]>
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 | ||
35 | #include <linux/module.h> | |
36 | #include <linux/config.h> | |
a6fb0934 | 37 | #include <linux/kthread.h> |
1da177e4 LT |
38 | #include <linux/linkage.h> |
39 | #include <linux/raid/md.h> | |
32a7627c | 40 | #include <linux/raid/bitmap.h> |
1da177e4 LT |
41 | #include <linux/sysctl.h> |
42 | #include <linux/devfs_fs_kernel.h> | |
43 | #include <linux/buffer_head.h> /* for invalidate_bdev */ | |
44 | #include <linux/suspend.h> | |
d7603b7e | 45 | #include <linux/poll.h> |
48c9c27b | 46 | #include <linux/mutex.h> |
16f17b39 | 47 | #include <linux/ctype.h> |
1da177e4 LT |
48 | |
49 | #include <linux/init.h> | |
50 | ||
32a7627c N |
51 | #include <linux/file.h> |
52 | ||
1da177e4 LT |
53 | #ifdef CONFIG_KMOD |
54 | #include <linux/kmod.h> | |
55 | #endif | |
56 | ||
57 | #include <asm/unaligned.h> | |
58 | ||
59 | #define MAJOR_NR MD_MAJOR | |
60 | #define MD_DRIVER | |
61 | ||
62 | /* 63 partitions with the alternate major number (mdp) */ | |
63 | #define MdpMinorShift 6 | |
64 | ||
65 | #define DEBUG 0 | |
66 | #define dprintk(x...) ((void)(DEBUG && printk(x))) | |
67 | ||
68 | ||
69 | #ifndef MODULE | |
70 | static void autostart_arrays (int part); | |
71 | #endif | |
72 | ||
2604b703 | 73 | static LIST_HEAD(pers_list); |
1da177e4 LT |
74 | static DEFINE_SPINLOCK(pers_lock); |
75 | ||
5e56341d AB |
76 | static void md_print_devices(void); |
77 | ||
78 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } | |
79 | ||
1da177e4 LT |
80 | /* |
81 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
82 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
83 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 84 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
85 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
86 | * speed limit - in case reconstruction slows down your system despite | |
87 | * idle IO detection. | |
88 | * | |
89 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 90 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
91 | */ |
92 | ||
93 | static int sysctl_speed_limit_min = 1000; | |
94 | static int sysctl_speed_limit_max = 200000; | |
88202a0c N |
95 | static inline int speed_min(mddev_t *mddev) |
96 | { | |
97 | return mddev->sync_speed_min ? | |
98 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
99 | } | |
100 | ||
101 | static inline int speed_max(mddev_t *mddev) | |
102 | { | |
103 | return mddev->sync_speed_max ? | |
104 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
105 | } | |
1da177e4 LT |
106 | |
107 | static struct ctl_table_header *raid_table_header; | |
108 | ||
109 | static ctl_table raid_table[] = { | |
110 | { | |
111 | .ctl_name = DEV_RAID_SPEED_LIMIT_MIN, | |
112 | .procname = "speed_limit_min", | |
113 | .data = &sysctl_speed_limit_min, | |
114 | .maxlen = sizeof(int), | |
115 | .mode = 0644, | |
116 | .proc_handler = &proc_dointvec, | |
117 | }, | |
118 | { | |
119 | .ctl_name = DEV_RAID_SPEED_LIMIT_MAX, | |
120 | .procname = "speed_limit_max", | |
121 | .data = &sysctl_speed_limit_max, | |
122 | .maxlen = sizeof(int), | |
123 | .mode = 0644, | |
124 | .proc_handler = &proc_dointvec, | |
125 | }, | |
126 | { .ctl_name = 0 } | |
127 | }; | |
128 | ||
129 | static ctl_table raid_dir_table[] = { | |
130 | { | |
131 | .ctl_name = DEV_RAID, | |
132 | .procname = "raid", | |
133 | .maxlen = 0, | |
134 | .mode = 0555, | |
135 | .child = raid_table, | |
136 | }, | |
137 | { .ctl_name = 0 } | |
138 | }; | |
139 | ||
140 | static ctl_table raid_root_table[] = { | |
141 | { | |
142 | .ctl_name = CTL_DEV, | |
143 | .procname = "dev", | |
144 | .maxlen = 0, | |
145 | .mode = 0555, | |
146 | .child = raid_dir_table, | |
147 | }, | |
148 | { .ctl_name = 0 } | |
149 | }; | |
150 | ||
151 | static struct block_device_operations md_fops; | |
152 | ||
f91de92e N |
153 | static int start_readonly; |
154 | ||
d7603b7e N |
155 | /* |
156 | * We have a system wide 'event count' that is incremented | |
157 | * on any 'interesting' event, and readers of /proc/mdstat | |
158 | * can use 'poll' or 'select' to find out when the event | |
159 | * count increases. | |
160 | * | |
161 | * Events are: | |
162 | * start array, stop array, error, add device, remove device, | |
163 | * start build, activate spare | |
164 | */ | |
2989ddbd | 165 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 166 | static atomic_t md_event_count; |
29269553 | 167 | void md_new_event(mddev_t *mddev) |
d7603b7e N |
168 | { |
169 | atomic_inc(&md_event_count); | |
170 | wake_up(&md_event_waiters); | |
4508a7a7 | 171 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
d7603b7e | 172 | } |
29269553 | 173 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 174 | |
c331eb04 N |
175 | /* Alternate version that can be called from interrupts |
176 | * when calling sysfs_notify isn't needed. | |
177 | */ | |
178 | void md_new_event_inintr(mddev_t *mddev) | |
179 | { | |
180 | atomic_inc(&md_event_count); | |
181 | wake_up(&md_event_waiters); | |
182 | } | |
183 | ||
1da177e4 LT |
184 | /* |
185 | * Enables to iterate over all existing md arrays | |
186 | * all_mddevs_lock protects this list. | |
187 | */ | |
188 | static LIST_HEAD(all_mddevs); | |
189 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
190 | ||
191 | ||
192 | /* | |
193 | * iterates through all used mddevs in the system. | |
194 | * We take care to grab the all_mddevs_lock whenever navigating | |
195 | * the list, and to always hold a refcount when unlocked. | |
196 | * Any code which breaks out of this loop while own | |
197 | * a reference to the current mddev and must mddev_put it. | |
198 | */ | |
199 | #define ITERATE_MDDEV(mddev,tmp) \ | |
200 | \ | |
201 | for (({ spin_lock(&all_mddevs_lock); \ | |
202 | tmp = all_mddevs.next; \ | |
203 | mddev = NULL;}); \ | |
204 | ({ if (tmp != &all_mddevs) \ | |
205 | mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ | |
206 | spin_unlock(&all_mddevs_lock); \ | |
207 | if (mddev) mddev_put(mddev); \ | |
208 | mddev = list_entry(tmp, mddev_t, all_mddevs); \ | |
209 | tmp != &all_mddevs;}); \ | |
210 | ({ spin_lock(&all_mddevs_lock); \ | |
211 | tmp = tmp->next;}) \ | |
212 | ) | |
213 | ||
214 | ||
215 | static int md_fail_request (request_queue_t *q, struct bio *bio) | |
216 | { | |
217 | bio_io_error(bio, bio->bi_size); | |
218 | return 0; | |
219 | } | |
220 | ||
221 | static inline mddev_t *mddev_get(mddev_t *mddev) | |
222 | { | |
223 | atomic_inc(&mddev->active); | |
224 | return mddev; | |
225 | } | |
226 | ||
227 | static void mddev_put(mddev_t *mddev) | |
228 | { | |
229 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
230 | return; | |
231 | if (!mddev->raid_disks && list_empty(&mddev->disks)) { | |
232 | list_del(&mddev->all_mddevs); | |
926ce2d8 | 233 | spin_unlock(&all_mddevs_lock); |
1312f40e | 234 | blk_cleanup_queue(mddev->queue); |
eae1701f | 235 | kobject_unregister(&mddev->kobj); |
926ce2d8 N |
236 | } else |
237 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
238 | } |
239 | ||
240 | static mddev_t * mddev_find(dev_t unit) | |
241 | { | |
242 | mddev_t *mddev, *new = NULL; | |
243 | ||
244 | retry: | |
245 | spin_lock(&all_mddevs_lock); | |
246 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
247 | if (mddev->unit == unit) { | |
248 | mddev_get(mddev); | |
249 | spin_unlock(&all_mddevs_lock); | |
990a8baf | 250 | kfree(new); |
1da177e4 LT |
251 | return mddev; |
252 | } | |
253 | ||
254 | if (new) { | |
255 | list_add(&new->all_mddevs, &all_mddevs); | |
256 | spin_unlock(&all_mddevs_lock); | |
257 | return new; | |
258 | } | |
259 | spin_unlock(&all_mddevs_lock); | |
260 | ||
9ffae0cf | 261 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
262 | if (!new) |
263 | return NULL; | |
264 | ||
1da177e4 LT |
265 | new->unit = unit; |
266 | if (MAJOR(unit) == MD_MAJOR) | |
267 | new->md_minor = MINOR(unit); | |
268 | else | |
269 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
270 | ||
df5b89b3 | 271 | mutex_init(&new->reconfig_mutex); |
1da177e4 LT |
272 | INIT_LIST_HEAD(&new->disks); |
273 | INIT_LIST_HEAD(&new->all_mddevs); | |
274 | init_timer(&new->safemode_timer); | |
275 | atomic_set(&new->active, 1); | |
06d91a5f | 276 | spin_lock_init(&new->write_lock); |
3d310eb7 | 277 | init_waitqueue_head(&new->sb_wait); |
1da177e4 LT |
278 | |
279 | new->queue = blk_alloc_queue(GFP_KERNEL); | |
280 | if (!new->queue) { | |
281 | kfree(new); | |
282 | return NULL; | |
283 | } | |
89e5c8b5 | 284 | set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags); |
1da177e4 LT |
285 | |
286 | blk_queue_make_request(new->queue, md_fail_request); | |
287 | ||
288 | goto retry; | |
289 | } | |
290 | ||
291 | static inline int mddev_lock(mddev_t * mddev) | |
292 | { | |
df5b89b3 | 293 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
294 | } |
295 | ||
1da177e4 LT |
296 | static inline int mddev_trylock(mddev_t * mddev) |
297 | { | |
df5b89b3 | 298 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
299 | } |
300 | ||
301 | static inline void mddev_unlock(mddev_t * mddev) | |
302 | { | |
df5b89b3 | 303 | mutex_unlock(&mddev->reconfig_mutex); |
1da177e4 | 304 | |
005eca5e | 305 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
306 | } |
307 | ||
2989ddbd | 308 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 LT |
309 | { |
310 | mdk_rdev_t * rdev; | |
311 | struct list_head *tmp; | |
312 | ||
313 | ITERATE_RDEV(mddev,rdev,tmp) { | |
314 | if (rdev->desc_nr == nr) | |
315 | return rdev; | |
316 | } | |
317 | return NULL; | |
318 | } | |
319 | ||
320 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
321 | { | |
322 | struct list_head *tmp; | |
323 | mdk_rdev_t *rdev; | |
324 | ||
325 | ITERATE_RDEV(mddev,rdev,tmp) { | |
326 | if (rdev->bdev->bd_dev == dev) | |
327 | return rdev; | |
328 | } | |
329 | return NULL; | |
330 | } | |
331 | ||
d9d166c2 | 332 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
333 | { |
334 | struct mdk_personality *pers; | |
d9d166c2 N |
335 | list_for_each_entry(pers, &pers_list, list) { |
336 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 337 | return pers; |
d9d166c2 N |
338 | if (strcmp(pers->name, clevel)==0) |
339 | return pers; | |
340 | } | |
2604b703 N |
341 | return NULL; |
342 | } | |
343 | ||
77933d72 | 344 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 LT |
345 | { |
346 | sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
347 | return MD_NEW_SIZE_BLOCKS(size); | |
348 | } | |
349 | ||
350 | static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size) | |
351 | { | |
352 | sector_t size; | |
353 | ||
354 | size = rdev->sb_offset; | |
355 | ||
356 | if (chunk_size) | |
357 | size &= ~((sector_t)chunk_size/1024 - 1); | |
358 | return size; | |
359 | } | |
360 | ||
361 | static int alloc_disk_sb(mdk_rdev_t * rdev) | |
362 | { | |
363 | if (rdev->sb_page) | |
364 | MD_BUG(); | |
365 | ||
366 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
367 | if (!rdev->sb_page) { | |
368 | printk(KERN_ALERT "md: out of memory.\n"); | |
369 | return -EINVAL; | |
370 | } | |
371 | ||
372 | return 0; | |
373 | } | |
374 | ||
375 | static void free_disk_sb(mdk_rdev_t * rdev) | |
376 | { | |
377 | if (rdev->sb_page) { | |
2d1f3b5d | 378 | put_page(rdev->sb_page); |
1da177e4 LT |
379 | rdev->sb_loaded = 0; |
380 | rdev->sb_page = NULL; | |
381 | rdev->sb_offset = 0; | |
382 | rdev->size = 0; | |
383 | } | |
384 | } | |
385 | ||
386 | ||
7bfa19f2 N |
387 | static int super_written(struct bio *bio, unsigned int bytes_done, int error) |
388 | { | |
389 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 390 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 N |
391 | if (bio->bi_size) |
392 | return 1; | |
393 | ||
394 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
a9701a30 | 395 | md_error(mddev, rdev); |
7bfa19f2 | 396 | |
a9701a30 N |
397 | if (atomic_dec_and_test(&mddev->pending_writes)) |
398 | wake_up(&mddev->sb_wait); | |
f8b58edf | 399 | bio_put(bio); |
7bfa19f2 N |
400 | return 0; |
401 | } | |
402 | ||
a9701a30 N |
403 | static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error) |
404 | { | |
405 | struct bio *bio2 = bio->bi_private; | |
406 | mdk_rdev_t *rdev = bio2->bi_private; | |
407 | mddev_t *mddev = rdev->mddev; | |
408 | if (bio->bi_size) | |
409 | return 1; | |
410 | ||
411 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
412 | error == -EOPNOTSUPP) { | |
413 | unsigned long flags; | |
414 | /* barriers don't appear to be supported :-( */ | |
415 | set_bit(BarriersNotsupp, &rdev->flags); | |
416 | mddev->barriers_work = 0; | |
417 | spin_lock_irqsave(&mddev->write_lock, flags); | |
418 | bio2->bi_next = mddev->biolist; | |
419 | mddev->biolist = bio2; | |
420 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
421 | wake_up(&mddev->sb_wait); | |
422 | bio_put(bio); | |
423 | return 0; | |
424 | } | |
425 | bio_put(bio2); | |
426 | bio->bi_private = rdev; | |
427 | return super_written(bio, bytes_done, error); | |
428 | } | |
429 | ||
7bfa19f2 N |
430 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
431 | sector_t sector, int size, struct page *page) | |
432 | { | |
433 | /* write first size bytes of page to sector of rdev | |
434 | * Increment mddev->pending_writes before returning | |
435 | * and decrement it on completion, waking up sb_wait | |
436 | * if zero is reached. | |
437 | * If an error occurred, call md_error | |
a9701a30 N |
438 | * |
439 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
440 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
441 | */ |
442 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
a9701a30 | 443 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC); |
7bfa19f2 N |
444 | |
445 | bio->bi_bdev = rdev->bdev; | |
446 | bio->bi_sector = sector; | |
447 | bio_add_page(bio, page, size, 0); | |
448 | bio->bi_private = rdev; | |
449 | bio->bi_end_io = super_written; | |
a9701a30 N |
450 | bio->bi_rw = rw; |
451 | ||
7bfa19f2 | 452 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
453 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
454 | struct bio *rbio; | |
455 | rw |= (1<<BIO_RW_BARRIER); | |
456 | rbio = bio_clone(bio, GFP_NOIO); | |
457 | rbio->bi_private = bio; | |
458 | rbio->bi_end_io = super_written_barrier; | |
459 | submit_bio(rw, rbio); | |
460 | } else | |
461 | submit_bio(rw, bio); | |
462 | } | |
463 | ||
464 | void md_super_wait(mddev_t *mddev) | |
465 | { | |
466 | /* wait for all superblock writes that were scheduled to complete. | |
467 | * if any had to be retried (due to BARRIER problems), retry them | |
468 | */ | |
469 | DEFINE_WAIT(wq); | |
470 | for(;;) { | |
471 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
472 | if (atomic_read(&mddev->pending_writes)==0) | |
473 | break; | |
474 | while (mddev->biolist) { | |
475 | struct bio *bio; | |
476 | spin_lock_irq(&mddev->write_lock); | |
477 | bio = mddev->biolist; | |
478 | mddev->biolist = bio->bi_next ; | |
479 | bio->bi_next = NULL; | |
480 | spin_unlock_irq(&mddev->write_lock); | |
481 | submit_bio(bio->bi_rw, bio); | |
482 | } | |
483 | schedule(); | |
484 | } | |
485 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
486 | } |
487 | ||
1da177e4 LT |
488 | static int bi_complete(struct bio *bio, unsigned int bytes_done, int error) |
489 | { | |
490 | if (bio->bi_size) | |
491 | return 1; | |
492 | ||
493 | complete((struct completion*)bio->bi_private); | |
494 | return 0; | |
495 | } | |
496 | ||
a654b9d8 | 497 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
498 | struct page *page, int rw) |
499 | { | |
baaa2c51 | 500 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
501 | struct completion event; |
502 | int ret; | |
503 | ||
504 | rw |= (1 << BIO_RW_SYNC); | |
505 | ||
506 | bio->bi_bdev = bdev; | |
507 | bio->bi_sector = sector; | |
508 | bio_add_page(bio, page, size, 0); | |
509 | init_completion(&event); | |
510 | bio->bi_private = &event; | |
511 | bio->bi_end_io = bi_complete; | |
512 | submit_bio(rw, bio); | |
513 | wait_for_completion(&event); | |
514 | ||
515 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
516 | bio_put(bio); | |
517 | return ret; | |
518 | } | |
a8745db2 | 519 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 520 | |
0002b271 | 521 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
522 | { |
523 | char b[BDEVNAME_SIZE]; | |
524 | if (!rdev->sb_page) { | |
525 | MD_BUG(); | |
526 | return -EINVAL; | |
527 | } | |
528 | if (rdev->sb_loaded) | |
529 | return 0; | |
530 | ||
531 | ||
0002b271 | 532 | if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ)) |
1da177e4 LT |
533 | goto fail; |
534 | rdev->sb_loaded = 1; | |
535 | return 0; | |
536 | ||
537 | fail: | |
538 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
539 | bdevname(rdev->bdev,b)); | |
540 | return -EINVAL; | |
541 | } | |
542 | ||
543 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
544 | { | |
545 | if ( (sb1->set_uuid0 == sb2->set_uuid0) && | |
546 | (sb1->set_uuid1 == sb2->set_uuid1) && | |
547 | (sb1->set_uuid2 == sb2->set_uuid2) && | |
548 | (sb1->set_uuid3 == sb2->set_uuid3)) | |
549 | ||
550 | return 1; | |
551 | ||
552 | return 0; | |
553 | } | |
554 | ||
555 | ||
556 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
557 | { | |
558 | int ret; | |
559 | mdp_super_t *tmp1, *tmp2; | |
560 | ||
561 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
562 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
563 | ||
564 | if (!tmp1 || !tmp2) { | |
565 | ret = 0; | |
566 | printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n"); | |
567 | goto abort; | |
568 | } | |
569 | ||
570 | *tmp1 = *sb1; | |
571 | *tmp2 = *sb2; | |
572 | ||
573 | /* | |
574 | * nr_disks is not constant | |
575 | */ | |
576 | tmp1->nr_disks = 0; | |
577 | tmp2->nr_disks = 0; | |
578 | ||
579 | if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4)) | |
580 | ret = 0; | |
581 | else | |
582 | ret = 1; | |
583 | ||
584 | abort: | |
990a8baf JJ |
585 | kfree(tmp1); |
586 | kfree(tmp2); | |
1da177e4 LT |
587 | return ret; |
588 | } | |
589 | ||
590 | static unsigned int calc_sb_csum(mdp_super_t * sb) | |
591 | { | |
592 | unsigned int disk_csum, csum; | |
593 | ||
594 | disk_csum = sb->sb_csum; | |
595 | sb->sb_csum = 0; | |
596 | csum = csum_partial((void *)sb, MD_SB_BYTES, 0); | |
597 | sb->sb_csum = disk_csum; | |
598 | return csum; | |
599 | } | |
600 | ||
601 | ||
602 | /* | |
603 | * Handle superblock details. | |
604 | * We want to be able to handle multiple superblock formats | |
605 | * so we have a common interface to them all, and an array of | |
606 | * different handlers. | |
607 | * We rely on user-space to write the initial superblock, and support | |
608 | * reading and updating of superblocks. | |
609 | * Interface methods are: | |
610 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
611 | * loads and validates a superblock on dev. | |
612 | * if refdev != NULL, compare superblocks on both devices | |
613 | * Return: | |
614 | * 0 - dev has a superblock that is compatible with refdev | |
615 | * 1 - dev has a superblock that is compatible and newer than refdev | |
616 | * so dev should be used as the refdev in future | |
617 | * -EINVAL superblock incompatible or invalid | |
618 | * -othererror e.g. -EIO | |
619 | * | |
620 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
621 | * Verify that dev is acceptable into mddev. | |
622 | * The first time, mddev->raid_disks will be 0, and data from | |
623 | * dev should be merged in. Subsequent calls check that dev | |
624 | * is new enough. Return 0 or -EINVAL | |
625 | * | |
626 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
627 | * Update the superblock for rdev with data in mddev | |
628 | * This does not write to disc. | |
629 | * | |
630 | */ | |
631 | ||
632 | struct super_type { | |
633 | char *name; | |
634 | struct module *owner; | |
635 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version); | |
636 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
637 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
638 | }; | |
639 | ||
640 | /* | |
641 | * load_super for 0.90.0 | |
642 | */ | |
643 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
644 | { | |
645 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
646 | mdp_super_t *sb; | |
647 | int ret; | |
648 | sector_t sb_offset; | |
649 | ||
650 | /* | |
651 | * Calculate the position of the superblock, | |
652 | * it's at the end of the disk. | |
653 | * | |
654 | * It also happens to be a multiple of 4Kb. | |
655 | */ | |
656 | sb_offset = calc_dev_sboffset(rdev->bdev); | |
657 | rdev->sb_offset = sb_offset; | |
658 | ||
0002b271 | 659 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
660 | if (ret) return ret; |
661 | ||
662 | ret = -EINVAL; | |
663 | ||
664 | bdevname(rdev->bdev, b); | |
665 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
666 | ||
667 | if (sb->md_magic != MD_SB_MAGIC) { | |
668 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
669 | b); | |
670 | goto abort; | |
671 | } | |
672 | ||
673 | if (sb->major_version != 0 || | |
f6705578 N |
674 | sb->minor_version < 90 || |
675 | sb->minor_version > 91) { | |
1da177e4 LT |
676 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
677 | sb->major_version, sb->minor_version, | |
678 | b); | |
679 | goto abort; | |
680 | } | |
681 | ||
682 | if (sb->raid_disks <= 0) | |
683 | goto abort; | |
684 | ||
685 | if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) { | |
686 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", | |
687 | b); | |
688 | goto abort; | |
689 | } | |
690 | ||
691 | rdev->preferred_minor = sb->md_minor; | |
692 | rdev->data_offset = 0; | |
0002b271 | 693 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
694 | |
695 | if (sb->level == LEVEL_MULTIPATH) | |
696 | rdev->desc_nr = -1; | |
697 | else | |
698 | rdev->desc_nr = sb->this_disk.number; | |
699 | ||
700 | if (refdev == 0) | |
701 | ret = 1; | |
702 | else { | |
703 | __u64 ev1, ev2; | |
704 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
705 | if (!uuid_equal(refsb, sb)) { | |
706 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
707 | b, bdevname(refdev->bdev,b2)); | |
708 | goto abort; | |
709 | } | |
710 | if (!sb_equal(refsb, sb)) { | |
711 | printk(KERN_WARNING "md: %s has same UUID" | |
712 | " but different superblock to %s\n", | |
713 | b, bdevname(refdev->bdev, b2)); | |
714 | goto abort; | |
715 | } | |
716 | ev1 = md_event(sb); | |
717 | ev2 = md_event(refsb); | |
718 | if (ev1 > ev2) | |
719 | ret = 1; | |
720 | else | |
721 | ret = 0; | |
722 | } | |
723 | rdev->size = calc_dev_size(rdev, sb->chunk_size); | |
724 | ||
2bf071bf N |
725 | if (rdev->size < sb->size && sb->level > 1) |
726 | /* "this cannot possibly happen" ... */ | |
727 | ret = -EINVAL; | |
728 | ||
1da177e4 LT |
729 | abort: |
730 | return ret; | |
731 | } | |
732 | ||
733 | /* | |
734 | * validate_super for 0.90.0 | |
735 | */ | |
736 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
737 | { | |
738 | mdp_disk_t *desc; | |
739 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 740 | __u64 ev1 = md_event(sb); |
1da177e4 | 741 | |
41158c7e | 742 | rdev->raid_disk = -1; |
b2d444d7 | 743 | rdev->flags = 0; |
1da177e4 LT |
744 | if (mddev->raid_disks == 0) { |
745 | mddev->major_version = 0; | |
746 | mddev->minor_version = sb->minor_version; | |
747 | mddev->patch_version = sb->patch_version; | |
748 | mddev->persistent = ! sb->not_persistent; | |
749 | mddev->chunk_size = sb->chunk_size; | |
750 | mddev->ctime = sb->ctime; | |
751 | mddev->utime = sb->utime; | |
752 | mddev->level = sb->level; | |
d9d166c2 | 753 | mddev->clevel[0] = 0; |
1da177e4 LT |
754 | mddev->layout = sb->layout; |
755 | mddev->raid_disks = sb->raid_disks; | |
756 | mddev->size = sb->size; | |
07d84d10 | 757 | mddev->events = ev1; |
9223214e | 758 | mddev->bitmap_offset = 0; |
36fa3063 | 759 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
1da177e4 | 760 | |
f6705578 N |
761 | if (mddev->minor_version >= 91) { |
762 | mddev->reshape_position = sb->reshape_position; | |
763 | mddev->delta_disks = sb->delta_disks; | |
764 | mddev->new_level = sb->new_level; | |
765 | mddev->new_layout = sb->new_layout; | |
766 | mddev->new_chunk = sb->new_chunk; | |
767 | } else { | |
768 | mddev->reshape_position = MaxSector; | |
769 | mddev->delta_disks = 0; | |
770 | mddev->new_level = mddev->level; | |
771 | mddev->new_layout = mddev->layout; | |
772 | mddev->new_chunk = mddev->chunk_size; | |
773 | } | |
774 | ||
1da177e4 LT |
775 | if (sb->state & (1<<MD_SB_CLEAN)) |
776 | mddev->recovery_cp = MaxSector; | |
777 | else { | |
778 | if (sb->events_hi == sb->cp_events_hi && | |
779 | sb->events_lo == sb->cp_events_lo) { | |
780 | mddev->recovery_cp = sb->recovery_cp; | |
781 | } else | |
782 | mddev->recovery_cp = 0; | |
783 | } | |
784 | ||
785 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
786 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
787 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
788 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
789 | ||
790 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
791 | |
792 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
793 | mddev->bitmap_file == NULL) { | |
c5a10f62 N |
794 | if (mddev->level != 1 && mddev->level != 4 |
795 | && mddev->level != 5 && mddev->level != 6 | |
6cce3b23 | 796 | && mddev->level != 10) { |
a654b9d8 | 797 | /* FIXME use a better test */ |
6cce3b23 | 798 | printk(KERN_WARNING "md: bitmaps not supported for this level.\n"); |
a654b9d8 N |
799 | return -EINVAL; |
800 | } | |
36fa3063 | 801 | mddev->bitmap_offset = mddev->default_bitmap_offset; |
a654b9d8 N |
802 | } |
803 | ||
41158c7e N |
804 | } else if (mddev->pers == NULL) { |
805 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
806 | ++ev1; |
807 | if (ev1 < mddev->events) | |
808 | return -EINVAL; | |
41158c7e N |
809 | } else if (mddev->bitmap) { |
810 | /* if adding to array with a bitmap, then we can accept an | |
811 | * older device ... but not too old. | |
812 | */ | |
41158c7e N |
813 | if (ev1 < mddev->bitmap->events_cleared) |
814 | return 0; | |
07d84d10 N |
815 | } else { |
816 | if (ev1 < mddev->events) | |
817 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
818 | return 0; | |
819 | } | |
41158c7e | 820 | |
1da177e4 | 821 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
822 | desc = sb->disks + rdev->desc_nr; |
823 | ||
824 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 825 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
826 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
827 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 828 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
829 | rdev->raid_disk = desc->raid_disk; |
830 | } | |
8ddf9efe N |
831 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
832 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 833 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 834 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
835 | return 0; |
836 | } | |
837 | ||
838 | /* | |
839 | * sync_super for 0.90.0 | |
840 | */ | |
841 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
842 | { | |
843 | mdp_super_t *sb; | |
844 | struct list_head *tmp; | |
845 | mdk_rdev_t *rdev2; | |
846 | int next_spare = mddev->raid_disks; | |
19133a42 | 847 | |
1da177e4 LT |
848 | |
849 | /* make rdev->sb match mddev data.. | |
850 | * | |
851 | * 1/ zero out disks | |
852 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
853 | * 3/ any empty disks < next_spare become removed | |
854 | * | |
855 | * disks[0] gets initialised to REMOVED because | |
856 | * we cannot be sure from other fields if it has | |
857 | * been initialised or not. | |
858 | */ | |
859 | int i; | |
860 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
861 | ||
61181565 N |
862 | rdev->sb_size = MD_SB_BYTES; |
863 | ||
1da177e4 LT |
864 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
865 | ||
866 | memset(sb, 0, sizeof(*sb)); | |
867 | ||
868 | sb->md_magic = MD_SB_MAGIC; | |
869 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
870 | sb->patch_version = mddev->patch_version; |
871 | sb->gvalid_words = 0; /* ignored */ | |
872 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
873 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
874 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
875 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
876 | ||
877 | sb->ctime = mddev->ctime; | |
878 | sb->level = mddev->level; | |
879 | sb->size = mddev->size; | |
880 | sb->raid_disks = mddev->raid_disks; | |
881 | sb->md_minor = mddev->md_minor; | |
882 | sb->not_persistent = !mddev->persistent; | |
883 | sb->utime = mddev->utime; | |
884 | sb->state = 0; | |
885 | sb->events_hi = (mddev->events>>32); | |
886 | sb->events_lo = (u32)mddev->events; | |
887 | ||
f6705578 N |
888 | if (mddev->reshape_position == MaxSector) |
889 | sb->minor_version = 90; | |
890 | else { | |
891 | sb->minor_version = 91; | |
892 | sb->reshape_position = mddev->reshape_position; | |
893 | sb->new_level = mddev->new_level; | |
894 | sb->delta_disks = mddev->delta_disks; | |
895 | sb->new_layout = mddev->new_layout; | |
896 | sb->new_chunk = mddev->new_chunk; | |
897 | } | |
898 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
899 | if (mddev->in_sync) |
900 | { | |
901 | sb->recovery_cp = mddev->recovery_cp; | |
902 | sb->cp_events_hi = (mddev->events>>32); | |
903 | sb->cp_events_lo = (u32)mddev->events; | |
904 | if (mddev->recovery_cp == MaxSector) | |
905 | sb->state = (1<< MD_SB_CLEAN); | |
906 | } else | |
907 | sb->recovery_cp = 0; | |
908 | ||
909 | sb->layout = mddev->layout; | |
910 | sb->chunk_size = mddev->chunk_size; | |
911 | ||
a654b9d8 N |
912 | if (mddev->bitmap && mddev->bitmap_file == NULL) |
913 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); | |
914 | ||
1da177e4 LT |
915 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
916 | ITERATE_RDEV(mddev,rdev2,tmp) { | |
917 | mdp_disk_t *d; | |
86e6ffdd | 918 | int desc_nr; |
b2d444d7 N |
919 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
920 | && !test_bit(Faulty, &rdev2->flags)) | |
86e6ffdd | 921 | desc_nr = rdev2->raid_disk; |
1da177e4 | 922 | else |
86e6ffdd | 923 | desc_nr = next_spare++; |
19133a42 | 924 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
925 | d = &sb->disks[rdev2->desc_nr]; |
926 | nr_disks++; | |
927 | d->number = rdev2->desc_nr; | |
928 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
929 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
b2d444d7 N |
930 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
931 | && !test_bit(Faulty, &rdev2->flags)) | |
1da177e4 LT |
932 | d->raid_disk = rdev2->raid_disk; |
933 | else | |
934 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 935 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 936 | d->state = (1<<MD_DISK_FAULTY); |
1be7892f | 937 | else if (test_bit(In_sync, &rdev2->flags)) { |
1da177e4 LT |
938 | d->state = (1<<MD_DISK_ACTIVE); |
939 | d->state |= (1<<MD_DISK_SYNC); | |
940 | active++; | |
941 | working++; | |
942 | } else { | |
943 | d->state = 0; | |
944 | spare++; | |
945 | working++; | |
946 | } | |
8ddf9efe N |
947 | if (test_bit(WriteMostly, &rdev2->flags)) |
948 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 949 | } |
1da177e4 LT |
950 | /* now set the "removed" and "faulty" bits on any missing devices */ |
951 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
952 | mdp_disk_t *d = &sb->disks[i]; | |
953 | if (d->state == 0 && d->number == 0) { | |
954 | d->number = i; | |
955 | d->raid_disk = i; | |
956 | d->state = (1<<MD_DISK_REMOVED); | |
957 | d->state |= (1<<MD_DISK_FAULTY); | |
958 | failed++; | |
959 | } | |
960 | } | |
961 | sb->nr_disks = nr_disks; | |
962 | sb->active_disks = active; | |
963 | sb->working_disks = working; | |
964 | sb->failed_disks = failed; | |
965 | sb->spare_disks = spare; | |
966 | ||
967 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
968 | sb->sb_csum = calc_sb_csum(sb); | |
969 | } | |
970 | ||
971 | /* | |
972 | * version 1 superblock | |
973 | */ | |
974 | ||
975 | static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb) | |
976 | { | |
977 | unsigned int disk_csum, csum; | |
978 | unsigned long long newcsum; | |
979 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
980 | unsigned int *isuper = (unsigned int*)sb; | |
981 | int i; | |
982 | ||
983 | disk_csum = sb->sb_csum; | |
984 | sb->sb_csum = 0; | |
985 | newcsum = 0; | |
986 | for (i=0; size>=4; size -= 4 ) | |
987 | newcsum += le32_to_cpu(*isuper++); | |
988 | ||
989 | if (size == 2) | |
990 | newcsum += le16_to_cpu(*(unsigned short*) isuper); | |
991 | ||
992 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
993 | sb->sb_csum = disk_csum; | |
994 | return cpu_to_le32(csum); | |
995 | } | |
996 | ||
997 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
998 | { | |
999 | struct mdp_superblock_1 *sb; | |
1000 | int ret; | |
1001 | sector_t sb_offset; | |
1002 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
0002b271 | 1003 | int bmask; |
1da177e4 LT |
1004 | |
1005 | /* | |
1006 | * Calculate the position of the superblock. | |
1007 | * It is always aligned to a 4K boundary and | |
1008 | * depeding on minor_version, it can be: | |
1009 | * 0: At least 8K, but less than 12K, from end of device | |
1010 | * 1: At start of device | |
1011 | * 2: 4K from start of device. | |
1012 | */ | |
1013 | switch(minor_version) { | |
1014 | case 0: | |
1015 | sb_offset = rdev->bdev->bd_inode->i_size >> 9; | |
1016 | sb_offset -= 8*2; | |
39730960 | 1017 | sb_offset &= ~(sector_t)(4*2-1); |
1da177e4 LT |
1018 | /* convert from sectors to K */ |
1019 | sb_offset /= 2; | |
1020 | break; | |
1021 | case 1: | |
1022 | sb_offset = 0; | |
1023 | break; | |
1024 | case 2: | |
1025 | sb_offset = 4; | |
1026 | break; | |
1027 | default: | |
1028 | return -EINVAL; | |
1029 | } | |
1030 | rdev->sb_offset = sb_offset; | |
1031 | ||
0002b271 N |
1032 | /* superblock is rarely larger than 1K, but it can be larger, |
1033 | * and it is safe to read 4k, so we do that | |
1034 | */ | |
1035 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1036 | if (ret) return ret; |
1037 | ||
1038 | ||
1039 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1040 | ||
1041 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1042 | sb->major_version != cpu_to_le32(1) || | |
1043 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
1044 | le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) || | |
71c0805c | 1045 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1046 | return -EINVAL; |
1047 | ||
1048 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1049 | printk("md: invalid superblock checksum on %s\n", | |
1050 | bdevname(rdev->bdev,b)); | |
1051 | return -EINVAL; | |
1052 | } | |
1053 | if (le64_to_cpu(sb->data_size) < 10) { | |
1054 | printk("md: data_size too small on %s\n", | |
1055 | bdevname(rdev->bdev,b)); | |
1056 | return -EINVAL; | |
1057 | } | |
1058 | rdev->preferred_minor = 0xffff; | |
1059 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1060 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1061 | |
0002b271 | 1062 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
720a3dc3 | 1063 | bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 N |
1064 | if (rdev->sb_size & bmask) |
1065 | rdev-> sb_size = (rdev->sb_size | bmask)+1; | |
1066 | ||
1da177e4 | 1067 | if (refdev == 0) |
8ed75463 | 1068 | ret = 1; |
1da177e4 LT |
1069 | else { |
1070 | __u64 ev1, ev2; | |
1071 | struct mdp_superblock_1 *refsb = | |
1072 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1073 | ||
1074 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1075 | sb->level != refsb->level || | |
1076 | sb->layout != refsb->layout || | |
1077 | sb->chunksize != refsb->chunksize) { | |
1078 | printk(KERN_WARNING "md: %s has strangely different" | |
1079 | " superblock to %s\n", | |
1080 | bdevname(rdev->bdev,b), | |
1081 | bdevname(refdev->bdev,b2)); | |
1082 | return -EINVAL; | |
1083 | } | |
1084 | ev1 = le64_to_cpu(sb->events); | |
1085 | ev2 = le64_to_cpu(refsb->events); | |
1086 | ||
1087 | if (ev1 > ev2) | |
8ed75463 N |
1088 | ret = 1; |
1089 | else | |
1090 | ret = 0; | |
1da177e4 LT |
1091 | } |
1092 | if (minor_version) | |
1093 | rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2; | |
1094 | else | |
1095 | rdev->size = rdev->sb_offset; | |
1096 | if (rdev->size < le64_to_cpu(sb->data_size)/2) | |
1097 | return -EINVAL; | |
1098 | rdev->size = le64_to_cpu(sb->data_size)/2; | |
1099 | if (le32_to_cpu(sb->chunksize)) | |
1100 | rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1); | |
2bf071bf N |
1101 | |
1102 | if (le32_to_cpu(sb->size) > rdev->size*2) | |
1103 | return -EINVAL; | |
8ed75463 | 1104 | return ret; |
1da177e4 LT |
1105 | } |
1106 | ||
1107 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1108 | { | |
1109 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1110 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1111 | |
41158c7e | 1112 | rdev->raid_disk = -1; |
b2d444d7 | 1113 | rdev->flags = 0; |
1da177e4 LT |
1114 | if (mddev->raid_disks == 0) { |
1115 | mddev->major_version = 1; | |
1116 | mddev->patch_version = 0; | |
1117 | mddev->persistent = 1; | |
1118 | mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9; | |
1119 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); | |
1120 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1121 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1122 | mddev->clevel[0] = 0; |
1da177e4 LT |
1123 | mddev->layout = le32_to_cpu(sb->layout); |
1124 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
1125 | mddev->size = le64_to_cpu(sb->size)/2; | |
07d84d10 | 1126 | mddev->events = ev1; |
9223214e | 1127 | mddev->bitmap_offset = 0; |
29fc7e3e | 1128 | mddev->default_bitmap_offset = 1024 >> 9; |
1da177e4 LT |
1129 | |
1130 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1131 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1132 | ||
1133 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1134 | |
71c0805c | 1135 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
a654b9d8 | 1136 | mddev->bitmap_file == NULL ) { |
6cce3b23 N |
1137 | if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6 |
1138 | && mddev->level != 10) { | |
1139 | printk(KERN_WARNING "md: bitmaps not supported for this level.\n"); | |
a654b9d8 N |
1140 | return -EINVAL; |
1141 | } | |
1142 | mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); | |
1143 | } | |
f6705578 N |
1144 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1145 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1146 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1147 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1148 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
1149 | mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9; | |
1150 | } else { | |
1151 | mddev->reshape_position = MaxSector; | |
1152 | mddev->delta_disks = 0; | |
1153 | mddev->new_level = mddev->level; | |
1154 | mddev->new_layout = mddev->layout; | |
1155 | mddev->new_chunk = mddev->chunk_size; | |
1156 | } | |
1157 | ||
41158c7e N |
1158 | } else if (mddev->pers == NULL) { |
1159 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1160 | ++ev1; |
1161 | if (ev1 < mddev->events) | |
1162 | return -EINVAL; | |
41158c7e N |
1163 | } else if (mddev->bitmap) { |
1164 | /* If adding to array with a bitmap, then we can accept an | |
1165 | * older device, but not too old. | |
1166 | */ | |
41158c7e N |
1167 | if (ev1 < mddev->bitmap->events_cleared) |
1168 | return 0; | |
07d84d10 N |
1169 | } else { |
1170 | if (ev1 < mddev->events) | |
1171 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1172 | return 0; | |
1173 | } | |
1da177e4 LT |
1174 | if (mddev->level != LEVEL_MULTIPATH) { |
1175 | int role; | |
1176 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1177 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1178 | switch(role) { | |
1179 | case 0xffff: /* spare */ | |
1da177e4 LT |
1180 | break; |
1181 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1182 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1183 | break; |
1184 | default: | |
5fd6c1dc N |
1185 | if ((le32_to_cpu(sb->feature_map) & |
1186 | MD_FEATURE_RECOVERY_OFFSET)) | |
1187 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1188 | else | |
1189 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1190 | rdev->raid_disk = role; |
1191 | break; | |
1192 | } | |
8ddf9efe N |
1193 | if (sb->devflags & WriteMostly1) |
1194 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1195 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1196 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1197 | |
1da177e4 LT |
1198 | return 0; |
1199 | } | |
1200 | ||
1201 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1202 | { | |
1203 | struct mdp_superblock_1 *sb; | |
1204 | struct list_head *tmp; | |
1205 | mdk_rdev_t *rdev2; | |
1206 | int max_dev, i; | |
1207 | /* make rdev->sb match mddev and rdev data. */ | |
1208 | ||
1209 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1210 | ||
1211 | sb->feature_map = 0; | |
1212 | sb->pad0 = 0; | |
5fd6c1dc | 1213 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1214 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1215 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1216 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1217 | ||
1218 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1219 | sb->events = cpu_to_le64(mddev->events); | |
1220 | if (mddev->in_sync) | |
1221 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1222 | else | |
1223 | sb->resync_offset = cpu_to_le64(0); | |
1224 | ||
4dbcdc75 N |
1225 | sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors); |
1226 | ||
f0ca340c | 1227 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
29fc7e3e | 1228 | sb->size = cpu_to_le64(mddev->size<<1); |
f0ca340c | 1229 | |
a654b9d8 N |
1230 | if (mddev->bitmap && mddev->bitmap_file == NULL) { |
1231 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); | |
71c0805c | 1232 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1233 | } |
5fd6c1dc N |
1234 | |
1235 | if (rdev->raid_disk >= 0 && | |
1236 | !test_bit(In_sync, &rdev->flags) && | |
1237 | rdev->recovery_offset > 0) { | |
1238 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1239 | sb->recovery_offset = cpu_to_le64(rdev->recovery_offset); | |
1240 | } | |
1241 | ||
f6705578 N |
1242 | if (mddev->reshape_position != MaxSector) { |
1243 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1244 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1245 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1246 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1247 | sb->new_level = cpu_to_le32(mddev->new_level); | |
1248 | sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9); | |
1249 | } | |
a654b9d8 | 1250 | |
1da177e4 LT |
1251 | max_dev = 0; |
1252 | ITERATE_RDEV(mddev,rdev2,tmp) | |
1253 | if (rdev2->desc_nr+1 > max_dev) | |
1254 | max_dev = rdev2->desc_nr+1; | |
1255 | ||
1256 | sb->max_dev = cpu_to_le32(max_dev); | |
1257 | for (i=0; i<max_dev;i++) | |
1258 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1259 | ||
1260 | ITERATE_RDEV(mddev,rdev2,tmp) { | |
1261 | i = rdev2->desc_nr; | |
b2d444d7 | 1262 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1263 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1264 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1265 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
5fd6c1dc N |
1266 | else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0) |
1267 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); | |
1da177e4 LT |
1268 | else |
1269 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1270 | } | |
1271 | ||
1da177e4 LT |
1272 | sb->sb_csum = calc_sb_1_csum(sb); |
1273 | } | |
1274 | ||
1275 | ||
75c96f85 | 1276 | static struct super_type super_types[] = { |
1da177e4 LT |
1277 | [0] = { |
1278 | .name = "0.90.0", | |
1279 | .owner = THIS_MODULE, | |
1280 | .load_super = super_90_load, | |
1281 | .validate_super = super_90_validate, | |
1282 | .sync_super = super_90_sync, | |
1283 | }, | |
1284 | [1] = { | |
1285 | .name = "md-1", | |
1286 | .owner = THIS_MODULE, | |
1287 | .load_super = super_1_load, | |
1288 | .validate_super = super_1_validate, | |
1289 | .sync_super = super_1_sync, | |
1290 | }, | |
1291 | }; | |
1292 | ||
1293 | static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev) | |
1294 | { | |
1295 | struct list_head *tmp; | |
1296 | mdk_rdev_t *rdev; | |
1297 | ||
1298 | ITERATE_RDEV(mddev,rdev,tmp) | |
1299 | if (rdev->bdev->bd_contains == dev->bdev->bd_contains) | |
1300 | return rdev; | |
1301 | ||
1302 | return NULL; | |
1303 | } | |
1304 | ||
1305 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1306 | { | |
1307 | struct list_head *tmp; | |
1308 | mdk_rdev_t *rdev; | |
1309 | ||
1310 | ITERATE_RDEV(mddev1,rdev,tmp) | |
1311 | if (match_dev_unit(mddev2, rdev)) | |
1312 | return 1; | |
1313 | ||
1314 | return 0; | |
1315 | } | |
1316 | ||
1317 | static LIST_HEAD(pending_raid_disks); | |
1318 | ||
1319 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) | |
1320 | { | |
1321 | mdk_rdev_t *same_pdev; | |
1322 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
f637b9f9 | 1323 | struct kobject *ko; |
1edf80d3 | 1324 | char *s; |
1da177e4 LT |
1325 | |
1326 | if (rdev->mddev) { | |
1327 | MD_BUG(); | |
1328 | return -EINVAL; | |
1329 | } | |
2bf071bf N |
1330 | /* make sure rdev->size exceeds mddev->size */ |
1331 | if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) { | |
1332 | if (mddev->pers) | |
1333 | /* Cannot change size, so fail */ | |
1334 | return -ENOSPC; | |
1335 | else | |
1336 | mddev->size = rdev->size; | |
1337 | } | |
1da177e4 LT |
1338 | same_pdev = match_dev_unit(mddev, rdev); |
1339 | if (same_pdev) | |
1340 | printk(KERN_WARNING | |
1341 | "%s: WARNING: %s appears to be on the same physical" | |
1342 | " disk as %s. True\n protection against single-disk" | |
1343 | " failure might be compromised.\n", | |
1344 | mdname(mddev), bdevname(rdev->bdev,b), | |
1345 | bdevname(same_pdev->bdev,b2)); | |
1346 | ||
1347 | /* Verify rdev->desc_nr is unique. | |
1348 | * If it is -1, assign a free number, else | |
1349 | * check number is not in use | |
1350 | */ | |
1351 | if (rdev->desc_nr < 0) { | |
1352 | int choice = 0; | |
1353 | if (mddev->pers) choice = mddev->raid_disks; | |
1354 | while (find_rdev_nr(mddev, choice)) | |
1355 | choice++; | |
1356 | rdev->desc_nr = choice; | |
1357 | } else { | |
1358 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1359 | return -EBUSY; | |
1360 | } | |
19133a42 N |
1361 | bdevname(rdev->bdev,b); |
1362 | if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0) | |
1363 | return -ENOMEM; | |
1edf80d3 NB |
1364 | while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL) |
1365 | *s = '!'; | |
1da177e4 LT |
1366 | |
1367 | list_add(&rdev->same_set, &mddev->disks); | |
1368 | rdev->mddev = mddev; | |
19133a42 | 1369 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1370 | |
9c791977 | 1371 | rdev->kobj.parent = &mddev->kobj; |
86e6ffdd N |
1372 | kobject_add(&rdev->kobj); |
1373 | ||
f637b9f9 N |
1374 | if (rdev->bdev->bd_part) |
1375 | ko = &rdev->bdev->bd_part->kobj; | |
1376 | else | |
1377 | ko = &rdev->bdev->bd_disk->kobj; | |
1378 | sysfs_create_link(&rdev->kobj, ko, "block"); | |
5463c790 | 1379 | bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk); |
1da177e4 LT |
1380 | return 0; |
1381 | } | |
1382 | ||
1383 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) | |
1384 | { | |
1385 | char b[BDEVNAME_SIZE]; | |
1386 | if (!rdev->mddev) { | |
1387 | MD_BUG(); | |
1388 | return; | |
1389 | } | |
5463c790 | 1390 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
1da177e4 LT |
1391 | list_del_init(&rdev->same_set); |
1392 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); | |
1393 | rdev->mddev = NULL; | |
86e6ffdd N |
1394 | sysfs_remove_link(&rdev->kobj, "block"); |
1395 | kobject_del(&rdev->kobj); | |
1da177e4 LT |
1396 | } |
1397 | ||
1398 | /* | |
1399 | * prevent the device from being mounted, repartitioned or | |
1400 | * otherwise reused by a RAID array (or any other kernel | |
1401 | * subsystem), by bd_claiming the device. | |
1402 | */ | |
1403 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev) | |
1404 | { | |
1405 | int err = 0; | |
1406 | struct block_device *bdev; | |
1407 | char b[BDEVNAME_SIZE]; | |
1408 | ||
1409 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); | |
1410 | if (IS_ERR(bdev)) { | |
1411 | printk(KERN_ERR "md: could not open %s.\n", | |
1412 | __bdevname(dev, b)); | |
1413 | return PTR_ERR(bdev); | |
1414 | } | |
1415 | err = bd_claim(bdev, rdev); | |
1416 | if (err) { | |
1417 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1418 | bdevname(bdev, b)); | |
1419 | blkdev_put(bdev); | |
1420 | return err; | |
1421 | } | |
1422 | rdev->bdev = bdev; | |
1423 | return err; | |
1424 | } | |
1425 | ||
1426 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1427 | { | |
1428 | struct block_device *bdev = rdev->bdev; | |
1429 | rdev->bdev = NULL; | |
1430 | if (!bdev) | |
1431 | MD_BUG(); | |
1432 | bd_release(bdev); | |
1433 | blkdev_put(bdev); | |
1434 | } | |
1435 | ||
1436 | void md_autodetect_dev(dev_t dev); | |
1437 | ||
1438 | static void export_rdev(mdk_rdev_t * rdev) | |
1439 | { | |
1440 | char b[BDEVNAME_SIZE]; | |
1441 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1442 | bdevname(rdev->bdev,b)); | |
1443 | if (rdev->mddev) | |
1444 | MD_BUG(); | |
1445 | free_disk_sb(rdev); | |
1446 | list_del_init(&rdev->same_set); | |
1447 | #ifndef MODULE | |
1448 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1449 | #endif | |
1450 | unlock_rdev(rdev); | |
86e6ffdd | 1451 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1452 | } |
1453 | ||
1454 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1455 | { | |
1456 | unbind_rdev_from_array(rdev); | |
1457 | export_rdev(rdev); | |
1458 | } | |
1459 | ||
1460 | static void export_array(mddev_t *mddev) | |
1461 | { | |
1462 | struct list_head *tmp; | |
1463 | mdk_rdev_t *rdev; | |
1464 | ||
1465 | ITERATE_RDEV(mddev,rdev,tmp) { | |
1466 | if (!rdev->mddev) { | |
1467 | MD_BUG(); | |
1468 | continue; | |
1469 | } | |
1470 | kick_rdev_from_array(rdev); | |
1471 | } | |
1472 | if (!list_empty(&mddev->disks)) | |
1473 | MD_BUG(); | |
1474 | mddev->raid_disks = 0; | |
1475 | mddev->major_version = 0; | |
1476 | } | |
1477 | ||
1478 | static void print_desc(mdp_disk_t *desc) | |
1479 | { | |
1480 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1481 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1482 | } | |
1483 | ||
1484 | static void print_sb(mdp_super_t *sb) | |
1485 | { | |
1486 | int i; | |
1487 | ||
1488 | printk(KERN_INFO | |
1489 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1490 | sb->major_version, sb->minor_version, sb->patch_version, | |
1491 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1492 | sb->ctime); | |
1493 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1494 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1495 | sb->md_minor, sb->layout, sb->chunk_size); | |
1496 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1497 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1498 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1499 | sb->failed_disks, sb->spare_disks, | |
1500 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1501 | ||
1502 | printk(KERN_INFO); | |
1503 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1504 | mdp_disk_t *desc; | |
1505 | ||
1506 | desc = sb->disks + i; | |
1507 | if (desc->number || desc->major || desc->minor || | |
1508 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1509 | printk(" D %2d: ", i); | |
1510 | print_desc(desc); | |
1511 | } | |
1512 | } | |
1513 | printk(KERN_INFO "md: THIS: "); | |
1514 | print_desc(&sb->this_disk); | |
1515 | ||
1516 | } | |
1517 | ||
1518 | static void print_rdev(mdk_rdev_t *rdev) | |
1519 | { | |
1520 | char b[BDEVNAME_SIZE]; | |
1521 | printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n", | |
1522 | bdevname(rdev->bdev,b), (unsigned long long)rdev->size, | |
b2d444d7 N |
1523 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
1524 | rdev->desc_nr); | |
1da177e4 LT |
1525 | if (rdev->sb_loaded) { |
1526 | printk(KERN_INFO "md: rdev superblock:\n"); | |
1527 | print_sb((mdp_super_t*)page_address(rdev->sb_page)); | |
1528 | } else | |
1529 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
1530 | } | |
1531 | ||
5e56341d | 1532 | static void md_print_devices(void) |
1da177e4 LT |
1533 | { |
1534 | struct list_head *tmp, *tmp2; | |
1535 | mdk_rdev_t *rdev; | |
1536 | mddev_t *mddev; | |
1537 | char b[BDEVNAME_SIZE]; | |
1538 | ||
1539 | printk("\n"); | |
1540 | printk("md: **********************************\n"); | |
1541 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
1542 | printk("md: **********************************\n"); | |
1543 | ITERATE_MDDEV(mddev,tmp) { | |
1da177e4 | 1544 | |
32a7627c N |
1545 | if (mddev->bitmap) |
1546 | bitmap_print_sb(mddev->bitmap); | |
1547 | else | |
1548 | printk("%s: ", mdname(mddev)); | |
1da177e4 LT |
1549 | ITERATE_RDEV(mddev,rdev,tmp2) |
1550 | printk("<%s>", bdevname(rdev->bdev,b)); | |
1551 | printk("\n"); | |
1552 | ||
1553 | ITERATE_RDEV(mddev,rdev,tmp2) | |
1554 | print_rdev(rdev); | |
1555 | } | |
1556 | printk("md: **********************************\n"); | |
1557 | printk("\n"); | |
1558 | } | |
1559 | ||
1560 | ||
42543769 | 1561 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 1562 | { |
42543769 N |
1563 | /* Update each superblock (in-memory image), but |
1564 | * if we are allowed to, skip spares which already | |
1565 | * have the right event counter, or have one earlier | |
1566 | * (which would mean they aren't being marked as dirty | |
1567 | * with the rest of the array) | |
1568 | */ | |
1da177e4 LT |
1569 | mdk_rdev_t *rdev; |
1570 | struct list_head *tmp; | |
1571 | ||
1572 | ITERATE_RDEV(mddev,rdev,tmp) { | |
42543769 N |
1573 | if (rdev->sb_events == mddev->events || |
1574 | (nospares && | |
1575 | rdev->raid_disk < 0 && | |
1576 | (rdev->sb_events&1)==0 && | |
1577 | rdev->sb_events+1 == mddev->events)) { | |
1578 | /* Don't update this superblock */ | |
1579 | rdev->sb_loaded = 2; | |
1580 | } else { | |
1581 | super_types[mddev->major_version]. | |
1582 | sync_super(mddev, rdev); | |
1583 | rdev->sb_loaded = 1; | |
1584 | } | |
1da177e4 LT |
1585 | } |
1586 | } | |
1587 | ||
f6705578 | 1588 | void md_update_sb(mddev_t * mddev) |
1da177e4 | 1589 | { |
7bfa19f2 | 1590 | int err; |
1da177e4 LT |
1591 | struct list_head *tmp; |
1592 | mdk_rdev_t *rdev; | |
06d91a5f | 1593 | int sync_req; |
42543769 | 1594 | int nospares = 0; |
1da177e4 | 1595 | |
1da177e4 | 1596 | repeat: |
a9701a30 | 1597 | spin_lock_irq(&mddev->write_lock); |
06d91a5f | 1598 | sync_req = mddev->in_sync; |
1da177e4 | 1599 | mddev->utime = get_seconds(); |
42543769 N |
1600 | if (mddev->sb_dirty == 3) |
1601 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
1602 | * though if events isn't the right even/odd, we will have to do | |
1603 | * spares after all | |
1604 | */ | |
1605 | nospares = 1; | |
1606 | ||
1607 | /* If this is just a dirty<->clean transition, and the array is clean | |
1608 | * and 'events' is odd, we can roll back to the previous clean state */ | |
1609 | if (mddev->sb_dirty == 3 | |
1610 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1611 | && (mddev->events & 1)) | |
1612 | mddev->events--; | |
1613 | else { | |
1614 | /* otherwise we have to go forward and ... */ | |
1615 | mddev->events ++; | |
1616 | if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ | |
1617 | /* .. if the array isn't clean, insist on an odd 'events' */ | |
1618 | if ((mddev->events&1)==0) { | |
1619 | mddev->events++; | |
1620 | nospares = 0; | |
1621 | } | |
1622 | } else { | |
1623 | /* otherwise insist on an even 'events' (for clean states) */ | |
1624 | if ((mddev->events&1)) { | |
1625 | mddev->events++; | |
1626 | nospares = 0; | |
1627 | } | |
1628 | } | |
1629 | } | |
1da177e4 LT |
1630 | |
1631 | if (!mddev->events) { | |
1632 | /* | |
1633 | * oops, this 64-bit counter should never wrap. | |
1634 | * Either we are in around ~1 trillion A.C., assuming | |
1635 | * 1 reboot per second, or we have a bug: | |
1636 | */ | |
1637 | MD_BUG(); | |
1638 | mddev->events --; | |
1639 | } | |
7bfa19f2 | 1640 | mddev->sb_dirty = 2; |
42543769 | 1641 | sync_sbs(mddev, nospares); |
1da177e4 LT |
1642 | |
1643 | /* | |
1644 | * do not write anything to disk if using | |
1645 | * nonpersistent superblocks | |
1646 | */ | |
06d91a5f N |
1647 | if (!mddev->persistent) { |
1648 | mddev->sb_dirty = 0; | |
a9701a30 | 1649 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1650 | wake_up(&mddev->sb_wait); |
1da177e4 | 1651 | return; |
06d91a5f | 1652 | } |
a9701a30 | 1653 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
1654 | |
1655 | dprintk(KERN_INFO | |
1656 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
1657 | mdname(mddev),mddev->in_sync); | |
1658 | ||
32a7627c | 1659 | err = bitmap_update_sb(mddev->bitmap); |
1da177e4 LT |
1660 | ITERATE_RDEV(mddev,rdev,tmp) { |
1661 | char b[BDEVNAME_SIZE]; | |
1662 | dprintk(KERN_INFO "md: "); | |
42543769 N |
1663 | if (rdev->sb_loaded != 1) |
1664 | continue; /* no noise on spare devices */ | |
b2d444d7 | 1665 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
1666 | dprintk("(skipping faulty "); |
1667 | ||
1668 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 1669 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 1670 | md_super_write(mddev,rdev, |
0002b271 | 1671 | rdev->sb_offset<<1, rdev->sb_size, |
7bfa19f2 N |
1672 | rdev->sb_page); |
1673 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
1674 | bdevname(rdev->bdev,b), | |
1675 | (unsigned long long)rdev->sb_offset); | |
42543769 | 1676 | rdev->sb_events = mddev->events; |
7bfa19f2 | 1677 | |
1da177e4 LT |
1678 | } else |
1679 | dprintk(")\n"); | |
7bfa19f2 | 1680 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
1681 | /* only need to write one superblock... */ |
1682 | break; | |
1683 | } | |
a9701a30 | 1684 | md_super_wait(mddev); |
7bfa19f2 N |
1685 | /* if there was a failure, sb_dirty was set to 1, and we re-write super */ |
1686 | ||
a9701a30 | 1687 | spin_lock_irq(&mddev->write_lock); |
7bfa19f2 | 1688 | if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) { |
06d91a5f | 1689 | /* have to write it out again */ |
a9701a30 | 1690 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
1691 | goto repeat; |
1692 | } | |
1693 | mddev->sb_dirty = 0; | |
a9701a30 | 1694 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1695 | wake_up(&mddev->sb_wait); |
06d91a5f | 1696 | |
1da177e4 | 1697 | } |
f6705578 | 1698 | EXPORT_SYMBOL_GPL(md_update_sb); |
1da177e4 | 1699 | |
bce74dac N |
1700 | /* words written to sysfs files may, or my not, be \n terminated. |
1701 | * We want to accept with case. For this we use cmd_match. | |
1702 | */ | |
1703 | static int cmd_match(const char *cmd, const char *str) | |
1704 | { | |
1705 | /* See if cmd, written into a sysfs file, matches | |
1706 | * str. They must either be the same, or cmd can | |
1707 | * have a trailing newline | |
1708 | */ | |
1709 | while (*cmd && *str && *cmd == *str) { | |
1710 | cmd++; | |
1711 | str++; | |
1712 | } | |
1713 | if (*cmd == '\n') | |
1714 | cmd++; | |
1715 | if (*str || *cmd) | |
1716 | return 0; | |
1717 | return 1; | |
1718 | } | |
1719 | ||
86e6ffdd N |
1720 | struct rdev_sysfs_entry { |
1721 | struct attribute attr; | |
1722 | ssize_t (*show)(mdk_rdev_t *, char *); | |
1723 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
1724 | }; | |
1725 | ||
1726 | static ssize_t | |
96de1e66 | 1727 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
1728 | { |
1729 | char *sep = ""; | |
1730 | int len=0; | |
1731 | ||
b2d444d7 | 1732 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
1733 | len+= sprintf(page+len, "%sfaulty",sep); |
1734 | sep = ","; | |
1735 | } | |
b2d444d7 | 1736 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
1737 | len += sprintf(page+len, "%sin_sync",sep); |
1738 | sep = ","; | |
1739 | } | |
b2d444d7 N |
1740 | if (!test_bit(Faulty, &rdev->flags) && |
1741 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
1742 | len += sprintf(page+len, "%sspare", sep); |
1743 | sep = ","; | |
1744 | } | |
1745 | return len+sprintf(page+len, "\n"); | |
1746 | } | |
1747 | ||
45dc2de1 N |
1748 | static ssize_t |
1749 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1750 | { | |
1751 | /* can write | |
1752 | * faulty - simulates and error | |
1753 | * remove - disconnects the device | |
1754 | */ | |
1755 | int err = -EINVAL; | |
1756 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
1757 | md_error(rdev->mddev, rdev); | |
1758 | err = 0; | |
1759 | } else if (cmd_match(buf, "remove")) { | |
1760 | if (rdev->raid_disk >= 0) | |
1761 | err = -EBUSY; | |
1762 | else { | |
1763 | mddev_t *mddev = rdev->mddev; | |
1764 | kick_rdev_from_array(rdev); | |
1765 | md_update_sb(mddev); | |
1766 | md_new_event(mddev); | |
1767 | err = 0; | |
1768 | } | |
1769 | } | |
1770 | return err ? err : len; | |
1771 | } | |
96de1e66 | 1772 | static struct rdev_sysfs_entry |
45dc2de1 | 1773 | rdev_state = __ATTR(state, 0644, state_show, state_store); |
86e6ffdd N |
1774 | |
1775 | static ssize_t | |
96de1e66 | 1776 | super_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
1777 | { |
1778 | if (rdev->sb_loaded && rdev->sb_size) { | |
1779 | memcpy(page, page_address(rdev->sb_page), rdev->sb_size); | |
1780 | return rdev->sb_size; | |
1781 | } else | |
1782 | return 0; | |
1783 | } | |
96de1e66 N |
1784 | static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super); |
1785 | ||
4dbcdc75 N |
1786 | static ssize_t |
1787 | errors_show(mdk_rdev_t *rdev, char *page) | |
1788 | { | |
1789 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
1790 | } | |
1791 | ||
1792 | static ssize_t | |
1793 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1794 | { | |
1795 | char *e; | |
1796 | unsigned long n = simple_strtoul(buf, &e, 10); | |
1797 | if (*buf && (*e == 0 || *e == '\n')) { | |
1798 | atomic_set(&rdev->corrected_errors, n); | |
1799 | return len; | |
1800 | } | |
1801 | return -EINVAL; | |
1802 | } | |
1803 | static struct rdev_sysfs_entry rdev_errors = | |
1804 | __ATTR(errors, 0644, errors_show, errors_store); | |
1805 | ||
014236d2 N |
1806 | static ssize_t |
1807 | slot_show(mdk_rdev_t *rdev, char *page) | |
1808 | { | |
1809 | if (rdev->raid_disk < 0) | |
1810 | return sprintf(page, "none\n"); | |
1811 | else | |
1812 | return sprintf(page, "%d\n", rdev->raid_disk); | |
1813 | } | |
1814 | ||
1815 | static ssize_t | |
1816 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1817 | { | |
1818 | char *e; | |
1819 | int slot = simple_strtoul(buf, &e, 10); | |
1820 | if (strncmp(buf, "none", 4)==0) | |
1821 | slot = -1; | |
1822 | else if (e==buf || (*e && *e!= '\n')) | |
1823 | return -EINVAL; | |
1824 | if (rdev->mddev->pers) | |
1825 | /* Cannot set slot in active array (yet) */ | |
1826 | return -EBUSY; | |
1827 | if (slot >= rdev->mddev->raid_disks) | |
1828 | return -ENOSPC; | |
1829 | rdev->raid_disk = slot; | |
1830 | /* assume it is working */ | |
1831 | rdev->flags = 0; | |
1832 | set_bit(In_sync, &rdev->flags); | |
1833 | return len; | |
1834 | } | |
1835 | ||
1836 | ||
1837 | static struct rdev_sysfs_entry rdev_slot = | |
1838 | __ATTR(slot, 0644, slot_show, slot_store); | |
1839 | ||
93c8cad0 N |
1840 | static ssize_t |
1841 | offset_show(mdk_rdev_t *rdev, char *page) | |
1842 | { | |
6961ece4 | 1843 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
1844 | } |
1845 | ||
1846 | static ssize_t | |
1847 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1848 | { | |
1849 | char *e; | |
1850 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
1851 | if (e==buf || (*e && *e != '\n')) | |
1852 | return -EINVAL; | |
1853 | if (rdev->mddev->pers) | |
1854 | return -EBUSY; | |
1855 | rdev->data_offset = offset; | |
1856 | return len; | |
1857 | } | |
1858 | ||
1859 | static struct rdev_sysfs_entry rdev_offset = | |
1860 | __ATTR(offset, 0644, offset_show, offset_store); | |
1861 | ||
83303b61 N |
1862 | static ssize_t |
1863 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
1864 | { | |
1865 | return sprintf(page, "%llu\n", (unsigned long long)rdev->size); | |
1866 | } | |
1867 | ||
1868 | static ssize_t | |
1869 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1870 | { | |
1871 | char *e; | |
1872 | unsigned long long size = simple_strtoull(buf, &e, 10); | |
1873 | if (e==buf || (*e && *e != '\n')) | |
1874 | return -EINVAL; | |
1875 | if (rdev->mddev->pers) | |
1876 | return -EBUSY; | |
1877 | rdev->size = size; | |
1878 | if (size < rdev->mddev->size || rdev->mddev->size == 0) | |
1879 | rdev->mddev->size = size; | |
1880 | return len; | |
1881 | } | |
1882 | ||
1883 | static struct rdev_sysfs_entry rdev_size = | |
1884 | __ATTR(size, 0644, rdev_size_show, rdev_size_store); | |
1885 | ||
86e6ffdd N |
1886 | static struct attribute *rdev_default_attrs[] = { |
1887 | &rdev_state.attr, | |
1888 | &rdev_super.attr, | |
4dbcdc75 | 1889 | &rdev_errors.attr, |
014236d2 | 1890 | &rdev_slot.attr, |
93c8cad0 | 1891 | &rdev_offset.attr, |
83303b61 | 1892 | &rdev_size.attr, |
86e6ffdd N |
1893 | NULL, |
1894 | }; | |
1895 | static ssize_t | |
1896 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
1897 | { | |
1898 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
1899 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
1900 | ||
1901 | if (!entry->show) | |
1902 | return -EIO; | |
1903 | return entry->show(rdev, page); | |
1904 | } | |
1905 | ||
1906 | static ssize_t | |
1907 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
1908 | const char *page, size_t length) | |
1909 | { | |
1910 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
1911 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
1912 | ||
1913 | if (!entry->store) | |
1914 | return -EIO; | |
1915 | return entry->store(rdev, page, length); | |
1916 | } | |
1917 | ||
1918 | static void rdev_free(struct kobject *ko) | |
1919 | { | |
1920 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
1921 | kfree(rdev); | |
1922 | } | |
1923 | static struct sysfs_ops rdev_sysfs_ops = { | |
1924 | .show = rdev_attr_show, | |
1925 | .store = rdev_attr_store, | |
1926 | }; | |
1927 | static struct kobj_type rdev_ktype = { | |
1928 | .release = rdev_free, | |
1929 | .sysfs_ops = &rdev_sysfs_ops, | |
1930 | .default_attrs = rdev_default_attrs, | |
1931 | }; | |
1932 | ||
1da177e4 LT |
1933 | /* |
1934 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
1935 | * | |
1936 | * mark the device faulty if: | |
1937 | * | |
1938 | * - the device is nonexistent (zero size) | |
1939 | * - the device has no valid superblock | |
1940 | * | |
1941 | * a faulty rdev _never_ has rdev->sb set. | |
1942 | */ | |
1943 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
1944 | { | |
1945 | char b[BDEVNAME_SIZE]; | |
1946 | int err; | |
1947 | mdk_rdev_t *rdev; | |
1948 | sector_t size; | |
1949 | ||
9ffae0cf | 1950 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
1951 | if (!rdev) { |
1952 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
1953 | return ERR_PTR(-ENOMEM); | |
1954 | } | |
1da177e4 LT |
1955 | |
1956 | if ((err = alloc_disk_sb(rdev))) | |
1957 | goto abort_free; | |
1958 | ||
1959 | err = lock_rdev(rdev, newdev); | |
1960 | if (err) | |
1961 | goto abort_free; | |
1962 | ||
86e6ffdd N |
1963 | rdev->kobj.parent = NULL; |
1964 | rdev->kobj.ktype = &rdev_ktype; | |
1965 | kobject_init(&rdev->kobj); | |
1966 | ||
1da177e4 | 1967 | rdev->desc_nr = -1; |
b2d444d7 | 1968 | rdev->flags = 0; |
1da177e4 | 1969 | rdev->data_offset = 0; |
42543769 | 1970 | rdev->sb_events = 0; |
1da177e4 | 1971 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 1972 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 1973 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
1974 | |
1975 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
1976 | if (!size) { | |
1977 | printk(KERN_WARNING | |
1978 | "md: %s has zero or unknown size, marking faulty!\n", | |
1979 | bdevname(rdev->bdev,b)); | |
1980 | err = -EINVAL; | |
1981 | goto abort_free; | |
1982 | } | |
1983 | ||
1984 | if (super_format >= 0) { | |
1985 | err = super_types[super_format]. | |
1986 | load_super(rdev, NULL, super_minor); | |
1987 | if (err == -EINVAL) { | |
1988 | printk(KERN_WARNING | |
1989 | "md: %s has invalid sb, not importing!\n", | |
1990 | bdevname(rdev->bdev,b)); | |
1991 | goto abort_free; | |
1992 | } | |
1993 | if (err < 0) { | |
1994 | printk(KERN_WARNING | |
1995 | "md: could not read %s's sb, not importing!\n", | |
1996 | bdevname(rdev->bdev,b)); | |
1997 | goto abort_free; | |
1998 | } | |
1999 | } | |
2000 | INIT_LIST_HEAD(&rdev->same_set); | |
2001 | ||
2002 | return rdev; | |
2003 | ||
2004 | abort_free: | |
2005 | if (rdev->sb_page) { | |
2006 | if (rdev->bdev) | |
2007 | unlock_rdev(rdev); | |
2008 | free_disk_sb(rdev); | |
2009 | } | |
2010 | kfree(rdev); | |
2011 | return ERR_PTR(err); | |
2012 | } | |
2013 | ||
2014 | /* | |
2015 | * Check a full RAID array for plausibility | |
2016 | */ | |
2017 | ||
2018 | ||
a757e64c | 2019 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2020 | { |
2021 | int i; | |
2022 | struct list_head *tmp; | |
2023 | mdk_rdev_t *rdev, *freshest; | |
2024 | char b[BDEVNAME_SIZE]; | |
2025 | ||
2026 | freshest = NULL; | |
2027 | ITERATE_RDEV(mddev,rdev,tmp) | |
2028 | switch (super_types[mddev->major_version]. | |
2029 | load_super(rdev, freshest, mddev->minor_version)) { | |
2030 | case 1: | |
2031 | freshest = rdev; | |
2032 | break; | |
2033 | case 0: | |
2034 | break; | |
2035 | default: | |
2036 | printk( KERN_ERR \ | |
2037 | "md: fatal superblock inconsistency in %s" | |
2038 | " -- removing from array\n", | |
2039 | bdevname(rdev->bdev,b)); | |
2040 | kick_rdev_from_array(rdev); | |
2041 | } | |
2042 | ||
2043 | ||
2044 | super_types[mddev->major_version]. | |
2045 | validate_super(mddev, freshest); | |
2046 | ||
2047 | i = 0; | |
2048 | ITERATE_RDEV(mddev,rdev,tmp) { | |
2049 | if (rdev != freshest) | |
2050 | if (super_types[mddev->major_version]. | |
2051 | validate_super(mddev, rdev)) { | |
2052 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2053 | " from array!\n", | |
2054 | bdevname(rdev->bdev,b)); | |
2055 | kick_rdev_from_array(rdev); | |
2056 | continue; | |
2057 | } | |
2058 | if (mddev->level == LEVEL_MULTIPATH) { | |
2059 | rdev->desc_nr = i++; | |
2060 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2061 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
2062 | } |
2063 | } | |
2064 | ||
2065 | ||
2066 | ||
2067 | if (mddev->recovery_cp != MaxSector && | |
2068 | mddev->level >= 1) | |
2069 | printk(KERN_ERR "md: %s: raid array is not clean" | |
2070 | " -- starting background reconstruction\n", | |
2071 | mdname(mddev)); | |
2072 | ||
1da177e4 LT |
2073 | } |
2074 | ||
16f17b39 N |
2075 | static ssize_t |
2076 | safe_delay_show(mddev_t *mddev, char *page) | |
2077 | { | |
2078 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2079 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2080 | } | |
2081 | static ssize_t | |
2082 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2083 | { | |
2084 | int scale=1; | |
2085 | int dot=0; | |
2086 | int i; | |
2087 | unsigned long msec; | |
2088 | char buf[30]; | |
2089 | char *e; | |
2090 | /* remove a period, and count digits after it */ | |
2091 | if (len >= sizeof(buf)) | |
2092 | return -EINVAL; | |
2093 | strlcpy(buf, cbuf, len); | |
2094 | buf[len] = 0; | |
2095 | for (i=0; i<len; i++) { | |
2096 | if (dot) { | |
2097 | if (isdigit(buf[i])) { | |
2098 | buf[i-1] = buf[i]; | |
2099 | scale *= 10; | |
2100 | } | |
2101 | buf[i] = 0; | |
2102 | } else if (buf[i] == '.') { | |
2103 | dot=1; | |
2104 | buf[i] = 0; | |
2105 | } | |
2106 | } | |
2107 | msec = simple_strtoul(buf, &e, 10); | |
2108 | if (e == buf || (*e && *e != '\n')) | |
2109 | return -EINVAL; | |
2110 | msec = (msec * 1000) / scale; | |
2111 | if (msec == 0) | |
2112 | mddev->safemode_delay = 0; | |
2113 | else { | |
2114 | mddev->safemode_delay = (msec*HZ)/1000; | |
2115 | if (mddev->safemode_delay == 0) | |
2116 | mddev->safemode_delay = 1; | |
2117 | } | |
2118 | return len; | |
2119 | } | |
2120 | static struct md_sysfs_entry md_safe_delay = | |
2121 | __ATTR(safe_mode_delay, 0644,safe_delay_show, safe_delay_store); | |
2122 | ||
eae1701f | 2123 | static ssize_t |
96de1e66 | 2124 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2125 | { |
2604b703 | 2126 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2127 | if (p) |
eae1701f | 2128 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2129 | else if (mddev->clevel[0]) |
2130 | return sprintf(page, "%s\n", mddev->clevel); | |
2131 | else if (mddev->level != LEVEL_NONE) | |
2132 | return sprintf(page, "%d\n", mddev->level); | |
2133 | else | |
2134 | return 0; | |
eae1701f N |
2135 | } |
2136 | ||
d9d166c2 N |
2137 | static ssize_t |
2138 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2139 | { | |
2140 | int rv = len; | |
2141 | if (mddev->pers) | |
2142 | return -EBUSY; | |
2143 | if (len == 0) | |
2144 | return 0; | |
2145 | if (len >= sizeof(mddev->clevel)) | |
2146 | return -ENOSPC; | |
2147 | strncpy(mddev->clevel, buf, len); | |
2148 | if (mddev->clevel[len-1] == '\n') | |
2149 | len--; | |
2150 | mddev->clevel[len] = 0; | |
2151 | mddev->level = LEVEL_NONE; | |
2152 | return rv; | |
2153 | } | |
2154 | ||
2155 | static struct md_sysfs_entry md_level = | |
2156 | __ATTR(level, 0644, level_show, level_store); | |
eae1701f | 2157 | |
d4dbd025 N |
2158 | |
2159 | static ssize_t | |
2160 | layout_show(mddev_t *mddev, char *page) | |
2161 | { | |
2162 | /* just a number, not meaningful for all levels */ | |
2163 | return sprintf(page, "%d\n", mddev->layout); | |
2164 | } | |
2165 | ||
2166 | static ssize_t | |
2167 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
2168 | { | |
2169 | char *e; | |
2170 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2171 | if (mddev->pers) | |
2172 | return -EBUSY; | |
2173 | ||
2174 | if (!*buf || (*e && *e != '\n')) | |
2175 | return -EINVAL; | |
2176 | ||
2177 | mddev->layout = n; | |
2178 | return len; | |
2179 | } | |
2180 | static struct md_sysfs_entry md_layout = | |
2181 | __ATTR(layout, 0655, layout_show, layout_store); | |
2182 | ||
2183 | ||
eae1701f | 2184 | static ssize_t |
96de1e66 | 2185 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 2186 | { |
bb636547 N |
2187 | if (mddev->raid_disks == 0) |
2188 | return 0; | |
eae1701f N |
2189 | return sprintf(page, "%d\n", mddev->raid_disks); |
2190 | } | |
2191 | ||
da943b99 N |
2192 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
2193 | ||
2194 | static ssize_t | |
2195 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
2196 | { | |
2197 | /* can only set raid_disks if array is not yet active */ | |
2198 | char *e; | |
2199 | int rv = 0; | |
2200 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2201 | ||
2202 | if (!*buf || (*e && *e != '\n')) | |
2203 | return -EINVAL; | |
2204 | ||
2205 | if (mddev->pers) | |
2206 | rv = update_raid_disks(mddev, n); | |
2207 | else | |
2208 | mddev->raid_disks = n; | |
2209 | return rv ? rv : len; | |
2210 | } | |
2211 | static struct md_sysfs_entry md_raid_disks = | |
2212 | __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store); | |
eae1701f | 2213 | |
3b34380a N |
2214 | static ssize_t |
2215 | chunk_size_show(mddev_t *mddev, char *page) | |
2216 | { | |
2217 | return sprintf(page, "%d\n", mddev->chunk_size); | |
2218 | } | |
2219 | ||
2220 | static ssize_t | |
2221 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
2222 | { | |
2223 | /* can only set chunk_size if array is not yet active */ | |
2224 | char *e; | |
2225 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2226 | ||
2227 | if (mddev->pers) | |
2228 | return -EBUSY; | |
2229 | if (!*buf || (*e && *e != '\n')) | |
2230 | return -EINVAL; | |
2231 | ||
2232 | mddev->chunk_size = n; | |
2233 | return len; | |
2234 | } | |
2235 | static struct md_sysfs_entry md_chunk_size = | |
2236 | __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store); | |
2237 | ||
9e653b63 N |
2238 | /* |
2239 | * The array state can be: | |
2240 | * | |
2241 | * clear | |
2242 | * No devices, no size, no level | |
2243 | * Equivalent to STOP_ARRAY ioctl | |
2244 | * inactive | |
2245 | * May have some settings, but array is not active | |
2246 | * all IO results in error | |
2247 | * When written, doesn't tear down array, but just stops it | |
2248 | * suspended (not supported yet) | |
2249 | * All IO requests will block. The array can be reconfigured. | |
2250 | * Writing this, if accepted, will block until array is quiessent | |
2251 | * readonly | |
2252 | * no resync can happen. no superblocks get written. | |
2253 | * write requests fail | |
2254 | * read-auto | |
2255 | * like readonly, but behaves like 'clean' on a write request. | |
2256 | * | |
2257 | * clean - no pending writes, but otherwise active. | |
2258 | * When written to inactive array, starts without resync | |
2259 | * If a write request arrives then | |
2260 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
2261 | * if not known, block and switch to write-pending | |
2262 | * If written to an active array that has pending writes, then fails. | |
2263 | * active | |
2264 | * fully active: IO and resync can be happening. | |
2265 | * When written to inactive array, starts with resync | |
2266 | * | |
2267 | * write-pending | |
2268 | * clean, but writes are blocked waiting for 'active' to be written. | |
2269 | * | |
2270 | * active-idle | |
2271 | * like active, but no writes have been seen for a while (100msec). | |
2272 | * | |
2273 | */ | |
2274 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
2275 | write_pending, active_idle, bad_word}; | |
2276 | char *array_states[] = { | |
2277 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", | |
2278 | "write-pending", "active-idle", NULL }; | |
2279 | ||
2280 | static int match_word(const char *word, char **list) | |
2281 | { | |
2282 | int n; | |
2283 | for (n=0; list[n]; n++) | |
2284 | if (cmd_match(word, list[n])) | |
2285 | break; | |
2286 | return n; | |
2287 | } | |
2288 | ||
2289 | static ssize_t | |
2290 | array_state_show(mddev_t *mddev, char *page) | |
2291 | { | |
2292 | enum array_state st = inactive; | |
2293 | ||
2294 | if (mddev->pers) | |
2295 | switch(mddev->ro) { | |
2296 | case 1: | |
2297 | st = readonly; | |
2298 | break; | |
2299 | case 2: | |
2300 | st = read_auto; | |
2301 | break; | |
2302 | case 0: | |
2303 | if (mddev->in_sync) | |
2304 | st = clean; | |
2305 | else if (mddev->safemode) | |
2306 | st = active_idle; | |
2307 | else | |
2308 | st = active; | |
2309 | } | |
2310 | else { | |
2311 | if (list_empty(&mddev->disks) && | |
2312 | mddev->raid_disks == 0 && | |
2313 | mddev->size == 0) | |
2314 | st = clear; | |
2315 | else | |
2316 | st = inactive; | |
2317 | } | |
2318 | return sprintf(page, "%s\n", array_states[st]); | |
2319 | } | |
2320 | ||
2321 | static int do_md_stop(mddev_t * mddev, int ro); | |
2322 | static int do_md_run(mddev_t * mddev); | |
2323 | static int restart_array(mddev_t *mddev); | |
2324 | ||
2325 | static ssize_t | |
2326 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
2327 | { | |
2328 | int err = -EINVAL; | |
2329 | enum array_state st = match_word(buf, array_states); | |
2330 | switch(st) { | |
2331 | case bad_word: | |
2332 | break; | |
2333 | case clear: | |
2334 | /* stopping an active array */ | |
2335 | if (mddev->pers) { | |
2336 | if (atomic_read(&mddev->active) > 1) | |
2337 | return -EBUSY; | |
2338 | err = do_md_stop(mddev, 0); | |
2339 | } | |
2340 | break; | |
2341 | case inactive: | |
2342 | /* stopping an active array */ | |
2343 | if (mddev->pers) { | |
2344 | if (atomic_read(&mddev->active) > 1) | |
2345 | return -EBUSY; | |
2346 | err = do_md_stop(mddev, 2); | |
2347 | } | |
2348 | break; | |
2349 | case suspended: | |
2350 | break; /* not supported yet */ | |
2351 | case readonly: | |
2352 | if (mddev->pers) | |
2353 | err = do_md_stop(mddev, 1); | |
2354 | else { | |
2355 | mddev->ro = 1; | |
2356 | err = do_md_run(mddev); | |
2357 | } | |
2358 | break; | |
2359 | case read_auto: | |
2360 | /* stopping an active array */ | |
2361 | if (mddev->pers) { | |
2362 | err = do_md_stop(mddev, 1); | |
2363 | if (err == 0) | |
2364 | mddev->ro = 2; /* FIXME mark devices writable */ | |
2365 | } else { | |
2366 | mddev->ro = 2; | |
2367 | err = do_md_run(mddev); | |
2368 | } | |
2369 | break; | |
2370 | case clean: | |
2371 | if (mddev->pers) { | |
2372 | restart_array(mddev); | |
2373 | spin_lock_irq(&mddev->write_lock); | |
2374 | if (atomic_read(&mddev->writes_pending) == 0) { | |
2375 | mddev->in_sync = 1; | |
2376 | mddev->sb_dirty = 1; | |
2377 | } | |
2378 | spin_unlock_irq(&mddev->write_lock); | |
2379 | } else { | |
2380 | mddev->ro = 0; | |
2381 | mddev->recovery_cp = MaxSector; | |
2382 | err = do_md_run(mddev); | |
2383 | } | |
2384 | break; | |
2385 | case active: | |
2386 | if (mddev->pers) { | |
2387 | restart_array(mddev); | |
2388 | mddev->sb_dirty = 0; | |
2389 | wake_up(&mddev->sb_wait); | |
2390 | err = 0; | |
2391 | } else { | |
2392 | mddev->ro = 0; | |
2393 | err = do_md_run(mddev); | |
2394 | } | |
2395 | break; | |
2396 | case write_pending: | |
2397 | case active_idle: | |
2398 | /* these cannot be set */ | |
2399 | break; | |
2400 | } | |
2401 | if (err) | |
2402 | return err; | |
2403 | else | |
2404 | return len; | |
2405 | } | |
2406 | static struct md_sysfs_entry md_array_state = __ATTR(array_state, 0644, array_state_show, array_state_store); | |
2407 | ||
6d7ff738 N |
2408 | static ssize_t |
2409 | null_show(mddev_t *mddev, char *page) | |
2410 | { | |
2411 | return -EINVAL; | |
2412 | } | |
2413 | ||
2414 | static ssize_t | |
2415 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
2416 | { | |
2417 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
2418 | /* The new device is added to the array. | |
2419 | * If the array has a persistent superblock, we read the | |
2420 | * superblock to initialise info and check validity. | |
2421 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
2422 | * which mainly checks size. | |
2423 | */ | |
2424 | char *e; | |
2425 | int major = simple_strtoul(buf, &e, 10); | |
2426 | int minor; | |
2427 | dev_t dev; | |
2428 | mdk_rdev_t *rdev; | |
2429 | int err; | |
2430 | ||
2431 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
2432 | return -EINVAL; | |
2433 | minor = simple_strtoul(e+1, &e, 10); | |
2434 | if (*e && *e != '\n') | |
2435 | return -EINVAL; | |
2436 | dev = MKDEV(major, minor); | |
2437 | if (major != MAJOR(dev) || | |
2438 | minor != MINOR(dev)) | |
2439 | return -EOVERFLOW; | |
2440 | ||
2441 | ||
2442 | if (mddev->persistent) { | |
2443 | rdev = md_import_device(dev, mddev->major_version, | |
2444 | mddev->minor_version); | |
2445 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
2446 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
2447 | mdk_rdev_t, same_set); | |
2448 | err = super_types[mddev->major_version] | |
2449 | .load_super(rdev, rdev0, mddev->minor_version); | |
2450 | if (err < 0) | |
2451 | goto out; | |
2452 | } | |
2453 | } else | |
2454 | rdev = md_import_device(dev, -1, -1); | |
2455 | ||
2456 | if (IS_ERR(rdev)) | |
2457 | return PTR_ERR(rdev); | |
2458 | err = bind_rdev_to_array(rdev, mddev); | |
2459 | out: | |
2460 | if (err) | |
2461 | export_rdev(rdev); | |
2462 | return err ? err : len; | |
2463 | } | |
2464 | ||
2465 | static struct md_sysfs_entry md_new_device = | |
2466 | __ATTR(new_dev, 0200, null_show, new_dev_store); | |
3b34380a | 2467 | |
a35b0d69 N |
2468 | static ssize_t |
2469 | size_show(mddev_t *mddev, char *page) | |
2470 | { | |
2471 | return sprintf(page, "%llu\n", (unsigned long long)mddev->size); | |
2472 | } | |
2473 | ||
2474 | static int update_size(mddev_t *mddev, unsigned long size); | |
2475 | ||
2476 | static ssize_t | |
2477 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
2478 | { | |
2479 | /* If array is inactive, we can reduce the component size, but | |
2480 | * not increase it (except from 0). | |
2481 | * If array is active, we can try an on-line resize | |
2482 | */ | |
2483 | char *e; | |
2484 | int err = 0; | |
2485 | unsigned long long size = simple_strtoull(buf, &e, 10); | |
2486 | if (!*buf || *buf == '\n' || | |
2487 | (*e && *e != '\n')) | |
2488 | return -EINVAL; | |
2489 | ||
2490 | if (mddev->pers) { | |
2491 | err = update_size(mddev, size); | |
2492 | md_update_sb(mddev); | |
2493 | } else { | |
2494 | if (mddev->size == 0 || | |
2495 | mddev->size > size) | |
2496 | mddev->size = size; | |
2497 | else | |
2498 | err = -ENOSPC; | |
2499 | } | |
2500 | return err ? err : len; | |
2501 | } | |
2502 | ||
2503 | static struct md_sysfs_entry md_size = | |
2504 | __ATTR(component_size, 0644, size_show, size_store); | |
2505 | ||
8bb93aac N |
2506 | |
2507 | /* Metdata version. | |
2508 | * This is either 'none' for arrays with externally managed metadata, | |
2509 | * or N.M for internally known formats | |
2510 | */ | |
2511 | static ssize_t | |
2512 | metadata_show(mddev_t *mddev, char *page) | |
2513 | { | |
2514 | if (mddev->persistent) | |
2515 | return sprintf(page, "%d.%d\n", | |
2516 | mddev->major_version, mddev->minor_version); | |
2517 | else | |
2518 | return sprintf(page, "none\n"); | |
2519 | } | |
2520 | ||
2521 | static ssize_t | |
2522 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
2523 | { | |
2524 | int major, minor; | |
2525 | char *e; | |
2526 | if (!list_empty(&mddev->disks)) | |
2527 | return -EBUSY; | |
2528 | ||
2529 | if (cmd_match(buf, "none")) { | |
2530 | mddev->persistent = 0; | |
2531 | mddev->major_version = 0; | |
2532 | mddev->minor_version = 90; | |
2533 | return len; | |
2534 | } | |
2535 | major = simple_strtoul(buf, &e, 10); | |
2536 | if (e==buf || *e != '.') | |
2537 | return -EINVAL; | |
2538 | buf = e+1; | |
2539 | minor = simple_strtoul(buf, &e, 10); | |
2540 | if (e==buf || *e != '\n') | |
2541 | return -EINVAL; | |
2542 | if (major >= sizeof(super_types)/sizeof(super_types[0]) || | |
2543 | super_types[major].name == NULL) | |
2544 | return -ENOENT; | |
2545 | mddev->major_version = major; | |
2546 | mddev->minor_version = minor; | |
2547 | mddev->persistent = 1; | |
2548 | return len; | |
2549 | } | |
2550 | ||
2551 | static struct md_sysfs_entry md_metadata = | |
2552 | __ATTR(metadata_version, 0644, metadata_show, metadata_store); | |
2553 | ||
24dd469d | 2554 | static ssize_t |
7eec314d | 2555 | action_show(mddev_t *mddev, char *page) |
24dd469d | 2556 | { |
7eec314d | 2557 | char *type = "idle"; |
31399d9e N |
2558 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
2559 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) { | |
ccfcc3c1 N |
2560 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
2561 | type = "reshape"; | |
2562 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
2563 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
2564 | type = "resync"; | |
2565 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
2566 | type = "check"; | |
2567 | else | |
2568 | type = "repair"; | |
2569 | } else | |
2570 | type = "recover"; | |
2571 | } | |
2572 | return sprintf(page, "%s\n", type); | |
2573 | } | |
2574 | ||
2575 | static ssize_t | |
7eec314d | 2576 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 2577 | { |
7eec314d N |
2578 | if (!mddev->pers || !mddev->pers->sync_request) |
2579 | return -EINVAL; | |
2580 | ||
bce74dac | 2581 | if (cmd_match(page, "idle")) { |
7eec314d N |
2582 | if (mddev->sync_thread) { |
2583 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2584 | md_unregister_thread(mddev->sync_thread); | |
2585 | mddev->sync_thread = NULL; | |
2586 | mddev->recovery = 0; | |
2587 | } | |
03c902e1 N |
2588 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
2589 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 2590 | return -EBUSY; |
03c902e1 | 2591 | else if (cmd_match(page, "resync") || cmd_match(page, "recover")) |
7eec314d | 2592 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
16484bf5 N |
2593 | else if (cmd_match(page, "reshape")) { |
2594 | int err; | |
2595 | if (mddev->pers->start_reshape == NULL) | |
2596 | return -EINVAL; | |
2597 | err = mddev->pers->start_reshape(mddev); | |
2598 | if (err) | |
2599 | return err; | |
2600 | } else { | |
bce74dac | 2601 | if (cmd_match(page, "check")) |
7eec314d | 2602 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 2603 | else if (!cmd_match(page, "repair")) |
7eec314d N |
2604 | return -EINVAL; |
2605 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
2606 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 2607 | } |
03c902e1 | 2608 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d N |
2609 | md_wakeup_thread(mddev->thread); |
2610 | return len; | |
2611 | } | |
2612 | ||
9d88883e | 2613 | static ssize_t |
96de1e66 | 2614 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
2615 | { |
2616 | return sprintf(page, "%llu\n", | |
2617 | (unsigned long long) mddev->resync_mismatches); | |
2618 | } | |
2619 | ||
96de1e66 | 2620 | static struct md_sysfs_entry |
7eec314d | 2621 | md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
24dd469d | 2622 | |
96de1e66 N |
2623 | |
2624 | static struct md_sysfs_entry | |
2625 | md_mismatches = __ATTR_RO(mismatch_cnt); | |
9d88883e | 2626 | |
88202a0c N |
2627 | static ssize_t |
2628 | sync_min_show(mddev_t *mddev, char *page) | |
2629 | { | |
2630 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
2631 | mddev->sync_speed_min ? "local": "system"); | |
2632 | } | |
2633 | ||
2634 | static ssize_t | |
2635 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
2636 | { | |
2637 | int min; | |
2638 | char *e; | |
2639 | if (strncmp(buf, "system", 6)==0) { | |
2640 | mddev->sync_speed_min = 0; | |
2641 | return len; | |
2642 | } | |
2643 | min = simple_strtoul(buf, &e, 10); | |
2644 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
2645 | return -EINVAL; | |
2646 | mddev->sync_speed_min = min; | |
2647 | return len; | |
2648 | } | |
2649 | ||
2650 | static struct md_sysfs_entry md_sync_min = | |
2651 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
2652 | ||
2653 | static ssize_t | |
2654 | sync_max_show(mddev_t *mddev, char *page) | |
2655 | { | |
2656 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
2657 | mddev->sync_speed_max ? "local": "system"); | |
2658 | } | |
2659 | ||
2660 | static ssize_t | |
2661 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
2662 | { | |
2663 | int max; | |
2664 | char *e; | |
2665 | if (strncmp(buf, "system", 6)==0) { | |
2666 | mddev->sync_speed_max = 0; | |
2667 | return len; | |
2668 | } | |
2669 | max = simple_strtoul(buf, &e, 10); | |
2670 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
2671 | return -EINVAL; | |
2672 | mddev->sync_speed_max = max; | |
2673 | return len; | |
2674 | } | |
2675 | ||
2676 | static struct md_sysfs_entry md_sync_max = | |
2677 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
2678 | ||
2679 | ||
2680 | static ssize_t | |
2681 | sync_speed_show(mddev_t *mddev, char *page) | |
2682 | { | |
2683 | unsigned long resync, dt, db; | |
2684 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active)); | |
2685 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
2686 | if (!dt) dt++; | |
2687 | db = resync - (mddev->resync_mark_cnt); | |
2688 | return sprintf(page, "%ld\n", db/dt/2); /* K/sec */ | |
2689 | } | |
2690 | ||
2691 | static struct md_sysfs_entry | |
2692 | md_sync_speed = __ATTR_RO(sync_speed); | |
2693 | ||
2694 | static ssize_t | |
2695 | sync_completed_show(mddev_t *mddev, char *page) | |
2696 | { | |
2697 | unsigned long max_blocks, resync; | |
2698 | ||
2699 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2700 | max_blocks = mddev->resync_max_sectors; | |
2701 | else | |
2702 | max_blocks = mddev->size << 1; | |
2703 | ||
2704 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active)); | |
2705 | return sprintf(page, "%lu / %lu\n", resync, max_blocks); | |
2706 | } | |
2707 | ||
2708 | static struct md_sysfs_entry | |
2709 | md_sync_completed = __ATTR_RO(sync_completed); | |
2710 | ||
e464eafd N |
2711 | static ssize_t |
2712 | suspend_lo_show(mddev_t *mddev, char *page) | |
2713 | { | |
2714 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
2715 | } | |
2716 | ||
2717 | static ssize_t | |
2718 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
2719 | { | |
2720 | char *e; | |
2721 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
2722 | ||
2723 | if (mddev->pers->quiesce == NULL) | |
2724 | return -EINVAL; | |
2725 | if (buf == e || (*e && *e != '\n')) | |
2726 | return -EINVAL; | |
2727 | if (new >= mddev->suspend_hi || | |
2728 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
2729 | mddev->suspend_lo = new; | |
2730 | mddev->pers->quiesce(mddev, 2); | |
2731 | return len; | |
2732 | } else | |
2733 | return -EINVAL; | |
2734 | } | |
2735 | static struct md_sysfs_entry md_suspend_lo = | |
2736 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
2737 | ||
2738 | ||
2739 | static ssize_t | |
2740 | suspend_hi_show(mddev_t *mddev, char *page) | |
2741 | { | |
2742 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
2743 | } | |
2744 | ||
2745 | static ssize_t | |
2746 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
2747 | { | |
2748 | char *e; | |
2749 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
2750 | ||
2751 | if (mddev->pers->quiesce == NULL) | |
2752 | return -EINVAL; | |
2753 | if (buf == e || (*e && *e != '\n')) | |
2754 | return -EINVAL; | |
2755 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
2756 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
2757 | mddev->suspend_hi = new; | |
2758 | mddev->pers->quiesce(mddev, 1); | |
2759 | mddev->pers->quiesce(mddev, 0); | |
2760 | return len; | |
2761 | } else | |
2762 | return -EINVAL; | |
2763 | } | |
2764 | static struct md_sysfs_entry md_suspend_hi = | |
2765 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
2766 | ||
2767 | ||
eae1701f N |
2768 | static struct attribute *md_default_attrs[] = { |
2769 | &md_level.attr, | |
d4dbd025 | 2770 | &md_layout.attr, |
eae1701f | 2771 | &md_raid_disks.attr, |
3b34380a | 2772 | &md_chunk_size.attr, |
a35b0d69 | 2773 | &md_size.attr, |
8bb93aac | 2774 | &md_metadata.attr, |
6d7ff738 | 2775 | &md_new_device.attr, |
16f17b39 | 2776 | &md_safe_delay.attr, |
9e653b63 | 2777 | &md_array_state.attr, |
411036fa N |
2778 | NULL, |
2779 | }; | |
2780 | ||
2781 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 2782 | &md_scan_mode.attr, |
9d88883e | 2783 | &md_mismatches.attr, |
88202a0c N |
2784 | &md_sync_min.attr, |
2785 | &md_sync_max.attr, | |
2786 | &md_sync_speed.attr, | |
2787 | &md_sync_completed.attr, | |
e464eafd N |
2788 | &md_suspend_lo.attr, |
2789 | &md_suspend_hi.attr, | |
eae1701f N |
2790 | NULL, |
2791 | }; | |
411036fa N |
2792 | static struct attribute_group md_redundancy_group = { |
2793 | .name = NULL, | |
2794 | .attrs = md_redundancy_attrs, | |
2795 | }; | |
2796 | ||
eae1701f N |
2797 | |
2798 | static ssize_t | |
2799 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2800 | { | |
2801 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
2802 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 2803 | ssize_t rv; |
eae1701f N |
2804 | |
2805 | if (!entry->show) | |
2806 | return -EIO; | |
5dc5cf7d IM |
2807 | rv = mddev_lock(mddev); |
2808 | if (!rv) { | |
2809 | rv = entry->show(mddev, page); | |
2810 | mddev_unlock(mddev); | |
2811 | } | |
96de1e66 | 2812 | return rv; |
eae1701f N |
2813 | } |
2814 | ||
2815 | static ssize_t | |
2816 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
2817 | const char *page, size_t length) | |
2818 | { | |
2819 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
2820 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 2821 | ssize_t rv; |
eae1701f N |
2822 | |
2823 | if (!entry->store) | |
2824 | return -EIO; | |
5dc5cf7d IM |
2825 | rv = mddev_lock(mddev); |
2826 | if (!rv) { | |
2827 | rv = entry->store(mddev, page, length); | |
2828 | mddev_unlock(mddev); | |
2829 | } | |
96de1e66 | 2830 | return rv; |
eae1701f N |
2831 | } |
2832 | ||
2833 | static void md_free(struct kobject *ko) | |
2834 | { | |
2835 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
2836 | kfree(mddev); | |
2837 | } | |
2838 | ||
2839 | static struct sysfs_ops md_sysfs_ops = { | |
2840 | .show = md_attr_show, | |
2841 | .store = md_attr_store, | |
2842 | }; | |
2843 | static struct kobj_type md_ktype = { | |
2844 | .release = md_free, | |
2845 | .sysfs_ops = &md_sysfs_ops, | |
2846 | .default_attrs = md_default_attrs, | |
2847 | }; | |
2848 | ||
1da177e4 LT |
2849 | int mdp_major = 0; |
2850 | ||
2851 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
2852 | { | |
48c9c27b | 2853 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
2854 | mddev_t *mddev = mddev_find(dev); |
2855 | struct gendisk *disk; | |
2856 | int partitioned = (MAJOR(dev) != MD_MAJOR); | |
2857 | int shift = partitioned ? MdpMinorShift : 0; | |
2858 | int unit = MINOR(dev) >> shift; | |
2859 | ||
2860 | if (!mddev) | |
2861 | return NULL; | |
2862 | ||
48c9c27b | 2863 | mutex_lock(&disks_mutex); |
1da177e4 | 2864 | if (mddev->gendisk) { |
48c9c27b | 2865 | mutex_unlock(&disks_mutex); |
1da177e4 LT |
2866 | mddev_put(mddev); |
2867 | return NULL; | |
2868 | } | |
2869 | disk = alloc_disk(1 << shift); | |
2870 | if (!disk) { | |
48c9c27b | 2871 | mutex_unlock(&disks_mutex); |
1da177e4 LT |
2872 | mddev_put(mddev); |
2873 | return NULL; | |
2874 | } | |
2875 | disk->major = MAJOR(dev); | |
2876 | disk->first_minor = unit << shift; | |
2877 | if (partitioned) { | |
2878 | sprintf(disk->disk_name, "md_d%d", unit); | |
2879 | sprintf(disk->devfs_name, "md/d%d", unit); | |
2880 | } else { | |
2881 | sprintf(disk->disk_name, "md%d", unit); | |
2882 | sprintf(disk->devfs_name, "md/%d", unit); | |
2883 | } | |
2884 | disk->fops = &md_fops; | |
2885 | disk->private_data = mddev; | |
2886 | disk->queue = mddev->queue; | |
2887 | add_disk(disk); | |
2888 | mddev->gendisk = disk; | |
48c9c27b | 2889 | mutex_unlock(&disks_mutex); |
9c791977 | 2890 | mddev->kobj.parent = &disk->kobj; |
eae1701f N |
2891 | mddev->kobj.k_name = NULL; |
2892 | snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md"); | |
2893 | mddev->kobj.ktype = &md_ktype; | |
2894 | kobject_register(&mddev->kobj); | |
1da177e4 LT |
2895 | return NULL; |
2896 | } | |
2897 | ||
1da177e4 LT |
2898 | static void md_safemode_timeout(unsigned long data) |
2899 | { | |
2900 | mddev_t *mddev = (mddev_t *) data; | |
2901 | ||
2902 | mddev->safemode = 1; | |
2903 | md_wakeup_thread(mddev->thread); | |
2904 | } | |
2905 | ||
6ff8d8ec | 2906 | static int start_dirty_degraded; |
1da177e4 LT |
2907 | |
2908 | static int do_md_run(mddev_t * mddev) | |
2909 | { | |
2604b703 | 2910 | int err; |
1da177e4 LT |
2911 | int chunk_size; |
2912 | struct list_head *tmp; | |
2913 | mdk_rdev_t *rdev; | |
2914 | struct gendisk *disk; | |
2604b703 | 2915 | struct mdk_personality *pers; |
1da177e4 LT |
2916 | char b[BDEVNAME_SIZE]; |
2917 | ||
a757e64c N |
2918 | if (list_empty(&mddev->disks)) |
2919 | /* cannot run an array with no devices.. */ | |
1da177e4 | 2920 | return -EINVAL; |
1da177e4 LT |
2921 | |
2922 | if (mddev->pers) | |
2923 | return -EBUSY; | |
2924 | ||
2925 | /* | |
2926 | * Analyze all RAID superblock(s) | |
2927 | */ | |
a757e64c N |
2928 | if (!mddev->raid_disks) |
2929 | analyze_sbs(mddev); | |
1da177e4 LT |
2930 | |
2931 | chunk_size = mddev->chunk_size; | |
2604b703 N |
2932 | |
2933 | if (chunk_size) { | |
1da177e4 LT |
2934 | if (chunk_size > MAX_CHUNK_SIZE) { |
2935 | printk(KERN_ERR "too big chunk_size: %d > %d\n", | |
2936 | chunk_size, MAX_CHUNK_SIZE); | |
2937 | return -EINVAL; | |
2938 | } | |
2939 | /* | |
2940 | * chunk-size has to be a power of 2 and multiples of PAGE_SIZE | |
2941 | */ | |
2942 | if ( (1 << ffz(~chunk_size)) != chunk_size) { | |
a757e64c | 2943 | printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size); |
1da177e4 LT |
2944 | return -EINVAL; |
2945 | } | |
2946 | if (chunk_size < PAGE_SIZE) { | |
2947 | printk(KERN_ERR "too small chunk_size: %d < %ld\n", | |
2948 | chunk_size, PAGE_SIZE); | |
2949 | return -EINVAL; | |
2950 | } | |
2951 | ||
2952 | /* devices must have minimum size of one chunk */ | |
2953 | ITERATE_RDEV(mddev,rdev,tmp) { | |
b2d444d7 | 2954 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2955 | continue; |
2956 | if (rdev->size < chunk_size / 1024) { | |
2957 | printk(KERN_WARNING | |
2958 | "md: Dev %s smaller than chunk_size:" | |
2959 | " %lluk < %dk\n", | |
2960 | bdevname(rdev->bdev,b), | |
2961 | (unsigned long long)rdev->size, | |
2962 | chunk_size / 1024); | |
2963 | return -EINVAL; | |
2964 | } | |
2965 | } | |
2966 | } | |
2967 | ||
1da177e4 | 2968 | #ifdef CONFIG_KMOD |
d9d166c2 N |
2969 | if (mddev->level != LEVEL_NONE) |
2970 | request_module("md-level-%d", mddev->level); | |
2971 | else if (mddev->clevel[0]) | |
2972 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
2973 | #endif |
2974 | ||
2975 | /* | |
2976 | * Drop all container device buffers, from now on | |
2977 | * the only valid external interface is through the md | |
2978 | * device. | |
2979 | * Also find largest hardsector size | |
2980 | */ | |
2981 | ITERATE_RDEV(mddev,rdev,tmp) { | |
b2d444d7 | 2982 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2983 | continue; |
2984 | sync_blockdev(rdev->bdev); | |
2985 | invalidate_bdev(rdev->bdev, 0); | |
2986 | } | |
2987 | ||
2988 | md_probe(mddev->unit, NULL, NULL); | |
2989 | disk = mddev->gendisk; | |
2990 | if (!disk) | |
2991 | return -ENOMEM; | |
2992 | ||
2993 | spin_lock(&pers_lock); | |
d9d166c2 | 2994 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 2995 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 2996 | spin_unlock(&pers_lock); |
d9d166c2 N |
2997 | if (mddev->level != LEVEL_NONE) |
2998 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
2999 | mddev->level); | |
3000 | else | |
3001 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
3002 | mddev->clevel); | |
1da177e4 LT |
3003 | return -EINVAL; |
3004 | } | |
2604b703 | 3005 | mddev->pers = pers; |
1da177e4 | 3006 | spin_unlock(&pers_lock); |
d9d166c2 N |
3007 | mddev->level = pers->level; |
3008 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
1da177e4 | 3009 | |
f6705578 | 3010 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 3011 | pers->start_reshape == NULL) { |
f6705578 N |
3012 | /* This personality cannot handle reshaping... */ |
3013 | mddev->pers = NULL; | |
3014 | module_put(pers->owner); | |
3015 | return -EINVAL; | |
3016 | } | |
3017 | ||
657390d2 | 3018 | mddev->recovery = 0; |
1da177e4 | 3019 | mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */ |
a9701a30 | 3020 | mddev->barriers_work = 1; |
6ff8d8ec | 3021 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 3022 | |
f91de92e N |
3023 | if (start_readonly) |
3024 | mddev->ro = 2; /* read-only, but switch on first write */ | |
3025 | ||
b15c2e57 N |
3026 | err = mddev->pers->run(mddev); |
3027 | if (!err && mddev->pers->sync_request) { | |
3028 | err = bitmap_create(mddev); | |
3029 | if (err) { | |
3030 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
3031 | mdname(mddev), err); | |
3032 | mddev->pers->stop(mddev); | |
3033 | } | |
3034 | } | |
1da177e4 LT |
3035 | if (err) { |
3036 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
3037 | module_put(mddev->pers->owner); | |
3038 | mddev->pers = NULL; | |
32a7627c N |
3039 | bitmap_destroy(mddev); |
3040 | return err; | |
1da177e4 | 3041 | } |
411036fa N |
3042 | if (mddev->pers->sync_request) |
3043 | sysfs_create_group(&mddev->kobj, &md_redundancy_group); | |
fd9d49ca N |
3044 | else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
3045 | mddev->ro = 0; | |
3046 | ||
1da177e4 LT |
3047 | atomic_set(&mddev->writes_pending,0); |
3048 | mddev->safemode = 0; | |
3049 | mddev->safemode_timer.function = md_safemode_timeout; | |
3050 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 3051 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 3052 | mddev->in_sync = 1; |
86e6ffdd N |
3053 | |
3054 | ITERATE_RDEV(mddev,rdev,tmp) | |
3055 | if (rdev->raid_disk >= 0) { | |
3056 | char nm[20]; | |
3057 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3058 | sysfs_create_link(&mddev->kobj, &rdev->kobj, nm); | |
3059 | } | |
1da177e4 LT |
3060 | |
3061 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
005eca5e | 3062 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
3063 | |
3064 | if (mddev->sb_dirty) | |
3065 | md_update_sb(mddev); | |
3066 | ||
3067 | set_capacity(disk, mddev->array_size<<1); | |
3068 | ||
3069 | /* If we call blk_queue_make_request here, it will | |
3070 | * re-initialise max_sectors etc which may have been | |
3071 | * refined inside -> run. So just set the bits we need to set. | |
3072 | * Most initialisation happended when we called | |
3073 | * blk_queue_make_request(..., md_fail_request) | |
3074 | * earlier. | |
3075 | */ | |
3076 | mddev->queue->queuedata = mddev; | |
3077 | mddev->queue->make_request_fn = mddev->pers->make_request; | |
3078 | ||
5fd6c1dc N |
3079 | /* If there is a partially-recovered drive we need to |
3080 | * start recovery here. If we leave it to md_check_recovery, | |
3081 | * it will remove the drives and not do the right thing | |
3082 | */ | |
3083 | if (mddev->degraded) { | |
3084 | struct list_head *rtmp; | |
3085 | int spares = 0; | |
3086 | ITERATE_RDEV(mddev,rdev,rtmp) | |
3087 | if (rdev->raid_disk >= 0 && | |
3088 | !test_bit(In_sync, &rdev->flags) && | |
3089 | !test_bit(Faulty, &rdev->flags)) | |
3090 | /* complete an interrupted recovery */ | |
3091 | spares++; | |
3092 | if (spares && mddev->pers->sync_request) { | |
3093 | mddev->recovery = 0; | |
3094 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3095 | mddev->sync_thread = md_register_thread(md_do_sync, | |
3096 | mddev, | |
3097 | "%s_resync"); | |
3098 | if (!mddev->sync_thread) { | |
3099 | printk(KERN_ERR "%s: could not start resync" | |
3100 | " thread...\n", | |
3101 | mdname(mddev)); | |
3102 | /* leave the spares where they are, it shouldn't hurt */ | |
3103 | mddev->recovery = 0; | |
3104 | } else | |
3105 | md_wakeup_thread(mddev->sync_thread); | |
3106 | } | |
3107 | } | |
3108 | ||
1da177e4 | 3109 | mddev->changed = 1; |
d7603b7e | 3110 | md_new_event(mddev); |
1da177e4 LT |
3111 | return 0; |
3112 | } | |
3113 | ||
3114 | static int restart_array(mddev_t *mddev) | |
3115 | { | |
3116 | struct gendisk *disk = mddev->gendisk; | |
3117 | int err; | |
3118 | ||
3119 | /* | |
3120 | * Complain if it has no devices | |
3121 | */ | |
3122 | err = -ENXIO; | |
3123 | if (list_empty(&mddev->disks)) | |
3124 | goto out; | |
3125 | ||
3126 | if (mddev->pers) { | |
3127 | err = -EBUSY; | |
3128 | if (!mddev->ro) | |
3129 | goto out; | |
3130 | ||
3131 | mddev->safemode = 0; | |
3132 | mddev->ro = 0; | |
3133 | set_disk_ro(disk, 0); | |
3134 | ||
3135 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
3136 | mdname(mddev)); | |
3137 | /* | |
3138 | * Kick recovery or resync if necessary | |
3139 | */ | |
3140 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3141 | md_wakeup_thread(mddev->thread); | |
5fd6c1dc | 3142 | md_wakeup_thread(mddev->sync_thread); |
1da177e4 | 3143 | err = 0; |
9e653b63 | 3144 | } else |
1da177e4 | 3145 | err = -EINVAL; |
1da177e4 LT |
3146 | |
3147 | out: | |
3148 | return err; | |
3149 | } | |
3150 | ||
acc55e22 N |
3151 | /* similar to deny_write_access, but accounts for our holding a reference |
3152 | * to the file ourselves */ | |
3153 | static int deny_bitmap_write_access(struct file * file) | |
3154 | { | |
3155 | struct inode *inode = file->f_mapping->host; | |
3156 | ||
3157 | spin_lock(&inode->i_lock); | |
3158 | if (atomic_read(&inode->i_writecount) > 1) { | |
3159 | spin_unlock(&inode->i_lock); | |
3160 | return -ETXTBSY; | |
3161 | } | |
3162 | atomic_set(&inode->i_writecount, -1); | |
3163 | spin_unlock(&inode->i_lock); | |
3164 | ||
3165 | return 0; | |
3166 | } | |
3167 | ||
3168 | static void restore_bitmap_write_access(struct file *file) | |
3169 | { | |
3170 | struct inode *inode = file->f_mapping->host; | |
3171 | ||
3172 | spin_lock(&inode->i_lock); | |
3173 | atomic_set(&inode->i_writecount, 1); | |
3174 | spin_unlock(&inode->i_lock); | |
3175 | } | |
3176 | ||
9e653b63 N |
3177 | /* mode: |
3178 | * 0 - completely stop and dis-assemble array | |
3179 | * 1 - switch to readonly | |
3180 | * 2 - stop but do not disassemble array | |
3181 | */ | |
3182 | static int do_md_stop(mddev_t * mddev, int mode) | |
1da177e4 LT |
3183 | { |
3184 | int err = 0; | |
3185 | struct gendisk *disk = mddev->gendisk; | |
3186 | ||
3187 | if (mddev->pers) { | |
3188 | if (atomic_read(&mddev->active)>2) { | |
3189 | printk("md: %s still in use.\n",mdname(mddev)); | |
3190 | return -EBUSY; | |
3191 | } | |
3192 | ||
3193 | if (mddev->sync_thread) { | |
5fd6c1dc | 3194 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 LT |
3195 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
3196 | md_unregister_thread(mddev->sync_thread); | |
3197 | mddev->sync_thread = NULL; | |
3198 | } | |
3199 | ||
3200 | del_timer_sync(&mddev->safemode_timer); | |
3201 | ||
3202 | invalidate_partition(disk, 0); | |
3203 | ||
9e653b63 N |
3204 | switch(mode) { |
3205 | case 1: /* readonly */ | |
1da177e4 | 3206 | err = -ENXIO; |
f91de92e | 3207 | if (mddev->ro==1) |
1da177e4 LT |
3208 | goto out; |
3209 | mddev->ro = 1; | |
9e653b63 N |
3210 | break; |
3211 | case 0: /* disassemble */ | |
3212 | case 2: /* stop */ | |
6b8b3e8a | 3213 | bitmap_flush(mddev); |
a9701a30 | 3214 | md_super_wait(mddev); |
1da177e4 LT |
3215 | if (mddev->ro) |
3216 | set_disk_ro(disk, 0); | |
3217 | blk_queue_make_request(mddev->queue, md_fail_request); | |
3218 | mddev->pers->stop(mddev); | |
411036fa N |
3219 | if (mddev->pers->sync_request) |
3220 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
3221 | ||
1da177e4 LT |
3222 | module_put(mddev->pers->owner); |
3223 | mddev->pers = NULL; | |
3224 | if (mddev->ro) | |
3225 | mddev->ro = 0; | |
3226 | } | |
5fd6c1dc | 3227 | if (!mddev->in_sync || mddev->sb_dirty) { |
1da177e4 LT |
3228 | /* mark array as shutdown cleanly */ |
3229 | mddev->in_sync = 1; | |
3230 | md_update_sb(mddev); | |
3231 | } | |
9e653b63 | 3232 | if (mode == 1) |
1da177e4 | 3233 | set_disk_ro(disk, 1); |
5fd6c1dc | 3234 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 | 3235 | } |
32a7627c | 3236 | |
1da177e4 LT |
3237 | /* |
3238 | * Free resources if final stop | |
3239 | */ | |
9e653b63 | 3240 | if (mode == 0) { |
86e6ffdd N |
3241 | mdk_rdev_t *rdev; |
3242 | struct list_head *tmp; | |
1da177e4 LT |
3243 | struct gendisk *disk; |
3244 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); | |
3245 | ||
978f946b N |
3246 | bitmap_destroy(mddev); |
3247 | if (mddev->bitmap_file) { | |
acc55e22 | 3248 | restore_bitmap_write_access(mddev->bitmap_file); |
978f946b N |
3249 | fput(mddev->bitmap_file); |
3250 | mddev->bitmap_file = NULL; | |
3251 | } | |
3252 | mddev->bitmap_offset = 0; | |
3253 | ||
86e6ffdd N |
3254 | ITERATE_RDEV(mddev,rdev,tmp) |
3255 | if (rdev->raid_disk >= 0) { | |
3256 | char nm[20]; | |
3257 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3258 | sysfs_remove_link(&mddev->kobj, nm); | |
3259 | } | |
3260 | ||
1da177e4 LT |
3261 | export_array(mddev); |
3262 | ||
3263 | mddev->array_size = 0; | |
9e653b63 N |
3264 | mddev->size = 0; |
3265 | mddev->raid_disks = 0; | |
3266 | ||
1da177e4 LT |
3267 | disk = mddev->gendisk; |
3268 | if (disk) | |
3269 | set_capacity(disk, 0); | |
3270 | mddev->changed = 1; | |
a8a55c38 | 3271 | } else if (mddev->pers) |
1da177e4 LT |
3272 | printk(KERN_INFO "md: %s switched to read-only mode.\n", |
3273 | mdname(mddev)); | |
3274 | err = 0; | |
d7603b7e | 3275 | md_new_event(mddev); |
1da177e4 LT |
3276 | out: |
3277 | return err; | |
3278 | } | |
3279 | ||
3280 | static void autorun_array(mddev_t *mddev) | |
3281 | { | |
3282 | mdk_rdev_t *rdev; | |
3283 | struct list_head *tmp; | |
3284 | int err; | |
3285 | ||
a757e64c | 3286 | if (list_empty(&mddev->disks)) |
1da177e4 | 3287 | return; |
1da177e4 LT |
3288 | |
3289 | printk(KERN_INFO "md: running: "); | |
3290 | ||
3291 | ITERATE_RDEV(mddev,rdev,tmp) { | |
3292 | char b[BDEVNAME_SIZE]; | |
3293 | printk("<%s>", bdevname(rdev->bdev,b)); | |
3294 | } | |
3295 | printk("\n"); | |
3296 | ||
3297 | err = do_md_run (mddev); | |
3298 | if (err) { | |
3299 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
3300 | do_md_stop (mddev, 0); | |
3301 | } | |
3302 | } | |
3303 | ||
3304 | /* | |
3305 | * lets try to run arrays based on all disks that have arrived | |
3306 | * until now. (those are in pending_raid_disks) | |
3307 | * | |
3308 | * the method: pick the first pending disk, collect all disks with | |
3309 | * the same UUID, remove all from the pending list and put them into | |
3310 | * the 'same_array' list. Then order this list based on superblock | |
3311 | * update time (freshest comes first), kick out 'old' disks and | |
3312 | * compare superblocks. If everything's fine then run it. | |
3313 | * | |
3314 | * If "unit" is allocated, then bump its reference count | |
3315 | */ | |
3316 | static void autorun_devices(int part) | |
3317 | { | |
1da177e4 LT |
3318 | struct list_head *tmp; |
3319 | mdk_rdev_t *rdev0, *rdev; | |
3320 | mddev_t *mddev; | |
3321 | char b[BDEVNAME_SIZE]; | |
3322 | ||
3323 | printk(KERN_INFO "md: autorun ...\n"); | |
3324 | while (!list_empty(&pending_raid_disks)) { | |
3325 | dev_t dev; | |
ad01c9e3 | 3326 | LIST_HEAD(candidates); |
1da177e4 LT |
3327 | rdev0 = list_entry(pending_raid_disks.next, |
3328 | mdk_rdev_t, same_set); | |
3329 | ||
3330 | printk(KERN_INFO "md: considering %s ...\n", | |
3331 | bdevname(rdev0->bdev,b)); | |
3332 | INIT_LIST_HEAD(&candidates); | |
3333 | ITERATE_RDEV_PENDING(rdev,tmp) | |
3334 | if (super_90_load(rdev, rdev0, 0) >= 0) { | |
3335 | printk(KERN_INFO "md: adding %s ...\n", | |
3336 | bdevname(rdev->bdev,b)); | |
3337 | list_move(&rdev->same_set, &candidates); | |
3338 | } | |
3339 | /* | |
3340 | * now we have a set of devices, with all of them having | |
3341 | * mostly sane superblocks. It's time to allocate the | |
3342 | * mddev. | |
3343 | */ | |
3344 | if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) { | |
3345 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", | |
3346 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
3347 | break; | |
3348 | } | |
3349 | if (part) | |
3350 | dev = MKDEV(mdp_major, | |
3351 | rdev0->preferred_minor << MdpMinorShift); | |
3352 | else | |
3353 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
3354 | ||
3355 | md_probe(dev, NULL, NULL); | |
3356 | mddev = mddev_find(dev); | |
3357 | if (!mddev) { | |
3358 | printk(KERN_ERR | |
3359 | "md: cannot allocate memory for md drive.\n"); | |
3360 | break; | |
3361 | } | |
3362 | if (mddev_lock(mddev)) | |
3363 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
3364 | mdname(mddev)); | |
3365 | else if (mddev->raid_disks || mddev->major_version | |
3366 | || !list_empty(&mddev->disks)) { | |
3367 | printk(KERN_WARNING | |
3368 | "md: %s already running, cannot run %s\n", | |
3369 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
3370 | mddev_unlock(mddev); | |
3371 | } else { | |
3372 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
3373 | ITERATE_RDEV_GENERIC(candidates,rdev,tmp) { | |
3374 | list_del_init(&rdev->same_set); | |
3375 | if (bind_rdev_to_array(rdev, mddev)) | |
3376 | export_rdev(rdev); | |
3377 | } | |
3378 | autorun_array(mddev); | |
3379 | mddev_unlock(mddev); | |
3380 | } | |
3381 | /* on success, candidates will be empty, on error | |
3382 | * it won't... | |
3383 | */ | |
3384 | ITERATE_RDEV_GENERIC(candidates,rdev,tmp) | |
3385 | export_rdev(rdev); | |
3386 | mddev_put(mddev); | |
3387 | } | |
3388 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
3389 | } | |
3390 | ||
3391 | /* | |
3392 | * import RAID devices based on one partition | |
3393 | * if possible, the array gets run as well. | |
3394 | */ | |
3395 | ||
3396 | static int autostart_array(dev_t startdev) | |
3397 | { | |
3398 | char b[BDEVNAME_SIZE]; | |
3399 | int err = -EINVAL, i; | |
3400 | mdp_super_t *sb = NULL; | |
3401 | mdk_rdev_t *start_rdev = NULL, *rdev; | |
3402 | ||
3403 | start_rdev = md_import_device(startdev, 0, 0); | |
3404 | if (IS_ERR(start_rdev)) | |
3405 | return err; | |
3406 | ||
3407 | ||
3408 | /* NOTE: this can only work for 0.90.0 superblocks */ | |
3409 | sb = (mdp_super_t*)page_address(start_rdev->sb_page); | |
3410 | if (sb->major_version != 0 || | |
3411 | sb->minor_version != 90 ) { | |
3412 | printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n"); | |
3413 | export_rdev(start_rdev); | |
3414 | return err; | |
3415 | } | |
3416 | ||
b2d444d7 | 3417 | if (test_bit(Faulty, &start_rdev->flags)) { |
1da177e4 LT |
3418 | printk(KERN_WARNING |
3419 | "md: can not autostart based on faulty %s!\n", | |
3420 | bdevname(start_rdev->bdev,b)); | |
3421 | export_rdev(start_rdev); | |
3422 | return err; | |
3423 | } | |
3424 | list_add(&start_rdev->same_set, &pending_raid_disks); | |
3425 | ||
3426 | for (i = 0; i < MD_SB_DISKS; i++) { | |
3427 | mdp_disk_t *desc = sb->disks + i; | |
3428 | dev_t dev = MKDEV(desc->major, desc->minor); | |
3429 | ||
3430 | if (!dev) | |
3431 | continue; | |
3432 | if (dev == startdev) | |
3433 | continue; | |
3434 | if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor) | |
3435 | continue; | |
3436 | rdev = md_import_device(dev, 0, 0); | |
3437 | if (IS_ERR(rdev)) | |
3438 | continue; | |
3439 | ||
3440 | list_add(&rdev->same_set, &pending_raid_disks); | |
3441 | } | |
3442 | ||
3443 | /* | |
3444 | * possibly return codes | |
3445 | */ | |
3446 | autorun_devices(0); | |
3447 | return 0; | |
3448 | ||
3449 | } | |
3450 | ||
3451 | ||
3452 | static int get_version(void __user * arg) | |
3453 | { | |
3454 | mdu_version_t ver; | |
3455 | ||
3456 | ver.major = MD_MAJOR_VERSION; | |
3457 | ver.minor = MD_MINOR_VERSION; | |
3458 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
3459 | ||
3460 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
3461 | return -EFAULT; | |
3462 | ||
3463 | return 0; | |
3464 | } | |
3465 | ||
3466 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
3467 | { | |
3468 | mdu_array_info_t info; | |
3469 | int nr,working,active,failed,spare; | |
3470 | mdk_rdev_t *rdev; | |
3471 | struct list_head *tmp; | |
3472 | ||
3473 | nr=working=active=failed=spare=0; | |
3474 | ITERATE_RDEV(mddev,rdev,tmp) { | |
3475 | nr++; | |
b2d444d7 | 3476 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
3477 | failed++; |
3478 | else { | |
3479 | working++; | |
b2d444d7 | 3480 | if (test_bit(In_sync, &rdev->flags)) |
1da177e4 LT |
3481 | active++; |
3482 | else | |
3483 | spare++; | |
3484 | } | |
3485 | } | |
3486 | ||
3487 | info.major_version = mddev->major_version; | |
3488 | info.minor_version = mddev->minor_version; | |
3489 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
3490 | info.ctime = mddev->ctime; | |
3491 | info.level = mddev->level; | |
3492 | info.size = mddev->size; | |
284ae7ca N |
3493 | if (info.size != mddev->size) /* overflow */ |
3494 | info.size = -1; | |
1da177e4 LT |
3495 | info.nr_disks = nr; |
3496 | info.raid_disks = mddev->raid_disks; | |
3497 | info.md_minor = mddev->md_minor; | |
3498 | info.not_persistent= !mddev->persistent; | |
3499 | ||
3500 | info.utime = mddev->utime; | |
3501 | info.state = 0; | |
3502 | if (mddev->in_sync) | |
3503 | info.state = (1<<MD_SB_CLEAN); | |
36fa3063 N |
3504 | if (mddev->bitmap && mddev->bitmap_offset) |
3505 | info.state = (1<<MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
3506 | info.active_disks = active; |
3507 | info.working_disks = working; | |
3508 | info.failed_disks = failed; | |
3509 | info.spare_disks = spare; | |
3510 | ||
3511 | info.layout = mddev->layout; | |
3512 | info.chunk_size = mddev->chunk_size; | |
3513 | ||
3514 | if (copy_to_user(arg, &info, sizeof(info))) | |
3515 | return -EFAULT; | |
3516 | ||
3517 | return 0; | |
3518 | } | |
3519 | ||
87162a28 | 3520 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
3521 | { |
3522 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
3523 | char *ptr, *buf = NULL; | |
3524 | int err = -ENOMEM; | |
3525 | ||
3526 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
3527 | if (!file) | |
3528 | goto out; | |
3529 | ||
3530 | /* bitmap disabled, zero the first byte and copy out */ | |
3531 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
3532 | file->pathname[0] = '\0'; | |
3533 | goto copy_out; | |
3534 | } | |
3535 | ||
3536 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
3537 | if (!buf) | |
3538 | goto out; | |
3539 | ||
3540 | ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname)); | |
3541 | if (!ptr) | |
3542 | goto out; | |
3543 | ||
3544 | strcpy(file->pathname, ptr); | |
3545 | ||
3546 | copy_out: | |
3547 | err = 0; | |
3548 | if (copy_to_user(arg, file, sizeof(*file))) | |
3549 | err = -EFAULT; | |
3550 | out: | |
3551 | kfree(buf); | |
3552 | kfree(file); | |
3553 | return err; | |
3554 | } | |
3555 | ||
1da177e4 LT |
3556 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
3557 | { | |
3558 | mdu_disk_info_t info; | |
3559 | unsigned int nr; | |
3560 | mdk_rdev_t *rdev; | |
3561 | ||
3562 | if (copy_from_user(&info, arg, sizeof(info))) | |
3563 | return -EFAULT; | |
3564 | ||
3565 | nr = info.number; | |
3566 | ||
3567 | rdev = find_rdev_nr(mddev, nr); | |
3568 | if (rdev) { | |
3569 | info.major = MAJOR(rdev->bdev->bd_dev); | |
3570 | info.minor = MINOR(rdev->bdev->bd_dev); | |
3571 | info.raid_disk = rdev->raid_disk; | |
3572 | info.state = 0; | |
b2d444d7 | 3573 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 3574 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 3575 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
3576 | info.state |= (1<<MD_DISK_ACTIVE); |
3577 | info.state |= (1<<MD_DISK_SYNC); | |
3578 | } | |
8ddf9efe N |
3579 | if (test_bit(WriteMostly, &rdev->flags)) |
3580 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
3581 | } else { |
3582 | info.major = info.minor = 0; | |
3583 | info.raid_disk = -1; | |
3584 | info.state = (1<<MD_DISK_REMOVED); | |
3585 | } | |
3586 | ||
3587 | if (copy_to_user(arg, &info, sizeof(info))) | |
3588 | return -EFAULT; | |
3589 | ||
3590 | return 0; | |
3591 | } | |
3592 | ||
3593 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
3594 | { | |
3595 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3596 | mdk_rdev_t *rdev; | |
3597 | dev_t dev = MKDEV(info->major,info->minor); | |
3598 | ||
3599 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
3600 | return -EOVERFLOW; | |
3601 | ||
3602 | if (!mddev->raid_disks) { | |
3603 | int err; | |
3604 | /* expecting a device which has a superblock */ | |
3605 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
3606 | if (IS_ERR(rdev)) { | |
3607 | printk(KERN_WARNING | |
3608 | "md: md_import_device returned %ld\n", | |
3609 | PTR_ERR(rdev)); | |
3610 | return PTR_ERR(rdev); | |
3611 | } | |
3612 | if (!list_empty(&mddev->disks)) { | |
3613 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3614 | mdk_rdev_t, same_set); | |
3615 | int err = super_types[mddev->major_version] | |
3616 | .load_super(rdev, rdev0, mddev->minor_version); | |
3617 | if (err < 0) { | |
3618 | printk(KERN_WARNING | |
3619 | "md: %s has different UUID to %s\n", | |
3620 | bdevname(rdev->bdev,b), | |
3621 | bdevname(rdev0->bdev,b2)); | |
3622 | export_rdev(rdev); | |
3623 | return -EINVAL; | |
3624 | } | |
3625 | } | |
3626 | err = bind_rdev_to_array(rdev, mddev); | |
3627 | if (err) | |
3628 | export_rdev(rdev); | |
3629 | return err; | |
3630 | } | |
3631 | ||
3632 | /* | |
3633 | * add_new_disk can be used once the array is assembled | |
3634 | * to add "hot spares". They must already have a superblock | |
3635 | * written | |
3636 | */ | |
3637 | if (mddev->pers) { | |
3638 | int err; | |
3639 | if (!mddev->pers->hot_add_disk) { | |
3640 | printk(KERN_WARNING | |
3641 | "%s: personality does not support diskops!\n", | |
3642 | mdname(mddev)); | |
3643 | return -EINVAL; | |
3644 | } | |
7b1e35f6 N |
3645 | if (mddev->persistent) |
3646 | rdev = md_import_device(dev, mddev->major_version, | |
3647 | mddev->minor_version); | |
3648 | else | |
3649 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
3650 | if (IS_ERR(rdev)) { |
3651 | printk(KERN_WARNING | |
3652 | "md: md_import_device returned %ld\n", | |
3653 | PTR_ERR(rdev)); | |
3654 | return PTR_ERR(rdev); | |
3655 | } | |
41158c7e N |
3656 | /* set save_raid_disk if appropriate */ |
3657 | if (!mddev->persistent) { | |
3658 | if (info->state & (1<<MD_DISK_SYNC) && | |
3659 | info->raid_disk < mddev->raid_disks) | |
3660 | rdev->raid_disk = info->raid_disk; | |
3661 | else | |
3662 | rdev->raid_disk = -1; | |
3663 | } else | |
3664 | super_types[mddev->major_version]. | |
3665 | validate_super(mddev, rdev); | |
3666 | rdev->saved_raid_disk = rdev->raid_disk; | |
3667 | ||
b2d444d7 | 3668 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
3669 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
3670 | set_bit(WriteMostly, &rdev->flags); | |
3671 | ||
1da177e4 LT |
3672 | rdev->raid_disk = -1; |
3673 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
3674 | if (!err && !mddev->pers->hot_remove_disk) { |
3675 | /* If there is hot_add_disk but no hot_remove_disk | |
3676 | * then added disks for geometry changes, | |
3677 | * and should be added immediately. | |
3678 | */ | |
3679 | super_types[mddev->major_version]. | |
3680 | validate_super(mddev, rdev); | |
3681 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
3682 | if (err) | |
3683 | unbind_rdev_from_array(rdev); | |
3684 | } | |
1da177e4 LT |
3685 | if (err) |
3686 | export_rdev(rdev); | |
c361777f N |
3687 | |
3688 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
005eca5e | 3689 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
3690 | return err; |
3691 | } | |
3692 | ||
3693 | /* otherwise, add_new_disk is only allowed | |
3694 | * for major_version==0 superblocks | |
3695 | */ | |
3696 | if (mddev->major_version != 0) { | |
3697 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
3698 | mdname(mddev)); | |
3699 | return -EINVAL; | |
3700 | } | |
3701 | ||
3702 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
3703 | int err; | |
3704 | rdev = md_import_device (dev, -1, 0); | |
3705 | if (IS_ERR(rdev)) { | |
3706 | printk(KERN_WARNING | |
3707 | "md: error, md_import_device() returned %ld\n", | |
3708 | PTR_ERR(rdev)); | |
3709 | return PTR_ERR(rdev); | |
3710 | } | |
3711 | rdev->desc_nr = info->number; | |
3712 | if (info->raid_disk < mddev->raid_disks) | |
3713 | rdev->raid_disk = info->raid_disk; | |
3714 | else | |
3715 | rdev->raid_disk = -1; | |
3716 | ||
b2d444d7 N |
3717 | rdev->flags = 0; |
3718 | ||
1da177e4 | 3719 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
3720 | if (info->state & (1<<MD_DISK_SYNC)) |
3721 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 3722 | |
8ddf9efe N |
3723 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
3724 | set_bit(WriteMostly, &rdev->flags); | |
3725 | ||
1da177e4 LT |
3726 | if (!mddev->persistent) { |
3727 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
3728 | rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
3729 | } else | |
3730 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
3731 | rdev->size = calc_dev_size(rdev, mddev->chunk_size); | |
3732 | ||
2bf071bf N |
3733 | err = bind_rdev_to_array(rdev, mddev); |
3734 | if (err) { | |
3735 | export_rdev(rdev); | |
3736 | return err; | |
3737 | } | |
1da177e4 LT |
3738 | } |
3739 | ||
3740 | return 0; | |
3741 | } | |
3742 | ||
3743 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
3744 | { | |
3745 | char b[BDEVNAME_SIZE]; | |
3746 | mdk_rdev_t *rdev; | |
3747 | ||
3748 | if (!mddev->pers) | |
3749 | return -ENODEV; | |
3750 | ||
3751 | rdev = find_rdev(mddev, dev); | |
3752 | if (!rdev) | |
3753 | return -ENXIO; | |
3754 | ||
3755 | if (rdev->raid_disk >= 0) | |
3756 | goto busy; | |
3757 | ||
3758 | kick_rdev_from_array(rdev); | |
3759 | md_update_sb(mddev); | |
d7603b7e | 3760 | md_new_event(mddev); |
1da177e4 LT |
3761 | |
3762 | return 0; | |
3763 | busy: | |
3764 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n", | |
3765 | bdevname(rdev->bdev,b), mdname(mddev)); | |
3766 | return -EBUSY; | |
3767 | } | |
3768 | ||
3769 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
3770 | { | |
3771 | char b[BDEVNAME_SIZE]; | |
3772 | int err; | |
3773 | unsigned int size; | |
3774 | mdk_rdev_t *rdev; | |
3775 | ||
3776 | if (!mddev->pers) | |
3777 | return -ENODEV; | |
3778 | ||
3779 | if (mddev->major_version != 0) { | |
3780 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
3781 | " version-0 superblocks.\n", | |
3782 | mdname(mddev)); | |
3783 | return -EINVAL; | |
3784 | } | |
3785 | if (!mddev->pers->hot_add_disk) { | |
3786 | printk(KERN_WARNING | |
3787 | "%s: personality does not support diskops!\n", | |
3788 | mdname(mddev)); | |
3789 | return -EINVAL; | |
3790 | } | |
3791 | ||
3792 | rdev = md_import_device (dev, -1, 0); | |
3793 | if (IS_ERR(rdev)) { | |
3794 | printk(KERN_WARNING | |
3795 | "md: error, md_import_device() returned %ld\n", | |
3796 | PTR_ERR(rdev)); | |
3797 | return -EINVAL; | |
3798 | } | |
3799 | ||
3800 | if (mddev->persistent) | |
3801 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
3802 | else | |
3803 | rdev->sb_offset = | |
3804 | rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
3805 | ||
3806 | size = calc_dev_size(rdev, mddev->chunk_size); | |
3807 | rdev->size = size; | |
3808 | ||
b2d444d7 | 3809 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
3810 | printk(KERN_WARNING |
3811 | "md: can not hot-add faulty %s disk to %s!\n", | |
3812 | bdevname(rdev->bdev,b), mdname(mddev)); | |
3813 | err = -EINVAL; | |
3814 | goto abort_export; | |
3815 | } | |
b2d444d7 | 3816 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 3817 | rdev->desc_nr = -1; |
2bf071bf N |
3818 | err = bind_rdev_to_array(rdev, mddev); |
3819 | if (err) | |
3820 | goto abort_export; | |
1da177e4 LT |
3821 | |
3822 | /* | |
3823 | * The rest should better be atomic, we can have disk failures | |
3824 | * noticed in interrupt contexts ... | |
3825 | */ | |
3826 | ||
3827 | if (rdev->desc_nr == mddev->max_disks) { | |
3828 | printk(KERN_WARNING "%s: can not hot-add to full array!\n", | |
3829 | mdname(mddev)); | |
3830 | err = -EBUSY; | |
3831 | goto abort_unbind_export; | |
3832 | } | |
3833 | ||
3834 | rdev->raid_disk = -1; | |
3835 | ||
3836 | md_update_sb(mddev); | |
3837 | ||
3838 | /* | |
3839 | * Kick recovery, maybe this spare has to be added to the | |
3840 | * array immediately. | |
3841 | */ | |
3842 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3843 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 3844 | md_new_event(mddev); |
1da177e4 LT |
3845 | return 0; |
3846 | ||
3847 | abort_unbind_export: | |
3848 | unbind_rdev_from_array(rdev); | |
3849 | ||
3850 | abort_export: | |
3851 | export_rdev(rdev); | |
3852 | return err; | |
3853 | } | |
3854 | ||
32a7627c N |
3855 | static int set_bitmap_file(mddev_t *mddev, int fd) |
3856 | { | |
3857 | int err; | |
3858 | ||
36fa3063 N |
3859 | if (mddev->pers) { |
3860 | if (!mddev->pers->quiesce) | |
3861 | return -EBUSY; | |
3862 | if (mddev->recovery || mddev->sync_thread) | |
3863 | return -EBUSY; | |
3864 | /* we should be able to change the bitmap.. */ | |
3865 | } | |
32a7627c | 3866 | |
32a7627c | 3867 | |
36fa3063 N |
3868 | if (fd >= 0) { |
3869 | if (mddev->bitmap) | |
3870 | return -EEXIST; /* cannot add when bitmap is present */ | |
3871 | mddev->bitmap_file = fget(fd); | |
32a7627c | 3872 | |
36fa3063 N |
3873 | if (mddev->bitmap_file == NULL) { |
3874 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", | |
3875 | mdname(mddev)); | |
3876 | return -EBADF; | |
3877 | } | |
3878 | ||
3879 | err = deny_bitmap_write_access(mddev->bitmap_file); | |
3880 | if (err) { | |
3881 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
3882 | mdname(mddev)); | |
3883 | fput(mddev->bitmap_file); | |
3884 | mddev->bitmap_file = NULL; | |
3885 | return err; | |
3886 | } | |
a654b9d8 | 3887 | mddev->bitmap_offset = 0; /* file overrides offset */ |
36fa3063 N |
3888 | } else if (mddev->bitmap == NULL) |
3889 | return -ENOENT; /* cannot remove what isn't there */ | |
3890 | err = 0; | |
3891 | if (mddev->pers) { | |
3892 | mddev->pers->quiesce(mddev, 1); | |
3893 | if (fd >= 0) | |
3894 | err = bitmap_create(mddev); | |
d7375ab3 | 3895 | if (fd < 0 || err) { |
36fa3063 | 3896 | bitmap_destroy(mddev); |
d7375ab3 N |
3897 | fd = -1; /* make sure to put the file */ |
3898 | } | |
36fa3063 | 3899 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
3900 | } |
3901 | if (fd < 0) { | |
acc55e22 N |
3902 | if (mddev->bitmap_file) { |
3903 | restore_bitmap_write_access(mddev->bitmap_file); | |
36fa3063 | 3904 | fput(mddev->bitmap_file); |
acc55e22 | 3905 | } |
36fa3063 N |
3906 | mddev->bitmap_file = NULL; |
3907 | } | |
3908 | ||
32a7627c N |
3909 | return err; |
3910 | } | |
3911 | ||
1da177e4 LT |
3912 | /* |
3913 | * set_array_info is used two different ways | |
3914 | * The original usage is when creating a new array. | |
3915 | * In this usage, raid_disks is > 0 and it together with | |
3916 | * level, size, not_persistent,layout,chunksize determine the | |
3917 | * shape of the array. | |
3918 | * This will always create an array with a type-0.90.0 superblock. | |
3919 | * The newer usage is when assembling an array. | |
3920 | * In this case raid_disks will be 0, and the major_version field is | |
3921 | * use to determine which style super-blocks are to be found on the devices. | |
3922 | * The minor and patch _version numbers are also kept incase the | |
3923 | * super_block handler wishes to interpret them. | |
3924 | */ | |
3925 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
3926 | { | |
3927 | ||
3928 | if (info->raid_disks == 0) { | |
3929 | /* just setting version number for superblock loading */ | |
3930 | if (info->major_version < 0 || | |
3931 | info->major_version >= sizeof(super_types)/sizeof(super_types[0]) || | |
3932 | super_types[info->major_version].name == NULL) { | |
3933 | /* maybe try to auto-load a module? */ | |
3934 | printk(KERN_INFO | |
3935 | "md: superblock version %d not known\n", | |
3936 | info->major_version); | |
3937 | return -EINVAL; | |
3938 | } | |
3939 | mddev->major_version = info->major_version; | |
3940 | mddev->minor_version = info->minor_version; | |
3941 | mddev->patch_version = info->patch_version; | |
3942 | return 0; | |
3943 | } | |
3944 | mddev->major_version = MD_MAJOR_VERSION; | |
3945 | mddev->minor_version = MD_MINOR_VERSION; | |
3946 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
3947 | mddev->ctime = get_seconds(); | |
3948 | ||
3949 | mddev->level = info->level; | |
17115e03 | 3950 | mddev->clevel[0] = 0; |
1da177e4 LT |
3951 | mddev->size = info->size; |
3952 | mddev->raid_disks = info->raid_disks; | |
3953 | /* don't set md_minor, it is determined by which /dev/md* was | |
3954 | * openned | |
3955 | */ | |
3956 | if (info->state & (1<<MD_SB_CLEAN)) | |
3957 | mddev->recovery_cp = MaxSector; | |
3958 | else | |
3959 | mddev->recovery_cp = 0; | |
3960 | mddev->persistent = ! info->not_persistent; | |
3961 | ||
3962 | mddev->layout = info->layout; | |
3963 | mddev->chunk_size = info->chunk_size; | |
3964 | ||
3965 | mddev->max_disks = MD_SB_DISKS; | |
3966 | ||
3967 | mddev->sb_dirty = 1; | |
3968 | ||
b2a2703c N |
3969 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
3970 | mddev->bitmap_offset = 0; | |
3971 | ||
f6705578 N |
3972 | mddev->reshape_position = MaxSector; |
3973 | ||
1da177e4 LT |
3974 | /* |
3975 | * Generate a 128 bit UUID | |
3976 | */ | |
3977 | get_random_bytes(mddev->uuid, 16); | |
3978 | ||
f6705578 N |
3979 | mddev->new_level = mddev->level; |
3980 | mddev->new_chunk = mddev->chunk_size; | |
3981 | mddev->new_layout = mddev->layout; | |
3982 | mddev->delta_disks = 0; | |
3983 | ||
1da177e4 LT |
3984 | return 0; |
3985 | } | |
3986 | ||
a35b0d69 N |
3987 | static int update_size(mddev_t *mddev, unsigned long size) |
3988 | { | |
3989 | mdk_rdev_t * rdev; | |
3990 | int rv; | |
3991 | struct list_head *tmp; | |
8ddeeae5 | 3992 | int fit = (size == 0); |
a35b0d69 N |
3993 | |
3994 | if (mddev->pers->resize == NULL) | |
3995 | return -EINVAL; | |
3996 | /* The "size" is the amount of each device that is used. | |
3997 | * This can only make sense for arrays with redundancy. | |
3998 | * linear and raid0 always use whatever space is available | |
3999 | * We can only consider changing the size if no resync | |
4000 | * or reconstruction is happening, and if the new size | |
4001 | * is acceptable. It must fit before the sb_offset or, | |
4002 | * if that is <data_offset, it must fit before the | |
4003 | * size of each device. | |
4004 | * If size is zero, we find the largest size that fits. | |
4005 | */ | |
4006 | if (mddev->sync_thread) | |
4007 | return -EBUSY; | |
4008 | ITERATE_RDEV(mddev,rdev,tmp) { | |
4009 | sector_t avail; | |
a35b0d69 N |
4010 | if (rdev->sb_offset > rdev->data_offset) |
4011 | avail = (rdev->sb_offset*2) - rdev->data_offset; | |
4012 | else | |
4013 | avail = get_capacity(rdev->bdev->bd_disk) | |
4014 | - rdev->data_offset; | |
4015 | if (fit && (size == 0 || size > avail/2)) | |
4016 | size = avail/2; | |
4017 | if (avail < ((sector_t)size << 1)) | |
4018 | return -ENOSPC; | |
4019 | } | |
4020 | rv = mddev->pers->resize(mddev, (sector_t)size *2); | |
4021 | if (!rv) { | |
4022 | struct block_device *bdev; | |
4023 | ||
4024 | bdev = bdget_disk(mddev->gendisk, 0); | |
4025 | if (bdev) { | |
1b1dcc1b | 4026 | mutex_lock(&bdev->bd_inode->i_mutex); |
6d89332b | 4027 | i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10); |
1b1dcc1b | 4028 | mutex_unlock(&bdev->bd_inode->i_mutex); |
a35b0d69 N |
4029 | bdput(bdev); |
4030 | } | |
4031 | } | |
4032 | return rv; | |
4033 | } | |
4034 | ||
da943b99 N |
4035 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
4036 | { | |
4037 | int rv; | |
4038 | /* change the number of raid disks */ | |
63c70c4f | 4039 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
4040 | return -EINVAL; |
4041 | if (raid_disks <= 0 || | |
4042 | raid_disks >= mddev->max_disks) | |
4043 | return -EINVAL; | |
63c70c4f | 4044 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 4045 | return -EBUSY; |
63c70c4f N |
4046 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
4047 | ||
4048 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
4049 | return rv; |
4050 | } | |
4051 | ||
4052 | ||
1da177e4 LT |
4053 | /* |
4054 | * update_array_info is used to change the configuration of an | |
4055 | * on-line array. | |
4056 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
4057 | * fields in the info are checked against the array. | |
4058 | * Any differences that cannot be handled will cause an error. | |
4059 | * Normally, only one change can be managed at a time. | |
4060 | */ | |
4061 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
4062 | { | |
4063 | int rv = 0; | |
4064 | int cnt = 0; | |
36fa3063 N |
4065 | int state = 0; |
4066 | ||
4067 | /* calculate expected state,ignoring low bits */ | |
4068 | if (mddev->bitmap && mddev->bitmap_offset) | |
4069 | state |= (1 << MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
4070 | |
4071 | if (mddev->major_version != info->major_version || | |
4072 | mddev->minor_version != info->minor_version || | |
4073 | /* mddev->patch_version != info->patch_version || */ | |
4074 | mddev->ctime != info->ctime || | |
4075 | mddev->level != info->level || | |
4076 | /* mddev->layout != info->layout || */ | |
4077 | !mddev->persistent != info->not_persistent|| | |
36fa3063 N |
4078 | mddev->chunk_size != info->chunk_size || |
4079 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ | |
4080 | ((state^info->state) & 0xfffffe00) | |
4081 | ) | |
1da177e4 LT |
4082 | return -EINVAL; |
4083 | /* Check there is only one change */ | |
284ae7ca | 4084 | if (info->size >= 0 && mddev->size != info->size) cnt++; |
1da177e4 LT |
4085 | if (mddev->raid_disks != info->raid_disks) cnt++; |
4086 | if (mddev->layout != info->layout) cnt++; | |
36fa3063 | 4087 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++; |
1da177e4 LT |
4088 | if (cnt == 0) return 0; |
4089 | if (cnt > 1) return -EINVAL; | |
4090 | ||
4091 | if (mddev->layout != info->layout) { | |
4092 | /* Change layout | |
4093 | * we don't need to do anything at the md level, the | |
4094 | * personality will take care of it all. | |
4095 | */ | |
4096 | if (mddev->pers->reconfig == NULL) | |
4097 | return -EINVAL; | |
4098 | else | |
4099 | return mddev->pers->reconfig(mddev, info->layout, -1); | |
4100 | } | |
284ae7ca | 4101 | if (info->size >= 0 && mddev->size != info->size) |
a35b0d69 N |
4102 | rv = update_size(mddev, info->size); |
4103 | ||
da943b99 N |
4104 | if (mddev->raid_disks != info->raid_disks) |
4105 | rv = update_raid_disks(mddev, info->raid_disks); | |
4106 | ||
36fa3063 N |
4107 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
4108 | if (mddev->pers->quiesce == NULL) | |
4109 | return -EINVAL; | |
4110 | if (mddev->recovery || mddev->sync_thread) | |
4111 | return -EBUSY; | |
4112 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
4113 | /* add the bitmap */ | |
4114 | if (mddev->bitmap) | |
4115 | return -EEXIST; | |
4116 | if (mddev->default_bitmap_offset == 0) | |
4117 | return -EINVAL; | |
4118 | mddev->bitmap_offset = mddev->default_bitmap_offset; | |
4119 | mddev->pers->quiesce(mddev, 1); | |
4120 | rv = bitmap_create(mddev); | |
4121 | if (rv) | |
4122 | bitmap_destroy(mddev); | |
4123 | mddev->pers->quiesce(mddev, 0); | |
4124 | } else { | |
4125 | /* remove the bitmap */ | |
4126 | if (!mddev->bitmap) | |
4127 | return -ENOENT; | |
4128 | if (mddev->bitmap->file) | |
4129 | return -EINVAL; | |
4130 | mddev->pers->quiesce(mddev, 1); | |
4131 | bitmap_destroy(mddev); | |
4132 | mddev->pers->quiesce(mddev, 0); | |
4133 | mddev->bitmap_offset = 0; | |
4134 | } | |
4135 | } | |
1da177e4 LT |
4136 | md_update_sb(mddev); |
4137 | return rv; | |
4138 | } | |
4139 | ||
4140 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
4141 | { | |
4142 | mdk_rdev_t *rdev; | |
4143 | ||
4144 | if (mddev->pers == NULL) | |
4145 | return -ENODEV; | |
4146 | ||
4147 | rdev = find_rdev(mddev, dev); | |
4148 | if (!rdev) | |
4149 | return -ENODEV; | |
4150 | ||
4151 | md_error(mddev, rdev); | |
4152 | return 0; | |
4153 | } | |
4154 | ||
a885c8c4 CH |
4155 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
4156 | { | |
4157 | mddev_t *mddev = bdev->bd_disk->private_data; | |
4158 | ||
4159 | geo->heads = 2; | |
4160 | geo->sectors = 4; | |
4161 | geo->cylinders = get_capacity(mddev->gendisk) / 8; | |
4162 | return 0; | |
4163 | } | |
4164 | ||
1da177e4 LT |
4165 | static int md_ioctl(struct inode *inode, struct file *file, |
4166 | unsigned int cmd, unsigned long arg) | |
4167 | { | |
4168 | int err = 0; | |
4169 | void __user *argp = (void __user *)arg; | |
1da177e4 LT |
4170 | mddev_t *mddev = NULL; |
4171 | ||
4172 | if (!capable(CAP_SYS_ADMIN)) | |
4173 | return -EACCES; | |
4174 | ||
4175 | /* | |
4176 | * Commands dealing with the RAID driver but not any | |
4177 | * particular array: | |
4178 | */ | |
4179 | switch (cmd) | |
4180 | { | |
4181 | case RAID_VERSION: | |
4182 | err = get_version(argp); | |
4183 | goto done; | |
4184 | ||
4185 | case PRINT_RAID_DEBUG: | |
4186 | err = 0; | |
4187 | md_print_devices(); | |
4188 | goto done; | |
4189 | ||
4190 | #ifndef MODULE | |
4191 | case RAID_AUTORUN: | |
4192 | err = 0; | |
4193 | autostart_arrays(arg); | |
4194 | goto done; | |
4195 | #endif | |
4196 | default:; | |
4197 | } | |
4198 | ||
4199 | /* | |
4200 | * Commands creating/starting a new array: | |
4201 | */ | |
4202 | ||
4203 | mddev = inode->i_bdev->bd_disk->private_data; | |
4204 | ||
4205 | if (!mddev) { | |
4206 | BUG(); | |
4207 | goto abort; | |
4208 | } | |
4209 | ||
4210 | ||
4211 | if (cmd == START_ARRAY) { | |
4212 | /* START_ARRAY doesn't need to lock the array as autostart_array | |
4213 | * does the locking, and it could even be a different array | |
4214 | */ | |
4215 | static int cnt = 3; | |
4216 | if (cnt > 0 ) { | |
4217 | printk(KERN_WARNING | |
4218 | "md: %s(pid %d) used deprecated START_ARRAY ioctl. " | |
e8a00334 | 4219 | "This will not be supported beyond July 2006\n", |
1da177e4 LT |
4220 | current->comm, current->pid); |
4221 | cnt--; | |
4222 | } | |
4223 | err = autostart_array(new_decode_dev(arg)); | |
4224 | if (err) { | |
4225 | printk(KERN_WARNING "md: autostart failed!\n"); | |
4226 | goto abort; | |
4227 | } | |
4228 | goto done; | |
4229 | } | |
4230 | ||
4231 | err = mddev_lock(mddev); | |
4232 | if (err) { | |
4233 | printk(KERN_INFO | |
4234 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
4235 | err, cmd); | |
4236 | goto abort; | |
4237 | } | |
4238 | ||
4239 | switch (cmd) | |
4240 | { | |
4241 | case SET_ARRAY_INFO: | |
4242 | { | |
4243 | mdu_array_info_t info; | |
4244 | if (!arg) | |
4245 | memset(&info, 0, sizeof(info)); | |
4246 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
4247 | err = -EFAULT; | |
4248 | goto abort_unlock; | |
4249 | } | |
4250 | if (mddev->pers) { | |
4251 | err = update_array_info(mddev, &info); | |
4252 | if (err) { | |
4253 | printk(KERN_WARNING "md: couldn't update" | |
4254 | " array info. %d\n", err); | |
4255 | goto abort_unlock; | |
4256 | } | |
4257 | goto done_unlock; | |
4258 | } | |
4259 | if (!list_empty(&mddev->disks)) { | |
4260 | printk(KERN_WARNING | |
4261 | "md: array %s already has disks!\n", | |
4262 | mdname(mddev)); | |
4263 | err = -EBUSY; | |
4264 | goto abort_unlock; | |
4265 | } | |
4266 | if (mddev->raid_disks) { | |
4267 | printk(KERN_WARNING | |
4268 | "md: array %s already initialised!\n", | |
4269 | mdname(mddev)); | |
4270 | err = -EBUSY; | |
4271 | goto abort_unlock; | |
4272 | } | |
4273 | err = set_array_info(mddev, &info); | |
4274 | if (err) { | |
4275 | printk(KERN_WARNING "md: couldn't set" | |
4276 | " array info. %d\n", err); | |
4277 | goto abort_unlock; | |
4278 | } | |
4279 | } | |
4280 | goto done_unlock; | |
4281 | ||
4282 | default:; | |
4283 | } | |
4284 | ||
4285 | /* | |
4286 | * Commands querying/configuring an existing array: | |
4287 | */ | |
32a7627c N |
4288 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
4289 | * RUN_ARRAY, and SET_BITMAP_FILE are allowed */ | |
4290 | if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
4291 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) { | |
1da177e4 LT |
4292 | err = -ENODEV; |
4293 | goto abort_unlock; | |
4294 | } | |
4295 | ||
4296 | /* | |
4297 | * Commands even a read-only array can execute: | |
4298 | */ | |
4299 | switch (cmd) | |
4300 | { | |
4301 | case GET_ARRAY_INFO: | |
4302 | err = get_array_info(mddev, argp); | |
4303 | goto done_unlock; | |
4304 | ||
32a7627c | 4305 | case GET_BITMAP_FILE: |
87162a28 | 4306 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
4307 | goto done_unlock; |
4308 | ||
1da177e4 LT |
4309 | case GET_DISK_INFO: |
4310 | err = get_disk_info(mddev, argp); | |
4311 | goto done_unlock; | |
4312 | ||
4313 | case RESTART_ARRAY_RW: | |
4314 | err = restart_array(mddev); | |
4315 | goto done_unlock; | |
4316 | ||
4317 | case STOP_ARRAY: | |
4318 | err = do_md_stop (mddev, 0); | |
4319 | goto done_unlock; | |
4320 | ||
4321 | case STOP_ARRAY_RO: | |
4322 | err = do_md_stop (mddev, 1); | |
4323 | goto done_unlock; | |
4324 | ||
4325 | /* | |
4326 | * We have a problem here : there is no easy way to give a CHS | |
4327 | * virtual geometry. We currently pretend that we have a 2 heads | |
4328 | * 4 sectors (with a BIG number of cylinders...). This drives | |
4329 | * dosfs just mad... ;-) | |
4330 | */ | |
1da177e4 LT |
4331 | } |
4332 | ||
4333 | /* | |
4334 | * The remaining ioctls are changing the state of the | |
f91de92e N |
4335 | * superblock, so we do not allow them on read-only arrays. |
4336 | * However non-MD ioctls (e.g. get-size) will still come through | |
4337 | * here and hit the 'default' below, so only disallow | |
4338 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 4339 | */ |
f91de92e N |
4340 | if (_IOC_TYPE(cmd) == MD_MAJOR && |
4341 | mddev->ro && mddev->pers) { | |
4342 | if (mddev->ro == 2) { | |
4343 | mddev->ro = 0; | |
4344 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4345 | md_wakeup_thread(mddev->thread); | |
4346 | ||
4347 | } else { | |
4348 | err = -EROFS; | |
4349 | goto abort_unlock; | |
4350 | } | |
1da177e4 LT |
4351 | } |
4352 | ||
4353 | switch (cmd) | |
4354 | { | |
4355 | case ADD_NEW_DISK: | |
4356 | { | |
4357 | mdu_disk_info_t info; | |
4358 | if (copy_from_user(&info, argp, sizeof(info))) | |
4359 | err = -EFAULT; | |
4360 | else | |
4361 | err = add_new_disk(mddev, &info); | |
4362 | goto done_unlock; | |
4363 | } | |
4364 | ||
4365 | case HOT_REMOVE_DISK: | |
4366 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
4367 | goto done_unlock; | |
4368 | ||
4369 | case HOT_ADD_DISK: | |
4370 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
4371 | goto done_unlock; | |
4372 | ||
4373 | case SET_DISK_FAULTY: | |
4374 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
4375 | goto done_unlock; | |
4376 | ||
4377 | case RUN_ARRAY: | |
4378 | err = do_md_run (mddev); | |
4379 | goto done_unlock; | |
4380 | ||
32a7627c N |
4381 | case SET_BITMAP_FILE: |
4382 | err = set_bitmap_file(mddev, (int)arg); | |
4383 | goto done_unlock; | |
4384 | ||
1da177e4 | 4385 | default: |
1da177e4 LT |
4386 | err = -EINVAL; |
4387 | goto abort_unlock; | |
4388 | } | |
4389 | ||
4390 | done_unlock: | |
4391 | abort_unlock: | |
4392 | mddev_unlock(mddev); | |
4393 | ||
4394 | return err; | |
4395 | done: | |
4396 | if (err) | |
4397 | MD_BUG(); | |
4398 | abort: | |
4399 | return err; | |
4400 | } | |
4401 | ||
4402 | static int md_open(struct inode *inode, struct file *file) | |
4403 | { | |
4404 | /* | |
4405 | * Succeed if we can lock the mddev, which confirms that | |
4406 | * it isn't being stopped right now. | |
4407 | */ | |
4408 | mddev_t *mddev = inode->i_bdev->bd_disk->private_data; | |
4409 | int err; | |
4410 | ||
4411 | if ((err = mddev_lock(mddev))) | |
4412 | goto out; | |
4413 | ||
4414 | err = 0; | |
4415 | mddev_get(mddev); | |
4416 | mddev_unlock(mddev); | |
4417 | ||
4418 | check_disk_change(inode->i_bdev); | |
4419 | out: | |
4420 | return err; | |
4421 | } | |
4422 | ||
4423 | static int md_release(struct inode *inode, struct file * file) | |
4424 | { | |
4425 | mddev_t *mddev = inode->i_bdev->bd_disk->private_data; | |
4426 | ||
4427 | if (!mddev) | |
4428 | BUG(); | |
4429 | mddev_put(mddev); | |
4430 | ||
4431 | return 0; | |
4432 | } | |
4433 | ||
4434 | static int md_media_changed(struct gendisk *disk) | |
4435 | { | |
4436 | mddev_t *mddev = disk->private_data; | |
4437 | ||
4438 | return mddev->changed; | |
4439 | } | |
4440 | ||
4441 | static int md_revalidate(struct gendisk *disk) | |
4442 | { | |
4443 | mddev_t *mddev = disk->private_data; | |
4444 | ||
4445 | mddev->changed = 0; | |
4446 | return 0; | |
4447 | } | |
4448 | static struct block_device_operations md_fops = | |
4449 | { | |
4450 | .owner = THIS_MODULE, | |
4451 | .open = md_open, | |
4452 | .release = md_release, | |
4453 | .ioctl = md_ioctl, | |
a885c8c4 | 4454 | .getgeo = md_getgeo, |
1da177e4 LT |
4455 | .media_changed = md_media_changed, |
4456 | .revalidate_disk= md_revalidate, | |
4457 | }; | |
4458 | ||
75c96f85 | 4459 | static int md_thread(void * arg) |
1da177e4 LT |
4460 | { |
4461 | mdk_thread_t *thread = arg; | |
4462 | ||
1da177e4 LT |
4463 | /* |
4464 | * md_thread is a 'system-thread', it's priority should be very | |
4465 | * high. We avoid resource deadlocks individually in each | |
4466 | * raid personality. (RAID5 does preallocation) We also use RR and | |
4467 | * the very same RT priority as kswapd, thus we will never get | |
4468 | * into a priority inversion deadlock. | |
4469 | * | |
4470 | * we definitely have to have equal or higher priority than | |
4471 | * bdflush, otherwise bdflush will deadlock if there are too | |
4472 | * many dirty RAID5 blocks. | |
4473 | */ | |
1da177e4 | 4474 | |
6985c43f | 4475 | allow_signal(SIGKILL); |
a6fb0934 | 4476 | while (!kthread_should_stop()) { |
1da177e4 | 4477 | |
93588e22 N |
4478 | /* We need to wait INTERRUPTIBLE so that |
4479 | * we don't add to the load-average. | |
4480 | * That means we need to be sure no signals are | |
4481 | * pending | |
4482 | */ | |
4483 | if (signal_pending(current)) | |
4484 | flush_signals(current); | |
4485 | ||
4486 | wait_event_interruptible_timeout | |
4487 | (thread->wqueue, | |
4488 | test_bit(THREAD_WAKEUP, &thread->flags) | |
4489 | || kthread_should_stop(), | |
4490 | thread->timeout); | |
3e1d1d28 | 4491 | try_to_freeze(); |
1da177e4 LT |
4492 | |
4493 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
4494 | ||
787453c2 | 4495 | thread->run(thread->mddev); |
1da177e4 | 4496 | } |
a6fb0934 | 4497 | |
1da177e4 LT |
4498 | return 0; |
4499 | } | |
4500 | ||
4501 | void md_wakeup_thread(mdk_thread_t *thread) | |
4502 | { | |
4503 | if (thread) { | |
4504 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
4505 | set_bit(THREAD_WAKEUP, &thread->flags); | |
4506 | wake_up(&thread->wqueue); | |
4507 | } | |
4508 | } | |
4509 | ||
4510 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
4511 | const char *name) | |
4512 | { | |
4513 | mdk_thread_t *thread; | |
1da177e4 | 4514 | |
9ffae0cf | 4515 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
4516 | if (!thread) |
4517 | return NULL; | |
4518 | ||
1da177e4 LT |
4519 | init_waitqueue_head(&thread->wqueue); |
4520 | ||
1da177e4 LT |
4521 | thread->run = run; |
4522 | thread->mddev = mddev; | |
32a7627c | 4523 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
6985c43f | 4524 | thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev)); |
a6fb0934 | 4525 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
4526 | kfree(thread); |
4527 | return NULL; | |
4528 | } | |
1da177e4 LT |
4529 | return thread; |
4530 | } | |
4531 | ||
1da177e4 LT |
4532 | void md_unregister_thread(mdk_thread_t *thread) |
4533 | { | |
d28446fe | 4534 | dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid); |
a6fb0934 N |
4535 | |
4536 | kthread_stop(thread->tsk); | |
1da177e4 LT |
4537 | kfree(thread); |
4538 | } | |
4539 | ||
4540 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
4541 | { | |
4542 | if (!mddev) { | |
4543 | MD_BUG(); | |
4544 | return; | |
4545 | } | |
4546 | ||
b2d444d7 | 4547 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4548 | return; |
32a7627c | 4549 | /* |
1da177e4 LT |
4550 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
4551 | mdname(mddev), | |
4552 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
4553 | __builtin_return_address(0),__builtin_return_address(1), | |
4554 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 4555 | */ |
1da177e4 LT |
4556 | if (!mddev->pers->error_handler) |
4557 | return; | |
4558 | mddev->pers->error_handler(mddev,rdev); | |
4559 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
4560 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4561 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 4562 | md_new_event_inintr(mddev); |
1da177e4 LT |
4563 | } |
4564 | ||
4565 | /* seq_file implementation /proc/mdstat */ | |
4566 | ||
4567 | static void status_unused(struct seq_file *seq) | |
4568 | { | |
4569 | int i = 0; | |
4570 | mdk_rdev_t *rdev; | |
4571 | struct list_head *tmp; | |
4572 | ||
4573 | seq_printf(seq, "unused devices: "); | |
4574 | ||
4575 | ITERATE_RDEV_PENDING(rdev,tmp) { | |
4576 | char b[BDEVNAME_SIZE]; | |
4577 | i++; | |
4578 | seq_printf(seq, "%s ", | |
4579 | bdevname(rdev->bdev,b)); | |
4580 | } | |
4581 | if (!i) | |
4582 | seq_printf(seq, "<none>"); | |
4583 | ||
4584 | seq_printf(seq, "\n"); | |
4585 | } | |
4586 | ||
4587 | ||
4588 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
4589 | { | |
4588b42e N |
4590 | sector_t max_blocks, resync, res; |
4591 | unsigned long dt, db, rt; | |
4592 | int scale; | |
4593 | unsigned int per_milli; | |
1da177e4 LT |
4594 | |
4595 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2; | |
4596 | ||
4597 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
4598 | max_blocks = mddev->resync_max_sectors >> 1; | |
4599 | else | |
4600 | max_blocks = mddev->size; | |
4601 | ||
4602 | /* | |
4603 | * Should not happen. | |
4604 | */ | |
4605 | if (!max_blocks) { | |
4606 | MD_BUG(); | |
4607 | return; | |
4608 | } | |
4588b42e N |
4609 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
4610 | * in a sector_t, and (max_blocks>>scale) will fit in a | |
4611 | * u32, as those are the requirements for sector_div. | |
4612 | * Thus 'scale' must be at least 10 | |
4613 | */ | |
4614 | scale = 10; | |
4615 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
4616 | while ( max_blocks/2 > (1ULL<<(scale+32))) | |
4617 | scale++; | |
4618 | } | |
4619 | res = (resync>>scale)*1000; | |
4620 | sector_div(res, (u32)((max_blocks>>scale)+1)); | |
4621 | ||
4622 | per_milli = res; | |
1da177e4 | 4623 | { |
4588b42e | 4624 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
4625 | seq_printf(seq, "["); |
4626 | for (i = 0; i < x; i++) | |
4627 | seq_printf(seq, "="); | |
4628 | seq_printf(seq, ">"); | |
4629 | for (i = 0; i < y; i++) | |
4630 | seq_printf(seq, "."); | |
4631 | seq_printf(seq, "] "); | |
4632 | } | |
4588b42e | 4633 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
4634 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
4635 | "reshape" : | |
1da177e4 | 4636 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? |
ccfcc3c1 | 4637 | "resync" : "recovery")), |
4588b42e N |
4638 | per_milli/10, per_milli % 10, |
4639 | (unsigned long long) resync, | |
4640 | (unsigned long long) max_blocks); | |
1da177e4 LT |
4641 | |
4642 | /* | |
4643 | * We do not want to overflow, so the order of operands and | |
4644 | * the * 100 / 100 trick are important. We do a +1 to be | |
4645 | * safe against division by zero. We only estimate anyway. | |
4646 | * | |
4647 | * dt: time from mark until now | |
4648 | * db: blocks written from mark until now | |
4649 | * rt: remaining time | |
4650 | */ | |
4651 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
4652 | if (!dt) dt++; | |
4653 | db = resync - (mddev->resync_mark_cnt/2); | |
4588b42e | 4654 | rt = (dt * ((unsigned long)(max_blocks-resync) / (db/100+1)))/100; |
1da177e4 LT |
4655 | |
4656 | seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6); | |
4657 | ||
4658 | seq_printf(seq, " speed=%ldK/sec", db/dt); | |
4659 | } | |
4660 | ||
4661 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
4662 | { | |
4663 | struct list_head *tmp; | |
4664 | loff_t l = *pos; | |
4665 | mddev_t *mddev; | |
4666 | ||
4667 | if (l >= 0x10000) | |
4668 | return NULL; | |
4669 | if (!l--) | |
4670 | /* header */ | |
4671 | return (void*)1; | |
4672 | ||
4673 | spin_lock(&all_mddevs_lock); | |
4674 | list_for_each(tmp,&all_mddevs) | |
4675 | if (!l--) { | |
4676 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
4677 | mddev_get(mddev); | |
4678 | spin_unlock(&all_mddevs_lock); | |
4679 | return mddev; | |
4680 | } | |
4681 | spin_unlock(&all_mddevs_lock); | |
4682 | if (!l--) | |
4683 | return (void*)2;/* tail */ | |
4684 | return NULL; | |
4685 | } | |
4686 | ||
4687 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4688 | { | |
4689 | struct list_head *tmp; | |
4690 | mddev_t *next_mddev, *mddev = v; | |
4691 | ||
4692 | ++*pos; | |
4693 | if (v == (void*)2) | |
4694 | return NULL; | |
4695 | ||
4696 | spin_lock(&all_mddevs_lock); | |
4697 | if (v == (void*)1) | |
4698 | tmp = all_mddevs.next; | |
4699 | else | |
4700 | tmp = mddev->all_mddevs.next; | |
4701 | if (tmp != &all_mddevs) | |
4702 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
4703 | else { | |
4704 | next_mddev = (void*)2; | |
4705 | *pos = 0x10000; | |
4706 | } | |
4707 | spin_unlock(&all_mddevs_lock); | |
4708 | ||
4709 | if (v != (void*)1) | |
4710 | mddev_put(mddev); | |
4711 | return next_mddev; | |
4712 | ||
4713 | } | |
4714 | ||
4715 | static void md_seq_stop(struct seq_file *seq, void *v) | |
4716 | { | |
4717 | mddev_t *mddev = v; | |
4718 | ||
4719 | if (mddev && v != (void*)1 && v != (void*)2) | |
4720 | mddev_put(mddev); | |
4721 | } | |
4722 | ||
d7603b7e N |
4723 | struct mdstat_info { |
4724 | int event; | |
4725 | }; | |
4726 | ||
1da177e4 LT |
4727 | static int md_seq_show(struct seq_file *seq, void *v) |
4728 | { | |
4729 | mddev_t *mddev = v; | |
4730 | sector_t size; | |
4731 | struct list_head *tmp2; | |
4732 | mdk_rdev_t *rdev; | |
d7603b7e | 4733 | struct mdstat_info *mi = seq->private; |
32a7627c | 4734 | struct bitmap *bitmap; |
1da177e4 LT |
4735 | |
4736 | if (v == (void*)1) { | |
2604b703 | 4737 | struct mdk_personality *pers; |
1da177e4 LT |
4738 | seq_printf(seq, "Personalities : "); |
4739 | spin_lock(&pers_lock); | |
2604b703 N |
4740 | list_for_each_entry(pers, &pers_list, list) |
4741 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
4742 | |
4743 | spin_unlock(&pers_lock); | |
4744 | seq_printf(seq, "\n"); | |
d7603b7e | 4745 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
4746 | return 0; |
4747 | } | |
4748 | if (v == (void*)2) { | |
4749 | status_unused(seq); | |
4750 | return 0; | |
4751 | } | |
4752 | ||
5dc5cf7d | 4753 | if (mddev_lock(mddev) < 0) |
1da177e4 | 4754 | return -EINTR; |
5dc5cf7d | 4755 | |
1da177e4 LT |
4756 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
4757 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
4758 | mddev->pers ? "" : "in"); | |
4759 | if (mddev->pers) { | |
f91de92e | 4760 | if (mddev->ro==1) |
1da177e4 | 4761 | seq_printf(seq, " (read-only)"); |
f91de92e N |
4762 | if (mddev->ro==2) |
4763 | seq_printf(seq, "(auto-read-only)"); | |
1da177e4 LT |
4764 | seq_printf(seq, " %s", mddev->pers->name); |
4765 | } | |
4766 | ||
4767 | size = 0; | |
4768 | ITERATE_RDEV(mddev,rdev,tmp2) { | |
4769 | char b[BDEVNAME_SIZE]; | |
4770 | seq_printf(seq, " %s[%d]", | |
4771 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
4772 | if (test_bit(WriteMostly, &rdev->flags)) |
4773 | seq_printf(seq, "(W)"); | |
b2d444d7 | 4774 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
4775 | seq_printf(seq, "(F)"); |
4776 | continue; | |
b325a32e N |
4777 | } else if (rdev->raid_disk < 0) |
4778 | seq_printf(seq, "(S)"); /* spare */ | |
1da177e4 LT |
4779 | size += rdev->size; |
4780 | } | |
4781 | ||
4782 | if (!list_empty(&mddev->disks)) { | |
4783 | if (mddev->pers) | |
4784 | seq_printf(seq, "\n %llu blocks", | |
4785 | (unsigned long long)mddev->array_size); | |
4786 | else | |
4787 | seq_printf(seq, "\n %llu blocks", | |
4788 | (unsigned long long)size); | |
4789 | } | |
1cd6bf19 N |
4790 | if (mddev->persistent) { |
4791 | if (mddev->major_version != 0 || | |
4792 | mddev->minor_version != 90) { | |
4793 | seq_printf(seq," super %d.%d", | |
4794 | mddev->major_version, | |
4795 | mddev->minor_version); | |
4796 | } | |
4797 | } else | |
4798 | seq_printf(seq, " super non-persistent"); | |
1da177e4 LT |
4799 | |
4800 | if (mddev->pers) { | |
4801 | mddev->pers->status (seq, mddev); | |
4802 | seq_printf(seq, "\n "); | |
8e1b39d6 N |
4803 | if (mddev->pers->sync_request) { |
4804 | if (mddev->curr_resync > 2) { | |
4805 | status_resync (seq, mddev); | |
4806 | seq_printf(seq, "\n "); | |
4807 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
4808 | seq_printf(seq, "\tresync=DELAYED\n "); | |
4809 | else if (mddev->recovery_cp < MaxSector) | |
4810 | seq_printf(seq, "\tresync=PENDING\n "); | |
4811 | } | |
32a7627c N |
4812 | } else |
4813 | seq_printf(seq, "\n "); | |
4814 | ||
4815 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
4816 | unsigned long chunk_kb; |
4817 | unsigned long flags; | |
32a7627c N |
4818 | spin_lock_irqsave(&bitmap->lock, flags); |
4819 | chunk_kb = bitmap->chunksize >> 10; | |
4820 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
4821 | "%lu%s chunk", | |
4822 | bitmap->pages - bitmap->missing_pages, | |
4823 | bitmap->pages, | |
4824 | (bitmap->pages - bitmap->missing_pages) | |
4825 | << (PAGE_SHIFT - 10), | |
4826 | chunk_kb ? chunk_kb : bitmap->chunksize, | |
4827 | chunk_kb ? "KB" : "B"); | |
78d742d8 N |
4828 | if (bitmap->file) { |
4829 | seq_printf(seq, ", file: "); | |
4830 | seq_path(seq, bitmap->file->f_vfsmnt, | |
4831 | bitmap->file->f_dentry," \t\n"); | |
32a7627c | 4832 | } |
78d742d8 | 4833 | |
32a7627c N |
4834 | seq_printf(seq, "\n"); |
4835 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
4836 | } |
4837 | ||
4838 | seq_printf(seq, "\n"); | |
4839 | } | |
4840 | mddev_unlock(mddev); | |
4841 | ||
4842 | return 0; | |
4843 | } | |
4844 | ||
4845 | static struct seq_operations md_seq_ops = { | |
4846 | .start = md_seq_start, | |
4847 | .next = md_seq_next, | |
4848 | .stop = md_seq_stop, | |
4849 | .show = md_seq_show, | |
4850 | }; | |
4851 | ||
4852 | static int md_seq_open(struct inode *inode, struct file *file) | |
4853 | { | |
4854 | int error; | |
d7603b7e N |
4855 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
4856 | if (mi == NULL) | |
4857 | return -ENOMEM; | |
1da177e4 LT |
4858 | |
4859 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
4860 | if (error) |
4861 | kfree(mi); | |
4862 | else { | |
4863 | struct seq_file *p = file->private_data; | |
4864 | p->private = mi; | |
4865 | mi->event = atomic_read(&md_event_count); | |
4866 | } | |
1da177e4 LT |
4867 | return error; |
4868 | } | |
4869 | ||
d7603b7e N |
4870 | static int md_seq_release(struct inode *inode, struct file *file) |
4871 | { | |
4872 | struct seq_file *m = file->private_data; | |
4873 | struct mdstat_info *mi = m->private; | |
4874 | m->private = NULL; | |
4875 | kfree(mi); | |
4876 | return seq_release(inode, file); | |
4877 | } | |
4878 | ||
4879 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) | |
4880 | { | |
4881 | struct seq_file *m = filp->private_data; | |
4882 | struct mdstat_info *mi = m->private; | |
4883 | int mask; | |
4884 | ||
4885 | poll_wait(filp, &md_event_waiters, wait); | |
4886 | ||
4887 | /* always allow read */ | |
4888 | mask = POLLIN | POLLRDNORM; | |
4889 | ||
4890 | if (mi->event != atomic_read(&md_event_count)) | |
4891 | mask |= POLLERR | POLLPRI; | |
4892 | return mask; | |
4893 | } | |
4894 | ||
1da177e4 LT |
4895 | static struct file_operations md_seq_fops = { |
4896 | .open = md_seq_open, | |
4897 | .read = seq_read, | |
4898 | .llseek = seq_lseek, | |
d7603b7e N |
4899 | .release = md_seq_release, |
4900 | .poll = mdstat_poll, | |
1da177e4 LT |
4901 | }; |
4902 | ||
2604b703 | 4903 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 4904 | { |
1da177e4 | 4905 | spin_lock(&pers_lock); |
2604b703 N |
4906 | list_add_tail(&p->list, &pers_list); |
4907 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
4908 | spin_unlock(&pers_lock); |
4909 | return 0; | |
4910 | } | |
4911 | ||
2604b703 | 4912 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 4913 | { |
2604b703 | 4914 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 4915 | spin_lock(&pers_lock); |
2604b703 | 4916 | list_del_init(&p->list); |
1da177e4 LT |
4917 | spin_unlock(&pers_lock); |
4918 | return 0; | |
4919 | } | |
4920 | ||
4921 | static int is_mddev_idle(mddev_t *mddev) | |
4922 | { | |
4923 | mdk_rdev_t * rdev; | |
4924 | struct list_head *tmp; | |
4925 | int idle; | |
4926 | unsigned long curr_events; | |
4927 | ||
4928 | idle = 1; | |
4929 | ITERATE_RDEV(mddev,rdev,tmp) { | |
4930 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; | |
a362357b JA |
4931 | curr_events = disk_stat_read(disk, sectors[0]) + |
4932 | disk_stat_read(disk, sectors[1]) - | |
1da177e4 | 4933 | atomic_read(&disk->sync_io); |
c0e48521 N |
4934 | /* The difference between curr_events and last_events |
4935 | * will be affected by any new non-sync IO (making | |
4936 | * curr_events bigger) and any difference in the amount of | |
4937 | * in-flight syncio (making current_events bigger or smaller) | |
4938 | * The amount in-flight is currently limited to | |
4939 | * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6 | |
4940 | * which is at most 4096 sectors. | |
4941 | * These numbers are fairly fragile and should be made | |
4942 | * more robust, probably by enforcing the | |
4943 | * 'window size' that md_do_sync sort-of uses. | |
4944 | * | |
1da177e4 LT |
4945 | * Note: the following is an unsigned comparison. |
4946 | */ | |
c0e48521 | 4947 | if ((curr_events - rdev->last_events + 4096) > 8192) { |
1da177e4 LT |
4948 | rdev->last_events = curr_events; |
4949 | idle = 0; | |
4950 | } | |
4951 | } | |
4952 | return idle; | |
4953 | } | |
4954 | ||
4955 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
4956 | { | |
4957 | /* another "blocks" (512byte) blocks have been synced */ | |
4958 | atomic_sub(blocks, &mddev->recovery_active); | |
4959 | wake_up(&mddev->recovery_wait); | |
4960 | if (!ok) { | |
4961 | set_bit(MD_RECOVERY_ERR, &mddev->recovery); | |
4962 | md_wakeup_thread(mddev->thread); | |
4963 | // stop recovery, signal do_sync .... | |
4964 | } | |
4965 | } | |
4966 | ||
4967 | ||
06d91a5f N |
4968 | /* md_write_start(mddev, bi) |
4969 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
4970 | * in superblock) before writing, schedule a superblock update |
4971 | * and wait for it to complete. | |
06d91a5f | 4972 | */ |
3d310eb7 | 4973 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 4974 | { |
06d91a5f | 4975 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 4976 | return; |
06d91a5f | 4977 | |
f91de92e N |
4978 | BUG_ON(mddev->ro == 1); |
4979 | if (mddev->ro == 2) { | |
4980 | /* need to switch to read/write */ | |
4981 | mddev->ro = 0; | |
4982 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4983 | md_wakeup_thread(mddev->thread); | |
4984 | } | |
06d91a5f | 4985 | atomic_inc(&mddev->writes_pending); |
06d91a5f | 4986 | if (mddev->in_sync) { |
a9701a30 | 4987 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
4988 | if (mddev->in_sync) { |
4989 | mddev->in_sync = 0; | |
42543769 | 4990 | mddev->sb_dirty = 3; |
3d310eb7 N |
4991 | md_wakeup_thread(mddev->thread); |
4992 | } | |
a9701a30 | 4993 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 4994 | } |
3d310eb7 | 4995 | wait_event(mddev->sb_wait, mddev->sb_dirty==0); |
1da177e4 LT |
4996 | } |
4997 | ||
4998 | void md_write_end(mddev_t *mddev) | |
4999 | { | |
5000 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
5001 | if (mddev->safemode == 2) | |
5002 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 5003 | else if (mddev->safemode_delay) |
1da177e4 LT |
5004 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
5005 | } | |
5006 | } | |
5007 | ||
75c96f85 | 5008 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
1da177e4 LT |
5009 | |
5010 | #define SYNC_MARKS 10 | |
5011 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 5012 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
5013 | { |
5014 | mddev_t *mddev2; | |
5015 | unsigned int currspeed = 0, | |
5016 | window; | |
57afd89f | 5017 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
5018 | unsigned long mark[SYNC_MARKS]; |
5019 | sector_t mark_cnt[SYNC_MARKS]; | |
5020 | int last_mark,m; | |
5021 | struct list_head *tmp; | |
5022 | sector_t last_check; | |
57afd89f | 5023 | int skipped = 0; |
5fd6c1dc N |
5024 | struct list_head *rtmp; |
5025 | mdk_rdev_t *rdev; | |
1da177e4 LT |
5026 | |
5027 | /* just incase thread restarts... */ | |
5028 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
5029 | return; | |
5fd6c1dc N |
5030 | if (mddev->ro) /* never try to sync a read-only array */ |
5031 | return; | |
1da177e4 LT |
5032 | |
5033 | /* we overload curr_resync somewhat here. | |
5034 | * 0 == not engaged in resync at all | |
5035 | * 2 == checking that there is no conflict with another sync | |
5036 | * 1 == like 2, but have yielded to allow conflicting resync to | |
5037 | * commense | |
5038 | * other == active in resync - this many blocks | |
5039 | * | |
5040 | * Before starting a resync we must have set curr_resync to | |
5041 | * 2, and then checked that every "conflicting" array has curr_resync | |
5042 | * less than ours. When we find one that is the same or higher | |
5043 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
5044 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
5045 | * This will mean we have to start checking from the beginning again. | |
5046 | * | |
5047 | */ | |
5048 | ||
5049 | do { | |
5050 | mddev->curr_resync = 2; | |
5051 | ||
5052 | try_again: | |
787453c2 | 5053 | if (kthread_should_stop()) { |
6985c43f | 5054 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
5055 | goto skip; |
5056 | } | |
5057 | ITERATE_MDDEV(mddev2,tmp) { | |
1da177e4 LT |
5058 | if (mddev2 == mddev) |
5059 | continue; | |
5060 | if (mddev2->curr_resync && | |
5061 | match_mddev_units(mddev,mddev2)) { | |
5062 | DEFINE_WAIT(wq); | |
5063 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
5064 | /* arbitrarily yield */ | |
5065 | mddev->curr_resync = 1; | |
5066 | wake_up(&resync_wait); | |
5067 | } | |
5068 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
5069 | /* no need to wait here, we can wait the next | |
5070 | * time 'round when curr_resync == 2 | |
5071 | */ | |
5072 | continue; | |
787453c2 N |
5073 | prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE); |
5074 | if (!kthread_should_stop() && | |
8712e553 | 5075 | mddev2->curr_resync >= mddev->curr_resync) { |
1da177e4 LT |
5076 | printk(KERN_INFO "md: delaying resync of %s" |
5077 | " until %s has finished resync (they" | |
5078 | " share one or more physical units)\n", | |
5079 | mdname(mddev), mdname(mddev2)); | |
5080 | mddev_put(mddev2); | |
5081 | schedule(); | |
5082 | finish_wait(&resync_wait, &wq); | |
5083 | goto try_again; | |
5084 | } | |
5085 | finish_wait(&resync_wait, &wq); | |
5086 | } | |
5087 | } | |
5088 | } while (mddev->curr_resync < 2); | |
5089 | ||
5fd6c1dc | 5090 | j = 0; |
9d88883e | 5091 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 5092 | /* resync follows the size requested by the personality, |
57afd89f | 5093 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
5094 | */ |
5095 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 5096 | mddev->resync_mismatches = 0; |
5fd6c1dc N |
5097 | /* we don't use the checkpoint if there's a bitmap */ |
5098 | if (!mddev->bitmap && | |
5099 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
5100 | j = mddev->recovery_cp; | |
ccfcc3c1 N |
5101 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5102 | max_sectors = mddev->size << 1; | |
5fd6c1dc | 5103 | else { |
1da177e4 LT |
5104 | /* recovery follows the physical size of devices */ |
5105 | max_sectors = mddev->size << 1; | |
5fd6c1dc N |
5106 | j = MaxSector; |
5107 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5108 | if (rdev->raid_disk >= 0 && | |
5109 | !test_bit(Faulty, &rdev->flags) && | |
5110 | !test_bit(In_sync, &rdev->flags) && | |
5111 | rdev->recovery_offset < j) | |
5112 | j = rdev->recovery_offset; | |
5113 | } | |
1da177e4 LT |
5114 | |
5115 | printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev)); | |
5116 | printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:" | |
88202a0c | 5117 | " %d KB/sec/disc.\n", speed_min(mddev)); |
338cec32 | 5118 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
1da177e4 | 5119 | "(but not more than %d KB/sec) for reconstruction.\n", |
88202a0c | 5120 | speed_max(mddev)); |
1da177e4 LT |
5121 | |
5122 | is_mddev_idle(mddev); /* this also initializes IO event counters */ | |
5fd6c1dc | 5123 | |
57afd89f | 5124 | io_sectors = 0; |
1da177e4 LT |
5125 | for (m = 0; m < SYNC_MARKS; m++) { |
5126 | mark[m] = jiffies; | |
57afd89f | 5127 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
5128 | } |
5129 | last_mark = 0; | |
5130 | mddev->resync_mark = mark[last_mark]; | |
5131 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
5132 | ||
5133 | /* | |
5134 | * Tune reconstruction: | |
5135 | */ | |
5136 | window = 32*(PAGE_SIZE/512); | |
5137 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
5138 | window/2,(unsigned long long) max_sectors/2); | |
5139 | ||
5140 | atomic_set(&mddev->recovery_active, 0); | |
5141 | init_waitqueue_head(&mddev->recovery_wait); | |
5142 | last_check = 0; | |
5143 | ||
5144 | if (j>2) { | |
5145 | printk(KERN_INFO | |
5146 | "md: resuming recovery of %s from checkpoint.\n", | |
5147 | mdname(mddev)); | |
5148 | mddev->curr_resync = j; | |
5149 | } | |
5150 | ||
5151 | while (j < max_sectors) { | |
57afd89f | 5152 | sector_t sectors; |
1da177e4 | 5153 | |
57afd89f N |
5154 | skipped = 0; |
5155 | sectors = mddev->pers->sync_request(mddev, j, &skipped, | |
88202a0c | 5156 | currspeed < speed_min(mddev)); |
57afd89f | 5157 | if (sectors == 0) { |
1da177e4 LT |
5158 | set_bit(MD_RECOVERY_ERR, &mddev->recovery); |
5159 | goto out; | |
5160 | } | |
57afd89f N |
5161 | |
5162 | if (!skipped) { /* actual IO requested */ | |
5163 | io_sectors += sectors; | |
5164 | atomic_add(sectors, &mddev->recovery_active); | |
5165 | } | |
5166 | ||
1da177e4 LT |
5167 | j += sectors; |
5168 | if (j>1) mddev->curr_resync = j; | |
d7603b7e N |
5169 | if (last_check == 0) |
5170 | /* this is the earliers that rebuilt will be | |
5171 | * visible in /proc/mdstat | |
5172 | */ | |
5173 | md_new_event(mddev); | |
57afd89f N |
5174 | |
5175 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
5176 | continue; |
5177 | ||
57afd89f | 5178 | last_check = io_sectors; |
1da177e4 LT |
5179 | |
5180 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) || | |
5181 | test_bit(MD_RECOVERY_ERR, &mddev->recovery)) | |
5182 | break; | |
5183 | ||
5184 | repeat: | |
5185 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
5186 | /* step marks */ | |
5187 | int next = (last_mark+1) % SYNC_MARKS; | |
5188 | ||
5189 | mddev->resync_mark = mark[next]; | |
5190 | mddev->resync_mark_cnt = mark_cnt[next]; | |
5191 | mark[next] = jiffies; | |
57afd89f | 5192 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
5193 | last_mark = next; |
5194 | } | |
5195 | ||
5196 | ||
787453c2 | 5197 | if (kthread_should_stop()) { |
1da177e4 LT |
5198 | /* |
5199 | * got a signal, exit. | |
5200 | */ | |
5201 | printk(KERN_INFO | |
5202 | "md: md_do_sync() got signal ... exiting\n"); | |
1da177e4 LT |
5203 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
5204 | goto out; | |
5205 | } | |
5206 | ||
5207 | /* | |
5208 | * this loop exits only if either when we are slower than | |
5209 | * the 'hard' speed limit, or the system was IO-idle for | |
5210 | * a jiffy. | |
5211 | * the system might be non-idle CPU-wise, but we only care | |
5212 | * about not overloading the IO subsystem. (things like an | |
5213 | * e2fsck being done on the RAID array should execute fast) | |
5214 | */ | |
5215 | mddev->queue->unplug_fn(mddev->queue); | |
5216 | cond_resched(); | |
5217 | ||
57afd89f N |
5218 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
5219 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 5220 | |
88202a0c N |
5221 | if (currspeed > speed_min(mddev)) { |
5222 | if ((currspeed > speed_max(mddev)) || | |
1da177e4 | 5223 | !is_mddev_idle(mddev)) { |
c0e48521 | 5224 | msleep(500); |
1da177e4 LT |
5225 | goto repeat; |
5226 | } | |
5227 | } | |
5228 | } | |
5229 | printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev)); | |
5230 | /* | |
5231 | * this also signals 'finished resyncing' to md_stop | |
5232 | */ | |
5233 | out: | |
5234 | mddev->queue->unplug_fn(mddev->queue); | |
5235 | ||
5236 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
5237 | ||
5238 | /* tell personality that we are finished */ | |
57afd89f | 5239 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 LT |
5240 | |
5241 | if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && | |
ccfcc3c1 N |
5242 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
5243 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && | |
5fd6c1dc N |
5244 | mddev->curr_resync > 2) { |
5245 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
5246 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5247 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
5248 | printk(KERN_INFO | |
5249 | "md: checkpointing recovery of %s.\n", | |
5250 | mdname(mddev)); | |
5251 | mddev->recovery_cp = mddev->curr_resync; | |
5252 | } | |
5253 | } else | |
5254 | mddev->recovery_cp = MaxSector; | |
5255 | } else { | |
5256 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5257 | mddev->curr_resync = MaxSector; | |
5258 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5259 | if (rdev->raid_disk >= 0 && | |
5260 | !test_bit(Faulty, &rdev->flags) && | |
5261 | !test_bit(In_sync, &rdev->flags) && | |
5262 | rdev->recovery_offset < mddev->curr_resync) | |
5263 | rdev->recovery_offset = mddev->curr_resync; | |
5264 | mddev->sb_dirty = 1; | |
5265 | } | |
1da177e4 LT |
5266 | } |
5267 | ||
1da177e4 LT |
5268 | skip: |
5269 | mddev->curr_resync = 0; | |
5270 | wake_up(&resync_wait); | |
5271 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
5272 | md_wakeup_thread(mddev->thread); | |
5273 | } | |
29269553 | 5274 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
5275 | |
5276 | ||
5277 | /* | |
5278 | * This routine is regularly called by all per-raid-array threads to | |
5279 | * deal with generic issues like resync and super-block update. | |
5280 | * Raid personalities that don't have a thread (linear/raid0) do not | |
5281 | * need this as they never do any recovery or update the superblock. | |
5282 | * | |
5283 | * It does not do any resync itself, but rather "forks" off other threads | |
5284 | * to do that as needed. | |
5285 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
5286 | * "->recovery" and create a thread at ->sync_thread. | |
5287 | * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR) | |
5288 | * and wakeups up this thread which will reap the thread and finish up. | |
5289 | * This thread also removes any faulty devices (with nr_pending == 0). | |
5290 | * | |
5291 | * The overall approach is: | |
5292 | * 1/ if the superblock needs updating, update it. | |
5293 | * 2/ If a recovery thread is running, don't do anything else. | |
5294 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
5295 | * 4/ If there are any faulty devices, remove them. | |
5296 | * 5/ If array is degraded, try to add spares devices | |
5297 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
5298 | */ | |
5299 | void md_check_recovery(mddev_t *mddev) | |
5300 | { | |
5301 | mdk_rdev_t *rdev; | |
5302 | struct list_head *rtmp; | |
5303 | ||
5304 | ||
5f40402d N |
5305 | if (mddev->bitmap) |
5306 | bitmap_daemon_work(mddev->bitmap); | |
1da177e4 LT |
5307 | |
5308 | if (mddev->ro) | |
5309 | return; | |
fca4d848 N |
5310 | |
5311 | if (signal_pending(current)) { | |
5312 | if (mddev->pers->sync_request) { | |
5313 | printk(KERN_INFO "md: %s in immediate safe mode\n", | |
5314 | mdname(mddev)); | |
5315 | mddev->safemode = 2; | |
5316 | } | |
5317 | flush_signals(current); | |
5318 | } | |
5319 | ||
1da177e4 LT |
5320 | if ( ! ( |
5321 | mddev->sb_dirty || | |
5322 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || | |
fca4d848 N |
5323 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
5324 | (mddev->safemode == 1) || | |
5325 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) | |
5326 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
5327 | )) |
5328 | return; | |
fca4d848 | 5329 | |
df5b89b3 | 5330 | if (mddev_trylock(mddev)) { |
1da177e4 | 5331 | int spares =0; |
fca4d848 | 5332 | |
a9701a30 | 5333 | spin_lock_irq(&mddev->write_lock); |
fca4d848 N |
5334 | if (mddev->safemode && !atomic_read(&mddev->writes_pending) && |
5335 | !mddev->in_sync && mddev->recovery_cp == MaxSector) { | |
5336 | mddev->in_sync = 1; | |
42543769 | 5337 | mddev->sb_dirty = 3; |
fca4d848 N |
5338 | } |
5339 | if (mddev->safemode == 1) | |
5340 | mddev->safemode = 0; | |
a9701a30 | 5341 | spin_unlock_irq(&mddev->write_lock); |
fca4d848 | 5342 | |
1da177e4 LT |
5343 | if (mddev->sb_dirty) |
5344 | md_update_sb(mddev); | |
06d91a5f | 5345 | |
06d91a5f | 5346 | |
1da177e4 LT |
5347 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
5348 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
5349 | /* resync/recovery still happening */ | |
5350 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5351 | goto unlock; | |
5352 | } | |
5353 | if (mddev->sync_thread) { | |
5354 | /* resync has finished, collect result */ | |
5355 | md_unregister_thread(mddev->sync_thread); | |
5356 | mddev->sync_thread = NULL; | |
5357 | if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && | |
5358 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5359 | /* success...*/ | |
5360 | /* activate any spares */ | |
5361 | mddev->pers->spare_active(mddev); | |
5362 | } | |
5363 | md_update_sb(mddev); | |
41158c7e N |
5364 | |
5365 | /* if array is no-longer degraded, then any saved_raid_disk | |
5366 | * information must be scrapped | |
5367 | */ | |
5368 | if (!mddev->degraded) | |
5369 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5370 | rdev->saved_raid_disk = -1; | |
5371 | ||
1da177e4 LT |
5372 | mddev->recovery = 0; |
5373 | /* flag recovery needed just to double check */ | |
5374 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
d7603b7e | 5375 | md_new_event(mddev); |
1da177e4 LT |
5376 | goto unlock; |
5377 | } | |
24dd469d N |
5378 | /* Clear some bits that don't mean anything, but |
5379 | * might be left set | |
5380 | */ | |
5381 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5382 | clear_bit(MD_RECOVERY_ERR, &mddev->recovery); | |
5383 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5384 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 5385 | |
5fd6c1dc N |
5386 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
5387 | goto unlock; | |
1da177e4 LT |
5388 | /* no recovery is running. |
5389 | * remove any failed drives, then | |
5390 | * add spares if possible. | |
5391 | * Spare are also removed and re-added, to allow | |
5392 | * the personality to fail the re-add. | |
5393 | */ | |
5394 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5395 | if (rdev->raid_disk >= 0 && | |
b2d444d7 | 5396 | (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) && |
1da177e4 | 5397 | atomic_read(&rdev->nr_pending)==0) { |
86e6ffdd N |
5398 | if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) { |
5399 | char nm[20]; | |
5400 | sprintf(nm,"rd%d", rdev->raid_disk); | |
5401 | sysfs_remove_link(&mddev->kobj, nm); | |
1da177e4 | 5402 | rdev->raid_disk = -1; |
86e6ffdd | 5403 | } |
1da177e4 LT |
5404 | } |
5405 | ||
5406 | if (mddev->degraded) { | |
5407 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5408 | if (rdev->raid_disk < 0 | |
b2d444d7 | 5409 | && !test_bit(Faulty, &rdev->flags)) { |
5fd6c1dc | 5410 | rdev->recovery_offset = 0; |
86e6ffdd N |
5411 | if (mddev->pers->hot_add_disk(mddev,rdev)) { |
5412 | char nm[20]; | |
5413 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5414 | sysfs_create_link(&mddev->kobj, &rdev->kobj, nm); | |
1da177e4 | 5415 | spares++; |
d7603b7e | 5416 | md_new_event(mddev); |
86e6ffdd | 5417 | } else |
1da177e4 LT |
5418 | break; |
5419 | } | |
5420 | } | |
5421 | ||
24dd469d N |
5422 | if (spares) { |
5423 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5424 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5425 | } else if (mddev->recovery_cp < MaxSector) { | |
5426 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5427 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
5428 | /* nothing to be done ... */ | |
1da177e4 | 5429 | goto unlock; |
24dd469d | 5430 | |
1da177e4 LT |
5431 | if (mddev->pers->sync_request) { |
5432 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
a654b9d8 N |
5433 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
5434 | /* We are adding a device or devices to an array | |
5435 | * which has the bitmap stored on all devices. | |
5436 | * So make sure all bitmap pages get written | |
5437 | */ | |
5438 | bitmap_write_all(mddev->bitmap); | |
5439 | } | |
1da177e4 LT |
5440 | mddev->sync_thread = md_register_thread(md_do_sync, |
5441 | mddev, | |
5442 | "%s_resync"); | |
5443 | if (!mddev->sync_thread) { | |
5444 | printk(KERN_ERR "%s: could not start resync" | |
5445 | " thread...\n", | |
5446 | mdname(mddev)); | |
5447 | /* leave the spares where they are, it shouldn't hurt */ | |
5448 | mddev->recovery = 0; | |
d7603b7e | 5449 | } else |
1da177e4 | 5450 | md_wakeup_thread(mddev->sync_thread); |
d7603b7e | 5451 | md_new_event(mddev); |
1da177e4 LT |
5452 | } |
5453 | unlock: | |
5454 | mddev_unlock(mddev); | |
5455 | } | |
5456 | } | |
5457 | ||
75c96f85 AB |
5458 | static int md_notify_reboot(struct notifier_block *this, |
5459 | unsigned long code, void *x) | |
1da177e4 LT |
5460 | { |
5461 | struct list_head *tmp; | |
5462 | mddev_t *mddev; | |
5463 | ||
5464 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
5465 | ||
5466 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
5467 | ||
5468 | ITERATE_MDDEV(mddev,tmp) | |
c71d4887 | 5469 | if (mddev_trylock(mddev)) { |
1da177e4 | 5470 | do_md_stop (mddev, 1); |
c71d4887 NB |
5471 | mddev_unlock(mddev); |
5472 | } | |
1da177e4 LT |
5473 | /* |
5474 | * certain more exotic SCSI devices are known to be | |
5475 | * volatile wrt too early system reboots. While the | |
5476 | * right place to handle this issue is the given | |
5477 | * driver, we do want to have a safe RAID driver ... | |
5478 | */ | |
5479 | mdelay(1000*1); | |
5480 | } | |
5481 | return NOTIFY_DONE; | |
5482 | } | |
5483 | ||
75c96f85 | 5484 | static struct notifier_block md_notifier = { |
1da177e4 LT |
5485 | .notifier_call = md_notify_reboot, |
5486 | .next = NULL, | |
5487 | .priority = INT_MAX, /* before any real devices */ | |
5488 | }; | |
5489 | ||
5490 | static void md_geninit(void) | |
5491 | { | |
5492 | struct proc_dir_entry *p; | |
5493 | ||
5494 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); | |
5495 | ||
5496 | p = create_proc_entry("mdstat", S_IRUGO, NULL); | |
5497 | if (p) | |
5498 | p->proc_fops = &md_seq_fops; | |
5499 | } | |
5500 | ||
75c96f85 | 5501 | static int __init md_init(void) |
1da177e4 LT |
5502 | { |
5503 | int minor; | |
5504 | ||
5505 | printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d," | |
5506 | " MD_SB_DISKS=%d\n", | |
5507 | MD_MAJOR_VERSION, MD_MINOR_VERSION, | |
5508 | MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS); | |
bd926c63 | 5509 | printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI, |
32a7627c | 5510 | BITMAP_MINOR); |
1da177e4 LT |
5511 | |
5512 | if (register_blkdev(MAJOR_NR, "md")) | |
5513 | return -1; | |
5514 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
5515 | unregister_blkdev(MAJOR_NR, "md"); | |
5516 | return -1; | |
5517 | } | |
5518 | devfs_mk_dir("md"); | |
5519 | blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE, | |
5520 | md_probe, NULL, NULL); | |
5521 | blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE, | |
5522 | md_probe, NULL, NULL); | |
5523 | ||
5524 | for (minor=0; minor < MAX_MD_DEVS; ++minor) | |
5525 | devfs_mk_bdev(MKDEV(MAJOR_NR, minor), | |
5526 | S_IFBLK|S_IRUSR|S_IWUSR, | |
5527 | "md/%d", minor); | |
5528 | ||
5529 | for (minor=0; minor < MAX_MD_DEVS; ++minor) | |
5530 | devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift), | |
5531 | S_IFBLK|S_IRUSR|S_IWUSR, | |
5532 | "md/mdp%d", minor); | |
5533 | ||
5534 | ||
5535 | register_reboot_notifier(&md_notifier); | |
5536 | raid_table_header = register_sysctl_table(raid_root_table, 1); | |
5537 | ||
5538 | md_geninit(); | |
5539 | return (0); | |
5540 | } | |
5541 | ||
5542 | ||
5543 | #ifndef MODULE | |
5544 | ||
5545 | /* | |
5546 | * Searches all registered partitions for autorun RAID arrays | |
5547 | * at boot time. | |
5548 | */ | |
5549 | static dev_t detected_devices[128]; | |
5550 | static int dev_cnt; | |
5551 | ||
5552 | void md_autodetect_dev(dev_t dev) | |
5553 | { | |
5554 | if (dev_cnt >= 0 && dev_cnt < 127) | |
5555 | detected_devices[dev_cnt++] = dev; | |
5556 | } | |
5557 | ||
5558 | ||
5559 | static void autostart_arrays(int part) | |
5560 | { | |
5561 | mdk_rdev_t *rdev; | |
5562 | int i; | |
5563 | ||
5564 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); | |
5565 | ||
5566 | for (i = 0; i < dev_cnt; i++) { | |
5567 | dev_t dev = detected_devices[i]; | |
5568 | ||
5569 | rdev = md_import_device(dev,0, 0); | |
5570 | if (IS_ERR(rdev)) | |
5571 | continue; | |
5572 | ||
b2d444d7 | 5573 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5574 | MD_BUG(); |
5575 | continue; | |
5576 | } | |
5577 | list_add(&rdev->same_set, &pending_raid_disks); | |
5578 | } | |
5579 | dev_cnt = 0; | |
5580 | ||
5581 | autorun_devices(part); | |
5582 | } | |
5583 | ||
5584 | #endif | |
5585 | ||
5586 | static __exit void md_exit(void) | |
5587 | { | |
5588 | mddev_t *mddev; | |
5589 | struct list_head *tmp; | |
5590 | int i; | |
5591 | blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS); | |
5592 | blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift); | |
5593 | for (i=0; i < MAX_MD_DEVS; i++) | |
5594 | devfs_remove("md/%d", i); | |
5595 | for (i=0; i < MAX_MD_DEVS; i++) | |
5596 | devfs_remove("md/d%d", i); | |
5597 | ||
5598 | devfs_remove("md"); | |
5599 | ||
5600 | unregister_blkdev(MAJOR_NR,"md"); | |
5601 | unregister_blkdev(mdp_major, "mdp"); | |
5602 | unregister_reboot_notifier(&md_notifier); | |
5603 | unregister_sysctl_table(raid_table_header); | |
5604 | remove_proc_entry("mdstat", NULL); | |
5605 | ITERATE_MDDEV(mddev,tmp) { | |
5606 | struct gendisk *disk = mddev->gendisk; | |
5607 | if (!disk) | |
5608 | continue; | |
5609 | export_array(mddev); | |
5610 | del_gendisk(disk); | |
5611 | put_disk(disk); | |
5612 | mddev->gendisk = NULL; | |
5613 | mddev_put(mddev); | |
5614 | } | |
5615 | } | |
5616 | ||
5617 | module_init(md_init) | |
5618 | module_exit(md_exit) | |
5619 | ||
f91de92e N |
5620 | static int get_ro(char *buffer, struct kernel_param *kp) |
5621 | { | |
5622 | return sprintf(buffer, "%d", start_readonly); | |
5623 | } | |
5624 | static int set_ro(const char *val, struct kernel_param *kp) | |
5625 | { | |
5626 | char *e; | |
5627 | int num = simple_strtoul(val, &e, 10); | |
5628 | if (*val && (*e == '\0' || *e == '\n')) { | |
5629 | start_readonly = num; | |
4dbcdc75 | 5630 | return 0; |
f91de92e N |
5631 | } |
5632 | return -EINVAL; | |
5633 | } | |
5634 | ||
5635 | module_param_call(start_ro, set_ro, get_ro, NULL, 0600); | |
6ff8d8ec N |
5636 | module_param(start_dirty_degraded, int, 0644); |
5637 | ||
f91de92e | 5638 | |
1da177e4 LT |
5639 | EXPORT_SYMBOL(register_md_personality); |
5640 | EXPORT_SYMBOL(unregister_md_personality); | |
5641 | EXPORT_SYMBOL(md_error); | |
5642 | EXPORT_SYMBOL(md_done_sync); | |
5643 | EXPORT_SYMBOL(md_write_start); | |
5644 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
5645 | EXPORT_SYMBOL(md_register_thread); |
5646 | EXPORT_SYMBOL(md_unregister_thread); | |
5647 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
5648 | EXPORT_SYMBOL(md_check_recovery); |
5649 | MODULE_LICENSE("GPL"); | |
aa1595e9 | 5650 | MODULE_ALIAS("md"); |
72008652 | 5651 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |