector_t sectors; |
1da177e4 | 6213 | mdk_rdev_t *rdev; |
d7603b7e | 6214 | struct mdstat_info *mi = seq->private; |
32a7627c | 6215 | struct bitmap *bitmap; |
1da177e4 LT |
6216 | |
6217 | if (v == (void*)1) { | |
2604b703 | 6218 | struct mdk_personality *pers; |
1da177e4 LT |
6219 | seq_printf(seq, "Personalities : "); |
6220 | spin_lock(&pers_lock); | |
2604b703 N |
6221 | list_for_each_entry(pers, &pers_list, list) |
6222 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6223 | |
6224 | spin_unlock(&pers_lock); | |
6225 | seq_printf(seq, "\n"); | |
d7603b7e | 6226 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
6227 | return 0; |
6228 | } | |
6229 | if (v == (void*)2) { | |
6230 | status_unused(seq); | |
6231 | return 0; | |
6232 | } | |
6233 | ||
5dc5cf7d | 6234 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6235 | return -EINTR; |
5dc5cf7d | 6236 | |
1da177e4 LT |
6237 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
6238 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
6239 | mddev->pers ? "" : "in"); | |
6240 | if (mddev->pers) { | |
f91de92e | 6241 | if (mddev->ro==1) |
1da177e4 | 6242 | seq_printf(seq, " (read-only)"); |
f91de92e | 6243 | if (mddev->ro==2) |
52720ae7 | 6244 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
6245 | seq_printf(seq, " %s", mddev->pers->name); |
6246 | } | |
6247 | ||
dd8ac336 | 6248 | sectors = 0; |
159ec1fc | 6249 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
6250 | char b[BDEVNAME_SIZE]; |
6251 | seq_printf(seq, " %s[%d]", | |
6252 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
6253 | if (test_bit(WriteMostly, &rdev->flags)) |
6254 | seq_printf(seq, "(W)"); | |
b2d444d7 | 6255 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6256 | seq_printf(seq, "(F)"); |
6257 | continue; | |
b325a32e N |
6258 | } else if (rdev->raid_disk < 0) |
6259 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 6260 | sectors += rdev->sectors; |
1da177e4 LT |
6261 | } |
6262 | ||
6263 | if (!list_empty(&mddev->disks)) { | |
6264 | if (mddev->pers) | |
6265 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
6266 | (unsigned long long) |
6267 | mddev->array_sectors / 2); | |
1da177e4 LT |
6268 | else |
6269 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 6270 | (unsigned long long)sectors / 2); |
1da177e4 | 6271 | } |
1cd6bf19 N |
6272 | if (mddev->persistent) { |
6273 | if (mddev->major_version != 0 || | |
6274 | mddev->minor_version != 90) { | |
6275 | seq_printf(seq," super %d.%d", | |
6276 | mddev->major_version, | |
6277 | mddev->minor_version); | |
6278 | } | |
e691063a N |
6279 | } else if (mddev->external) |
6280 | seq_printf(seq, " super external:%s", | |
6281 | mddev->metadata_type); | |
6282 | else | |
1cd6bf19 | 6283 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
6284 | |
6285 | if (mddev->pers) { | |
d710e138 | 6286 | mddev->pers->status(seq, mddev); |
1da177e4 | 6287 | seq_printf(seq, "\n "); |
8e1b39d6 N |
6288 | if (mddev->pers->sync_request) { |
6289 | if (mddev->curr_resync > 2) { | |
d710e138 | 6290 | status_resync(seq, mddev); |
8e1b39d6 N |
6291 | seq_printf(seq, "\n "); |
6292 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
6293 | seq_printf(seq, "\tresync=DELAYED\n "); | |
6294 | else if (mddev->recovery_cp < MaxSector) | |
6295 | seq_printf(seq, "\tresync=PENDING\n "); | |
6296 | } | |
32a7627c N |
6297 | } else |
6298 | seq_printf(seq, "\n "); | |
6299 | ||
6300 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
6301 | unsigned long chunk_kb; |
6302 | unsigned long flags; | |
32a7627c | 6303 | spin_lock_irqsave(&bitmap->lock, flags); |
42a04b50 | 6304 | chunk_kb = mddev->bitmap_info.chunksize >> 10; |
32a7627c N |
6305 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " |
6306 | "%lu%s chunk", | |
6307 | bitmap->pages - bitmap->missing_pages, | |
6308 | bitmap->pages, | |
6309 | (bitmap->pages - bitmap->missing_pages) | |
6310 | << (PAGE_SHIFT - 10), | |
42a04b50 | 6311 | chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, |
32a7627c | 6312 | chunk_kb ? "KB" : "B"); |
78d742d8 N |
6313 | if (bitmap->file) { |
6314 | seq_printf(seq, ", file: "); | |
c32c2f63 | 6315 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 6316 | } |
78d742d8 | 6317 | |
32a7627c N |
6318 | seq_printf(seq, "\n"); |
6319 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
6320 | } |
6321 | ||
6322 | seq_printf(seq, "\n"); | |
6323 | } | |
6324 | mddev_unlock(mddev); | |
6325 | ||
6326 | return 0; | |
6327 | } | |
6328 | ||
110518bc | 6329 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6330 | .start = md_seq_start, |
6331 | .next = md_seq_next, | |
6332 | .stop = md_seq_stop, | |
6333 | .show = md_seq_show, | |
6334 | }; | |
6335 | ||
6336 | static int md_seq_open(struct inode *inode, struct file *file) | |
6337 | { | |
6338 | int error; | |
d7603b7e N |
6339 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6340 | if (mi == NULL) | |
6341 | return -ENOMEM; | |
1da177e4 LT |
6342 | |
6343 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6344 | if (error) |
6345 | kfree(mi); | |
6346 | else { | |
6347 | struct seq_file *p = file->private_data; | |
6348 | p->private = mi; | |
6349 | mi->event = atomic_read(&md_event_count); | |
6350 | } | |
1da177e4 LT |
6351 | return error; |
6352 | } | |
6353 | ||
d7603b7e N |
6354 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6355 | { | |
6356 | struct seq_file *m = filp->private_data; | |
6357 | struct mdstat_info *mi = m->private; | |
6358 | int mask; | |
6359 | ||
6360 | poll_wait(filp, &md_event_waiters, wait); | |
6361 | ||
6362 | /* always allow read */ | |
6363 | mask = POLLIN | POLLRDNORM; | |
6364 | ||
6365 | if (mi->event != atomic_read(&md_event_count)) | |
6366 | mask |= POLLERR | POLLPRI; | |
6367 | return mask; | |
6368 | } | |
6369 | ||
fa027c2a | 6370 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6371 | .owner = THIS_MODULE, |
1da177e4 LT |
6372 | .open = md_seq_open, |
6373 | .read = seq_read, | |
6374 | .llseek = seq_lseek, | |
c3f94b40 | 6375 | .release = seq_release_private, |
d7603b7e | 6376 | .poll = mdstat_poll, |
1da177e4 LT |
6377 | }; |
6378 | ||
2604b703 | 6379 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6380 | { |
1da177e4 | 6381 | spin_lock(&pers_lock); |
2604b703 N |
6382 | list_add_tail(&p->list, &pers_list); |
6383 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6384 | spin_unlock(&pers_lock); |
6385 | return 0; | |
6386 | } | |
6387 | ||
2604b703 | 6388 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6389 | { |
2604b703 | 6390 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6391 | spin_lock(&pers_lock); |
2604b703 | 6392 | list_del_init(&p->list); |
1da177e4 LT |
6393 | spin_unlock(&pers_lock); |
6394 | return 0; | |
6395 | } | |
6396 | ||
eea1bf38 | 6397 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6398 | { |
6399 | mdk_rdev_t * rdev; | |
1da177e4 | 6400 | int idle; |
eea1bf38 | 6401 | int curr_events; |
1da177e4 LT |
6402 | |
6403 | idle = 1; | |
4b80991c N |
6404 | rcu_read_lock(); |
6405 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6406 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6407 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6408 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6409 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6410 | /* sync IO will cause sync_io to increase before the disk_stats |
6411 | * as sync_io is counted when a request starts, and | |
6412 | * disk_stats is counted when it completes. | |
6413 | * So resync activity will cause curr_events to be smaller than | |
6414 | * when there was no such activity. | |
6415 | * non-sync IO will cause disk_stat to increase without | |
6416 | * increasing sync_io so curr_events will (eventually) | |
6417 | * be larger than it was before. Once it becomes | |
6418 | * substantially larger, the test below will cause | |
6419 | * the array to appear non-idle, and resync will slow | |
6420 | * down. | |
6421 | * If there is a lot of outstanding resync activity when | |
6422 | * we set last_event to curr_events, then all that activity | |
6423 | * completing might cause the array to appear non-idle | |
6424 | * and resync will be slowed down even though there might | |
6425 | * not have been non-resync activity. This will only | |
6426 | * happen once though. 'last_events' will soon reflect | |
6427 | * the state where there is little or no outstanding | |
6428 | * resync requests, and further resync activity will | |
6429 | * always make curr_events less than last_events. | |
c0e48521 | 6430 | * |
1da177e4 | 6431 | */ |
eea1bf38 | 6432 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6433 | rdev->last_events = curr_events; |
6434 | idle = 0; | |
6435 | } | |
6436 | } | |
4b80991c | 6437 | rcu_read_unlock(); |
1da177e4 LT |
6438 | return idle; |
6439 | } | |
6440 | ||
6441 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6442 | { | |
6443 | /* another "blocks" (512byte) blocks have been synced */ | |
6444 | atomic_sub(blocks, &mddev->recovery_active); | |
6445 | wake_up(&mddev->recovery_wait); | |
6446 | if (!ok) { | |
dfc70645 | 6447 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6448 | md_wakeup_thread(mddev->thread); |
6449 | // stop recovery, signal do_sync .... | |
6450 | } | |
6451 | } | |
6452 | ||
6453 | ||
06d91a5f N |
6454 | /* md_write_start(mddev, bi) |
6455 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6456 | * in superblock) before writing, schedule a superblock update |
6457 | * and wait for it to complete. | |
06d91a5f | 6458 | */ |
3d310eb7 | 6459 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6460 | { |
0fd62b86 | 6461 | int did_change = 0; |
06d91a5f | 6462 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6463 | return; |
06d91a5f | 6464 | |
f91de92e N |
6465 | BUG_ON(mddev->ro == 1); |
6466 | if (mddev->ro == 2) { | |
6467 | /* need to switch to read/write */ | |
6468 | mddev->ro = 0; | |
6469 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6470 | md_wakeup_thread(mddev->thread); | |
25156198 | 6471 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6472 | did_change = 1; |
f91de92e | 6473 | } |
06d91a5f | 6474 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6475 | if (mddev->safemode == 1) |
6476 | mddev->safemode = 0; | |
06d91a5f | 6477 | if (mddev->in_sync) { |
a9701a30 | 6478 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6479 | if (mddev->in_sync) { |
6480 | mddev->in_sync = 0; | |
850b2b42 | 6481 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3d310eb7 | 6482 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6483 | did_change = 1; |
3d310eb7 | 6484 | } |
a9701a30 | 6485 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6486 | } |
0fd62b86 | 6487 | if (did_change) |
b62b7590 | 6488 | sysfs_notify_dirent(mddev->sysfs_state); |
09a44cc1 N |
6489 | wait_event(mddev->sb_wait, |
6490 | !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && | |
6491 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1da177e4 LT |
6492 | } |
6493 | ||
6494 | void md_write_end(mddev_t *mddev) | |
6495 | { | |
6496 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6497 | if (mddev->safemode == 2) | |
6498 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6499 | else if (mddev->safemode_delay) |
1da177e4 LT |
6500 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6501 | } | |
6502 | } | |
6503 | ||
2a2275d6 N |
6504 | /* md_allow_write(mddev) |
6505 | * Calling this ensures that the array is marked 'active' so that writes | |
6506 | * may proceed without blocking. It is important to call this before | |
6507 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6508 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6509 | * |
6510 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6511 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6512 | */ |
b5470dc5 | 6513 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6514 | { |
6515 | if (!mddev->pers) | |
b5470dc5 | 6516 | return 0; |
2a2275d6 | 6517 | if (mddev->ro) |
b5470dc5 | 6518 | return 0; |
1a0fd497 | 6519 | if (!mddev->pers->sync_request) |
b5470dc5 | 6520 | return 0; |
2a2275d6 N |
6521 | |
6522 | spin_lock_irq(&mddev->write_lock); | |
6523 | if (mddev->in_sync) { | |
6524 | mddev->in_sync = 0; | |
6525 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6526 | if (mddev->safemode_delay && | |
6527 | mddev->safemode == 0) | |
6528 | mddev->safemode = 1; | |
6529 | spin_unlock_irq(&mddev->write_lock); | |
6530 | md_update_sb(mddev, 0); | |
b62b7590 | 6531 | sysfs_notify_dirent(mddev->sysfs_state); |
2a2275d6 N |
6532 | } else |
6533 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 DW |
6534 | |
6535 | if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
6536 | return -EAGAIN; | |
6537 | else | |
6538 | return 0; | |
2a2275d6 N |
6539 | } |
6540 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6541 | ||
1da177e4 LT |
6542 | #define SYNC_MARKS 10 |
6543 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6544 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6545 | { |
6546 | mddev_t *mddev2; | |
6547 | unsigned int currspeed = 0, | |
6548 | window; | |
57afd89f | 6549 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6550 | unsigned long mark[SYNC_MARKS]; |
6551 | sector_t mark_cnt[SYNC_MARKS]; | |
6552 | int last_mark,m; | |
6553 | struct list_head *tmp; | |
6554 | sector_t last_check; | |
57afd89f | 6555 | int skipped = 0; |
5fd6c1dc | 6556 | mdk_rdev_t *rdev; |
61df9d91 | 6557 | char *desc; |
1da177e4 LT |
6558 | |
6559 | /* just incase thread restarts... */ | |
6560 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6561 | return; | |
5fd6c1dc N |
6562 | if (mddev->ro) /* never try to sync a read-only array */ |
6563 | return; | |
1da177e4 | 6564 | |
61df9d91 N |
6565 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6566 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6567 | desc = "data-check"; | |
6568 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6569 | desc = "requested-resync"; | |
6570 | else | |
6571 | desc = "resync"; | |
6572 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6573 | desc = "reshape"; | |
6574 | else | |
6575 | desc = "recovery"; | |
6576 | ||
1da177e4 LT |
6577 | /* we overload curr_resync somewhat here. |
6578 | * 0 == not engaged in resync at all | |
6579 | * 2 == checking that there is no conflict with another sync | |
6580 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6581 | * commense | |
6582 | * other == active in resync - this many blocks | |
6583 | * | |
6584 | * Before starting a resync we must have set curr_resync to | |
6585 | * 2, and then checked that every "conflicting" array has curr_resync | |
6586 | * less than ours. When we find one that is the same or higher | |
6587 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6588 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6589 | * This will mean we have to start checking from the beginning again. | |
6590 | * | |
6591 | */ | |
6592 | ||
6593 | do { | |
6594 | mddev->curr_resync = 2; | |
6595 | ||
6596 | try_again: | |
404e4b43 | 6597 | if (kthread_should_stop()) |
6985c43f | 6598 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
404e4b43 N |
6599 | |
6600 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
1da177e4 | 6601 | goto skip; |
29ac4aa3 | 6602 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6603 | if (mddev2 == mddev) |
6604 | continue; | |
90b08710 BS |
6605 | if (!mddev->parallel_resync |
6606 | && mddev2->curr_resync | |
6607 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6608 | DEFINE_WAIT(wq); |
6609 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6610 | /* arbitrarily yield */ | |
6611 | mddev->curr_resync = 1; | |
6612 | wake_up(&resync_wait); | |
6613 | } | |
6614 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6615 | /* no need to wait here, we can wait the next | |
6616 | * time 'round when curr_resync == 2 | |
6617 | */ | |
6618 | continue; | |
9744197c N |
6619 | /* We need to wait 'interruptible' so as not to |
6620 | * contribute to the load average, and not to | |
6621 | * be caught by 'softlockup' | |
6622 | */ | |
6623 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6624 | if (!kthread_should_stop() && |
8712e553 | 6625 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6626 | printk(KERN_INFO "md: delaying %s of %s" |
6627 | " until %s has finished (they" | |
1da177e4 | 6628 | " share one or more physical units)\n", |
61df9d91 | 6629 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6630 | mddev_put(mddev2); |
9744197c N |
6631 | if (signal_pending(current)) |
6632 | flush_signals(current); | |
1da177e4 LT |
6633 | schedule(); |
6634 | finish_wait(&resync_wait, &wq); | |
6635 | goto try_again; | |
6636 | } | |
6637 | finish_wait(&resync_wait, &wq); | |
6638 | } | |
6639 | } | |
6640 | } while (mddev->curr_resync < 2); | |
6641 | ||
5fd6c1dc | 6642 | j = 0; |
9d88883e | 6643 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6644 | /* resync follows the size requested by the personality, |
57afd89f | 6645 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6646 | */ |
6647 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6648 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6649 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6650 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6651 | j = mddev->resync_min; | |
6652 | else if (!mddev->bitmap) | |
5fd6c1dc | 6653 | j = mddev->recovery_cp; |
5e96ee65 | 6654 | |
ccfcc3c1 | 6655 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6656 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6657 | else { |
1da177e4 | 6658 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6659 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6660 | j = MaxSector; |
4e59ca7d DW |
6661 | rcu_read_lock(); |
6662 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6663 | if (rdev->raid_disk >= 0 && |
6664 | !test_bit(Faulty, &rdev->flags) && | |
6665 | !test_bit(In_sync, &rdev->flags) && | |
6666 | rdev->recovery_offset < j) | |
6667 | j = rdev->recovery_offset; | |
4e59ca7d | 6668 | rcu_read_unlock(); |
5fd6c1dc | 6669 | } |
1da177e4 | 6670 | |
61df9d91 N |
6671 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6672 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6673 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6674 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6675 | "(but not more than %d KB/sec) for %s.\n", |
6676 | speed_max(mddev), desc); | |
1da177e4 | 6677 | |
eea1bf38 | 6678 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6679 | |
57afd89f | 6680 | io_sectors = 0; |
1da177e4 LT |
6681 | for (m = 0; m < SYNC_MARKS; m++) { |
6682 | mark[m] = jiffies; | |
57afd89f | 6683 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6684 | } |
6685 | last_mark = 0; | |
6686 | mddev->resync_mark = mark[last_mark]; | |
6687 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6688 | ||
6689 | /* | |
6690 | * Tune reconstruction: | |
6691 | */ | |
6692 | window = 32*(PAGE_SIZE/512); | |
6693 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6694 | window/2,(unsigned long long) max_sectors/2); | |
6695 | ||
6696 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6697 | last_check = 0; |
6698 | ||
6699 | if (j>2) { | |
6700 | printk(KERN_INFO | |
61df9d91 N |
6701 | "md: resuming %s of %s from checkpoint.\n", |
6702 | desc, mdname(mddev)); | |
1da177e4 LT |
6703 | mddev->curr_resync = j; |
6704 | } | |
efa59339 | 6705 | mddev->curr_resync_completed = mddev->curr_resync; |
1da177e4 LT |
6706 | |
6707 | while (j < max_sectors) { | |
57afd89f | 6708 | sector_t sectors; |
1da177e4 | 6709 | |
57afd89f | 6710 | skipped = 0; |
97e4f42d | 6711 | |
7a91ee1f N |
6712 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6713 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6714 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6715 | > (max_sectors >> 4)) || | |
6716 | (j - mddev->curr_resync_completed)*2 | |
6717 | >= mddev->resync_max - mddev->curr_resync_completed | |
6718 | )) { | |
97e4f42d N |
6719 | /* time to update curr_resync_completed */ |
6720 | blk_unplug(mddev->queue); | |
6721 | wait_event(mddev->recovery_wait, | |
6722 | atomic_read(&mddev->recovery_active) == 0); | |
6723 | mddev->curr_resync_completed = | |
6724 | mddev->curr_resync; | |
6725 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
acb180b0 | 6726 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6727 | } |
acb180b0 | 6728 | |
e62e58a5 N |
6729 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6730 | /* As this condition is controlled by user-space, | |
6731 | * we can block indefinitely, so use '_interruptible' | |
6732 | * to avoid triggering warnings. | |
6733 | */ | |
6734 | flush_signals(current); /* just in case */ | |
6735 | wait_event_interruptible(mddev->recovery_wait, | |
6736 | mddev->resync_max > j | |
6737 | || kthread_should_stop()); | |
6738 | } | |
acb180b0 N |
6739 | |
6740 | if (kthread_should_stop()) | |
6741 | goto interrupted; | |
6742 | ||
57afd89f | 6743 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6744 | currspeed < speed_min(mddev)); |
57afd89f | 6745 | if (sectors == 0) { |
dfc70645 | 6746 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6747 | goto out; |
6748 | } | |
57afd89f N |
6749 | |
6750 | if (!skipped) { /* actual IO requested */ | |
6751 | io_sectors += sectors; | |
6752 | atomic_add(sectors, &mddev->recovery_active); | |
6753 | } | |
6754 | ||
1da177e4 LT |
6755 | j += sectors; |
6756 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6757 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6758 | if (last_check == 0) |
6759 | /* this is the earliers that rebuilt will be | |
6760 | * visible in /proc/mdstat | |
6761 | */ | |
6762 | md_new_event(mddev); | |
57afd89f N |
6763 | |
6764 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6765 | continue; |
6766 | ||
57afd89f | 6767 | last_check = io_sectors; |
1da177e4 | 6768 | |
dfc70645 | 6769 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6770 | break; |
6771 | ||
6772 | repeat: | |
6773 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6774 | /* step marks */ | |
6775 | int next = (last_mark+1) % SYNC_MARKS; | |
6776 | ||
6777 | mddev->resync_mark = mark[next]; | |
6778 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6779 | mark[next] = jiffies; | |
57afd89f | 6780 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6781 | last_mark = next; |
6782 | } | |
6783 | ||
6784 | ||
c6207277 N |
6785 | if (kthread_should_stop()) |
6786 | goto interrupted; | |
6787 | ||
1da177e4 LT |
6788 | |
6789 | /* | |
6790 | * this loop exits only if either when we are slower than | |
6791 | * the 'hard' speed limit, or the system was IO-idle for | |
6792 | * a jiffy. | |
6793 | * the system might be non-idle CPU-wise, but we only care | |
6794 | * about not overloading the IO subsystem. (things like an | |
6795 | * e2fsck being done on the RAID array should execute fast) | |
6796 | */ | |
2ad8b1ef | 6797 | blk_unplug(mddev->queue); |
1da177e4 LT |
6798 | cond_resched(); |
6799 | ||
57afd89f N |
6800 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6801 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6802 | |
88202a0c N |
6803 | if (currspeed > speed_min(mddev)) { |
6804 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6805 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6806 | msleep(500); |
1da177e4 LT |
6807 | goto repeat; |
6808 | } | |
6809 | } | |
6810 | } | |
61df9d91 | 6811 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6812 | /* |
6813 | * this also signals 'finished resyncing' to md_stop | |
6814 | */ | |
6815 | out: | |
2ad8b1ef | 6816 | blk_unplug(mddev->queue); |
1da177e4 LT |
6817 | |
6818 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
6819 | ||
6820 | /* tell personality that we are finished */ | |
57afd89f | 6821 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6822 | |
dfc70645 | 6823 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6824 | mddev->curr_resync > 2) { |
6825 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6826 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6827 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6828 | printk(KERN_INFO | |
61df9d91 N |
6829 | "md: checkpointing %s of %s.\n", |
6830 | desc, mdname(mddev)); | |
5fd6c1dc N |
6831 | mddev->recovery_cp = mddev->curr_resync; |
6832 | } | |
6833 | } else | |
6834 | mddev->recovery_cp = MaxSector; | |
6835 | } else { | |
6836 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6837 | mddev->curr_resync = MaxSector; | |
4e59ca7d DW |
6838 | rcu_read_lock(); |
6839 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6840 | if (rdev->raid_disk >= 0 && |
6841 | !test_bit(Faulty, &rdev->flags) && | |
6842 | !test_bit(In_sync, &rdev->flags) && | |
6843 | rdev->recovery_offset < mddev->curr_resync) | |
6844 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 6845 | rcu_read_unlock(); |
5fd6c1dc | 6846 | } |
1da177e4 | 6847 | } |
17571284 | 6848 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6849 | |
1da177e4 | 6850 | skip: |
c07b70ad N |
6851 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
6852 | /* We completed so min/max setting can be forgotten if used. */ | |
6853 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6854 | mddev->resync_min = 0; | |
6855 | mddev->resync_max = MaxSector; | |
6856 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6857 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 | 6858 | mddev->curr_resync = 0; |
efa59339 N |
6859 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
6860 | mddev->curr_resync_completed = 0; | |
c6207277 | 6861 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
1da177e4 LT |
6862 | wake_up(&resync_wait); |
6863 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
6864 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
6865 | return; |
6866 | ||
6867 | interrupted: | |
6868 | /* | |
6869 | * got a signal, exit. | |
6870 | */ | |
6871 | printk(KERN_INFO | |
6872 | "md: md_do_sync() got signal ... exiting\n"); | |
6873 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6874 | goto out; | |
6875 | ||
1da177e4 | 6876 | } |
29269553 | 6877 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
6878 | |
6879 | ||
b4c4c7b8 N |
6880 | static int remove_and_add_spares(mddev_t *mddev) |
6881 | { | |
6882 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
6883 | int spares = 0; |
6884 | ||
97e4f42d N |
6885 | mddev->curr_resync_completed = 0; |
6886 | ||
159ec1fc | 6887 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 6888 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 6889 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
6890 | (test_bit(Faulty, &rdev->flags) || |
6891 | ! test_bit(In_sync, &rdev->flags)) && | |
6892 | atomic_read(&rdev->nr_pending)==0) { | |
6893 | if (mddev->pers->hot_remove_disk( | |
6894 | mddev, rdev->raid_disk)==0) { | |
6895 | char nm[20]; | |
6896 | sprintf(nm,"rd%d", rdev->raid_disk); | |
6897 | sysfs_remove_link(&mddev->kobj, nm); | |
6898 | rdev->raid_disk = -1; | |
6899 | } | |
6900 | } | |
6901 | ||
4044ba58 | 6902 | if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { |
159ec1fc | 6903 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 6904 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
6905 | !test_bit(In_sync, &rdev->flags) && |
6906 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 6907 | spares++; |
b4c4c7b8 N |
6908 | if (rdev->raid_disk < 0 |
6909 | && !test_bit(Faulty, &rdev->flags)) { | |
6910 | rdev->recovery_offset = 0; | |
199050ea NB |
6911 | if (mddev->pers-> |
6912 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
6913 | char nm[20]; |
6914 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
6915 | if (sysfs_create_link(&mddev->kobj, |
6916 | &rdev->kobj, nm)) | |
6917 | printk(KERN_WARNING | |
6918 | "md: cannot register " | |
6919 | "%s for %s\n", | |
6920 | nm, mdname(mddev)); | |
b4c4c7b8 N |
6921 | spares++; |
6922 | md_new_event(mddev); | |
93be75ff | 6923 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
b4c4c7b8 N |
6924 | } else |
6925 | break; | |
6926 | } | |
dfc70645 | 6927 | } |
b4c4c7b8 N |
6928 | } |
6929 | return spares; | |
6930 | } | |
1da177e4 LT |
6931 | /* |
6932 | * This routine is regularly called by all per-raid-array threads to | |
6933 | * deal with generic issues like resync and super-block update. | |
6934 | * Raid personalities that don't have a thread (linear/raid0) do not | |
6935 | * need this as they never do any recovery or update the superblock. | |
6936 | * | |
6937 | * It does not do any resync itself, but rather "forks" off other threads | |
6938 | * to do that as needed. | |
6939 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
6940 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 6941 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
6942 | * and wakeups up this thread which will reap the thread and finish up. |
6943 | * This thread also removes any faulty devices (with nr_pending == 0). | |
6944 | * | |
6945 | * The overall approach is: | |
6946 | * 1/ if the superblock needs updating, update it. | |
6947 | * 2/ If a recovery thread is running, don't do anything else. | |
6948 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
6949 | * 4/ If there are any faulty devices, remove them. | |
6950 | * 5/ If array is degraded, try to add spares devices | |
6951 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
6952 | */ | |
6953 | void md_check_recovery(mddev_t *mddev) | |
6954 | { | |
6955 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6956 | |
6957 | ||
5f40402d | 6958 | if (mddev->bitmap) |
aa5cbd10 | 6959 | bitmap_daemon_work(mddev); |
1da177e4 LT |
6960 | |
6961 | if (mddev->ro) | |
6962 | return; | |
fca4d848 N |
6963 | |
6964 | if (signal_pending(current)) { | |
31a59e34 | 6965 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
6966 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
6967 | mdname(mddev)); | |
6968 | mddev->safemode = 2; | |
6969 | } | |
6970 | flush_signals(current); | |
6971 | } | |
6972 | ||
c89a8eee N |
6973 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
6974 | return; | |
1da177e4 | 6975 | if ( ! ( |
e691063a | 6976 | (mddev->flags && !mddev->external) || |
1da177e4 | 6977 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 6978 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 6979 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
6980 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
6981 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
6982 | )) |
6983 | return; | |
fca4d848 | 6984 | |
df5b89b3 | 6985 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 6986 | int spares = 0; |
fca4d848 | 6987 | |
c89a8eee N |
6988 | if (mddev->ro) { |
6989 | /* Only thing we do on a ro array is remove | |
6990 | * failed devices. | |
6991 | */ | |
6992 | remove_and_add_spares(mddev); | |
6993 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6994 | goto unlock; | |
6995 | } | |
6996 | ||
31a59e34 | 6997 | if (!mddev->external) { |
0fd62b86 | 6998 | int did_change = 0; |
31a59e34 N |
6999 | spin_lock_irq(&mddev->write_lock); |
7000 | if (mddev->safemode && | |
7001 | !atomic_read(&mddev->writes_pending) && | |
7002 | !mddev->in_sync && | |
7003 | mddev->recovery_cp == MaxSector) { | |
7004 | mddev->in_sync = 1; | |
0fd62b86 | 7005 | did_change = 1; |
31a59e34 N |
7006 | if (mddev->persistent) |
7007 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
7008 | } | |
7009 | if (mddev->safemode == 1) | |
7010 | mddev->safemode = 0; | |
7011 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7012 | if (did_change) |
b62b7590 | 7013 | sysfs_notify_dirent(mddev->sysfs_state); |
fca4d848 | 7014 | } |
fca4d848 | 7015 | |
850b2b42 N |
7016 | if (mddev->flags) |
7017 | md_update_sb(mddev, 0); | |
06d91a5f | 7018 | |
1da177e4 LT |
7019 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7020 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7021 | /* resync/recovery still happening */ | |
7022 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7023 | goto unlock; | |
7024 | } | |
7025 | if (mddev->sync_thread) { | |
7026 | /* resync has finished, collect result */ | |
7027 | md_unregister_thread(mddev->sync_thread); | |
7028 | mddev->sync_thread = NULL; | |
56ac36d7 DW |
7029 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
7030 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1da177e4 LT |
7031 | /* success...*/ |
7032 | /* activate any spares */ | |
a99ac971 NB |
7033 | if (mddev->pers->spare_active(mddev)) |
7034 | sysfs_notify(&mddev->kobj, NULL, | |
7035 | "degraded"); | |
1da177e4 | 7036 | } |
cea9c228 N |
7037 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7038 | mddev->pers->finish_reshape) | |
7039 | mddev->pers->finish_reshape(mddev); | |
850b2b42 | 7040 | md_update_sb(mddev, 1); |
41158c7e N |
7041 | |
7042 | /* if array is no-longer degraded, then any saved_raid_disk | |
7043 | * information must be scrapped | |
7044 | */ | |
7045 | if (!mddev->degraded) | |
159ec1fc | 7046 | list_for_each_entry(rdev, &mddev->disks, same_set) |
41158c7e N |
7047 | rdev->saved_raid_disk = -1; |
7048 | ||
1da177e4 LT |
7049 | mddev->recovery = 0; |
7050 | /* flag recovery needed just to double check */ | |
7051 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
0c3573f1 | 7052 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 7053 | md_new_event(mddev); |
1da177e4 LT |
7054 | goto unlock; |
7055 | } | |
72a23c21 NB |
7056 | /* Set RUNNING before clearing NEEDED to avoid |
7057 | * any transients in the value of "sync_action". | |
7058 | */ | |
7059 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7060 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
7061 | /* Clear some bits that don't mean anything, but |
7062 | * might be left set | |
7063 | */ | |
24dd469d N |
7064 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7065 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7066 | |
5fd6c1dc N |
7067 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
7068 | goto unlock; | |
1da177e4 LT |
7069 | /* no recovery is running. |
7070 | * remove any failed drives, then | |
7071 | * add spares if possible. | |
7072 | * Spare are also removed and re-added, to allow | |
7073 | * the personality to fail the re-add. | |
7074 | */ | |
1da177e4 | 7075 | |
b4c4c7b8 | 7076 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7077 | if (mddev->pers->check_reshape == NULL || |
7078 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7079 | /* Cannot proceed */ |
7080 | goto unlock; | |
7081 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7082 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 7083 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
7084 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7085 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7086 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7087 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7088 | } else if (mddev->recovery_cp < MaxSector) { |
7089 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7090 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7091 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7092 | /* nothing to be done ... */ | |
1da177e4 | 7093 | goto unlock; |
24dd469d | 7094 | |
1da177e4 | 7095 | if (mddev->pers->sync_request) { |
a654b9d8 N |
7096 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
7097 | /* We are adding a device or devices to an array | |
7098 | * which has the bitmap stored on all devices. | |
7099 | * So make sure all bitmap pages get written | |
7100 | */ | |
7101 | bitmap_write_all(mddev->bitmap); | |
7102 | } | |
1da177e4 LT |
7103 | mddev->sync_thread = md_register_thread(md_do_sync, |
7104 | mddev, | |
0da3c619 | 7105 | "resync"); |
1da177e4 LT |
7106 | if (!mddev->sync_thread) { |
7107 | printk(KERN_ERR "%s: could not start resync" | |
7108 | " thread...\n", | |
7109 | mdname(mddev)); | |
7110 | /* leave the spares where they are, it shouldn't hurt */ | |
7111 | mddev->recovery = 0; | |
d7603b7e | 7112 | } else |
1da177e4 | 7113 | md_wakeup_thread(mddev->sync_thread); |
0c3573f1 | 7114 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 7115 | md_new_event(mddev); |
1da177e4 LT |
7116 | } |
7117 | unlock: | |
72a23c21 NB |
7118 | if (!mddev->sync_thread) { |
7119 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7120 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7121 | &mddev->recovery)) | |
0c3573f1 N |
7122 | if (mddev->sysfs_action) |
7123 | sysfs_notify_dirent(mddev->sysfs_action); | |
72a23c21 | 7124 | } |
1da177e4 LT |
7125 | mddev_unlock(mddev); |
7126 | } | |
7127 | } | |
7128 | ||
6bfe0b49 DW |
7129 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
7130 | { | |
3c0ee63a | 7131 | sysfs_notify_dirent(rdev->sysfs_state); |
6bfe0b49 DW |
7132 | wait_event_timeout(rdev->blocked_wait, |
7133 | !test_bit(Blocked, &rdev->flags), | |
7134 | msecs_to_jiffies(5000)); | |
7135 | rdev_dec_pending(rdev, mddev); | |
7136 | } | |
7137 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7138 | ||
75c96f85 AB |
7139 | static int md_notify_reboot(struct notifier_block *this, |
7140 | unsigned long code, void *x) | |
1da177e4 LT |
7141 | { |
7142 | struct list_head *tmp; | |
7143 | mddev_t *mddev; | |
7144 | ||
7145 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
7146 | ||
7147 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
7148 | ||
29ac4aa3 | 7149 | for_each_mddev(mddev, tmp) |
c71d4887 | 7150 | if (mddev_trylock(mddev)) { |
2b25000b N |
7151 | /* Force a switch to readonly even array |
7152 | * appears to still be in use. Hence | |
7153 | * the '100'. | |
7154 | */ | |
a4bd82d0 | 7155 | md_set_readonly(mddev, 100); |
c71d4887 NB |
7156 | mddev_unlock(mddev); |
7157 | } | |
1da177e4 LT |
7158 | /* |
7159 | * certain more exotic SCSI devices are known to be | |
7160 | * volatile wrt too early system reboots. While the | |
7161 | * right place to handle this issue is the given | |
7162 | * driver, we do want to have a safe RAID driver ... | |
7163 | */ | |
7164 | mdelay(1000*1); | |
7165 | } | |
7166 | return NOTIFY_DONE; | |
7167 | } | |
7168 | ||
75c96f85 | 7169 | static struct notifier_block md_notifier = { |
1da177e4 LT |
7170 | .notifier_call = md_notify_reboot, |
7171 | .next = NULL, | |
7172 | .priority = INT_MAX, /* before any real devices */ | |
7173 | }; | |
7174 | ||
7175 | static void md_geninit(void) | |
7176 | { | |
1da177e4 LT |
7177 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
7178 | ||
c7705f34 | 7179 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
7180 | } |
7181 | ||
75c96f85 | 7182 | static int __init md_init(void) |
1da177e4 | 7183 | { |
3dbd8c2e | 7184 | if (register_blkdev(MD_MAJOR, "md")) |
1da177e4 LT |
7185 | return -1; |
7186 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
3dbd8c2e | 7187 | unregister_blkdev(MD_MAJOR, "md"); |
1da177e4 LT |
7188 | return -1; |
7189 | } | |
3dbd8c2e | 7190 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
7191 | md_probe, NULL, NULL); |
7192 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
7193 | md_probe, NULL, NULL); |
7194 | ||
1da177e4 | 7195 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 7196 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
7197 | |
7198 | md_geninit(); | |
d710e138 | 7199 | return 0; |
1da177e4 LT |
7200 | } |
7201 | ||
7202 | ||
7203 | #ifndef MODULE | |
7204 | ||
7205 | /* | |
7206 | * Searches all registered partitions for autorun RAID arrays | |
7207 | * at boot time. | |
7208 | */ | |
4d936ec1 ME |
7209 | |
7210 | static LIST_HEAD(all_detected_devices); | |
7211 | struct detected_devices_node { | |
7212 | struct list_head list; | |
7213 | dev_t dev; | |
7214 | }; | |
1da177e4 LT |
7215 | |
7216 | void md_autodetect_dev(dev_t dev) | |
7217 | { | |
4d936ec1 ME |
7218 | struct detected_devices_node *node_detected_dev; |
7219 | ||
7220 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
7221 | if (node_detected_dev) { | |
7222 | node_detected_dev->dev = dev; | |
7223 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
7224 | } else { | |
7225 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
7226 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
7227 | } | |
1da177e4 LT |
7228 | } |
7229 | ||
7230 | ||
7231 | static void autostart_arrays(int part) | |
7232 | { | |
7233 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
7234 | struct detected_devices_node *node_detected_dev; |
7235 | dev_t dev; | |
7236 | int i_scanned, i_passed; | |
1da177e4 | 7237 | |
4d936ec1 ME |
7238 | i_scanned = 0; |
7239 | i_passed = 0; | |
1da177e4 | 7240 | |
4d936ec1 | 7241 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 7242 | |
4d936ec1 ME |
7243 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
7244 | i_scanned++; | |
7245 | node_detected_dev = list_entry(all_detected_devices.next, | |
7246 | struct detected_devices_node, list); | |
7247 | list_del(&node_detected_dev->list); | |
7248 | dev = node_detected_dev->dev; | |
7249 | kfree(node_detected_dev); | |
df968c4e | 7250 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
7251 | if (IS_ERR(rdev)) |
7252 | continue; | |
7253 | ||
b2d444d7 | 7254 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7255 | MD_BUG(); |
7256 | continue; | |
7257 | } | |
d0fae18f | 7258 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 7259 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 7260 | i_passed++; |
1da177e4 | 7261 | } |
4d936ec1 ME |
7262 | |
7263 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
7264 | i_scanned, i_passed); | |
1da177e4 LT |
7265 | |
7266 | autorun_devices(part); | |
7267 | } | |
7268 | ||
fdee8ae4 | 7269 | #endif /* !MODULE */ |
1da177e4 LT |
7270 | |
7271 | static __exit void md_exit(void) | |
7272 | { | |
7273 | mddev_t *mddev; | |
7274 | struct list_head *tmp; | |
8ab5e4c1 | 7275 | |
3dbd8c2e | 7276 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 7277 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 7278 | |
3dbd8c2e | 7279 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
7280 | unregister_blkdev(mdp_major, "mdp"); |
7281 | unregister_reboot_notifier(&md_notifier); | |
7282 | unregister_sysctl_table(raid_table_header); | |
7283 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 7284 | for_each_mddev(mddev, tmp) { |
1da177e4 | 7285 | export_array(mddev); |
d3374825 | 7286 | mddev->hold_active = 0; |
1da177e4 LT |
7287 | } |
7288 | } | |
7289 | ||
685784aa | 7290 | subsys_initcall(md_init); |
1da177e4 LT |
7291 | module_exit(md_exit) |
7292 | ||
f91de92e N |
7293 | static int get_ro(char *buffer, struct kernel_param *kp) |
7294 | { | |
7295 | return sprintf(buffer, "%d", start_readonly); | |
7296 | } | |
7297 | static int set_ro(const char *val, struct kernel_param *kp) | |
7298 | { | |
7299 | char *e; | |
7300 | int num = simple_strtoul(val, &e, 10); | |
7301 | if (*val && (*e == '\0' || *e == '\n')) { | |
7302 | start_readonly = num; | |
4dbcdc75 | 7303 | return 0; |
f91de92e N |
7304 | } |
7305 | return -EINVAL; | |
7306 | } | |
7307 | ||
80ca3a44 N |
7308 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
7309 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 7310 | |
efeb53c0 | 7311 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 7312 | |
1da177e4 LT |
7313 | EXPORT_SYMBOL(register_md_personality); |
7314 | EXPORT_SYMBOL(unregister_md_personality); | |
7315 | EXPORT_SYMBOL(md_error); | |
7316 | EXPORT_SYMBOL(md_done_sync); | |
7317 | EXPORT_SYMBOL(md_write_start); | |
7318 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
7319 | EXPORT_SYMBOL(md_register_thread); |
7320 | EXPORT_SYMBOL(md_unregister_thread); | |
7321 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
7322 | EXPORT_SYMBOL(md_check_recovery); |
7323 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7324 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 7325 | MODULE_ALIAS("md"); |
72008652 | 7326 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |