]>
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 | ||
575 | if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4)) | |
576 | ret = 0; | |
577 | else | |
578 | ret = 1; | |
579 | ||
580 | abort: | |
990a8baf JJ |
581 | kfree(tmp1); |
582 | kfree(tmp2); | |
1da177e4 LT |
583 | return ret; |
584 | } | |
585 | ||
4d167f09 N |
586 | |
587 | static u32 md_csum_fold(u32 csum) | |
588 | { | |
589 | csum = (csum & 0xffff) + (csum >> 16); | |
590 | return (csum & 0xffff) + (csum >> 16); | |
591 | } | |
592 | ||
1da177e4 LT |
593 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
594 | { | |
4d167f09 N |
595 | u64 newcsum = 0; |
596 | u32 *sb32 = (u32*)sb; | |
597 | int i; | |
1da177e4 LT |
598 | unsigned int disk_csum, csum; |
599 | ||
600 | disk_csum = sb->sb_csum; | |
601 | sb->sb_csum = 0; | |
4d167f09 N |
602 | |
603 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
604 | newcsum += sb32[i]; | |
605 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
606 | ||
607 | ||
608 | #ifdef CONFIG_ALPHA | |
609 | /* This used to use csum_partial, which was wrong for several | |
610 | * reasons including that different results are returned on | |
611 | * different architectures. It isn't critical that we get exactly | |
612 | * the same return value as before (we always csum_fold before | |
613 | * testing, and that removes any differences). However as we | |
614 | * know that csum_partial always returned a 16bit value on | |
615 | * alphas, do a fold to maximise conformity to previous behaviour. | |
616 | */ | |
617 | sb->sb_csum = md_csum_fold(disk_csum); | |
618 | #else | |
1da177e4 | 619 | sb->sb_csum = disk_csum; |
4d167f09 | 620 | #endif |
1da177e4 LT |
621 | return csum; |
622 | } | |
623 | ||
624 | ||
625 | /* | |
626 | * Handle superblock details. | |
627 | * We want to be able to handle multiple superblock formats | |
628 | * so we have a common interface to them all, and an array of | |
629 | * different handlers. | |
630 | * We rely on user-space to write the initial superblock, and support | |
631 | * reading and updating of superblocks. | |
632 | * Interface methods are: | |
633 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
634 | * loads and validates a superblock on dev. | |
635 | * if refdev != NULL, compare superblocks on both devices | |
636 | * Return: | |
637 | * 0 - dev has a superblock that is compatible with refdev | |
638 | * 1 - dev has a superblock that is compatible and newer than refdev | |
639 | * so dev should be used as the refdev in future | |
640 | * -EINVAL superblock incompatible or invalid | |
641 | * -othererror e.g. -EIO | |
642 | * | |
643 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
644 | * Verify that dev is acceptable into mddev. | |
645 | * The first time, mddev->raid_disks will be 0, and data from | |
646 | * dev should be merged in. Subsequent calls check that dev | |
647 | * is new enough. Return 0 or -EINVAL | |
648 | * | |
649 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
650 | * Update the superblock for rdev with data in mddev | |
651 | * This does not write to disc. | |
652 | * | |
653 | */ | |
654 | ||
655 | struct super_type { | |
0cd17fec CW |
656 | char *name; |
657 | struct module *owner; | |
658 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
659 | int minor_version); | |
660 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
661 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
662 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
663 | unsigned long long size); | |
1da177e4 LT |
664 | }; |
665 | ||
666 | /* | |
667 | * load_super for 0.90.0 | |
668 | */ | |
669 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
670 | { | |
671 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
672 | mdp_super_t *sb; | |
673 | int ret; | |
674 | sector_t sb_offset; | |
675 | ||
676 | /* | |
677 | * Calculate the position of the superblock, | |
678 | * it's at the end of the disk. | |
679 | * | |
680 | * It also happens to be a multiple of 4Kb. | |
681 | */ | |
682 | sb_offset = calc_dev_sboffset(rdev->bdev); | |
683 | rdev->sb_offset = sb_offset; | |
684 | ||
0002b271 | 685 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
686 | if (ret) return ret; |
687 | ||
688 | ret = -EINVAL; | |
689 | ||
690 | bdevname(rdev->bdev, b); | |
691 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
692 | ||
693 | if (sb->md_magic != MD_SB_MAGIC) { | |
694 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
695 | b); | |
696 | goto abort; | |
697 | } | |
698 | ||
699 | if (sb->major_version != 0 || | |
f6705578 N |
700 | sb->minor_version < 90 || |
701 | sb->minor_version > 91) { | |
1da177e4 LT |
702 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
703 | sb->major_version, sb->minor_version, | |
704 | b); | |
705 | goto abort; | |
706 | } | |
707 | ||
708 | if (sb->raid_disks <= 0) | |
709 | goto abort; | |
710 | ||
4d167f09 | 711 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
712 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
713 | b); | |
714 | goto abort; | |
715 | } | |
716 | ||
717 | rdev->preferred_minor = sb->md_minor; | |
718 | rdev->data_offset = 0; | |
0002b271 | 719 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 | 720 | |
e11e93fa N |
721 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) { |
722 | if (sb->level != 1 && sb->level != 4 | |
723 | && sb->level != 5 && sb->level != 6 | |
724 | && sb->level != 10) { | |
725 | /* FIXME use a better test */ | |
726 | printk(KERN_WARNING | |
727 | "md: bitmaps not supported for this level.\n"); | |
728 | goto abort; | |
729 | } | |
730 | } | |
731 | ||
1da177e4 LT |
732 | if (sb->level == LEVEL_MULTIPATH) |
733 | rdev->desc_nr = -1; | |
734 | else | |
735 | rdev->desc_nr = sb->this_disk.number; | |
736 | ||
9a7b2b0f | 737 | if (!refdev) { |
1da177e4 | 738 | ret = 1; |
9a7b2b0f | 739 | } else { |
1da177e4 LT |
740 | __u64 ev1, ev2; |
741 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
742 | if (!uuid_equal(refsb, sb)) { | |
743 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
744 | b, bdevname(refdev->bdev,b2)); | |
745 | goto abort; | |
746 | } | |
747 | if (!sb_equal(refsb, sb)) { | |
748 | printk(KERN_WARNING "md: %s has same UUID" | |
749 | " but different superblock to %s\n", | |
750 | b, bdevname(refdev->bdev, b2)); | |
751 | goto abort; | |
752 | } | |
753 | ev1 = md_event(sb); | |
754 | ev2 = md_event(refsb); | |
755 | if (ev1 > ev2) | |
756 | ret = 1; | |
757 | else | |
758 | ret = 0; | |
759 | } | |
760 | rdev->size = calc_dev_size(rdev, sb->chunk_size); | |
761 | ||
2bf071bf N |
762 | if (rdev->size < sb->size && sb->level > 1) |
763 | /* "this cannot possibly happen" ... */ | |
764 | ret = -EINVAL; | |
765 | ||
1da177e4 LT |
766 | abort: |
767 | return ret; | |
768 | } | |
769 | ||
770 | /* | |
771 | * validate_super for 0.90.0 | |
772 | */ | |
773 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
774 | { | |
775 | mdp_disk_t *desc; | |
776 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 777 | __u64 ev1 = md_event(sb); |
1da177e4 | 778 | |
41158c7e | 779 | rdev->raid_disk = -1; |
c5d79adb N |
780 | clear_bit(Faulty, &rdev->flags); |
781 | clear_bit(In_sync, &rdev->flags); | |
782 | clear_bit(WriteMostly, &rdev->flags); | |
783 | clear_bit(BarriersNotsupp, &rdev->flags); | |
784 | ||
1da177e4 LT |
785 | if (mddev->raid_disks == 0) { |
786 | mddev->major_version = 0; | |
787 | mddev->minor_version = sb->minor_version; | |
788 | mddev->patch_version = sb->patch_version; | |
e691063a | 789 | mddev->external = 0; |
1da177e4 LT |
790 | mddev->chunk_size = sb->chunk_size; |
791 | mddev->ctime = sb->ctime; | |
792 | mddev->utime = sb->utime; | |
793 | mddev->level = sb->level; | |
d9d166c2 | 794 | mddev->clevel[0] = 0; |
1da177e4 LT |
795 | mddev->layout = sb->layout; |
796 | mddev->raid_disks = sb->raid_disks; | |
797 | mddev->size = sb->size; | |
07d84d10 | 798 | mddev->events = ev1; |
9223214e | 799 | mddev->bitmap_offset = 0; |
36fa3063 | 800 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
1da177e4 | 801 | |
f6705578 N |
802 | if (mddev->minor_version >= 91) { |
803 | mddev->reshape_position = sb->reshape_position; | |
804 | mddev->delta_disks = sb->delta_disks; | |
805 | mddev->new_level = sb->new_level; | |
806 | mddev->new_layout = sb->new_layout; | |
807 | mddev->new_chunk = sb->new_chunk; | |
808 | } else { | |
809 | mddev->reshape_position = MaxSector; | |
810 | mddev->delta_disks = 0; | |
811 | mddev->new_level = mddev->level; | |
812 | mddev->new_layout = mddev->layout; | |
813 | mddev->new_chunk = mddev->chunk_size; | |
814 | } | |
815 | ||
1da177e4 LT |
816 | if (sb->state & (1<<MD_SB_CLEAN)) |
817 | mddev->recovery_cp = MaxSector; | |
818 | else { | |
819 | if (sb->events_hi == sb->cp_events_hi && | |
820 | sb->events_lo == sb->cp_events_lo) { | |
821 | mddev->recovery_cp = sb->recovery_cp; | |
822 | } else | |
823 | mddev->recovery_cp = 0; | |
824 | } | |
825 | ||
826 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
827 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
828 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
829 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
830 | ||
831 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
832 | |
833 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
e11e93fa | 834 | mddev->bitmap_file == NULL) |
36fa3063 | 835 | mddev->bitmap_offset = mddev->default_bitmap_offset; |
a654b9d8 | 836 | |
41158c7e N |
837 | } else if (mddev->pers == NULL) { |
838 | /* Insist on good event counter while assembling */ | |
1da177e4 LT |
839 | ++ev1; |
840 | if (ev1 < mddev->events) | |
841 | return -EINVAL; | |
41158c7e N |
842 | } else if (mddev->bitmap) { |
843 | /* if adding to array with a bitmap, then we can accept an | |
844 | * older device ... but not too old. | |
845 | */ | |
41158c7e N |
846 | if (ev1 < mddev->bitmap->events_cleared) |
847 | return 0; | |
07d84d10 N |
848 | } else { |
849 | if (ev1 < mddev->events) | |
850 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
851 | return 0; | |
852 | } | |
41158c7e | 853 | |
1da177e4 | 854 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
855 | desc = sb->disks + rdev->desc_nr; |
856 | ||
857 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 858 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
859 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
860 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 861 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
862 | rdev->raid_disk = desc->raid_disk; |
863 | } | |
8ddf9efe N |
864 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
865 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 866 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 867 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
868 | return 0; |
869 | } | |
870 | ||
871 | /* | |
872 | * sync_super for 0.90.0 | |
873 | */ | |
874 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
875 | { | |
876 | mdp_super_t *sb; | |
877 | struct list_head *tmp; | |
878 | mdk_rdev_t *rdev2; | |
879 | int next_spare = mddev->raid_disks; | |
19133a42 | 880 | |
1da177e4 LT |
881 | |
882 | /* make rdev->sb match mddev data.. | |
883 | * | |
884 | * 1/ zero out disks | |
885 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
886 | * 3/ any empty disks < next_spare become removed | |
887 | * | |
888 | * disks[0] gets initialised to REMOVED because | |
889 | * we cannot be sure from other fields if it has | |
890 | * been initialised or not. | |
891 | */ | |
892 | int i; | |
893 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
894 | ||
61181565 N |
895 | rdev->sb_size = MD_SB_BYTES; |
896 | ||
1da177e4 LT |
897 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
898 | ||
899 | memset(sb, 0, sizeof(*sb)); | |
900 | ||
901 | sb->md_magic = MD_SB_MAGIC; | |
902 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
903 | sb->patch_version = mddev->patch_version; |
904 | sb->gvalid_words = 0; /* ignored */ | |
905 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
906 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
907 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
908 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
909 | ||
910 | sb->ctime = mddev->ctime; | |
911 | sb->level = mddev->level; | |
912 | sb->size = mddev->size; | |
913 | sb->raid_disks = mddev->raid_disks; | |
914 | sb->md_minor = mddev->md_minor; | |
e691063a | 915 | sb->not_persistent = 0; |
1da177e4 LT |
916 | sb->utime = mddev->utime; |
917 | sb->state = 0; | |
918 | sb->events_hi = (mddev->events>>32); | |
919 | sb->events_lo = (u32)mddev->events; | |
920 | ||
f6705578 N |
921 | if (mddev->reshape_position == MaxSector) |
922 | sb->minor_version = 90; | |
923 | else { | |
924 | sb->minor_version = 91; | |
925 | sb->reshape_position = mddev->reshape_position; | |
926 | sb->new_level = mddev->new_level; | |
927 | sb->delta_disks = mddev->delta_disks; | |
928 | sb->new_layout = mddev->new_layout; | |
929 | sb->new_chunk = mddev->new_chunk; | |
930 | } | |
931 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
932 | if (mddev->in_sync) |
933 | { | |
934 | sb->recovery_cp = mddev->recovery_cp; | |
935 | sb->cp_events_hi = (mddev->events>>32); | |
936 | sb->cp_events_lo = (u32)mddev->events; | |
937 | if (mddev->recovery_cp == MaxSector) | |
938 | sb->state = (1<< MD_SB_CLEAN); | |
939 | } else | |
940 | sb->recovery_cp = 0; | |
941 | ||
942 | sb->layout = mddev->layout; | |
943 | sb->chunk_size = mddev->chunk_size; | |
944 | ||
a654b9d8 N |
945 | if (mddev->bitmap && mddev->bitmap_file == NULL) |
946 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); | |
947 | ||
1da177e4 | 948 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
d089c6af | 949 | rdev_for_each(rdev2, tmp, mddev) { |
1da177e4 | 950 | mdp_disk_t *d; |
86e6ffdd | 951 | int desc_nr; |
b2d444d7 N |
952 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
953 | && !test_bit(Faulty, &rdev2->flags)) | |
86e6ffdd | 954 | desc_nr = rdev2->raid_disk; |
1da177e4 | 955 | else |
86e6ffdd | 956 | desc_nr = next_spare++; |
19133a42 | 957 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
958 | d = &sb->disks[rdev2->desc_nr]; |
959 | nr_disks++; | |
960 | d->number = rdev2->desc_nr; | |
961 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
962 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
b2d444d7 N |
963 | if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags) |
964 | && !test_bit(Faulty, &rdev2->flags)) | |
1da177e4 LT |
965 | d->raid_disk = rdev2->raid_disk; |
966 | else | |
967 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 968 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 969 | d->state = (1<<MD_DISK_FAULTY); |
1be7892f | 970 | else if (test_bit(In_sync, &rdev2->flags)) { |
1da177e4 LT |
971 | d->state = (1<<MD_DISK_ACTIVE); |
972 | d->state |= (1<<MD_DISK_SYNC); | |
973 | active++; | |
974 | working++; | |
975 | } else { | |
976 | d->state = 0; | |
977 | spare++; | |
978 | working++; | |
979 | } | |
8ddf9efe N |
980 | if (test_bit(WriteMostly, &rdev2->flags)) |
981 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 982 | } |
1da177e4 LT |
983 | /* now set the "removed" and "faulty" bits on any missing devices */ |
984 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
985 | mdp_disk_t *d = &sb->disks[i]; | |
986 | if (d->state == 0 && d->number == 0) { | |
987 | d->number = i; | |
988 | d->raid_disk = i; | |
989 | d->state = (1<<MD_DISK_REMOVED); | |
990 | d->state |= (1<<MD_DISK_FAULTY); | |
991 | failed++; | |
992 | } | |
993 | } | |
994 | sb->nr_disks = nr_disks; | |
995 | sb->active_disks = active; | |
996 | sb->working_disks = working; | |
997 | sb->failed_disks = failed; | |
998 | sb->spare_disks = spare; | |
999 | ||
1000 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1001 | sb->sb_csum = calc_sb_csum(sb); | |
1002 | } | |
1003 | ||
0cd17fec CW |
1004 | /* |
1005 | * rdev_size_change for 0.90.0 | |
1006 | */ | |
1007 | static unsigned long long | |
1008 | super_90_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size) | |
1009 | { | |
1010 | if (size && size < rdev->mddev->size) | |
1011 | return 0; /* component must fit device */ | |
1012 | size *= 2; /* convert to sectors */ | |
1013 | if (rdev->mddev->bitmap_offset) | |
1014 | return 0; /* can't move bitmap */ | |
1015 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
1016 | if (!size || size > rdev->sb_offset*2) | |
1017 | size = rdev->sb_offset*2; | |
1018 | md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size, | |
1019 | rdev->sb_page); | |
1020 | md_super_wait(rdev->mddev); | |
1021 | return size/2; /* kB for sysfs */ | |
1022 | } | |
1023 | ||
1024 | ||
1da177e4 LT |
1025 | /* |
1026 | * version 1 superblock | |
1027 | */ | |
1028 | ||
1c05b4bc | 1029 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1030 | { |
1c05b4bc N |
1031 | __le32 disk_csum; |
1032 | u32 csum; | |
1da177e4 LT |
1033 | unsigned long long newcsum; |
1034 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1035 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1036 | int i; |
1037 | ||
1038 | disk_csum = sb->sb_csum; | |
1039 | sb->sb_csum = 0; | |
1040 | newcsum = 0; | |
1041 | for (i=0; size>=4; size -= 4 ) | |
1042 | newcsum += le32_to_cpu(*isuper++); | |
1043 | ||
1044 | if (size == 2) | |
1c05b4bc | 1045 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1046 | |
1047 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1048 | sb->sb_csum = disk_csum; | |
1049 | return cpu_to_le32(csum); | |
1050 | } | |
1051 | ||
1052 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1053 | { | |
1054 | struct mdp_superblock_1 *sb; | |
1055 | int ret; | |
1056 | sector_t sb_offset; | |
1057 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
0002b271 | 1058 | int bmask; |
1da177e4 LT |
1059 | |
1060 | /* | |
1061 | * Calculate the position of the superblock. | |
1062 | * It is always aligned to a 4K boundary and | |
1063 | * depeding on minor_version, it can be: | |
1064 | * 0: At least 8K, but less than 12K, from end of device | |
1065 | * 1: At start of device | |
1066 | * 2: 4K from start of device. | |
1067 | */ | |
1068 | switch(minor_version) { | |
1069 | case 0: | |
1070 | sb_offset = rdev->bdev->bd_inode->i_size >> 9; | |
1071 | sb_offset -= 8*2; | |
39730960 | 1072 | sb_offset &= ~(sector_t)(4*2-1); |
1da177e4 LT |
1073 | /* convert from sectors to K */ |
1074 | sb_offset /= 2; | |
1075 | break; | |
1076 | case 1: | |
1077 | sb_offset = 0; | |
1078 | break; | |
1079 | case 2: | |
1080 | sb_offset = 4; | |
1081 | break; | |
1082 | default: | |
1083 | return -EINVAL; | |
1084 | } | |
1085 | rdev->sb_offset = sb_offset; | |
1086 | ||
0002b271 N |
1087 | /* superblock is rarely larger than 1K, but it can be larger, |
1088 | * and it is safe to read 4k, so we do that | |
1089 | */ | |
1090 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1091 | if (ret) return ret; |
1092 | ||
1093 | ||
1094 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1095 | ||
1096 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1097 | sb->major_version != cpu_to_le32(1) || | |
1098 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
1099 | le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) || | |
71c0805c | 1100 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1101 | return -EINVAL; |
1102 | ||
1103 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1104 | printk("md: invalid superblock checksum on %s\n", | |
1105 | bdevname(rdev->bdev,b)); | |
1106 | return -EINVAL; | |
1107 | } | |
1108 | if (le64_to_cpu(sb->data_size) < 10) { | |
1109 | printk("md: data_size too small on %s\n", | |
1110 | bdevname(rdev->bdev,b)); | |
1111 | return -EINVAL; | |
1112 | } | |
e11e93fa N |
1113 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) { |
1114 | if (sb->level != cpu_to_le32(1) && | |
1115 | sb->level != cpu_to_le32(4) && | |
1116 | sb->level != cpu_to_le32(5) && | |
1117 | sb->level != cpu_to_le32(6) && | |
1118 | sb->level != cpu_to_le32(10)) { | |
1119 | printk(KERN_WARNING | |
1120 | "md: bitmaps not supported for this level.\n"); | |
1121 | return -EINVAL; | |
1122 | } | |
1123 | } | |
1124 | ||
1da177e4 LT |
1125 | rdev->preferred_minor = 0xffff; |
1126 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1127 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1128 | |
0002b271 | 1129 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
720a3dc3 | 1130 | bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1131 | if (rdev->sb_size & bmask) |
a1801f85 N |
1132 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1133 | ||
1134 | if (minor_version | |
1135 | && rdev->data_offset < sb_offset + (rdev->sb_size/512)) | |
1136 | return -EINVAL; | |
0002b271 | 1137 | |
31b65a0d N |
1138 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1139 | rdev->desc_nr = -1; | |
1140 | else | |
1141 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1142 | ||
9a7b2b0f | 1143 | if (!refdev) { |
8ed75463 | 1144 | ret = 1; |
9a7b2b0f | 1145 | } else { |
1da177e4 LT |
1146 | __u64 ev1, ev2; |
1147 | struct mdp_superblock_1 *refsb = | |
1148 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1149 | ||
1150 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1151 | sb->level != refsb->level || | |
1152 | sb->layout != refsb->layout || | |
1153 | sb->chunksize != refsb->chunksize) { | |
1154 | printk(KERN_WARNING "md: %s has strangely different" | |
1155 | " superblock to %s\n", | |
1156 | bdevname(rdev->bdev,b), | |
1157 | bdevname(refdev->bdev,b2)); | |
1158 | return -EINVAL; | |
1159 | } | |
1160 | ev1 = le64_to_cpu(sb->events); | |
1161 | ev2 = le64_to_cpu(refsb->events); | |
1162 | ||
1163 | if (ev1 > ev2) | |
8ed75463 N |
1164 | ret = 1; |
1165 | else | |
1166 | ret = 0; | |
1da177e4 | 1167 | } |
a1801f85 | 1168 | if (minor_version) |
1da177e4 LT |
1169 | rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2; |
1170 | else | |
1171 | rdev->size = rdev->sb_offset; | |
1172 | if (rdev->size < le64_to_cpu(sb->data_size)/2) | |
1173 | return -EINVAL; | |
1174 | rdev->size = le64_to_cpu(sb->data_size)/2; | |
1175 | if (le32_to_cpu(sb->chunksize)) | |
1176 | rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1); | |
2bf071bf | 1177 | |
1c05b4bc | 1178 | if (le64_to_cpu(sb->size) > rdev->size*2) |
2bf071bf | 1179 | return -EINVAL; |
8ed75463 | 1180 | return ret; |
1da177e4 LT |
1181 | } |
1182 | ||
1183 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1184 | { | |
1185 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1186 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1187 | |
41158c7e | 1188 | rdev->raid_disk = -1; |
c5d79adb N |
1189 | clear_bit(Faulty, &rdev->flags); |
1190 | clear_bit(In_sync, &rdev->flags); | |
1191 | clear_bit(WriteMostly, &rdev->flags); | |
1192 | clear_bit(BarriersNotsupp, &rdev->flags); | |
1193 | ||
1da177e4 LT |
1194 | if (mddev->raid_disks == 0) { |
1195 | mddev->major_version = 1; | |
1196 | mddev->patch_version = 0; | |
e691063a | 1197 | mddev->external = 0; |
1da177e4 LT |
1198 | mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9; |
1199 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); | |
1200 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1201 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1202 | mddev->clevel[0] = 0; |
1da177e4 LT |
1203 | mddev->layout = le32_to_cpu(sb->layout); |
1204 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
1205 | mddev->size = le64_to_cpu(sb->size)/2; | |
07d84d10 | 1206 | mddev->events = ev1; |
9223214e | 1207 | mddev->bitmap_offset = 0; |
29fc7e3e | 1208 | mddev->default_bitmap_offset = 1024 >> 9; |
1da177e4 LT |
1209 | |
1210 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1211 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1212 | ||
1213 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1214 | |
71c0805c | 1215 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
e11e93fa | 1216 | mddev->bitmap_file == NULL ) |
a654b9d8 | 1217 | mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); |
e11e93fa | 1218 | |
f6705578 N |
1219 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1220 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1221 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1222 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1223 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
1224 | mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9; | |
1225 | } else { | |
1226 | mddev->reshape_position = MaxSector; | |
1227 | mddev->delta_disks = 0; | |
1228 | mddev->new_level = mddev->level; | |
1229 | mddev->new_layout = mddev->layout; | |
1230 | mddev->new_chunk = mddev->chunk_size; | |
1231 | } | |
1232 | ||
41158c7e N |
1233 | } else if (mddev->pers == NULL) { |
1234 | /* Insist of good event counter while assembling */ | |
1da177e4 LT |
1235 | ++ev1; |
1236 | if (ev1 < mddev->events) | |
1237 | return -EINVAL; | |
41158c7e N |
1238 | } else if (mddev->bitmap) { |
1239 | /* If adding to array with a bitmap, then we can accept an | |
1240 | * older device, but not too old. | |
1241 | */ | |
41158c7e N |
1242 | if (ev1 < mddev->bitmap->events_cleared) |
1243 | return 0; | |
07d84d10 N |
1244 | } else { |
1245 | if (ev1 < mddev->events) | |
1246 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1247 | return 0; | |
1248 | } | |
1da177e4 LT |
1249 | if (mddev->level != LEVEL_MULTIPATH) { |
1250 | int role; | |
1da177e4 LT |
1251 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); |
1252 | switch(role) { | |
1253 | case 0xffff: /* spare */ | |
1da177e4 LT |
1254 | break; |
1255 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1256 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1257 | break; |
1258 | default: | |
5fd6c1dc N |
1259 | if ((le32_to_cpu(sb->feature_map) & |
1260 | MD_FEATURE_RECOVERY_OFFSET)) | |
1261 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1262 | else | |
1263 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1264 | rdev->raid_disk = role; |
1265 | break; | |
1266 | } | |
8ddf9efe N |
1267 | if (sb->devflags & WriteMostly1) |
1268 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1269 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1270 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1271 | |
1da177e4 LT |
1272 | return 0; |
1273 | } | |
1274 | ||
1275 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1276 | { | |
1277 | struct mdp_superblock_1 *sb; | |
1278 | struct list_head *tmp; | |
1279 | mdk_rdev_t *rdev2; | |
1280 | int max_dev, i; | |
1281 | /* make rdev->sb match mddev and rdev data. */ | |
1282 | ||
1283 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1284 | ||
1285 | sb->feature_map = 0; | |
1286 | sb->pad0 = 0; | |
5fd6c1dc | 1287 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1288 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1289 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1290 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1291 | ||
1292 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1293 | sb->events = cpu_to_le64(mddev->events); | |
1294 | if (mddev->in_sync) | |
1295 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1296 | else | |
1297 | sb->resync_offset = cpu_to_le64(0); | |
1298 | ||
1c05b4bc | 1299 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1300 | |
f0ca340c | 1301 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
29fc7e3e | 1302 | sb->size = cpu_to_le64(mddev->size<<1); |
f0ca340c | 1303 | |
a654b9d8 N |
1304 | if (mddev->bitmap && mddev->bitmap_file == NULL) { |
1305 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); | |
71c0805c | 1306 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1307 | } |
5fd6c1dc N |
1308 | |
1309 | if (rdev->raid_disk >= 0 && | |
1310 | !test_bit(In_sync, &rdev->flags) && | |
1311 | rdev->recovery_offset > 0) { | |
1312 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1313 | sb->recovery_offset = cpu_to_le64(rdev->recovery_offset); | |
1314 | } | |
1315 | ||
f6705578 N |
1316 | if (mddev->reshape_position != MaxSector) { |
1317 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1318 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1319 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1320 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1321 | sb->new_level = cpu_to_le32(mddev->new_level); | |
1322 | sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9); | |
1323 | } | |
a654b9d8 | 1324 | |
1da177e4 | 1325 | max_dev = 0; |
d089c6af | 1326 | rdev_for_each(rdev2, tmp, mddev) |
1da177e4 LT |
1327 | if (rdev2->desc_nr+1 > max_dev) |
1328 | max_dev = rdev2->desc_nr+1; | |
a778b73f N |
1329 | |
1330 | if (max_dev > le32_to_cpu(sb->max_dev)) | |
1331 | sb->max_dev = cpu_to_le32(max_dev); | |
1da177e4 LT |
1332 | for (i=0; i<max_dev;i++) |
1333 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1334 | ||
d089c6af | 1335 | rdev_for_each(rdev2, tmp, mddev) { |
1da177e4 | 1336 | i = rdev2->desc_nr; |
b2d444d7 | 1337 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1338 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1339 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1340 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
5fd6c1dc N |
1341 | else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0) |
1342 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); | |
1da177e4 LT |
1343 | else |
1344 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1345 | } | |
1346 | ||
1da177e4 LT |
1347 | sb->sb_csum = calc_sb_1_csum(sb); |
1348 | } | |
1349 | ||
0cd17fec CW |
1350 | static unsigned long long |
1351 | super_1_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size) | |
1352 | { | |
1353 | struct mdp_superblock_1 *sb; | |
1354 | unsigned long long max_size; | |
1355 | if (size && size < rdev->mddev->size) | |
1356 | return 0; /* component must fit device */ | |
1357 | size *= 2; /* convert to sectors */ | |
1358 | if (rdev->sb_offset < rdev->data_offset/2) { | |
1359 | /* minor versions 1 and 2; superblock before data */ | |
1360 | max_size = (rdev->bdev->bd_inode->i_size >> 9); | |
1361 | max_size -= rdev->data_offset; | |
1362 | if (!size || size > max_size) | |
1363 | size = max_size; | |
1364 | } else if (rdev->mddev->bitmap_offset) { | |
1365 | /* minor version 0 with bitmap we can't move */ | |
1366 | return 0; | |
1367 | } else { | |
1368 | /* minor version 0; superblock after data */ | |
1369 | sector_t sb_offset; | |
1370 | sb_offset = (rdev->bdev->bd_inode->i_size >> 9) - 8*2; | |
1371 | sb_offset &= ~(sector_t)(4*2 - 1); | |
1372 | max_size = rdev->size*2 + sb_offset - rdev->sb_offset*2; | |
1373 | if (!size || size > max_size) | |
1374 | size = max_size; | |
1375 | rdev->sb_offset = sb_offset/2; | |
1376 | } | |
1377 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
1378 | sb->data_size = cpu_to_le64(size); | |
1379 | sb->super_offset = rdev->sb_offset*2; | |
1380 | sb->sb_csum = calc_sb_1_csum(sb); | |
1381 | md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size, | |
1382 | rdev->sb_page); | |
1383 | md_super_wait(rdev->mddev); | |
1384 | return size/2; /* kB for sysfs */ | |
1385 | } | |
1da177e4 | 1386 | |
75c96f85 | 1387 | static struct super_type super_types[] = { |
1da177e4 LT |
1388 | [0] = { |
1389 | .name = "0.90.0", | |
1390 | .owner = THIS_MODULE, | |
0cd17fec CW |
1391 | .load_super = super_90_load, |
1392 | .validate_super = super_90_validate, | |
1393 | .sync_super = super_90_sync, | |
1394 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1395 | }, |
1396 | [1] = { | |
1397 | .name = "md-1", | |
1398 | .owner = THIS_MODULE, | |
0cd17fec CW |
1399 | .load_super = super_1_load, |
1400 | .validate_super = super_1_validate, | |
1401 | .sync_super = super_1_sync, | |
1402 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1403 | }, |
1404 | }; | |
1da177e4 LT |
1405 | |
1406 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1407 | { | |
7dd5e7c3 N |
1408 | struct list_head *tmp, *tmp2; |
1409 | mdk_rdev_t *rdev, *rdev2; | |
1da177e4 | 1410 | |
d089c6af N |
1411 | rdev_for_each(rdev, tmp, mddev1) |
1412 | rdev_for_each(rdev2, tmp2, mddev2) | |
7dd5e7c3 N |
1413 | if (rdev->bdev->bd_contains == |
1414 | rdev2->bdev->bd_contains) | |
1415 | return 1; | |
1da177e4 LT |
1416 | |
1417 | return 0; | |
1418 | } | |
1419 | ||
1420 | static LIST_HEAD(pending_raid_disks); | |
1421 | ||
1422 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) | |
1423 | { | |
7dd5e7c3 | 1424 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1425 | struct kobject *ko; |
1edf80d3 | 1426 | char *s; |
5e55e2f5 | 1427 | int err; |
1da177e4 LT |
1428 | |
1429 | if (rdev->mddev) { | |
1430 | MD_BUG(); | |
1431 | return -EINVAL; | |
1432 | } | |
11e2ede0 DW |
1433 | |
1434 | /* prevent duplicates */ | |
1435 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1436 | return -EEXIST; | |
1437 | ||
2bf071bf N |
1438 | /* make sure rdev->size exceeds mddev->size */ |
1439 | if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) { | |
a778b73f N |
1440 | if (mddev->pers) { |
1441 | /* Cannot change size, so fail | |
1442 | * If mddev->level <= 0, then we don't care | |
1443 | * about aligning sizes (e.g. linear) | |
1444 | */ | |
1445 | if (mddev->level > 0) | |
1446 | return -ENOSPC; | |
1447 | } else | |
2bf071bf N |
1448 | mddev->size = rdev->size; |
1449 | } | |
1da177e4 LT |
1450 | |
1451 | /* Verify rdev->desc_nr is unique. | |
1452 | * If it is -1, assign a free number, else | |
1453 | * check number is not in use | |
1454 | */ | |
1455 | if (rdev->desc_nr < 0) { | |
1456 | int choice = 0; | |
1457 | if (mddev->pers) choice = mddev->raid_disks; | |
1458 | while (find_rdev_nr(mddev, choice)) | |
1459 | choice++; | |
1460 | rdev->desc_nr = choice; | |
1461 | } else { | |
1462 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1463 | return -EBUSY; | |
1464 | } | |
19133a42 | 1465 | bdevname(rdev->bdev,b); |
649316b2 | 1466 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1467 | *s = '!'; |
649316b2 | 1468 | |
1da177e4 | 1469 | rdev->mddev = mddev; |
19133a42 | 1470 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1471 | |
b2d6db58 | 1472 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1473 | goto fail; |
86e6ffdd | 1474 | |
f637b9f9 | 1475 | if (rdev->bdev->bd_part) |
edfaa7c3 | 1476 | ko = &rdev->bdev->bd_part->dev.kobj; |
f637b9f9 | 1477 | else |
edfaa7c3 | 1478 | ko = &rdev->bdev->bd_disk->dev.kobj; |
5e55e2f5 N |
1479 | if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) { |
1480 | kobject_del(&rdev->kobj); | |
1481 | goto fail; | |
1482 | } | |
1483 | list_add(&rdev->same_set, &mddev->disks); | |
c5d79adb | 1484 | bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk); |
1da177e4 | 1485 | return 0; |
5e55e2f5 N |
1486 | |
1487 | fail: | |
1488 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1489 | b, mdname(mddev)); | |
1490 | return err; | |
1da177e4 LT |
1491 | } |
1492 | ||
177a99b2 | 1493 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1494 | { |
1495 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1496 | kobject_del(&rdev->kobj); | |
177a99b2 | 1497 | kobject_put(&rdev->kobj); |
5792a285 N |
1498 | } |
1499 | ||
1da177e4 LT |
1500 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1501 | { | |
1502 | char b[BDEVNAME_SIZE]; | |
1503 | if (!rdev->mddev) { | |
1504 | MD_BUG(); | |
1505 | return; | |
1506 | } | |
5463c790 | 1507 | bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk); |
1da177e4 LT |
1508 | list_del_init(&rdev->same_set); |
1509 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); | |
1510 | rdev->mddev = NULL; | |
86e6ffdd | 1511 | sysfs_remove_link(&rdev->kobj, "block"); |
5792a285 N |
1512 | |
1513 | /* We need to delay this, otherwise we can deadlock when | |
1514 | * writing to 'remove' to "dev/state" | |
1515 | */ | |
177a99b2 N |
1516 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1517 | kobject_get(&rdev->kobj); | |
5792a285 | 1518 | schedule_work(&rdev->del_work); |
1da177e4 LT |
1519 | } |
1520 | ||
1521 | /* | |
1522 | * prevent the device from being mounted, repartitioned or | |
1523 | * otherwise reused by a RAID array (or any other kernel | |
1524 | * subsystem), by bd_claiming the device. | |
1525 | */ | |
c5d79adb | 1526 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1527 | { |
1528 | int err = 0; | |
1529 | struct block_device *bdev; | |
1530 | char b[BDEVNAME_SIZE]; | |
1531 | ||
2e7b651d | 1532 | bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); |
1da177e4 LT |
1533 | if (IS_ERR(bdev)) { |
1534 | printk(KERN_ERR "md: could not open %s.\n", | |
1535 | __bdevname(dev, b)); | |
1536 | return PTR_ERR(bdev); | |
1537 | } | |
c5d79adb | 1538 | err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev); |
1da177e4 LT |
1539 | if (err) { |
1540 | printk(KERN_ERR "md: could not bd_claim %s.\n", | |
1541 | bdevname(bdev, b)); | |
2e7b651d | 1542 | blkdev_put(bdev); |
1da177e4 LT |
1543 | return err; |
1544 | } | |
c5d79adb N |
1545 | if (!shared) |
1546 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1547 | rdev->bdev = bdev; |
1548 | return err; | |
1549 | } | |
1550 | ||
1551 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1552 | { | |
1553 | struct block_device *bdev = rdev->bdev; | |
1554 | rdev->bdev = NULL; | |
1555 | if (!bdev) | |
1556 | MD_BUG(); | |
1557 | bd_release(bdev); | |
2e7b651d | 1558 | blkdev_put(bdev); |
1da177e4 LT |
1559 | } |
1560 | ||
1561 | void md_autodetect_dev(dev_t dev); | |
1562 | ||
1563 | static void export_rdev(mdk_rdev_t * rdev) | |
1564 | { | |
1565 | char b[BDEVNAME_SIZE]; | |
1566 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1567 | bdevname(rdev->bdev,b)); | |
1568 | if (rdev->mddev) | |
1569 | MD_BUG(); | |
1570 | free_disk_sb(rdev); | |
1571 | list_del_init(&rdev->same_set); | |
1572 | #ifndef MODULE | |
d0fae18f N |
1573 | if (test_bit(AutoDetected, &rdev->flags)) |
1574 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1575 | #endif |
1576 | unlock_rdev(rdev); | |
86e6ffdd | 1577 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1578 | } |
1579 | ||
1580 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1581 | { | |
1582 | unbind_rdev_from_array(rdev); | |
1583 | export_rdev(rdev); | |
1584 | } | |
1585 | ||
1586 | static void export_array(mddev_t *mddev) | |
1587 | { | |
1588 | struct list_head *tmp; | |
1589 | mdk_rdev_t *rdev; | |
1590 | ||
d089c6af | 1591 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1592 | if (!rdev->mddev) { |
1593 | MD_BUG(); | |
1594 | continue; | |
1595 | } | |
1596 | kick_rdev_from_array(rdev); | |
1597 | } | |
1598 | if (!list_empty(&mddev->disks)) | |
1599 | MD_BUG(); | |
1600 | mddev->raid_disks = 0; | |
1601 | mddev->major_version = 0; | |
1602 | } | |
1603 | ||
1604 | static void print_desc(mdp_disk_t *desc) | |
1605 | { | |
1606 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
1607 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
1608 | } | |
1609 | ||
1610 | static void print_sb(mdp_super_t *sb) | |
1611 | { | |
1612 | int i; | |
1613 | ||
1614 | printk(KERN_INFO | |
1615 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
1616 | sb->major_version, sb->minor_version, sb->patch_version, | |
1617 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
1618 | sb->ctime); | |
1619 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
1620 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
1621 | sb->md_minor, sb->layout, sb->chunk_size); | |
1622 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
1623 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
1624 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
1625 | sb->failed_disks, sb->spare_disks, | |
1626 | sb->sb_csum, (unsigned long)sb->events_lo); | |
1627 | ||
1628 | printk(KERN_INFO); | |
1629 | for (i = 0; i < MD_SB_DISKS; i++) { | |
1630 | mdp_disk_t *desc; | |
1631 | ||
1632 | desc = sb->disks + i; | |
1633 | if (desc->number || desc->major || desc->minor || | |
1634 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
1635 | printk(" D %2d: ", i); | |
1636 | print_desc(desc); | |
1637 | } | |
1638 | } | |
1639 | printk(KERN_INFO "md: THIS: "); | |
1640 | print_desc(&sb->this_disk); | |
1641 | ||
1642 | } | |
1643 | ||
1644 | static void print_rdev(mdk_rdev_t *rdev) | |
1645 | { | |
1646 | char b[BDEVNAME_SIZE]; | |
1647 | printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n", | |
1648 | bdevname(rdev->bdev,b), (unsigned long long)rdev->size, | |
b2d444d7 N |
1649 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
1650 | rdev->desc_nr); | |
1da177e4 LT |
1651 | if (rdev->sb_loaded) { |
1652 | printk(KERN_INFO "md: rdev superblock:\n"); | |
1653 | print_sb((mdp_super_t*)page_address(rdev->sb_page)); | |
1654 | } else | |
1655 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
1656 | } | |
1657 | ||
5e56341d | 1658 | static void md_print_devices(void) |
1da177e4 LT |
1659 | { |
1660 | struct list_head *tmp, *tmp2; | |
1661 | mdk_rdev_t *rdev; | |
1662 | mddev_t *mddev; | |
1663 | char b[BDEVNAME_SIZE]; | |
1664 | ||
1665 | printk("\n"); | |
1666 | printk("md: **********************************\n"); | |
1667 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
1668 | printk("md: **********************************\n"); | |
29ac4aa3 | 1669 | for_each_mddev(mddev, tmp) { |
1da177e4 | 1670 | |
32a7627c N |
1671 | if (mddev->bitmap) |
1672 | bitmap_print_sb(mddev->bitmap); | |
1673 | else | |
1674 | printk("%s: ", mdname(mddev)); | |
d089c6af | 1675 | rdev_for_each(rdev, tmp2, mddev) |
1da177e4 LT |
1676 | printk("<%s>", bdevname(rdev->bdev,b)); |
1677 | printk("\n"); | |
1678 | ||
d089c6af | 1679 | rdev_for_each(rdev, tmp2, mddev) |
1da177e4 LT |
1680 | print_rdev(rdev); |
1681 | } | |
1682 | printk("md: **********************************\n"); | |
1683 | printk("\n"); | |
1684 | } | |
1685 | ||
1686 | ||
42543769 | 1687 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 1688 | { |
42543769 N |
1689 | /* Update each superblock (in-memory image), but |
1690 | * if we are allowed to, skip spares which already | |
1691 | * have the right event counter, or have one earlier | |
1692 | * (which would mean they aren't being marked as dirty | |
1693 | * with the rest of the array) | |
1694 | */ | |
1da177e4 LT |
1695 | mdk_rdev_t *rdev; |
1696 | struct list_head *tmp; | |
1697 | ||
d089c6af | 1698 | rdev_for_each(rdev, tmp, mddev) { |
42543769 N |
1699 | if (rdev->sb_events == mddev->events || |
1700 | (nospares && | |
1701 | rdev->raid_disk < 0 && | |
1702 | (rdev->sb_events&1)==0 && | |
1703 | rdev->sb_events+1 == mddev->events)) { | |
1704 | /* Don't update this superblock */ | |
1705 | rdev->sb_loaded = 2; | |
1706 | } else { | |
1707 | super_types[mddev->major_version]. | |
1708 | sync_super(mddev, rdev); | |
1709 | rdev->sb_loaded = 1; | |
1710 | } | |
1da177e4 LT |
1711 | } |
1712 | } | |
1713 | ||
850b2b42 | 1714 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 1715 | { |
1da177e4 LT |
1716 | struct list_head *tmp; |
1717 | mdk_rdev_t *rdev; | |
06d91a5f | 1718 | int sync_req; |
42543769 | 1719 | int nospares = 0; |
1da177e4 | 1720 | |
8377bc80 N |
1721 | if (mddev->external) |
1722 | return; | |
1da177e4 | 1723 | repeat: |
a9701a30 | 1724 | spin_lock_irq(&mddev->write_lock); |
84692195 | 1725 | |
850b2b42 N |
1726 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
1727 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) | |
1728 | force_change = 1; | |
1729 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
1730 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
1731 | * though if events isn't the right even/odd, we will have to do | |
1732 | * spares after all | |
1733 | */ | |
1734 | nospares = 1; | |
1735 | if (force_change) | |
1736 | nospares = 0; | |
1737 | if (mddev->degraded) | |
84692195 N |
1738 | /* If the array is degraded, then skipping spares is both |
1739 | * dangerous and fairly pointless. | |
1740 | * Dangerous because a device that was removed from the array | |
1741 | * might have a event_count that still looks up-to-date, | |
1742 | * so it can be re-added without a resync. | |
1743 | * Pointless because if there are any spares to skip, | |
1744 | * then a recovery will happen and soon that array won't | |
1745 | * be degraded any more and the spare can go back to sleep then. | |
1746 | */ | |
850b2b42 | 1747 | nospares = 0; |
84692195 | 1748 | |
06d91a5f | 1749 | sync_req = mddev->in_sync; |
1da177e4 | 1750 | mddev->utime = get_seconds(); |
42543769 N |
1751 | |
1752 | /* If this is just a dirty<->clean transition, and the array is clean | |
1753 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 1754 | if (nospares |
42543769 | 1755 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
1031be7a N |
1756 | && (mddev->events & 1) |
1757 | && mddev->events != 1) | |
42543769 N |
1758 | mddev->events--; |
1759 | else { | |
1760 | /* otherwise we have to go forward and ... */ | |
1761 | mddev->events ++; | |
1762 | if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */ | |
1763 | /* .. if the array isn't clean, insist on an odd 'events' */ | |
1764 | if ((mddev->events&1)==0) { | |
1765 | mddev->events++; | |
1766 | nospares = 0; | |
1767 | } | |
1768 | } else { | |
1769 | /* otherwise insist on an even 'events' (for clean states) */ | |
1770 | if ((mddev->events&1)) { | |
1771 | mddev->events++; | |
1772 | nospares = 0; | |
1773 | } | |
1774 | } | |
1775 | } | |
1da177e4 LT |
1776 | |
1777 | if (!mddev->events) { | |
1778 | /* | |
1779 | * oops, this 64-bit counter should never wrap. | |
1780 | * Either we are in around ~1 trillion A.C., assuming | |
1781 | * 1 reboot per second, or we have a bug: | |
1782 | */ | |
1783 | MD_BUG(); | |
1784 | mddev->events --; | |
1785 | } | |
1da177e4 LT |
1786 | |
1787 | /* | |
1788 | * do not write anything to disk if using | |
1789 | * nonpersistent superblocks | |
1790 | */ | |
06d91a5f | 1791 | if (!mddev->persistent) { |
e691063a N |
1792 | if (!mddev->external) |
1793 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
1794 | ||
a9701a30 | 1795 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1796 | wake_up(&mddev->sb_wait); |
1da177e4 | 1797 | return; |
06d91a5f | 1798 | } |
e691063a | 1799 | sync_sbs(mddev, nospares); |
a9701a30 | 1800 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
1801 | |
1802 | dprintk(KERN_INFO | |
1803 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
1804 | mdname(mddev),mddev->in_sync); | |
1805 | ||
4ad13663 | 1806 | bitmap_update_sb(mddev->bitmap); |
d089c6af | 1807 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1808 | char b[BDEVNAME_SIZE]; |
1809 | dprintk(KERN_INFO "md: "); | |
42543769 N |
1810 | if (rdev->sb_loaded != 1) |
1811 | continue; /* no noise on spare devices */ | |
b2d444d7 | 1812 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
1813 | dprintk("(skipping faulty "); |
1814 | ||
1815 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 1816 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 1817 | md_super_write(mddev,rdev, |
0002b271 | 1818 | rdev->sb_offset<<1, rdev->sb_size, |
7bfa19f2 N |
1819 | rdev->sb_page); |
1820 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
1821 | bdevname(rdev->bdev,b), | |
1822 | (unsigned long long)rdev->sb_offset); | |
42543769 | 1823 | rdev->sb_events = mddev->events; |
7bfa19f2 | 1824 | |
1da177e4 LT |
1825 | } else |
1826 | dprintk(")\n"); | |
7bfa19f2 | 1827 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
1828 | /* only need to write one superblock... */ |
1829 | break; | |
1830 | } | |
a9701a30 | 1831 | md_super_wait(mddev); |
850b2b42 | 1832 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 1833 | |
a9701a30 | 1834 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
1835 | if (mddev->in_sync != sync_req || |
1836 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 1837 | /* have to write it out again */ |
a9701a30 | 1838 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
1839 | goto repeat; |
1840 | } | |
850b2b42 | 1841 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 1842 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 1843 | wake_up(&mddev->sb_wait); |
06d91a5f | 1844 | |
1da177e4 LT |
1845 | } |
1846 | ||
7f6ce769 | 1847 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
1848 | * We want to accept with case. For this we use cmd_match. |
1849 | */ | |
1850 | static int cmd_match(const char *cmd, const char *str) | |
1851 | { | |
1852 | /* See if cmd, written into a sysfs file, matches | |
1853 | * str. They must either be the same, or cmd can | |
1854 | * have a trailing newline | |
1855 | */ | |
1856 | while (*cmd && *str && *cmd == *str) { | |
1857 | cmd++; | |
1858 | str++; | |
1859 | } | |
1860 | if (*cmd == '\n') | |
1861 | cmd++; | |
1862 | if (*str || *cmd) | |
1863 | return 0; | |
1864 | return 1; | |
1865 | } | |
1866 | ||
86e6ffdd N |
1867 | struct rdev_sysfs_entry { |
1868 | struct attribute attr; | |
1869 | ssize_t (*show)(mdk_rdev_t *, char *); | |
1870 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
1871 | }; | |
1872 | ||
1873 | static ssize_t | |
96de1e66 | 1874 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
1875 | { |
1876 | char *sep = ""; | |
20a49ff6 | 1877 | size_t len = 0; |
86e6ffdd | 1878 | |
b2d444d7 | 1879 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
1880 | len+= sprintf(page+len, "%sfaulty",sep); |
1881 | sep = ","; | |
1882 | } | |
b2d444d7 | 1883 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
1884 | len += sprintf(page+len, "%sin_sync",sep); |
1885 | sep = ","; | |
1886 | } | |
f655675b N |
1887 | if (test_bit(WriteMostly, &rdev->flags)) { |
1888 | len += sprintf(page+len, "%swrite_mostly",sep); | |
1889 | sep = ","; | |
1890 | } | |
6bfe0b49 DW |
1891 | if (test_bit(Blocked, &rdev->flags)) { |
1892 | len += sprintf(page+len, "%sblocked", sep); | |
1893 | sep = ","; | |
1894 | } | |
b2d444d7 N |
1895 | if (!test_bit(Faulty, &rdev->flags) && |
1896 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
1897 | len += sprintf(page+len, "%sspare", sep); |
1898 | sep = ","; | |
1899 | } | |
1900 | return len+sprintf(page+len, "\n"); | |
1901 | } | |
1902 | ||
45dc2de1 N |
1903 | static ssize_t |
1904 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1905 | { | |
1906 | /* can write | |
1907 | * faulty - simulates and error | |
1908 | * remove - disconnects the device | |
f655675b N |
1909 | * writemostly - sets write_mostly |
1910 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
1911 | * blocked - sets the Blocked flag |
1912 | * -blocked - clears the Blocked flag | |
45dc2de1 N |
1913 | */ |
1914 | int err = -EINVAL; | |
1915 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
1916 | md_error(rdev->mddev, rdev); | |
1917 | err = 0; | |
1918 | } else if (cmd_match(buf, "remove")) { | |
1919 | if (rdev->raid_disk >= 0) | |
1920 | err = -EBUSY; | |
1921 | else { | |
1922 | mddev_t *mddev = rdev->mddev; | |
1923 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
1924 | if (mddev->pers) |
1925 | md_update_sb(mddev, 1); | |
45dc2de1 N |
1926 | md_new_event(mddev); |
1927 | err = 0; | |
1928 | } | |
f655675b N |
1929 | } else if (cmd_match(buf, "writemostly")) { |
1930 | set_bit(WriteMostly, &rdev->flags); | |
1931 | err = 0; | |
1932 | } else if (cmd_match(buf, "-writemostly")) { | |
1933 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
1934 | err = 0; |
1935 | } else if (cmd_match(buf, "blocked")) { | |
1936 | set_bit(Blocked, &rdev->flags); | |
1937 | err = 0; | |
1938 | } else if (cmd_match(buf, "-blocked")) { | |
1939 | clear_bit(Blocked, &rdev->flags); | |
1940 | wake_up(&rdev->blocked_wait); | |
1941 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
1942 | md_wakeup_thread(rdev->mddev->thread); | |
1943 | ||
f655675b | 1944 | err = 0; |
45dc2de1 | 1945 | } |
52664732 NB |
1946 | if (!err) |
1947 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
45dc2de1 N |
1948 | return err ? err : len; |
1949 | } | |
80ca3a44 N |
1950 | static struct rdev_sysfs_entry rdev_state = |
1951 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 1952 | |
4dbcdc75 N |
1953 | static ssize_t |
1954 | errors_show(mdk_rdev_t *rdev, char *page) | |
1955 | { | |
1956 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
1957 | } | |
1958 | ||
1959 | static ssize_t | |
1960 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1961 | { | |
1962 | char *e; | |
1963 | unsigned long n = simple_strtoul(buf, &e, 10); | |
1964 | if (*buf && (*e == 0 || *e == '\n')) { | |
1965 | atomic_set(&rdev->corrected_errors, n); | |
1966 | return len; | |
1967 | } | |
1968 | return -EINVAL; | |
1969 | } | |
1970 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 1971 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 1972 | |
014236d2 N |
1973 | static ssize_t |
1974 | slot_show(mdk_rdev_t *rdev, char *page) | |
1975 | { | |
1976 | if (rdev->raid_disk < 0) | |
1977 | return sprintf(page, "none\n"); | |
1978 | else | |
1979 | return sprintf(page, "%d\n", rdev->raid_disk); | |
1980 | } | |
1981 | ||
1982 | static ssize_t | |
1983 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
1984 | { | |
1985 | char *e; | |
c303da6d N |
1986 | int err; |
1987 | char nm[20]; | |
014236d2 N |
1988 | int slot = simple_strtoul(buf, &e, 10); |
1989 | if (strncmp(buf, "none", 4)==0) | |
1990 | slot = -1; | |
1991 | else if (e==buf || (*e && *e!= '\n')) | |
1992 | return -EINVAL; | |
6c2fce2e | 1993 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
1994 | /* Setting 'slot' on an active array requires also |
1995 | * updating the 'rd%d' link, and communicating | |
1996 | * with the personality with ->hot_*_disk. | |
1997 | * For now we only support removing | |
1998 | * failed/spare devices. This normally happens automatically, | |
1999 | * but not when the metadata is externally managed. | |
2000 | */ | |
c303da6d N |
2001 | if (rdev->raid_disk == -1) |
2002 | return -EEXIST; | |
2003 | /* personality does all needed checks */ | |
2004 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2005 | return -EINVAL; | |
2006 | err = rdev->mddev->pers-> | |
2007 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2008 | if (err) | |
2009 | return err; | |
2010 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2011 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
2012 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2013 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2014 | } else if (rdev->mddev->pers) { |
2015 | mdk_rdev_t *rdev2; | |
2016 | struct list_head *tmp; | |
2017 | /* Activating a spare .. or possibly reactivating | |
2018 | * if we every get bitmaps working here. | |
2019 | */ | |
2020 | ||
2021 | if (rdev->raid_disk != -1) | |
2022 | return -EBUSY; | |
2023 | ||
2024 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2025 | return -EINVAL; | |
2026 | ||
2027 | rdev_for_each(rdev2, tmp, rdev->mddev) | |
2028 | if (rdev2->raid_disk == slot) | |
2029 | return -EEXIST; | |
2030 | ||
2031 | rdev->raid_disk = slot; | |
2032 | if (test_bit(In_sync, &rdev->flags)) | |
2033 | rdev->saved_raid_disk = slot; | |
2034 | else | |
2035 | rdev->saved_raid_disk = -1; | |
2036 | err = rdev->mddev->pers-> | |
2037 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2038 | if (err) { |
6c2fce2e | 2039 | rdev->raid_disk = -1; |
6c2fce2e | 2040 | return err; |
52664732 NB |
2041 | } else |
2042 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
6c2fce2e NB |
2043 | sprintf(nm, "rd%d", rdev->raid_disk); |
2044 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
2045 | printk(KERN_WARNING | |
2046 | "md: cannot register " | |
2047 | "%s for %s\n", | |
2048 | nm, mdname(rdev->mddev)); | |
2049 | ||
2050 | /* don't wakeup anyone, leave that to userspace. */ | |
c303da6d N |
2051 | } else { |
2052 | if (slot >= rdev->mddev->raid_disks) | |
2053 | return -ENOSPC; | |
2054 | rdev->raid_disk = slot; | |
2055 | /* assume it is working */ | |
c5d79adb N |
2056 | clear_bit(Faulty, &rdev->flags); |
2057 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2058 | set_bit(In_sync, &rdev->flags); |
52664732 | 2059 | sysfs_notify(&rdev->kobj, NULL, "state"); |
c303da6d | 2060 | } |
014236d2 N |
2061 | return len; |
2062 | } | |
2063 | ||
2064 | ||
2065 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2066 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2067 | |
93c8cad0 N |
2068 | static ssize_t |
2069 | offset_show(mdk_rdev_t *rdev, char *page) | |
2070 | { | |
6961ece4 | 2071 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2072 | } |
2073 | ||
2074 | static ssize_t | |
2075 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2076 | { | |
2077 | char *e; | |
2078 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2079 | if (e==buf || (*e && *e != '\n')) | |
2080 | return -EINVAL; | |
8ed0a521 | 2081 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2082 | return -EBUSY; |
c5d79adb N |
2083 | if (rdev->size && rdev->mddev->external) |
2084 | /* Must set offset before size, so overlap checks | |
2085 | * can be sane */ | |
2086 | return -EBUSY; | |
93c8cad0 N |
2087 | rdev->data_offset = offset; |
2088 | return len; | |
2089 | } | |
2090 | ||
2091 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2092 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2093 | |
83303b61 N |
2094 | static ssize_t |
2095 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2096 | { | |
2097 | return sprintf(page, "%llu\n", (unsigned long long)rdev->size); | |
2098 | } | |
2099 | ||
c5d79adb N |
2100 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2101 | { | |
2102 | /* check if two start/length pairs overlap */ | |
2103 | if (s1+l1 <= s2) | |
2104 | return 0; | |
2105 | if (s2+l2 <= s1) | |
2106 | return 0; | |
2107 | return 1; | |
2108 | } | |
2109 | ||
83303b61 N |
2110 | static ssize_t |
2111 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2112 | { | |
2113 | char *e; | |
2114 | unsigned long long size = simple_strtoull(buf, &e, 10); | |
c5d79adb | 2115 | unsigned long long oldsize = rdev->size; |
27c529bb N |
2116 | mddev_t *my_mddev = rdev->mddev; |
2117 | ||
83303b61 N |
2118 | if (e==buf || (*e && *e != '\n')) |
2119 | return -EINVAL; | |
0cd17fec CW |
2120 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
2121 | if (rdev->mddev->persistent) { | |
2122 | size = super_types[rdev->mddev->major_version]. | |
2123 | rdev_size_change(rdev, size); | |
2124 | if (!size) | |
2125 | return -EBUSY; | |
2126 | } else if (!size) { | |
2127 | size = (rdev->bdev->bd_inode->i_size >> 10); | |
2128 | size -= rdev->data_offset/2; | |
2129 | } | |
2130 | if (size < rdev->mddev->size) | |
2131 | return -EINVAL; /* component must fit device */ | |
2132 | } | |
2133 | ||
83303b61 | 2134 | rdev->size = size; |
c5d79adb N |
2135 | if (size > oldsize && rdev->mddev->external) { |
2136 | /* need to check that all other rdevs with the same ->bdev | |
2137 | * do not overlap. We need to unlock the mddev to avoid | |
2138 | * a deadlock. We have already changed rdev->size, and if | |
2139 | * we have to change it back, we will have the lock again. | |
2140 | */ | |
2141 | mddev_t *mddev; | |
2142 | int overlap = 0; | |
2143 | struct list_head *tmp, *tmp2; | |
2144 | ||
27c529bb | 2145 | mddev_unlock(my_mddev); |
29ac4aa3 | 2146 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2147 | mdk_rdev_t *rdev2; |
2148 | ||
2149 | mddev_lock(mddev); | |
d089c6af | 2150 | rdev_for_each(rdev2, tmp2, mddev) |
c5d79adb N |
2151 | if (test_bit(AllReserved, &rdev2->flags) || |
2152 | (rdev->bdev == rdev2->bdev && | |
2153 | rdev != rdev2 && | |
2154 | overlaps(rdev->data_offset, rdev->size, | |
2155 | rdev2->data_offset, rdev2->size))) { | |
2156 | overlap = 1; | |
2157 | break; | |
2158 | } | |
2159 | mddev_unlock(mddev); | |
2160 | if (overlap) { | |
2161 | mddev_put(mddev); | |
2162 | break; | |
2163 | } | |
2164 | } | |
27c529bb | 2165 | mddev_lock(my_mddev); |
c5d79adb N |
2166 | if (overlap) { |
2167 | /* Someone else could have slipped in a size | |
2168 | * change here, but doing so is just silly. | |
2169 | * We put oldsize back because we *know* it is | |
2170 | * safe, and trust userspace not to race with | |
2171 | * itself | |
2172 | */ | |
2173 | rdev->size = oldsize; | |
2174 | return -EBUSY; | |
2175 | } | |
2176 | } | |
27c529bb N |
2177 | if (size < my_mddev->size || my_mddev->size == 0) |
2178 | my_mddev->size = size; | |
83303b61 N |
2179 | return len; |
2180 | } | |
2181 | ||
2182 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2183 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2184 | |
86e6ffdd N |
2185 | static struct attribute *rdev_default_attrs[] = { |
2186 | &rdev_state.attr, | |
4dbcdc75 | 2187 | &rdev_errors.attr, |
014236d2 | 2188 | &rdev_slot.attr, |
93c8cad0 | 2189 | &rdev_offset.attr, |
83303b61 | 2190 | &rdev_size.attr, |
86e6ffdd N |
2191 | NULL, |
2192 | }; | |
2193 | static ssize_t | |
2194 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2195 | { | |
2196 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2197 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2198 | mddev_t *mddev = rdev->mddev; |
2199 | ssize_t rv; | |
86e6ffdd N |
2200 | |
2201 | if (!entry->show) | |
2202 | return -EIO; | |
27c529bb N |
2203 | |
2204 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2205 | if (!rv) { | |
2206 | if (rdev->mddev == NULL) | |
2207 | rv = -EBUSY; | |
2208 | else | |
2209 | rv = entry->show(rdev, page); | |
2210 | mddev_unlock(mddev); | |
2211 | } | |
2212 | return rv; | |
86e6ffdd N |
2213 | } |
2214 | ||
2215 | static ssize_t | |
2216 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2217 | const char *page, size_t length) | |
2218 | { | |
2219 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2220 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2221 | ssize_t rv; |
2222 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2223 | |
2224 | if (!entry->store) | |
2225 | return -EIO; | |
67463acb N |
2226 | if (!capable(CAP_SYS_ADMIN)) |
2227 | return -EACCES; | |
27c529bb | 2228 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2229 | if (!rv) { |
27c529bb N |
2230 | if (rdev->mddev == NULL) |
2231 | rv = -EBUSY; | |
2232 | else | |
2233 | rv = entry->store(rdev, page, length); | |
6a51830e | 2234 | mddev_unlock(mddev); |
ca388059 N |
2235 | } |
2236 | return rv; | |
86e6ffdd N |
2237 | } |
2238 | ||
2239 | static void rdev_free(struct kobject *ko) | |
2240 | { | |
2241 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2242 | kfree(rdev); | |
2243 | } | |
2244 | static struct sysfs_ops rdev_sysfs_ops = { | |
2245 | .show = rdev_attr_show, | |
2246 | .store = rdev_attr_store, | |
2247 | }; | |
2248 | static struct kobj_type rdev_ktype = { | |
2249 | .release = rdev_free, | |
2250 | .sysfs_ops = &rdev_sysfs_ops, | |
2251 | .default_attrs = rdev_default_attrs, | |
2252 | }; | |
2253 | ||
1da177e4 LT |
2254 | /* |
2255 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2256 | * | |
2257 | * mark the device faulty if: | |
2258 | * | |
2259 | * - the device is nonexistent (zero size) | |
2260 | * - the device has no valid superblock | |
2261 | * | |
2262 | * a faulty rdev _never_ has rdev->sb set. | |
2263 | */ | |
2264 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2265 | { | |
2266 | char b[BDEVNAME_SIZE]; | |
2267 | int err; | |
2268 | mdk_rdev_t *rdev; | |
2269 | sector_t size; | |
2270 | ||
9ffae0cf | 2271 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2272 | if (!rdev) { |
2273 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2274 | return ERR_PTR(-ENOMEM); | |
2275 | } | |
1da177e4 LT |
2276 | |
2277 | if ((err = alloc_disk_sb(rdev))) | |
2278 | goto abort_free; | |
2279 | ||
c5d79adb | 2280 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2281 | if (err) |
2282 | goto abort_free; | |
2283 | ||
f9cb074b | 2284 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2285 | |
1da177e4 | 2286 | rdev->desc_nr = -1; |
2b6e8459 | 2287 | rdev->saved_raid_disk = -1; |
3f9d7b0d | 2288 | rdev->raid_disk = -1; |
b2d444d7 | 2289 | rdev->flags = 0; |
1da177e4 | 2290 | rdev->data_offset = 0; |
42543769 | 2291 | rdev->sb_events = 0; |
1da177e4 | 2292 | atomic_set(&rdev->nr_pending, 0); |
ba22dcbf | 2293 | atomic_set(&rdev->read_errors, 0); |
4dbcdc75 | 2294 | atomic_set(&rdev->corrected_errors, 0); |
1da177e4 LT |
2295 | |
2296 | size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
2297 | if (!size) { | |
2298 | printk(KERN_WARNING | |
2299 | "md: %s has zero or unknown size, marking faulty!\n", | |
2300 | bdevname(rdev->bdev,b)); | |
2301 | err = -EINVAL; | |
2302 | goto abort_free; | |
2303 | } | |
2304 | ||
2305 | if (super_format >= 0) { | |
2306 | err = super_types[super_format]. | |
2307 | load_super(rdev, NULL, super_minor); | |
2308 | if (err == -EINVAL) { | |
df968c4e N |
2309 | printk(KERN_WARNING |
2310 | "md: %s does not have a valid v%d.%d " | |
2311 | "superblock, not importing!\n", | |
2312 | bdevname(rdev->bdev,b), | |
2313 | super_format, super_minor); | |
1da177e4 LT |
2314 | goto abort_free; |
2315 | } | |
2316 | if (err < 0) { | |
2317 | printk(KERN_WARNING | |
2318 | "md: could not read %s's sb, not importing!\n", | |
2319 | bdevname(rdev->bdev,b)); | |
2320 | goto abort_free; | |
2321 | } | |
2322 | } | |
6bfe0b49 | 2323 | |
1da177e4 | 2324 | INIT_LIST_HEAD(&rdev->same_set); |
6bfe0b49 | 2325 | init_waitqueue_head(&rdev->blocked_wait); |
1da177e4 LT |
2326 | |
2327 | return rdev; | |
2328 | ||
2329 | abort_free: | |
2330 | if (rdev->sb_page) { | |
2331 | if (rdev->bdev) | |
2332 | unlock_rdev(rdev); | |
2333 | free_disk_sb(rdev); | |
2334 | } | |
2335 | kfree(rdev); | |
2336 | return ERR_PTR(err); | |
2337 | } | |
2338 | ||
2339 | /* | |
2340 | * Check a full RAID array for plausibility | |
2341 | */ | |
2342 | ||
2343 | ||
a757e64c | 2344 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2345 | { |
2346 | int i; | |
2347 | struct list_head *tmp; | |
2348 | mdk_rdev_t *rdev, *freshest; | |
2349 | char b[BDEVNAME_SIZE]; | |
2350 | ||
2351 | freshest = NULL; | |
d089c6af | 2352 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2353 | switch (super_types[mddev->major_version]. |
2354 | load_super(rdev, freshest, mddev->minor_version)) { | |
2355 | case 1: | |
2356 | freshest = rdev; | |
2357 | break; | |
2358 | case 0: | |
2359 | break; | |
2360 | default: | |
2361 | printk( KERN_ERR \ | |
2362 | "md: fatal superblock inconsistency in %s" | |
2363 | " -- removing from array\n", | |
2364 | bdevname(rdev->bdev,b)); | |
2365 | kick_rdev_from_array(rdev); | |
2366 | } | |
2367 | ||
2368 | ||
2369 | super_types[mddev->major_version]. | |
2370 | validate_super(mddev, freshest); | |
2371 | ||
2372 | i = 0; | |
d089c6af | 2373 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
2374 | if (rdev != freshest) |
2375 | if (super_types[mddev->major_version]. | |
2376 | validate_super(mddev, rdev)) { | |
2377 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2378 | " from array!\n", | |
2379 | bdevname(rdev->bdev,b)); | |
2380 | kick_rdev_from_array(rdev); | |
2381 | continue; | |
2382 | } | |
2383 | if (mddev->level == LEVEL_MULTIPATH) { | |
2384 | rdev->desc_nr = i++; | |
2385 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2386 | set_bit(In_sync, &rdev->flags); |
a778b73f N |
2387 | } else if (rdev->raid_disk >= mddev->raid_disks) { |
2388 | rdev->raid_disk = -1; | |
2389 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2390 | } |
2391 | } | |
2392 | ||
2393 | ||
2394 | ||
2395 | if (mddev->recovery_cp != MaxSector && | |
2396 | mddev->level >= 1) | |
2397 | printk(KERN_ERR "md: %s: raid array is not clean" | |
2398 | " -- starting background reconstruction\n", | |
2399 | mdname(mddev)); | |
2400 | ||
1da177e4 LT |
2401 | } |
2402 | ||
16f17b39 N |
2403 | static ssize_t |
2404 | safe_delay_show(mddev_t *mddev, char *page) | |
2405 | { | |
2406 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2407 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2408 | } | |
2409 | static ssize_t | |
2410 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2411 | { | |
2412 | int scale=1; | |
2413 | int dot=0; | |
2414 | int i; | |
2415 | unsigned long msec; | |
2416 | char buf[30]; | |
2417 | char *e; | |
2418 | /* remove a period, and count digits after it */ | |
2419 | if (len >= sizeof(buf)) | |
2420 | return -EINVAL; | |
2421 | strlcpy(buf, cbuf, len); | |
2422 | buf[len] = 0; | |
2423 | for (i=0; i<len; i++) { | |
2424 | if (dot) { | |
2425 | if (isdigit(buf[i])) { | |
2426 | buf[i-1] = buf[i]; | |
2427 | scale *= 10; | |
2428 | } | |
2429 | buf[i] = 0; | |
2430 | } else if (buf[i] == '.') { | |
2431 | dot=1; | |
2432 | buf[i] = 0; | |
2433 | } | |
2434 | } | |
2435 | msec = simple_strtoul(buf, &e, 10); | |
2436 | if (e == buf || (*e && *e != '\n')) | |
2437 | return -EINVAL; | |
2438 | msec = (msec * 1000) / scale; | |
2439 | if (msec == 0) | |
2440 | mddev->safemode_delay = 0; | |
2441 | else { | |
2442 | mddev->safemode_delay = (msec*HZ)/1000; | |
2443 | if (mddev->safemode_delay == 0) | |
2444 | mddev->safemode_delay = 1; | |
2445 | } | |
2446 | return len; | |
2447 | } | |
2448 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2449 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2450 | |
eae1701f | 2451 | static ssize_t |
96de1e66 | 2452 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2453 | { |
2604b703 | 2454 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2455 | if (p) |
eae1701f | 2456 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2457 | else if (mddev->clevel[0]) |
2458 | return sprintf(page, "%s\n", mddev->clevel); | |
2459 | else if (mddev->level != LEVEL_NONE) | |
2460 | return sprintf(page, "%d\n", mddev->level); | |
2461 | else | |
2462 | return 0; | |
eae1701f N |
2463 | } |
2464 | ||
d9d166c2 N |
2465 | static ssize_t |
2466 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2467 | { | |
20a49ff6 | 2468 | ssize_t rv = len; |
d9d166c2 N |
2469 | if (mddev->pers) |
2470 | return -EBUSY; | |
2471 | if (len == 0) | |
2472 | return 0; | |
2473 | if (len >= sizeof(mddev->clevel)) | |
2474 | return -ENOSPC; | |
2475 | strncpy(mddev->clevel, buf, len); | |
2476 | if (mddev->clevel[len-1] == '\n') | |
2477 | len--; | |
2478 | mddev->clevel[len] = 0; | |
2479 | mddev->level = LEVEL_NONE; | |
2480 | return rv; | |
2481 | } | |
2482 | ||
2483 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 2484 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 2485 | |
d4dbd025 N |
2486 | |
2487 | static ssize_t | |
2488 | layout_show(mddev_t *mddev, char *page) | |
2489 | { | |
2490 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
2491 | if (mddev->reshape_position != MaxSector && |
2492 | mddev->layout != mddev->new_layout) | |
2493 | return sprintf(page, "%d (%d)\n", | |
2494 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
2495 | return sprintf(page, "%d\n", mddev->layout); |
2496 | } | |
2497 | ||
2498 | static ssize_t | |
2499 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
2500 | { | |
2501 | char *e; | |
2502 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
2503 | |
2504 | if (!*buf || (*e && *e != '\n')) | |
2505 | return -EINVAL; | |
2506 | ||
08a02ecd N |
2507 | if (mddev->pers) |
2508 | return -EBUSY; | |
2509 | if (mddev->reshape_position != MaxSector) | |
2510 | mddev->new_layout = n; | |
2511 | else | |
2512 | mddev->layout = n; | |
d4dbd025 N |
2513 | return len; |
2514 | } | |
2515 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 2516 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
2517 | |
2518 | ||
eae1701f | 2519 | static ssize_t |
96de1e66 | 2520 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 2521 | { |
bb636547 N |
2522 | if (mddev->raid_disks == 0) |
2523 | return 0; | |
08a02ecd N |
2524 | if (mddev->reshape_position != MaxSector && |
2525 | mddev->delta_disks != 0) | |
2526 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
2527 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
2528 | return sprintf(page, "%d\n", mddev->raid_disks); |
2529 | } | |
2530 | ||
da943b99 N |
2531 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
2532 | ||
2533 | static ssize_t | |
2534 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
2535 | { | |
da943b99 N |
2536 | char *e; |
2537 | int rv = 0; | |
2538 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2539 | ||
2540 | if (!*buf || (*e && *e != '\n')) | |
2541 | return -EINVAL; | |
2542 | ||
2543 | if (mddev->pers) | |
2544 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
2545 | else if (mddev->reshape_position != MaxSector) { |
2546 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
2547 | mddev->delta_disks = n - olddisks; | |
2548 | mddev->raid_disks = n; | |
2549 | } else | |
da943b99 N |
2550 | mddev->raid_disks = n; |
2551 | return rv ? rv : len; | |
2552 | } | |
2553 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 2554 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 2555 | |
3b34380a N |
2556 | static ssize_t |
2557 | chunk_size_show(mddev_t *mddev, char *page) | |
2558 | { | |
08a02ecd N |
2559 | if (mddev->reshape_position != MaxSector && |
2560 | mddev->chunk_size != mddev->new_chunk) | |
2561 | return sprintf(page, "%d (%d)\n", mddev->new_chunk, | |
2562 | mddev->chunk_size); | |
3b34380a N |
2563 | return sprintf(page, "%d\n", mddev->chunk_size); |
2564 | } | |
2565 | ||
2566 | static ssize_t | |
2567 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
2568 | { | |
2569 | /* can only set chunk_size if array is not yet active */ | |
2570 | char *e; | |
2571 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2572 | ||
3b34380a N |
2573 | if (!*buf || (*e && *e != '\n')) |
2574 | return -EINVAL; | |
2575 | ||
08a02ecd N |
2576 | if (mddev->pers) |
2577 | return -EBUSY; | |
2578 | else if (mddev->reshape_position != MaxSector) | |
2579 | mddev->new_chunk = n; | |
2580 | else | |
2581 | mddev->chunk_size = n; | |
3b34380a N |
2582 | return len; |
2583 | } | |
2584 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 2585 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 2586 | |
a94213b1 N |
2587 | static ssize_t |
2588 | resync_start_show(mddev_t *mddev, char *page) | |
2589 | { | |
2590 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); | |
2591 | } | |
2592 | ||
2593 | static ssize_t | |
2594 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
2595 | { | |
a94213b1 N |
2596 | char *e; |
2597 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
2598 | ||
2599 | if (mddev->pers) | |
2600 | return -EBUSY; | |
2601 | if (!*buf || (*e && *e != '\n')) | |
2602 | return -EINVAL; | |
2603 | ||
2604 | mddev->recovery_cp = n; | |
2605 | return len; | |
2606 | } | |
2607 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 2608 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 2609 | |
9e653b63 N |
2610 | /* |
2611 | * The array state can be: | |
2612 | * | |
2613 | * clear | |
2614 | * No devices, no size, no level | |
2615 | * Equivalent to STOP_ARRAY ioctl | |
2616 | * inactive | |
2617 | * May have some settings, but array is not active | |
2618 | * all IO results in error | |
2619 | * When written, doesn't tear down array, but just stops it | |
2620 | * suspended (not supported yet) | |
2621 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 2622 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
2623 | * readonly |
2624 | * no resync can happen. no superblocks get written. | |
2625 | * write requests fail | |
2626 | * read-auto | |
2627 | * like readonly, but behaves like 'clean' on a write request. | |
2628 | * | |
2629 | * clean - no pending writes, but otherwise active. | |
2630 | * When written to inactive array, starts without resync | |
2631 | * If a write request arrives then | |
2632 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
2633 | * if not known, block and switch to write-pending | |
2634 | * If written to an active array that has pending writes, then fails. | |
2635 | * active | |
2636 | * fully active: IO and resync can be happening. | |
2637 | * When written to inactive array, starts with resync | |
2638 | * | |
2639 | * write-pending | |
2640 | * clean, but writes are blocked waiting for 'active' to be written. | |
2641 | * | |
2642 | * active-idle | |
2643 | * like active, but no writes have been seen for a while (100msec). | |
2644 | * | |
2645 | */ | |
2646 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
2647 | write_pending, active_idle, bad_word}; | |
05381954 | 2648 | static char *array_states[] = { |
9e653b63 N |
2649 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
2650 | "write-pending", "active-idle", NULL }; | |
2651 | ||
2652 | static int match_word(const char *word, char **list) | |
2653 | { | |
2654 | int n; | |
2655 | for (n=0; list[n]; n++) | |
2656 | if (cmd_match(word, list[n])) | |
2657 | break; | |
2658 | return n; | |
2659 | } | |
2660 | ||
2661 | static ssize_t | |
2662 | array_state_show(mddev_t *mddev, char *page) | |
2663 | { | |
2664 | enum array_state st = inactive; | |
2665 | ||
2666 | if (mddev->pers) | |
2667 | switch(mddev->ro) { | |
2668 | case 1: | |
2669 | st = readonly; | |
2670 | break; | |
2671 | case 2: | |
2672 | st = read_auto; | |
2673 | break; | |
2674 | case 0: | |
2675 | if (mddev->in_sync) | |
2676 | st = clean; | |
e691063a N |
2677 | else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) |
2678 | st = write_pending; | |
9e653b63 N |
2679 | else if (mddev->safemode) |
2680 | st = active_idle; | |
2681 | else | |
2682 | st = active; | |
2683 | } | |
2684 | else { | |
2685 | if (list_empty(&mddev->disks) && | |
2686 | mddev->raid_disks == 0 && | |
2687 | mddev->size == 0) | |
2688 | st = clear; | |
2689 | else | |
2690 | st = inactive; | |
2691 | } | |
2692 | return sprintf(page, "%s\n", array_states[st]); | |
2693 | } | |
2694 | ||
2695 | static int do_md_stop(mddev_t * mddev, int ro); | |
2696 | static int do_md_run(mddev_t * mddev); | |
2697 | static int restart_array(mddev_t *mddev); | |
2698 | ||
2699 | static ssize_t | |
2700 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
2701 | { | |
2702 | int err = -EINVAL; | |
2703 | enum array_state st = match_word(buf, array_states); | |
2704 | switch(st) { | |
2705 | case bad_word: | |
2706 | break; | |
2707 | case clear: | |
2708 | /* stopping an active array */ | |
e691063a N |
2709 | if (atomic_read(&mddev->active) > 1) |
2710 | return -EBUSY; | |
2711 | err = do_md_stop(mddev, 0); | |
9e653b63 N |
2712 | break; |
2713 | case inactive: | |
2714 | /* stopping an active array */ | |
2715 | if (mddev->pers) { | |
2716 | if (atomic_read(&mddev->active) > 1) | |
2717 | return -EBUSY; | |
2718 | err = do_md_stop(mddev, 2); | |
e691063a N |
2719 | } else |
2720 | err = 0; /* already inactive */ | |
9e653b63 N |
2721 | break; |
2722 | case suspended: | |
2723 | break; /* not supported yet */ | |
2724 | case readonly: | |
2725 | if (mddev->pers) | |
2726 | err = do_md_stop(mddev, 1); | |
2727 | else { | |
2728 | mddev->ro = 1; | |
648b629e | 2729 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
2730 | err = do_md_run(mddev); |
2731 | } | |
2732 | break; | |
2733 | case read_auto: | |
9e653b63 | 2734 | if (mddev->pers) { |
648b629e N |
2735 | if (mddev->ro != 1) |
2736 | err = do_md_stop(mddev, 1); | |
2737 | else | |
2738 | err = restart_array(mddev); | |
2739 | if (err == 0) { | |
2740 | mddev->ro = 2; | |
2741 | set_disk_ro(mddev->gendisk, 0); | |
2742 | } | |
9e653b63 N |
2743 | } else { |
2744 | mddev->ro = 2; | |
2745 | err = do_md_run(mddev); | |
2746 | } | |
2747 | break; | |
2748 | case clean: | |
2749 | if (mddev->pers) { | |
2750 | restart_array(mddev); | |
2751 | spin_lock_irq(&mddev->write_lock); | |
2752 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
2753 | if (mddev->in_sync == 0) { |
2754 | mddev->in_sync = 1; | |
31a59e34 N |
2755 | if (mddev->safemode == 1) |
2756 | mddev->safemode = 0; | |
e691063a N |
2757 | if (mddev->persistent) |
2758 | set_bit(MD_CHANGE_CLEAN, | |
2759 | &mddev->flags); | |
2760 | } | |
2761 | err = 0; | |
2762 | } else | |
2763 | err = -EBUSY; | |
9e653b63 N |
2764 | spin_unlock_irq(&mddev->write_lock); |
2765 | } else { | |
2766 | mddev->ro = 0; | |
2767 | mddev->recovery_cp = MaxSector; | |
2768 | err = do_md_run(mddev); | |
2769 | } | |
2770 | break; | |
2771 | case active: | |
2772 | if (mddev->pers) { | |
2773 | restart_array(mddev); | |
e691063a N |
2774 | if (mddev->external) |
2775 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
9e653b63 N |
2776 | wake_up(&mddev->sb_wait); |
2777 | err = 0; | |
2778 | } else { | |
2779 | mddev->ro = 0; | |
648b629e | 2780 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
2781 | err = do_md_run(mddev); |
2782 | } | |
2783 | break; | |
2784 | case write_pending: | |
2785 | case active_idle: | |
2786 | /* these cannot be set */ | |
2787 | break; | |
2788 | } | |
2789 | if (err) | |
2790 | return err; | |
0fd62b86 NB |
2791 | else { |
2792 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
9e653b63 | 2793 | return len; |
0fd62b86 | 2794 | } |
9e653b63 | 2795 | } |
80ca3a44 N |
2796 | static struct md_sysfs_entry md_array_state = |
2797 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 2798 | |
6d7ff738 N |
2799 | static ssize_t |
2800 | null_show(mddev_t *mddev, char *page) | |
2801 | { | |
2802 | return -EINVAL; | |
2803 | } | |
2804 | ||
2805 | static ssize_t | |
2806 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
2807 | { | |
2808 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
2809 | /* The new device is added to the array. | |
2810 | * If the array has a persistent superblock, we read the | |
2811 | * superblock to initialise info and check validity. | |
2812 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
2813 | * which mainly checks size. | |
2814 | */ | |
2815 | char *e; | |
2816 | int major = simple_strtoul(buf, &e, 10); | |
2817 | int minor; | |
2818 | dev_t dev; | |
2819 | mdk_rdev_t *rdev; | |
2820 | int err; | |
2821 | ||
2822 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
2823 | return -EINVAL; | |
2824 | minor = simple_strtoul(e+1, &e, 10); | |
2825 | if (*e && *e != '\n') | |
2826 | return -EINVAL; | |
2827 | dev = MKDEV(major, minor); | |
2828 | if (major != MAJOR(dev) || | |
2829 | minor != MINOR(dev)) | |
2830 | return -EOVERFLOW; | |
2831 | ||
2832 | ||
2833 | if (mddev->persistent) { | |
2834 | rdev = md_import_device(dev, mddev->major_version, | |
2835 | mddev->minor_version); | |
2836 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
2837 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
2838 | mdk_rdev_t, same_set); | |
2839 | err = super_types[mddev->major_version] | |
2840 | .load_super(rdev, rdev0, mddev->minor_version); | |
2841 | if (err < 0) | |
2842 | goto out; | |
2843 | } | |
c5d79adb N |
2844 | } else if (mddev->external) |
2845 | rdev = md_import_device(dev, -2, -1); | |
2846 | else | |
6d7ff738 N |
2847 | rdev = md_import_device(dev, -1, -1); |
2848 | ||
2849 | if (IS_ERR(rdev)) | |
2850 | return PTR_ERR(rdev); | |
2851 | err = bind_rdev_to_array(rdev, mddev); | |
2852 | out: | |
2853 | if (err) | |
2854 | export_rdev(rdev); | |
2855 | return err ? err : len; | |
2856 | } | |
2857 | ||
2858 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 2859 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 2860 | |
9b1d1dac PC |
2861 | static ssize_t |
2862 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
2863 | { | |
2864 | char *end; | |
2865 | unsigned long chunk, end_chunk; | |
2866 | ||
2867 | if (!mddev->bitmap) | |
2868 | goto out; | |
2869 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
2870 | while (*buf) { | |
2871 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
2872 | if (buf == end) break; | |
2873 | if (*end == '-') { /* range */ | |
2874 | buf = end + 1; | |
2875 | end_chunk = simple_strtoul(buf, &end, 0); | |
2876 | if (buf == end) break; | |
2877 | } | |
2878 | if (*end && !isspace(*end)) break; | |
2879 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
2880 | buf = end; | |
2881 | while (isspace(*buf)) buf++; | |
2882 | } | |
2883 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
2884 | out: | |
2885 | return len; | |
2886 | } | |
2887 | ||
2888 | static struct md_sysfs_entry md_bitmap = | |
2889 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
2890 | ||
a35b0d69 N |
2891 | static ssize_t |
2892 | size_show(mddev_t *mddev, char *page) | |
2893 | { | |
2894 | return sprintf(page, "%llu\n", (unsigned long long)mddev->size); | |
2895 | } | |
2896 | ||
2897 | static int update_size(mddev_t *mddev, unsigned long size); | |
2898 | ||
2899 | static ssize_t | |
2900 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
2901 | { | |
2902 | /* If array is inactive, we can reduce the component size, but | |
2903 | * not increase it (except from 0). | |
2904 | * If array is active, we can try an on-line resize | |
2905 | */ | |
2906 | char *e; | |
2907 | int err = 0; | |
2908 | unsigned long long size = simple_strtoull(buf, &e, 10); | |
2909 | if (!*buf || *buf == '\n' || | |
2910 | (*e && *e != '\n')) | |
2911 | return -EINVAL; | |
2912 | ||
2913 | if (mddev->pers) { | |
2914 | err = update_size(mddev, size); | |
850b2b42 | 2915 | md_update_sb(mddev, 1); |
a35b0d69 N |
2916 | } else { |
2917 | if (mddev->size == 0 || | |
2918 | mddev->size > size) | |
2919 | mddev->size = size; | |
2920 | else | |
2921 | err = -ENOSPC; | |
2922 | } | |
2923 | return err ? err : len; | |
2924 | } | |
2925 | ||
2926 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 2927 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 2928 | |
8bb93aac N |
2929 | |
2930 | /* Metdata version. | |
e691063a N |
2931 | * This is one of |
2932 | * 'none' for arrays with no metadata (good luck...) | |
2933 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
2934 | * or N.M for internally known formats |
2935 | */ | |
2936 | static ssize_t | |
2937 | metadata_show(mddev_t *mddev, char *page) | |
2938 | { | |
2939 | if (mddev->persistent) | |
2940 | return sprintf(page, "%d.%d\n", | |
2941 | mddev->major_version, mddev->minor_version); | |
e691063a N |
2942 | else if (mddev->external) |
2943 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
2944 | else |
2945 | return sprintf(page, "none\n"); | |
2946 | } | |
2947 | ||
2948 | static ssize_t | |
2949 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
2950 | { | |
2951 | int major, minor; | |
2952 | char *e; | |
2953 | if (!list_empty(&mddev->disks)) | |
2954 | return -EBUSY; | |
2955 | ||
2956 | if (cmd_match(buf, "none")) { | |
2957 | mddev->persistent = 0; | |
e691063a N |
2958 | mddev->external = 0; |
2959 | mddev->major_version = 0; | |
2960 | mddev->minor_version = 90; | |
2961 | return len; | |
2962 | } | |
2963 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 2964 | size_t namelen = len-9; |
e691063a N |
2965 | if (namelen >= sizeof(mddev->metadata_type)) |
2966 | namelen = sizeof(mddev->metadata_type)-1; | |
2967 | strncpy(mddev->metadata_type, buf+9, namelen); | |
2968 | mddev->metadata_type[namelen] = 0; | |
2969 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
2970 | mddev->metadata_type[--namelen] = 0; | |
2971 | mddev->persistent = 0; | |
2972 | mddev->external = 1; | |
8bb93aac N |
2973 | mddev->major_version = 0; |
2974 | mddev->minor_version = 90; | |
2975 | return len; | |
2976 | } | |
2977 | major = simple_strtoul(buf, &e, 10); | |
2978 | if (e==buf || *e != '.') | |
2979 | return -EINVAL; | |
2980 | buf = e+1; | |
2981 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 2982 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 2983 | return -EINVAL; |
50511da3 | 2984 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
2985 | return -ENOENT; |
2986 | mddev->major_version = major; | |
2987 | mddev->minor_version = minor; | |
2988 | mddev->persistent = 1; | |
e691063a | 2989 | mddev->external = 0; |
8bb93aac N |
2990 | return len; |
2991 | } | |
2992 | ||
2993 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 2994 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 2995 | |
24dd469d | 2996 | static ssize_t |
7eec314d | 2997 | action_show(mddev_t *mddev, char *page) |
24dd469d | 2998 | { |
7eec314d | 2999 | char *type = "idle"; |
31399d9e | 3000 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
2b12ab6d | 3001 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3002 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3003 | type = "reshape"; | |
3004 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3005 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3006 | type = "resync"; | |
3007 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3008 | type = "check"; | |
3009 | else | |
3010 | type = "repair"; | |
72a23c21 | 3011 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3012 | type = "recover"; |
3013 | } | |
3014 | return sprintf(page, "%s\n", type); | |
3015 | } | |
3016 | ||
3017 | static ssize_t | |
7eec314d | 3018 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3019 | { |
7eec314d N |
3020 | if (!mddev->pers || !mddev->pers->sync_request) |
3021 | return -EINVAL; | |
3022 | ||
bce74dac | 3023 | if (cmd_match(page, "idle")) { |
7eec314d N |
3024 | if (mddev->sync_thread) { |
3025 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
3026 | md_unregister_thread(mddev->sync_thread); | |
3027 | mddev->sync_thread = NULL; | |
3028 | mddev->recovery = 0; | |
3029 | } | |
03c902e1 N |
3030 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3031 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3032 | return -EBUSY; |
72a23c21 NB |
3033 | else if (cmd_match(page, "resync")) |
3034 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3035 | else if (cmd_match(page, "recover")) { | |
3036 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3037 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3038 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3039 | int err; |
3040 | if (mddev->pers->start_reshape == NULL) | |
3041 | return -EINVAL; | |
3042 | err = mddev->pers->start_reshape(mddev); | |
3043 | if (err) | |
3044 | return err; | |
a99ac971 | 3045 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3046 | } else { |
bce74dac | 3047 | if (cmd_match(page, "check")) |
7eec314d | 3048 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3049 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3050 | return -EINVAL; |
3051 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3052 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3053 | } |
03c902e1 | 3054 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3055 | md_wakeup_thread(mddev->thread); |
72a23c21 | 3056 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
24dd469d N |
3057 | return len; |
3058 | } | |
3059 | ||
9d88883e | 3060 | static ssize_t |
96de1e66 | 3061 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3062 | { |
3063 | return sprintf(page, "%llu\n", | |
3064 | (unsigned long long) mddev->resync_mismatches); | |
3065 | } | |
3066 | ||
80ca3a44 N |
3067 | static struct md_sysfs_entry md_scan_mode = |
3068 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3069 | |
96de1e66 | 3070 | |
80ca3a44 | 3071 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3072 | |
88202a0c N |
3073 | static ssize_t |
3074 | sync_min_show(mddev_t *mddev, char *page) | |
3075 | { | |
3076 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3077 | mddev->sync_speed_min ? "local": "system"); | |
3078 | } | |
3079 | ||
3080 | static ssize_t | |
3081 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3082 | { | |
3083 | int min; | |
3084 | char *e; | |
3085 | if (strncmp(buf, "system", 6)==0) { | |
3086 | mddev->sync_speed_min = 0; | |
3087 | return len; | |
3088 | } | |
3089 | min = simple_strtoul(buf, &e, 10); | |
3090 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3091 | return -EINVAL; | |
3092 | mddev->sync_speed_min = min; | |
3093 | return len; | |
3094 | } | |
3095 | ||
3096 | static struct md_sysfs_entry md_sync_min = | |
3097 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3098 | ||
3099 | static ssize_t | |
3100 | sync_max_show(mddev_t *mddev, char *page) | |
3101 | { | |
3102 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3103 | mddev->sync_speed_max ? "local": "system"); | |
3104 | } | |
3105 | ||
3106 | static ssize_t | |
3107 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3108 | { | |
3109 | int max; | |
3110 | char *e; | |
3111 | if (strncmp(buf, "system", 6)==0) { | |
3112 | mddev->sync_speed_max = 0; | |
3113 | return len; | |
3114 | } | |
3115 | max = simple_strtoul(buf, &e, 10); | |
3116 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3117 | return -EINVAL; | |
3118 | mddev->sync_speed_max = max; | |
3119 | return len; | |
3120 | } | |
3121 | ||
3122 | static struct md_sysfs_entry md_sync_max = | |
3123 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3124 | ||
d7f3d291 IP |
3125 | static ssize_t |
3126 | degraded_show(mddev_t *mddev, char *page) | |
3127 | { | |
3128 | return sprintf(page, "%d\n", mddev->degraded); | |
3129 | } | |
3130 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3131 | |
90b08710 BS |
3132 | static ssize_t |
3133 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3134 | { | |
3135 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3136 | } | |
3137 | ||
3138 | static ssize_t | |
3139 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3140 | { | |
3141 | long n; | |
3142 | ||
3143 | if (strict_strtol(buf, 10, &n)) | |
3144 | return -EINVAL; | |
3145 | ||
3146 | if (n != 0 && n != 1) | |
3147 | return -EINVAL; | |
3148 | ||
3149 | mddev->parallel_resync = n; | |
3150 | ||
3151 | if (mddev->sync_thread) | |
3152 | wake_up(&resync_wait); | |
3153 | ||
3154 | return len; | |
3155 | } | |
3156 | ||
3157 | /* force parallel resync, even with shared block devices */ | |
3158 | static struct md_sysfs_entry md_sync_force_parallel = | |
3159 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3160 | sync_force_parallel_show, sync_force_parallel_store); | |
3161 | ||
88202a0c N |
3162 | static ssize_t |
3163 | sync_speed_show(mddev_t *mddev, char *page) | |
3164 | { | |
3165 | unsigned long resync, dt, db; | |
9687a60c AN |
3166 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3167 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3168 | if (!dt) dt++; |
9687a60c AN |
3169 | db = resync - mddev->resync_mark_cnt; |
3170 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3171 | } |
3172 | ||
80ca3a44 | 3173 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3174 | |
3175 | static ssize_t | |
3176 | sync_completed_show(mddev_t *mddev, char *page) | |
3177 | { | |
3178 | unsigned long max_blocks, resync; | |
3179 | ||
3180 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
3181 | max_blocks = mddev->resync_max_sectors; | |
3182 | else | |
3183 | max_blocks = mddev->size << 1; | |
3184 | ||
3185 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active)); | |
3186 | return sprintf(page, "%lu / %lu\n", resync, max_blocks); | |
3187 | } | |
3188 | ||
80ca3a44 | 3189 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3190 | |
5e96ee65 NB |
3191 | static ssize_t |
3192 | min_sync_show(mddev_t *mddev, char *page) | |
3193 | { | |
3194 | return sprintf(page, "%llu\n", | |
3195 | (unsigned long long)mddev->resync_min); | |
3196 | } | |
3197 | static ssize_t | |
3198 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3199 | { | |
3200 | unsigned long long min; | |
3201 | if (strict_strtoull(buf, 10, &min)) | |
3202 | return -EINVAL; | |
3203 | if (min > mddev->resync_max) | |
3204 | return -EINVAL; | |
3205 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3206 | return -EBUSY; | |
3207 | ||
3208 | /* Must be a multiple of chunk_size */ | |
3209 | if (mddev->chunk_size) { | |
3210 | if (min & (sector_t)((mddev->chunk_size>>9)-1)) | |
3211 | return -EINVAL; | |
3212 | } | |
3213 | mddev->resync_min = min; | |
3214 | ||
3215 | return len; | |
3216 | } | |
3217 | ||
3218 | static struct md_sysfs_entry md_min_sync = | |
3219 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3220 | ||
c6207277 N |
3221 | static ssize_t |
3222 | max_sync_show(mddev_t *mddev, char *page) | |
3223 | { | |
3224 | if (mddev->resync_max == MaxSector) | |
3225 | return sprintf(page, "max\n"); | |
3226 | else | |
3227 | return sprintf(page, "%llu\n", | |
3228 | (unsigned long long)mddev->resync_max); | |
3229 | } | |
3230 | static ssize_t | |
3231 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3232 | { | |
3233 | if (strncmp(buf, "max", 3) == 0) | |
3234 | mddev->resync_max = MaxSector; | |
3235 | else { | |
5e96ee65 NB |
3236 | unsigned long long max; |
3237 | if (strict_strtoull(buf, 10, &max)) | |
3238 | return -EINVAL; | |
3239 | if (max < mddev->resync_min) | |
c6207277 N |
3240 | return -EINVAL; |
3241 | if (max < mddev->resync_max && | |
3242 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3243 | return -EBUSY; | |
3244 | ||
3245 | /* Must be a multiple of chunk_size */ | |
3246 | if (mddev->chunk_size) { | |
3247 | if (max & (sector_t)((mddev->chunk_size>>9)-1)) | |
3248 | return -EINVAL; | |
3249 | } | |
3250 | mddev->resync_max = max; | |
3251 | } | |
3252 | wake_up(&mddev->recovery_wait); | |
3253 | return len; | |
3254 | } | |
3255 | ||
3256 | static struct md_sysfs_entry md_max_sync = | |
3257 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
3258 | ||
e464eafd N |
3259 | static ssize_t |
3260 | suspend_lo_show(mddev_t *mddev, char *page) | |
3261 | { | |
3262 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
3263 | } | |
3264 | ||
3265 | static ssize_t | |
3266 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
3267 | { | |
3268 | char *e; | |
3269 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3270 | ||
3271 | if (mddev->pers->quiesce == NULL) | |
3272 | return -EINVAL; | |
3273 | if (buf == e || (*e && *e != '\n')) | |
3274 | return -EINVAL; | |
3275 | if (new >= mddev->suspend_hi || | |
3276 | (new > mddev->suspend_lo && new < mddev->suspend_hi)) { | |
3277 | mddev->suspend_lo = new; | |
3278 | mddev->pers->quiesce(mddev, 2); | |
3279 | return len; | |
3280 | } else | |
3281 | return -EINVAL; | |
3282 | } | |
3283 | static struct md_sysfs_entry md_suspend_lo = | |
3284 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
3285 | ||
3286 | ||
3287 | static ssize_t | |
3288 | suspend_hi_show(mddev_t *mddev, char *page) | |
3289 | { | |
3290 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
3291 | } | |
3292 | ||
3293 | static ssize_t | |
3294 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
3295 | { | |
3296 | char *e; | |
3297 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3298 | ||
3299 | if (mddev->pers->quiesce == NULL) | |
3300 | return -EINVAL; | |
3301 | if (buf == e || (*e && *e != '\n')) | |
3302 | return -EINVAL; | |
3303 | if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) || | |
3304 | (new > mddev->suspend_lo && new > mddev->suspend_hi)) { | |
3305 | mddev->suspend_hi = new; | |
3306 | mddev->pers->quiesce(mddev, 1); | |
3307 | mddev->pers->quiesce(mddev, 0); | |
3308 | return len; | |
3309 | } else | |
3310 | return -EINVAL; | |
3311 | } | |
3312 | static struct md_sysfs_entry md_suspend_hi = | |
3313 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
3314 | ||
08a02ecd N |
3315 | static ssize_t |
3316 | reshape_position_show(mddev_t *mddev, char *page) | |
3317 | { | |
3318 | if (mddev->reshape_position != MaxSector) | |
3319 | return sprintf(page, "%llu\n", | |
3320 | (unsigned long long)mddev->reshape_position); | |
3321 | strcpy(page, "none\n"); | |
3322 | return 5; | |
3323 | } | |
3324 | ||
3325 | static ssize_t | |
3326 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
3327 | { | |
3328 | char *e; | |
3329 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
3330 | if (mddev->pers) | |
3331 | return -EBUSY; | |
3332 | if (buf == e || (*e && *e != '\n')) | |
3333 | return -EINVAL; | |
3334 | mddev->reshape_position = new; | |
3335 | mddev->delta_disks = 0; | |
3336 | mddev->new_level = mddev->level; | |
3337 | mddev->new_layout = mddev->layout; | |
3338 | mddev->new_chunk = mddev->chunk_size; | |
3339 | return len; | |
3340 | } | |
3341 | ||
3342 | static struct md_sysfs_entry md_reshape_position = | |
3343 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
3344 | reshape_position_store); | |
3345 | ||
e464eafd | 3346 | |
eae1701f N |
3347 | static struct attribute *md_default_attrs[] = { |
3348 | &md_level.attr, | |
d4dbd025 | 3349 | &md_layout.attr, |
eae1701f | 3350 | &md_raid_disks.attr, |
3b34380a | 3351 | &md_chunk_size.attr, |
a35b0d69 | 3352 | &md_size.attr, |
a94213b1 | 3353 | &md_resync_start.attr, |
8bb93aac | 3354 | &md_metadata.attr, |
6d7ff738 | 3355 | &md_new_device.attr, |
16f17b39 | 3356 | &md_safe_delay.attr, |
9e653b63 | 3357 | &md_array_state.attr, |
08a02ecd | 3358 | &md_reshape_position.attr, |
411036fa N |
3359 | NULL, |
3360 | }; | |
3361 | ||
3362 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 3363 | &md_scan_mode.attr, |
9d88883e | 3364 | &md_mismatches.attr, |
88202a0c N |
3365 | &md_sync_min.attr, |
3366 | &md_sync_max.attr, | |
3367 | &md_sync_speed.attr, | |
90b08710 | 3368 | &md_sync_force_parallel.attr, |
88202a0c | 3369 | &md_sync_completed.attr, |
5e96ee65 | 3370 | &md_min_sync.attr, |
c6207277 | 3371 | &md_max_sync.attr, |
e464eafd N |
3372 | &md_suspend_lo.attr, |
3373 | &md_suspend_hi.attr, | |
9b1d1dac | 3374 | &md_bitmap.attr, |
d7f3d291 | 3375 | &md_degraded.attr, |
eae1701f N |
3376 | NULL, |
3377 | }; | |
411036fa N |
3378 | static struct attribute_group md_redundancy_group = { |
3379 | .name = NULL, | |
3380 | .attrs = md_redundancy_attrs, | |
3381 | }; | |
3382 | ||
eae1701f N |
3383 | |
3384 | static ssize_t | |
3385 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3386 | { | |
3387 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3388 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3389 | ssize_t rv; |
eae1701f N |
3390 | |
3391 | if (!entry->show) | |
3392 | return -EIO; | |
5dc5cf7d IM |
3393 | rv = mddev_lock(mddev); |
3394 | if (!rv) { | |
3395 | rv = entry->show(mddev, page); | |
3396 | mddev_unlock(mddev); | |
3397 | } | |
96de1e66 | 3398 | return rv; |
eae1701f N |
3399 | } |
3400 | ||
3401 | static ssize_t | |
3402 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
3403 | const char *page, size_t length) | |
3404 | { | |
3405 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
3406 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 3407 | ssize_t rv; |
eae1701f N |
3408 | |
3409 | if (!entry->store) | |
3410 | return -EIO; | |
67463acb N |
3411 | if (!capable(CAP_SYS_ADMIN)) |
3412 | return -EACCES; | |
5dc5cf7d IM |
3413 | rv = mddev_lock(mddev); |
3414 | if (!rv) { | |
3415 | rv = entry->store(mddev, page, length); | |
3416 | mddev_unlock(mddev); | |
3417 | } | |
96de1e66 | 3418 | return rv; |
eae1701f N |
3419 | } |
3420 | ||
3421 | static void md_free(struct kobject *ko) | |
3422 | { | |
3423 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
3424 | kfree(mddev); | |
3425 | } | |
3426 | ||
3427 | static struct sysfs_ops md_sysfs_ops = { | |
3428 | .show = md_attr_show, | |
3429 | .store = md_attr_store, | |
3430 | }; | |
3431 | static struct kobj_type md_ktype = { | |
3432 | .release = md_free, | |
3433 | .sysfs_ops = &md_sysfs_ops, | |
3434 | .default_attrs = md_default_attrs, | |
3435 | }; | |
3436 | ||
1da177e4 LT |
3437 | int mdp_major = 0; |
3438 | ||
3439 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
3440 | { | |
48c9c27b | 3441 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
3442 | mddev_t *mddev = mddev_find(dev); |
3443 | struct gendisk *disk; | |
3444 | int partitioned = (MAJOR(dev) != MD_MAJOR); | |
3445 | int shift = partitioned ? MdpMinorShift : 0; | |
3446 | int unit = MINOR(dev) >> shift; | |
3830c62f | 3447 | int error; |
1da177e4 LT |
3448 | |
3449 | if (!mddev) | |
3450 | return NULL; | |
3451 | ||
48c9c27b | 3452 | mutex_lock(&disks_mutex); |
1da177e4 | 3453 | if (mddev->gendisk) { |
48c9c27b | 3454 | mutex_unlock(&disks_mutex); |
1da177e4 LT |
3455 | mddev_put(mddev); |
3456 | return NULL; | |
3457 | } | |
3458 | disk = alloc_disk(1 << shift); | |
3459 | if (!disk) { | |
48c9c27b | 3460 | mutex_unlock(&disks_mutex); |
1da177e4 LT |
3461 | mddev_put(mddev); |
3462 | return NULL; | |
3463 | } | |
3464 | disk->major = MAJOR(dev); | |
3465 | disk->first_minor = unit << shift; | |
ce7b0f46 | 3466 | if (partitioned) |
1da177e4 | 3467 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 3468 | else |
1da177e4 | 3469 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
3470 | disk->fops = &md_fops; |
3471 | disk->private_data = mddev; | |
3472 | disk->queue = mddev->queue; | |
3473 | add_disk(disk); | |
3474 | mddev->gendisk = disk; | |
edfaa7c3 | 3475 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj, |
3830c62f | 3476 | "%s", "md"); |
f48ed538 | 3477 | mutex_unlock(&disks_mutex); |
3830c62f | 3478 | if (error) |
5e55e2f5 N |
3479 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
3480 | disk->disk_name); | |
3830c62f GKH |
3481 | else |
3482 | kobject_uevent(&mddev->kobj, KOBJ_ADD); | |
1da177e4 LT |
3483 | return NULL; |
3484 | } | |
3485 | ||
1da177e4 LT |
3486 | static void md_safemode_timeout(unsigned long data) |
3487 | { | |
3488 | mddev_t *mddev = (mddev_t *) data; | |
3489 | ||
0fd62b86 NB |
3490 | if (!atomic_read(&mddev->writes_pending)) { |
3491 | mddev->safemode = 1; | |
3492 | if (mddev->external) | |
3493 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
3494 | } | |
1da177e4 LT |
3495 | md_wakeup_thread(mddev->thread); |
3496 | } | |
3497 | ||
6ff8d8ec | 3498 | static int start_dirty_degraded; |
1da177e4 LT |
3499 | |
3500 | static int do_md_run(mddev_t * mddev) | |
3501 | { | |
2604b703 | 3502 | int err; |
1da177e4 LT |
3503 | int chunk_size; |
3504 | struct list_head *tmp; | |
3505 | mdk_rdev_t *rdev; | |
3506 | struct gendisk *disk; | |
2604b703 | 3507 | struct mdk_personality *pers; |
1da177e4 LT |
3508 | char b[BDEVNAME_SIZE]; |
3509 | ||
a757e64c N |
3510 | if (list_empty(&mddev->disks)) |
3511 | /* cannot run an array with no devices.. */ | |
1da177e4 | 3512 | return -EINVAL; |
1da177e4 LT |
3513 | |
3514 | if (mddev->pers) | |
3515 | return -EBUSY; | |
3516 | ||
3517 | /* | |
3518 | * Analyze all RAID superblock(s) | |
3519 | */ | |
1ec4a939 N |
3520 | if (!mddev->raid_disks) { |
3521 | if (!mddev->persistent) | |
3522 | return -EINVAL; | |
a757e64c | 3523 | analyze_sbs(mddev); |
1ec4a939 | 3524 | } |
1da177e4 LT |
3525 | |
3526 | chunk_size = mddev->chunk_size; | |
2604b703 N |
3527 | |
3528 | if (chunk_size) { | |
1da177e4 LT |
3529 | if (chunk_size > MAX_CHUNK_SIZE) { |
3530 | printk(KERN_ERR "too big chunk_size: %d > %d\n", | |
3531 | chunk_size, MAX_CHUNK_SIZE); | |
3532 | return -EINVAL; | |
3533 | } | |
3534 | /* | |
3535 | * chunk-size has to be a power of 2 and multiples of PAGE_SIZE | |
3536 | */ | |
3537 | if ( (1 << ffz(~chunk_size)) != chunk_size) { | |
a757e64c | 3538 | printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size); |
1da177e4 LT |
3539 | return -EINVAL; |
3540 | } | |
3541 | if (chunk_size < PAGE_SIZE) { | |
3542 | printk(KERN_ERR "too small chunk_size: %d < %ld\n", | |
3543 | chunk_size, PAGE_SIZE); | |
3544 | return -EINVAL; | |
3545 | } | |
3546 | ||
3547 | /* devices must have minimum size of one chunk */ | |
d089c6af | 3548 | rdev_for_each(rdev, tmp, mddev) { |
b2d444d7 | 3549 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
3550 | continue; |
3551 | if (rdev->size < chunk_size / 1024) { | |
3552 | printk(KERN_WARNING | |
3553 | "md: Dev %s smaller than chunk_size:" | |
3554 | " %lluk < %dk\n", | |
3555 | bdevname(rdev->bdev,b), | |
3556 | (unsigned long long)rdev->size, | |
3557 | chunk_size / 1024); | |
3558 | return -EINVAL; | |
3559 | } | |
3560 | } | |
3561 | } | |
3562 | ||
1da177e4 | 3563 | #ifdef CONFIG_KMOD |
d9d166c2 N |
3564 | if (mddev->level != LEVEL_NONE) |
3565 | request_module("md-level-%d", mddev->level); | |
3566 | else if (mddev->clevel[0]) | |
3567 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
3568 | #endif |
3569 | ||
3570 | /* | |
3571 | * Drop all container device buffers, from now on | |
3572 | * the only valid external interface is through the md | |
3573 | * device. | |
1da177e4 | 3574 | */ |
d089c6af | 3575 | rdev_for_each(rdev, tmp, mddev) { |
b2d444d7 | 3576 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
3577 | continue; |
3578 | sync_blockdev(rdev->bdev); | |
f98393a6 | 3579 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
3580 | |
3581 | /* perform some consistency tests on the device. | |
3582 | * We don't want the data to overlap the metadata, | |
3583 | * Internal Bitmap issues has handled elsewhere. | |
3584 | */ | |
3585 | if (rdev->data_offset < rdev->sb_offset) { | |
3586 | if (mddev->size && | |
3587 | rdev->data_offset + mddev->size*2 | |
3588 | > rdev->sb_offset*2) { | |
3589 | printk("md: %s: data overlaps metadata\n", | |
3590 | mdname(mddev)); | |
3591 | return -EINVAL; | |
3592 | } | |
3593 | } else { | |
3594 | if (rdev->sb_offset*2 + rdev->sb_size/512 | |
3595 | > rdev->data_offset) { | |
3596 | printk("md: %s: metadata overlaps data\n", | |
3597 | mdname(mddev)); | |
3598 | return -EINVAL; | |
3599 | } | |
3600 | } | |
52664732 | 3601 | sysfs_notify(&rdev->kobj, NULL, "state"); |
1da177e4 LT |
3602 | } |
3603 | ||
3604 | md_probe(mddev->unit, NULL, NULL); | |
3605 | disk = mddev->gendisk; | |
3606 | if (!disk) | |
3607 | return -ENOMEM; | |
3608 | ||
3609 | spin_lock(&pers_lock); | |
d9d166c2 | 3610 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 3611 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 3612 | spin_unlock(&pers_lock); |
d9d166c2 N |
3613 | if (mddev->level != LEVEL_NONE) |
3614 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
3615 | mddev->level); | |
3616 | else | |
3617 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
3618 | mddev->clevel); | |
1da177e4 LT |
3619 | return -EINVAL; |
3620 | } | |
2604b703 | 3621 | mddev->pers = pers; |
1da177e4 | 3622 | spin_unlock(&pers_lock); |
d9d166c2 N |
3623 | mddev->level = pers->level; |
3624 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
1da177e4 | 3625 | |
f6705578 | 3626 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 3627 | pers->start_reshape == NULL) { |
f6705578 N |
3628 | /* This personality cannot handle reshaping... */ |
3629 | mddev->pers = NULL; | |
3630 | module_put(pers->owner); | |
3631 | return -EINVAL; | |
3632 | } | |
3633 | ||
7dd5e7c3 N |
3634 | if (pers->sync_request) { |
3635 | /* Warn if this is a potentially silly | |
3636 | * configuration. | |
3637 | */ | |
3638 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3639 | mdk_rdev_t *rdev2; | |
3640 | struct list_head *tmp2; | |
3641 | int warned = 0; | |
d089c6af N |
3642 | rdev_for_each(rdev, tmp, mddev) { |
3643 | rdev_for_each(rdev2, tmp2, mddev) { | |
7dd5e7c3 N |
3644 | if (rdev < rdev2 && |
3645 | rdev->bdev->bd_contains == | |
3646 | rdev2->bdev->bd_contains) { | |
3647 | printk(KERN_WARNING | |
3648 | "%s: WARNING: %s appears to be" | |
3649 | " on the same physical disk as" | |
3650 | " %s.\n", | |
3651 | mdname(mddev), | |
3652 | bdevname(rdev->bdev,b), | |
3653 | bdevname(rdev2->bdev,b2)); | |
3654 | warned = 1; | |
3655 | } | |
3656 | } | |
3657 | } | |
3658 | if (warned) | |
3659 | printk(KERN_WARNING | |
3660 | "True protection against single-disk" | |
3661 | " failure might be compromised.\n"); | |
3662 | } | |
3663 | ||
657390d2 | 3664 | mddev->recovery = 0; |
1da177e4 | 3665 | mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */ |
a9701a30 | 3666 | mddev->barriers_work = 1; |
6ff8d8ec | 3667 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 3668 | |
f91de92e N |
3669 | if (start_readonly) |
3670 | mddev->ro = 2; /* read-only, but switch on first write */ | |
3671 | ||
b15c2e57 | 3672 | err = mddev->pers->run(mddev); |
13e53df3 AN |
3673 | if (err) |
3674 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
3675 | else if (mddev->pers->sync_request) { | |
b15c2e57 N |
3676 | err = bitmap_create(mddev); |
3677 | if (err) { | |
3678 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
3679 | mdname(mddev), err); | |
3680 | mddev->pers->stop(mddev); | |
3681 | } | |
3682 | } | |
1da177e4 | 3683 | if (err) { |
1da177e4 LT |
3684 | module_put(mddev->pers->owner); |
3685 | mddev->pers = NULL; | |
32a7627c N |
3686 | bitmap_destroy(mddev); |
3687 | return err; | |
1da177e4 | 3688 | } |
5e55e2f5 N |
3689 | if (mddev->pers->sync_request) { |
3690 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3691 | printk(KERN_WARNING | |
3692 | "md: cannot register extra attributes for %s\n", | |
3693 | mdname(mddev)); | |
3694 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ | |
fd9d49ca N |
3695 | mddev->ro = 0; |
3696 | ||
1da177e4 LT |
3697 | atomic_set(&mddev->writes_pending,0); |
3698 | mddev->safemode = 0; | |
3699 | mddev->safemode_timer.function = md_safemode_timeout; | |
3700 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 3701 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 3702 | mddev->in_sync = 1; |
86e6ffdd | 3703 | |
d089c6af | 3704 | rdev_for_each(rdev, tmp, mddev) |
86e6ffdd N |
3705 | if (rdev->raid_disk >= 0) { |
3706 | char nm[20]; | |
3707 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
3708 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
3709 | printk("md: cannot register %s for %s\n", | |
3710 | nm, mdname(mddev)); | |
86e6ffdd | 3711 | } |
1da177e4 LT |
3712 | |
3713 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3714 | ||
850b2b42 N |
3715 | if (mddev->flags) |
3716 | md_update_sb(mddev, 0); | |
1da177e4 LT |
3717 | |
3718 | set_capacity(disk, mddev->array_size<<1); | |
3719 | ||
3720 | /* If we call blk_queue_make_request here, it will | |
3721 | * re-initialise max_sectors etc which may have been | |
3722 | * refined inside -> run. So just set the bits we need to set. | |
3723 | * Most initialisation happended when we called | |
3724 | * blk_queue_make_request(..., md_fail_request) | |
3725 | * earlier. | |
3726 | */ | |
3727 | mddev->queue->queuedata = mddev; | |
3728 | mddev->queue->make_request_fn = mddev->pers->make_request; | |
3729 | ||
5fd6c1dc N |
3730 | /* If there is a partially-recovered drive we need to |
3731 | * start recovery here. If we leave it to md_check_recovery, | |
3732 | * it will remove the drives and not do the right thing | |
3733 | */ | |
0b8c9de0 | 3734 | if (mddev->degraded && !mddev->sync_thread) { |
5fd6c1dc N |
3735 | struct list_head *rtmp; |
3736 | int spares = 0; | |
d089c6af | 3737 | rdev_for_each(rdev, rtmp, mddev) |
5fd6c1dc N |
3738 | if (rdev->raid_disk >= 0 && |
3739 | !test_bit(In_sync, &rdev->flags) && | |
3740 | !test_bit(Faulty, &rdev->flags)) | |
3741 | /* complete an interrupted recovery */ | |
3742 | spares++; | |
3743 | if (spares && mddev->pers->sync_request) { | |
3744 | mddev->recovery = 0; | |
3745 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3746 | mddev->sync_thread = md_register_thread(md_do_sync, | |
3747 | mddev, | |
3748 | "%s_resync"); | |
3749 | if (!mddev->sync_thread) { | |
3750 | printk(KERN_ERR "%s: could not start resync" | |
3751 | " thread...\n", | |
3752 | mdname(mddev)); | |
3753 | /* leave the spares where they are, it shouldn't hurt */ | |
3754 | mddev->recovery = 0; | |
0b8c9de0 | 3755 | } |
5fd6c1dc N |
3756 | } |
3757 | } | |
0b8c9de0 N |
3758 | md_wakeup_thread(mddev->thread); |
3759 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 3760 | |
44ce6294 | 3761 | mddev->changed = 1; |
d7603b7e | 3762 | md_new_event(mddev); |
0fd62b86 | 3763 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
72a23c21 | 3764 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
a99ac971 | 3765 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
edfaa7c3 | 3766 | kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE); |
1da177e4 LT |
3767 | return 0; |
3768 | } | |
3769 | ||
3770 | static int restart_array(mddev_t *mddev) | |
3771 | { | |
3772 | struct gendisk *disk = mddev->gendisk; | |
3773 | int err; | |
3774 | ||
3775 | /* | |
3776 | * Complain if it has no devices | |
3777 | */ | |
3778 | err = -ENXIO; | |
3779 | if (list_empty(&mddev->disks)) | |
3780 | goto out; | |
3781 | ||
3782 | if (mddev->pers) { | |
3783 | err = -EBUSY; | |
3784 | if (!mddev->ro) | |
3785 | goto out; | |
3786 | ||
3787 | mddev->safemode = 0; | |
3788 | mddev->ro = 0; | |
3789 | set_disk_ro(disk, 0); | |
3790 | ||
3791 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
3792 | mdname(mddev)); | |
3793 | /* | |
3794 | * Kick recovery or resync if necessary | |
3795 | */ | |
3796 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3797 | md_wakeup_thread(mddev->thread); | |
5fd6c1dc | 3798 | md_wakeup_thread(mddev->sync_thread); |
1da177e4 | 3799 | err = 0; |
0fd62b86 NB |
3800 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
3801 | ||
9e653b63 | 3802 | } else |
1da177e4 | 3803 | err = -EINVAL; |
1da177e4 LT |
3804 | |
3805 | out: | |
3806 | return err; | |
3807 | } | |
3808 | ||
acc55e22 N |
3809 | /* similar to deny_write_access, but accounts for our holding a reference |
3810 | * to the file ourselves */ | |
3811 | static int deny_bitmap_write_access(struct file * file) | |
3812 | { | |
3813 | struct inode *inode = file->f_mapping->host; | |
3814 | ||
3815 | spin_lock(&inode->i_lock); | |
3816 | if (atomic_read(&inode->i_writecount) > 1) { | |
3817 | spin_unlock(&inode->i_lock); | |
3818 | return -ETXTBSY; | |
3819 | } | |
3820 | atomic_set(&inode->i_writecount, -1); | |
3821 | spin_unlock(&inode->i_lock); | |
3822 | ||
3823 | return 0; | |
3824 | } | |
3825 | ||
3826 | static void restore_bitmap_write_access(struct file *file) | |
3827 | { | |
3828 | struct inode *inode = file->f_mapping->host; | |
3829 | ||
3830 | spin_lock(&inode->i_lock); | |
3831 | atomic_set(&inode->i_writecount, 1); | |
3832 | spin_unlock(&inode->i_lock); | |
3833 | } | |
3834 | ||
9e653b63 N |
3835 | /* mode: |
3836 | * 0 - completely stop and dis-assemble array | |
3837 | * 1 - switch to readonly | |
3838 | * 2 - stop but do not disassemble array | |
3839 | */ | |
3840 | static int do_md_stop(mddev_t * mddev, int mode) | |
1da177e4 LT |
3841 | { |
3842 | int err = 0; | |
3843 | struct gendisk *disk = mddev->gendisk; | |
3844 | ||
3845 | if (mddev->pers) { | |
3846 | if (atomic_read(&mddev->active)>2) { | |
3847 | printk("md: %s still in use.\n",mdname(mddev)); | |
3848 | return -EBUSY; | |
3849 | } | |
3850 | ||
3851 | if (mddev->sync_thread) { | |
5fd6c1dc | 3852 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 LT |
3853 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
3854 | md_unregister_thread(mddev->sync_thread); | |
3855 | mddev->sync_thread = NULL; | |
3856 | } | |
3857 | ||
3858 | del_timer_sync(&mddev->safemode_timer); | |
3859 | ||
3860 | invalidate_partition(disk, 0); | |
3861 | ||
9e653b63 N |
3862 | switch(mode) { |
3863 | case 1: /* readonly */ | |
1da177e4 | 3864 | err = -ENXIO; |
f91de92e | 3865 | if (mddev->ro==1) |
1da177e4 LT |
3866 | goto out; |
3867 | mddev->ro = 1; | |
9e653b63 N |
3868 | break; |
3869 | case 0: /* disassemble */ | |
3870 | case 2: /* stop */ | |
6b8b3e8a | 3871 | bitmap_flush(mddev); |
a9701a30 | 3872 | md_super_wait(mddev); |
1da177e4 LT |
3873 | if (mddev->ro) |
3874 | set_disk_ro(disk, 0); | |
3875 | blk_queue_make_request(mddev->queue, md_fail_request); | |
3876 | mddev->pers->stop(mddev); | |
d1b5380c N |
3877 | mddev->queue->merge_bvec_fn = NULL; |
3878 | mddev->queue->unplug_fn = NULL; | |
041ae52e | 3879 | mddev->queue->backing_dev_info.congested_fn = NULL; |
411036fa N |
3880 | if (mddev->pers->sync_request) |
3881 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
3882 | ||
1da177e4 LT |
3883 | module_put(mddev->pers->owner); |
3884 | mddev->pers = NULL; | |
4f54b0e9 DW |
3885 | /* tell userspace to handle 'inactive' */ |
3886 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
0d4ca600 N |
3887 | |
3888 | set_capacity(disk, 0); | |
44ce6294 | 3889 | mddev->changed = 1; |
0d4ca600 | 3890 | |
1da177e4 LT |
3891 | if (mddev->ro) |
3892 | mddev->ro = 0; | |
3893 | } | |
850b2b42 | 3894 | if (!mddev->in_sync || mddev->flags) { |
1da177e4 LT |
3895 | /* mark array as shutdown cleanly */ |
3896 | mddev->in_sync = 1; | |
850b2b42 | 3897 | md_update_sb(mddev, 1); |
1da177e4 | 3898 | } |
9e653b63 | 3899 | if (mode == 1) |
1da177e4 | 3900 | set_disk_ro(disk, 1); |
5fd6c1dc | 3901 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
1da177e4 | 3902 | } |
32a7627c | 3903 | |
1da177e4 LT |
3904 | /* |
3905 | * Free resources if final stop | |
3906 | */ | |
9e653b63 | 3907 | if (mode == 0) { |
86e6ffdd N |
3908 | mdk_rdev_t *rdev; |
3909 | struct list_head *tmp; | |
0d4ca600 | 3910 | |
1da177e4 LT |
3911 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
3912 | ||
978f946b N |
3913 | bitmap_destroy(mddev); |
3914 | if (mddev->bitmap_file) { | |
acc55e22 | 3915 | restore_bitmap_write_access(mddev->bitmap_file); |
978f946b N |
3916 | fput(mddev->bitmap_file); |
3917 | mddev->bitmap_file = NULL; | |
3918 | } | |
3919 | mddev->bitmap_offset = 0; | |
3920 | ||
d089c6af | 3921 | rdev_for_each(rdev, tmp, mddev) |
86e6ffdd N |
3922 | if (rdev->raid_disk >= 0) { |
3923 | char nm[20]; | |
3924 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3925 | sysfs_remove_link(&mddev->kobj, nm); | |
3926 | } | |
3927 | ||
177a99b2 | 3928 | /* make sure all md_delayed_delete calls have finished */ |
5792a285 N |
3929 | flush_scheduled_work(); |
3930 | ||
1da177e4 LT |
3931 | export_array(mddev); |
3932 | ||
3933 | mddev->array_size = 0; | |
9e653b63 N |
3934 | mddev->size = 0; |
3935 | mddev->raid_disks = 0; | |
a94213b1 | 3936 | mddev->recovery_cp = 0; |
5e96ee65 | 3937 | mddev->resync_min = 0; |
c6207277 | 3938 | mddev->resync_max = MaxSector; |
08a02ecd | 3939 | mddev->reshape_position = MaxSector; |
e691063a | 3940 | mddev->external = 0; |
1ec4a939 | 3941 | mddev->persistent = 0; |
d897dbf9 N |
3942 | mddev->level = LEVEL_NONE; |
3943 | mddev->clevel[0] = 0; | |
3944 | mddev->flags = 0; | |
3945 | mddev->ro = 0; | |
3946 | mddev->metadata_type[0] = 0; | |
3947 | mddev->chunk_size = 0; | |
3948 | mddev->ctime = mddev->utime = 0; | |
3949 | mddev->layout = 0; | |
3950 | mddev->max_disks = 0; | |
3951 | mddev->events = 0; | |
3952 | mddev->delta_disks = 0; | |
3953 | mddev->new_level = LEVEL_NONE; | |
3954 | mddev->new_layout = 0; | |
3955 | mddev->new_chunk = 0; | |
3956 | mddev->curr_resync = 0; | |
3957 | mddev->resync_mismatches = 0; | |
3958 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
3959 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
3960 | mddev->recovery = 0; | |
3961 | mddev->in_sync = 0; | |
3962 | mddev->changed = 0; | |
3963 | mddev->degraded = 0; | |
3964 | mddev->barriers_work = 0; | |
3965 | mddev->safemode = 0; | |
9e653b63 | 3966 | |
a8a55c38 | 3967 | } else if (mddev->pers) |
1da177e4 LT |
3968 | printk(KERN_INFO "md: %s switched to read-only mode.\n", |
3969 | mdname(mddev)); | |
3970 | err = 0; | |
d7603b7e | 3971 | md_new_event(mddev); |
0fd62b86 | 3972 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
1da177e4 LT |
3973 | out: |
3974 | return err; | |
3975 | } | |
3976 | ||
fdee8ae4 | 3977 | #ifndef MODULE |
1da177e4 LT |
3978 | static void autorun_array(mddev_t *mddev) |
3979 | { | |
3980 | mdk_rdev_t *rdev; | |
3981 | struct list_head *tmp; | |
3982 | int err; | |
3983 | ||
a757e64c | 3984 | if (list_empty(&mddev->disks)) |
1da177e4 | 3985 | return; |
1da177e4 LT |
3986 | |
3987 | printk(KERN_INFO "md: running: "); | |
3988 | ||
d089c6af | 3989 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
3990 | char b[BDEVNAME_SIZE]; |
3991 | printk("<%s>", bdevname(rdev->bdev,b)); | |
3992 | } | |
3993 | printk("\n"); | |
3994 | ||
3995 | err = do_md_run (mddev); | |
3996 | if (err) { | |
3997 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
3998 | do_md_stop (mddev, 0); | |
3999 | } | |
4000 | } | |
4001 | ||
4002 | /* | |
4003 | * lets try to run arrays based on all disks that have arrived | |
4004 | * until now. (those are in pending_raid_disks) | |
4005 | * | |
4006 | * the method: pick the first pending disk, collect all disks with | |
4007 | * the same UUID, remove all from the pending list and put them into | |
4008 | * the 'same_array' list. Then order this list based on superblock | |
4009 | * update time (freshest comes first), kick out 'old' disks and | |
4010 | * compare superblocks. If everything's fine then run it. | |
4011 | * | |
4012 | * If "unit" is allocated, then bump its reference count | |
4013 | */ | |
4014 | static void autorun_devices(int part) | |
4015 | { | |
1da177e4 LT |
4016 | struct list_head *tmp; |
4017 | mdk_rdev_t *rdev0, *rdev; | |
4018 | mddev_t *mddev; | |
4019 | char b[BDEVNAME_SIZE]; | |
4020 | ||
4021 | printk(KERN_INFO "md: autorun ...\n"); | |
4022 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4023 | int unit; |
1da177e4 | 4024 | dev_t dev; |
ad01c9e3 | 4025 | LIST_HEAD(candidates); |
1da177e4 LT |
4026 | rdev0 = list_entry(pending_raid_disks.next, |
4027 | mdk_rdev_t, same_set); | |
4028 | ||
4029 | printk(KERN_INFO "md: considering %s ...\n", | |
4030 | bdevname(rdev0->bdev,b)); | |
4031 | INIT_LIST_HEAD(&candidates); | |
73c34431 | 4032 | rdev_for_each_list(rdev, tmp, pending_raid_disks) |
1da177e4 LT |
4033 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4034 | printk(KERN_INFO "md: adding %s ...\n", | |
4035 | bdevname(rdev->bdev,b)); | |
4036 | list_move(&rdev->same_set, &candidates); | |
4037 | } | |
4038 | /* | |
4039 | * now we have a set of devices, with all of them having | |
4040 | * mostly sane superblocks. It's time to allocate the | |
4041 | * mddev. | |
4042 | */ | |
e8703fe1 N |
4043 | if (part) { |
4044 | dev = MKDEV(mdp_major, | |
4045 | rdev0->preferred_minor << MdpMinorShift); | |
4046 | unit = MINOR(dev) >> MdpMinorShift; | |
4047 | } else { | |
4048 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4049 | unit = MINOR(dev); | |
4050 | } | |
4051 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4052 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4053 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4054 | break; | |
4055 | } | |
1da177e4 LT |
4056 | |
4057 | md_probe(dev, NULL, NULL); | |
4058 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4059 | if (!mddev || !mddev->gendisk) { |
4060 | if (mddev) | |
4061 | mddev_put(mddev); | |
4062 | printk(KERN_ERR | |
1da177e4 LT |
4063 | "md: cannot allocate memory for md drive.\n"); |
4064 | break; | |
4065 | } | |
4066 | if (mddev_lock(mddev)) | |
4067 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4068 | mdname(mddev)); | |
4069 | else if (mddev->raid_disks || mddev->major_version | |
4070 | || !list_empty(&mddev->disks)) { | |
4071 | printk(KERN_WARNING | |
4072 | "md: %s already running, cannot run %s\n", | |
4073 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4074 | mddev_unlock(mddev); | |
4075 | } else { | |
4076 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4077 | mddev->persistent = 1; |
73c34431 | 4078 | rdev_for_each_list(rdev, tmp, candidates) { |
1da177e4 LT |
4079 | list_del_init(&rdev->same_set); |
4080 | if (bind_rdev_to_array(rdev, mddev)) | |
4081 | export_rdev(rdev); | |
4082 | } | |
4083 | autorun_array(mddev); | |
4084 | mddev_unlock(mddev); | |
4085 | } | |
4086 | /* on success, candidates will be empty, on error | |
4087 | * it won't... | |
4088 | */ | |
73c34431 | 4089 | rdev_for_each_list(rdev, tmp, candidates) |
1da177e4 LT |
4090 | export_rdev(rdev); |
4091 | mddev_put(mddev); | |
4092 | } | |
4093 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4094 | } | |
fdee8ae4 | 4095 | #endif /* !MODULE */ |
1da177e4 | 4096 | |
1da177e4 LT |
4097 | static int get_version(void __user * arg) |
4098 | { | |
4099 | mdu_version_t ver; | |
4100 | ||
4101 | ver.major = MD_MAJOR_VERSION; | |
4102 | ver.minor = MD_MINOR_VERSION; | |
4103 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4104 | ||
4105 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4106 | return -EFAULT; | |
4107 | ||
4108 | return 0; | |
4109 | } | |
4110 | ||
4111 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4112 | { | |
4113 | mdu_array_info_t info; | |
4114 | int nr,working,active,failed,spare; | |
4115 | mdk_rdev_t *rdev; | |
4116 | struct list_head *tmp; | |
4117 | ||
4118 | nr=working=active=failed=spare=0; | |
d089c6af | 4119 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 | 4120 | nr++; |
b2d444d7 | 4121 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4122 | failed++; |
4123 | else { | |
4124 | working++; | |
b2d444d7 | 4125 | if (test_bit(In_sync, &rdev->flags)) |
1da177e4 LT |
4126 | active++; |
4127 | else | |
4128 | spare++; | |
4129 | } | |
4130 | } | |
4131 | ||
4132 | info.major_version = mddev->major_version; | |
4133 | info.minor_version = mddev->minor_version; | |
4134 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
4135 | info.ctime = mddev->ctime; | |
4136 | info.level = mddev->level; | |
4137 | info.size = mddev->size; | |
284ae7ca N |
4138 | if (info.size != mddev->size) /* overflow */ |
4139 | info.size = -1; | |
1da177e4 LT |
4140 | info.nr_disks = nr; |
4141 | info.raid_disks = mddev->raid_disks; | |
4142 | info.md_minor = mddev->md_minor; | |
4143 | info.not_persistent= !mddev->persistent; | |
4144 | ||
4145 | info.utime = mddev->utime; | |
4146 | info.state = 0; | |
4147 | if (mddev->in_sync) | |
4148 | info.state = (1<<MD_SB_CLEAN); | |
36fa3063 N |
4149 | if (mddev->bitmap && mddev->bitmap_offset) |
4150 | info.state = (1<<MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
4151 | info.active_disks = active; |
4152 | info.working_disks = working; | |
4153 | info.failed_disks = failed; | |
4154 | info.spare_disks = spare; | |
4155 | ||
4156 | info.layout = mddev->layout; | |
4157 | info.chunk_size = mddev->chunk_size; | |
4158 | ||
4159 | if (copy_to_user(arg, &info, sizeof(info))) | |
4160 | return -EFAULT; | |
4161 | ||
4162 | return 0; | |
4163 | } | |
4164 | ||
87162a28 | 4165 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
4166 | { |
4167 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
4168 | char *ptr, *buf = NULL; | |
4169 | int err = -ENOMEM; | |
4170 | ||
b5470dc5 DW |
4171 | if (md_allow_write(mddev)) |
4172 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
4173 | else | |
4174 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 4175 | |
32a7627c N |
4176 | if (!file) |
4177 | goto out; | |
4178 | ||
4179 | /* bitmap disabled, zero the first byte and copy out */ | |
4180 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
4181 | file->pathname[0] = '\0'; | |
4182 | goto copy_out; | |
4183 | } | |
4184 | ||
4185 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
4186 | if (!buf) | |
4187 | goto out; | |
4188 | ||
6bcfd601 CH |
4189 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
4190 | if (IS_ERR(ptr)) | |
32a7627c N |
4191 | goto out; |
4192 | ||
4193 | strcpy(file->pathname, ptr); | |
4194 | ||
4195 | copy_out: | |
4196 | err = 0; | |
4197 | if (copy_to_user(arg, file, sizeof(*file))) | |
4198 | err = -EFAULT; | |
4199 | out: | |
4200 | kfree(buf); | |
4201 | kfree(file); | |
4202 | return err; | |
4203 | } | |
4204 | ||
1da177e4 LT |
4205 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
4206 | { | |
4207 | mdu_disk_info_t info; | |
4208 | unsigned int nr; | |
4209 | mdk_rdev_t *rdev; | |
4210 | ||
4211 | if (copy_from_user(&info, arg, sizeof(info))) | |
4212 | return -EFAULT; | |
4213 | ||
4214 | nr = info.number; | |
4215 | ||
4216 | rdev = find_rdev_nr(mddev, nr); | |
4217 | if (rdev) { | |
4218 | info.major = MAJOR(rdev->bdev->bd_dev); | |
4219 | info.minor = MINOR(rdev->bdev->bd_dev); | |
4220 | info.raid_disk = rdev->raid_disk; | |
4221 | info.state = 0; | |
b2d444d7 | 4222 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 4223 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 4224 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4225 | info.state |= (1<<MD_DISK_ACTIVE); |
4226 | info.state |= (1<<MD_DISK_SYNC); | |
4227 | } | |
8ddf9efe N |
4228 | if (test_bit(WriteMostly, &rdev->flags)) |
4229 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
4230 | } else { |
4231 | info.major = info.minor = 0; | |
4232 | info.raid_disk = -1; | |
4233 | info.state = (1<<MD_DISK_REMOVED); | |
4234 | } | |
4235 | ||
4236 | if (copy_to_user(arg, &info, sizeof(info))) | |
4237 | return -EFAULT; | |
4238 | ||
4239 | return 0; | |
4240 | } | |
4241 | ||
4242 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
4243 | { | |
4244 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4245 | mdk_rdev_t *rdev; | |
4246 | dev_t dev = MKDEV(info->major,info->minor); | |
4247 | ||
4248 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
4249 | return -EOVERFLOW; | |
4250 | ||
4251 | if (!mddev->raid_disks) { | |
4252 | int err; | |
4253 | /* expecting a device which has a superblock */ | |
4254 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
4255 | if (IS_ERR(rdev)) { | |
4256 | printk(KERN_WARNING | |
4257 | "md: md_import_device returned %ld\n", | |
4258 | PTR_ERR(rdev)); | |
4259 | return PTR_ERR(rdev); | |
4260 | } | |
4261 | if (!list_empty(&mddev->disks)) { | |
4262 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
4263 | mdk_rdev_t, same_set); | |
4264 | int err = super_types[mddev->major_version] | |
4265 | .load_super(rdev, rdev0, mddev->minor_version); | |
4266 | if (err < 0) { | |
4267 | printk(KERN_WARNING | |
4268 | "md: %s has different UUID to %s\n", | |
4269 | bdevname(rdev->bdev,b), | |
4270 | bdevname(rdev0->bdev,b2)); | |
4271 | export_rdev(rdev); | |
4272 | return -EINVAL; | |
4273 | } | |
4274 | } | |
4275 | err = bind_rdev_to_array(rdev, mddev); | |
4276 | if (err) | |
4277 | export_rdev(rdev); | |
4278 | return err; | |
4279 | } | |
4280 | ||
4281 | /* | |
4282 | * add_new_disk can be used once the array is assembled | |
4283 | * to add "hot spares". They must already have a superblock | |
4284 | * written | |
4285 | */ | |
4286 | if (mddev->pers) { | |
4287 | int err; | |
4288 | if (!mddev->pers->hot_add_disk) { | |
4289 | printk(KERN_WARNING | |
4290 | "%s: personality does not support diskops!\n", | |
4291 | mdname(mddev)); | |
4292 | return -EINVAL; | |
4293 | } | |
7b1e35f6 N |
4294 | if (mddev->persistent) |
4295 | rdev = md_import_device(dev, mddev->major_version, | |
4296 | mddev->minor_version); | |
4297 | else | |
4298 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
4299 | if (IS_ERR(rdev)) { |
4300 | printk(KERN_WARNING | |
4301 | "md: md_import_device returned %ld\n", | |
4302 | PTR_ERR(rdev)); | |
4303 | return PTR_ERR(rdev); | |
4304 | } | |
41158c7e N |
4305 | /* set save_raid_disk if appropriate */ |
4306 | if (!mddev->persistent) { | |
4307 | if (info->state & (1<<MD_DISK_SYNC) && | |
4308 | info->raid_disk < mddev->raid_disks) | |
4309 | rdev->raid_disk = info->raid_disk; | |
4310 | else | |
4311 | rdev->raid_disk = -1; | |
4312 | } else | |
4313 | super_types[mddev->major_version]. | |
4314 | validate_super(mddev, rdev); | |
4315 | rdev->saved_raid_disk = rdev->raid_disk; | |
4316 | ||
b2d444d7 | 4317 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
4318 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4319 | set_bit(WriteMostly, &rdev->flags); | |
4320 | ||
1da177e4 LT |
4321 | rdev->raid_disk = -1; |
4322 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
4323 | if (!err && !mddev->pers->hot_remove_disk) { |
4324 | /* If there is hot_add_disk but no hot_remove_disk | |
4325 | * then added disks for geometry changes, | |
4326 | * and should be added immediately. | |
4327 | */ | |
4328 | super_types[mddev->major_version]. | |
4329 | validate_super(mddev, rdev); | |
4330 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
4331 | if (err) | |
4332 | unbind_rdev_from_array(rdev); | |
4333 | } | |
1da177e4 LT |
4334 | if (err) |
4335 | export_rdev(rdev); | |
52664732 NB |
4336 | else |
4337 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
c361777f | 4338 | |
17571284 | 4339 | md_update_sb(mddev, 1); |
72a23c21 NB |
4340 | if (mddev->degraded) |
4341 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 4342 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 4343 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4344 | return err; |
4345 | } | |
4346 | ||
4347 | /* otherwise, add_new_disk is only allowed | |
4348 | * for major_version==0 superblocks | |
4349 | */ | |
4350 | if (mddev->major_version != 0) { | |
4351 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
4352 | mdname(mddev)); | |
4353 | return -EINVAL; | |
4354 | } | |
4355 | ||
4356 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
4357 | int err; | |
4358 | rdev = md_import_device (dev, -1, 0); | |
4359 | if (IS_ERR(rdev)) { | |
4360 | printk(KERN_WARNING | |
4361 | "md: error, md_import_device() returned %ld\n", | |
4362 | PTR_ERR(rdev)); | |
4363 | return PTR_ERR(rdev); | |
4364 | } | |
4365 | rdev->desc_nr = info->number; | |
4366 | if (info->raid_disk < mddev->raid_disks) | |
4367 | rdev->raid_disk = info->raid_disk; | |
4368 | else | |
4369 | rdev->raid_disk = -1; | |
4370 | ||
1da177e4 | 4371 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
4372 | if (info->state & (1<<MD_DISK_SYNC)) |
4373 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 4374 | |
8ddf9efe N |
4375 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
4376 | set_bit(WriteMostly, &rdev->flags); | |
4377 | ||
1da177e4 LT |
4378 | if (!mddev->persistent) { |
4379 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
4380 | rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
4381 | } else | |
4382 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
4383 | rdev->size = calc_dev_size(rdev, mddev->chunk_size); | |
4384 | ||
2bf071bf N |
4385 | err = bind_rdev_to_array(rdev, mddev); |
4386 | if (err) { | |
4387 | export_rdev(rdev); | |
4388 | return err; | |
4389 | } | |
1da177e4 LT |
4390 | } |
4391 | ||
4392 | return 0; | |
4393 | } | |
4394 | ||
4395 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
4396 | { | |
4397 | char b[BDEVNAME_SIZE]; | |
4398 | mdk_rdev_t *rdev; | |
4399 | ||
1da177e4 LT |
4400 | rdev = find_rdev(mddev, dev); |
4401 | if (!rdev) | |
4402 | return -ENXIO; | |
4403 | ||
4404 | if (rdev->raid_disk >= 0) | |
4405 | goto busy; | |
4406 | ||
4407 | kick_rdev_from_array(rdev); | |
850b2b42 | 4408 | md_update_sb(mddev, 1); |
d7603b7e | 4409 | md_new_event(mddev); |
1da177e4 LT |
4410 | |
4411 | return 0; | |
4412 | busy: | |
fdefa4d8 | 4413 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
4414 | bdevname(rdev->bdev,b), mdname(mddev)); |
4415 | return -EBUSY; | |
4416 | } | |
4417 | ||
4418 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
4419 | { | |
4420 | char b[BDEVNAME_SIZE]; | |
4421 | int err; | |
4422 | unsigned int size; | |
4423 | mdk_rdev_t *rdev; | |
4424 | ||
4425 | if (!mddev->pers) | |
4426 | return -ENODEV; | |
4427 | ||
4428 | if (mddev->major_version != 0) { | |
4429 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
4430 | " version-0 superblocks.\n", | |
4431 | mdname(mddev)); | |
4432 | return -EINVAL; | |
4433 | } | |
4434 | if (!mddev->pers->hot_add_disk) { | |
4435 | printk(KERN_WARNING | |
4436 | "%s: personality does not support diskops!\n", | |
4437 | mdname(mddev)); | |
4438 | return -EINVAL; | |
4439 | } | |
4440 | ||
4441 | rdev = md_import_device (dev, -1, 0); | |
4442 | if (IS_ERR(rdev)) { | |
4443 | printk(KERN_WARNING | |
4444 | "md: error, md_import_device() returned %ld\n", | |
4445 | PTR_ERR(rdev)); | |
4446 | return -EINVAL; | |
4447 | } | |
4448 | ||
4449 | if (mddev->persistent) | |
4450 | rdev->sb_offset = calc_dev_sboffset(rdev->bdev); | |
4451 | else | |
4452 | rdev->sb_offset = | |
4453 | rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; | |
4454 | ||
4455 | size = calc_dev_size(rdev, mddev->chunk_size); | |
4456 | rdev->size = size; | |
4457 | ||
b2d444d7 | 4458 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
4459 | printk(KERN_WARNING |
4460 | "md: can not hot-add faulty %s disk to %s!\n", | |
4461 | bdevname(rdev->bdev,b), mdname(mddev)); | |
4462 | err = -EINVAL; | |
4463 | goto abort_export; | |
4464 | } | |
b2d444d7 | 4465 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4466 | rdev->desc_nr = -1; |
5842730d | 4467 | rdev->saved_raid_disk = -1; |
2bf071bf N |
4468 | err = bind_rdev_to_array(rdev, mddev); |
4469 | if (err) | |
4470 | goto abort_export; | |
1da177e4 LT |
4471 | |
4472 | /* | |
4473 | * The rest should better be atomic, we can have disk failures | |
4474 | * noticed in interrupt contexts ... | |
4475 | */ | |
4476 | ||
4477 | if (rdev->desc_nr == mddev->max_disks) { | |
4478 | printk(KERN_WARNING "%s: can not hot-add to full array!\n", | |
4479 | mdname(mddev)); | |
4480 | err = -EBUSY; | |
4481 | goto abort_unbind_export; | |
4482 | } | |
4483 | ||
4484 | rdev->raid_disk = -1; | |
4485 | ||
850b2b42 | 4486 | md_update_sb(mddev, 1); |
1da177e4 LT |
4487 | |
4488 | /* | |
4489 | * Kick recovery, maybe this spare has to be added to the | |
4490 | * array immediately. | |
4491 | */ | |
4492 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4493 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 4494 | md_new_event(mddev); |
1da177e4 LT |
4495 | return 0; |
4496 | ||
4497 | abort_unbind_export: | |
4498 | unbind_rdev_from_array(rdev); | |
4499 | ||
4500 | abort_export: | |
4501 | export_rdev(rdev); | |
4502 | return err; | |
4503 | } | |
4504 | ||
32a7627c N |
4505 | static int set_bitmap_file(mddev_t *mddev, int fd) |
4506 | { | |
4507 | int err; | |
4508 | ||
36fa3063 N |
4509 | if (mddev->pers) { |
4510 | if (!mddev->pers->quiesce) | |
4511 | return -EBUSY; | |
4512 | if (mddev->recovery || mddev->sync_thread) | |
4513 | return -EBUSY; | |
4514 | /* we should be able to change the bitmap.. */ | |
4515 | } | |
32a7627c | 4516 | |
32a7627c | 4517 | |
36fa3063 N |
4518 | if (fd >= 0) { |
4519 | if (mddev->bitmap) | |
4520 | return -EEXIST; /* cannot add when bitmap is present */ | |
4521 | mddev->bitmap_file = fget(fd); | |
32a7627c | 4522 | |
36fa3063 N |
4523 | if (mddev->bitmap_file == NULL) { |
4524 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", | |
4525 | mdname(mddev)); | |
4526 | return -EBADF; | |
4527 | } | |
4528 | ||
4529 | err = deny_bitmap_write_access(mddev->bitmap_file); | |
4530 | if (err) { | |
4531 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
4532 | mdname(mddev)); | |
4533 | fput(mddev->bitmap_file); | |
4534 | mddev->bitmap_file = NULL; | |
4535 | return err; | |
4536 | } | |
a654b9d8 | 4537 | mddev->bitmap_offset = 0; /* file overrides offset */ |
36fa3063 N |
4538 | } else if (mddev->bitmap == NULL) |
4539 | return -ENOENT; /* cannot remove what isn't there */ | |
4540 | err = 0; | |
4541 | if (mddev->pers) { | |
4542 | mddev->pers->quiesce(mddev, 1); | |
4543 | if (fd >= 0) | |
4544 | err = bitmap_create(mddev); | |
d7375ab3 | 4545 | if (fd < 0 || err) { |
36fa3063 | 4546 | bitmap_destroy(mddev); |
d7375ab3 N |
4547 | fd = -1; /* make sure to put the file */ |
4548 | } | |
36fa3063 | 4549 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
4550 | } |
4551 | if (fd < 0) { | |
acc55e22 N |
4552 | if (mddev->bitmap_file) { |
4553 | restore_bitmap_write_access(mddev->bitmap_file); | |
36fa3063 | 4554 | fput(mddev->bitmap_file); |
acc55e22 | 4555 | } |
36fa3063 N |
4556 | mddev->bitmap_file = NULL; |
4557 | } | |
4558 | ||
32a7627c N |
4559 | return err; |
4560 | } | |
4561 | ||
1da177e4 LT |
4562 | /* |
4563 | * set_array_info is used two different ways | |
4564 | * The original usage is when creating a new array. | |
4565 | * In this usage, raid_disks is > 0 and it together with | |
4566 | * level, size, not_persistent,layout,chunksize determine the | |
4567 | * shape of the array. | |
4568 | * This will always create an array with a type-0.90.0 superblock. | |
4569 | * The newer usage is when assembling an array. | |
4570 | * In this case raid_disks will be 0, and the major_version field is | |
4571 | * use to determine which style super-blocks are to be found on the devices. | |
4572 | * The minor and patch _version numbers are also kept incase the | |
4573 | * super_block handler wishes to interpret them. | |
4574 | */ | |
4575 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
4576 | { | |
4577 | ||
4578 | if (info->raid_disks == 0) { | |
4579 | /* just setting version number for superblock loading */ | |
4580 | if (info->major_version < 0 || | |
50511da3 | 4581 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
4582 | super_types[info->major_version].name == NULL) { |
4583 | /* maybe try to auto-load a module? */ | |
4584 | printk(KERN_INFO | |
4585 | "md: superblock version %d not known\n", | |
4586 | info->major_version); | |
4587 | return -EINVAL; | |
4588 | } | |
4589 | mddev->major_version = info->major_version; | |
4590 | mddev->minor_version = info->minor_version; | |
4591 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 4592 | mddev->persistent = !info->not_persistent; |
1da177e4 LT |
4593 | return 0; |
4594 | } | |
4595 | mddev->major_version = MD_MAJOR_VERSION; | |
4596 | mddev->minor_version = MD_MINOR_VERSION; | |
4597 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
4598 | mddev->ctime = get_seconds(); | |
4599 | ||
4600 | mddev->level = info->level; | |
17115e03 | 4601 | mddev->clevel[0] = 0; |
1da177e4 LT |
4602 | mddev->size = info->size; |
4603 | mddev->raid_disks = info->raid_disks; | |
4604 | /* don't set md_minor, it is determined by which /dev/md* was | |
4605 | * openned | |
4606 | */ | |
4607 | if (info->state & (1<<MD_SB_CLEAN)) | |
4608 | mddev->recovery_cp = MaxSector; | |
4609 | else | |
4610 | mddev->recovery_cp = 0; | |
4611 | mddev->persistent = ! info->not_persistent; | |
e691063a | 4612 | mddev->external = 0; |
1da177e4 LT |
4613 | |
4614 | mddev->layout = info->layout; | |
4615 | mddev->chunk_size = info->chunk_size; | |
4616 | ||
4617 | mddev->max_disks = MD_SB_DISKS; | |
4618 | ||
e691063a N |
4619 | if (mddev->persistent) |
4620 | mddev->flags = 0; | |
850b2b42 | 4621 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 4622 | |
b2a2703c N |
4623 | mddev->default_bitmap_offset = MD_SB_BYTES >> 9; |
4624 | mddev->bitmap_offset = 0; | |
4625 | ||
f6705578 N |
4626 | mddev->reshape_position = MaxSector; |
4627 | ||
1da177e4 LT |
4628 | /* |
4629 | * Generate a 128 bit UUID | |
4630 | */ | |
4631 | get_random_bytes(mddev->uuid, 16); | |
4632 | ||
f6705578 N |
4633 | mddev->new_level = mddev->level; |
4634 | mddev->new_chunk = mddev->chunk_size; | |
4635 | mddev->new_layout = mddev->layout; | |
4636 | mddev->delta_disks = 0; | |
4637 | ||
1da177e4 LT |
4638 | return 0; |
4639 | } | |
4640 | ||
a35b0d69 N |
4641 | static int update_size(mddev_t *mddev, unsigned long size) |
4642 | { | |
4643 | mdk_rdev_t * rdev; | |
4644 | int rv; | |
4645 | struct list_head *tmp; | |
8ddeeae5 | 4646 | int fit = (size == 0); |
a35b0d69 N |
4647 | |
4648 | if (mddev->pers->resize == NULL) | |
4649 | return -EINVAL; | |
4650 | /* The "size" is the amount of each device that is used. | |
4651 | * This can only make sense for arrays with redundancy. | |
4652 | * linear and raid0 always use whatever space is available | |
4653 | * We can only consider changing the size if no resync | |
4654 | * or reconstruction is happening, and if the new size | |
4655 | * is acceptable. It must fit before the sb_offset or, | |
4656 | * if that is <data_offset, it must fit before the | |
4657 | * size of each device. | |
4658 | * If size is zero, we find the largest size that fits. | |
4659 | */ | |
4660 | if (mddev->sync_thread) | |
4661 | return -EBUSY; | |
d089c6af | 4662 | rdev_for_each(rdev, tmp, mddev) { |
a35b0d69 | 4663 | sector_t avail; |
01ab5662 N |
4664 | avail = rdev->size * 2; |
4665 | ||
a35b0d69 N |
4666 | if (fit && (size == 0 || size > avail/2)) |
4667 | size = avail/2; | |
4668 | if (avail < ((sector_t)size << 1)) | |
4669 | return -ENOSPC; | |
4670 | } | |
4671 | rv = mddev->pers->resize(mddev, (sector_t)size *2); | |
4672 | if (!rv) { | |
4673 | struct block_device *bdev; | |
4674 | ||
4675 | bdev = bdget_disk(mddev->gendisk, 0); | |
4676 | if (bdev) { | |
1b1dcc1b | 4677 | mutex_lock(&bdev->bd_inode->i_mutex); |
6d89332b | 4678 | i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10); |
1b1dcc1b | 4679 | mutex_unlock(&bdev->bd_inode->i_mutex); |
a35b0d69 N |
4680 | bdput(bdev); |
4681 | } | |
4682 | } | |
4683 | return rv; | |
4684 | } | |
4685 | ||
da943b99 N |
4686 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
4687 | { | |
4688 | int rv; | |
4689 | /* change the number of raid disks */ | |
63c70c4f | 4690 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
4691 | return -EINVAL; |
4692 | if (raid_disks <= 0 || | |
4693 | raid_disks >= mddev->max_disks) | |
4694 | return -EINVAL; | |
63c70c4f | 4695 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 4696 | return -EBUSY; |
63c70c4f N |
4697 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
4698 | ||
4699 | rv = mddev->pers->check_reshape(mddev); | |
da943b99 N |
4700 | return rv; |
4701 | } | |
4702 | ||
4703 | ||
1da177e4 LT |
4704 | /* |
4705 | * update_array_info is used to change the configuration of an | |
4706 | * on-line array. | |
4707 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
4708 | * fields in the info are checked against the array. | |
4709 | * Any differences that cannot be handled will cause an error. | |
4710 | * Normally, only one change can be managed at a time. | |
4711 | */ | |
4712 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
4713 | { | |
4714 | int rv = 0; | |
4715 | int cnt = 0; | |
36fa3063 N |
4716 | int state = 0; |
4717 | ||
4718 | /* calculate expected state,ignoring low bits */ | |
4719 | if (mddev->bitmap && mddev->bitmap_offset) | |
4720 | state |= (1 << MD_SB_BITMAP_PRESENT); | |
1da177e4 LT |
4721 | |
4722 | if (mddev->major_version != info->major_version || | |
4723 | mddev->minor_version != info->minor_version || | |
4724 | /* mddev->patch_version != info->patch_version || */ | |
4725 | mddev->ctime != info->ctime || | |
4726 | mddev->level != info->level || | |
4727 | /* mddev->layout != info->layout || */ | |
4728 | !mddev->persistent != info->not_persistent|| | |
36fa3063 N |
4729 | mddev->chunk_size != info->chunk_size || |
4730 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ | |
4731 | ((state^info->state) & 0xfffffe00) | |
4732 | ) | |
1da177e4 LT |
4733 | return -EINVAL; |
4734 | /* Check there is only one change */ | |
284ae7ca | 4735 | if (info->size >= 0 && mddev->size != info->size) cnt++; |
1da177e4 LT |
4736 | if (mddev->raid_disks != info->raid_disks) cnt++; |
4737 | if (mddev->layout != info->layout) cnt++; | |
36fa3063 | 4738 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++; |
1da177e4 LT |
4739 | if (cnt == 0) return 0; |
4740 | if (cnt > 1) return -EINVAL; | |
4741 | ||
4742 | if (mddev->layout != info->layout) { | |
4743 | /* Change layout | |
4744 | * we don't need to do anything at the md level, the | |
4745 | * personality will take care of it all. | |
4746 | */ | |
4747 | if (mddev->pers->reconfig == NULL) | |
4748 | return -EINVAL; | |
4749 | else | |
4750 | return mddev->pers->reconfig(mddev, info->layout, -1); | |
4751 | } | |
284ae7ca | 4752 | if (info->size >= 0 && mddev->size != info->size) |
a35b0d69 N |
4753 | rv = update_size(mddev, info->size); |
4754 | ||
da943b99 N |
4755 | if (mddev->raid_disks != info->raid_disks) |
4756 | rv = update_raid_disks(mddev, info->raid_disks); | |
4757 | ||
36fa3063 N |
4758 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
4759 | if (mddev->pers->quiesce == NULL) | |
4760 | return -EINVAL; | |
4761 | if (mddev->recovery || mddev->sync_thread) | |
4762 | return -EBUSY; | |
4763 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
4764 | /* add the bitmap */ | |
4765 | if (mddev->bitmap) | |
4766 | return -EEXIST; | |
4767 | if (mddev->default_bitmap_offset == 0) | |
4768 | return -EINVAL; | |
4769 | mddev->bitmap_offset = mddev->default_bitmap_offset; | |
4770 | mddev->pers->quiesce(mddev, 1); | |
4771 | rv = bitmap_create(mddev); | |
4772 | if (rv) | |
4773 | bitmap_destroy(mddev); | |
4774 | mddev->pers->quiesce(mddev, 0); | |
4775 | } else { | |
4776 | /* remove the bitmap */ | |
4777 | if (!mddev->bitmap) | |
4778 | return -ENOENT; | |
4779 | if (mddev->bitmap->file) | |
4780 | return -EINVAL; | |
4781 | mddev->pers->quiesce(mddev, 1); | |
4782 | bitmap_destroy(mddev); | |
4783 | mddev->pers->quiesce(mddev, 0); | |
4784 | mddev->bitmap_offset = 0; | |
4785 | } | |
4786 | } | |
850b2b42 | 4787 | md_update_sb(mddev, 1); |
1da177e4 LT |
4788 | return rv; |
4789 | } | |
4790 | ||
4791 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
4792 | { | |
4793 | mdk_rdev_t *rdev; | |
4794 | ||
4795 | if (mddev->pers == NULL) | |
4796 | return -ENODEV; | |
4797 | ||
4798 | rdev = find_rdev(mddev, dev); | |
4799 | if (!rdev) | |
4800 | return -ENODEV; | |
4801 | ||
4802 | md_error(mddev, rdev); | |
4803 | return 0; | |
4804 | } | |
4805 | ||
2f9618ce AN |
4806 | /* |
4807 | * We have a problem here : there is no easy way to give a CHS | |
4808 | * virtual geometry. We currently pretend that we have a 2 heads | |
4809 | * 4 sectors (with a BIG number of cylinders...). This drives | |
4810 | * dosfs just mad... ;-) | |
4811 | */ | |
a885c8c4 CH |
4812 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
4813 | { | |
4814 | mddev_t *mddev = bdev->bd_disk->private_data; | |
4815 | ||
4816 | geo->heads = 2; | |
4817 | geo->sectors = 4; | |
4818 | geo->cylinders = get_capacity(mddev->gendisk) / 8; | |
4819 | return 0; | |
4820 | } | |
4821 | ||
1da177e4 LT |
4822 | static int md_ioctl(struct inode *inode, struct file *file, |
4823 | unsigned int cmd, unsigned long arg) | |
4824 | { | |
4825 | int err = 0; | |
4826 | void __user *argp = (void __user *)arg; | |
1da177e4 LT |
4827 | mddev_t *mddev = NULL; |
4828 | ||
4829 | if (!capable(CAP_SYS_ADMIN)) | |
4830 | return -EACCES; | |
4831 | ||
4832 | /* | |
4833 | * Commands dealing with the RAID driver but not any | |
4834 | * particular array: | |
4835 | */ | |
4836 | switch (cmd) | |
4837 | { | |
4838 | case RAID_VERSION: | |
4839 | err = get_version(argp); | |
4840 | goto done; | |
4841 | ||
4842 | case PRINT_RAID_DEBUG: | |
4843 | err = 0; | |
4844 | md_print_devices(); | |
4845 | goto done; | |
4846 | ||
4847 | #ifndef MODULE | |
4848 | case RAID_AUTORUN: | |
4849 | err = 0; | |
4850 | autostart_arrays(arg); | |
4851 | goto done; | |
4852 | #endif | |
4853 | default:; | |
4854 | } | |
4855 | ||
4856 | /* | |
4857 | * Commands creating/starting a new array: | |
4858 | */ | |
4859 | ||
4860 | mddev = inode->i_bdev->bd_disk->private_data; | |
4861 | ||
4862 | if (!mddev) { | |
4863 | BUG(); | |
4864 | goto abort; | |
4865 | } | |
4866 | ||
1da177e4 LT |
4867 | err = mddev_lock(mddev); |
4868 | if (err) { | |
4869 | printk(KERN_INFO | |
4870 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
4871 | err, cmd); | |
4872 | goto abort; | |
4873 | } | |
4874 | ||
4875 | switch (cmd) | |
4876 | { | |
4877 | case SET_ARRAY_INFO: | |
4878 | { | |
4879 | mdu_array_info_t info; | |
4880 | if (!arg) | |
4881 | memset(&info, 0, sizeof(info)); | |
4882 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
4883 | err = -EFAULT; | |
4884 | goto abort_unlock; | |
4885 | } | |
4886 | if (mddev->pers) { | |
4887 | err = update_array_info(mddev, &info); | |
4888 | if (err) { | |
4889 | printk(KERN_WARNING "md: couldn't update" | |
4890 | " array info. %d\n", err); | |
4891 | goto abort_unlock; | |
4892 | } | |
4893 | goto done_unlock; | |
4894 | } | |
4895 | if (!list_empty(&mddev->disks)) { | |
4896 | printk(KERN_WARNING | |
4897 | "md: array %s already has disks!\n", | |
4898 | mdname(mddev)); | |
4899 | err = -EBUSY; | |
4900 | goto abort_unlock; | |
4901 | } | |
4902 | if (mddev->raid_disks) { | |
4903 | printk(KERN_WARNING | |
4904 | "md: array %s already initialised!\n", | |
4905 | mdname(mddev)); | |
4906 | err = -EBUSY; | |
4907 | goto abort_unlock; | |
4908 | } | |
4909 | err = set_array_info(mddev, &info); | |
4910 | if (err) { | |
4911 | printk(KERN_WARNING "md: couldn't set" | |
4912 | " array info. %d\n", err); | |
4913 | goto abort_unlock; | |
4914 | } | |
4915 | } | |
4916 | goto done_unlock; | |
4917 | ||
4918 | default:; | |
4919 | } | |
4920 | ||
4921 | /* | |
4922 | * Commands querying/configuring an existing array: | |
4923 | */ | |
32a7627c | 4924 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 4925 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
4926 | if ((!mddev->raid_disks && !mddev->external) |
4927 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
4928 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
4929 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
4930 | err = -ENODEV; |
4931 | goto abort_unlock; | |
4932 | } | |
4933 | ||
4934 | /* | |
4935 | * Commands even a read-only array can execute: | |
4936 | */ | |
4937 | switch (cmd) | |
4938 | { | |
4939 | case GET_ARRAY_INFO: | |
4940 | err = get_array_info(mddev, argp); | |
4941 | goto done_unlock; | |
4942 | ||
32a7627c | 4943 | case GET_BITMAP_FILE: |
87162a28 | 4944 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
4945 | goto done_unlock; |
4946 | ||
1da177e4 LT |
4947 | case GET_DISK_INFO: |
4948 | err = get_disk_info(mddev, argp); | |
4949 | goto done_unlock; | |
4950 | ||
4951 | case RESTART_ARRAY_RW: | |
4952 | err = restart_array(mddev); | |
4953 | goto done_unlock; | |
4954 | ||
4955 | case STOP_ARRAY: | |
4956 | err = do_md_stop (mddev, 0); | |
4957 | goto done_unlock; | |
4958 | ||
4959 | case STOP_ARRAY_RO: | |
4960 | err = do_md_stop (mddev, 1); | |
4961 | goto done_unlock; | |
4962 | ||
1da177e4 LT |
4963 | } |
4964 | ||
4965 | /* | |
4966 | * The remaining ioctls are changing the state of the | |
f91de92e N |
4967 | * superblock, so we do not allow them on read-only arrays. |
4968 | * However non-MD ioctls (e.g. get-size) will still come through | |
4969 | * here and hit the 'default' below, so only disallow | |
4970 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 4971 | */ |
bb57fc64 | 4972 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
4973 | if (mddev->ro == 2) { |
4974 | mddev->ro = 0; | |
0fd62b86 NB |
4975 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
4976 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4977 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
4978 | } else { |
4979 | err = -EROFS; | |
4980 | goto abort_unlock; | |
4981 | } | |
1da177e4 LT |
4982 | } |
4983 | ||
4984 | switch (cmd) | |
4985 | { | |
4986 | case ADD_NEW_DISK: | |
4987 | { | |
4988 | mdu_disk_info_t info; | |
4989 | if (copy_from_user(&info, argp, sizeof(info))) | |
4990 | err = -EFAULT; | |
4991 | else | |
4992 | err = add_new_disk(mddev, &info); | |
4993 | goto done_unlock; | |
4994 | } | |
4995 | ||
4996 | case HOT_REMOVE_DISK: | |
4997 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
4998 | goto done_unlock; | |
4999 | ||
5000 | case HOT_ADD_DISK: | |
5001 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5002 | goto done_unlock; | |
5003 | ||
5004 | case SET_DISK_FAULTY: | |
5005 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5006 | goto done_unlock; | |
5007 | ||
5008 | case RUN_ARRAY: | |
5009 | err = do_md_run (mddev); | |
5010 | goto done_unlock; | |
5011 | ||
32a7627c N |
5012 | case SET_BITMAP_FILE: |
5013 | err = set_bitmap_file(mddev, (int)arg); | |
5014 | goto done_unlock; | |
5015 | ||
1da177e4 | 5016 | default: |
1da177e4 LT |
5017 | err = -EINVAL; |
5018 | goto abort_unlock; | |
5019 | } | |
5020 | ||
5021 | done_unlock: | |
5022 | abort_unlock: | |
5023 | mddev_unlock(mddev); | |
5024 | ||
5025 | return err; | |
5026 | done: | |
5027 | if (err) | |
5028 | MD_BUG(); | |
5029 | abort: | |
5030 | return err; | |
5031 | } | |
5032 | ||
5033 | static int md_open(struct inode *inode, struct file *file) | |
5034 | { | |
5035 | /* | |
5036 | * Succeed if we can lock the mddev, which confirms that | |
5037 | * it isn't being stopped right now. | |
5038 | */ | |
5039 | mddev_t *mddev = inode->i_bdev->bd_disk->private_data; | |
5040 | int err; | |
5041 | ||
d63a5a74 | 5042 | if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1))) |
1da177e4 LT |
5043 | goto out; |
5044 | ||
5045 | err = 0; | |
5046 | mddev_get(mddev); | |
5047 | mddev_unlock(mddev); | |
5048 | ||
5049 | check_disk_change(inode->i_bdev); | |
5050 | out: | |
5051 | return err; | |
5052 | } | |
5053 | ||
5054 | static int md_release(struct inode *inode, struct file * file) | |
5055 | { | |
5056 | mddev_t *mddev = inode->i_bdev->bd_disk->private_data; | |
5057 | ||
52e5f9d1 | 5058 | BUG_ON(!mddev); |
1da177e4 LT |
5059 | mddev_put(mddev); |
5060 | ||
5061 | return 0; | |
5062 | } | |
5063 | ||
44ce6294 LT |
5064 | static int md_media_changed(struct gendisk *disk) |
5065 | { | |
5066 | mddev_t *mddev = disk->private_data; | |
5067 | ||
5068 | return mddev->changed; | |
5069 | } | |
5070 | ||
5071 | static int md_revalidate(struct gendisk *disk) | |
5072 | { | |
5073 | mddev_t *mddev = disk->private_data; | |
5074 | ||
5075 | mddev->changed = 0; | |
5076 | return 0; | |
5077 | } | |
1da177e4 LT |
5078 | static struct block_device_operations md_fops = |
5079 | { | |
5080 | .owner = THIS_MODULE, | |
5081 | .open = md_open, | |
5082 | .release = md_release, | |
5083 | .ioctl = md_ioctl, | |
a885c8c4 | 5084 | .getgeo = md_getgeo, |
44ce6294 LT |
5085 | .media_changed = md_media_changed, |
5086 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
5087 | }; |
5088 | ||
75c96f85 | 5089 | static int md_thread(void * arg) |
1da177e4 LT |
5090 | { |
5091 | mdk_thread_t *thread = arg; | |
5092 | ||
1da177e4 LT |
5093 | /* |
5094 | * md_thread is a 'system-thread', it's priority should be very | |
5095 | * high. We avoid resource deadlocks individually in each | |
5096 | * raid personality. (RAID5 does preallocation) We also use RR and | |
5097 | * the very same RT priority as kswapd, thus we will never get | |
5098 | * into a priority inversion deadlock. | |
5099 | * | |
5100 | * we definitely have to have equal or higher priority than | |
5101 | * bdflush, otherwise bdflush will deadlock if there are too | |
5102 | * many dirty RAID5 blocks. | |
5103 | */ | |
1da177e4 | 5104 | |
6985c43f | 5105 | allow_signal(SIGKILL); |
a6fb0934 | 5106 | while (!kthread_should_stop()) { |
1da177e4 | 5107 | |
93588e22 N |
5108 | /* We need to wait INTERRUPTIBLE so that |
5109 | * we don't add to the load-average. | |
5110 | * That means we need to be sure no signals are | |
5111 | * pending | |
5112 | */ | |
5113 | if (signal_pending(current)) | |
5114 | flush_signals(current); | |
5115 | ||
5116 | wait_event_interruptible_timeout | |
5117 | (thread->wqueue, | |
5118 | test_bit(THREAD_WAKEUP, &thread->flags) | |
5119 | || kthread_should_stop(), | |
5120 | thread->timeout); | |
1da177e4 LT |
5121 | |
5122 | clear_bit(THREAD_WAKEUP, &thread->flags); | |
5123 | ||
787453c2 | 5124 | thread->run(thread->mddev); |
1da177e4 | 5125 | } |
a6fb0934 | 5126 | |
1da177e4 LT |
5127 | return 0; |
5128 | } | |
5129 | ||
5130 | void md_wakeup_thread(mdk_thread_t *thread) | |
5131 | { | |
5132 | if (thread) { | |
5133 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
5134 | set_bit(THREAD_WAKEUP, &thread->flags); | |
5135 | wake_up(&thread->wqueue); | |
5136 | } | |
5137 | } | |
5138 | ||
5139 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
5140 | const char *name) | |
5141 | { | |
5142 | mdk_thread_t *thread; | |
1da177e4 | 5143 | |
9ffae0cf | 5144 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
5145 | if (!thread) |
5146 | return NULL; | |
5147 | ||
1da177e4 LT |
5148 | init_waitqueue_head(&thread->wqueue); |
5149 | ||
1da177e4 LT |
5150 | thread->run = run; |
5151 | thread->mddev = mddev; | |
32a7627c | 5152 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
6985c43f | 5153 | thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev)); |
a6fb0934 | 5154 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
5155 | kfree(thread); |
5156 | return NULL; | |
5157 | } | |
1da177e4 LT |
5158 | return thread; |
5159 | } | |
5160 | ||
1da177e4 LT |
5161 | void md_unregister_thread(mdk_thread_t *thread) |
5162 | { | |
ba25f9dc | 5163 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
5164 | |
5165 | kthread_stop(thread->tsk); | |
1da177e4 LT |
5166 | kfree(thread); |
5167 | } | |
5168 | ||
5169 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
5170 | { | |
5171 | if (!mddev) { | |
5172 | MD_BUG(); | |
5173 | return; | |
5174 | } | |
5175 | ||
b2d444d7 | 5176 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5177 | return; |
6bfe0b49 DW |
5178 | |
5179 | if (mddev->external) | |
5180 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 5181 | /* |
1da177e4 LT |
5182 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
5183 | mdname(mddev), | |
5184 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
5185 | __builtin_return_address(0),__builtin_return_address(1), | |
5186 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 5187 | */ |
d0a0a5ee AM |
5188 | if (!mddev->pers) |
5189 | return; | |
1da177e4 LT |
5190 | if (!mddev->pers->error_handler) |
5191 | return; | |
5192 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
5193 | if (mddev->degraded) |
5194 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
52664732 | 5195 | set_bit(StateChanged, &rdev->flags); |
1da177e4 LT |
5196 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
5197 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5198 | md_wakeup_thread(mddev->thread); | |
c331eb04 | 5199 | md_new_event_inintr(mddev); |
1da177e4 LT |
5200 | } |
5201 | ||
5202 | /* seq_file implementation /proc/mdstat */ | |
5203 | ||
5204 | static void status_unused(struct seq_file *seq) | |
5205 | { | |
5206 | int i = 0; | |
5207 | mdk_rdev_t *rdev; | |
5208 | struct list_head *tmp; | |
5209 | ||
5210 | seq_printf(seq, "unused devices: "); | |
5211 | ||
73c34431 | 5212 | rdev_for_each_list(rdev, tmp, pending_raid_disks) { |
1da177e4 LT |
5213 | char b[BDEVNAME_SIZE]; |
5214 | i++; | |
5215 | seq_printf(seq, "%s ", | |
5216 | bdevname(rdev->bdev,b)); | |
5217 | } | |
5218 | if (!i) | |
5219 | seq_printf(seq, "<none>"); | |
5220 | ||
5221 | seq_printf(seq, "\n"); | |
5222 | } | |
5223 | ||
5224 | ||
5225 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
5226 | { | |
4588b42e N |
5227 | sector_t max_blocks, resync, res; |
5228 | unsigned long dt, db, rt; | |
5229 | int scale; | |
5230 | unsigned int per_milli; | |
1da177e4 LT |
5231 | |
5232 | resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2; | |
5233 | ||
5234 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
5235 | max_blocks = mddev->resync_max_sectors >> 1; | |
5236 | else | |
5237 | max_blocks = mddev->size; | |
5238 | ||
5239 | /* | |
5240 | * Should not happen. | |
5241 | */ | |
5242 | if (!max_blocks) { | |
5243 | MD_BUG(); | |
5244 | return; | |
5245 | } | |
4588b42e N |
5246 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
5247 | * in a sector_t, and (max_blocks>>scale) will fit in a | |
5248 | * u32, as those are the requirements for sector_div. | |
5249 | * Thus 'scale' must be at least 10 | |
5250 | */ | |
5251 | scale = 10; | |
5252 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
5253 | while ( max_blocks/2 > (1ULL<<(scale+32))) | |
5254 | scale++; | |
5255 | } | |
5256 | res = (resync>>scale)*1000; | |
5257 | sector_div(res, (u32)((max_blocks>>scale)+1)); | |
5258 | ||
5259 | per_milli = res; | |
1da177e4 | 5260 | { |
4588b42e | 5261 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
5262 | seq_printf(seq, "["); |
5263 | for (i = 0; i < x; i++) | |
5264 | seq_printf(seq, "="); | |
5265 | seq_printf(seq, ">"); | |
5266 | for (i = 0; i < y; i++) | |
5267 | seq_printf(seq, "."); | |
5268 | seq_printf(seq, "] "); | |
5269 | } | |
4588b42e | 5270 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
5271 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
5272 | "reshape" : | |
61df9d91 N |
5273 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
5274 | "check" : | |
5275 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
5276 | "resync" : "recovery"))), | |
5277 | per_milli/10, per_milli % 10, | |
4588b42e N |
5278 | (unsigned long long) resync, |
5279 | (unsigned long long) max_blocks); | |
1da177e4 LT |
5280 | |
5281 | /* | |
5282 | * We do not want to overflow, so the order of operands and | |
5283 | * the * 100 / 100 trick are important. We do a +1 to be | |
5284 | * safe against division by zero. We only estimate anyway. | |
5285 | * | |
5286 | * dt: time from mark until now | |
5287 | * db: blocks written from mark until now | |
5288 | * rt: remaining time | |
5289 | */ | |
5290 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
5291 | if (!dt) dt++; | |
ff4e8d9a N |
5292 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
5293 | - mddev->resync_mark_cnt; | |
5294 | rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100; | |
1da177e4 LT |
5295 | |
5296 | seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6); | |
5297 | ||
ff4e8d9a | 5298 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
5299 | } |
5300 | ||
5301 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
5302 | { | |
5303 | struct list_head *tmp; | |
5304 | loff_t l = *pos; | |
5305 | mddev_t *mddev; | |
5306 | ||
5307 | if (l >= 0x10000) | |
5308 | return NULL; | |
5309 | if (!l--) | |
5310 | /* header */ | |
5311 | return (void*)1; | |
5312 | ||
5313 | spin_lock(&all_mddevs_lock); | |
5314 | list_for_each(tmp,&all_mddevs) | |
5315 | if (!l--) { | |
5316 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
5317 | mddev_get(mddev); | |
5318 | spin_unlock(&all_mddevs_lock); | |
5319 | return mddev; | |
5320 | } | |
5321 | spin_unlock(&all_mddevs_lock); | |
5322 | if (!l--) | |
5323 | return (void*)2;/* tail */ | |
5324 | return NULL; | |
5325 | } | |
5326 | ||
5327 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
5328 | { | |
5329 | struct list_head *tmp; | |
5330 | mddev_t *next_mddev, *mddev = v; | |
5331 | ||
5332 | ++*pos; | |
5333 | if (v == (void*)2) | |
5334 | return NULL; | |
5335 | ||
5336 | spin_lock(&all_mddevs_lock); | |
5337 | if (v == (void*)1) | |
5338 | tmp = all_mddevs.next; | |
5339 | else | |
5340 | tmp = mddev->all_mddevs.next; | |
5341 | if (tmp != &all_mddevs) | |
5342 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
5343 | else { | |
5344 | next_mddev = (void*)2; | |
5345 | *pos = 0x10000; | |
5346 | } | |
5347 | spin_unlock(&all_mddevs_lock); | |
5348 | ||
5349 | if (v != (void*)1) | |
5350 | mddev_put(mddev); | |
5351 | return next_mddev; | |
5352 | ||
5353 | } | |
5354 | ||
5355 | static void md_seq_stop(struct seq_file *seq, void *v) | |
5356 | { | |
5357 | mddev_t *mddev = v; | |
5358 | ||
5359 | if (mddev && v != (void*)1 && v != (void*)2) | |
5360 | mddev_put(mddev); | |
5361 | } | |
5362 | ||
d7603b7e N |
5363 | struct mdstat_info { |
5364 | int event; | |
5365 | }; | |
5366 | ||
1da177e4 LT |
5367 | static int md_seq_show(struct seq_file *seq, void *v) |
5368 | { | |
5369 | mddev_t *mddev = v; | |
5370 | sector_t size; | |
5371 | struct list_head *tmp2; | |
5372 | mdk_rdev_t *rdev; | |
d7603b7e | 5373 | struct mdstat_info *mi = seq->private; |
32a7627c | 5374 | struct bitmap *bitmap; |
1da177e4 LT |
5375 | |
5376 | if (v == (void*)1) { | |
2604b703 | 5377 | struct mdk_personality *pers; |
1da177e4 LT |
5378 | seq_printf(seq, "Personalities : "); |
5379 | spin_lock(&pers_lock); | |
2604b703 N |
5380 | list_for_each_entry(pers, &pers_list, list) |
5381 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
5382 | |
5383 | spin_unlock(&pers_lock); | |
5384 | seq_printf(seq, "\n"); | |
d7603b7e | 5385 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
5386 | return 0; |
5387 | } | |
5388 | if (v == (void*)2) { | |
5389 | status_unused(seq); | |
5390 | return 0; | |
5391 | } | |
5392 | ||
5dc5cf7d | 5393 | if (mddev_lock(mddev) < 0) |
1da177e4 | 5394 | return -EINTR; |
5dc5cf7d | 5395 | |
1da177e4 LT |
5396 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
5397 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
5398 | mddev->pers ? "" : "in"); | |
5399 | if (mddev->pers) { | |
f91de92e | 5400 | if (mddev->ro==1) |
1da177e4 | 5401 | seq_printf(seq, " (read-only)"); |
f91de92e | 5402 | if (mddev->ro==2) |
52720ae7 | 5403 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
5404 | seq_printf(seq, " %s", mddev->pers->name); |
5405 | } | |
5406 | ||
5407 | size = 0; | |
d089c6af | 5408 | rdev_for_each(rdev, tmp2, mddev) { |
1da177e4 LT |
5409 | char b[BDEVNAME_SIZE]; |
5410 | seq_printf(seq, " %s[%d]", | |
5411 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
5412 | if (test_bit(WriteMostly, &rdev->flags)) |
5413 | seq_printf(seq, "(W)"); | |
b2d444d7 | 5414 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5415 | seq_printf(seq, "(F)"); |
5416 | continue; | |
b325a32e N |
5417 | } else if (rdev->raid_disk < 0) |
5418 | seq_printf(seq, "(S)"); /* spare */ | |
1da177e4 LT |
5419 | size += rdev->size; |
5420 | } | |
5421 | ||
5422 | if (!list_empty(&mddev->disks)) { | |
5423 | if (mddev->pers) | |
5424 | seq_printf(seq, "\n %llu blocks", | |
5425 | (unsigned long long)mddev->array_size); | |
5426 | else | |
5427 | seq_printf(seq, "\n %llu blocks", | |
5428 | (unsigned long long)size); | |
5429 | } | |
1cd6bf19 N |
5430 | if (mddev->persistent) { |
5431 | if (mddev->major_version != 0 || | |
5432 | mddev->minor_version != 90) { | |
5433 | seq_printf(seq," super %d.%d", | |
5434 | mddev->major_version, | |
5435 | mddev->minor_version); | |
5436 | } | |
e691063a N |
5437 | } else if (mddev->external) |
5438 | seq_printf(seq, " super external:%s", | |
5439 | mddev->metadata_type); | |
5440 | else | |
1cd6bf19 | 5441 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
5442 | |
5443 | if (mddev->pers) { | |
5444 | mddev->pers->status (seq, mddev); | |
5445 | seq_printf(seq, "\n "); | |
8e1b39d6 N |
5446 | if (mddev->pers->sync_request) { |
5447 | if (mddev->curr_resync > 2) { | |
5448 | status_resync (seq, mddev); | |
5449 | seq_printf(seq, "\n "); | |
5450 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
5451 | seq_printf(seq, "\tresync=DELAYED\n "); | |
5452 | else if (mddev->recovery_cp < MaxSector) | |
5453 | seq_printf(seq, "\tresync=PENDING\n "); | |
5454 | } | |
32a7627c N |
5455 | } else |
5456 | seq_printf(seq, "\n "); | |
5457 | ||
5458 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
5459 | unsigned long chunk_kb; |
5460 | unsigned long flags; | |
32a7627c N |
5461 | spin_lock_irqsave(&bitmap->lock, flags); |
5462 | chunk_kb = bitmap->chunksize >> 10; | |
5463 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
5464 | "%lu%s chunk", | |
5465 | bitmap->pages - bitmap->missing_pages, | |
5466 | bitmap->pages, | |
5467 | (bitmap->pages - bitmap->missing_pages) | |
5468 | << (PAGE_SHIFT - 10), | |
5469 | chunk_kb ? chunk_kb : bitmap->chunksize, | |
5470 | chunk_kb ? "KB" : "B"); | |
78d742d8 N |
5471 | if (bitmap->file) { |
5472 | seq_printf(seq, ", file: "); | |
c32c2f63 | 5473 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 5474 | } |
78d742d8 | 5475 | |
32a7627c N |
5476 | seq_printf(seq, "\n"); |
5477 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
5478 | } |
5479 | ||
5480 | seq_printf(seq, "\n"); | |
5481 | } | |
5482 | mddev_unlock(mddev); | |
5483 | ||
5484 | return 0; | |
5485 | } | |
5486 | ||
5487 | static struct seq_operations md_seq_ops = { | |
5488 | .start = md_seq_start, | |
5489 | .next = md_seq_next, | |
5490 | .stop = md_seq_stop, | |
5491 | .show = md_seq_show, | |
5492 | }; | |
5493 | ||
5494 | static int md_seq_open(struct inode *inode, struct file *file) | |
5495 | { | |
5496 | int error; | |
d7603b7e N |
5497 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
5498 | if (mi == NULL) | |
5499 | return -ENOMEM; | |
1da177e4 LT |
5500 | |
5501 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
5502 | if (error) |
5503 | kfree(mi); | |
5504 | else { | |
5505 | struct seq_file *p = file->private_data; | |
5506 | p->private = mi; | |
5507 | mi->event = atomic_read(&md_event_count); | |
5508 | } | |
1da177e4 LT |
5509 | return error; |
5510 | } | |
5511 | ||
d7603b7e N |
5512 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
5513 | { | |
5514 | struct seq_file *m = filp->private_data; | |
5515 | struct mdstat_info *mi = m->private; | |
5516 | int mask; | |
5517 | ||
5518 | poll_wait(filp, &md_event_waiters, wait); | |
5519 | ||
5520 | /* always allow read */ | |
5521 | mask = POLLIN | POLLRDNORM; | |
5522 | ||
5523 | if (mi->event != atomic_read(&md_event_count)) | |
5524 | mask |= POLLERR | POLLPRI; | |
5525 | return mask; | |
5526 | } | |
5527 | ||
fa027c2a | 5528 | static const struct file_operations md_seq_fops = { |
e24650c2 | 5529 | .owner = THIS_MODULE, |
1da177e4 LT |
5530 | .open = md_seq_open, |
5531 | .read = seq_read, | |
5532 | .llseek = seq_lseek, | |
c3f94b40 | 5533 | .release = seq_release_private, |
d7603b7e | 5534 | .poll = mdstat_poll, |
1da177e4 LT |
5535 | }; |
5536 | ||
2604b703 | 5537 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 5538 | { |
1da177e4 | 5539 | spin_lock(&pers_lock); |
2604b703 N |
5540 | list_add_tail(&p->list, &pers_list); |
5541 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
5542 | spin_unlock(&pers_lock); |
5543 | return 0; | |
5544 | } | |
5545 | ||
2604b703 | 5546 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 5547 | { |
2604b703 | 5548 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 5549 | spin_lock(&pers_lock); |
2604b703 | 5550 | list_del_init(&p->list); |
1da177e4 LT |
5551 | spin_unlock(&pers_lock); |
5552 | return 0; | |
5553 | } | |
5554 | ||
5555 | static int is_mddev_idle(mddev_t *mddev) | |
5556 | { | |
5557 | mdk_rdev_t * rdev; | |
5558 | struct list_head *tmp; | |
5559 | int idle; | |
713f6ab1 | 5560 | long curr_events; |
1da177e4 LT |
5561 | |
5562 | idle = 1; | |
d089c6af | 5563 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 | 5564 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
a362357b JA |
5565 | curr_events = disk_stat_read(disk, sectors[0]) + |
5566 | disk_stat_read(disk, sectors[1]) - | |
1da177e4 | 5567 | atomic_read(&disk->sync_io); |
713f6ab1 N |
5568 | /* sync IO will cause sync_io to increase before the disk_stats |
5569 | * as sync_io is counted when a request starts, and | |
5570 | * disk_stats is counted when it completes. | |
5571 | * So resync activity will cause curr_events to be smaller than | |
5572 | * when there was no such activity. | |
5573 | * non-sync IO will cause disk_stat to increase without | |
5574 | * increasing sync_io so curr_events will (eventually) | |
5575 | * be larger than it was before. Once it becomes | |
5576 | * substantially larger, the test below will cause | |
5577 | * the array to appear non-idle, and resync will slow | |
5578 | * down. | |
5579 | * If there is a lot of outstanding resync activity when | |
5580 | * we set last_event to curr_events, then all that activity | |
5581 | * completing might cause the array to appear non-idle | |
5582 | * and resync will be slowed down even though there might | |
5583 | * not have been non-resync activity. This will only | |
5584 | * happen once though. 'last_events' will soon reflect | |
5585 | * the state where there is little or no outstanding | |
5586 | * resync requests, and further resync activity will | |
5587 | * always make curr_events less than last_events. | |
c0e48521 | 5588 | * |
1da177e4 | 5589 | */ |
713f6ab1 | 5590 | if (curr_events - rdev->last_events > 4096) { |
1da177e4 LT |
5591 | rdev->last_events = curr_events; |
5592 | idle = 0; | |
5593 | } | |
5594 | } | |
5595 | return idle; | |
5596 | } | |
5597 | ||
5598 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
5599 | { | |
5600 | /* another "blocks" (512byte) blocks have been synced */ | |
5601 | atomic_sub(blocks, &mddev->recovery_active); | |
5602 | wake_up(&mddev->recovery_wait); | |
5603 | if (!ok) { | |
dfc70645 | 5604 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
5605 | md_wakeup_thread(mddev->thread); |
5606 | // stop recovery, signal do_sync .... | |
5607 | } | |
5608 | } | |
5609 | ||
5610 | ||
06d91a5f N |
5611 | /* md_write_start(mddev, bi) |
5612 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
5613 | * in superblock) before writing, schedule a superblock update |
5614 | * and wait for it to complete. | |
06d91a5f | 5615 | */ |
3d310eb7 | 5616 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 5617 | { |
0fd62b86 | 5618 | int did_change = 0; |
06d91a5f | 5619 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 5620 | return; |
06d91a5f | 5621 | |
f91de92e N |
5622 | BUG_ON(mddev->ro == 1); |
5623 | if (mddev->ro == 2) { | |
5624 | /* need to switch to read/write */ | |
5625 | mddev->ro = 0; | |
5626 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5627 | md_wakeup_thread(mddev->thread); | |
25156198 | 5628 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 5629 | did_change = 1; |
f91de92e | 5630 | } |
06d91a5f | 5631 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
5632 | if (mddev->safemode == 1) |
5633 | mddev->safemode = 0; | |
06d91a5f | 5634 | if (mddev->in_sync) { |
a9701a30 | 5635 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
5636 | if (mddev->in_sync) { |
5637 | mddev->in_sync = 0; | |
850b2b42 | 5638 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3d310eb7 | 5639 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 5640 | did_change = 1; |
3d310eb7 | 5641 | } |
a9701a30 | 5642 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 5643 | } |
0fd62b86 NB |
5644 | if (did_change) |
5645 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
09a44cc1 N |
5646 | wait_event(mddev->sb_wait, |
5647 | !test_bit(MD_CHANGE_CLEAN, &mddev->flags) && | |
5648 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1da177e4 LT |
5649 | } |
5650 | ||
5651 | void md_write_end(mddev_t *mddev) | |
5652 | { | |
5653 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
5654 | if (mddev->safemode == 2) | |
5655 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 5656 | else if (mddev->safemode_delay) |
1da177e4 LT |
5657 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
5658 | } | |
5659 | } | |
5660 | ||
2a2275d6 N |
5661 | /* md_allow_write(mddev) |
5662 | * Calling this ensures that the array is marked 'active' so that writes | |
5663 | * may proceed without blocking. It is important to call this before | |
5664 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
5665 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
5666 | * |
5667 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
5668 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 5669 | */ |
b5470dc5 | 5670 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
5671 | { |
5672 | if (!mddev->pers) | |
b5470dc5 | 5673 | return 0; |
2a2275d6 | 5674 | if (mddev->ro) |
b5470dc5 | 5675 | return 0; |
1a0fd497 | 5676 | if (!mddev->pers->sync_request) |
b5470dc5 | 5677 | return 0; |
2a2275d6 N |
5678 | |
5679 | spin_lock_irq(&mddev->write_lock); | |
5680 | if (mddev->in_sync) { | |
5681 | mddev->in_sync = 0; | |
5682 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
5683 | if (mddev->safemode_delay && | |
5684 | mddev->safemode == 0) | |
5685 | mddev->safemode = 1; | |
5686 | spin_unlock_irq(&mddev->write_lock); | |
5687 | md_update_sb(mddev, 0); | |
09a44cc1 | 5688 | sysfs_notify(&mddev->kobj, NULL, "array_state"); |
2a2275d6 N |
5689 | } else |
5690 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 DW |
5691 | |
5692 | if (test_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
5693 | return -EAGAIN; | |
5694 | else | |
5695 | return 0; | |
2a2275d6 N |
5696 | } |
5697 | EXPORT_SYMBOL_GPL(md_allow_write); | |
5698 | ||
1da177e4 LT |
5699 | #define SYNC_MARKS 10 |
5700 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 5701 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
5702 | { |
5703 | mddev_t *mddev2; | |
5704 | unsigned int currspeed = 0, | |
5705 | window; | |
57afd89f | 5706 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
5707 | unsigned long mark[SYNC_MARKS]; |
5708 | sector_t mark_cnt[SYNC_MARKS]; | |
5709 | int last_mark,m; | |
5710 | struct list_head *tmp; | |
5711 | sector_t last_check; | |
57afd89f | 5712 | int skipped = 0; |
5fd6c1dc N |
5713 | struct list_head *rtmp; |
5714 | mdk_rdev_t *rdev; | |
61df9d91 | 5715 | char *desc; |
1da177e4 LT |
5716 | |
5717 | /* just incase thread restarts... */ | |
5718 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
5719 | return; | |
5fd6c1dc N |
5720 | if (mddev->ro) /* never try to sync a read-only array */ |
5721 | return; | |
1da177e4 | 5722 | |
61df9d91 N |
5723 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
5724 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
5725 | desc = "data-check"; | |
5726 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
5727 | desc = "requested-resync"; | |
5728 | else | |
5729 | desc = "resync"; | |
5730 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
5731 | desc = "reshape"; | |
5732 | else | |
5733 | desc = "recovery"; | |
5734 | ||
1da177e4 LT |
5735 | /* we overload curr_resync somewhat here. |
5736 | * 0 == not engaged in resync at all | |
5737 | * 2 == checking that there is no conflict with another sync | |
5738 | * 1 == like 2, but have yielded to allow conflicting resync to | |
5739 | * commense | |
5740 | * other == active in resync - this many blocks | |
5741 | * | |
5742 | * Before starting a resync we must have set curr_resync to | |
5743 | * 2, and then checked that every "conflicting" array has curr_resync | |
5744 | * less than ours. When we find one that is the same or higher | |
5745 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
5746 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
5747 | * This will mean we have to start checking from the beginning again. | |
5748 | * | |
5749 | */ | |
5750 | ||
5751 | do { | |
5752 | mddev->curr_resync = 2; | |
5753 | ||
5754 | try_again: | |
787453c2 | 5755 | if (kthread_should_stop()) { |
6985c43f | 5756 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
5757 | goto skip; |
5758 | } | |
29ac4aa3 | 5759 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
5760 | if (mddev2 == mddev) |
5761 | continue; | |
90b08710 BS |
5762 | if (!mddev->parallel_resync |
5763 | && mddev2->curr_resync | |
5764 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
5765 | DEFINE_WAIT(wq); |
5766 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
5767 | /* arbitrarily yield */ | |
5768 | mddev->curr_resync = 1; | |
5769 | wake_up(&resync_wait); | |
5770 | } | |
5771 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
5772 | /* no need to wait here, we can wait the next | |
5773 | * time 'round when curr_resync == 2 | |
5774 | */ | |
5775 | continue; | |
787453c2 N |
5776 | prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE); |
5777 | if (!kthread_should_stop() && | |
8712e553 | 5778 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
5779 | printk(KERN_INFO "md: delaying %s of %s" |
5780 | " until %s has finished (they" | |
1da177e4 | 5781 | " share one or more physical units)\n", |
61df9d91 | 5782 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 LT |
5783 | mddev_put(mddev2); |
5784 | schedule(); | |
5785 | finish_wait(&resync_wait, &wq); | |
5786 | goto try_again; | |
5787 | } | |
5788 | finish_wait(&resync_wait, &wq); | |
5789 | } | |
5790 | } | |
5791 | } while (mddev->curr_resync < 2); | |
5792 | ||
5fd6c1dc | 5793 | j = 0; |
9d88883e | 5794 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 5795 | /* resync follows the size requested by the personality, |
57afd89f | 5796 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
5797 | */ |
5798 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 5799 | mddev->resync_mismatches = 0; |
5fd6c1dc | 5800 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
5801 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
5802 | j = mddev->resync_min; | |
5803 | else if (!mddev->bitmap) | |
5fd6c1dc | 5804 | j = mddev->recovery_cp; |
5e96ee65 | 5805 | |
ccfcc3c1 N |
5806 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5807 | max_sectors = mddev->size << 1; | |
5fd6c1dc | 5808 | else { |
1da177e4 LT |
5809 | /* recovery follows the physical size of devices */ |
5810 | max_sectors = mddev->size << 1; | |
5fd6c1dc | 5811 | j = MaxSector; |
d089c6af | 5812 | rdev_for_each(rdev, rtmp, mddev) |
5fd6c1dc N |
5813 | if (rdev->raid_disk >= 0 && |
5814 | !test_bit(Faulty, &rdev->flags) && | |
5815 | !test_bit(In_sync, &rdev->flags) && | |
5816 | rdev->recovery_offset < j) | |
5817 | j = rdev->recovery_offset; | |
5818 | } | |
1da177e4 | 5819 | |
61df9d91 N |
5820 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
5821 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
5822 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 5823 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
5824 | "(but not more than %d KB/sec) for %s.\n", |
5825 | speed_max(mddev), desc); | |
1da177e4 LT |
5826 | |
5827 | is_mddev_idle(mddev); /* this also initializes IO event counters */ | |
5fd6c1dc | 5828 | |
57afd89f | 5829 | io_sectors = 0; |
1da177e4 LT |
5830 | for (m = 0; m < SYNC_MARKS; m++) { |
5831 | mark[m] = jiffies; | |
57afd89f | 5832 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
5833 | } |
5834 | last_mark = 0; | |
5835 | mddev->resync_mark = mark[last_mark]; | |
5836 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
5837 | ||
5838 | /* | |
5839 | * Tune reconstruction: | |
5840 | */ | |
5841 | window = 32*(PAGE_SIZE/512); | |
5842 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
5843 | window/2,(unsigned long long) max_sectors/2); | |
5844 | ||
5845 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
5846 | last_check = 0; |
5847 | ||
5848 | if (j>2) { | |
5849 | printk(KERN_INFO | |
61df9d91 N |
5850 | "md: resuming %s of %s from checkpoint.\n", |
5851 | desc, mdname(mddev)); | |
1da177e4 LT |
5852 | mddev->curr_resync = j; |
5853 | } | |
5854 | ||
5855 | while (j < max_sectors) { | |
57afd89f | 5856 | sector_t sectors; |
1da177e4 | 5857 | |
57afd89f | 5858 | skipped = 0; |
c6207277 N |
5859 | if (j >= mddev->resync_max) { |
5860 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
5861 | wait_event(mddev->recovery_wait, | |
5862 | mddev->resync_max > j | |
5863 | || kthread_should_stop()); | |
5864 | } | |
5865 | if (kthread_should_stop()) | |
5866 | goto interrupted; | |
57afd89f | 5867 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 5868 | currspeed < speed_min(mddev)); |
57afd89f | 5869 | if (sectors == 0) { |
dfc70645 | 5870 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
5871 | goto out; |
5872 | } | |
57afd89f N |
5873 | |
5874 | if (!skipped) { /* actual IO requested */ | |
5875 | io_sectors += sectors; | |
5876 | atomic_add(sectors, &mddev->recovery_active); | |
5877 | } | |
5878 | ||
1da177e4 LT |
5879 | j += sectors; |
5880 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 5881 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
5882 | if (last_check == 0) |
5883 | /* this is the earliers that rebuilt will be | |
5884 | * visible in /proc/mdstat | |
5885 | */ | |
5886 | md_new_event(mddev); | |
57afd89f N |
5887 | |
5888 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
5889 | continue; |
5890 | ||
57afd89f | 5891 | last_check = io_sectors; |
1da177e4 | 5892 | |
dfc70645 | 5893 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
5894 | break; |
5895 | ||
5896 | repeat: | |
5897 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
5898 | /* step marks */ | |
5899 | int next = (last_mark+1) % SYNC_MARKS; | |
5900 | ||
5901 | mddev->resync_mark = mark[next]; | |
5902 | mddev->resync_mark_cnt = mark_cnt[next]; | |
5903 | mark[next] = jiffies; | |
57afd89f | 5904 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
5905 | last_mark = next; |
5906 | } | |
5907 | ||
5908 | ||
c6207277 N |
5909 | if (kthread_should_stop()) |
5910 | goto interrupted; | |
5911 | ||
1da177e4 LT |
5912 | |
5913 | /* | |
5914 | * this loop exits only if either when we are slower than | |
5915 | * the 'hard' speed limit, or the system was IO-idle for | |
5916 | * a jiffy. | |
5917 | * the system might be non-idle CPU-wise, but we only care | |
5918 | * about not overloading the IO subsystem. (things like an | |
5919 | * e2fsck being done on the RAID array should execute fast) | |
5920 | */ | |
2ad8b1ef | 5921 | blk_unplug(mddev->queue); |
1da177e4 LT |
5922 | cond_resched(); |
5923 | ||
57afd89f N |
5924 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
5925 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 5926 | |
88202a0c N |
5927 | if (currspeed > speed_min(mddev)) { |
5928 | if ((currspeed > speed_max(mddev)) || | |
1da177e4 | 5929 | !is_mddev_idle(mddev)) { |
c0e48521 | 5930 | msleep(500); |
1da177e4 LT |
5931 | goto repeat; |
5932 | } | |
5933 | } | |
5934 | } | |
61df9d91 | 5935 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
5936 | /* |
5937 | * this also signals 'finished resyncing' to md_stop | |
5938 | */ | |
5939 | out: | |
2ad8b1ef | 5940 | blk_unplug(mddev->queue); |
1da177e4 LT |
5941 | |
5942 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
5943 | ||
5944 | /* tell personality that we are finished */ | |
57afd89f | 5945 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 5946 | |
dfc70645 | 5947 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
5948 | mddev->curr_resync > 2) { |
5949 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
5950 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5951 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
5952 | printk(KERN_INFO | |
61df9d91 N |
5953 | "md: checkpointing %s of %s.\n", |
5954 | desc, mdname(mddev)); | |
5fd6c1dc N |
5955 | mddev->recovery_cp = mddev->curr_resync; |
5956 | } | |
5957 | } else | |
5958 | mddev->recovery_cp = MaxSector; | |
5959 | } else { | |
5960 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5961 | mddev->curr_resync = MaxSector; | |
d089c6af | 5962 | rdev_for_each(rdev, rtmp, mddev) |
5fd6c1dc N |
5963 | if (rdev->raid_disk >= 0 && |
5964 | !test_bit(Faulty, &rdev->flags) && | |
5965 | !test_bit(In_sync, &rdev->flags) && | |
5966 | rdev->recovery_offset < mddev->curr_resync) | |
5967 | rdev->recovery_offset = mddev->curr_resync; | |
5fd6c1dc | 5968 | } |
1da177e4 | 5969 | } |
17571284 | 5970 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5971 | |
1da177e4 LT |
5972 | skip: |
5973 | mddev->curr_resync = 0; | |
5e96ee65 | 5974 | mddev->resync_min = 0; |
c6207277 N |
5975 | mddev->resync_max = MaxSector; |
5976 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
1da177e4 LT |
5977 | wake_up(&resync_wait); |
5978 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
5979 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
5980 | return; |
5981 | ||
5982 | interrupted: | |
5983 | /* | |
5984 | * got a signal, exit. | |
5985 | */ | |
5986 | printk(KERN_INFO | |
5987 | "md: md_do_sync() got signal ... exiting\n"); | |
5988 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5989 | goto out; | |
5990 | ||
1da177e4 | 5991 | } |
29269553 | 5992 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
5993 | |
5994 | ||
b4c4c7b8 N |
5995 | static int remove_and_add_spares(mddev_t *mddev) |
5996 | { | |
5997 | mdk_rdev_t *rdev; | |
5998 | struct list_head *rtmp; | |
5999 | int spares = 0; | |
6000 | ||
d089c6af | 6001 | rdev_for_each(rdev, rtmp, mddev) |
b4c4c7b8 | 6002 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 6003 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
6004 | (test_bit(Faulty, &rdev->flags) || |
6005 | ! test_bit(In_sync, &rdev->flags)) && | |
6006 | atomic_read(&rdev->nr_pending)==0) { | |
6007 | if (mddev->pers->hot_remove_disk( | |
6008 | mddev, rdev->raid_disk)==0) { | |
6009 | char nm[20]; | |
6010 | sprintf(nm,"rd%d", rdev->raid_disk); | |
6011 | sysfs_remove_link(&mddev->kobj, nm); | |
6012 | rdev->raid_disk = -1; | |
6013 | } | |
6014 | } | |
6015 | ||
6016 | if (mddev->degraded) { | |
dfc70645 N |
6017 | rdev_for_each(rdev, rtmp, mddev) { |
6018 | if (rdev->raid_disk >= 0 && | |
6019 | !test_bit(In_sync, &rdev->flags)) | |
6020 | spares++; | |
b4c4c7b8 N |
6021 | if (rdev->raid_disk < 0 |
6022 | && !test_bit(Faulty, &rdev->flags)) { | |
6023 | rdev->recovery_offset = 0; | |
199050ea NB |
6024 | if (mddev->pers-> |
6025 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
6026 | char nm[20]; |
6027 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
6028 | if (sysfs_create_link(&mddev->kobj, |
6029 | &rdev->kobj, nm)) | |
6030 | printk(KERN_WARNING | |
6031 | "md: cannot register " | |
6032 | "%s for %s\n", | |
6033 | nm, mdname(mddev)); | |
b4c4c7b8 N |
6034 | spares++; |
6035 | md_new_event(mddev); | |
6036 | } else | |
6037 | break; | |
6038 | } | |
dfc70645 | 6039 | } |
b4c4c7b8 N |
6040 | } |
6041 | return spares; | |
6042 | } | |
1da177e4 LT |
6043 | /* |
6044 | * This routine is regularly called by all per-raid-array threads to | |
6045 | * deal with generic issues like resync and super-block update. | |
6046 | * Raid personalities that don't have a thread (linear/raid0) do not | |
6047 | * need this as they never do any recovery or update the superblock. | |
6048 | * | |
6049 | * It does not do any resync itself, but rather "forks" off other threads | |
6050 | * to do that as needed. | |
6051 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
6052 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 6053 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
6054 | * and wakeups up this thread which will reap the thread and finish up. |
6055 | * This thread also removes any faulty devices (with nr_pending == 0). | |
6056 | * | |
6057 | * The overall approach is: | |
6058 | * 1/ if the superblock needs updating, update it. | |
6059 | * 2/ If a recovery thread is running, don't do anything else. | |
6060 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
6061 | * 4/ If there are any faulty devices, remove them. | |
6062 | * 5/ If array is degraded, try to add spares devices | |
6063 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
6064 | */ | |
6065 | void md_check_recovery(mddev_t *mddev) | |
6066 | { | |
6067 | mdk_rdev_t *rdev; | |
6068 | struct list_head *rtmp; | |
6069 | ||
6070 | ||
5f40402d N |
6071 | if (mddev->bitmap) |
6072 | bitmap_daemon_work(mddev->bitmap); | |
1da177e4 LT |
6073 | |
6074 | if (mddev->ro) | |
6075 | return; | |
fca4d848 N |
6076 | |
6077 | if (signal_pending(current)) { | |
31a59e34 | 6078 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
6079 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
6080 | mdname(mddev)); | |
6081 | mddev->safemode = 2; | |
6082 | } | |
6083 | flush_signals(current); | |
6084 | } | |
6085 | ||
1da177e4 | 6086 | if ( ! ( |
e691063a | 6087 | (mddev->flags && !mddev->external) || |
1da177e4 | 6088 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 6089 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 6090 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
6091 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
6092 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
6093 | )) |
6094 | return; | |
fca4d848 | 6095 | |
df5b89b3 | 6096 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 6097 | int spares = 0; |
fca4d848 | 6098 | |
31a59e34 | 6099 | if (!mddev->external) { |
0fd62b86 | 6100 | int did_change = 0; |
31a59e34 N |
6101 | spin_lock_irq(&mddev->write_lock); |
6102 | if (mddev->safemode && | |
6103 | !atomic_read(&mddev->writes_pending) && | |
6104 | !mddev->in_sync && | |
6105 | mddev->recovery_cp == MaxSector) { | |
6106 | mddev->in_sync = 1; | |
0fd62b86 | 6107 | did_change = 1; |
31a59e34 N |
6108 | if (mddev->persistent) |
6109 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
6110 | } | |
6111 | if (mddev->safemode == 1) | |
6112 | mddev->safemode = 0; | |
6113 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 NB |
6114 | if (did_change) |
6115 | sysfs_notify(&mddev->kobj, NULL, "array_state"); | |
fca4d848 | 6116 | } |
fca4d848 | 6117 | |
850b2b42 N |
6118 | if (mddev->flags) |
6119 | md_update_sb(mddev, 0); | |
06d91a5f | 6120 | |
52664732 NB |
6121 | rdev_for_each(rdev, rtmp, mddev) |
6122 | if (test_and_clear_bit(StateChanged, &rdev->flags)) | |
6123 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
6124 | ||
06d91a5f | 6125 | |
1da177e4 LT |
6126 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
6127 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
6128 | /* resync/recovery still happening */ | |
6129 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6130 | goto unlock; | |
6131 | } | |
6132 | if (mddev->sync_thread) { | |
6133 | /* resync has finished, collect result */ | |
6134 | md_unregister_thread(mddev->sync_thread); | |
6135 | mddev->sync_thread = NULL; | |
dfc70645 | 6136 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
1da177e4 LT |
6137 | /* success...*/ |
6138 | /* activate any spares */ | |
a99ac971 NB |
6139 | if (mddev->pers->spare_active(mddev)) |
6140 | sysfs_notify(&mddev->kobj, NULL, | |
6141 | "degraded"); | |
1da177e4 | 6142 | } |
850b2b42 | 6143 | md_update_sb(mddev, 1); |
41158c7e N |
6144 | |
6145 | /* if array is no-longer degraded, then any saved_raid_disk | |
6146 | * information must be scrapped | |
6147 | */ | |
6148 | if (!mddev->degraded) | |
d089c6af | 6149 | rdev_for_each(rdev, rtmp, mddev) |
41158c7e N |
6150 | rdev->saved_raid_disk = -1; |
6151 | ||
1da177e4 LT |
6152 | mddev->recovery = 0; |
6153 | /* flag recovery needed just to double check */ | |
6154 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
72a23c21 | 6155 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
d7603b7e | 6156 | md_new_event(mddev); |
1da177e4 LT |
6157 | goto unlock; |
6158 | } | |
72a23c21 NB |
6159 | /* Set RUNNING before clearing NEEDED to avoid |
6160 | * any transients in the value of "sync_action". | |
6161 | */ | |
6162 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6163 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
6164 | /* Clear some bits that don't mean anything, but |
6165 | * might be left set | |
6166 | */ | |
24dd469d N |
6167 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6168 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 6169 | |
5fd6c1dc N |
6170 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6171 | goto unlock; | |
1da177e4 LT |
6172 | /* no recovery is running. |
6173 | * remove any failed drives, then | |
6174 | * add spares if possible. | |
6175 | * Spare are also removed and re-added, to allow | |
6176 | * the personality to fail the re-add. | |
6177 | */ | |
1da177e4 | 6178 | |
b4c4c7b8 N |
6179 | if (mddev->reshape_position != MaxSector) { |
6180 | if (mddev->pers->check_reshape(mddev) != 0) | |
6181 | /* Cannot proceed */ | |
6182 | goto unlock; | |
6183 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 6184 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 6185 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
6186 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
6187 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
72a23c21 | 6188 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6189 | } else if (mddev->recovery_cp < MaxSector) { |
6190 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 6191 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
6192 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
6193 | /* nothing to be done ... */ | |
1da177e4 | 6194 | goto unlock; |
24dd469d | 6195 | |
1da177e4 | 6196 | if (mddev->pers->sync_request) { |
a654b9d8 N |
6197 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
6198 | /* We are adding a device or devices to an array | |
6199 | * which has the bitmap stored on all devices. | |
6200 | * So make sure all bitmap pages get written | |
6201 | */ | |
6202 | bitmap_write_all(mddev->bitmap); | |
6203 | } | |
1da177e4 LT |
6204 | mddev->sync_thread = md_register_thread(md_do_sync, |
6205 | mddev, | |
6206 | "%s_resync"); | |
6207 | if (!mddev->sync_thread) { | |
6208 | printk(KERN_ERR "%s: could not start resync" | |
6209 | " thread...\n", | |
6210 | mdname(mddev)); | |
6211 | /* leave the spares where they are, it shouldn't hurt */ | |
6212 | mddev->recovery = 0; | |
d7603b7e | 6213 | } else |
1da177e4 | 6214 | md_wakeup_thread(mddev->sync_thread); |
72a23c21 | 6215 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); |
d7603b7e | 6216 | md_new_event(mddev); |
1da177e4 LT |
6217 | } |
6218 | unlock: | |
72a23c21 NB |
6219 | if (!mddev->sync_thread) { |
6220 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6221 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
6222 | &mddev->recovery)) | |
6223 | sysfs_notify(&mddev->kobj, NULL, "sync_action"); | |
6224 | } | |
1da177e4 LT |
6225 | mddev_unlock(mddev); |
6226 | } | |
6227 | } | |
6228 | ||
6bfe0b49 DW |
6229 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
6230 | { | |
6231 | sysfs_notify(&rdev->kobj, NULL, "state"); | |
6232 | wait_event_timeout(rdev->blocked_wait, | |
6233 | !test_bit(Blocked, &rdev->flags), | |
6234 | msecs_to_jiffies(5000)); | |
6235 | rdev_dec_pending(rdev, mddev); | |
6236 | } | |
6237 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
6238 | ||
75c96f85 AB |
6239 | static int md_notify_reboot(struct notifier_block *this, |
6240 | unsigned long code, void *x) | |
1da177e4 LT |
6241 | { |
6242 | struct list_head *tmp; | |
6243 | mddev_t *mddev; | |
6244 | ||
6245 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
6246 | ||
6247 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
6248 | ||
29ac4aa3 | 6249 | for_each_mddev(mddev, tmp) |
c71d4887 | 6250 | if (mddev_trylock(mddev)) { |
1da177e4 | 6251 | do_md_stop (mddev, 1); |
c71d4887 NB |
6252 | mddev_unlock(mddev); |
6253 | } | |
1da177e4 LT |
6254 | /* |
6255 | * certain more exotic SCSI devices are known to be | |
6256 | * volatile wrt too early system reboots. While the | |
6257 | * right place to handle this issue is the given | |
6258 | * driver, we do want to have a safe RAID driver ... | |
6259 | */ | |
6260 | mdelay(1000*1); | |
6261 | } | |
6262 | return NOTIFY_DONE; | |
6263 | } | |
6264 | ||
75c96f85 | 6265 | static struct notifier_block md_notifier = { |
1da177e4 LT |
6266 | .notifier_call = md_notify_reboot, |
6267 | .next = NULL, | |
6268 | .priority = INT_MAX, /* before any real devices */ | |
6269 | }; | |
6270 | ||
6271 | static void md_geninit(void) | |
6272 | { | |
1da177e4 LT |
6273 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
6274 | ||
c7705f34 | 6275 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
6276 | } |
6277 | ||
75c96f85 | 6278 | static int __init md_init(void) |
1da177e4 | 6279 | { |
1da177e4 LT |
6280 | if (register_blkdev(MAJOR_NR, "md")) |
6281 | return -1; | |
6282 | if ((mdp_major=register_blkdev(0, "mdp"))<=0) { | |
6283 | unregister_blkdev(MAJOR_NR, "md"); | |
6284 | return -1; | |
6285 | } | |
e8703fe1 N |
6286 | blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE, |
6287 | md_probe, NULL, NULL); | |
6288 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
6289 | md_probe, NULL, NULL); |
6290 | ||
1da177e4 | 6291 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 6292 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
6293 | |
6294 | md_geninit(); | |
6295 | return (0); | |
6296 | } | |
6297 | ||
6298 | ||
6299 | #ifndef MODULE | |
6300 | ||
6301 | /* | |
6302 | * Searches all registered partitions for autorun RAID arrays | |
6303 | * at boot time. | |
6304 | */ | |
4d936ec1 ME |
6305 | |
6306 | static LIST_HEAD(all_detected_devices); | |
6307 | struct detected_devices_node { | |
6308 | struct list_head list; | |
6309 | dev_t dev; | |
6310 | }; | |
1da177e4 LT |
6311 | |
6312 | void md_autodetect_dev(dev_t dev) | |
6313 | { | |
4d936ec1 ME |
6314 | struct detected_devices_node *node_detected_dev; |
6315 | ||
6316 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
6317 | if (node_detected_dev) { | |
6318 | node_detected_dev->dev = dev; | |
6319 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
6320 | } else { | |
6321 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
6322 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
6323 | } | |
1da177e4 LT |
6324 | } |
6325 | ||
6326 | ||
6327 | static void autostart_arrays(int part) | |
6328 | { | |
6329 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
6330 | struct detected_devices_node *node_detected_dev; |
6331 | dev_t dev; | |
6332 | int i_scanned, i_passed; | |
1da177e4 | 6333 | |
4d936ec1 ME |
6334 | i_scanned = 0; |
6335 | i_passed = 0; | |
1da177e4 | 6336 | |
4d936ec1 | 6337 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 6338 | |
4d936ec1 ME |
6339 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
6340 | i_scanned++; | |
6341 | node_detected_dev = list_entry(all_detected_devices.next, | |
6342 | struct detected_devices_node, list); | |
6343 | list_del(&node_detected_dev->list); | |
6344 | dev = node_detected_dev->dev; | |
6345 | kfree(node_detected_dev); | |
df968c4e | 6346 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
6347 | if (IS_ERR(rdev)) |
6348 | continue; | |
6349 | ||
b2d444d7 | 6350 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6351 | MD_BUG(); |
6352 | continue; | |
6353 | } | |
d0fae18f | 6354 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 6355 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 6356 | i_passed++; |
1da177e4 | 6357 | } |
4d936ec1 ME |
6358 | |
6359 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
6360 | i_scanned, i_passed); | |
1da177e4 LT |
6361 | |
6362 | autorun_devices(part); | |
6363 | } | |
6364 | ||
fdee8ae4 | 6365 | #endif /* !MODULE */ |
1da177e4 LT |
6366 | |
6367 | static __exit void md_exit(void) | |
6368 | { | |
6369 | mddev_t *mddev; | |
6370 | struct list_head *tmp; | |
8ab5e4c1 | 6371 | |
e8703fe1 N |
6372 | blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS); |
6373 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); | |
1da177e4 LT |
6374 | |
6375 | unregister_blkdev(MAJOR_NR,"md"); | |
6376 | unregister_blkdev(mdp_major, "mdp"); | |
6377 | unregister_reboot_notifier(&md_notifier); | |
6378 | unregister_sysctl_table(raid_table_header); | |
6379 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 6380 | for_each_mddev(mddev, tmp) { |
1da177e4 LT |
6381 | struct gendisk *disk = mddev->gendisk; |
6382 | if (!disk) | |
6383 | continue; | |
6384 | export_array(mddev); | |
6385 | del_gendisk(disk); | |
6386 | put_disk(disk); | |
6387 | mddev->gendisk = NULL; | |
6388 | mddev_put(mddev); | |
6389 | } | |
6390 | } | |
6391 | ||
685784aa | 6392 | subsys_initcall(md_init); |
1da177e4 LT |
6393 | module_exit(md_exit) |
6394 | ||
f91de92e N |
6395 | static int get_ro(char *buffer, struct kernel_param *kp) |
6396 | { | |
6397 | return sprintf(buffer, "%d", start_readonly); | |
6398 | } | |
6399 | static int set_ro(const char *val, struct kernel_param *kp) | |
6400 | { | |
6401 | char *e; | |
6402 | int num = simple_strtoul(val, &e, 10); | |
6403 | if (*val && (*e == '\0' || *e == '\n')) { | |
6404 | start_readonly = num; | |
4dbcdc75 | 6405 | return 0; |
f91de92e N |
6406 | } |
6407 | return -EINVAL; | |
6408 | } | |
6409 | ||
80ca3a44 N |
6410 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
6411 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 6412 | |
f91de92e | 6413 | |
1da177e4 LT |
6414 | EXPORT_SYMBOL(register_md_personality); |
6415 | EXPORT_SYMBOL(unregister_md_personality); | |
6416 | EXPORT_SYMBOL(md_error); | |
6417 | EXPORT_SYMBOL(md_done_sync); | |
6418 | EXPORT_SYMBOL(md_write_start); | |
6419 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
6420 | EXPORT_SYMBOL(md_register_thread); |
6421 | EXPORT_SYMBOL(md_unregister_thread); | |
6422 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
6423 | EXPORT_SYMBOL(md_check_recovery); |
6424 | MODULE_LICENSE("GPL"); | |
aa1595e9 | 6425 | MODULE_ALIAS("md"); |
72008652 | 6426 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |