]>
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> | |
a6fb0934 | 36 | #include <linux/kthread.h> |
1da177e4 LT |
37 | #include <linux/linkage.h> |
38 | #include <linux/raid/md.h> | |
32a7627c | 39 | #include <linux/raid/bitmap.h> |
1da177e4 | 40 | #include <linux/sysctl.h> |
1da177e4 LT |
41 | #include <linux/buffer_head.h> /* for invalidate_bdev */ |
42 | #include <linux/suspend.h> | |
d7603b7e | 43 | #include <linux/poll.h> |
48c9c27b | 44 | #include <linux/mutex.h> |
16f17b39 | 45 | #include <linux/ctype.h> |
1da177e4 LT |
46 | |
47 | #include <linux/init.h> | |
48 | ||
32a7627c N |
49 | #include <linux/file.h> |
50 | ||
1da177e4 LT |
51 | #ifdef CONFIG_KMOD |
52 | #include <linux/kmod.h> | |
53 | #endif | |
54 | ||
55 | #include <asm/unaligned.h> | |
56 | ||
57 | #define MAJOR_NR MD_MAJOR | |
58 | #define MD_DRIVER | |
59 | ||
60 | /* 63 partitions with the alternate major number (mdp) */ | |
61 | #define MdpMinorShift 6 | |
62 | ||
63 | #define DEBUG 0 | |
64 | #define dprintk(x...) ((void)(DEBUG && printk(x))) | |
65 | ||
66 | ||
67 | #ifndef MODULE | |
68 | static void autostart_arrays (int part); | |
69 | #endif | |
70 | ||
2604b703 | 71 | static LIST_HEAD(pers_list); |
1da177e4 LT |
72 | static DEFINE_SPINLOCK(pers_lock); |
73 | ||
5e56341d AB |
74 | static void md_print_devices(void); |
75 | ||
76 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } | |
77 | ||
1da177e4 LT |
78 | /* |
79 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
80 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
81 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 82 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
83 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
84 | * speed limit - in case reconstruction slows down your system despite | |
85 | * idle IO detection. | |
86 | * | |
87 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 88 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
89 | */ |
90 | ||
91 | static int sysctl_speed_limit_min = 1000; | |
92 | static int sysctl_speed_limit_max = 200000; | |
88202a0c N |
93 | static inline int speed_min(mddev_t *mddev) |
94 | { | |
95 | return mddev->sync_speed_min ? | |
96 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
97 | } | |
98 | ||
99 | static inline int speed_max(mddev_t *mddev) | |
100 | { | |
101 | return mddev->sync_speed_max ? | |
102 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
103 | } | |
1da177e4 LT |
104 | |
105 | static struct ctl_table_header *raid_table_header; | |
106 | ||
107 | static ctl_table raid_table[] = { | |
108 | { | |
109 | .ctl_name = DEV_RAID_SPEED_LIMIT_MIN, | |
110 | .procname = "speed_limit_min", | |
111 | .data = &sysctl_speed_limit_min, | |
112 | .maxlen = sizeof(int), | |
113 | .mode = 0644, | |
114 | .proc_handler = &proc_dointvec, | |
115 | }, | |
116 | { | |
117 | .ctl_name = DEV_RAID_SPEED_LIMIT_MAX, | |
118 | .procname = "speed_limit_max", | |
119 | .data = &sysctl_speed_limit_max, | |
120 | .maxlen = sizeof(int), | |
121 | .mode = 0644, | |
122 | .proc_handler = &proc_dointvec, | |
123 | }, | |
124 | { .ctl_name = 0 } | |
125 | }; | |
126 | ||
127 | static ctl_table raid_dir_table[] = { | |
128 | { | |
129 | .ctl_name = DEV_RAID, | |
130 | .procname = "raid", | |
131 | .maxlen = 0, | |
132 | .mode = 0555, | |
133 | .child = raid_table, | |
134 | }, | |
135 | { .ctl_name = 0 } | |
136 | }; | |
137 | ||
138 | static ctl_table raid_root_table[] = { | |
139 | { | |
140 | .ctl_name = CTL_DEV, | |
141 | .procname = "dev", | |
142 | .maxlen = 0, | |
143 | .mode = 0555, | |
144 | .child = raid_dir_table, | |
145 | }, | |
146 | { .ctl_name = 0 } | |
147 | }; | |
148 | ||
149 | static struct block_device_operations md_fops; | |
150 | ||
f91de92e N |
151 | static int start_readonly; |
152 | ||
d7603b7e N |
153 | /* |
154 | * We have a system wide 'event count' that is incremented | |
155 | * on any 'interesting' event, and readers of /proc/mdstat | |
156 | * can use 'poll' or 'select' to find out when the event | |
157 | * count increases. | |
158 | * | |
159 | * Events are: | |
160 | * start array, stop array, error, add device, remove device, | |
161 | * start build, activate spare | |
162 | */ | |
2989ddbd | 163 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 164 | static atomic_t md_event_count; |
29269553 | 165 | void md_new_event(mddev_t *mddev) |
d7603b7e N |
166 | { |
167 | atomic_inc(&md_event_count); | |
168 | wake_up(&md_event_waiters); | |
4508a7a7 | 169 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
d7603b7e | 170 | } |
29269553 | 171 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 172 | |
c331eb04 N |
173 | /* Alternate version that can be called from interrupts |
174 | * when calling sysfs_notify isn't needed. | |
175 | */ | |
05381954 | 176 | static void md_new_event_inintr(mddev_t *mddev) |
c331eb04 N |
177 | { |
178 | atomic_inc(&md_event_count); | |
179 | wake_up(&md_event_waiters); | |
180 | } | |
181 | ||
1da177e4 LT |
182 | /* |
183 | * Enables to iterate over all existing md arrays | |
184 | * all_mddevs_lock protects this list. | |
185 | */ | |
186 | static LIST_HEAD(all_mddevs); | |
187 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
188 | ||
189 | ||
190 | /* | |
191 | * iterates through all used mddevs in the system. | |
192 | * We take care to grab the all_mddevs_lock whenever navigating | |
193 | * the list, and to always hold a refcount when unlocked. | |
194 | * Any code which breaks out of this loop while own | |
195 | * a reference to the current mddev and must mddev_put it. | |
196 | */ | |
197 | #define ITERATE_MDDEV(mddev,tmp) \ | |
198 | \ | |
199 | for (({ spin_lock(&all_mddevs_lock); \ | |
200 | tmp = all_mddevs.next; \ | |
201 | mddev = NULL;}); \ | |
202 | ({ if (tmp != &all_mddevs) \ | |
203 | mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ | |
204 | spin_unlock(&all_mddevs_lock); \ | |
205 | if (mddev) mddev_put(mddev); \ | |
206 | mddev = list_entry(tmp, mddev_t, all_mddevs); \ | |
207 | tmp != &all_mddevs;}); \ | |
208 | ({ spin_lock(&all_mddevs_lock); \ | |
209 | tmp = tmp->next;}) \ | |
210 | ) | |
211 | ||
212 | ||
213 | static int md_fail_request (request_queue_t *q, struct bio *bio) | |
214 | { | |
215 | bio_io_error(bio, bio->bi_size); | |
216 | return 0; | |
217 | } | |
218 | ||
219 | static inline mddev_t *mddev_get(mddev_t *mddev) | |
220 | { | |
221 | atomic_inc(&mddev->active); | |
222 | return mddev; | |
223 | } | |
224 | ||
225 | static void mddev_put(mddev_t *mddev) | |
226 | { | |
227 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
228 | return; | |
229 | if (!mddev->raid_disks && list_empty(&mddev->disks)) { | |
230 | list_del(&mddev->all_mddevs); | |
926ce2d8 | 231 | spin_unlock(&all_mddevs_lock); |
1312f40e | 232 | blk_cleanup_queue(mddev->queue); |
eae1701f | 233 | kobject_unregister(&mddev->kobj); |
926ce2d8 N |
234 | } else |
235 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
236 | } |
237 | ||
238 | static mddev_t * mddev_find(dev_t unit) | |
239 | { | |
240 | mddev_t *mddev, *new = NULL; | |
241 | ||
242 | retry: | |
243 | spin_lock(&all_mddevs_lock); | |
244 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
245 | if (mddev->unit == unit) { | |
246 | mddev_get(mddev); | |
247 | spin_unlock(&all_mddevs_lock); | |
990a8baf | 248 | kfree(new); |
1da177e4 LT |
249 | return mddev; |
250 | } | |
251 | ||
252 | if (new) { | |
253 | list_add(&new->all_mddevs, &all_mddevs); | |
254 | spin_unlock(&all_mddevs_lock); | |
255 | return new; | |
256 | } | |
257 | spin_unlock(&all_mddevs_lock); | |
258 | ||
9ffae0cf | 259 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
260 | if (!new) |
261 | return NULL; | |
262 | ||
1da177e4 LT |
263 | new->unit = unit; |
264 | if (MAJOR(unit) == MD_MAJOR) | |
265 | new->md_minor = MINOR(unit); | |
266 | else | |
267 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
268 | ||
df5b89b3 | 269 | mutex_init(&new->reconfig_mutex); |
1da177e4 LT |
270 | INIT_LIST_HEAD(&new->disks); |
271 | INIT_LIST_HEAD(&new->all_mddevs); | |
272 | init_timer(&new->safemode_timer); | |
273 | atomic_set(&new->active, 1); | |
06d91a5f | 274 | spin_lock_init(&new->write_lock); |
3d310eb7 | 275 | init_waitqueue_head(&new->sb_wait); |
1da177e4 LT |
276 | |
277 | new->queue = blk_alloc_queue(GFP_KERNEL); | |
278 | if (!new->queue) { | |
279 | kfree(new); | |
280 | return NULL; | |
281 | } | |
89e5c8b5 | 282 | set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags); |
1da177e4 LT |
283 | |
284 | blk_queue_make_request(new->queue, md_fail_request); | |
285 | ||
286 | goto retry; | |
287 | } | |
288 | ||
289 | static inline int mddev_lock(mddev_t * mddev) | |
290 | { | |
df5b89b3 | 291 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
292 | } |
293 | ||
1da177e4 LT |
294 | static inline int mddev_trylock(mddev_t * mddev) |
295 | { | |
df5b89b3 | 296 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
297 | } |
298 | ||
299 | static inline void mddev_unlock(mddev_t * mddev) | |
300 | { | |
df5b89b3 | 301 | mutex_unlock(&mddev->reconfig_mutex); |
1da177e4 | 302 | |
005eca5e | 303 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
304 | } |
305 | ||
2989ddbd | 306 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 LT |
307 | { |
308 | mdk_rdev_t * rdev; | |
309 | struct list_head *tmp; | |
310 | ||
311 | ITERATE_RDEV(mddev,rdev,tmp) { | |
312 | if (rdev->desc_nr == nr) | |
313 | return rdev; | |
314 | } | |
315 | return NULL; | |
316 | } | |
317 | ||
318 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
319 | { | |
320 | struct list_head *tmp; | |
321 | mdk_rdev_t *rdev; | |
322 | ||
323 | ITERATE_RDEV(mddev,rdev,tmp) { | |
324 | if (rdev->bdev->bd_dev == dev) | |
325 | return rdev; | |
326 | } | |
327 | return NULL; | |
328 | } | |
329 | ||
d9d166c2 | 330 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
331 | { |
332 | struct mdk_personality *pers; | |
d9d166c2 N |
333 | list_for_each_entry(pers, &pers_list, list) { |
334 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 335 | return pers; |
d9d166c2 N |
336 | if (strcmp(pers->name, clevel)==0) |
337 | return pers; | |
338 | } | |
2604b703 N |
339 | return NULL; |
340 | } | |
341 | ||
77933d72 | 342 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 LT |
343 | { |
344 | sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
345 | return MD_NEW_SIZE_BLOCKS(size); | |
346 | } | |
347 | ||
348 | static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size) | |
349 | { | |
350 | sector_t size; | |
351 | ||
352 | size = rdev->sb_offset; | |
353 | ||
354 | if (chunk_size) | |
355 | size &= ~((sector_t)chunk_size/1024 - 1); | |
356 | return size; | |
357 | } | |
358 | ||
359 | static int alloc_disk_sb(mdk_rdev_t * rdev) | |
360 | { | |
361 | if (rdev->sb_page) | |
362 | MD_BUG(); | |
363 | ||
364 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
365 | if (!rdev->sb_page) { | |
366 | printk(KERN_ALERT "md: out of memory.\n"); | |
367 | return -EINVAL; | |
368 | } | |
369 | ||
370 | return 0; | |
371 | } | |
372 | ||
373 | static void free_disk_sb(mdk_rdev_t * rdev) | |
374 | { | |
375 | if (rdev->sb_page) { | |
2d1f3b5d | 376 | put_page(rdev->sb_page); |
1da177e4 LT |
377 | rdev->sb_loaded = 0; |
378 | rdev->sb_page = NULL; | |
379 | rdev->sb_offset = 0; | |
380 | rdev->size = 0; | |
381 | } | |
382 | } | |
383 | ||
384 | ||
7bfa19f2 N |
385 | static int super_written(struct bio *bio, unsigned int bytes_done, int error) |
386 | { | |
387 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 388 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 N |
389 | if (bio->bi_size) |
390 | return 1; | |
391 | ||
392 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
a9701a30 | 393 | md_error(mddev, rdev); |
7bfa19f2 | 394 | |
a9701a30 N |
395 | if (atomic_dec_and_test(&mddev->pending_writes)) |
396 | wake_up(&mddev->sb_wait); | |
f8b58edf | 397 | bio_put(bio); |
7bfa19f2 N |
398 | return 0; |
399 | } | |
400 | ||
a9701a30 N |
401 | static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error) |
402 | { | |
403 | struct bio *bio2 = bio->bi_private; | |
404 | mdk_rdev_t *rdev = bio2->bi_private; | |
405 | mddev_t *mddev = rdev->mddev; | |
406 | if (bio->bi_size) | |
407 | return 1; | |
408 | ||
409 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
410 | error == -EOPNOTSUPP) { | |
411 | unsigned long flags; | |
412 | /* barriers don't appear to be supported :-( */ | |
413 | set_bit(BarriersNotsupp, &rdev->flags); | |
414 | mddev->barriers_work = 0; | |
415 | spin_lock_irqsave(&mddev->write_lock, flags); | |
416 | bio2->bi_next = mddev->biolist; | |
417 | mddev->biolist = bio2; | |
418 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
419 | wake_up(&mddev->sb_wait); | |
420 | bio_put(bio); | |
421 | return 0; | |
422 | } | |
423 | bio_put(bio2); | |
424 | bio->bi_private = rdev; | |
425 | return super_written(bio, bytes_done, error); | |
426 | } | |
427 | ||
7bfa19f2 N |
428 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
429 | sector_t sector, int size, struct page *page) | |
430 | { | |
431 | /* write first size bytes of page to sector of rdev | |
432 | * Increment mddev->pending_writes before returning | |
433 | * and decrement it on completion, waking up sb_wait | |
434 | * if zero is reached. | |
435 | * If an error occurred, call md_error | |
a9701a30 N |
436 | * |
437 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
438 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
439 | */ |
440 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
a9701a30 | 441 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC); |
7bfa19f2 N |
442 | |
443 | bio->bi_bdev = rdev->bdev; | |
444 | bio->bi_sector = sector; | |
445 | bio_add_page(bio, page, size, 0); | |
446 | bio->bi_private = rdev; | |
447 | bio->bi_end_io = super_written; | |
a9701a30 N |
448 | bio->bi_rw = rw; |
449 | ||
7bfa19f2 | 450 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
451 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
452 | struct bio *rbio; | |
453 | rw |= (1<<BIO_RW_BARRIER); | |
454 | rbio = bio_clone(bio, GFP_NOIO); | |
455 | rbio->bi_private = bio; | |
456 | rbio->bi_end_io = super_written_barrier; | |
457 | submit_bio(rw, rbio); | |
458 | } else | |
459 | submit_bio(rw, bio); | |
460 | } | |
461 | ||
462 | void md_super_wait(mddev_t *mddev) | |
463 | { | |
464 | /* wait for all superblock writes that were scheduled to complete. | |
465 | * if any had to be retried (due to BARRIER problems), retry them | |
466 | */ | |
467 | DEFINE_WAIT(wq); | |
468 | for(;;) { | |
469 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
470 | if (atomic_read(&mddev->pending_writes)==0) | |
471 | break; | |
472 | while (mddev->biolist) { | |
473 | struct bio *bio; | |
474 | spin_lock_irq(&mddev->write_lock); | |
475 | bio = mddev->biolist; | |
476 | mddev->biolist = bio->bi_next ; | |
477 | bio->bi_next = NULL; | |
478 | spin_unlock_irq(&mddev->write_lock); | |
479 | submit_bio(bio->bi_rw, bio); | |
480 | } | |
481 | schedule(); | |
482 | } | |
483 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
484 | } |
485 | ||
1da177e4 LT |
486 | static int bi_complete(struct bio *bio, unsigned int bytes_done, int error) |
487 | { | |
488 | if (bio->bi_size) | |
489 | return 1; | |
490 | ||
491 | complete((struct completion*)bio->bi_private); | |
492 | return 0; | |
493 | } | |
494 | ||
a654b9d8 | 495 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
496 | struct page *page, int rw) |
497 | { | |
baaa2c51 | 498 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
499 | struct completion event; |
500 | int ret; | |
501 | ||
502 | rw |= (1 << BIO_RW_SYNC); | |
503 | ||
504 | bio->bi_bdev = bdev; | |
505 | bio->bi_sector = sector; | |
506 | bio_add_page(bio, page, size, 0); | |
507 | init_completion(&event); | |
508 | bio->bi_private = &event; | |
509 | bio->bi_end_io = bi_complete; | |
510 | submit_bio(rw, bio); | |
511 | wait_for_completion(&event); | |
512 | ||
513 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
514 | bio_put(bio); | |
515 | return ret; | |
516 | } | |
a8745db2 | 517 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 518 | |
0002b271 | 519 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
520 | { |
521 | char b[BDEVNAME_SIZE]; | |
522 | if (!rdev->sb_page) { | |
523 | MD_BUG(); | |
524 | return -EINVAL; | |
525 | } | |
526 | if (rdev->sb_loaded) | |
527 | return 0; | |
528 | ||
529 | ||
0002b271 | 530 | if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ)) |
1da177e4 LT |
531 | goto fail; |
532 | rdev->sb_loaded = 1; | |
533 | return 0; | |
534 | ||
535 | fail: | |
536 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
537 | bdevname(rdev->bdev,b)); | |
538 | return -EINVAL; | |
539 | } | |
540 | ||
541 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
542 | { | |
543 | if ( (sb1->set_uuid0 == sb2->set_uuid0) && | |
544 | (sb1->set_uuid1 == sb2->set_uuid1) && | |
545 | (sb1->set_uuid2 == sb2->set_uuid2) && | |
546 | (sb1->set_uuid3 == sb2->set_uuid3)) | |
547 | ||
548 | return 1; | |
549 | ||
550 | return 0; | |
551 | } | |
552 | ||
553 | ||
554 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
555 | { | |
556 | int ret; | |
557 | mdp_super_t *tmp1, *tmp2; | |
558 | ||
559 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
560 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
561 | ||
562 | if (!tmp1 || !tmp2) { | |
563 | ret = 0; | |
564 | printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n"); | |
565 | goto abort; | |
566 | } | |
567 | ||
568 | *tmp1 = *sb1; | |
569 | *tmp2 = *sb2; | |
570 | ||
571 | /* | |
572 | * nr_disks is not constant | |
573 | */ | |
574 | tmp1->nr_disks = 0; | |
575 | tmp2->nr_disks = 0; | |
576 | ||
577 | if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4)) | |
578 | ret = 0; | |
579 | else | |
580 | ret = 1; | |
581 | ||
582 | abort: | |
990a8baf JJ |
583 | kfree(tmp1); |
584 | kfree(tmp2); | |
1da177e4 LT |
585 | return ret; |
586 | } | |
587 | ||
588 | static unsigned int calc_sb_csum(mdp_super_t * sb) | |
589 | { | |
590 | unsigned int disk_csum, csum; | |
591 | ||
592 | disk_csum = sb->sb_csum; | |
593 | sb->sb_csum = 0; | |
594 | csum = csum_partial((void *)sb, MD_SB_BYTES, 0); | |
595 | sb->sb_csum = disk_csum; | |
596 | return csum; | |
597 | } | |
598 | ||
599 | ||
600 | /* | |
601 | * Handle superblock details. | |
602 | * We want to be able to handle multiple superblock formats | |
603 | * so we have a common interface to them all, and an array of | |
604 | * different handlers. | |
605 | * We rely on user-space to write the initial superblock, and support | |
606 | * reading and updating of superblocks. | |
607 | * Interface methods are: | |
608 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
609 | * loads and validates a superblock on dev. | |
610 | * if refdev != NULL, compare superblocks on both devices | |
611 | * Return: | |
612 | * 0 - dev has a superblock that is compatible with refdev | |
613 | * 1 - dev has a superblock that is compatible and newer than refdev | |
614 | * so dev should be used as the refdev in future | |
615 | * -EINVAL superblock incompatible or invalid | |
616 | * -othererror e.g. -EIO | |
617 | * | |
618 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
619 | * Verify that dev is acceptable into mddev. | |
620 | * The first time, mddev->raid_disks will be 0, and data from | |
621 | * dev should be merged in. Subsequent calls check that dev | |
622 | * is new enough. Return 0 or -EINVAL | |
623 | * | |
624 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
625 | * Update the superblock for rdev with data in mddev | |
626 | * This does not write to disc. | |
627 | * | |
628 | */ | |
629 | ||
630 | struct super_type { | |
631 | char *name; | |
632 | struct module *owner; | |
633 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version); | |
634 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
635 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
636 | }; | |
637 | ||
638 | /* | |
639 | * load_super for 0.90.0 | |
640 | */ | |
641 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
642 | { | |
643 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
644 | mdp_super_t *sb; | |
645 | int ret; | |
646 | sector_t sb_offset; | |
647 | ||
648 | /* | |
649 | * Calculate the position of the superblock, | |
650 | * it's at the end of the disk. | |
651 | * | |
652 | * It also happens to be a multiple of 4Kb. | |
653 | */ | |
654 | sb_offset = calc_dev_sboffset(rdev->bdev); | |
655 | rdev->sb_offset = sb_offset; | |
656 | ||
0002b271 | 657 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
658 | if (ret) return ret; |
659 | ||
660 | ret = -EINVAL; | |
661 | ||
662 | bdevname(rdev->bdev, b); | |
663 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
664 | ||
665 | if (sb->md_magic != MD_SB_MAGIC) { | |
666 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
667 | b); | |
668 | goto abort; | |
669 | } | |
670 | ||
671 | if (sb->major_version != 0 || | |
f6705578 N |
672 | sb->minor_version < 90 || |
673 | sb->minor_version > 91) { | |
1da177e4 LT |
674 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
675 | sb->major_version, sb->minor_version, | |
676 | b); | |
677 | goto abort; | |
678 | } | |
679 | ||
680 | if (sb->raid_disks <= 0) | |
681 | goto abort; | |
682 | ||
683 | if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) { | |
684 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", | |
685 | b); | |
686 | goto abort; | |
687 | } | |
688 | ||
689 | rdev->preferred_minor = sb->md_minor; | |
690 | rdev->data_offset = 0; | |
0002b271 | 691 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
692 | |
693 | if (sb->level == LEVEL_MULTIPATH) | |
694 | rdev->desc_nr = -1; | |
695 | else | |
696 | rdev->desc_nr = sb->this_disk.number; | |
697 | ||
698 | if (refdev == 0) | |
699 | ret = 1; | |
700 | else { | |
701 | __u64 ev1, ev2; | |
702 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
703 | if (!uuid_equal(refsb, sb)) { | |
704 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
705 | b, bdevname(refdev->bdev,b2)); | |
706 | goto abort; | |
707 | } | |
708 | if (!sb_equal(refsb, sb)) { | |
709 | printk(KERN_WARNING "md: %s has same UUID" | |
710 | " but different superblock to %s\n", | |
711 | b, bdevname(refdev->bdev, b2)); | |
712 | goto abort; | |
713 | } | |
714 | ev1 = md_event(sb); | |
715 | ev2 = md_event(refsb); | |
716 | if (ev1 > ev2) | |
717 | ret = 1; | |
718 | else | |
719 | ret = 0; | |
720 | } | |
721 | rdev->size = calc_dev_size(rdev, sb->chunk_size); | |
722 | ||
2bf071bf N |
723 | if (rdev->size < sb->size && sb->level > 1) |
724 | /* "this cannot possibly happen" ... */ | |
725 | ret = -EINVAL; | |
726 | ||
1da177e4 LT |
727 | abort: |
728 | return ret; | |
729 | } | |
730 | ||
731 | /* | |
732 | * validate_super for 0.90.0 | |
733 | */ | |
734 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
735 | { | |
736 | mdp_disk_t *desc; | |
737 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 738 | __u64 ev1 = md_event(sb); |
1da177e4 | 739 | |
41158c7e | 740 | rdev->raid_disk = -1; |
b2d444d7 | 741 | rdev->flags = 0; |
1da177e4 LT |
742 | if (mddev->raid_disks == 0) { |
743 | mddev->major_version = 0; | |
744 | mddev->minor_version = sb->minor_version; | |
745 | mddev->patch_version = sb->patch_version; | |
746 | mddev->persistent = ! sb->not_persistent; | |
747 | mddev->chunk_size = sb->chunk_size; | |
748 | mddev->ctime = sb->ctime; | |
749 | mddev->utime = sb->utime; | |
750 | mddev->level = sb->level; | |
d9d166c2 | 751 | mddev->clevel[0] = 0; |
1da177e4 LT |
752 | mddev->layout = sb->layout; |
753 | mddev->raid_disks = sb->raid_disks; | |
754 | mddev->size = sb->size; | |
07d84d10 | 755 | mddev->events = ev1; |
9223214e | 756 | mddev->bitmap_offset = 0; |
36fa3063 | 757 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
1da177e4 | 758 | |
f6705578 N |
759 | if (mddev->minor_version >= 91) { |
760 | mddev->reshape_position = sb->reshape_position; | |
761 | mddev->delta_disks = sb->delta_disks; | |
762 | mddev->new_level = sb->new_level; | |
763 | mddev->new_layout = sb->new_layout; | |
764 | mddev->new_chunk = sb->new_chunk; | |
765 | } else { | |
766 | mddev->reshape_position = MaxSector; | |
767 | mddev->delta_disks = 0; | |
768 | mddev->new_level = mddev->level; | |
769 | mddev->new_layout = mddev->layout; | |
770 | mddev->new_chunk = mddev->chunk_size; | |
771 | } | |
772 | ||
1da177e4 LT |
773 | if (sb->state & (1<<MD_SB_CLEAN)) |
774 | mddev->recovery_cp = MaxSector; | |
775 | else { | |
776 | if (sb->events_hi == sb->cp_events_hi && | |
777 | sb->events_lo == sb->cp_events_lo) { | |
778 | mddev->recovery_cp = sb->recovery_cp; | |
779 | } else | |
780 | mddev->recovery_cp = 0; | |
781 | } | |
782 | ||
783 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
784 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
785 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
786 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
787 | ||
788 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
789 | |
790 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
791 | mddev->bitmap_file == NULL) { | |
c5a10f62 N |
792 | if (mddev->level != 1 && mddev->level != 4 |
793 | && mddev->level != 5 && mddev->level != 6 | |
6cce3b23 | 794 | && mddev->level != 10) { |
a654b9d8 | 795 | /* FIXME use a better test */ |
6cce3b23 | 796 | printk(KERN_WARNING "md: bitmaps not supported for this level.\n"); |
a654b9d8 N |
797 | return -EINVAL; |
798 | } | |
36fa3063 | 799 | mddev->bitmap_offset = mddev->default_bitmap_offset; |
a654b9d8 N |
800 | } |
801 | ||
41158c7e N |
802 | } else if (mddev->pers == NULL) { |
803 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
804 | ++ev1; |
805 | if (ev1 < mddev->events) | |
806 | return -EINVAL; | |
41158c7e N |
807 | } else if (mddev->bitmap) { |
808 | /* if adding to array with a bitmap, then we can accept an | |
809 | * older device ... but not too old. | |
810 | */ | |
41158c7e N |
811 | if (ev1 < mddev->bitmap->events_cleared) |
812 | return 0; | |
07d84d10 N |
813 | } else { |
814 | if (ev1 < mddev->events) | |
815 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
816 | return 0; | |
817 | } | |
41158c7e | 818 | |
1da177e4 | 819 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
820 | desc = sb->disks + rdev->desc_nr; |
821 | ||
822 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 823 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
824 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
825 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 826 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
827 | rdev->raid_disk = desc->raid_disk; |
828 | } | |
8ddf9efe N |
829 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
830 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 831 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 832 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
833 | return 0; |
834 | } | |
835 | ||
836 | /* | |
837 | * sync_super for 0.90.0 | |
838 | */ | |
839 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
840 | { | |
841 | mdp_super_t *sb; | |
842 | struct list_head *tmp; | |
843 | mdk_rdev_t *rdev2; | |
844 | int next_spare = mddev->raid_disks; | |
19133a42 | 845 | |
1da177e4 LT |
846 | |
847 | /* make rdev->sb match mddev data.. | |
848 | * | |
849 | * 1/ zero out disks | |
850 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
851 | * 3/ any empty disks < next_spare become removed | |
852 | * | |
853 | * disks[0] gets initialised to REMOVED because | |
854 | * we cannot be sure from other fields if it has | |
855 | * been initialised or not. | |
856 | */ | |
857 | int i; | |
858 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
859 | ||
61181565 N |
860 | rdev->sb_size = MD_SB_BYTES; |
861 | ||
1da177e4 LT |
862 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
863 | ||
864 | memset(sb, 0, sizeof(*sb)); | |
865 | ||
866 | sb->md_magic = MD_SB_MAGIC; | |
867 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
868 | sb->patch_version = mddev->patch_version; |
869 | sb->gvalid_words = 0; /* ignored */ | |
870 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
871 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
872 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
873 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
874 | ||
875 | sb->ctime = mddev->ctime; | |
876 | sb->level = mddev->level; | |
877 | sb->size = mddev->size; | |
878 | sb->raid_disks = mddev->raid_disks; | |
879 | sb->md_minor = mddev->md_minor; | |
880 | sb->not_persistent = !mddev->persistent; | |
881 | sb->utime = mddev->utime; | |
882 | sb->state = 0; | |
883 | sb->events_hi = (mddev->events>>32); | |
884 | sb->events_lo = (u32)mddev->events; | |
885 | ||
f6705578 N |
886 | if (mddev->reshape_position == MaxSector) |
887 | sb->minor_version = 90; | |
888 | else { | |
889 | sb->minor_version = 91; | |
890 | sb->reshape_position = mddev->reshape_position; | |
891 | sb->new_level = mddev->new_level; | |
892 | sb->delta_disks = mddev->delta_disks; | |
893 | sb->new_layout = mddev->new_layout; | |
894 | sb->new_chunk = mddev->new_chunk; | |
895 | } | |
896 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
897 | if (mddev->in_sync) |
898 | { | |
899 | sb->recovery_cp = mddev->recovery_cp; | |
900 | sb->cp_events_hi = (mddev->events>>32); | |
901 | sb->cp_events_lo = (u32)mddev->events; | |
902 | if (mddev->recovery_cp == MaxSector) | |
903 | sb->state = (1<< MD_SB_CLEAN); | |
904 | } else | |
905 | sb->recovery_cp = 0; | |
906 | ||
907 | sb->layout = mddev->layout; | |
908 | sb->chunk_size = mddev->chunk_size; | |
909 | ||
a654b9d8 N |
910 | if (mddev->bitmap && mddev->bitmap_file == NULL) |
911 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); | |
912 | ||
1da177e4 LT |
913 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
914 | ITERATE_RDEV(mddev,rdev2,tmp) { | |
915 | mdp_disk_t *d; | |
86e6ffdd | 916 | int desc_nr; |
b2d444d7 N |
917 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
918 | && !test_bit(Faulty, &rdev2->flags)) | |
86e6ffdd | 919 | desc_nr = rdev2->raid_disk; |
1da177e4 | 920 | else |
86e6ffdd | 921 | desc_nr = next_spare++; |
19133a42 | 922 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
923 | d = &sb->disks[rdev2->desc_nr]; |
924 | nr_disks++; | |
925 | d->number = rdev2->desc_nr; | |
926 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
927 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
b2d444d7 N |
928 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
929 | && !test_bit(Faulty, &rdev2->flags)) | |
1da177e4 LT |
930 | d->raid_disk = rdev2->raid_disk; |
931 | else | |
932 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 933 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 934 | d->state = (1<<MD_DISK_FAULTY); |
1be7892f | 935 | else if (test_bit(In_sync, &rdev2->flags)) { |
1da177e4 LT |
936 | d->state = (1<<MD_DISK_ACTIVE); |
937 | d->state |= (1<<MD_DISK_SYNC); | |
938 | active++; | |
939 | working++; | |
940 | } else { | |
941 | d->state = 0; | |
942 | spare++; | |
943 | working++; | |
944 | } | |
8ddf9efe N |
945 | if (test_bit(WriteMostly, &rdev2->flags)) |
946 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 947 | } |
1da177e4 LT |
948 | /* now set the "removed" and "faulty" bits on any missing devices */ |
949 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
950 | mdp_disk_t *d = &sb->disks[i]; | |
951 | if (d->state == 0 && d->number == 0) { | |
952 | d->number = i; | |
953 | d->raid_disk = i; | |
954 | d->state = (1<<MD_DISK_REMOVED); | |
955 | d->state |= (1<<MD_DISK_FAULTY); | |
956 | failed++; | |
957 | } | |
958 | } | |
959 | sb->nr_disks = nr_disks; | |
960 | sb->active_disks = active; | |
961 | sb->working_disks = working; | |
962 | sb->failed_disks = failed; | |
963 | sb->spare_disks = spare; | |
964 | ||
965 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
966 | sb->sb_csum = calc_sb_csum(sb); | |
967 | } | |
968 | ||
969 | /* | |
970 | * version 1 superblock | |
971 | */ | |
972 | ||
973 | static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb) | |
974 | { | |
975 | unsigned int disk_csum, csum; | |
976 | unsigned long long newcsum; | |
977 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
978 | unsigned int *isuper = (unsigned int*)sb; | |
979 | int i; | |
980 | ||
981 | disk_csum = sb->sb_csum; | |
982 | sb->sb_csum = 0; | |
983 | newcsum = 0; | |
984 | for (i=0; size>=4; size -= 4 ) | |
985 | newcsum += le32_to_cpu(*isuper++); | |
986 | ||
987 | if (size == 2) | |
988 | newcsum += le16_to_cpu(*(unsigned short*) isuper); | |
989 | ||
990 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
991 | sb->sb_csum = disk_csum; | |
992 | return cpu_to_le32(csum); | |
993 | } | |
994 | ||
995 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
996 | { | |
997 | struct mdp_superblock_1 *sb; | |
998 | int ret; | |
999 | sector_t sb_offset; | |
1000 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
0002b271 | 1001 | int bmask; |
1da177e4 LT |
1002 | |
1003 | /* | |
1004 | * Calculate the position of the superblock. | |
1005 | * It is always aligned to a 4K boundary and | |
1006 | * depeding on minor_version, it can be: | |
1007 | * 0: At least 8K, but less than 12K, from end of device | |
1008 | * 1: At start of device | |
1009 | * 2: 4K from start of device. | |
1010 | */ | |
1011 | switch(minor_version) { | |
1012 | case 0: | |
1013 | sb_offset = rdev->bdev->bd_inode->i_size >> 9; | |
1014 | sb_offset -= 8*2; | |
39730960 | 1015 | sb_offset &= ~(sector_t)(4*2-1); |
1da177e4 LT |
1016 | /* convert from sectors to K */ |
1017 | sb_offset /= 2; | |
1018 | break; | |
1019 | case 1: | |
1020 | sb_offset = 0; | |
1021 | break; | |
1022 | case 2: | |
1023 | sb_offset = 4; | |
1024 | break; | |
1025 | default: | |
1026 | return -EINVAL; | |
1027 | } | |
1028 | rdev->sb_offset = sb_offset; | |
1029 | ||
0002b271 N |
1030 | /* superblock is rarely larger than 1K, but it can be larger, |
1031 | * and it is safe to read 4k, so we do that | |
1032 | */ | |
1033 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1034 | if (ret) return ret; |
1035 | ||
1036 | ||
1037 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1038 | ||
1039 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1040 | sb->major_version != cpu_to_le32(1) || | |
1041 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
1042 | le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) || | |
71c0805c | 1043 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1044 | return -EINVAL; |
1045 | ||
1046 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1047 | printk("md: invalid superblock checksum on %s\n", | |
1048 | bdevname(rdev->bdev,b)); | |
1049 | return -EINVAL; | |
1050 | } | |
1051 | if (le64_to_cpu(sb->data_size) < 10) { | |
1052 | printk("md: data_size too small on %s\n", | |
1053 | bdevname(rdev->bdev,b)); | |
1054 | return -EINVAL; | |
1055 | } | |
1056 | rdev->preferred_minor = 0xffff; | |
1057 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1058 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1059 | |
0002b271 | 1060 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
720a3dc3 | 1061 | bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 N |
1062 | if (rdev->sb_size & bmask) |
1063 | rdev-> sb_size = (rdev->sb_size | bmask)+1; | |
1064 | ||
31b65a0d N |
1065 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1066 | rdev->desc_nr = -1; | |
1067 | else | |
1068 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1069 | ||
1da177e4 | 1070 | if (refdev == 0) |
8ed75463 | 1071 | ret = 1; |
1da177e4 LT |
1072 | else { |
1073 | __u64 ev1, ev2; | |
1074 | struct mdp_superblock_1 *refsb = | |
1075 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1076 | ||
1077 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1078 | sb->level != refsb->level || | |
1079 | sb->layout != refsb->layout || | |
1080 | sb->chunksize != refsb->chunksize) { | |
1081 | printk(KERN_WARNING "md: %s has strangely different" | |
1082 | " superblock to %s\n", | |
1083 | bdevname(rdev->bdev,b), | |
1084 | bdevname(refdev->bdev,b2)); | |
1085 | return -EINVAL; | |
1086 | } | |
1087 | ev1 = le64_to_cpu(sb->events); | |
1088 | ev2 = le64_to_cpu(refsb->events); | |
1089 | ||
1090 | if (ev1 > ev2) | |
8ed75463 N |
1091 | ret = 1; |
1092 | else | |
1093 | ret = 0; | |
1da177e4 LT |
1094 | } |
1095 | if (minor_version) | |
1096 | rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2; | |
1097 | else | |
1098 | rdev->size = rdev->sb_offset; | |
1099 | if (rdev->size < le64_to_cpu(sb->data_size)/2) | |
1100 | return -EINVAL; | |
1101 | rdev->size = le64_to_cpu(sb->data_size)/2; | |
1102 | if (le32_to_cpu(sb->chunksize)) | |
1103 | rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1); | |
2bf071bf N |
1104 | |
1105 | if (le32_to_cpu(sb->size) > rdev->size*2) | |
1106 | return -EINVAL; | |
8ed75463 | 1107 | return ret; |
1da177e4 LT |
1108 | } |
1109 | ||
1110 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1111 | { | |
1112 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1113 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1114 | |
41158c7e | 1115 | rdev->raid_disk = -1; |
b2d444d7 | 1116 | rdev->flags = 0; |
1da177e4 LT |
1117 | if (mddev->raid_disks == 0) { |
1118 | mddev->major_version = 1; | |
1119 | mddev->patch_version = 0; | |
1120 | mddev->persistent = 1; | |
1121 | mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9; | |
1122 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); | |
1123 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1124 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1125 | mddev->clevel[0] = 0; |
1da177e4 LT |
1126 | mddev->layout = le32_to_cpu(sb->layout); |
1127 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
1128 | mddev->size = le64_to_cpu(sb->size)/2; | |
07d84d10 | 1129 | mddev->events = ev1; |
9223214e | 1130 | mddev->bitmap_offset = 0; |
29fc7e3e | 1131 | mddev->default_bitmap_offset = 1024 >> 9; |
1da177e4 LT |
1132 | |
1133 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1134 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1135 | ||
1136 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1137 | |
71c0805c | 1138 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
a654b9d8 | 1139 | mddev->bitmap_file == NULL ) { |
6cce3b23 N |
1140 | if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6 |
1141 | && mddev->level != 10) { | |
1142 | printk(KERN_WARNING "md: bitmaps not supported for this level.\n"); | |
a654b9d8 N |
1143 | return -EINVAL; |
1144 | } | |
1145 | mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); | |
1146 | } | |
f6705578 N |
1147 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1148 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1149 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1150 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1151 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
1152 | mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9; | |
1153 | } else { | |
1154 | mddev->reshape_position = MaxSector; | |
1155 | mddev->delta_disks = 0; | |
1156 | mddev->new_level = mddev->level; | |
1157 | mddev->new_layout = mddev->layout; | |
1158 | mddev->new_chunk = mddev->chunk_size; | |
1159 | } | |
1160 | ||
41158c7e N |
1161 | } else if (mddev->pers == NULL) { |
1162 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1163 | ++ev1; |
1164 | if (ev1 < mddev->events) | |
1165 | return -EINVAL; | |
41158c7e N |
1166 | } else if (mddev->bitmap) { |
1167 | /* If adding to array with a bitmap, then we can accept an | |
1168 | * older device, but not too old. | |
1169 | */ | |
41158c7e N |
1170 | if (ev1 < mddev->bitmap->events_cleared) |
1171 | return 0; | |
07d84d10 N |
1172 | } else { |
1173 | if (ev1 < mddev->events) | |
1174 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1175 | return 0; | |
1176 | } | |
1da177e4 LT |
1177 | if (mddev->level != LEVEL_MULTIPATH) { |
1178 | int role; | |
1da177e4 LT |
1179 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); |
1180 | switch(role) { | |
1181 | case 0xffff: /* spare */ | |
1da177e4 LT |
1182 | break; |
1183 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1184 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1185 | break; |
1186 | default: | |
5fd6c1dc N |
1187 | if ((le32_to_cpu(sb->feature_map) & |
1188 | MD_FEATURE_RECOVERY_OFFSET)) | |
1189 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1190 | else | |
1191 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1192 | rdev->raid_disk = role; |
1193 | break; | |
1194 | } | |
8ddf9efe N |
1195 | if (sb->devflags & WriteMostly1) |
1196 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1197 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1198 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1199 | |
1da177e4 LT |
1200 | return 0; |
1201 | } | |
1202 | ||
1203 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1204 | { | |
1205 | struct mdp_superblock_1 *sb; | |
1206 | struct list_head *tmp; | |
1207 | mdk_rdev_t *rdev2; | |
1208 | int max_dev, i; | |
1209 | /* make rdev->sb match mddev and rdev data. */ | |
1210 | ||
1211 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1212 | ||
1213 | sb->feature_map = 0; | |
1214 | sb->pad0 = 0; | |
5fd6c1dc | 1215 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1216 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1217 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1218 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1219 | ||
1220 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1221 | sb->events = cpu_to_le64(mddev->events); | |
1222 | if (mddev->in_sync) | |
1223 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1224 | else | |
1225 | sb->resync_offset = cpu_to_le64(0); | |
1226 | ||
4dbcdc75 N |
1227 | sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors); |
1228 | ||
f0ca340c | 1229 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
29fc7e3e | 1230 | sb->size = cpu_to_le64(mddev->size<<1); |
f0ca340c | 1231 | |
a654b9d8 N |
1232 | if (mddev->bitmap && mddev->bitmap_file == NULL) { |
1233 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); | |
71c0805c | 1234 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1235 | } |
5fd6c1dc N |
1236 | |
1237 | if (rdev->raid_disk >= 0 && | |
1238 | !test_bit(In_sync, &rdev->flags) && | |
1239 | rdev->recovery_offset > 0) { | |
1240 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1241 | sb->recovery_offset = cpu_to_le64(rdev->recovery_offset); | |
1242 | } | |
1243 | ||
f6705578 N |
1244 | if (mddev->reshape_position != MaxSector) { |
1245 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1246 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1247 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1248 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1249 | sb->new_level = cpu_to_le32(mddev->new_level); | |
1250 | sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9); | |
1251 | } | |
a654b9d8 | 1252 | |
1da177e4 LT |
1253 | max_dev = 0; |
1254 | ITERATE_RDEV(mddev,rdev2,tmp) | |
1255 | if (rdev2->desc_nr+1 > max_dev) | |
1256 | max_dev = rdev2->desc_nr+1; | |
1257 | ||
1258 | sb->max_dev = cpu_to_le32(max_dev); | |
1259 | for (i=0; i<max_dev;i++) | |
1260 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1261 | ||
1262 | ITERATE_RDEV(mddev,rdev2,tmp) { | |
1263 | i = rdev2->desc_nr; | |
b2d444d7 | 1264 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1265 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1266 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1267 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
5fd6c1dc N |
1268 | else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0) |
1269 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); | |
1da177e4 LT |
1270 | else |
1271 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1272 | } | |
1273 | ||
1da177e4 LT |
1274 | sb->sb_csum = calc_sb_1_csum(sb); |
1275 | } | |
1276 | ||
1277 | ||
75c96f85 | 1278 | static struct super_type super_types[] = { |
1da177e4 LT |
1279 | [0] = { |
1280 | .name = "0.90.0", | |
1281 | .owner = THIS_MODULE, | |
1282 | .load_super = super_90_load, | |
1283 | .validate_super = super_90_validate, | |
1284 | .sync_super = super_90_sync, | |
1285 | }, | |
1286 | [1] = { | |
1287 | .name = "md-1", | |
1288 | .owner = THIS_MODULE, | |
1289 | .load_super = super_1_load, | |
1290 | .validate_super = super_1_validate, | |
1291 | .sync_super = super_1_sync, | |
1292 | }, | |
1293 | }; | |
1294 | ||
1295 | static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev) | |
1296 | { | |
1297 | struct list_head *tmp; | |
1298 | mdk_rdev_t *rdev; | |
1299 | ||
1300 | ITERATE_RDEV(mddev,rdev,tmp) | |
1301 | if (rdev->bdev->bd_contains == dev->bdev->bd_contains) | |
1302 | return rdev; | |
1303 | ||
1304 | return NULL; | |
1305 | } | |
1306 | ||
1307 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1308 | { | |
1309 | struct list_head *tmp; | |
1310 | mdk_rdev_t *rdev; | |
1311 | ||
1312 | ITERATE_RDEV(mddev1,rdev,tmp) | |
1313 | if (match_dev_unit(mddev2, rdev)) | |
1314 | return 1; | |
1315 | ||
1316 | return 0; | |
1317 | } | |
1318 | ||
1319 | static LIST_HEAD(pending_raid_disks); | |
1320 | ||
1321 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) | |
1322 | { | |
1323 | mdk_rdev_t *same_pdev; | |
1324 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
f637b9f9 | 1325 | struct kobject *ko; |
1edf80d3 | 1326 | char *s; |
1da177e4 LT |
1327 | |
1328 | if (rdev->mddev) { | |
1329 | MD_BUG(); | |
1330 | return -EINVAL; | |
1331 | } | |
2bf071bf N |
1332 | /* make sure rdev->size exceeds mddev->size */ |
1333 | if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) { | |
1334 | if (mddev->pers) | |
1335 | /* Cannot change size, so fail */ | |
1336 | return -ENOSPC; | |
1337 | else | |
1338 | mddev->size = rdev->size; | |
1339 | } | |
1da177e4 LT |
1340 | same_pdev = match_dev_unit(mddev, rdev); |
1341 | if (same_pdev) | |
1342 | printk(KERN_WARNING | |
1343 | "%s: WARNING: %s appears to be on the same physical" | |
1344 | " disk as %s. True\n protection against single-disk" | |
1345 | " failure might be compromised.\n", | |
1346 | mdname(mddev), bdevname(rdev->bdev,b), | |
1347 | bdevname(same_pdev->bdev,b2)); | |
1348 | ||
1349 | /* Verify rdev->desc_nr is unique. | |
1350 | * If it is -1, assign a free number, else | |
1351 | * check number is not in use | |
1352 | */ | |
1353 | if (rdev->desc_nr < 0) { | |
1354 | int choice = 0; | |
1355 | if (mddev->pers) choice = mddev->raid_disks; | |
1356 | while (find_rdev_nr(mddev, choice)) | |
1357 | choice++; | |
1358 | rdev->desc_nr = choice; | |
1359 | } else { | |
1360 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1361 | return -EBUSY; | |
1362 | } | |
19133a42 N |
1363 | bdevname(rdev->bdev,b); |
1364 | if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0) | |
1365 | return -ENOMEM; | |
1edf80d3 NB |
1366 | while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL) |
1367 | *s = '!'; | |
1da177e4 LT |
1368 | |
1369 | list_add(&rdev->same_set, &mddev->disks); | |
1370 | rdev->mddev = mddev; | |
19133a42 | 1371 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1372 | |
9c791977 | 1373 | rdev->kobj.parent = &mddev->kobj; |
86e6ffdd N |
1374 | kobject_add(&rdev->kobj); |
1375 | ||
f637b9f9 N |
1376 | if (rdev->bdev->bd_part) |
1377 | ko = &rdev->bdev->bd_part->kobj; | |
1378 | else | |
1379 | ko = &rdev->bdev->bd_disk->kobj; | |
1380 | sysfs_create_link(&rdev->kobj, ko, "block"); | |
5463c790 | 1381 | bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk); |
1da177e4 LT |
1382 | return 0; |
1383 | } | |
1384 | ||
1385 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) | |
1386 | { | |
1387 | char b[BDEVNAME_SIZE]; | |
1388 | if (!rdev->mddev) { | |
1389 | MD_BUG(); | |
1390 | return; | |
1391 | } | |
5463c790 | 1392 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
1da177e4 LT |
1393 | list_del_init(&rdev->same_set); |
1394 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); | |
1395 | rdev->mddev = NULL; | |
86e6ffdd N |
1396 | sysfs_remove_link(&rdev->kobj, "block"); |
1397 | kobject_del(&rdev->kobj); | |
1da177e4 LT |
1398 | } |
1399 | ||
1400 | /* | |
1401 | * prevent the device from being mounted, repartitioned or | |
1402 | * otherwise reused by a RAID array (or any other kernel | |
1403 | * subsystem), by bd_claiming the device. | |
1404 | */ | |
1405 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev) | |
1406 | { | |
1407 | int err = 0; | |
1408 | struct block_device *bdev; | |
1409 | char b[BDEVNAME_SIZE]; | |
1410 | ||
663d440e | 1411 | bdev = open_partition_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1412 | if (IS_ERR(bdev)) { |
1413 | printk(KERN_ERR "md: could not open %s.\n", | |
1414 | __bdevname(dev, b)); | |
1415 | return PTR_ERR(bdev); | |
1416 | } | |
1417 | err = bd_claim(bdev, rdev); | |
1418 | if (err) { | |
1419 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1420 | bdevname(bdev, b)); | |
663d440e | 1421 | blkdev_put_partition(bdev); |
1da177e4 LT |
1422 | return err; |
1423 | } | |
1424 | rdev->bdev = bdev; | |
1425 | return err; | |
1426 | } | |
1427 | ||
1428 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1429 | { | |
1430 | struct block_device *bdev = rdev->bdev; | |
1431 | rdev->bdev = NULL; | |
1432 | if (!bdev) | |
1433 | MD_BUG(); | |
1434 | bd_release(bdev); | |
663d440e | 1435 | blkdev_put_partition(bdev); |
1da177e4 LT |
1436 | } |
1437 | ||
1438 | void md_autodetect_dev(dev_t dev); | |
1439 | ||
1440 | static void export_rdev(mdk_rdev_t * rdev) | |
1441 | { | |
1442 | char b[BDEVNAME_SIZE]; | |
1443 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1444 | bdevname(rdev->bdev,b)); | |
1445 | if (rdev->mddev) | |
1446 | MD_BUG(); | |
1447 | free_disk_sb(rdev); | |
1448 | list_del_init(&rdev->same_set); | |
1449 | #ifndef MODULE | |
1450 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1451 | #endif | |
1452 | unlock_rdev(rdev); | |
86e6ffdd | 1453 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1454 | } |
1455 | ||
1456 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1457 | { | |
1458 | unbind_rdev_from_array(rdev); | |
1459 | export_rdev(rdev); | |
1460 | } | |
1461 | ||
1462 | static void export_array(mddev_t *mddev) | |
1463 | { | |
1464 | struct list_head *tmp; | |
1465 | mdk_rdev_t *rdev; | |
1466 | ||
1467 | ITERATE_RDEV(mddev,rdev,tmp) { | |
1468 | if (!rdev->mddev) { | |
1469 | MD_BUG(); | |
1470 | continue; | |
1471 | } | |
1472 | kick_rdev_from_array(rdev); | |
1473 | } | |
1474 | if (!list_empty(&mddev->disks)) | |
1475 | MD_BUG(); | |
1476 | mddev->raid_disks = 0; | |
1477 | mddev->major_version = 0; | |
1478 | } | |
1479 | ||
1480 | static void print_desc(mdp_disk_t *desc) | |
1481 | { | |
1482 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1483 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1484 | } | |
1485 | ||
1486 | static void print_sb(mdp_super_t *sb) | |
1487 | { | |
1488 | int i; | |
1489 | ||
1490 | printk(KERN_INFO | |
1491 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1492 | sb->major_version, sb->minor_version, sb->patch_version, | |
1493 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1494 | sb->ctime); | |
1495 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1496 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1497 | sb->md_minor, sb->layout, sb->chunk_size); | |
1498 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1499 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1500 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1501 | sb->failed_disks, sb->spare_disks, | |
1502 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1503 | ||
1504 | printk(KERN_INFO); | |
1505 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1506 | mdp_disk_t *desc; | |
1507 | ||
1508 | desc = sb->disks + i; | |
1509 | if (desc->number || desc->major || desc->minor || | |
1510 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1511 | printk(" D %2d: ", i); | |
1512 | print_desc(desc); | |
1513 | } | |
1514 | } | |
1515 | printk(KERN_INFO "md: THIS: "); | |
1516 | print_desc(&sb->this_disk); | |
1517 | ||
1518 | } | |
1519 | ||
1520 | static void print_rdev(mdk_rdev_t *rdev) | |
1521 | { | |
1522 | char b[BDEVNAME_SIZE]; | |
1523 | printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n", | |
1524 | bdevname(rdev->bdev,b), (unsigned long long)rdev->size, | |
b2d444d7 N |
1525 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
1526 | rdev->desc_nr); | |
1da177e4 LT |
1527 | if (rdev->sb_loaded) { |
1528 | printk(KERN_INFO "md: rdev superblock:\n"); | |
1529 | print_sb((mdp_super_t*)page_address(rdev->sb_page)); | |
1530 | } else | |
1531 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
1532 | } | |
1533 | ||
5e56341d | 1534 | static void md_print_devices(void) |
1da177e4 LT |
1535 | { |
1536 | struct list_head *tmp, *tmp2; | |
1537 | mdk_rdev_t *rdev; | |
1538 | mddev_t *mddev; | |
1539 | char b[BDEVNAME_SIZE]; | |
1540 | ||
1541 | printk("\n"); | |
1542 | printk("md: **********************************\n"); | |
1543 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
1544 | printk("md: **********************************\n"); | |
1545 | ITERATE_MDDEV(mddev,tmp) { | |
1da177e4 | 1546 | |
32a7627c N |
1547 | if (mddev->bitmap) |
1548 | bitmap_print_sb(mddev->bitmap); | |
1549 | else | |
1550 | printk("%s: ", mdname(mddev)); | |
1da177e4 LT |
1551 | ITERATE_RDEV(mddev,rdev,tmp2) |
1552 | printk("<%s>", bdevname(rdev->bdev,b)); | |
1553 | printk("\n"); | |
1554 | ||
1555 | ITERATE_RDEV(mddev,rdev,tmp2) | |
1556 | print_rdev(rdev); | |
1557 | } | |
1558 | printk("md: **********************************\n"); | |
1559 | printk("\n"); | |
1560 | } | |
1561 | ||
1562 | ||
42543769 | 1563 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 1564 | { |
42543769 N |
1565 | /* Update each superblock (in-memory image), but |
1566 | * if we are allowed to, skip spares which already | |
1567 | * have the right event counter, or have one earlier | |
1568 | * (which would mean they aren't being marked as dirty | |
1569 | * with the rest of the array) | |
1570 | */ | |
1da177e4 LT |
1571 | mdk_rdev_t *rdev; |
1572 | struct list_head *tmp; | |
1573 | ||
1574 | ITERATE_RDEV(mddev,rdev,tmp) { | |
42543769 N |
1575 | if (rdev->sb_events == mddev->events || |
1576 | (nospares && | |
1577 | rdev->raid_disk < 0 && | |
1578 | (rdev->sb_events&1)==0 && | |
1579 | rdev->sb_events+1 == mddev->events)) { | |
1580 | /* Don't update this superblock */ | |
1581 | rdev->sb_loaded = 2; | |
1582 | } else { | |
1583 | super_types[mddev->major_version]. | |
1584 | sync_super(mddev, rdev); | |
1585 | rdev->sb_loaded = 1; | |
1586 | } | |
1da177e4 LT |
1587 | } |
1588 | } | |
1589 | ||
f6705578 | 1590 | void md_update_sb(mddev_t * mddev) |
1da177e4 | 1591 | { |
7bfa19f2 | 1592 | int err; |
1da177e4 LT |
1593 | struct list_head *tmp; |
1594 | mdk_rdev_t *rdev; | |
06d91a5f | 1595 | int sync_req; |
42543769 | 1596 | int nospares = 0; |
1da177e4 | 1597 | |
1da177e4 | 1598 | repeat: |
a9701a30 | 1599 | spin_lock_irq(&mddev->write_lock); |
06d91a5f | 1600 | sync_req = mddev->in_sync; |
1da177e4 | 1601 | mddev->utime = get_seconds(); |
42543769 N |
1602 | if (mddev->sb_dirty == 3) |
1603 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
1604 | * though if events isn't the right even/odd, we will have to do | |
1605 | * spares after all | |
1606 | */ | |
1607 | nospares = 1; | |
1608 | ||
1609 | /* If this is just a dirty<->clean transition, and the array is clean | |
1610 | * and 'events' is odd, we can roll back to the previous clean state */ | |
1611 | if (mddev->sb_dirty == 3 | |
1612 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1613 | && (mddev->events & 1)) | |
1614 | mddev->events--; | |
1615 | else { | |
1616 | /* otherwise we have to go forward and ... */ | |
1617 | mddev->events ++; | |
1618 | if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ | |
1619 | /* .. if the array isn't clean, insist on an odd 'events' */ | |
1620 | if ((mddev->events&1)==0) { | |
1621 | mddev->events++; | |
1622 | nospares = 0; | |
1623 | } | |
1624 | } else { | |
1625 | /* otherwise insist on an even 'events' (for clean states) */ | |
1626 | if ((mddev->events&1)) { | |
1627 | mddev->events++; | |
1628 | nospares = 0; | |
1629 | } | |
1630 | } | |
1631 | } | |
1da177e4 LT |
1632 | |
1633 | if (!mddev->events) { | |
1634 | /* | |
1635 | * oops, this 64-bit counter should never wrap. | |
1636 | * Either we are in around ~1 trillion A.C., assuming | |
1637 | * 1 reboot per second, or we have a bug: | |
1638 | */ | |
1639 | MD_BUG(); | |
1640 | mddev->events --; | |
1641 | } | |
7bfa19f2 | 1642 | mddev->sb_dirty = 2; |
42543769 | 1643 | sync_sbs(mddev, nospares); |
1da177e4 LT |
1644 | |
1645 | /* | |
1646 | * do not write anything to disk if using | |
1647 | * nonpersistent superblocks | |
1648 | */ | |
06d91a5f N |
1649 | if (!mddev->persistent) { |
1650 | mddev->sb_dirty = 0; | |
a9701a30 | 1651 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1652 | wake_up(&mddev->sb_wait); |
1da177e4 | 1653 | return; |
06d91a5f | 1654 | } |
a9701a30 | 1655 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
1656 | |
1657 | dprintk(KERN_INFO | |
1658 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
1659 | mdname(mddev),mddev->in_sync); | |
1660 | ||
32a7627c | 1661 | err = bitmap_update_sb(mddev->bitmap); |
1da177e4 LT |
1662 | ITERATE_RDEV(mddev,rdev,tmp) { |
1663 | char b[BDEVNAME_SIZE]; | |
1664 | dprintk(KERN_INFO "md: "); | |
42543769 N |
1665 | if (rdev->sb_loaded != 1) |
1666 | continue; /* no noise on spare devices */ | |
b2d444d7 | 1667 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
1668 | dprintk("(skipping faulty "); |
1669 | ||
1670 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 1671 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 1672 | md_super_write(mddev,rdev, |
0002b271 | 1673 | rdev->sb_offset<<1, rdev->sb_size, |
7bfa19f2 N |
1674 | rdev->sb_page); |
1675 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
1676 | bdevname(rdev->bdev,b), | |
1677 | (unsigned long long)rdev->sb_offset); | |
42543769 | 1678 | rdev->sb_events = mddev->events; |
7bfa19f2 | 1679 | |
1da177e4 LT |
1680 | } else |
1681 | dprintk(")\n"); | |
7bfa19f2 | 1682 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
1683 | /* only need to write one superblock... */ |
1684 | break; | |
1685 | } | |
a9701a30 | 1686 | md_super_wait(mddev); |
7bfa19f2 N |
1687 | /* if there was a failure, sb_dirty was set to 1, and we re-write super */ |
1688 | ||
a9701a30 | 1689 | spin_lock_irq(&mddev->write_lock); |
7bfa19f2 | 1690 | if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) { |
06d91a5f | 1691 | /* have to write it out again */ |
a9701a30 | 1692 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
1693 | goto repeat; |
1694 | } | |
1695 | mddev->sb_dirty = 0; | |
a9701a30 | 1696 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1697 | wake_up(&mddev->sb_wait); |
06d91a5f | 1698 | |
1da177e4 | 1699 | } |
f6705578 | 1700 | EXPORT_SYMBOL_GPL(md_update_sb); |
1da177e4 | 1701 | |
bce74dac N |
1702 | /* words written to sysfs files may, or my not, be \n terminated. |
1703 | * We want to accept with case. For this we use cmd_match. | |
1704 | */ | |
1705 | static int cmd_match(const char *cmd, const char *str) | |
1706 | { | |
1707 | /* See if cmd, written into a sysfs file, matches | |
1708 | * str. They must either be the same, or cmd can | |
1709 | * have a trailing newline | |
1710 | */ | |
1711 | while (*cmd && *str && *cmd == *str) { | |
1712 | cmd++; | |
1713 | str++; | |
1714 | } | |
1715 | if (*cmd == '\n') | |
1716 | cmd++; | |
1717 | if (*str || *cmd) | |
1718 | return 0; | |
1719 | return 1; | |
1720 | } | |
1721 | ||
86e6ffdd N |
1722 | struct rdev_sysfs_entry { |
1723 | struct attribute attr; | |
1724 | ssize_t (*show)(mdk_rdev_t *, char *); | |
1725 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
1726 | }; | |
1727 | ||
1728 | static ssize_t | |
96de1e66 | 1729 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
1730 | { |
1731 | char *sep = ""; | |
1732 | int len=0; | |
1733 | ||
b2d444d7 | 1734 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
1735 | len+= sprintf(page+len, "%sfaulty",sep); |
1736 | sep = ","; | |
1737 | } | |
b2d444d7 | 1738 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
1739 | len += sprintf(page+len, "%sin_sync",sep); |
1740 | sep = ","; | |
1741 | } | |
f655675b N |
1742 | if (test_bit(WriteMostly, &rdev->flags)) { |
1743 | len += sprintf(page+len, "%swrite_mostly",sep); | |
1744 | sep = ","; | |
1745 | } | |
b2d444d7 N |
1746 | if (!test_bit(Faulty, &rdev->flags) && |
1747 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
1748 | len += sprintf(page+len, "%sspare", sep); |
1749 | sep = ","; | |
1750 | } | |
1751 | return len+sprintf(page+len, "\n"); | |
1752 | } | |
1753 | ||
45dc2de1 N |
1754 | static ssize_t |
1755 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1756 | { | |
1757 | /* can write | |
1758 | * faulty - simulates and error | |
1759 | * remove - disconnects the device | |
f655675b N |
1760 | * writemostly - sets write_mostly |
1761 | * -writemostly - clears write_mostly | |
45dc2de1 N |
1762 | */ |
1763 | int err = -EINVAL; | |
1764 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
1765 | md_error(rdev->mddev, rdev); | |
1766 | err = 0; | |
1767 | } else if (cmd_match(buf, "remove")) { | |
1768 | if (rdev->raid_disk >= 0) | |
1769 | err = -EBUSY; | |
1770 | else { | |
1771 | mddev_t *mddev = rdev->mddev; | |
1772 | kick_rdev_from_array(rdev); | |
1773 | md_update_sb(mddev); | |
1774 | md_new_event(mddev); | |
1775 | err = 0; | |
1776 | } | |
f655675b N |
1777 | } else if (cmd_match(buf, "writemostly")) { |
1778 | set_bit(WriteMostly, &rdev->flags); | |
1779 | err = 0; | |
1780 | } else if (cmd_match(buf, "-writemostly")) { | |
1781 | clear_bit(WriteMostly, &rdev->flags); | |
1782 | err = 0; | |
45dc2de1 N |
1783 | } |
1784 | return err ? err : len; | |
1785 | } | |
96de1e66 | 1786 | static struct rdev_sysfs_entry |
45dc2de1 | 1787 | rdev_state = __ATTR(state, 0644, state_show, state_store); |
86e6ffdd N |
1788 | |
1789 | static ssize_t | |
96de1e66 | 1790 | super_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
1791 | { |
1792 | if (rdev->sb_loaded && rdev->sb_size) { | |
1793 | memcpy(page, page_address(rdev->sb_page), rdev->sb_size); | |
1794 | return rdev->sb_size; | |
1795 | } else | |
1796 | return 0; | |
1797 | } | |
96de1e66 N |
1798 | static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super); |
1799 | ||
4dbcdc75 N |
1800 | static ssize_t |
1801 | errors_show(mdk_rdev_t *rdev, char *page) | |
1802 | { | |
1803 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
1804 | } | |
1805 | ||
1806 | static ssize_t | |
1807 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1808 | { | |
1809 | char *e; | |
1810 | unsigned long n = simple_strtoul(buf, &e, 10); | |
1811 | if (*buf && (*e == 0 || *e == '\n')) { | |
1812 | atomic_set(&rdev->corrected_errors, n); | |
1813 | return len; | |
1814 | } | |
1815 | return -EINVAL; | |
1816 | } | |
1817 | static struct rdev_sysfs_entry rdev_errors = | |
1818 | __ATTR(errors, 0644, errors_show, errors_store); | |
1819 | ||
014236d2 N |
1820 | static ssize_t |
1821 | slot_show(mdk_rdev_t *rdev, char *page) | |
1822 | { | |
1823 | if (rdev->raid_disk < 0) | |
1824 | return sprintf(page, "none\n"); | |
1825 | else | |
1826 | return sprintf(page, "%d\n", rdev->raid_disk); | |
1827 | } | |
1828 | ||
1829 | static ssize_t | |
1830 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1831 | { | |
1832 | char *e; | |
1833 | int slot = simple_strtoul(buf, &e, 10); | |
1834 | if (strncmp(buf, "none", 4)==0) | |
1835 | slot = -1; | |
1836 | else if (e==buf || (*e && *e!= '\n')) | |
1837 | return -EINVAL; | |
1838 | if (rdev->mddev->pers) | |
1839 | /* Cannot set slot in active array (yet) */ | |
1840 | return -EBUSY; | |
1841 | if (slot >= rdev->mddev->raid_disks) | |
1842 | return -ENOSPC; | |
1843 | rdev->raid_disk = slot; | |
1844 | /* assume it is working */ | |
1845 | rdev->flags = 0; | |
1846 | set_bit(In_sync, &rdev->flags); | |
1847 | return len; | |
1848 | } | |
1849 | ||
1850 | ||
1851 | static struct rdev_sysfs_entry rdev_slot = | |
1852 | __ATTR(slot, 0644, slot_show, slot_store); | |
1853 | ||
93c8cad0 N |
1854 | static ssize_t |
1855 | offset_show(mdk_rdev_t *rdev, char *page) | |
1856 | { | |
6961ece4 | 1857 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
1858 | } |
1859 | ||
1860 | static ssize_t | |
1861 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1862 | { | |
1863 | char *e; | |
1864 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
1865 | if (e==buf || (*e && *e != '\n')) | |
1866 | return -EINVAL; | |
1867 | if (rdev->mddev->pers) | |
1868 | return -EBUSY; | |
1869 | rdev->data_offset = offset; | |
1870 | return len; | |
1871 | } | |
1872 | ||
1873 | static struct rdev_sysfs_entry rdev_offset = | |
1874 | __ATTR(offset, 0644, offset_show, offset_store); | |
1875 | ||
83303b61 N |
1876 | static ssize_t |
1877 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
1878 | { | |
1879 | return sprintf(page, "%llu\n", (unsigned long long)rdev->size); | |
1880 | } | |
1881 | ||
1882 | static ssize_t | |
1883 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1884 | { | |
1885 | char *e; | |
1886 | unsigned long long size = simple_strtoull(buf, &e, 10); | |
1887 | if (e==buf || (*e && *e != '\n')) | |
1888 | return -EINVAL; | |
1889 | if (rdev->mddev->pers) | |
1890 | return -EBUSY; | |
1891 | rdev->size = size; | |
1892 | if (size < rdev->mddev->size || rdev->mddev->size == 0) | |
1893 | rdev->mddev->size = size; | |
1894 | return len; | |
1895 | } | |
1896 | ||
1897 | static struct rdev_sysfs_entry rdev_size = | |
1898 | __ATTR(size, 0644, rdev_size_show, rdev_size_store); | |
1899 | ||
86e6ffdd N |
1900 | static struct attribute *rdev_default_attrs[] = { |
1901 | &rdev_state.attr, | |
1902 | &rdev_super.attr, | |
4dbcdc75 | 1903 | &rdev_errors.attr, |
014236d2 | 1904 | &rdev_slot.attr, |
93c8cad0 | 1905 | &rdev_offset.attr, |
83303b61 | 1906 | &rdev_size.attr, |
86e6ffdd N |
1907 | NULL, |
1908 | }; | |
1909 | static ssize_t | |
1910 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
1911 | { | |
1912 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
1913 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
1914 | ||
1915 | if (!entry->show) | |
1916 | return -EIO; | |
1917 | return entry->show(rdev, page); | |
1918 | } | |
1919 | ||
1920 | static ssize_t | |
1921 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
1922 | const char *page, size_t length) | |
1923 | { | |
1924 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
1925 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
1926 | ||
1927 | if (!entry->store) | |
1928 | return -EIO; | |
1929 | return entry->store(rdev, page, length); | |
1930 | } | |
1931 | ||
1932 | static void rdev_free(struct kobject *ko) | |
1933 | { | |
1934 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
1935 | kfree(rdev); | |
1936 | } | |
1937 | static struct sysfs_ops rdev_sysfs_ops = { | |
1938 | .show = rdev_attr_show, | |
1939 | .store = rdev_attr_store, | |
1940 | }; | |
1941 | static struct kobj_type rdev_ktype = { | |
1942 | .release = rdev_free, | |
1943 | .sysfs_ops = &rdev_sysfs_ops, | |
1944 | .default_attrs = rdev_default_attrs, | |
1945 | }; | |
1946 | ||
1da177e4 LT |
1947 | /* |
1948 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
1949 | * | |
1950 | * mark the device faulty if: | |
1951 | * | |
1952 | * - the device is nonexistent (zero size) | |
1953 | * - the device has no valid superblock | |
1954 | * | |
1955 | * a faulty rdev _never_ has rdev->sb set. | |
1956 | */ | |
1957 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
1958 | { | |
1959 | char b[BDEVNAME_SIZE]; | |
1960 | int err; | |
1961 | mdk_rdev_t *rdev; | |
1962 | sector_t size; | |
1963 | ||
9ffae0cf | 1964 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
1965 | if (!rdev) { |
1966 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
1967 | return ERR_PTR(-ENOMEM); | |
1968 | } | |
1da177e4 LT |
1969 | |
1970 | if ((err = alloc_disk_sb(rdev))) | |
1971 | goto abort_free; | |
1972 | ||
1973 | err = lock_rdev(rdev, newdev); | |
1974 | if (err) | |
1975 | goto abort_free; | |
1976 | ||
86e6ffdd N |
1977 | rdev->kobj.parent = NULL; |
1978 | rdev->kobj.ktype = &rdev_ktype; | |
1979 | kobject_init(&rdev->kobj); | |
1980 | ||
1da177e4 | 1981 | rdev->desc_nr = -1; |
b2d444d7 | 1982 | rdev->flags = 0; |
1da177e4 | 1983 | rdev->data_offset = 0; |
42543769 | 1984 | rdev->sb_events = 0; |
1da177e4 | 1985 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 1986 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 1987 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
1988 | |
1989 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
1990 | if (!size) { | |
1991 | printk(KERN_WARNING | |
1992 | "md: %s has zero or unknown size, marking faulty!\n", | |
1993 | bdevname(rdev->bdev,b)); | |
1994 | err = -EINVAL; | |
1995 | goto abort_free; | |
1996 | } | |
1997 | ||
1998 | if (super_format >= 0) { | |
1999 | err = super_types[super_format]. | |
2000 | load_super(rdev, NULL, super_minor); | |
2001 | if (err == -EINVAL) { | |
2002 | printk(KERN_WARNING | |
2003 | "md: %s has invalid sb, not importing!\n", | |
2004 | bdevname(rdev->bdev,b)); | |
2005 | goto abort_free; | |
2006 | } | |
2007 | if (err < 0) { | |
2008 | printk(KERN_WARNING | |
2009 | "md: could not read %s's sb, not importing!\n", | |
2010 | bdevname(rdev->bdev,b)); | |
2011 | goto abort_free; | |
2012 | } | |
2013 | } | |
2014 | INIT_LIST_HEAD(&rdev->same_set); | |
2015 | ||
2016 | return rdev; | |
2017 | ||
2018 | abort_free: | |
2019 | if (rdev->sb_page) { | |
2020 | if (rdev->bdev) | |
2021 | unlock_rdev(rdev); | |
2022 | free_disk_sb(rdev); | |
2023 | } | |
2024 | kfree(rdev); | |
2025 | return ERR_PTR(err); | |
2026 | } | |
2027 | ||
2028 | /* | |
2029 | * Check a full RAID array for plausibility | |
2030 | */ | |
2031 | ||
2032 | ||
a757e64c | 2033 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2034 | { |
2035 | int i; | |
2036 | struct list_head *tmp; | |
2037 | mdk_rdev_t *rdev, *freshest; | |
2038 | char b[BDEVNAME_SIZE]; | |
2039 | ||
2040 | freshest = NULL; | |
2041 | ITERATE_RDEV(mddev,rdev,tmp) | |
2042 | switch (super_types[mddev->major_version]. | |
2043 | load_super(rdev, freshest, mddev->minor_version)) { | |
2044 | case 1: | |
2045 | freshest = rdev; | |
2046 | break; | |
2047 | case 0: | |
2048 | break; | |
2049 | default: | |
2050 | printk( KERN_ERR \ | |
2051 | "md: fatal superblock inconsistency in %s" | |
2052 | " -- removing from array\n", | |
2053 | bdevname(rdev->bdev,b)); | |
2054 | kick_rdev_from_array(rdev); | |
2055 | } | |
2056 | ||
2057 | ||
2058 | super_types[mddev->major_version]. | |
2059 | validate_super(mddev, freshest); | |
2060 | ||
2061 | i = 0; | |
2062 | ITERATE_RDEV(mddev,rdev,tmp) { | |
2063 | if (rdev != freshest) | |
2064 | if (super_types[mddev->major_version]. | |
2065 | validate_super(mddev, rdev)) { | |
2066 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2067 | " from array!\n", | |
2068 | bdevname(rdev->bdev,b)); | |
2069 | kick_rdev_from_array(rdev); | |
2070 | continue; | |
2071 | } | |
2072 | if (mddev->level == LEVEL_MULTIPATH) { | |
2073 | rdev->desc_nr = i++; | |
2074 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2075 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
2076 | } |
2077 | } | |
2078 | ||
2079 | ||
2080 | ||
2081 | if (mddev->recovery_cp != MaxSector && | |
2082 | mddev->level >= 1) | |
2083 | printk(KERN_ERR "md: %s: raid array is not clean" | |
2084 | " -- starting background reconstruction\n", | |
2085 | mdname(mddev)); | |
2086 | ||
1da177e4 LT |
2087 | } |
2088 | ||
16f17b39 N |
2089 | static ssize_t |
2090 | safe_delay_show(mddev_t *mddev, char *page) | |
2091 | { | |
2092 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2093 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2094 | } | |
2095 | static ssize_t | |
2096 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2097 | { | |
2098 | int scale=1; | |
2099 | int dot=0; | |
2100 | int i; | |
2101 | unsigned long msec; | |
2102 | char buf[30]; | |
2103 | char *e; | |
2104 | /* remove a period, and count digits after it */ | |
2105 | if (len >= sizeof(buf)) | |
2106 | return -EINVAL; | |
2107 | strlcpy(buf, cbuf, len); | |
2108 | buf[len] = 0; | |
2109 | for (i=0; i<len; i++) { | |
2110 | if (dot) { | |
2111 | if (isdigit(buf[i])) { | |
2112 | buf[i-1] = buf[i]; | |
2113 | scale *= 10; | |
2114 | } | |
2115 | buf[i] = 0; | |
2116 | } else if (buf[i] == '.') { | |
2117 | dot=1; | |
2118 | buf[i] = 0; | |
2119 | } | |
2120 | } | |
2121 | msec = simple_strtoul(buf, &e, 10); | |
2122 | if (e == buf || (*e && *e != '\n')) | |
2123 | return -EINVAL; | |
2124 | msec = (msec * 1000) / scale; | |
2125 | if (msec == 0) | |
2126 | mddev->safemode_delay = 0; | |
2127 | else { | |
2128 | mddev->safemode_delay = (msec*HZ)/1000; | |
2129 | if (mddev->safemode_delay == 0) | |
2130 | mddev->safemode_delay = 1; | |
2131 | } | |
2132 | return len; | |
2133 | } | |
2134 | static struct md_sysfs_entry md_safe_delay = | |
2135 | __ATTR(safe_mode_delay, 0644,safe_delay_show, safe_delay_store); | |
2136 | ||
eae1701f | 2137 | static ssize_t |
96de1e66 | 2138 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2139 | { |
2604b703 | 2140 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2141 | if (p) |
eae1701f | 2142 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2143 | else if (mddev->clevel[0]) |
2144 | return sprintf(page, "%s\n", mddev->clevel); | |
2145 | else if (mddev->level != LEVEL_NONE) | |
2146 | return sprintf(page, "%d\n", mddev->level); | |
2147 | else | |
2148 | return 0; | |
eae1701f N |
2149 | } |
2150 | ||
d9d166c2 N |
2151 | static ssize_t |
2152 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2153 | { | |
2154 | int rv = len; | |
2155 | if (mddev->pers) | |
2156 | return -EBUSY; | |
2157 | if (len == 0) | |
2158 | return 0; | |
2159 | if (len >= sizeof(mddev->clevel)) | |
2160 | return -ENOSPC; | |
2161 | strncpy(mddev->clevel, buf, len); | |
2162 | if (mddev->clevel[len-1] == '\n') | |
2163 | len--; | |
2164 | mddev->clevel[len] = 0; | |
2165 | mddev->level = LEVEL_NONE; | |
2166 | return rv; | |
2167 | } | |
2168 | ||
2169 | static struct md_sysfs_entry md_level = | |
2170 | __ATTR(level, 0644, level_show, level_store); | |
eae1701f | 2171 | |
d4dbd025 N |
2172 | |
2173 | static ssize_t | |
2174 | layout_show(mddev_t *mddev, char *page) | |
2175 | { | |
2176 | /* just a number, not meaningful for all levels */ | |
2177 | return sprintf(page, "%d\n", mddev->layout); | |
2178 | } | |
2179 | ||
2180 | static ssize_t | |
2181 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
2182 | { | |
2183 | char *e; | |
2184 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2185 | if (mddev->pers) | |
2186 | return -EBUSY; | |
2187 | ||
2188 | if (!*buf || (*e && *e != '\n')) | |
2189 | return -EINVAL; | |
2190 | ||
2191 | mddev->layout = n; | |
2192 | return len; | |
2193 | } | |
2194 | static struct md_sysfs_entry md_layout = | |
2195 | __ATTR(layout, 0655, layout_show, layout_store); | |
2196 | ||
2197 | ||
eae1701f | 2198 | static ssize_t |
96de1e66 | 2199 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 2200 | { |
bb636547 N |
2201 | if (mddev->raid_disks == 0) |
2202 | return 0; | |
eae1701f N |
2203 | return sprintf(page, "%d\n", mddev->raid_disks); |
2204 | } | |
2205 | ||
da943b99 N |
2206 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
2207 | ||
2208 | static ssize_t | |
2209 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
2210 | { | |
2211 | /* can only set raid_disks if array is not yet active */ | |
2212 | char *e; | |
2213 | int rv = 0; | |
2214 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2215 | ||
2216 | if (!*buf || (*e && *e != '\n')) | |
2217 | return -EINVAL; | |
2218 | ||
2219 | if (mddev->pers) | |
2220 | rv = update_raid_disks(mddev, n); | |
2221 | else | |
2222 | mddev->raid_disks = n; | |
2223 | return rv ? rv : len; | |
2224 | } | |
2225 | static struct md_sysfs_entry md_raid_disks = | |
2226 | __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store); | |
eae1701f | 2227 | |
3b34380a N |
2228 | static ssize_t |
2229 | chunk_size_show(mddev_t *mddev, char *page) | |
2230 | { | |
2231 | return sprintf(page, "%d\n", mddev->chunk_size); | |
2232 | } | |
2233 | ||
2234 | static ssize_t | |
2235 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
2236 | { | |
2237 | /* can only set chunk_size if array is not yet active */ | |
2238 | char *e; | |
2239 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2240 | ||
2241 | if (mddev->pers) | |
2242 | return -EBUSY; | |
2243 | if (!*buf || (*e && *e != '\n')) | |
2244 | return -EINVAL; | |
2245 | ||
2246 | mddev->chunk_size = n; | |
2247 | return len; | |
2248 | } | |
2249 | static struct md_sysfs_entry md_chunk_size = | |
2250 | __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store); | |
2251 | ||
a94213b1 N |
2252 | static ssize_t |
2253 | resync_start_show(mddev_t *mddev, char *page) | |
2254 | { | |
2255 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); | |
2256 | } | |
2257 | ||
2258 | static ssize_t | |
2259 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
2260 | { | |
2261 | /* can only set chunk_size if array is not yet active */ | |
2262 | char *e; | |
2263 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
2264 | ||
2265 | if (mddev->pers) | |
2266 | return -EBUSY; | |
2267 | if (!*buf || (*e && *e != '\n')) | |
2268 | return -EINVAL; | |
2269 | ||
2270 | mddev->recovery_cp = n; | |
2271 | return len; | |
2272 | } | |
2273 | static struct md_sysfs_entry md_resync_start = | |
2274 | __ATTR(resync_start, 0644, resync_start_show, resync_start_store); | |
2275 | ||
9e653b63 N |
2276 | /* |
2277 | * The array state can be: | |
2278 | * | |
2279 | * clear | |
2280 | * No devices, no size, no level | |
2281 | * Equivalent to STOP_ARRAY ioctl | |
2282 | * inactive | |
2283 | * May have some settings, but array is not active | |
2284 | * all IO results in error | |
2285 | * When written, doesn't tear down array, but just stops it | |
2286 | * suspended (not supported yet) | |
2287 | * All IO requests will block. The array can be reconfigured. | |
2288 | * Writing this, if accepted, will block until array is quiessent | |
2289 | * readonly | |
2290 | * no resync can happen. no superblocks get written. | |
2291 | * write requests fail | |
2292 | * read-auto | |
2293 | * like readonly, but behaves like 'clean' on a write request. | |
2294 | * | |
2295 | * clean - no pending writes, but otherwise active. | |
2296 | * When written to inactive array, starts without resync | |
2297 | * If a write request arrives then | |
2298 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
2299 | * if not known, block and switch to write-pending | |
2300 | * If written to an active array that has pending writes, then fails. | |
2301 | * active | |
2302 | * fully active: IO and resync can be happening. | |
2303 | * When written to inactive array, starts with resync | |
2304 | * | |
2305 | * write-pending | |
2306 | * clean, but writes are blocked waiting for 'active' to be written. | |
2307 | * | |
2308 | * active-idle | |
2309 | * like active, but no writes have been seen for a while (100msec). | |
2310 | * | |
2311 | */ | |
2312 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
2313 | write_pending, active_idle, bad_word}; | |
05381954 | 2314 | static char *array_states[] = { |
9e653b63 N |
2315 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
2316 | "write-pending", "active-idle", NULL }; | |
2317 | ||
2318 | static int match_word(const char *word, char **list) | |
2319 | { | |
2320 | int n; | |
2321 | for (n=0; list[n]; n++) | |
2322 | if (cmd_match(word, list[n])) | |
2323 | break; | |
2324 | return n; | |
2325 | } | |
2326 | ||
2327 | static ssize_t | |
2328 | array_state_show(mddev_t *mddev, char *page) | |
2329 | { | |
2330 | enum array_state st = inactive; | |
2331 | ||
2332 | if (mddev->pers) | |
2333 | switch(mddev->ro) { | |
2334 | case 1: | |
2335 | st = readonly; | |
2336 | break; | |
2337 | case 2: | |
2338 | st = read_auto; | |
2339 | break; | |
2340 | case 0: | |
2341 | if (mddev->in_sync) | |
2342 | st = clean; | |
2343 | else if (mddev->safemode) | |
2344 | st = active_idle; | |
2345 | else | |
2346 | st = active; | |
2347 | } | |
2348 | else { | |
2349 | if (list_empty(&mddev->disks) && | |
2350 | mddev->raid_disks == 0 && | |
2351 | mddev->size == 0) | |
2352 | st = clear; | |
2353 | else | |
2354 | st = inactive; | |
2355 | } | |
2356 | return sprintf(page, "%s\n", array_states[st]); | |
2357 | } | |
2358 | ||
2359 | static int do_md_stop(mddev_t * mddev, int ro); | |
2360 | static int do_md_run(mddev_t * mddev); | |
2361 | static int restart_array(mddev_t *mddev); | |
2362 | ||
2363 | static ssize_t | |
2364 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
2365 | { | |
2366 | int err = -EINVAL; | |
2367 | enum array_state st = match_word(buf, array_states); | |
2368 | switch(st) { | |
2369 | case bad_word: | |
2370 | break; | |
2371 | case clear: | |
2372 | /* stopping an active array */ | |
2373 | if (mddev->pers) { | |
2374 | if (atomic_read(&mddev->active) > 1) | |
2375 | return -EBUSY; | |
2376 | err = do_md_stop(mddev, 0); | |
2377 | } | |
2378 | break; | |
2379 | case inactive: | |
2380 | /* stopping an active array */ | |
2381 | if (mddev->pers) { | |
2382 | if (atomic_read(&mddev->active) > 1) | |
2383 | return -EBUSY; | |
2384 | err = do_md_stop(mddev, 2); | |
2385 | } | |
2386 | break; | |
2387 | case suspended: | |
2388 | break; /* not supported yet */ | |
2389 | case readonly: | |
2390 | if (mddev->pers) | |
2391 | err = do_md_stop(mddev, 1); | |
2392 | else { | |
2393 | mddev->ro = 1; | |
2394 | err = do_md_run(mddev); | |
2395 | } | |
2396 | break; | |
2397 | case read_auto: | |
2398 | /* stopping an active array */ | |
2399 | if (mddev->pers) { | |
2400 | err = do_md_stop(mddev, 1); | |
2401 | if (err == 0) | |
2402 | mddev->ro = 2; /* FIXME mark devices writable */ | |
2403 | } else { | |
2404 | mddev->ro = 2; | |
2405 | err = do_md_run(mddev); | |
2406 | } | |
2407 | break; | |
2408 | case clean: | |
2409 | if (mddev->pers) { | |
2410 | restart_array(mddev); | |
2411 | spin_lock_irq(&mddev->write_lock); | |
2412 | if (atomic_read(&mddev->writes_pending) == 0) { | |
2413 | mddev->in_sync = 1; | |
2414 | mddev->sb_dirty = 1; | |
2415 | } | |
2416 | spin_unlock_irq(&mddev->write_lock); | |
2417 | } else { | |
2418 | mddev->ro = 0; | |
2419 | mddev->recovery_cp = MaxSector; | |
2420 | err = do_md_run(mddev); | |
2421 | } | |
2422 | break; | |
2423 | case active: | |
2424 | if (mddev->pers) { | |
2425 | restart_array(mddev); | |
2426 | mddev->sb_dirty = 0; | |
2427 | wake_up(&mddev->sb_wait); | |
2428 | err = 0; | |
2429 | } else { | |
2430 | mddev->ro = 0; | |
2431 | err = do_md_run(mddev); | |
2432 | } | |
2433 | break; | |
2434 | case write_pending: | |
2435 | case active_idle: | |
2436 | /* these cannot be set */ | |
2437 | break; | |
2438 | } | |
2439 | if (err) | |
2440 | return err; | |
2441 | else | |
2442 | return len; | |
2443 | } | |
2444 | static struct md_sysfs_entry md_array_state = __ATTR(array_state, 0644, array_state_show, array_state_store); | |
2445 | ||
6d7ff738 N |
2446 | static ssize_t |
2447 | null_show(mddev_t *mddev, char *page) | |
2448 | { | |
2449 | return -EINVAL; | |
2450 | } | |
2451 | ||
2452 | static ssize_t | |
2453 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
2454 | { | |
2455 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
2456 | /* The new device is added to the array. | |
2457 | * If the array has a persistent superblock, we read the | |
2458 | * superblock to initialise info and check validity. | |
2459 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
2460 | * which mainly checks size. | |
2461 | */ | |
2462 | char *e; | |
2463 | int major = simple_strtoul(buf, &e, 10); | |
2464 | int minor; | |
2465 | dev_t dev; | |
2466 | mdk_rdev_t *rdev; | |
2467 | int err; | |
2468 | ||
2469 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
2470 | return -EINVAL; | |
2471 | minor = simple_strtoul(e+1, &e, 10); | |
2472 | if (*e && *e != '\n') | |
2473 | return -EINVAL; | |
2474 | dev = MKDEV(major, minor); | |
2475 | if (major != MAJOR(dev) || | |
2476 | minor != MINOR(dev)) | |
2477 | return -EOVERFLOW; | |
2478 | ||
2479 | ||
2480 | if (mddev->persistent) { | |
2481 | rdev = md_import_device(dev, mddev->major_version, | |
2482 | mddev->minor_version); | |
2483 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
2484 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
2485 | mdk_rdev_t, same_set); | |
2486 | err = super_types[mddev->major_version] | |
2487 | .load_super(rdev, rdev0, mddev->minor_version); | |
2488 | if (err < 0) | |
2489 | goto out; | |
2490 | } | |
2491 | } else | |
2492 | rdev = md_import_device(dev, -1, -1); | |
2493 | ||
2494 | if (IS_ERR(rdev)) | |
2495 | return PTR_ERR(rdev); | |
2496 | err = bind_rdev_to_array(rdev, mddev); | |
2497 | out: | |
2498 | if (err) | |
2499 | export_rdev(rdev); | |
2500 | return err ? err : len; | |
2501 | } | |
2502 | ||
2503 | static struct md_sysfs_entry md_new_device = | |
2504 | __ATTR(new_dev, 0200, null_show, new_dev_store); | |
3b34380a | 2505 | |
a35b0d69 N |
2506 | static ssize_t |
2507 | size_show(mddev_t *mddev, char *page) | |
2508 | { | |
2509 | return sprintf(page, "%llu\n", (unsigned long long)mddev->size); | |
2510 | } | |
2511 | ||
2512 | static int update_size(mddev_t *mddev, unsigned long size); | |
2513 | ||
2514 | static ssize_t | |
2515 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
2516 | { | |
2517 | /* If array is inactive, we can reduce the component size, but | |
2518 | * not increase it (except from 0). | |
2519 | * If array is active, we can try an on-line resize | |
2520 | */ | |
2521 | char *e; | |
2522 | int err = 0; | |
2523 | unsigned long long size = simple_strtoull(buf, &e, 10); | |
2524 | if (!*buf || *buf == '\n' || | |
2525 | (*e && *e != '\n')) | |
2526 | return -EINVAL; | |
2527 | ||
2528 | if (mddev->pers) { | |
2529 | err = update_size(mddev, size); | |
2530 | md_update_sb(mddev); | |
2531 | } else { | |
2532 | if (mddev->size == 0 || | |
2533 | mddev->size > size) | |
2534 | mddev->size = size; | |
2535 | else | |
2536 | err = -ENOSPC; | |
2537 | } | |
2538 | return err ? err : len; | |
2539 | } | |
2540 | ||
2541 | static struct md_sysfs_entry md_size = | |
2542 | __ATTR(component_size, 0644, size_show, size_store); | |
2543 | ||
8bb93aac N |
2544 | |
2545 | /* Metdata version. | |
2546 | * This is either 'none' for arrays with externally managed metadata, | |
2547 | * or N.M for internally known formats | |
2548 | */ | |
2549 | static ssize_t | |
2550 | metadata_show(mddev_t *mddev, char *page) | |
2551 | { | |
2552 | if (mddev->persistent) | |
2553 | return sprintf(page, "%d.%d\n", | |
2554 | mddev->major_version, mddev->minor_version); | |
2555 | else | |
2556 | return sprintf(page, "none\n"); | |
2557 | } | |
2558 | ||
2559 | static ssize_t | |
2560 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
2561 | { | |
2562 | int major, minor; | |
2563 | char *e; | |
2564 | if (!list_empty(&mddev->disks)) | |
2565 | return -EBUSY; | |
2566 | ||
2567 | if (cmd_match(buf, "none")) { | |
2568 | mddev->persistent = 0; | |
2569 | mddev->major_version = 0; | |
2570 | mddev->minor_version = 90; | |
2571 | return len; | |
2572 | } | |
2573 | major = simple_strtoul(buf, &e, 10); | |
2574 | if (e==buf || *e != '.') | |
2575 | return -EINVAL; | |
2576 | buf = e+1; | |
2577 | minor = simple_strtoul(buf, &e, 10); | |
2578 | if (e==buf || *e != '\n') | |
2579 | return -EINVAL; | |
2580 | if (major >= sizeof(super_types)/sizeof(super_types[0]) || | |
2581 | super_types[major].name == NULL) | |
2582 | return -ENOENT; | |
2583 | mddev->major_version = major; | |
2584 | mddev->minor_version = minor; | |
2585 | mddev->persistent = 1; | |
2586 | return len; | |
2587 | } | |
2588 | ||
2589 | static struct md_sysfs_entry md_metadata = | |
2590 | __ATTR(metadata_version, 0644, metadata_show, metadata_store); | |
2591 | ||
24dd469d | 2592 | static ssize_t |
7eec314d | 2593 | action_show(mddev_t *mddev, char *page) |
24dd469d | 2594 | { |
7eec314d | 2595 | char *type = "idle"; |
31399d9e N |
2596 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
2597 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) { | |
ccfcc3c1 N |
2598 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
2599 | type = "reshape"; | |
2600 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
2601 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
2602 | type = "resync"; | |
2603 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
2604 | type = "check"; | |
2605 | else | |
2606 | type = "repair"; | |
2607 | } else | |
2608 | type = "recover"; | |
2609 | } | |
2610 | return sprintf(page, "%s\n", type); | |
2611 | } | |
2612 | ||
2613 | static ssize_t | |
7eec314d | 2614 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 2615 | { |
7eec314d N |
2616 | if (!mddev->pers || !mddev->pers->sync_request) |
2617 | return -EINVAL; | |
2618 | ||
bce74dac | 2619 | if (cmd_match(page, "idle")) { |
7eec314d N |
2620 | if (mddev->sync_thread) { |
2621 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2622 | md_unregister_thread(mddev->sync_thread); | |
2623 | mddev->sync_thread = NULL; | |
2624 | mddev->recovery = 0; | |
2625 | } | |
03c902e1 N |
2626 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
2627 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 2628 | return -EBUSY; |
03c902e1 | 2629 | else if (cmd_match(page, "resync") || cmd_match(page, "recover")) |
7eec314d | 2630 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
16484bf5 N |
2631 | else if (cmd_match(page, "reshape")) { |
2632 | int err; | |
2633 | if (mddev->pers->start_reshape == NULL) | |
2634 | return -EINVAL; | |
2635 | err = mddev->pers->start_reshape(mddev); | |
2636 | if (err) | |
2637 | return err; | |
2638 | } else { | |
bce74dac | 2639 | if (cmd_match(page, "check")) |
7eec314d | 2640 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 2641 | else if (!cmd_match(page, "repair")) |
7eec314d N |
2642 | return -EINVAL; |
2643 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
2644 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 2645 | } |
03c902e1 | 2646 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d N |
2647 | md_wakeup_thread(mddev->thread); |
2648 | return len; | |
2649 | } | |
2650 | ||
9d88883e | 2651 | static ssize_t |
96de1e66 | 2652 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
2653 | { |
2654 | return sprintf(page, "%llu\n", | |
2655 | (unsigned long long) mddev->resync_mismatches); | |
2656 | } | |
2657 | ||
96de1e66 | 2658 | static struct md_sysfs_entry |
7eec314d | 2659 | md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
24dd469d | 2660 | |
96de1e66 N |
2661 | |
2662 | static struct md_sysfs_entry | |
2663 | md_mismatches = __ATTR_RO(mismatch_cnt); | |
9d88883e | 2664 | |
88202a0c N |
2665 | static ssize_t |
2666 | sync_min_show(mddev_t *mddev, char *page) | |
2667 | { | |
2668 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
2669 | mddev->sync_speed_min ? "local": "system"); | |
2670 | } | |
2671 | ||
2672 | static ssize_t | |
2673 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
2674 | { | |
2675 | int min; | |
2676 | char *e; | |
2677 | if (strncmp(buf, "system", 6)==0) { | |
2678 | mddev->sync_speed_min = 0; | |
2679 | return len; | |
2680 | } | |
2681 | min = simple_strtoul(buf, &e, 10); | |
2682 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
2683 | return -EINVAL; | |
2684 | mddev->sync_speed_min = min; | |
2685 | return len; | |
2686 | } | |
2687 | ||
2688 | static struct md_sysfs_entry md_sync_min = | |
2689 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
2690 | ||
2691 | static ssize_t | |
2692 | sync_max_show(mddev_t *mddev, char *page) | |
2693 | { | |
2694 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
2695 | mddev->sync_speed_max ? "local": "system"); | |
2696 | } | |
2697 | ||
2698 | static ssize_t | |
2699 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
2700 | { | |
2701 | int max; | |
2702 | char *e; | |
2703 | if (strncmp(buf, "system", 6)==0) { | |
2704 | mddev->sync_speed_max = 0; | |
2705 | return len; | |
2706 | } | |
2707 | max = simple_strtoul(buf, &e, 10); | |
2708 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
2709 | return -EINVAL; | |
2710 | mddev->sync_speed_max = max; | |
2711 | return len; | |
2712 | } | |
2713 | ||
2714 | static struct md_sysfs_entry md_sync_max = | |
2715 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
2716 | ||
2717 | ||
2718 | static ssize_t | |
2719 | sync_speed_show(mddev_t *mddev, char *page) | |
2720 | { | |
2721 | unsigned long resync, dt, db; | |
2722 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active)); | |
2723 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
2724 | if (!dt) dt++; | |
2725 | db = resync - (mddev->resync_mark_cnt); | |
2726 | return sprintf(page, "%ld\n", db/dt/2); /* K/sec */ | |
2727 | } | |
2728 | ||
2729 | static struct md_sysfs_entry | |
2730 | md_sync_speed = __ATTR_RO(sync_speed); | |
2731 | ||
2732 | static ssize_t | |
2733 | sync_completed_show(mddev_t *mddev, char *page) | |
2734 | { | |
2735 | unsigned long max_blocks, resync; | |
2736 | ||
2737 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2738 | max_blocks = mddev->resync_max_sectors; | |
2739 | else | |
2740 | max_blocks = mddev->size << 1; | |
2741 | ||
2742 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active)); | |
2743 | return sprintf(page, "%lu / %lu\n", resync, max_blocks); | |
2744 | } | |
2745 | ||
2746 | static struct md_sysfs_entry | |
2747 | md_sync_completed = __ATTR_RO(sync_completed); | |
2748 | ||
e464eafd N |
2749 | static ssize_t |
2750 | suspend_lo_show(mddev_t *mddev, char *page) | |
2751 | { | |
2752 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
2753 | } | |
2754 | ||
2755 | static ssize_t | |
2756 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
2757 | { | |
2758 | char *e; | |
2759 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
2760 | ||
2761 | if (mddev->pers->quiesce == NULL) | |
2762 | return -EINVAL; | |
2763 | if (buf == e || (*e && *e != '\n')) | |
2764 | return -EINVAL; | |
2765 | if (new >= mddev->suspend_hi || | |
2766 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
2767 | mddev->suspend_lo = new; | |
2768 | mddev->pers->quiesce(mddev, 2); | |
2769 | return len; | |
2770 | } else | |
2771 | return -EINVAL; | |
2772 | } | |
2773 | static struct md_sysfs_entry md_suspend_lo = | |
2774 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
2775 | ||
2776 | ||
2777 | static ssize_t | |
2778 | suspend_hi_show(mddev_t *mddev, char *page) | |
2779 | { | |
2780 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
2781 | } | |
2782 | ||
2783 | static ssize_t | |
2784 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
2785 | { | |
2786 | char *e; | |
2787 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
2788 | ||
2789 | if (mddev->pers->quiesce == NULL) | |
2790 | return -EINVAL; | |
2791 | if (buf == e || (*e && *e != '\n')) | |
2792 | return -EINVAL; | |
2793 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
2794 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
2795 | mddev->suspend_hi = new; | |
2796 | mddev->pers->quiesce(mddev, 1); | |
2797 | mddev->pers->quiesce(mddev, 0); | |
2798 | return len; | |
2799 | } else | |
2800 | return -EINVAL; | |
2801 | } | |
2802 | static struct md_sysfs_entry md_suspend_hi = | |
2803 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
2804 | ||
2805 | ||
eae1701f N |
2806 | static struct attribute *md_default_attrs[] = { |
2807 | &md_level.attr, | |
d4dbd025 | 2808 | &md_layout.attr, |
eae1701f | 2809 | &md_raid_disks.attr, |
3b34380a | 2810 | &md_chunk_size.attr, |
a35b0d69 | 2811 | &md_size.attr, |
a94213b1 | 2812 | &md_resync_start.attr, |
8bb93aac | 2813 | &md_metadata.attr, |
6d7ff738 | 2814 | &md_new_device.attr, |
16f17b39 | 2815 | &md_safe_delay.attr, |
9e653b63 | 2816 | &md_array_state.attr, |
411036fa N |
2817 | NULL, |
2818 | }; | |
2819 | ||
2820 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 2821 | &md_scan_mode.attr, |
9d88883e | 2822 | &md_mismatches.attr, |
88202a0c N |
2823 | &md_sync_min.attr, |
2824 | &md_sync_max.attr, | |
2825 | &md_sync_speed.attr, | |
2826 | &md_sync_completed.attr, | |
e464eafd N |
2827 | &md_suspend_lo.attr, |
2828 | &md_suspend_hi.attr, | |
eae1701f N |
2829 | NULL, |
2830 | }; | |
411036fa N |
2831 | static struct attribute_group md_redundancy_group = { |
2832 | .name = NULL, | |
2833 | .attrs = md_redundancy_attrs, | |
2834 | }; | |
2835 | ||
eae1701f N |
2836 | |
2837 | static ssize_t | |
2838 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2839 | { | |
2840 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
2841 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 2842 | ssize_t rv; |
eae1701f N |
2843 | |
2844 | if (!entry->show) | |
2845 | return -EIO; | |
5dc5cf7d IM |
2846 | rv = mddev_lock(mddev); |
2847 | if (!rv) { | |
2848 | rv = entry->show(mddev, page); | |
2849 | mddev_unlock(mddev); | |
2850 | } | |
96de1e66 | 2851 | return rv; |
eae1701f N |
2852 | } |
2853 | ||
2854 | static ssize_t | |
2855 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
2856 | const char *page, size_t length) | |
2857 | { | |
2858 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
2859 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 2860 | ssize_t rv; |
eae1701f N |
2861 | |
2862 | if (!entry->store) | |
2863 | return -EIO; | |
5dc5cf7d IM |
2864 | rv = mddev_lock(mddev); |
2865 | if (!rv) { | |
2866 | rv = entry->store(mddev, page, length); | |
2867 | mddev_unlock(mddev); | |
2868 | } | |
96de1e66 | 2869 | return rv; |
eae1701f N |
2870 | } |
2871 | ||
2872 | static void md_free(struct kobject *ko) | |
2873 | { | |
2874 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
2875 | kfree(mddev); | |
2876 | } | |
2877 | ||
2878 | static struct sysfs_ops md_sysfs_ops = { | |
2879 | .show = md_attr_show, | |
2880 | .store = md_attr_store, | |
2881 | }; | |
2882 | static struct kobj_type md_ktype = { | |
2883 | .release = md_free, | |
2884 | .sysfs_ops = &md_sysfs_ops, | |
2885 | .default_attrs = md_default_attrs, | |
2886 | }; | |
2887 | ||
1da177e4 LT |
2888 | int mdp_major = 0; |
2889 | ||
2890 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
2891 | { | |
48c9c27b | 2892 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
2893 | mddev_t *mddev = mddev_find(dev); |
2894 | struct gendisk *disk; | |
2895 | int partitioned = (MAJOR(dev) != MD_MAJOR); | |
2896 | int shift = partitioned ? MdpMinorShift : 0; | |
2897 | int unit = MINOR(dev) >> shift; | |
2898 | ||
2899 | if (!mddev) | |
2900 | return NULL; | |
2901 | ||
48c9c27b | 2902 | mutex_lock(&disks_mutex); |
1da177e4 | 2903 | if (mddev->gendisk) { |
48c9c27b | 2904 | mutex_unlock(&disks_mutex); |
1da177e4 LT |
2905 | mddev_put(mddev); |
2906 | return NULL; | |
2907 | } | |
2908 | disk = alloc_disk(1 << shift); | |
2909 | if (!disk) { | |
48c9c27b | 2910 | mutex_unlock(&disks_mutex); |
1da177e4 LT |
2911 | mddev_put(mddev); |
2912 | return NULL; | |
2913 | } | |
2914 | disk->major = MAJOR(dev); | |
2915 | disk->first_minor = unit << shift; | |
ce7b0f46 | 2916 | if (partitioned) |
1da177e4 | 2917 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 2918 | else |
1da177e4 | 2919 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
2920 | disk->fops = &md_fops; |
2921 | disk->private_data = mddev; | |
2922 | disk->queue = mddev->queue; | |
2923 | add_disk(disk); | |
2924 | mddev->gendisk = disk; | |
48c9c27b | 2925 | mutex_unlock(&disks_mutex); |
9c791977 | 2926 | mddev->kobj.parent = &disk->kobj; |
eae1701f N |
2927 | mddev->kobj.k_name = NULL; |
2928 | snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md"); | |
2929 | mddev->kobj.ktype = &md_ktype; | |
2930 | kobject_register(&mddev->kobj); | |
1da177e4 LT |
2931 | return NULL; |
2932 | } | |
2933 | ||
1da177e4 LT |
2934 | static void md_safemode_timeout(unsigned long data) |
2935 | { | |
2936 | mddev_t *mddev = (mddev_t *) data; | |
2937 | ||
2938 | mddev->safemode = 1; | |
2939 | md_wakeup_thread(mddev->thread); | |
2940 | } | |
2941 | ||
6ff8d8ec | 2942 | static int start_dirty_degraded; |
1da177e4 LT |
2943 | |
2944 | static int do_md_run(mddev_t * mddev) | |
2945 | { | |
2604b703 | 2946 | int err; |
1da177e4 LT |
2947 | int chunk_size; |
2948 | struct list_head *tmp; | |
2949 | mdk_rdev_t *rdev; | |
2950 | struct gendisk *disk; | |
2604b703 | 2951 | struct mdk_personality *pers; |
1da177e4 LT |
2952 | char b[BDEVNAME_SIZE]; |
2953 | ||
a757e64c N |
2954 | if (list_empty(&mddev->disks)) |
2955 | /* cannot run an array with no devices.. */ | |
1da177e4 | 2956 | return -EINVAL; |
1da177e4 LT |
2957 | |
2958 | if (mddev->pers) | |
2959 | return -EBUSY; | |
2960 | ||
2961 | /* | |
2962 | * Analyze all RAID superblock(s) | |
2963 | */ | |
a757e64c N |
2964 | if (!mddev->raid_disks) |
2965 | analyze_sbs(mddev); | |
1da177e4 LT |
2966 | |
2967 | chunk_size = mddev->chunk_size; | |
2604b703 N |
2968 | |
2969 | if (chunk_size) { | |
1da177e4 LT |
2970 | if (chunk_size > MAX_CHUNK_SIZE) { |
2971 | printk(KERN_ERR "too big chunk_size: %d > %d\n", | |
2972 | chunk_size, MAX_CHUNK_SIZE); | |
2973 | return -EINVAL; | |
2974 | } | |
2975 | /* | |
2976 | * chunk-size has to be a power of 2 and multiples of PAGE_SIZE | |
2977 | */ | |
2978 | if ( (1 << ffz(~chunk_size)) != chunk_size) { | |
a757e64c | 2979 | printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size); |
1da177e4 LT |
2980 | return -EINVAL; |
2981 | } | |
2982 | if (chunk_size < PAGE_SIZE) { | |
2983 | printk(KERN_ERR "too small chunk_size: %d < %ld\n", | |
2984 | chunk_size, PAGE_SIZE); | |
2985 | return -EINVAL; | |
2986 | } | |
2987 | ||
2988 | /* devices must have minimum size of one chunk */ | |
2989 | ITERATE_RDEV(mddev,rdev,tmp) { | |
b2d444d7 | 2990 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2991 | continue; |
2992 | if (rdev->size < chunk_size / 1024) { | |
2993 | printk(KERN_WARNING | |
2994 | "md: Dev %s smaller than chunk_size:" | |
2995 | " %lluk < %dk\n", | |
2996 | bdevname(rdev->bdev,b), | |
2997 | (unsigned long long)rdev->size, | |
2998 | chunk_size / 1024); | |
2999 | return -EINVAL; | |
3000 | } | |
3001 | } | |
3002 | } | |
3003 | ||
1da177e4 | 3004 | #ifdef CONFIG_KMOD |
d9d166c2 N |
3005 | if (mddev->level != LEVEL_NONE) |
3006 | request_module("md-level-%d", mddev->level); | |
3007 | else if (mddev->clevel[0]) | |
3008 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
3009 | #endif |
3010 | ||
3011 | /* | |
3012 | * Drop all container device buffers, from now on | |
3013 | * the only valid external interface is through the md | |
3014 | * device. | |
3015 | * Also find largest hardsector size | |
3016 | */ | |
3017 | ITERATE_RDEV(mddev,rdev,tmp) { | |
b2d444d7 | 3018 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
3019 | continue; |
3020 | sync_blockdev(rdev->bdev); | |
3021 | invalidate_bdev(rdev->bdev, 0); | |
3022 | } | |
3023 | ||
3024 | md_probe(mddev->unit, NULL, NULL); | |
3025 | disk = mddev->gendisk; | |
3026 | if (!disk) | |
3027 | return -ENOMEM; | |
3028 | ||
3029 | spin_lock(&pers_lock); | |
d9d166c2 | 3030 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 3031 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 3032 | spin_unlock(&pers_lock); |
d9d166c2 N |
3033 | if (mddev->level != LEVEL_NONE) |
3034 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
3035 | mddev->level); | |
3036 | else | |
3037 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
3038 | mddev->clevel); | |
1da177e4 LT |
3039 | return -EINVAL; |
3040 | } | |
2604b703 | 3041 | mddev->pers = pers; |
1da177e4 | 3042 | spin_unlock(&pers_lock); |
d9d166c2 N |
3043 | mddev->level = pers->level; |
3044 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
1da177e4 | 3045 | |
f6705578 | 3046 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 3047 | pers->start_reshape == NULL) { |
f6705578 N |
3048 | /* This personality cannot handle reshaping... */ |
3049 | mddev->pers = NULL; | |
3050 | module_put(pers->owner); | |
3051 | return -EINVAL; | |
3052 | } | |
3053 | ||
657390d2 | 3054 | mddev->recovery = 0; |
1da177e4 | 3055 | mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */ |
a9701a30 | 3056 | mddev->barriers_work = 1; |
6ff8d8ec | 3057 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 3058 | |
f91de92e N |
3059 | if (start_readonly) |
3060 | mddev->ro = 2; /* read-only, but switch on first write */ | |
3061 | ||
b15c2e57 N |
3062 | err = mddev->pers->run(mddev); |
3063 | if (!err && mddev->pers->sync_request) { | |
3064 | err = bitmap_create(mddev); | |
3065 | if (err) { | |
3066 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
3067 | mdname(mddev), err); | |
3068 | mddev->pers->stop(mddev); | |
3069 | } | |
3070 | } | |
1da177e4 LT |
3071 | if (err) { |
3072 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
3073 | module_put(mddev->pers->owner); | |
3074 | mddev->pers = NULL; | |
32a7627c N |
3075 | bitmap_destroy(mddev); |
3076 | return err; | |
1da177e4 | 3077 | } |
411036fa N |
3078 | if (mddev->pers->sync_request) |
3079 | sysfs_create_group(&mddev->kobj, &md_redundancy_group); | |
fd9d49ca N |
3080 | else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
3081 | mddev->ro = 0; | |
3082 | ||
1da177e4 LT |
3083 | atomic_set(&mddev->writes_pending,0); |
3084 | mddev->safemode = 0; | |
3085 | mddev->safemode_timer.function = md_safemode_timeout; | |
3086 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 3087 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 3088 | mddev->in_sync = 1; |
86e6ffdd N |
3089 | |
3090 | ITERATE_RDEV(mddev,rdev,tmp) | |
3091 | if (rdev->raid_disk >= 0) { | |
3092 | char nm[20]; | |
3093 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3094 | sysfs_create_link(&mddev->kobj, &rdev->kobj, nm); | |
3095 | } | |
1da177e4 LT |
3096 | |
3097 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3098 | ||
3099 | if (mddev->sb_dirty) | |
3100 | md_update_sb(mddev); | |
3101 | ||
3102 | set_capacity(disk, mddev->array_size<<1); | |
3103 | ||
3104 | /* If we call blk_queue_make_request here, it will | |
3105 | * re-initialise max_sectors etc which may have been | |
3106 | * refined inside -> run. So just set the bits we need to set. | |
3107 | * Most initialisation happended when we called | |
3108 | * blk_queue_make_request(..., md_fail_request) | |
3109 | * earlier. | |
3110 | */ | |
3111 | mddev->queue->queuedata = mddev; | |
3112 | mddev->queue->make_request_fn = mddev->pers->make_request; | |
3113 | ||
5fd6c1dc N |
3114 | /* If there is a partially-recovered drive we need to |
3115 | * start recovery here. If we leave it to md_check_recovery, | |
3116 | * it will remove the drives and not do the right thing | |
3117 | */ | |
0b8c9de0 | 3118 | if (mddev->degraded && !mddev->sync_thread) { |
5fd6c1dc N |
3119 | struct list_head *rtmp; |
3120 | int spares = 0; | |
3121 | ITERATE_RDEV(mddev,rdev,rtmp) | |
3122 | if (rdev->raid_disk >= 0 && | |
3123 | !test_bit(In_sync, &rdev->flags) && | |
3124 | !test_bit(Faulty, &rdev->flags)) | |
3125 | /* complete an interrupted recovery */ | |
3126 | spares++; | |
3127 | if (spares && mddev->pers->sync_request) { | |
3128 | mddev->recovery = 0; | |
3129 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3130 | mddev->sync_thread = md_register_thread(md_do_sync, | |
3131 | mddev, | |
3132 | "%s_resync"); | |
3133 | if (!mddev->sync_thread) { | |
3134 | printk(KERN_ERR "%s: could not start resync" | |
3135 | " thread...\n", | |
3136 | mdname(mddev)); | |
3137 | /* leave the spares where they are, it shouldn't hurt */ | |
3138 | mddev->recovery = 0; | |
0b8c9de0 | 3139 | } |
5fd6c1dc N |
3140 | } |
3141 | } | |
0b8c9de0 N |
3142 | md_wakeup_thread(mddev->thread); |
3143 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 3144 | |
1da177e4 | 3145 | mddev->changed = 1; |
d7603b7e | 3146 | md_new_event(mddev); |
1da177e4 LT |
3147 | return 0; |
3148 | } | |
3149 | ||
3150 | static int restart_array(mddev_t *mddev) | |
3151 | { | |
3152 | struct gendisk *disk = mddev->gendisk; | |
3153 | int err; | |
3154 | ||
3155 | /* | |
3156 | * Complain if it has no devices | |
3157 | */ | |
3158 | err = -ENXIO; | |
3159 | if (list_empty(&mddev->disks)) | |
3160 | goto out; | |
3161 | ||
3162 | if (mddev->pers) { | |
3163 | err = -EBUSY; | |
3164 | if (!mddev->ro) | |
3165 | goto out; | |
3166 | ||
3167 | mddev->safemode = 0; | |
3168 | mddev->ro = 0; | |
3169 | set_disk_ro(disk, 0); | |
3170 | ||
3171 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
3172 | mdname(mddev)); | |
3173 | /* | |
3174 | * Kick recovery or resync if necessary | |
3175 | */ | |
3176 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3177 | md_wakeup_thread(mddev->thread); | |
5fd6c1dc | 3178 | md_wakeup_thread(mddev->sync_thread); |
1da177e4 | 3179 | err = 0; |
9e653b63 | 3180 | } else |
1da177e4 | 3181 | err = -EINVAL; |
1da177e4 LT |
3182 | |
3183 | out: | |
3184 | return err; | |
3185 | } | |
3186 | ||
acc55e22 N |
3187 | /* similar to deny_write_access, but accounts for our holding a reference |
3188 | * to the file ourselves */ | |
3189 | static int deny_bitmap_write_access(struct file * file) | |
3190 | { | |
3191 | struct inode *inode = file->f_mapping->host; | |
3192 | ||
3193 | spin_lock(&inode->i_lock); | |
3194 | if (atomic_read(&inode->i_writecount) > 1) { | |
3195 | spin_unlock(&inode->i_lock); | |
3196 | return -ETXTBSY; | |
3197 | } | |
3198 | atomic_set(&inode->i_writecount, -1); | |
3199 | spin_unlock(&inode->i_lock); | |
3200 | ||
3201 | return 0; | |
3202 | } | |
3203 | ||
3204 | static void restore_bitmap_write_access(struct file *file) | |
3205 | { | |
3206 | struct inode *inode = file->f_mapping->host; | |
3207 | ||
3208 | spin_lock(&inode->i_lock); | |
3209 | atomic_set(&inode->i_writecount, 1); | |
3210 | spin_unlock(&inode->i_lock); | |
3211 | } | |
3212 | ||
9e653b63 N |
3213 | /* mode: |
3214 | * 0 - completely stop and dis-assemble array | |
3215 | * 1 - switch to readonly | |
3216 | * 2 - stop but do not disassemble array | |
3217 | */ | |
3218 | static int do_md_stop(mddev_t * mddev, int mode) | |
1da177e4 LT |
3219 | { |
3220 | int err = 0; | |
3221 | struct gendisk *disk = mddev->gendisk; | |
3222 | ||
3223 | if (mddev->pers) { | |
3224 | if (atomic_read(&mddev->active)>2) { | |
3225 | printk("md: %s still in use.\n",mdname(mddev)); | |
3226 | return -EBUSY; | |
3227 | } | |
3228 | ||
3229 | if (mddev->sync_thread) { | |
5fd6c1dc | 3230 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 LT |
3231 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
3232 | md_unregister_thread(mddev->sync_thread); | |
3233 | mddev->sync_thread = NULL; | |
3234 | } | |
3235 | ||
3236 | del_timer_sync(&mddev->safemode_timer); | |
3237 | ||
3238 | invalidate_partition(disk, 0); | |
3239 | ||
9e653b63 N |
3240 | switch(mode) { |
3241 | case 1: /* readonly */ | |
1da177e4 | 3242 | err = -ENXIO; |
f91de92e | 3243 | if (mddev->ro==1) |
1da177e4 LT |
3244 | goto out; |
3245 | mddev->ro = 1; | |
9e653b63 N |
3246 | break; |
3247 | case 0: /* disassemble */ | |
3248 | case 2: /* stop */ | |
6b8b3e8a | 3249 | bitmap_flush(mddev); |
a9701a30 | 3250 | md_super_wait(mddev); |
1da177e4 LT |
3251 | if (mddev->ro) |
3252 | set_disk_ro(disk, 0); | |
3253 | blk_queue_make_request(mddev->queue, md_fail_request); | |
3254 | mddev->pers->stop(mddev); | |
411036fa N |
3255 | if (mddev->pers->sync_request) |
3256 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
3257 | ||
1da177e4 LT |
3258 | module_put(mddev->pers->owner); |
3259 | mddev->pers = NULL; | |
3260 | if (mddev->ro) | |
3261 | mddev->ro = 0; | |
3262 | } | |
5fd6c1dc | 3263 | if (!mddev->in_sync || mddev->sb_dirty) { |
1da177e4 LT |
3264 | /* mark array as shutdown cleanly */ |
3265 | mddev->in_sync = 1; | |
3266 | md_update_sb(mddev); | |
3267 | } | |
9e653b63 | 3268 | if (mode == 1) |
1da177e4 | 3269 | set_disk_ro(disk, 1); |
5fd6c1dc | 3270 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 | 3271 | } |
32a7627c | 3272 | |
1da177e4 LT |
3273 | /* |
3274 | * Free resources if final stop | |
3275 | */ | |
9e653b63 | 3276 | if (mode == 0) { |
86e6ffdd N |
3277 | mdk_rdev_t *rdev; |
3278 | struct list_head *tmp; | |
1da177e4 LT |
3279 | struct gendisk *disk; |
3280 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); | |
3281 | ||
978f946b N |
3282 | bitmap_destroy(mddev); |
3283 | if (mddev->bitmap_file) { | |
acc55e22 | 3284 | restore_bitmap_write_access(mddev->bitmap_file); |
978f946b N |
3285 | fput(mddev->bitmap_file); |
3286 | mddev->bitmap_file = NULL; | |
3287 | } | |
3288 | mddev->bitmap_offset = 0; | |
3289 | ||
86e6ffdd N |
3290 | ITERATE_RDEV(mddev,rdev,tmp) |
3291 | if (rdev->raid_disk >= 0) { | |
3292 | char nm[20]; | |
3293 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3294 | sysfs_remove_link(&mddev->kobj, nm); | |
3295 | } | |
3296 | ||
1da177e4 LT |
3297 | export_array(mddev); |
3298 | ||
3299 | mddev->array_size = 0; | |
9e653b63 N |
3300 | mddev->size = 0; |
3301 | mddev->raid_disks = 0; | |
a94213b1 | 3302 | mddev->recovery_cp = 0; |
9e653b63 | 3303 | |
1da177e4 LT |
3304 | disk = mddev->gendisk; |
3305 | if (disk) | |
3306 | set_capacity(disk, 0); | |
3307 | mddev->changed = 1; | |
a8a55c38 | 3308 | } else if (mddev->pers) |
1da177e4 LT |
3309 | printk(KERN_INFO "md: %s switched to read-only mode.\n", |
3310 | mdname(mddev)); | |
3311 | err = 0; | |
d7603b7e | 3312 | md_new_event(mddev); |
1da177e4 LT |
3313 | out: |
3314 | return err; | |
3315 | } | |
3316 | ||
3317 | static void autorun_array(mddev_t *mddev) | |
3318 | { | |
3319 | mdk_rdev_t *rdev; | |
3320 | struct list_head *tmp; | |
3321 | int err; | |
3322 | ||
a757e64c | 3323 | if (list_empty(&mddev->disks)) |
1da177e4 | 3324 | return; |
1da177e4 LT |
3325 | |
3326 | printk(KERN_INFO "md: running: "); | |
3327 | ||
3328 | ITERATE_RDEV(mddev,rdev,tmp) { | |
3329 | char b[BDEVNAME_SIZE]; | |
3330 | printk("<%s>", bdevname(rdev->bdev,b)); | |
3331 | } | |
3332 | printk("\n"); | |
3333 | ||
3334 | err = do_md_run (mddev); | |
3335 | if (err) { | |
3336 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
3337 | do_md_stop (mddev, 0); | |
3338 | } | |
3339 | } | |
3340 | ||
3341 | /* | |
3342 | * lets try to run arrays based on all disks that have arrived | |
3343 | * until now. (those are in pending_raid_disks) | |
3344 | * | |
3345 | * the method: pick the first pending disk, collect all disks with | |
3346 | * the same UUID, remove all from the pending list and put them into | |
3347 | * the 'same_array' list. Then order this list based on superblock | |
3348 | * update time (freshest comes first), kick out 'old' disks and | |
3349 | * compare superblocks. If everything's fine then run it. | |
3350 | * | |
3351 | * If "unit" is allocated, then bump its reference count | |
3352 | */ | |
3353 | static void autorun_devices(int part) | |
3354 | { | |
1da177e4 LT |
3355 | struct list_head *tmp; |
3356 | mdk_rdev_t *rdev0, *rdev; | |
3357 | mddev_t *mddev; | |
3358 | char b[BDEVNAME_SIZE]; | |
3359 | ||
3360 | printk(KERN_INFO "md: autorun ...\n"); | |
3361 | while (!list_empty(&pending_raid_disks)) { | |
3362 | dev_t dev; | |
ad01c9e3 | 3363 | LIST_HEAD(candidates); |
1da177e4 LT |
3364 | rdev0 = list_entry(pending_raid_disks.next, |
3365 | mdk_rdev_t, same_set); | |
3366 | ||
3367 | printk(KERN_INFO "md: considering %s ...\n", | |
3368 | bdevname(rdev0->bdev,b)); | |
3369 | INIT_LIST_HEAD(&candidates); | |
3370 | ITERATE_RDEV_PENDING(rdev,tmp) | |
3371 | if (super_90_load(rdev, rdev0, 0) >= 0) { | |
3372 | printk(KERN_INFO "md: adding %s ...\n", | |
3373 | bdevname(rdev->bdev,b)); | |
3374 | list_move(&rdev->same_set, &candidates); | |
3375 | } | |
3376 | /* | |
3377 | * now we have a set of devices, with all of them having | |
3378 | * mostly sane superblocks. It's time to allocate the | |
3379 | * mddev. | |
3380 | */ | |
3381 | if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) { | |
3382 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", | |
3383 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
3384 | break; | |
3385 | } | |
3386 | if (part) | |
3387 | dev = MKDEV(mdp_major, | |
3388 | rdev0->preferred_minor << MdpMinorShift); | |
3389 | else | |
3390 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
3391 | ||
3392 | md_probe(dev, NULL, NULL); | |
3393 | mddev = mddev_find(dev); | |
3394 | if (!mddev) { | |
3395 | printk(KERN_ERR | |
3396 | "md: cannot allocate memory for md drive.\n"); | |
3397 | break; | |
3398 | } | |
3399 | if (mddev_lock(mddev)) | |
3400 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
3401 | mdname(mddev)); | |
3402 | else if (mddev->raid_disks || mddev->major_version | |
3403 | || !list_empty(&mddev->disks)) { | |
3404 | printk(KERN_WARNING | |
3405 | "md: %s already running, cannot run %s\n", | |
3406 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
3407 | mddev_unlock(mddev); | |
3408 | } else { | |
3409 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
3410 | ITERATE_RDEV_GENERIC(candidates,rdev,tmp) { | |
3411 | list_del_init(&rdev->same_set); | |
3412 | if (bind_rdev_to_array(rdev, mddev)) | |
3413 | export_rdev(rdev); | |
3414 | } | |
3415 | autorun_array(mddev); | |
3416 | mddev_unlock(mddev); | |
3417 | } | |
3418 | /* on success, candidates will be empty, on error | |
3419 | * it won't... | |
3420 | */ | |
3421 | ITERATE_RDEV_GENERIC(candidates,rdev,tmp) | |
3422 | export_rdev(rdev); | |
3423 | mddev_put(mddev); | |
3424 | } | |
3425 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
3426 | } | |
3427 | ||
3428 | /* | |
3429 | * import RAID devices based on one partition | |
3430 | * if possible, the array gets run as well. | |
3431 | */ | |
3432 | ||
3433 | static int autostart_array(dev_t startdev) | |
3434 | { | |
3435 | char b[BDEVNAME_SIZE]; | |
3436 | int err = -EINVAL, i; | |
3437 | mdp_super_t *sb = NULL; | |
3438 | mdk_rdev_t *start_rdev = NULL, *rdev; | |
3439 | ||
3440 | start_rdev = md_import_device(startdev, 0, 0); | |
3441 | if (IS_ERR(start_rdev)) | |
3442 | return err; | |
3443 | ||
3444 | ||
3445 | /* NOTE: this can only work for 0.90.0 superblocks */ | |
3446 | sb = (mdp_super_t*)page_address(start_rdev->sb_page); | |
3447 | if (sb->major_version != 0 || | |
3448 | sb->minor_version != 90 ) { | |
3449 | printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n"); | |
3450 | export_rdev(start_rdev); | |
3451 | return err; | |
3452 | } | |
3453 | ||
b2d444d7 | 3454 | if (test_bit(Faulty, &start_rdev->flags)) { |
1da177e4 LT |
3455 | printk(KERN_WARNING |
3456 | "md: can not autostart based on faulty %s!\n", | |
3457 | bdevname(start_rdev->bdev,b)); | |
3458 | export_rdev(start_rdev); | |
3459 | return err; | |
3460 | } | |
3461 | list_add(&start_rdev->same_set, &pending_raid_disks); | |
3462 | ||
3463 | for (i = 0; i < MD_SB_DISKS; i++) { | |
3464 | mdp_disk_t *desc = sb->disks + i; | |
3465 | dev_t dev = MKDEV(desc->major, desc->minor); | |
3466 | ||
3467 | if (!dev) | |
3468 | continue; | |
3469 | if (dev == startdev) | |
3470 | continue; | |
3471 | if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor) | |
3472 | continue; | |
3473 | rdev = md_import_device(dev, 0, 0); | |
3474 | if (IS_ERR(rdev)) | |
3475 | continue; | |
3476 | ||
3477 | list_add(&rdev->same_set, &pending_raid_disks); | |
3478 | } | |
3479 | ||
3480 | /* | |
3481 | * possibly return codes | |
3482 | */ | |
3483 | autorun_devices(0); | |
3484 | return 0; | |
3485 | ||
3486 | } | |
3487 | ||
3488 | ||
3489 | static int get_version(void __user * arg) | |
3490 | { | |
3491 | mdu_version_t ver; | |
3492 | ||
3493 | ver.major = MD_MAJOR_VERSION; | |
3494 | ver.minor = MD_MINOR_VERSION; | |
3495 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
3496 | ||
3497 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
3498 | return -EFAULT; | |
3499 | ||
3500 | return 0; | |
3501 | } | |
3502 | ||
3503 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
3504 | { | |
3505 | mdu_array_info_t info; | |
3506 | int nr,working,active,failed,spare; | |
3507 | mdk_rdev_t *rdev; | |
3508 | struct list_head *tmp; | |
3509 | ||
3510 | nr=working=active=failed=spare=0; | |
3511 | ITERATE_RDEV(mddev,rdev,tmp) { | |
3512 | nr++; | |
b2d444d7 | 3513 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
3514 | failed++; |
3515 | else { | |
3516 | working++; | |
b2d444d7 | 3517 | if (test_bit(In_sync, &rdev->flags)) |
1da177e4 LT |
3518 | active++; |
3519 | else | |
3520 | spare++; | |
3521 | } | |
3522 | } | |
3523 | ||
3524 | info.major_version = mddev->major_version; | |
3525 | info.minor_version = mddev->minor_version; | |
3526 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
3527 | info.ctime = mddev->ctime; | |
3528 | info.level = mddev->level; | |
3529 | info.size = mddev->size; | |
284ae7ca N |
3530 | if (info.size != mddev->size) /* overflow */ |
3531 | info.size = -1; | |
1da177e4 LT |
3532 | info.nr_disks = nr; |
3533 | info.raid_disks = mddev->raid_disks; | |
3534 | info.md_minor = mddev->md_minor; | |
3535 | info.not_persistent= !mddev->persistent; | |
3536 | ||
3537 | info.utime = mddev->utime; | |
3538 | info.state = 0; | |
3539 | if (mddev->in_sync) | |
3540 | info.state = (1<<MD_SB_CLEAN); | |
36fa3063 N |
3541 | if (mddev->bitmap && mddev->bitmap_offset) |
3542 | info.state = (1<<MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
3543 | info.active_disks = active; |
3544 | info.working_disks = working; | |
3545 | info.failed_disks = failed; | |
3546 | info.spare_disks = spare; | |
3547 | ||
3548 | info.layout = mddev->layout; | |
3549 | info.chunk_size = mddev->chunk_size; | |
3550 | ||
3551 | if (copy_to_user(arg, &info, sizeof(info))) | |
3552 | return -EFAULT; | |
3553 | ||
3554 | return 0; | |
3555 | } | |
3556 | ||
87162a28 | 3557 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
3558 | { |
3559 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
3560 | char *ptr, *buf = NULL; | |
3561 | int err = -ENOMEM; | |
3562 | ||
3563 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
3564 | if (!file) | |
3565 | goto out; | |
3566 | ||
3567 | /* bitmap disabled, zero the first byte and copy out */ | |
3568 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
3569 | file->pathname[0] = '\0'; | |
3570 | goto copy_out; | |
3571 | } | |
3572 | ||
3573 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
3574 | if (!buf) | |
3575 | goto out; | |
3576 | ||
3577 | ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname)); | |
3578 | if (!ptr) | |
3579 | goto out; | |
3580 | ||
3581 | strcpy(file->pathname, ptr); | |
3582 | ||
3583 | copy_out: | |
3584 | err = 0; | |
3585 | if (copy_to_user(arg, file, sizeof(*file))) | |
3586 | err = -EFAULT; | |
3587 | out: | |
3588 | kfree(buf); | |
3589 | kfree(file); | |
3590 | return err; | |
3591 | } | |
3592 | ||
1da177e4 LT |
3593 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
3594 | { | |
3595 | mdu_disk_info_t info; | |
3596 | unsigned int nr; | |
3597 | mdk_rdev_t *rdev; | |
3598 | ||
3599 | if (copy_from_user(&info, arg, sizeof(info))) | |
3600 | return -EFAULT; | |
3601 | ||
3602 | nr = info.number; | |
3603 | ||
3604 | rdev = find_rdev_nr(mddev, nr); | |
3605 | if (rdev) { | |
3606 | info.major = MAJOR(rdev->bdev->bd_dev); | |
3607 | info.minor = MINOR(rdev->bdev->bd_dev); | |
3608 | info.raid_disk = rdev->raid_disk; | |
3609 | info.state = 0; | |
b2d444d7 | 3610 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 3611 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 3612 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
3613 | info.state |= (1<<MD_DISK_ACTIVE); |
3614 | info.state |= (1<<MD_DISK_SYNC); | |
3615 | } | |
8ddf9efe N |
3616 | if (test_bit(WriteMostly, &rdev->flags)) |
3617 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
3618 | } else { |
3619 | info.major = info.minor = 0; | |
3620 | info.raid_disk = -1; | |
3621 | info.state = (1<<MD_DISK_REMOVED); | |
3622 | } | |
3623 | ||
3624 | if (copy_to_user(arg, &info, sizeof(info))) | |
3625 | return -EFAULT; | |
3626 | ||
3627 | return 0; | |
3628 | } | |
3629 | ||
3630 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
3631 | { | |
3632 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3633 | mdk_rdev_t *rdev; | |
3634 | dev_t dev = MKDEV(info->major,info->minor); | |
3635 | ||
3636 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
3637 | return -EOVERFLOW; | |
3638 | ||
3639 | if (!mddev->raid_disks) { | |
3640 | int err; | |
3641 | /* expecting a device which has a superblock */ | |
3642 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
3643 | if (IS_ERR(rdev)) { | |
3644 | printk(KERN_WARNING | |
3645 | "md: md_import_device returned %ld\n", | |
3646 | PTR_ERR(rdev)); | |
3647 | return PTR_ERR(rdev); | |
3648 | } | |
3649 | if (!list_empty(&mddev->disks)) { | |
3650 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3651 | mdk_rdev_t, same_set); | |
3652 | int err = super_types[mddev->major_version] | |
3653 | .load_super(rdev, rdev0, mddev->minor_version); | |
3654 | if (err < 0) { | |
3655 | printk(KERN_WARNING | |
3656 | "md: %s has different UUID to %s\n", | |
3657 | bdevname(rdev->bdev,b), | |
3658 | bdevname(rdev0->bdev,b2)); | |
3659 | export_rdev(rdev); | |
3660 | return -EINVAL; | |
3661 | } | |
3662 | } | |
3663 | err = bind_rdev_to_array(rdev, mddev); | |
3664 | if (err) | |
3665 | export_rdev(rdev); | |
3666 | return err; | |
3667 | } | |
3668 | ||
3669 | /* | |
3670 | * add_new_disk can be used once the array is assembled | |
3671 | * to add "hot spares". They must already have a superblock | |
3672 | * written | |
3673 | */ | |
3674 | if (mddev->pers) { | |
3675 | int err; | |
3676 | if (!mddev->pers->hot_add_disk) { | |
3677 | printk(KERN_WARNING | |
3678 | "%s: personality does not support diskops!\n", | |
3679 | mdname(mddev)); | |
3680 | return -EINVAL; | |
3681 | } | |
7b1e35f6 N |
3682 | if (mddev->persistent) |
3683 | rdev = md_import_device(dev, mddev->major_version, | |
3684 | mddev->minor_version); | |
3685 | else | |
3686 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
3687 | if (IS_ERR(rdev)) { |
3688 | printk(KERN_WARNING | |
3689 | "md: md_import_device returned %ld\n", | |
3690 | PTR_ERR(rdev)); | |
3691 | return PTR_ERR(rdev); | |
3692 | } | |
41158c7e N |
3693 | /* set save_raid_disk if appropriate */ |
3694 | if (!mddev->persistent) { | |
3695 | if (info->state & (1<<MD_DISK_SYNC) && | |
3696 | info->raid_disk < mddev->raid_disks) | |
3697 | rdev->raid_disk = info->raid_disk; | |
3698 | else | |
3699 | rdev->raid_disk = -1; | |
3700 | } else | |
3701 | super_types[mddev->major_version]. | |
3702 | validate_super(mddev, rdev); | |
3703 | rdev->saved_raid_disk = rdev->raid_disk; | |
3704 | ||
b2d444d7 | 3705 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
3706 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
3707 | set_bit(WriteMostly, &rdev->flags); | |
3708 | ||
1da177e4 LT |
3709 | rdev->raid_disk = -1; |
3710 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
3711 | if (!err && !mddev->pers->hot_remove_disk) { |
3712 | /* If there is hot_add_disk but no hot_remove_disk | |
3713 | * then added disks for geometry changes, | |
3714 | * and should be added immediately. | |
3715 | */ | |
3716 | super_types[mddev->major_version]. | |
3717 | validate_super(mddev, rdev); | |
3718 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
3719 | if (err) | |
3720 | unbind_rdev_from_array(rdev); | |
3721 | } | |
1da177e4 LT |
3722 | if (err) |
3723 | export_rdev(rdev); | |
c361777f N |
3724 | |
3725 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
005eca5e | 3726 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
3727 | return err; |
3728 | } | |
3729 | ||
3730 | /* otherwise, add_new_disk is only allowed | |
3731 | * for major_version==0 superblocks | |
3732 | */ | |
3733 | if (mddev->major_version != 0) { | |
3734 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
3735 | mdname(mddev)); | |
3736 | return -EINVAL; | |
3737 | } | |
3738 | ||
3739 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
3740 | int err; | |
3741 | rdev = md_import_device (dev, -1, 0); | |
3742 | if (IS_ERR(rdev)) { | |
3743 | printk(KERN_WARNING | |
3744 | "md: error, md_import_device() returned %ld\n", | |
3745 | PTR_ERR(rdev)); | |
3746 | return PTR_ERR(rdev); | |
3747 | } | |
3748 | rdev->desc_nr = info->number; | |
3749 | if (info->raid_disk < mddev->raid_disks) | |
3750 | rdev->raid_disk = info->raid_disk; | |
3751 | else | |
3752 | rdev->raid_disk = -1; | |
3753 | ||
b2d444d7 N |
3754 | rdev->flags = 0; |
3755 | ||
1da177e4 | 3756 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
3757 | if (info->state & (1<<MD_DISK_SYNC)) |
3758 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 3759 | |
8ddf9efe N |
3760 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
3761 | set_bit(WriteMostly, &rdev->flags); | |
3762 | ||
1da177e4 LT |
3763 | if (!mddev->persistent) { |
3764 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
3765 | rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
3766 | } else | |
3767 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
3768 | rdev->size = calc_dev_size(rdev, mddev->chunk_size); | |
3769 | ||
2bf071bf N |
3770 | err = bind_rdev_to_array(rdev, mddev); |
3771 | if (err) { | |
3772 | export_rdev(rdev); | |
3773 | return err; | |
3774 | } | |
1da177e4 LT |
3775 | } |
3776 | ||
3777 | return 0; | |
3778 | } | |
3779 | ||
3780 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
3781 | { | |
3782 | char b[BDEVNAME_SIZE]; | |
3783 | mdk_rdev_t *rdev; | |
3784 | ||
3785 | if (!mddev->pers) | |
3786 | return -ENODEV; | |
3787 | ||
3788 | rdev = find_rdev(mddev, dev); | |
3789 | if (!rdev) | |
3790 | return -ENXIO; | |
3791 | ||
3792 | if (rdev->raid_disk >= 0) | |
3793 | goto busy; | |
3794 | ||
3795 | kick_rdev_from_array(rdev); | |
3796 | md_update_sb(mddev); | |
d7603b7e | 3797 | md_new_event(mddev); |
1da177e4 LT |
3798 | |
3799 | return 0; | |
3800 | busy: | |
3801 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n", | |
3802 | bdevname(rdev->bdev,b), mdname(mddev)); | |
3803 | return -EBUSY; | |
3804 | } | |
3805 | ||
3806 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
3807 | { | |
3808 | char b[BDEVNAME_SIZE]; | |
3809 | int err; | |
3810 | unsigned int size; | |
3811 | mdk_rdev_t *rdev; | |
3812 | ||
3813 | if (!mddev->pers) | |
3814 | return -ENODEV; | |
3815 | ||
3816 | if (mddev->major_version != 0) { | |
3817 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
3818 | " version-0 superblocks.\n", | |
3819 | mdname(mddev)); | |
3820 | return -EINVAL; | |
3821 | } | |
3822 | if (!mddev->pers->hot_add_disk) { | |
3823 | printk(KERN_WARNING | |
3824 | "%s: personality does not support diskops!\n", | |
3825 | mdname(mddev)); | |
3826 | return -EINVAL; | |
3827 | } | |
3828 | ||
3829 | rdev = md_import_device (dev, -1, 0); | |
3830 | if (IS_ERR(rdev)) { | |
3831 | printk(KERN_WARNING | |
3832 | "md: error, md_import_device() returned %ld\n", | |
3833 | PTR_ERR(rdev)); | |
3834 | return -EINVAL; | |
3835 | } | |
3836 | ||
3837 | if (mddev->persistent) | |
3838 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
3839 | else | |
3840 | rdev->sb_offset = | |
3841 | rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
3842 | ||
3843 | size = calc_dev_size(rdev, mddev->chunk_size); | |
3844 | rdev->size = size; | |
3845 | ||
b2d444d7 | 3846 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
3847 | printk(KERN_WARNING |
3848 | "md: can not hot-add faulty %s disk to %s!\n", | |
3849 | bdevname(rdev->bdev,b), mdname(mddev)); | |
3850 | err = -EINVAL; | |
3851 | goto abort_export; | |
3852 | } | |
b2d444d7 | 3853 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 3854 | rdev->desc_nr = -1; |
2bf071bf N |
3855 | err = bind_rdev_to_array(rdev, mddev); |
3856 | if (err) | |
3857 | goto abort_export; | |
1da177e4 LT |
3858 | |
3859 | /* | |
3860 | * The rest should better be atomic, we can have disk failures | |
3861 | * noticed in interrupt contexts ... | |
3862 | */ | |
3863 | ||
3864 | if (rdev->desc_nr == mddev->max_disks) { | |
3865 | printk(KERN_WARNING "%s: can not hot-add to full array!\n", | |
3866 | mdname(mddev)); | |
3867 | err = -EBUSY; | |
3868 | goto abort_unbind_export; | |
3869 | } | |
3870 | ||
3871 | rdev->raid_disk = -1; | |
3872 | ||
3873 | md_update_sb(mddev); | |
3874 | ||
3875 | /* | |
3876 | * Kick recovery, maybe this spare has to be added to the | |
3877 | * array immediately. | |
3878 | */ | |
3879 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3880 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 3881 | md_new_event(mddev); |
1da177e4 LT |
3882 | return 0; |
3883 | ||
3884 | abort_unbind_export: | |
3885 | unbind_rdev_from_array(rdev); | |
3886 | ||
3887 | abort_export: | |
3888 | export_rdev(rdev); | |
3889 | return err; | |
3890 | } | |
3891 | ||
32a7627c N |
3892 | static int set_bitmap_file(mddev_t *mddev, int fd) |
3893 | { | |
3894 | int err; | |
3895 | ||
36fa3063 N |
3896 | if (mddev->pers) { |
3897 | if (!mddev->pers->quiesce) | |
3898 | return -EBUSY; | |
3899 | if (mddev->recovery || mddev->sync_thread) | |
3900 | return -EBUSY; | |
3901 | /* we should be able to change the bitmap.. */ | |
3902 | } | |
32a7627c | 3903 | |
32a7627c | 3904 | |
36fa3063 N |
3905 | if (fd >= 0) { |
3906 | if (mddev->bitmap) | |
3907 | return -EEXIST; /* cannot add when bitmap is present */ | |
3908 | mddev->bitmap_file = fget(fd); | |
32a7627c | 3909 | |
36fa3063 N |
3910 | if (mddev->bitmap_file == NULL) { |
3911 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", | |
3912 | mdname(mddev)); | |
3913 | return -EBADF; | |
3914 | } | |
3915 | ||
3916 | err = deny_bitmap_write_access(mddev->bitmap_file); | |
3917 | if (err) { | |
3918 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
3919 | mdname(mddev)); | |
3920 | fput(mddev->bitmap_file); | |
3921 | mddev->bitmap_file = NULL; | |
3922 | return err; | |
3923 | } | |
a654b9d8 | 3924 | mddev->bitmap_offset = 0; /* file overrides offset */ |
36fa3063 N |
3925 | } else if (mddev->bitmap == NULL) |
3926 | return -ENOENT; /* cannot remove what isn't there */ | |
3927 | err = 0; | |
3928 | if (mddev->pers) { | |
3929 | mddev->pers->quiesce(mddev, 1); | |
3930 | if (fd >= 0) | |
3931 | err = bitmap_create(mddev); | |
d7375ab3 | 3932 | if (fd < 0 || err) { |
36fa3063 | 3933 | bitmap_destroy(mddev); |
d7375ab3 N |
3934 | fd = -1; /* make sure to put the file */ |
3935 | } | |
36fa3063 | 3936 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
3937 | } |
3938 | if (fd < 0) { | |
acc55e22 N |
3939 | if (mddev->bitmap_file) { |
3940 | restore_bitmap_write_access(mddev->bitmap_file); | |
36fa3063 | 3941 | fput(mddev->bitmap_file); |
acc55e22 | 3942 | } |
36fa3063 N |
3943 | mddev->bitmap_file = NULL; |
3944 | } | |
3945 | ||
32a7627c N |
3946 | return err; |
3947 | } | |
3948 | ||
1da177e4 LT |
3949 | /* |
3950 | * set_array_info is used two different ways | |
3951 | * The original usage is when creating a new array. | |
3952 | * In this usage, raid_disks is > 0 and it together with | |
3953 | * level, size, not_persistent,layout,chunksize determine the | |
3954 | * shape of the array. | |
3955 | * This will always create an array with a type-0.90.0 superblock. | |
3956 | * The newer usage is when assembling an array. | |
3957 | * In this case raid_disks will be 0, and the major_version field is | |
3958 | * use to determine which style super-blocks are to be found on the devices. | |
3959 | * The minor and patch _version numbers are also kept incase the | |
3960 | * super_block handler wishes to interpret them. | |
3961 | */ | |
3962 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
3963 | { | |
3964 | ||
3965 | if (info->raid_disks == 0) { | |
3966 | /* just setting version number for superblock loading */ | |
3967 | if (info->major_version < 0 || | |
3968 | info->major_version >= sizeof(super_types)/sizeof(super_types[0]) || | |
3969 | super_types[info->major_version].name == NULL) { | |
3970 | /* maybe try to auto-load a module? */ | |
3971 | printk(KERN_INFO | |
3972 | "md: superblock version %d not known\n", | |
3973 | info->major_version); | |
3974 | return -EINVAL; | |
3975 | } | |
3976 | mddev->major_version = info->major_version; | |
3977 | mddev->minor_version = info->minor_version; | |
3978 | mddev->patch_version = info->patch_version; | |
3979 | return 0; | |
3980 | } | |
3981 | mddev->major_version = MD_MAJOR_VERSION; | |
3982 | mddev->minor_version = MD_MINOR_VERSION; | |
3983 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
3984 | mddev->ctime = get_seconds(); | |
3985 | ||
3986 | mddev->level = info->level; | |
17115e03 | 3987 | mddev->clevel[0] = 0; |
1da177e4 LT |
3988 | mddev->size = info->size; |
3989 | mddev->raid_disks = info->raid_disks; | |
3990 | /* don't set md_minor, it is determined by which /dev/md* was | |
3991 | * openned | |
3992 | */ | |
3993 | if (info->state & (1<<MD_SB_CLEAN)) | |
3994 | mddev->recovery_cp = MaxSector; | |
3995 | else | |
3996 | mddev->recovery_cp = 0; | |
3997 | mddev->persistent = ! info->not_persistent; | |
3998 | ||
3999 | mddev->layout = info->layout; | |
4000 | mddev->chunk_size = info->chunk_size; | |
4001 | ||
4002 | mddev->max_disks = MD_SB_DISKS; | |
4003 | ||
4004 | mddev->sb_dirty = 1; | |
4005 | ||
b2a2703c N |
4006 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
4007 | mddev->bitmap_offset = 0; | |
4008 | ||
f6705578 N |
4009 | mddev->reshape_position = MaxSector; |
4010 | ||
1da177e4 LT |
4011 | /* |
4012 | * Generate a 128 bit UUID | |
4013 | */ | |
4014 | get_random_bytes(mddev->uuid, 16); | |
4015 | ||
f6705578 N |
4016 | mddev->new_level = mddev->level; |
4017 | mddev->new_chunk = mddev->chunk_size; | |
4018 | mddev->new_layout = mddev->layout; | |
4019 | mddev->delta_disks = 0; | |
4020 | ||
1da177e4 LT |
4021 | return 0; |
4022 | } | |
4023 | ||
a35b0d69 N |
4024 | static int update_size(mddev_t *mddev, unsigned long size) |
4025 | { | |
4026 | mdk_rdev_t * rdev; | |
4027 | int rv; | |
4028 | struct list_head *tmp; | |
8ddeeae5 | 4029 | int fit = (size == 0); |
a35b0d69 N |
4030 | |
4031 | if (mddev->pers->resize == NULL) | |
4032 | return -EINVAL; | |
4033 | /* The "size" is the amount of each device that is used. | |
4034 | * This can only make sense for arrays with redundancy. | |
4035 | * linear and raid0 always use whatever space is available | |
4036 | * We can only consider changing the size if no resync | |
4037 | * or reconstruction is happening, and if the new size | |
4038 | * is acceptable. It must fit before the sb_offset or, | |
4039 | * if that is <data_offset, it must fit before the | |
4040 | * size of each device. | |
4041 | * If size is zero, we find the largest size that fits. | |
4042 | */ | |
4043 | if (mddev->sync_thread) | |
4044 | return -EBUSY; | |
4045 | ITERATE_RDEV(mddev,rdev,tmp) { | |
4046 | sector_t avail; | |
a35b0d69 N |
4047 | if (rdev->sb_offset > rdev->data_offset) |
4048 | avail = (rdev->sb_offset*2) - rdev->data_offset; | |
4049 | else | |
4050 | avail = get_capacity(rdev->bdev->bd_disk) | |
4051 | - rdev->data_offset; | |
4052 | if (fit && (size == 0 || size > avail/2)) | |
4053 | size = avail/2; | |
4054 | if (avail < ((sector_t)size << 1)) | |
4055 | return -ENOSPC; | |
4056 | } | |
4057 | rv = mddev->pers->resize(mddev, (sector_t)size *2); | |
4058 | if (!rv) { | |
4059 | struct block_device *bdev; | |
4060 | ||
4061 | bdev = bdget_disk(mddev->gendisk, 0); | |
4062 | if (bdev) { | |
1b1dcc1b | 4063 | mutex_lock(&bdev->bd_inode->i_mutex); |
6d89332b | 4064 | i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10); |
1b1dcc1b | 4065 | mutex_unlock(&bdev->bd_inode->i_mutex); |
a35b0d69 N |
4066 | bdput(bdev); |
4067 | } | |
4068 | } | |
4069 | return rv; | |
4070 | } | |
4071 | ||
da943b99 N |
4072 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
4073 | { | |
4074 | int rv; | |
4075 | /* change the number of raid disks */ | |
63c70c4f | 4076 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
4077 | return -EINVAL; |
4078 | if (raid_disks <= 0 || | |
4079 | raid_disks >= mddev->max_disks) | |
4080 | return -EINVAL; | |
63c70c4f | 4081 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 4082 | return -EBUSY; |
63c70c4f N |
4083 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
4084 | ||
4085 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
4086 | return rv; |
4087 | } | |
4088 | ||
4089 | ||
1da177e4 LT |
4090 | /* |
4091 | * update_array_info is used to change the configuration of an | |
4092 | * on-line array. | |
4093 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
4094 | * fields in the info are checked against the array. | |
4095 | * Any differences that cannot be handled will cause an error. | |
4096 | * Normally, only one change can be managed at a time. | |
4097 | */ | |
4098 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
4099 | { | |
4100 | int rv = 0; | |
4101 | int cnt = 0; | |
36fa3063 N |
4102 | int state = 0; |
4103 | ||
4104 | /* calculate expected state,ignoring low bits */ | |
4105 | if (mddev->bitmap && mddev->bitmap_offset) | |
4106 | state |= (1 << MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
4107 | |
4108 | if (mddev->major_version != info->major_version || | |
4109 | mddev->minor_version != info->minor_version || | |
4110 | /* mddev->patch_version != info->patch_version || */ | |
4111 | mddev->ctime != info->ctime || | |
4112 | mddev->level != info->level || | |
4113 | /* mddev->layout != info->layout || */ | |
4114 | !mddev->persistent != info->not_persistent|| | |
36fa3063 N |
4115 | mddev->chunk_size != info->chunk_size || |
4116 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ | |
4117 | ((state^info->state) & 0xfffffe00) | |
4118 | ) | |
1da177e4 LT |
4119 | return -EINVAL; |
4120 | /* Check there is only one change */ | |
284ae7ca | 4121 | if (info->size >= 0 && mddev->size != info->size) cnt++; |
1da177e4 LT |
4122 | if (mddev->raid_disks != info->raid_disks) cnt++; |
4123 | if (mddev->layout != info->layout) cnt++; | |
36fa3063 | 4124 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++; |
1da177e4 LT |
4125 | if (cnt == 0) return 0; |
4126 | if (cnt > 1) return -EINVAL; | |
4127 | ||
4128 | if (mddev->layout != info->layout) { | |
4129 | /* Change layout | |
4130 | * we don't need to do anything at the md level, the | |
4131 | * personality will take care of it all. | |
4132 | */ | |
4133 | if (mddev->pers->reconfig == NULL) | |
4134 | return -EINVAL; | |
4135 | else | |
4136 | return mddev->pers->reconfig(mddev, info->layout, -1); | |
4137 | } | |
284ae7ca | 4138 | if (info->size >= 0 && mddev->size != info->size) |
a35b0d69 N |
4139 | rv = update_size(mddev, info->size); |
4140 | ||
da943b99 N |
4141 | if (mddev->raid_disks != info->raid_disks) |
4142 | rv = update_raid_disks(mddev, info->raid_disks); | |
4143 | ||
36fa3063 N |
4144 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
4145 | if (mddev->pers->quiesce == NULL) | |
4146 | return -EINVAL; | |
4147 | if (mddev->recovery || mddev->sync_thread) | |
4148 | return -EBUSY; | |
4149 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
4150 | /* add the bitmap */ | |
4151 | if (mddev->bitmap) | |
4152 | return -EEXIST; | |
4153 | if (mddev->default_bitmap_offset == 0) | |
4154 | return -EINVAL; | |
4155 | mddev->bitmap_offset = mddev->default_bitmap_offset; | |
4156 | mddev->pers->quiesce(mddev, 1); | |
4157 | rv = bitmap_create(mddev); | |
4158 | if (rv) | |
4159 | bitmap_destroy(mddev); | |
4160 | mddev->pers->quiesce(mddev, 0); | |
4161 | } else { | |
4162 | /* remove the bitmap */ | |
4163 | if (!mddev->bitmap) | |
4164 | return -ENOENT; | |
4165 | if (mddev->bitmap->file) | |
4166 | return -EINVAL; | |
4167 | mddev->pers->quiesce(mddev, 1); | |
4168 | bitmap_destroy(mddev); | |
4169 | mddev->pers->quiesce(mddev, 0); | |
4170 | mddev->bitmap_offset = 0; | |
4171 | } | |
4172 | } | |
1da177e4 LT |
4173 | md_update_sb(mddev); |
4174 | return rv; | |
4175 | } | |
4176 | ||
4177 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
4178 | { | |
4179 | mdk_rdev_t *rdev; | |
4180 | ||
4181 | if (mddev->pers == NULL) | |
4182 | return -ENODEV; | |
4183 | ||
4184 | rdev = find_rdev(mddev, dev); | |
4185 | if (!rdev) | |
4186 | return -ENODEV; | |
4187 | ||
4188 | md_error(mddev, rdev); | |
4189 | return 0; | |
4190 | } | |
4191 | ||
a885c8c4 CH |
4192 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
4193 | { | |
4194 | mddev_t *mddev = bdev->bd_disk->private_data; | |
4195 | ||
4196 | geo->heads = 2; | |
4197 | geo->sectors = 4; | |
4198 | geo->cylinders = get_capacity(mddev->gendisk) / 8; | |
4199 | return 0; | |
4200 | } | |
4201 | ||
1da177e4 LT |
4202 | static int md_ioctl(struct inode *inode, struct file *file, |
4203 | unsigned int cmd, unsigned long arg) | |
4204 | { | |
4205 | int err = 0; | |
4206 | void __user *argp = (void __user *)arg; | |
1da177e4 LT |
4207 | mddev_t *mddev = NULL; |
4208 | ||
4209 | if (!capable(CAP_SYS_ADMIN)) | |
4210 | return -EACCES; | |
4211 | ||
4212 | /* | |
4213 | * Commands dealing with the RAID driver but not any | |
4214 | * particular array: | |
4215 | */ | |
4216 | switch (cmd) | |
4217 | { | |
4218 | case RAID_VERSION: | |
4219 | err = get_version(argp); | |
4220 | goto done; | |
4221 | ||
4222 | case PRINT_RAID_DEBUG: | |
4223 | err = 0; | |
4224 | md_print_devices(); | |
4225 | goto done; | |
4226 | ||
4227 | #ifndef MODULE | |
4228 | case RAID_AUTORUN: | |
4229 | err = 0; | |
4230 | autostart_arrays(arg); | |
4231 | goto done; | |
4232 | #endif | |
4233 | default:; | |
4234 | } | |
4235 | ||
4236 | /* | |
4237 | * Commands creating/starting a new array: | |
4238 | */ | |
4239 | ||
4240 | mddev = inode->i_bdev->bd_disk->private_data; | |
4241 | ||
4242 | if (!mddev) { | |
4243 | BUG(); | |
4244 | goto abort; | |
4245 | } | |
4246 | ||
4247 | ||
4248 | if (cmd == START_ARRAY) { | |
4249 | /* START_ARRAY doesn't need to lock the array as autostart_array | |
4250 | * does the locking, and it could even be a different array | |
4251 | */ | |
4252 | static int cnt = 3; | |
4253 | if (cnt > 0 ) { | |
4254 | printk(KERN_WARNING | |
4255 | "md: %s(pid %d) used deprecated START_ARRAY ioctl. " | |
e8a00334 | 4256 | "This will not be supported beyond July 2006\n", |
1da177e4 LT |
4257 | current->comm, current->pid); |
4258 | cnt--; | |
4259 | } | |
4260 | err = autostart_array(new_decode_dev(arg)); | |
4261 | if (err) { | |
4262 | printk(KERN_WARNING "md: autostart failed!\n"); | |
4263 | goto abort; | |
4264 | } | |
4265 | goto done; | |
4266 | } | |
4267 | ||
4268 | err = mddev_lock(mddev); | |
4269 | if (err) { | |
4270 | printk(KERN_INFO | |
4271 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
4272 | err, cmd); | |
4273 | goto abort; | |
4274 | } | |
4275 | ||
4276 | switch (cmd) | |
4277 | { | |
4278 | case SET_ARRAY_INFO: | |
4279 | { | |
4280 | mdu_array_info_t info; | |
4281 | if (!arg) | |
4282 | memset(&info, 0, sizeof(info)); | |
4283 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
4284 | err = -EFAULT; | |
4285 | goto abort_unlock; | |
4286 | } | |
4287 | if (mddev->pers) { | |
4288 | err = update_array_info(mddev, &info); | |
4289 | if (err) { | |
4290 | printk(KERN_WARNING "md: couldn't update" | |
4291 | " array info. %d\n", err); | |
4292 | goto abort_unlock; | |
4293 | } | |
4294 | goto done_unlock; | |
4295 | } | |
4296 | if (!list_empty(&mddev->disks)) { | |
4297 | printk(KERN_WARNING | |
4298 | "md: array %s already has disks!\n", | |
4299 | mdname(mddev)); | |
4300 | err = -EBUSY; | |
4301 | goto abort_unlock; | |
4302 | } | |
4303 | if (mddev->raid_disks) { | |
4304 | printk(KERN_WARNING | |
4305 | "md: array %s already initialised!\n", | |
4306 | mdname(mddev)); | |
4307 | err = -EBUSY; | |
4308 | goto abort_unlock; | |
4309 | } | |
4310 | err = set_array_info(mddev, &info); | |
4311 | if (err) { | |
4312 | printk(KERN_WARNING "md: couldn't set" | |
4313 | " array info. %d\n", err); | |
4314 | goto abort_unlock; | |
4315 | } | |
4316 | } | |
4317 | goto done_unlock; | |
4318 | ||
4319 | default:; | |
4320 | } | |
4321 | ||
4322 | /* | |
4323 | * Commands querying/configuring an existing array: | |
4324 | */ | |
32a7627c N |
4325 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
4326 | * RUN_ARRAY, and SET_BITMAP_FILE are allowed */ | |
4327 | if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
4328 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) { | |
1da177e4 LT |
4329 | err = -ENODEV; |
4330 | goto abort_unlock; | |
4331 | } | |
4332 | ||
4333 | /* | |
4334 | * Commands even a read-only array can execute: | |
4335 | */ | |
4336 | switch (cmd) | |
4337 | { | |
4338 | case GET_ARRAY_INFO: | |
4339 | err = get_array_info(mddev, argp); | |
4340 | goto done_unlock; | |
4341 | ||
32a7627c | 4342 | case GET_BITMAP_FILE: |
87162a28 | 4343 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
4344 | goto done_unlock; |
4345 | ||
1da177e4 LT |
4346 | case GET_DISK_INFO: |
4347 | err = get_disk_info(mddev, argp); | |
4348 | goto done_unlock; | |
4349 | ||
4350 | case RESTART_ARRAY_RW: | |
4351 | err = restart_array(mddev); | |
4352 | goto done_unlock; | |
4353 | ||
4354 | case STOP_ARRAY: | |
4355 | err = do_md_stop (mddev, 0); | |
4356 | goto done_unlock; | |
4357 | ||
4358 | case STOP_ARRAY_RO: | |
4359 | err = do_md_stop (mddev, 1); | |
4360 | goto done_unlock; | |
4361 | ||
4362 | /* | |
4363 | * We have a problem here : there is no easy way to give a CHS | |
4364 | * virtual geometry. We currently pretend that we have a 2 heads | |
4365 | * 4 sectors (with a BIG number of cylinders...). This drives | |
4366 | * dosfs just mad... ;-) | |
4367 | */ | |
1da177e4 LT |
4368 | } |
4369 | ||
4370 | /* | |
4371 | * The remaining ioctls are changing the state of the | |
f91de92e N |
4372 | * superblock, so we do not allow them on read-only arrays. |
4373 | * However non-MD ioctls (e.g. get-size) will still come through | |
4374 | * here and hit the 'default' below, so only disallow | |
4375 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 4376 | */ |
f91de92e N |
4377 | if (_IOC_TYPE(cmd) == MD_MAJOR && |
4378 | mddev->ro && mddev->pers) { | |
4379 | if (mddev->ro == 2) { | |
4380 | mddev->ro = 0; | |
4381 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4382 | md_wakeup_thread(mddev->thread); | |
4383 | ||
4384 | } else { | |
4385 | err = -EROFS; | |
4386 | goto abort_unlock; | |
4387 | } | |
1da177e4 LT |
4388 | } |
4389 | ||
4390 | switch (cmd) | |
4391 | { | |
4392 | case ADD_NEW_DISK: | |
4393 | { | |
4394 | mdu_disk_info_t info; | |
4395 | if (copy_from_user(&info, argp, sizeof(info))) | |
4396 | err = -EFAULT; | |
4397 | else | |
4398 | err = add_new_disk(mddev, &info); | |
4399 | goto done_unlock; | |
4400 | } | |
4401 | ||
4402 | case HOT_REMOVE_DISK: | |
4403 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
4404 | goto done_unlock; | |
4405 | ||
4406 | case HOT_ADD_DISK: | |
4407 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
4408 | goto done_unlock; | |
4409 | ||
4410 | case SET_DISK_FAULTY: | |
4411 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
4412 | goto done_unlock; | |
4413 | ||
4414 | case RUN_ARRAY: | |
4415 | err = do_md_run (mddev); | |
4416 | goto done_unlock; | |
4417 | ||
32a7627c N |
4418 | case SET_BITMAP_FILE: |
4419 | err = set_bitmap_file(mddev, (int)arg); | |
4420 | goto done_unlock; | |
4421 | ||
1da177e4 | 4422 | default: |
1da177e4 LT |
4423 | err = -EINVAL; |
4424 | goto abort_unlock; | |
4425 | } | |
4426 | ||
4427 | done_unlock: | |
4428 | abort_unlock: | |
4429 | mddev_unlock(mddev); | |
4430 | ||
4431 | return err; | |
4432 | done: | |
4433 | if (err) | |
4434 | MD_BUG(); | |
4435 | abort: | |
4436 | return err; | |
4437 | } | |
4438 | ||
4439 | static int md_open(struct inode *inode, struct file *file) | |
4440 | { | |
4441 | /* | |
4442 | * Succeed if we can lock the mddev, which confirms that | |
4443 | * it isn't being stopped right now. | |
4444 | */ | |
4445 | mddev_t *mddev = inode->i_bdev->bd_disk->private_data; | |
4446 | int err; | |
4447 | ||
4448 | if ((err = mddev_lock(mddev))) | |
4449 | goto out; | |
4450 | ||
4451 | err = 0; | |
4452 | mddev_get(mddev); | |
4453 | mddev_unlock(mddev); | |
4454 | ||
4455 | check_disk_change(inode->i_bdev); | |
4456 | out: | |
4457 | return err; | |
4458 | } | |
4459 | ||
4460 | static int md_release(struct inode *inode, struct file * file) | |
4461 | { | |
4462 | mddev_t *mddev = inode->i_bdev->bd_disk->private_data; | |
4463 | ||
4464 | if (!mddev) | |
4465 | BUG(); | |
4466 | mddev_put(mddev); | |
4467 | ||
4468 | return 0; | |
4469 | } | |
4470 | ||
4471 | static int md_media_changed(struct gendisk *disk) | |
4472 | { | |
4473 | mddev_t *mddev = disk->private_data; | |
4474 | ||
4475 | return mddev->changed; | |
4476 | } | |
4477 | ||
4478 | static int md_revalidate(struct gendisk *disk) | |
4479 | { | |
4480 | mddev_t *mddev = disk->private_data; | |
4481 | ||
4482 | mddev->changed = 0; | |
4483 | return 0; | |
4484 | } | |
4485 | static struct block_device_operations md_fops = | |
4486 | { | |
4487 | .owner = THIS_MODULE, | |
4488 | .open = md_open, | |
4489 | .release = md_release, | |
4490 | .ioctl = md_ioctl, | |
a885c8c4 | 4491 | .getgeo = md_getgeo, |
1da177e4 LT |
4492 | .media_changed = md_media_changed, |
4493 | .revalidate_disk= md_revalidate, | |
4494 | }; | |
4495 | ||
75c96f85 | 4496 | static int md_thread(void * arg) |
1da177e4 LT |
4497 | { |
4498 | mdk_thread_t *thread = arg; | |
4499 | ||
1da177e4 LT |
4500 | /* |
4501 | * md_thread is a 'system-thread', it's priority should be very | |
4502 | * high. We avoid resource deadlocks individually in each | |
4503 | * raid personality. (RAID5 does preallocation) We also use RR and | |
4504 | * the very same RT priority as kswapd, thus we will never get | |
4505 | * into a priority inversion deadlock. | |
4506 | * | |
4507 | * we definitely have to have equal or higher priority than | |
4508 | * bdflush, otherwise bdflush will deadlock if there are too | |
4509 | * many dirty RAID5 blocks. | |
4510 | */ | |
1da177e4 | 4511 | |
6985c43f | 4512 | allow_signal(SIGKILL); |
a6fb0934 | 4513 | while (!kthread_should_stop()) { |
1da177e4 | 4514 | |
93588e22 N |
4515 | /* We need to wait INTERRUPTIBLE so that |
4516 | * we don't add to the load-average. | |
4517 | * That means we need to be sure no signals are | |
4518 | * pending | |
4519 | */ | |
4520 | if (signal_pending(current)) | |
4521 | flush_signals(current); | |
4522 | ||
4523 | wait_event_interruptible_timeout | |
4524 | (thread->wqueue, | |
4525 | test_bit(THREAD_WAKEUP, &thread->flags) | |
4526 | || kthread_should_stop(), | |
4527 | thread->timeout); | |
3e1d1d28 | 4528 | try_to_freeze(); |
1da177e4 LT |
4529 | |
4530 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
4531 | ||
787453c2 | 4532 | thread->run(thread->mddev); |
1da177e4 | 4533 | } |
a6fb0934 | 4534 | |
1da177e4 LT |
4535 | return 0; |
4536 | } | |
4537 | ||
4538 | void md_wakeup_thread(mdk_thread_t *thread) | |
4539 | { | |
4540 | if (thread) { | |
4541 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
4542 | set_bit(THREAD_WAKEUP, &thread->flags); | |
4543 | wake_up(&thread->wqueue); | |
4544 | } | |
4545 | } | |
4546 | ||
4547 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
4548 | const char *name) | |
4549 | { | |
4550 | mdk_thread_t *thread; | |
1da177e4 | 4551 | |
9ffae0cf | 4552 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
4553 | if (!thread) |
4554 | return NULL; | |
4555 | ||
1da177e4 LT |
4556 | init_waitqueue_head(&thread->wqueue); |
4557 | ||
1da177e4 LT |
4558 | thread->run = run; |
4559 | thread->mddev = mddev; | |
32a7627c | 4560 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
6985c43f | 4561 | thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev)); |
a6fb0934 | 4562 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
4563 | kfree(thread); |
4564 | return NULL; | |
4565 | } | |
1da177e4 LT |
4566 | return thread; |
4567 | } | |
4568 | ||
1da177e4 LT |
4569 | void md_unregister_thread(mdk_thread_t *thread) |
4570 | { | |
d28446fe | 4571 | dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid); |
a6fb0934 N |
4572 | |
4573 | kthread_stop(thread->tsk); | |
1da177e4 LT |
4574 | kfree(thread); |
4575 | } | |
4576 | ||
4577 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
4578 | { | |
4579 | if (!mddev) { | |
4580 | MD_BUG(); | |
4581 | return; | |
4582 | } | |
4583 | ||
b2d444d7 | 4584 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4585 | return; |
32a7627c | 4586 | /* |
1da177e4 LT |
4587 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
4588 | mdname(mddev), | |
4589 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
4590 | __builtin_return_address(0),__builtin_return_address(1), | |
4591 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 4592 | */ |
1da177e4 LT |
4593 | if (!mddev->pers->error_handler) |
4594 | return; | |
4595 | mddev->pers->error_handler(mddev,rdev); | |
4596 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
4597 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4598 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 4599 | md_new_event_inintr(mddev); |
1da177e4 LT |
4600 | } |
4601 | ||
4602 | /* seq_file implementation /proc/mdstat */ | |
4603 | ||
4604 | static void status_unused(struct seq_file *seq) | |
4605 | { | |
4606 | int i = 0; | |
4607 | mdk_rdev_t *rdev; | |
4608 | struct list_head *tmp; | |
4609 | ||
4610 | seq_printf(seq, "unused devices: "); | |
4611 | ||
4612 | ITERATE_RDEV_PENDING(rdev,tmp) { | |
4613 | char b[BDEVNAME_SIZE]; | |
4614 | i++; | |
4615 | seq_printf(seq, "%s ", | |
4616 | bdevname(rdev->bdev,b)); | |
4617 | } | |
4618 | if (!i) | |
4619 | seq_printf(seq, "<none>"); | |
4620 | ||
4621 | seq_printf(seq, "\n"); | |
4622 | } | |
4623 | ||
4624 | ||
4625 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
4626 | { | |
4588b42e N |
4627 | sector_t max_blocks, resync, res; |
4628 | unsigned long dt, db, rt; | |
4629 | int scale; | |
4630 | unsigned int per_milli; | |
1da177e4 LT |
4631 | |
4632 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2; | |
4633 | ||
4634 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
4635 | max_blocks = mddev->resync_max_sectors >> 1; | |
4636 | else | |
4637 | max_blocks = mddev->size; | |
4638 | ||
4639 | /* | |
4640 | * Should not happen. | |
4641 | */ | |
4642 | if (!max_blocks) { | |
4643 | MD_BUG(); | |
4644 | return; | |
4645 | } | |
4588b42e N |
4646 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
4647 | * in a sector_t, and (max_blocks>>scale) will fit in a | |
4648 | * u32, as those are the requirements for sector_div. | |
4649 | * Thus 'scale' must be at least 10 | |
4650 | */ | |
4651 | scale = 10; | |
4652 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
4653 | while ( max_blocks/2 > (1ULL<<(scale+32))) | |
4654 | scale++; | |
4655 | } | |
4656 | res = (resync>>scale)*1000; | |
4657 | sector_div(res, (u32)((max_blocks>>scale)+1)); | |
4658 | ||
4659 | per_milli = res; | |
1da177e4 | 4660 | { |
4588b42e | 4661 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
4662 | seq_printf(seq, "["); |
4663 | for (i = 0; i < x; i++) | |
4664 | seq_printf(seq, "="); | |
4665 | seq_printf(seq, ">"); | |
4666 | for (i = 0; i < y; i++) | |
4667 | seq_printf(seq, "."); | |
4668 | seq_printf(seq, "] "); | |
4669 | } | |
4588b42e | 4670 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
4671 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
4672 | "reshape" : | |
1da177e4 | 4673 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? |
ccfcc3c1 | 4674 | "resync" : "recovery")), |
4588b42e N |
4675 | per_milli/10, per_milli % 10, |
4676 | (unsigned long long) resync, | |
4677 | (unsigned long long) max_blocks); | |
1da177e4 LT |
4678 | |
4679 | /* | |
4680 | * We do not want to overflow, so the order of operands and | |
4681 | * the * 100 / 100 trick are important. We do a +1 to be | |
4682 | * safe against division by zero. We only estimate anyway. | |
4683 | * | |
4684 | * dt: time from mark until now | |
4685 | * db: blocks written from mark until now | |
4686 | * rt: remaining time | |
4687 | */ | |
4688 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
4689 | if (!dt) dt++; | |
4690 | db = resync - (mddev->resync_mark_cnt/2); | |
4588b42e | 4691 | rt = (dt * ((unsigned long)(max_blocks-resync) / (db/100+1)))/100; |
1da177e4 LT |
4692 | |
4693 | seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6); | |
4694 | ||
4695 | seq_printf(seq, " speed=%ldK/sec", db/dt); | |
4696 | } | |
4697 | ||
4698 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
4699 | { | |
4700 | struct list_head *tmp; | |
4701 | loff_t l = *pos; | |
4702 | mddev_t *mddev; | |
4703 | ||
4704 | if (l >= 0x10000) | |
4705 | return NULL; | |
4706 | if (!l--) | |
4707 | /* header */ | |
4708 | return (void*)1; | |
4709 | ||
4710 | spin_lock(&all_mddevs_lock); | |
4711 | list_for_each(tmp,&all_mddevs) | |
4712 | if (!l--) { | |
4713 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
4714 | mddev_get(mddev); | |
4715 | spin_unlock(&all_mddevs_lock); | |
4716 | return mddev; | |
4717 | } | |
4718 | spin_unlock(&all_mddevs_lock); | |
4719 | if (!l--) | |
4720 | return (void*)2;/* tail */ | |
4721 | return NULL; | |
4722 | } | |
4723 | ||
4724 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4725 | { | |
4726 | struct list_head *tmp; | |
4727 | mddev_t *next_mddev, *mddev = v; | |
4728 | ||
4729 | ++*pos; | |
4730 | if (v == (void*)2) | |
4731 | return NULL; | |
4732 | ||
4733 | spin_lock(&all_mddevs_lock); | |
4734 | if (v == (void*)1) | |
4735 | tmp = all_mddevs.next; | |
4736 | else | |
4737 | tmp = mddev->all_mddevs.next; | |
4738 | if (tmp != &all_mddevs) | |
4739 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
4740 | else { | |
4741 | next_mddev = (void*)2; | |
4742 | *pos = 0x10000; | |
4743 | } | |
4744 | spin_unlock(&all_mddevs_lock); | |
4745 | ||
4746 | if (v != (void*)1) | |
4747 | mddev_put(mddev); | |
4748 | return next_mddev; | |
4749 | ||
4750 | } | |
4751 | ||
4752 | static void md_seq_stop(struct seq_file *seq, void *v) | |
4753 | { | |
4754 | mddev_t *mddev = v; | |
4755 | ||
4756 | if (mddev && v != (void*)1 && v != (void*)2) | |
4757 | mddev_put(mddev); | |
4758 | } | |
4759 | ||
d7603b7e N |
4760 | struct mdstat_info { |
4761 | int event; | |
4762 | }; | |
4763 | ||
1da177e4 LT |
4764 | static int md_seq_show(struct seq_file *seq, void *v) |
4765 | { | |
4766 | mddev_t *mddev = v; | |
4767 | sector_t size; | |
4768 | struct list_head *tmp2; | |
4769 | mdk_rdev_t *rdev; | |
d7603b7e | 4770 | struct mdstat_info *mi = seq->private; |
32a7627c | 4771 | struct bitmap *bitmap; |
1da177e4 LT |
4772 | |
4773 | if (v == (void*)1) { | |
2604b703 | 4774 | struct mdk_personality *pers; |
1da177e4 LT |
4775 | seq_printf(seq, "Personalities : "); |
4776 | spin_lock(&pers_lock); | |
2604b703 N |
4777 | list_for_each_entry(pers, &pers_list, list) |
4778 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
4779 | |
4780 | spin_unlock(&pers_lock); | |
4781 | seq_printf(seq, "\n"); | |
d7603b7e | 4782 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
4783 | return 0; |
4784 | } | |
4785 | if (v == (void*)2) { | |
4786 | status_unused(seq); | |
4787 | return 0; | |
4788 | } | |
4789 | ||
5dc5cf7d | 4790 | if (mddev_lock(mddev) < 0) |
1da177e4 | 4791 | return -EINTR; |
5dc5cf7d | 4792 | |
1da177e4 LT |
4793 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
4794 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
4795 | mddev->pers ? "" : "in"); | |
4796 | if (mddev->pers) { | |
f91de92e | 4797 | if (mddev->ro==1) |
1da177e4 | 4798 | seq_printf(seq, " (read-only)"); |
f91de92e N |
4799 | if (mddev->ro==2) |
4800 | seq_printf(seq, "(auto-read-only)"); | |
1da177e4 LT |
4801 | seq_printf(seq, " %s", mddev->pers->name); |
4802 | } | |
4803 | ||
4804 | size = 0; | |
4805 | ITERATE_RDEV(mddev,rdev,tmp2) { | |
4806 | char b[BDEVNAME_SIZE]; | |
4807 | seq_printf(seq, " %s[%d]", | |
4808 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
4809 | if (test_bit(WriteMostly, &rdev->flags)) |
4810 | seq_printf(seq, "(W)"); | |
b2d444d7 | 4811 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
4812 | seq_printf(seq, "(F)"); |
4813 | continue; | |
b325a32e N |
4814 | } else if (rdev->raid_disk < 0) |
4815 | seq_printf(seq, "(S)"); /* spare */ | |
1da177e4 LT |
4816 | size += rdev->size; |
4817 | } | |
4818 | ||
4819 | if (!list_empty(&mddev->disks)) { | |
4820 | if (mddev->pers) | |
4821 | seq_printf(seq, "\n %llu blocks", | |
4822 | (unsigned long long)mddev->array_size); | |
4823 | else | |
4824 | seq_printf(seq, "\n %llu blocks", | |
4825 | (unsigned long long)size); | |
4826 | } | |
1cd6bf19 N |
4827 | if (mddev->persistent) { |
4828 | if (mddev->major_version != 0 || | |
4829 | mddev->minor_version != 90) { | |
4830 | seq_printf(seq," super %d.%d", | |
4831 | mddev->major_version, | |
4832 | mddev->minor_version); | |
4833 | } | |
4834 | } else | |
4835 | seq_printf(seq, " super non-persistent"); | |
1da177e4 LT |
4836 | |
4837 | if (mddev->pers) { | |
4838 | mddev->pers->status (seq, mddev); | |
4839 | seq_printf(seq, "\n "); | |
8e1b39d6 N |
4840 | if (mddev->pers->sync_request) { |
4841 | if (mddev->curr_resync > 2) { | |
4842 | status_resync (seq, mddev); | |
4843 | seq_printf(seq, "\n "); | |
4844 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
4845 | seq_printf(seq, "\tresync=DELAYED\n "); | |
4846 | else if (mddev->recovery_cp < MaxSector) | |
4847 | seq_printf(seq, "\tresync=PENDING\n "); | |
4848 | } | |
32a7627c N |
4849 | } else |
4850 | seq_printf(seq, "\n "); | |
4851 | ||
4852 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
4853 | unsigned long chunk_kb; |
4854 | unsigned long flags; | |
32a7627c N |
4855 | spin_lock_irqsave(&bitmap->lock, flags); |
4856 | chunk_kb = bitmap->chunksize >> 10; | |
4857 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
4858 | "%lu%s chunk", | |
4859 | bitmap->pages - bitmap->missing_pages, | |
4860 | bitmap->pages, | |
4861 | (bitmap->pages - bitmap->missing_pages) | |
4862 | << (PAGE_SHIFT - 10), | |
4863 | chunk_kb ? chunk_kb : bitmap->chunksize, | |
4864 | chunk_kb ? "KB" : "B"); | |
78d742d8 N |
4865 | if (bitmap->file) { |
4866 | seq_printf(seq, ", file: "); | |
4867 | seq_path(seq, bitmap->file->f_vfsmnt, | |
4868 | bitmap->file->f_dentry," \t\n"); | |
32a7627c | 4869 | } |
78d742d8 | 4870 | |
32a7627c N |
4871 | seq_printf(seq, "\n"); |
4872 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
4873 | } |
4874 | ||
4875 | seq_printf(seq, "\n"); | |
4876 | } | |
4877 | mddev_unlock(mddev); | |
4878 | ||
4879 | return 0; | |
4880 | } | |
4881 | ||
4882 | static struct seq_operations md_seq_ops = { | |
4883 | .start = md_seq_start, | |
4884 | .next = md_seq_next, | |
4885 | .stop = md_seq_stop, | |
4886 | .show = md_seq_show, | |
4887 | }; | |
4888 | ||
4889 | static int md_seq_open(struct inode *inode, struct file *file) | |
4890 | { | |
4891 | int error; | |
d7603b7e N |
4892 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
4893 | if (mi == NULL) | |
4894 | return -ENOMEM; | |
1da177e4 LT |
4895 | |
4896 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
4897 | if (error) |
4898 | kfree(mi); | |
4899 | else { | |
4900 | struct seq_file *p = file->private_data; | |
4901 | p->private = mi; | |
4902 | mi->event = atomic_read(&md_event_count); | |
4903 | } | |
1da177e4 LT |
4904 | return error; |
4905 | } | |
4906 | ||
d7603b7e N |
4907 | static int md_seq_release(struct inode *inode, struct file *file) |
4908 | { | |
4909 | struct seq_file *m = file->private_data; | |
4910 | struct mdstat_info *mi = m->private; | |
4911 | m->private = NULL; | |
4912 | kfree(mi); | |
4913 | return seq_release(inode, file); | |
4914 | } | |
4915 | ||
4916 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) | |
4917 | { | |
4918 | struct seq_file *m = filp->private_data; | |
4919 | struct mdstat_info *mi = m->private; | |
4920 | int mask; | |
4921 | ||
4922 | poll_wait(filp, &md_event_waiters, wait); | |
4923 | ||
4924 | /* always allow read */ | |
4925 | mask = POLLIN | POLLRDNORM; | |
4926 | ||
4927 | if (mi->event != atomic_read(&md_event_count)) | |
4928 | mask |= POLLERR | POLLPRI; | |
4929 | return mask; | |
4930 | } | |
4931 | ||
1da177e4 LT |
4932 | static struct file_operations md_seq_fops = { |
4933 | .open = md_seq_open, | |
4934 | .read = seq_read, | |
4935 | .llseek = seq_lseek, | |
d7603b7e N |
4936 | .release = md_seq_release, |
4937 | .poll = mdstat_poll, | |
1da177e4 LT |
4938 | }; |
4939 | ||
2604b703 | 4940 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 4941 | { |
1da177e4 | 4942 | spin_lock(&pers_lock); |
2604b703 N |
4943 | list_add_tail(&p->list, &pers_list); |
4944 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
4945 | spin_unlock(&pers_lock); |
4946 | return 0; | |
4947 | } | |
4948 | ||
2604b703 | 4949 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 4950 | { |
2604b703 | 4951 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 4952 | spin_lock(&pers_lock); |
2604b703 | 4953 | list_del_init(&p->list); |
1da177e4 LT |
4954 | spin_unlock(&pers_lock); |
4955 | return 0; | |
4956 | } | |
4957 | ||
4958 | static int is_mddev_idle(mddev_t *mddev) | |
4959 | { | |
4960 | mdk_rdev_t * rdev; | |
4961 | struct list_head *tmp; | |
4962 | int idle; | |
4963 | unsigned long curr_events; | |
4964 | ||
4965 | idle = 1; | |
4966 | ITERATE_RDEV(mddev,rdev,tmp) { | |
4967 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; | |
a362357b JA |
4968 | curr_events = disk_stat_read(disk, sectors[0]) + |
4969 | disk_stat_read(disk, sectors[1]) - | |
1da177e4 | 4970 | atomic_read(&disk->sync_io); |
c0e48521 N |
4971 | /* The difference between curr_events and last_events |
4972 | * will be affected by any new non-sync IO (making | |
4973 | * curr_events bigger) and any difference in the amount of | |
4974 | * in-flight syncio (making current_events bigger or smaller) | |
4975 | * The amount in-flight is currently limited to | |
4976 | * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6 | |
4977 | * which is at most 4096 sectors. | |
4978 | * These numbers are fairly fragile and should be made | |
4979 | * more robust, probably by enforcing the | |
4980 | * 'window size' that md_do_sync sort-of uses. | |
4981 | * | |
1da177e4 LT |
4982 | * Note: the following is an unsigned comparison. |
4983 | */ | |
c0e48521 | 4984 | if ((curr_events - rdev->last_events + 4096) > 8192) { |
1da177e4 LT |
4985 | rdev->last_events = curr_events; |
4986 | idle = 0; | |
4987 | } | |
4988 | } | |
4989 | return idle; | |
4990 | } | |
4991 | ||
4992 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
4993 | { | |
4994 | /* another "blocks" (512byte) blocks have been synced */ | |
4995 | atomic_sub(blocks, &mddev->recovery_active); | |
4996 | wake_up(&mddev->recovery_wait); | |
4997 | if (!ok) { | |
4998 | set_bit(MD_RECOVERY_ERR, &mddev->recovery); | |
4999 | md_wakeup_thread(mddev->thread); | |
5000 | // stop recovery, signal do_sync .... | |
5001 | } | |
5002 | } | |
5003 | ||
5004 | ||
06d91a5f N |
5005 | /* md_write_start(mddev, bi) |
5006 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
5007 | * in superblock) before writing, schedule a superblock update |
5008 | * and wait for it to complete. | |
06d91a5f | 5009 | */ |
3d310eb7 | 5010 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 5011 | { |
06d91a5f | 5012 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 5013 | return; |
06d91a5f | 5014 | |
f91de92e N |
5015 | BUG_ON(mddev->ro == 1); |
5016 | if (mddev->ro == 2) { | |
5017 | /* need to switch to read/write */ | |
5018 | mddev->ro = 0; | |
5019 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5020 | md_wakeup_thread(mddev->thread); | |
5021 | } | |
06d91a5f | 5022 | atomic_inc(&mddev->writes_pending); |
06d91a5f | 5023 | if (mddev->in_sync) { |
a9701a30 | 5024 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
5025 | if (mddev->in_sync) { |
5026 | mddev->in_sync = 0; | |
42543769 | 5027 | mddev->sb_dirty = 3; |
3d310eb7 N |
5028 | md_wakeup_thread(mddev->thread); |
5029 | } | |
a9701a30 | 5030 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 5031 | } |
3d310eb7 | 5032 | wait_event(mddev->sb_wait, mddev->sb_dirty==0); |
1da177e4 LT |
5033 | } |
5034 | ||
5035 | void md_write_end(mddev_t *mddev) | |
5036 | { | |
5037 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
5038 | if (mddev->safemode == 2) | |
5039 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 5040 | else if (mddev->safemode_delay) |
1da177e4 LT |
5041 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
5042 | } | |
5043 | } | |
5044 | ||
75c96f85 | 5045 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
1da177e4 LT |
5046 | |
5047 | #define SYNC_MARKS 10 | |
5048 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 5049 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
5050 | { |
5051 | mddev_t *mddev2; | |
5052 | unsigned int currspeed = 0, | |
5053 | window; | |
57afd89f | 5054 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
5055 | unsigned long mark[SYNC_MARKS]; |
5056 | sector_t mark_cnt[SYNC_MARKS]; | |
5057 | int last_mark,m; | |
5058 | struct list_head *tmp; | |
5059 | sector_t last_check; | |
57afd89f | 5060 | int skipped = 0; |
5fd6c1dc N |
5061 | struct list_head *rtmp; |
5062 | mdk_rdev_t *rdev; | |
1da177e4 LT |
5063 | |
5064 | /* just incase thread restarts... */ | |
5065 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
5066 | return; | |
5fd6c1dc N |
5067 | if (mddev->ro) /* never try to sync a read-only array */ |
5068 | return; | |
1da177e4 LT |
5069 | |
5070 | /* we overload curr_resync somewhat here. | |
5071 | * 0 == not engaged in resync at all | |
5072 | * 2 == checking that there is no conflict with another sync | |
5073 | * 1 == like 2, but have yielded to allow conflicting resync to | |
5074 | * commense | |
5075 | * other == active in resync - this many blocks | |
5076 | * | |
5077 | * Before starting a resync we must have set curr_resync to | |
5078 | * 2, and then checked that every "conflicting" array has curr_resync | |
5079 | * less than ours. When we find one that is the same or higher | |
5080 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
5081 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
5082 | * This will mean we have to start checking from the beginning again. | |
5083 | * | |
5084 | */ | |
5085 | ||
5086 | do { | |
5087 | mddev->curr_resync = 2; | |
5088 | ||
5089 | try_again: | |
787453c2 | 5090 | if (kthread_should_stop()) { |
6985c43f | 5091 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
5092 | goto skip; |
5093 | } | |
5094 | ITERATE_MDDEV(mddev2,tmp) { | |
1da177e4 LT |
5095 | if (mddev2 == mddev) |
5096 | continue; | |
5097 | if (mddev2->curr_resync && | |
5098 | match_mddev_units(mddev,mddev2)) { | |
5099 | DEFINE_WAIT(wq); | |
5100 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
5101 | /* arbitrarily yield */ | |
5102 | mddev->curr_resync = 1; | |
5103 | wake_up(&resync_wait); | |
5104 | } | |
5105 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
5106 | /* no need to wait here, we can wait the next | |
5107 | * time 'round when curr_resync == 2 | |
5108 | */ | |
5109 | continue; | |
787453c2 N |
5110 | prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE); |
5111 | if (!kthread_should_stop() && | |
8712e553 | 5112 | mddev2->curr_resync >= mddev->curr_resync) { |
1da177e4 LT |
5113 | printk(KERN_INFO "md: delaying resync of %s" |
5114 | " until %s has finished resync (they" | |
5115 | " share one or more physical units)\n", | |
5116 | mdname(mddev), mdname(mddev2)); | |
5117 | mddev_put(mddev2); | |
5118 | schedule(); | |
5119 | finish_wait(&resync_wait, &wq); | |
5120 | goto try_again; | |
5121 | } | |
5122 | finish_wait(&resync_wait, &wq); | |
5123 | } | |
5124 | } | |
5125 | } while (mddev->curr_resync < 2); | |
5126 | ||
5fd6c1dc | 5127 | j = 0; |
9d88883e | 5128 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 5129 | /* resync follows the size requested by the personality, |
57afd89f | 5130 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
5131 | */ |
5132 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 5133 | mddev->resync_mismatches = 0; |
5fd6c1dc N |
5134 | /* we don't use the checkpoint if there's a bitmap */ |
5135 | if (!mddev->bitmap && | |
5136 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
5137 | j = mddev->recovery_cp; | |
ccfcc3c1 N |
5138 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5139 | max_sectors = mddev->size << 1; | |
5fd6c1dc | 5140 | else { |
1da177e4 LT |
5141 | /* recovery follows the physical size of devices */ |
5142 | max_sectors = mddev->size << 1; | |
5fd6c1dc N |
5143 | j = MaxSector; |
5144 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5145 | if (rdev->raid_disk >= 0 && | |
5146 | !test_bit(Faulty, &rdev->flags) && | |
5147 | !test_bit(In_sync, &rdev->flags) && | |
5148 | rdev->recovery_offset < j) | |
5149 | j = rdev->recovery_offset; | |
5150 | } | |
1da177e4 LT |
5151 | |
5152 | printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev)); | |
5153 | printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:" | |
88202a0c | 5154 | " %d KB/sec/disc.\n", speed_min(mddev)); |
338cec32 | 5155 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
1da177e4 | 5156 | "(but not more than %d KB/sec) for reconstruction.\n", |
88202a0c | 5157 | speed_max(mddev)); |
1da177e4 LT |
5158 | |
5159 | is_mddev_idle(mddev); /* this also initializes IO event counters */ | |
5fd6c1dc | 5160 | |
57afd89f | 5161 | io_sectors = 0; |
1da177e4 LT |
5162 | for (m = 0; m < SYNC_MARKS; m++) { |
5163 | mark[m] = jiffies; | |
57afd89f | 5164 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
5165 | } |
5166 | last_mark = 0; | |
5167 | mddev->resync_mark = mark[last_mark]; | |
5168 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
5169 | ||
5170 | /* | |
5171 | * Tune reconstruction: | |
5172 | */ | |
5173 | window = 32*(PAGE_SIZE/512); | |
5174 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
5175 | window/2,(unsigned long long) max_sectors/2); | |
5176 | ||
5177 | atomic_set(&mddev->recovery_active, 0); | |
5178 | init_waitqueue_head(&mddev->recovery_wait); | |
5179 | last_check = 0; | |
5180 | ||
5181 | if (j>2) { | |
5182 | printk(KERN_INFO | |
5183 | "md: resuming recovery of %s from checkpoint.\n", | |
5184 | mdname(mddev)); | |
5185 | mddev->curr_resync = j; | |
5186 | } | |
5187 | ||
5188 | while (j < max_sectors) { | |
57afd89f | 5189 | sector_t sectors; |
1da177e4 | 5190 | |
57afd89f N |
5191 | skipped = 0; |
5192 | sectors = mddev->pers->sync_request(mddev, j, &skipped, | |
88202a0c | 5193 | currspeed < speed_min(mddev)); |
57afd89f | 5194 | if (sectors == 0) { |
1da177e4 LT |
5195 | set_bit(MD_RECOVERY_ERR, &mddev->recovery); |
5196 | goto out; | |
5197 | } | |
57afd89f N |
5198 | |
5199 | if (!skipped) { /* actual IO requested */ | |
5200 | io_sectors += sectors; | |
5201 | atomic_add(sectors, &mddev->recovery_active); | |
5202 | } | |
5203 | ||
1da177e4 LT |
5204 | j += sectors; |
5205 | if (j>1) mddev->curr_resync = j; | |
d7603b7e N |
5206 | if (last_check == 0) |
5207 | /* this is the earliers that rebuilt will be | |
5208 | * visible in /proc/mdstat | |
5209 | */ | |
5210 | md_new_event(mddev); | |
57afd89f N |
5211 | |
5212 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
5213 | continue; |
5214 | ||
57afd89f | 5215 | last_check = io_sectors; |
1da177e4 LT |
5216 | |
5217 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) || | |
5218 | test_bit(MD_RECOVERY_ERR, &mddev->recovery)) | |
5219 | break; | |
5220 | ||
5221 | repeat: | |
5222 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
5223 | /* step marks */ | |
5224 | int next = (last_mark+1) % SYNC_MARKS; | |
5225 | ||
5226 | mddev->resync_mark = mark[next]; | |
5227 | mddev->resync_mark_cnt = mark_cnt[next]; | |
5228 | mark[next] = jiffies; | |
57afd89f | 5229 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
5230 | last_mark = next; |
5231 | } | |
5232 | ||
5233 | ||
787453c2 | 5234 | if (kthread_should_stop()) { |
1da177e4 LT |
5235 | /* |
5236 | * got a signal, exit. | |
5237 | */ | |
5238 | printk(KERN_INFO | |
5239 | "md: md_do_sync() got signal ... exiting\n"); | |
1da177e4 LT |
5240 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
5241 | goto out; | |
5242 | } | |
5243 | ||
5244 | /* | |
5245 | * this loop exits only if either when we are slower than | |
5246 | * the 'hard' speed limit, or the system was IO-idle for | |
5247 | * a jiffy. | |
5248 | * the system might be non-idle CPU-wise, but we only care | |
5249 | * about not overloading the IO subsystem. (things like an | |
5250 | * e2fsck being done on the RAID array should execute fast) | |
5251 | */ | |
5252 | mddev->queue->unplug_fn(mddev->queue); | |
5253 | cond_resched(); | |
5254 | ||
57afd89f N |
5255 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
5256 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 5257 | |
88202a0c N |
5258 | if (currspeed > speed_min(mddev)) { |
5259 | if ((currspeed > speed_max(mddev)) || | |
1da177e4 | 5260 | !is_mddev_idle(mddev)) { |
c0e48521 | 5261 | msleep(500); |
1da177e4 LT |
5262 | goto repeat; |
5263 | } | |
5264 | } | |
5265 | } | |
5266 | printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev)); | |
5267 | /* | |
5268 | * this also signals 'finished resyncing' to md_stop | |
5269 | */ | |
5270 | out: | |
5271 | mddev->queue->unplug_fn(mddev->queue); | |
5272 | ||
5273 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
5274 | ||
5275 | /* tell personality that we are finished */ | |
57afd89f | 5276 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 LT |
5277 | |
5278 | if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && | |
ccfcc3c1 N |
5279 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
5280 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && | |
5fd6c1dc N |
5281 | mddev->curr_resync > 2) { |
5282 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
5283 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5284 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
5285 | printk(KERN_INFO | |
5286 | "md: checkpointing recovery of %s.\n", | |
5287 | mdname(mddev)); | |
5288 | mddev->recovery_cp = mddev->curr_resync; | |
5289 | } | |
5290 | } else | |
5291 | mddev->recovery_cp = MaxSector; | |
5292 | } else { | |
5293 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5294 | mddev->curr_resync = MaxSector; | |
5295 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5296 | if (rdev->raid_disk >= 0 && | |
5297 | !test_bit(Faulty, &rdev->flags) && | |
5298 | !test_bit(In_sync, &rdev->flags) && | |
5299 | rdev->recovery_offset < mddev->curr_resync) | |
5300 | rdev->recovery_offset = mddev->curr_resync; | |
5301 | mddev->sb_dirty = 1; | |
5302 | } | |
1da177e4 LT |
5303 | } |
5304 | ||
1da177e4 LT |
5305 | skip: |
5306 | mddev->curr_resync = 0; | |
5307 | wake_up(&resync_wait); | |
5308 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
5309 | md_wakeup_thread(mddev->thread); | |
5310 | } | |
29269553 | 5311 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
5312 | |
5313 | ||
5314 | /* | |
5315 | * This routine is regularly called by all per-raid-array threads to | |
5316 | * deal with generic issues like resync and super-block update. | |
5317 | * Raid personalities that don't have a thread (linear/raid0) do not | |
5318 | * need this as they never do any recovery or update the superblock. | |
5319 | * | |
5320 | * It does not do any resync itself, but rather "forks" off other threads | |
5321 | * to do that as needed. | |
5322 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
5323 | * "->recovery" and create a thread at ->sync_thread. | |
5324 | * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR) | |
5325 | * and wakeups up this thread which will reap the thread and finish up. | |
5326 | * This thread also removes any faulty devices (with nr_pending == 0). | |
5327 | * | |
5328 | * The overall approach is: | |
5329 | * 1/ if the superblock needs updating, update it. | |
5330 | * 2/ If a recovery thread is running, don't do anything else. | |
5331 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
5332 | * 4/ If there are any faulty devices, remove them. | |
5333 | * 5/ If array is degraded, try to add spares devices | |
5334 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
5335 | */ | |
5336 | void md_check_recovery(mddev_t *mddev) | |
5337 | { | |
5338 | mdk_rdev_t *rdev; | |
5339 | struct list_head *rtmp; | |
5340 | ||
5341 | ||
5f40402d N |
5342 | if (mddev->bitmap) |
5343 | bitmap_daemon_work(mddev->bitmap); | |
1da177e4 LT |
5344 | |
5345 | if (mddev->ro) | |
5346 | return; | |
fca4d848 N |
5347 | |
5348 | if (signal_pending(current)) { | |
5349 | if (mddev->pers->sync_request) { | |
5350 | printk(KERN_INFO "md: %s in immediate safe mode\n", | |
5351 | mdname(mddev)); | |
5352 | mddev->safemode = 2; | |
5353 | } | |
5354 | flush_signals(current); | |
5355 | } | |
5356 | ||
1da177e4 LT |
5357 | if ( ! ( |
5358 | mddev->sb_dirty || | |
5359 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || | |
fca4d848 N |
5360 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
5361 | (mddev->safemode == 1) || | |
5362 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) | |
5363 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
5364 | )) |
5365 | return; | |
fca4d848 | 5366 | |
df5b89b3 | 5367 | if (mddev_trylock(mddev)) { |
1da177e4 | 5368 | int spares =0; |
fca4d848 | 5369 | |
a9701a30 | 5370 | spin_lock_irq(&mddev->write_lock); |
fca4d848 N |
5371 | if (mddev->safemode && !atomic_read(&mddev->writes_pending) && |
5372 | !mddev->in_sync && mddev->recovery_cp == MaxSector) { | |
5373 | mddev->in_sync = 1; | |
42543769 | 5374 | mddev->sb_dirty = 3; |
fca4d848 N |
5375 | } |
5376 | if (mddev->safemode == 1) | |
5377 | mddev->safemode = 0; | |
a9701a30 | 5378 | spin_unlock_irq(&mddev->write_lock); |
fca4d848 | 5379 | |
1da177e4 LT |
5380 | if (mddev->sb_dirty) |
5381 | md_update_sb(mddev); | |
06d91a5f | 5382 | |
06d91a5f | 5383 | |
1da177e4 LT |
5384 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
5385 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
5386 | /* resync/recovery still happening */ | |
5387 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5388 | goto unlock; | |
5389 | } | |
5390 | if (mddev->sync_thread) { | |
5391 | /* resync has finished, collect result */ | |
5392 | md_unregister_thread(mddev->sync_thread); | |
5393 | mddev->sync_thread = NULL; | |
5394 | if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && | |
5395 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5396 | /* success...*/ | |
5397 | /* activate any spares */ | |
5398 | mddev->pers->spare_active(mddev); | |
5399 | } | |
5400 | md_update_sb(mddev); | |
41158c7e N |
5401 | |
5402 | /* if array is no-longer degraded, then any saved_raid_disk | |
5403 | * information must be scrapped | |
5404 | */ | |
5405 | if (!mddev->degraded) | |
5406 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5407 | rdev->saved_raid_disk = -1; | |
5408 | ||
1da177e4 LT |
5409 | mddev->recovery = 0; |
5410 | /* flag recovery needed just to double check */ | |
5411 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
d7603b7e | 5412 | md_new_event(mddev); |
1da177e4 LT |
5413 | goto unlock; |
5414 | } | |
24dd469d N |
5415 | /* Clear some bits that don't mean anything, but |
5416 | * might be left set | |
5417 | */ | |
5418 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5419 | clear_bit(MD_RECOVERY_ERR, &mddev->recovery); | |
5420 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5421 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 5422 | |
5fd6c1dc N |
5423 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
5424 | goto unlock; | |
1da177e4 LT |
5425 | /* no recovery is running. |
5426 | * remove any failed drives, then | |
5427 | * add spares if possible. | |
5428 | * Spare are also removed and re-added, to allow | |
5429 | * the personality to fail the re-add. | |
5430 | */ | |
5431 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5432 | if (rdev->raid_disk >= 0 && | |
b2d444d7 | 5433 | (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) && |
1da177e4 | 5434 | atomic_read(&rdev->nr_pending)==0) { |
86e6ffdd N |
5435 | if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) { |
5436 | char nm[20]; | |
5437 | sprintf(nm,"rd%d", rdev->raid_disk); | |
5438 | sysfs_remove_link(&mddev->kobj, nm); | |
1da177e4 | 5439 | rdev->raid_disk = -1; |
86e6ffdd | 5440 | } |
1da177e4 LT |
5441 | } |
5442 | ||
5443 | if (mddev->degraded) { | |
5444 | ITERATE_RDEV(mddev,rdev,rtmp) | |
5445 | if (rdev->raid_disk < 0 | |
b2d444d7 | 5446 | && !test_bit(Faulty, &rdev->flags)) { |
5fd6c1dc | 5447 | rdev->recovery_offset = 0; |
86e6ffdd N |
5448 | if (mddev->pers->hot_add_disk(mddev,rdev)) { |
5449 | char nm[20]; | |
5450 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5451 | sysfs_create_link(&mddev->kobj, &rdev->kobj, nm); | |
1da177e4 | 5452 | spares++; |
d7603b7e | 5453 | md_new_event(mddev); |
86e6ffdd | 5454 | } else |
1da177e4 LT |
5455 | break; |
5456 | } | |
5457 | } | |
5458 | ||
24dd469d N |
5459 | if (spares) { |
5460 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5461 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5462 | } else if (mddev->recovery_cp < MaxSector) { | |
5463 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5464 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
5465 | /* nothing to be done ... */ | |
1da177e4 | 5466 | goto unlock; |
24dd469d | 5467 | |
1da177e4 LT |
5468 | if (mddev->pers->sync_request) { |
5469 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
a654b9d8 N |
5470 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
5471 | /* We are adding a device or devices to an array | |
5472 | * which has the bitmap stored on all devices. | |
5473 | * So make sure all bitmap pages get written | |
5474 | */ | |
5475 | bitmap_write_all(mddev->bitmap); | |
5476 | } | |
1da177e4 LT |
5477 | mddev->sync_thread = md_register_thread(md_do_sync, |
5478 | mddev, | |
5479 | "%s_resync"); | |
5480 | if (!mddev->sync_thread) { | |
5481 | printk(KERN_ERR "%s: could not start resync" | |
5482 | " thread...\n", | |
5483 | mdname(mddev)); | |
5484 | /* leave the spares where they are, it shouldn't hurt */ | |
5485 | mddev->recovery = 0; | |
d7603b7e | 5486 | } else |
1da177e4 | 5487 | md_wakeup_thread(mddev->sync_thread); |
d7603b7e | 5488 | md_new_event(mddev); |
1da177e4 LT |
5489 | } |
5490 | unlock: | |
5491 | mddev_unlock(mddev); | |
5492 | } | |
5493 | } | |
5494 | ||
75c96f85 AB |
5495 | static int md_notify_reboot(struct notifier_block *this, |
5496 | unsigned long code, void *x) | |
1da177e4 LT |
5497 | { |
5498 | struct list_head *tmp; | |
5499 | mddev_t *mddev; | |
5500 | ||
5501 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
5502 | ||
5503 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
5504 | ||
5505 | ITERATE_MDDEV(mddev,tmp) | |
c71d4887 | 5506 | if (mddev_trylock(mddev)) { |
1da177e4 | 5507 | do_md_stop (mddev, 1); |
c71d4887 NB |
5508 | mddev_unlock(mddev); |
5509 | } | |
1da177e4 LT |
5510 | /* |
5511 | * certain more exotic SCSI devices are known to be | |
5512 | * volatile wrt too early system reboots. While the | |
5513 | * right place to handle this issue is the given | |
5514 | * driver, we do want to have a safe RAID driver ... | |
5515 | */ | |
5516 | mdelay(1000*1); | |
5517 | } | |
5518 | return NOTIFY_DONE; | |
5519 | } | |
5520 | ||
75c96f85 | 5521 | static struct notifier_block md_notifier = { |
1da177e4 LT |
5522 | .notifier_call = md_notify_reboot, |
5523 | .next = NULL, | |
5524 | .priority = INT_MAX, /* before any real devices */ | |
5525 | }; | |
5526 | ||
5527 | static void md_geninit(void) | |
5528 | { | |
5529 | struct proc_dir_entry *p; | |
5530 | ||
5531 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); | |
5532 | ||
5533 | p = create_proc_entry("mdstat", S_IRUGO, NULL); | |
5534 | if (p) | |
5535 | p->proc_fops = &md_seq_fops; | |
5536 | } | |
5537 | ||
75c96f85 | 5538 | static int __init md_init(void) |
1da177e4 | 5539 | { |
1da177e4 LT |
5540 | printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d," |
5541 | " MD_SB_DISKS=%d\n", | |
5542 | MD_MAJOR_VERSION, MD_MINOR_VERSION, | |
5543 | MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS); | |
bd926c63 | 5544 | printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI, |
32a7627c | 5545 | BITMAP_MINOR); |
1da177e4 LT |
5546 | |
5547 | if (register_blkdev(MAJOR_NR, "md")) | |
5548 | return -1; | |
5549 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
5550 | unregister_blkdev(MAJOR_NR, "md"); | |
5551 | return -1; | |
5552 | } | |
1da177e4 LT |
5553 | blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE, |
5554 | md_probe, NULL, NULL); | |
5555 | blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE, | |
5556 | md_probe, NULL, NULL); | |
5557 | ||
1da177e4 LT |
5558 | register_reboot_notifier(&md_notifier); |
5559 | raid_table_header = register_sysctl_table(raid_root_table, 1); | |
5560 | ||
5561 | md_geninit(); | |
5562 | return (0); | |
5563 | } | |
5564 | ||
5565 | ||
5566 | #ifndef MODULE | |
5567 | ||
5568 | /* | |
5569 | * Searches all registered partitions for autorun RAID arrays | |
5570 | * at boot time. | |
5571 | */ | |
5572 | static dev_t detected_devices[128]; | |
5573 | static int dev_cnt; | |
5574 | ||
5575 | void md_autodetect_dev(dev_t dev) | |
5576 | { | |
5577 | if (dev_cnt >= 0 && dev_cnt < 127) | |
5578 | detected_devices[dev_cnt++] = dev; | |
5579 | } | |
5580 | ||
5581 | ||
5582 | static void autostart_arrays(int part) | |
5583 | { | |
5584 | mdk_rdev_t *rdev; | |
5585 | int i; | |
5586 | ||
5587 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); | |
5588 | ||
5589 | for (i = 0; i < dev_cnt; i++) { | |
5590 | dev_t dev = detected_devices[i]; | |
5591 | ||
5592 | rdev = md_import_device(dev,0, 0); | |
5593 | if (IS_ERR(rdev)) | |
5594 | continue; | |
5595 | ||
b2d444d7 | 5596 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5597 | MD_BUG(); |
5598 | continue; | |
5599 | } | |
5600 | list_add(&rdev->same_set, &pending_raid_disks); | |
5601 | } | |
5602 | dev_cnt = 0; | |
5603 | ||
5604 | autorun_devices(part); | |
5605 | } | |
5606 | ||
5607 | #endif | |
5608 | ||
5609 | static __exit void md_exit(void) | |
5610 | { | |
5611 | mddev_t *mddev; | |
5612 | struct list_head *tmp; | |
8ab5e4c1 | 5613 | |
1da177e4 LT |
5614 | blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS); |
5615 | blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift); | |
1da177e4 LT |
5616 | |
5617 | unregister_blkdev(MAJOR_NR,"md"); | |
5618 | unregister_blkdev(mdp_major, "mdp"); | |
5619 | unregister_reboot_notifier(&md_notifier); | |
5620 | unregister_sysctl_table(raid_table_header); | |
5621 | remove_proc_entry("mdstat", NULL); | |
5622 | ITERATE_MDDEV(mddev,tmp) { | |
5623 | struct gendisk *disk = mddev->gendisk; | |
5624 | if (!disk) | |
5625 | continue; | |
5626 | export_array(mddev); | |
5627 | del_gendisk(disk); | |
5628 | put_disk(disk); | |
5629 | mddev->gendisk = NULL; | |
5630 | mddev_put(mddev); | |
5631 | } | |
5632 | } | |
5633 | ||
5634 | module_init(md_init) | |
5635 | module_exit(md_exit) | |
5636 | ||
f91de92e N |
5637 | static int get_ro(char *buffer, struct kernel_param *kp) |
5638 | { | |
5639 | return sprintf(buffer, "%d", start_readonly); | |
5640 | } | |
5641 | static int set_ro(const char *val, struct kernel_param *kp) | |
5642 | { | |
5643 | char *e; | |
5644 | int num = simple_strtoul(val, &e, 10); | |
5645 | if (*val && (*e == '\0' || *e == '\n')) { | |
5646 | start_readonly = num; | |
4dbcdc75 | 5647 | return 0; |
f91de92e N |
5648 | } |
5649 | return -EINVAL; | |
5650 | } | |
5651 | ||
5652 | module_param_call(start_ro, set_ro, get_ro, NULL, 0600); | |
6ff8d8ec N |
5653 | module_param(start_dirty_degraded, int, 0644); |
5654 | ||
f91de92e | 5655 | |
1da177e4 LT |
5656 | EXPORT_SYMBOL(register_md_personality); |
5657 | EXPORT_SYMBOL(unregister_md_personality); | |
5658 | EXPORT_SYMBOL(md_error); | |
5659 | EXPORT_SYMBOL(md_done_sync); | |
5660 | EXPORT_SYMBOL(md_write_start); | |
5661 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
5662 | EXPORT_SYMBOL(md_register_thread); |
5663 | EXPORT_SYMBOL(md_unregister_thread); | |
5664 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
5665 | EXPORT_SYMBOL(md_check_recovery); |
5666 | MODULE_LICENSE("GPL"); | |
aa1595e9 | 5667 | MODULE_ALIAS("md"); |
72008652 | 5668 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |