]>
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 | |
409c57f3 | 243 | rv = mddev->pers->make_request(q, 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 | ||
257 | static void mddev_suspend(mddev_t *mddev) | |
258 | { | |
259 | BUG_ON(mddev->suspended); | |
260 | mddev->suspended = 1; | |
261 | synchronize_rcu(); | |
262 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
263 | mddev->pers->quiesce(mddev, 1); | |
264 | md_unregister_thread(mddev->thread); | |
265 | mddev->thread = NULL; | |
266 | /* we now know that no code is executing in the personality module, | |
267 | * except possibly the tail end of a ->bi_end_io function, but that | |
268 | * is certain to complete before the module has a chance to get | |
269 | * unloaded | |
270 | */ | |
271 | } | |
272 | ||
273 | static void mddev_resume(mddev_t *mddev) | |
274 | { | |
275 | mddev->suspended = 0; | |
276 | wake_up(&mddev->sb_wait); | |
277 | mddev->pers->quiesce(mddev, 0); | |
1da177e4 LT |
278 | } |
279 | ||
3fa841d7 N |
280 | int mddev_congested(mddev_t *mddev, int bits) |
281 | { | |
a2826aa9 N |
282 | if (mddev->barrier) |
283 | return 1; | |
3fa841d7 N |
284 | return mddev->suspended; |
285 | } | |
286 | EXPORT_SYMBOL(mddev_congested); | |
287 | ||
a2826aa9 N |
288 | /* |
289 | * Generic barrier handling for md | |
290 | */ | |
291 | ||
292 | #define POST_REQUEST_BARRIER ((void*)1) | |
293 | ||
294 | static void md_end_barrier(struct bio *bio, int err) | |
295 | { | |
296 | mdk_rdev_t *rdev = bio->bi_private; | |
297 | mddev_t *mddev = rdev->mddev; | |
298 | if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER) | |
299 | set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags); | |
300 | ||
301 | rdev_dec_pending(rdev, mddev); | |
302 | ||
303 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
304 | if (mddev->barrier == POST_REQUEST_BARRIER) { | |
305 | /* This was a post-request barrier */ | |
306 | mddev->barrier = NULL; | |
307 | wake_up(&mddev->sb_wait); | |
308 | } else | |
309 | /* The pre-request barrier has finished */ | |
310 | schedule_work(&mddev->barrier_work); | |
311 | } | |
312 | bio_put(bio); | |
313 | } | |
314 | ||
315 | static void submit_barriers(mddev_t *mddev) | |
316 | { | |
317 | mdk_rdev_t *rdev; | |
318 | ||
319 | rcu_read_lock(); | |
320 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
321 | if (rdev->raid_disk >= 0 && | |
322 | !test_bit(Faulty, &rdev->flags)) { | |
323 | /* Take two references, one is dropped | |
324 | * when request finishes, one after | |
325 | * we reclaim rcu_read_lock | |
326 | */ | |
327 | struct bio *bi; | |
328 | atomic_inc(&rdev->nr_pending); | |
329 | atomic_inc(&rdev->nr_pending); | |
330 | rcu_read_unlock(); | |
331 | bi = bio_alloc(GFP_KERNEL, 0); | |
332 | bi->bi_end_io = md_end_barrier; | |
333 | bi->bi_private = rdev; | |
334 | bi->bi_bdev = rdev->bdev; | |
335 | atomic_inc(&mddev->flush_pending); | |
336 | submit_bio(WRITE_BARRIER, bi); | |
337 | rcu_read_lock(); | |
338 | rdev_dec_pending(rdev, mddev); | |
339 | } | |
340 | rcu_read_unlock(); | |
341 | } | |
342 | ||
343 | static void md_submit_barrier(struct work_struct *ws) | |
344 | { | |
345 | mddev_t *mddev = container_of(ws, mddev_t, barrier_work); | |
346 | struct bio *bio = mddev->barrier; | |
347 | ||
348 | atomic_set(&mddev->flush_pending, 1); | |
349 | ||
350 | if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags)) | |
351 | bio_endio(bio, -EOPNOTSUPP); | |
352 | else if (bio->bi_size == 0) | |
353 | /* an empty barrier - all done */ | |
354 | bio_endio(bio, 0); | |
355 | else { | |
356 | bio->bi_rw &= ~(1<<BIO_RW_BARRIER); | |
357 | if (mddev->pers->make_request(mddev->queue, bio)) | |
358 | generic_make_request(bio); | |
359 | mddev->barrier = POST_REQUEST_BARRIER; | |
360 | submit_barriers(mddev); | |
361 | } | |
362 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
363 | mddev->barrier = NULL; | |
364 | wake_up(&mddev->sb_wait); | |
365 | } | |
366 | } | |
367 | ||
368 | void md_barrier_request(mddev_t *mddev, struct bio *bio) | |
369 | { | |
370 | spin_lock_irq(&mddev->write_lock); | |
371 | wait_event_lock_irq(mddev->sb_wait, | |
372 | !mddev->barrier, | |
373 | mddev->write_lock, /*nothing*/); | |
374 | mddev->barrier = bio; | |
375 | spin_unlock_irq(&mddev->write_lock); | |
376 | ||
377 | atomic_set(&mddev->flush_pending, 1); | |
378 | INIT_WORK(&mddev->barrier_work, md_submit_barrier); | |
379 | ||
380 | submit_barriers(mddev); | |
381 | ||
382 | if (atomic_dec_and_test(&mddev->flush_pending)) | |
383 | schedule_work(&mddev->barrier_work); | |
384 | } | |
385 | EXPORT_SYMBOL(md_barrier_request); | |
409c57f3 | 386 | |
1da177e4 LT |
387 | static inline mddev_t *mddev_get(mddev_t *mddev) |
388 | { | |
389 | atomic_inc(&mddev->active); | |
390 | return mddev; | |
391 | } | |
392 | ||
5fd3a17e | 393 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 394 | |
1da177e4 LT |
395 | static void mddev_put(mddev_t *mddev) |
396 | { | |
397 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
398 | return; | |
d3374825 | 399 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
400 | mddev->ctime == 0 && !mddev->hold_active) { |
401 | /* Array is not configured at all, and not held active, | |
402 | * so destroy it */ | |
1da177e4 | 403 | list_del(&mddev->all_mddevs); |
d3374825 N |
404 | if (mddev->gendisk) { |
405 | /* we did a probe so need to clean up. | |
406 | * Call schedule_work inside the spinlock | |
407 | * so that flush_scheduled_work() after | |
408 | * mddev_find will succeed in waiting for the | |
409 | * work to be done. | |
410 | */ | |
411 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
412 | schedule_work(&mddev->del_work); | |
413 | } else | |
414 | kfree(mddev); | |
415 | } | |
416 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
417 | } |
418 | ||
419 | static mddev_t * mddev_find(dev_t unit) | |
420 | { | |
421 | mddev_t *mddev, *new = NULL; | |
422 | ||
423 | retry: | |
424 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
425 | |
426 | if (unit) { | |
427 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
428 | if (mddev->unit == unit) { | |
429 | mddev_get(mddev); | |
430 | spin_unlock(&all_mddevs_lock); | |
431 | kfree(new); | |
432 | return mddev; | |
433 | } | |
434 | ||
435 | if (new) { | |
436 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 437 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
438 | new->hold_active = UNTIL_IOCTL; |
439 | return new; | |
1da177e4 | 440 | } |
efeb53c0 N |
441 | } else if (new) { |
442 | /* find an unused unit number */ | |
443 | static int next_minor = 512; | |
444 | int start = next_minor; | |
445 | int is_free = 0; | |
446 | int dev = 0; | |
447 | while (!is_free) { | |
448 | dev = MKDEV(MD_MAJOR, next_minor); | |
449 | next_minor++; | |
450 | if (next_minor > MINORMASK) | |
451 | next_minor = 0; | |
452 | if (next_minor == start) { | |
453 | /* Oh dear, all in use. */ | |
454 | spin_unlock(&all_mddevs_lock); | |
455 | kfree(new); | |
456 | return NULL; | |
457 | } | |
458 | ||
459 | is_free = 1; | |
460 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
461 | if (mddev->unit == dev) { | |
462 | is_free = 0; | |
463 | break; | |
464 | } | |
465 | } | |
466 | new->unit = dev; | |
467 | new->md_minor = MINOR(dev); | |
468 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
469 | list_add(&new->all_mddevs, &all_mddevs); |
470 | spin_unlock(&all_mddevs_lock); | |
471 | return new; | |
472 | } | |
473 | spin_unlock(&all_mddevs_lock); | |
474 | ||
9ffae0cf | 475 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
476 | if (!new) |
477 | return NULL; | |
478 | ||
1da177e4 LT |
479 | new->unit = unit; |
480 | if (MAJOR(unit) == MD_MAJOR) | |
481 | new->md_minor = MINOR(unit); | |
482 | else | |
483 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
484 | ||
c8c00a69 | 485 | mutex_init(&new->open_mutex); |
df5b89b3 | 486 | mutex_init(&new->reconfig_mutex); |
c3d9714e | 487 | mutex_init(&new->bitmap_info.mutex); |
1da177e4 LT |
488 | INIT_LIST_HEAD(&new->disks); |
489 | INIT_LIST_HEAD(&new->all_mddevs); | |
490 | init_timer(&new->safemode_timer); | |
491 | atomic_set(&new->active, 1); | |
f2ea68cf | 492 | atomic_set(&new->openers, 0); |
409c57f3 | 493 | atomic_set(&new->active_io, 0); |
06d91a5f | 494 | spin_lock_init(&new->write_lock); |
a2826aa9 | 495 | atomic_set(&new->flush_pending, 0); |
3d310eb7 | 496 | init_waitqueue_head(&new->sb_wait); |
a6d8113a | 497 | init_waitqueue_head(&new->recovery_wait); |
08a02ecd | 498 | new->reshape_position = MaxSector; |
5e96ee65 | 499 | new->resync_min = 0; |
c6207277 | 500 | new->resync_max = MaxSector; |
d897dbf9 | 501 | new->level = LEVEL_NONE; |
1da177e4 | 502 | |
1da177e4 LT |
503 | goto retry; |
504 | } | |
505 | ||
506 | static inline int mddev_lock(mddev_t * mddev) | |
507 | { | |
df5b89b3 | 508 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
509 | } |
510 | ||
b522adcd DW |
511 | static inline int mddev_is_locked(mddev_t *mddev) |
512 | { | |
513 | return mutex_is_locked(&mddev->reconfig_mutex); | |
514 | } | |
515 | ||
1da177e4 LT |
516 | static inline int mddev_trylock(mddev_t * mddev) |
517 | { | |
df5b89b3 | 518 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
519 | } |
520 | ||
b6eb127d N |
521 | static struct attribute_group md_redundancy_group; |
522 | ||
a64c876f | 523 | static void mddev_unlock(mddev_t * mddev) |
1da177e4 | 524 | { |
a64c876f | 525 | if (mddev->to_remove) { |
b6eb127d N |
526 | /* These cannot be removed under reconfig_mutex as |
527 | * an access to the files will try to take reconfig_mutex | |
528 | * while holding the file unremovable, which leads to | |
529 | * a deadlock. | |
530 | * So hold open_mutex instead - we are allowed to take | |
531 | * it while holding reconfig_mutex, and md_run can | |
532 | * use it to wait for the remove to complete. | |
533 | */ | |
a64c876f N |
534 | struct attribute_group *to_remove = mddev->to_remove; |
535 | mddev->to_remove = NULL; | |
b6eb127d N |
536 | mutex_lock(&mddev->open_mutex); |
537 | mutex_unlock(&mddev->reconfig_mutex); | |
538 | ||
a64c876f N |
539 | if (to_remove != &md_redundancy_group) |
540 | sysfs_remove_group(&mddev->kobj, to_remove); | |
541 | if (mddev->pers == NULL || | |
542 | mddev->pers->sync_request == NULL) { | |
543 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
544 | if (mddev->sysfs_action) | |
545 | sysfs_put(mddev->sysfs_action); | |
546 | mddev->sysfs_action = NULL; | |
547 | } | |
b6eb127d N |
548 | mutex_unlock(&mddev->open_mutex); |
549 | } else | |
550 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 551 | |
005eca5e | 552 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
553 | } |
554 | ||
2989ddbd | 555 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 556 | { |
159ec1fc | 557 | mdk_rdev_t *rdev; |
1da177e4 | 558 | |
159ec1fc | 559 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
560 | if (rdev->desc_nr == nr) |
561 | return rdev; | |
159ec1fc | 562 | |
1da177e4 LT |
563 | return NULL; |
564 | } | |
565 | ||
566 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
567 | { | |
1da177e4 LT |
568 | mdk_rdev_t *rdev; |
569 | ||
159ec1fc | 570 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
571 | if (rdev->bdev->bd_dev == dev) |
572 | return rdev; | |
159ec1fc | 573 | |
1da177e4 LT |
574 | return NULL; |
575 | } | |
576 | ||
d9d166c2 | 577 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
578 | { |
579 | struct mdk_personality *pers; | |
d9d166c2 N |
580 | list_for_each_entry(pers, &pers_list, list) { |
581 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 582 | return pers; |
d9d166c2 N |
583 | if (strcmp(pers->name, clevel)==0) |
584 | return pers; | |
585 | } | |
2604b703 N |
586 | return NULL; |
587 | } | |
588 | ||
b73df2d3 | 589 | /* return the offset of the super block in 512byte sectors */ |
77933d72 | 590 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 | 591 | { |
b73df2d3 AN |
592 | sector_t num_sectors = bdev->bd_inode->i_size / 512; |
593 | return MD_NEW_SIZE_SECTORS(num_sectors); | |
1da177e4 LT |
594 | } |
595 | ||
1da177e4 LT |
596 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
597 | { | |
598 | if (rdev->sb_page) | |
599 | MD_BUG(); | |
600 | ||
601 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
602 | if (!rdev->sb_page) { | |
603 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 604 | return -ENOMEM; |
1da177e4 LT |
605 | } |
606 | ||
607 | return 0; | |
608 | } | |
609 | ||
610 | static void free_disk_sb(mdk_rdev_t * rdev) | |
611 | { | |
612 | if (rdev->sb_page) { | |
2d1f3b5d | 613 | put_page(rdev->sb_page); |
1da177e4 LT |
614 | rdev->sb_loaded = 0; |
615 | rdev->sb_page = NULL; | |
0f420358 | 616 | rdev->sb_start = 0; |
dd8ac336 | 617 | rdev->sectors = 0; |
1da177e4 LT |
618 | } |
619 | } | |
620 | ||
621 | ||
6712ecf8 | 622 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
623 | { |
624 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 625 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 626 | |
3a0f5bbb N |
627 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
628 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
629 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
630 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 631 | md_error(mddev, rdev); |
3a0f5bbb | 632 | } |
7bfa19f2 | 633 | |
a9701a30 N |
634 | if (atomic_dec_and_test(&mddev->pending_writes)) |
635 | wake_up(&mddev->sb_wait); | |
f8b58edf | 636 | bio_put(bio); |
7bfa19f2 N |
637 | } |
638 | ||
6712ecf8 | 639 | static void super_written_barrier(struct bio *bio, int error) |
a9701a30 N |
640 | { |
641 | struct bio *bio2 = bio->bi_private; | |
642 | mdk_rdev_t *rdev = bio2->bi_private; | |
643 | mddev_t *mddev = rdev->mddev; | |
a9701a30 N |
644 | |
645 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
646 | error == -EOPNOTSUPP) { | |
647 | unsigned long flags; | |
648 | /* barriers don't appear to be supported :-( */ | |
649 | set_bit(BarriersNotsupp, &rdev->flags); | |
650 | mddev->barriers_work = 0; | |
651 | spin_lock_irqsave(&mddev->write_lock, flags); | |
652 | bio2->bi_next = mddev->biolist; | |
653 | mddev->biolist = bio2; | |
654 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
655 | wake_up(&mddev->sb_wait); | |
656 | bio_put(bio); | |
6712ecf8 N |
657 | } else { |
658 | bio_put(bio2); | |
659 | bio->bi_private = rdev; | |
660 | super_written(bio, error); | |
a9701a30 | 661 | } |
a9701a30 N |
662 | } |
663 | ||
7bfa19f2 N |
664 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
665 | sector_t sector, int size, struct page *page) | |
666 | { | |
667 | /* write first size bytes of page to sector of rdev | |
668 | * Increment mddev->pending_writes before returning | |
669 | * and decrement it on completion, waking up sb_wait | |
670 | * if zero is reached. | |
671 | * If an error occurred, call md_error | |
a9701a30 N |
672 | * |
673 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
674 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
675 | */ |
676 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
93dbb393 | 677 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG); |
7bfa19f2 N |
678 | |
679 | bio->bi_bdev = rdev->bdev; | |
680 | bio->bi_sector = sector; | |
681 | bio_add_page(bio, page, size, 0); | |
682 | bio->bi_private = rdev; | |
683 | bio->bi_end_io = super_written; | |
a9701a30 N |
684 | bio->bi_rw = rw; |
685 | ||
7bfa19f2 | 686 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
687 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
688 | struct bio *rbio; | |
689 | rw |= (1<<BIO_RW_BARRIER); | |
690 | rbio = bio_clone(bio, GFP_NOIO); | |
691 | rbio->bi_private = bio; | |
692 | rbio->bi_end_io = super_written_barrier; | |
693 | submit_bio(rw, rbio); | |
694 | } else | |
695 | submit_bio(rw, bio); | |
696 | } | |
697 | ||
698 | void md_super_wait(mddev_t *mddev) | |
699 | { | |
700 | /* wait for all superblock writes that were scheduled to complete. | |
701 | * if any had to be retried (due to BARRIER problems), retry them | |
702 | */ | |
703 | DEFINE_WAIT(wq); | |
704 | for(;;) { | |
705 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
706 | if (atomic_read(&mddev->pending_writes)==0) | |
707 | break; | |
708 | while (mddev->biolist) { | |
709 | struct bio *bio; | |
710 | spin_lock_irq(&mddev->write_lock); | |
711 | bio = mddev->biolist; | |
712 | mddev->biolist = bio->bi_next ; | |
713 | bio->bi_next = NULL; | |
714 | spin_unlock_irq(&mddev->write_lock); | |
715 | submit_bio(bio->bi_rw, bio); | |
716 | } | |
717 | schedule(); | |
718 | } | |
719 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
720 | } |
721 | ||
6712ecf8 | 722 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 723 | { |
1da177e4 | 724 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
725 | } |
726 | ||
a654b9d8 | 727 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
728 | struct page *page, int rw) |
729 | { | |
baaa2c51 | 730 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
731 | struct completion event; |
732 | int ret; | |
733 | ||
93dbb393 | 734 | rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG); |
1da177e4 LT |
735 | |
736 | bio->bi_bdev = bdev; | |
737 | bio->bi_sector = sector; | |
738 | bio_add_page(bio, page, size, 0); | |
739 | init_completion(&event); | |
740 | bio->bi_private = &event; | |
741 | bio->bi_end_io = bi_complete; | |
742 | submit_bio(rw, bio); | |
743 | wait_for_completion(&event); | |
744 | ||
745 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
746 | bio_put(bio); | |
747 | return ret; | |
748 | } | |
a8745db2 | 749 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 750 | |
0002b271 | 751 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
752 | { |
753 | char b[BDEVNAME_SIZE]; | |
754 | if (!rdev->sb_page) { | |
755 | MD_BUG(); | |
756 | return -EINVAL; | |
757 | } | |
758 | if (rdev->sb_loaded) | |
759 | return 0; | |
760 | ||
761 | ||
0f420358 | 762 | if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ)) |
1da177e4 LT |
763 | goto fail; |
764 | rdev->sb_loaded = 1; | |
765 | return 0; | |
766 | ||
767 | fail: | |
768 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
769 | bdevname(rdev->bdev,b)); | |
770 | return -EINVAL; | |
771 | } | |
772 | ||
773 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
774 | { | |
05710466 AN |
775 | return sb1->set_uuid0 == sb2->set_uuid0 && |
776 | sb1->set_uuid1 == sb2->set_uuid1 && | |
777 | sb1->set_uuid2 == sb2->set_uuid2 && | |
778 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
779 | } |
780 | ||
1da177e4 LT |
781 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
782 | { | |
783 | int ret; | |
784 | mdp_super_t *tmp1, *tmp2; | |
785 | ||
786 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
787 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
788 | ||
789 | if (!tmp1 || !tmp2) { | |
790 | ret = 0; | |
35020f1a | 791 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
792 | goto abort; |
793 | } | |
794 | ||
795 | *tmp1 = *sb1; | |
796 | *tmp2 = *sb2; | |
797 | ||
798 | /* | |
799 | * nr_disks is not constant | |
800 | */ | |
801 | tmp1->nr_disks = 0; | |
802 | tmp2->nr_disks = 0; | |
803 | ||
ce0c8e05 | 804 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 805 | abort: |
990a8baf JJ |
806 | kfree(tmp1); |
807 | kfree(tmp2); | |
1da177e4 LT |
808 | return ret; |
809 | } | |
810 | ||
4d167f09 N |
811 | |
812 | static u32 md_csum_fold(u32 csum) | |
813 | { | |
814 | csum = (csum & 0xffff) + (csum >> 16); | |
815 | return (csum & 0xffff) + (csum >> 16); | |
816 | } | |
817 | ||
1da177e4 LT |
818 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
819 | { | |
4d167f09 N |
820 | u64 newcsum = 0; |
821 | u32 *sb32 = (u32*)sb; | |
822 | int i; | |
1da177e4 LT |
823 | unsigned int disk_csum, csum; |
824 | ||
825 | disk_csum = sb->sb_csum; | |
826 | sb->sb_csum = 0; | |
4d167f09 N |
827 | |
828 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
829 | newcsum += sb32[i]; | |
830 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
831 | ||
832 | ||
833 | #ifdef CONFIG_ALPHA | |
834 | /* This used to use csum_partial, which was wrong for several | |
835 | * reasons including that different results are returned on | |
836 | * different architectures. It isn't critical that we get exactly | |
837 | * the same return value as before (we always csum_fold before | |
838 | * testing, and that removes any differences). However as we | |
839 | * know that csum_partial always returned a 16bit value on | |
840 | * alphas, do a fold to maximise conformity to previous behaviour. | |
841 | */ | |
842 | sb->sb_csum = md_csum_fold(disk_csum); | |
843 | #else | |
1da177e4 | 844 | sb->sb_csum = disk_csum; |
4d167f09 | 845 | #endif |
1da177e4 LT |
846 | return csum; |
847 | } | |
848 | ||
849 | ||
850 | /* | |
851 | * Handle superblock details. | |
852 | * We want to be able to handle multiple superblock formats | |
853 | * so we have a common interface to them all, and an array of | |
854 | * different handlers. | |
855 | * We rely on user-space to write the initial superblock, and support | |
856 | * reading and updating of superblocks. | |
857 | * Interface methods are: | |
858 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
859 | * loads and validates a superblock on dev. | |
860 | * if refdev != NULL, compare superblocks on both devices | |
861 | * Return: | |
862 | * 0 - dev has a superblock that is compatible with refdev | |
863 | * 1 - dev has a superblock that is compatible and newer than refdev | |
864 | * so dev should be used as the refdev in future | |
865 | * -EINVAL superblock incompatible or invalid | |
866 | * -othererror e.g. -EIO | |
867 | * | |
868 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
869 | * Verify that dev is acceptable into mddev. | |
870 | * The first time, mddev->raid_disks will be 0, and data from | |
871 | * dev should be merged in. Subsequent calls check that dev | |
872 | * is new enough. Return 0 or -EINVAL | |
873 | * | |
874 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
875 | * Update the superblock for rdev with data in mddev | |
876 | * This does not write to disc. | |
877 | * | |
878 | */ | |
879 | ||
880 | struct super_type { | |
0cd17fec CW |
881 | char *name; |
882 | struct module *owner; | |
883 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
884 | int minor_version); | |
885 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
886 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
887 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 888 | sector_t num_sectors); |
1da177e4 LT |
889 | }; |
890 | ||
0894cc30 AN |
891 | /* |
892 | * Check that the given mddev has no bitmap. | |
893 | * | |
894 | * This function is called from the run method of all personalities that do not | |
895 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
896 | * has a bitmap. Otherwise, it returns 0. | |
897 | * | |
898 | */ | |
899 | int md_check_no_bitmap(mddev_t *mddev) | |
900 | { | |
c3d9714e | 901 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
902 | return 0; |
903 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
904 | mdname(mddev), mddev->pers->name); | |
905 | return 1; | |
906 | } | |
907 | EXPORT_SYMBOL(md_check_no_bitmap); | |
908 | ||
1da177e4 LT |
909 | /* |
910 | * load_super for 0.90.0 | |
911 | */ | |
912 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
913 | { | |
914 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
915 | mdp_super_t *sb; | |
916 | int ret; | |
1da177e4 LT |
917 | |
918 | /* | |
0f420358 | 919 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
920 | * it's at the end of the disk. |
921 | * | |
922 | * It also happens to be a multiple of 4Kb. | |
923 | */ | |
0f420358 | 924 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 925 | |
0002b271 | 926 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
927 | if (ret) return ret; |
928 | ||
929 | ret = -EINVAL; | |
930 | ||
931 | bdevname(rdev->bdev, b); | |
932 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
933 | ||
934 | if (sb->md_magic != MD_SB_MAGIC) { | |
935 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
936 | b); | |
937 | goto abort; | |
938 | } | |
939 | ||
940 | if (sb->major_version != 0 || | |
f6705578 N |
941 | sb->minor_version < 90 || |
942 | sb->minor_version > 91) { | |
1da177e4 LT |
943 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
944 | sb->major_version, sb->minor_version, | |
945 | b); | |
946 | goto abort; | |
947 | } | |
948 | ||
949 | if (sb->raid_disks <= 0) | |
950 | goto abort; | |
951 | ||
4d167f09 | 952 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
953 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
954 | b); | |
955 | goto abort; | |
956 | } | |
957 | ||
958 | rdev->preferred_minor = sb->md_minor; | |
959 | rdev->data_offset = 0; | |
0002b271 | 960 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
961 | |
962 | if (sb->level == LEVEL_MULTIPATH) | |
963 | rdev->desc_nr = -1; | |
964 | else | |
965 | rdev->desc_nr = sb->this_disk.number; | |
966 | ||
9a7b2b0f | 967 | if (!refdev) { |
1da177e4 | 968 | ret = 1; |
9a7b2b0f | 969 | } else { |
1da177e4 LT |
970 | __u64 ev1, ev2; |
971 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
972 | if (!uuid_equal(refsb, sb)) { | |
973 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
974 | b, bdevname(refdev->bdev,b2)); | |
975 | goto abort; | |
976 | } | |
977 | if (!sb_equal(refsb, sb)) { | |
978 | printk(KERN_WARNING "md: %s has same UUID" | |
979 | " but different superblock to %s\n", | |
980 | b, bdevname(refdev->bdev, b2)); | |
981 | goto abort; | |
982 | } | |
983 | ev1 = md_event(sb); | |
984 | ev2 = md_event(refsb); | |
985 | if (ev1 > ev2) | |
986 | ret = 1; | |
987 | else | |
988 | ret = 0; | |
989 | } | |
8190e754 | 990 | rdev->sectors = rdev->sb_start; |
1da177e4 | 991 | |
dd8ac336 | 992 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
993 | /* "this cannot possibly happen" ... */ |
994 | ret = -EINVAL; | |
995 | ||
1da177e4 LT |
996 | abort: |
997 | return ret; | |
998 | } | |
999 | ||
1000 | /* | |
1001 | * validate_super for 0.90.0 | |
1002 | */ | |
1003 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1004 | { | |
1005 | mdp_disk_t *desc; | |
1006 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 1007 | __u64 ev1 = md_event(sb); |
1da177e4 | 1008 | |
41158c7e | 1009 | rdev->raid_disk = -1; |
c5d79adb N |
1010 | clear_bit(Faulty, &rdev->flags); |
1011 | clear_bit(In_sync, &rdev->flags); | |
1012 | clear_bit(WriteMostly, &rdev->flags); | |
1013 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1014 | ||
1da177e4 LT |
1015 | if (mddev->raid_disks == 0) { |
1016 | mddev->major_version = 0; | |
1017 | mddev->minor_version = sb->minor_version; | |
1018 | mddev->patch_version = sb->patch_version; | |
e691063a | 1019 | mddev->external = 0; |
9d8f0363 | 1020 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1021 | mddev->ctime = sb->ctime; |
1022 | mddev->utime = sb->utime; | |
1023 | mddev->level = sb->level; | |
d9d166c2 | 1024 | mddev->clevel[0] = 0; |
1da177e4 LT |
1025 | mddev->layout = sb->layout; |
1026 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 1027 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 1028 | mddev->events = ev1; |
c3d9714e N |
1029 | mddev->bitmap_info.offset = 0; |
1030 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; | |
1da177e4 | 1031 | |
f6705578 N |
1032 | if (mddev->minor_version >= 91) { |
1033 | mddev->reshape_position = sb->reshape_position; | |
1034 | mddev->delta_disks = sb->delta_disks; | |
1035 | mddev->new_level = sb->new_level; | |
1036 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1037 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
1038 | } else { |
1039 | mddev->reshape_position = MaxSector; | |
1040 | mddev->delta_disks = 0; | |
1041 | mddev->new_level = mddev->level; | |
1042 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1043 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1044 | } |
1045 | ||
1da177e4 LT |
1046 | if (sb->state & (1<<MD_SB_CLEAN)) |
1047 | mddev->recovery_cp = MaxSector; | |
1048 | else { | |
1049 | if (sb->events_hi == sb->cp_events_hi && | |
1050 | sb->events_lo == sb->cp_events_lo) { | |
1051 | mddev->recovery_cp = sb->recovery_cp; | |
1052 | } else | |
1053 | mddev->recovery_cp = 0; | |
1054 | } | |
1055 | ||
1056 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1057 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1058 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1059 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1060 | ||
1061 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1062 | |
1063 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
c3d9714e N |
1064 | mddev->bitmap_info.file == NULL) |
1065 | mddev->bitmap_info.offset = | |
1066 | mddev->bitmap_info.default_offset; | |
a654b9d8 | 1067 | |
41158c7e N |
1068 | } else if (mddev->pers == NULL) { |
1069 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
1070 | ++ev1; |
1071 | if (ev1 < mddev->events) | |
1072 | return -EINVAL; | |
41158c7e N |
1073 | } else if (mddev->bitmap) { |
1074 | /* if adding to array with a bitmap, then we can accept an | |
1075 | * older device ... but not too old. | |
1076 | */ | |
41158c7e N |
1077 | if (ev1 < mddev->bitmap->events_cleared) |
1078 | return 0; | |
07d84d10 N |
1079 | } else { |
1080 | if (ev1 < mddev->events) | |
1081 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1082 | return 0; | |
1083 | } | |
41158c7e | 1084 | |
1da177e4 | 1085 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1086 | desc = sb->disks + rdev->desc_nr; |
1087 | ||
1088 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1089 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1090 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1091 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1092 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1093 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
1094 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1095 | /* active but not in sync implies recovery up to | |
1096 | * reshape position. We don't know exactly where | |
1097 | * that is, so set to zero for now */ | |
1098 | if (mddev->minor_version >= 91) { | |
1099 | rdev->recovery_offset = 0; | |
1100 | rdev->raid_disk = desc->raid_disk; | |
1101 | } | |
1da177e4 | 1102 | } |
8ddf9efe N |
1103 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1104 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1105 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1106 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1107 | return 0; |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * sync_super for 0.90.0 | |
1112 | */ | |
1113 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1114 | { | |
1115 | mdp_super_t *sb; | |
1da177e4 LT |
1116 | mdk_rdev_t *rdev2; |
1117 | int next_spare = mddev->raid_disks; | |
19133a42 | 1118 | |
1da177e4 LT |
1119 | |
1120 | /* make rdev->sb match mddev data.. | |
1121 | * | |
1122 | * 1/ zero out disks | |
1123 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1124 | * 3/ any empty disks < next_spare become removed | |
1125 | * | |
1126 | * disks[0] gets initialised to REMOVED because | |
1127 | * we cannot be sure from other fields if it has | |
1128 | * been initialised or not. | |
1129 | */ | |
1130 | int i; | |
1131 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1132 | ||
61181565 N |
1133 | rdev->sb_size = MD_SB_BYTES; |
1134 | ||
1da177e4 LT |
1135 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
1136 | ||
1137 | memset(sb, 0, sizeof(*sb)); | |
1138 | ||
1139 | sb->md_magic = MD_SB_MAGIC; | |
1140 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1141 | sb->patch_version = mddev->patch_version; |
1142 | sb->gvalid_words = 0; /* ignored */ | |
1143 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1144 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1145 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1146 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1147 | ||
1148 | sb->ctime = mddev->ctime; | |
1149 | sb->level = mddev->level; | |
58c0fed4 | 1150 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1151 | sb->raid_disks = mddev->raid_disks; |
1152 | sb->md_minor = mddev->md_minor; | |
e691063a | 1153 | sb->not_persistent = 0; |
1da177e4 LT |
1154 | sb->utime = mddev->utime; |
1155 | sb->state = 0; | |
1156 | sb->events_hi = (mddev->events>>32); | |
1157 | sb->events_lo = (u32)mddev->events; | |
1158 | ||
f6705578 N |
1159 | if (mddev->reshape_position == MaxSector) |
1160 | sb->minor_version = 90; | |
1161 | else { | |
1162 | sb->minor_version = 91; | |
1163 | sb->reshape_position = mddev->reshape_position; | |
1164 | sb->new_level = mddev->new_level; | |
1165 | sb->delta_disks = mddev->delta_disks; | |
1166 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1167 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1168 | } |
1169 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1170 | if (mddev->in_sync) |
1171 | { | |
1172 | sb->recovery_cp = mddev->recovery_cp; | |
1173 | sb->cp_events_hi = (mddev->events>>32); | |
1174 | sb->cp_events_lo = (u32)mddev->events; | |
1175 | if (mddev->recovery_cp == MaxSector) | |
1176 | sb->state = (1<< MD_SB_CLEAN); | |
1177 | } else | |
1178 | sb->recovery_cp = 0; | |
1179 | ||
1180 | sb->layout = mddev->layout; | |
9d8f0363 | 1181 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1182 | |
c3d9714e | 1183 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1184 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1185 | ||
1da177e4 | 1186 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1187 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1188 | mdp_disk_t *d; |
86e6ffdd | 1189 | int desc_nr; |
0261cd9f N |
1190 | int is_active = test_bit(In_sync, &rdev2->flags); |
1191 | ||
1192 | if (rdev2->raid_disk >= 0 && | |
1193 | sb->minor_version >= 91) | |
1194 | /* we have nowhere to store the recovery_offset, | |
1195 | * but if it is not below the reshape_position, | |
1196 | * we can piggy-back on that. | |
1197 | */ | |
1198 | is_active = 1; | |
1199 | if (rdev2->raid_disk < 0 || | |
1200 | test_bit(Faulty, &rdev2->flags)) | |
1201 | is_active = 0; | |
1202 | if (is_active) | |
86e6ffdd | 1203 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1204 | else |
86e6ffdd | 1205 | desc_nr = next_spare++; |
19133a42 | 1206 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1207 | d = &sb->disks[rdev2->desc_nr]; |
1208 | nr_disks++; | |
1209 | d->number = rdev2->desc_nr; | |
1210 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1211 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1212 | if (is_active) |
1da177e4 LT |
1213 | d->raid_disk = rdev2->raid_disk; |
1214 | else | |
1215 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1216 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1217 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1218 | else if (is_active) { |
1da177e4 | 1219 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1220 | if (test_bit(In_sync, &rdev2->flags)) |
1221 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1222 | active++; |
1223 | working++; | |
1224 | } else { | |
1225 | d->state = 0; | |
1226 | spare++; | |
1227 | working++; | |
1228 | } | |
8ddf9efe N |
1229 | if (test_bit(WriteMostly, &rdev2->flags)) |
1230 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1231 | } |
1da177e4 LT |
1232 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1233 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1234 | mdp_disk_t *d = &sb->disks[i]; | |
1235 | if (d->state == 0 && d->number == 0) { | |
1236 | d->number = i; | |
1237 | d->raid_disk = i; | |
1238 | d->state = (1<<MD_DISK_REMOVED); | |
1239 | d->state |= (1<<MD_DISK_FAULTY); | |
1240 | failed++; | |
1241 | } | |
1242 | } | |
1243 | sb->nr_disks = nr_disks; | |
1244 | sb->active_disks = active; | |
1245 | sb->working_disks = working; | |
1246 | sb->failed_disks = failed; | |
1247 | sb->spare_disks = spare; | |
1248 | ||
1249 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1250 | sb->sb_csum = calc_sb_csum(sb); | |
1251 | } | |
1252 | ||
0cd17fec CW |
1253 | /* |
1254 | * rdev_size_change for 0.90.0 | |
1255 | */ | |
1256 | static unsigned long long | |
15f4a5fd | 1257 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1258 | { |
58c0fed4 | 1259 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1260 | return 0; /* component must fit device */ |
c3d9714e | 1261 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1262 | return 0; /* can't move bitmap */ |
0f420358 | 1263 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
15f4a5fd AN |
1264 | if (!num_sectors || num_sectors > rdev->sb_start) |
1265 | num_sectors = rdev->sb_start; | |
0f420358 | 1266 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1267 | rdev->sb_page); |
1268 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1269 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec CW |
1270 | } |
1271 | ||
1272 | ||
1da177e4 LT |
1273 | /* |
1274 | * version 1 superblock | |
1275 | */ | |
1276 | ||
1c05b4bc | 1277 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1278 | { |
1c05b4bc N |
1279 | __le32 disk_csum; |
1280 | u32 csum; | |
1da177e4 LT |
1281 | unsigned long long newcsum; |
1282 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1283 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1284 | int i; |
1285 | ||
1286 | disk_csum = sb->sb_csum; | |
1287 | sb->sb_csum = 0; | |
1288 | newcsum = 0; | |
1289 | for (i=0; size>=4; size -= 4 ) | |
1290 | newcsum += le32_to_cpu(*isuper++); | |
1291 | ||
1292 | if (size == 2) | |
1c05b4bc | 1293 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1294 | |
1295 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1296 | sb->sb_csum = disk_csum; | |
1297 | return cpu_to_le32(csum); | |
1298 | } | |
1299 | ||
1300 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1301 | { | |
1302 | struct mdp_superblock_1 *sb; | |
1303 | int ret; | |
0f420358 | 1304 | sector_t sb_start; |
1da177e4 | 1305 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1306 | int bmask; |
1da177e4 LT |
1307 | |
1308 | /* | |
0f420358 | 1309 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1310 | * It is always aligned to a 4K boundary and |
1311 | * depeding on minor_version, it can be: | |
1312 | * 0: At least 8K, but less than 12K, from end of device | |
1313 | * 1: At start of device | |
1314 | * 2: 4K from start of device. | |
1315 | */ | |
1316 | switch(minor_version) { | |
1317 | case 0: | |
0f420358 AN |
1318 | sb_start = rdev->bdev->bd_inode->i_size >> 9; |
1319 | sb_start -= 8*2; | |
1320 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1321 | break; |
1322 | case 1: | |
0f420358 | 1323 | sb_start = 0; |
1da177e4 LT |
1324 | break; |
1325 | case 2: | |
0f420358 | 1326 | sb_start = 8; |
1da177e4 LT |
1327 | break; |
1328 | default: | |
1329 | return -EINVAL; | |
1330 | } | |
0f420358 | 1331 | rdev->sb_start = sb_start; |
1da177e4 | 1332 | |
0002b271 N |
1333 | /* superblock is rarely larger than 1K, but it can be larger, |
1334 | * and it is safe to read 4k, so we do that | |
1335 | */ | |
1336 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1337 | if (ret) return ret; |
1338 | ||
1339 | ||
1340 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1341 | ||
1342 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1343 | sb->major_version != cpu_to_le32(1) || | |
1344 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1345 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1346 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1347 | return -EINVAL; |
1348 | ||
1349 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1350 | printk("md: invalid superblock checksum on %s\n", | |
1351 | bdevname(rdev->bdev,b)); | |
1352 | return -EINVAL; | |
1353 | } | |
1354 | if (le64_to_cpu(sb->data_size) < 10) { | |
1355 | printk("md: data_size too small on %s\n", | |
1356 | bdevname(rdev->bdev,b)); | |
1357 | return -EINVAL; | |
1358 | } | |
e11e93fa | 1359 | |
1da177e4 LT |
1360 | rdev->preferred_minor = 0xffff; |
1361 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1362 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1363 | |
0002b271 | 1364 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1365 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1366 | if (rdev->sb_size & bmask) |
a1801f85 N |
1367 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1368 | ||
1369 | if (minor_version | |
0f420358 | 1370 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1371 | return -EINVAL; |
0002b271 | 1372 | |
31b65a0d N |
1373 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1374 | rdev->desc_nr = -1; | |
1375 | else | |
1376 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1377 | ||
9a7b2b0f | 1378 | if (!refdev) { |
8ed75463 | 1379 | ret = 1; |
9a7b2b0f | 1380 | } else { |
1da177e4 LT |
1381 | __u64 ev1, ev2; |
1382 | struct mdp_superblock_1 *refsb = | |
1383 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1384 | ||
1385 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1386 | sb->level != refsb->level || | |
1387 | sb->layout != refsb->layout || | |
1388 | sb->chunksize != refsb->chunksize) { | |
1389 | printk(KERN_WARNING "md: %s has strangely different" | |
1390 | " superblock to %s\n", | |
1391 | bdevname(rdev->bdev,b), | |
1392 | bdevname(refdev->bdev,b2)); | |
1393 | return -EINVAL; | |
1394 | } | |
1395 | ev1 = le64_to_cpu(sb->events); | |
1396 | ev2 = le64_to_cpu(refsb->events); | |
1397 | ||
1398 | if (ev1 > ev2) | |
8ed75463 N |
1399 | ret = 1; |
1400 | else | |
1401 | ret = 0; | |
1da177e4 | 1402 | } |
a1801f85 | 1403 | if (minor_version) |
dd8ac336 AN |
1404 | rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) - |
1405 | le64_to_cpu(sb->data_offset); | |
1da177e4 | 1406 | else |
dd8ac336 AN |
1407 | rdev->sectors = rdev->sb_start; |
1408 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1409 | return -EINVAL; |
dd8ac336 | 1410 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1411 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1412 | return -EINVAL; |
8ed75463 | 1413 | return ret; |
1da177e4 LT |
1414 | } |
1415 | ||
1416 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1417 | { | |
1418 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1419 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1420 | |
41158c7e | 1421 | rdev->raid_disk = -1; |
c5d79adb N |
1422 | clear_bit(Faulty, &rdev->flags); |
1423 | clear_bit(In_sync, &rdev->flags); | |
1424 | clear_bit(WriteMostly, &rdev->flags); | |
1425 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1426 | ||
1da177e4 LT |
1427 | if (mddev->raid_disks == 0) { |
1428 | mddev->major_version = 1; | |
1429 | mddev->patch_version = 0; | |
e691063a | 1430 | mddev->external = 0; |
9d8f0363 | 1431 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1432 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1433 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1434 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1435 | mddev->clevel[0] = 0; |
1da177e4 LT |
1436 | mddev->layout = le32_to_cpu(sb->layout); |
1437 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1438 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1439 | mddev->events = ev1; |
c3d9714e N |
1440 | mddev->bitmap_info.offset = 0; |
1441 | mddev->bitmap_info.default_offset = 1024 >> 9; | |
1da177e4 LT |
1442 | |
1443 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1444 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1445 | ||
1446 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1447 | |
71c0805c | 1448 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
c3d9714e N |
1449 | mddev->bitmap_info.file == NULL ) |
1450 | mddev->bitmap_info.offset = | |
1451 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
e11e93fa | 1452 | |
f6705578 N |
1453 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1454 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1455 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1456 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1457 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1458 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1459 | } else { |
1460 | mddev->reshape_position = MaxSector; | |
1461 | mddev->delta_disks = 0; | |
1462 | mddev->new_level = mddev->level; | |
1463 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1464 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1465 | } |
1466 | ||
41158c7e N |
1467 | } else if (mddev->pers == NULL) { |
1468 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1469 | ++ev1; |
1470 | if (ev1 < mddev->events) | |
1471 | return -EINVAL; | |
41158c7e N |
1472 | } else if (mddev->bitmap) { |
1473 | /* If adding to array with a bitmap, then we can accept an | |
1474 | * older device, but not too old. | |
1475 | */ | |
41158c7e N |
1476 | if (ev1 < mddev->bitmap->events_cleared) |
1477 | return 0; | |
07d84d10 N |
1478 | } else { |
1479 | if (ev1 < mddev->events) | |
1480 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1481 | return 0; | |
1482 | } | |
1da177e4 LT |
1483 | if (mddev->level != LEVEL_MULTIPATH) { |
1484 | int role; | |
3673f305 N |
1485 | if (rdev->desc_nr < 0 || |
1486 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1487 | role = 0xffff; | |
1488 | rdev->desc_nr = -1; | |
1489 | } else | |
1490 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1491 | switch(role) { |
1492 | case 0xffff: /* spare */ | |
1da177e4 LT |
1493 | break; |
1494 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1495 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1496 | break; |
1497 | default: | |
5fd6c1dc N |
1498 | if ((le32_to_cpu(sb->feature_map) & |
1499 | MD_FEATURE_RECOVERY_OFFSET)) | |
1500 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1501 | else | |
1502 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1503 | rdev->raid_disk = role; |
1504 | break; | |
1505 | } | |
8ddf9efe N |
1506 | if (sb->devflags & WriteMostly1) |
1507 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1508 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1509 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1510 | |
1da177e4 LT |
1511 | return 0; |
1512 | } | |
1513 | ||
1514 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1515 | { | |
1516 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1517 | mdk_rdev_t *rdev2; |
1518 | int max_dev, i; | |
1519 | /* make rdev->sb match mddev and rdev data. */ | |
1520 | ||
1521 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1522 | ||
1523 | sb->feature_map = 0; | |
1524 | sb->pad0 = 0; | |
5fd6c1dc | 1525 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1526 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1527 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1528 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1529 | ||
1530 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1531 | sb->events = cpu_to_le64(mddev->events); | |
1532 | if (mddev->in_sync) | |
1533 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1534 | else | |
1535 | sb->resync_offset = cpu_to_le64(0); | |
1536 | ||
1c05b4bc | 1537 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1538 | |
f0ca340c | 1539 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1540 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1541 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1542 | sb->level = cpu_to_le32(mddev->level); |
1543 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1544 | |
c3d9714e N |
1545 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1546 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1547 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1548 | } |
5fd6c1dc N |
1549 | |
1550 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1551 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1552 | sb->feature_map |= |
1553 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1554 | sb->recovery_offset = | |
1555 | cpu_to_le64(rdev->recovery_offset); | |
5fd6c1dc N |
1556 | } |
1557 | ||
f6705578 N |
1558 | if (mddev->reshape_position != MaxSector) { |
1559 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1560 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1561 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1562 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1563 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1564 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1565 | } |
a654b9d8 | 1566 | |
1da177e4 | 1567 | max_dev = 0; |
159ec1fc | 1568 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1569 | if (rdev2->desc_nr+1 > max_dev) |
1570 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1571 | |
70471daf N |
1572 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1573 | int bmask; | |
a778b73f | 1574 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1575 | rdev->sb_size = max_dev * 2 + 256; |
1576 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1577 | if (rdev->sb_size & bmask) | |
1578 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
1579 | } | |
1da177e4 LT |
1580 | for (i=0; i<max_dev;i++) |
1581 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1582 | ||
159ec1fc | 1583 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1584 | i = rdev2->desc_nr; |
b2d444d7 | 1585 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1586 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1587 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1588 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1589 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1590 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1591 | else |
1592 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1593 | } | |
1594 | ||
1da177e4 LT |
1595 | sb->sb_csum = calc_sb_1_csum(sb); |
1596 | } | |
1597 | ||
0cd17fec | 1598 | static unsigned long long |
15f4a5fd | 1599 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1600 | { |
1601 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1602 | sector_t max_sectors; |
58c0fed4 | 1603 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1604 | return 0; /* component must fit device */ |
0f420358 | 1605 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1606 | /* minor versions 1 and 2; superblock before data */ |
15f4a5fd AN |
1607 | max_sectors = rdev->bdev->bd_inode->i_size >> 9; |
1608 | max_sectors -= rdev->data_offset; | |
1609 | if (!num_sectors || num_sectors > max_sectors) | |
1610 | num_sectors = max_sectors; | |
c3d9714e | 1611 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1612 | /* minor version 0 with bitmap we can't move */ |
1613 | return 0; | |
1614 | } else { | |
1615 | /* minor version 0; superblock after data */ | |
0f420358 AN |
1616 | sector_t sb_start; |
1617 | sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; | |
1618 | sb_start &= ~(sector_t)(4*2 - 1); | |
dd8ac336 | 1619 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1620 | if (!num_sectors || num_sectors > max_sectors) |
1621 | num_sectors = max_sectors; | |
0f420358 | 1622 | rdev->sb_start = sb_start; |
0cd17fec CW |
1623 | } |
1624 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1625 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1626 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1627 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1628 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1629 | rdev->sb_page); |
1630 | md_super_wait(rdev->mddev); | |
15f4a5fd | 1631 | return num_sectors / 2; /* kB for sysfs */ |
0cd17fec | 1632 | } |
1da177e4 | 1633 | |
75c96f85 | 1634 | static struct super_type super_types[] = { |
1da177e4 LT |
1635 | [0] = { |
1636 | .name = "0.90.0", | |
1637 | .owner = THIS_MODULE, | |
0cd17fec CW |
1638 | .load_super = super_90_load, |
1639 | .validate_super = super_90_validate, | |
1640 | .sync_super = super_90_sync, | |
1641 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1642 | }, |
1643 | [1] = { | |
1644 | .name = "md-1", | |
1645 | .owner = THIS_MODULE, | |
0cd17fec CW |
1646 | .load_super = super_1_load, |
1647 | .validate_super = super_1_validate, | |
1648 | .sync_super = super_1_sync, | |
1649 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1650 | }, |
1651 | }; | |
1da177e4 LT |
1652 | |
1653 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1654 | { | |
7dd5e7c3 | 1655 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1656 | |
4b80991c N |
1657 | rcu_read_lock(); |
1658 | rdev_for_each_rcu(rdev, mddev1) | |
1659 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1660 | if (rdev->bdev->bd_contains == |
4b80991c N |
1661 | rdev2->bdev->bd_contains) { |
1662 | rcu_read_unlock(); | |
7dd5e7c3 | 1663 | return 1; |
4b80991c N |
1664 | } |
1665 | rcu_read_unlock(); | |
1da177e4 LT |
1666 | return 0; |
1667 | } | |
1668 | ||
1669 | static LIST_HEAD(pending_raid_disks); | |
1670 | ||
ac5e7113 AN |
1671 | /* |
1672 | * Try to register data integrity profile for an mddev | |
1673 | * | |
1674 | * This is called when an array is started and after a disk has been kicked | |
1675 | * from the array. It only succeeds if all working and active component devices | |
1676 | * are integrity capable with matching profiles. | |
1677 | */ | |
1678 | int md_integrity_register(mddev_t *mddev) | |
1679 | { | |
1680 | mdk_rdev_t *rdev, *reference = NULL; | |
1681 | ||
1682 | if (list_empty(&mddev->disks)) | |
1683 | return 0; /* nothing to do */ | |
1684 | if (blk_get_integrity(mddev->gendisk)) | |
1685 | return 0; /* already registered */ | |
1686 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
1687 | /* skip spares and non-functional disks */ | |
1688 | if (test_bit(Faulty, &rdev->flags)) | |
1689 | continue; | |
1690 | if (rdev->raid_disk < 0) | |
1691 | continue; | |
1692 | /* | |
1693 | * If at least one rdev is not integrity capable, we can not | |
1694 | * enable data integrity for the md device. | |
1695 | */ | |
1696 | if (!bdev_get_integrity(rdev->bdev)) | |
1697 | return -EINVAL; | |
1698 | if (!reference) { | |
1699 | /* Use the first rdev as the reference */ | |
1700 | reference = rdev; | |
1701 | continue; | |
1702 | } | |
1703 | /* does this rdev's profile match the reference profile? */ | |
1704 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1705 | rdev->bdev->bd_disk) < 0) | |
1706 | return -EINVAL; | |
1707 | } | |
1708 | /* | |
1709 | * All component devices are integrity capable and have matching | |
1710 | * profiles, register the common profile for the md device. | |
1711 | */ | |
1712 | if (blk_integrity_register(mddev->gendisk, | |
1713 | bdev_get_integrity(reference->bdev)) != 0) { | |
1714 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1715 | mdname(mddev)); | |
1716 | return -EINVAL; | |
1717 | } | |
1718 | printk(KERN_NOTICE "md: data integrity on %s enabled\n", | |
1719 | mdname(mddev)); | |
1720 | return 0; | |
1721 | } | |
1722 | EXPORT_SYMBOL(md_integrity_register); | |
1723 | ||
1724 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
1725 | void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) | |
3f9d99c1 | 1726 | { |
3f9d99c1 | 1727 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); |
ac5e7113 | 1728 | struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); |
3f9d99c1 | 1729 | |
ac5e7113 | 1730 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1731 | return; |
ac5e7113 | 1732 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1733 | return; |
ac5e7113 AN |
1734 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1735 | rdev->bdev->bd_disk) >= 0) | |
1736 | return; | |
1737 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1738 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1739 | } |
ac5e7113 | 1740 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1741 | |
1da177e4 LT |
1742 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1743 | { | |
7dd5e7c3 | 1744 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1745 | struct kobject *ko; |
1edf80d3 | 1746 | char *s; |
5e55e2f5 | 1747 | int err; |
1da177e4 LT |
1748 | |
1749 | if (rdev->mddev) { | |
1750 | MD_BUG(); | |
1751 | return -EINVAL; | |
1752 | } | |
11e2ede0 DW |
1753 | |
1754 | /* prevent duplicates */ | |
1755 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1756 | return -EEXIST; | |
1757 | ||
dd8ac336 AN |
1758 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1759 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1760 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1761 | if (mddev->pers) { |
1762 | /* Cannot change size, so fail | |
1763 | * If mddev->level <= 0, then we don't care | |
1764 | * about aligning sizes (e.g. linear) | |
1765 | */ | |
1766 | if (mddev->level > 0) | |
1767 | return -ENOSPC; | |
1768 | } else | |
dd8ac336 | 1769 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1770 | } |
1da177e4 LT |
1771 | |
1772 | /* Verify rdev->desc_nr is unique. | |
1773 | * If it is -1, assign a free number, else | |
1774 | * check number is not in use | |
1775 | */ | |
1776 | if (rdev->desc_nr < 0) { | |
1777 | int choice = 0; | |
1778 | if (mddev->pers) choice = mddev->raid_disks; | |
1779 | while (find_rdev_nr(mddev, choice)) | |
1780 | choice++; | |
1781 | rdev->desc_nr = choice; | |
1782 | } else { | |
1783 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1784 | return -EBUSY; | |
1785 | } | |
de01dfad N |
1786 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1787 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1788 | mdname(mddev), mddev->max_disks); | |
1789 | return -EBUSY; | |
1790 | } | |
19133a42 | 1791 | bdevname(rdev->bdev,b); |
649316b2 | 1792 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1793 | *s = '!'; |
649316b2 | 1794 | |
1da177e4 | 1795 | rdev->mddev = mddev; |
19133a42 | 1796 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1797 | |
b2d6db58 | 1798 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1799 | goto fail; |
86e6ffdd | 1800 | |
0762b8bd | 1801 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
5e55e2f5 N |
1802 | if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) { |
1803 | kobject_del(&rdev->kobj); | |
1804 | goto fail; | |
1805 | } | |
3c0ee63a N |
1806 | rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state"); |
1807 | ||
4b80991c | 1808 | list_add_rcu(&rdev->same_set, &mddev->disks); |
c5d79adb | 1809 | bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); |
4044ba58 N |
1810 | |
1811 | /* May as well allow recovery to be retried once */ | |
1812 | mddev->recovery_disabled = 0; | |
3f9d99c1 | 1813 | |
1da177e4 | 1814 | return 0; |
5e55e2f5 N |
1815 | |
1816 | fail: | |
1817 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1818 | b, mdname(mddev)); | |
1819 | return err; | |
1da177e4 LT |
1820 | } |
1821 | ||
177a99b2 | 1822 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1823 | { |
1824 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1825 | kobject_del(&rdev->kobj); | |
177a99b2 | 1826 | kobject_put(&rdev->kobj); |
5792a285 N |
1827 | } |
1828 | ||
1da177e4 LT |
1829 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1830 | { | |
1831 | char b[BDEVNAME_SIZE]; | |
1832 | if (!rdev->mddev) { | |
1833 | MD_BUG(); | |
1834 | return; | |
1835 | } | |
5463c790 | 1836 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1837 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1838 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1839 | rdev->mddev = NULL; | |
86e6ffdd | 1840 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1841 | sysfs_put(rdev->sysfs_state); |
1842 | rdev->sysfs_state = NULL; | |
5792a285 | 1843 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1844 | * writing to 'remove' to "dev/state". We also need |
1845 | * to delay it due to rcu usage. | |
5792a285 | 1846 | */ |
4b80991c | 1847 | synchronize_rcu(); |
177a99b2 N |
1848 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1849 | kobject_get(&rdev->kobj); | |
5792a285 | 1850 | schedule_work(&rdev->del_work); |
1da177e4 LT |
1851 | } |
1852 | ||
1853 | /* | |
1854 | * prevent the device from being mounted, repartitioned or | |
1855 | * otherwise reused by a RAID array (or any other kernel | |
1856 | * subsystem), by bd_claiming the device. | |
1857 | */ | |
c5d79adb | 1858 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1859 | { |
1860 | int err = 0; | |
1861 | struct block_device *bdev; | |
1862 | char b[BDEVNAME_SIZE]; | |
1863 | ||
2e7b651d | 1864 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1865 | if (IS_ERR(bdev)) { |
1866 | printk(KERN_ERR "md: could not open %s.\n", | |
1867 | __bdevname(dev, b)); | |
1868 | return PTR_ERR(bdev); | |
1869 | } | |
c5d79adb | 1870 | err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); |
1da177e4 LT |
1871 | if (err) { |
1872 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1873 | bdevname(bdev, b)); | |
9a1c3542 | 1874 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1875 | return err; |
1876 | } | |
c5d79adb N |
1877 | if (!shared) |
1878 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1879 | rdev->bdev = bdev; |
1880 | return err; | |
1881 | } | |
1882 | ||
1883 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1884 | { | |
1885 | struct block_device *bdev = rdev->bdev; | |
1886 | rdev->bdev = NULL; | |
1887 | if (!bdev) | |
1888 | MD_BUG(); | |
1889 | bd_release(bdev); | |
9a1c3542 | 1890 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1891 | } |
1892 | ||
1893 | void md_autodetect_dev(dev_t dev); | |
1894 | ||
1895 | static void export_rdev(mdk_rdev_t * rdev) | |
1896 | { | |
1897 | char b[BDEVNAME_SIZE]; | |
1898 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1899 | bdevname(rdev->bdev,b)); | |
1900 | if (rdev->mddev) | |
1901 | MD_BUG(); | |
1902 | free_disk_sb(rdev); | |
1da177e4 | 1903 | #ifndef MODULE |
d0fae18f N |
1904 | if (test_bit(AutoDetected, &rdev->flags)) |
1905 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1906 | #endif |
1907 | unlock_rdev(rdev); | |
86e6ffdd | 1908 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1909 | } |
1910 | ||
1911 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1912 | { | |
1913 | unbind_rdev_from_array(rdev); | |
1914 | export_rdev(rdev); | |
1915 | } | |
1916 | ||
1917 | static void export_array(mddev_t *mddev) | |
1918 | { | |
159ec1fc | 1919 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 1920 | |
d089c6af | 1921 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1922 | if (!rdev->mddev) { |
1923 | MD_BUG(); | |
1924 | continue; | |
1925 | } | |
1926 | kick_rdev_from_array(rdev); | |
1927 | } | |
1928 | if (!list_empty(&mddev->disks)) | |
1929 | MD_BUG(); | |
1930 | mddev->raid_disks = 0; | |
1931 | mddev->major_version = 0; | |
1932 | } | |
1933 | ||
1934 | static void print_desc(mdp_disk_t *desc) | |
1935 | { | |
1936 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1937 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1938 | } | |
1939 | ||
cd2ac932 | 1940 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
1941 | { |
1942 | int i; | |
1943 | ||
1944 | printk(KERN_INFO | |
1945 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1946 | sb->major_version, sb->minor_version, sb->patch_version, | |
1947 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1948 | sb->ctime); | |
1949 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1950 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1951 | sb->md_minor, sb->layout, sb->chunk_size); | |
1952 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1953 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1954 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1955 | sb->failed_disks, sb->spare_disks, | |
1956 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1957 | ||
1958 | printk(KERN_INFO); | |
1959 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1960 | mdp_disk_t *desc; | |
1961 | ||
1962 | desc = sb->disks + i; | |
1963 | if (desc->number || desc->major || desc->minor || | |
1964 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1965 | printk(" D %2d: ", i); | |
1966 | print_desc(desc); | |
1967 | } | |
1968 | } | |
1969 | printk(KERN_INFO "md: THIS: "); | |
1970 | print_desc(&sb->this_disk); | |
cd2ac932 | 1971 | } |
1da177e4 | 1972 | |
cd2ac932 CR |
1973 | static void print_sb_1(struct mdp_superblock_1 *sb) |
1974 | { | |
1975 | __u8 *uuid; | |
1976 | ||
1977 | uuid = sb->set_uuid; | |
ad361c98 | 1978 | printk(KERN_INFO |
7b75c2f8 | 1979 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 1980 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
1981 | le32_to_cpu(sb->major_version), |
1982 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 1983 | uuid, |
cd2ac932 CR |
1984 | sb->set_name, |
1985 | (unsigned long long)le64_to_cpu(sb->ctime) | |
1986 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
1987 | ||
1988 | uuid = sb->device_uuid; | |
ad361c98 JP |
1989 | printk(KERN_INFO |
1990 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 1991 | " RO:%llu\n" |
7b75c2f8 | 1992 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
1993 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
1994 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
1995 | le32_to_cpu(sb->level), |
1996 | (unsigned long long)le64_to_cpu(sb->size), | |
1997 | le32_to_cpu(sb->raid_disks), | |
1998 | le32_to_cpu(sb->layout), | |
1999 | le32_to_cpu(sb->chunksize), | |
2000 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2001 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2002 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2003 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2004 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2005 | uuid, |
cd2ac932 CR |
2006 | sb->devflags, |
2007 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2008 | (unsigned long long)le64_to_cpu(sb->events), | |
2009 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2010 | le32_to_cpu(sb->sb_csum), | |
2011 | le32_to_cpu(sb->max_dev) | |
2012 | ); | |
1da177e4 LT |
2013 | } |
2014 | ||
cd2ac932 | 2015 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
2016 | { |
2017 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2018 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2019 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2020 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2021 | rdev->desc_nr); | |
1da177e4 | 2022 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2023 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2024 | switch (major_version) { | |
2025 | case 0: | |
2026 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
2027 | break; | |
2028 | case 1: | |
2029 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
2030 | break; | |
2031 | } | |
1da177e4 LT |
2032 | } else |
2033 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2034 | } | |
2035 | ||
5e56341d | 2036 | static void md_print_devices(void) |
1da177e4 | 2037 | { |
159ec1fc | 2038 | struct list_head *tmp; |
1da177e4 LT |
2039 | mdk_rdev_t *rdev; |
2040 | mddev_t *mddev; | |
2041 | char b[BDEVNAME_SIZE]; | |
2042 | ||
2043 | printk("\n"); | |
2044 | printk("md: **********************************\n"); | |
2045 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2046 | printk("md: **********************************\n"); | |
29ac4aa3 | 2047 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2048 | |
32a7627c N |
2049 | if (mddev->bitmap) |
2050 | bitmap_print_sb(mddev->bitmap); | |
2051 | else | |
2052 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 2053 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
2054 | printk("<%s>", bdevname(rdev->bdev,b)); |
2055 | printk("\n"); | |
2056 | ||
159ec1fc | 2057 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 2058 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2059 | } |
2060 | printk("md: **********************************\n"); | |
2061 | printk("\n"); | |
2062 | } | |
2063 | ||
2064 | ||
42543769 | 2065 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 2066 | { |
42543769 N |
2067 | /* Update each superblock (in-memory image), but |
2068 | * if we are allowed to, skip spares which already | |
2069 | * have the right event counter, or have one earlier | |
2070 | * (which would mean they aren't being marked as dirty | |
2071 | * with the rest of the array) | |
2072 | */ | |
1da177e4 | 2073 | mdk_rdev_t *rdev; |
1da177e4 | 2074 | |
5e865106 N |
2075 | /* First make sure individual recovery_offsets are correct */ |
2076 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
2077 | if (rdev->raid_disk >= 0 && | |
2078 | !test_bit(In_sync, &rdev->flags) && | |
2079 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2080 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2081 | ||
2082 | } | |
159ec1fc | 2083 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
2084 | if (rdev->sb_events == mddev->events || |
2085 | (nospares && | |
2086 | rdev->raid_disk < 0 && | |
2087 | (rdev->sb_events&1)==0 && | |
2088 | rdev->sb_events+1 == mddev->events)) { | |
2089 | /* Don't update this superblock */ | |
2090 | rdev->sb_loaded = 2; | |
2091 | } else { | |
2092 | super_types[mddev->major_version]. | |
2093 | sync_super(mddev, rdev); | |
2094 | rdev->sb_loaded = 1; | |
2095 | } | |
1da177e4 LT |
2096 | } |
2097 | } | |
2098 | ||
850b2b42 | 2099 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 2100 | { |
1da177e4 | 2101 | mdk_rdev_t *rdev; |
06d91a5f | 2102 | int sync_req; |
42543769 | 2103 | int nospares = 0; |
1da177e4 | 2104 | |
1b57f132 | 2105 | mddev->utime = get_seconds(); |
8377bc80 N |
2106 | if (mddev->external) |
2107 | return; | |
1da177e4 | 2108 | repeat: |
a9701a30 | 2109 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2110 | |
850b2b42 N |
2111 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2112 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) | |
2113 | force_change = 1; | |
2114 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2115 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2116 | * though if events isn't the right even/odd, we will have to do | |
2117 | * spares after all | |
2118 | */ | |
2119 | nospares = 1; | |
2120 | if (force_change) | |
2121 | nospares = 0; | |
2122 | if (mddev->degraded) | |
84692195 N |
2123 | /* If the array is degraded, then skipping spares is both |
2124 | * dangerous and fairly pointless. | |
2125 | * Dangerous because a device that was removed from the array | |
2126 | * might have a event_count that still looks up-to-date, | |
2127 | * so it can be re-added without a resync. | |
2128 | * Pointless because if there are any spares to skip, | |
2129 | * then a recovery will happen and soon that array won't | |
2130 | * be degraded any more and the spare can go back to sleep then. | |
2131 | */ | |
850b2b42 | 2132 | nospares = 0; |
84692195 | 2133 | |
06d91a5f | 2134 | sync_req = mddev->in_sync; |
42543769 N |
2135 | |
2136 | /* If this is just a dirty<->clean transition, and the array is clean | |
2137 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2138 | if (nospares |
42543769 | 2139 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
1031be7a N |
2140 | && (mddev->events & 1) |
2141 | && mddev->events != 1) | |
42543769 N |
2142 | mddev->events--; |
2143 | else { | |
2144 | /* otherwise we have to go forward and ... */ | |
2145 | mddev->events ++; | |
2146 | if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ | |
51d5668c N |
2147 | /* .. if the array isn't clean, an 'even' event must also go |
2148 | * to spares. */ | |
2149 | if ((mddev->events&1)==0) | |
42543769 | 2150 | nospares = 0; |
42543769 | 2151 | } else { |
51d5668c N |
2152 | /* otherwise an 'odd' event must go to spares */ |
2153 | if ((mddev->events&1)) | |
42543769 | 2154 | nospares = 0; |
42543769 N |
2155 | } |
2156 | } | |
1da177e4 LT |
2157 | |
2158 | if (!mddev->events) { | |
2159 | /* | |
2160 | * oops, this 64-bit counter should never wrap. | |
2161 | * Either we are in around ~1 trillion A.C., assuming | |
2162 | * 1 reboot per second, or we have a bug: | |
2163 | */ | |
2164 | MD_BUG(); | |
2165 | mddev->events --; | |
2166 | } | |
1da177e4 LT |
2167 | |
2168 | /* | |
2169 | * do not write anything to disk if using | |
2170 | * nonpersistent superblocks | |
2171 | */ | |
06d91a5f | 2172 | if (!mddev->persistent) { |
e691063a N |
2173 | if (!mddev->external) |
2174 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
2175 | ||
a9701a30 | 2176 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2177 | wake_up(&mddev->sb_wait); |
1da177e4 | 2178 | return; |
06d91a5f | 2179 | } |
e691063a | 2180 | sync_sbs(mddev, nospares); |
a9701a30 | 2181 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2182 | |
2183 | dprintk(KERN_INFO | |
2184 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2185 | mdname(mddev),mddev->in_sync); | |
2186 | ||
4ad13663 | 2187 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2188 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2189 | char b[BDEVNAME_SIZE]; |
2190 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2191 | if (rdev->sb_loaded != 1) |
2192 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2193 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2194 | dprintk("(skipping faulty "); |
2195 | ||
2196 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2197 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2198 | md_super_write(mddev,rdev, |
0f420358 | 2199 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2200 | rdev->sb_page); |
2201 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2202 | bdevname(rdev->bdev,b), | |
0f420358 | 2203 | (unsigned long long)rdev->sb_start); |
42543769 | 2204 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2205 | |
1da177e4 LT |
2206 | } else |
2207 | dprintk(")\n"); | |
7bfa19f2 | 2208 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2209 | /* only need to write one superblock... */ |
2210 | break; | |
2211 | } | |
a9701a30 | 2212 | md_super_wait(mddev); |
850b2b42 | 2213 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2214 | |
a9701a30 | 2215 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2216 | if (mddev->in_sync != sync_req || |
2217 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2218 | /* have to write it out again */ |
a9701a30 | 2219 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2220 | goto repeat; |
2221 | } | |
850b2b42 | 2222 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2223 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2224 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2225 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2226 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2227 | |
1da177e4 LT |
2228 | } |
2229 | ||
7f6ce769 | 2230 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2231 | * We want to accept with case. For this we use cmd_match. |
2232 | */ | |
2233 | static int cmd_match(const char *cmd, const char *str) | |
2234 | { | |
2235 | /* See if cmd, written into a sysfs file, matches | |
2236 | * str. They must either be the same, or cmd can | |
2237 | * have a trailing newline | |
2238 | */ | |
2239 | while (*cmd && *str && *cmd == *str) { | |
2240 | cmd++; | |
2241 | str++; | |
2242 | } | |
2243 | if (*cmd == '\n') | |
2244 | cmd++; | |
2245 | if (*str || *cmd) | |
2246 | return 0; | |
2247 | return 1; | |
2248 | } | |
2249 | ||
86e6ffdd N |
2250 | struct rdev_sysfs_entry { |
2251 | struct attribute attr; | |
2252 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2253 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2254 | }; | |
2255 | ||
2256 | static ssize_t | |
96de1e66 | 2257 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2258 | { |
2259 | char *sep = ""; | |
20a49ff6 | 2260 | size_t len = 0; |
86e6ffdd | 2261 | |
b2d444d7 | 2262 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2263 | len+= sprintf(page+len, "%sfaulty",sep); |
2264 | sep = ","; | |
2265 | } | |
b2d444d7 | 2266 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2267 | len += sprintf(page+len, "%sin_sync",sep); |
2268 | sep = ","; | |
2269 | } | |
f655675b N |
2270 | if (test_bit(WriteMostly, &rdev->flags)) { |
2271 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2272 | sep = ","; | |
2273 | } | |
6bfe0b49 DW |
2274 | if (test_bit(Blocked, &rdev->flags)) { |
2275 | len += sprintf(page+len, "%sblocked", sep); | |
2276 | sep = ","; | |
2277 | } | |
b2d444d7 N |
2278 | if (!test_bit(Faulty, &rdev->flags) && |
2279 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2280 | len += sprintf(page+len, "%sspare", sep); |
2281 | sep = ","; | |
2282 | } | |
2283 | return len+sprintf(page+len, "\n"); | |
2284 | } | |
2285 | ||
45dc2de1 N |
2286 | static ssize_t |
2287 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2288 | { | |
2289 | /* can write | |
2290 | * faulty - simulates and error | |
2291 | * remove - disconnects the device | |
f655675b N |
2292 | * writemostly - sets write_mostly |
2293 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2294 | * blocked - sets the Blocked flag |
2295 | * -blocked - clears the Blocked flag | |
6d56e278 | 2296 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2297 | */ |
2298 | int err = -EINVAL; | |
2299 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2300 | md_error(rdev->mddev, rdev); | |
2301 | err = 0; | |
2302 | } else if (cmd_match(buf, "remove")) { | |
2303 | if (rdev->raid_disk >= 0) | |
2304 | err = -EBUSY; | |
2305 | else { | |
2306 | mddev_t *mddev = rdev->mddev; | |
2307 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2308 | if (mddev->pers) |
2309 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2310 | md_new_event(mddev); |
2311 | err = 0; | |
2312 | } | |
f655675b N |
2313 | } else if (cmd_match(buf, "writemostly")) { |
2314 | set_bit(WriteMostly, &rdev->flags); | |
2315 | err = 0; | |
2316 | } else if (cmd_match(buf, "-writemostly")) { | |
2317 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2318 | err = 0; |
2319 | } else if (cmd_match(buf, "blocked")) { | |
2320 | set_bit(Blocked, &rdev->flags); | |
2321 | err = 0; | |
2322 | } else if (cmd_match(buf, "-blocked")) { | |
2323 | clear_bit(Blocked, &rdev->flags); | |
2324 | wake_up(&rdev->blocked_wait); | |
2325 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2326 | md_wakeup_thread(rdev->mddev->thread); | |
2327 | ||
6d56e278 N |
2328 | err = 0; |
2329 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2330 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2331 | err = 0; |
45dc2de1 | 2332 | } |
3c0ee63a N |
2333 | if (!err && rdev->sysfs_state) |
2334 | sysfs_notify_dirent(rdev->sysfs_state); | |
45dc2de1 N |
2335 | return err ? err : len; |
2336 | } | |
80ca3a44 N |
2337 | static struct rdev_sysfs_entry rdev_state = |
2338 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2339 | |
4dbcdc75 N |
2340 | static ssize_t |
2341 | errors_show(mdk_rdev_t *rdev, char *page) | |
2342 | { | |
2343 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2344 | } | |
2345 | ||
2346 | static ssize_t | |
2347 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2348 | { | |
2349 | char *e; | |
2350 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2351 | if (*buf && (*e == 0 || *e == '\n')) { | |
2352 | atomic_set(&rdev->corrected_errors, n); | |
2353 | return len; | |
2354 | } | |
2355 | return -EINVAL; | |
2356 | } | |
2357 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2358 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2359 | |
014236d2 N |
2360 | static ssize_t |
2361 | slot_show(mdk_rdev_t *rdev, char *page) | |
2362 | { | |
2363 | if (rdev->raid_disk < 0) | |
2364 | return sprintf(page, "none\n"); | |
2365 | else | |
2366 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2367 | } | |
2368 | ||
2369 | static ssize_t | |
2370 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2371 | { | |
2372 | char *e; | |
c303da6d N |
2373 | int err; |
2374 | char nm[20]; | |
014236d2 N |
2375 | int slot = simple_strtoul(buf, &e, 10); |
2376 | if (strncmp(buf, "none", 4)==0) | |
2377 | slot = -1; | |
2378 | else if (e==buf || (*e && *e!= '\n')) | |
2379 | return -EINVAL; | |
6c2fce2e | 2380 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2381 | /* Setting 'slot' on an active array requires also |
2382 | * updating the 'rd%d' link, and communicating | |
2383 | * with the personality with ->hot_*_disk. | |
2384 | * For now we only support removing | |
2385 | * failed/spare devices. This normally happens automatically, | |
2386 | * but not when the metadata is externally managed. | |
2387 | */ | |
c303da6d N |
2388 | if (rdev->raid_disk == -1) |
2389 | return -EEXIST; | |
2390 | /* personality does all needed checks */ | |
2391 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2392 | return -EINVAL; | |
2393 | err = rdev->mddev->pers-> | |
2394 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2395 | if (err) | |
2396 | return err; | |
2397 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2398 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
b7103107 | 2399 | rdev->raid_disk = -1; |
c303da6d N |
2400 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2401 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2402 | } else if (rdev->mddev->pers) { |
2403 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2404 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2405 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2406 | */ |
2407 | ||
2408 | if (rdev->raid_disk != -1) | |
2409 | return -EBUSY; | |
2410 | ||
2411 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2412 | return -EINVAL; | |
2413 | ||
159ec1fc | 2414 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2415 | if (rdev2->raid_disk == slot) |
2416 | return -EEXIST; | |
2417 | ||
2418 | rdev->raid_disk = slot; | |
2419 | if (test_bit(In_sync, &rdev->flags)) | |
2420 | rdev->saved_raid_disk = slot; | |
2421 | else | |
2422 | rdev->saved_raid_disk = -1; | |
2423 | err = rdev->mddev->pers-> | |
2424 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2425 | if (err) { |
6c2fce2e | 2426 | rdev->raid_disk = -1; |
6c2fce2e | 2427 | return err; |
52664732 | 2428 | } else |
3c0ee63a | 2429 | sysfs_notify_dirent(rdev->sysfs_state); |
6c2fce2e NB |
2430 | sprintf(nm, "rd%d", rdev->raid_disk); |
2431 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
2432 | printk(KERN_WARNING | |
2433 | "md: cannot register " | |
2434 | "%s for %s\n", | |
2435 | nm, mdname(rdev->mddev)); | |
2436 | ||
2437 | /* don't wakeup anyone, leave that to userspace. */ | |
c303da6d N |
2438 | } else { |
2439 | if (slot >= rdev->mddev->raid_disks) | |
2440 | return -ENOSPC; | |
2441 | rdev->raid_disk = slot; | |
2442 | /* assume it is working */ | |
c5d79adb N |
2443 | clear_bit(Faulty, &rdev->flags); |
2444 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2445 | set_bit(In_sync, &rdev->flags); |
3c0ee63a | 2446 | sysfs_notify_dirent(rdev->sysfs_state); |
c303da6d | 2447 | } |
014236d2 N |
2448 | return len; |
2449 | } | |
2450 | ||
2451 | ||
2452 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2453 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2454 | |
93c8cad0 N |
2455 | static ssize_t |
2456 | offset_show(mdk_rdev_t *rdev, char *page) | |
2457 | { | |
6961ece4 | 2458 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2459 | } |
2460 | ||
2461 | static ssize_t | |
2462 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2463 | { | |
2464 | char *e; | |
2465 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2466 | if (e==buf || (*e && *e != '\n')) | |
2467 | return -EINVAL; | |
8ed0a521 | 2468 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2469 | return -EBUSY; |
dd8ac336 | 2470 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2471 | /* Must set offset before size, so overlap checks |
2472 | * can be sane */ | |
2473 | return -EBUSY; | |
93c8cad0 N |
2474 | rdev->data_offset = offset; |
2475 | return len; | |
2476 | } | |
2477 | ||
2478 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2479 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2480 | |
83303b61 N |
2481 | static ssize_t |
2482 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2483 | { | |
dd8ac336 | 2484 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2485 | } |
2486 | ||
c5d79adb N |
2487 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2488 | { | |
2489 | /* check if two start/length pairs overlap */ | |
2490 | if (s1+l1 <= s2) | |
2491 | return 0; | |
2492 | if (s2+l2 <= s1) | |
2493 | return 0; | |
2494 | return 1; | |
2495 | } | |
2496 | ||
b522adcd DW |
2497 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2498 | { | |
2499 | unsigned long long blocks; | |
2500 | sector_t new; | |
2501 | ||
2502 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2503 | return -EINVAL; | |
2504 | ||
2505 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2506 | return -EINVAL; /* sector conversion overflow */ | |
2507 | ||
2508 | new = blocks * 2; | |
2509 | if (new != blocks * 2) | |
2510 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2511 | ||
2512 | *sectors = new; | |
2513 | return 0; | |
2514 | } | |
2515 | ||
83303b61 N |
2516 | static ssize_t |
2517 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2518 | { | |
27c529bb | 2519 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2520 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2521 | sector_t sectors; |
27c529bb | 2522 | |
b522adcd | 2523 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2524 | return -EINVAL; |
0cd17fec | 2525 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2526 | if (my_mddev->persistent) { |
dd8ac336 AN |
2527 | sectors = super_types[my_mddev->major_version]. |
2528 | rdev_size_change(rdev, sectors); | |
2529 | if (!sectors) | |
0cd17fec | 2530 | return -EBUSY; |
dd8ac336 AN |
2531 | } else if (!sectors) |
2532 | sectors = (rdev->bdev->bd_inode->i_size >> 9) - | |
2533 | rdev->data_offset; | |
0cd17fec | 2534 | } |
dd8ac336 | 2535 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2536 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2537 | |
dd8ac336 AN |
2538 | rdev->sectors = sectors; |
2539 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2540 | /* need to check that all other rdevs with the same ->bdev |
2541 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2542 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2543 | * we have to change it back, we will have the lock again. |
2544 | */ | |
2545 | mddev_t *mddev; | |
2546 | int overlap = 0; | |
159ec1fc | 2547 | struct list_head *tmp; |
c5d79adb | 2548 | |
27c529bb | 2549 | mddev_unlock(my_mddev); |
29ac4aa3 | 2550 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2551 | mdk_rdev_t *rdev2; |
2552 | ||
2553 | mddev_lock(mddev); | |
159ec1fc | 2554 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
c5d79adb N |
2555 | if (test_bit(AllReserved, &rdev2->flags) || |
2556 | (rdev->bdev == rdev2->bdev && | |
2557 | rdev != rdev2 && | |
dd8ac336 | 2558 | overlaps(rdev->data_offset, rdev->sectors, |
d07bd3bc | 2559 | rdev2->data_offset, |
dd8ac336 | 2560 | rdev2->sectors))) { |
c5d79adb N |
2561 | overlap = 1; |
2562 | break; | |
2563 | } | |
2564 | mddev_unlock(mddev); | |
2565 | if (overlap) { | |
2566 | mddev_put(mddev); | |
2567 | break; | |
2568 | } | |
2569 | } | |
27c529bb | 2570 | mddev_lock(my_mddev); |
c5d79adb N |
2571 | if (overlap) { |
2572 | /* Someone else could have slipped in a size | |
2573 | * change here, but doing so is just silly. | |
dd8ac336 | 2574 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2575 | * safe, and trust userspace not to race with |
2576 | * itself | |
2577 | */ | |
dd8ac336 | 2578 | rdev->sectors = oldsectors; |
c5d79adb N |
2579 | return -EBUSY; |
2580 | } | |
2581 | } | |
83303b61 N |
2582 | return len; |
2583 | } | |
2584 | ||
2585 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2586 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2587 | |
06e3c817 DW |
2588 | |
2589 | static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) | |
2590 | { | |
2591 | unsigned long long recovery_start = rdev->recovery_offset; | |
2592 | ||
2593 | if (test_bit(In_sync, &rdev->flags) || | |
2594 | recovery_start == MaxSector) | |
2595 | return sprintf(page, "none\n"); | |
2596 | ||
2597 | return sprintf(page, "%llu\n", recovery_start); | |
2598 | } | |
2599 | ||
2600 | static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2601 | { | |
2602 | unsigned long long recovery_start; | |
2603 | ||
2604 | if (cmd_match(buf, "none")) | |
2605 | recovery_start = MaxSector; | |
2606 | else if (strict_strtoull(buf, 10, &recovery_start)) | |
2607 | return -EINVAL; | |
2608 | ||
2609 | if (rdev->mddev->pers && | |
2610 | rdev->raid_disk >= 0) | |
2611 | return -EBUSY; | |
2612 | ||
2613 | rdev->recovery_offset = recovery_start; | |
2614 | if (recovery_start == MaxSector) | |
2615 | set_bit(In_sync, &rdev->flags); | |
2616 | else | |
2617 | clear_bit(In_sync, &rdev->flags); | |
2618 | return len; | |
2619 | } | |
2620 | ||
2621 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2622 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2623 | ||
86e6ffdd N |
2624 | static struct attribute *rdev_default_attrs[] = { |
2625 | &rdev_state.attr, | |
4dbcdc75 | 2626 | &rdev_errors.attr, |
014236d2 | 2627 | &rdev_slot.attr, |
93c8cad0 | 2628 | &rdev_offset.attr, |
83303b61 | 2629 | &rdev_size.attr, |
06e3c817 | 2630 | &rdev_recovery_start.attr, |
86e6ffdd N |
2631 | NULL, |
2632 | }; | |
2633 | static ssize_t | |
2634 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2635 | { | |
2636 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2637 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2638 | mddev_t *mddev = rdev->mddev; |
2639 | ssize_t rv; | |
86e6ffdd N |
2640 | |
2641 | if (!entry->show) | |
2642 | return -EIO; | |
27c529bb N |
2643 | |
2644 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2645 | if (!rv) { | |
2646 | if (rdev->mddev == NULL) | |
2647 | rv = -EBUSY; | |
2648 | else | |
2649 | rv = entry->show(rdev, page); | |
2650 | mddev_unlock(mddev); | |
2651 | } | |
2652 | return rv; | |
86e6ffdd N |
2653 | } |
2654 | ||
2655 | static ssize_t | |
2656 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2657 | const char *page, size_t length) | |
2658 | { | |
2659 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2660 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2661 | ssize_t rv; |
2662 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2663 | |
2664 | if (!entry->store) | |
2665 | return -EIO; | |
67463acb N |
2666 | if (!capable(CAP_SYS_ADMIN)) |
2667 | return -EACCES; | |
27c529bb | 2668 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2669 | if (!rv) { |
27c529bb N |
2670 | if (rdev->mddev == NULL) |
2671 | rv = -EBUSY; | |
2672 | else | |
2673 | rv = entry->store(rdev, page, length); | |
6a51830e | 2674 | mddev_unlock(mddev); |
ca388059 N |
2675 | } |
2676 | return rv; | |
86e6ffdd N |
2677 | } |
2678 | ||
2679 | static void rdev_free(struct kobject *ko) | |
2680 | { | |
2681 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2682 | kfree(rdev); | |
2683 | } | |
2684 | static struct sysfs_ops rdev_sysfs_ops = { | |
2685 | .show = rdev_attr_show, | |
2686 | .store = rdev_attr_store, | |
2687 | }; | |
2688 | static struct kobj_type rdev_ktype = { | |
2689 | .release = rdev_free, | |
2690 | .sysfs_ops = &rdev_sysfs_ops, | |
2691 | .default_attrs = rdev_default_attrs, | |
2692 | }; | |
2693 | ||
1da177e4 LT |
2694 | /* |
2695 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2696 | * | |
2697 | * mark the device faulty if: | |
2698 | * | |
2699 | * - the device is nonexistent (zero size) | |
2700 | * - the device has no valid superblock | |
2701 | * | |
2702 | * a faulty rdev _never_ has rdev->sb set. | |
2703 | */ | |
2704 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2705 | { | |
2706 | char b[BDEVNAME_SIZE]; | |
2707 | int err; | |
2708 | mdk_rdev_t *rdev; | |
2709 | sector_t size; | |
2710 | ||
9ffae0cf | 2711 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2712 | if (!rdev) { |
2713 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2714 | return ERR_PTR(-ENOMEM); | |
2715 | } | |
1da177e4 LT |
2716 | |
2717 | if ((err = alloc_disk_sb(rdev))) | |
2718 | goto abort_free; | |
2719 | ||
c5d79adb | 2720 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2721 | if (err) |
2722 | goto abort_free; | |
2723 | ||
f9cb074b | 2724 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2725 | |
1da177e4 | 2726 | rdev->desc_nr = -1; |
2b6e8459 | 2727 | rdev->saved_raid_disk = -1; |
3f9d7b0d | 2728 | rdev->raid_disk = -1; |
b2d444d7 | 2729 | rdev->flags = 0; |
1da177e4 | 2730 | rdev->data_offset = 0; |
42543769 | 2731 | rdev->sb_events = 0; |
1e50915f RB |
2732 | rdev->last_read_error.tv_sec = 0; |
2733 | rdev->last_read_error.tv_nsec = 0; | |
1da177e4 | 2734 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 2735 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 2736 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
2737 | |
2738 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
2739 | if (!size) { | |
2740 | printk(KERN_WARNING | |
2741 | "md: %s has zero or unknown size, marking faulty!\n", | |
2742 | bdevname(rdev->bdev,b)); | |
2743 | err = -EINVAL; | |
2744 | goto abort_free; | |
2745 | } | |
2746 | ||
2747 | if (super_format >= 0) { | |
2748 | err = super_types[super_format]. | |
2749 | load_super(rdev, NULL, super_minor); | |
2750 | if (err == -EINVAL) { | |
df968c4e N |
2751 | printk(KERN_WARNING |
2752 | "md: %s does not have a valid v%d.%d " | |
2753 | "superblock, not importing!\n", | |
2754 | bdevname(rdev->bdev,b), | |
2755 | super_format, super_minor); | |
1da177e4 LT |
2756 | goto abort_free; |
2757 | } | |
2758 | if (err < 0) { | |
2759 | printk(KERN_WARNING | |
2760 | "md: could not read %s's sb, not importing!\n", | |
2761 | bdevname(rdev->bdev,b)); | |
2762 | goto abort_free; | |
2763 | } | |
2764 | } | |
6bfe0b49 | 2765 | |
1da177e4 | 2766 | INIT_LIST_HEAD(&rdev->same_set); |
6bfe0b49 | 2767 | init_waitqueue_head(&rdev->blocked_wait); |
1da177e4 LT |
2768 | |
2769 | return rdev; | |
2770 | ||
2771 | abort_free: | |
2772 | if (rdev->sb_page) { | |
2773 | if (rdev->bdev) | |
2774 | unlock_rdev(rdev); | |
2775 | free_disk_sb(rdev); | |
2776 | } | |
2777 | kfree(rdev); | |
2778 | return ERR_PTR(err); | |
2779 | } | |
2780 | ||
2781 | /* | |
2782 | * Check a full RAID array for plausibility | |
2783 | */ | |
2784 | ||
2785 | ||
a757e64c | 2786 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2787 | { |
2788 | int i; | |
159ec1fc | 2789 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2790 | char b[BDEVNAME_SIZE]; |
2791 | ||
2792 | freshest = NULL; | |
d089c6af | 2793 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2794 | switch (super_types[mddev->major_version]. |
2795 | load_super(rdev, freshest, mddev->minor_version)) { | |
2796 | case 1: | |
2797 | freshest = rdev; | |
2798 | break; | |
2799 | case 0: | |
2800 | break; | |
2801 | default: | |
2802 | printk( KERN_ERR \ | |
2803 | "md: fatal superblock inconsistency in %s" | |
2804 | " -- removing from array\n", | |
2805 | bdevname(rdev->bdev,b)); | |
2806 | kick_rdev_from_array(rdev); | |
2807 | } | |
2808 | ||
2809 | ||
2810 | super_types[mddev->major_version]. | |
2811 | validate_super(mddev, freshest); | |
2812 | ||
2813 | i = 0; | |
d089c6af | 2814 | rdev_for_each(rdev, tmp, mddev) { |
233fca36 N |
2815 | if (mddev->max_disks && |
2816 | (rdev->desc_nr >= mddev->max_disks || | |
2817 | i > mddev->max_disks)) { | |
de01dfad N |
2818 | printk(KERN_WARNING |
2819 | "md: %s: %s: only %d devices permitted\n", | |
2820 | mdname(mddev), bdevname(rdev->bdev, b), | |
2821 | mddev->max_disks); | |
2822 | kick_rdev_from_array(rdev); | |
2823 | continue; | |
2824 | } | |
1da177e4 LT |
2825 | if (rdev != freshest) |
2826 | if (super_types[mddev->major_version]. | |
2827 | validate_super(mddev, rdev)) { | |
2828 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2829 | " from array!\n", | |
2830 | bdevname(rdev->bdev,b)); | |
2831 | kick_rdev_from_array(rdev); | |
2832 | continue; | |
2833 | } | |
2834 | if (mddev->level == LEVEL_MULTIPATH) { | |
2835 | rdev->desc_nr = i++; | |
2836 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2837 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 2838 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
2839 | rdev->raid_disk = -1; |
2840 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2841 | } |
2842 | } | |
1da177e4 LT |
2843 | } |
2844 | ||
72e02075 N |
2845 | /* Read a fixed-point number. |
2846 | * Numbers in sysfs attributes should be in "standard" units where | |
2847 | * possible, so time should be in seconds. | |
2848 | * However we internally use a a much smaller unit such as | |
2849 | * milliseconds or jiffies. | |
2850 | * This function takes a decimal number with a possible fractional | |
2851 | * component, and produces an integer which is the result of | |
2852 | * multiplying that number by 10^'scale'. | |
2853 | * all without any floating-point arithmetic. | |
2854 | */ | |
2855 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
2856 | { | |
2857 | unsigned long result = 0; | |
2858 | long decimals = -1; | |
2859 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
2860 | if (*cp == '.') | |
2861 | decimals = 0; | |
2862 | else if (decimals < scale) { | |
2863 | unsigned int value; | |
2864 | value = *cp - '0'; | |
2865 | result = result * 10 + value; | |
2866 | if (decimals >= 0) | |
2867 | decimals++; | |
2868 | } | |
2869 | cp++; | |
2870 | } | |
2871 | if (*cp == '\n') | |
2872 | cp++; | |
2873 | if (*cp) | |
2874 | return -EINVAL; | |
2875 | if (decimals < 0) | |
2876 | decimals = 0; | |
2877 | while (decimals < scale) { | |
2878 | result *= 10; | |
2879 | decimals ++; | |
2880 | } | |
2881 | *res = result; | |
2882 | return 0; | |
2883 | } | |
2884 | ||
2885 | ||
19052c0e N |
2886 | static void md_safemode_timeout(unsigned long data); |
2887 | ||
16f17b39 N |
2888 | static ssize_t |
2889 | safe_delay_show(mddev_t *mddev, char *page) | |
2890 | { | |
2891 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2892 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2893 | } | |
2894 | static ssize_t | |
2895 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2896 | { | |
16f17b39 | 2897 | unsigned long msec; |
97ce0a7f | 2898 | |
72e02075 | 2899 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 2900 | return -EINVAL; |
16f17b39 N |
2901 | if (msec == 0) |
2902 | mddev->safemode_delay = 0; | |
2903 | else { | |
19052c0e | 2904 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2905 | mddev->safemode_delay = (msec*HZ)/1000; |
2906 | if (mddev->safemode_delay == 0) | |
2907 | mddev->safemode_delay = 1; | |
19052c0e N |
2908 | if (mddev->safemode_delay < old_delay) |
2909 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2910 | } |
2911 | return len; | |
2912 | } | |
2913 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2914 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2915 | |
eae1701f | 2916 | static ssize_t |
96de1e66 | 2917 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2918 | { |
2604b703 | 2919 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2920 | if (p) |
eae1701f | 2921 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2922 | else if (mddev->clevel[0]) |
2923 | return sprintf(page, "%s\n", mddev->clevel); | |
2924 | else if (mddev->level != LEVEL_NONE) | |
2925 | return sprintf(page, "%d\n", mddev->level); | |
2926 | else | |
2927 | return 0; | |
eae1701f N |
2928 | } |
2929 | ||
d9d166c2 N |
2930 | static ssize_t |
2931 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2932 | { | |
245f46c2 | 2933 | char level[16]; |
20a49ff6 | 2934 | ssize_t rv = len; |
245f46c2 N |
2935 | struct mdk_personality *pers; |
2936 | void *priv; | |
3a981b03 | 2937 | mdk_rdev_t *rdev; |
245f46c2 N |
2938 | |
2939 | if (mddev->pers == NULL) { | |
2940 | if (len == 0) | |
2941 | return 0; | |
2942 | if (len >= sizeof(mddev->clevel)) | |
2943 | return -ENOSPC; | |
2944 | strncpy(mddev->clevel, buf, len); | |
2945 | if (mddev->clevel[len-1] == '\n') | |
2946 | len--; | |
2947 | mddev->clevel[len] = 0; | |
2948 | mddev->level = LEVEL_NONE; | |
2949 | return rv; | |
2950 | } | |
2951 | ||
2952 | /* request to change the personality. Need to ensure: | |
2953 | * - array is not engaged in resync/recovery/reshape | |
2954 | * - old personality can be suspended | |
2955 | * - new personality will access other array. | |
2956 | */ | |
2957 | ||
2958 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) | |
d9d166c2 | 2959 | return -EBUSY; |
245f46c2 N |
2960 | |
2961 | if (!mddev->pers->quiesce) { | |
2962 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
2963 | mdname(mddev), mddev->pers->name); | |
2964 | return -EINVAL; | |
2965 | } | |
2966 | ||
2967 | /* Now find the new personality */ | |
2968 | if (len == 0 || len >= sizeof(level)) | |
2969 | return -EINVAL; | |
2970 | strncpy(level, buf, len); | |
2971 | if (level[len-1] == '\n') | |
d9d166c2 | 2972 | len--; |
245f46c2 N |
2973 | level[len] = 0; |
2974 | ||
2975 | request_module("md-%s", level); | |
2976 | spin_lock(&pers_lock); | |
2977 | pers = find_pers(LEVEL_NONE, level); | |
2978 | if (!pers || !try_module_get(pers->owner)) { | |
2979 | spin_unlock(&pers_lock); | |
2980 | printk(KERN_WARNING "md: personality %s not loaded\n", level); | |
2981 | return -EINVAL; | |
2982 | } | |
2983 | spin_unlock(&pers_lock); | |
2984 | ||
2985 | if (pers == mddev->pers) { | |
2986 | /* Nothing to do! */ | |
2987 | module_put(pers->owner); | |
2988 | return rv; | |
2989 | } | |
2990 | if (!pers->takeover) { | |
2991 | module_put(pers->owner); | |
2992 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
2993 | mdname(mddev), level); | |
2994 | return -EINVAL; | |
2995 | } | |
2996 | ||
2997 | /* ->takeover must set new_* and/or delta_disks | |
2998 | * if it succeeds, and may set them when it fails. | |
2999 | */ | |
3000 | priv = pers->takeover(mddev); | |
3001 | if (IS_ERR(priv)) { | |
3002 | mddev->new_level = mddev->level; | |
3003 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3004 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3005 | mddev->raid_disks -= mddev->delta_disks; |
3006 | mddev->delta_disks = 0; | |
3007 | module_put(pers->owner); | |
3008 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
3009 | mdname(mddev), level); | |
3010 | return PTR_ERR(priv); | |
3011 | } | |
3012 | ||
3013 | /* Looks like we have a winner */ | |
3014 | mddev_suspend(mddev); | |
3015 | mddev->pers->stop(mddev); | |
a64c876f N |
3016 | |
3017 | if (mddev->pers->sync_request == NULL && | |
3018 | pers->sync_request != NULL) { | |
3019 | /* need to add the md_redundancy_group */ | |
3020 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3021 | printk(KERN_WARNING | |
3022 | "md: cannot register extra attributes for %s\n", | |
3023 | mdname(mddev)); | |
3024 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); | |
3025 | } | |
3026 | if (mddev->pers->sync_request != NULL && | |
3027 | pers->sync_request == NULL) { | |
3028 | /* need to remove the md_redundancy_group */ | |
3029 | if (mddev->to_remove == NULL) | |
3030 | mddev->to_remove = &md_redundancy_group; | |
3031 | } | |
3032 | ||
54071b38 TM |
3033 | if (mddev->pers->sync_request == NULL && |
3034 | mddev->external) { | |
3035 | /* We are converting from a no-redundancy array | |
3036 | * to a redundancy array and metadata is managed | |
3037 | * externally so we need to be sure that writes | |
3038 | * won't block due to a need to transition | |
3039 | * clean->dirty | |
3040 | * until external management is started. | |
3041 | */ | |
3042 | mddev->in_sync = 0; | |
3043 | mddev->safemode_delay = 0; | |
3044 | mddev->safemode = 0; | |
3045 | } | |
3046 | ||
245f46c2 | 3047 | module_put(mddev->pers->owner); |
3a981b03 N |
3048 | /* Invalidate devices that are now superfluous */ |
3049 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
3050 | if (rdev->raid_disk >= mddev->raid_disks) { | |
3051 | rdev->raid_disk = -1; | |
3052 | clear_bit(In_sync, &rdev->flags); | |
3053 | } | |
245f46c2 N |
3054 | mddev->pers = pers; |
3055 | mddev->private = priv; | |
3056 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3057 | mddev->level = mddev->new_level; | |
3058 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3059 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3060 | mddev->delta_disks = 0; |
9af204cf TM |
3061 | if (mddev->pers->sync_request == NULL) { |
3062 | /* this is now an array without redundancy, so | |
3063 | * it must always be in_sync | |
3064 | */ | |
3065 | mddev->in_sync = 1; | |
3066 | del_timer_sync(&mddev->safemode_timer); | |
3067 | } | |
245f46c2 N |
3068 | pers->run(mddev); |
3069 | mddev_resume(mddev); | |
3070 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3071 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3072 | md_wakeup_thread(mddev->thread); | |
5cac7861 | 3073 | sysfs_notify(&mddev->kobj, NULL, "level"); |
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); |
9e653b63 N |
3312 | static int do_md_run(mddev_t * mddev); |
3313 | static int restart_array(mddev_t *mddev); | |
3314 | ||
3315 | static ssize_t | |
3316 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3317 | { | |
3318 | int err = -EINVAL; | |
3319 | enum array_state st = match_word(buf, array_states); | |
3320 | switch(st) { | |
3321 | case bad_word: | |
3322 | break; | |
3323 | case clear: | |
3324 | /* stopping an active array */ | |
f2ea68cf | 3325 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3326 | return -EBUSY; |
df5b20cf | 3327 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3328 | break; |
3329 | case inactive: | |
3330 | /* stopping an active array */ | |
3331 | if (mddev->pers) { | |
f2ea68cf | 3332 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3333 | return -EBUSY; |
df5b20cf | 3334 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3335 | } else |
3336 | err = 0; /* already inactive */ | |
9e653b63 N |
3337 | break; |
3338 | case suspended: | |
3339 | break; /* not supported yet */ | |
3340 | case readonly: | |
3341 | if (mddev->pers) | |
df5b20cf | 3342 | err = do_md_stop(mddev, 1, 0); |
9e653b63 N |
3343 | else { |
3344 | mddev->ro = 1; | |
648b629e | 3345 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3346 | err = do_md_run(mddev); |
3347 | } | |
3348 | break; | |
3349 | case read_auto: | |
9e653b63 | 3350 | if (mddev->pers) { |
80268ee9 | 3351 | if (mddev->ro == 0) |
df5b20cf | 3352 | err = do_md_stop(mddev, 1, 0); |
80268ee9 | 3353 | else if (mddev->ro == 1) |
648b629e N |
3354 | err = restart_array(mddev); |
3355 | if (err == 0) { | |
3356 | mddev->ro = 2; | |
3357 | set_disk_ro(mddev->gendisk, 0); | |
3358 | } | |
9e653b63 N |
3359 | } else { |
3360 | mddev->ro = 2; | |
3361 | err = do_md_run(mddev); | |
3362 | } | |
3363 | break; | |
3364 | case clean: | |
3365 | if (mddev->pers) { | |
3366 | restart_array(mddev); | |
3367 | spin_lock_irq(&mddev->write_lock); | |
3368 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3369 | if (mddev->in_sync == 0) { |
3370 | mddev->in_sync = 1; | |
31a59e34 N |
3371 | if (mddev->safemode == 1) |
3372 | mddev->safemode = 0; | |
e691063a N |
3373 | if (mddev->persistent) |
3374 | set_bit(MD_CHANGE_CLEAN, | |
3375 | &mddev->flags); | |
3376 | } | |
3377 | err = 0; | |
3378 | } else | |
3379 | err = -EBUSY; | |
9e653b63 | 3380 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3381 | } else |
3382 | err = -EINVAL; | |
9e653b63 N |
3383 | break; |
3384 | case active: | |
3385 | if (mddev->pers) { | |
3386 | restart_array(mddev); | |
e691063a N |
3387 | if (mddev->external) |
3388 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
9e653b63 N |
3389 | wake_up(&mddev->sb_wait); |
3390 | err = 0; | |
3391 | } else { | |
3392 | mddev->ro = 0; | |
648b629e | 3393 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3394 | err = do_md_run(mddev); |
3395 | } | |
3396 | break; | |
3397 | case write_pending: | |
3398 | case active_idle: | |
3399 | /* these cannot be set */ | |
3400 | break; | |
3401 | } | |
3402 | if (err) | |
3403 | return err; | |
0fd62b86 | 3404 | else { |
b62b7590 | 3405 | sysfs_notify_dirent(mddev->sysfs_state); |
9e653b63 | 3406 | return len; |
0fd62b86 | 3407 | } |
9e653b63 | 3408 | } |
80ca3a44 N |
3409 | static struct md_sysfs_entry md_array_state = |
3410 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3411 | |
1e50915f RB |
3412 | static ssize_t |
3413 | max_corrected_read_errors_show(mddev_t *mddev, char *page) { | |
3414 | return sprintf(page, "%d\n", | |
3415 | atomic_read(&mddev->max_corr_read_errors)); | |
3416 | } | |
3417 | ||
3418 | static ssize_t | |
3419 | max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) | |
3420 | { | |
3421 | char *e; | |
3422 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3423 | ||
3424 | if (*buf && (*e == 0 || *e == '\n')) { | |
3425 | atomic_set(&mddev->max_corr_read_errors, n); | |
3426 | return len; | |
3427 | } | |
3428 | return -EINVAL; | |
3429 | } | |
3430 | ||
3431 | static struct md_sysfs_entry max_corr_read_errors = | |
3432 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3433 | max_corrected_read_errors_store); | |
3434 | ||
6d7ff738 N |
3435 | static ssize_t |
3436 | null_show(mddev_t *mddev, char *page) | |
3437 | { | |
3438 | return -EINVAL; | |
3439 | } | |
3440 | ||
3441 | static ssize_t | |
3442 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3443 | { | |
3444 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3445 | /* The new device is added to the array. | |
3446 | * If the array has a persistent superblock, we read the | |
3447 | * superblock to initialise info and check validity. | |
3448 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3449 | * which mainly checks size. | |
3450 | */ | |
3451 | char *e; | |
3452 | int major = simple_strtoul(buf, &e, 10); | |
3453 | int minor; | |
3454 | dev_t dev; | |
3455 | mdk_rdev_t *rdev; | |
3456 | int err; | |
3457 | ||
3458 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3459 | return -EINVAL; | |
3460 | minor = simple_strtoul(e+1, &e, 10); | |
3461 | if (*e && *e != '\n') | |
3462 | return -EINVAL; | |
3463 | dev = MKDEV(major, minor); | |
3464 | if (major != MAJOR(dev) || | |
3465 | minor != MINOR(dev)) | |
3466 | return -EOVERFLOW; | |
3467 | ||
3468 | ||
3469 | if (mddev->persistent) { | |
3470 | rdev = md_import_device(dev, mddev->major_version, | |
3471 | mddev->minor_version); | |
3472 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3473 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3474 | mdk_rdev_t, same_set); | |
3475 | err = super_types[mddev->major_version] | |
3476 | .load_super(rdev, rdev0, mddev->minor_version); | |
3477 | if (err < 0) | |
3478 | goto out; | |
3479 | } | |
c5d79adb N |
3480 | } else if (mddev->external) |
3481 | rdev = md_import_device(dev, -2, -1); | |
3482 | else | |
6d7ff738 N |
3483 | rdev = md_import_device(dev, -1, -1); |
3484 | ||
3485 | if (IS_ERR(rdev)) | |
3486 | return PTR_ERR(rdev); | |
3487 | err = bind_rdev_to_array(rdev, mddev); | |
3488 | out: | |
3489 | if (err) | |
3490 | export_rdev(rdev); | |
3491 | return err ? err : len; | |
3492 | } | |
3493 | ||
3494 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3495 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3496 | |
9b1d1dac PC |
3497 | static ssize_t |
3498 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3499 | { | |
3500 | char *end; | |
3501 | unsigned long chunk, end_chunk; | |
3502 | ||
3503 | if (!mddev->bitmap) | |
3504 | goto out; | |
3505 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3506 | while (*buf) { | |
3507 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3508 | if (buf == end) break; | |
3509 | if (*end == '-') { /* range */ | |
3510 | buf = end + 1; | |
3511 | end_chunk = simple_strtoul(buf, &end, 0); | |
3512 | if (buf == end) break; | |
3513 | } | |
3514 | if (*end && !isspace(*end)) break; | |
3515 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3516 | buf = skip_spaces(end); |
9b1d1dac PC |
3517 | } |
3518 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3519 | out: | |
3520 | return len; | |
3521 | } | |
3522 | ||
3523 | static struct md_sysfs_entry md_bitmap = | |
3524 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3525 | ||
a35b0d69 N |
3526 | static ssize_t |
3527 | size_show(mddev_t *mddev, char *page) | |
3528 | { | |
58c0fed4 AN |
3529 | return sprintf(page, "%llu\n", |
3530 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3531 | } |
3532 | ||
d71f9f88 | 3533 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3534 | |
3535 | static ssize_t | |
3536 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3537 | { | |
3538 | /* If array is inactive, we can reduce the component size, but | |
3539 | * not increase it (except from 0). | |
3540 | * If array is active, we can try an on-line resize | |
3541 | */ | |
b522adcd DW |
3542 | sector_t sectors; |
3543 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3544 | |
58c0fed4 AN |
3545 | if (err < 0) |
3546 | return err; | |
a35b0d69 | 3547 | if (mddev->pers) { |
58c0fed4 | 3548 | err = update_size(mddev, sectors); |
850b2b42 | 3549 | md_update_sb(mddev, 1); |
a35b0d69 | 3550 | } else { |
58c0fed4 AN |
3551 | if (mddev->dev_sectors == 0 || |
3552 | mddev->dev_sectors > sectors) | |
3553 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3554 | else |
3555 | err = -ENOSPC; | |
3556 | } | |
3557 | return err ? err : len; | |
3558 | } | |
3559 | ||
3560 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3561 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3562 | |
8bb93aac N |
3563 | |
3564 | /* Metdata version. | |
e691063a N |
3565 | * This is one of |
3566 | * 'none' for arrays with no metadata (good luck...) | |
3567 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3568 | * or N.M for internally known formats |
3569 | */ | |
3570 | static ssize_t | |
3571 | metadata_show(mddev_t *mddev, char *page) | |
3572 | { | |
3573 | if (mddev->persistent) | |
3574 | return sprintf(page, "%d.%d\n", | |
3575 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3576 | else if (mddev->external) |
3577 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3578 | else |
3579 | return sprintf(page, "none\n"); | |
3580 | } | |
3581 | ||
3582 | static ssize_t | |
3583 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3584 | { | |
3585 | int major, minor; | |
3586 | char *e; | |
ea43ddd8 N |
3587 | /* Changing the details of 'external' metadata is |
3588 | * always permitted. Otherwise there must be | |
3589 | * no devices attached to the array. | |
3590 | */ | |
3591 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3592 | ; | |
3593 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3594 | return -EBUSY; |
3595 | ||
3596 | if (cmd_match(buf, "none")) { | |
3597 | mddev->persistent = 0; | |
e691063a N |
3598 | mddev->external = 0; |
3599 | mddev->major_version = 0; | |
3600 | mddev->minor_version = 90; | |
3601 | return len; | |
3602 | } | |
3603 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3604 | size_t namelen = len-9; |
e691063a N |
3605 | if (namelen >= sizeof(mddev->metadata_type)) |
3606 | namelen = sizeof(mddev->metadata_type)-1; | |
3607 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3608 | mddev->metadata_type[namelen] = 0; | |
3609 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3610 | mddev->metadata_type[--namelen] = 0; | |
3611 | mddev->persistent = 0; | |
3612 | mddev->external = 1; | |
8bb93aac N |
3613 | mddev->major_version = 0; |
3614 | mddev->minor_version = 90; | |
3615 | return len; | |
3616 | } | |
3617 | major = simple_strtoul(buf, &e, 10); | |
3618 | if (e==buf || *e != '.') | |
3619 | return -EINVAL; | |
3620 | buf = e+1; | |
3621 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3622 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3623 | return -EINVAL; |
50511da3 | 3624 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3625 | return -ENOENT; |
3626 | mddev->major_version = major; | |
3627 | mddev->minor_version = minor; | |
3628 | mddev->persistent = 1; | |
e691063a | 3629 | mddev->external = 0; |
8bb93aac N |
3630 | return len; |
3631 | } | |
3632 | ||
3633 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3634 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3635 | |
24dd469d | 3636 | static ssize_t |
7eec314d | 3637 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3638 | { |
7eec314d | 3639 | char *type = "idle"; |
b6a9ce68 N |
3640 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3641 | type = "frozen"; | |
3642 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3643 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3644 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3645 | type = "reshape"; | |
3646 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3647 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3648 | type = "resync"; | |
3649 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3650 | type = "check"; | |
3651 | else | |
3652 | type = "repair"; | |
72a23c21 | 3653 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3654 | type = "recover"; |
3655 | } | |
3656 | return sprintf(page, "%s\n", type); | |
3657 | } | |
3658 | ||
3659 | static ssize_t | |
7eec314d | 3660 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3661 | { |
7eec314d N |
3662 | if (!mddev->pers || !mddev->pers->sync_request) |
3663 | return -EINVAL; | |
3664 | ||
b6a9ce68 N |
3665 | if (cmd_match(page, "frozen")) |
3666 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3667 | else | |
3668 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3669 | ||
3670 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3671 | if (mddev->sync_thread) { |
3672 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
3673 | md_unregister_thread(mddev->sync_thread); | |
3674 | mddev->sync_thread = NULL; | |
3675 | mddev->recovery = 0; | |
3676 | } | |
03c902e1 N |
3677 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3678 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3679 | return -EBUSY; |
72a23c21 NB |
3680 | else if (cmd_match(page, "resync")) |
3681 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3682 | else if (cmd_match(page, "recover")) { | |
3683 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3684 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3685 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3686 | int err; |
3687 | if (mddev->pers->start_reshape == NULL) | |
3688 | return -EINVAL; | |
3689 | err = mddev->pers->start_reshape(mddev); | |
3690 | if (err) | |
3691 | return err; | |
a99ac971 | 3692 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3693 | } else { |
bce74dac | 3694 | if (cmd_match(page, "check")) |
7eec314d | 3695 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3696 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3697 | return -EINVAL; |
3698 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3699 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3700 | } |
03c902e1 | 3701 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3702 | md_wakeup_thread(mddev->thread); |
0c3573f1 | 3703 | sysfs_notify_dirent(mddev->sysfs_action); |
24dd469d N |
3704 | return len; |
3705 | } | |
3706 | ||
9d88883e | 3707 | static ssize_t |
96de1e66 | 3708 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3709 | { |
3710 | return sprintf(page, "%llu\n", | |
3711 | (unsigned long long) mddev->resync_mismatches); | |
3712 | } | |
3713 | ||
80ca3a44 N |
3714 | static struct md_sysfs_entry md_scan_mode = |
3715 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3716 | |
96de1e66 | 3717 | |
80ca3a44 | 3718 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3719 | |
88202a0c N |
3720 | static ssize_t |
3721 | sync_min_show(mddev_t *mddev, char *page) | |
3722 | { | |
3723 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3724 | mddev->sync_speed_min ? "local": "system"); | |
3725 | } | |
3726 | ||
3727 | static ssize_t | |
3728 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3729 | { | |
3730 | int min; | |
3731 | char *e; | |
3732 | if (strncmp(buf, "system", 6)==0) { | |
3733 | mddev->sync_speed_min = 0; | |
3734 | return len; | |
3735 | } | |
3736 | min = simple_strtoul(buf, &e, 10); | |
3737 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3738 | return -EINVAL; | |
3739 | mddev->sync_speed_min = min; | |
3740 | return len; | |
3741 | } | |
3742 | ||
3743 | static struct md_sysfs_entry md_sync_min = | |
3744 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3745 | ||
3746 | static ssize_t | |
3747 | sync_max_show(mddev_t *mddev, char *page) | |
3748 | { | |
3749 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3750 | mddev->sync_speed_max ? "local": "system"); | |
3751 | } | |
3752 | ||
3753 | static ssize_t | |
3754 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3755 | { | |
3756 | int max; | |
3757 | char *e; | |
3758 | if (strncmp(buf, "system", 6)==0) { | |
3759 | mddev->sync_speed_max = 0; | |
3760 | return len; | |
3761 | } | |
3762 | max = simple_strtoul(buf, &e, 10); | |
3763 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3764 | return -EINVAL; | |
3765 | mddev->sync_speed_max = max; | |
3766 | return len; | |
3767 | } | |
3768 | ||
3769 | static struct md_sysfs_entry md_sync_max = | |
3770 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3771 | ||
d7f3d291 IP |
3772 | static ssize_t |
3773 | degraded_show(mddev_t *mddev, char *page) | |
3774 | { | |
3775 | return sprintf(page, "%d\n", mddev->degraded); | |
3776 | } | |
3777 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3778 | |
90b08710 BS |
3779 | static ssize_t |
3780 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3781 | { | |
3782 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3783 | } | |
3784 | ||
3785 | static ssize_t | |
3786 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3787 | { | |
3788 | long n; | |
3789 | ||
3790 | if (strict_strtol(buf, 10, &n)) | |
3791 | return -EINVAL; | |
3792 | ||
3793 | if (n != 0 && n != 1) | |
3794 | return -EINVAL; | |
3795 | ||
3796 | mddev->parallel_resync = n; | |
3797 | ||
3798 | if (mddev->sync_thread) | |
3799 | wake_up(&resync_wait); | |
3800 | ||
3801 | return len; | |
3802 | } | |
3803 | ||
3804 | /* force parallel resync, even with shared block devices */ | |
3805 | static struct md_sysfs_entry md_sync_force_parallel = | |
3806 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3807 | sync_force_parallel_show, sync_force_parallel_store); | |
3808 | ||
88202a0c N |
3809 | static ssize_t |
3810 | sync_speed_show(mddev_t *mddev, char *page) | |
3811 | { | |
3812 | unsigned long resync, dt, db; | |
d1a7c503 N |
3813 | if (mddev->curr_resync == 0) |
3814 | return sprintf(page, "none\n"); | |
9687a60c AN |
3815 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3816 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3817 | if (!dt) dt++; |
9687a60c AN |
3818 | db = resync - mddev->resync_mark_cnt; |
3819 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3820 | } |
3821 | ||
80ca3a44 | 3822 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3823 | |
3824 | static ssize_t | |
3825 | sync_completed_show(mddev_t *mddev, char *page) | |
3826 | { | |
58c0fed4 | 3827 | unsigned long max_sectors, resync; |
88202a0c | 3828 | |
acb180b0 N |
3829 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3830 | return sprintf(page, "none\n"); | |
3831 | ||
88202a0c | 3832 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3833 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3834 | else |
58c0fed4 | 3835 | max_sectors = mddev->dev_sectors; |
88202a0c | 3836 | |
acb180b0 | 3837 | resync = mddev->curr_resync_completed; |
58c0fed4 | 3838 | return sprintf(page, "%lu / %lu\n", resync, max_sectors); |
88202a0c N |
3839 | } |
3840 | ||
80ca3a44 | 3841 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3842 | |
5e96ee65 NB |
3843 | static ssize_t |
3844 | min_sync_show(mddev_t *mddev, char *page) | |
3845 | { | |
3846 | return sprintf(page, "%llu\n", | |
3847 | (unsigned long long)mddev->resync_min); | |
3848 | } | |
3849 | static ssize_t | |
3850 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3851 | { | |
3852 | unsigned long long min; | |
3853 | if (strict_strtoull(buf, 10, &min)) | |
3854 | return -EINVAL; | |
3855 | if (min > mddev->resync_max) | |
3856 | return -EINVAL; | |
3857 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3858 | return -EBUSY; | |
3859 | ||
3860 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3861 | if (mddev->chunk_sectors) { |
2ac06c33 | 3862 | sector_t temp = min; |
9d8f0363 | 3863 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3864 | return -EINVAL; |
3865 | } | |
3866 | mddev->resync_min = min; | |
3867 | ||
3868 | return len; | |
3869 | } | |
3870 | ||
3871 | static struct md_sysfs_entry md_min_sync = | |
3872 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3873 | ||
c6207277 N |
3874 | static ssize_t |
3875 | max_sync_show(mddev_t *mddev, char *page) | |
3876 | { | |
3877 | if (mddev->resync_max == MaxSector) | |
3878 | return sprintf(page, "max\n"); | |
3879 | else | |
3880 | return sprintf(page, "%llu\n", | |
3881 | (unsigned long long)mddev->resync_max); | |
3882 | } | |
3883 | static ssize_t | |
3884 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3885 | { | |
3886 | if (strncmp(buf, "max", 3) == 0) | |
3887 | mddev->resync_max = MaxSector; | |
3888 | else { | |
5e96ee65 NB |
3889 | unsigned long long max; |
3890 | if (strict_strtoull(buf, 10, &max)) | |
3891 | return -EINVAL; | |
3892 | if (max < mddev->resync_min) | |
c6207277 N |
3893 | return -EINVAL; |
3894 | if (max < mddev->resync_max && | |
4d484a4a | 3895 | mddev->ro == 0 && |
c6207277 N |
3896 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3897 | return -EBUSY; | |
3898 | ||
3899 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3900 | if (mddev->chunk_sectors) { |
2ac06c33 | 3901 | sector_t temp = max; |
9d8f0363 | 3902 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
3903 | return -EINVAL; |
3904 | } | |
3905 | mddev->resync_max = max; | |
3906 | } | |
3907 | wake_up(&mddev->recovery_wait); | |
3908 | return len; | |
3909 | } | |
3910 | ||
3911 | static struct md_sysfs_entry md_max_sync = | |
3912 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
3913 | ||
e464eafd N |
3914 | static ssize_t |
3915 | suspend_lo_show(mddev_t *mddev, char *page) | |
3916 | { | |
3917 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
3918 | } | |
3919 | ||
3920 | static ssize_t | |
3921 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
3922 | { | |
3923 | char *e; | |
3924 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3925 | ||
b8d966ef N |
3926 | if (mddev->pers == NULL || |
3927 | mddev->pers->quiesce == NULL) | |
e464eafd N |
3928 | return -EINVAL; |
3929 | if (buf == e || (*e && *e != '\n')) | |
3930 | return -EINVAL; | |
3931 | if (new >= mddev->suspend_hi || | |
3932 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
3933 | mddev->suspend_lo = new; | |
3934 | mddev->pers->quiesce(mddev, 2); | |
3935 | return len; | |
3936 | } else | |
3937 | return -EINVAL; | |
3938 | } | |
3939 | static struct md_sysfs_entry md_suspend_lo = | |
3940 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
3941 | ||
3942 | ||
3943 | static ssize_t | |
3944 | suspend_hi_show(mddev_t *mddev, char *page) | |
3945 | { | |
3946 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
3947 | } | |
3948 | ||
3949 | static ssize_t | |
3950 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
3951 | { | |
3952 | char *e; | |
3953 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3954 | ||
b8d966ef N |
3955 | if (mddev->pers == NULL || |
3956 | mddev->pers->quiesce == NULL) | |
e464eafd N |
3957 | return -EINVAL; |
3958 | if (buf == e || (*e && *e != '\n')) | |
3959 | return -EINVAL; | |
3960 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
3961 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
3962 | mddev->suspend_hi = new; | |
3963 | mddev->pers->quiesce(mddev, 1); | |
3964 | mddev->pers->quiesce(mddev, 0); | |
3965 | return len; | |
3966 | } else | |
3967 | return -EINVAL; | |
3968 | } | |
3969 | static struct md_sysfs_entry md_suspend_hi = | |
3970 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
3971 | ||
08a02ecd N |
3972 | static ssize_t |
3973 | reshape_position_show(mddev_t *mddev, char *page) | |
3974 | { | |
3975 | if (mddev->reshape_position != MaxSector) | |
3976 | return sprintf(page, "%llu\n", | |
3977 | (unsigned long long)mddev->reshape_position); | |
3978 | strcpy(page, "none\n"); | |
3979 | return 5; | |
3980 | } | |
3981 | ||
3982 | static ssize_t | |
3983 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
3984 | { | |
3985 | char *e; | |
3986 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3987 | if (mddev->pers) | |
3988 | return -EBUSY; | |
3989 | if (buf == e || (*e && *e != '\n')) | |
3990 | return -EINVAL; | |
3991 | mddev->reshape_position = new; | |
3992 | mddev->delta_disks = 0; | |
3993 | mddev->new_level = mddev->level; | |
3994 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3995 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
3996 | return len; |
3997 | } | |
3998 | ||
3999 | static struct md_sysfs_entry md_reshape_position = | |
4000 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4001 | reshape_position_store); | |
4002 | ||
b522adcd DW |
4003 | static ssize_t |
4004 | array_size_show(mddev_t *mddev, char *page) | |
4005 | { | |
4006 | if (mddev->external_size) | |
4007 | return sprintf(page, "%llu\n", | |
4008 | (unsigned long long)mddev->array_sectors/2); | |
4009 | else | |
4010 | return sprintf(page, "default\n"); | |
4011 | } | |
4012 | ||
4013 | static ssize_t | |
4014 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
4015 | { | |
4016 | sector_t sectors; | |
4017 | ||
4018 | if (strncmp(buf, "default", 7) == 0) { | |
4019 | if (mddev->pers) | |
4020 | sectors = mddev->pers->size(mddev, 0, 0); | |
4021 | else | |
4022 | sectors = mddev->array_sectors; | |
4023 | ||
4024 | mddev->external_size = 0; | |
4025 | } else { | |
4026 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4027 | return -EINVAL; | |
4028 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4029 | return -E2BIG; |
b522adcd DW |
4030 | |
4031 | mddev->external_size = 1; | |
4032 | } | |
4033 | ||
4034 | mddev->array_sectors = sectors; | |
4035 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e N |
4036 | if (mddev->pers) |
4037 | revalidate_disk(mddev->gendisk); | |
b522adcd DW |
4038 | |
4039 | return len; | |
4040 | } | |
4041 | ||
4042 | static struct md_sysfs_entry md_array_size = | |
4043 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4044 | array_size_store); | |
e464eafd | 4045 | |
eae1701f N |
4046 | static struct attribute *md_default_attrs[] = { |
4047 | &md_level.attr, | |
d4dbd025 | 4048 | &md_layout.attr, |
eae1701f | 4049 | &md_raid_disks.attr, |
3b34380a | 4050 | &md_chunk_size.attr, |
a35b0d69 | 4051 | &md_size.attr, |
a94213b1 | 4052 | &md_resync_start.attr, |
8bb93aac | 4053 | &md_metadata.attr, |
6d7ff738 | 4054 | &md_new_device.attr, |
16f17b39 | 4055 | &md_safe_delay.attr, |
9e653b63 | 4056 | &md_array_state.attr, |
08a02ecd | 4057 | &md_reshape_position.attr, |
b522adcd | 4058 | &md_array_size.attr, |
1e50915f | 4059 | &max_corr_read_errors.attr, |
411036fa N |
4060 | NULL, |
4061 | }; | |
4062 | ||
4063 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4064 | &md_scan_mode.attr, |
9d88883e | 4065 | &md_mismatches.attr, |
88202a0c N |
4066 | &md_sync_min.attr, |
4067 | &md_sync_max.attr, | |
4068 | &md_sync_speed.attr, | |
90b08710 | 4069 | &md_sync_force_parallel.attr, |
88202a0c | 4070 | &md_sync_completed.attr, |
5e96ee65 | 4071 | &md_min_sync.attr, |
c6207277 | 4072 | &md_max_sync.attr, |
e464eafd N |
4073 | &md_suspend_lo.attr, |
4074 | &md_suspend_hi.attr, | |
9b1d1dac | 4075 | &md_bitmap.attr, |
d7f3d291 | 4076 | &md_degraded.attr, |
eae1701f N |
4077 | NULL, |
4078 | }; | |
411036fa N |
4079 | static struct attribute_group md_redundancy_group = { |
4080 | .name = NULL, | |
4081 | .attrs = md_redundancy_attrs, | |
4082 | }; | |
4083 | ||
eae1701f N |
4084 | |
4085 | static ssize_t | |
4086 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4087 | { | |
4088 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4089 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4090 | ssize_t rv; |
eae1701f N |
4091 | |
4092 | if (!entry->show) | |
4093 | return -EIO; | |
5dc5cf7d IM |
4094 | rv = mddev_lock(mddev); |
4095 | if (!rv) { | |
4096 | rv = entry->show(mddev, page); | |
4097 | mddev_unlock(mddev); | |
4098 | } | |
96de1e66 | 4099 | return rv; |
eae1701f N |
4100 | } |
4101 | ||
4102 | static ssize_t | |
4103 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4104 | const char *page, size_t length) | |
4105 | { | |
4106 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4107 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4108 | ssize_t rv; |
eae1701f N |
4109 | |
4110 | if (!entry->store) | |
4111 | return -EIO; | |
67463acb N |
4112 | if (!capable(CAP_SYS_ADMIN)) |
4113 | return -EACCES; | |
5dc5cf7d | 4114 | rv = mddev_lock(mddev); |
d3374825 N |
4115 | if (mddev->hold_active == UNTIL_IOCTL) |
4116 | mddev->hold_active = 0; | |
5dc5cf7d IM |
4117 | if (!rv) { |
4118 | rv = entry->store(mddev, page, length); | |
4119 | mddev_unlock(mddev); | |
4120 | } | |
96de1e66 | 4121 | return rv; |
eae1701f N |
4122 | } |
4123 | ||
4124 | static void md_free(struct kobject *ko) | |
4125 | { | |
4126 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
4127 | |
4128 | if (mddev->sysfs_state) | |
4129 | sysfs_put(mddev->sysfs_state); | |
4130 | ||
4131 | if (mddev->gendisk) { | |
4132 | del_gendisk(mddev->gendisk); | |
4133 | put_disk(mddev->gendisk); | |
4134 | } | |
4135 | if (mddev->queue) | |
4136 | blk_cleanup_queue(mddev->queue); | |
4137 | ||
eae1701f N |
4138 | kfree(mddev); |
4139 | } | |
4140 | ||
4141 | static struct sysfs_ops md_sysfs_ops = { | |
4142 | .show = md_attr_show, | |
4143 | .store = md_attr_store, | |
4144 | }; | |
4145 | static struct kobj_type md_ktype = { | |
4146 | .release = md_free, | |
4147 | .sysfs_ops = &md_sysfs_ops, | |
4148 | .default_attrs = md_default_attrs, | |
4149 | }; | |
4150 | ||
1da177e4 LT |
4151 | int mdp_major = 0; |
4152 | ||
5fd3a17e DW |
4153 | static void mddev_delayed_delete(struct work_struct *ws) |
4154 | { | |
4155 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
4156 | ||
43a70507 | 4157 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4158 | kobject_del(&mddev->kobj); |
4159 | kobject_put(&mddev->kobj); | |
4160 | } | |
4161 | ||
efeb53c0 | 4162 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4163 | { |
48c9c27b | 4164 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
4165 | mddev_t *mddev = mddev_find(dev); |
4166 | struct gendisk *disk; | |
efeb53c0 N |
4167 | int partitioned; |
4168 | int shift; | |
4169 | int unit; | |
3830c62f | 4170 | int error; |
1da177e4 LT |
4171 | |
4172 | if (!mddev) | |
efeb53c0 N |
4173 | return -ENODEV; |
4174 | ||
4175 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4176 | shift = partitioned ? MdpMinorShift : 0; | |
4177 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4178 | |
d3374825 N |
4179 | /* wait for any previous instance if this device |
4180 | * to be completed removed (mddev_delayed_delete). | |
4181 | */ | |
4182 | flush_scheduled_work(); | |
4183 | ||
48c9c27b | 4184 | mutex_lock(&disks_mutex); |
0909dc44 N |
4185 | error = -EEXIST; |
4186 | if (mddev->gendisk) | |
4187 | goto abort; | |
efeb53c0 N |
4188 | |
4189 | if (name) { | |
4190 | /* Need to ensure that 'name' is not a duplicate. | |
4191 | */ | |
4192 | mddev_t *mddev2; | |
4193 | spin_lock(&all_mddevs_lock); | |
4194 | ||
4195 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4196 | if (mddev2->gendisk && | |
4197 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4198 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4199 | goto abort; |
efeb53c0 N |
4200 | } |
4201 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4202 | } |
8b765398 | 4203 | |
0909dc44 | 4204 | error = -ENOMEM; |
8b765398 | 4205 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4206 | if (!mddev->queue) |
4207 | goto abort; | |
409c57f3 N |
4208 | mddev->queue->queuedata = mddev; |
4209 | ||
8b765398 N |
4210 | /* Can be unlocked because the queue is new: no concurrency */ |
4211 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); | |
4212 | ||
409c57f3 | 4213 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 4214 | |
1da177e4 LT |
4215 | disk = alloc_disk(1 << shift); |
4216 | if (!disk) { | |
8b765398 N |
4217 | blk_cleanup_queue(mddev->queue); |
4218 | mddev->queue = NULL; | |
0909dc44 | 4219 | goto abort; |
1da177e4 | 4220 | } |
efeb53c0 | 4221 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4222 | disk->first_minor = unit << shift; |
efeb53c0 N |
4223 | if (name) |
4224 | strcpy(disk->disk_name, name); | |
4225 | else if (partitioned) | |
1da177e4 | 4226 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4227 | else |
1da177e4 | 4228 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4229 | disk->fops = &md_fops; |
4230 | disk->private_data = mddev; | |
4231 | disk->queue = mddev->queue; | |
92850bbd | 4232 | /* Allow extended partitions. This makes the |
d3374825 | 4233 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4234 | * remove it now. |
4235 | */ | |
4236 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 LT |
4237 | add_disk(disk); |
4238 | mddev->gendisk = disk; | |
ed9e1982 TH |
4239 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4240 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4241 | if (error) { |
4242 | /* This isn't possible, but as kobject_init_and_add is marked | |
4243 | * __must_check, we must do something with the result | |
4244 | */ | |
5e55e2f5 N |
4245 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4246 | disk->disk_name); | |
0909dc44 N |
4247 | error = 0; |
4248 | } | |
43a70507 N |
4249 | if (sysfs_create_group(&mddev->kobj, &md_bitmap_group)) |
4250 | printk(KERN_DEBUG "pointless warning\n"); | |
0909dc44 N |
4251 | abort: |
4252 | mutex_unlock(&disks_mutex); | |
4253 | if (!error) { | |
3830c62f | 4254 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
b62b7590 N |
4255 | mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state"); |
4256 | } | |
d3374825 | 4257 | mddev_put(mddev); |
0909dc44 | 4258 | return error; |
efeb53c0 N |
4259 | } |
4260 | ||
4261 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4262 | { | |
4263 | md_alloc(dev, NULL); | |
1da177e4 LT |
4264 | return NULL; |
4265 | } | |
4266 | ||
efeb53c0 N |
4267 | static int add_named_array(const char *val, struct kernel_param *kp) |
4268 | { | |
4269 | /* val must be "md_*" where * is not all digits. | |
4270 | * We allocate an array with a large free minor number, and | |
4271 | * set the name to val. val must not already be an active name. | |
4272 | */ | |
4273 | int len = strlen(val); | |
4274 | char buf[DISK_NAME_LEN]; | |
4275 | ||
4276 | while (len && val[len-1] == '\n') | |
4277 | len--; | |
4278 | if (len >= DISK_NAME_LEN) | |
4279 | return -E2BIG; | |
4280 | strlcpy(buf, val, len+1); | |
4281 | if (strncmp(buf, "md_", 3) != 0) | |
4282 | return -EINVAL; | |
4283 | return md_alloc(0, buf); | |
4284 | } | |
4285 | ||
1da177e4 LT |
4286 | static void md_safemode_timeout(unsigned long data) |
4287 | { | |
4288 | mddev_t *mddev = (mddev_t *) data; | |
4289 | ||
0fd62b86 NB |
4290 | if (!atomic_read(&mddev->writes_pending)) { |
4291 | mddev->safemode = 1; | |
4292 | if (mddev->external) | |
b62b7590 | 4293 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 | 4294 | } |
1da177e4 LT |
4295 | md_wakeup_thread(mddev->thread); |
4296 | } | |
4297 | ||
6ff8d8ec | 4298 | static int start_dirty_degraded; |
1da177e4 LT |
4299 | |
4300 | static int do_md_run(mddev_t * mddev) | |
4301 | { | |
2604b703 | 4302 | int err; |
1da177e4 LT |
4303 | mdk_rdev_t *rdev; |
4304 | struct gendisk *disk; | |
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 | disk = mddev->gendisk; |
1da177e4 LT |
4370 | |
4371 | spin_lock(&pers_lock); | |
d9d166c2 | 4372 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4373 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4374 | spin_unlock(&pers_lock); |
d9d166c2 N |
4375 | if (mddev->level != LEVEL_NONE) |
4376 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4377 | mddev->level); | |
4378 | else | |
4379 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4380 | mddev->clevel); | |
1da177e4 LT |
4381 | return -EINVAL; |
4382 | } | |
2604b703 | 4383 | mddev->pers = pers; |
1da177e4 | 4384 | spin_unlock(&pers_lock); |
34817e8c N |
4385 | if (mddev->level != pers->level) { |
4386 | mddev->level = pers->level; | |
4387 | mddev->new_level = pers->level; | |
4388 | } | |
d9d166c2 | 4389 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4390 | |
f6705578 | 4391 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4392 | pers->start_reshape == NULL) { |
f6705578 N |
4393 | /* This personality cannot handle reshaping... */ |
4394 | mddev->pers = NULL; | |
4395 | module_put(pers->owner); | |
4396 | return -EINVAL; | |
4397 | } | |
4398 | ||
7dd5e7c3 N |
4399 | if (pers->sync_request) { |
4400 | /* Warn if this is a potentially silly | |
4401 | * configuration. | |
4402 | */ | |
4403 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4404 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4405 | int warned = 0; |
159ec1fc CR |
4406 | |
4407 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4408 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4409 | if (rdev < rdev2 && |
4410 | rdev->bdev->bd_contains == | |
4411 | rdev2->bdev->bd_contains) { | |
4412 | printk(KERN_WARNING | |
4413 | "%s: WARNING: %s appears to be" | |
4414 | " on the same physical disk as" | |
4415 | " %s.\n", | |
4416 | mdname(mddev), | |
4417 | bdevname(rdev->bdev,b), | |
4418 | bdevname(rdev2->bdev,b2)); | |
4419 | warned = 1; | |
4420 | } | |
4421 | } | |
159ec1fc | 4422 | |
7dd5e7c3 N |
4423 | if (warned) |
4424 | printk(KERN_WARNING | |
4425 | "True protection against single-disk" | |
4426 | " failure might be compromised.\n"); | |
4427 | } | |
4428 | ||
657390d2 | 4429 | mddev->recovery = 0; |
58c0fed4 AN |
4430 | /* may be over-ridden by personality */ |
4431 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4432 | ||
a9701a30 | 4433 | mddev->barriers_work = 1; |
6ff8d8ec | 4434 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4435 | |
0f9552b5 | 4436 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
4437 | mddev->ro = 2; /* read-only, but switch on first write */ |
4438 | ||
b15c2e57 | 4439 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4440 | if (err) |
4441 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4442 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4443 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4444 | " but 'external_size' not in effect?\n", __func__); | |
4445 | printk(KERN_ERR | |
4446 | "md: invalid array_size %llu > default size %llu\n", | |
4447 | (unsigned long long)mddev->array_sectors / 2, | |
4448 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4449 | err = -EINVAL; | |
4450 | mddev->pers->stop(mddev); | |
4451 | } | |
4452 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4453 | err = bitmap_create(mddev); |
4454 | if (err) { | |
4455 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4456 | mdname(mddev), err); | |
4457 | mddev->pers->stop(mddev); | |
4458 | } | |
4459 | } | |
1da177e4 | 4460 | if (err) { |
1da177e4 LT |
4461 | module_put(mddev->pers->owner); |
4462 | mddev->pers = NULL; | |
32a7627c N |
4463 | bitmap_destroy(mddev); |
4464 | return err; | |
1da177e4 | 4465 | } |
5e55e2f5 N |
4466 | if (mddev->pers->sync_request) { |
4467 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
4468 | printk(KERN_WARNING | |
4469 | "md: cannot register extra attributes for %s\n", | |
4470 | mdname(mddev)); | |
0c3573f1 | 4471 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4472 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4473 | mddev->ro = 0; |
4474 | ||
1da177e4 | 4475 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
4476 | atomic_set(&mddev->max_corr_read_errors, |
4477 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
4478 | mddev->safemode = 0; |
4479 | mddev->safemode_timer.function = md_safemode_timeout; | |
4480 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4481 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4482 | mddev->in_sync = 1; |
86e6ffdd | 4483 | |
159ec1fc | 4484 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4485 | if (rdev->raid_disk >= 0) { |
4486 | char nm[20]; | |
4487 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
4488 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
4489 | printk("md: cannot register %s for %s\n", | |
4490 | nm, mdname(mddev)); | |
86e6ffdd | 4491 | } |
1da177e4 LT |
4492 | |
4493 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4494 | ||
850b2b42 N |
4495 | if (mddev->flags) |
4496 | md_update_sb(mddev, 0); | |
1da177e4 | 4497 | |
f233ea5c | 4498 | set_capacity(disk, mddev->array_sectors); |
1da177e4 | 4499 | |
0b8c9de0 N |
4500 | md_wakeup_thread(mddev->thread); |
4501 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4502 | |
449aad3e | 4503 | revalidate_disk(mddev->gendisk); |
44ce6294 | 4504 | mddev->changed = 1; |
d7603b7e | 4505 | md_new_event(mddev); |
b62b7590 | 4506 | sysfs_notify_dirent(mddev->sysfs_state); |
0c3573f1 N |
4507 | if (mddev->sysfs_action) |
4508 | sysfs_notify_dirent(mddev->sysfs_action); | |
a99ac971 | 4509 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
ed9e1982 | 4510 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
1da177e4 LT |
4511 | return 0; |
4512 | } | |
4513 | ||
4514 | static int restart_array(mddev_t *mddev) | |
4515 | { | |
4516 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4517 | |
80fab1d7 | 4518 | /* Complain if it has no devices */ |
1da177e4 | 4519 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4520 | return -ENXIO; |
4521 | if (!mddev->pers) | |
4522 | return -EINVAL; | |
4523 | if (!mddev->ro) | |
4524 | return -EBUSY; | |
4525 | mddev->safemode = 0; | |
4526 | mddev->ro = 0; | |
4527 | set_disk_ro(disk, 0); | |
4528 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4529 | mdname(mddev)); | |
4530 | /* Kick recovery or resync if necessary */ | |
4531 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4532 | md_wakeup_thread(mddev->thread); | |
4533 | md_wakeup_thread(mddev->sync_thread); | |
b62b7590 | 4534 | sysfs_notify_dirent(mddev->sysfs_state); |
80fab1d7 | 4535 | return 0; |
1da177e4 LT |
4536 | } |
4537 | ||
acc55e22 N |
4538 | /* similar to deny_write_access, but accounts for our holding a reference |
4539 | * to the file ourselves */ | |
4540 | static int deny_bitmap_write_access(struct file * file) | |
4541 | { | |
4542 | struct inode *inode = file->f_mapping->host; | |
4543 | ||
4544 | spin_lock(&inode->i_lock); | |
4545 | if (atomic_read(&inode->i_writecount) > 1) { | |
4546 | spin_unlock(&inode->i_lock); | |
4547 | return -ETXTBSY; | |
4548 | } | |
4549 | atomic_set(&inode->i_writecount, -1); | |
4550 | spin_unlock(&inode->i_lock); | |
4551 | ||
4552 | return 0; | |
4553 | } | |
4554 | ||
43a70507 | 4555 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
4556 | { |
4557 | struct inode *inode = file->f_mapping->host; | |
4558 | ||
4559 | spin_lock(&inode->i_lock); | |
4560 | atomic_set(&inode->i_writecount, 1); | |
4561 | spin_unlock(&inode->i_lock); | |
4562 | } | |
4563 | ||
9e653b63 N |
4564 | /* mode: |
4565 | * 0 - completely stop and dis-assemble array | |
4566 | * 1 - switch to readonly | |
4567 | * 2 - stop but do not disassemble array | |
4568 | */ | |
df5b20cf | 4569 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 LT |
4570 | { |
4571 | int err = 0; | |
4572 | struct gendisk *disk = mddev->gendisk; | |
c4647292 | 4573 | mdk_rdev_t *rdev; |
1da177e4 | 4574 | |
c8c00a69 | 4575 | mutex_lock(&mddev->open_mutex); |
f2ea68cf | 4576 | if (atomic_read(&mddev->openers) > is_open) { |
df5b20cf | 4577 | printk("md: %s still in use.\n",mdname(mddev)); |
c8c00a69 N |
4578 | err = -EBUSY; |
4579 | } else if (mddev->pers) { | |
1da177e4 LT |
4580 | |
4581 | if (mddev->sync_thread) { | |
5fd6c1dc | 4582 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 LT |
4583 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4584 | md_unregister_thread(mddev->sync_thread); | |
4585 | mddev->sync_thread = NULL; | |
4586 | } | |
4587 | ||
4588 | del_timer_sync(&mddev->safemode_timer); | |
4589 | ||
9e653b63 N |
4590 | switch(mode) { |
4591 | case 1: /* readonly */ | |
1da177e4 | 4592 | err = -ENXIO; |
f91de92e | 4593 | if (mddev->ro==1) |
1da177e4 LT |
4594 | goto out; |
4595 | mddev->ro = 1; | |
9e653b63 N |
4596 | break; |
4597 | case 0: /* disassemble */ | |
4598 | case 2: /* stop */ | |
6b8b3e8a | 4599 | bitmap_flush(mddev); |
a9701a30 | 4600 | md_super_wait(mddev); |
1da177e4 LT |
4601 | if (mddev->ro) |
4602 | set_disk_ro(disk, 0); | |
409c57f3 | 4603 | |
1da177e4 | 4604 | mddev->pers->stop(mddev); |
d1b5380c N |
4605 | mddev->queue->merge_bvec_fn = NULL; |
4606 | mddev->queue->unplug_fn = NULL; | |
041ae52e | 4607 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4608 | module_put(mddev->pers->owner); |
a64c876f N |
4609 | if (mddev->pers->sync_request && mddev->to_remove == NULL) |
4610 | mddev->to_remove = &md_redundancy_group; | |
1da177e4 | 4611 | mddev->pers = NULL; |
4f54b0e9 | 4612 | /* tell userspace to handle 'inactive' */ |
b62b7590 | 4613 | sysfs_notify_dirent(mddev->sysfs_state); |
0d4ca600 | 4614 | |
c4647292 N |
4615 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4616 | if (rdev->raid_disk >= 0) { | |
4617 | char nm[20]; | |
4618 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4619 | sysfs_remove_link(&mddev->kobj, nm); | |
4620 | } | |
4621 | ||
0d4ca600 | 4622 | set_capacity(disk, 0); |
44ce6294 | 4623 | mddev->changed = 1; |
0d4ca600 | 4624 | |
1da177e4 LT |
4625 | if (mddev->ro) |
4626 | mddev->ro = 0; | |
4627 | } | |
850b2b42 | 4628 | if (!mddev->in_sync || mddev->flags) { |
1da177e4 LT |
4629 | /* mark array as shutdown cleanly */ |
4630 | mddev->in_sync = 1; | |
850b2b42 | 4631 | md_update_sb(mddev, 1); |
1da177e4 | 4632 | } |
9e653b63 | 4633 | if (mode == 1) |
1da177e4 | 4634 | set_disk_ro(disk, 1); |
5fd6c1dc | 4635 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
80ffb3cc | 4636 | err = 0; |
1da177e4 | 4637 | } |
c8c00a69 N |
4638 | out: |
4639 | mutex_unlock(&mddev->open_mutex); | |
4640 | if (err) | |
4641 | return err; | |
1da177e4 LT |
4642 | /* |
4643 | * Free resources if final stop | |
4644 | */ | |
9e653b63 | 4645 | if (mode == 0) { |
0d4ca600 | 4646 | |
1da177e4 LT |
4647 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4648 | ||
978f946b | 4649 | bitmap_destroy(mddev); |
c3d9714e N |
4650 | if (mddev->bitmap_info.file) { |
4651 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
4652 | fput(mddev->bitmap_info.file); | |
4653 | mddev->bitmap_info.file = NULL; | |
978f946b | 4654 | } |
c3d9714e | 4655 | mddev->bitmap_info.offset = 0; |
978f946b | 4656 | |
1da177e4 LT |
4657 | export_array(mddev); |
4658 | ||
f233ea5c | 4659 | mddev->array_sectors = 0; |
b522adcd | 4660 | mddev->external_size = 0; |
58c0fed4 | 4661 | mddev->dev_sectors = 0; |
9e653b63 | 4662 | mddev->raid_disks = 0; |
a94213b1 | 4663 | mddev->recovery_cp = 0; |
5e96ee65 | 4664 | mddev->resync_min = 0; |
c6207277 | 4665 | mddev->resync_max = MaxSector; |
08a02ecd | 4666 | mddev->reshape_position = MaxSector; |
e691063a | 4667 | mddev->external = 0; |
1ec4a939 | 4668 | mddev->persistent = 0; |
d897dbf9 N |
4669 | mddev->level = LEVEL_NONE; |
4670 | mddev->clevel[0] = 0; | |
4671 | mddev->flags = 0; | |
4672 | mddev->ro = 0; | |
4673 | mddev->metadata_type[0] = 0; | |
9d8f0363 | 4674 | mddev->chunk_sectors = 0; |
d897dbf9 N |
4675 | mddev->ctime = mddev->utime = 0; |
4676 | mddev->layout = 0; | |
4677 | mddev->max_disks = 0; | |
4678 | mddev->events = 0; | |
4679 | mddev->delta_disks = 0; | |
4680 | mddev->new_level = LEVEL_NONE; | |
4681 | mddev->new_layout = 0; | |
664e7c41 | 4682 | mddev->new_chunk_sectors = 0; |
d897dbf9 N |
4683 | mddev->curr_resync = 0; |
4684 | mddev->resync_mismatches = 0; | |
4685 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4686 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4687 | mddev->recovery = 0; | |
4688 | mddev->in_sync = 0; | |
4689 | mddev->changed = 0; | |
4690 | mddev->degraded = 0; | |
4691 | mddev->barriers_work = 0; | |
4692 | mddev->safemode = 0; | |
c3d9714e N |
4693 | mddev->bitmap_info.offset = 0; |
4694 | mddev->bitmap_info.default_offset = 0; | |
42a04b50 N |
4695 | mddev->bitmap_info.chunksize = 0; |
4696 | mddev->bitmap_info.daemon_sleep = 0; | |
4697 | mddev->bitmap_info.max_write_behind = 0; | |
934d9c23 | 4698 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4699 | if (mddev->hold_active == UNTIL_STOP) |
4700 | mddev->hold_active = 0; | |
9e653b63 | 4701 | |
a8a55c38 | 4702 | } else if (mddev->pers) |
1da177e4 LT |
4703 | printk(KERN_INFO "md: %s switched to read-only mode.\n", |
4704 | mdname(mddev)); | |
4705 | err = 0; | |
3f9d99c1 | 4706 | blk_integrity_unregister(disk); |
d7603b7e | 4707 | md_new_event(mddev); |
b62b7590 | 4708 | sysfs_notify_dirent(mddev->sysfs_state); |
1da177e4 LT |
4709 | return err; |
4710 | } | |
4711 | ||
fdee8ae4 | 4712 | #ifndef MODULE |
1da177e4 LT |
4713 | static void autorun_array(mddev_t *mddev) |
4714 | { | |
4715 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4716 | int err; |
4717 | ||
a757e64c | 4718 | if (list_empty(&mddev->disks)) |
1da177e4 | 4719 | return; |
1da177e4 LT |
4720 | |
4721 | printk(KERN_INFO "md: running: "); | |
4722 | ||
159ec1fc | 4723 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4724 | char b[BDEVNAME_SIZE]; |
4725 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4726 | } | |
4727 | printk("\n"); | |
4728 | ||
d710e138 | 4729 | err = do_md_run(mddev); |
1da177e4 LT |
4730 | if (err) { |
4731 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4732 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4733 | } |
4734 | } | |
4735 | ||
4736 | /* | |
4737 | * lets try to run arrays based on all disks that have arrived | |
4738 | * until now. (those are in pending_raid_disks) | |
4739 | * | |
4740 | * the method: pick the first pending disk, collect all disks with | |
4741 | * the same UUID, remove all from the pending list and put them into | |
4742 | * the 'same_array' list. Then order this list based on superblock | |
4743 | * update time (freshest comes first), kick out 'old' disks and | |
4744 | * compare superblocks. If everything's fine then run it. | |
4745 | * | |
4746 | * If "unit" is allocated, then bump its reference count | |
4747 | */ | |
4748 | static void autorun_devices(int part) | |
4749 | { | |
159ec1fc | 4750 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4751 | mddev_t *mddev; |
4752 | char b[BDEVNAME_SIZE]; | |
4753 | ||
4754 | printk(KERN_INFO "md: autorun ...\n"); | |
4755 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4756 | int unit; |
1da177e4 | 4757 | dev_t dev; |
ad01c9e3 | 4758 | LIST_HEAD(candidates); |
1da177e4 LT |
4759 | rdev0 = list_entry(pending_raid_disks.next, |
4760 | mdk_rdev_t, same_set); | |
4761 | ||
4762 | printk(KERN_INFO "md: considering %s ...\n", | |
4763 | bdevname(rdev0->bdev,b)); | |
4764 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4765 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4766 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4767 | printk(KERN_INFO "md: adding %s ...\n", | |
4768 | bdevname(rdev->bdev,b)); | |
4769 | list_move(&rdev->same_set, &candidates); | |
4770 | } | |
4771 | /* | |
4772 | * now we have a set of devices, with all of them having | |
4773 | * mostly sane superblocks. It's time to allocate the | |
4774 | * mddev. | |
4775 | */ | |
e8703fe1 N |
4776 | if (part) { |
4777 | dev = MKDEV(mdp_major, | |
4778 | rdev0->preferred_minor << MdpMinorShift); | |
4779 | unit = MINOR(dev) >> MdpMinorShift; | |
4780 | } else { | |
4781 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4782 | unit = MINOR(dev); | |
4783 | } | |
4784 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4785 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4786 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4787 | break; | |
4788 | } | |
1da177e4 LT |
4789 | |
4790 | md_probe(dev, NULL, NULL); | |
4791 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4792 | if (!mddev || !mddev->gendisk) { |
4793 | if (mddev) | |
4794 | mddev_put(mddev); | |
4795 | printk(KERN_ERR | |
1da177e4 LT |
4796 | "md: cannot allocate memory for md drive.\n"); |
4797 | break; | |
4798 | } | |
4799 | if (mddev_lock(mddev)) | |
4800 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4801 | mdname(mddev)); | |
4802 | else if (mddev->raid_disks || mddev->major_version | |
4803 | || !list_empty(&mddev->disks)) { | |
4804 | printk(KERN_WARNING | |
4805 | "md: %s already running, cannot run %s\n", | |
4806 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4807 | mddev_unlock(mddev); | |
4808 | } else { | |
4809 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4810 | mddev->persistent = 1; |
159ec1fc | 4811 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4812 | list_del_init(&rdev->same_set); |
4813 | if (bind_rdev_to_array(rdev, mddev)) | |
4814 | export_rdev(rdev); | |
4815 | } | |
4816 | autorun_array(mddev); | |
4817 | mddev_unlock(mddev); | |
4818 | } | |
4819 | /* on success, candidates will be empty, on error | |
4820 | * it won't... | |
4821 | */ | |
159ec1fc | 4822 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4823 | list_del_init(&rdev->same_set); |
1da177e4 | 4824 | export_rdev(rdev); |
4b80991c | 4825 | } |
1da177e4 LT |
4826 | mddev_put(mddev); |
4827 | } | |
4828 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4829 | } | |
fdee8ae4 | 4830 | #endif /* !MODULE */ |
1da177e4 | 4831 | |
1da177e4 LT |
4832 | static int get_version(void __user * arg) |
4833 | { | |
4834 | mdu_version_t ver; | |
4835 | ||
4836 | ver.major = MD_MAJOR_VERSION; | |
4837 | ver.minor = MD_MINOR_VERSION; | |
4838 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4839 | ||
4840 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4841 | return -EFAULT; | |
4842 | ||
4843 | return 0; | |
4844 | } | |
4845 | ||
4846 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4847 | { | |
4848 | mdu_array_info_t info; | |
a9f326eb | 4849 | int nr,working,insync,failed,spare; |
1da177e4 | 4850 | mdk_rdev_t *rdev; |
1da177e4 | 4851 | |
a9f326eb | 4852 | nr=working=insync=failed=spare=0; |
159ec1fc | 4853 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4854 | nr++; |
b2d444d7 | 4855 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4856 | failed++; |
4857 | else { | |
4858 | working++; | |
b2d444d7 | 4859 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 4860 | insync++; |
1da177e4 LT |
4861 | else |
4862 | spare++; | |
4863 | } | |
4864 | } | |
4865 | ||
4866 | info.major_version = mddev->major_version; | |
4867 | info.minor_version = mddev->minor_version; | |
4868 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
4869 | info.ctime = mddev->ctime; | |
4870 | info.level = mddev->level; | |
58c0fed4 AN |
4871 | info.size = mddev->dev_sectors / 2; |
4872 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 4873 | info.size = -1; |
1da177e4 LT |
4874 | info.nr_disks = nr; |
4875 | info.raid_disks = mddev->raid_disks; | |
4876 | info.md_minor = mddev->md_minor; | |
4877 | info.not_persistent= !mddev->persistent; | |
4878 | ||
4879 | info.utime = mddev->utime; | |
4880 | info.state = 0; | |
4881 | if (mddev->in_sync) | |
4882 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 4883 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 4884 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 4885 | info.active_disks = insync; |
1da177e4 LT |
4886 | info.working_disks = working; |
4887 | info.failed_disks = failed; | |
4888 | info.spare_disks = spare; | |
4889 | ||
4890 | info.layout = mddev->layout; | |
9d8f0363 | 4891 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
4892 | |
4893 | if (copy_to_user(arg, &info, sizeof(info))) | |
4894 | return -EFAULT; | |
4895 | ||
4896 | return 0; | |
4897 | } | |
4898 | ||
87162a28 | 4899 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
4900 | { |
4901 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
4902 | char *ptr, *buf = NULL; | |
4903 | int err = -ENOMEM; | |
4904 | ||
b5470dc5 DW |
4905 | if (md_allow_write(mddev)) |
4906 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
4907 | else | |
4908 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 4909 | |
32a7627c N |
4910 | if (!file) |
4911 | goto out; | |
4912 | ||
4913 | /* bitmap disabled, zero the first byte and copy out */ | |
4914 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
4915 | file->pathname[0] = '\0'; | |
4916 | goto copy_out; | |
4917 | } | |
4918 | ||
4919 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
4920 | if (!buf) | |
4921 | goto out; | |
4922 | ||
6bcfd601 CH |
4923 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
4924 | if (IS_ERR(ptr)) | |
32a7627c N |
4925 | goto out; |
4926 | ||
4927 | strcpy(file->pathname, ptr); | |
4928 | ||
4929 | copy_out: | |
4930 | err = 0; | |
4931 | if (copy_to_user(arg, file, sizeof(*file))) | |
4932 | err = -EFAULT; | |
4933 | out: | |
4934 | kfree(buf); | |
4935 | kfree(file); | |
4936 | return err; | |
4937 | } | |
4938 | ||
1da177e4 LT |
4939 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
4940 | { | |
4941 | mdu_disk_info_t info; | |
1da177e4 LT |
4942 | mdk_rdev_t *rdev; |
4943 | ||
4944 | if (copy_from_user(&info, arg, sizeof(info))) | |
4945 | return -EFAULT; | |
4946 | ||
26ef379f | 4947 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
4948 | if (rdev) { |
4949 | info.major = MAJOR(rdev->bdev->bd_dev); | |
4950 | info.minor = MINOR(rdev->bdev->bd_dev); | |
4951 | info.raid_disk = rdev->raid_disk; | |
4952 | info.state = 0; | |
b2d444d7 | 4953 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4954 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 4955 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4956 | info.state |= (1<<MD_DISK_ACTIVE); |
4957 | info.state |= (1<<MD_DISK_SYNC); | |
4958 | } | |
8ddf9efe N |
4959 | if (test_bit(WriteMostly, &rdev->flags)) |
4960 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
4961 | } else { |
4962 | info.major = info.minor = 0; | |
4963 | info.raid_disk = -1; | |
4964 | info.state = (1<<MD_DISK_REMOVED); | |
4965 | } | |
4966 | ||
4967 | if (copy_to_user(arg, &info, sizeof(info))) | |
4968 | return -EFAULT; | |
4969 | ||
4970 | return 0; | |
4971 | } | |
4972 | ||
4973 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
4974 | { | |
4975 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4976 | mdk_rdev_t *rdev; | |
4977 | dev_t dev = MKDEV(info->major,info->minor); | |
4978 | ||
4979 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
4980 | return -EOVERFLOW; | |
4981 | ||
4982 | if (!mddev->raid_disks) { | |
4983 | int err; | |
4984 | /* expecting a device which has a superblock */ | |
4985 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
4986 | if (IS_ERR(rdev)) { | |
4987 | printk(KERN_WARNING | |
4988 | "md: md_import_device returned %ld\n", | |
4989 | PTR_ERR(rdev)); | |
4990 | return PTR_ERR(rdev); | |
4991 | } | |
4992 | if (!list_empty(&mddev->disks)) { | |
4993 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
4994 | mdk_rdev_t, same_set); | |
a9f326eb | 4995 | err = super_types[mddev->major_version] |
1da177e4 LT |
4996 | .load_super(rdev, rdev0, mddev->minor_version); |
4997 | if (err < 0) { | |
4998 | printk(KERN_WARNING | |
4999 | "md: %s has different UUID to %s\n", | |
5000 | bdevname(rdev->bdev,b), | |
5001 | bdevname(rdev0->bdev,b2)); | |
5002 | export_rdev(rdev); | |
5003 | return -EINVAL; | |
5004 | } | |
5005 | } | |
5006 | err = bind_rdev_to_array(rdev, mddev); | |
5007 | if (err) | |
5008 | export_rdev(rdev); | |
5009 | return err; | |
5010 | } | |
5011 | ||
5012 | /* | |
5013 | * add_new_disk can be used once the array is assembled | |
5014 | * to add "hot spares". They must already have a superblock | |
5015 | * written | |
5016 | */ | |
5017 | if (mddev->pers) { | |
5018 | int err; | |
5019 | if (!mddev->pers->hot_add_disk) { | |
5020 | printk(KERN_WARNING | |
5021 | "%s: personality does not support diskops!\n", | |
5022 | mdname(mddev)); | |
5023 | return -EINVAL; | |
5024 | } | |
7b1e35f6 N |
5025 | if (mddev->persistent) |
5026 | rdev = md_import_device(dev, mddev->major_version, | |
5027 | mddev->minor_version); | |
5028 | else | |
5029 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5030 | if (IS_ERR(rdev)) { |
5031 | printk(KERN_WARNING | |
5032 | "md: md_import_device returned %ld\n", | |
5033 | PTR_ERR(rdev)); | |
5034 | return PTR_ERR(rdev); | |
5035 | } | |
41158c7e N |
5036 | /* set save_raid_disk if appropriate */ |
5037 | if (!mddev->persistent) { | |
5038 | if (info->state & (1<<MD_DISK_SYNC) && | |
5039 | info->raid_disk < mddev->raid_disks) | |
5040 | rdev->raid_disk = info->raid_disk; | |
5041 | else | |
5042 | rdev->raid_disk = -1; | |
5043 | } else | |
5044 | super_types[mddev->major_version]. | |
5045 | validate_super(mddev, rdev); | |
5046 | rdev->saved_raid_disk = rdev->raid_disk; | |
5047 | ||
b2d444d7 | 5048 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5049 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5050 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5051 | else |
5052 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5053 | |
1da177e4 LT |
5054 | rdev->raid_disk = -1; |
5055 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5056 | if (!err && !mddev->pers->hot_remove_disk) { |
5057 | /* If there is hot_add_disk but no hot_remove_disk | |
5058 | * then added disks for geometry changes, | |
5059 | * and should be added immediately. | |
5060 | */ | |
5061 | super_types[mddev->major_version]. | |
5062 | validate_super(mddev, rdev); | |
5063 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5064 | if (err) | |
5065 | unbind_rdev_from_array(rdev); | |
5066 | } | |
1da177e4 LT |
5067 | if (err) |
5068 | export_rdev(rdev); | |
52664732 | 5069 | else |
3c0ee63a | 5070 | sysfs_notify_dirent(rdev->sysfs_state); |
c361777f | 5071 | |
17571284 | 5072 | md_update_sb(mddev, 1); |
72a23c21 NB |
5073 | if (mddev->degraded) |
5074 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5075 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 5076 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5077 | return err; |
5078 | } | |
5079 | ||
5080 | /* otherwise, add_new_disk is only allowed | |
5081 | * for major_version==0 superblocks | |
5082 | */ | |
5083 | if (mddev->major_version != 0) { | |
5084 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5085 | mdname(mddev)); | |
5086 | return -EINVAL; | |
5087 | } | |
5088 | ||
5089 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5090 | int err; | |
d710e138 | 5091 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5092 | if (IS_ERR(rdev)) { |
5093 | printk(KERN_WARNING | |
5094 | "md: error, md_import_device() returned %ld\n", | |
5095 | PTR_ERR(rdev)); | |
5096 | return PTR_ERR(rdev); | |
5097 | } | |
5098 | rdev->desc_nr = info->number; | |
5099 | if (info->raid_disk < mddev->raid_disks) | |
5100 | rdev->raid_disk = info->raid_disk; | |
5101 | else | |
5102 | rdev->raid_disk = -1; | |
5103 | ||
1da177e4 | 5104 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5105 | if (info->state & (1<<MD_DISK_SYNC)) |
5106 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5107 | |
8ddf9efe N |
5108 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5109 | set_bit(WriteMostly, &rdev->flags); | |
5110 | ||
1da177e4 LT |
5111 | if (!mddev->persistent) { |
5112 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
0f420358 | 5113 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 5114 | } else |
0f420358 | 5115 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
8190e754 | 5116 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5117 | |
2bf071bf N |
5118 | err = bind_rdev_to_array(rdev, mddev); |
5119 | if (err) { | |
5120 | export_rdev(rdev); | |
5121 | return err; | |
5122 | } | |
1da177e4 LT |
5123 | } |
5124 | ||
5125 | return 0; | |
5126 | } | |
5127 | ||
5128 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
5129 | { | |
5130 | char b[BDEVNAME_SIZE]; | |
5131 | mdk_rdev_t *rdev; | |
5132 | ||
1da177e4 LT |
5133 | rdev = find_rdev(mddev, dev); |
5134 | if (!rdev) | |
5135 | return -ENXIO; | |
5136 | ||
5137 | if (rdev->raid_disk >= 0) | |
5138 | goto busy; | |
5139 | ||
5140 | kick_rdev_from_array(rdev); | |
850b2b42 | 5141 | md_update_sb(mddev, 1); |
d7603b7e | 5142 | md_new_event(mddev); |
1da177e4 LT |
5143 | |
5144 | return 0; | |
5145 | busy: | |
fdefa4d8 | 5146 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5147 | bdevname(rdev->bdev,b), mdname(mddev)); |
5148 | return -EBUSY; | |
5149 | } | |
5150 | ||
5151 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
5152 | { | |
5153 | char b[BDEVNAME_SIZE]; | |
5154 | int err; | |
1da177e4 LT |
5155 | mdk_rdev_t *rdev; |
5156 | ||
5157 | if (!mddev->pers) | |
5158 | return -ENODEV; | |
5159 | ||
5160 | if (mddev->major_version != 0) { | |
5161 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5162 | " version-0 superblocks.\n", | |
5163 | mdname(mddev)); | |
5164 | return -EINVAL; | |
5165 | } | |
5166 | if (!mddev->pers->hot_add_disk) { | |
5167 | printk(KERN_WARNING | |
5168 | "%s: personality does not support diskops!\n", | |
5169 | mdname(mddev)); | |
5170 | return -EINVAL; | |
5171 | } | |
5172 | ||
d710e138 | 5173 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5174 | if (IS_ERR(rdev)) { |
5175 | printk(KERN_WARNING | |
5176 | "md: error, md_import_device() returned %ld\n", | |
5177 | PTR_ERR(rdev)); | |
5178 | return -EINVAL; | |
5179 | } | |
5180 | ||
5181 | if (mddev->persistent) | |
0f420358 | 5182 | rdev->sb_start = calc_dev_sboffset(rdev->bdev); |
1da177e4 | 5183 | else |
0f420358 | 5184 | rdev->sb_start = rdev->bdev->bd_inode->i_size / 512; |
1da177e4 | 5185 | |
8190e754 | 5186 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5187 | |
b2d444d7 | 5188 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5189 | printk(KERN_WARNING |
5190 | "md: can not hot-add faulty %s disk to %s!\n", | |
5191 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5192 | err = -EINVAL; | |
5193 | goto abort_export; | |
5194 | } | |
b2d444d7 | 5195 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5196 | rdev->desc_nr = -1; |
5842730d | 5197 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5198 | err = bind_rdev_to_array(rdev, mddev); |
5199 | if (err) | |
5200 | goto abort_export; | |
1da177e4 LT |
5201 | |
5202 | /* | |
5203 | * The rest should better be atomic, we can have disk failures | |
5204 | * noticed in interrupt contexts ... | |
5205 | */ | |
5206 | ||
1da177e4 LT |
5207 | rdev->raid_disk = -1; |
5208 | ||
850b2b42 | 5209 | md_update_sb(mddev, 1); |
1da177e4 LT |
5210 | |
5211 | /* | |
5212 | * Kick recovery, maybe this spare has to be added to the | |
5213 | * array immediately. | |
5214 | */ | |
5215 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5216 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5217 | md_new_event(mddev); |
1da177e4 LT |
5218 | return 0; |
5219 | ||
1da177e4 LT |
5220 | abort_export: |
5221 | export_rdev(rdev); | |
5222 | return err; | |
5223 | } | |
5224 | ||
32a7627c N |
5225 | static int set_bitmap_file(mddev_t *mddev, int fd) |
5226 | { | |
5227 | int err; | |
5228 | ||
36fa3063 N |
5229 | if (mddev->pers) { |
5230 | if (!mddev->pers->quiesce) | |
5231 | return -EBUSY; | |
5232 | if (mddev->recovery || mddev->sync_thread) | |
5233 | return -EBUSY; | |
5234 | /* we should be able to change the bitmap.. */ | |
5235 | } | |
32a7627c | 5236 | |
32a7627c | 5237 | |
36fa3063 N |
5238 | if (fd >= 0) { |
5239 | if (mddev->bitmap) | |
5240 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5241 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5242 | |
c3d9714e | 5243 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5244 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5245 | mdname(mddev)); | |
5246 | return -EBADF; | |
5247 | } | |
5248 | ||
c3d9714e | 5249 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
5250 | if (err) { |
5251 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5252 | mdname(mddev)); | |
c3d9714e N |
5253 | fput(mddev->bitmap_info.file); |
5254 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5255 | return err; |
5256 | } | |
c3d9714e | 5257 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5258 | } else if (mddev->bitmap == NULL) |
5259 | return -ENOENT; /* cannot remove what isn't there */ | |
5260 | err = 0; | |
5261 | if (mddev->pers) { | |
5262 | mddev->pers->quiesce(mddev, 1); | |
5263 | if (fd >= 0) | |
5264 | err = bitmap_create(mddev); | |
d7375ab3 | 5265 | if (fd < 0 || err) { |
36fa3063 | 5266 | bitmap_destroy(mddev); |
d7375ab3 N |
5267 | fd = -1; /* make sure to put the file */ |
5268 | } | |
36fa3063 | 5269 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5270 | } |
5271 | if (fd < 0) { | |
c3d9714e N |
5272 | if (mddev->bitmap_info.file) { |
5273 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5274 | fput(mddev->bitmap_info.file); | |
acc55e22 | 5275 | } |
c3d9714e | 5276 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
5277 | } |
5278 | ||
32a7627c N |
5279 | return err; |
5280 | } | |
5281 | ||
1da177e4 LT |
5282 | /* |
5283 | * set_array_info is used two different ways | |
5284 | * The original usage is when creating a new array. | |
5285 | * In this usage, raid_disks is > 0 and it together with | |
5286 | * level, size, not_persistent,layout,chunksize determine the | |
5287 | * shape of the array. | |
5288 | * This will always create an array with a type-0.90.0 superblock. | |
5289 | * The newer usage is when assembling an array. | |
5290 | * In this case raid_disks will be 0, and the major_version field is | |
5291 | * use to determine which style super-blocks are to be found on the devices. | |
5292 | * The minor and patch _version numbers are also kept incase the | |
5293 | * super_block handler wishes to interpret them. | |
5294 | */ | |
5295 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5296 | { | |
5297 | ||
5298 | if (info->raid_disks == 0) { | |
5299 | /* just setting version number for superblock loading */ | |
5300 | if (info->major_version < 0 || | |
50511da3 | 5301 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5302 | super_types[info->major_version].name == NULL) { |
5303 | /* maybe try to auto-load a module? */ | |
5304 | printk(KERN_INFO | |
5305 | "md: superblock version %d not known\n", | |
5306 | info->major_version); | |
5307 | return -EINVAL; | |
5308 | } | |
5309 | mddev->major_version = info->major_version; | |
5310 | mddev->minor_version = info->minor_version; | |
5311 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5312 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
5313 | /* ensure mddev_put doesn't delete this now that there |
5314 | * is some minimal configuration. | |
5315 | */ | |
5316 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
5317 | return 0; |
5318 | } | |
5319 | mddev->major_version = MD_MAJOR_VERSION; | |
5320 | mddev->minor_version = MD_MINOR_VERSION; | |
5321 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5322 | mddev->ctime = get_seconds(); | |
5323 | ||
5324 | mddev->level = info->level; | |
17115e03 | 5325 | mddev->clevel[0] = 0; |
58c0fed4 | 5326 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5327 | mddev->raid_disks = info->raid_disks; |
5328 | /* don't set md_minor, it is determined by which /dev/md* was | |
5329 | * openned | |
5330 | */ | |
5331 | if (info->state & (1<<MD_SB_CLEAN)) | |
5332 | mddev->recovery_cp = MaxSector; | |
5333 | else | |
5334 | mddev->recovery_cp = 0; | |
5335 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5336 | mddev->external = 0; |
1da177e4 LT |
5337 | |
5338 | mddev->layout = info->layout; | |
9d8f0363 | 5339 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5340 | |
5341 | mddev->max_disks = MD_SB_DISKS; | |
5342 | ||
e691063a N |
5343 | if (mddev->persistent) |
5344 | mddev->flags = 0; | |
850b2b42 | 5345 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5346 | |
c3d9714e N |
5347 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
5348 | mddev->bitmap_info.offset = 0; | |
b2a2703c | 5349 | |
f6705578 N |
5350 | mddev->reshape_position = MaxSector; |
5351 | ||
1da177e4 LT |
5352 | /* |
5353 | * Generate a 128 bit UUID | |
5354 | */ | |
5355 | get_random_bytes(mddev->uuid, 16); | |
5356 | ||
f6705578 | 5357 | mddev->new_level = mddev->level; |
664e7c41 | 5358 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5359 | mddev->new_layout = mddev->layout; |
5360 | mddev->delta_disks = 0; | |
5361 | ||
1da177e4 LT |
5362 | return 0; |
5363 | } | |
5364 | ||
1f403624 DW |
5365 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5366 | { | |
b522adcd DW |
5367 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5368 | ||
5369 | if (mddev->external_size) | |
5370 | return; | |
5371 | ||
1f403624 DW |
5372 | mddev->array_sectors = array_sectors; |
5373 | } | |
5374 | EXPORT_SYMBOL(md_set_array_sectors); | |
5375 | ||
d71f9f88 | 5376 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5377 | { |
159ec1fc | 5378 | mdk_rdev_t *rdev; |
a35b0d69 | 5379 | int rv; |
d71f9f88 | 5380 | int fit = (num_sectors == 0); |
a35b0d69 N |
5381 | |
5382 | if (mddev->pers->resize == NULL) | |
5383 | return -EINVAL; | |
d71f9f88 AN |
5384 | /* The "num_sectors" is the number of sectors of each device that |
5385 | * is used. This can only make sense for arrays with redundancy. | |
5386 | * linear and raid0 always use whatever space is available. We can only | |
5387 | * consider changing this number if no resync or reconstruction is | |
5388 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5389 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5390 | * of each device. If num_sectors is zero, we find the largest size |
5391 | * that fits. | |
5392 | ||
a35b0d69 N |
5393 | */ |
5394 | if (mddev->sync_thread) | |
5395 | return -EBUSY; | |
dba034ee N |
5396 | if (mddev->bitmap) |
5397 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5398 | * grow, and re-add. | |
5399 | */ | |
5400 | return -EBUSY; | |
159ec1fc | 5401 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5402 | sector_t avail = rdev->sectors; |
01ab5662 | 5403 | |
d71f9f88 AN |
5404 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5405 | num_sectors = avail; | |
5406 | if (avail < num_sectors) | |
a35b0d69 N |
5407 | return -ENOSPC; |
5408 | } | |
d71f9f88 | 5409 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5410 | if (!rv) |
5411 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5412 | return rv; |
5413 | } | |
5414 | ||
da943b99 N |
5415 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5416 | { | |
5417 | int rv; | |
5418 | /* change the number of raid disks */ | |
63c70c4f | 5419 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5420 | return -EINVAL; |
5421 | if (raid_disks <= 0 || | |
233fca36 | 5422 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 5423 | return -EINVAL; |
63c70c4f | 5424 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5425 | return -EBUSY; |
63c70c4f N |
5426 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5427 | ||
5428 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
5429 | return rv; |
5430 | } | |
5431 | ||
5432 | ||
1da177e4 LT |
5433 | /* |
5434 | * update_array_info is used to change the configuration of an | |
5435 | * on-line array. | |
5436 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5437 | * fields in the info are checked against the array. | |
5438 | * Any differences that cannot be handled will cause an error. | |
5439 | * Normally, only one change can be managed at a time. | |
5440 | */ | |
5441 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5442 | { | |
5443 | int rv = 0; | |
5444 | int cnt = 0; | |
36fa3063 N |
5445 | int state = 0; |
5446 | ||
5447 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 5448 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5449 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
5450 | |
5451 | if (mddev->major_version != info->major_version || | |
5452 | mddev->minor_version != info->minor_version || | |
5453 | /* mddev->patch_version != info->patch_version || */ | |
5454 | mddev->ctime != info->ctime || | |
5455 | mddev->level != info->level || | |
5456 | /* mddev->layout != info->layout || */ | |
5457 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5458 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5459 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5460 | ((state^info->state) & 0xfffffe00) | |
5461 | ) | |
1da177e4 LT |
5462 | return -EINVAL; |
5463 | /* Check there is only one change */ | |
58c0fed4 AN |
5464 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5465 | cnt++; | |
5466 | if (mddev->raid_disks != info->raid_disks) | |
5467 | cnt++; | |
5468 | if (mddev->layout != info->layout) | |
5469 | cnt++; | |
5470 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5471 | cnt++; | |
5472 | if (cnt == 0) | |
5473 | return 0; | |
5474 | if (cnt > 1) | |
5475 | return -EINVAL; | |
1da177e4 LT |
5476 | |
5477 | if (mddev->layout != info->layout) { | |
5478 | /* Change layout | |
5479 | * we don't need to do anything at the md level, the | |
5480 | * personality will take care of it all. | |
5481 | */ | |
50ac168a | 5482 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5483 | return -EINVAL; |
597a711b N |
5484 | else { |
5485 | mddev->new_layout = info->layout; | |
50ac168a | 5486 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5487 | if (rv) |
5488 | mddev->new_layout = mddev->layout; | |
5489 | return rv; | |
5490 | } | |
1da177e4 | 5491 | } |
58c0fed4 | 5492 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5493 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5494 | |
da943b99 N |
5495 | if (mddev->raid_disks != info->raid_disks) |
5496 | rv = update_raid_disks(mddev, info->raid_disks); | |
5497 | ||
36fa3063 N |
5498 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5499 | if (mddev->pers->quiesce == NULL) | |
5500 | return -EINVAL; | |
5501 | if (mddev->recovery || mddev->sync_thread) | |
5502 | return -EBUSY; | |
5503 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5504 | /* add the bitmap */ | |
5505 | if (mddev->bitmap) | |
5506 | return -EEXIST; | |
c3d9714e | 5507 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 5508 | return -EINVAL; |
c3d9714e N |
5509 | mddev->bitmap_info.offset = |
5510 | mddev->bitmap_info.default_offset; | |
36fa3063 N |
5511 | mddev->pers->quiesce(mddev, 1); |
5512 | rv = bitmap_create(mddev); | |
5513 | if (rv) | |
5514 | bitmap_destroy(mddev); | |
5515 | mddev->pers->quiesce(mddev, 0); | |
5516 | } else { | |
5517 | /* remove the bitmap */ | |
5518 | if (!mddev->bitmap) | |
5519 | return -ENOENT; | |
5520 | if (mddev->bitmap->file) | |
5521 | return -EINVAL; | |
5522 | mddev->pers->quiesce(mddev, 1); | |
5523 | bitmap_destroy(mddev); | |
5524 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 5525 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
5526 | } |
5527 | } | |
850b2b42 | 5528 | md_update_sb(mddev, 1); |
1da177e4 LT |
5529 | return rv; |
5530 | } | |
5531 | ||
5532 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5533 | { | |
5534 | mdk_rdev_t *rdev; | |
5535 | ||
5536 | if (mddev->pers == NULL) | |
5537 | return -ENODEV; | |
5538 | ||
5539 | rdev = find_rdev(mddev, dev); | |
5540 | if (!rdev) | |
5541 | return -ENODEV; | |
5542 | ||
5543 | md_error(mddev, rdev); | |
5544 | return 0; | |
5545 | } | |
5546 | ||
2f9618ce AN |
5547 | /* |
5548 | * We have a problem here : there is no easy way to give a CHS | |
5549 | * virtual geometry. We currently pretend that we have a 2 heads | |
5550 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5551 | * dosfs just mad... ;-) | |
5552 | */ | |
a885c8c4 CH |
5553 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5554 | { | |
5555 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5556 | ||
5557 | geo->heads = 2; | |
5558 | geo->sectors = 4; | |
5559 | geo->cylinders = get_capacity(mddev->gendisk) / 8; | |
5560 | return 0; | |
5561 | } | |
5562 | ||
a39907fa | 5563 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5564 | unsigned int cmd, unsigned long arg) |
5565 | { | |
5566 | int err = 0; | |
5567 | void __user *argp = (void __user *)arg; | |
1da177e4 | 5568 | mddev_t *mddev = NULL; |
e2218350 | 5569 | int ro; |
1da177e4 LT |
5570 | |
5571 | if (!capable(CAP_SYS_ADMIN)) | |
5572 | return -EACCES; | |
5573 | ||
5574 | /* | |
5575 | * Commands dealing with the RAID driver but not any | |
5576 | * particular array: | |
5577 | */ | |
5578 | switch (cmd) | |
5579 | { | |
5580 | case RAID_VERSION: | |
5581 | err = get_version(argp); | |
5582 | goto done; | |
5583 | ||
5584 | case PRINT_RAID_DEBUG: | |
5585 | err = 0; | |
5586 | md_print_devices(); | |
5587 | goto done; | |
5588 | ||
5589 | #ifndef MODULE | |
5590 | case RAID_AUTORUN: | |
5591 | err = 0; | |
5592 | autostart_arrays(arg); | |
5593 | goto done; | |
5594 | #endif | |
5595 | default:; | |
5596 | } | |
5597 | ||
5598 | /* | |
5599 | * Commands creating/starting a new array: | |
5600 | */ | |
5601 | ||
a39907fa | 5602 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5603 | |
5604 | if (!mddev) { | |
5605 | BUG(); | |
5606 | goto abort; | |
5607 | } | |
5608 | ||
1da177e4 LT |
5609 | err = mddev_lock(mddev); |
5610 | if (err) { | |
5611 | printk(KERN_INFO | |
5612 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5613 | err, cmd); | |
5614 | goto abort; | |
5615 | } | |
5616 | ||
5617 | switch (cmd) | |
5618 | { | |
5619 | case SET_ARRAY_INFO: | |
5620 | { | |
5621 | mdu_array_info_t info; | |
5622 | if (!arg) | |
5623 | memset(&info, 0, sizeof(info)); | |
5624 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5625 | err = -EFAULT; | |
5626 | goto abort_unlock; | |
5627 | } | |
5628 | if (mddev->pers) { | |
5629 | err = update_array_info(mddev, &info); | |
5630 | if (err) { | |
5631 | printk(KERN_WARNING "md: couldn't update" | |
5632 | " array info. %d\n", err); | |
5633 | goto abort_unlock; | |
5634 | } | |
5635 | goto done_unlock; | |
5636 | } | |
5637 | if (!list_empty(&mddev->disks)) { | |
5638 | printk(KERN_WARNING | |
5639 | "md: array %s already has disks!\n", | |
5640 | mdname(mddev)); | |
5641 | err = -EBUSY; | |
5642 | goto abort_unlock; | |
5643 | } | |
5644 | if (mddev->raid_disks) { | |
5645 | printk(KERN_WARNING | |
5646 | "md: array %s already initialised!\n", | |
5647 | mdname(mddev)); | |
5648 | err = -EBUSY; | |
5649 | goto abort_unlock; | |
5650 | } | |
5651 | err = set_array_info(mddev, &info); | |
5652 | if (err) { | |
5653 | printk(KERN_WARNING "md: couldn't set" | |
5654 | " array info. %d\n", err); | |
5655 | goto abort_unlock; | |
5656 | } | |
5657 | } | |
5658 | goto done_unlock; | |
5659 | ||
5660 | default:; | |
5661 | } | |
5662 | ||
5663 | /* | |
5664 | * Commands querying/configuring an existing array: | |
5665 | */ | |
32a7627c | 5666 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5667 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5668 | if ((!mddev->raid_disks && !mddev->external) |
5669 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5670 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5671 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5672 | err = -ENODEV; |
5673 | goto abort_unlock; | |
5674 | } | |
5675 | ||
5676 | /* | |
5677 | * Commands even a read-only array can execute: | |
5678 | */ | |
5679 | switch (cmd) | |
5680 | { | |
5681 | case GET_ARRAY_INFO: | |
5682 | err = get_array_info(mddev, argp); | |
5683 | goto done_unlock; | |
5684 | ||
32a7627c | 5685 | case GET_BITMAP_FILE: |
87162a28 | 5686 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5687 | goto done_unlock; |
5688 | ||
1da177e4 LT |
5689 | case GET_DISK_INFO: |
5690 | err = get_disk_info(mddev, argp); | |
5691 | goto done_unlock; | |
5692 | ||
5693 | case RESTART_ARRAY_RW: | |
5694 | err = restart_array(mddev); | |
5695 | goto done_unlock; | |
5696 | ||
5697 | case STOP_ARRAY: | |
d710e138 | 5698 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5699 | goto done_unlock; |
5700 | ||
5701 | case STOP_ARRAY_RO: | |
d710e138 | 5702 | err = do_md_stop(mddev, 1, 1); |
1da177e4 LT |
5703 | goto done_unlock; |
5704 | ||
e2218350 DW |
5705 | case BLKROSET: |
5706 | if (get_user(ro, (int __user *)(arg))) { | |
5707 | err = -EFAULT; | |
5708 | goto done_unlock; | |
5709 | } | |
5710 | err = -EINVAL; | |
5711 | ||
5712 | /* if the bdev is going readonly the value of mddev->ro | |
5713 | * does not matter, no writes are coming | |
5714 | */ | |
5715 | if (ro) | |
5716 | goto done_unlock; | |
5717 | ||
5718 | /* are we are already prepared for writes? */ | |
5719 | if (mddev->ro != 1) | |
5720 | goto done_unlock; | |
5721 | ||
5722 | /* transitioning to readauto need only happen for | |
5723 | * arrays that call md_write_start | |
5724 | */ | |
5725 | if (mddev->pers) { | |
5726 | err = restart_array(mddev); | |
5727 | if (err == 0) { | |
5728 | mddev->ro = 2; | |
5729 | set_disk_ro(mddev->gendisk, 0); | |
5730 | } | |
5731 | } | |
5732 | goto done_unlock; | |
1da177e4 LT |
5733 | } |
5734 | ||
5735 | /* | |
5736 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5737 | * superblock, so we do not allow them on read-only arrays. |
5738 | * However non-MD ioctls (e.g. get-size) will still come through | |
5739 | * here and hit the 'default' below, so only disallow | |
5740 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5741 | */ |
bb57fc64 | 5742 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5743 | if (mddev->ro == 2) { |
5744 | mddev->ro = 0; | |
b62b7590 | 5745 | sysfs_notify_dirent(mddev->sysfs_state); |
0fd62b86 NB |
5746 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5747 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5748 | } else { |
5749 | err = -EROFS; | |
5750 | goto abort_unlock; | |
5751 | } | |
1da177e4 LT |
5752 | } |
5753 | ||
5754 | switch (cmd) | |
5755 | { | |
5756 | case ADD_NEW_DISK: | |
5757 | { | |
5758 | mdu_disk_info_t info; | |
5759 | if (copy_from_user(&info, argp, sizeof(info))) | |
5760 | err = -EFAULT; | |
5761 | else | |
5762 | err = add_new_disk(mddev, &info); | |
5763 | goto done_unlock; | |
5764 | } | |
5765 | ||
5766 | case HOT_REMOVE_DISK: | |
5767 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5768 | goto done_unlock; | |
5769 | ||
5770 | case HOT_ADD_DISK: | |
5771 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5772 | goto done_unlock; | |
5773 | ||
5774 | case SET_DISK_FAULTY: | |
5775 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5776 | goto done_unlock; | |
5777 | ||
5778 | case RUN_ARRAY: | |
d710e138 | 5779 | err = do_md_run(mddev); |
1da177e4 LT |
5780 | goto done_unlock; |
5781 | ||
32a7627c N |
5782 | case SET_BITMAP_FILE: |
5783 | err = set_bitmap_file(mddev, (int)arg); | |
5784 | goto done_unlock; | |
5785 | ||
1da177e4 | 5786 | default: |
1da177e4 LT |
5787 | err = -EINVAL; |
5788 | goto abort_unlock; | |
5789 | } | |
5790 | ||
5791 | done_unlock: | |
5792 | abort_unlock: | |
d3374825 N |
5793 | if (mddev->hold_active == UNTIL_IOCTL && |
5794 | err != -EINVAL) | |
5795 | mddev->hold_active = 0; | |
1da177e4 LT |
5796 | mddev_unlock(mddev); |
5797 | ||
5798 | return err; | |
5799 | done: | |
5800 | if (err) | |
5801 | MD_BUG(); | |
5802 | abort: | |
5803 | return err; | |
5804 | } | |
aa98aa31 AB |
5805 | #ifdef CONFIG_COMPAT |
5806 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
5807 | unsigned int cmd, unsigned long arg) | |
5808 | { | |
5809 | switch (cmd) { | |
5810 | case HOT_REMOVE_DISK: | |
5811 | case HOT_ADD_DISK: | |
5812 | case SET_DISK_FAULTY: | |
5813 | case SET_BITMAP_FILE: | |
5814 | /* These take in integer arg, do not convert */ | |
5815 | break; | |
5816 | default: | |
5817 | arg = (unsigned long)compat_ptr(arg); | |
5818 | break; | |
5819 | } | |
5820 | ||
5821 | return md_ioctl(bdev, mode, cmd, arg); | |
5822 | } | |
5823 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 5824 | |
a39907fa | 5825 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
5826 | { |
5827 | /* | |
5828 | * Succeed if we can lock the mddev, which confirms that | |
5829 | * it isn't being stopped right now. | |
5830 | */ | |
d3374825 | 5831 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
5832 | int err; |
5833 | ||
d3374825 N |
5834 | if (mddev->gendisk != bdev->bd_disk) { |
5835 | /* we are racing with mddev_put which is discarding this | |
5836 | * bd_disk. | |
5837 | */ | |
5838 | mddev_put(mddev); | |
5839 | /* Wait until bdev->bd_disk is definitely gone */ | |
5840 | flush_scheduled_work(); | |
5841 | /* Then retry the open from the top */ | |
5842 | return -ERESTARTSYS; | |
5843 | } | |
5844 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
5845 | ||
c8c00a69 | 5846 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
5847 | goto out; |
5848 | ||
5849 | err = 0; | |
f2ea68cf | 5850 | atomic_inc(&mddev->openers); |
c8c00a69 | 5851 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 5852 | |
a39907fa | 5853 | check_disk_change(bdev); |
1da177e4 LT |
5854 | out: |
5855 | return err; | |
5856 | } | |
5857 | ||
a39907fa | 5858 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 5859 | { |
a39907fa | 5860 | mddev_t *mddev = disk->private_data; |
1da177e4 | 5861 | |
52e5f9d1 | 5862 | BUG_ON(!mddev); |
f2ea68cf | 5863 | atomic_dec(&mddev->openers); |
1da177e4 LT |
5864 | mddev_put(mddev); |
5865 | ||
5866 | return 0; | |
5867 | } | |
5868 | ||
44ce6294 LT |
5869 | static int md_media_changed(struct gendisk *disk) |
5870 | { | |
5871 | mddev_t *mddev = disk->private_data; | |
5872 | ||
5873 | return mddev->changed; | |
5874 | } | |
5875 | ||
5876 | static int md_revalidate(struct gendisk *disk) | |
5877 | { | |
5878 | mddev_t *mddev = disk->private_data; | |
5879 | ||
5880 | mddev->changed = 0; | |
5881 | return 0; | |
5882 | } | |
83d5cde4 | 5883 | static const struct block_device_operations md_fops = |
1da177e4 LT |
5884 | { |
5885 | .owner = THIS_MODULE, | |
a39907fa AV |
5886 | .open = md_open, |
5887 | .release = md_release, | |
b492b852 | 5888 | .ioctl = md_ioctl, |
aa98aa31 AB |
5889 | #ifdef CONFIG_COMPAT |
5890 | .compat_ioctl = md_compat_ioctl, | |
5891 | #endif | |
a885c8c4 | 5892 | .getgeo = md_getgeo, |
44ce6294 LT |
5893 | .media_changed = md_media_changed, |
5894 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
5895 | }; |
5896 | ||
75c96f85 | 5897 | static int md_thread(void * arg) |
1da177e4 LT |
5898 | { |
5899 | mdk_thread_t *thread = arg; | |
5900 | ||
1da177e4 LT |
5901 | /* |
5902 | * md_thread is a 'system-thread', it's priority should be very | |
5903 | * high. We avoid resource deadlocks individually in each | |
5904 | * raid personality. (RAID5 does preallocation) We also use RR and | |
5905 | * the very same RT priority as kswapd, thus we will never get | |
5906 | * into a priority inversion deadlock. | |
5907 | * | |
5908 | * we definitely have to have equal or higher priority than | |
5909 | * bdflush, otherwise bdflush will deadlock if there are too | |
5910 | * many dirty RAID5 blocks. | |
5911 | */ | |
1da177e4 | 5912 | |
6985c43f | 5913 | allow_signal(SIGKILL); |
a6fb0934 | 5914 | while (!kthread_should_stop()) { |
1da177e4 | 5915 | |
93588e22 N |
5916 | /* We need to wait INTERRUPTIBLE so that |
5917 | * we don't add to the load-average. | |
5918 | * That means we need to be sure no signals are | |
5919 | * pending | |
5920 | */ | |
5921 | if (signal_pending(current)) | |
5922 | flush_signals(current); | |
5923 | ||
5924 | wait_event_interruptible_timeout | |
5925 | (thread->wqueue, | |
5926 | test_bit(THREAD_WAKEUP, &thread->flags) | |
5927 | || kthread_should_stop(), | |
5928 | thread->timeout); | |
1da177e4 LT |
5929 | |
5930 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
5931 | ||
787453c2 | 5932 | thread->run(thread->mddev); |
1da177e4 | 5933 | } |
a6fb0934 | 5934 | |
1da177e4 LT |
5935 | return 0; |
5936 | } | |
5937 | ||
5938 | void md_wakeup_thread(mdk_thread_t *thread) | |
5939 | { | |
5940 | if (thread) { | |
5941 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
5942 | set_bit(THREAD_WAKEUP, &thread->flags); | |
5943 | wake_up(&thread->wqueue); | |
5944 | } | |
5945 | } | |
5946 | ||
5947 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
5948 | const char *name) | |
5949 | { | |
5950 | mdk_thread_t *thread; | |
1da177e4 | 5951 | |
9ffae0cf | 5952 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
5953 | if (!thread) |
5954 | return NULL; | |
5955 | ||
1da177e4 LT |
5956 | init_waitqueue_head(&thread->wqueue); |
5957 | ||
1da177e4 LT |
5958 | thread->run = run; |
5959 | thread->mddev = mddev; | |
32a7627c | 5960 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
5961 | thread->tsk = kthread_run(md_thread, thread, |
5962 | "%s_%s", | |
5963 | mdname(thread->mddev), | |
5964 | name ?: mddev->pers->name); | |
a6fb0934 | 5965 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
5966 | kfree(thread); |
5967 | return NULL; | |
5968 | } | |
1da177e4 LT |
5969 | return thread; |
5970 | } | |
5971 | ||
1da177e4 LT |
5972 | void md_unregister_thread(mdk_thread_t *thread) |
5973 | { | |
e0cf8f04 N |
5974 | if (!thread) |
5975 | return; | |
ba25f9dc | 5976 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
5977 | |
5978 | kthread_stop(thread->tsk); | |
1da177e4 LT |
5979 | kfree(thread); |
5980 | } | |
5981 | ||
5982 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
5983 | { | |
5984 | if (!mddev) { | |
5985 | MD_BUG(); | |
5986 | return; | |
5987 | } | |
5988 | ||
b2d444d7 | 5989 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5990 | return; |
6bfe0b49 DW |
5991 | |
5992 | if (mddev->external) | |
5993 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 5994 | /* |
1da177e4 LT |
5995 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
5996 | mdname(mddev), | |
5997 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
5998 | __builtin_return_address(0),__builtin_return_address(1), | |
5999 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 6000 | */ |
d0a0a5ee AM |
6001 | if (!mddev->pers) |
6002 | return; | |
1da177e4 LT |
6003 | if (!mddev->pers->error_handler) |
6004 | return; | |
6005 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6006 | if (mddev->degraded) |
6007 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c0cc75f8 | 6008 | sysfs_notify_dirent(rdev->sysfs_state); |
1da177e4 LT |
6009 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6010 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6011 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 6012 | md_new_event_inintr(mddev); |
1da177e4 LT |
6013 | } |
6014 | ||
6015 | /* seq_file implementation /proc/mdstat */ | |
6016 | ||
6017 | static void status_unused(struct seq_file *seq) | |
6018 | { | |
6019 | int i = 0; | |
6020 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6021 | |
6022 | seq_printf(seq, "unused devices: "); | |
6023 | ||
159ec1fc | 6024 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6025 | char b[BDEVNAME_SIZE]; |
6026 | i++; | |
6027 | seq_printf(seq, "%s ", | |
6028 | bdevname(rdev->bdev,b)); | |
6029 | } | |
6030 | if (!i) | |
6031 | seq_printf(seq, "<none>"); | |
6032 | ||
6033 | seq_printf(seq, "\n"); | |
6034 | } | |
6035 | ||
6036 | ||
6037 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
6038 | { | |
dd71cf6b N |
6039 | sector_t max_sectors, resync, res; |
6040 | unsigned long dt, db; | |
6041 | sector_t rt; | |
4588b42e N |
6042 | int scale; |
6043 | unsigned int per_milli; | |
1da177e4 | 6044 | |
dd71cf6b | 6045 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6046 | |
6047 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 6048 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6049 | else |
dd71cf6b | 6050 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6051 | |
6052 | /* | |
6053 | * Should not happen. | |
6054 | */ | |
dd71cf6b | 6055 | if (!max_sectors) { |
1da177e4 LT |
6056 | MD_BUG(); |
6057 | return; | |
6058 | } | |
4588b42e | 6059 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6060 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6061 | * u32, as those are the requirements for sector_div. |
6062 | * Thus 'scale' must be at least 10 | |
6063 | */ | |
6064 | scale = 10; | |
6065 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6066 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6067 | scale++; |
6068 | } | |
6069 | res = (resync>>scale)*1000; | |
dd71cf6b | 6070 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6071 | |
6072 | per_milli = res; | |
1da177e4 | 6073 | { |
4588b42e | 6074 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6075 | seq_printf(seq, "["); |
6076 | for (i = 0; i < x; i++) | |
6077 | seq_printf(seq, "="); | |
6078 | seq_printf(seq, ">"); | |
6079 | for (i = 0; i < y; i++) | |
6080 | seq_printf(seq, "."); | |
6081 | seq_printf(seq, "] "); | |
6082 | } | |
4588b42e | 6083 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6084 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6085 | "reshape" : | |
61df9d91 N |
6086 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6087 | "check" : | |
6088 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6089 | "resync" : "recovery"))), | |
6090 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6091 | (unsigned long long) resync/2, |
6092 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6093 | |
6094 | /* | |
1da177e4 LT |
6095 | * dt: time from mark until now |
6096 | * db: blocks written from mark until now | |
6097 | * rt: remaining time | |
dd71cf6b N |
6098 | * |
6099 | * rt is a sector_t, so could be 32bit or 64bit. | |
6100 | * So we divide before multiply in case it is 32bit and close | |
6101 | * to the limit. | |
6102 | * We scale the divisor (db) by 32 to avoid loosing precision | |
6103 | * near the end of resync when the number of remaining sectors | |
6104 | * is close to 'db'. | |
6105 | * We then divide rt by 32 after multiplying by db to compensate. | |
6106 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6107 | */ |
6108 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6109 | if (!dt) dt++; | |
ff4e8d9a N |
6110 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6111 | - mddev->resync_mark_cnt; | |
1da177e4 | 6112 | |
dd71cf6b N |
6113 | rt = max_sectors - resync; /* number of remaining sectors */ |
6114 | sector_div(rt, db/32+1); | |
6115 | rt *= dt; | |
6116 | rt >>= 5; | |
6117 | ||
6118 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6119 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6120 | |
ff4e8d9a | 6121 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6122 | } |
6123 | ||
6124 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6125 | { | |
6126 | struct list_head *tmp; | |
6127 | loff_t l = *pos; | |
6128 | mddev_t *mddev; | |
6129 | ||
6130 | if (l >= 0x10000) | |
6131 | return NULL; | |
6132 | if (!l--) | |
6133 | /* header */ | |
6134 | return (void*)1; | |
6135 | ||
6136 | spin_lock(&all_mddevs_lock); | |
6137 | list_for_each(tmp,&all_mddevs) | |
6138 | if (!l--) { | |
6139 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
6140 | mddev_get(mddev); | |
6141 | spin_unlock(&all_mddevs_lock); | |
6142 | return mddev; | |
6143 | } | |
6144 | spin_unlock(&all_mddevs_lock); | |
6145 | if (!l--) | |
6146 | return (void*)2;/* tail */ | |
6147 | return NULL; | |
6148 | } | |
6149 | ||
6150 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6151 | { | |
6152 | struct list_head *tmp; | |
6153 | mddev_t *next_mddev, *mddev = v; | |
6154 | ||
6155 | ++*pos; | |
6156 | if (v == (void*)2) | |
6157 | return NULL; | |
6158 | ||
6159 | spin_lock(&all_mddevs_lock); | |
6160 | if (v == (void*)1) | |
6161 | tmp = all_mddevs.next; | |
6162 | else | |
6163 | tmp = mddev->all_mddevs.next; | |
6164 | if (tmp != &all_mddevs) | |
6165 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
6166 | else { | |
6167 | next_mddev = (void*)2; | |
6168 | *pos = 0x10000; | |
6169 | } | |
6170 | spin_unlock(&all_mddevs_lock); | |
6171 | ||
6172 | if (v != (void*)1) | |
6173 | mddev_put(mddev); | |
6174 | return next_mddev; | |
6175 | ||
6176 | } | |
6177 | ||
6178 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6179 | { | |
6180 | mddev_t *mddev = v; | |
6181 | ||
6182 | if (mddev && v != (void*)1 && v != (void*)2) | |
6183 | mddev_put(mddev); | |
6184 | } | |
6185 | ||
d7603b7e N |
6186 | struct mdstat_info { |
6187 | int event; | |
6188 | }; | |
6189 | ||
1da177e4 LT |
6190 | static int md_seq_show(struct seq_file *seq, void *v) |
6191 | { | |
6192 | mddev_t *mddev = v; | |
dd8ac336 | 6193 | sector_t sectors; |
1da177e4 | 6194 | mdk_rdev_t *rdev; |
d7603b7e | 6195 | struct mdstat_info *mi = seq->private; |
32a7627c | 6196 | struct bitmap *bitmap; |
1da177e4 LT |
6197 | |
6198 | if (v == (void*)1) { | |
2604b703 | 6199 | struct mdk_personality *pers; |
1da177e4 LT |
6200 | seq_printf(seq, "Personalities : "); |
6201 | spin_lock(&pers_lock); | |
2604b703 N |
6202 | list_for_each_entry(pers, &pers_list, list) |
6203 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6204 | |
6205 | spin_unlock(&pers_lock); | |
6206 | seq_printf(seq, "\n"); | |
d7603b7e | 6207 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
6208 | return 0; |
6209 | } | |
6210 | if (v == (void*)2) { | |
6211 | status_unused(seq); | |
6212 | return 0; | |
6213 | } | |
6214 | ||
5dc5cf7d | 6215 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6216 | return -EINTR; |
5dc5cf7d | 6217 | |
1da177e4 LT |
6218 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
6219 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
6220 | mddev->pers ? "" : "in"); | |
6221 | if (mddev->pers) { | |
f91de92e | 6222 | if (mddev->ro==1) |
1da177e4 | 6223 | seq_printf(seq, " (read-only)"); |
f91de92e | 6224 | if (mddev->ro==2) |
52720ae7 | 6225 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
6226 | seq_printf(seq, " %s", mddev->pers->name); |
6227 | } | |
6228 | ||
dd8ac336 | 6229 | sectors = 0; |
159ec1fc | 6230 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
6231 | char b[BDEVNAME_SIZE]; |
6232 | seq_printf(seq, " %s[%d]", | |
6233 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
6234 | if (test_bit(WriteMostly, &rdev->flags)) |
6235 | seq_printf(seq, "(W)"); | |
b2d444d7 | 6236 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6237 | seq_printf(seq, "(F)"); |
6238 | continue; | |
b325a32e N |
6239 | } else if (rdev->raid_disk < 0) |
6240 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 6241 | sectors += rdev->sectors; |
1da177e4 LT |
6242 | } |
6243 | ||
6244 | if (!list_empty(&mddev->disks)) { | |
6245 | if (mddev->pers) | |
6246 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
6247 | (unsigned long long) |
6248 | mddev->array_sectors / 2); | |
1da177e4 LT |
6249 | else |
6250 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 6251 | (unsigned long long)sectors / 2); |
1da177e4 | 6252 | } |
1cd6bf19 N |
6253 | if (mddev->persistent) { |
6254 | if (mddev->major_version != 0 || | |
6255 | mddev->minor_version != 90) { | |
6256 | seq_printf(seq," super %d.%d", | |
6257 | mddev->major_version, | |
6258 | mddev->minor_version); | |
6259 | } | |
e691063a N |
6260 | } else if (mddev->external) |
6261 | seq_printf(seq, " super external:%s", | |
6262 | mddev->metadata_type); | |
6263 | else | |
1cd6bf19 | 6264 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
6265 | |
6266 | if (mddev->pers) { | |
d710e138 | 6267 | mddev->pers->status(seq, mddev); |
1da177e4 | 6268 | seq_printf(seq, "\n "); |
8e1b39d6 N |
6269 | if (mddev->pers->sync_request) { |
6270 | if (mddev->curr_resync > 2) { | |
d710e138 | 6271 | status_resync(seq, mddev); |
8e1b39d6 N |
6272 | seq_printf(seq, "\n "); |
6273 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
6274 | seq_printf(seq, "\tresync=DELAYED\n "); | |
6275 | else if (mddev->recovery_cp < MaxSector) | |
6276 | seq_printf(seq, "\tresync=PENDING\n "); | |
6277 | } | |
32a7627c N |
6278 | } else |
6279 | seq_printf(seq, "\n "); | |
6280 | ||
6281 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
6282 | unsigned long chunk_kb; |
6283 | unsigned long flags; | |
32a7627c | 6284 | spin_lock_irqsave(&bitmap->lock, flags); |
42a04b50 | 6285 | chunk_kb = mddev->bitmap_info.chunksize >> 10; |
32a7627c N |
6286 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " |
6287 | "%lu%s chunk", | |
6288 | bitmap->pages - bitmap->missing_pages, | |
6289 | bitmap->pages, | |
6290 | (bitmap->pages - bitmap->missing_pages) | |
6291 | << (PAGE_SHIFT - 10), | |
42a04b50 | 6292 | chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, |
32a7627c | 6293 | chunk_kb ? "KB" : "B"); |
78d742d8 N |
6294 | if (bitmap->file) { |
6295 | seq_printf(seq, ", file: "); | |
c32c2f63 | 6296 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 6297 | } |
78d742d8 | 6298 | |
32a7627c N |
6299 | seq_printf(seq, "\n"); |
6300 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
6301 | } |
6302 | ||
6303 | seq_printf(seq, "\n"); | |
6304 | } | |
6305 | mddev_unlock(mddev); | |
6306 | ||
6307 | return 0; | |
6308 | } | |
6309 | ||
110518bc | 6310 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6311 | .start = md_seq_start, |
6312 | .next = md_seq_next, | |
6313 | .stop = md_seq_stop, | |
6314 | .show = md_seq_show, | |
6315 | }; | |
6316 | ||
6317 | static int md_seq_open(struct inode *inode, struct file *file) | |
6318 | { | |
6319 | int error; | |
d7603b7e N |
6320 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6321 | if (mi == NULL) | |
6322 | return -ENOMEM; | |
1da177e4 LT |
6323 | |
6324 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6325 | if (error) |
6326 | kfree(mi); | |
6327 | else { | |
6328 | struct seq_file *p = file->private_data; | |
6329 | p->private = mi; | |
6330 | mi->event = atomic_read(&md_event_count); | |
6331 | } | |
1da177e4 LT |
6332 | return error; |
6333 | } | |
6334 | ||
d7603b7e N |
6335 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6336 | { | |
6337 | struct seq_file *m = filp->private_data; | |
6338 | struct mdstat_info *mi = m->private; | |
6339 | int mask; | |
6340 | ||
6341 | poll_wait(filp, &md_event_waiters, wait); | |
6342 | ||
6343 | /* always allow read */ | |
6344 | mask = POLLIN | POLLRDNORM; | |
6345 | ||
6346 | if (mi->event != atomic_read(&md_event_count)) | |
6347 | mask |= POLLERR | POLLPRI; | |
6348 | return mask; | |
6349 | } | |
6350 | ||
fa027c2a | 6351 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6352 | .owner = THIS_MODULE, |
1da177e4 LT |
6353 | .open = md_seq_open, |
6354 | .read = seq_read, | |
6355 | .llseek = seq_lseek, | |
c3f94b40 | 6356 | .release = seq_release_private, |
d7603b7e | 6357 | .poll = mdstat_poll, |
1da177e4 LT |
6358 | }; |
6359 | ||
2604b703 | 6360 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6361 | { |
1da177e4 | 6362 | spin_lock(&pers_lock); |
2604b703 N |
6363 | list_add_tail(&p->list, &pers_list); |
6364 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6365 | spin_unlock(&pers_lock); |
6366 | return 0; | |
6367 | } | |
6368 | ||
2604b703 | 6369 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6370 | { |
2604b703 | 6371 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6372 | spin_lock(&pers_lock); |
2604b703 | 6373 | list_del_init(&p->list); |
1da177e4 LT |
6374 | spin_unlock(&pers_lock); |
6375 | return 0; | |
6376 | } | |
6377 | ||
eea1bf38 | 6378 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6379 | { |
6380 | mdk_rdev_t * rdev; | |
1da177e4 | 6381 | int idle; |
eea1bf38 | 6382 | int curr_events; |
1da177e4 LT |
6383 | |
6384 | idle = 1; | |
4b80991c N |
6385 | rcu_read_lock(); |
6386 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6387 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6388 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6389 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6390 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6391 | /* sync IO will cause sync_io to increase before the disk_stats |
6392 | * as sync_io is counted when a request starts, and | |
6393 | * disk_stats is counted when it completes. | |
6394 | * So resync activity will cause curr_events to be smaller than | |
6395 | * when there was no such activity. | |
6396 | * non-sync IO will cause disk_stat to increase without | |
6397 | * increasing sync_io so curr_events will (eventually) | |
6398 | * be larger than it was before. Once it becomes | |
6399 | * substantially larger, the test below will cause | |
6400 | * the array to appear non-idle, and resync will slow | |
6401 | * down. | |
6402 | * If there is a lot of outstanding resync activity when | |
6403 | * we set last_event to curr_events, then all that activity | |
6404 | * completing might cause the array to appear non-idle | |
6405 | * and resync will be slowed down even though there might | |
6406 | * not have been non-resync activity. This will only | |
6407 | * happen once though. 'last_events' will soon reflect | |
6408 | * the state where there is little or no outstanding | |
6409 | * resync requests, and further resync activity will | |
6410 | * always make curr_events less than last_events. | |
c0e48521 | 6411 | * |
1da177e4 | 6412 | */ |
eea1bf38 | 6413 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6414 | rdev->last_events = curr_events; |
6415 | idle = 0; | |
6416 | } | |
6417 | } | |
4b80991c | 6418 | rcu_read_unlock(); |
1da177e4 LT |
6419 | return idle; |
6420 | } | |
6421 | ||
6422 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6423 | { | |
6424 | /* another "blocks" (512byte) blocks have been synced */ | |
6425 | atomic_sub(blocks, &mddev->recovery_active); | |
6426 | wake_up(&mddev->recovery_wait); | |
6427 | if (!ok) { | |
dfc70645 | 6428 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6429 | md_wakeup_thread(mddev->thread); |
6430 | // stop recovery, signal do_sync .... | |
6431 | } | |
6432 | } | |
6433 | ||
6434 | ||
06d91a5f N |
6435 | /* md_write_start(mddev, bi) |
6436 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6437 | * in superblock) before writing, schedule a superblock update |
6438 | * and wait for it to complete. | |
06d91a5f | 6439 | */ |
3d310eb7 | 6440 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6441 | { |
0fd62b86 | 6442 | int did_change = 0; |
06d91a5f | 6443 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6444 | return; |
06d91a5f | 6445 | |
f91de92e N |
6446 | BUG_ON(mddev->ro == 1); |
6447 | if (mddev->ro == 2) { | |
6448 | /* need to switch to read/write */ | |
6449 | mddev->ro = 0; | |
6450 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6451 | md_wakeup_thread(mddev->thread); | |
25156198 | 6452 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6453 | did_change = 1; |
f91de92e | 6454 | } |
06d91a5f | 6455 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6456 | if (mddev->safemode == 1) |
6457 | mddev->safemode = 0; | |
06d91a5f | 6458 | if (mddev->in_sync) { |
a9701a30 | 6459 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6460 | if (mddev->in_sync) { |
6461 | mddev->in_sync = 0; | |
850b2b42 | 6462 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3d310eb7 | 6463 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6464 | did_change = 1; |
3d310eb7 | 6465 | } |
a9701a30 | 6466 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6467 | } |
0fd62b86 | 6468 | if (did_change) |
b62b7590 | 6469 | sysfs_notify_dirent(mddev->sysfs_state); |
09a44cc1 N |
6470 | wait_event(mddev->sb_wait, |
6471 | !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && | |
6472 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1da177e4 LT |
6473 | } |
6474 | ||
6475 | void md_write_end(mddev_t *mddev) | |
6476 | { | |
6477 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6478 | if (mddev->safemode == 2) | |
6479 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6480 | else if (mddev->safemode_delay) |
1da177e4 LT |
6481 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6482 | } | |
6483 | } | |
6484 | ||
2a2275d6 N |
6485 | /* md_allow_write(mddev) |
6486 | * Calling this ensures that the array is marked 'active' so that writes | |
6487 | * may proceed without blocking. It is important to call this before | |
6488 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6489 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6490 | * |
6491 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6492 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6493 | */ |
b5470dc5 | 6494 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6495 | { |
6496 | if (!mddev->pers) | |
b5470dc5 | 6497 | return 0; |
2a2275d6 | 6498 | if (mddev->ro) |
b5470dc5 | 6499 | return 0; |
1a0fd497 | 6500 | if (!mddev->pers->sync_request) |
b5470dc5 | 6501 | return 0; |
2a2275d6 N |
6502 | |
6503 | spin_lock_irq(&mddev->write_lock); | |
6504 | if (mddev->in_sync) { | |
6505 | mddev->in_sync = 0; | |
6506 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6507 | if (mddev->safemode_delay && | |
6508 | mddev->safemode == 0) | |
6509 | mddev->safemode = 1; | |
6510 | spin_unlock_irq(&mddev->write_lock); | |
6511 | md_update_sb(mddev, 0); | |
b62b7590 | 6512 | sysfs_notify_dirent(mddev->sysfs_state); |
2a2275d6 N |
6513 | } else |
6514 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 DW |
6515 | |
6516 | if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
6517 | return -EAGAIN; | |
6518 | else | |
6519 | return 0; | |
2a2275d6 N |
6520 | } |
6521 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6522 | ||
1da177e4 LT |
6523 | #define SYNC_MARKS 10 |
6524 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6525 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6526 | { |
6527 | mddev_t *mddev2; | |
6528 | unsigned int currspeed = 0, | |
6529 | window; | |
57afd89f | 6530 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6531 | unsigned long mark[SYNC_MARKS]; |
6532 | sector_t mark_cnt[SYNC_MARKS]; | |
6533 | int last_mark,m; | |
6534 | struct list_head *tmp; | |
6535 | sector_t last_check; | |
57afd89f | 6536 | int skipped = 0; |
5fd6c1dc | 6537 | mdk_rdev_t *rdev; |
61df9d91 | 6538 | char *desc; |
1da177e4 LT |
6539 | |
6540 | /* just incase thread restarts... */ | |
6541 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6542 | return; | |
5fd6c1dc N |
6543 | if (mddev->ro) /* never try to sync a read-only array */ |
6544 | return; | |
1da177e4 | 6545 | |
61df9d91 N |
6546 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6547 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6548 | desc = "data-check"; | |
6549 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6550 | desc = "requested-resync"; | |
6551 | else | |
6552 | desc = "resync"; | |
6553 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6554 | desc = "reshape"; | |
6555 | else | |
6556 | desc = "recovery"; | |
6557 | ||
1da177e4 LT |
6558 | /* we overload curr_resync somewhat here. |
6559 | * 0 == not engaged in resync at all | |
6560 | * 2 == checking that there is no conflict with another sync | |
6561 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6562 | * commense | |
6563 | * other == active in resync - this many blocks | |
6564 | * | |
6565 | * Before starting a resync we must have set curr_resync to | |
6566 | * 2, and then checked that every "conflicting" array has curr_resync | |
6567 | * less than ours. When we find one that is the same or higher | |
6568 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6569 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6570 | * This will mean we have to start checking from the beginning again. | |
6571 | * | |
6572 | */ | |
6573 | ||
6574 | do { | |
6575 | mddev->curr_resync = 2; | |
6576 | ||
6577 | try_again: | |
404e4b43 | 6578 | if (kthread_should_stop()) |
6985c43f | 6579 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
404e4b43 N |
6580 | |
6581 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
1da177e4 | 6582 | goto skip; |
29ac4aa3 | 6583 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6584 | if (mddev2 == mddev) |
6585 | continue; | |
90b08710 BS |
6586 | if (!mddev->parallel_resync |
6587 | && mddev2->curr_resync | |
6588 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6589 | DEFINE_WAIT(wq); |
6590 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6591 | /* arbitrarily yield */ | |
6592 | mddev->curr_resync = 1; | |
6593 | wake_up(&resync_wait); | |
6594 | } | |
6595 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6596 | /* no need to wait here, we can wait the next | |
6597 | * time 'round when curr_resync == 2 | |
6598 | */ | |
6599 | continue; | |
9744197c N |
6600 | /* We need to wait 'interruptible' so as not to |
6601 | * contribute to the load average, and not to | |
6602 | * be caught by 'softlockup' | |
6603 | */ | |
6604 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6605 | if (!kthread_should_stop() && |
8712e553 | 6606 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6607 | printk(KERN_INFO "md: delaying %s of %s" |
6608 | " until %s has finished (they" | |
1da177e4 | 6609 | " share one or more physical units)\n", |
61df9d91 | 6610 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6611 | mddev_put(mddev2); |
9744197c N |
6612 | if (signal_pending(current)) |
6613 | flush_signals(current); | |
1da177e4 LT |
6614 | schedule(); |
6615 | finish_wait(&resync_wait, &wq); | |
6616 | goto try_again; | |
6617 | } | |
6618 | finish_wait(&resync_wait, &wq); | |
6619 | } | |
6620 | } | |
6621 | } while (mddev->curr_resync < 2); | |
6622 | ||
5fd6c1dc | 6623 | j = 0; |
9d88883e | 6624 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6625 | /* resync follows the size requested by the personality, |
57afd89f | 6626 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6627 | */ |
6628 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6629 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6630 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6631 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6632 | j = mddev->resync_min; | |
6633 | else if (!mddev->bitmap) | |
5fd6c1dc | 6634 | j = mddev->recovery_cp; |
5e96ee65 | 6635 | |
ccfcc3c1 | 6636 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6637 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6638 | else { |
1da177e4 | 6639 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6640 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6641 | j = MaxSector; |
4e59ca7d DW |
6642 | rcu_read_lock(); |
6643 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6644 | if (rdev->raid_disk >= 0 && |
6645 | !test_bit(Faulty, &rdev->flags) && | |
6646 | !test_bit(In_sync, &rdev->flags) && | |
6647 | rdev->recovery_offset < j) | |
6648 | j = rdev->recovery_offset; | |
4e59ca7d | 6649 | rcu_read_unlock(); |
5fd6c1dc | 6650 | } |
1da177e4 | 6651 | |
61df9d91 N |
6652 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6653 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6654 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6655 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6656 | "(but not more than %d KB/sec) for %s.\n", |
6657 | speed_max(mddev), desc); | |
1da177e4 | 6658 | |
eea1bf38 | 6659 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6660 | |
57afd89f | 6661 | io_sectors = 0; |
1da177e4 LT |
6662 | for (m = 0; m < SYNC_MARKS; m++) { |
6663 | mark[m] = jiffies; | |
57afd89f | 6664 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6665 | } |
6666 | last_mark = 0; | |
6667 | mddev->resync_mark = mark[last_mark]; | |
6668 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6669 | ||
6670 | /* | |
6671 | * Tune reconstruction: | |
6672 | */ | |
6673 | window = 32*(PAGE_SIZE/512); | |
6674 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6675 | window/2,(unsigned long long) max_sectors/2); | |
6676 | ||
6677 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6678 | last_check = 0; |
6679 | ||
6680 | if (j>2) { | |
6681 | printk(KERN_INFO | |
61df9d91 N |
6682 | "md: resuming %s of %s from checkpoint.\n", |
6683 | desc, mdname(mddev)); | |
1da177e4 LT |
6684 | mddev->curr_resync = j; |
6685 | } | |
efa59339 | 6686 | mddev->curr_resync_completed = mddev->curr_resync; |
1da177e4 LT |
6687 | |
6688 | while (j < max_sectors) { | |
57afd89f | 6689 | sector_t sectors; |
1da177e4 | 6690 | |
57afd89f | 6691 | skipped = 0; |
97e4f42d | 6692 | |
7a91ee1f N |
6693 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6694 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6695 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6696 | > (max_sectors >> 4)) || | |
6697 | (j - mddev->curr_resync_completed)*2 | |
6698 | >= mddev->resync_max - mddev->curr_resync_completed | |
6699 | )) { | |
97e4f42d N |
6700 | /* time to update curr_resync_completed */ |
6701 | blk_unplug(mddev->queue); | |
6702 | wait_event(mddev->recovery_wait, | |
6703 | atomic_read(&mddev->recovery_active) == 0); | |
6704 | mddev->curr_resync_completed = | |
6705 | mddev->curr_resync; | |
6706 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
acb180b0 | 6707 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6708 | } |
acb180b0 | 6709 | |
e62e58a5 N |
6710 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6711 | /* As this condition is controlled by user-space, | |
6712 | * we can block indefinitely, so use '_interruptible' | |
6713 | * to avoid triggering warnings. | |
6714 | */ | |
6715 | flush_signals(current); /* just in case */ | |
6716 | wait_event_interruptible(mddev->recovery_wait, | |
6717 | mddev->resync_max > j | |
6718 | || kthread_should_stop()); | |
6719 | } | |
acb180b0 N |
6720 | |
6721 | if (kthread_should_stop()) | |
6722 | goto interrupted; | |
6723 | ||
57afd89f | 6724 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6725 | currspeed < speed_min(mddev)); |
57afd89f | 6726 | if (sectors == 0) { |
dfc70645 | 6727 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6728 | goto out; |
6729 | } | |
57afd89f N |
6730 | |
6731 | if (!skipped) { /* actual IO requested */ | |
6732 | io_sectors += sectors; | |
6733 | atomic_add(sectors, &mddev->recovery_active); | |
6734 | } | |
6735 | ||
1da177e4 LT |
6736 | j += sectors; |
6737 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6738 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6739 | if (last_check == 0) |
6740 | /* this is the earliers that rebuilt will be | |
6741 | * visible in /proc/mdstat | |
6742 | */ | |
6743 | md_new_event(mddev); | |
57afd89f N |
6744 | |
6745 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6746 | continue; |
6747 | ||
57afd89f | 6748 | last_check = io_sectors; |
1da177e4 | 6749 | |
dfc70645 | 6750 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6751 | break; |
6752 | ||
6753 | repeat: | |
6754 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6755 | /* step marks */ | |
6756 | int next = (last_mark+1) % SYNC_MARKS; | |
6757 | ||
6758 | mddev->resync_mark = mark[next]; | |
6759 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6760 | mark[next] = jiffies; | |
57afd89f | 6761 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6762 | last_mark = next; |
6763 | } | |
6764 | ||
6765 | ||
c6207277 N |
6766 | if (kthread_should_stop()) |
6767 | goto interrupted; | |
6768 | ||
1da177e4 LT |
6769 | |
6770 | /* | |
6771 | * this loop exits only if either when we are slower than | |
6772 | * the 'hard' speed limit, or the system was IO-idle for | |
6773 | * a jiffy. | |
6774 | * the system might be non-idle CPU-wise, but we only care | |
6775 | * about not overloading the IO subsystem. (things like an | |
6776 | * e2fsck being done on the RAID array should execute fast) | |
6777 | */ | |
2ad8b1ef | 6778 | blk_unplug(mddev->queue); |
1da177e4 LT |
6779 | cond_resched(); |
6780 | ||
57afd89f N |
6781 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6782 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6783 | |
88202a0c N |
6784 | if (currspeed > speed_min(mddev)) { |
6785 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6786 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6787 | msleep(500); |
1da177e4 LT |
6788 | goto repeat; |
6789 | } | |
6790 | } | |
6791 | } | |
61df9d91 | 6792 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6793 | /* |
6794 | * this also signals 'finished resyncing' to md_stop | |
6795 | */ | |
6796 | out: | |
2ad8b1ef | 6797 | blk_unplug(mddev->queue); |
1da177e4 LT |
6798 | |
6799 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
6800 | ||
6801 | /* tell personality that we are finished */ | |
57afd89f | 6802 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6803 | |
dfc70645 | 6804 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6805 | mddev->curr_resync > 2) { |
6806 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6807 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6808 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6809 | printk(KERN_INFO | |
61df9d91 N |
6810 | "md: checkpointing %s of %s.\n", |
6811 | desc, mdname(mddev)); | |
5fd6c1dc N |
6812 | mddev->recovery_cp = mddev->curr_resync; |
6813 | } | |
6814 | } else | |
6815 | mddev->recovery_cp = MaxSector; | |
6816 | } else { | |
6817 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6818 | mddev->curr_resync = MaxSector; | |
4e59ca7d DW |
6819 | rcu_read_lock(); |
6820 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6821 | if (rdev->raid_disk >= 0 && |
6822 | !test_bit(Faulty, &rdev->flags) && | |
6823 | !test_bit(In_sync, &rdev->flags) && | |
6824 | rdev->recovery_offset < mddev->curr_resync) | |
6825 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 6826 | rcu_read_unlock(); |
5fd6c1dc | 6827 | } |
1da177e4 | 6828 | } |
17571284 | 6829 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6830 | |
1da177e4 | 6831 | skip: |
c07b70ad N |
6832 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
6833 | /* We completed so min/max setting can be forgotten if used. */ | |
6834 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6835 | mddev->resync_min = 0; | |
6836 | mddev->resync_max = MaxSector; | |
6837 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6838 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 | 6839 | mddev->curr_resync = 0; |
efa59339 N |
6840 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
6841 | mddev->curr_resync_completed = 0; | |
c6207277 | 6842 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
1da177e4 LT |
6843 | wake_up(&resync_wait); |
6844 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
6845 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
6846 | return; |
6847 | ||
6848 | interrupted: | |
6849 | /* | |
6850 | * got a signal, exit. | |
6851 | */ | |
6852 | printk(KERN_INFO | |
6853 | "md: md_do_sync() got signal ... exiting\n"); | |
6854 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6855 | goto out; | |
6856 | ||
1da177e4 | 6857 | } |
29269553 | 6858 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
6859 | |
6860 | ||
b4c4c7b8 N |
6861 | static int remove_and_add_spares(mddev_t *mddev) |
6862 | { | |
6863 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
6864 | int spares = 0; |
6865 | ||
97e4f42d N |
6866 | mddev->curr_resync_completed = 0; |
6867 | ||
159ec1fc | 6868 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 6869 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 6870 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
6871 | (test_bit(Faulty, &rdev->flags) || |
6872 | ! test_bit(In_sync, &rdev->flags)) && | |
6873 | atomic_read(&rdev->nr_pending)==0) { | |
6874 | if (mddev->pers->hot_remove_disk( | |
6875 | mddev, rdev->raid_disk)==0) { | |
6876 | char nm[20]; | |
6877 | sprintf(nm,"rd%d", rdev->raid_disk); | |
6878 | sysfs_remove_link(&mddev->kobj, nm); | |
6879 | rdev->raid_disk = -1; | |
6880 | } | |
6881 | } | |
6882 | ||
4044ba58 | 6883 | if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { |
159ec1fc | 6884 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 6885 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
6886 | !test_bit(In_sync, &rdev->flags) && |
6887 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 6888 | spares++; |
b4c4c7b8 N |
6889 | if (rdev->raid_disk < 0 |
6890 | && !test_bit(Faulty, &rdev->flags)) { | |
6891 | rdev->recovery_offset = 0; | |
199050ea NB |
6892 | if (mddev->pers-> |
6893 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
6894 | char nm[20]; |
6895 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
6896 | if (sysfs_create_link(&mddev->kobj, |
6897 | &rdev->kobj, nm)) | |
6898 | printk(KERN_WARNING | |
6899 | "md: cannot register " | |
6900 | "%s for %s\n", | |
6901 | nm, mdname(mddev)); | |
b4c4c7b8 N |
6902 | spares++; |
6903 | md_new_event(mddev); | |
93be75ff | 6904 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
b4c4c7b8 N |
6905 | } else |
6906 | break; | |
6907 | } | |
dfc70645 | 6908 | } |
b4c4c7b8 N |
6909 | } |
6910 | return spares; | |
6911 | } | |
1da177e4 LT |
6912 | /* |
6913 | * This routine is regularly called by all per-raid-array threads to | |
6914 | * deal with generic issues like resync and super-block update. | |
6915 | * Raid personalities that don't have a thread (linear/raid0) do not | |
6916 | * need this as they never do any recovery or update the superblock. | |
6917 | * | |
6918 | * It does not do any resync itself, but rather "forks" off other threads | |
6919 | * to do that as needed. | |
6920 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
6921 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 6922 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
6923 | * and wakeups up this thread which will reap the thread and finish up. |
6924 | * This thread also removes any faulty devices (with nr_pending == 0). | |
6925 | * | |
6926 | * The overall approach is: | |
6927 | * 1/ if the superblock needs updating, update it. | |
6928 | * 2/ If a recovery thread is running, don't do anything else. | |
6929 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
6930 | * 4/ If there are any faulty devices, remove them. | |
6931 | * 5/ If array is degraded, try to add spares devices | |
6932 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
6933 | */ | |
6934 | void md_check_recovery(mddev_t *mddev) | |
6935 | { | |
6936 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6937 | |
6938 | ||
5f40402d | 6939 | if (mddev->bitmap) |
aa5cbd10 | 6940 | bitmap_daemon_work(mddev); |
1da177e4 LT |
6941 | |
6942 | if (mddev->ro) | |
6943 | return; | |
fca4d848 N |
6944 | |
6945 | if (signal_pending(current)) { | |
31a59e34 | 6946 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
6947 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
6948 | mdname(mddev)); | |
6949 | mddev->safemode = 2; | |
6950 | } | |
6951 | flush_signals(current); | |
6952 | } | |
6953 | ||
c89a8eee N |
6954 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
6955 | return; | |
1da177e4 | 6956 | if ( ! ( |
e691063a | 6957 | (mddev->flags && !mddev->external) || |
1da177e4 | 6958 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 6959 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 6960 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
6961 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
6962 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
6963 | )) |
6964 | return; | |
fca4d848 | 6965 | |
df5b89b3 | 6966 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 6967 | int spares = 0; |
fca4d848 | 6968 | |
c89a8eee N |
6969 | if (mddev->ro) { |
6970 | /* Only thing we do on a ro array is remove | |
6971 | * failed devices. | |
6972 | */ | |
6973 | remove_and_add_spares(mddev); | |
6974 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6975 | goto unlock; | |
6976 | } | |
6977 | ||
31a59e34 | 6978 | if (!mddev->external) { |
0fd62b86 | 6979 | int did_change = 0; |
31a59e34 N |
6980 | spin_lock_irq(&mddev->write_lock); |
6981 | if (mddev->safemode && | |
6982 | !atomic_read(&mddev->writes_pending) && | |
6983 | !mddev->in_sync && | |
6984 | mddev->recovery_cp == MaxSector) { | |
6985 | mddev->in_sync = 1; | |
0fd62b86 | 6986 | did_change = 1; |
31a59e34 N |
6987 | if (mddev->persistent) |
6988 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6989 | } | |
6990 | if (mddev->safemode == 1) | |
6991 | mddev->safemode = 0; | |
6992 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 6993 | if (did_change) |
b62b7590 | 6994 | sysfs_notify_dirent(mddev->sysfs_state); |
fca4d848 | 6995 | } |
fca4d848 | 6996 | |
850b2b42 N |
6997 | if (mddev->flags) |
6998 | md_update_sb(mddev, 0); | |
06d91a5f | 6999 | |
1da177e4 LT |
7000 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7001 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7002 | /* resync/recovery still happening */ | |
7003 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7004 | goto unlock; | |
7005 | } | |
7006 | if (mddev->sync_thread) { | |
7007 | /* resync has finished, collect result */ | |
7008 | md_unregister_thread(mddev->sync_thread); | |
7009 | mddev->sync_thread = NULL; | |
56ac36d7 DW |
7010 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
7011 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1da177e4 LT |
7012 | /* success...*/ |
7013 | /* activate any spares */ | |
a99ac971 NB |
7014 | if (mddev->pers->spare_active(mddev)) |
7015 | sysfs_notify(&mddev->kobj, NULL, | |
7016 | "degraded"); | |
1da177e4 | 7017 | } |
cea9c228 N |
7018 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7019 | mddev->pers->finish_reshape) | |
7020 | mddev->pers->finish_reshape(mddev); | |
850b2b42 | 7021 | md_update_sb(mddev, 1); |
41158c7e N |
7022 | |
7023 | /* if array is no-longer degraded, then any saved_raid_disk | |
7024 | * information must be scrapped | |
7025 | */ | |
7026 | if (!mddev->degraded) | |
159ec1fc | 7027 | list_for_each_entry(rdev, &mddev->disks, same_set) |
41158c7e N |
7028 | rdev->saved_raid_disk = -1; |
7029 | ||
1da177e4 LT |
7030 | mddev->recovery = 0; |
7031 | /* flag recovery needed just to double check */ | |
7032 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
0c3573f1 | 7033 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 7034 | md_new_event(mddev); |
1da177e4 LT |
7035 | goto unlock; |
7036 | } | |
72a23c21 NB |
7037 | /* Set RUNNING before clearing NEEDED to avoid |
7038 | * any transients in the value of "sync_action". | |
7039 | */ | |
7040 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7041 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
7042 | /* Clear some bits that don't mean anything, but |
7043 | * might be left set | |
7044 | */ | |
24dd469d N |
7045 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7046 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7047 | |
5fd6c1dc N |
7048 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
7049 | goto unlock; | |
1da177e4 LT |
7050 | /* no recovery is running. |
7051 | * remove any failed drives, then | |
7052 | * add spares if possible. | |
7053 | * Spare are also removed and re-added, to allow | |
7054 | * the personality to fail the re-add. | |
7055 | */ | |
1da177e4 | 7056 | |
b4c4c7b8 | 7057 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7058 | if (mddev->pers->check_reshape == NULL || |
7059 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7060 | /* Cannot proceed */ |
7061 | goto unlock; | |
7062 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7063 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 7064 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
7065 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7066 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7067 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7068 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7069 | } else if (mddev->recovery_cp < MaxSector) { |
7070 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7071 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7072 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7073 | /* nothing to be done ... */ | |
1da177e4 | 7074 | goto unlock; |
24dd469d | 7075 | |
1da177e4 | 7076 | if (mddev->pers->sync_request) { |
a654b9d8 N |
7077 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
7078 | /* We are adding a device or devices to an array | |
7079 | * which has the bitmap stored on all devices. | |
7080 | * So make sure all bitmap pages get written | |
7081 | */ | |
7082 | bitmap_write_all(mddev->bitmap); | |
7083 | } | |
1da177e4 LT |
7084 | mddev->sync_thread = md_register_thread(md_do_sync, |
7085 | mddev, | |
0da3c619 | 7086 | "resync"); |
1da177e4 LT |
7087 | if (!mddev->sync_thread) { |
7088 | printk(KERN_ERR "%s: could not start resync" | |
7089 | " thread...\n", | |
7090 | mdname(mddev)); | |
7091 | /* leave the spares where they are, it shouldn't hurt */ | |
7092 | mddev->recovery = 0; | |
d7603b7e | 7093 | } else |
1da177e4 | 7094 | md_wakeup_thread(mddev->sync_thread); |
0c3573f1 | 7095 | sysfs_notify_dirent(mddev->sysfs_action); |
d7603b7e | 7096 | md_new_event(mddev); |
1da177e4 LT |
7097 | } |
7098 | unlock: | |
72a23c21 NB |
7099 | if (!mddev->sync_thread) { |
7100 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7101 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7102 | &mddev->recovery)) | |
0c3573f1 N |
7103 | if (mddev->sysfs_action) |
7104 | sysfs_notify_dirent(mddev->sysfs_action); | |
72a23c21 | 7105 | } |
1da177e4 LT |
7106 | mddev_unlock(mddev); |
7107 | } | |
7108 | } | |
7109 | ||
6bfe0b49 DW |
7110 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
7111 | { | |
3c0ee63a | 7112 | sysfs_notify_dirent(rdev->sysfs_state); |
6bfe0b49 DW |
7113 | wait_event_timeout(rdev->blocked_wait, |
7114 | !test_bit(Blocked, &rdev->flags), | |
7115 | msecs_to_jiffies(5000)); | |
7116 | rdev_dec_pending(rdev, mddev); | |
7117 | } | |
7118 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7119 | ||
75c96f85 AB |
7120 | static int md_notify_reboot(struct notifier_block *this, |
7121 | unsigned long code, void *x) | |
1da177e4 LT |
7122 | { |
7123 | struct list_head *tmp; | |
7124 | mddev_t *mddev; | |
7125 | ||
7126 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
7127 | ||
7128 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
7129 | ||
29ac4aa3 | 7130 | for_each_mddev(mddev, tmp) |
c71d4887 | 7131 | if (mddev_trylock(mddev)) { |
2b25000b N |
7132 | /* Force a switch to readonly even array |
7133 | * appears to still be in use. Hence | |
7134 | * the '100'. | |
7135 | */ | |
d710e138 | 7136 | do_md_stop(mddev, 1, 100); |
c71d4887 NB |
7137 | mddev_unlock(mddev); |
7138 | } | |
1da177e4 LT |
7139 | /* |
7140 | * certain more exotic SCSI devices are known to be | |
7141 | * volatile wrt too early system reboots. While the | |
7142 | * right place to handle this issue is the given | |
7143 | * driver, we do want to have a safe RAID driver ... | |
7144 | */ | |
7145 | mdelay(1000*1); | |
7146 | } | |
7147 | return NOTIFY_DONE; | |
7148 | } | |
7149 | ||
75c96f85 | 7150 | static struct notifier_block md_notifier = { |
1da177e4 LT |
7151 | .notifier_call = md_notify_reboot, |
7152 | .next = NULL, | |
7153 | .priority = INT_MAX, /* before any real devices */ | |
7154 | }; | |
7155 | ||
7156 | static void md_geninit(void) | |
7157 | { | |
1da177e4 LT |
7158 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
7159 | ||
c7705f34 | 7160 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
7161 | } |
7162 | ||
75c96f85 | 7163 | static int __init md_init(void) |
1da177e4 | 7164 | { |
3dbd8c2e | 7165 | if (register_blkdev(MD_MAJOR, "md")) |
1da177e4 LT |
7166 | return -1; |
7167 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
3dbd8c2e | 7168 | unregister_blkdev(MD_MAJOR, "md"); |
1da177e4 LT |
7169 | return -1; |
7170 | } | |
3dbd8c2e | 7171 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
7172 | md_probe, NULL, NULL); |
7173 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
7174 | md_probe, NULL, NULL); |
7175 | ||
1da177e4 | 7176 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 7177 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
7178 | |
7179 | md_geninit(); | |
d710e138 | 7180 | return 0; |
1da177e4 LT |
7181 | } |
7182 | ||
7183 | ||
7184 | #ifndef MODULE | |
7185 | ||
7186 | /* | |
7187 | * Searches all registered partitions for autorun RAID arrays | |
7188 | * at boot time. | |
7189 | */ | |
4d936ec1 ME |
7190 | |
7191 | static LIST_HEAD(all_detected_devices); | |
7192 | struct detected_devices_node { | |
7193 | struct list_head list; | |
7194 | dev_t dev; | |
7195 | }; | |
1da177e4 LT |
7196 | |
7197 | void md_autodetect_dev(dev_t dev) | |
7198 | { | |
4d936ec1 ME |
7199 | struct detected_devices_node *node_detected_dev; |
7200 | ||
7201 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
7202 | if (node_detected_dev) { | |
7203 | node_detected_dev->dev = dev; | |
7204 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
7205 | } else { | |
7206 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
7207 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
7208 | } | |
1da177e4 LT |
7209 | } |
7210 | ||
7211 | ||
7212 | static void autostart_arrays(int part) | |
7213 | { | |
7214 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
7215 | struct detected_devices_node *node_detected_dev; |
7216 | dev_t dev; | |
7217 | int i_scanned, i_passed; | |
1da177e4 | 7218 | |
4d936ec1 ME |
7219 | i_scanned = 0; |
7220 | i_passed = 0; | |
1da177e4 | 7221 | |
4d936ec1 | 7222 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 7223 | |
4d936ec1 ME |
7224 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
7225 | i_scanned++; | |
7226 | node_detected_dev = list_entry(all_detected_devices.next, | |
7227 | struct detected_devices_node, list); | |
7228 | list_del(&node_detected_dev->list); | |
7229 | dev = node_detected_dev->dev; | |
7230 | kfree(node_detected_dev); | |
df968c4e | 7231 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
7232 | if (IS_ERR(rdev)) |
7233 | continue; | |
7234 | ||
b2d444d7 | 7235 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7236 | MD_BUG(); |
7237 | continue; | |
7238 | } | |
d0fae18f | 7239 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 7240 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 7241 | i_passed++; |
1da177e4 | 7242 | } |
4d936ec1 ME |
7243 | |
7244 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
7245 | i_scanned, i_passed); | |
1da177e4 LT |
7246 | |
7247 | autorun_devices(part); | |
7248 | } | |
7249 | ||
fdee8ae4 | 7250 | #endif /* !MODULE */ |
1da177e4 LT |
7251 | |
7252 | static __exit void md_exit(void) | |
7253 | { | |
7254 | mddev_t *mddev; | |
7255 | struct list_head *tmp; | |
8ab5e4c1 | 7256 | |
3dbd8c2e | 7257 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 7258 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 7259 | |
3dbd8c2e | 7260 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
7261 | unregister_blkdev(mdp_major, "mdp"); |
7262 | unregister_reboot_notifier(&md_notifier); | |
7263 | unregister_sysctl_table(raid_table_header); | |
7264 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 7265 | for_each_mddev(mddev, tmp) { |
1da177e4 | 7266 | export_array(mddev); |
d3374825 | 7267 | mddev->hold_active = 0; |
1da177e4 LT |
7268 | } |
7269 | } | |
7270 | ||
685784aa | 7271 | subsys_initcall(md_init); |
1da177e4 LT |
7272 | module_exit(md_exit) |
7273 | ||
f91de92e N |
7274 | static int get_ro(char *buffer, struct kernel_param *kp) |
7275 | { | |
7276 | return sprintf(buffer, "%d", start_readonly); | |
7277 | } | |
7278 | static int set_ro(const char *val, struct kernel_param *kp) | |
7279 | { | |
7280 | char *e; | |
7281 | int num = simple_strtoul(val, &e, 10); | |
7282 | if (*val && (*e == '\0' || *e == '\n')) { | |
7283 | start_readonly = num; | |
4dbcdc75 | 7284 | return 0; |
f91de92e N |
7285 | } |
7286 | return -EINVAL; | |
7287 | } | |
7288 | ||
80ca3a44 N |
7289 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
7290 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 7291 | |
efeb53c0 | 7292 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 7293 | |
1da177e4 LT |
7294 | EXPORT_SYMBOL(register_md_personality); |
7295 | EXPORT_SYMBOL(unregister_md_personality); | |
7296 | EXPORT_SYMBOL(md_error); | |
7297 | EXPORT_SYMBOL(md_done_sync); | |
7298 | EXPORT_SYMBOL(md_write_start); | |
7299 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
7300 | EXPORT_SYMBOL(md_register_thread); |
7301 | EXPORT_SYMBOL(md_unregister_thread); | |
7302 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
7303 | EXPORT_SYMBOL(md_check_recovery); |
7304 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7305 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 7306 | MODULE_ALIAS("md"); |
72008652 | 7307 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |