]>
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> | |
50511da3 | 36 | #include <linux/kernel.h> |
a6fb0934 | 37 | #include <linux/kthread.h> |
1da177e4 LT |
38 | #include <linux/linkage.h> |
39 | #include <linux/raid/md.h> | |
32a7627c | 40 | #include <linux/raid/bitmap.h> |
1da177e4 | 41 | #include <linux/sysctl.h> |
1da177e4 | 42 | #include <linux/buffer_head.h> /* for invalidate_bdev */ |
d7603b7e | 43 | #include <linux/poll.h> |
48c9c27b | 44 | #include <linux/mutex.h> |
16f17b39 | 45 | #include <linux/ctype.h> |
7dfb7103 | 46 | #include <linux/freezer.h> |
1da177e4 LT |
47 | |
48 | #include <linux/init.h> | |
49 | ||
32a7627c N |
50 | #include <linux/file.h> |
51 | ||
1da177e4 LT |
52 | #ifdef CONFIG_KMOD |
53 | #include <linux/kmod.h> | |
54 | #endif | |
55 | ||
56 | #include <asm/unaligned.h> | |
57 | ||
58 | #define MAJOR_NR MD_MAJOR | |
59 | #define MD_DRIVER | |
60 | ||
61 | /* 63 partitions with the alternate major number (mdp) */ | |
62 | #define MdpMinorShift 6 | |
63 | ||
64 | #define DEBUG 0 | |
65 | #define dprintk(x...) ((void)(DEBUG && printk(x))) | |
66 | ||
67 | ||
68 | #ifndef MODULE | |
69 | static void autostart_arrays (int part); | |
70 | #endif | |
71 | ||
2604b703 | 72 | static LIST_HEAD(pers_list); |
1da177e4 LT |
73 | static DEFINE_SPINLOCK(pers_lock); |
74 | ||
5e56341d AB |
75 | static void md_print_devices(void); |
76 | ||
90b08710 BS |
77 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
78 | ||
5e56341d AB |
79 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } |
80 | ||
1da177e4 LT |
81 | /* |
82 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
83 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
84 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 85 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
86 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
87 | * speed limit - in case reconstruction slows down your system despite | |
88 | * idle IO detection. | |
89 | * | |
90 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 91 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
92 | */ |
93 | ||
94 | static int sysctl_speed_limit_min = 1000; | |
95 | static int sysctl_speed_limit_max = 200000; | |
88202a0c N |
96 | static inline int speed_min(mddev_t *mddev) |
97 | { | |
98 | return mddev->sync_speed_min ? | |
99 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
100 | } | |
101 | ||
102 | static inline int speed_max(mddev_t *mddev) | |
103 | { | |
104 | return mddev->sync_speed_max ? | |
105 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
106 | } | |
1da177e4 LT |
107 | |
108 | static struct ctl_table_header *raid_table_header; | |
109 | ||
110 | static ctl_table raid_table[] = { | |
111 | { | |
112 | .ctl_name = DEV_RAID_SPEED_LIMIT_MIN, | |
113 | .procname = "speed_limit_min", | |
114 | .data = &sysctl_speed_limit_min, | |
115 | .maxlen = sizeof(int), | |
80ca3a44 | 116 | .mode = S_IRUGO|S_IWUSR, |
1da177e4 LT |
117 | .proc_handler = &proc_dointvec, |
118 | }, | |
119 | { | |
120 | .ctl_name = DEV_RAID_SPEED_LIMIT_MAX, | |
121 | .procname = "speed_limit_max", | |
122 | .data = &sysctl_speed_limit_max, | |
123 | .maxlen = sizeof(int), | |
80ca3a44 | 124 | .mode = S_IRUGO|S_IWUSR, |
1da177e4 LT |
125 | .proc_handler = &proc_dointvec, |
126 | }, | |
127 | { .ctl_name = 0 } | |
128 | }; | |
129 | ||
130 | static ctl_table raid_dir_table[] = { | |
131 | { | |
132 | .ctl_name = DEV_RAID, | |
133 | .procname = "raid", | |
134 | .maxlen = 0, | |
80ca3a44 | 135 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
136 | .child = raid_table, |
137 | }, | |
138 | { .ctl_name = 0 } | |
139 | }; | |
140 | ||
141 | static ctl_table raid_root_table[] = { | |
142 | { | |
143 | .ctl_name = CTL_DEV, | |
144 | .procname = "dev", | |
145 | .maxlen = 0, | |
146 | .mode = 0555, | |
147 | .child = raid_dir_table, | |
148 | }, | |
149 | { .ctl_name = 0 } | |
150 | }; | |
151 | ||
152 | static struct block_device_operations md_fops; | |
153 | ||
f91de92e N |
154 | static int start_readonly; |
155 | ||
d7603b7e N |
156 | /* |
157 | * We have a system wide 'event count' that is incremented | |
158 | * on any 'interesting' event, and readers of /proc/mdstat | |
159 | * can use 'poll' or 'select' to find out when the event | |
160 | * count increases. | |
161 | * | |
162 | * Events are: | |
163 | * start array, stop array, error, add device, remove device, | |
164 | * start build, activate spare | |
165 | */ | |
2989ddbd | 166 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 167 | static atomic_t md_event_count; |
29269553 | 168 | void md_new_event(mddev_t *mddev) |
d7603b7e N |
169 | { |
170 | atomic_inc(&md_event_count); | |
171 | wake_up(&md_event_waiters); | |
172 | } | |
29269553 | 173 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 174 | |
c331eb04 N |
175 | /* Alternate version that can be called from interrupts |
176 | * when calling sysfs_notify isn't needed. | |
177 | */ | |
05381954 | 178 | static void md_new_event_inintr(mddev_t *mddev) |
c331eb04 N |
179 | { |
180 | atomic_inc(&md_event_count); | |
181 | wake_up(&md_event_waiters); | |
182 | } | |
183 | ||
1da177e4 LT |
184 | /* |
185 | * Enables to iterate over all existing md arrays | |
186 | * all_mddevs_lock protects this list. | |
187 | */ | |
188 | static LIST_HEAD(all_mddevs); | |
189 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
190 | ||
191 | ||
192 | /* | |
193 | * iterates through all used mddevs in the system. | |
194 | * We take care to grab the all_mddevs_lock whenever navigating | |
195 | * the list, and to always hold a refcount when unlocked. | |
196 | * Any code which breaks out of this loop while own | |
197 | * a reference to the current mddev and must mddev_put it. | |
198 | */ | |
29ac4aa3 | 199 | #define for_each_mddev(mddev,tmp) \ |
1da177e4 LT |
200 | \ |
201 | for (({ spin_lock(&all_mddevs_lock); \ | |
202 | tmp = all_mddevs.next; \ | |
203 | mddev = NULL;}); \ | |
204 | ({ if (tmp != &all_mddevs) \ | |
205 | mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ | |
206 | spin_unlock(&all_mddevs_lock); \ | |
207 | if (mddev) mddev_put(mddev); \ | |
208 | mddev = list_entry(tmp, mddev_t, all_mddevs); \ | |
209 | tmp != &all_mddevs;}); \ | |
210 | ({ spin_lock(&all_mddevs_lock); \ | |
211 | tmp = tmp->next;}) \ | |
212 | ) | |
213 | ||
214 | ||
165125e1 | 215 | static int md_fail_request (struct request_queue *q, struct bio *bio) |
1da177e4 | 216 | { |
6712ecf8 | 217 | bio_io_error(bio); |
1da177e4 LT |
218 | return 0; |
219 | } | |
220 | ||
221 | static inline mddev_t *mddev_get(mddev_t *mddev) | |
222 | { | |
223 | atomic_inc(&mddev->active); | |
224 | return mddev; | |
225 | } | |
226 | ||
227 | static void mddev_put(mddev_t *mddev) | |
228 | { | |
229 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
230 | return; | |
231 | if (!mddev->raid_disks && list_empty(&mddev->disks)) { | |
232 | list_del(&mddev->all_mddevs); | |
926ce2d8 | 233 | spin_unlock(&all_mddevs_lock); |
1312f40e | 234 | blk_cleanup_queue(mddev->queue); |
c10997f6 | 235 | kobject_put(&mddev->kobj); |
926ce2d8 N |
236 | } else |
237 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
238 | } |
239 | ||
240 | static mddev_t * mddev_find(dev_t unit) | |
241 | { | |
242 | mddev_t *mddev, *new = NULL; | |
243 | ||
244 | retry: | |
245 | spin_lock(&all_mddevs_lock); | |
246 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
247 | if (mddev->unit == unit) { | |
248 | mddev_get(mddev); | |
249 | spin_unlock(&all_mddevs_lock); | |
990a8baf | 250 | kfree(new); |
1da177e4 LT |
251 | return mddev; |
252 | } | |
253 | ||
254 | if (new) { | |
255 | list_add(&new->all_mddevs, &all_mddevs); | |
256 | spin_unlock(&all_mddevs_lock); | |
257 | return new; | |
258 | } | |
259 | spin_unlock(&all_mddevs_lock); | |
260 | ||
9ffae0cf | 261 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
262 | if (!new) |
263 | return NULL; | |
264 | ||
1da177e4 LT |
265 | new->unit = unit; |
266 | if (MAJOR(unit) == MD_MAJOR) | |
267 | new->md_minor = MINOR(unit); | |
268 | else | |
269 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
270 | ||
df5b89b3 | 271 | mutex_init(&new->reconfig_mutex); |
1da177e4 LT |
272 | INIT_LIST_HEAD(&new->disks); |
273 | INIT_LIST_HEAD(&new->all_mddevs); | |
274 | init_timer(&new->safemode_timer); | |
275 | atomic_set(&new->active, 1); | |
06d91a5f | 276 | spin_lock_init(&new->write_lock); |
3d310eb7 | 277 | init_waitqueue_head(&new->sb_wait); |
a6d8113a | 278 | init_waitqueue_head(&new->recovery_wait); |
08a02ecd | 279 | new->reshape_position = MaxSector; |
5e96ee65 | 280 | new->resync_min = 0; |
c6207277 | 281 | new->resync_max = MaxSector; |
d897dbf9 | 282 | new->level = LEVEL_NONE; |
1da177e4 LT |
283 | |
284 | new->queue = blk_alloc_queue(GFP_KERNEL); | |
285 | if (!new->queue) { | |
286 | kfree(new); | |
287 | return NULL; | |
288 | } | |
75ad23bc NP |
289 | /* Can be unlocked because the queue is new: no concurrency */ |
290 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue); | |
1da177e4 LT |
291 | |
292 | blk_queue_make_request(new->queue, md_fail_request); | |
293 | ||
294 | goto retry; | |
295 | } | |
296 | ||
297 | static inline int mddev_lock(mddev_t * mddev) | |
298 | { | |
df5b89b3 | 299 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
300 | } |
301 | ||
1da177e4 LT |
302 | static inline int mddev_trylock(mddev_t * mddev) |
303 | { | |
df5b89b3 | 304 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
305 | } |
306 | ||
307 | static inline void mddev_unlock(mddev_t * mddev) | |
308 | { | |
df5b89b3 | 309 | mutex_unlock(&mddev->reconfig_mutex); |
1da177e4 | 310 | |
005eca5e | 311 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
312 | } |
313 | ||
2989ddbd | 314 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 LT |
315 | { |
316 | mdk_rdev_t * rdev; | |
317 | struct list_head *tmp; | |
318 | ||
d089c6af | 319 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
320 | if (rdev->desc_nr == nr) |
321 | return rdev; | |
322 | } | |
323 | return NULL; | |
324 | } | |
325 | ||
326 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
327 | { | |
328 | struct list_head *tmp; | |
329 | mdk_rdev_t *rdev; | |
330 | ||
d089c6af | 331 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
332 | if (rdev->bdev->bd_dev == dev) |
333 | return rdev; | |
334 | } | |
335 | return NULL; | |
336 | } | |
337 | ||
d9d166c2 | 338 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
339 | { |
340 | struct mdk_personality *pers; | |
d9d166c2 N |
341 | list_for_each_entry(pers, &pers_list, list) { |
342 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 343 | return pers; |
d9d166c2 N |
344 | if (strcmp(pers->name, clevel)==0) |
345 | return pers; | |
346 | } | |
2604b703 N |
347 | return NULL; |
348 | } | |
349 | ||
77933d72 | 350 | static inline sector_t calc_dev_sboffset(struct block_device *bdev) |
1da177e4 LT |
351 | { |
352 | sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
353 | return MD_NEW_SIZE_BLOCKS(size); | |
354 | } | |
355 | ||
356 | static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size) | |
357 | { | |
358 | sector_t size; | |
359 | ||
360 | size = rdev->sb_offset; | |
361 | ||
362 | if (chunk_size) | |
363 | size &= ~((sector_t)chunk_size/1024 - 1); | |
364 | return size; | |
365 | } | |
366 | ||
367 | static int alloc_disk_sb(mdk_rdev_t * rdev) | |
368 | { | |
369 | if (rdev->sb_page) | |
370 | MD_BUG(); | |
371 | ||
372 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
373 | if (!rdev->sb_page) { | |
374 | printk(KERN_ALERT "md: out of memory.\n"); | |
375 | return -EINVAL; | |
376 | } | |
377 | ||
378 | return 0; | |
379 | } | |
380 | ||
381 | static void free_disk_sb(mdk_rdev_t * rdev) | |
382 | { | |
383 | if (rdev->sb_page) { | |
2d1f3b5d | 384 | put_page(rdev->sb_page); |
1da177e4 LT |
385 | rdev->sb_loaded = 0; |
386 | rdev->sb_page = NULL; | |
387 | rdev->sb_offset = 0; | |
388 | rdev->size = 0; | |
389 | } | |
390 | } | |
391 | ||
392 | ||
6712ecf8 | 393 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
394 | { |
395 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 396 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 397 | |
3a0f5bbb N |
398 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
399 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
400 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
401 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 402 | md_error(mddev, rdev); |
3a0f5bbb | 403 | } |
7bfa19f2 | 404 | |
a9701a30 N |
405 | if (atomic_dec_and_test(&mddev->pending_writes)) |
406 | wake_up(&mddev->sb_wait); | |
f8b58edf | 407 | bio_put(bio); |
7bfa19f2 N |
408 | } |
409 | ||
6712ecf8 | 410 | static void super_written_barrier(struct bio *bio, int error) |
a9701a30 N |
411 | { |
412 | struct bio *bio2 = bio->bi_private; | |
413 | mdk_rdev_t *rdev = bio2->bi_private; | |
414 | mddev_t *mddev = rdev->mddev; | |
a9701a30 N |
415 | |
416 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
417 | error == -EOPNOTSUPP) { | |
418 | unsigned long flags; | |
419 | /* barriers don't appear to be supported :-( */ | |
420 | set_bit(BarriersNotsupp, &rdev->flags); | |
421 | mddev->barriers_work = 0; | |
422 | spin_lock_irqsave(&mddev->write_lock, flags); | |
423 | bio2->bi_next = mddev->biolist; | |
424 | mddev->biolist = bio2; | |
425 | spin_unlock_irqrestore(&mddev->write_lock, flags); | |
426 | wake_up(&mddev->sb_wait); | |
427 | bio_put(bio); | |
6712ecf8 N |
428 | } else { |
429 | bio_put(bio2); | |
430 | bio->bi_private = rdev; | |
431 | super_written(bio, error); | |
a9701a30 | 432 | } |
a9701a30 N |
433 | } |
434 | ||
7bfa19f2 N |
435 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, |
436 | sector_t sector, int size, struct page *page) | |
437 | { | |
438 | /* write first size bytes of page to sector of rdev | |
439 | * Increment mddev->pending_writes before returning | |
440 | * and decrement it on completion, waking up sb_wait | |
441 | * if zero is reached. | |
442 | * If an error occurred, call md_error | |
a9701a30 N |
443 | * |
444 | * As we might need to resubmit the request if BIO_RW_BARRIER | |
445 | * causes ENOTSUPP, we allocate a spare bio... | |
7bfa19f2 N |
446 | */ |
447 | struct bio *bio = bio_alloc(GFP_NOIO, 1); | |
a9701a30 | 448 | int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC); |
7bfa19f2 N |
449 | |
450 | bio->bi_bdev = rdev->bdev; | |
451 | bio->bi_sector = sector; | |
452 | bio_add_page(bio, page, size, 0); | |
453 | bio->bi_private = rdev; | |
454 | bio->bi_end_io = super_written; | |
a9701a30 N |
455 | bio->bi_rw = rw; |
456 | ||
7bfa19f2 | 457 | atomic_inc(&mddev->pending_writes); |
a9701a30 N |
458 | if (!test_bit(BarriersNotsupp, &rdev->flags)) { |
459 | struct bio *rbio; | |
460 | rw |= (1<<BIO_RW_BARRIER); | |
461 | rbio = bio_clone(bio, GFP_NOIO); | |
462 | rbio->bi_private = bio; | |
463 | rbio->bi_end_io = super_written_barrier; | |
464 | submit_bio(rw, rbio); | |
465 | } else | |
466 | submit_bio(rw, bio); | |
467 | } | |
468 | ||
469 | void md_super_wait(mddev_t *mddev) | |
470 | { | |
471 | /* wait for all superblock writes that were scheduled to complete. | |
472 | * if any had to be retried (due to BARRIER problems), retry them | |
473 | */ | |
474 | DEFINE_WAIT(wq); | |
475 | for(;;) { | |
476 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
477 | if (atomic_read(&mddev->pending_writes)==0) | |
478 | break; | |
479 | while (mddev->biolist) { | |
480 | struct bio *bio; | |
481 | spin_lock_irq(&mddev->write_lock); | |
482 | bio = mddev->biolist; | |
483 | mddev->biolist = bio->bi_next ; | |
484 | bio->bi_next = NULL; | |
485 | spin_unlock_irq(&mddev->write_lock); | |
486 | submit_bio(bio->bi_rw, bio); | |
487 | } | |
488 | schedule(); | |
489 | } | |
490 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
491 | } |
492 | ||
6712ecf8 | 493 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 494 | { |
1da177e4 | 495 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
496 | } |
497 | ||
a654b9d8 | 498 | int sync_page_io(struct block_device *bdev, sector_t sector, int size, |
1da177e4 LT |
499 | struct page *page, int rw) |
500 | { | |
baaa2c51 | 501 | struct bio *bio = bio_alloc(GFP_NOIO, 1); |
1da177e4 LT |
502 | struct completion event; |
503 | int ret; | |
504 | ||
505 | rw |= (1 << BIO_RW_SYNC); | |
506 | ||
507 | bio->bi_bdev = bdev; | |
508 | bio->bi_sector = sector; | |
509 | bio_add_page(bio, page, size, 0); | |
510 | init_completion(&event); | |
511 | bio->bi_private = &event; | |
512 | bio->bi_end_io = bi_complete; | |
513 | submit_bio(rw, bio); | |
514 | wait_for_completion(&event); | |
515 | ||
516 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
517 | bio_put(bio); | |
518 | return ret; | |
519 | } | |
a8745db2 | 520 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 521 | |
0002b271 | 522 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
523 | { |
524 | char b[BDEVNAME_SIZE]; | |
525 | if (!rdev->sb_page) { | |
526 | MD_BUG(); | |
527 | return -EINVAL; | |
528 | } | |
529 | if (rdev->sb_loaded) | |
530 | return 0; | |
531 | ||
532 | ||
0002b271 | 533 | if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ)) |
1da177e4 LT |
534 | goto fail; |
535 | rdev->sb_loaded = 1; | |
536 | return 0; | |
537 | ||
538 | fail: | |
539 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
540 | bdevname(rdev->bdev,b)); | |
541 | return -EINVAL; | |
542 | } | |
543 | ||
544 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
545 | { | |
05710466 AN |
546 | return sb1->set_uuid0 == sb2->set_uuid0 && |
547 | sb1->set_uuid1 == sb2->set_uuid1 && | |
548 | sb1->set_uuid2 == sb2->set_uuid2 && | |
549 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
550 | } |
551 | ||
1da177e4 LT |
552 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
553 | { | |
554 | int ret; | |
555 | mdp_super_t *tmp1, *tmp2; | |
556 | ||
557 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
558 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
559 | ||
560 | if (!tmp1 || !tmp2) { | |
561 | ret = 0; | |
35020f1a | 562 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
563 | goto abort; |
564 | } | |
565 | ||
566 | *tmp1 = *sb1; | |
567 | *tmp2 = *sb2; | |
568 | ||
569 | /* | |
570 | * nr_disks is not constant | |
571 | */ | |
572 | tmp1->nr_disks = 0; | |
573 | tmp2->nr_disks = 0; | |
574 | ||
ce0c8e05 | 575 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 576 | abort: |
990a8baf JJ |
577 | kfree(tmp1); |
578 | kfree(tmp2); | |
1da177e4 LT |
579 | return ret; |
580 | } | |
581 | ||
4d167f09 N |
582 | |
583 | static u32 md_csum_fold(u32 csum) | |
584 | { | |
585 | csum = (csum & 0xffff) + (csum >> 16); | |
586 | return (csum & 0xffff) + (csum >> 16); | |
587 | } | |
588 | ||
1da177e4 LT |
589 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
590 | { | |
4d167f09 N |
591 | u64 newcsum = 0; |
592 | u32 *sb32 = (u32*)sb; | |
593 | int i; | |
1da177e4 LT |
594 | unsigned int disk_csum, csum; |
595 | ||
596 | disk_csum = sb->sb_csum; | |
597 | sb->sb_csum = 0; | |
4d167f09 N |
598 | |
599 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
600 | newcsum += sb32[i]; | |
601 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
602 | ||
603 | ||
604 | #ifdef CONFIG_ALPHA | |
605 | /* This used to use csum_partial, which was wrong for several | |
606 | * reasons including that different results are returned on | |
607 | * different architectures. It isn't critical that we get exactly | |
608 | * the same return value as before (we always csum_fold before | |
609 | * testing, and that removes any differences). However as we | |
610 | * know that csum_partial always returned a 16bit value on | |
611 | * alphas, do a fold to maximise conformity to previous behaviour. | |
612 | */ | |
613 | sb->sb_csum = md_csum_fold(disk_csum); | |
614 | #else | |
1da177e4 | 615 | sb->sb_csum = disk_csum; |
4d167f09 | 616 | #endif |
1da177e4 LT |
617 | return csum; |
618 | } | |
619 | ||
620 | ||
621 | /* | |
622 | * Handle superblock details. | |
623 | * We want to be able to handle multiple superblock formats | |
624 | * so we have a common interface to them all, and an array of | |
625 | * different handlers. | |
626 | * We rely on user-space to write the initial superblock, and support | |
627 | * reading and updating of superblocks. | |
628 | * Interface methods are: | |
629 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
630 | * loads and validates a superblock on dev. | |
631 | * if refdev != NULL, compare superblocks on both devices | |
632 | * Return: | |
633 | * 0 - dev has a superblock that is compatible with refdev | |
634 | * 1 - dev has a superblock that is compatible and newer than refdev | |
635 | * so dev should be used as the refdev in future | |
636 | * -EINVAL superblock incompatible or invalid | |
637 | * -othererror e.g. -EIO | |
638 | * | |
639 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
640 | * Verify that dev is acceptable into mddev. | |
641 | * The first time, mddev->raid_disks will be 0, and data from | |
642 | * dev should be merged in. Subsequent calls check that dev | |
643 | * is new enough. Return 0 or -EINVAL | |
644 | * | |
645 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
646 | * Update the superblock for rdev with data in mddev | |
647 | * This does not write to disc. | |
648 | * | |
649 | */ | |
650 | ||
651 | struct super_type { | |
0cd17fec CW |
652 | char *name; |
653 | struct module *owner; | |
654 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
655 | int minor_version); | |
656 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
657 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
658 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
659 | unsigned long long size); | |
1da177e4 LT |
660 | }; |
661 | ||
662 | /* | |
663 | * load_super for 0.90.0 | |
664 | */ | |
665 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
666 | { | |
667 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
668 | mdp_super_t *sb; | |
669 | int ret; | |
670 | sector_t sb_offset; | |
671 | ||
672 | /* | |
673 | * Calculate the position of the superblock, | |
674 | * it's at the end of the disk. | |
675 | * | |
676 | * It also happens to be a multiple of 4Kb. | |
677 | */ | |
678 | sb_offset = calc_dev_sboffset(rdev->bdev); | |
679 | rdev->sb_offset = sb_offset; | |
680 | ||
0002b271 | 681 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
682 | if (ret) return ret; |
683 | ||
684 | ret = -EINVAL; | |
685 | ||
686 | bdevname(rdev->bdev, b); | |
687 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
688 | ||
689 | if (sb->md_magic != MD_SB_MAGIC) { | |
690 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
691 | b); | |
692 | goto abort; | |
693 | } | |
694 | ||
695 | if (sb->major_version != 0 || | |
f6705578 N |
696 | sb->minor_version < 90 || |
697 | sb->minor_version > 91) { | |
1da177e4 LT |
698 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
699 | sb->major_version, sb->minor_version, | |
700 | b); | |
701 | goto abort; | |
702 | } | |
703 | ||
704 | if (sb->raid_disks <= 0) | |
705 | goto abort; | |
706 | ||
4d167f09 | 707 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
708 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
709 | b); | |
710 | goto abort; | |
711 | } | |
712 | ||
713 | rdev->preferred_minor = sb->md_minor; | |
714 | rdev->data_offset = 0; | |
0002b271 | 715 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 | 716 | |
e11e93fa N |
717 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) { |
718 | if (sb->level != 1 && sb->level != 4 | |
719 | && sb->level != 5 && sb->level != 6 | |
720 | && sb->level != 10) { | |
721 | /* FIXME use a better test */ | |
722 | printk(KERN_WARNING | |
723 | "md: bitmaps not supported for this level.\n"); | |
724 | goto abort; | |
725 | } | |
726 | } | |
727 | ||
1da177e4 LT |
728 | if (sb->level == LEVEL_MULTIPATH) |
729 | rdev->desc_nr = -1; | |
730 | else | |
731 | rdev->desc_nr = sb->this_disk.number; | |
732 | ||
9a7b2b0f | 733 | if (!refdev) { |
1da177e4 | 734 | ret = 1; |
9a7b2b0f | 735 | } else { |
1da177e4 LT |
736 | __u64 ev1, ev2; |
737 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
738 | if (!uuid_equal(refsb, sb)) { | |
739 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
740 | b, bdevname(refdev->bdev,b2)); | |
741 | goto abort; | |
742 | } | |
743 | if (!sb_equal(refsb, sb)) { | |
744 | printk(KERN_WARNING "md: %s has same UUID" | |
745 | " but different superblock to %s\n", | |
746 | b, bdevname(refdev->bdev, b2)); | |
747 | goto abort; | |
748 | } | |
749 | ev1 = md_event(sb); | |
750 | ev2 = md_event(refsb); | |
751 | if (ev1 > ev2) | |
752 | ret = 1; | |
753 | else | |
754 | ret = 0; | |
755 | } | |
756 | rdev->size = calc_dev_size(rdev, sb->chunk_size); | |
757 | ||
2bf071bf N |
758 | if (rdev->size < sb->size && sb->level > 1) |
759 | /* "this cannot possibly happen" ... */ | |
760 | ret = -EINVAL; | |
761 | ||
1da177e4 LT |
762 | abort: |
763 | return ret; | |
764 | } | |
765 | ||
766 | /* | |
767 | * validate_super for 0.90.0 | |
768 | */ | |
769 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
770 | { | |
771 | mdp_disk_t *desc; | |
772 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 773 | __u64 ev1 = md_event(sb); |
1da177e4 | 774 | |
41158c7e | 775 | rdev->raid_disk = -1; |
c5d79adb N |
776 | clear_bit(Faulty, &rdev->flags); |
777 | clear_bit(In_sync, &rdev->flags); | |
778 | clear_bit(WriteMostly, &rdev->flags); | |
779 | clear_bit(BarriersNotsupp, &rdev->flags); | |
780 | ||
1da177e4 LT |
781 | if (mddev->raid_disks == 0) { |
782 | mddev->major_version = 0; | |
783 | mddev->minor_version = sb->minor_version; | |
784 | mddev->patch_version = sb->patch_version; | |
e691063a | 785 | mddev->external = 0; |
1da177e4 LT |
786 | mddev->chunk_size = sb->chunk_size; |
787 | mddev->ctime = sb->ctime; | |
788 | mddev->utime = sb->utime; | |
789 | mddev->level = sb->level; | |
d9d166c2 | 790 | mddev->clevel[0] = 0; |
1da177e4 LT |
791 | mddev->layout = sb->layout; |
792 | mddev->raid_disks = sb->raid_disks; | |
793 | mddev->size = sb->size; | |
07d84d10 | 794 | mddev->events = ev1; |
9223214e | 795 | mddev->bitmap_offset = 0; |
36fa3063 | 796 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
1da177e4 | 797 | |
f6705578 N |
798 | if (mddev->minor_version >= 91) { |
799 | mddev->reshape_position = sb->reshape_position; | |
800 | mddev->delta_disks = sb->delta_disks; | |
801 | mddev->new_level = sb->new_level; | |
802 | mddev->new_layout = sb->new_layout; | |
803 | mddev->new_chunk = sb->new_chunk; | |
804 | } else { | |
805 | mddev->reshape_position = MaxSector; | |
806 | mddev->delta_disks = 0; | |
807 | mddev->new_level = mddev->level; | |
808 | mddev->new_layout = mddev->layout; | |
809 | mddev->new_chunk = mddev->chunk_size; | |
810 | } | |
811 | ||
1da177e4 LT |
812 | if (sb->state & (1<<MD_SB_CLEAN)) |
813 | mddev->recovery_cp = MaxSector; | |
814 | else { | |
815 | if (sb->events_hi == sb->cp_events_hi && | |
816 | sb->events_lo == sb->cp_events_lo) { | |
817 | mddev->recovery_cp = sb->recovery_cp; | |
818 | } else | |
819 | mddev->recovery_cp = 0; | |
820 | } | |
821 | ||
822 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
823 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
824 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
825 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
826 | ||
827 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
828 | |
829 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
e11e93fa | 830 | mddev->bitmap_file == NULL) |
36fa3063 | 831 | mddev->bitmap_offset = mddev->default_bitmap_offset; |
a654b9d8 | 832 | |
41158c7e N |
833 | } else if (mddev->pers == NULL) { |
834 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
835 | ++ev1; |
836 | if (ev1 < mddev->events) | |
837 | return -EINVAL; | |
41158c7e N |
838 | } else if (mddev->bitmap) { |
839 | /* if adding to array with a bitmap, then we can accept an | |
840 | * older device ... but not too old. | |
841 | */ | |
41158c7e N |
842 | if (ev1 < mddev->bitmap->events_cleared) |
843 | return 0; | |
07d84d10 N |
844 | } else { |
845 | if (ev1 < mddev->events) | |
846 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
847 | return 0; | |
848 | } | |
41158c7e | 849 | |
1da177e4 | 850 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
851 | desc = sb->disks + rdev->desc_nr; |
852 | ||
853 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 854 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
855 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
856 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 857 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
858 | rdev->raid_disk = desc->raid_disk; |
859 | } | |
8ddf9efe N |
860 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
861 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 862 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 863 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
864 | return 0; |
865 | } | |
866 | ||
867 | /* | |
868 | * sync_super for 0.90.0 | |
869 | */ | |
870 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
871 | { | |
872 | mdp_super_t *sb; | |
873 | struct list_head *tmp; | |
874 | mdk_rdev_t *rdev2; | |
875 | int next_spare = mddev->raid_disks; | |
19133a42 | 876 | |
1da177e4 LT |
877 | |
878 | /* make rdev->sb match mddev data.. | |
879 | * | |
880 | * 1/ zero out disks | |
881 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
882 | * 3/ any empty disks < next_spare become removed | |
883 | * | |
884 | * disks[0] gets initialised to REMOVED because | |
885 | * we cannot be sure from other fields if it has | |
886 | * been initialised or not. | |
887 | */ | |
888 | int i; | |
889 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
890 | ||
61181565 N |
891 | rdev->sb_size = MD_SB_BYTES; |
892 | ||
1da177e4 LT |
893 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
894 | ||
895 | memset(sb, 0, sizeof(*sb)); | |
896 | ||
897 | sb->md_magic = MD_SB_MAGIC; | |
898 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
899 | sb->patch_version = mddev->patch_version; |
900 | sb->gvalid_words = 0; /* ignored */ | |
901 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
902 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
903 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
904 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
905 | ||
906 | sb->ctime = mddev->ctime; | |
907 | sb->level = mddev->level; | |
908 | sb->size = mddev->size; | |
909 | sb->raid_disks = mddev->raid_disks; | |
910 | sb->md_minor = mddev->md_minor; | |
e691063a | 911 | sb->not_persistent = 0; |
1da177e4 LT |
912 | sb->utime = mddev->utime; |
913 | sb->state = 0; | |
914 | sb->events_hi = (mddev->events>>32); | |
915 | sb->events_lo = (u32)mddev->events; | |
916 | ||
f6705578 N |
917 | if (mddev->reshape_position == MaxSector) |
918 | sb->minor_version = 90; | |
919 | else { | |
920 | sb->minor_version = 91; | |
921 | sb->reshape_position = mddev->reshape_position; | |
922 | sb->new_level = mddev->new_level; | |
923 | sb->delta_disks = mddev->delta_disks; | |
924 | sb->new_layout = mddev->new_layout; | |
925 | sb->new_chunk = mddev->new_chunk; | |
926 | } | |
927 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
928 | if (mddev->in_sync) |
929 | { | |
930 | sb->recovery_cp = mddev->recovery_cp; | |
931 | sb->cp_events_hi = (mddev->events>>32); | |
932 | sb->cp_events_lo = (u32)mddev->events; | |
933 | if (mddev->recovery_cp == MaxSector) | |
934 | sb->state = (1<< MD_SB_CLEAN); | |
935 | } else | |
936 | sb->recovery_cp = 0; | |
937 | ||
938 | sb->layout = mddev->layout; | |
939 | sb->chunk_size = mddev->chunk_size; | |
940 | ||
a654b9d8 N |
941 | if (mddev->bitmap && mddev->bitmap_file == NULL) |
942 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); | |
943 | ||
1da177e4 | 944 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
d089c6af | 945 | rdev_for_each(rdev2, tmp, mddev) { |
1da177e4 | 946 | mdp_disk_t *d; |
86e6ffdd | 947 | int desc_nr; |
b2d444d7 N |
948 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
949 | && !test_bit(Faulty, &rdev2->flags)) | |
86e6ffdd | 950 | desc_nr = rdev2->raid_disk; |
1da177e4 | 951 | else |
86e6ffdd | 952 | desc_nr = next_spare++; |
19133a42 | 953 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
954 | d = &sb->disks[rdev2->desc_nr]; |
955 | nr_disks++; | |
956 | d->number = rdev2->desc_nr; | |
957 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
958 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
b2d444d7 N |
959 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
960 | && !test_bit(Faulty, &rdev2->flags)) | |
1da177e4 LT |
961 | d->raid_disk = rdev2->raid_disk; |
962 | else | |
963 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 964 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 965 | d->state = (1<<MD_DISK_FAULTY); |
1be7892f | 966 | else if (test_bit(In_sync, &rdev2->flags)) { |
1da177e4 LT |
967 | d->state = (1<<MD_DISK_ACTIVE); |
968 | d->state |= (1<<MD_DISK_SYNC); | |
969 | active++; | |
970 | working++; | |
971 | } else { | |
972 | d->state = 0; | |
973 | spare++; | |
974 | working++; | |
975 | } | |
8ddf9efe N |
976 | if (test_bit(WriteMostly, &rdev2->flags)) |
977 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 978 | } |
1da177e4 LT |
979 | /* now set the "removed" and "faulty" bits on any missing devices */ |
980 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
981 | mdp_disk_t *d = &sb->disks[i]; | |
982 | if (d->state == 0 && d->number == 0) { | |
983 | d->number = i; | |
984 | d->raid_disk = i; | |
985 | d->state = (1<<MD_DISK_REMOVED); | |
986 | d->state |= (1<<MD_DISK_FAULTY); | |
987 | failed++; | |
988 | } | |
989 | } | |
990 | sb->nr_disks = nr_disks; | |
991 | sb->active_disks = active; | |
992 | sb->working_disks = working; | |
993 | sb->failed_disks = failed; | |
994 | sb->spare_disks = spare; | |
995 | ||
996 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
997 | sb->sb_csum = calc_sb_csum(sb); | |
998 | } | |
999 | ||
0cd17fec CW |
1000 | /* |
1001 | * rdev_size_change for 0.90.0 | |
1002 | */ | |
1003 | static unsigned long long | |
1004 | super_90_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size) | |
1005 | { | |
1006 | if (size && size < rdev->mddev->size) | |
1007 | return 0; /* component must fit device */ | |
1008 | size *= 2; /* convert to sectors */ | |
1009 | if (rdev->mddev->bitmap_offset) | |
1010 | return 0; /* can't move bitmap */ | |
1011 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
1012 | if (!size || size > rdev->sb_offset*2) | |
1013 | size = rdev->sb_offset*2; | |
1014 | md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size, | |
1015 | rdev->sb_page); | |
1016 | md_super_wait(rdev->mddev); | |
1017 | return size/2; /* kB for sysfs */ | |
1018 | } | |
1019 | ||
1020 | ||
1da177e4 LT |
1021 | /* |
1022 | * version 1 superblock | |
1023 | */ | |
1024 | ||
1c05b4bc | 1025 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1026 | { |
1c05b4bc N |
1027 | __le32 disk_csum; |
1028 | u32 csum; | |
1da177e4 LT |
1029 | unsigned long long newcsum; |
1030 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1031 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1032 | int i; |
1033 | ||
1034 | disk_csum = sb->sb_csum; | |
1035 | sb->sb_csum = 0; | |
1036 | newcsum = 0; | |
1037 | for (i=0; size>=4; size -= 4 ) | |
1038 | newcsum += le32_to_cpu(*isuper++); | |
1039 | ||
1040 | if (size == 2) | |
1c05b4bc | 1041 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1042 | |
1043 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1044 | sb->sb_csum = disk_csum; | |
1045 | return cpu_to_le32(csum); | |
1046 | } | |
1047 | ||
1048 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1049 | { | |
1050 | struct mdp_superblock_1 *sb; | |
1051 | int ret; | |
1052 | sector_t sb_offset; | |
1053 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
0002b271 | 1054 | int bmask; |
1da177e4 LT |
1055 | |
1056 | /* | |
1057 | * Calculate the position of the superblock. | |
1058 | * It is always aligned to a 4K boundary and | |
1059 | * depeding on minor_version, it can be: | |
1060 | * 0: At least 8K, but less than 12K, from end of device | |
1061 | * 1: At start of device | |
1062 | * 2: 4K from start of device. | |
1063 | */ | |
1064 | switch(minor_version) { | |
1065 | case 0: | |
1066 | sb_offset = rdev->bdev->bd_inode->i_size >> 9; | |
1067 | sb_offset -= 8*2; | |
39730960 | 1068 | sb_offset &= ~(sector_t)(4*2-1); |
1da177e4 LT |
1069 | /* convert from sectors to K */ |
1070 | sb_offset /= 2; | |
1071 | break; | |
1072 | case 1: | |
1073 | sb_offset = 0; | |
1074 | break; | |
1075 | case 2: | |
1076 | sb_offset = 4; | |
1077 | break; | |
1078 | default: | |
1079 | return -EINVAL; | |
1080 | } | |
1081 | rdev->sb_offset = sb_offset; | |
1082 | ||
0002b271 N |
1083 | /* superblock is rarely larger than 1K, but it can be larger, |
1084 | * and it is safe to read 4k, so we do that | |
1085 | */ | |
1086 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1087 | if (ret) return ret; |
1088 | ||
1089 | ||
1090 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1091 | ||
1092 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1093 | sb->major_version != cpu_to_le32(1) || | |
1094 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
1095 | le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) || | |
71c0805c | 1096 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1097 | return -EINVAL; |
1098 | ||
1099 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1100 | printk("md: invalid superblock checksum on %s\n", | |
1101 | bdevname(rdev->bdev,b)); | |
1102 | return -EINVAL; | |
1103 | } | |
1104 | if (le64_to_cpu(sb->data_size) < 10) { | |
1105 | printk("md: data_size too small on %s\n", | |
1106 | bdevname(rdev->bdev,b)); | |
1107 | return -EINVAL; | |
1108 | } | |
e11e93fa N |
1109 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) { |
1110 | if (sb->level != cpu_to_le32(1) && | |
1111 | sb->level != cpu_to_le32(4) && | |
1112 | sb->level != cpu_to_le32(5) && | |
1113 | sb->level != cpu_to_le32(6) && | |
1114 | sb->level != cpu_to_le32(10)) { | |
1115 | printk(KERN_WARNING | |
1116 | "md: bitmaps not supported for this level.\n"); | |
1117 | return -EINVAL; | |
1118 | } | |
1119 | } | |
1120 | ||
1da177e4 LT |
1121 | rdev->preferred_minor = 0xffff; |
1122 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1123 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1124 | |
0002b271 | 1125 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
720a3dc3 | 1126 | bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1127 | if (rdev->sb_size & bmask) |
a1801f85 N |
1128 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1129 | ||
1130 | if (minor_version | |
1131 | && rdev->data_offset < sb_offset + (rdev->sb_size/512)) | |
1132 | return -EINVAL; | |
0002b271 | 1133 | |
31b65a0d N |
1134 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1135 | rdev->desc_nr = -1; | |
1136 | else | |
1137 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1138 | ||
9a7b2b0f | 1139 | if (!refdev) { |
8ed75463 | 1140 | ret = 1; |
9a7b2b0f | 1141 | } else { |
1da177e4 LT |
1142 | __u64 ev1, ev2; |
1143 | struct mdp_superblock_1 *refsb = | |
1144 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1145 | ||
1146 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1147 | sb->level != refsb->level || | |
1148 | sb->layout != refsb->layout || | |
1149 | sb->chunksize != refsb->chunksize) { | |
1150 | printk(KERN_WARNING "md: %s has strangely different" | |
1151 | " superblock to %s\n", | |
1152 | bdevname(rdev->bdev,b), | |
1153 | bdevname(refdev->bdev,b2)); | |
1154 | return -EINVAL; | |
1155 | } | |
1156 | ev1 = le64_to_cpu(sb->events); | |
1157 | ev2 = le64_to_cpu(refsb->events); | |
1158 | ||
1159 | if (ev1 > ev2) | |
8ed75463 N |
1160 | ret = 1; |
1161 | else | |
1162 | ret = 0; | |
1da177e4 | 1163 | } |
a1801f85 | 1164 | if (minor_version) |
1da177e4 LT |
1165 | rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2; |
1166 | else | |
1167 | rdev->size = rdev->sb_offset; | |
1168 | if (rdev->size < le64_to_cpu(sb->data_size)/2) | |
1169 | return -EINVAL; | |
1170 | rdev->size = le64_to_cpu(sb->data_size)/2; | |
1171 | if (le32_to_cpu(sb->chunksize)) | |
1172 | rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1); | |
2bf071bf | 1173 | |
1c05b4bc | 1174 | if (le64_to_cpu(sb->size) > rdev->size*2) |
2bf071bf | 1175 | return -EINVAL; |
8ed75463 | 1176 | return ret; |
1da177e4 LT |
1177 | } |
1178 | ||
1179 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1180 | { | |
1181 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1182 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1183 | |
41158c7e | 1184 | rdev->raid_disk = -1; |
c5d79adb N |
1185 | clear_bit(Faulty, &rdev->flags); |
1186 | clear_bit(In_sync, &rdev->flags); | |
1187 | clear_bit(WriteMostly, &rdev->flags); | |
1188 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1189 | ||
1da177e4 LT |
1190 | if (mddev->raid_disks == 0) { |
1191 | mddev->major_version = 1; | |
1192 | mddev->patch_version = 0; | |
e691063a | 1193 | mddev->external = 0; |
1da177e4 LT |
1194 | mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9; |
1195 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); | |
1196 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1197 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1198 | mddev->clevel[0] = 0; |
1da177e4 LT |
1199 | mddev->layout = le32_to_cpu(sb->layout); |
1200 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
1201 | mddev->size = le64_to_cpu(sb->size)/2; | |
07d84d10 | 1202 | mddev->events = ev1; |
9223214e | 1203 | mddev->bitmap_offset = 0; |
29fc7e3e | 1204 | mddev->default_bitmap_offset = 1024 >> 9; |
1da177e4 LT |
1205 | |
1206 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1207 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1208 | ||
1209 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1210 | |
71c0805c | 1211 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
e11e93fa | 1212 | mddev->bitmap_file == NULL ) |
a654b9d8 | 1213 | mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); |
e11e93fa | 1214 | |
f6705578 N |
1215 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1216 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1217 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1218 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1219 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
1220 | mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9; | |
1221 | } else { | |
1222 | mddev->reshape_position = MaxSector; | |
1223 | mddev->delta_disks = 0; | |
1224 | mddev->new_level = mddev->level; | |
1225 | mddev->new_layout = mddev->layout; | |
1226 | mddev->new_chunk = mddev->chunk_size; | |
1227 | } | |
1228 | ||
41158c7e N |
1229 | } else if (mddev->pers == NULL) { |
1230 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1231 | ++ev1; |
1232 | if (ev1 < mddev->events) | |
1233 | return -EINVAL; | |
41158c7e N |
1234 | } else if (mddev->bitmap) { |
1235 | /* If adding to array with a bitmap, then we can accept an | |
1236 | * older device, but not too old. | |
1237 | */ | |
41158c7e N |
1238 | if (ev1 < mddev->bitmap->events_cleared) |
1239 | return 0; | |
07d84d10 N |
1240 | } else { |
1241 | if (ev1 < mddev->events) | |
1242 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1243 | return 0; | |
1244 | } | |
1da177e4 LT |
1245 | if (mddev->level != LEVEL_MULTIPATH) { |
1246 | int role; | |
1da177e4 LT |
1247 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); |
1248 | switch(role) { | |
1249 | case 0xffff: /* spare */ | |
1da177e4 LT |
1250 | break; |
1251 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1252 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1253 | break; |
1254 | default: | |
5fd6c1dc N |
1255 | if ((le32_to_cpu(sb->feature_map) & |
1256 | MD_FEATURE_RECOVERY_OFFSET)) | |
1257 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1258 | else | |
1259 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1260 | rdev->raid_disk = role; |
1261 | break; | |
1262 | } | |
8ddf9efe N |
1263 | if (sb->devflags & WriteMostly1) |
1264 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1265 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1266 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1267 | |
1da177e4 LT |
1268 | return 0; |
1269 | } | |
1270 | ||
1271 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1272 | { | |
1273 | struct mdp_superblock_1 *sb; | |
1274 | struct list_head *tmp; | |
1275 | mdk_rdev_t *rdev2; | |
1276 | int max_dev, i; | |
1277 | /* make rdev->sb match mddev and rdev data. */ | |
1278 | ||
1279 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1280 | ||
1281 | sb->feature_map = 0; | |
1282 | sb->pad0 = 0; | |
5fd6c1dc | 1283 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1284 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1285 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1286 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1287 | ||
1288 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1289 | sb->events = cpu_to_le64(mddev->events); | |
1290 | if (mddev->in_sync) | |
1291 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1292 | else | |
1293 | sb->resync_offset = cpu_to_le64(0); | |
1294 | ||
1c05b4bc | 1295 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1296 | |
f0ca340c | 1297 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
29fc7e3e | 1298 | sb->size = cpu_to_le64(mddev->size<<1); |
f0ca340c | 1299 | |
a654b9d8 N |
1300 | if (mddev->bitmap && mddev->bitmap_file == NULL) { |
1301 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); | |
71c0805c | 1302 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1303 | } |
5fd6c1dc N |
1304 | |
1305 | if (rdev->raid_disk >= 0 && | |
1306 | !test_bit(In_sync, &rdev->flags) && | |
1307 | rdev->recovery_offset > 0) { | |
1308 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1309 | sb->recovery_offset = cpu_to_le64(rdev->recovery_offset); | |
1310 | } | |
1311 | ||
f6705578 N |
1312 | if (mddev->reshape_position != MaxSector) { |
1313 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1314 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1315 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1316 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1317 | sb->new_level = cpu_to_le32(mddev->new_level); | |
1318 | sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9); | |
1319 | } | |
a654b9d8 | 1320 | |
1da177e4 | 1321 | max_dev = 0; |
d089c6af | 1322 | rdev_for_each(rdev2, tmp, mddev) |
1da177e4 LT |
1323 | if (rdev2->desc_nr+1 > max_dev) |
1324 | max_dev = rdev2->desc_nr+1; | |
a778b73f N |
1325 | |
1326 | if (max_dev > le32_to_cpu(sb->max_dev)) | |
1327 | sb->max_dev = cpu_to_le32(max_dev); | |
1da177e4 LT |
1328 | for (i=0; i<max_dev;i++) |
1329 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1330 | ||
d089c6af | 1331 | rdev_for_each(rdev2, tmp, mddev) { |
1da177e4 | 1332 | i = rdev2->desc_nr; |
b2d444d7 | 1333 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1334 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1335 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1336 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
5fd6c1dc N |
1337 | else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0) |
1338 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); | |
1da177e4 LT |
1339 | else |
1340 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1341 | } | |
1342 | ||
1da177e4 LT |
1343 | sb->sb_csum = calc_sb_1_csum(sb); |
1344 | } | |
1345 | ||
0cd17fec CW |
1346 | static unsigned long long |
1347 | super_1_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size) | |
1348 | { | |
1349 | struct mdp_superblock_1 *sb; | |
1350 | unsigned long long max_size; | |
1351 | if (size && size < rdev->mddev->size) | |
1352 | return 0; /* component must fit device */ | |
1353 | size *= 2; /* convert to sectors */ | |
1354 | if (rdev->sb_offset < rdev->data_offset/2) { | |
1355 | /* minor versions 1 and 2; superblock before data */ | |
1356 | max_size = (rdev->bdev->bd_inode->i_size >> 9); | |
1357 | max_size -= rdev->data_offset; | |
1358 | if (!size || size > max_size) | |
1359 | size = max_size; | |
1360 | } else if (rdev->mddev->bitmap_offset) { | |
1361 | /* minor version 0 with bitmap we can't move */ | |
1362 | return 0; | |
1363 | } else { | |
1364 | /* minor version 0; superblock after data */ | |
1365 | sector_t sb_offset; | |
1366 | sb_offset = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; | |
1367 | sb_offset &= ~(sector_t)(4*2 - 1); | |
1368 | max_size = rdev->size*2 + sb_offset - rdev->sb_offset*2; | |
1369 | if (!size || size > max_size) | |
1370 | size = max_size; | |
1371 | rdev->sb_offset = sb_offset/2; | |
1372 | } | |
1373 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
1374 | sb->data_size = cpu_to_le64(size); | |
1375 | sb->super_offset = rdev->sb_offset*2; | |
1376 | sb->sb_csum = calc_sb_1_csum(sb); | |
1377 | md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size, | |
1378 | rdev->sb_page); | |
1379 | md_super_wait(rdev->mddev); | |
1380 | return size/2; /* kB for sysfs */ | |
1381 | } | |
1da177e4 | 1382 | |
75c96f85 | 1383 | static struct super_type super_types[] = { |
1da177e4 LT |
1384 | [0] = { |
1385 | .name = "0.90.0", | |
1386 | .owner = THIS_MODULE, | |
0cd17fec CW |
1387 | .load_super = super_90_load, |
1388 | .validate_super = super_90_validate, | |
1389 | .sync_super = super_90_sync, | |
1390 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1391 | }, |
1392 | [1] = { | |
1393 | .name = "md-1", | |
1394 | .owner = THIS_MODULE, | |
0cd17fec CW |
1395 | .load_super = super_1_load, |
1396 | .validate_super = super_1_validate, | |
1397 | .sync_super = super_1_sync, | |
1398 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1399 | }, |
1400 | }; | |
1da177e4 LT |
1401 | |
1402 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1403 | { | |
7dd5e7c3 N |
1404 | struct list_head *tmp, *tmp2; |
1405 | mdk_rdev_t *rdev, *rdev2; | |
1da177e4 | 1406 | |
d089c6af N |
1407 | rdev_for_each(rdev, tmp, mddev1) |
1408 | rdev_for_each(rdev2, tmp2, mddev2) | |
7dd5e7c3 N |
1409 | if (rdev->bdev->bd_contains == |
1410 | rdev2->bdev->bd_contains) | |
1411 | return 1; | |
1da177e4 LT |
1412 | |
1413 | return 0; | |
1414 | } | |
1415 | ||
1416 | static LIST_HEAD(pending_raid_disks); | |
1417 | ||
1418 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) | |
1419 | { | |
7dd5e7c3 | 1420 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1421 | struct kobject *ko; |
1edf80d3 | 1422 | char *s; |
5e55e2f5 | 1423 | int err; |
1da177e4 LT |
1424 | |
1425 | if (rdev->mddev) { | |
1426 | MD_BUG(); | |
1427 | return -EINVAL; | |
1428 | } | |
11e2ede0 DW |
1429 | |
1430 | /* prevent duplicates */ | |
1431 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1432 | return -EEXIST; | |
1433 | ||
2bf071bf N |
1434 | /* make sure rdev->size exceeds mddev->size */ |
1435 | if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) { | |
a778b73f N |
1436 | if (mddev->pers) { |
1437 | /* Cannot change size, so fail | |
1438 | * If mddev->level <= 0, then we don't care | |
1439 | * about aligning sizes (e.g. linear) | |
1440 | */ | |
1441 | if (mddev->level > 0) | |
1442 | return -ENOSPC; | |
1443 | } else | |
2bf071bf N |
1444 | mddev->size = rdev->size; |
1445 | } | |
1da177e4 LT |
1446 | |
1447 | /* Verify rdev->desc_nr is unique. | |
1448 | * If it is -1, assign a free number, else | |
1449 | * check number is not in use | |
1450 | */ | |
1451 | if (rdev->desc_nr < 0) { | |
1452 | int choice = 0; | |
1453 | if (mddev->pers) choice = mddev->raid_disks; | |
1454 | while (find_rdev_nr(mddev, choice)) | |
1455 | choice++; | |
1456 | rdev->desc_nr = choice; | |
1457 | } else { | |
1458 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1459 | return -EBUSY; | |
1460 | } | |
19133a42 | 1461 | bdevname(rdev->bdev,b); |
649316b2 | 1462 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1463 | *s = '!'; |
649316b2 | 1464 | |
1da177e4 | 1465 | rdev->mddev = mddev; |
19133a42 | 1466 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1467 | |
b2d6db58 | 1468 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1469 | goto fail; |
86e6ffdd | 1470 | |
f637b9f9 | 1471 | if (rdev->bdev->bd_part) |
edfaa7c3 | 1472 | ko = &rdev->bdev->bd_part->dev.kobj; |
f637b9f9 | 1473 | else |
edfaa7c3 | 1474 | ko = &rdev->bdev->bd_disk->dev.kobj; |
5e55e2f5 N |
1475 | if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) { |
1476 | kobject_del(&rdev->kobj); | |
1477 | goto fail; | |
1478 | } | |
1479 | list_add(&rdev->same_set, &mddev->disks); | |
c5d79adb | 1480 | bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); |
1da177e4 | 1481 | return 0; |
5e55e2f5 N |
1482 | |
1483 | fail: | |
1484 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1485 | b, mdname(mddev)); | |
1486 | return err; | |
1da177e4 LT |
1487 | } |
1488 | ||
177a99b2 | 1489 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1490 | { |
1491 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1492 | kobject_del(&rdev->kobj); | |
177a99b2 | 1493 | kobject_put(&rdev->kobj); |
5792a285 N |
1494 | } |
1495 | ||
1da177e4 LT |
1496 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1497 | { | |
1498 | char b[BDEVNAME_SIZE]; | |
1499 | if (!rdev->mddev) { | |
1500 | MD_BUG(); | |
1501 | return; | |
1502 | } | |
5463c790 | 1503 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
1da177e4 LT |
1504 | list_del_init(&rdev->same_set); |
1505 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); | |
1506 | rdev->mddev = NULL; | |
86e6ffdd | 1507 | sysfs_remove_link(&rdev->kobj, "block"); |
5792a285 N |
1508 | |
1509 | /* We need to delay this, otherwise we can deadlock when | |
1510 | * writing to 'remove' to "dev/state" | |
1511 | */ | |
177a99b2 N |
1512 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1513 | kobject_get(&rdev->kobj); | |
5792a285 | 1514 | schedule_work(&rdev->del_work); |
1da177e4 LT |
1515 | } |
1516 | ||
1517 | /* | |
1518 | * prevent the device from being mounted, repartitioned or | |
1519 | * otherwise reused by a RAID array (or any other kernel | |
1520 | * subsystem), by bd_claiming the device. | |
1521 | */ | |
c5d79adb | 1522 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1523 | { |
1524 | int err = 0; | |
1525 | struct block_device *bdev; | |
1526 | char b[BDEVNAME_SIZE]; | |
1527 | ||
2e7b651d | 1528 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1529 | if (IS_ERR(bdev)) { |
1530 | printk(KERN_ERR "md: could not open %s.\n", | |
1531 | __bdevname(dev, b)); | |
1532 | return PTR_ERR(bdev); | |
1533 | } | |
c5d79adb | 1534 | err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); |
1da177e4 LT |
1535 | if (err) { |
1536 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1537 | bdevname(bdev, b)); | |
2e7b651d | 1538 | blkdev_put(bdev); |
1da177e4 LT |
1539 | return err; |
1540 | } | |
c5d79adb N |
1541 | if (!shared) |
1542 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1543 | rdev->bdev = bdev; |
1544 | return err; | |
1545 | } | |
1546 | ||
1547 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1548 | { | |
1549 | struct block_device *bdev = rdev->bdev; | |
1550 | rdev->bdev = NULL; | |
1551 | if (!bdev) | |
1552 | MD_BUG(); | |
1553 | bd_release(bdev); | |
2e7b651d | 1554 | blkdev_put(bdev); |
1da177e4 LT |
1555 | } |
1556 | ||
1557 | void md_autodetect_dev(dev_t dev); | |
1558 | ||
1559 | static void export_rdev(mdk_rdev_t * rdev) | |
1560 | { | |
1561 | char b[BDEVNAME_SIZE]; | |
1562 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1563 | bdevname(rdev->bdev,b)); | |
1564 | if (rdev->mddev) | |
1565 | MD_BUG(); | |
1566 | free_disk_sb(rdev); | |
1567 | list_del_init(&rdev->same_set); | |
1568 | #ifndef MODULE | |
d0fae18f N |
1569 | if (test_bit(AutoDetected, &rdev->flags)) |
1570 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1571 | #endif |
1572 | unlock_rdev(rdev); | |
86e6ffdd | 1573 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1574 | } |
1575 | ||
1576 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1577 | { | |
1578 | unbind_rdev_from_array(rdev); | |
1579 | export_rdev(rdev); | |
1580 | } | |
1581 | ||
1582 | static void export_array(mddev_t *mddev) | |
1583 | { | |
1584 | struct list_head *tmp; | |
1585 | mdk_rdev_t *rdev; | |
1586 | ||
d089c6af | 1587 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1588 | if (!rdev->mddev) { |
1589 | MD_BUG(); | |
1590 | continue; | |
1591 | } | |
1592 | kick_rdev_from_array(rdev); | |
1593 | } | |
1594 | if (!list_empty(&mddev->disks)) | |
1595 | MD_BUG(); | |
1596 | mddev->raid_disks = 0; | |
1597 | mddev->major_version = 0; | |
1598 | } | |
1599 | ||
1600 | static void print_desc(mdp_disk_t *desc) | |
1601 | { | |
1602 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1603 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1604 | } | |
1605 | ||
1606 | static void print_sb(mdp_super_t *sb) | |
1607 | { | |
1608 | int i; | |
1609 | ||
1610 | printk(KERN_INFO | |
1611 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1612 | sb->major_version, sb->minor_version, sb->patch_version, | |
1613 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1614 | sb->ctime); | |
1615 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1616 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1617 | sb->md_minor, sb->layout, sb->chunk_size); | |
1618 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1619 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1620 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1621 | sb->failed_disks, sb->spare_disks, | |
1622 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1623 | ||
1624 | printk(KERN_INFO); | |
1625 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1626 | mdp_disk_t *desc; | |
1627 | ||
1628 | desc = sb->disks + i; | |
1629 | if (desc->number || desc->major || desc->minor || | |
1630 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1631 | printk(" D %2d: ", i); | |
1632 | print_desc(desc); | |
1633 | } | |
1634 | } | |
1635 | printk(KERN_INFO "md: THIS: "); | |
1636 | print_desc(&sb->this_disk); | |
1637 | ||
1638 | } | |
1639 | ||
1640 | static void print_rdev(mdk_rdev_t *rdev) | |
1641 | { | |
1642 | char b[BDEVNAME_SIZE]; | |
1643 | printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n", | |
1644 | bdevname(rdev->bdev,b), (unsigned long long)rdev->size, | |
b2d444d7 N |
1645 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
1646 | rdev->desc_nr); | |
1da177e4 LT |
1647 | if (rdev->sb_loaded) { |
1648 | printk(KERN_INFO "md: rdev superblock:\n"); | |
1649 | print_sb((mdp_super_t*)page_address(rdev->sb_page)); | |
1650 | } else | |
1651 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
1652 | } | |
1653 | ||
5e56341d | 1654 | static void md_print_devices(void) |
1da177e4 LT |
1655 | { |
1656 | struct list_head *tmp, *tmp2; | |
1657 | mdk_rdev_t *rdev; | |
1658 | mddev_t *mddev; | |
1659 | char b[BDEVNAME_SIZE]; | |
1660 | ||
1661 | printk("\n"); | |
1662 | printk("md: **********************************\n"); | |
1663 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
1664 | printk("md: **********************************\n"); | |
29ac4aa3 | 1665 | for_each_mddev(mddev, tmp) { |
1da177e4 | 1666 | |
32a7627c N |
1667 | if (mddev->bitmap) |
1668 | bitmap_print_sb(mddev->bitmap); | |
1669 | else | |
1670 | printk("%s: ", mdname(mddev)); | |
d089c6af | 1671 | rdev_for_each(rdev, tmp2, mddev) |
1da177e4 LT |
1672 | printk("<%s>", bdevname(rdev->bdev,b)); |
1673 | printk("\n"); | |
1674 | ||
d089c6af | 1675 | rdev_for_each(rdev, tmp2, mddev) |
1da177e4 LT |
1676 | print_rdev(rdev); |
1677 | } | |
1678 | printk("md: **********************************\n"); | |
1679 | printk("\n"); | |
1680 | } | |
1681 | ||
1682 | ||
42543769 | 1683 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 1684 | { |
42543769 N |
1685 | /* Update each superblock (in-memory image), but |
1686 | * if we are allowed to, skip spares which already | |
1687 | * have the right event counter, or have one earlier | |
1688 | * (which would mean they aren't being marked as dirty | |
1689 | * with the rest of the array) | |
1690 | */ | |
1da177e4 LT |
1691 | mdk_rdev_t *rdev; |
1692 | struct list_head *tmp; | |
1693 | ||
d089c6af | 1694 | rdev_for_each(rdev, tmp, mddev) { |
42543769 N |
1695 | if (rdev->sb_events == mddev->events || |
1696 | (nospares && | |
1697 | rdev->raid_disk < 0 && | |
1698 | (rdev->sb_events&1)==0 && | |
1699 | rdev->sb_events+1 == mddev->events)) { | |
1700 | /* Don't update this superblock */ | |
1701 | rdev->sb_loaded = 2; | |
1702 | } else { | |
1703 | super_types[mddev->major_version]. | |
1704 | sync_super(mddev, rdev); | |
1705 | rdev->sb_loaded = 1; | |
1706 | } | |
1da177e4 LT |
1707 | } |
1708 | } | |
1709 | ||
850b2b42 | 1710 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 1711 | { |
1da177e4 LT |
1712 | struct list_head *tmp; |
1713 | mdk_rdev_t *rdev; | |
06d91a5f | 1714 | int sync_req; |
42543769 | 1715 | int nospares = 0; |
1da177e4 | 1716 | |
8377bc80 N |
1717 | if (mddev->external) |
1718 | return; | |
1da177e4 | 1719 | repeat: |
a9701a30 | 1720 | spin_lock_irq(&mddev->write_lock); |
84692195 | 1721 | |
850b2b42 N |
1722 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
1723 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) | |
1724 | force_change = 1; | |
1725 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
1726 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
1727 | * though if events isn't the right even/odd, we will have to do | |
1728 | * spares after all | |
1729 | */ | |
1730 | nospares = 1; | |
1731 | if (force_change) | |
1732 | nospares = 0; | |
1733 | if (mddev->degraded) | |
84692195 N |
1734 | /* If the array is degraded, then skipping spares is both |
1735 | * dangerous and fairly pointless. | |
1736 | * Dangerous because a device that was removed from the array | |
1737 | * might have a event_count that still looks up-to-date, | |
1738 | * so it can be re-added without a resync. | |
1739 | * Pointless because if there are any spares to skip, | |
1740 | * then a recovery will happen and soon that array won't | |
1741 | * be degraded any more and the spare can go back to sleep then. | |
1742 | */ | |
850b2b42 | 1743 | nospares = 0; |
84692195 | 1744 | |
06d91a5f | 1745 | sync_req = mddev->in_sync; |
1da177e4 | 1746 | mddev->utime = get_seconds(); |
42543769 N |
1747 | |
1748 | /* If this is just a dirty<->clean transition, and the array is clean | |
1749 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 1750 | if (nospares |
42543769 | 1751 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
1031be7a N |
1752 | && (mddev->events & 1) |
1753 | && mddev->events != 1) | |
42543769 N |
1754 | mddev->events--; |
1755 | else { | |
1756 | /* otherwise we have to go forward and ... */ | |
1757 | mddev->events ++; | |
1758 | if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ | |
1759 | /* .. if the array isn't clean, insist on an odd 'events' */ | |
1760 | if ((mddev->events&1)==0) { | |
1761 | mddev->events++; | |
1762 | nospares = 0; | |
1763 | } | |
1764 | } else { | |
1765 | /* otherwise insist on an even 'events' (for clean states) */ | |
1766 | if ((mddev->events&1)) { | |
1767 | mddev->events++; | |
1768 | nospares = 0; | |
1769 | } | |
1770 | } | |
1771 | } | |
1da177e4 LT |
1772 | |
1773 | if (!mddev->events) { | |
1774 | /* | |
1775 | * oops, this 64-bit counter should never wrap. | |
1776 | * Either we are in around ~1 trillion A.C., assuming | |
1777 | * 1 reboot per second, or we have a bug: | |
1778 | */ | |
1779 | MD_BUG(); | |
1780 | mddev->events --; | |
1781 | } | |
1da177e4 LT |
1782 | |
1783 | /* | |
1784 | * do not write anything to disk if using | |
1785 | * nonpersistent superblocks | |
1786 | */ | |
06d91a5f | 1787 | if (!mddev->persistent) { |
e691063a N |
1788 | if (!mddev->external) |
1789 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
1790 | ||
a9701a30 | 1791 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1792 | wake_up(&mddev->sb_wait); |
1da177e4 | 1793 | return; |
06d91a5f | 1794 | } |
e691063a | 1795 | sync_sbs(mddev, nospares); |
a9701a30 | 1796 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
1797 | |
1798 | dprintk(KERN_INFO | |
1799 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
1800 | mdname(mddev),mddev->in_sync); | |
1801 | ||
4ad13663 | 1802 | bitmap_update_sb(mddev->bitmap); |
d089c6af | 1803 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1804 | char b[BDEVNAME_SIZE]; |
1805 | dprintk(KERN_INFO "md: "); | |
42543769 N |
1806 | if (rdev->sb_loaded != 1) |
1807 | continue; /* no noise on spare devices */ | |
b2d444d7 | 1808 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
1809 | dprintk("(skipping faulty "); |
1810 | ||
1811 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 1812 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 1813 | md_super_write(mddev,rdev, |
0002b271 | 1814 | rdev->sb_offset<<1, rdev->sb_size, |
7bfa19f2 N |
1815 | rdev->sb_page); |
1816 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
1817 | bdevname(rdev->bdev,b), | |
1818 | (unsigned long long)rdev->sb_offset); | |
42543769 | 1819 | rdev->sb_events = mddev->events; |
7bfa19f2 | 1820 | |
1da177e4 LT |
1821 | } else |
1822 | dprintk(")\n"); | |
7bfa19f2 | 1823 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
1824 | /* only need to write one superblock... */ |
1825 | break; | |
1826 | } | |
a9701a30 | 1827 | md_super_wait(mddev); |
850b2b42 | 1828 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 1829 | |
a9701a30 | 1830 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
1831 | if (mddev->in_sync != sync_req || |
1832 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 1833 | /* have to write it out again */ |
a9701a30 | 1834 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
1835 | goto repeat; |
1836 | } | |
850b2b42 | 1837 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 1838 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1839 | wake_up(&mddev->sb_wait); |
06d91a5f | 1840 | |
1da177e4 LT |
1841 | } |
1842 | ||
7f6ce769 | 1843 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
1844 | * We want to accept with case. For this we use cmd_match. |
1845 | */ | |
1846 | static int cmd_match(const char *cmd, const char *str) | |
1847 | { | |
1848 | /* See if cmd, written into a sysfs file, matches | |
1849 | * str. They must either be the same, or cmd can | |
1850 | * have a trailing newline | |
1851 | */ | |
1852 | while (*cmd && *str && *cmd == *str) { | |
1853 | cmd++; | |
1854 | str++; | |
1855 | } | |
1856 | if (*cmd == '\n') | |
1857 | cmd++; | |
1858 | if (*str || *cmd) | |
1859 | return 0; | |
1860 | return 1; | |
1861 | } | |
1862 | ||
86e6ffdd N |
1863 | struct rdev_sysfs_entry { |
1864 | struct attribute attr; | |
1865 | ssize_t (*show)(mdk_rdev_t *, char *); | |
1866 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
1867 | }; | |
1868 | ||
1869 | static ssize_t | |
96de1e66 | 1870 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
1871 | { |
1872 | char *sep = ""; | |
20a49ff6 | 1873 | size_t len = 0; |
86e6ffdd | 1874 | |
b2d444d7 | 1875 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
1876 | len+= sprintf(page+len, "%sfaulty",sep); |
1877 | sep = ","; | |
1878 | } | |
b2d444d7 | 1879 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
1880 | len += sprintf(page+len, "%sin_sync",sep); |
1881 | sep = ","; | |
1882 | } | |
f655675b N |
1883 | if (test_bit(WriteMostly, &rdev->flags)) { |
1884 | len += sprintf(page+len, "%swrite_mostly",sep); | |
1885 | sep = ","; | |
1886 | } | |
6bfe0b49 DW |
1887 | if (test_bit(Blocked, &rdev->flags)) { |
1888 | len += sprintf(page+len, "%sblocked", sep); | |
1889 | sep = ","; | |
1890 | } | |
b2d444d7 N |
1891 | if (!test_bit(Faulty, &rdev->flags) && |
1892 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
1893 | len += sprintf(page+len, "%sspare", sep); |
1894 | sep = ","; | |
1895 | } | |
1896 | return len+sprintf(page+len, "\n"); | |
1897 | } | |
1898 | ||
45dc2de1 N |
1899 | static ssize_t |
1900 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1901 | { | |
1902 | /* can write | |
1903 | * faulty - simulates and error | |
1904 | * remove - disconnects the device | |
f655675b N |
1905 | * writemostly - sets write_mostly |
1906 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
1907 | * blocked - sets the Blocked flag |
1908 | * -blocked - clears the Blocked flag | |
45dc2de1 N |
1909 | */ |
1910 | int err = -EINVAL; | |
1911 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
1912 | md_error(rdev->mddev, rdev); | |
1913 | err = 0; | |
1914 | } else if (cmd_match(buf, "remove")) { | |
1915 | if (rdev->raid_disk >= 0) | |
1916 | err = -EBUSY; | |
1917 | else { | |
1918 | mddev_t *mddev = rdev->mddev; | |
1919 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
1920 | if (mddev->pers) |
1921 | md_update_sb(mddev, 1); | |
45dc2de1 N |
1922 | md_new_event(mddev); |
1923 | err = 0; | |
1924 | } | |
f655675b N |
1925 | } else if (cmd_match(buf, "writemostly")) { |
1926 | set_bit(WriteMostly, &rdev->flags); | |
1927 | err = 0; | |
1928 | } else if (cmd_match(buf, "-writemostly")) { | |
1929 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
1930 | err = 0; |
1931 | } else if (cmd_match(buf, "blocked")) { | |
1932 | set_bit(Blocked, &rdev->flags); | |
1933 | err = 0; | |
1934 | } else if (cmd_match(buf, "-blocked")) { | |
1935 | clear_bit(Blocked, &rdev->flags); | |
1936 | wake_up(&rdev->blocked_wait); | |
1937 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
1938 | md_wakeup_thread(rdev->mddev->thread); | |
1939 | ||
f655675b | 1940 | err = 0; |
45dc2de1 | 1941 | } |
52664732 NB |
1942 | if (!err) |
1943 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
45dc2de1 N |
1944 | return err ? err : len; |
1945 | } | |
80ca3a44 N |
1946 | static struct rdev_sysfs_entry rdev_state = |
1947 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 1948 | |
4dbcdc75 N |
1949 | static ssize_t |
1950 | errors_show(mdk_rdev_t *rdev, char *page) | |
1951 | { | |
1952 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
1953 | } | |
1954 | ||
1955 | static ssize_t | |
1956 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1957 | { | |
1958 | char *e; | |
1959 | unsigned long n = simple_strtoul(buf, &e, 10); | |
1960 | if (*buf && (*e == 0 || *e == '\n')) { | |
1961 | atomic_set(&rdev->corrected_errors, n); | |
1962 | return len; | |
1963 | } | |
1964 | return -EINVAL; | |
1965 | } | |
1966 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 1967 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 1968 | |
014236d2 N |
1969 | static ssize_t |
1970 | slot_show(mdk_rdev_t *rdev, char *page) | |
1971 | { | |
1972 | if (rdev->raid_disk < 0) | |
1973 | return sprintf(page, "none\n"); | |
1974 | else | |
1975 | return sprintf(page, "%d\n", rdev->raid_disk); | |
1976 | } | |
1977 | ||
1978 | static ssize_t | |
1979 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1980 | { | |
1981 | char *e; | |
c303da6d N |
1982 | int err; |
1983 | char nm[20]; | |
014236d2 N |
1984 | int slot = simple_strtoul(buf, &e, 10); |
1985 | if (strncmp(buf, "none", 4)==0) | |
1986 | slot = -1; | |
1987 | else if (e==buf || (*e && *e!= '\n')) | |
1988 | return -EINVAL; | |
6c2fce2e | 1989 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
1990 | /* Setting 'slot' on an active array requires also |
1991 | * updating the 'rd%d' link, and communicating | |
1992 | * with the personality with ->hot_*_disk. | |
1993 | * For now we only support removing | |
1994 | * failed/spare devices. This normally happens automatically, | |
1995 | * but not when the metadata is externally managed. | |
1996 | */ | |
c303da6d N |
1997 | if (rdev->raid_disk == -1) |
1998 | return -EEXIST; | |
1999 | /* personality does all needed checks */ | |
2000 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2001 | return -EINVAL; | |
2002 | err = rdev->mddev->pers-> | |
2003 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2004 | if (err) | |
2005 | return err; | |
2006 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2007 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
2008 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2009 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2010 | } else if (rdev->mddev->pers) { |
2011 | mdk_rdev_t *rdev2; | |
2012 | struct list_head *tmp; | |
2013 | /* Activating a spare .. or possibly reactivating | |
2014 | * if we every get bitmaps working here. | |
2015 | */ | |
2016 | ||
2017 | if (rdev->raid_disk != -1) | |
2018 | return -EBUSY; | |
2019 | ||
2020 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2021 | return -EINVAL; | |
2022 | ||
2023 | rdev_for_each(rdev2, tmp, rdev->mddev) | |
2024 | if (rdev2->raid_disk == slot) | |
2025 | return -EEXIST; | |
2026 | ||
2027 | rdev->raid_disk = slot; | |
2028 | if (test_bit(In_sync, &rdev->flags)) | |
2029 | rdev->saved_raid_disk = slot; | |
2030 | else | |
2031 | rdev->saved_raid_disk = -1; | |
2032 | err = rdev->mddev->pers-> | |
2033 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2034 | if (err) { |
6c2fce2e | 2035 | rdev->raid_disk = -1; |
6c2fce2e | 2036 | return err; |
52664732 NB |
2037 | } else |
2038 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
6c2fce2e NB |
2039 | sprintf(nm, "rd%d", rdev->raid_disk); |
2040 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
2041 | printk(KERN_WARNING | |
2042 | "md: cannot register " | |
2043 | "%s for %s\n", | |
2044 | nm, mdname(rdev->mddev)); | |
2045 | ||
2046 | /* don't wakeup anyone, leave that to userspace. */ | |
c303da6d N |
2047 | } else { |
2048 | if (slot >= rdev->mddev->raid_disks) | |
2049 | return -ENOSPC; | |
2050 | rdev->raid_disk = slot; | |
2051 | /* assume it is working */ | |
c5d79adb N |
2052 | clear_bit(Faulty, &rdev->flags); |
2053 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2054 | set_bit(In_sync, &rdev->flags); |
52664732 | 2055 | sysfs_notify(&rdev->kobj, NULL, "state"); |
c303da6d | 2056 | } |
014236d2 N |
2057 | return len; |
2058 | } | |
2059 | ||
2060 | ||
2061 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2062 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2063 | |
93c8cad0 N |
2064 | static ssize_t |
2065 | offset_show(mdk_rdev_t *rdev, char *page) | |
2066 | { | |
6961ece4 | 2067 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2068 | } |
2069 | ||
2070 | static ssize_t | |
2071 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2072 | { | |
2073 | char *e; | |
2074 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2075 | if (e==buf || (*e && *e != '\n')) | |
2076 | return -EINVAL; | |
8ed0a521 | 2077 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2078 | return -EBUSY; |
c5d79adb N |
2079 | if (rdev->size && rdev->mddev->external) |
2080 | /* Must set offset before size, so overlap checks | |
2081 | * can be sane */ | |
2082 | return -EBUSY; | |
93c8cad0 N |
2083 | rdev->data_offset = offset; |
2084 | return len; | |
2085 | } | |
2086 | ||
2087 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2088 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2089 | |
83303b61 N |
2090 | static ssize_t |
2091 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2092 | { | |
2093 | return sprintf(page, "%llu\n", (unsigned long long)rdev->size); | |
2094 | } | |
2095 | ||
c5d79adb N |
2096 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2097 | { | |
2098 | /* check if two start/length pairs overlap */ | |
2099 | if (s1+l1 <= s2) | |
2100 | return 0; | |
2101 | if (s2+l2 <= s1) | |
2102 | return 0; | |
2103 | return 1; | |
2104 | } | |
2105 | ||
83303b61 N |
2106 | static ssize_t |
2107 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2108 | { | |
2109 | char *e; | |
2110 | unsigned long long size = simple_strtoull(buf, &e, 10); | |
c5d79adb | 2111 | unsigned long long oldsize = rdev->size; |
27c529bb N |
2112 | mddev_t *my_mddev = rdev->mddev; |
2113 | ||
83303b61 N |
2114 | if (e==buf || (*e && *e != '\n')) |
2115 | return -EINVAL; | |
0cd17fec CW |
2116 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
2117 | if (rdev->mddev->persistent) { | |
2118 | size = super_types[rdev->mddev->major_version]. | |
2119 | rdev_size_change(rdev, size); | |
2120 | if (!size) | |
2121 | return -EBUSY; | |
2122 | } else if (!size) { | |
2123 | size = (rdev->bdev->bd_inode->i_size >> 10); | |
2124 | size -= rdev->data_offset/2; | |
2125 | } | |
2126 | if (size < rdev->mddev->size) | |
2127 | return -EINVAL; /* component must fit device */ | |
2128 | } | |
2129 | ||
83303b61 | 2130 | rdev->size = size; |
c5d79adb N |
2131 | if (size > oldsize && rdev->mddev->external) { |
2132 | /* need to check that all other rdevs with the same ->bdev | |
2133 | * do not overlap. We need to unlock the mddev to avoid | |
2134 | * a deadlock. We have already changed rdev->size, and if | |
2135 | * we have to change it back, we will have the lock again. | |
2136 | */ | |
2137 | mddev_t *mddev; | |
2138 | int overlap = 0; | |
2139 | struct list_head *tmp, *tmp2; | |
2140 | ||
27c529bb | 2141 | mddev_unlock(my_mddev); |
29ac4aa3 | 2142 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2143 | mdk_rdev_t *rdev2; |
2144 | ||
2145 | mddev_lock(mddev); | |
d089c6af | 2146 | rdev_for_each(rdev2, tmp2, mddev) |
c5d79adb N |
2147 | if (test_bit(AllReserved, &rdev2->flags) || |
2148 | (rdev->bdev == rdev2->bdev && | |
2149 | rdev != rdev2 && | |
2150 | overlaps(rdev->data_offset, rdev->size, | |
2151 | rdev2->data_offset, rdev2->size))) { | |
2152 | overlap = 1; | |
2153 | break; | |
2154 | } | |
2155 | mddev_unlock(mddev); | |
2156 | if (overlap) { | |
2157 | mddev_put(mddev); | |
2158 | break; | |
2159 | } | |
2160 | } | |
27c529bb | 2161 | mddev_lock(my_mddev); |
c5d79adb N |
2162 | if (overlap) { |
2163 | /* Someone else could have slipped in a size | |
2164 | * change here, but doing so is just silly. | |
2165 | * We put oldsize back because we *know* it is | |
2166 | * safe, and trust userspace not to race with | |
2167 | * itself | |
2168 | */ | |
2169 | rdev->size = oldsize; | |
2170 | return -EBUSY; | |
2171 | } | |
2172 | } | |
27c529bb N |
2173 | if (size < my_mddev->size || my_mddev->size == 0) |
2174 | my_mddev->size = size; | |
83303b61 N |
2175 | return len; |
2176 | } | |
2177 | ||
2178 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2179 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2180 | |
86e6ffdd N |
2181 | static struct attribute *rdev_default_attrs[] = { |
2182 | &rdev_state.attr, | |
4dbcdc75 | 2183 | &rdev_errors.attr, |
014236d2 | 2184 | &rdev_slot.attr, |
93c8cad0 | 2185 | &rdev_offset.attr, |
83303b61 | 2186 | &rdev_size.attr, |
86e6ffdd N |
2187 | NULL, |
2188 | }; | |
2189 | static ssize_t | |
2190 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2191 | { | |
2192 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2193 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2194 | mddev_t *mddev = rdev->mddev; |
2195 | ssize_t rv; | |
86e6ffdd N |
2196 | |
2197 | if (!entry->show) | |
2198 | return -EIO; | |
27c529bb N |
2199 | |
2200 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2201 | if (!rv) { | |
2202 | if (rdev->mddev == NULL) | |
2203 | rv = -EBUSY; | |
2204 | else | |
2205 | rv = entry->show(rdev, page); | |
2206 | mddev_unlock(mddev); | |
2207 | } | |
2208 | return rv; | |
86e6ffdd N |
2209 | } |
2210 | ||
2211 | static ssize_t | |
2212 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2213 | const char *page, size_t length) | |
2214 | { | |
2215 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2216 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2217 | ssize_t rv; |
2218 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2219 | |
2220 | if (!entry->store) | |
2221 | return -EIO; | |
67463acb N |
2222 | if (!capable(CAP_SYS_ADMIN)) |
2223 | return -EACCES; | |
27c529bb | 2224 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2225 | if (!rv) { |
27c529bb N |
2226 | if (rdev->mddev == NULL) |
2227 | rv = -EBUSY; | |
2228 | else | |
2229 | rv = entry->store(rdev, page, length); | |
6a51830e | 2230 | mddev_unlock(mddev); |
ca388059 N |
2231 | } |
2232 | return rv; | |
86e6ffdd N |
2233 | } |
2234 | ||
2235 | static void rdev_free(struct kobject *ko) | |
2236 | { | |
2237 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2238 | kfree(rdev); | |
2239 | } | |
2240 | static struct sysfs_ops rdev_sysfs_ops = { | |
2241 | .show = rdev_attr_show, | |
2242 | .store = rdev_attr_store, | |
2243 | }; | |
2244 | static struct kobj_type rdev_ktype = { | |
2245 | .release = rdev_free, | |
2246 | .sysfs_ops = &rdev_sysfs_ops, | |
2247 | .default_attrs = rdev_default_attrs, | |
2248 | }; | |
2249 | ||
1da177e4 LT |
2250 | /* |
2251 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2252 | * | |
2253 | * mark the device faulty if: | |
2254 | * | |
2255 | * - the device is nonexistent (zero size) | |
2256 | * - the device has no valid superblock | |
2257 | * | |
2258 | * a faulty rdev _never_ has rdev->sb set. | |
2259 | */ | |
2260 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2261 | { | |
2262 | char b[BDEVNAME_SIZE]; | |
2263 | int err; | |
2264 | mdk_rdev_t *rdev; | |
2265 | sector_t size; | |
2266 | ||
9ffae0cf | 2267 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2268 | if (!rdev) { |
2269 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2270 | return ERR_PTR(-ENOMEM); | |
2271 | } | |
1da177e4 LT |
2272 | |
2273 | if ((err = alloc_disk_sb(rdev))) | |
2274 | goto abort_free; | |
2275 | ||
c5d79adb | 2276 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2277 | if (err) |
2278 | goto abort_free; | |
2279 | ||
f9cb074b | 2280 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2281 | |
1da177e4 | 2282 | rdev->desc_nr = -1; |
2b6e8459 | 2283 | rdev->saved_raid_disk = -1; |
3f9d7b0d | 2284 | rdev->raid_disk = -1; |
b2d444d7 | 2285 | rdev->flags = 0; |
1da177e4 | 2286 | rdev->data_offset = 0; |
42543769 | 2287 | rdev->sb_events = 0; |
1da177e4 | 2288 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 2289 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 2290 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
2291 | |
2292 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
2293 | if (!size) { | |
2294 | printk(KERN_WARNING | |
2295 | "md: %s has zero or unknown size, marking faulty!\n", | |
2296 | bdevname(rdev->bdev,b)); | |
2297 | err = -EINVAL; | |
2298 | goto abort_free; | |
2299 | } | |
2300 | ||
2301 | if (super_format >= 0) { | |
2302 | err = super_types[super_format]. | |
2303 | load_super(rdev, NULL, super_minor); | |
2304 | if (err == -EINVAL) { | |
df968c4e N |
2305 | printk(KERN_WARNING |
2306 | "md: %s does not have a valid v%d.%d " | |
2307 | "superblock, not importing!\n", | |
2308 | bdevname(rdev->bdev,b), | |
2309 | super_format, super_minor); | |
1da177e4 LT |
2310 | goto abort_free; |
2311 | } | |
2312 | if (err < 0) { | |
2313 | printk(KERN_WARNING | |
2314 | "md: could not read %s's sb, not importing!\n", | |
2315 | bdevname(rdev->bdev,b)); | |
2316 | goto abort_free; | |
2317 | } | |
2318 | } | |
6bfe0b49 | 2319 | |
1da177e4 | 2320 | INIT_LIST_HEAD(&rdev->same_set); |
6bfe0b49 | 2321 | init_waitqueue_head(&rdev->blocked_wait); |
1da177e4 LT |
2322 | |
2323 | return rdev; | |
2324 | ||
2325 | abort_free: | |
2326 | if (rdev->sb_page) { | |
2327 | if (rdev->bdev) | |
2328 | unlock_rdev(rdev); | |
2329 | free_disk_sb(rdev); | |
2330 | } | |
2331 | kfree(rdev); | |
2332 | return ERR_PTR(err); | |
2333 | } | |
2334 | ||
2335 | /* | |
2336 | * Check a full RAID array for plausibility | |
2337 | */ | |
2338 | ||
2339 | ||
a757e64c | 2340 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2341 | { |
2342 | int i; | |
2343 | struct list_head *tmp; | |
2344 | mdk_rdev_t *rdev, *freshest; | |
2345 | char b[BDEVNAME_SIZE]; | |
2346 | ||
2347 | freshest = NULL; | |
d089c6af | 2348 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2349 | switch (super_types[mddev->major_version]. |
2350 | load_super(rdev, freshest, mddev->minor_version)) { | |
2351 | case 1: | |
2352 | freshest = rdev; | |
2353 | break; | |
2354 | case 0: | |
2355 | break; | |
2356 | default: | |
2357 | printk( KERN_ERR \ | |
2358 | "md: fatal superblock inconsistency in %s" | |
2359 | " -- removing from array\n", | |
2360 | bdevname(rdev->bdev,b)); | |
2361 | kick_rdev_from_array(rdev); | |
2362 | } | |
2363 | ||
2364 | ||
2365 | super_types[mddev->major_version]. | |
2366 | validate_super(mddev, freshest); | |
2367 | ||
2368 | i = 0; | |
d089c6af | 2369 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
2370 | if (rdev != freshest) |
2371 | if (super_types[mddev->major_version]. | |
2372 | validate_super(mddev, rdev)) { | |
2373 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2374 | " from array!\n", | |
2375 | bdevname(rdev->bdev,b)); | |
2376 | kick_rdev_from_array(rdev); | |
2377 | continue; | |
2378 | } | |
2379 | if (mddev->level == LEVEL_MULTIPATH) { | |
2380 | rdev->desc_nr = i++; | |
2381 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2382 | set_bit(In_sync, &rdev->flags); |
a778b73f N |
2383 | } else if (rdev->raid_disk >= mddev->raid_disks) { |
2384 | rdev->raid_disk = -1; | |
2385 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2386 | } |
2387 | } | |
2388 | ||
2389 | ||
2390 | ||
2391 | if (mddev->recovery_cp != MaxSector && | |
2392 | mddev->level >= 1) | |
2393 | printk(KERN_ERR "md: %s: raid array is not clean" | |
2394 | " -- starting background reconstruction\n", | |
2395 | mdname(mddev)); | |
2396 | ||
1da177e4 LT |
2397 | } |
2398 | ||
16f17b39 N |
2399 | static ssize_t |
2400 | safe_delay_show(mddev_t *mddev, char *page) | |
2401 | { | |
2402 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2403 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2404 | } | |
2405 | static ssize_t | |
2406 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2407 | { | |
2408 | int scale=1; | |
2409 | int dot=0; | |
2410 | int i; | |
2411 | unsigned long msec; | |
2412 | char buf[30]; | |
2413 | char *e; | |
2414 | /* remove a period, and count digits after it */ | |
2415 | if (len >= sizeof(buf)) | |
2416 | return -EINVAL; | |
2417 | strlcpy(buf, cbuf, len); | |
2418 | buf[len] = 0; | |
2419 | for (i=0; i<len; i++) { | |
2420 | if (dot) { | |
2421 | if (isdigit(buf[i])) { | |
2422 | buf[i-1] = buf[i]; | |
2423 | scale *= 10; | |
2424 | } | |
2425 | buf[i] = 0; | |
2426 | } else if (buf[i] == '.') { | |
2427 | dot=1; | |
2428 | buf[i] = 0; | |
2429 | } | |
2430 | } | |
2431 | msec = simple_strtoul(buf, &e, 10); | |
2432 | if (e == buf || (*e && *e != '\n')) | |
2433 | return -EINVAL; | |
2434 | msec = (msec * 1000) / scale; | |
2435 | if (msec == 0) | |
2436 | mddev->safemode_delay = 0; | |
2437 | else { | |
2438 | mddev->safemode_delay = (msec*HZ)/1000; | |
2439 | if (mddev->safemode_delay == 0) | |
2440 | mddev->safemode_delay = 1; | |
2441 | } | |
2442 | return len; | |
2443 | } | |
2444 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2445 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2446 | |
eae1701f | 2447 | static ssize_t |
96de1e66 | 2448 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2449 | { |
2604b703 | 2450 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2451 | if (p) |
eae1701f | 2452 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2453 | else if (mddev->clevel[0]) |
2454 | return sprintf(page, "%s\n", mddev->clevel); | |
2455 | else if (mddev->level != LEVEL_NONE) | |
2456 | return sprintf(page, "%d\n", mddev->level); | |
2457 | else | |
2458 | return 0; | |
eae1701f N |
2459 | } |
2460 | ||
d9d166c2 N |
2461 | static ssize_t |
2462 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2463 | { | |
20a49ff6 | 2464 | ssize_t rv = len; |
d9d166c2 N |
2465 | if (mddev->pers) |
2466 | return -EBUSY; | |
2467 | if (len == 0) | |
2468 | return 0; | |
2469 | if (len >= sizeof(mddev->clevel)) | |
2470 | return -ENOSPC; | |
2471 | strncpy(mddev->clevel, buf, len); | |
2472 | if (mddev->clevel[len-1] == '\n') | |
2473 | len--; | |
2474 | mddev->clevel[len] = 0; | |
2475 | mddev->level = LEVEL_NONE; | |
2476 | return rv; | |
2477 | } | |
2478 | ||
2479 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 2480 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 2481 | |
d4dbd025 N |
2482 | |
2483 | static ssize_t | |
2484 | layout_show(mddev_t *mddev, char *page) | |
2485 | { | |
2486 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
2487 | if (mddev->reshape_position != MaxSector && |
2488 | mddev->layout != mddev->new_layout) | |
2489 | return sprintf(page, "%d (%d)\n", | |
2490 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
2491 | return sprintf(page, "%d\n", mddev->layout); |
2492 | } | |
2493 | ||
2494 | static ssize_t | |
2495 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
2496 | { | |
2497 | char *e; | |
2498 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
2499 | |
2500 | if (!*buf || (*e && *e != '\n')) | |
2501 | return -EINVAL; | |
2502 | ||
08a02ecd N |
2503 | if (mddev->pers) |
2504 | return -EBUSY; | |
2505 | if (mddev->reshape_position != MaxSector) | |
2506 | mddev->new_layout = n; | |
2507 | else | |
2508 | mddev->layout = n; | |
d4dbd025 N |
2509 | return len; |
2510 | } | |
2511 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 2512 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
2513 | |
2514 | ||
eae1701f | 2515 | static ssize_t |
96de1e66 | 2516 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 2517 | { |
bb636547 N |
2518 | if (mddev->raid_disks == 0) |
2519 | return 0; | |
08a02ecd N |
2520 | if (mddev->reshape_position != MaxSector && |
2521 | mddev->delta_disks != 0) | |
2522 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
2523 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
2524 | return sprintf(page, "%d\n", mddev->raid_disks); |
2525 | } | |
2526 | ||
da943b99 N |
2527 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
2528 | ||
2529 | static ssize_t | |
2530 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
2531 | { | |
da943b99 N |
2532 | char *e; |
2533 | int rv = 0; | |
2534 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2535 | ||
2536 | if (!*buf || (*e && *e != '\n')) | |
2537 | return -EINVAL; | |
2538 | ||
2539 | if (mddev->pers) | |
2540 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
2541 | else if (mddev->reshape_position != MaxSector) { |
2542 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
2543 | mddev->delta_disks = n - olddisks; | |
2544 | mddev->raid_disks = n; | |
2545 | } else | |
da943b99 N |
2546 | mddev->raid_disks = n; |
2547 | return rv ? rv : len; | |
2548 | } | |
2549 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 2550 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 2551 | |
3b34380a N |
2552 | static ssize_t |
2553 | chunk_size_show(mddev_t *mddev, char *page) | |
2554 | { | |
08a02ecd N |
2555 | if (mddev->reshape_position != MaxSector && |
2556 | mddev->chunk_size != mddev->new_chunk) | |
2557 | return sprintf(page, "%d (%d)\n", mddev->new_chunk, | |
2558 | mddev->chunk_size); | |
3b34380a N |
2559 | return sprintf(page, "%d\n", mddev->chunk_size); |
2560 | } | |
2561 | ||
2562 | static ssize_t | |
2563 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
2564 | { | |
2565 | /* can only set chunk_size if array is not yet active */ | |
2566 | char *e; | |
2567 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2568 | ||
3b34380a N |
2569 | if (!*buf || (*e && *e != '\n')) |
2570 | return -EINVAL; | |
2571 | ||
08a02ecd N |
2572 | if (mddev->pers) |
2573 | return -EBUSY; | |
2574 | else if (mddev->reshape_position != MaxSector) | |
2575 | mddev->new_chunk = n; | |
2576 | else | |
2577 | mddev->chunk_size = n; | |
3b34380a N |
2578 | return len; |
2579 | } | |
2580 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 2581 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 2582 | |
a94213b1 N |
2583 | static ssize_t |
2584 | resync_start_show(mddev_t *mddev, char *page) | |
2585 | { | |
2586 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); | |
2587 | } | |
2588 | ||
2589 | static ssize_t | |
2590 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
2591 | { | |
a94213b1 N |
2592 | char *e; |
2593 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
2594 | ||
2595 | if (mddev->pers) | |
2596 | return -EBUSY; | |
2597 | if (!*buf || (*e && *e != '\n')) | |
2598 | return -EINVAL; | |
2599 | ||
2600 | mddev->recovery_cp = n; | |
2601 | return len; | |
2602 | } | |
2603 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 2604 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 2605 | |
9e653b63 N |
2606 | /* |
2607 | * The array state can be: | |
2608 | * | |
2609 | * clear | |
2610 | * No devices, no size, no level | |
2611 | * Equivalent to STOP_ARRAY ioctl | |
2612 | * inactive | |
2613 | * May have some settings, but array is not active | |
2614 | * all IO results in error | |
2615 | * When written, doesn't tear down array, but just stops it | |
2616 | * suspended (not supported yet) | |
2617 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 2618 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
2619 | * readonly |
2620 | * no resync can happen. no superblocks get written. | |
2621 | * write requests fail | |
2622 | * read-auto | |
2623 | * like readonly, but behaves like 'clean' on a write request. | |
2624 | * | |
2625 | * clean - no pending writes, but otherwise active. | |
2626 | * When written to inactive array, starts without resync | |
2627 | * If a write request arrives then | |
2628 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
2629 | * if not known, block and switch to write-pending | |
2630 | * If written to an active array that has pending writes, then fails. | |
2631 | * active | |
2632 | * fully active: IO and resync can be happening. | |
2633 | * When written to inactive array, starts with resync | |
2634 | * | |
2635 | * write-pending | |
2636 | * clean, but writes are blocked waiting for 'active' to be written. | |
2637 | * | |
2638 | * active-idle | |
2639 | * like active, but no writes have been seen for a while (100msec). | |
2640 | * | |
2641 | */ | |
2642 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
2643 | write_pending, active_idle, bad_word}; | |
05381954 | 2644 | static char *array_states[] = { |
9e653b63 N |
2645 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
2646 | "write-pending", "active-idle", NULL }; | |
2647 | ||
2648 | static int match_word(const char *word, char **list) | |
2649 | { | |
2650 | int n; | |
2651 | for (n=0; list[n]; n++) | |
2652 | if (cmd_match(word, list[n])) | |
2653 | break; | |
2654 | return n; | |
2655 | } | |
2656 | ||
2657 | static ssize_t | |
2658 | array_state_show(mddev_t *mddev, char *page) | |
2659 | { | |
2660 | enum array_state st = inactive; | |
2661 | ||
2662 | if (mddev->pers) | |
2663 | switch(mddev->ro) { | |
2664 | case 1: | |
2665 | st = readonly; | |
2666 | break; | |
2667 | case 2: | |
2668 | st = read_auto; | |
2669 | break; | |
2670 | case 0: | |
2671 | if (mddev->in_sync) | |
2672 | st = clean; | |
e691063a N |
2673 | else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) |
2674 | st = write_pending; | |
9e653b63 N |
2675 | else if (mddev->safemode) |
2676 | st = active_idle; | |
2677 | else | |
2678 | st = active; | |
2679 | } | |
2680 | else { | |
2681 | if (list_empty(&mddev->disks) && | |
2682 | mddev->raid_disks == 0 && | |
2683 | mddev->size == 0) | |
2684 | st = clear; | |
2685 | else | |
2686 | st = inactive; | |
2687 | } | |
2688 | return sprintf(page, "%s\n", array_states[st]); | |
2689 | } | |
2690 | ||
2691 | static int do_md_stop(mddev_t * mddev, int ro); | |
2692 | static int do_md_run(mddev_t * mddev); | |
2693 | static int restart_array(mddev_t *mddev); | |
2694 | ||
2695 | static ssize_t | |
2696 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
2697 | { | |
2698 | int err = -EINVAL; | |
2699 | enum array_state st = match_word(buf, array_states); | |
2700 | switch(st) { | |
2701 | case bad_word: | |
2702 | break; | |
2703 | case clear: | |
2704 | /* stopping an active array */ | |
e691063a N |
2705 | if (atomic_read(&mddev->active) > 1) |
2706 | return -EBUSY; | |
2707 | err = do_md_stop(mddev, 0); | |
9e653b63 N |
2708 | break; |
2709 | case inactive: | |
2710 | /* stopping an active array */ | |
2711 | if (mddev->pers) { | |
2712 | if (atomic_read(&mddev->active) > 1) | |
2713 | return -EBUSY; | |
2714 | err = do_md_stop(mddev, 2); | |
e691063a N |
2715 | } else |
2716 | err = 0; /* already inactive */ | |
9e653b63 N |
2717 | break; |
2718 | case suspended: | |
2719 | break; /* not supported yet */ | |
2720 | case readonly: | |
2721 | if (mddev->pers) | |
2722 | err = do_md_stop(mddev, 1); | |
2723 | else { | |
2724 | mddev->ro = 1; | |
648b629e | 2725 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
2726 | err = do_md_run(mddev); |
2727 | } | |
2728 | break; | |
2729 | case read_auto: | |
9e653b63 | 2730 | if (mddev->pers) { |
648b629e N |
2731 | if (mddev->ro != 1) |
2732 | err = do_md_stop(mddev, 1); | |
2733 | else | |
2734 | err = restart_array(mddev); | |
2735 | if (err == 0) { | |
2736 | mddev->ro = 2; | |
2737 | set_disk_ro(mddev->gendisk, 0); | |
2738 | } | |
9e653b63 N |
2739 | } else { |
2740 | mddev->ro = 2; | |
2741 | err = do_md_run(mddev); | |
2742 | } | |
2743 | break; | |
2744 | case clean: | |
2745 | if (mddev->pers) { | |
2746 | restart_array(mddev); | |
2747 | spin_lock_irq(&mddev->write_lock); | |
2748 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
2749 | if (mddev->in_sync == 0) { |
2750 | mddev->in_sync = 1; | |
31a59e34 N |
2751 | if (mddev->safemode == 1) |
2752 | mddev->safemode = 0; | |
e691063a N |
2753 | if (mddev->persistent) |
2754 | set_bit(MD_CHANGE_CLEAN, | |
2755 | &mddev->flags); | |
2756 | } | |
2757 | err = 0; | |
2758 | } else | |
2759 | err = -EBUSY; | |
9e653b63 N |
2760 | spin_unlock_irq(&mddev->write_lock); |
2761 | } else { | |
2762 | mddev->ro = 0; | |
2763 | mddev->recovery_cp = MaxSector; | |
2764 | err = do_md_run(mddev); | |
2765 | } | |
2766 | break; | |
2767 | case active: | |
2768 | if (mddev->pers) { | |
2769 | restart_array(mddev); | |
e691063a N |
2770 | if (mddev->external) |
2771 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
9e653b63 N |
2772 | wake_up(&mddev->sb_wait); |
2773 | err = 0; | |
2774 | } else { | |
2775 | mddev->ro = 0; | |
648b629e | 2776 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
2777 | err = do_md_run(mddev); |
2778 | } | |
2779 | break; | |
2780 | case write_pending: | |
2781 | case active_idle: | |
2782 | /* these cannot be set */ | |
2783 | break; | |
2784 | } | |
2785 | if (err) | |
2786 | return err; | |
0fd62b86 NB |
2787 | else { |
2788 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
9e653b63 | 2789 | return len; |
0fd62b86 | 2790 | } |
9e653b63 | 2791 | } |
80ca3a44 N |
2792 | static struct md_sysfs_entry md_array_state = |
2793 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 2794 | |
6d7ff738 N |
2795 | static ssize_t |
2796 | null_show(mddev_t *mddev, char *page) | |
2797 | { | |
2798 | return -EINVAL; | |
2799 | } | |
2800 | ||
2801 | static ssize_t | |
2802 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
2803 | { | |
2804 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
2805 | /* The new device is added to the array. | |
2806 | * If the array has a persistent superblock, we read the | |
2807 | * superblock to initialise info and check validity. | |
2808 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
2809 | * which mainly checks size. | |
2810 | */ | |
2811 | char *e; | |
2812 | int major = simple_strtoul(buf, &e, 10); | |
2813 | int minor; | |
2814 | dev_t dev; | |
2815 | mdk_rdev_t *rdev; | |
2816 | int err; | |
2817 | ||
2818 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
2819 | return -EINVAL; | |
2820 | minor = simple_strtoul(e+1, &e, 10); | |
2821 | if (*e && *e != '\n') | |
2822 | return -EINVAL; | |
2823 | dev = MKDEV(major, minor); | |
2824 | if (major != MAJOR(dev) || | |
2825 | minor != MINOR(dev)) | |
2826 | return -EOVERFLOW; | |
2827 | ||
2828 | ||
2829 | if (mddev->persistent) { | |
2830 | rdev = md_import_device(dev, mddev->major_version, | |
2831 | mddev->minor_version); | |
2832 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
2833 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
2834 | mdk_rdev_t, same_set); | |
2835 | err = super_types[mddev->major_version] | |
2836 | .load_super(rdev, rdev0, mddev->minor_version); | |
2837 | if (err < 0) | |
2838 | goto out; | |
2839 | } | |
c5d79adb N |
2840 | } else if (mddev->external) |
2841 | rdev = md_import_device(dev, -2, -1); | |
2842 | else | |
6d7ff738 N |
2843 | rdev = md_import_device(dev, -1, -1); |
2844 | ||
2845 | if (IS_ERR(rdev)) | |
2846 | return PTR_ERR(rdev); | |
2847 | err = bind_rdev_to_array(rdev, mddev); | |
2848 | out: | |
2849 | if (err) | |
2850 | export_rdev(rdev); | |
2851 | return err ? err : len; | |
2852 | } | |
2853 | ||
2854 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 2855 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 2856 | |
9b1d1dac PC |
2857 | static ssize_t |
2858 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
2859 | { | |
2860 | char *end; | |
2861 | unsigned long chunk, end_chunk; | |
2862 | ||
2863 | if (!mddev->bitmap) | |
2864 | goto out; | |
2865 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
2866 | while (*buf) { | |
2867 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
2868 | if (buf == end) break; | |
2869 | if (*end == '-') { /* range */ | |
2870 | buf = end + 1; | |
2871 | end_chunk = simple_strtoul(buf, &end, 0); | |
2872 | if (buf == end) break; | |
2873 | } | |
2874 | if (*end && !isspace(*end)) break; | |
2875 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
2876 | buf = end; | |
2877 | while (isspace(*buf)) buf++; | |
2878 | } | |
2879 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
2880 | out: | |
2881 | return len; | |
2882 | } | |
2883 | ||
2884 | static struct md_sysfs_entry md_bitmap = | |
2885 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
2886 | ||
a35b0d69 N |
2887 | static ssize_t |
2888 | size_show(mddev_t *mddev, char *page) | |
2889 | { | |
2890 | return sprintf(page, "%llu\n", (unsigned long long)mddev->size); | |
2891 | } | |
2892 | ||
2893 | static int update_size(mddev_t *mddev, unsigned long size); | |
2894 | ||
2895 | static ssize_t | |
2896 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
2897 | { | |
2898 | /* If array is inactive, we can reduce the component size, but | |
2899 | * not increase it (except from 0). | |
2900 | * If array is active, we can try an on-line resize | |
2901 | */ | |
2902 | char *e; | |
2903 | int err = 0; | |
2904 | unsigned long long size = simple_strtoull(buf, &e, 10); | |
2905 | if (!*buf || *buf == '\n' || | |
2906 | (*e && *e != '\n')) | |
2907 | return -EINVAL; | |
2908 | ||
2909 | if (mddev->pers) { | |
2910 | err = update_size(mddev, size); | |
850b2b42 | 2911 | md_update_sb(mddev, 1); |
a35b0d69 N |
2912 | } else { |
2913 | if (mddev->size == 0 || | |
2914 | mddev->size > size) | |
2915 | mddev->size = size; | |
2916 | else | |
2917 | err = -ENOSPC; | |
2918 | } | |
2919 | return err ? err : len; | |
2920 | } | |
2921 | ||
2922 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 2923 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 2924 | |
8bb93aac N |
2925 | |
2926 | /* Metdata version. | |
e691063a N |
2927 | * This is one of |
2928 | * 'none' for arrays with no metadata (good luck...) | |
2929 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
2930 | * or N.M for internally known formats |
2931 | */ | |
2932 | static ssize_t | |
2933 | metadata_show(mddev_t *mddev, char *page) | |
2934 | { | |
2935 | if (mddev->persistent) | |
2936 | return sprintf(page, "%d.%d\n", | |
2937 | mddev->major_version, mddev->minor_version); | |
e691063a N |
2938 | else if (mddev->external) |
2939 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
2940 | else |
2941 | return sprintf(page, "none\n"); | |
2942 | } | |
2943 | ||
2944 | static ssize_t | |
2945 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
2946 | { | |
2947 | int major, minor; | |
2948 | char *e; | |
2949 | if (!list_empty(&mddev->disks)) | |
2950 | return -EBUSY; | |
2951 | ||
2952 | if (cmd_match(buf, "none")) { | |
2953 | mddev->persistent = 0; | |
e691063a N |
2954 | mddev->external = 0; |
2955 | mddev->major_version = 0; | |
2956 | mddev->minor_version = 90; | |
2957 | return len; | |
2958 | } | |
2959 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 2960 | size_t namelen = len-9; |
e691063a N |
2961 | if (namelen >= sizeof(mddev->metadata_type)) |
2962 | namelen = sizeof(mddev->metadata_type)-1; | |
2963 | strncpy(mddev->metadata_type, buf+9, namelen); | |
2964 | mddev->metadata_type[namelen] = 0; | |
2965 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
2966 | mddev->metadata_type[--namelen] = 0; | |
2967 | mddev->persistent = 0; | |
2968 | mddev->external = 1; | |
8bb93aac N |
2969 | mddev->major_version = 0; |
2970 | mddev->minor_version = 90; | |
2971 | return len; | |
2972 | } | |
2973 | major = simple_strtoul(buf, &e, 10); | |
2974 | if (e==buf || *e != '.') | |
2975 | return -EINVAL; | |
2976 | buf = e+1; | |
2977 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 2978 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 2979 | return -EINVAL; |
50511da3 | 2980 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
2981 | return -ENOENT; |
2982 | mddev->major_version = major; | |
2983 | mddev->minor_version = minor; | |
2984 | mddev->persistent = 1; | |
e691063a | 2985 | mddev->external = 0; |
8bb93aac N |
2986 | return len; |
2987 | } | |
2988 | ||
2989 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 2990 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 2991 | |
24dd469d | 2992 | static ssize_t |
7eec314d | 2993 | action_show(mddev_t *mddev, char *page) |
24dd469d | 2994 | { |
7eec314d | 2995 | char *type = "idle"; |
31399d9e | 2996 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
2b12ab6d | 2997 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
2998 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
2999 | type = "reshape"; | |
3000 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3001 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3002 | type = "resync"; | |
3003 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3004 | type = "check"; | |
3005 | else | |
3006 | type = "repair"; | |
72a23c21 | 3007 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3008 | type = "recover"; |
3009 | } | |
3010 | return sprintf(page, "%s\n", type); | |
3011 | } | |
3012 | ||
3013 | static ssize_t | |
7eec314d | 3014 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3015 | { |
7eec314d N |
3016 | if (!mddev->pers || !mddev->pers->sync_request) |
3017 | return -EINVAL; | |
3018 | ||
bce74dac | 3019 | if (cmd_match(page, "idle")) { |
7eec314d N |
3020 | if (mddev->sync_thread) { |
3021 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
3022 | md_unregister_thread(mddev->sync_thread); | |
3023 | mddev->sync_thread = NULL; | |
3024 | mddev->recovery = 0; | |
3025 | } | |
03c902e1 N |
3026 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3027 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3028 | return -EBUSY; |
72a23c21 NB |
3029 | else if (cmd_match(page, "resync")) |
3030 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3031 | else if (cmd_match(page, "recover")) { | |
3032 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3033 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3034 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3035 | int err; |
3036 | if (mddev->pers->start_reshape == NULL) | |
3037 | return -EINVAL; | |
3038 | err = mddev->pers->start_reshape(mddev); | |
3039 | if (err) | |
3040 | return err; | |
a99ac971 | 3041 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3042 | } else { |
bce74dac | 3043 | if (cmd_match(page, "check")) |
7eec314d | 3044 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3045 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3046 | return -EINVAL; |
3047 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3048 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3049 | } |
03c902e1 | 3050 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3051 | md_wakeup_thread(mddev->thread); |
72a23c21 | 3052 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
24dd469d N |
3053 | return len; |
3054 | } | |
3055 | ||
9d88883e | 3056 | static ssize_t |
96de1e66 | 3057 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3058 | { |
3059 | return sprintf(page, "%llu\n", | |
3060 | (unsigned long long) mddev->resync_mismatches); | |
3061 | } | |
3062 | ||
80ca3a44 N |
3063 | static struct md_sysfs_entry md_scan_mode = |
3064 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3065 | |
96de1e66 | 3066 | |
80ca3a44 | 3067 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3068 | |
88202a0c N |
3069 | static ssize_t |
3070 | sync_min_show(mddev_t *mddev, char *page) | |
3071 | { | |
3072 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3073 | mddev->sync_speed_min ? "local": "system"); | |
3074 | } | |
3075 | ||
3076 | static ssize_t | |
3077 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3078 | { | |
3079 | int min; | |
3080 | char *e; | |
3081 | if (strncmp(buf, "system", 6)==0) { | |
3082 | mddev->sync_speed_min = 0; | |
3083 | return len; | |
3084 | } | |
3085 | min = simple_strtoul(buf, &e, 10); | |
3086 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3087 | return -EINVAL; | |
3088 | mddev->sync_speed_min = min; | |
3089 | return len; | |
3090 | } | |
3091 | ||
3092 | static struct md_sysfs_entry md_sync_min = | |
3093 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3094 | ||
3095 | static ssize_t | |
3096 | sync_max_show(mddev_t *mddev, char *page) | |
3097 | { | |
3098 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3099 | mddev->sync_speed_max ? "local": "system"); | |
3100 | } | |
3101 | ||
3102 | static ssize_t | |
3103 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3104 | { | |
3105 | int max; | |
3106 | char *e; | |
3107 | if (strncmp(buf, "system", 6)==0) { | |
3108 | mddev->sync_speed_max = 0; | |
3109 | return len; | |
3110 | } | |
3111 | max = simple_strtoul(buf, &e, 10); | |
3112 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3113 | return -EINVAL; | |
3114 | mddev->sync_speed_max = max; | |
3115 | return len; | |
3116 | } | |
3117 | ||
3118 | static struct md_sysfs_entry md_sync_max = | |
3119 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3120 | ||
d7f3d291 IP |
3121 | static ssize_t |
3122 | degraded_show(mddev_t *mddev, char *page) | |
3123 | { | |
3124 | return sprintf(page, "%d\n", mddev->degraded); | |
3125 | } | |
3126 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3127 | |
90b08710 BS |
3128 | static ssize_t |
3129 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3130 | { | |
3131 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3132 | } | |
3133 | ||
3134 | static ssize_t | |
3135 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3136 | { | |
3137 | long n; | |
3138 | ||
3139 | if (strict_strtol(buf, 10, &n)) | |
3140 | return -EINVAL; | |
3141 | ||
3142 | if (n != 0 && n != 1) | |
3143 | return -EINVAL; | |
3144 | ||
3145 | mddev->parallel_resync = n; | |
3146 | ||
3147 | if (mddev->sync_thread) | |
3148 | wake_up(&resync_wait); | |
3149 | ||
3150 | return len; | |
3151 | } | |
3152 | ||
3153 | /* force parallel resync, even with shared block devices */ | |
3154 | static struct md_sysfs_entry md_sync_force_parallel = | |
3155 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3156 | sync_force_parallel_show, sync_force_parallel_store); | |
3157 | ||
88202a0c N |
3158 | static ssize_t |
3159 | sync_speed_show(mddev_t *mddev, char *page) | |
3160 | { | |
3161 | unsigned long resync, dt, db; | |
9687a60c AN |
3162 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3163 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3164 | if (!dt) dt++; |
9687a60c AN |
3165 | db = resync - mddev->resync_mark_cnt; |
3166 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3167 | } |
3168 | ||
80ca3a44 | 3169 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3170 | |
3171 | static ssize_t | |
3172 | sync_completed_show(mddev_t *mddev, char *page) | |
3173 | { | |
3174 | unsigned long max_blocks, resync; | |
3175 | ||
3176 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
3177 | max_blocks = mddev->resync_max_sectors; | |
3178 | else | |
3179 | max_blocks = mddev->size << 1; | |
3180 | ||
3181 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active)); | |
3182 | return sprintf(page, "%lu / %lu\n", resync, max_blocks); | |
3183 | } | |
3184 | ||
80ca3a44 | 3185 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3186 | |
5e96ee65 NB |
3187 | static ssize_t |
3188 | min_sync_show(mddev_t *mddev, char *page) | |
3189 | { | |
3190 | return sprintf(page, "%llu\n", | |
3191 | (unsigned long long)mddev->resync_min); | |
3192 | } | |
3193 | static ssize_t | |
3194 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3195 | { | |
3196 | unsigned long long min; | |
3197 | if (strict_strtoull(buf, 10, &min)) | |
3198 | return -EINVAL; | |
3199 | if (min > mddev->resync_max) | |
3200 | return -EINVAL; | |
3201 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3202 | return -EBUSY; | |
3203 | ||
3204 | /* Must be a multiple of chunk_size */ | |
3205 | if (mddev->chunk_size) { | |
3206 | if (min & (sector_t)((mddev->chunk_size>>9)-1)) | |
3207 | return -EINVAL; | |
3208 | } | |
3209 | mddev->resync_min = min; | |
3210 | ||
3211 | return len; | |
3212 | } | |
3213 | ||
3214 | static struct md_sysfs_entry md_min_sync = | |
3215 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3216 | ||
c6207277 N |
3217 | static ssize_t |
3218 | max_sync_show(mddev_t *mddev, char *page) | |
3219 | { | |
3220 | if (mddev->resync_max == MaxSector) | |
3221 | return sprintf(page, "max\n"); | |
3222 | else | |
3223 | return sprintf(page, "%llu\n", | |
3224 | (unsigned long long)mddev->resync_max); | |
3225 | } | |
3226 | static ssize_t | |
3227 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3228 | { | |
3229 | if (strncmp(buf, "max", 3) == 0) | |
3230 | mddev->resync_max = MaxSector; | |
3231 | else { | |
5e96ee65 NB |
3232 | unsigned long long max; |
3233 | if (strict_strtoull(buf, 10, &max)) | |
3234 | return -EINVAL; | |
3235 | if (max < mddev->resync_min) | |
c6207277 N |
3236 | return -EINVAL; |
3237 | if (max < mddev->resync_max && | |
3238 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3239 | return -EBUSY; | |
3240 | ||
3241 | /* Must be a multiple of chunk_size */ | |
3242 | if (mddev->chunk_size) { | |
3243 | if (max & (sector_t)((mddev->chunk_size>>9)-1)) | |
3244 | return -EINVAL; | |
3245 | } | |
3246 | mddev->resync_max = max; | |
3247 | } | |
3248 | wake_up(&mddev->recovery_wait); | |
3249 | return len; | |
3250 | } | |
3251 | ||
3252 | static struct md_sysfs_entry md_max_sync = | |
3253 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
3254 | ||
e464eafd N |
3255 | static ssize_t |
3256 | suspend_lo_show(mddev_t *mddev, char *page) | |
3257 | { | |
3258 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
3259 | } | |
3260 | ||
3261 | static ssize_t | |
3262 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
3263 | { | |
3264 | char *e; | |
3265 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3266 | ||
3267 | if (mddev->pers->quiesce == NULL) | |
3268 | return -EINVAL; | |
3269 | if (buf == e || (*e && *e != '\n')) | |
3270 | return -EINVAL; | |
3271 | if (new >= mddev->suspend_hi || | |
3272 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
3273 | mddev->suspend_lo = new; | |
3274 | mddev->pers->quiesce(mddev, 2); | |
3275 | return len; | |
3276 | } else | |
3277 | return -EINVAL; | |
3278 | } | |
3279 | static struct md_sysfs_entry md_suspend_lo = | |
3280 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
3281 | ||
3282 | ||
3283 | static ssize_t | |
3284 | suspend_hi_show(mddev_t *mddev, char *page) | |
3285 | { | |
3286 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
3287 | } | |
3288 | ||
3289 | static ssize_t | |
3290 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
3291 | { | |
3292 | char *e; | |
3293 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3294 | ||
3295 | if (mddev->pers->quiesce == NULL) | |
3296 | return -EINVAL; | |
3297 | if (buf == e || (*e && *e != '\n')) | |
3298 | return -EINVAL; | |
3299 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
3300 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
3301 | mddev->suspend_hi = new; | |
3302 | mddev->pers->quiesce(mddev, 1); | |
3303 | mddev->pers->quiesce(mddev, 0); | |
3304 | return len; | |
3305 | } else | |
3306 | return -EINVAL; | |
3307 | } | |
3308 | static struct md_sysfs_entry md_suspend_hi = | |
3309 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
3310 | ||
08a02ecd N |
3311 | static ssize_t |
3312 | reshape_position_show(mddev_t *mddev, char *page) | |
3313 | { | |
3314 | if (mddev->reshape_position != MaxSector) | |
3315 | return sprintf(page, "%llu\n", | |
3316 | (unsigned long long)mddev->reshape_position); | |
3317 | strcpy(page, "none\n"); | |
3318 | return 5; | |
3319 | } | |
3320 | ||
3321 | static ssize_t | |
3322 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
3323 | { | |
3324 | char *e; | |
3325 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3326 | if (mddev->pers) | |
3327 | return -EBUSY; | |
3328 | if (buf == e || (*e && *e != '\n')) | |
3329 | return -EINVAL; | |
3330 | mddev->reshape_position = new; | |
3331 | mddev->delta_disks = 0; | |
3332 | mddev->new_level = mddev->level; | |
3333 | mddev->new_layout = mddev->layout; | |
3334 | mddev->new_chunk = mddev->chunk_size; | |
3335 | return len; | |
3336 | } | |
3337 | ||
3338 | static struct md_sysfs_entry md_reshape_position = | |
3339 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
3340 | reshape_position_store); | |
3341 | ||
e464eafd | 3342 | |
eae1701f N |
3343 | static struct attribute *md_default_attrs[] = { |
3344 | &md_level.attr, | |
d4dbd025 | 3345 | &md_layout.attr, |
eae1701f | 3346 | &md_raid_disks.attr, |
3b34380a | 3347 | &md_chunk_size.attr, |
a35b0d69 | 3348 | &md_size.attr, |
a94213b1 | 3349 | &md_resync_start.attr, |
8bb93aac | 3350 | &md_metadata.attr, |
6d7ff738 | 3351 | &md_new_device.attr, |
16f17b39 | 3352 | &md_safe_delay.attr, |
9e653b63 | 3353 | &md_array_state.attr, |
08a02ecd | 3354 | &md_reshape_position.attr, |
411036fa N |
3355 | NULL, |
3356 | }; | |
3357 | ||
3358 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 3359 | &md_scan_mode.attr, |
9d88883e | 3360 | &md_mismatches.attr, |
88202a0c N |
3361 | &md_sync_min.attr, |
3362 | &md_sync_max.attr, | |
3363 | &md_sync_speed.attr, | |
90b08710 | 3364 | &md_sync_force_parallel.attr, |
88202a0c | 3365 | &md_sync_completed.attr, |
5e96ee65 | 3366 | &md_min_sync.attr, |
c6207277 | 3367 | &md_max_sync.attr, |
e464eafd N |
3368 | &md_suspend_lo.attr, |
3369 | &md_suspend_hi.attr, | |
9b1d1dac | 3370 | &md_bitmap.attr, |
d7f3d291 | 3371 | &md_degraded.attr, |
eae1701f N |
3372 | NULL, |
3373 | }; | |
411036fa N |
3374 | static struct attribute_group md_redundancy_group = { |
3375 | .name = NULL, | |
3376 | .attrs = md_redundancy_attrs, | |
3377 | }; | |
3378 | ||
eae1701f N |
3379 | |
3380 | static ssize_t | |
3381 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3382 | { | |
3383 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3384 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3385 | ssize_t rv; |
eae1701f N |
3386 | |
3387 | if (!entry->show) | |
3388 | return -EIO; | |
5dc5cf7d IM |
3389 | rv = mddev_lock(mddev); |
3390 | if (!rv) { | |
3391 | rv = entry->show(mddev, page); | |
3392 | mddev_unlock(mddev); | |
3393 | } | |
96de1e66 | 3394 | return rv; |
eae1701f N |
3395 | } |
3396 | ||
3397 | static ssize_t | |
3398 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
3399 | const char *page, size_t length) | |
3400 | { | |
3401 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3402 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3403 | ssize_t rv; |
eae1701f N |
3404 | |
3405 | if (!entry->store) | |
3406 | return -EIO; | |
67463acb N |
3407 | if (!capable(CAP_SYS_ADMIN)) |
3408 | return -EACCES; | |
5dc5cf7d IM |
3409 | rv = mddev_lock(mddev); |
3410 | if (!rv) { | |
3411 | rv = entry->store(mddev, page, length); | |
3412 | mddev_unlock(mddev); | |
3413 | } | |
96de1e66 | 3414 | return rv; |
eae1701f N |
3415 | } |
3416 | ||
3417 | static void md_free(struct kobject *ko) | |
3418 | { | |
3419 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
3420 | kfree(mddev); | |
3421 | } | |
3422 | ||
3423 | static struct sysfs_ops md_sysfs_ops = { | |
3424 | .show = md_attr_show, | |
3425 | .store = md_attr_store, | |
3426 | }; | |
3427 | static struct kobj_type md_ktype = { | |
3428 | .release = md_free, | |
3429 | .sysfs_ops = &md_sysfs_ops, | |
3430 | .default_attrs = md_default_attrs, | |
3431 | }; | |
3432 | ||
1da177e4 LT |
3433 | int mdp_major = 0; |
3434 | ||
3435 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
3436 | { | |
48c9c27b | 3437 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
3438 | mddev_t *mddev = mddev_find(dev); |
3439 | struct gendisk *disk; | |
3440 | int partitioned = (MAJOR(dev) != MD_MAJOR); | |
3441 | int shift = partitioned ? MdpMinorShift : 0; | |
3442 | int unit = MINOR(dev) >> shift; | |
3830c62f | 3443 | int error; |
1da177e4 LT |
3444 | |
3445 | if (!mddev) | |
3446 | return NULL; | |
3447 | ||
48c9c27b | 3448 | mutex_lock(&disks_mutex); |
1da177e4 | 3449 | if (mddev->gendisk) { |
48c9c27b | 3450 | mutex_unlock(&disks_mutex); |
1da177e4 LT |
3451 | mddev_put(mddev); |
3452 | return NULL; | |
3453 | } | |
3454 | disk = alloc_disk(1 << shift); | |
3455 | if (!disk) { | |
48c9c27b | 3456 | mutex_unlock(&disks_mutex); |
1da177e4 LT |
3457 | mddev_put(mddev); |
3458 | return NULL; | |
3459 | } | |
3460 | disk->major = MAJOR(dev); | |
3461 | disk->first_minor = unit << shift; | |
ce7b0f46 | 3462 | if (partitioned) |
1da177e4 | 3463 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 3464 | else |
1da177e4 | 3465 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
3466 | disk->fops = &md_fops; |
3467 | disk->private_data = mddev; | |
3468 | disk->queue = mddev->queue; | |
3469 | add_disk(disk); | |
3470 | mddev->gendisk = disk; | |
edfaa7c3 | 3471 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj, |
3830c62f | 3472 | "%s", "md"); |
f48ed538 | 3473 | mutex_unlock(&disks_mutex); |
3830c62f | 3474 | if (error) |
5e55e2f5 N |
3475 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
3476 | disk->disk_name); | |
3830c62f GKH |
3477 | else |
3478 | kobject_uevent(&mddev->kobj, KOBJ_ADD); | |
1da177e4 LT |
3479 | return NULL; |
3480 | } | |
3481 | ||
1da177e4 LT |
3482 | static void md_safemode_timeout(unsigned long data) |
3483 | { | |
3484 | mddev_t *mddev = (mddev_t *) data; | |
3485 | ||
0fd62b86 NB |
3486 | if (!atomic_read(&mddev->writes_pending)) { |
3487 | mddev->safemode = 1; | |
3488 | if (mddev->external) | |
3489 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
3490 | } | |
1da177e4 LT |
3491 | md_wakeup_thread(mddev->thread); |
3492 | } | |
3493 | ||
6ff8d8ec | 3494 | static int start_dirty_degraded; |
1da177e4 LT |
3495 | |
3496 | static int do_md_run(mddev_t * mddev) | |
3497 | { | |
2604b703 | 3498 | int err; |
1da177e4 LT |
3499 | int chunk_size; |
3500 | struct list_head *tmp; | |
3501 | mdk_rdev_t *rdev; | |
3502 | struct gendisk *disk; | |
2604b703 | 3503 | struct mdk_personality *pers; |
1da177e4 LT |
3504 | char b[BDEVNAME_SIZE]; |
3505 | ||
a757e64c N |
3506 | if (list_empty(&mddev->disks)) |
3507 | /* cannot run an array with no devices.. */ | |
1da177e4 | 3508 | return -EINVAL; |
1da177e4 LT |
3509 | |
3510 | if (mddev->pers) | |
3511 | return -EBUSY; | |
3512 | ||
3513 | /* | |
3514 | * Analyze all RAID superblock(s) | |
3515 | */ | |
1ec4a939 N |
3516 | if (!mddev->raid_disks) { |
3517 | if (!mddev->persistent) | |
3518 | return -EINVAL; | |
a757e64c | 3519 | analyze_sbs(mddev); |
1ec4a939 | 3520 | } |
1da177e4 LT |
3521 | |
3522 | chunk_size = mddev->chunk_size; | |
2604b703 N |
3523 | |
3524 | if (chunk_size) { | |
1da177e4 LT |
3525 | if (chunk_size > MAX_CHUNK_SIZE) { |
3526 | printk(KERN_ERR "too big chunk_size: %d > %d\n", | |
3527 | chunk_size, MAX_CHUNK_SIZE); | |
3528 | return -EINVAL; | |
3529 | } | |
3530 | /* | |
3531 | * chunk-size has to be a power of 2 and multiples of PAGE_SIZE | |
3532 | */ | |
3533 | if ( (1 << ffz(~chunk_size)) != chunk_size) { | |
a757e64c | 3534 | printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size); |
1da177e4 LT |
3535 | return -EINVAL; |
3536 | } | |
3537 | if (chunk_size < PAGE_SIZE) { | |
3538 | printk(KERN_ERR "too small chunk_size: %d < %ld\n", | |
3539 | chunk_size, PAGE_SIZE); | |
3540 | return -EINVAL; | |
3541 | } | |
3542 | ||
3543 | /* devices must have minimum size of one chunk */ | |
d089c6af | 3544 | rdev_for_each(rdev, tmp, mddev) { |
b2d444d7 | 3545 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
3546 | continue; |
3547 | if (rdev->size < chunk_size / 1024) { | |
3548 | printk(KERN_WARNING | |
3549 | "md: Dev %s smaller than chunk_size:" | |
3550 | " %lluk < %dk\n", | |
3551 | bdevname(rdev->bdev,b), | |
3552 | (unsigned long long)rdev->size, | |
3553 | chunk_size / 1024); | |
3554 | return -EINVAL; | |
3555 | } | |
3556 | } | |
3557 | } | |
3558 | ||
1da177e4 | 3559 | #ifdef CONFIG_KMOD |
d9d166c2 N |
3560 | if (mddev->level != LEVEL_NONE) |
3561 | request_module("md-level-%d", mddev->level); | |
3562 | else if (mddev->clevel[0]) | |
3563 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
3564 | #endif |
3565 | ||
3566 | /* | |
3567 | * Drop all container device buffers, from now on | |
3568 | * the only valid external interface is through the md | |
3569 | * device. | |
1da177e4 | 3570 | */ |
d089c6af | 3571 | rdev_for_each(rdev, tmp, mddev) { |
b2d444d7 | 3572 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
3573 | continue; |
3574 | sync_blockdev(rdev->bdev); | |
f98393a6 | 3575 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
3576 | |
3577 | /* perform some consistency tests on the device. | |
3578 | * We don't want the data to overlap the metadata, | |
3579 | * Internal Bitmap issues has handled elsewhere. | |
3580 | */ | |
3581 | if (rdev->data_offset < rdev->sb_offset) { | |
3582 | if (mddev->size && | |
3583 | rdev->data_offset + mddev->size*2 | |
3584 | > rdev->sb_offset*2) { | |
3585 | printk("md: %s: data overlaps metadata\n", | |
3586 | mdname(mddev)); | |
3587 | return -EINVAL; | |
3588 | } | |
3589 | } else { | |
3590 | if (rdev->sb_offset*2 + rdev->sb_size/512 | |
3591 | > rdev->data_offset) { | |
3592 | printk("md: %s: metadata overlaps data\n", | |
3593 | mdname(mddev)); | |
3594 | return -EINVAL; | |
3595 | } | |
3596 | } | |
52664732 | 3597 | sysfs_notify(&rdev->kobj, NULL, "state"); |
1da177e4 LT |
3598 | } |
3599 | ||
3600 | md_probe(mddev->unit, NULL, NULL); | |
3601 | disk = mddev->gendisk; | |
3602 | if (!disk) | |
3603 | return -ENOMEM; | |
3604 | ||
3605 | spin_lock(&pers_lock); | |
d9d166c2 | 3606 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 3607 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 3608 | spin_unlock(&pers_lock); |
d9d166c2 N |
3609 | if (mddev->level != LEVEL_NONE) |
3610 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
3611 | mddev->level); | |
3612 | else | |
3613 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
3614 | mddev->clevel); | |
1da177e4 LT |
3615 | return -EINVAL; |
3616 | } | |
2604b703 | 3617 | mddev->pers = pers; |
1da177e4 | 3618 | spin_unlock(&pers_lock); |
d9d166c2 N |
3619 | mddev->level = pers->level; |
3620 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
1da177e4 | 3621 | |
f6705578 | 3622 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 3623 | pers->start_reshape == NULL) { |
f6705578 N |
3624 | /* This personality cannot handle reshaping... */ |
3625 | mddev->pers = NULL; | |
3626 | module_put(pers->owner); | |
3627 | return -EINVAL; | |
3628 | } | |
3629 | ||
7dd5e7c3 N |
3630 | if (pers->sync_request) { |
3631 | /* Warn if this is a potentially silly | |
3632 | * configuration. | |
3633 | */ | |
3634 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3635 | mdk_rdev_t *rdev2; | |
3636 | struct list_head *tmp2; | |
3637 | int warned = 0; | |
d089c6af N |
3638 | rdev_for_each(rdev, tmp, mddev) { |
3639 | rdev_for_each(rdev2, tmp2, mddev) { | |
7dd5e7c3 N |
3640 | if (rdev < rdev2 && |
3641 | rdev->bdev->bd_contains == | |
3642 | rdev2->bdev->bd_contains) { | |
3643 | printk(KERN_WARNING | |
3644 | "%s: WARNING: %s appears to be" | |
3645 | " on the same physical disk as" | |
3646 | " %s.\n", | |
3647 | mdname(mddev), | |
3648 | bdevname(rdev->bdev,b), | |
3649 | bdevname(rdev2->bdev,b2)); | |
3650 | warned = 1; | |
3651 | } | |
3652 | } | |
3653 | } | |
3654 | if (warned) | |
3655 | printk(KERN_WARNING | |
3656 | "True protection against single-disk" | |
3657 | " failure might be compromised.\n"); | |
3658 | } | |
3659 | ||
657390d2 | 3660 | mddev->recovery = 0; |
1da177e4 | 3661 | mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */ |
a9701a30 | 3662 | mddev->barriers_work = 1; |
6ff8d8ec | 3663 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 3664 | |
f91de92e N |
3665 | if (start_readonly) |
3666 | mddev->ro = 2; /* read-only, but switch on first write */ | |
3667 | ||
b15c2e57 | 3668 | err = mddev->pers->run(mddev); |
13e53df3 AN |
3669 | if (err) |
3670 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
3671 | else if (mddev->pers->sync_request) { | |
b15c2e57 N |
3672 | err = bitmap_create(mddev); |
3673 | if (err) { | |
3674 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
3675 | mdname(mddev), err); | |
3676 | mddev->pers->stop(mddev); | |
3677 | } | |
3678 | } | |
1da177e4 | 3679 | if (err) { |
1da177e4 LT |
3680 | module_put(mddev->pers->owner); |
3681 | mddev->pers = NULL; | |
32a7627c N |
3682 | bitmap_destroy(mddev); |
3683 | return err; | |
1da177e4 | 3684 | } |
5e55e2f5 N |
3685 | if (mddev->pers->sync_request) { |
3686 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3687 | printk(KERN_WARNING | |
3688 | "md: cannot register extra attributes for %s\n", | |
3689 | mdname(mddev)); | |
3690 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ | |
fd9d49ca N |
3691 | mddev->ro = 0; |
3692 | ||
1da177e4 LT |
3693 | atomic_set(&mddev->writes_pending,0); |
3694 | mddev->safemode = 0; | |
3695 | mddev->safemode_timer.function = md_safemode_timeout; | |
3696 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 3697 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 3698 | mddev->in_sync = 1; |
86e6ffdd | 3699 | |
d089c6af | 3700 | rdev_for_each(rdev, tmp, mddev) |
86e6ffdd N |
3701 | if (rdev->raid_disk >= 0) { |
3702 | char nm[20]; | |
3703 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
3704 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
3705 | printk("md: cannot register %s for %s\n", | |
3706 | nm, mdname(mddev)); | |
86e6ffdd | 3707 | } |
1da177e4 LT |
3708 | |
3709 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3710 | ||
850b2b42 N |
3711 | if (mddev->flags) |
3712 | md_update_sb(mddev, 0); | |
1da177e4 LT |
3713 | |
3714 | set_capacity(disk, mddev->array_size<<1); | |
3715 | ||
3716 | /* If we call blk_queue_make_request here, it will | |
3717 | * re-initialise max_sectors etc which may have been | |
3718 | * refined inside -> run. So just set the bits we need to set. | |
3719 | * Most initialisation happended when we called | |
3720 | * blk_queue_make_request(..., md_fail_request) | |
3721 | * earlier. | |
3722 | */ | |
3723 | mddev->queue->queuedata = mddev; | |
3724 | mddev->queue->make_request_fn = mddev->pers->make_request; | |
3725 | ||
5fd6c1dc N |
3726 | /* If there is a partially-recovered drive we need to |
3727 | * start recovery here. If we leave it to md_check_recovery, | |
3728 | * it will remove the drives and not do the right thing | |
3729 | */ | |
0b8c9de0 | 3730 | if (mddev->degraded && !mddev->sync_thread) { |
5fd6c1dc N |
3731 | struct list_head *rtmp; |
3732 | int spares = 0; | |
d089c6af | 3733 | rdev_for_each(rdev, rtmp, mddev) |
5fd6c1dc N |
3734 | if (rdev->raid_disk >= 0 && |
3735 | !test_bit(In_sync, &rdev->flags) && | |
3736 | !test_bit(Faulty, &rdev->flags)) | |
3737 | /* complete an interrupted recovery */ | |
3738 | spares++; | |
3739 | if (spares && mddev->pers->sync_request) { | |
3740 | mddev->recovery = 0; | |
3741 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3742 | mddev->sync_thread = md_register_thread(md_do_sync, | |
3743 | mddev, | |
3744 | "%s_resync"); | |
3745 | if (!mddev->sync_thread) { | |
3746 | printk(KERN_ERR "%s: could not start resync" | |
3747 | " thread...\n", | |
3748 | mdname(mddev)); | |
3749 | /* leave the spares where they are, it shouldn't hurt */ | |
3750 | mddev->recovery = 0; | |
0b8c9de0 | 3751 | } |
5fd6c1dc N |
3752 | } |
3753 | } | |
0b8c9de0 N |
3754 | md_wakeup_thread(mddev->thread); |
3755 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 3756 | |
44ce6294 | 3757 | mddev->changed = 1; |
d7603b7e | 3758 | md_new_event(mddev); |
0fd62b86 | 3759 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
72a23c21 | 3760 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
a99ac971 | 3761 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
edfaa7c3 | 3762 | kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE); |
1da177e4 LT |
3763 | return 0; |
3764 | } | |
3765 | ||
3766 | static int restart_array(mddev_t *mddev) | |
3767 | { | |
3768 | struct gendisk *disk = mddev->gendisk; | |
3769 | int err; | |
3770 | ||
3771 | /* | |
3772 | * Complain if it has no devices | |
3773 | */ | |
3774 | err = -ENXIO; | |
3775 | if (list_empty(&mddev->disks)) | |
3776 | goto out; | |
3777 | ||
3778 | if (mddev->pers) { | |
3779 | err = -EBUSY; | |
3780 | if (!mddev->ro) | |
3781 | goto out; | |
3782 | ||
3783 | mddev->safemode = 0; | |
3784 | mddev->ro = 0; | |
3785 | set_disk_ro(disk, 0); | |
3786 | ||
3787 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
3788 | mdname(mddev)); | |
3789 | /* | |
3790 | * Kick recovery or resync if necessary | |
3791 | */ | |
3792 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3793 | md_wakeup_thread(mddev->thread); | |
5fd6c1dc | 3794 | md_wakeup_thread(mddev->sync_thread); |
1da177e4 | 3795 | err = 0; |
0fd62b86 NB |
3796 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
3797 | ||
9e653b63 | 3798 | } else |
1da177e4 | 3799 | err = -EINVAL; |
1da177e4 LT |
3800 | |
3801 | out: | |
3802 | return err; | |
3803 | } | |
3804 | ||
acc55e22 N |
3805 | /* similar to deny_write_access, but accounts for our holding a reference |
3806 | * to the file ourselves */ | |
3807 | static int deny_bitmap_write_access(struct file * file) | |
3808 | { | |
3809 | struct inode *inode = file->f_mapping->host; | |
3810 | ||
3811 | spin_lock(&inode->i_lock); | |
3812 | if (atomic_read(&inode->i_writecount) > 1) { | |
3813 | spin_unlock(&inode->i_lock); | |
3814 | return -ETXTBSY; | |
3815 | } | |
3816 | atomic_set(&inode->i_writecount, -1); | |
3817 | spin_unlock(&inode->i_lock); | |
3818 | ||
3819 | return 0; | |
3820 | } | |
3821 | ||
3822 | static void restore_bitmap_write_access(struct file *file) | |
3823 | { | |
3824 | struct inode *inode = file->f_mapping->host; | |
3825 | ||
3826 | spin_lock(&inode->i_lock); | |
3827 | atomic_set(&inode->i_writecount, 1); | |
3828 | spin_unlock(&inode->i_lock); | |
3829 | } | |
3830 | ||
9e653b63 N |
3831 | /* mode: |
3832 | * 0 - completely stop and dis-assemble array | |
3833 | * 1 - switch to readonly | |
3834 | * 2 - stop but do not disassemble array | |
3835 | */ | |
3836 | static int do_md_stop(mddev_t * mddev, int mode) | |
1da177e4 LT |
3837 | { |
3838 | int err = 0; | |
3839 | struct gendisk *disk = mddev->gendisk; | |
3840 | ||
3841 | if (mddev->pers) { | |
3842 | if (atomic_read(&mddev->active)>2) { | |
3843 | printk("md: %s still in use.\n",mdname(mddev)); | |
3844 | return -EBUSY; | |
3845 | } | |
3846 | ||
3847 | if (mddev->sync_thread) { | |
5fd6c1dc | 3848 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 LT |
3849 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
3850 | md_unregister_thread(mddev->sync_thread); | |
3851 | mddev->sync_thread = NULL; | |
3852 | } | |
3853 | ||
3854 | del_timer_sync(&mddev->safemode_timer); | |
3855 | ||
3856 | invalidate_partition(disk, 0); | |
3857 | ||
9e653b63 N |
3858 | switch(mode) { |
3859 | case 1: /* readonly */ | |
1da177e4 | 3860 | err = -ENXIO; |
f91de92e | 3861 | if (mddev->ro==1) |
1da177e4 LT |
3862 | goto out; |
3863 | mddev->ro = 1; | |
9e653b63 N |
3864 | break; |
3865 | case 0: /* disassemble */ | |
3866 | case 2: /* stop */ | |
6b8b3e8a | 3867 | bitmap_flush(mddev); |
a9701a30 | 3868 | md_super_wait(mddev); |
1da177e4 LT |
3869 | if (mddev->ro) |
3870 | set_disk_ro(disk, 0); | |
3871 | blk_queue_make_request(mddev->queue, md_fail_request); | |
3872 | mddev->pers->stop(mddev); | |
d1b5380c N |
3873 | mddev->queue->merge_bvec_fn = NULL; |
3874 | mddev->queue->unplug_fn = NULL; | |
041ae52e | 3875 | mddev->queue->backing_dev_info.congested_fn = NULL; |
411036fa N |
3876 | if (mddev->pers->sync_request) |
3877 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
3878 | ||
1da177e4 LT |
3879 | module_put(mddev->pers->owner); |
3880 | mddev->pers = NULL; | |
4f54b0e9 DW |
3881 | /* tell userspace to handle 'inactive' */ |
3882 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
0d4ca600 N |
3883 | |
3884 | set_capacity(disk, 0); | |
44ce6294 | 3885 | mddev->changed = 1; |
0d4ca600 | 3886 | |
1da177e4 LT |
3887 | if (mddev->ro) |
3888 | mddev->ro = 0; | |
3889 | } | |
850b2b42 | 3890 | if (!mddev->in_sync || mddev->flags) { |
1da177e4 LT |
3891 | /* mark array as shutdown cleanly */ |
3892 | mddev->in_sync = 1; | |
850b2b42 | 3893 | md_update_sb(mddev, 1); |
1da177e4 | 3894 | } |
9e653b63 | 3895 | if (mode == 1) |
1da177e4 | 3896 | set_disk_ro(disk, 1); |
5fd6c1dc | 3897 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 | 3898 | } |
32a7627c | 3899 | |
1da177e4 LT |
3900 | /* |
3901 | * Free resources if final stop | |
3902 | */ | |
9e653b63 | 3903 | if (mode == 0) { |
86e6ffdd N |
3904 | mdk_rdev_t *rdev; |
3905 | struct list_head *tmp; | |
0d4ca600 | 3906 | |
1da177e4 LT |
3907 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
3908 | ||
978f946b N |
3909 | bitmap_destroy(mddev); |
3910 | if (mddev->bitmap_file) { | |
acc55e22 | 3911 | restore_bitmap_write_access(mddev->bitmap_file); |
978f946b N |
3912 | fput(mddev->bitmap_file); |
3913 | mddev->bitmap_file = NULL; | |
3914 | } | |
3915 | mddev->bitmap_offset = 0; | |
3916 | ||
d089c6af | 3917 | rdev_for_each(rdev, tmp, mddev) |
86e6ffdd N |
3918 | if (rdev->raid_disk >= 0) { |
3919 | char nm[20]; | |
3920 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3921 | sysfs_remove_link(&mddev->kobj, nm); | |
3922 | } | |
3923 | ||
177a99b2 | 3924 | /* make sure all md_delayed_delete calls have finished */ |
5792a285 N |
3925 | flush_scheduled_work(); |
3926 | ||
1da177e4 LT |
3927 | export_array(mddev); |
3928 | ||
3929 | mddev->array_size = 0; | |
9e653b63 N |
3930 | mddev->size = 0; |
3931 | mddev->raid_disks = 0; | |
a94213b1 | 3932 | mddev->recovery_cp = 0; |
5e96ee65 | 3933 | mddev->resync_min = 0; |
c6207277 | 3934 | mddev->resync_max = MaxSector; |
08a02ecd | 3935 | mddev->reshape_position = MaxSector; |
e691063a | 3936 | mddev->external = 0; |
1ec4a939 | 3937 | mddev->persistent = 0; |
d897dbf9 N |
3938 | mddev->level = LEVEL_NONE; |
3939 | mddev->clevel[0] = 0; | |
3940 | mddev->flags = 0; | |
3941 | mddev->ro = 0; | |
3942 | mddev->metadata_type[0] = 0; | |
3943 | mddev->chunk_size = 0; | |
3944 | mddev->ctime = mddev->utime = 0; | |
3945 | mddev->layout = 0; | |
3946 | mddev->max_disks = 0; | |
3947 | mddev->events = 0; | |
3948 | mddev->delta_disks = 0; | |
3949 | mddev->new_level = LEVEL_NONE; | |
3950 | mddev->new_layout = 0; | |
3951 | mddev->new_chunk = 0; | |
3952 | mddev->curr_resync = 0; | |
3953 | mddev->resync_mismatches = 0; | |
3954 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
3955 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
3956 | mddev->recovery = 0; | |
3957 | mddev->in_sync = 0; | |
3958 | mddev->changed = 0; | |
3959 | mddev->degraded = 0; | |
3960 | mddev->barriers_work = 0; | |
3961 | mddev->safemode = 0; | |
9e653b63 | 3962 | |
a8a55c38 | 3963 | } else if (mddev->pers) |
1da177e4 LT |
3964 | printk(KERN_INFO "md: %s switched to read-only mode.\n", |
3965 | mdname(mddev)); | |
3966 | err = 0; | |
d7603b7e | 3967 | md_new_event(mddev); |
0fd62b86 | 3968 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
1da177e4 LT |
3969 | out: |
3970 | return err; | |
3971 | } | |
3972 | ||
fdee8ae4 | 3973 | #ifndef MODULE |
1da177e4 LT |
3974 | static void autorun_array(mddev_t *mddev) |
3975 | { | |
3976 | mdk_rdev_t *rdev; | |
3977 | struct list_head *tmp; | |
3978 | int err; | |
3979 | ||
a757e64c | 3980 | if (list_empty(&mddev->disks)) |
1da177e4 | 3981 | return; |
1da177e4 LT |
3982 | |
3983 | printk(KERN_INFO "md: running: "); | |
3984 | ||
d089c6af | 3985 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
3986 | char b[BDEVNAME_SIZE]; |
3987 | printk("<%s>", bdevname(rdev->bdev,b)); | |
3988 | } | |
3989 | printk("\n"); | |
3990 | ||
3991 | err = do_md_run (mddev); | |
3992 | if (err) { | |
3993 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
3994 | do_md_stop (mddev, 0); | |
3995 | } | |
3996 | } | |
3997 | ||
3998 | /* | |
3999 | * lets try to run arrays based on all disks that have arrived | |
4000 | * until now. (those are in pending_raid_disks) | |
4001 | * | |
4002 | * the method: pick the first pending disk, collect all disks with | |
4003 | * the same UUID, remove all from the pending list and put them into | |
4004 | * the 'same_array' list. Then order this list based on superblock | |
4005 | * update time (freshest comes first), kick out 'old' disks and | |
4006 | * compare superblocks. If everything's fine then run it. | |
4007 | * | |
4008 | * If "unit" is allocated, then bump its reference count | |
4009 | */ | |
4010 | static void autorun_devices(int part) | |
4011 | { | |
1da177e4 LT |
4012 | struct list_head *tmp; |
4013 | mdk_rdev_t *rdev0, *rdev; | |
4014 | mddev_t *mddev; | |
4015 | char b[BDEVNAME_SIZE]; | |
4016 | ||
4017 | printk(KERN_INFO "md: autorun ...\n"); | |
4018 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4019 | int unit; |
1da177e4 | 4020 | dev_t dev; |
ad01c9e3 | 4021 | LIST_HEAD(candidates); |
1da177e4 LT |
4022 | rdev0 = list_entry(pending_raid_disks.next, |
4023 | mdk_rdev_t, same_set); | |
4024 | ||
4025 | printk(KERN_INFO "md: considering %s ...\n", | |
4026 | bdevname(rdev0->bdev,b)); | |
4027 | INIT_LIST_HEAD(&candidates); | |
73c34431 | 4028 | rdev_for_each_list(rdev, tmp, pending_raid_disks) |
1da177e4 LT |
4029 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4030 | printk(KERN_INFO "md: adding %s ...\n", | |
4031 | bdevname(rdev->bdev,b)); | |
4032 | list_move(&rdev->same_set, &candidates); | |
4033 | } | |
4034 | /* | |
4035 | * now we have a set of devices, with all of them having | |
4036 | * mostly sane superblocks. It's time to allocate the | |
4037 | * mddev. | |
4038 | */ | |
e8703fe1 N |
4039 | if (part) { |
4040 | dev = MKDEV(mdp_major, | |
4041 | rdev0->preferred_minor << MdpMinorShift); | |
4042 | unit = MINOR(dev) >> MdpMinorShift; | |
4043 | } else { | |
4044 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4045 | unit = MINOR(dev); | |
4046 | } | |
4047 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4048 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4049 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4050 | break; | |
4051 | } | |
1da177e4 LT |
4052 | |
4053 | md_probe(dev, NULL, NULL); | |
4054 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4055 | if (!mddev || !mddev->gendisk) { |
4056 | if (mddev) | |
4057 | mddev_put(mddev); | |
4058 | printk(KERN_ERR | |
1da177e4 LT |
4059 | "md: cannot allocate memory for md drive.\n"); |
4060 | break; | |
4061 | } | |
4062 | if (mddev_lock(mddev)) | |
4063 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4064 | mdname(mddev)); | |
4065 | else if (mddev->raid_disks || mddev->major_version | |
4066 | || !list_empty(&mddev->disks)) { | |
4067 | printk(KERN_WARNING | |
4068 | "md: %s already running, cannot run %s\n", | |
4069 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4070 | mddev_unlock(mddev); | |
4071 | } else { | |
4072 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4073 | mddev->persistent = 1; |
73c34431 | 4074 | rdev_for_each_list(rdev, tmp, candidates) { |
1da177e4 LT |
4075 | list_del_init(&rdev->same_set); |
4076 | if (bind_rdev_to_array(rdev, mddev)) | |
4077 | export_rdev(rdev); | |
4078 | } | |
4079 | autorun_array(mddev); | |
4080 | mddev_unlock(mddev); | |
4081 | } | |
4082 | /* on success, candidates will be empty, on error | |
4083 | * it won't... | |
4084 | */ | |
73c34431 | 4085 | rdev_for_each_list(rdev, tmp, candidates) |
1da177e4 LT |
4086 | export_rdev(rdev); |
4087 | mddev_put(mddev); | |
4088 | } | |
4089 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4090 | } | |
fdee8ae4 | 4091 | #endif /* !MODULE */ |
1da177e4 | 4092 | |
1da177e4 LT |
4093 | static int get_version(void __user * arg) |
4094 | { | |
4095 | mdu_version_t ver; | |
4096 | ||
4097 | ver.major = MD_MAJOR_VERSION; | |
4098 | ver.minor = MD_MINOR_VERSION; | |
4099 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4100 | ||
4101 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4102 | return -EFAULT; | |
4103 | ||
4104 | return 0; | |
4105 | } | |
4106 | ||
4107 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4108 | { | |
4109 | mdu_array_info_t info; | |
4110 | int nr,working,active,failed,spare; | |
4111 | mdk_rdev_t *rdev; | |
4112 | struct list_head *tmp; | |
4113 | ||
4114 | nr=working=active=failed=spare=0; | |
d089c6af | 4115 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 | 4116 | nr++; |
b2d444d7 | 4117 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4118 | failed++; |
4119 | else { | |
4120 | working++; | |
b2d444d7 | 4121 | if (test_bit(In_sync, &rdev->flags)) |
1da177e4 LT |
4122 | active++; |
4123 | else | |
4124 | spare++; | |
4125 | } | |
4126 | } | |
4127 | ||
4128 | info.major_version = mddev->major_version; | |
4129 | info.minor_version = mddev->minor_version; | |
4130 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
4131 | info.ctime = mddev->ctime; | |
4132 | info.level = mddev->level; | |
4133 | info.size = mddev->size; | |
284ae7ca N |
4134 | if (info.size != mddev->size) /* overflow */ |
4135 | info.size = -1; | |
1da177e4 LT |
4136 | info.nr_disks = nr; |
4137 | info.raid_disks = mddev->raid_disks; | |
4138 | info.md_minor = mddev->md_minor; | |
4139 | info.not_persistent= !mddev->persistent; | |
4140 | ||
4141 | info.utime = mddev->utime; | |
4142 | info.state = 0; | |
4143 | if (mddev->in_sync) | |
4144 | info.state = (1<<MD_SB_CLEAN); | |
36fa3063 N |
4145 | if (mddev->bitmap && mddev->bitmap_offset) |
4146 | info.state = (1<<MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
4147 | info.active_disks = active; |
4148 | info.working_disks = working; | |
4149 | info.failed_disks = failed; | |
4150 | info.spare_disks = spare; | |
4151 | ||
4152 | info.layout = mddev->layout; | |
4153 | info.chunk_size = mddev->chunk_size; | |
4154 | ||
4155 | if (copy_to_user(arg, &info, sizeof(info))) | |
4156 | return -EFAULT; | |
4157 | ||
4158 | return 0; | |
4159 | } | |
4160 | ||
87162a28 | 4161 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
4162 | { |
4163 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
4164 | char *ptr, *buf = NULL; | |
4165 | int err = -ENOMEM; | |
4166 | ||
b5470dc5 DW |
4167 | if (md_allow_write(mddev)) |
4168 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
4169 | else | |
4170 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 4171 | |
32a7627c N |
4172 | if (!file) |
4173 | goto out; | |
4174 | ||
4175 | /* bitmap disabled, zero the first byte and copy out */ | |
4176 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
4177 | file->pathname[0] = '\0'; | |
4178 | goto copy_out; | |
4179 | } | |
4180 | ||
4181 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
4182 | if (!buf) | |
4183 | goto out; | |
4184 | ||
6bcfd601 CH |
4185 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
4186 | if (IS_ERR(ptr)) | |
32a7627c N |
4187 | goto out; |
4188 | ||
4189 | strcpy(file->pathname, ptr); | |
4190 | ||
4191 | copy_out: | |
4192 | err = 0; | |
4193 | if (copy_to_user(arg, file, sizeof(*file))) | |
4194 | err = -EFAULT; | |
4195 | out: | |
4196 | kfree(buf); | |
4197 | kfree(file); | |
4198 | return err; | |
4199 | } | |
4200 | ||
1da177e4 LT |
4201 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
4202 | { | |
4203 | mdu_disk_info_t info; | |
4204 | unsigned int nr; | |
4205 | mdk_rdev_t *rdev; | |
4206 | ||
4207 | if (copy_from_user(&info, arg, sizeof(info))) | |
4208 | return -EFAULT; | |
4209 | ||
4210 | nr = info.number; | |
4211 | ||
4212 | rdev = find_rdev_nr(mddev, nr); | |
4213 | if (rdev) { | |
4214 | info.major = MAJOR(rdev->bdev->bd_dev); | |
4215 | info.minor = MINOR(rdev->bdev->bd_dev); | |
4216 | info.raid_disk = rdev->raid_disk; | |
4217 | info.state = 0; | |
b2d444d7 | 4218 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4219 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 4220 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4221 | info.state |= (1<<MD_DISK_ACTIVE); |
4222 | info.state |= (1<<MD_DISK_SYNC); | |
4223 | } | |
8ddf9efe N |
4224 | if (test_bit(WriteMostly, &rdev->flags)) |
4225 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
4226 | } else { |
4227 | info.major = info.minor = 0; | |
4228 | info.raid_disk = -1; | |
4229 | info.state = (1<<MD_DISK_REMOVED); | |
4230 | } | |
4231 | ||
4232 | if (copy_to_user(arg, &info, sizeof(info))) | |
4233 | return -EFAULT; | |
4234 | ||
4235 | return 0; | |
4236 | } | |
4237 | ||
4238 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
4239 | { | |
4240 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4241 | mdk_rdev_t *rdev; | |
4242 | dev_t dev = MKDEV(info->major,info->minor); | |
4243 | ||
4244 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
4245 | return -EOVERFLOW; | |
4246 | ||
4247 | if (!mddev->raid_disks) { | |
4248 | int err; | |
4249 | /* expecting a device which has a superblock */ | |
4250 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
4251 | if (IS_ERR(rdev)) { | |
4252 | printk(KERN_WARNING | |
4253 | "md: md_import_device returned %ld\n", | |
4254 | PTR_ERR(rdev)); | |
4255 | return PTR_ERR(rdev); | |
4256 | } | |
4257 | if (!list_empty(&mddev->disks)) { | |
4258 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
4259 | mdk_rdev_t, same_set); | |
4260 | int err = super_types[mddev->major_version] | |
4261 | .load_super(rdev, rdev0, mddev->minor_version); | |
4262 | if (err < 0) { | |
4263 | printk(KERN_WARNING | |
4264 | "md: %s has different UUID to %s\n", | |
4265 | bdevname(rdev->bdev,b), | |
4266 | bdevname(rdev0->bdev,b2)); | |
4267 | export_rdev(rdev); | |
4268 | return -EINVAL; | |
4269 | } | |
4270 | } | |
4271 | err = bind_rdev_to_array(rdev, mddev); | |
4272 | if (err) | |
4273 | export_rdev(rdev); | |
4274 | return err; | |
4275 | } | |
4276 | ||
4277 | /* | |
4278 | * add_new_disk can be used once the array is assembled | |
4279 | * to add "hot spares". They must already have a superblock | |
4280 | * written | |
4281 | */ | |
4282 | if (mddev->pers) { | |
4283 | int err; | |
4284 | if (!mddev->pers->hot_add_disk) { | |
4285 | printk(KERN_WARNING | |
4286 | "%s: personality does not support diskops!\n", | |
4287 | mdname(mddev)); | |
4288 | return -EINVAL; | |
4289 | } | |
7b1e35f6 N |
4290 | if (mddev->persistent) |
4291 | rdev = md_import_device(dev, mddev->major_version, | |
4292 | mddev->minor_version); | |
4293 | else | |
4294 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
4295 | if (IS_ERR(rdev)) { |
4296 | printk(KERN_WARNING | |
4297 | "md: md_import_device returned %ld\n", | |
4298 | PTR_ERR(rdev)); | |
4299 | return PTR_ERR(rdev); | |
4300 | } | |
41158c7e N |
4301 | /* set save_raid_disk if appropriate */ |
4302 | if (!mddev->persistent) { | |
4303 | if (info->state & (1<<MD_DISK_SYNC) && | |
4304 | info->raid_disk < mddev->raid_disks) | |
4305 | rdev->raid_disk = info->raid_disk; | |
4306 | else | |
4307 | rdev->raid_disk = -1; | |
4308 | } else | |
4309 | super_types[mddev->major_version]. | |
4310 | validate_super(mddev, rdev); | |
4311 | rdev->saved_raid_disk = rdev->raid_disk; | |
4312 | ||
b2d444d7 | 4313 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
4314 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4315 | set_bit(WriteMostly, &rdev->flags); | |
4316 | ||
1da177e4 LT |
4317 | rdev->raid_disk = -1; |
4318 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
4319 | if (!err && !mddev->pers->hot_remove_disk) { |
4320 | /* If there is hot_add_disk but no hot_remove_disk | |
4321 | * then added disks for geometry changes, | |
4322 | * and should be added immediately. | |
4323 | */ | |
4324 | super_types[mddev->major_version]. | |
4325 | validate_super(mddev, rdev); | |
4326 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
4327 | if (err) | |
4328 | unbind_rdev_from_array(rdev); | |
4329 | } | |
1da177e4 LT |
4330 | if (err) |
4331 | export_rdev(rdev); | |
52664732 NB |
4332 | else |
4333 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
c361777f | 4334 | |
17571284 | 4335 | md_update_sb(mddev, 1); |
72a23c21 NB |
4336 | if (mddev->degraded) |
4337 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 4338 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 4339 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4340 | return err; |
4341 | } | |
4342 | ||
4343 | /* otherwise, add_new_disk is only allowed | |
4344 | * for major_version==0 superblocks | |
4345 | */ | |
4346 | if (mddev->major_version != 0) { | |
4347 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
4348 | mdname(mddev)); | |
4349 | return -EINVAL; | |
4350 | } | |
4351 | ||
4352 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
4353 | int err; | |
4354 | rdev = md_import_device (dev, -1, 0); | |
4355 | if (IS_ERR(rdev)) { | |
4356 | printk(KERN_WARNING | |
4357 | "md: error, md_import_device() returned %ld\n", | |
4358 | PTR_ERR(rdev)); | |
4359 | return PTR_ERR(rdev); | |
4360 | } | |
4361 | rdev->desc_nr = info->number; | |
4362 | if (info->raid_disk < mddev->raid_disks) | |
4363 | rdev->raid_disk = info->raid_disk; | |
4364 | else | |
4365 | rdev->raid_disk = -1; | |
4366 | ||
1da177e4 | 4367 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
4368 | if (info->state & (1<<MD_DISK_SYNC)) |
4369 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 4370 | |
8ddf9efe N |
4371 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4372 | set_bit(WriteMostly, &rdev->flags); | |
4373 | ||
1da177e4 LT |
4374 | if (!mddev->persistent) { |
4375 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
4376 | rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
4377 | } else | |
4378 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
4379 | rdev->size = calc_dev_size(rdev, mddev->chunk_size); | |
4380 | ||
2bf071bf N |
4381 | err = bind_rdev_to_array(rdev, mddev); |
4382 | if (err) { | |
4383 | export_rdev(rdev); | |
4384 | return err; | |
4385 | } | |
1da177e4 LT |
4386 | } |
4387 | ||
4388 | return 0; | |
4389 | } | |
4390 | ||
4391 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
4392 | { | |
4393 | char b[BDEVNAME_SIZE]; | |
4394 | mdk_rdev_t *rdev; | |
4395 | ||
1da177e4 LT |
4396 | rdev = find_rdev(mddev, dev); |
4397 | if (!rdev) | |
4398 | return -ENXIO; | |
4399 | ||
4400 | if (rdev->raid_disk >= 0) | |
4401 | goto busy; | |
4402 | ||
4403 | kick_rdev_from_array(rdev); | |
850b2b42 | 4404 | md_update_sb(mddev, 1); |
d7603b7e | 4405 | md_new_event(mddev); |
1da177e4 LT |
4406 | |
4407 | return 0; | |
4408 | busy: | |
fdefa4d8 | 4409 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
4410 | bdevname(rdev->bdev,b), mdname(mddev)); |
4411 | return -EBUSY; | |
4412 | } | |
4413 | ||
4414 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
4415 | { | |
4416 | char b[BDEVNAME_SIZE]; | |
4417 | int err; | |
4418 | unsigned int size; | |
4419 | mdk_rdev_t *rdev; | |
4420 | ||
4421 | if (!mddev->pers) | |
4422 | return -ENODEV; | |
4423 | ||
4424 | if (mddev->major_version != 0) { | |
4425 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
4426 | " version-0 superblocks.\n", | |
4427 | mdname(mddev)); | |
4428 | return -EINVAL; | |
4429 | } | |
4430 | if (!mddev->pers->hot_add_disk) { | |
4431 | printk(KERN_WARNING | |
4432 | "%s: personality does not support diskops!\n", | |
4433 | mdname(mddev)); | |
4434 | return -EINVAL; | |
4435 | } | |
4436 | ||
4437 | rdev = md_import_device (dev, -1, 0); | |
4438 | if (IS_ERR(rdev)) { | |
4439 | printk(KERN_WARNING | |
4440 | "md: error, md_import_device() returned %ld\n", | |
4441 | PTR_ERR(rdev)); | |
4442 | return -EINVAL; | |
4443 | } | |
4444 | ||
4445 | if (mddev->persistent) | |
4446 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
4447 | else | |
4448 | rdev->sb_offset = | |
4449 | rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
4450 | ||
4451 | size = calc_dev_size(rdev, mddev->chunk_size); | |
4452 | rdev->size = size; | |
4453 | ||
b2d444d7 | 4454 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
4455 | printk(KERN_WARNING |
4456 | "md: can not hot-add faulty %s disk to %s!\n", | |
4457 | bdevname(rdev->bdev,b), mdname(mddev)); | |
4458 | err = -EINVAL; | |
4459 | goto abort_export; | |
4460 | } | |
b2d444d7 | 4461 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4462 | rdev->desc_nr = -1; |
5842730d | 4463 | rdev->saved_raid_disk = -1; |
2bf071bf N |
4464 | err = bind_rdev_to_array(rdev, mddev); |
4465 | if (err) | |
4466 | goto abort_export; | |
1da177e4 LT |
4467 | |
4468 | /* | |
4469 | * The rest should better be atomic, we can have disk failures | |
4470 | * noticed in interrupt contexts ... | |
4471 | */ | |
4472 | ||
4473 | if (rdev->desc_nr == mddev->max_disks) { | |
4474 | printk(KERN_WARNING "%s: can not hot-add to full array!\n", | |
4475 | mdname(mddev)); | |
4476 | err = -EBUSY; | |
4477 | goto abort_unbind_export; | |
4478 | } | |
4479 | ||
4480 | rdev->raid_disk = -1; | |
4481 | ||
850b2b42 | 4482 | md_update_sb(mddev, 1); |
1da177e4 LT |
4483 | |
4484 | /* | |
4485 | * Kick recovery, maybe this spare has to be added to the | |
4486 | * array immediately. | |
4487 | */ | |
4488 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4489 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 4490 | md_new_event(mddev); |
1da177e4 LT |
4491 | return 0; |
4492 | ||
4493 | abort_unbind_export: | |
4494 | unbind_rdev_from_array(rdev); | |
4495 | ||
4496 | abort_export: | |
4497 | export_rdev(rdev); | |
4498 | return err; | |
4499 | } | |
4500 | ||
32a7627c N |
4501 | static int set_bitmap_file(mddev_t *mddev, int fd) |
4502 | { | |
4503 | int err; | |
4504 | ||
36fa3063 N |
4505 | if (mddev->pers) { |
4506 | if (!mddev->pers->quiesce) | |
4507 | return -EBUSY; | |
4508 | if (mddev->recovery || mddev->sync_thread) | |
4509 | return -EBUSY; | |
4510 | /* we should be able to change the bitmap.. */ | |
4511 | } | |
32a7627c | 4512 | |
32a7627c | 4513 | |
36fa3063 N |
4514 | if (fd >= 0) { |
4515 | if (mddev->bitmap) | |
4516 | return -EEXIST; /* cannot add when bitmap is present */ | |
4517 | mddev->bitmap_file = fget(fd); | |
32a7627c | 4518 | |
36fa3063 N |
4519 | if (mddev->bitmap_file == NULL) { |
4520 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", | |
4521 | mdname(mddev)); | |
4522 | return -EBADF; | |
4523 | } | |
4524 | ||
4525 | err = deny_bitmap_write_access(mddev->bitmap_file); | |
4526 | if (err) { | |
4527 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
4528 | mdname(mddev)); | |
4529 | fput(mddev->bitmap_file); | |
4530 | mddev->bitmap_file = NULL; | |
4531 | return err; | |
4532 | } | |
a654b9d8 | 4533 | mddev->bitmap_offset = 0; /* file overrides offset */ |
36fa3063 N |
4534 | } else if (mddev->bitmap == NULL) |
4535 | return -ENOENT; /* cannot remove what isn't there */ | |
4536 | err = 0; | |
4537 | if (mddev->pers) { | |
4538 | mddev->pers->quiesce(mddev, 1); | |
4539 | if (fd >= 0) | |
4540 | err = bitmap_create(mddev); | |
d7375ab3 | 4541 | if (fd < 0 || err) { |
36fa3063 | 4542 | bitmap_destroy(mddev); |
d7375ab3 N |
4543 | fd = -1; /* make sure to put the file */ |
4544 | } | |
36fa3063 | 4545 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
4546 | } |
4547 | if (fd < 0) { | |
acc55e22 N |
4548 | if (mddev->bitmap_file) { |
4549 | restore_bitmap_write_access(mddev->bitmap_file); | |
36fa3063 | 4550 | fput(mddev->bitmap_file); |
acc55e22 | 4551 | } |
36fa3063 N |
4552 | mddev->bitmap_file = NULL; |
4553 | } | |
4554 | ||
32a7627c N |
4555 | return err; |
4556 | } | |
4557 | ||
1da177e4 LT |
4558 | /* |
4559 | * set_array_info is used two different ways | |
4560 | * The original usage is when creating a new array. | |
4561 | * In this usage, raid_disks is > 0 and it together with | |
4562 | * level, size, not_persistent,layout,chunksize determine the | |
4563 | * shape of the array. | |
4564 | * This will always create an array with a type-0.90.0 superblock. | |
4565 | * The newer usage is when assembling an array. | |
4566 | * In this case raid_disks will be 0, and the major_version field is | |
4567 | * use to determine which style super-blocks are to be found on the devices. | |
4568 | * The minor and patch _version numbers are also kept incase the | |
4569 | * super_block handler wishes to interpret them. | |
4570 | */ | |
4571 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
4572 | { | |
4573 | ||
4574 | if (info->raid_disks == 0) { | |
4575 | /* just setting version number for superblock loading */ | |
4576 | if (info->major_version < 0 || | |
50511da3 | 4577 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
4578 | super_types[info->major_version].name == NULL) { |
4579 | /* maybe try to auto-load a module? */ | |
4580 | printk(KERN_INFO | |
4581 | "md: superblock version %d not known\n", | |
4582 | info->major_version); | |
4583 | return -EINVAL; | |
4584 | } | |
4585 | mddev->major_version = info->major_version; | |
4586 | mddev->minor_version = info->minor_version; | |
4587 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 4588 | mddev->persistent = !info->not_persistent; |
1da177e4 LT |
4589 | return 0; |
4590 | } | |
4591 | mddev->major_version = MD_MAJOR_VERSION; | |
4592 | mddev->minor_version = MD_MINOR_VERSION; | |
4593 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
4594 | mddev->ctime = get_seconds(); | |
4595 | ||
4596 | mddev->level = info->level; | |
17115e03 | 4597 | mddev->clevel[0] = 0; |
1da177e4 LT |
4598 | mddev->size = info->size; |
4599 | mddev->raid_disks = info->raid_disks; | |
4600 | /* don't set md_minor, it is determined by which /dev/md* was | |
4601 | * openned | |
4602 | */ | |
4603 | if (info->state & (1<<MD_SB_CLEAN)) | |
4604 | mddev->recovery_cp = MaxSector; | |
4605 | else | |
4606 | mddev->recovery_cp = 0; | |
4607 | mddev->persistent = ! info->not_persistent; | |
e691063a | 4608 | mddev->external = 0; |
1da177e4 LT |
4609 | |
4610 | mddev->layout = info->layout; | |
4611 | mddev->chunk_size = info->chunk_size; | |
4612 | ||
4613 | mddev->max_disks = MD_SB_DISKS; | |
4614 | ||
e691063a N |
4615 | if (mddev->persistent) |
4616 | mddev->flags = 0; | |
850b2b42 | 4617 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 4618 | |
b2a2703c N |
4619 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
4620 | mddev->bitmap_offset = 0; | |
4621 | ||
f6705578 N |
4622 | mddev->reshape_position = MaxSector; |
4623 | ||
1da177e4 LT |
4624 | /* |
4625 | * Generate a 128 bit UUID | |
4626 | */ | |
4627 | get_random_bytes(mddev->uuid, 16); | |
4628 | ||
f6705578 N |
4629 | mddev->new_level = mddev->level; |
4630 | mddev->new_chunk = mddev->chunk_size; | |
4631 | mddev->new_layout = mddev->layout; | |
4632 | mddev->delta_disks = 0; | |
4633 | ||
1da177e4 LT |
4634 | return 0; |
4635 | } | |
4636 | ||
a35b0d69 N |
4637 | static int update_size(mddev_t *mddev, unsigned long size) |
4638 | { | |
4639 | mdk_rdev_t * rdev; | |
4640 | int rv; | |
4641 | struct list_head *tmp; | |
8ddeeae5 | 4642 | int fit = (size == 0); |
a35b0d69 N |
4643 | |
4644 | if (mddev->pers->resize == NULL) | |
4645 | return -EINVAL; | |
4646 | /* The "size" is the amount of each device that is used. | |
4647 | * This can only make sense for arrays with redundancy. | |
4648 | * linear and raid0 always use whatever space is available | |
4649 | * We can only consider changing the size if no resync | |
4650 | * or reconstruction is happening, and if the new size | |
4651 | * is acceptable. It must fit before the sb_offset or, | |
4652 | * if that is <data_offset, it must fit before the | |
4653 | * size of each device. | |
4654 | * If size is zero, we find the largest size that fits. | |
4655 | */ | |
4656 | if (mddev->sync_thread) | |
4657 | return -EBUSY; | |
d089c6af | 4658 | rdev_for_each(rdev, tmp, mddev) { |
a35b0d69 | 4659 | sector_t avail; |
01ab5662 N |
4660 | avail = rdev->size * 2; |
4661 | ||
a35b0d69 N |
4662 | if (fit && (size == 0 || size > avail/2)) |
4663 | size = avail/2; | |
4664 | if (avail < ((sector_t)size << 1)) | |
4665 | return -ENOSPC; | |
4666 | } | |
4667 | rv = mddev->pers->resize(mddev, (sector_t)size *2); | |
4668 | if (!rv) { | |
4669 | struct block_device *bdev; | |
4670 | ||
4671 | bdev = bdget_disk(mddev->gendisk, 0); | |
4672 | if (bdev) { | |
1b1dcc1b | 4673 | mutex_lock(&bdev->bd_inode->i_mutex); |
6d89332b | 4674 | i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10); |
1b1dcc1b | 4675 | mutex_unlock(&bdev->bd_inode->i_mutex); |
a35b0d69 N |
4676 | bdput(bdev); |
4677 | } | |
4678 | } | |
4679 | return rv; | |
4680 | } | |
4681 | ||
da943b99 N |
4682 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
4683 | { | |
4684 | int rv; | |
4685 | /* change the number of raid disks */ | |
63c70c4f | 4686 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
4687 | return -EINVAL; |
4688 | if (raid_disks <= 0 || | |
4689 | raid_disks >= mddev->max_disks) | |
4690 | return -EINVAL; | |
63c70c4f | 4691 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 4692 | return -EBUSY; |
63c70c4f N |
4693 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
4694 | ||
4695 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
4696 | return rv; |
4697 | } | |
4698 | ||
4699 | ||
1da177e4 LT |
4700 | /* |
4701 | * update_array_info is used to change the configuration of an | |
4702 | * on-line array. | |
4703 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
4704 | * fields in the info are checked against the array. | |
4705 | * Any differences that cannot be handled will cause an error. | |
4706 | * Normally, only one change can be managed at a time. | |
4707 | */ | |
4708 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
4709 | { | |
4710 | int rv = 0; | |
4711 | int cnt = 0; | |
36fa3063 N |
4712 | int state = 0; |
4713 | ||
4714 | /* calculate expected state,ignoring low bits */ | |
4715 | if (mddev->bitmap && mddev->bitmap_offset) | |
4716 | state |= (1 << MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
4717 | |
4718 | if (mddev->major_version != info->major_version || | |
4719 | mddev->minor_version != info->minor_version || | |
4720 | /* mddev->patch_version != info->patch_version || */ | |
4721 | mddev->ctime != info->ctime || | |
4722 | mddev->level != info->level || | |
4723 | /* mddev->layout != info->layout || */ | |
4724 | !mddev->persistent != info->not_persistent|| | |
36fa3063 N |
4725 | mddev->chunk_size != info->chunk_size || |
4726 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ | |
4727 | ((state^info->state) & 0xfffffe00) | |
4728 | ) | |
1da177e4 LT |
4729 | return -EINVAL; |
4730 | /* Check there is only one change */ | |
284ae7ca | 4731 | if (info->size >= 0 && mddev->size != info->size) cnt++; |
1da177e4 LT |
4732 | if (mddev->raid_disks != info->raid_disks) cnt++; |
4733 | if (mddev->layout != info->layout) cnt++; | |
36fa3063 | 4734 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++; |
1da177e4 LT |
4735 | if (cnt == 0) return 0; |
4736 | if (cnt > 1) return -EINVAL; | |
4737 | ||
4738 | if (mddev->layout != info->layout) { | |
4739 | /* Change layout | |
4740 | * we don't need to do anything at the md level, the | |
4741 | * personality will take care of it all. | |
4742 | */ | |
4743 | if (mddev->pers->reconfig == NULL) | |
4744 | return -EINVAL; | |
4745 | else | |
4746 | return mddev->pers->reconfig(mddev, info->layout, -1); | |
4747 | } | |
284ae7ca | 4748 | if (info->size >= 0 && mddev->size != info->size) |
a35b0d69 N |
4749 | rv = update_size(mddev, info->size); |
4750 | ||
da943b99 N |
4751 | if (mddev->raid_disks != info->raid_disks) |
4752 | rv = update_raid_disks(mddev, info->raid_disks); | |
4753 | ||
36fa3063 N |
4754 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
4755 | if (mddev->pers->quiesce == NULL) | |
4756 | return -EINVAL; | |
4757 | if (mddev->recovery || mddev->sync_thread) | |
4758 | return -EBUSY; | |
4759 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
4760 | /* add the bitmap */ | |
4761 | if (mddev->bitmap) | |
4762 | return -EEXIST; | |
4763 | if (mddev->default_bitmap_offset == 0) | |
4764 | return -EINVAL; | |
4765 | mddev->bitmap_offset = mddev->default_bitmap_offset; | |
4766 | mddev->pers->quiesce(mddev, 1); | |
4767 | rv = bitmap_create(mddev); | |
4768 | if (rv) | |
4769 | bitmap_destroy(mddev); | |
4770 | mddev->pers->quiesce(mddev, 0); | |
4771 | } else { | |
4772 | /* remove the bitmap */ | |
4773 | if (!mddev->bitmap) | |
4774 | return -ENOENT; | |
4775 | if (mddev->bitmap->file) | |
4776 | return -EINVAL; | |
4777 | mddev->pers->quiesce(mddev, 1); | |
4778 | bitmap_destroy(mddev); | |
4779 | mddev->pers->quiesce(mddev, 0); | |
4780 | mddev->bitmap_offset = 0; | |
4781 | } | |
4782 | } | |
850b2b42 | 4783 | md_update_sb(mddev, 1); |
1da177e4 LT |
4784 | return rv; |
4785 | } | |
4786 | ||
4787 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
4788 | { | |
4789 | mdk_rdev_t *rdev; | |
4790 | ||
4791 | if (mddev->pers == NULL) | |
4792 | return -ENODEV; | |
4793 | ||
4794 | rdev = find_rdev(mddev, dev); | |
4795 | if (!rdev) | |
4796 | return -ENODEV; | |
4797 | ||
4798 | md_error(mddev, rdev); | |
4799 | return 0; | |
4800 | } | |
4801 | ||
2f9618ce AN |
4802 | /* |
4803 | * We have a problem here : there is no easy way to give a CHS | |
4804 | * virtual geometry. We currently pretend that we have a 2 heads | |
4805 | * 4 sectors (with a BIG number of cylinders...). This drives | |
4806 | * dosfs just mad... ;-) | |
4807 | */ | |
a885c8c4 CH |
4808 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
4809 | { | |
4810 | mddev_t *mddev = bdev->bd_disk->private_data; | |
4811 | ||
4812 | geo->heads = 2; | |
4813 | geo->sectors = 4; | |
4814 | geo->cylinders = get_capacity(mddev->gendisk) / 8; | |
4815 | return 0; | |
4816 | } | |
4817 | ||
1da177e4 LT |
4818 | static int md_ioctl(struct inode *inode, struct file *file, |
4819 | unsigned int cmd, unsigned long arg) | |
4820 | { | |
4821 | int err = 0; | |
4822 | void __user *argp = (void __user *)arg; | |
1da177e4 LT |
4823 | mddev_t *mddev = NULL; |
4824 | ||
4825 | if (!capable(CAP_SYS_ADMIN)) | |
4826 | return -EACCES; | |
4827 | ||
4828 | /* | |
4829 | * Commands dealing with the RAID driver but not any | |
4830 | * particular array: | |
4831 | */ | |
4832 | switch (cmd) | |
4833 | { | |
4834 | case RAID_VERSION: | |
4835 | err = get_version(argp); | |
4836 | goto done; | |
4837 | ||
4838 | case PRINT_RAID_DEBUG: | |
4839 | err = 0; | |
4840 | md_print_devices(); | |
4841 | goto done; | |
4842 | ||
4843 | #ifndef MODULE | |
4844 | case RAID_AUTORUN: | |
4845 | err = 0; | |
4846 | autostart_arrays(arg); | |
4847 | goto done; | |
4848 | #endif | |
4849 | default:; | |
4850 | } | |
4851 | ||
4852 | /* | |
4853 | * Commands creating/starting a new array: | |
4854 | */ | |
4855 | ||
4856 | mddev = inode->i_bdev->bd_disk->private_data; | |
4857 | ||
4858 | if (!mddev) { | |
4859 | BUG(); | |
4860 | goto abort; | |
4861 | } | |
4862 | ||
1da177e4 LT |
4863 | err = mddev_lock(mddev); |
4864 | if (err) { | |
4865 | printk(KERN_INFO | |
4866 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
4867 | err, cmd); | |
4868 | goto abort; | |
4869 | } | |
4870 | ||
4871 | switch (cmd) | |
4872 | { | |
4873 | case SET_ARRAY_INFO: | |
4874 | { | |
4875 | mdu_array_info_t info; | |
4876 | if (!arg) | |
4877 | memset(&info, 0, sizeof(info)); | |
4878 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
4879 | err = -EFAULT; | |
4880 | goto abort_unlock; | |
4881 | } | |
4882 | if (mddev->pers) { | |
4883 | err = update_array_info(mddev, &info); | |
4884 | if (err) { | |
4885 | printk(KERN_WARNING "md: couldn't update" | |
4886 | " array info. %d\n", err); | |
4887 | goto abort_unlock; | |
4888 | } | |
4889 | goto done_unlock; | |
4890 | } | |
4891 | if (!list_empty(&mddev->disks)) { | |
4892 | printk(KERN_WARNING | |
4893 | "md: array %s already has disks!\n", | |
4894 | mdname(mddev)); | |
4895 | err = -EBUSY; | |
4896 | goto abort_unlock; | |
4897 | } | |
4898 | if (mddev->raid_disks) { | |
4899 | printk(KERN_WARNING | |
4900 | "md: array %s already initialised!\n", | |
4901 | mdname(mddev)); | |
4902 | err = -EBUSY; | |
4903 | goto abort_unlock; | |
4904 | } | |
4905 | err = set_array_info(mddev, &info); | |
4906 | if (err) { | |
4907 | printk(KERN_WARNING "md: couldn't set" | |
4908 | " array info. %d\n", err); | |
4909 | goto abort_unlock; | |
4910 | } | |
4911 | } | |
4912 | goto done_unlock; | |
4913 | ||
4914 | default:; | |
4915 | } | |
4916 | ||
4917 | /* | |
4918 | * Commands querying/configuring an existing array: | |
4919 | */ | |
32a7627c | 4920 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 4921 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
4922 | if ((!mddev->raid_disks && !mddev->external) |
4923 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
4924 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
4925 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
4926 | err = -ENODEV; |
4927 | goto abort_unlock; | |
4928 | } | |
4929 | ||
4930 | /* | |
4931 | * Commands even a read-only array can execute: | |
4932 | */ | |
4933 | switch (cmd) | |
4934 | { | |
4935 | case GET_ARRAY_INFO: | |
4936 | err = get_array_info(mddev, argp); | |
4937 | goto done_unlock; | |
4938 | ||
32a7627c | 4939 | case GET_BITMAP_FILE: |
87162a28 | 4940 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
4941 | goto done_unlock; |
4942 | ||
1da177e4 LT |
4943 | case GET_DISK_INFO: |
4944 | err = get_disk_info(mddev, argp); | |
4945 | goto done_unlock; | |
4946 | ||
4947 | case RESTART_ARRAY_RW: | |
4948 | err = restart_array(mddev); | |
4949 | goto done_unlock; | |
4950 | ||
4951 | case STOP_ARRAY: | |
4952 | err = do_md_stop (mddev, 0); | |
4953 | goto done_unlock; | |
4954 | ||
4955 | case STOP_ARRAY_RO: | |
4956 | err = do_md_stop (mddev, 1); | |
4957 | goto done_unlock; | |
4958 | ||
1da177e4 LT |
4959 | } |
4960 | ||
4961 | /* | |
4962 | * The remaining ioctls are changing the state of the | |
f91de92e N |
4963 | * superblock, so we do not allow them on read-only arrays. |
4964 | * However non-MD ioctls (e.g. get-size) will still come through | |
4965 | * here and hit the 'default' below, so only disallow | |
4966 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 4967 | */ |
bb57fc64 | 4968 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
4969 | if (mddev->ro == 2) { |
4970 | mddev->ro = 0; | |
0fd62b86 NB |
4971 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
4972 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4973 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
4974 | } else { |
4975 | err = -EROFS; | |
4976 | goto abort_unlock; | |
4977 | } | |
1da177e4 LT |
4978 | } |
4979 | ||
4980 | switch (cmd) | |
4981 | { | |
4982 | case ADD_NEW_DISK: | |
4983 | { | |
4984 | mdu_disk_info_t info; | |
4985 | if (copy_from_user(&info, argp, sizeof(info))) | |
4986 | err = -EFAULT; | |
4987 | else | |
4988 | err = add_new_disk(mddev, &info); | |
4989 | goto done_unlock; | |
4990 | } | |
4991 | ||
4992 | case HOT_REMOVE_DISK: | |
4993 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
4994 | goto done_unlock; | |
4995 | ||
4996 | case HOT_ADD_DISK: | |
4997 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
4998 | goto done_unlock; | |
4999 | ||
5000 | case SET_DISK_FAULTY: | |
5001 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5002 | goto done_unlock; | |
5003 | ||
5004 | case RUN_ARRAY: | |
5005 | err = do_md_run (mddev); | |
5006 | goto done_unlock; | |
5007 | ||
32a7627c N |
5008 | case SET_BITMAP_FILE: |
5009 | err = set_bitmap_file(mddev, (int)arg); | |
5010 | goto done_unlock; | |
5011 | ||
1da177e4 | 5012 | default: |
1da177e4 LT |
5013 | err = -EINVAL; |
5014 | goto abort_unlock; | |
5015 | } | |
5016 | ||
5017 | done_unlock: | |
5018 | abort_unlock: | |
5019 | mddev_unlock(mddev); | |
5020 | ||
5021 | return err; | |
5022 | done: | |
5023 | if (err) | |
5024 | MD_BUG(); | |
5025 | abort: | |
5026 | return err; | |
5027 | } | |
5028 | ||
5029 | static int md_open(struct inode *inode, struct file *file) | |
5030 | { | |
5031 | /* | |
5032 | * Succeed if we can lock the mddev, which confirms that | |
5033 | * it isn't being stopped right now. | |
5034 | */ | |
5035 | mddev_t *mddev = inode->i_bdev->bd_disk->private_data; | |
5036 | int err; | |
5037 | ||
d63a5a74 | 5038 | if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1))) |
1da177e4 LT |
5039 | goto out; |
5040 | ||
5041 | err = 0; | |
5042 | mddev_get(mddev); | |
5043 | mddev_unlock(mddev); | |
5044 | ||
5045 | check_disk_change(inode->i_bdev); | |
5046 | out: | |
5047 | return err; | |
5048 | } | |
5049 | ||
5050 | static int md_release(struct inode *inode, struct file * file) | |
5051 | { | |
5052 | mddev_t *mddev = inode->i_bdev->bd_disk->private_data; | |
5053 | ||
52e5f9d1 | 5054 | BUG_ON(!mddev); |
1da177e4 LT |
5055 | mddev_put(mddev); |
5056 | ||
5057 | return 0; | |
5058 | } | |
5059 | ||
44ce6294 LT |
5060 | static int md_media_changed(struct gendisk *disk) |
5061 | { | |
5062 | mddev_t *mddev = disk->private_data; | |
5063 | ||
5064 | return mddev->changed; | |
5065 | } | |
5066 | ||
5067 | static int md_revalidate(struct gendisk *disk) | |
5068 | { | |
5069 | mddev_t *mddev = disk->private_data; | |
5070 | ||
5071 | mddev->changed = 0; | |
5072 | return 0; | |
5073 | } | |
1da177e4 LT |
5074 | static struct block_device_operations md_fops = |
5075 | { | |
5076 | .owner = THIS_MODULE, | |
5077 | .open = md_open, | |
5078 | .release = md_release, | |
5079 | .ioctl = md_ioctl, | |
a885c8c4 | 5080 | .getgeo = md_getgeo, |
44ce6294 LT |
5081 | .media_changed = md_media_changed, |
5082 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
5083 | }; |
5084 | ||
75c96f85 | 5085 | static int md_thread(void * arg) |
1da177e4 LT |
5086 | { |
5087 | mdk_thread_t *thread = arg; | |
5088 | ||
1da177e4 LT |
5089 | /* |
5090 | * md_thread is a 'system-thread', it's priority should be very | |
5091 | * high. We avoid resource deadlocks individually in each | |
5092 | * raid personality. (RAID5 does preallocation) We also use RR and | |
5093 | * the very same RT priority as kswapd, thus we will never get | |
5094 | * into a priority inversion deadlock. | |
5095 | * | |
5096 | * we definitely have to have equal or higher priority than | |
5097 | * bdflush, otherwise bdflush will deadlock if there are too | |
5098 | * many dirty RAID5 blocks. | |
5099 | */ | |
1da177e4 | 5100 | |
6985c43f | 5101 | allow_signal(SIGKILL); |
a6fb0934 | 5102 | while (!kthread_should_stop()) { |
1da177e4 | 5103 | |
93588e22 N |
5104 | /* We need to wait INTERRUPTIBLE so that |
5105 | * we don't add to the load-average. | |
5106 | * That means we need to be sure no signals are | |
5107 | * pending | |
5108 | */ | |
5109 | if (signal_pending(current)) | |
5110 | flush_signals(current); | |
5111 | ||
5112 | wait_event_interruptible_timeout | |
5113 | (thread->wqueue, | |
5114 | test_bit(THREAD_WAKEUP, &thread->flags) | |
5115 | || kthread_should_stop(), | |
5116 | thread->timeout); | |
1da177e4 LT |
5117 | |
5118 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
5119 | ||
787453c2 | 5120 | thread->run(thread->mddev); |
1da177e4 | 5121 | } |
a6fb0934 | 5122 | |
1da177e4 LT |
5123 | return 0; |
5124 | } | |
5125 | ||
5126 | void md_wakeup_thread(mdk_thread_t *thread) | |
5127 | { | |
5128 | if (thread) { | |
5129 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
5130 | set_bit(THREAD_WAKEUP, &thread->flags); | |
5131 | wake_up(&thread->wqueue); | |
5132 | } | |
5133 | } | |
5134 | ||
5135 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
5136 | const char *name) | |
5137 | { | |
5138 | mdk_thread_t *thread; | |
1da177e4 | 5139 | |
9ffae0cf | 5140 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
5141 | if (!thread) |
5142 | return NULL; | |
5143 | ||
1da177e4 LT |
5144 | init_waitqueue_head(&thread->wqueue); |
5145 | ||
1da177e4 LT |
5146 | thread->run = run; |
5147 | thread->mddev = mddev; | |
32a7627c | 5148 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
6985c43f | 5149 | thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev)); |
a6fb0934 | 5150 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
5151 | kfree(thread); |
5152 | return NULL; | |
5153 | } | |
1da177e4 LT |
5154 | return thread; |
5155 | } | |
5156 | ||
1da177e4 LT |
5157 | void md_unregister_thread(mdk_thread_t *thread) |
5158 | { | |
ba25f9dc | 5159 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
5160 | |
5161 | kthread_stop(thread->tsk); | |
1da177e4 LT |
5162 | kfree(thread); |
5163 | } | |
5164 | ||
5165 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
5166 | { | |
5167 | if (!mddev) { | |
5168 | MD_BUG(); | |
5169 | return; | |
5170 | } | |
5171 | ||
b2d444d7 | 5172 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5173 | return; |
6bfe0b49 DW |
5174 | |
5175 | if (mddev->external) | |
5176 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 5177 | /* |
1da177e4 LT |
5178 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
5179 | mdname(mddev), | |
5180 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
5181 | __builtin_return_address(0),__builtin_return_address(1), | |
5182 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 5183 | */ |
d0a0a5ee AM |
5184 | if (!mddev->pers) |
5185 | return; | |
1da177e4 LT |
5186 | if (!mddev->pers->error_handler) |
5187 | return; | |
5188 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
5189 | if (mddev->degraded) |
5190 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
52664732 | 5191 | set_bit(StateChanged, &rdev->flags); |
1da177e4 LT |
5192 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
5193 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5194 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 5195 | md_new_event_inintr(mddev); |
1da177e4 LT |
5196 | } |
5197 | ||
5198 | /* seq_file implementation /proc/mdstat */ | |
5199 | ||
5200 | static void status_unused(struct seq_file *seq) | |
5201 | { | |
5202 | int i = 0; | |
5203 | mdk_rdev_t *rdev; | |
5204 | struct list_head *tmp; | |
5205 | ||
5206 | seq_printf(seq, "unused devices: "); | |
5207 | ||
73c34431 | 5208 | rdev_for_each_list(rdev, tmp, pending_raid_disks) { |
1da177e4 LT |
5209 | char b[BDEVNAME_SIZE]; |
5210 | i++; | |
5211 | seq_printf(seq, "%s ", | |
5212 | bdevname(rdev->bdev,b)); | |
5213 | } | |
5214 | if (!i) | |
5215 | seq_printf(seq, "<none>"); | |
5216 | ||
5217 | seq_printf(seq, "\n"); | |
5218 | } | |
5219 | ||
5220 | ||
5221 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
5222 | { | |
4588b42e N |
5223 | sector_t max_blocks, resync, res; |
5224 | unsigned long dt, db, rt; | |
5225 | int scale; | |
5226 | unsigned int per_milli; | |
1da177e4 LT |
5227 | |
5228 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2; | |
5229 | ||
5230 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
5231 | max_blocks = mddev->resync_max_sectors >> 1; | |
5232 | else | |
5233 | max_blocks = mddev->size; | |
5234 | ||
5235 | /* | |
5236 | * Should not happen. | |
5237 | */ | |
5238 | if (!max_blocks) { | |
5239 | MD_BUG(); | |
5240 | return; | |
5241 | } | |
4588b42e N |
5242 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
5243 | * in a sector_t, and (max_blocks>>scale) will fit in a | |
5244 | * u32, as those are the requirements for sector_div. | |
5245 | * Thus 'scale' must be at least 10 | |
5246 | */ | |
5247 | scale = 10; | |
5248 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
5249 | while ( max_blocks/2 > (1ULL<<(scale+32))) | |
5250 | scale++; | |
5251 | } | |
5252 | res = (resync>>scale)*1000; | |
5253 | sector_div(res, (u32)((max_blocks>>scale)+1)); | |
5254 | ||
5255 | per_milli = res; | |
1da177e4 | 5256 | { |
4588b42e | 5257 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
5258 | seq_printf(seq, "["); |
5259 | for (i = 0; i < x; i++) | |
5260 | seq_printf(seq, "="); | |
5261 | seq_printf(seq, ">"); | |
5262 | for (i = 0; i < y; i++) | |
5263 | seq_printf(seq, "."); | |
5264 | seq_printf(seq, "] "); | |
5265 | } | |
4588b42e | 5266 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
5267 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
5268 | "reshape" : | |
61df9d91 N |
5269 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
5270 | "check" : | |
5271 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
5272 | "resync" : "recovery"))), | |
5273 | per_milli/10, per_milli % 10, | |
4588b42e N |
5274 | (unsigned long long) resync, |
5275 | (unsigned long long) max_blocks); | |
1da177e4 LT |
5276 | |
5277 | /* | |
5278 | * We do not want to overflow, so the order of operands and | |
5279 | * the * 100 / 100 trick are important. We do a +1 to be | |
5280 | * safe against division by zero. We only estimate anyway. | |
5281 | * | |
5282 | * dt: time from mark until now | |
5283 | * db: blocks written from mark until now | |
5284 | * rt: remaining time | |
5285 | */ | |
5286 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
5287 | if (!dt) dt++; | |
ff4e8d9a N |
5288 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
5289 | - mddev->resync_mark_cnt; | |
5290 | rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100; | |
1da177e4 LT |
5291 | |
5292 | seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6); | |
5293 | ||
ff4e8d9a | 5294 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
5295 | } |
5296 | ||
5297 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
5298 | { | |
5299 | struct list_head *tmp; | |
5300 | loff_t l = *pos; | |
5301 | mddev_t *mddev; | |
5302 | ||
5303 | if (l >= 0x10000) | |
5304 | return NULL; | |
5305 | if (!l--) | |
5306 | /* header */ | |
5307 | return (void*)1; | |
5308 | ||
5309 | spin_lock(&all_mddevs_lock); | |
5310 | list_for_each(tmp,&all_mddevs) | |
5311 | if (!l--) { | |
5312 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
5313 | mddev_get(mddev); | |
5314 | spin_unlock(&all_mddevs_lock); | |
5315 | return mddev; | |
5316 | } | |
5317 | spin_unlock(&all_mddevs_lock); | |
5318 | if (!l--) | |
5319 | return (void*)2;/* tail */ | |
5320 | return NULL; | |
5321 | } | |
5322 | ||
5323 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
5324 | { | |
5325 | struct list_head *tmp; | |
5326 | mddev_t *next_mddev, *mddev = v; | |
5327 | ||
5328 | ++*pos; | |
5329 | if (v == (void*)2) | |
5330 | return NULL; | |
5331 | ||
5332 | spin_lock(&all_mddevs_lock); | |
5333 | if (v == (void*)1) | |
5334 | tmp = all_mddevs.next; | |
5335 | else | |
5336 | tmp = mddev->all_mddevs.next; | |
5337 | if (tmp != &all_mddevs) | |
5338 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
5339 | else { | |
5340 | next_mddev = (void*)2; | |
5341 | *pos = 0x10000; | |
5342 | } | |
5343 | spin_unlock(&all_mddevs_lock); | |
5344 | ||
5345 | if (v != (void*)1) | |
5346 | mddev_put(mddev); | |
5347 | return next_mddev; | |
5348 | ||
5349 | } | |
5350 | ||
5351 | static void md_seq_stop(struct seq_file *seq, void *v) | |
5352 | { | |
5353 | mddev_t *mddev = v; | |
5354 | ||
5355 | if (mddev && v != (void*)1 && v != (void*)2) | |
5356 | mddev_put(mddev); | |
5357 | } | |
5358 | ||
d7603b7e N |
5359 | struct mdstat_info { |
5360 | int event; | |
5361 | }; | |
5362 | ||
1da177e4 LT |
5363 | static int md_seq_show(struct seq_file *seq, void *v) |
5364 | { | |
5365 | mddev_t *mddev = v; | |
5366 | sector_t size; | |
5367 | struct list_head *tmp2; | |
5368 | mdk_rdev_t *rdev; | |
d7603b7e | 5369 | struct mdstat_info *mi = seq->private; |
32a7627c | 5370 | struct bitmap *bitmap; |
1da177e4 LT |
5371 | |
5372 | if (v == (void*)1) { | |
2604b703 | 5373 | struct mdk_personality *pers; |
1da177e4 LT |
5374 | seq_printf(seq, "Personalities : "); |
5375 | spin_lock(&pers_lock); | |
2604b703 N |
5376 | list_for_each_entry(pers, &pers_list, list) |
5377 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
5378 | |
5379 | spin_unlock(&pers_lock); | |
5380 | seq_printf(seq, "\n"); | |
d7603b7e | 5381 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
5382 | return 0; |
5383 | } | |
5384 | if (v == (void*)2) { | |
5385 | status_unused(seq); | |
5386 | return 0; | |
5387 | } | |
5388 | ||
5dc5cf7d | 5389 | if (mddev_lock(mddev) < 0) |
1da177e4 | 5390 | return -EINTR; |
5dc5cf7d | 5391 | |
1da177e4 LT |
5392 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
5393 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
5394 | mddev->pers ? "" : "in"); | |
5395 | if (mddev->pers) { | |
f91de92e | 5396 | if (mddev->ro==1) |
1da177e4 | 5397 | seq_printf(seq, " (read-only)"); |
f91de92e | 5398 | if (mddev->ro==2) |
52720ae7 | 5399 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
5400 | seq_printf(seq, " %s", mddev->pers->name); |
5401 | } | |
5402 | ||
5403 | size = 0; | |
d089c6af | 5404 | rdev_for_each(rdev, tmp2, mddev) { |
1da177e4 LT |
5405 | char b[BDEVNAME_SIZE]; |
5406 | seq_printf(seq, " %s[%d]", | |
5407 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
5408 | if (test_bit(WriteMostly, &rdev->flags)) |
5409 | seq_printf(seq, "(W)"); | |
b2d444d7 | 5410 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5411 | seq_printf(seq, "(F)"); |
5412 | continue; | |
b325a32e N |
5413 | } else if (rdev->raid_disk < 0) |
5414 | seq_printf(seq, "(S)"); /* spare */ | |
1da177e4 LT |
5415 | size += rdev->size; |
5416 | } | |
5417 | ||
5418 | if (!list_empty(&mddev->disks)) { | |
5419 | if (mddev->pers) | |
5420 | seq_printf(seq, "\n %llu blocks", | |
5421 | (unsigned long long)mddev->array_size); | |
5422 | else | |
5423 | seq_printf(seq, "\n %llu blocks", | |
5424 | (unsigned long long)size); | |
5425 | } | |
1cd6bf19 N |
5426 | if (mddev->persistent) { |
5427 | if (mddev->major_version != 0 || | |
5428 | mddev->minor_version != 90) { | |
5429 | seq_printf(seq," super %d.%d", | |
5430 | mddev->major_version, | |
5431 | mddev->minor_version); | |
5432 | } | |
e691063a N |
5433 | } else if (mddev->external) |
5434 | seq_printf(seq, " super external:%s", | |
5435 | mddev->metadata_type); | |
5436 | else | |
1cd6bf19 | 5437 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
5438 | |
5439 | if (mddev->pers) { | |
5440 | mddev->pers->status (seq, mddev); | |
5441 | seq_printf(seq, "\n "); | |
8e1b39d6 N |
5442 | if (mddev->pers->sync_request) { |
5443 | if (mddev->curr_resync > 2) { | |
5444 | status_resync (seq, mddev); | |
5445 | seq_printf(seq, "\n "); | |
5446 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
5447 | seq_printf(seq, "\tresync=DELAYED\n "); | |
5448 | else if (mddev->recovery_cp < MaxSector) | |
5449 | seq_printf(seq, "\tresync=PENDING\n "); | |
5450 | } | |
32a7627c N |
5451 | } else |
5452 | seq_printf(seq, "\n "); | |
5453 | ||
5454 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
5455 | unsigned long chunk_kb; |
5456 | unsigned long flags; | |
32a7627c N |
5457 | spin_lock_irqsave(&bitmap->lock, flags); |
5458 | chunk_kb = bitmap->chunksize >> 10; | |
5459 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
5460 | "%lu%s chunk", | |
5461 | bitmap->pages - bitmap->missing_pages, | |
5462 | bitmap->pages, | |
5463 | (bitmap->pages - bitmap->missing_pages) | |
5464 | << (PAGE_SHIFT - 10), | |
5465 | chunk_kb ? chunk_kb : bitmap->chunksize, | |
5466 | chunk_kb ? "KB" : "B"); | |
78d742d8 N |
5467 | if (bitmap->file) { |
5468 | seq_printf(seq, ", file: "); | |
c32c2f63 | 5469 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 5470 | } |
78d742d8 | 5471 | |
32a7627c N |
5472 | seq_printf(seq, "\n"); |
5473 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
5474 | } |
5475 | ||
5476 | seq_printf(seq, "\n"); | |
5477 | } | |
5478 | mddev_unlock(mddev); | |
5479 | ||
5480 | return 0; | |
5481 | } | |
5482 | ||
5483 | static struct seq_operations md_seq_ops = { | |
5484 | .start = md_seq_start, | |
5485 | .next = md_seq_next, | |
5486 | .stop = md_seq_stop, | |
5487 | .show = md_seq_show, | |
5488 | }; | |
5489 | ||
5490 | static int md_seq_open(struct inode *inode, struct file *file) | |
5491 | { | |
5492 | int error; | |
d7603b7e N |
5493 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
5494 | if (mi == NULL) | |
5495 | return -ENOMEM; | |
1da177e4 LT |
5496 | |
5497 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
5498 | if (error) |
5499 | kfree(mi); | |
5500 | else { | |
5501 | struct seq_file *p = file->private_data; | |
5502 | p->private = mi; | |
5503 | mi->event = atomic_read(&md_event_count); | |
5504 | } | |
1da177e4 LT |
5505 | return error; |
5506 | } | |
5507 | ||
d7603b7e N |
5508 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
5509 | { | |
5510 | struct seq_file *m = filp->private_data; | |
5511 | struct mdstat_info *mi = m->private; | |
5512 | int mask; | |
5513 | ||
5514 | poll_wait(filp, &md_event_waiters, wait); | |
5515 | ||
5516 | /* always allow read */ | |
5517 | mask = POLLIN | POLLRDNORM; | |
5518 | ||
5519 | if (mi->event != atomic_read(&md_event_count)) | |
5520 | mask |= POLLERR | POLLPRI; | |
5521 | return mask; | |
5522 | } | |
5523 | ||
fa027c2a | 5524 | static const struct file_operations md_seq_fops = { |
e24650c2 | 5525 | .owner = THIS_MODULE, |
1da177e4 LT |
5526 | .open = md_seq_open, |
5527 | .read = seq_read, | |
5528 | .llseek = seq_lseek, | |
c3f94b40 | 5529 | .release = seq_release_private, |
d7603b7e | 5530 | .poll = mdstat_poll, |
1da177e4 LT |
5531 | }; |
5532 | ||
2604b703 | 5533 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 5534 | { |
1da177e4 | 5535 | spin_lock(&pers_lock); |
2604b703 N |
5536 | list_add_tail(&p->list, &pers_list); |
5537 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
5538 | spin_unlock(&pers_lock); |
5539 | return 0; | |
5540 | } | |
5541 | ||
2604b703 | 5542 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 5543 | { |
2604b703 | 5544 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 5545 | spin_lock(&pers_lock); |
2604b703 | 5546 | list_del_init(&p->list); |
1da177e4 LT |
5547 | spin_unlock(&pers_lock); |
5548 | return 0; | |
5549 | } | |
5550 | ||
5551 | static int is_mddev_idle(mddev_t *mddev) | |
5552 | { | |
5553 | mdk_rdev_t * rdev; | |
5554 | struct list_head *tmp; | |
5555 | int idle; | |
713f6ab1 | 5556 | long curr_events; |
1da177e4 LT |
5557 | |
5558 | idle = 1; | |
d089c6af | 5559 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 | 5560 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
a362357b JA |
5561 | curr_events = disk_stat_read(disk, sectors[0]) + |
5562 | disk_stat_read(disk, sectors[1]) - | |
1da177e4 | 5563 | atomic_read(&disk->sync_io); |
713f6ab1 N |
5564 | /* sync IO will cause sync_io to increase before the disk_stats |
5565 | * as sync_io is counted when a request starts, and | |
5566 | * disk_stats is counted when it completes. | |
5567 | * So resync activity will cause curr_events to be smaller than | |
5568 | * when there was no such activity. | |
5569 | * non-sync IO will cause disk_stat to increase without | |
5570 | * increasing sync_io so curr_events will (eventually) | |
5571 | * be larger than it was before. Once it becomes | |
5572 | * substantially larger, the test below will cause | |
5573 | * the array to appear non-idle, and resync will slow | |
5574 | * down. | |
5575 | * If there is a lot of outstanding resync activity when | |
5576 | * we set last_event to curr_events, then all that activity | |
5577 | * completing might cause the array to appear non-idle | |
5578 | * and resync will be slowed down even though there might | |
5579 | * not have been non-resync activity. This will only | |
5580 | * happen once though. 'last_events' will soon reflect | |
5581 | * the state where there is little or no outstanding | |
5582 | * resync requests, and further resync activity will | |
5583 | * always make curr_events less than last_events. | |
c0e48521 | 5584 | * |
1da177e4 | 5585 | */ |
713f6ab1 | 5586 | if (curr_events - rdev->last_events > 4096) { |
1da177e4 LT |
5587 | rdev->last_events = curr_events; |
5588 | idle = 0; | |
5589 | } | |
5590 | } | |
5591 | return idle; | |
5592 | } | |
5593 | ||
5594 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
5595 | { | |
5596 | /* another "blocks" (512byte) blocks have been synced */ | |
5597 | atomic_sub(blocks, &mddev->recovery_active); | |
5598 | wake_up(&mddev->recovery_wait); | |
5599 | if (!ok) { | |
dfc70645 | 5600 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
5601 | md_wakeup_thread(mddev->thread); |
5602 | // stop recovery, signal do_sync .... | |
5603 | } | |
5604 | } | |
5605 | ||
5606 | ||
06d91a5f N |
5607 | /* md_write_start(mddev, bi) |
5608 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
5609 | * in superblock) before writing, schedule a superblock update |
5610 | * and wait for it to complete. | |
06d91a5f | 5611 | */ |
3d310eb7 | 5612 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 5613 | { |
0fd62b86 | 5614 | int did_change = 0; |
06d91a5f | 5615 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 5616 | return; |
06d91a5f | 5617 | |
f91de92e N |
5618 | BUG_ON(mddev->ro == 1); |
5619 | if (mddev->ro == 2) { | |
5620 | /* need to switch to read/write */ | |
5621 | mddev->ro = 0; | |
5622 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5623 | md_wakeup_thread(mddev->thread); | |
25156198 | 5624 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 5625 | did_change = 1; |
f91de92e | 5626 | } |
06d91a5f | 5627 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
5628 | if (mddev->safemode == 1) |
5629 | mddev->safemode = 0; | |
06d91a5f | 5630 | if (mddev->in_sync) { |
a9701a30 | 5631 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
5632 | if (mddev->in_sync) { |
5633 | mddev->in_sync = 0; | |
850b2b42 | 5634 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3d310eb7 | 5635 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 5636 | did_change = 1; |
3d310eb7 | 5637 | } |
a9701a30 | 5638 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 5639 | } |
0fd62b86 NB |
5640 | if (did_change) |
5641 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
09a44cc1 N |
5642 | wait_event(mddev->sb_wait, |
5643 | !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && | |
5644 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1da177e4 LT |
5645 | } |
5646 | ||
5647 | void md_write_end(mddev_t *mddev) | |
5648 | { | |
5649 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
5650 | if (mddev->safemode == 2) | |
5651 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 5652 | else if (mddev->safemode_delay) |
1da177e4 LT |
5653 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
5654 | } | |
5655 | } | |
5656 | ||
2a2275d6 N |
5657 | /* md_allow_write(mddev) |
5658 | * Calling this ensures that the array is marked 'active' so that writes | |
5659 | * may proceed without blocking. It is important to call this before | |
5660 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
5661 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
5662 | * |
5663 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
5664 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 5665 | */ |
b5470dc5 | 5666 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
5667 | { |
5668 | if (!mddev->pers) | |
b5470dc5 | 5669 | return 0; |
2a2275d6 | 5670 | if (mddev->ro) |
b5470dc5 | 5671 | return 0; |
1a0fd497 | 5672 | if (!mddev->pers->sync_request) |
b5470dc5 | 5673 | return 0; |
2a2275d6 N |
5674 | |
5675 | spin_lock_irq(&mddev->write_lock); | |
5676 | if (mddev->in_sync) { | |
5677 | mddev->in_sync = 0; | |
5678 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
5679 | if (mddev->safemode_delay && | |
5680 | mddev->safemode == 0) | |
5681 | mddev->safemode = 1; | |
5682 | spin_unlock_irq(&mddev->write_lock); | |
5683 | md_update_sb(mddev, 0); | |
09a44cc1 | 5684 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
2a2275d6 N |
5685 | } else |
5686 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 DW |
5687 | |
5688 | if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
5689 | return -EAGAIN; | |
5690 | else | |
5691 | return 0; | |
2a2275d6 N |
5692 | } |
5693 | EXPORT_SYMBOL_GPL(md_allow_write); | |
5694 | ||
1da177e4 LT |
5695 | #define SYNC_MARKS 10 |
5696 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 5697 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
5698 | { |
5699 | mddev_t *mddev2; | |
5700 | unsigned int currspeed = 0, | |
5701 | window; | |
57afd89f | 5702 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
5703 | unsigned long mark[SYNC_MARKS]; |
5704 | sector_t mark_cnt[SYNC_MARKS]; | |
5705 | int last_mark,m; | |
5706 | struct list_head *tmp; | |
5707 | sector_t last_check; | |
57afd89f | 5708 | int skipped = 0; |
5fd6c1dc N |
5709 | struct list_head *rtmp; |
5710 | mdk_rdev_t *rdev; | |
61df9d91 | 5711 | char *desc; |
1da177e4 LT |
5712 | |
5713 | /* just incase thread restarts... */ | |
5714 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
5715 | return; | |
5fd6c1dc N |
5716 | if (mddev->ro) /* never try to sync a read-only array */ |
5717 | return; | |
1da177e4 | 5718 | |
61df9d91 N |
5719 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
5720 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
5721 | desc = "data-check"; | |
5722 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
5723 | desc = "requested-resync"; | |
5724 | else | |
5725 | desc = "resync"; | |
5726 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
5727 | desc = "reshape"; | |
5728 | else | |
5729 | desc = "recovery"; | |
5730 | ||
1da177e4 LT |
5731 | /* we overload curr_resync somewhat here. |
5732 | * 0 == not engaged in resync at all | |
5733 | * 2 == checking that there is no conflict with another sync | |
5734 | * 1 == like 2, but have yielded to allow conflicting resync to | |
5735 | * commense | |
5736 | * other == active in resync - this many blocks | |
5737 | * | |
5738 | * Before starting a resync we must have set curr_resync to | |
5739 | * 2, and then checked that every "conflicting" array has curr_resync | |
5740 | * less than ours. When we find one that is the same or higher | |
5741 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
5742 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
5743 | * This will mean we have to start checking from the beginning again. | |
5744 | * | |
5745 | */ | |
5746 | ||
5747 | do { | |
5748 | mddev->curr_resync = 2; | |
5749 | ||
5750 | try_again: | |
787453c2 | 5751 | if (kthread_should_stop()) { |
6985c43f | 5752 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
5753 | goto skip; |
5754 | } | |
29ac4aa3 | 5755 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
5756 | if (mddev2 == mddev) |
5757 | continue; | |
90b08710 BS |
5758 | if (!mddev->parallel_resync |
5759 | && mddev2->curr_resync | |
5760 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
5761 | DEFINE_WAIT(wq); |
5762 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
5763 | /* arbitrarily yield */ | |
5764 | mddev->curr_resync = 1; | |
5765 | wake_up(&resync_wait); | |
5766 | } | |
5767 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
5768 | /* no need to wait here, we can wait the next | |
5769 | * time 'round when curr_resync == 2 | |
5770 | */ | |
5771 | continue; | |
787453c2 N |
5772 | prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE); |
5773 | if (!kthread_should_stop() && | |
8712e553 | 5774 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
5775 | printk(KERN_INFO "md: delaying %s of %s" |
5776 | " until %s has finished (they" | |
1da177e4 | 5777 | " share one or more physical units)\n", |
61df9d91 | 5778 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 LT |
5779 | mddev_put(mddev2); |
5780 | schedule(); | |
5781 | finish_wait(&resync_wait, &wq); | |
5782 | goto try_again; | |
5783 | } | |
5784 | finish_wait(&resync_wait, &wq); | |
5785 | } | |
5786 | } | |
5787 | } while (mddev->curr_resync < 2); | |
5788 | ||
5fd6c1dc | 5789 | j = 0; |
9d88883e | 5790 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 5791 | /* resync follows the size requested by the personality, |
57afd89f | 5792 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
5793 | */ |
5794 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 5795 | mddev->resync_mismatches = 0; |
5fd6c1dc | 5796 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
5797 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
5798 | j = mddev->resync_min; | |
5799 | else if (!mddev->bitmap) | |
5fd6c1dc | 5800 | j = mddev->recovery_cp; |
5e96ee65 | 5801 | |
ccfcc3c1 N |
5802 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5803 | max_sectors = mddev->size << 1; | |
5fd6c1dc | 5804 | else { |
1da177e4 LT |
5805 | /* recovery follows the physical size of devices */ |
5806 | max_sectors = mddev->size << 1; | |
5fd6c1dc | 5807 | j = MaxSector; |
d089c6af | 5808 | rdev_for_each(rdev, rtmp, mddev) |
5fd6c1dc N |
5809 | if (rdev->raid_disk >= 0 && |
5810 | !test_bit(Faulty, &rdev->flags) && | |
5811 | !test_bit(In_sync, &rdev->flags) && | |
5812 | rdev->recovery_offset < j) | |
5813 | j = rdev->recovery_offset; | |
5814 | } | |
1da177e4 | 5815 | |
61df9d91 N |
5816 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
5817 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
5818 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 5819 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
5820 | "(but not more than %d KB/sec) for %s.\n", |
5821 | speed_max(mddev), desc); | |
1da177e4 LT |
5822 | |
5823 | is_mddev_idle(mddev); /* this also initializes IO event counters */ | |
5fd6c1dc | 5824 | |
57afd89f | 5825 | io_sectors = 0; |
1da177e4 LT |
5826 | for (m = 0; m < SYNC_MARKS; m++) { |
5827 | mark[m] = jiffies; | |
57afd89f | 5828 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
5829 | } |
5830 | last_mark = 0; | |
5831 | mddev->resync_mark = mark[last_mark]; | |
5832 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
5833 | ||
5834 | /* | |
5835 | * Tune reconstruction: | |
5836 | */ | |
5837 | window = 32*(PAGE_SIZE/512); | |
5838 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
5839 | window/2,(unsigned long long) max_sectors/2); | |
5840 | ||
5841 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
5842 | last_check = 0; |
5843 | ||
5844 | if (j>2) { | |
5845 | printk(KERN_INFO | |
61df9d91 N |
5846 | "md: resuming %s of %s from checkpoint.\n", |
5847 | desc, mdname(mddev)); | |
1da177e4 LT |
5848 | mddev->curr_resync = j; |
5849 | } | |
5850 | ||
5851 | while (j < max_sectors) { | |
57afd89f | 5852 | sector_t sectors; |
1da177e4 | 5853 | |
57afd89f | 5854 | skipped = 0; |
c6207277 N |
5855 | if (j >= mddev->resync_max) { |
5856 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
5857 | wait_event(mddev->recovery_wait, | |
5858 | mddev->resync_max > j | |
5859 | || kthread_should_stop()); | |
5860 | } | |
5861 | if (kthread_should_stop()) | |
5862 | goto interrupted; | |
57afd89f | 5863 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 5864 | currspeed < speed_min(mddev)); |
57afd89f | 5865 | if (sectors == 0) { |
dfc70645 | 5866 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
5867 | goto out; |
5868 | } | |
57afd89f N |
5869 | |
5870 | if (!skipped) { /* actual IO requested */ | |
5871 | io_sectors += sectors; | |
5872 | atomic_add(sectors, &mddev->recovery_active); | |
5873 | } | |
5874 | ||
1da177e4 LT |
5875 | j += sectors; |
5876 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 5877 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
5878 | if (last_check == 0) |
5879 | /* this is the earliers that rebuilt will be | |
5880 | * visible in /proc/mdstat | |
5881 | */ | |
5882 | md_new_event(mddev); | |
57afd89f N |
5883 | |
5884 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
5885 | continue; |
5886 | ||
57afd89f | 5887 | last_check = io_sectors; |
1da177e4 | 5888 | |
dfc70645 | 5889 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
5890 | break; |
5891 | ||
5892 | repeat: | |
5893 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
5894 | /* step marks */ | |
5895 | int next = (last_mark+1) % SYNC_MARKS; | |
5896 | ||
5897 | mddev->resync_mark = mark[next]; | |
5898 | mddev->resync_mark_cnt = mark_cnt[next]; | |
5899 | mark[next] = jiffies; | |
57afd89f | 5900 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
5901 | last_mark = next; |
5902 | } | |
5903 | ||
5904 | ||
c6207277 N |
5905 | if (kthread_should_stop()) |
5906 | goto interrupted; | |
5907 | ||
1da177e4 LT |
5908 | |
5909 | /* | |
5910 | * this loop exits only if either when we are slower than | |
5911 | * the 'hard' speed limit, or the system was IO-idle for | |
5912 | * a jiffy. | |
5913 | * the system might be non-idle CPU-wise, but we only care | |
5914 | * about not overloading the IO subsystem. (things like an | |
5915 | * e2fsck being done on the RAID array should execute fast) | |
5916 | */ | |
2ad8b1ef | 5917 | blk_unplug(mddev->queue); |
1da177e4 LT |
5918 | cond_resched(); |
5919 | ||
57afd89f N |
5920 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
5921 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 5922 | |
88202a0c N |
5923 | if (currspeed > speed_min(mddev)) { |
5924 | if ((currspeed > speed_max(mddev)) || | |
1da177e4 | 5925 | !is_mddev_idle(mddev)) { |
c0e48521 | 5926 | msleep(500); |
1da177e4 LT |
5927 | goto repeat; |
5928 | } | |
5929 | } | |
5930 | } | |
61df9d91 | 5931 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
5932 | /* |
5933 | * this also signals 'finished resyncing' to md_stop | |
5934 | */ | |
5935 | out: | |
2ad8b1ef | 5936 | blk_unplug(mddev->queue); |
1da177e4 LT |
5937 | |
5938 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
5939 | ||
5940 | /* tell personality that we are finished */ | |
57afd89f | 5941 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 5942 | |
dfc70645 | 5943 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
5944 | mddev->curr_resync > 2) { |
5945 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
5946 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5947 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
5948 | printk(KERN_INFO | |
61df9d91 N |
5949 | "md: checkpointing %s of %s.\n", |
5950 | desc, mdname(mddev)); | |
5fd6c1dc N |
5951 | mddev->recovery_cp = mddev->curr_resync; |
5952 | } | |
5953 | } else | |
5954 | mddev->recovery_cp = MaxSector; | |
5955 | } else { | |
5956 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5957 | mddev->curr_resync = MaxSector; | |
d089c6af | 5958 | rdev_for_each(rdev, rtmp, mddev) |
5fd6c1dc N |
5959 | if (rdev->raid_disk >= 0 && |
5960 | !test_bit(Faulty, &rdev->flags) && | |
5961 | !test_bit(In_sync, &rdev->flags) && | |
5962 | rdev->recovery_offset < mddev->curr_resync) | |
5963 | rdev->recovery_offset = mddev->curr_resync; | |
5fd6c1dc | 5964 | } |
1da177e4 | 5965 | } |
17571284 | 5966 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5967 | |
1da177e4 LT |
5968 | skip: |
5969 | mddev->curr_resync = 0; | |
5e96ee65 | 5970 | mddev->resync_min = 0; |
c6207277 N |
5971 | mddev->resync_max = MaxSector; |
5972 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
1da177e4 LT |
5973 | wake_up(&resync_wait); |
5974 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
5975 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
5976 | return; |
5977 | ||
5978 | interrupted: | |
5979 | /* | |
5980 | * got a signal, exit. | |
5981 | */ | |
5982 | printk(KERN_INFO | |
5983 | "md: md_do_sync() got signal ... exiting\n"); | |
5984 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5985 | goto out; | |
5986 | ||
1da177e4 | 5987 | } |
29269553 | 5988 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
5989 | |
5990 | ||
b4c4c7b8 N |
5991 | static int remove_and_add_spares(mddev_t *mddev) |
5992 | { | |
5993 | mdk_rdev_t *rdev; | |
5994 | struct list_head *rtmp; | |
5995 | int spares = 0; | |
5996 | ||
d089c6af | 5997 | rdev_for_each(rdev, rtmp, mddev) |
b4c4c7b8 | 5998 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 5999 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
6000 | (test_bit(Faulty, &rdev->flags) || |
6001 | ! test_bit(In_sync, &rdev->flags)) && | |
6002 | atomic_read(&rdev->nr_pending)==0) { | |
6003 | if (mddev->pers->hot_remove_disk( | |
6004 | mddev, rdev->raid_disk)==0) { | |
6005 | char nm[20]; | |
6006 | sprintf(nm,"rd%d", rdev->raid_disk); | |
6007 | sysfs_remove_link(&mddev->kobj, nm); | |
6008 | rdev->raid_disk = -1; | |
6009 | } | |
6010 | } | |
6011 | ||
6012 | if (mddev->degraded) { | |
dfc70645 N |
6013 | rdev_for_each(rdev, rtmp, mddev) { |
6014 | if (rdev->raid_disk >= 0 && | |
6015 | !test_bit(In_sync, &rdev->flags)) | |
6016 | spares++; | |
b4c4c7b8 N |
6017 | if (rdev->raid_disk < 0 |
6018 | && !test_bit(Faulty, &rdev->flags)) { | |
6019 | rdev->recovery_offset = 0; | |
199050ea NB |
6020 | if (mddev->pers-> |
6021 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
6022 | char nm[20]; |
6023 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
6024 | if (sysfs_create_link(&mddev->kobj, |
6025 | &rdev->kobj, nm)) | |
6026 | printk(KERN_WARNING | |
6027 | "md: cannot register " | |
6028 | "%s for %s\n", | |
6029 | nm, mdname(mddev)); | |
b4c4c7b8 N |
6030 | spares++; |
6031 | md_new_event(mddev); | |
6032 | } else | |
6033 | break; | |
6034 | } | |
dfc70645 | 6035 | } |
b4c4c7b8 N |
6036 | } |
6037 | return spares; | |
6038 | } | |
1da177e4 LT |
6039 | /* |
6040 | * This routine is regularly called by all per-raid-array threads to | |
6041 | * deal with generic issues like resync and super-block update. | |
6042 | * Raid personalities that don't have a thread (linear/raid0) do not | |
6043 | * need this as they never do any recovery or update the superblock. | |
6044 | * | |
6045 | * It does not do any resync itself, but rather "forks" off other threads | |
6046 | * to do that as needed. | |
6047 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
6048 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 6049 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
6050 | * and wakeups up this thread which will reap the thread and finish up. |
6051 | * This thread also removes any faulty devices (with nr_pending == 0). | |
6052 | * | |
6053 | * The overall approach is: | |
6054 | * 1/ if the superblock needs updating, update it. | |
6055 | * 2/ If a recovery thread is running, don't do anything else. | |
6056 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
6057 | * 4/ If there are any faulty devices, remove them. | |
6058 | * 5/ If array is degraded, try to add spares devices | |
6059 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
6060 | */ | |
6061 | void md_check_recovery(mddev_t *mddev) | |
6062 | { | |
6063 | mdk_rdev_t *rdev; | |
6064 | struct list_head *rtmp; | |
6065 | ||
6066 | ||
5f40402d N |
6067 | if (mddev->bitmap) |
6068 | bitmap_daemon_work(mddev->bitmap); | |
1da177e4 LT |
6069 | |
6070 | if (mddev->ro) | |
6071 | return; | |
fca4d848 N |
6072 | |
6073 | if (signal_pending(current)) { | |
31a59e34 | 6074 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
6075 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
6076 | mdname(mddev)); | |
6077 | mddev->safemode = 2; | |
6078 | } | |
6079 | flush_signals(current); | |
6080 | } | |
6081 | ||
1da177e4 | 6082 | if ( ! ( |
e691063a | 6083 | (mddev->flags && !mddev->external) || |
1da177e4 | 6084 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 6085 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 6086 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
6087 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
6088 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
6089 | )) |
6090 | return; | |
fca4d848 | 6091 | |
df5b89b3 | 6092 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 6093 | int spares = 0; |
fca4d848 | 6094 | |
31a59e34 | 6095 | if (!mddev->external) { |
0fd62b86 | 6096 | int did_change = 0; |
31a59e34 N |
6097 | spin_lock_irq(&mddev->write_lock); |
6098 | if (mddev->safemode && | |
6099 | !atomic_read(&mddev->writes_pending) && | |
6100 | !mddev->in_sync && | |
6101 | mddev->recovery_cp == MaxSector) { | |
6102 | mddev->in_sync = 1; | |
0fd62b86 | 6103 | did_change = 1; |
31a59e34 N |
6104 | if (mddev->persistent) |
6105 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6106 | } | |
6107 | if (mddev->safemode == 1) | |
6108 | mddev->safemode = 0; | |
6109 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 NB |
6110 | if (did_change) |
6111 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
fca4d848 | 6112 | } |
fca4d848 | 6113 | |
850b2b42 N |
6114 | if (mddev->flags) |
6115 | md_update_sb(mddev, 0); | |
06d91a5f | 6116 | |
52664732 NB |
6117 | rdev_for_each(rdev, rtmp, mddev) |
6118 | if (test_and_clear_bit(StateChanged, &rdev->flags)) | |
6119 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
6120 | ||
06d91a5f | 6121 | |
1da177e4 LT |
6122 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
6123 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
6124 | /* resync/recovery still happening */ | |
6125 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6126 | goto unlock; | |
6127 | } | |
6128 | if (mddev->sync_thread) { | |
6129 | /* resync has finished, collect result */ | |
6130 | md_unregister_thread(mddev->sync_thread); | |
6131 | mddev->sync_thread = NULL; | |
dfc70645 | 6132 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
1da177e4 LT |
6133 | /* success...*/ |
6134 | /* activate any spares */ | |
a99ac971 NB |
6135 | if (mddev->pers->spare_active(mddev)) |
6136 | sysfs_notify(&mddev->kobj, NULL, | |
6137 | "degraded"); | |
1da177e4 | 6138 | } |
850b2b42 | 6139 | md_update_sb(mddev, 1); |
41158c7e N |
6140 | |
6141 | /* if array is no-longer degraded, then any saved_raid_disk | |
6142 | * information must be scrapped | |
6143 | */ | |
6144 | if (!mddev->degraded) | |
d089c6af | 6145 | rdev_for_each(rdev, rtmp, mddev) |
41158c7e N |
6146 | rdev->saved_raid_disk = -1; |
6147 | ||
1da177e4 LT |
6148 | mddev->recovery = 0; |
6149 | /* flag recovery needed just to double check */ | |
6150 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
72a23c21 | 6151 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
d7603b7e | 6152 | md_new_event(mddev); |
1da177e4 LT |
6153 | goto unlock; |
6154 | } | |
72a23c21 NB |
6155 | /* Set RUNNING before clearing NEEDED to avoid |
6156 | * any transients in the value of "sync_action". | |
6157 | */ | |
6158 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6159 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
6160 | /* Clear some bits that don't mean anything, but |
6161 | * might be left set | |
6162 | */ | |
24dd469d N |
6163 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6164 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 6165 | |
5fd6c1dc N |
6166 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6167 | goto unlock; | |
1da177e4 LT |
6168 | /* no recovery is running. |
6169 | * remove any failed drives, then | |
6170 | * add spares if possible. | |
6171 | * Spare are also removed and re-added, to allow | |
6172 | * the personality to fail the re-add. | |
6173 | */ | |
1da177e4 | 6174 | |
b4c4c7b8 N |
6175 | if (mddev->reshape_position != MaxSector) { |
6176 | if (mddev->pers->check_reshape(mddev) != 0) | |
6177 | /* Cannot proceed */ | |
6178 | goto unlock; | |
6179 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 6180 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 6181 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
6182 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
6183 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
72a23c21 | 6184 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6185 | } else if (mddev->recovery_cp < MaxSector) { |
6186 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 6187 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6188 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
6189 | /* nothing to be done ... */ | |
1da177e4 | 6190 | goto unlock; |
24dd469d | 6191 | |
1da177e4 | 6192 | if (mddev->pers->sync_request) { |
a654b9d8 N |
6193 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
6194 | /* We are adding a device or devices to an array | |
6195 | * which has the bitmap stored on all devices. | |
6196 | * So make sure all bitmap pages get written | |
6197 | */ | |
6198 | bitmap_write_all(mddev->bitmap); | |
6199 | } | |
1da177e4 LT |
6200 | mddev->sync_thread = md_register_thread(md_do_sync, |
6201 | mddev, | |
6202 | "%s_resync"); | |
6203 | if (!mddev->sync_thread) { | |
6204 | printk(KERN_ERR "%s: could not start resync" | |
6205 | " thread...\n", | |
6206 | mdname(mddev)); | |
6207 | /* leave the spares where they are, it shouldn't hurt */ | |
6208 | mddev->recovery = 0; | |
d7603b7e | 6209 | } else |
1da177e4 | 6210 | md_wakeup_thread(mddev->sync_thread); |
72a23c21 | 6211 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
d7603b7e | 6212 | md_new_event(mddev); |
1da177e4 LT |
6213 | } |
6214 | unlock: | |
72a23c21 NB |
6215 | if (!mddev->sync_thread) { |
6216 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6217 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
6218 | &mddev->recovery)) | |
6219 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); | |
6220 | } | |
1da177e4 LT |
6221 | mddev_unlock(mddev); |
6222 | } | |
6223 | } | |
6224 | ||
6bfe0b49 DW |
6225 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
6226 | { | |
6227 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
6228 | wait_event_timeout(rdev->blocked_wait, | |
6229 | !test_bit(Blocked, &rdev->flags), | |
6230 | msecs_to_jiffies(5000)); | |
6231 | rdev_dec_pending(rdev, mddev); | |
6232 | } | |
6233 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
6234 | ||
75c96f85 AB |
6235 | static int md_notify_reboot(struct notifier_block *this, |
6236 | unsigned long code, void *x) | |
1da177e4 LT |
6237 | { |
6238 | struct list_head *tmp; | |
6239 | mddev_t *mddev; | |
6240 | ||
6241 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
6242 | ||
6243 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
6244 | ||
29ac4aa3 | 6245 | for_each_mddev(mddev, tmp) |
c71d4887 | 6246 | if (mddev_trylock(mddev)) { |
1da177e4 | 6247 | do_md_stop (mddev, 1); |
c71d4887 NB |
6248 | mddev_unlock(mddev); |
6249 | } | |
1da177e4 LT |
6250 | /* |
6251 | * certain more exotic SCSI devices are known to be | |
6252 | * volatile wrt too early system reboots. While the | |
6253 | * right place to handle this issue is the given | |
6254 | * driver, we do want to have a safe RAID driver ... | |
6255 | */ | |
6256 | mdelay(1000*1); | |
6257 | } | |
6258 | return NOTIFY_DONE; | |
6259 | } | |
6260 | ||
75c96f85 | 6261 | static struct notifier_block md_notifier = { |
1da177e4 LT |
6262 | .notifier_call = md_notify_reboot, |
6263 | .next = NULL, | |
6264 | .priority = INT_MAX, /* before any real devices */ | |
6265 | }; | |
6266 | ||
6267 | static void md_geninit(void) | |
6268 | { | |
1da177e4 LT |
6269 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
6270 | ||
c7705f34 | 6271 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
6272 | } |
6273 | ||
75c96f85 | 6274 | static int __init md_init(void) |
1da177e4 | 6275 | { |
1da177e4 LT |
6276 | if (register_blkdev(MAJOR_NR, "md")) |
6277 | return -1; | |
6278 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
6279 | unregister_blkdev(MAJOR_NR, "md"); | |
6280 | return -1; | |
6281 | } | |
e8703fe1 N |
6282 | blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE, |
6283 | md_probe, NULL, NULL); | |
6284 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
6285 | md_probe, NULL, NULL); |
6286 | ||
1da177e4 | 6287 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 6288 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
6289 | |
6290 | md_geninit(); | |
6291 | return (0); | |
6292 | } | |
6293 | ||
6294 | ||
6295 | #ifndef MODULE | |
6296 | ||
6297 | /* | |
6298 | * Searches all registered partitions for autorun RAID arrays | |
6299 | * at boot time. | |
6300 | */ | |
4d936ec1 ME |
6301 | |
6302 | static LIST_HEAD(all_detected_devices); | |
6303 | struct detected_devices_node { | |
6304 | struct list_head list; | |
6305 | dev_t dev; | |
6306 | }; | |
1da177e4 LT |
6307 | |
6308 | void md_autodetect_dev(dev_t dev) | |
6309 | { | |
4d936ec1 ME |
6310 | struct detected_devices_node *node_detected_dev; |
6311 | ||
6312 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
6313 | if (node_detected_dev) { | |
6314 | node_detected_dev->dev = dev; | |
6315 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
6316 | } else { | |
6317 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
6318 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
6319 | } | |
1da177e4 LT |
6320 | } |
6321 | ||
6322 | ||
6323 | static void autostart_arrays(int part) | |
6324 | { | |
6325 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
6326 | struct detected_devices_node *node_detected_dev; |
6327 | dev_t dev; | |
6328 | int i_scanned, i_passed; | |
1da177e4 | 6329 | |
4d936ec1 ME |
6330 | i_scanned = 0; |
6331 | i_passed = 0; | |
1da177e4 | 6332 | |
4d936ec1 | 6333 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 6334 | |
4d936ec1 ME |
6335 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
6336 | i_scanned++; | |
6337 | node_detected_dev = list_entry(all_detected_devices.next, | |
6338 | struct detected_devices_node, list); | |
6339 | list_del(&node_detected_dev->list); | |
6340 | dev = node_detected_dev->dev; | |
6341 | kfree(node_detected_dev); | |
df968c4e | 6342 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
6343 | if (IS_ERR(rdev)) |
6344 | continue; | |
6345 | ||
b2d444d7 | 6346 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6347 | MD_BUG(); |
6348 | continue; | |
6349 | } | |
d0fae18f | 6350 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 6351 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 6352 | i_passed++; |
1da177e4 | 6353 | } |
4d936ec1 ME |
6354 | |
6355 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
6356 | i_scanned, i_passed); | |
1da177e4 LT |
6357 | |
6358 | autorun_devices(part); | |
6359 | } | |
6360 | ||
fdee8ae4 | 6361 | #endif /* !MODULE */ |
1da177e4 LT |
6362 | |
6363 | static __exit void md_exit(void) | |
6364 | { | |
6365 | mddev_t *mddev; | |
6366 | struct list_head *tmp; | |
8ab5e4c1 | 6367 | |
e8703fe1 N |
6368 | blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS); |
6369 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); | |
1da177e4 LT |
6370 | |
6371 | unregister_blkdev(MAJOR_NR,"md"); | |
6372 | unregister_blkdev(mdp_major, "mdp"); | |
6373 | unregister_reboot_notifier(&md_notifier); | |
6374 | unregister_sysctl_table(raid_table_header); | |
6375 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 6376 | for_each_mddev(mddev, tmp) { |
1da177e4 LT |
6377 | struct gendisk *disk = mddev->gendisk; |
6378 | if (!disk) | |
6379 | continue; | |
6380 | export_array(mddev); | |
6381 | del_gendisk(disk); | |
6382 | put_disk(disk); | |
6383 | mddev->gendisk = NULL; | |
6384 | mddev_put(mddev); | |
6385 | } | |
6386 | } | |
6387 | ||
685784aa | 6388 | subsys_initcall(md_init); |
1da177e4 LT |
6389 | module_exit(md_exit) |
6390 | ||
f91de92e N |
6391 | static int get_ro(char *buffer, struct kernel_param *kp) |
6392 | { | |
6393 | return sprintf(buffer, "%d", start_readonly); | |
6394 | } | |
6395 | static int set_ro(const char *val, struct kernel_param *kp) | |
6396 | { | |
6397 | char *e; | |
6398 | int num = simple_strtoul(val, &e, 10); | |
6399 | if (*val && (*e == '\0' || *e == '\n')) { | |
6400 | start_readonly = num; | |
4dbcdc75 | 6401 | return 0; |
f91de92e N |
6402 | } |
6403 | return -EINVAL; | |
6404 | } | |
6405 | ||
80ca3a44 N |
6406 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
6407 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 6408 | |
f91de92e | 6409 | |
1da177e4 LT |
6410 | EXPORT_SYMBOL(register_md_personality); |
6411 | EXPORT_SYMBOL(unregister_md_personality); | |
6412 | EXPORT_SYMBOL(md_error); | |
6413 | EXPORT_SYMBOL(md_done_sync); | |
6414 | EXPORT_SYMBOL(md_write_start); | |
6415 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
6416 | EXPORT_SYMBOL(md_register_thread); |
6417 | EXPORT_SYMBOL(md_unregister_thread); | |
6418 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
6419 | EXPORT_SYMBOL(md_check_recovery); |
6420 | MODULE_LICENSE("GPL"); | |
aa1595e9 | 6421 | MODULE_ALIAS("md"); |
72008652 | 6422 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |