Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | md.c : Multiple Devices driver for Linux | |
3 | Copyright (C) 1998, 1999, 2000 Ingo Molnar | |
4 | ||
5 | completely rewritten, based on the MD driver code from Marc Zyngier | |
6 | ||
7 | Changes: | |
8 | ||
9 | - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar | |
10 | - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com> | |
11 | - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net> | |
12 | - kerneld support by Boris Tobotras <boris@xtalk.msk.su> | |
13 | - kmod support by: Cyrus Durgin | |
14 | - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com> | |
15 | - Devfs support by Richard Gooch <rgooch@atnf.csiro.au> | |
16 | ||
17 | - lots of fixes and improvements to the RAID1/RAID5 and generic | |
18 | RAID code (such as request based resynchronization): | |
19 | ||
20 | Neil Brown <neilb@cse.unsw.edu.au>. | |
21 | ||
32a7627c N |
22 | - persistent bitmap code |
23 | Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. | |
24 | ||
1da177e4 LT |
25 | This program is free software; you can redistribute it and/or modify |
26 | it under the terms of the GNU General Public License as published by | |
27 | the Free Software Foundation; either version 2, or (at your option) | |
28 | any later version. | |
29 | ||
30 | You should have received a copy of the GNU General Public License | |
31 | (for example /usr/src/linux/COPYING); if not, write to the Free | |
32 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
33 | */ | |
34 | ||
a6fb0934 | 35 | #include <linux/kthread.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
1da177e4 | 37 | #include <linux/sysctl.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
ff01bb48 | 39 | #include <linux/fs.h> |
d7603b7e | 40 | #include <linux/poll.h> |
16f17b39 | 41 | #include <linux/ctype.h> |
e7d2860b | 42 | #include <linux/string.h> |
fb4d8c76 N |
43 | #include <linux/hdreg.h> |
44 | #include <linux/proc_fs.h> | |
45 | #include <linux/random.h> | |
056075c7 | 46 | #include <linux/module.h> |
fb4d8c76 | 47 | #include <linux/reboot.h> |
32a7627c | 48 | #include <linux/file.h> |
aa98aa31 | 49 | #include <linux/compat.h> |
25570727 | 50 | #include <linux/delay.h> |
bff61975 N |
51 | #include <linux/raid/md_p.h> |
52 | #include <linux/raid/md_u.h> | |
5a0e3ad6 | 53 | #include <linux/slab.h> |
43b2e5d8 | 54 | #include "md.h" |
ef740c37 | 55 | #include "bitmap.h" |
1da177e4 | 56 | |
1da177e4 | 57 | #ifndef MODULE |
d710e138 | 58 | static void autostart_arrays(int part); |
1da177e4 LT |
59 | #endif |
60 | ||
01f96c0a N |
61 | /* pers_list is a list of registered personalities protected |
62 | * by pers_lock. | |
63 | * pers_lock does extra service to protect accesses to | |
64 | * mddev->thread when the mutex cannot be held. | |
65 | */ | |
2604b703 | 66 | static LIST_HEAD(pers_list); |
1da177e4 LT |
67 | static DEFINE_SPINLOCK(pers_lock); |
68 | ||
5e56341d AB |
69 | static void md_print_devices(void); |
70 | ||
90b08710 | 71 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
72 | static struct workqueue_struct *md_wq; |
73 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 74 | |
5e56341d AB |
75 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } |
76 | ||
1e50915f RB |
77 | /* |
78 | * Default number of read corrections we'll attempt on an rdev | |
79 | * before ejecting it from the array. We divide the read error | |
80 | * count by 2 for every hour elapsed between read errors. | |
81 | */ | |
82 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
83 | /* |
84 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
85 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
86 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 87 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
88 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
89 | * speed limit - in case reconstruction slows down your system despite | |
90 | * idle IO detection. | |
91 | * | |
92 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 93 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
94 | */ |
95 | ||
96 | static int sysctl_speed_limit_min = 1000; | |
97 | static int sysctl_speed_limit_max = 200000; | |
fd01b88c | 98 | static inline int speed_min(struct mddev *mddev) |
88202a0c N |
99 | { |
100 | return mddev->sync_speed_min ? | |
101 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
102 | } | |
103 | ||
fd01b88c | 104 | static inline int speed_max(struct mddev *mddev) |
88202a0c N |
105 | { |
106 | return mddev->sync_speed_max ? | |
107 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
108 | } | |
1da177e4 LT |
109 | |
110 | static struct ctl_table_header *raid_table_header; | |
111 | ||
112 | static ctl_table raid_table[] = { | |
113 | { | |
1da177e4 LT |
114 | .procname = "speed_limit_min", |
115 | .data = &sysctl_speed_limit_min, | |
116 | .maxlen = sizeof(int), | |
80ca3a44 | 117 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 118 | .proc_handler = proc_dointvec, |
1da177e4 LT |
119 | }, |
120 | { | |
1da177e4 LT |
121 | .procname = "speed_limit_max", |
122 | .data = &sysctl_speed_limit_max, | |
123 | .maxlen = sizeof(int), | |
80ca3a44 | 124 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 125 | .proc_handler = proc_dointvec, |
1da177e4 | 126 | }, |
894d2491 | 127 | { } |
1da177e4 LT |
128 | }; |
129 | ||
130 | static ctl_table raid_dir_table[] = { | |
131 | { | |
1da177e4 LT |
132 | .procname = "raid", |
133 | .maxlen = 0, | |
80ca3a44 | 134 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
135 | .child = raid_table, |
136 | }, | |
894d2491 | 137 | { } |
1da177e4 LT |
138 | }; |
139 | ||
140 | static ctl_table raid_root_table[] = { | |
141 | { | |
1da177e4 LT |
142 | .procname = "dev", |
143 | .maxlen = 0, | |
144 | .mode = 0555, | |
145 | .child = raid_dir_table, | |
146 | }, | |
894d2491 | 147 | { } |
1da177e4 LT |
148 | }; |
149 | ||
83d5cde4 | 150 | static const struct block_device_operations md_fops; |
1da177e4 | 151 | |
f91de92e N |
152 | static int start_readonly; |
153 | ||
a167f663 N |
154 | /* bio_clone_mddev |
155 | * like bio_clone, but with a local bio set | |
156 | */ | |
157 | ||
a167f663 | 158 | struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, |
fd01b88c | 159 | struct mddev *mddev) |
a167f663 N |
160 | { |
161 | struct bio *b; | |
a167f663 N |
162 | |
163 | if (!mddev || !mddev->bio_set) | |
164 | return bio_alloc(gfp_mask, nr_iovecs); | |
165 | ||
395c72a7 | 166 | b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set); |
a167f663 N |
167 | if (!b) |
168 | return NULL; | |
a167f663 N |
169 | return b; |
170 | } | |
171 | EXPORT_SYMBOL_GPL(bio_alloc_mddev); | |
172 | ||
173 | struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, | |
fd01b88c | 174 | struct mddev *mddev) |
a167f663 | 175 | { |
a167f663 N |
176 | if (!mddev || !mddev->bio_set) |
177 | return bio_clone(bio, gfp_mask); | |
178 | ||
bf800ef1 | 179 | return bio_clone_bioset(bio, gfp_mask, mddev->bio_set); |
a167f663 N |
180 | } |
181 | EXPORT_SYMBOL_GPL(bio_clone_mddev); | |
182 | ||
d2eb35ac N |
183 | void md_trim_bio(struct bio *bio, int offset, int size) |
184 | { | |
185 | /* 'bio' is a cloned bio which we need to trim to match | |
186 | * the given offset and size. | |
187 | * This requires adjusting bi_sector, bi_size, and bi_io_vec | |
188 | */ | |
189 | int i; | |
190 | struct bio_vec *bvec; | |
191 | int sofar = 0; | |
192 | ||
193 | size <<= 9; | |
194 | if (offset == 0 && size == bio->bi_size) | |
195 | return; | |
196 | ||
197 | bio->bi_sector += offset; | |
198 | bio->bi_size = size; | |
199 | offset <<= 9; | |
200 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
201 | ||
202 | while (bio->bi_idx < bio->bi_vcnt && | |
203 | bio->bi_io_vec[bio->bi_idx].bv_len <= offset) { | |
204 | /* remove this whole bio_vec */ | |
205 | offset -= bio->bi_io_vec[bio->bi_idx].bv_len; | |
206 | bio->bi_idx++; | |
207 | } | |
208 | if (bio->bi_idx < bio->bi_vcnt) { | |
209 | bio->bi_io_vec[bio->bi_idx].bv_offset += offset; | |
210 | bio->bi_io_vec[bio->bi_idx].bv_len -= offset; | |
211 | } | |
212 | /* avoid any complications with bi_idx being non-zero*/ | |
213 | if (bio->bi_idx) { | |
214 | memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx, | |
215 | (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec)); | |
216 | bio->bi_vcnt -= bio->bi_idx; | |
217 | bio->bi_idx = 0; | |
218 | } | |
219 | /* Make sure vcnt and last bv are not too big */ | |
220 | bio_for_each_segment(bvec, bio, i) { | |
221 | if (sofar + bvec->bv_len > size) | |
222 | bvec->bv_len = size - sofar; | |
223 | if (bvec->bv_len == 0) { | |
224 | bio->bi_vcnt = i; | |
225 | break; | |
226 | } | |
227 | sofar += bvec->bv_len; | |
228 | } | |
229 | } | |
230 | EXPORT_SYMBOL_GPL(md_trim_bio); | |
231 | ||
d7603b7e N |
232 | /* |
233 | * We have a system wide 'event count' that is incremented | |
234 | * on any 'interesting' event, and readers of /proc/mdstat | |
235 | * can use 'poll' or 'select' to find out when the event | |
236 | * count increases. | |
237 | * | |
238 | * Events are: | |
239 | * start array, stop array, error, add device, remove device, | |
240 | * start build, activate spare | |
241 | */ | |
2989ddbd | 242 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 243 | static atomic_t md_event_count; |
fd01b88c | 244 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
245 | { |
246 | atomic_inc(&md_event_count); | |
247 | wake_up(&md_event_waiters); | |
248 | } | |
29269553 | 249 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 250 | |
c331eb04 N |
251 | /* Alternate version that can be called from interrupts |
252 | * when calling sysfs_notify isn't needed. | |
253 | */ | |
fd01b88c | 254 | static void md_new_event_inintr(struct mddev *mddev) |
c331eb04 N |
255 | { |
256 | atomic_inc(&md_event_count); | |
257 | wake_up(&md_event_waiters); | |
258 | } | |
259 | ||
1da177e4 LT |
260 | /* |
261 | * Enables to iterate over all existing md arrays | |
262 | * all_mddevs_lock protects this list. | |
263 | */ | |
264 | static LIST_HEAD(all_mddevs); | |
265 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
266 | ||
267 | ||
268 | /* | |
269 | * iterates through all used mddevs in the system. | |
270 | * We take care to grab the all_mddevs_lock whenever navigating | |
271 | * the list, and to always hold a refcount when unlocked. | |
272 | * Any code which breaks out of this loop while own | |
273 | * a reference to the current mddev and must mddev_put it. | |
274 | */ | |
fd01b88c | 275 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 LT |
276 | \ |
277 | for (({ spin_lock(&all_mddevs_lock); \ | |
fd01b88c N |
278 | _tmp = all_mddevs.next; \ |
279 | _mddev = NULL;}); \ | |
280 | ({ if (_tmp != &all_mddevs) \ | |
281 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 282 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
283 | if (_mddev) mddev_put(_mddev); \ |
284 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
285 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 286 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 287 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
288 | ) |
289 | ||
290 | ||
409c57f3 N |
291 | /* Rather than calling directly into the personality make_request function, |
292 | * IO requests come here first so that we can check if the device is | |
293 | * being suspended pending a reconfiguration. | |
294 | * We hold a refcount over the call to ->make_request. By the time that | |
295 | * call has finished, the bio has been linked into some internal structure | |
296 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
297 | */ | |
5a7bbad2 | 298 | static void md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 299 | { |
49077326 | 300 | const int rw = bio_data_dir(bio); |
fd01b88c | 301 | struct mddev *mddev = q->queuedata; |
49077326 | 302 | int cpu; |
e91ece55 | 303 | unsigned int sectors; |
49077326 | 304 | |
0ca69886 N |
305 | if (mddev == NULL || mddev->pers == NULL |
306 | || !mddev->ready) { | |
409c57f3 | 307 | bio_io_error(bio); |
5a7bbad2 | 308 | return; |
409c57f3 | 309 | } |
bbfa57c0 SR |
310 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
311 | bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS); | |
312 | return; | |
313 | } | |
0ca69886 | 314 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 315 | rcu_read_lock(); |
e9c7469b | 316 | if (mddev->suspended) { |
409c57f3 N |
317 | DEFINE_WAIT(__wait); |
318 | for (;;) { | |
319 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
320 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 321 | if (!mddev->suspended) |
409c57f3 N |
322 | break; |
323 | rcu_read_unlock(); | |
324 | schedule(); | |
325 | rcu_read_lock(); | |
326 | } | |
327 | finish_wait(&mddev->sb_wait, &__wait); | |
328 | } | |
329 | atomic_inc(&mddev->active_io); | |
330 | rcu_read_unlock(); | |
49077326 | 331 | |
e91ece55 CM |
332 | /* |
333 | * save the sectors now since our bio can | |
334 | * go away inside make_request | |
335 | */ | |
336 | sectors = bio_sectors(bio); | |
5a7bbad2 | 337 | mddev->pers->make_request(mddev, bio); |
49077326 N |
338 | |
339 | cpu = part_stat_lock(); | |
340 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
e91ece55 | 341 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); |
49077326 N |
342 | part_stat_unlock(); |
343 | ||
409c57f3 N |
344 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
345 | wake_up(&mddev->sb_wait); | |
409c57f3 N |
346 | } |
347 | ||
9e35b99c N |
348 | /* mddev_suspend makes sure no new requests are submitted |
349 | * to the device, and that any requests that have been submitted | |
350 | * are completely handled. | |
351 | * Once ->stop is called and completes, the module will be completely | |
352 | * unused. | |
353 | */ | |
fd01b88c | 354 | void mddev_suspend(struct mddev *mddev) |
409c57f3 N |
355 | { |
356 | BUG_ON(mddev->suspended); | |
357 | mddev->suspended = 1; | |
358 | synchronize_rcu(); | |
359 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
360 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
361 | |
362 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 363 | } |
390ee602 | 364 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 365 | |
fd01b88c | 366 | void mddev_resume(struct mddev *mddev) |
409c57f3 N |
367 | { |
368 | mddev->suspended = 0; | |
369 | wake_up(&mddev->sb_wait); | |
370 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 371 | |
47525e59 | 372 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
373 | md_wakeup_thread(mddev->thread); |
374 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 375 | } |
390ee602 | 376 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 377 | |
fd01b88c | 378 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 N |
379 | { |
380 | return mddev->suspended; | |
381 | } | |
382 | EXPORT_SYMBOL(mddev_congested); | |
383 | ||
a2826aa9 | 384 | /* |
e9c7469b | 385 | * Generic flush handling for md |
a2826aa9 N |
386 | */ |
387 | ||
e9c7469b | 388 | static void md_end_flush(struct bio *bio, int err) |
a2826aa9 | 389 | { |
3cb03002 | 390 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 391 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
392 | |
393 | rdev_dec_pending(rdev, mddev); | |
394 | ||
395 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 396 | /* The pre-request flush has finished */ |
e804ac78 | 397 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
398 | } |
399 | bio_put(bio); | |
400 | } | |
401 | ||
a7a07e69 N |
402 | static void md_submit_flush_data(struct work_struct *ws); |
403 | ||
a035fc3e | 404 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 405 | { |
fd01b88c | 406 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 407 | struct md_rdev *rdev; |
a2826aa9 | 408 | |
a7a07e69 N |
409 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
410 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 411 | rcu_read_lock(); |
dafb20fa | 412 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
413 | if (rdev->raid_disk >= 0 && |
414 | !test_bit(Faulty, &rdev->flags)) { | |
415 | /* Take two references, one is dropped | |
416 | * when request finishes, one after | |
417 | * we reclaim rcu_read_lock | |
418 | */ | |
419 | struct bio *bi; | |
420 | atomic_inc(&rdev->nr_pending); | |
421 | atomic_inc(&rdev->nr_pending); | |
422 | rcu_read_unlock(); | |
b5e1b8ce | 423 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 424 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
425 | bi->bi_private = rdev; |
426 | bi->bi_bdev = rdev->bdev; | |
427 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 428 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
429 | rcu_read_lock(); |
430 | rdev_dec_pending(rdev, mddev); | |
431 | } | |
432 | rcu_read_unlock(); | |
a7a07e69 N |
433 | if (atomic_dec_and_test(&mddev->flush_pending)) |
434 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
435 | } |
436 | ||
e9c7469b | 437 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 438 | { |
fd01b88c | 439 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 440 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 441 | |
e9c7469b | 442 | if (bio->bi_size == 0) |
a2826aa9 N |
443 | /* an empty barrier - all done */ |
444 | bio_endio(bio, 0); | |
445 | else { | |
e9c7469b | 446 | bio->bi_rw &= ~REQ_FLUSH; |
5a7bbad2 | 447 | mddev->pers->make_request(mddev, bio); |
a2826aa9 | 448 | } |
2b74e12e N |
449 | |
450 | mddev->flush_bio = NULL; | |
451 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
452 | } |
453 | ||
fd01b88c | 454 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 N |
455 | { |
456 | spin_lock_irq(&mddev->write_lock); | |
457 | wait_event_lock_irq(mddev->sb_wait, | |
e9c7469b | 458 | !mddev->flush_bio, |
eed8c02e | 459 | mddev->write_lock); |
e9c7469b | 460 | mddev->flush_bio = bio; |
a2826aa9 N |
461 | spin_unlock_irq(&mddev->write_lock); |
462 | ||
a035fc3e N |
463 | INIT_WORK(&mddev->flush_work, submit_flushes); |
464 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 465 | } |
e9c7469b | 466 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 467 | |
74018dc3 | 468 | void md_unplug(struct blk_plug_cb *cb, bool from_schedule) |
97658cdd | 469 | { |
9cbb1750 N |
470 | struct mddev *mddev = cb->data; |
471 | md_wakeup_thread(mddev->thread); | |
472 | kfree(cb); | |
97658cdd | 473 | } |
9cbb1750 | 474 | EXPORT_SYMBOL(md_unplug); |
2ac87401 | 475 | |
fd01b88c | 476 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
477 | { |
478 | atomic_inc(&mddev->active); | |
479 | return mddev; | |
480 | } | |
481 | ||
5fd3a17e | 482 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 483 | |
fd01b88c | 484 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 485 | { |
a167f663 N |
486 | struct bio_set *bs = NULL; |
487 | ||
1da177e4 LT |
488 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
489 | return; | |
d3374825 | 490 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
491 | mddev->ctime == 0 && !mddev->hold_active) { |
492 | /* Array is not configured at all, and not held active, | |
493 | * so destroy it */ | |
af8a2434 | 494 | list_del_init(&mddev->all_mddevs); |
a167f663 N |
495 | bs = mddev->bio_set; |
496 | mddev->bio_set = NULL; | |
d3374825 | 497 | if (mddev->gendisk) { |
e804ac78 TH |
498 | /* We did a probe so need to clean up. Call |
499 | * queue_work inside the spinlock so that | |
500 | * flush_workqueue() after mddev_find will | |
501 | * succeed in waiting for the work to be done. | |
d3374825 N |
502 | */ |
503 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 504 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
505 | } else |
506 | kfree(mddev); | |
507 | } | |
508 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
509 | if (bs) |
510 | bioset_free(bs); | |
1da177e4 LT |
511 | } |
512 | ||
fd01b88c | 513 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
514 | { |
515 | mutex_init(&mddev->open_mutex); | |
516 | mutex_init(&mddev->reconfig_mutex); | |
517 | mutex_init(&mddev->bitmap_info.mutex); | |
518 | INIT_LIST_HEAD(&mddev->disks); | |
519 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
520 | init_timer(&mddev->safemode_timer); | |
521 | atomic_set(&mddev->active, 1); | |
522 | atomic_set(&mddev->openers, 0); | |
523 | atomic_set(&mddev->active_io, 0); | |
524 | spin_lock_init(&mddev->write_lock); | |
525 | atomic_set(&mddev->flush_pending, 0); | |
526 | init_waitqueue_head(&mddev->sb_wait); | |
527 | init_waitqueue_head(&mddev->recovery_wait); | |
528 | mddev->reshape_position = MaxSector; | |
2c810cdd | 529 | mddev->reshape_backwards = 0; |
fafd7fb0 N |
530 | mddev->resync_min = 0; |
531 | mddev->resync_max = MaxSector; | |
532 | mddev->level = LEVEL_NONE; | |
533 | } | |
390ee602 | 534 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 535 | |
fd01b88c | 536 | static struct mddev * mddev_find(dev_t unit) |
1da177e4 | 537 | { |
fd01b88c | 538 | struct mddev *mddev, *new = NULL; |
1da177e4 | 539 | |
8f5f02c4 N |
540 | if (unit && MAJOR(unit) != MD_MAJOR) |
541 | unit &= ~((1<<MdpMinorShift)-1); | |
542 | ||
1da177e4 LT |
543 | retry: |
544 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
545 | |
546 | if (unit) { | |
547 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
548 | if (mddev->unit == unit) { | |
549 | mddev_get(mddev); | |
550 | spin_unlock(&all_mddevs_lock); | |
551 | kfree(new); | |
552 | return mddev; | |
553 | } | |
554 | ||
555 | if (new) { | |
556 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 557 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
558 | new->hold_active = UNTIL_IOCTL; |
559 | return new; | |
1da177e4 | 560 | } |
efeb53c0 N |
561 | } else if (new) { |
562 | /* find an unused unit number */ | |
563 | static int next_minor = 512; | |
564 | int start = next_minor; | |
565 | int is_free = 0; | |
566 | int dev = 0; | |
567 | while (!is_free) { | |
568 | dev = MKDEV(MD_MAJOR, next_minor); | |
569 | next_minor++; | |
570 | if (next_minor > MINORMASK) | |
571 | next_minor = 0; | |
572 | if (next_minor == start) { | |
573 | /* Oh dear, all in use. */ | |
574 | spin_unlock(&all_mddevs_lock); | |
575 | kfree(new); | |
576 | return NULL; | |
577 | } | |
578 | ||
579 | is_free = 1; | |
580 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
581 | if (mddev->unit == dev) { | |
582 | is_free = 0; | |
583 | break; | |
584 | } | |
585 | } | |
586 | new->unit = dev; | |
587 | new->md_minor = MINOR(dev); | |
588 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
589 | list_add(&new->all_mddevs, &all_mddevs); |
590 | spin_unlock(&all_mddevs_lock); | |
591 | return new; | |
592 | } | |
593 | spin_unlock(&all_mddevs_lock); | |
594 | ||
9ffae0cf | 595 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
596 | if (!new) |
597 | return NULL; | |
598 | ||
1da177e4 LT |
599 | new->unit = unit; |
600 | if (MAJOR(unit) == MD_MAJOR) | |
601 | new->md_minor = MINOR(unit); | |
602 | else | |
603 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
604 | ||
fafd7fb0 | 605 | mddev_init(new); |
1da177e4 | 606 | |
1da177e4 LT |
607 | goto retry; |
608 | } | |
609 | ||
fd01b88c | 610 | static inline int mddev_lock(struct mddev * mddev) |
1da177e4 | 611 | { |
df5b89b3 | 612 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
613 | } |
614 | ||
fd01b88c | 615 | static inline int mddev_is_locked(struct mddev *mddev) |
b522adcd DW |
616 | { |
617 | return mutex_is_locked(&mddev->reconfig_mutex); | |
618 | } | |
619 | ||
fd01b88c | 620 | static inline int mddev_trylock(struct mddev * mddev) |
1da177e4 | 621 | { |
df5b89b3 | 622 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
623 | } |
624 | ||
b6eb127d N |
625 | static struct attribute_group md_redundancy_group; |
626 | ||
fd01b88c | 627 | static void mddev_unlock(struct mddev * mddev) |
1da177e4 | 628 | { |
a64c876f | 629 | if (mddev->to_remove) { |
b6eb127d N |
630 | /* These cannot be removed under reconfig_mutex as |
631 | * an access to the files will try to take reconfig_mutex | |
632 | * while holding the file unremovable, which leads to | |
633 | * a deadlock. | |
bb4f1e9d N |
634 | * So hold set sysfs_active while the remove in happeing, |
635 | * and anything else which might set ->to_remove or my | |
636 | * otherwise change the sysfs namespace will fail with | |
637 | * -EBUSY if sysfs_active is still set. | |
638 | * We set sysfs_active under reconfig_mutex and elsewhere | |
639 | * test it under the same mutex to ensure its correct value | |
640 | * is seen. | |
b6eb127d | 641 | */ |
a64c876f N |
642 | struct attribute_group *to_remove = mddev->to_remove; |
643 | mddev->to_remove = NULL; | |
bb4f1e9d | 644 | mddev->sysfs_active = 1; |
b6eb127d N |
645 | mutex_unlock(&mddev->reconfig_mutex); |
646 | ||
00bcb4ac N |
647 | if (mddev->kobj.sd) { |
648 | if (to_remove != &md_redundancy_group) | |
649 | sysfs_remove_group(&mddev->kobj, to_remove); | |
650 | if (mddev->pers == NULL || | |
651 | mddev->pers->sync_request == NULL) { | |
652 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
653 | if (mddev->sysfs_action) | |
654 | sysfs_put(mddev->sysfs_action); | |
655 | mddev->sysfs_action = NULL; | |
656 | } | |
a64c876f | 657 | } |
bb4f1e9d | 658 | mddev->sysfs_active = 0; |
b6eb127d N |
659 | } else |
660 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 661 | |
751e67ca CD |
662 | /* As we've dropped the mutex we need a spinlock to |
663 | * make sure the thread doesn't disappear | |
01f96c0a N |
664 | */ |
665 | spin_lock(&pers_lock); | |
005eca5e | 666 | md_wakeup_thread(mddev->thread); |
01f96c0a | 667 | spin_unlock(&pers_lock); |
1da177e4 LT |
668 | } |
669 | ||
fd01b88c | 670 | static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr) |
1da177e4 | 671 | { |
3cb03002 | 672 | struct md_rdev *rdev; |
1da177e4 | 673 | |
dafb20fa | 674 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
675 | if (rdev->desc_nr == nr) |
676 | return rdev; | |
159ec1fc | 677 | |
1da177e4 LT |
678 | return NULL; |
679 | } | |
680 | ||
1ca69c4b N |
681 | static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr) |
682 | { | |
683 | struct md_rdev *rdev; | |
684 | ||
685 | rdev_for_each_rcu(rdev, mddev) | |
686 | if (rdev->desc_nr == nr) | |
687 | return rdev; | |
688 | ||
689 | return NULL; | |
690 | } | |
691 | ||
692 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 693 | { |
3cb03002 | 694 | struct md_rdev *rdev; |
1da177e4 | 695 | |
dafb20fa | 696 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
697 | if (rdev->bdev->bd_dev == dev) |
698 | return rdev; | |
159ec1fc | 699 | |
1da177e4 LT |
700 | return NULL; |
701 | } | |
702 | ||
1ca69c4b N |
703 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
704 | { | |
705 | struct md_rdev *rdev; | |
706 | ||
707 | rdev_for_each_rcu(rdev, mddev) | |
708 | if (rdev->bdev->bd_dev == dev) | |
709 | return rdev; | |
710 | ||
711 | return NULL; | |
712 | } | |
713 | ||
84fc4b56 | 714 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 715 | { |
84fc4b56 | 716 | struct md_personality *pers; |
d9d166c2 N |
717 | list_for_each_entry(pers, &pers_list, list) { |
718 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 719 | return pers; |
d9d166c2 N |
720 | if (strcmp(pers->name, clevel)==0) |
721 | return pers; | |
722 | } | |
2604b703 N |
723 | return NULL; |
724 | } | |
725 | ||
b73df2d3 | 726 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 727 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 728 | { |
57b2caa3 | 729 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 730 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
731 | } |
732 | ||
3cb03002 | 733 | static int alloc_disk_sb(struct md_rdev * rdev) |
1da177e4 LT |
734 | { |
735 | if (rdev->sb_page) | |
736 | MD_BUG(); | |
737 | ||
738 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
739 | if (!rdev->sb_page) { | |
740 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 741 | return -ENOMEM; |
1da177e4 LT |
742 | } |
743 | ||
744 | return 0; | |
745 | } | |
746 | ||
545c8795 | 747 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
748 | { |
749 | if (rdev->sb_page) { | |
2d1f3b5d | 750 | put_page(rdev->sb_page); |
1da177e4 LT |
751 | rdev->sb_loaded = 0; |
752 | rdev->sb_page = NULL; | |
0f420358 | 753 | rdev->sb_start = 0; |
dd8ac336 | 754 | rdev->sectors = 0; |
1da177e4 | 755 | } |
2699b672 N |
756 | if (rdev->bb_page) { |
757 | put_page(rdev->bb_page); | |
758 | rdev->bb_page = NULL; | |
759 | } | |
4fa2f327 N |
760 | kfree(rdev->badblocks.page); |
761 | rdev->badblocks.page = NULL; | |
1da177e4 | 762 | } |
545c8795 | 763 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 764 | |
6712ecf8 | 765 | static void super_written(struct bio *bio, int error) |
7bfa19f2 | 766 | { |
3cb03002 | 767 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 768 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 769 | |
3a0f5bbb N |
770 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
771 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
772 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
773 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 774 | md_error(mddev, rdev); |
3a0f5bbb | 775 | } |
7bfa19f2 | 776 | |
a9701a30 N |
777 | if (atomic_dec_and_test(&mddev->pending_writes)) |
778 | wake_up(&mddev->sb_wait); | |
f8b58edf | 779 | bio_put(bio); |
7bfa19f2 N |
780 | } |
781 | ||
fd01b88c | 782 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
783 | sector_t sector, int size, struct page *page) |
784 | { | |
785 | /* write first size bytes of page to sector of rdev | |
786 | * Increment mddev->pending_writes before returning | |
787 | * and decrement it on completion, waking up sb_wait | |
788 | * if zero is reached. | |
789 | * If an error occurred, call md_error | |
790 | */ | |
a167f663 | 791 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 792 | |
a6ff7e08 | 793 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
7bfa19f2 N |
794 | bio->bi_sector = sector; |
795 | bio_add_page(bio, page, size, 0); | |
796 | bio->bi_private = rdev; | |
797 | bio->bi_end_io = super_written; | |
a9701a30 | 798 | |
7bfa19f2 | 799 | atomic_inc(&mddev->pending_writes); |
a5bf4df0 | 800 | submit_bio(WRITE_FLUSH_FUA, bio); |
a9701a30 N |
801 | } |
802 | ||
fd01b88c | 803 | void md_super_wait(struct mddev *mddev) |
a9701a30 | 804 | { |
e9c7469b | 805 | /* wait for all superblock writes that were scheduled to complete */ |
a9701a30 N |
806 | DEFINE_WAIT(wq); |
807 | for(;;) { | |
808 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
809 | if (atomic_read(&mddev->pending_writes)==0) | |
810 | break; | |
a9701a30 N |
811 | schedule(); |
812 | } | |
813 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
814 | } |
815 | ||
6712ecf8 | 816 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 817 | { |
1da177e4 | 818 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
819 | } |
820 | ||
3cb03002 | 821 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
ccebd4c4 | 822 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 823 | { |
a167f663 | 824 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
825 | struct completion event; |
826 | int ret; | |
827 | ||
721a9602 | 828 | rw |= REQ_SYNC; |
1da177e4 | 829 | |
a6ff7e08 JB |
830 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
831 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 JB |
832 | if (metadata_op) |
833 | bio->bi_sector = sector + rdev->sb_start; | |
1fdd6fc9 N |
834 | else if (rdev->mddev->reshape_position != MaxSector && |
835 | (rdev->mddev->reshape_backwards == | |
836 | (sector >= rdev->mddev->reshape_position))) | |
837 | bio->bi_sector = sector + rdev->new_data_offset; | |
ccebd4c4 JB |
838 | else |
839 | bio->bi_sector = sector + rdev->data_offset; | |
1da177e4 LT |
840 | bio_add_page(bio, page, size, 0); |
841 | init_completion(&event); | |
842 | bio->bi_private = &event; | |
843 | bio->bi_end_io = bi_complete; | |
844 | submit_bio(rw, bio); | |
845 | wait_for_completion(&event); | |
846 | ||
847 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
848 | bio_put(bio); | |
849 | return ret; | |
850 | } | |
a8745db2 | 851 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 852 | |
3cb03002 | 853 | static int read_disk_sb(struct md_rdev * rdev, int size) |
1da177e4 LT |
854 | { |
855 | char b[BDEVNAME_SIZE]; | |
856 | if (!rdev->sb_page) { | |
857 | MD_BUG(); | |
858 | return -EINVAL; | |
859 | } | |
860 | if (rdev->sb_loaded) | |
861 | return 0; | |
862 | ||
863 | ||
ccebd4c4 | 864 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
865 | goto fail; |
866 | rdev->sb_loaded = 1; | |
867 | return 0; | |
868 | ||
869 | fail: | |
870 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
871 | bdevname(rdev->bdev,b)); | |
872 | return -EINVAL; | |
873 | } | |
874 | ||
875 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
876 | { | |
05710466 AN |
877 | return sb1->set_uuid0 == sb2->set_uuid0 && |
878 | sb1->set_uuid1 == sb2->set_uuid1 && | |
879 | sb1->set_uuid2 == sb2->set_uuid2 && | |
880 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
881 | } |
882 | ||
1da177e4 LT |
883 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
884 | { | |
885 | int ret; | |
886 | mdp_super_t *tmp1, *tmp2; | |
887 | ||
888 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
889 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
890 | ||
891 | if (!tmp1 || !tmp2) { | |
892 | ret = 0; | |
35020f1a | 893 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
894 | goto abort; |
895 | } | |
896 | ||
897 | *tmp1 = *sb1; | |
898 | *tmp2 = *sb2; | |
899 | ||
900 | /* | |
901 | * nr_disks is not constant | |
902 | */ | |
903 | tmp1->nr_disks = 0; | |
904 | tmp2->nr_disks = 0; | |
905 | ||
ce0c8e05 | 906 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 907 | abort: |
990a8baf JJ |
908 | kfree(tmp1); |
909 | kfree(tmp2); | |
1da177e4 LT |
910 | return ret; |
911 | } | |
912 | ||
4d167f09 N |
913 | |
914 | static u32 md_csum_fold(u32 csum) | |
915 | { | |
916 | csum = (csum & 0xffff) + (csum >> 16); | |
917 | return (csum & 0xffff) + (csum >> 16); | |
918 | } | |
919 | ||
1da177e4 LT |
920 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
921 | { | |
4d167f09 N |
922 | u64 newcsum = 0; |
923 | u32 *sb32 = (u32*)sb; | |
924 | int i; | |
1da177e4 LT |
925 | unsigned int disk_csum, csum; |
926 | ||
927 | disk_csum = sb->sb_csum; | |
928 | sb->sb_csum = 0; | |
4d167f09 N |
929 | |
930 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
931 | newcsum += sb32[i]; | |
932 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
933 | ||
934 | ||
935 | #ifdef CONFIG_ALPHA | |
936 | /* This used to use csum_partial, which was wrong for several | |
937 | * reasons including that different results are returned on | |
938 | * different architectures. It isn't critical that we get exactly | |
939 | * the same return value as before (we always csum_fold before | |
940 | * testing, and that removes any differences). However as we | |
941 | * know that csum_partial always returned a 16bit value on | |
942 | * alphas, do a fold to maximise conformity to previous behaviour. | |
943 | */ | |
944 | sb->sb_csum = md_csum_fold(disk_csum); | |
945 | #else | |
1da177e4 | 946 | sb->sb_csum = disk_csum; |
4d167f09 | 947 | #endif |
1da177e4 LT |
948 | return csum; |
949 | } | |
950 | ||
951 | ||
952 | /* | |
953 | * Handle superblock details. | |
954 | * We want to be able to handle multiple superblock formats | |
955 | * so we have a common interface to them all, and an array of | |
956 | * different handlers. | |
957 | * We rely on user-space to write the initial superblock, and support | |
958 | * reading and updating of superblocks. | |
959 | * Interface methods are: | |
3cb03002 | 960 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
961 | * loads and validates a superblock on dev. |
962 | * if refdev != NULL, compare superblocks on both devices | |
963 | * Return: | |
964 | * 0 - dev has a superblock that is compatible with refdev | |
965 | * 1 - dev has a superblock that is compatible and newer than refdev | |
966 | * so dev should be used as the refdev in future | |
967 | * -EINVAL superblock incompatible or invalid | |
968 | * -othererror e.g. -EIO | |
969 | * | |
fd01b88c | 970 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
971 | * Verify that dev is acceptable into mddev. |
972 | * The first time, mddev->raid_disks will be 0, and data from | |
973 | * dev should be merged in. Subsequent calls check that dev | |
974 | * is new enough. Return 0 or -EINVAL | |
975 | * | |
fd01b88c | 976 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
977 | * Update the superblock for rdev with data in mddev |
978 | * This does not write to disc. | |
979 | * | |
980 | */ | |
981 | ||
982 | struct super_type { | |
0cd17fec CW |
983 | char *name; |
984 | struct module *owner; | |
c6563a8c N |
985 | int (*load_super)(struct md_rdev *rdev, |
986 | struct md_rdev *refdev, | |
0cd17fec | 987 | int minor_version); |
c6563a8c N |
988 | int (*validate_super)(struct mddev *mddev, |
989 | struct md_rdev *rdev); | |
990 | void (*sync_super)(struct mddev *mddev, | |
991 | struct md_rdev *rdev); | |
3cb03002 | 992 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 993 | sector_t num_sectors); |
c6563a8c N |
994 | int (*allow_new_offset)(struct md_rdev *rdev, |
995 | unsigned long long new_offset); | |
1da177e4 LT |
996 | }; |
997 | ||
0894cc30 AN |
998 | /* |
999 | * Check that the given mddev has no bitmap. | |
1000 | * | |
1001 | * This function is called from the run method of all personalities that do not | |
1002 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
1003 | * has a bitmap. Otherwise, it returns 0. | |
1004 | * | |
1005 | */ | |
fd01b88c | 1006 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 1007 | { |
c3d9714e | 1008 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
1009 | return 0; |
1010 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
1011 | mdname(mddev), mddev->pers->name); | |
1012 | return 1; | |
1013 | } | |
1014 | EXPORT_SYMBOL(md_check_no_bitmap); | |
1015 | ||
1da177e4 LT |
1016 | /* |
1017 | * load_super for 0.90.0 | |
1018 | */ | |
3cb03002 | 1019 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1020 | { |
1021 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
1022 | mdp_super_t *sb; | |
1023 | int ret; | |
1da177e4 LT |
1024 | |
1025 | /* | |
0f420358 | 1026 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
1027 | * it's at the end of the disk. |
1028 | * | |
1029 | * It also happens to be a multiple of 4Kb. | |
1030 | */ | |
57b2caa3 | 1031 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 1032 | |
0002b271 | 1033 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
1034 | if (ret) return ret; |
1035 | ||
1036 | ret = -EINVAL; | |
1037 | ||
1038 | bdevname(rdev->bdev, b); | |
65a06f06 | 1039 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1040 | |
1041 | if (sb->md_magic != MD_SB_MAGIC) { | |
1042 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
1043 | b); | |
1044 | goto abort; | |
1045 | } | |
1046 | ||
1047 | if (sb->major_version != 0 || | |
f6705578 N |
1048 | sb->minor_version < 90 || |
1049 | sb->minor_version > 91) { | |
1da177e4 LT |
1050 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
1051 | sb->major_version, sb->minor_version, | |
1052 | b); | |
1053 | goto abort; | |
1054 | } | |
1055 | ||
1056 | if (sb->raid_disks <= 0) | |
1057 | goto abort; | |
1058 | ||
4d167f09 | 1059 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
1060 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
1061 | b); | |
1062 | goto abort; | |
1063 | } | |
1064 | ||
1065 | rdev->preferred_minor = sb->md_minor; | |
1066 | rdev->data_offset = 0; | |
c6563a8c | 1067 | rdev->new_data_offset = 0; |
0002b271 | 1068 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1069 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1070 | |
1071 | if (sb->level == LEVEL_MULTIPATH) | |
1072 | rdev->desc_nr = -1; | |
1073 | else | |
1074 | rdev->desc_nr = sb->this_disk.number; | |
1075 | ||
9a7b2b0f | 1076 | if (!refdev) { |
1da177e4 | 1077 | ret = 1; |
9a7b2b0f | 1078 | } else { |
1da177e4 | 1079 | __u64 ev1, ev2; |
65a06f06 | 1080 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1081 | if (!uuid_equal(refsb, sb)) { |
1082 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1083 | b, bdevname(refdev->bdev,b2)); | |
1084 | goto abort; | |
1085 | } | |
1086 | if (!sb_equal(refsb, sb)) { | |
1087 | printk(KERN_WARNING "md: %s has same UUID" | |
1088 | " but different superblock to %s\n", | |
1089 | b, bdevname(refdev->bdev, b2)); | |
1090 | goto abort; | |
1091 | } | |
1092 | ev1 = md_event(sb); | |
1093 | ev2 = md_event(refsb); | |
1094 | if (ev1 > ev2) | |
1095 | ret = 1; | |
1096 | else | |
1097 | ret = 0; | |
1098 | } | |
8190e754 | 1099 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1100 | /* Limit to 4TB as metadata cannot record more than that. |
1101 | * (not needed for Linear and RAID0 as metadata doesn't | |
1102 | * record this size) | |
1103 | */ | |
1104 | if (rdev->sectors >= (2ULL << 32) && sb->level >= 1) | |
27a7b260 | 1105 | rdev->sectors = (2ULL << 32) - 2; |
1da177e4 | 1106 | |
27a7b260 | 1107 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1108 | /* "this cannot possibly happen" ... */ |
1109 | ret = -EINVAL; | |
1110 | ||
1da177e4 LT |
1111 | abort: |
1112 | return ret; | |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * validate_super for 0.90.0 | |
1117 | */ | |
fd01b88c | 1118 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1119 | { |
1120 | mdp_disk_t *desc; | |
65a06f06 | 1121 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1122 | __u64 ev1 = md_event(sb); |
1da177e4 | 1123 | |
41158c7e | 1124 | rdev->raid_disk = -1; |
c5d79adb N |
1125 | clear_bit(Faulty, &rdev->flags); |
1126 | clear_bit(In_sync, &rdev->flags); | |
1127 | clear_bit(WriteMostly, &rdev->flags); | |
c5d79adb | 1128 | |
1da177e4 LT |
1129 | if (mddev->raid_disks == 0) { |
1130 | mddev->major_version = 0; | |
1131 | mddev->minor_version = sb->minor_version; | |
1132 | mddev->patch_version = sb->patch_version; | |
e691063a | 1133 | mddev->external = 0; |
9d8f0363 | 1134 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1135 | mddev->ctime = sb->ctime; |
1136 | mddev->utime = sb->utime; | |
1137 | mddev->level = sb->level; | |
d9d166c2 | 1138 | mddev->clevel[0] = 0; |
1da177e4 LT |
1139 | mddev->layout = sb->layout; |
1140 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1141 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1142 | mddev->events = ev1; |
c3d9714e | 1143 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1144 | mddev->bitmap_info.space = 0; |
1145 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1146 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1147 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1148 | mddev->reshape_backwards = 0; |
1da177e4 | 1149 | |
f6705578 N |
1150 | if (mddev->minor_version >= 91) { |
1151 | mddev->reshape_position = sb->reshape_position; | |
1152 | mddev->delta_disks = sb->delta_disks; | |
1153 | mddev->new_level = sb->new_level; | |
1154 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1155 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1156 | if (mddev->delta_disks < 0) |
1157 | mddev->reshape_backwards = 1; | |
f6705578 N |
1158 | } else { |
1159 | mddev->reshape_position = MaxSector; | |
1160 | mddev->delta_disks = 0; | |
1161 | mddev->new_level = mddev->level; | |
1162 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1163 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1164 | } |
1165 | ||
1da177e4 LT |
1166 | if (sb->state & (1<<MD_SB_CLEAN)) |
1167 | mddev->recovery_cp = MaxSector; | |
1168 | else { | |
1169 | if (sb->events_hi == sb->cp_events_hi && | |
1170 | sb->events_lo == sb->cp_events_lo) { | |
1171 | mddev->recovery_cp = sb->recovery_cp; | |
1172 | } else | |
1173 | mddev->recovery_cp = 0; | |
1174 | } | |
1175 | ||
1176 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1177 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1178 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1179 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1180 | ||
1181 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1182 | |
1183 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1184 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1185 | mddev->bitmap_info.offset = |
1186 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
1187 | mddev->bitmap_info.space = |
1188 | mddev->bitmap_info.space; | |
1189 | } | |
a654b9d8 | 1190 | |
41158c7e | 1191 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1192 | /* Insist on good event counter while assembling, except |
1193 | * for spares (which don't need an event count) */ | |
1da177e4 | 1194 | ++ev1; |
be6800a7 N |
1195 | if (sb->disks[rdev->desc_nr].state & ( |
1196 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
1197 | if (ev1 < mddev->events) | |
1198 | return -EINVAL; | |
41158c7e N |
1199 | } else if (mddev->bitmap) { |
1200 | /* if adding to array with a bitmap, then we can accept an | |
1201 | * older device ... but not too old. | |
1202 | */ | |
41158c7e N |
1203 | if (ev1 < mddev->bitmap->events_cleared) |
1204 | return 0; | |
07d84d10 N |
1205 | } else { |
1206 | if (ev1 < mddev->events) | |
1207 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1208 | return 0; | |
1209 | } | |
41158c7e | 1210 | |
1da177e4 | 1211 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1212 | desc = sb->disks + rdev->desc_nr; |
1213 | ||
1214 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1215 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1216 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1217 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1218 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1219 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
1220 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1221 | /* active but not in sync implies recovery up to | |
1222 | * reshape position. We don't know exactly where | |
1223 | * that is, so set to zero for now */ | |
1224 | if (mddev->minor_version >= 91) { | |
1225 | rdev->recovery_offset = 0; | |
1226 | rdev->raid_disk = desc->raid_disk; | |
1227 | } | |
1da177e4 | 1228 | } |
8ddf9efe N |
1229 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1230 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1231 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1232 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1233 | return 0; |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * sync_super for 0.90.0 | |
1238 | */ | |
fd01b88c | 1239 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1240 | { |
1241 | mdp_super_t *sb; | |
3cb03002 | 1242 | struct md_rdev *rdev2; |
1da177e4 | 1243 | int next_spare = mddev->raid_disks; |
19133a42 | 1244 | |
1da177e4 LT |
1245 | |
1246 | /* make rdev->sb match mddev data.. | |
1247 | * | |
1248 | * 1/ zero out disks | |
1249 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1250 | * 3/ any empty disks < next_spare become removed | |
1251 | * | |
1252 | * disks[0] gets initialised to REMOVED because | |
1253 | * we cannot be sure from other fields if it has | |
1254 | * been initialised or not. | |
1255 | */ | |
1256 | int i; | |
1257 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1258 | ||
61181565 N |
1259 | rdev->sb_size = MD_SB_BYTES; |
1260 | ||
65a06f06 | 1261 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1262 | |
1263 | memset(sb, 0, sizeof(*sb)); | |
1264 | ||
1265 | sb->md_magic = MD_SB_MAGIC; | |
1266 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1267 | sb->patch_version = mddev->patch_version; |
1268 | sb->gvalid_words = 0; /* ignored */ | |
1269 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1270 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1271 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1272 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1273 | ||
1274 | sb->ctime = mddev->ctime; | |
1275 | sb->level = mddev->level; | |
58c0fed4 | 1276 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1277 | sb->raid_disks = mddev->raid_disks; |
1278 | sb->md_minor = mddev->md_minor; | |
e691063a | 1279 | sb->not_persistent = 0; |
1da177e4 LT |
1280 | sb->utime = mddev->utime; |
1281 | sb->state = 0; | |
1282 | sb->events_hi = (mddev->events>>32); | |
1283 | sb->events_lo = (u32)mddev->events; | |
1284 | ||
f6705578 N |
1285 | if (mddev->reshape_position == MaxSector) |
1286 | sb->minor_version = 90; | |
1287 | else { | |
1288 | sb->minor_version = 91; | |
1289 | sb->reshape_position = mddev->reshape_position; | |
1290 | sb->new_level = mddev->new_level; | |
1291 | sb->delta_disks = mddev->delta_disks; | |
1292 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1293 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1294 | } |
1295 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1296 | if (mddev->in_sync) |
1297 | { | |
1298 | sb->recovery_cp = mddev->recovery_cp; | |
1299 | sb->cp_events_hi = (mddev->events>>32); | |
1300 | sb->cp_events_lo = (u32)mddev->events; | |
1301 | if (mddev->recovery_cp == MaxSector) | |
1302 | sb->state = (1<< MD_SB_CLEAN); | |
1303 | } else | |
1304 | sb->recovery_cp = 0; | |
1305 | ||
1306 | sb->layout = mddev->layout; | |
9d8f0363 | 1307 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1308 | |
c3d9714e | 1309 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1310 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1311 | ||
1da177e4 | 1312 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1313 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1314 | mdp_disk_t *d; |
86e6ffdd | 1315 | int desc_nr; |
0261cd9f N |
1316 | int is_active = test_bit(In_sync, &rdev2->flags); |
1317 | ||
1318 | if (rdev2->raid_disk >= 0 && | |
1319 | sb->minor_version >= 91) | |
1320 | /* we have nowhere to store the recovery_offset, | |
1321 | * but if it is not below the reshape_position, | |
1322 | * we can piggy-back on that. | |
1323 | */ | |
1324 | is_active = 1; | |
1325 | if (rdev2->raid_disk < 0 || | |
1326 | test_bit(Faulty, &rdev2->flags)) | |
1327 | is_active = 0; | |
1328 | if (is_active) | |
86e6ffdd | 1329 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1330 | else |
86e6ffdd | 1331 | desc_nr = next_spare++; |
19133a42 | 1332 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1333 | d = &sb->disks[rdev2->desc_nr]; |
1334 | nr_disks++; | |
1335 | d->number = rdev2->desc_nr; | |
1336 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1337 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1338 | if (is_active) |
1da177e4 LT |
1339 | d->raid_disk = rdev2->raid_disk; |
1340 | else | |
1341 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1342 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1343 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1344 | else if (is_active) { |
1da177e4 | 1345 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1346 | if (test_bit(In_sync, &rdev2->flags)) |
1347 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1348 | active++; |
1349 | working++; | |
1350 | } else { | |
1351 | d->state = 0; | |
1352 | spare++; | |
1353 | working++; | |
1354 | } | |
8ddf9efe N |
1355 | if (test_bit(WriteMostly, &rdev2->flags)) |
1356 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1357 | } |
1da177e4 LT |
1358 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1359 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1360 | mdp_disk_t *d = &sb->disks[i]; | |
1361 | if (d->state == 0 && d->number == 0) { | |
1362 | d->number = i; | |
1363 | d->raid_disk = i; | |
1364 | d->state = (1<<MD_DISK_REMOVED); | |
1365 | d->state |= (1<<MD_DISK_FAULTY); | |
1366 | failed++; | |
1367 | } | |
1368 | } | |
1369 | sb->nr_disks = nr_disks; | |
1370 | sb->active_disks = active; | |
1371 | sb->working_disks = working; | |
1372 | sb->failed_disks = failed; | |
1373 | sb->spare_disks = spare; | |
1374 | ||
1375 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1376 | sb->sb_csum = calc_sb_csum(sb); | |
1377 | } | |
1378 | ||
0cd17fec CW |
1379 | /* |
1380 | * rdev_size_change for 0.90.0 | |
1381 | */ | |
1382 | static unsigned long long | |
3cb03002 | 1383 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1384 | { |
58c0fed4 | 1385 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1386 | return 0; /* component must fit device */ |
c3d9714e | 1387 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1388 | return 0; /* can't move bitmap */ |
57b2caa3 | 1389 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1390 | if (!num_sectors || num_sectors > rdev->sb_start) |
1391 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1392 | /* Limit to 4TB as metadata cannot record more than that. |
1393 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1394 | */ | |
667a5313 | 1395 | if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1) |
27a7b260 | 1396 | num_sectors = (2ULL << 32) - 2; |
0f420358 | 1397 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1398 | rdev->sb_page); |
1399 | md_super_wait(rdev->mddev); | |
c26a44ed | 1400 | return num_sectors; |
0cd17fec CW |
1401 | } |
1402 | ||
c6563a8c N |
1403 | static int |
1404 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1405 | { | |
1406 | /* non-zero offset changes not possible with v0.90 */ | |
1407 | return new_offset == 0; | |
1408 | } | |
0cd17fec | 1409 | |
1da177e4 LT |
1410 | /* |
1411 | * version 1 superblock | |
1412 | */ | |
1413 | ||
1c05b4bc | 1414 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1415 | { |
1c05b4bc N |
1416 | __le32 disk_csum; |
1417 | u32 csum; | |
1da177e4 LT |
1418 | unsigned long long newcsum; |
1419 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1420 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1421 | |
1422 | disk_csum = sb->sb_csum; | |
1423 | sb->sb_csum = 0; | |
1424 | newcsum = 0; | |
1f3c9907 | 1425 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1426 | newcsum += le32_to_cpu(*isuper++); |
1427 | ||
1428 | if (size == 2) | |
1c05b4bc | 1429 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1430 | |
1431 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1432 | sb->sb_csum = disk_csum; | |
1433 | return cpu_to_le32(csum); | |
1434 | } | |
1435 | ||
2699b672 N |
1436 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, |
1437 | int acknowledged); | |
3cb03002 | 1438 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1439 | { |
1440 | struct mdp_superblock_1 *sb; | |
1441 | int ret; | |
0f420358 | 1442 | sector_t sb_start; |
c6563a8c | 1443 | sector_t sectors; |
1da177e4 | 1444 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1445 | int bmask; |
1da177e4 LT |
1446 | |
1447 | /* | |
0f420358 | 1448 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1449 | * It is always aligned to a 4K boundary and |
1450 | * depeding on minor_version, it can be: | |
1451 | * 0: At least 8K, but less than 12K, from end of device | |
1452 | * 1: At start of device | |
1453 | * 2: 4K from start of device. | |
1454 | */ | |
1455 | switch(minor_version) { | |
1456 | case 0: | |
77304d2a | 1457 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1458 | sb_start -= 8*2; |
1459 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1460 | break; |
1461 | case 1: | |
0f420358 | 1462 | sb_start = 0; |
1da177e4 LT |
1463 | break; |
1464 | case 2: | |
0f420358 | 1465 | sb_start = 8; |
1da177e4 LT |
1466 | break; |
1467 | default: | |
1468 | return -EINVAL; | |
1469 | } | |
0f420358 | 1470 | rdev->sb_start = sb_start; |
1da177e4 | 1471 | |
0002b271 N |
1472 | /* superblock is rarely larger than 1K, but it can be larger, |
1473 | * and it is safe to read 4k, so we do that | |
1474 | */ | |
1475 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1476 | if (ret) return ret; |
1477 | ||
1478 | ||
65a06f06 | 1479 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1480 | |
1481 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1482 | sb->major_version != cpu_to_le32(1) || | |
1483 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1484 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1485 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1486 | return -EINVAL; |
1487 | ||
1488 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1489 | printk("md: invalid superblock checksum on %s\n", | |
1490 | bdevname(rdev->bdev,b)); | |
1491 | return -EINVAL; | |
1492 | } | |
1493 | if (le64_to_cpu(sb->data_size) < 10) { | |
1494 | printk("md: data_size too small on %s\n", | |
1495 | bdevname(rdev->bdev,b)); | |
1496 | return -EINVAL; | |
1497 | } | |
c6563a8c N |
1498 | if (sb->pad0 || |
1499 | sb->pad3[0] || | |
1500 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1501 | /* Some padding is non-zero, might be a new feature */ | |
1502 | return -EINVAL; | |
e11e93fa | 1503 | |
1da177e4 LT |
1504 | rdev->preferred_minor = 0xffff; |
1505 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1506 | rdev->new_data_offset = rdev->data_offset; |
1507 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1508 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1509 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1510 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1511 | |
0002b271 | 1512 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1513 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1514 | if (rdev->sb_size & bmask) |
a1801f85 N |
1515 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1516 | ||
1517 | if (minor_version | |
0f420358 | 1518 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1519 | return -EINVAL; |
c6563a8c N |
1520 | if (minor_version |
1521 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1522 | return -EINVAL; | |
0002b271 | 1523 | |
31b65a0d N |
1524 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1525 | rdev->desc_nr = -1; | |
1526 | else | |
1527 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1528 | ||
2699b672 N |
1529 | if (!rdev->bb_page) { |
1530 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1531 | if (!rdev->bb_page) | |
1532 | return -ENOMEM; | |
1533 | } | |
1534 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1535 | rdev->badblocks.count == 0) { | |
1536 | /* need to load the bad block list. | |
1537 | * Currently we limit it to one page. | |
1538 | */ | |
1539 | s32 offset; | |
1540 | sector_t bb_sector; | |
1541 | u64 *bbp; | |
1542 | int i; | |
1543 | int sectors = le16_to_cpu(sb->bblog_size); | |
1544 | if (sectors > (PAGE_SIZE / 512)) | |
1545 | return -EINVAL; | |
1546 | offset = le32_to_cpu(sb->bblog_offset); | |
1547 | if (offset == 0) | |
1548 | return -EINVAL; | |
1549 | bb_sector = (long long)offset; | |
1550 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
1551 | rdev->bb_page, READ, true)) | |
1552 | return -EIO; | |
1553 | bbp = (u64 *)page_address(rdev->bb_page); | |
1554 | rdev->badblocks.shift = sb->bblog_shift; | |
1555 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1556 | u64 bb = le64_to_cpu(*bbp); | |
1557 | int count = bb & (0x3ff); | |
1558 | u64 sector = bb >> 10; | |
1559 | sector <<= sb->bblog_shift; | |
1560 | count <<= sb->bblog_shift; | |
1561 | if (bb + 1 == 0) | |
1562 | break; | |
1563 | if (md_set_badblocks(&rdev->badblocks, | |
1564 | sector, count, 1) == 0) | |
1565 | return -EINVAL; | |
1566 | } | |
1567 | } else if (sb->bblog_offset == 0) | |
1568 | rdev->badblocks.shift = -1; | |
1569 | ||
9a7b2b0f | 1570 | if (!refdev) { |
8ed75463 | 1571 | ret = 1; |
9a7b2b0f | 1572 | } else { |
1da177e4 | 1573 | __u64 ev1, ev2; |
65a06f06 | 1574 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1575 | |
1576 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1577 | sb->level != refsb->level || | |
1578 | sb->layout != refsb->layout || | |
1579 | sb->chunksize != refsb->chunksize) { | |
1580 | printk(KERN_WARNING "md: %s has strangely different" | |
1581 | " superblock to %s\n", | |
1582 | bdevname(rdev->bdev,b), | |
1583 | bdevname(refdev->bdev,b2)); | |
1584 | return -EINVAL; | |
1585 | } | |
1586 | ev1 = le64_to_cpu(sb->events); | |
1587 | ev2 = le64_to_cpu(refsb->events); | |
1588 | ||
1589 | if (ev1 > ev2) | |
8ed75463 N |
1590 | ret = 1; |
1591 | else | |
1592 | ret = 0; | |
1da177e4 | 1593 | } |
c6563a8c N |
1594 | if (minor_version) { |
1595 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1596 | sectors -= rdev->data_offset; | |
1597 | } else | |
1598 | sectors = rdev->sb_start; | |
1599 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1600 | return -EINVAL; |
dd8ac336 | 1601 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1602 | return ret; |
1da177e4 LT |
1603 | } |
1604 | ||
fd01b88c | 1605 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1606 | { |
65a06f06 | 1607 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1608 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1609 | |
41158c7e | 1610 | rdev->raid_disk = -1; |
c5d79adb N |
1611 | clear_bit(Faulty, &rdev->flags); |
1612 | clear_bit(In_sync, &rdev->flags); | |
1613 | clear_bit(WriteMostly, &rdev->flags); | |
c5d79adb | 1614 | |
1da177e4 LT |
1615 | if (mddev->raid_disks == 0) { |
1616 | mddev->major_version = 1; | |
1617 | mddev->patch_version = 0; | |
e691063a | 1618 | mddev->external = 0; |
9d8f0363 | 1619 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1620 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1621 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1622 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1623 | mddev->clevel[0] = 0; |
1da177e4 LT |
1624 | mddev->layout = le32_to_cpu(sb->layout); |
1625 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1626 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1627 | mddev->events = ev1; |
c3d9714e | 1628 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1629 | mddev->bitmap_info.space = 0; |
1630 | /* Default location for bitmap is 1K after superblock | |
1631 | * using 3K - total of 4K | |
1632 | */ | |
c3d9714e | 1633 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1634 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1635 | mddev->reshape_backwards = 0; |
1636 | ||
1da177e4 LT |
1637 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1638 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1639 | ||
1640 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1641 | |
71c0805c | 1642 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1643 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1644 | mddev->bitmap_info.offset = |
1645 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1646 | /* Metadata doesn't record how much space is available. |
1647 | * For 1.0, we assume we can use up to the superblock | |
1648 | * if before, else to 4K beyond superblock. | |
1649 | * For others, assume no change is possible. | |
1650 | */ | |
1651 | if (mddev->minor_version > 0) | |
1652 | mddev->bitmap_info.space = 0; | |
1653 | else if (mddev->bitmap_info.offset > 0) | |
1654 | mddev->bitmap_info.space = | |
1655 | 8 - mddev->bitmap_info.offset; | |
1656 | else | |
1657 | mddev->bitmap_info.space = | |
1658 | -mddev->bitmap_info.offset; | |
1659 | } | |
e11e93fa | 1660 | |
f6705578 N |
1661 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1662 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1663 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1664 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1665 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1666 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1667 | if (mddev->delta_disks < 0 || |
1668 | (mddev->delta_disks == 0 && | |
1669 | (le32_to_cpu(sb->feature_map) | |
1670 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1671 | mddev->reshape_backwards = 1; | |
f6705578 N |
1672 | } else { |
1673 | mddev->reshape_position = MaxSector; | |
1674 | mddev->delta_disks = 0; | |
1675 | mddev->new_level = mddev->level; | |
1676 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1677 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1678 | } |
1679 | ||
41158c7e | 1680 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1681 | /* Insist of good event counter while assembling, except for |
1682 | * spares (which don't need an event count) */ | |
1da177e4 | 1683 | ++ev1; |
be6800a7 N |
1684 | if (rdev->desc_nr >= 0 && |
1685 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1686 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1687 | if (ev1 < mddev->events) | |
1688 | return -EINVAL; | |
41158c7e N |
1689 | } else if (mddev->bitmap) { |
1690 | /* If adding to array with a bitmap, then we can accept an | |
1691 | * older device, but not too old. | |
1692 | */ | |
41158c7e N |
1693 | if (ev1 < mddev->bitmap->events_cleared) |
1694 | return 0; | |
07d84d10 N |
1695 | } else { |
1696 | if (ev1 < mddev->events) | |
1697 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1698 | return 0; | |
1699 | } | |
1da177e4 LT |
1700 | if (mddev->level != LEVEL_MULTIPATH) { |
1701 | int role; | |
3673f305 N |
1702 | if (rdev->desc_nr < 0 || |
1703 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1704 | role = 0xffff; | |
1705 | rdev->desc_nr = -1; | |
1706 | } else | |
1707 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1708 | switch(role) { |
1709 | case 0xffff: /* spare */ | |
1da177e4 LT |
1710 | break; |
1711 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1712 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1713 | break; |
1714 | default: | |
5fd6c1dc N |
1715 | if ((le32_to_cpu(sb->feature_map) & |
1716 | MD_FEATURE_RECOVERY_OFFSET)) | |
1717 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1718 | else | |
1719 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1720 | rdev->raid_disk = role; |
1721 | break; | |
1722 | } | |
8ddf9efe N |
1723 | if (sb->devflags & WriteMostly1) |
1724 | set_bit(WriteMostly, &rdev->flags); | |
2d78f8c4 N |
1725 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1726 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1727 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1728 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1729 | |
1da177e4 LT |
1730 | return 0; |
1731 | } | |
1732 | ||
fd01b88c | 1733 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1734 | { |
1735 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1736 | struct md_rdev *rdev2; |
1da177e4 LT |
1737 | int max_dev, i; |
1738 | /* make rdev->sb match mddev and rdev data. */ | |
1739 | ||
65a06f06 | 1740 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1741 | |
1742 | sb->feature_map = 0; | |
1743 | sb->pad0 = 0; | |
5fd6c1dc | 1744 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1745 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1746 | ||
1747 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1748 | sb->events = cpu_to_le64(mddev->events); | |
1749 | if (mddev->in_sync) | |
1750 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1751 | else | |
1752 | sb->resync_offset = cpu_to_le64(0); | |
1753 | ||
1c05b4bc | 1754 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1755 | |
f0ca340c | 1756 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1757 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1758 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1759 | sb->level = cpu_to_le32(mddev->level); |
1760 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1761 | |
aeb9b211 N |
1762 | if (test_bit(WriteMostly, &rdev->flags)) |
1763 | sb->devflags |= WriteMostly1; | |
1764 | else | |
1765 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1766 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1767 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1768 | |
c3d9714e N |
1769 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1770 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1771 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1772 | } |
5fd6c1dc N |
1773 | |
1774 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1775 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1776 | sb->feature_map |= |
1777 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1778 | sb->recovery_offset = | |
1779 | cpu_to_le64(rdev->recovery_offset); | |
5fd6c1dc | 1780 | } |
2d78f8c4 N |
1781 | if (test_bit(Replacement, &rdev->flags)) |
1782 | sb->feature_map |= | |
1783 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1784 | |
f6705578 N |
1785 | if (mddev->reshape_position != MaxSector) { |
1786 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1787 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1788 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1789 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1790 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1791 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1792 | if (mddev->delta_disks == 0 && |
1793 | mddev->reshape_backwards) | |
1794 | sb->feature_map | |
1795 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1796 | if (rdev->new_data_offset != rdev->data_offset) { |
1797 | sb->feature_map | |
1798 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1799 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1800 | - rdev->data_offset)); | |
1801 | } | |
f6705578 | 1802 | } |
a654b9d8 | 1803 | |
2699b672 N |
1804 | if (rdev->badblocks.count == 0) |
1805 | /* Nothing to do for bad blocks*/ ; | |
1806 | else if (sb->bblog_offset == 0) | |
1807 | /* Cannot record bad blocks on this device */ | |
1808 | md_error(mddev, rdev); | |
1809 | else { | |
1810 | struct badblocks *bb = &rdev->badblocks; | |
1811 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1812 | u64 *p = bb->page; | |
1813 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1814 | if (bb->changed) { | |
1815 | unsigned seq; | |
1816 | ||
1817 | retry: | |
1818 | seq = read_seqbegin(&bb->lock); | |
1819 | ||
1820 | memset(bbp, 0xff, PAGE_SIZE); | |
1821 | ||
1822 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1823 | u64 internal_bb = p[i]; |
2699b672 N |
1824 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1825 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1826 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1827 | } |
d0962936 | 1828 | bb->changed = 0; |
2699b672 N |
1829 | if (read_seqretry(&bb->lock, seq)) |
1830 | goto retry; | |
1831 | ||
1832 | bb->sector = (rdev->sb_start + | |
1833 | (int)le32_to_cpu(sb->bblog_offset)); | |
1834 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1835 | } |
1836 | } | |
1837 | ||
1da177e4 | 1838 | max_dev = 0; |
dafb20fa | 1839 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1840 | if (rdev2->desc_nr+1 > max_dev) |
1841 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1842 | |
70471daf N |
1843 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1844 | int bmask; | |
a778b73f | 1845 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1846 | rdev->sb_size = max_dev * 2 + 256; |
1847 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1848 | if (rdev->sb_size & bmask) | |
1849 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1850 | } else |
1851 | max_dev = le32_to_cpu(sb->max_dev); | |
1852 | ||
1da177e4 LT |
1853 | for (i=0; i<max_dev;i++) |
1854 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1855 | ||
dafb20fa | 1856 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1857 | i = rdev2->desc_nr; |
b2d444d7 | 1858 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1859 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1860 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1861 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1862 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1863 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1864 | else |
1865 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1866 | } | |
1867 | ||
1da177e4 LT |
1868 | sb->sb_csum = calc_sb_1_csum(sb); |
1869 | } | |
1870 | ||
0cd17fec | 1871 | static unsigned long long |
3cb03002 | 1872 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1873 | { |
1874 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1875 | sector_t max_sectors; |
58c0fed4 | 1876 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1877 | return 0; /* component must fit device */ |
c6563a8c N |
1878 | if (rdev->data_offset != rdev->new_data_offset) |
1879 | return 0; /* too confusing */ | |
0f420358 | 1880 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1881 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1882 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1883 | max_sectors -= rdev->data_offset; |
1884 | if (!num_sectors || num_sectors > max_sectors) | |
1885 | num_sectors = max_sectors; | |
c3d9714e | 1886 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1887 | /* minor version 0 with bitmap we can't move */ |
1888 | return 0; | |
1889 | } else { | |
1890 | /* minor version 0; superblock after data */ | |
0f420358 | 1891 | sector_t sb_start; |
77304d2a | 1892 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1893 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1894 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1895 | if (!num_sectors || num_sectors > max_sectors) |
1896 | num_sectors = max_sectors; | |
0f420358 | 1897 | rdev->sb_start = sb_start; |
0cd17fec | 1898 | } |
65a06f06 | 1899 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1900 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1901 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1902 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1903 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1904 | rdev->sb_page); |
1905 | md_super_wait(rdev->mddev); | |
c26a44ed | 1906 | return num_sectors; |
c6563a8c N |
1907 | |
1908 | } | |
1909 | ||
1910 | static int | |
1911 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1912 | unsigned long long new_offset) | |
1913 | { | |
1914 | /* All necessary checks on new >= old have been done */ | |
1915 | struct bitmap *bitmap; | |
1916 | if (new_offset >= rdev->data_offset) | |
1917 | return 1; | |
1918 | ||
1919 | /* with 1.0 metadata, there is no metadata to tread on | |
1920 | * so we can always move back */ | |
1921 | if (rdev->mddev->minor_version == 0) | |
1922 | return 1; | |
1923 | ||
1924 | /* otherwise we must be sure not to step on | |
1925 | * any metadata, so stay: | |
1926 | * 36K beyond start of superblock | |
1927 | * beyond end of badblocks | |
1928 | * beyond write-intent bitmap | |
1929 | */ | |
1930 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1931 | return 0; | |
1932 | bitmap = rdev->mddev->bitmap; | |
1933 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1934 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1935 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1936 | return 0; |
1937 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1938 | return 0; | |
1939 | ||
1940 | return 1; | |
0cd17fec | 1941 | } |
1da177e4 | 1942 | |
75c96f85 | 1943 | static struct super_type super_types[] = { |
1da177e4 LT |
1944 | [0] = { |
1945 | .name = "0.90.0", | |
1946 | .owner = THIS_MODULE, | |
0cd17fec CW |
1947 | .load_super = super_90_load, |
1948 | .validate_super = super_90_validate, | |
1949 | .sync_super = super_90_sync, | |
1950 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1951 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1952 | }, |
1953 | [1] = { | |
1954 | .name = "md-1", | |
1955 | .owner = THIS_MODULE, | |
0cd17fec CW |
1956 | .load_super = super_1_load, |
1957 | .validate_super = super_1_validate, | |
1958 | .sync_super = super_1_sync, | |
1959 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1960 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1961 | }, |
1962 | }; | |
1da177e4 | 1963 | |
fd01b88c | 1964 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1965 | { |
1966 | if (mddev->sync_super) { | |
1967 | mddev->sync_super(mddev, rdev); | |
1968 | return; | |
1969 | } | |
1970 | ||
1971 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1972 | ||
1973 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1974 | } | |
1975 | ||
fd01b88c | 1976 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1977 | { |
3cb03002 | 1978 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1979 | |
4b80991c N |
1980 | rcu_read_lock(); |
1981 | rdev_for_each_rcu(rdev, mddev1) | |
1982 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1983 | if (rdev->bdev->bd_contains == |
4b80991c N |
1984 | rdev2->bdev->bd_contains) { |
1985 | rcu_read_unlock(); | |
7dd5e7c3 | 1986 | return 1; |
4b80991c N |
1987 | } |
1988 | rcu_read_unlock(); | |
1da177e4 LT |
1989 | return 0; |
1990 | } | |
1991 | ||
1992 | static LIST_HEAD(pending_raid_disks); | |
1993 | ||
ac5e7113 AN |
1994 | /* |
1995 | * Try to register data integrity profile for an mddev | |
1996 | * | |
1997 | * This is called when an array is started and after a disk has been kicked | |
1998 | * from the array. It only succeeds if all working and active component devices | |
1999 | * are integrity capable with matching profiles. | |
2000 | */ | |
fd01b88c | 2001 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 2002 | { |
3cb03002 | 2003 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
2004 | |
2005 | if (list_empty(&mddev->disks)) | |
2006 | return 0; /* nothing to do */ | |
629acb6a JB |
2007 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
2008 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 2009 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
2010 | /* skip spares and non-functional disks */ |
2011 | if (test_bit(Faulty, &rdev->flags)) | |
2012 | continue; | |
2013 | if (rdev->raid_disk < 0) | |
2014 | continue; | |
ac5e7113 AN |
2015 | if (!reference) { |
2016 | /* Use the first rdev as the reference */ | |
2017 | reference = rdev; | |
2018 | continue; | |
2019 | } | |
2020 | /* does this rdev's profile match the reference profile? */ | |
2021 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
2022 | rdev->bdev->bd_disk) < 0) | |
2023 | return -EINVAL; | |
2024 | } | |
89078d57 MP |
2025 | if (!reference || !bdev_get_integrity(reference->bdev)) |
2026 | return 0; | |
ac5e7113 AN |
2027 | /* |
2028 | * All component devices are integrity capable and have matching | |
2029 | * profiles, register the common profile for the md device. | |
2030 | */ | |
2031 | if (blk_integrity_register(mddev->gendisk, | |
2032 | bdev_get_integrity(reference->bdev)) != 0) { | |
2033 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
2034 | mdname(mddev)); | |
2035 | return -EINVAL; | |
2036 | } | |
a91a2785 MP |
2037 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
2038 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
2039 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
2040 | mdname(mddev)); | |
2041 | return -EINVAL; | |
2042 | } | |
ac5e7113 AN |
2043 | return 0; |
2044 | } | |
2045 | EXPORT_SYMBOL(md_integrity_register); | |
2046 | ||
2047 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
fd01b88c | 2048 | void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) |
3f9d99c1 | 2049 | { |
2863b9eb JB |
2050 | struct blk_integrity *bi_rdev; |
2051 | struct blk_integrity *bi_mddev; | |
2052 | ||
2053 | if (!mddev->gendisk) | |
2054 | return; | |
2055 | ||
2056 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2057 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2058 | |
ac5e7113 | 2059 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 2060 | return; |
ac5e7113 | 2061 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 2062 | return; |
ac5e7113 AN |
2063 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
2064 | rdev->bdev->bd_disk) >= 0) | |
2065 | return; | |
2066 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
2067 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 2068 | } |
ac5e7113 | 2069 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2070 | |
fd01b88c | 2071 | static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev) |
1da177e4 | 2072 | { |
7dd5e7c3 | 2073 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2074 | struct kobject *ko; |
1edf80d3 | 2075 | char *s; |
5e55e2f5 | 2076 | int err; |
1da177e4 LT |
2077 | |
2078 | if (rdev->mddev) { | |
2079 | MD_BUG(); | |
2080 | return -EINVAL; | |
2081 | } | |
11e2ede0 DW |
2082 | |
2083 | /* prevent duplicates */ | |
2084 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2085 | return -EEXIST; | |
2086 | ||
dd8ac336 AN |
2087 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
2088 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
2089 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2090 | if (mddev->pers) { |
2091 | /* Cannot change size, so fail | |
2092 | * If mddev->level <= 0, then we don't care | |
2093 | * about aligning sizes (e.g. linear) | |
2094 | */ | |
2095 | if (mddev->level > 0) | |
2096 | return -ENOSPC; | |
2097 | } else | |
dd8ac336 | 2098 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2099 | } |
1da177e4 LT |
2100 | |
2101 | /* Verify rdev->desc_nr is unique. | |
2102 | * If it is -1, assign a free number, else | |
2103 | * check number is not in use | |
2104 | */ | |
2105 | if (rdev->desc_nr < 0) { | |
2106 | int choice = 0; | |
2107 | if (mddev->pers) choice = mddev->raid_disks; | |
2108 | while (find_rdev_nr(mddev, choice)) | |
2109 | choice++; | |
2110 | rdev->desc_nr = choice; | |
2111 | } else { | |
2112 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
2113 | return -EBUSY; | |
2114 | } | |
de01dfad N |
2115 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
2116 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
2117 | mdname(mddev), mddev->max_disks); | |
2118 | return -EBUSY; | |
2119 | } | |
19133a42 | 2120 | bdevname(rdev->bdev,b); |
649316b2 | 2121 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 2122 | *s = '!'; |
649316b2 | 2123 | |
1da177e4 | 2124 | rdev->mddev = mddev; |
19133a42 | 2125 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 2126 | |
b2d6db58 | 2127 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2128 | goto fail; |
86e6ffdd | 2129 | |
0762b8bd | 2130 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2131 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2132 | /* failure here is OK */; | |
2133 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2134 | |
4b80991c | 2135 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2136 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2137 | |
2138 | /* May as well allow recovery to be retried once */ | |
5389042f | 2139 | mddev->recovery_disabled++; |
3f9d99c1 | 2140 | |
1da177e4 | 2141 | return 0; |
5e55e2f5 N |
2142 | |
2143 | fail: | |
2144 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
2145 | b, mdname(mddev)); | |
2146 | return err; | |
1da177e4 LT |
2147 | } |
2148 | ||
177a99b2 | 2149 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2150 | { |
3cb03002 | 2151 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2152 | kobject_del(&rdev->kobj); |
177a99b2 | 2153 | kobject_put(&rdev->kobj); |
5792a285 N |
2154 | } |
2155 | ||
3cb03002 | 2156 | static void unbind_rdev_from_array(struct md_rdev * rdev) |
1da177e4 LT |
2157 | { |
2158 | char b[BDEVNAME_SIZE]; | |
2159 | if (!rdev->mddev) { | |
2160 | MD_BUG(); | |
2161 | return; | |
2162 | } | |
49731baa | 2163 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2164 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
2165 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
2166 | rdev->mddev = NULL; | |
86e6ffdd | 2167 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2168 | sysfs_put(rdev->sysfs_state); |
2169 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2170 | rdev->badblocks.count = 0; |
5792a285 | 2171 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2172 | * writing to 'remove' to "dev/state". We also need |
2173 | * to delay it due to rcu usage. | |
5792a285 | 2174 | */ |
4b80991c | 2175 | synchronize_rcu(); |
177a99b2 N |
2176 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2177 | kobject_get(&rdev->kobj); | |
e804ac78 | 2178 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2179 | } |
2180 | ||
2181 | /* | |
2182 | * prevent the device from being mounted, repartitioned or | |
2183 | * otherwise reused by a RAID array (or any other kernel | |
2184 | * subsystem), by bd_claiming the device. | |
2185 | */ | |
3cb03002 | 2186 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2187 | { |
2188 | int err = 0; | |
2189 | struct block_device *bdev; | |
2190 | char b[BDEVNAME_SIZE]; | |
2191 | ||
d4d77629 | 2192 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2193 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 LT |
2194 | if (IS_ERR(bdev)) { |
2195 | printk(KERN_ERR "md: could not open %s.\n", | |
2196 | __bdevname(dev, b)); | |
2197 | return PTR_ERR(bdev); | |
2198 | } | |
1da177e4 LT |
2199 | rdev->bdev = bdev; |
2200 | return err; | |
2201 | } | |
2202 | ||
3cb03002 | 2203 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2204 | { |
2205 | struct block_device *bdev = rdev->bdev; | |
2206 | rdev->bdev = NULL; | |
2207 | if (!bdev) | |
2208 | MD_BUG(); | |
e525fd89 | 2209 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2210 | } |
2211 | ||
2212 | void md_autodetect_dev(dev_t dev); | |
2213 | ||
3cb03002 | 2214 | static void export_rdev(struct md_rdev * rdev) |
1da177e4 LT |
2215 | { |
2216 | char b[BDEVNAME_SIZE]; | |
2217 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
2218 | bdevname(rdev->bdev,b)); | |
2219 | if (rdev->mddev) | |
2220 | MD_BUG(); | |
545c8795 | 2221 | md_rdev_clear(rdev); |
1da177e4 | 2222 | #ifndef MODULE |
d0fae18f N |
2223 | if (test_bit(AutoDetected, &rdev->flags)) |
2224 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2225 | #endif |
2226 | unlock_rdev(rdev); | |
86e6ffdd | 2227 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2228 | } |
2229 | ||
3cb03002 | 2230 | static void kick_rdev_from_array(struct md_rdev * rdev) |
1da177e4 LT |
2231 | { |
2232 | unbind_rdev_from_array(rdev); | |
2233 | export_rdev(rdev); | |
2234 | } | |
2235 | ||
fd01b88c | 2236 | static void export_array(struct mddev *mddev) |
1da177e4 | 2237 | { |
3cb03002 | 2238 | struct md_rdev *rdev, *tmp; |
1da177e4 | 2239 | |
dafb20fa | 2240 | rdev_for_each_safe(rdev, tmp, mddev) { |
1da177e4 LT |
2241 | if (!rdev->mddev) { |
2242 | MD_BUG(); | |
2243 | continue; | |
2244 | } | |
2245 | kick_rdev_from_array(rdev); | |
2246 | } | |
2247 | if (!list_empty(&mddev->disks)) | |
2248 | MD_BUG(); | |
2249 | mddev->raid_disks = 0; | |
2250 | mddev->major_version = 0; | |
2251 | } | |
2252 | ||
2253 | static void print_desc(mdp_disk_t *desc) | |
2254 | { | |
2255 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
2256 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
2257 | } | |
2258 | ||
cd2ac932 | 2259 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
2260 | { |
2261 | int i; | |
2262 | ||
2263 | printk(KERN_INFO | |
2264 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
2265 | sb->major_version, sb->minor_version, sb->patch_version, | |
2266 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
2267 | sb->ctime); | |
2268 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
2269 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
2270 | sb->md_minor, sb->layout, sb->chunk_size); | |
2271 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
2272 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
2273 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
2274 | sb->failed_disks, sb->spare_disks, | |
2275 | sb->sb_csum, (unsigned long)sb->events_lo); | |
2276 | ||
2277 | printk(KERN_INFO); | |
2278 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2279 | mdp_disk_t *desc; | |
2280 | ||
2281 | desc = sb->disks + i; | |
2282 | if (desc->number || desc->major || desc->minor || | |
2283 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
2284 | printk(" D %2d: ", i); | |
2285 | print_desc(desc); | |
2286 | } | |
2287 | } | |
2288 | printk(KERN_INFO "md: THIS: "); | |
2289 | print_desc(&sb->this_disk); | |
cd2ac932 | 2290 | } |
1da177e4 | 2291 | |
cd2ac932 CR |
2292 | static void print_sb_1(struct mdp_superblock_1 *sb) |
2293 | { | |
2294 | __u8 *uuid; | |
2295 | ||
2296 | uuid = sb->set_uuid; | |
ad361c98 | 2297 | printk(KERN_INFO |
7b75c2f8 | 2298 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 2299 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
2300 | le32_to_cpu(sb->major_version), |
2301 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 2302 | uuid, |
cd2ac932 CR |
2303 | sb->set_name, |
2304 | (unsigned long long)le64_to_cpu(sb->ctime) | |
2305 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
2306 | ||
2307 | uuid = sb->device_uuid; | |
ad361c98 JP |
2308 | printk(KERN_INFO |
2309 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 2310 | " RO:%llu\n" |
7b75c2f8 | 2311 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
2312 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
2313 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
2314 | le32_to_cpu(sb->level), |
2315 | (unsigned long long)le64_to_cpu(sb->size), | |
2316 | le32_to_cpu(sb->raid_disks), | |
2317 | le32_to_cpu(sb->layout), | |
2318 | le32_to_cpu(sb->chunksize), | |
2319 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2320 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2321 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2322 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2323 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2324 | uuid, |
cd2ac932 CR |
2325 | sb->devflags, |
2326 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2327 | (unsigned long long)le64_to_cpu(sb->events), | |
2328 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2329 | le32_to_cpu(sb->sb_csum), | |
2330 | le32_to_cpu(sb->max_dev) | |
2331 | ); | |
1da177e4 LT |
2332 | } |
2333 | ||
3cb03002 | 2334 | static void print_rdev(struct md_rdev *rdev, int major_version) |
1da177e4 LT |
2335 | { |
2336 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2337 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2338 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2339 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2340 | rdev->desc_nr); | |
1da177e4 | 2341 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2342 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2343 | switch (major_version) { | |
2344 | case 0: | |
65a06f06 | 2345 | print_sb_90(page_address(rdev->sb_page)); |
cd2ac932 CR |
2346 | break; |
2347 | case 1: | |
65a06f06 | 2348 | print_sb_1(page_address(rdev->sb_page)); |
cd2ac932 CR |
2349 | break; |
2350 | } | |
1da177e4 LT |
2351 | } else |
2352 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2353 | } | |
2354 | ||
5e56341d | 2355 | static void md_print_devices(void) |
1da177e4 | 2356 | { |
159ec1fc | 2357 | struct list_head *tmp; |
3cb03002 | 2358 | struct md_rdev *rdev; |
fd01b88c | 2359 | struct mddev *mddev; |
1da177e4 LT |
2360 | char b[BDEVNAME_SIZE]; |
2361 | ||
2362 | printk("\n"); | |
2363 | printk("md: **********************************\n"); | |
2364 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2365 | printk("md: **********************************\n"); | |
29ac4aa3 | 2366 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2367 | |
32a7627c N |
2368 | if (mddev->bitmap) |
2369 | bitmap_print_sb(mddev->bitmap); | |
2370 | else | |
2371 | printk("%s: ", mdname(mddev)); | |
dafb20fa | 2372 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
2373 | printk("<%s>", bdevname(rdev->bdev,b)); |
2374 | printk("\n"); | |
2375 | ||
dafb20fa | 2376 | rdev_for_each(rdev, mddev) |
cd2ac932 | 2377 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2378 | } |
2379 | printk("md: **********************************\n"); | |
2380 | printk("\n"); | |
2381 | } | |
2382 | ||
2383 | ||
fd01b88c | 2384 | static void sync_sbs(struct mddev * mddev, int nospares) |
1da177e4 | 2385 | { |
42543769 N |
2386 | /* Update each superblock (in-memory image), but |
2387 | * if we are allowed to, skip spares which already | |
2388 | * have the right event counter, or have one earlier | |
2389 | * (which would mean they aren't being marked as dirty | |
2390 | * with the rest of the array) | |
2391 | */ | |
3cb03002 | 2392 | struct md_rdev *rdev; |
dafb20fa | 2393 | rdev_for_each(rdev, mddev) { |
42543769 N |
2394 | if (rdev->sb_events == mddev->events || |
2395 | (nospares && | |
2396 | rdev->raid_disk < 0 && | |
42543769 N |
2397 | rdev->sb_events+1 == mddev->events)) { |
2398 | /* Don't update this superblock */ | |
2399 | rdev->sb_loaded = 2; | |
2400 | } else { | |
076f968b | 2401 | sync_super(mddev, rdev); |
42543769 N |
2402 | rdev->sb_loaded = 1; |
2403 | } | |
1da177e4 LT |
2404 | } |
2405 | } | |
2406 | ||
fd01b88c | 2407 | static void md_update_sb(struct mddev * mddev, int force_change) |
1da177e4 | 2408 | { |
3cb03002 | 2409 | struct md_rdev *rdev; |
06d91a5f | 2410 | int sync_req; |
42543769 | 2411 | int nospares = 0; |
2699b672 | 2412 | int any_badblocks_changed = 0; |
1da177e4 | 2413 | |
1da177e4 | 2414 | repeat: |
3a3a5ddb | 2415 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2416 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2417 | if (rdev->raid_disk >= 0 && |
2418 | mddev->delta_disks >= 0 && | |
2419 | !test_bit(In_sync, &rdev->flags) && | |
2420 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2421 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2422 | ||
2423 | } | |
bd52b746 | 2424 | if (!mddev->persistent) { |
070dc6dd | 2425 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2426 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
de393cde | 2427 | if (!mddev->external) { |
d97a41dc | 2428 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
dafb20fa | 2429 | rdev_for_each(rdev, mddev) { |
de393cde | 2430 | if (rdev->badblocks.changed) { |
d0962936 | 2431 | rdev->badblocks.changed = 0; |
de393cde N |
2432 | md_ack_all_badblocks(&rdev->badblocks); |
2433 | md_error(mddev, rdev); | |
2434 | } | |
2435 | clear_bit(Blocked, &rdev->flags); | |
2436 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2437 | wake_up(&rdev->blocked_wait); | |
2438 | } | |
2439 | } | |
3a3a5ddb N |
2440 | wake_up(&mddev->sb_wait); |
2441 | return; | |
2442 | } | |
2443 | ||
a9701a30 | 2444 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2445 | |
3a3a5ddb N |
2446 | mddev->utime = get_seconds(); |
2447 | ||
850b2b42 N |
2448 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2449 | force_change = 1; | |
2450 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2451 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2452 | * though if events isn't the right even/odd, we will have to do | |
2453 | * spares after all | |
2454 | */ | |
2455 | nospares = 1; | |
2456 | if (force_change) | |
2457 | nospares = 0; | |
2458 | if (mddev->degraded) | |
84692195 N |
2459 | /* If the array is degraded, then skipping spares is both |
2460 | * dangerous and fairly pointless. | |
2461 | * Dangerous because a device that was removed from the array | |
2462 | * might have a event_count that still looks up-to-date, | |
2463 | * so it can be re-added without a resync. | |
2464 | * Pointless because if there are any spares to skip, | |
2465 | * then a recovery will happen and soon that array won't | |
2466 | * be degraded any more and the spare can go back to sleep then. | |
2467 | */ | |
850b2b42 | 2468 | nospares = 0; |
84692195 | 2469 | |
06d91a5f | 2470 | sync_req = mddev->in_sync; |
42543769 N |
2471 | |
2472 | /* If this is just a dirty<->clean transition, and the array is clean | |
2473 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2474 | if (nospares |
42543769 | 2475 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2476 | && mddev->can_decrease_events |
2477 | && mddev->events != 1) { | |
42543769 | 2478 | mddev->events--; |
a8707c08 N |
2479 | mddev->can_decrease_events = 0; |
2480 | } else { | |
42543769 N |
2481 | /* otherwise we have to go forward and ... */ |
2482 | mddev->events ++; | |
a8707c08 | 2483 | mddev->can_decrease_events = nospares; |
42543769 | 2484 | } |
1da177e4 LT |
2485 | |
2486 | if (!mddev->events) { | |
2487 | /* | |
2488 | * oops, this 64-bit counter should never wrap. | |
2489 | * Either we are in around ~1 trillion A.C., assuming | |
2490 | * 1 reboot per second, or we have a bug: | |
2491 | */ | |
2492 | MD_BUG(); | |
2493 | mddev->events --; | |
2494 | } | |
2699b672 | 2495 | |
dafb20fa | 2496 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2497 | if (rdev->badblocks.changed) |
2498 | any_badblocks_changed++; | |
de393cde N |
2499 | if (test_bit(Faulty, &rdev->flags)) |
2500 | set_bit(FaultRecorded, &rdev->flags); | |
2501 | } | |
2699b672 | 2502 | |
e691063a | 2503 | sync_sbs(mddev, nospares); |
a9701a30 | 2504 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 | 2505 | |
36a4e1fe N |
2506 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2507 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2508 | |
4ad13663 | 2509 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2510 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2511 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2512 | |
42543769 N |
2513 | if (rdev->sb_loaded != 1) |
2514 | continue; /* no noise on spare devices */ | |
1da177e4 | 2515 | |
d70ed2e4 AW |
2516 | if (!test_bit(Faulty, &rdev->flags) && |
2517 | rdev->saved_raid_disk == -1) { | |
7bfa19f2 | 2518 | md_super_write(mddev,rdev, |
0f420358 | 2519 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2520 | rdev->sb_page); |
36a4e1fe N |
2521 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2522 | bdevname(rdev->bdev, b), | |
2523 | (unsigned long long)rdev->sb_start); | |
42543769 | 2524 | rdev->sb_events = mddev->events; |
2699b672 N |
2525 | if (rdev->badblocks.size) { |
2526 | md_super_write(mddev, rdev, | |
2527 | rdev->badblocks.sector, | |
2528 | rdev->badblocks.size << 9, | |
2529 | rdev->bb_page); | |
2530 | rdev->badblocks.size = 0; | |
2531 | } | |
7bfa19f2 | 2532 | |
d70ed2e4 | 2533 | } else if (test_bit(Faulty, &rdev->flags)) |
36a4e1fe N |
2534 | pr_debug("md: %s (skipping faulty)\n", |
2535 | bdevname(rdev->bdev, b)); | |
d70ed2e4 AW |
2536 | else |
2537 | pr_debug("(skipping incremental s/r "); | |
2538 | ||
7bfa19f2 | 2539 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2540 | /* only need to write one superblock... */ |
2541 | break; | |
2542 | } | |
a9701a30 | 2543 | md_super_wait(mddev); |
850b2b42 | 2544 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2545 | |
a9701a30 | 2546 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2547 | if (mddev->in_sync != sync_req || |
2548 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2549 | /* have to write it out again */ |
a9701a30 | 2550 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2551 | goto repeat; |
2552 | } | |
850b2b42 | 2553 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2554 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2555 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2556 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2557 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2558 | |
dafb20fa | 2559 | rdev_for_each(rdev, mddev) { |
de393cde N |
2560 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2561 | clear_bit(Blocked, &rdev->flags); | |
2562 | ||
2563 | if (any_badblocks_changed) | |
2699b672 | 2564 | md_ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2565 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2566 | wake_up(&rdev->blocked_wait); | |
2567 | } | |
1da177e4 LT |
2568 | } |
2569 | ||
7f6ce769 | 2570 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2571 | * We want to accept with case. For this we use cmd_match. |
2572 | */ | |
2573 | static int cmd_match(const char *cmd, const char *str) | |
2574 | { | |
2575 | /* See if cmd, written into a sysfs file, matches | |
2576 | * str. They must either be the same, or cmd can | |
2577 | * have a trailing newline | |
2578 | */ | |
2579 | while (*cmd && *str && *cmd == *str) { | |
2580 | cmd++; | |
2581 | str++; | |
2582 | } | |
2583 | if (*cmd == '\n') | |
2584 | cmd++; | |
2585 | if (*str || *cmd) | |
2586 | return 0; | |
2587 | return 1; | |
2588 | } | |
2589 | ||
86e6ffdd N |
2590 | struct rdev_sysfs_entry { |
2591 | struct attribute attr; | |
3cb03002 N |
2592 | ssize_t (*show)(struct md_rdev *, char *); |
2593 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2594 | }; |
2595 | ||
2596 | static ssize_t | |
3cb03002 | 2597 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd N |
2598 | { |
2599 | char *sep = ""; | |
20a49ff6 | 2600 | size_t len = 0; |
86e6ffdd | 2601 | |
de393cde N |
2602 | if (test_bit(Faulty, &rdev->flags) || |
2603 | rdev->badblocks.unacked_exist) { | |
86e6ffdd N |
2604 | len+= sprintf(page+len, "%sfaulty",sep); |
2605 | sep = ","; | |
2606 | } | |
b2d444d7 | 2607 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2608 | len += sprintf(page+len, "%sin_sync",sep); |
2609 | sep = ","; | |
2610 | } | |
f655675b N |
2611 | if (test_bit(WriteMostly, &rdev->flags)) { |
2612 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2613 | sep = ","; | |
2614 | } | |
de393cde | 2615 | if (test_bit(Blocked, &rdev->flags) || |
52c64152 N |
2616 | (rdev->badblocks.unacked_exist |
2617 | && !test_bit(Faulty, &rdev->flags))) { | |
6bfe0b49 DW |
2618 | len += sprintf(page+len, "%sblocked", sep); |
2619 | sep = ","; | |
2620 | } | |
b2d444d7 N |
2621 | if (!test_bit(Faulty, &rdev->flags) && |
2622 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2623 | len += sprintf(page+len, "%sspare", sep); |
2624 | sep = ","; | |
2625 | } | |
d7a9d443 N |
2626 | if (test_bit(WriteErrorSeen, &rdev->flags)) { |
2627 | len += sprintf(page+len, "%swrite_error", sep); | |
2628 | sep = ","; | |
2629 | } | |
2d78f8c4 N |
2630 | if (test_bit(WantReplacement, &rdev->flags)) { |
2631 | len += sprintf(page+len, "%swant_replacement", sep); | |
2632 | sep = ","; | |
2633 | } | |
2634 | if (test_bit(Replacement, &rdev->flags)) { | |
2635 | len += sprintf(page+len, "%sreplacement", sep); | |
2636 | sep = ","; | |
2637 | } | |
2638 | ||
86e6ffdd N |
2639 | return len+sprintf(page+len, "\n"); |
2640 | } | |
2641 | ||
45dc2de1 | 2642 | static ssize_t |
3cb03002 | 2643 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2644 | { |
2645 | /* can write | |
de393cde | 2646 | * faulty - simulates an error |
45dc2de1 | 2647 | * remove - disconnects the device |
f655675b N |
2648 | * writemostly - sets write_mostly |
2649 | * -writemostly - clears write_mostly | |
de393cde N |
2650 | * blocked - sets the Blocked flags |
2651 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2652 | * insync - sets Insync providing device isn't active |
d7a9d443 N |
2653 | * write_error - sets WriteErrorSeen |
2654 | * -write_error - clears WriteErrorSeen | |
45dc2de1 N |
2655 | */ |
2656 | int err = -EINVAL; | |
2657 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2658 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2659 | if (test_bit(Faulty, &rdev->flags)) |
2660 | err = 0; | |
2661 | else | |
2662 | err = -EBUSY; | |
45dc2de1 N |
2663 | } else if (cmd_match(buf, "remove")) { |
2664 | if (rdev->raid_disk >= 0) | |
2665 | err = -EBUSY; | |
2666 | else { | |
fd01b88c | 2667 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2668 | kick_rdev_from_array(rdev); |
3f9d7b0d N |
2669 | if (mddev->pers) |
2670 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2671 | md_new_event(mddev); |
2672 | err = 0; | |
2673 | } | |
f655675b N |
2674 | } else if (cmd_match(buf, "writemostly")) { |
2675 | set_bit(WriteMostly, &rdev->flags); | |
2676 | err = 0; | |
2677 | } else if (cmd_match(buf, "-writemostly")) { | |
2678 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2679 | err = 0; |
2680 | } else if (cmd_match(buf, "blocked")) { | |
2681 | set_bit(Blocked, &rdev->flags); | |
2682 | err = 0; | |
2683 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2684 | if (!test_bit(Faulty, &rdev->flags) && |
7da64a0a | 2685 | rdev->badblocks.unacked_exist) { |
de393cde N |
2686 | /* metadata handler doesn't understand badblocks, |
2687 | * so we need to fail the device | |
2688 | */ | |
2689 | md_error(rdev->mddev, rdev); | |
2690 | } | |
6bfe0b49 | 2691 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2692 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2693 | wake_up(&rdev->blocked_wait); |
2694 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2695 | md_wakeup_thread(rdev->mddev->thread); | |
2696 | ||
6d56e278 N |
2697 | err = 0; |
2698 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2699 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2700 | err = 0; |
d7a9d443 N |
2701 | } else if (cmd_match(buf, "write_error")) { |
2702 | set_bit(WriteErrorSeen, &rdev->flags); | |
2703 | err = 0; | |
2704 | } else if (cmd_match(buf, "-write_error")) { | |
2705 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2706 | err = 0; | |
2d78f8c4 N |
2707 | } else if (cmd_match(buf, "want_replacement")) { |
2708 | /* Any non-spare device that is not a replacement can | |
2709 | * become want_replacement at any time, but we then need to | |
2710 | * check if recovery is needed. | |
2711 | */ | |
2712 | if (rdev->raid_disk >= 0 && | |
2713 | !test_bit(Replacement, &rdev->flags)) | |
2714 | set_bit(WantReplacement, &rdev->flags); | |
2715 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2716 | md_wakeup_thread(rdev->mddev->thread); | |
2717 | err = 0; | |
2718 | } else if (cmd_match(buf, "-want_replacement")) { | |
2719 | /* Clearing 'want_replacement' is always allowed. | |
2720 | * Once replacements starts it is too late though. | |
2721 | */ | |
2722 | err = 0; | |
2723 | clear_bit(WantReplacement, &rdev->flags); | |
2724 | } else if (cmd_match(buf, "replacement")) { | |
2725 | /* Can only set a device as a replacement when array has not | |
2726 | * yet been started. Once running, replacement is automatic | |
2727 | * from spares, or by assigning 'slot'. | |
2728 | */ | |
2729 | if (rdev->mddev->pers) | |
2730 | err = -EBUSY; | |
2731 | else { | |
2732 | set_bit(Replacement, &rdev->flags); | |
2733 | err = 0; | |
2734 | } | |
2735 | } else if (cmd_match(buf, "-replacement")) { | |
2736 | /* Similarly, can only clear Replacement before start */ | |
2737 | if (rdev->mddev->pers) | |
2738 | err = -EBUSY; | |
2739 | else { | |
2740 | clear_bit(Replacement, &rdev->flags); | |
2741 | err = 0; | |
2742 | } | |
45dc2de1 | 2743 | } |
00bcb4ac N |
2744 | if (!err) |
2745 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2746 | return err ? err : len; |
2747 | } | |
80ca3a44 N |
2748 | static struct rdev_sysfs_entry rdev_state = |
2749 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2750 | |
4dbcdc75 | 2751 | static ssize_t |
3cb03002 | 2752 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2753 | { |
2754 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2755 | } | |
2756 | ||
2757 | static ssize_t | |
3cb03002 | 2758 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 N |
2759 | { |
2760 | char *e; | |
2761 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2762 | if (*buf && (*e == 0 || *e == '\n')) { | |
2763 | atomic_set(&rdev->corrected_errors, n); | |
2764 | return len; | |
2765 | } | |
2766 | return -EINVAL; | |
2767 | } | |
2768 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2769 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2770 | |
014236d2 | 2771 | static ssize_t |
3cb03002 | 2772 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 N |
2773 | { |
2774 | if (rdev->raid_disk < 0) | |
2775 | return sprintf(page, "none\n"); | |
2776 | else | |
2777 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2778 | } | |
2779 | ||
2780 | static ssize_t | |
3cb03002 | 2781 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 N |
2782 | { |
2783 | char *e; | |
c303da6d | 2784 | int err; |
014236d2 N |
2785 | int slot = simple_strtoul(buf, &e, 10); |
2786 | if (strncmp(buf, "none", 4)==0) | |
2787 | slot = -1; | |
2788 | else if (e==buf || (*e && *e!= '\n')) | |
2789 | return -EINVAL; | |
6c2fce2e | 2790 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2791 | /* Setting 'slot' on an active array requires also |
2792 | * updating the 'rd%d' link, and communicating | |
2793 | * with the personality with ->hot_*_disk. | |
2794 | * For now we only support removing | |
2795 | * failed/spare devices. This normally happens automatically, | |
2796 | * but not when the metadata is externally managed. | |
2797 | */ | |
c303da6d N |
2798 | if (rdev->raid_disk == -1) |
2799 | return -EEXIST; | |
2800 | /* personality does all needed checks */ | |
01393f3d | 2801 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d N |
2802 | return -EINVAL; |
2803 | err = rdev->mddev->pers-> | |
b8321b68 | 2804 | hot_remove_disk(rdev->mddev, rdev); |
c303da6d N |
2805 | if (err) |
2806 | return err; | |
36fad858 | 2807 | sysfs_unlink_rdev(rdev->mddev, rdev); |
b7103107 | 2808 | rdev->raid_disk = -1; |
c303da6d N |
2809 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2810 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2811 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2812 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2813 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2814 | */ |
2815 | ||
2816 | if (rdev->raid_disk != -1) | |
2817 | return -EBUSY; | |
2818 | ||
c6751b2b N |
2819 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2820 | return -EBUSY; | |
2821 | ||
6c2fce2e NB |
2822 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2823 | return -EINVAL; | |
2824 | ||
ba1b41b6 N |
2825 | if (slot >= rdev->mddev->raid_disks && |
2826 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2827 | return -ENOSPC; | |
2828 | ||
6c2fce2e NB |
2829 | rdev->raid_disk = slot; |
2830 | if (test_bit(In_sync, &rdev->flags)) | |
2831 | rdev->saved_raid_disk = slot; | |
2832 | else | |
2833 | rdev->saved_raid_disk = -1; | |
d30519fc | 2834 | clear_bit(In_sync, &rdev->flags); |
6c2fce2e NB |
2835 | err = rdev->mddev->pers-> |
2836 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2837 | if (err) { |
6c2fce2e | 2838 | rdev->raid_disk = -1; |
6c2fce2e | 2839 | return err; |
52664732 | 2840 | } else |
00bcb4ac | 2841 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
36fad858 | 2842 | if (sysfs_link_rdev(rdev->mddev, rdev)) |
00bcb4ac | 2843 | /* failure here is OK */; |
6c2fce2e | 2844 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2845 | } else { |
ba1b41b6 N |
2846 | if (slot >= rdev->mddev->raid_disks && |
2847 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2848 | return -ENOSPC; |
2849 | rdev->raid_disk = slot; | |
2850 | /* assume it is working */ | |
c5d79adb N |
2851 | clear_bit(Faulty, &rdev->flags); |
2852 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2853 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2854 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2855 | } |
014236d2 N |
2856 | return len; |
2857 | } | |
2858 | ||
2859 | ||
2860 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2861 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2862 | |
93c8cad0 | 2863 | static ssize_t |
3cb03002 | 2864 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2865 | { |
6961ece4 | 2866 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2867 | } |
2868 | ||
2869 | static ssize_t | |
3cb03002 | 2870 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2871 | { |
c6563a8c N |
2872 | unsigned long long offset; |
2873 | if (strict_strtoull(buf, 10, &offset) < 0) | |
93c8cad0 | 2874 | return -EINVAL; |
8ed0a521 | 2875 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2876 | return -EBUSY; |
dd8ac336 | 2877 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2878 | /* Must set offset before size, so overlap checks |
2879 | * can be sane */ | |
2880 | return -EBUSY; | |
93c8cad0 | 2881 | rdev->data_offset = offset; |
25f7fd47 | 2882 | rdev->new_data_offset = offset; |
93c8cad0 N |
2883 | return len; |
2884 | } | |
2885 | ||
2886 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2887 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2888 | |
c6563a8c N |
2889 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2890 | { | |
2891 | return sprintf(page, "%llu\n", | |
2892 | (unsigned long long)rdev->new_data_offset); | |
2893 | } | |
2894 | ||
2895 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2896 | const char *buf, size_t len) | |
2897 | { | |
2898 | unsigned long long new_offset; | |
2899 | struct mddev *mddev = rdev->mddev; | |
2900 | ||
2901 | if (strict_strtoull(buf, 10, &new_offset) < 0) | |
2902 | return -EINVAL; | |
2903 | ||
2904 | if (mddev->sync_thread) | |
2905 | return -EBUSY; | |
2906 | if (new_offset == rdev->data_offset) | |
2907 | /* reset is always permitted */ | |
2908 | ; | |
2909 | else if (new_offset > rdev->data_offset) { | |
2910 | /* must not push array size beyond rdev_sectors */ | |
2911 | if (new_offset - rdev->data_offset | |
2912 | + mddev->dev_sectors > rdev->sectors) | |
2913 | return -E2BIG; | |
2914 | } | |
2915 | /* Metadata worries about other space details. */ | |
2916 | ||
2917 | /* decreasing the offset is inconsistent with a backwards | |
2918 | * reshape. | |
2919 | */ | |
2920 | if (new_offset < rdev->data_offset && | |
2921 | mddev->reshape_backwards) | |
2922 | return -EINVAL; | |
2923 | /* Increasing offset is inconsistent with forwards | |
2924 | * reshape. reshape_direction should be set to | |
2925 | * 'backwards' first. | |
2926 | */ | |
2927 | if (new_offset > rdev->data_offset && | |
2928 | !mddev->reshape_backwards) | |
2929 | return -EINVAL; | |
2930 | ||
2931 | if (mddev->pers && mddev->persistent && | |
2932 | !super_types[mddev->major_version] | |
2933 | .allow_new_offset(rdev, new_offset)) | |
2934 | return -E2BIG; | |
2935 | rdev->new_data_offset = new_offset; | |
2936 | if (new_offset > rdev->data_offset) | |
2937 | mddev->reshape_backwards = 1; | |
2938 | else if (new_offset < rdev->data_offset) | |
2939 | mddev->reshape_backwards = 0; | |
2940 | ||
2941 | return len; | |
2942 | } | |
2943 | static struct rdev_sysfs_entry rdev_new_offset = | |
2944 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2945 | ||
83303b61 | 2946 | static ssize_t |
3cb03002 | 2947 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2948 | { |
dd8ac336 | 2949 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2950 | } |
2951 | ||
c5d79adb N |
2952 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2953 | { | |
2954 | /* check if two start/length pairs overlap */ | |
2955 | if (s1+l1 <= s2) | |
2956 | return 0; | |
2957 | if (s2+l2 <= s1) | |
2958 | return 0; | |
2959 | return 1; | |
2960 | } | |
2961 | ||
b522adcd DW |
2962 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2963 | { | |
2964 | unsigned long long blocks; | |
2965 | sector_t new; | |
2966 | ||
2967 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2968 | return -EINVAL; | |
2969 | ||
2970 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2971 | return -EINVAL; /* sector conversion overflow */ | |
2972 | ||
2973 | new = blocks * 2; | |
2974 | if (new != blocks * 2) | |
2975 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2976 | ||
2977 | *sectors = new; | |
2978 | return 0; | |
2979 | } | |
2980 | ||
83303b61 | 2981 | static ssize_t |
3cb03002 | 2982 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2983 | { |
fd01b88c | 2984 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2985 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2986 | sector_t sectors; |
27c529bb | 2987 | |
b522adcd | 2988 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2989 | return -EINVAL; |
c6563a8c N |
2990 | if (rdev->data_offset != rdev->new_data_offset) |
2991 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2992 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2993 | if (my_mddev->persistent) { |
dd8ac336 AN |
2994 | sectors = super_types[my_mddev->major_version]. |
2995 | rdev_size_change(rdev, sectors); | |
2996 | if (!sectors) | |
0cd17fec | 2997 | return -EBUSY; |
dd8ac336 | 2998 | } else if (!sectors) |
77304d2a | 2999 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 3000 | rdev->data_offset; |
a6468539 N |
3001 | if (!my_mddev->pers->resize) |
3002 | /* Cannot change size for RAID0 or Linear etc */ | |
3003 | return -EINVAL; | |
0cd17fec | 3004 | } |
dd8ac336 | 3005 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 3006 | return -EINVAL; /* component must fit device */ |
0cd17fec | 3007 | |
dd8ac336 AN |
3008 | rdev->sectors = sectors; |
3009 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
3010 | /* need to check that all other rdevs with the same ->bdev |
3011 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 3012 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
3013 | * we have to change it back, we will have the lock again. |
3014 | */ | |
fd01b88c | 3015 | struct mddev *mddev; |
c5d79adb | 3016 | int overlap = 0; |
159ec1fc | 3017 | struct list_head *tmp; |
c5d79adb | 3018 | |
27c529bb | 3019 | mddev_unlock(my_mddev); |
29ac4aa3 | 3020 | for_each_mddev(mddev, tmp) { |
3cb03002 | 3021 | struct md_rdev *rdev2; |
c5d79adb N |
3022 | |
3023 | mddev_lock(mddev); | |
dafb20fa | 3024 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
3025 | if (rdev->bdev == rdev2->bdev && |
3026 | rdev != rdev2 && | |
3027 | overlaps(rdev->data_offset, rdev->sectors, | |
3028 | rdev2->data_offset, | |
3029 | rdev2->sectors)) { | |
c5d79adb N |
3030 | overlap = 1; |
3031 | break; | |
3032 | } | |
3033 | mddev_unlock(mddev); | |
3034 | if (overlap) { | |
3035 | mddev_put(mddev); | |
3036 | break; | |
3037 | } | |
3038 | } | |
27c529bb | 3039 | mddev_lock(my_mddev); |
c5d79adb N |
3040 | if (overlap) { |
3041 | /* Someone else could have slipped in a size | |
3042 | * change here, but doing so is just silly. | |
dd8ac336 | 3043 | * We put oldsectors back because we *know* it is |
c5d79adb N |
3044 | * safe, and trust userspace not to race with |
3045 | * itself | |
3046 | */ | |
dd8ac336 | 3047 | rdev->sectors = oldsectors; |
c5d79adb N |
3048 | return -EBUSY; |
3049 | } | |
3050 | } | |
83303b61 N |
3051 | return len; |
3052 | } | |
3053 | ||
3054 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 3055 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 3056 | |
06e3c817 | 3057 | |
3cb03002 | 3058 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
3059 | { |
3060 | unsigned long long recovery_start = rdev->recovery_offset; | |
3061 | ||
3062 | if (test_bit(In_sync, &rdev->flags) || | |
3063 | recovery_start == MaxSector) | |
3064 | return sprintf(page, "none\n"); | |
3065 | ||
3066 | return sprintf(page, "%llu\n", recovery_start); | |
3067 | } | |
3068 | ||
3cb03002 | 3069 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
3070 | { |
3071 | unsigned long long recovery_start; | |
3072 | ||
3073 | if (cmd_match(buf, "none")) | |
3074 | recovery_start = MaxSector; | |
3075 | else if (strict_strtoull(buf, 10, &recovery_start)) | |
3076 | return -EINVAL; | |
3077 | ||
3078 | if (rdev->mddev->pers && | |
3079 | rdev->raid_disk >= 0) | |
3080 | return -EBUSY; | |
3081 | ||
3082 | rdev->recovery_offset = recovery_start; | |
3083 | if (recovery_start == MaxSector) | |
3084 | set_bit(In_sync, &rdev->flags); | |
3085 | else | |
3086 | clear_bit(In_sync, &rdev->flags); | |
3087 | return len; | |
3088 | } | |
3089 | ||
3090 | static struct rdev_sysfs_entry rdev_recovery_start = | |
3091 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
3092 | ||
16c791a5 N |
3093 | |
3094 | static ssize_t | |
3095 | badblocks_show(struct badblocks *bb, char *page, int unack); | |
3096 | static ssize_t | |
3097 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); | |
3098 | ||
3cb03002 | 3099 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3100 | { |
3101 | return badblocks_show(&rdev->badblocks, page, 0); | |
3102 | } | |
3cb03002 | 3103 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 3104 | { |
de393cde N |
3105 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
3106 | /* Maybe that ack was all we needed */ | |
3107 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
3108 | wake_up(&rdev->blocked_wait); | |
3109 | return rv; | |
16c791a5 N |
3110 | } |
3111 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
3112 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
3113 | ||
3114 | ||
3cb03002 | 3115 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3116 | { |
3117 | return badblocks_show(&rdev->badblocks, page, 1); | |
3118 | } | |
3cb03002 | 3119 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
3120 | { |
3121 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
3122 | } | |
3123 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
3124 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
3125 | ||
86e6ffdd N |
3126 | static struct attribute *rdev_default_attrs[] = { |
3127 | &rdev_state.attr, | |
4dbcdc75 | 3128 | &rdev_errors.attr, |
014236d2 | 3129 | &rdev_slot.attr, |
93c8cad0 | 3130 | &rdev_offset.attr, |
c6563a8c | 3131 | &rdev_new_offset.attr, |
83303b61 | 3132 | &rdev_size.attr, |
06e3c817 | 3133 | &rdev_recovery_start.attr, |
16c791a5 N |
3134 | &rdev_bad_blocks.attr, |
3135 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
3136 | NULL, |
3137 | }; | |
3138 | static ssize_t | |
3139 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3140 | { | |
3141 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3142 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
fd01b88c | 3143 | struct mddev *mddev = rdev->mddev; |
27c529bb | 3144 | ssize_t rv; |
86e6ffdd N |
3145 | |
3146 | if (!entry->show) | |
3147 | return -EIO; | |
27c529bb N |
3148 | |
3149 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
3150 | if (!rv) { | |
3151 | if (rdev->mddev == NULL) | |
3152 | rv = -EBUSY; | |
3153 | else | |
3154 | rv = entry->show(rdev, page); | |
3155 | mddev_unlock(mddev); | |
3156 | } | |
3157 | return rv; | |
86e6ffdd N |
3158 | } |
3159 | ||
3160 | static ssize_t | |
3161 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3162 | const char *page, size_t length) | |
3163 | { | |
3164 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3165 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3166 | ssize_t rv; |
fd01b88c | 3167 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3168 | |
3169 | if (!entry->store) | |
3170 | return -EIO; | |
67463acb N |
3171 | if (!capable(CAP_SYS_ADMIN)) |
3172 | return -EACCES; | |
27c529bb | 3173 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3174 | if (!rv) { |
27c529bb N |
3175 | if (rdev->mddev == NULL) |
3176 | rv = -EBUSY; | |
3177 | else | |
3178 | rv = entry->store(rdev, page, length); | |
6a51830e | 3179 | mddev_unlock(mddev); |
ca388059 N |
3180 | } |
3181 | return rv; | |
86e6ffdd N |
3182 | } |
3183 | ||
3184 | static void rdev_free(struct kobject *ko) | |
3185 | { | |
3cb03002 | 3186 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3187 | kfree(rdev); |
3188 | } | |
52cf25d0 | 3189 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3190 | .show = rdev_attr_show, |
3191 | .store = rdev_attr_store, | |
3192 | }; | |
3193 | static struct kobj_type rdev_ktype = { | |
3194 | .release = rdev_free, | |
3195 | .sysfs_ops = &rdev_sysfs_ops, | |
3196 | .default_attrs = rdev_default_attrs, | |
3197 | }; | |
3198 | ||
3cb03002 | 3199 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3200 | { |
3201 | rdev->desc_nr = -1; | |
3202 | rdev->saved_raid_disk = -1; | |
3203 | rdev->raid_disk = -1; | |
3204 | rdev->flags = 0; | |
3205 | rdev->data_offset = 0; | |
c6563a8c | 3206 | rdev->new_data_offset = 0; |
e8bb9a83 N |
3207 | rdev->sb_events = 0; |
3208 | rdev->last_read_error.tv_sec = 0; | |
3209 | rdev->last_read_error.tv_nsec = 0; | |
2699b672 N |
3210 | rdev->sb_loaded = 0; |
3211 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3212 | atomic_set(&rdev->nr_pending, 0); |
3213 | atomic_set(&rdev->read_errors, 0); | |
3214 | atomic_set(&rdev->corrected_errors, 0); | |
3215 | ||
3216 | INIT_LIST_HEAD(&rdev->same_set); | |
3217 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3218 | |
3219 | /* Add space to store bad block list. | |
3220 | * This reserves the space even on arrays where it cannot | |
3221 | * be used - I wonder if that matters | |
3222 | */ | |
3223 | rdev->badblocks.count = 0; | |
3224 | rdev->badblocks.shift = 0; | |
3225 | rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
3226 | seqlock_init(&rdev->badblocks.lock); | |
3227 | if (rdev->badblocks.page == NULL) | |
3228 | return -ENOMEM; | |
3229 | ||
3230 | return 0; | |
e8bb9a83 N |
3231 | } |
3232 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3233 | /* |
3234 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3235 | * | |
3236 | * mark the device faulty if: | |
3237 | * | |
3238 | * - the device is nonexistent (zero size) | |
3239 | * - the device has no valid superblock | |
3240 | * | |
3241 | * a faulty rdev _never_ has rdev->sb set. | |
3242 | */ | |
3cb03002 | 3243 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3244 | { |
3245 | char b[BDEVNAME_SIZE]; | |
3246 | int err; | |
3cb03002 | 3247 | struct md_rdev *rdev; |
1da177e4 LT |
3248 | sector_t size; |
3249 | ||
9ffae0cf | 3250 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
3251 | if (!rdev) { |
3252 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
3253 | return ERR_PTR(-ENOMEM); | |
3254 | } | |
1da177e4 | 3255 | |
2230dfe4 N |
3256 | err = md_rdev_init(rdev); |
3257 | if (err) | |
3258 | goto abort_free; | |
3259 | err = alloc_disk_sb(rdev); | |
3260 | if (err) | |
1da177e4 LT |
3261 | goto abort_free; |
3262 | ||
c5d79adb | 3263 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3264 | if (err) |
3265 | goto abort_free; | |
3266 | ||
f9cb074b | 3267 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3268 | |
77304d2a | 3269 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 LT |
3270 | if (!size) { |
3271 | printk(KERN_WARNING | |
3272 | "md: %s has zero or unknown size, marking faulty!\n", | |
3273 | bdevname(rdev->bdev,b)); | |
3274 | err = -EINVAL; | |
3275 | goto abort_free; | |
3276 | } | |
3277 | ||
3278 | if (super_format >= 0) { | |
3279 | err = super_types[super_format]. | |
3280 | load_super(rdev, NULL, super_minor); | |
3281 | if (err == -EINVAL) { | |
df968c4e N |
3282 | printk(KERN_WARNING |
3283 | "md: %s does not have a valid v%d.%d " | |
3284 | "superblock, not importing!\n", | |
3285 | bdevname(rdev->bdev,b), | |
3286 | super_format, super_minor); | |
1da177e4 LT |
3287 | goto abort_free; |
3288 | } | |
3289 | if (err < 0) { | |
3290 | printk(KERN_WARNING | |
3291 | "md: could not read %s's sb, not importing!\n", | |
3292 | bdevname(rdev->bdev,b)); | |
3293 | goto abort_free; | |
3294 | } | |
3295 | } | |
9f2f3830 N |
3296 | if (super_format == -1) |
3297 | /* hot-add for 0.90, or non-persistent: so no badblocks */ | |
3298 | rdev->badblocks.shift = -1; | |
6bfe0b49 | 3299 | |
1da177e4 LT |
3300 | return rdev; |
3301 | ||
3302 | abort_free: | |
2699b672 N |
3303 | if (rdev->bdev) |
3304 | unlock_rdev(rdev); | |
545c8795 | 3305 | md_rdev_clear(rdev); |
1da177e4 LT |
3306 | kfree(rdev); |
3307 | return ERR_PTR(err); | |
3308 | } | |
3309 | ||
3310 | /* | |
3311 | * Check a full RAID array for plausibility | |
3312 | */ | |
3313 | ||
3314 | ||
fd01b88c | 3315 | static void analyze_sbs(struct mddev * mddev) |
1da177e4 LT |
3316 | { |
3317 | int i; | |
3cb03002 | 3318 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3319 | char b[BDEVNAME_SIZE]; |
3320 | ||
3321 | freshest = NULL; | |
dafb20fa | 3322 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3323 | switch (super_types[mddev->major_version]. |
3324 | load_super(rdev, freshest, mddev->minor_version)) { | |
3325 | case 1: | |
3326 | freshest = rdev; | |
3327 | break; | |
3328 | case 0: | |
3329 | break; | |
3330 | default: | |
3331 | printk( KERN_ERR \ | |
3332 | "md: fatal superblock inconsistency in %s" | |
3333 | " -- removing from array\n", | |
3334 | bdevname(rdev->bdev,b)); | |
3335 | kick_rdev_from_array(rdev); | |
3336 | } | |
3337 | ||
3338 | ||
3339 | super_types[mddev->major_version]. | |
3340 | validate_super(mddev, freshest); | |
3341 | ||
3342 | i = 0; | |
dafb20fa | 3343 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3344 | if (mddev->max_disks && |
3345 | (rdev->desc_nr >= mddev->max_disks || | |
3346 | i > mddev->max_disks)) { | |
de01dfad N |
3347 | printk(KERN_WARNING |
3348 | "md: %s: %s: only %d devices permitted\n", | |
3349 | mdname(mddev), bdevname(rdev->bdev, b), | |
3350 | mddev->max_disks); | |
3351 | kick_rdev_from_array(rdev); | |
3352 | continue; | |
3353 | } | |
1da177e4 LT |
3354 | if (rdev != freshest) |
3355 | if (super_types[mddev->major_version]. | |
3356 | validate_super(mddev, rdev)) { | |
3357 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
3358 | " from array!\n", | |
3359 | bdevname(rdev->bdev,b)); | |
3360 | kick_rdev_from_array(rdev); | |
3361 | continue; | |
3362 | } | |
3363 | if (mddev->level == LEVEL_MULTIPATH) { | |
3364 | rdev->desc_nr = i++; | |
3365 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3366 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 3367 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
3368 | rdev->raid_disk = -1; |
3369 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3370 | } |
3371 | } | |
1da177e4 LT |
3372 | } |
3373 | ||
72e02075 N |
3374 | /* Read a fixed-point number. |
3375 | * Numbers in sysfs attributes should be in "standard" units where | |
3376 | * possible, so time should be in seconds. | |
3377 | * However we internally use a a much smaller unit such as | |
3378 | * milliseconds or jiffies. | |
3379 | * This function takes a decimal number with a possible fractional | |
3380 | * component, and produces an integer which is the result of | |
3381 | * multiplying that number by 10^'scale'. | |
3382 | * all without any floating-point arithmetic. | |
3383 | */ | |
3384 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3385 | { | |
3386 | unsigned long result = 0; | |
3387 | long decimals = -1; | |
3388 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3389 | if (*cp == '.') | |
3390 | decimals = 0; | |
3391 | else if (decimals < scale) { | |
3392 | unsigned int value; | |
3393 | value = *cp - '0'; | |
3394 | result = result * 10 + value; | |
3395 | if (decimals >= 0) | |
3396 | decimals++; | |
3397 | } | |
3398 | cp++; | |
3399 | } | |
3400 | if (*cp == '\n') | |
3401 | cp++; | |
3402 | if (*cp) | |
3403 | return -EINVAL; | |
3404 | if (decimals < 0) | |
3405 | decimals = 0; | |
3406 | while (decimals < scale) { | |
3407 | result *= 10; | |
3408 | decimals ++; | |
3409 | } | |
3410 | *res = result; | |
3411 | return 0; | |
3412 | } | |
3413 | ||
3414 | ||
19052c0e N |
3415 | static void md_safemode_timeout(unsigned long data); |
3416 | ||
16f17b39 | 3417 | static ssize_t |
fd01b88c | 3418 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3419 | { |
3420 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3421 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3422 | } | |
3423 | static ssize_t | |
fd01b88c | 3424 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3425 | { |
16f17b39 | 3426 | unsigned long msec; |
97ce0a7f | 3427 | |
72e02075 | 3428 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3429 | return -EINVAL; |
16f17b39 N |
3430 | if (msec == 0) |
3431 | mddev->safemode_delay = 0; | |
3432 | else { | |
19052c0e | 3433 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
3434 | mddev->safemode_delay = (msec*HZ)/1000; |
3435 | if (mddev->safemode_delay == 0) | |
3436 | mddev->safemode_delay = 1; | |
19052c0e N |
3437 | if (mddev->safemode_delay < old_delay) |
3438 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
3439 | } |
3440 | return len; | |
3441 | } | |
3442 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3443 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3444 | |
eae1701f | 3445 | static ssize_t |
fd01b88c | 3446 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3447 | { |
84fc4b56 | 3448 | struct md_personality *p = mddev->pers; |
d9d166c2 | 3449 | if (p) |
eae1701f | 3450 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
3451 | else if (mddev->clevel[0]) |
3452 | return sprintf(page, "%s\n", mddev->clevel); | |
3453 | else if (mddev->level != LEVEL_NONE) | |
3454 | return sprintf(page, "%d\n", mddev->level); | |
3455 | else | |
3456 | return 0; | |
eae1701f N |
3457 | } |
3458 | ||
d9d166c2 | 3459 | static ssize_t |
fd01b88c | 3460 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3461 | { |
f2859af6 | 3462 | char clevel[16]; |
20a49ff6 | 3463 | ssize_t rv = len; |
84fc4b56 | 3464 | struct md_personality *pers; |
f2859af6 | 3465 | long level; |
245f46c2 | 3466 | void *priv; |
3cb03002 | 3467 | struct md_rdev *rdev; |
245f46c2 N |
3468 | |
3469 | if (mddev->pers == NULL) { | |
3470 | if (len == 0) | |
3471 | return 0; | |
3472 | if (len >= sizeof(mddev->clevel)) | |
3473 | return -ENOSPC; | |
3474 | strncpy(mddev->clevel, buf, len); | |
3475 | if (mddev->clevel[len-1] == '\n') | |
3476 | len--; | |
3477 | mddev->clevel[len] = 0; | |
3478 | mddev->level = LEVEL_NONE; | |
3479 | return rv; | |
3480 | } | |
3481 | ||
3482 | /* request to change the personality. Need to ensure: | |
3483 | * - array is not engaged in resync/recovery/reshape | |
3484 | * - old personality can be suspended | |
3485 | * - new personality will access other array. | |
3486 | */ | |
3487 | ||
bb4f1e9d N |
3488 | if (mddev->sync_thread || |
3489 | mddev->reshape_position != MaxSector || | |
3490 | mddev->sysfs_active) | |
d9d166c2 | 3491 | return -EBUSY; |
245f46c2 N |
3492 | |
3493 | if (!mddev->pers->quiesce) { | |
3494 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3495 | mdname(mddev), mddev->pers->name); | |
3496 | return -EINVAL; | |
3497 | } | |
3498 | ||
3499 | /* Now find the new personality */ | |
f2859af6 | 3500 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3501 | return -EINVAL; |
f2859af6 DW |
3502 | strncpy(clevel, buf, len); |
3503 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3504 | len--; |
f2859af6 DW |
3505 | clevel[len] = 0; |
3506 | if (strict_strtol(clevel, 10, &level)) | |
3507 | level = LEVEL_NONE; | |
245f46c2 | 3508 | |
f2859af6 DW |
3509 | if (request_module("md-%s", clevel) != 0) |
3510 | request_module("md-level-%s", clevel); | |
245f46c2 | 3511 | spin_lock(&pers_lock); |
f2859af6 | 3512 | pers = find_pers(level, clevel); |
245f46c2 N |
3513 | if (!pers || !try_module_get(pers->owner)) { |
3514 | spin_unlock(&pers_lock); | |
f2859af6 | 3515 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3516 | return -EINVAL; |
3517 | } | |
3518 | spin_unlock(&pers_lock); | |
3519 | ||
3520 | if (pers == mddev->pers) { | |
3521 | /* Nothing to do! */ | |
3522 | module_put(pers->owner); | |
3523 | return rv; | |
3524 | } | |
3525 | if (!pers->takeover) { | |
3526 | module_put(pers->owner); | |
3527 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3528 | mdname(mddev), clevel); |
245f46c2 N |
3529 | return -EINVAL; |
3530 | } | |
3531 | ||
dafb20fa | 3532 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3533 | rdev->new_raid_disk = rdev->raid_disk; |
3534 | ||
245f46c2 N |
3535 | /* ->takeover must set new_* and/or delta_disks |
3536 | * if it succeeds, and may set them when it fails. | |
3537 | */ | |
3538 | priv = pers->takeover(mddev); | |
3539 | if (IS_ERR(priv)) { | |
3540 | mddev->new_level = mddev->level; | |
3541 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3542 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3543 | mddev->raid_disks -= mddev->delta_disks; |
3544 | mddev->delta_disks = 0; | |
2c810cdd | 3545 | mddev->reshape_backwards = 0; |
245f46c2 N |
3546 | module_put(pers->owner); |
3547 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3548 | mdname(mddev), clevel); |
245f46c2 N |
3549 | return PTR_ERR(priv); |
3550 | } | |
3551 | ||
3552 | /* Looks like we have a winner */ | |
3553 | mddev_suspend(mddev); | |
3554 | mddev->pers->stop(mddev); | |
a64c876f N |
3555 | |
3556 | if (mddev->pers->sync_request == NULL && | |
3557 | pers->sync_request != NULL) { | |
3558 | /* need to add the md_redundancy_group */ | |
3559 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3560 | printk(KERN_WARNING | |
3561 | "md: cannot register extra attributes for %s\n", | |
3562 | mdname(mddev)); | |
19fdb9ee | 3563 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); |
a64c876f N |
3564 | } |
3565 | if (mddev->pers->sync_request != NULL && | |
3566 | pers->sync_request == NULL) { | |
3567 | /* need to remove the md_redundancy_group */ | |
3568 | if (mddev->to_remove == NULL) | |
3569 | mddev->to_remove = &md_redundancy_group; | |
3570 | } | |
3571 | ||
54071b38 TM |
3572 | if (mddev->pers->sync_request == NULL && |
3573 | mddev->external) { | |
3574 | /* We are converting from a no-redundancy array | |
3575 | * to a redundancy array and metadata is managed | |
3576 | * externally so we need to be sure that writes | |
3577 | * won't block due to a need to transition | |
3578 | * clean->dirty | |
3579 | * until external management is started. | |
3580 | */ | |
3581 | mddev->in_sync = 0; | |
3582 | mddev->safemode_delay = 0; | |
3583 | mddev->safemode = 0; | |
3584 | } | |
3585 | ||
dafb20fa | 3586 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3587 | if (rdev->raid_disk < 0) |
3588 | continue; | |
bf2cb0da | 3589 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3590 | rdev->new_raid_disk = -1; |
3591 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3592 | continue; | |
36fad858 | 3593 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3594 | } |
dafb20fa | 3595 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3596 | if (rdev->raid_disk < 0) |
3597 | continue; | |
3598 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3599 | continue; | |
3600 | rdev->raid_disk = rdev->new_raid_disk; | |
3601 | if (rdev->raid_disk < 0) | |
3a981b03 | 3602 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3603 | else { |
36fad858 NK |
3604 | if (sysfs_link_rdev(mddev, rdev)) |
3605 | printk(KERN_WARNING "md: cannot register rd%d" | |
3606 | " for %s after level change\n", | |
3607 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3608 | } |
e93f68a1 N |
3609 | } |
3610 | ||
3611 | module_put(mddev->pers->owner); | |
245f46c2 N |
3612 | mddev->pers = pers; |
3613 | mddev->private = priv; | |
3614 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3615 | mddev->level = mddev->new_level; | |
3616 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3617 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3618 | mddev->delta_disks = 0; |
2c810cdd | 3619 | mddev->reshape_backwards = 0; |
fee68723 | 3620 | mddev->degraded = 0; |
9af204cf TM |
3621 | if (mddev->pers->sync_request == NULL) { |
3622 | /* this is now an array without redundancy, so | |
3623 | * it must always be in_sync | |
3624 | */ | |
3625 | mddev->in_sync = 1; | |
3626 | del_timer_sync(&mddev->safemode_timer); | |
3627 | } | |
245f46c2 | 3628 | pers->run(mddev); |
245f46c2 | 3629 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
47525e59 | 3630 | mddev_resume(mddev); |
5cac7861 | 3631 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3632 | md_new_event(mddev); |
d9d166c2 N |
3633 | return rv; |
3634 | } | |
3635 | ||
3636 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3637 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3638 | |
d4dbd025 N |
3639 | |
3640 | static ssize_t | |
fd01b88c | 3641 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3642 | { |
3643 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3644 | if (mddev->reshape_position != MaxSector && |
3645 | mddev->layout != mddev->new_layout) | |
3646 | return sprintf(page, "%d (%d)\n", | |
3647 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3648 | return sprintf(page, "%d\n", mddev->layout); |
3649 | } | |
3650 | ||
3651 | static ssize_t | |
fd01b88c | 3652 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 N |
3653 | { |
3654 | char *e; | |
3655 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3656 | |
3657 | if (!*buf || (*e && *e != '\n')) | |
3658 | return -EINVAL; | |
3659 | ||
b3546035 N |
3660 | if (mddev->pers) { |
3661 | int err; | |
50ac168a | 3662 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3663 | return -EBUSY; |
597a711b | 3664 | mddev->new_layout = n; |
50ac168a | 3665 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3666 | if (err) { |
3667 | mddev->new_layout = mddev->layout; | |
b3546035 | 3668 | return err; |
597a711b | 3669 | } |
b3546035 | 3670 | } else { |
08a02ecd | 3671 | mddev->new_layout = n; |
b3546035 N |
3672 | if (mddev->reshape_position == MaxSector) |
3673 | mddev->layout = n; | |
3674 | } | |
d4dbd025 N |
3675 | return len; |
3676 | } | |
3677 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3678 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3679 | |
3680 | ||
eae1701f | 3681 | static ssize_t |
fd01b88c | 3682 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3683 | { |
bb636547 N |
3684 | if (mddev->raid_disks == 0) |
3685 | return 0; | |
08a02ecd N |
3686 | if (mddev->reshape_position != MaxSector && |
3687 | mddev->delta_disks != 0) | |
3688 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3689 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3690 | return sprintf(page, "%d\n", mddev->raid_disks); |
3691 | } | |
3692 | ||
fd01b88c | 3693 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3694 | |
3695 | static ssize_t | |
fd01b88c | 3696 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3697 | { |
da943b99 N |
3698 | char *e; |
3699 | int rv = 0; | |
3700 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3701 | ||
3702 | if (!*buf || (*e && *e != '\n')) | |
3703 | return -EINVAL; | |
3704 | ||
3705 | if (mddev->pers) | |
3706 | rv = update_raid_disks(mddev, n); | |
08a02ecd | 3707 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3708 | struct md_rdev *rdev; |
08a02ecd | 3709 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c N |
3710 | |
3711 | rdev_for_each(rdev, mddev) { | |
3712 | if (olddisks < n && | |
3713 | rdev->data_offset < rdev->new_data_offset) | |
3714 | return -EINVAL; | |
3715 | if (olddisks > n && | |
3716 | rdev->data_offset > rdev->new_data_offset) | |
3717 | return -EINVAL; | |
3718 | } | |
08a02ecd N |
3719 | mddev->delta_disks = n - olddisks; |
3720 | mddev->raid_disks = n; | |
2c810cdd | 3721 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3722 | } else |
da943b99 N |
3723 | mddev->raid_disks = n; |
3724 | return rv ? rv : len; | |
3725 | } | |
3726 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3727 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3728 | |
3b34380a | 3729 | static ssize_t |
fd01b88c | 3730 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3731 | { |
08a02ecd | 3732 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3733 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3734 | return sprintf(page, "%d (%d)\n", | |
3735 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3736 | mddev->chunk_sectors << 9); |
3737 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3738 | } |
3739 | ||
3740 | static ssize_t | |
fd01b88c | 3741 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3742 | { |
3b34380a N |
3743 | char *e; |
3744 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3745 | ||
3b34380a N |
3746 | if (!*buf || (*e && *e != '\n')) |
3747 | return -EINVAL; | |
3748 | ||
b3546035 N |
3749 | if (mddev->pers) { |
3750 | int err; | |
50ac168a | 3751 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3752 | return -EBUSY; |
597a711b | 3753 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3754 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3755 | if (err) { |
3756 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3757 | return err; |
597a711b | 3758 | } |
b3546035 | 3759 | } else { |
664e7c41 | 3760 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3761 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3762 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3763 | } |
3b34380a N |
3764 | return len; |
3765 | } | |
3766 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3767 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3768 | |
a94213b1 | 3769 | static ssize_t |
fd01b88c | 3770 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3771 | { |
d1a7c503 N |
3772 | if (mddev->recovery_cp == MaxSector) |
3773 | return sprintf(page, "none\n"); | |
a94213b1 N |
3774 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3775 | } | |
3776 | ||
3777 | static ssize_t | |
fd01b88c | 3778 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3779 | { |
a94213b1 N |
3780 | char *e; |
3781 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3782 | ||
b098636c | 3783 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
a94213b1 | 3784 | return -EBUSY; |
06e3c817 DW |
3785 | if (cmd_match(buf, "none")) |
3786 | n = MaxSector; | |
3787 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3788 | return -EINVAL; |
3789 | ||
3790 | mddev->recovery_cp = n; | |
db07d85e N |
3791 | if (mddev->pers) |
3792 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
a94213b1 N |
3793 | return len; |
3794 | } | |
3795 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3796 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3797 | |
9e653b63 N |
3798 | /* |
3799 | * The array state can be: | |
3800 | * | |
3801 | * clear | |
3802 | * No devices, no size, no level | |
3803 | * Equivalent to STOP_ARRAY ioctl | |
3804 | * inactive | |
3805 | * May have some settings, but array is not active | |
3806 | * all IO results in error | |
3807 | * When written, doesn't tear down array, but just stops it | |
3808 | * suspended (not supported yet) | |
3809 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3810 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3811 | * readonly |
3812 | * no resync can happen. no superblocks get written. | |
3813 | * write requests fail | |
3814 | * read-auto | |
3815 | * like readonly, but behaves like 'clean' on a write request. | |
3816 | * | |
3817 | * clean - no pending writes, but otherwise active. | |
3818 | * When written to inactive array, starts without resync | |
3819 | * If a write request arrives then | |
3820 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3821 | * if not known, block and switch to write-pending | |
3822 | * If written to an active array that has pending writes, then fails. | |
3823 | * active | |
3824 | * fully active: IO and resync can be happening. | |
3825 | * When written to inactive array, starts with resync | |
3826 | * | |
3827 | * write-pending | |
3828 | * clean, but writes are blocked waiting for 'active' to be written. | |
3829 | * | |
3830 | * active-idle | |
3831 | * like active, but no writes have been seen for a while (100msec). | |
3832 | * | |
3833 | */ | |
3834 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3835 | write_pending, active_idle, bad_word}; | |
05381954 | 3836 | static char *array_states[] = { |
9e653b63 N |
3837 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3838 | "write-pending", "active-idle", NULL }; | |
3839 | ||
3840 | static int match_word(const char *word, char **list) | |
3841 | { | |
3842 | int n; | |
3843 | for (n=0; list[n]; n++) | |
3844 | if (cmd_match(word, list[n])) | |
3845 | break; | |
3846 | return n; | |
3847 | } | |
3848 | ||
3849 | static ssize_t | |
fd01b88c | 3850 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3851 | { |
3852 | enum array_state st = inactive; | |
3853 | ||
3854 | if (mddev->pers) | |
3855 | switch(mddev->ro) { | |
3856 | case 1: | |
3857 | st = readonly; | |
3858 | break; | |
3859 | case 2: | |
3860 | st = read_auto; | |
3861 | break; | |
3862 | case 0: | |
3863 | if (mddev->in_sync) | |
3864 | st = clean; | |
070dc6dd | 3865 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3866 | st = write_pending; |
9e653b63 N |
3867 | else if (mddev->safemode) |
3868 | st = active_idle; | |
3869 | else | |
3870 | st = active; | |
3871 | } | |
3872 | else { | |
3873 | if (list_empty(&mddev->disks) && | |
3874 | mddev->raid_disks == 0 && | |
58c0fed4 | 3875 | mddev->dev_sectors == 0) |
9e653b63 N |
3876 | st = clear; |
3877 | else | |
3878 | st = inactive; | |
3879 | } | |
3880 | return sprintf(page, "%s\n", array_states[st]); | |
3881 | } | |
3882 | ||
a05b7ea0 N |
3883 | static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev); |
3884 | static int md_set_readonly(struct mddev * mddev, struct block_device *bdev); | |
fd01b88c N |
3885 | static int do_md_run(struct mddev * mddev); |
3886 | static int restart_array(struct mddev *mddev); | |
9e653b63 N |
3887 | |
3888 | static ssize_t | |
fd01b88c | 3889 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 N |
3890 | { |
3891 | int err = -EINVAL; | |
3892 | enum array_state st = match_word(buf, array_states); | |
3893 | switch(st) { | |
3894 | case bad_word: | |
3895 | break; | |
3896 | case clear: | |
3897 | /* stopping an active array */ | |
a05b7ea0 | 3898 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3899 | break; |
3900 | case inactive: | |
3901 | /* stopping an active array */ | |
90cf195d | 3902 | if (mddev->pers) |
a05b7ea0 | 3903 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3904 | else |
e691063a | 3905 | err = 0; /* already inactive */ |
9e653b63 N |
3906 | break; |
3907 | case suspended: | |
3908 | break; /* not supported yet */ | |
3909 | case readonly: | |
3910 | if (mddev->pers) | |
a05b7ea0 | 3911 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3912 | else { |
3913 | mddev->ro = 1; | |
648b629e | 3914 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3915 | err = do_md_run(mddev); |
3916 | } | |
3917 | break; | |
3918 | case read_auto: | |
9e653b63 | 3919 | if (mddev->pers) { |
80268ee9 | 3920 | if (mddev->ro == 0) |
a05b7ea0 | 3921 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3922 | else if (mddev->ro == 1) |
648b629e N |
3923 | err = restart_array(mddev); |
3924 | if (err == 0) { | |
3925 | mddev->ro = 2; | |
3926 | set_disk_ro(mddev->gendisk, 0); | |
3927 | } | |
9e653b63 N |
3928 | } else { |
3929 | mddev->ro = 2; | |
3930 | err = do_md_run(mddev); | |
3931 | } | |
3932 | break; | |
3933 | case clean: | |
3934 | if (mddev->pers) { | |
3935 | restart_array(mddev); | |
3936 | spin_lock_irq(&mddev->write_lock); | |
3937 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3938 | if (mddev->in_sync == 0) { |
3939 | mddev->in_sync = 1; | |
31a59e34 N |
3940 | if (mddev->safemode == 1) |
3941 | mddev->safemode = 0; | |
070dc6dd | 3942 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3943 | } |
3944 | err = 0; | |
3945 | } else | |
3946 | err = -EBUSY; | |
9e653b63 | 3947 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3948 | } else |
3949 | err = -EINVAL; | |
9e653b63 N |
3950 | break; |
3951 | case active: | |
3952 | if (mddev->pers) { | |
3953 | restart_array(mddev); | |
070dc6dd | 3954 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3955 | wake_up(&mddev->sb_wait); |
3956 | err = 0; | |
3957 | } else { | |
3958 | mddev->ro = 0; | |
648b629e | 3959 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3960 | err = do_md_run(mddev); |
3961 | } | |
3962 | break; | |
3963 | case write_pending: | |
3964 | case active_idle: | |
3965 | /* these cannot be set */ | |
3966 | break; | |
3967 | } | |
3968 | if (err) | |
3969 | return err; | |
0fd62b86 | 3970 | else { |
1d23f178 N |
3971 | if (mddev->hold_active == UNTIL_IOCTL) |
3972 | mddev->hold_active = 0; | |
00bcb4ac | 3973 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3974 | return len; |
0fd62b86 | 3975 | } |
9e653b63 | 3976 | } |
80ca3a44 N |
3977 | static struct md_sysfs_entry md_array_state = |
3978 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3979 | |
1e50915f | 3980 | static ssize_t |
fd01b88c | 3981 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
3982 | return sprintf(page, "%d\n", |
3983 | atomic_read(&mddev->max_corr_read_errors)); | |
3984 | } | |
3985 | ||
3986 | static ssize_t | |
fd01b88c | 3987 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f RB |
3988 | { |
3989 | char *e; | |
3990 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3991 | ||
3992 | if (*buf && (*e == 0 || *e == '\n')) { | |
3993 | atomic_set(&mddev->max_corr_read_errors, n); | |
3994 | return len; | |
3995 | } | |
3996 | return -EINVAL; | |
3997 | } | |
3998 | ||
3999 | static struct md_sysfs_entry max_corr_read_errors = | |
4000 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
4001 | max_corrected_read_errors_store); | |
4002 | ||
6d7ff738 | 4003 | static ssize_t |
fd01b88c | 4004 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
4005 | { |
4006 | return -EINVAL; | |
4007 | } | |
4008 | ||
4009 | static ssize_t | |
fd01b88c | 4010 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
4011 | { |
4012 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
4013 | /* The new device is added to the array. | |
4014 | * If the array has a persistent superblock, we read the | |
4015 | * superblock to initialise info and check validity. | |
4016 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
4017 | * which mainly checks size. | |
4018 | */ | |
4019 | char *e; | |
4020 | int major = simple_strtoul(buf, &e, 10); | |
4021 | int minor; | |
4022 | dev_t dev; | |
3cb03002 | 4023 | struct md_rdev *rdev; |
6d7ff738 N |
4024 | int err; |
4025 | ||
4026 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
4027 | return -EINVAL; | |
4028 | minor = simple_strtoul(e+1, &e, 10); | |
4029 | if (*e && *e != '\n') | |
4030 | return -EINVAL; | |
4031 | dev = MKDEV(major, minor); | |
4032 | if (major != MAJOR(dev) || | |
4033 | minor != MINOR(dev)) | |
4034 | return -EOVERFLOW; | |
4035 | ||
4036 | ||
4037 | if (mddev->persistent) { | |
4038 | rdev = md_import_device(dev, mddev->major_version, | |
4039 | mddev->minor_version); | |
4040 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4041 | struct md_rdev *rdev0 |
4042 | = list_entry(mddev->disks.next, | |
4043 | struct md_rdev, same_set); | |
6d7ff738 N |
4044 | err = super_types[mddev->major_version] |
4045 | .load_super(rdev, rdev0, mddev->minor_version); | |
4046 | if (err < 0) | |
4047 | goto out; | |
4048 | } | |
c5d79adb N |
4049 | } else if (mddev->external) |
4050 | rdev = md_import_device(dev, -2, -1); | |
4051 | else | |
6d7ff738 N |
4052 | rdev = md_import_device(dev, -1, -1); |
4053 | ||
4054 | if (IS_ERR(rdev)) | |
4055 | return PTR_ERR(rdev); | |
4056 | err = bind_rdev_to_array(rdev, mddev); | |
4057 | out: | |
4058 | if (err) | |
4059 | export_rdev(rdev); | |
4060 | return err ? err : len; | |
4061 | } | |
4062 | ||
4063 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4064 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4065 | |
9b1d1dac | 4066 | static ssize_t |
fd01b88c | 4067 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4068 | { |
4069 | char *end; | |
4070 | unsigned long chunk, end_chunk; | |
4071 | ||
4072 | if (!mddev->bitmap) | |
4073 | goto out; | |
4074 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4075 | while (*buf) { | |
4076 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4077 | if (buf == end) break; | |
4078 | if (*end == '-') { /* range */ | |
4079 | buf = end + 1; | |
4080 | end_chunk = simple_strtoul(buf, &end, 0); | |
4081 | if (buf == end) break; | |
4082 | } | |
4083 | if (*end && !isspace(*end)) break; | |
4084 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4085 | buf = skip_spaces(end); |
9b1d1dac PC |
4086 | } |
4087 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4088 | out: | |
4089 | return len; | |
4090 | } | |
4091 | ||
4092 | static struct md_sysfs_entry md_bitmap = | |
4093 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4094 | ||
a35b0d69 | 4095 | static ssize_t |
fd01b88c | 4096 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4097 | { |
58c0fed4 AN |
4098 | return sprintf(page, "%llu\n", |
4099 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4100 | } |
4101 | ||
fd01b88c | 4102 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4103 | |
4104 | static ssize_t | |
fd01b88c | 4105 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4106 | { |
4107 | /* If array is inactive, we can reduce the component size, but | |
4108 | * not increase it (except from 0). | |
4109 | * If array is active, we can try an on-line resize | |
4110 | */ | |
b522adcd DW |
4111 | sector_t sectors; |
4112 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4113 | |
58c0fed4 AN |
4114 | if (err < 0) |
4115 | return err; | |
a35b0d69 | 4116 | if (mddev->pers) { |
58c0fed4 | 4117 | err = update_size(mddev, sectors); |
850b2b42 | 4118 | md_update_sb(mddev, 1); |
a35b0d69 | 4119 | } else { |
58c0fed4 AN |
4120 | if (mddev->dev_sectors == 0 || |
4121 | mddev->dev_sectors > sectors) | |
4122 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4123 | else |
4124 | err = -ENOSPC; | |
4125 | } | |
4126 | return err ? err : len; | |
4127 | } | |
4128 | ||
4129 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4130 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4131 | |
8bb93aac | 4132 | |
83f0d77a | 4133 | /* Metadata version. |
e691063a N |
4134 | * This is one of |
4135 | * 'none' for arrays with no metadata (good luck...) | |
4136 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4137 | * or N.M for internally known formats |
4138 | */ | |
4139 | static ssize_t | |
fd01b88c | 4140 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4141 | { |
4142 | if (mddev->persistent) | |
4143 | return sprintf(page, "%d.%d\n", | |
4144 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4145 | else if (mddev->external) |
4146 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4147 | else |
4148 | return sprintf(page, "none\n"); | |
4149 | } | |
4150 | ||
4151 | static ssize_t | |
fd01b88c | 4152 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4153 | { |
4154 | int major, minor; | |
4155 | char *e; | |
ea43ddd8 N |
4156 | /* Changing the details of 'external' metadata is |
4157 | * always permitted. Otherwise there must be | |
4158 | * no devices attached to the array. | |
4159 | */ | |
4160 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
4161 | ; | |
4162 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
4163 | return -EBUSY; |
4164 | ||
4165 | if (cmd_match(buf, "none")) { | |
4166 | mddev->persistent = 0; | |
e691063a N |
4167 | mddev->external = 0; |
4168 | mddev->major_version = 0; | |
4169 | mddev->minor_version = 90; | |
4170 | return len; | |
4171 | } | |
4172 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4173 | size_t namelen = len-9; |
e691063a N |
4174 | if (namelen >= sizeof(mddev->metadata_type)) |
4175 | namelen = sizeof(mddev->metadata_type)-1; | |
4176 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4177 | mddev->metadata_type[namelen] = 0; | |
4178 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4179 | mddev->metadata_type[--namelen] = 0; | |
4180 | mddev->persistent = 0; | |
4181 | mddev->external = 1; | |
8bb93aac N |
4182 | mddev->major_version = 0; |
4183 | mddev->minor_version = 90; | |
4184 | return len; | |
4185 | } | |
4186 | major = simple_strtoul(buf, &e, 10); | |
4187 | if (e==buf || *e != '.') | |
4188 | return -EINVAL; | |
4189 | buf = e+1; | |
4190 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4191 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 4192 | return -EINVAL; |
50511da3 | 4193 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
4194 | return -ENOENT; |
4195 | mddev->major_version = major; | |
4196 | mddev->minor_version = minor; | |
4197 | mddev->persistent = 1; | |
e691063a | 4198 | mddev->external = 0; |
8bb93aac N |
4199 | return len; |
4200 | } | |
4201 | ||
4202 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 4203 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4204 | |
24dd469d | 4205 | static ssize_t |
fd01b88c | 4206 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4207 | { |
7eec314d | 4208 | char *type = "idle"; |
b6a9ce68 N |
4209 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
4210 | type = "frozen"; | |
4211 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 4212 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
4213 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4214 | type = "reshape"; | |
4215 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
4216 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
4217 | type = "resync"; | |
4218 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
4219 | type = "check"; | |
4220 | else | |
4221 | type = "repair"; | |
72a23c21 | 4222 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
4223 | type = "recover"; |
4224 | } | |
4225 | return sprintf(page, "%s\n", type); | |
4226 | } | |
4227 | ||
fd01b88c | 4228 | static void reap_sync_thread(struct mddev *mddev); |
7ebc0be7 | 4229 | |
24dd469d | 4230 | static ssize_t |
fd01b88c | 4231 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4232 | { |
7eec314d N |
4233 | if (!mddev->pers || !mddev->pers->sync_request) |
4234 | return -EINVAL; | |
4235 | ||
b6a9ce68 N |
4236 | if (cmd_match(page, "frozen")) |
4237 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4238 | else | |
4239 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4240 | ||
4241 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
4242 | if (mddev->sync_thread) { |
4243 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 4244 | reap_sync_thread(mddev); |
7eec314d | 4245 | } |
03c902e1 N |
4246 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
4247 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 4248 | return -EBUSY; |
72a23c21 NB |
4249 | else if (cmd_match(page, "resync")) |
4250 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4251 | else if (cmd_match(page, "recover")) { | |
4252 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 4253 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 4254 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4255 | int err; |
4256 | if (mddev->pers->start_reshape == NULL) | |
4257 | return -EINVAL; | |
4258 | err = mddev->pers->start_reshape(mddev); | |
4259 | if (err) | |
4260 | return err; | |
a99ac971 | 4261 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4262 | } else { |
bce74dac | 4263 | if (cmd_match(page, "check")) |
7eec314d | 4264 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4265 | else if (!cmd_match(page, "repair")) |
7eec314d N |
4266 | return -EINVAL; |
4267 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
4268 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4269 | } |
48c26ddc N |
4270 | if (mddev->ro == 2) { |
4271 | /* A write to sync_action is enough to justify | |
4272 | * canceling read-auto mode | |
4273 | */ | |
4274 | mddev->ro = 0; | |
4275 | md_wakeup_thread(mddev->sync_thread); | |
4276 | } | |
03c902e1 | 4277 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4278 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4279 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4280 | return len; |
4281 | } | |
4282 | ||
9d88883e | 4283 | static ssize_t |
fd01b88c | 4284 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4285 | { |
4286 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4287 | (unsigned long long) |
4288 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4289 | } |
4290 | ||
80ca3a44 N |
4291 | static struct md_sysfs_entry md_scan_mode = |
4292 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 4293 | |
96de1e66 | 4294 | |
80ca3a44 | 4295 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4296 | |
88202a0c | 4297 | static ssize_t |
fd01b88c | 4298 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4299 | { |
4300 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4301 | mddev->sync_speed_min ? "local": "system"); | |
4302 | } | |
4303 | ||
4304 | static ssize_t | |
fd01b88c | 4305 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4306 | { |
4307 | int min; | |
4308 | char *e; | |
4309 | if (strncmp(buf, "system", 6)==0) { | |
4310 | mddev->sync_speed_min = 0; | |
4311 | return len; | |
4312 | } | |
4313 | min = simple_strtoul(buf, &e, 10); | |
4314 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
4315 | return -EINVAL; | |
4316 | mddev->sync_speed_min = min; | |
4317 | return len; | |
4318 | } | |
4319 | ||
4320 | static struct md_sysfs_entry md_sync_min = | |
4321 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4322 | ||
4323 | static ssize_t | |
fd01b88c | 4324 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4325 | { |
4326 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4327 | mddev->sync_speed_max ? "local": "system"); | |
4328 | } | |
4329 | ||
4330 | static ssize_t | |
fd01b88c | 4331 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4332 | { |
4333 | int max; | |
4334 | char *e; | |
4335 | if (strncmp(buf, "system", 6)==0) { | |
4336 | mddev->sync_speed_max = 0; | |
4337 | return len; | |
4338 | } | |
4339 | max = simple_strtoul(buf, &e, 10); | |
4340 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
4341 | return -EINVAL; | |
4342 | mddev->sync_speed_max = max; | |
4343 | return len; | |
4344 | } | |
4345 | ||
4346 | static struct md_sysfs_entry md_sync_max = | |
4347 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4348 | ||
d7f3d291 | 4349 | static ssize_t |
fd01b88c | 4350 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4351 | { |
4352 | return sprintf(page, "%d\n", mddev->degraded); | |
4353 | } | |
4354 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4355 | |
90b08710 | 4356 | static ssize_t |
fd01b88c | 4357 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4358 | { |
4359 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4360 | } | |
4361 | ||
4362 | static ssize_t | |
fd01b88c | 4363 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4364 | { |
4365 | long n; | |
4366 | ||
4367 | if (strict_strtol(buf, 10, &n)) | |
4368 | return -EINVAL; | |
4369 | ||
4370 | if (n != 0 && n != 1) | |
4371 | return -EINVAL; | |
4372 | ||
4373 | mddev->parallel_resync = n; | |
4374 | ||
4375 | if (mddev->sync_thread) | |
4376 | wake_up(&resync_wait); | |
4377 | ||
4378 | return len; | |
4379 | } | |
4380 | ||
4381 | /* force parallel resync, even with shared block devices */ | |
4382 | static struct md_sysfs_entry md_sync_force_parallel = | |
4383 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4384 | sync_force_parallel_show, sync_force_parallel_store); | |
4385 | ||
88202a0c | 4386 | static ssize_t |
fd01b88c | 4387 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4388 | { |
4389 | unsigned long resync, dt, db; | |
d1a7c503 N |
4390 | if (mddev->curr_resync == 0) |
4391 | return sprintf(page, "none\n"); | |
9687a60c AN |
4392 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4393 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4394 | if (!dt) dt++; |
9687a60c AN |
4395 | db = resync - mddev->resync_mark_cnt; |
4396 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4397 | } |
4398 | ||
80ca3a44 | 4399 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4400 | |
4401 | static ssize_t | |
fd01b88c | 4402 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4403 | { |
13ae864b | 4404 | unsigned long long max_sectors, resync; |
88202a0c | 4405 | |
acb180b0 N |
4406 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4407 | return sprintf(page, "none\n"); | |
4408 | ||
72f36d59 N |
4409 | if (mddev->curr_resync == 1 || |
4410 | mddev->curr_resync == 2) | |
4411 | return sprintf(page, "delayed\n"); | |
4412 | ||
c804cdec N |
4413 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4414 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4415 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4416 | else |
58c0fed4 | 4417 | max_sectors = mddev->dev_sectors; |
88202a0c | 4418 | |
acb180b0 | 4419 | resync = mddev->curr_resync_completed; |
13ae864b | 4420 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4421 | } |
4422 | ||
80ca3a44 | 4423 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 4424 | |
5e96ee65 | 4425 | static ssize_t |
fd01b88c | 4426 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4427 | { |
4428 | return sprintf(page, "%llu\n", | |
4429 | (unsigned long long)mddev->resync_min); | |
4430 | } | |
4431 | static ssize_t | |
fd01b88c | 4432 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4433 | { |
4434 | unsigned long long min; | |
4435 | if (strict_strtoull(buf, 10, &min)) | |
4436 | return -EINVAL; | |
4437 | if (min > mddev->resync_max) | |
4438 | return -EINVAL; | |
4439 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4440 | return -EBUSY; | |
4441 | ||
4442 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4443 | if (mddev->chunk_sectors) { |
2ac06c33 | 4444 | sector_t temp = min; |
9d8f0363 | 4445 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
4446 | return -EINVAL; |
4447 | } | |
4448 | mddev->resync_min = min; | |
4449 | ||
4450 | return len; | |
4451 | } | |
4452 | ||
4453 | static struct md_sysfs_entry md_min_sync = | |
4454 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4455 | ||
c6207277 | 4456 | static ssize_t |
fd01b88c | 4457 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4458 | { |
4459 | if (mddev->resync_max == MaxSector) | |
4460 | return sprintf(page, "max\n"); | |
4461 | else | |
4462 | return sprintf(page, "%llu\n", | |
4463 | (unsigned long long)mddev->resync_max); | |
4464 | } | |
4465 | static ssize_t | |
fd01b88c | 4466 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 N |
4467 | { |
4468 | if (strncmp(buf, "max", 3) == 0) | |
4469 | mddev->resync_max = MaxSector; | |
4470 | else { | |
5e96ee65 NB |
4471 | unsigned long long max; |
4472 | if (strict_strtoull(buf, 10, &max)) | |
4473 | return -EINVAL; | |
4474 | if (max < mddev->resync_min) | |
c6207277 N |
4475 | return -EINVAL; |
4476 | if (max < mddev->resync_max && | |
4d484a4a | 4477 | mddev->ro == 0 && |
c6207277 N |
4478 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4479 | return -EBUSY; | |
4480 | ||
4481 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4482 | if (mddev->chunk_sectors) { |
2ac06c33 | 4483 | sector_t temp = max; |
9d8f0363 | 4484 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
4485 | return -EINVAL; |
4486 | } | |
4487 | mddev->resync_max = max; | |
4488 | } | |
4489 | wake_up(&mddev->recovery_wait); | |
4490 | return len; | |
4491 | } | |
4492 | ||
4493 | static struct md_sysfs_entry md_max_sync = | |
4494 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4495 | ||
e464eafd | 4496 | static ssize_t |
fd01b88c | 4497 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4498 | { |
4499 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4500 | } | |
4501 | ||
4502 | static ssize_t | |
fd01b88c | 4503 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4504 | { |
4505 | char *e; | |
4506 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4507 | unsigned long long old = mddev->suspend_lo; |
e464eafd | 4508 | |
b8d966ef N |
4509 | if (mddev->pers == NULL || |
4510 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4511 | return -EINVAL; |
4512 | if (buf == e || (*e && *e != '\n')) | |
4513 | return -EINVAL; | |
23ddff37 N |
4514 | |
4515 | mddev->suspend_lo = new; | |
4516 | if (new >= old) | |
4517 | /* Shrinking suspended region */ | |
e464eafd | 4518 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4519 | else { |
4520 | /* Expanding suspended region - need to wait */ | |
4521 | mddev->pers->quiesce(mddev, 1); | |
4522 | mddev->pers->quiesce(mddev, 0); | |
4523 | } | |
4524 | return len; | |
e464eafd N |
4525 | } |
4526 | static struct md_sysfs_entry md_suspend_lo = | |
4527 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4528 | ||
4529 | ||
4530 | static ssize_t | |
fd01b88c | 4531 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4532 | { |
4533 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4534 | } | |
4535 | ||
4536 | static ssize_t | |
fd01b88c | 4537 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4538 | { |
4539 | char *e; | |
4540 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4541 | unsigned long long old = mddev->suspend_hi; |
e464eafd | 4542 | |
b8d966ef N |
4543 | if (mddev->pers == NULL || |
4544 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4545 | return -EINVAL; |
4546 | if (buf == e || (*e && *e != '\n')) | |
4547 | return -EINVAL; | |
23ddff37 N |
4548 | |
4549 | mddev->suspend_hi = new; | |
4550 | if (new <= old) | |
4551 | /* Shrinking suspended region */ | |
4552 | mddev->pers->quiesce(mddev, 2); | |
4553 | else { | |
4554 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4555 | mddev->pers->quiesce(mddev, 1); |
4556 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 N |
4557 | } |
4558 | return len; | |
e464eafd N |
4559 | } |
4560 | static struct md_sysfs_entry md_suspend_hi = | |
4561 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4562 | ||
08a02ecd | 4563 | static ssize_t |
fd01b88c | 4564 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4565 | { |
4566 | if (mddev->reshape_position != MaxSector) | |
4567 | return sprintf(page, "%llu\n", | |
4568 | (unsigned long long)mddev->reshape_position); | |
4569 | strcpy(page, "none\n"); | |
4570 | return 5; | |
4571 | } | |
4572 | ||
4573 | static ssize_t | |
fd01b88c | 4574 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4575 | { |
c6563a8c | 4576 | struct md_rdev *rdev; |
08a02ecd N |
4577 | char *e; |
4578 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4579 | if (mddev->pers) | |
4580 | return -EBUSY; | |
4581 | if (buf == e || (*e && *e != '\n')) | |
4582 | return -EINVAL; | |
4583 | mddev->reshape_position = new; | |
4584 | mddev->delta_disks = 0; | |
2c810cdd | 4585 | mddev->reshape_backwards = 0; |
08a02ecd N |
4586 | mddev->new_level = mddev->level; |
4587 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4588 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4589 | rdev_for_each(rdev, mddev) |
4590 | rdev->new_data_offset = rdev->data_offset; | |
08a02ecd N |
4591 | return len; |
4592 | } | |
4593 | ||
4594 | static struct md_sysfs_entry md_reshape_position = | |
4595 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4596 | reshape_position_store); | |
4597 | ||
2c810cdd N |
4598 | static ssize_t |
4599 | reshape_direction_show(struct mddev *mddev, char *page) | |
4600 | { | |
4601 | return sprintf(page, "%s\n", | |
4602 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4603 | } | |
4604 | ||
4605 | static ssize_t | |
4606 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4607 | { | |
4608 | int backwards = 0; | |
4609 | if (cmd_match(buf, "forwards")) | |
4610 | backwards = 0; | |
4611 | else if (cmd_match(buf, "backwards")) | |
4612 | backwards = 1; | |
4613 | else | |
4614 | return -EINVAL; | |
4615 | if (mddev->reshape_backwards == backwards) | |
4616 | return len; | |
4617 | ||
4618 | /* check if we are allowed to change */ | |
4619 | if (mddev->delta_disks) | |
4620 | return -EBUSY; | |
4621 | ||
4622 | if (mddev->persistent && | |
4623 | mddev->major_version == 0) | |
4624 | return -EINVAL; | |
4625 | ||
4626 | mddev->reshape_backwards = backwards; | |
4627 | return len; | |
4628 | } | |
4629 | ||
4630 | static struct md_sysfs_entry md_reshape_direction = | |
4631 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4632 | reshape_direction_store); | |
4633 | ||
b522adcd | 4634 | static ssize_t |
fd01b88c | 4635 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4636 | { |
4637 | if (mddev->external_size) | |
4638 | return sprintf(page, "%llu\n", | |
4639 | (unsigned long long)mddev->array_sectors/2); | |
4640 | else | |
4641 | return sprintf(page, "default\n"); | |
4642 | } | |
4643 | ||
4644 | static ssize_t | |
fd01b88c | 4645 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4646 | { |
4647 | sector_t sectors; | |
4648 | ||
4649 | if (strncmp(buf, "default", 7) == 0) { | |
4650 | if (mddev->pers) | |
4651 | sectors = mddev->pers->size(mddev, 0, 0); | |
4652 | else | |
4653 | sectors = mddev->array_sectors; | |
4654 | ||
4655 | mddev->external_size = 0; | |
4656 | } else { | |
4657 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4658 | return -EINVAL; | |
4659 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4660 | return -E2BIG; |
b522adcd DW |
4661 | |
4662 | mddev->external_size = 1; | |
4663 | } | |
4664 | ||
4665 | mddev->array_sectors = sectors; | |
cbe6ef1d N |
4666 | if (mddev->pers) { |
4667 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 4668 | revalidate_disk(mddev->gendisk); |
cbe6ef1d | 4669 | } |
b522adcd DW |
4670 | return len; |
4671 | } | |
4672 | ||
4673 | static struct md_sysfs_entry md_array_size = | |
4674 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4675 | array_size_store); | |
e464eafd | 4676 | |
eae1701f N |
4677 | static struct attribute *md_default_attrs[] = { |
4678 | &md_level.attr, | |
d4dbd025 | 4679 | &md_layout.attr, |
eae1701f | 4680 | &md_raid_disks.attr, |
3b34380a | 4681 | &md_chunk_size.attr, |
a35b0d69 | 4682 | &md_size.attr, |
a94213b1 | 4683 | &md_resync_start.attr, |
8bb93aac | 4684 | &md_metadata.attr, |
6d7ff738 | 4685 | &md_new_device.attr, |
16f17b39 | 4686 | &md_safe_delay.attr, |
9e653b63 | 4687 | &md_array_state.attr, |
08a02ecd | 4688 | &md_reshape_position.attr, |
2c810cdd | 4689 | &md_reshape_direction.attr, |
b522adcd | 4690 | &md_array_size.attr, |
1e50915f | 4691 | &max_corr_read_errors.attr, |
411036fa N |
4692 | NULL, |
4693 | }; | |
4694 | ||
4695 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4696 | &md_scan_mode.attr, |
9d88883e | 4697 | &md_mismatches.attr, |
88202a0c N |
4698 | &md_sync_min.attr, |
4699 | &md_sync_max.attr, | |
4700 | &md_sync_speed.attr, | |
90b08710 | 4701 | &md_sync_force_parallel.attr, |
88202a0c | 4702 | &md_sync_completed.attr, |
5e96ee65 | 4703 | &md_min_sync.attr, |
c6207277 | 4704 | &md_max_sync.attr, |
e464eafd N |
4705 | &md_suspend_lo.attr, |
4706 | &md_suspend_hi.attr, | |
9b1d1dac | 4707 | &md_bitmap.attr, |
d7f3d291 | 4708 | &md_degraded.attr, |
eae1701f N |
4709 | NULL, |
4710 | }; | |
411036fa N |
4711 | static struct attribute_group md_redundancy_group = { |
4712 | .name = NULL, | |
4713 | .attrs = md_redundancy_attrs, | |
4714 | }; | |
4715 | ||
eae1701f N |
4716 | |
4717 | static ssize_t | |
4718 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4719 | { | |
4720 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4721 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4722 | ssize_t rv; |
eae1701f N |
4723 | |
4724 | if (!entry->show) | |
4725 | return -EIO; | |
af8a2434 N |
4726 | spin_lock(&all_mddevs_lock); |
4727 | if (list_empty(&mddev->all_mddevs)) { | |
4728 | spin_unlock(&all_mddevs_lock); | |
4729 | return -EBUSY; | |
4730 | } | |
4731 | mddev_get(mddev); | |
4732 | spin_unlock(&all_mddevs_lock); | |
4733 | ||
5dc5cf7d IM |
4734 | rv = mddev_lock(mddev); |
4735 | if (!rv) { | |
4736 | rv = entry->show(mddev, page); | |
4737 | mddev_unlock(mddev); | |
4738 | } | |
af8a2434 | 4739 | mddev_put(mddev); |
96de1e66 | 4740 | return rv; |
eae1701f N |
4741 | } |
4742 | ||
4743 | static ssize_t | |
4744 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4745 | const char *page, size_t length) | |
4746 | { | |
4747 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4748 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4749 | ssize_t rv; |
eae1701f N |
4750 | |
4751 | if (!entry->store) | |
4752 | return -EIO; | |
67463acb N |
4753 | if (!capable(CAP_SYS_ADMIN)) |
4754 | return -EACCES; | |
af8a2434 N |
4755 | spin_lock(&all_mddevs_lock); |
4756 | if (list_empty(&mddev->all_mddevs)) { | |
4757 | spin_unlock(&all_mddevs_lock); | |
4758 | return -EBUSY; | |
4759 | } | |
4760 | mddev_get(mddev); | |
4761 | spin_unlock(&all_mddevs_lock); | |
a7a3f08d N |
4762 | if (entry->store == new_dev_store) |
4763 | flush_workqueue(md_misc_wq); | |
5dc5cf7d IM |
4764 | rv = mddev_lock(mddev); |
4765 | if (!rv) { | |
4766 | rv = entry->store(mddev, page, length); | |
4767 | mddev_unlock(mddev); | |
4768 | } | |
af8a2434 | 4769 | mddev_put(mddev); |
96de1e66 | 4770 | return rv; |
eae1701f N |
4771 | } |
4772 | ||
4773 | static void md_free(struct kobject *ko) | |
4774 | { | |
fd01b88c | 4775 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4776 | |
4777 | if (mddev->sysfs_state) | |
4778 | sysfs_put(mddev->sysfs_state); | |
4779 | ||
4780 | if (mddev->gendisk) { | |
4781 | del_gendisk(mddev->gendisk); | |
4782 | put_disk(mddev->gendisk); | |
4783 | } | |
4784 | if (mddev->queue) | |
4785 | blk_cleanup_queue(mddev->queue); | |
4786 | ||
eae1701f N |
4787 | kfree(mddev); |
4788 | } | |
4789 | ||
52cf25d0 | 4790 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4791 | .show = md_attr_show, |
4792 | .store = md_attr_store, | |
4793 | }; | |
4794 | static struct kobj_type md_ktype = { | |
4795 | .release = md_free, | |
4796 | .sysfs_ops = &md_sysfs_ops, | |
4797 | .default_attrs = md_default_attrs, | |
4798 | }; | |
4799 | ||
1da177e4 LT |
4800 | int mdp_major = 0; |
4801 | ||
5fd3a17e DW |
4802 | static void mddev_delayed_delete(struct work_struct *ws) |
4803 | { | |
fd01b88c | 4804 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4805 | |
43a70507 | 4806 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4807 | kobject_del(&mddev->kobj); |
4808 | kobject_put(&mddev->kobj); | |
4809 | } | |
4810 | ||
efeb53c0 | 4811 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4812 | { |
48c9c27b | 4813 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 4814 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 4815 | struct gendisk *disk; |
efeb53c0 N |
4816 | int partitioned; |
4817 | int shift; | |
4818 | int unit; | |
3830c62f | 4819 | int error; |
1da177e4 LT |
4820 | |
4821 | if (!mddev) | |
efeb53c0 N |
4822 | return -ENODEV; |
4823 | ||
4824 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4825 | shift = partitioned ? MdpMinorShift : 0; | |
4826 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4827 | |
e804ac78 TH |
4828 | /* wait for any previous instance of this device to be |
4829 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4830 | */ |
e804ac78 | 4831 | flush_workqueue(md_misc_wq); |
d3374825 | 4832 | |
48c9c27b | 4833 | mutex_lock(&disks_mutex); |
0909dc44 N |
4834 | error = -EEXIST; |
4835 | if (mddev->gendisk) | |
4836 | goto abort; | |
efeb53c0 N |
4837 | |
4838 | if (name) { | |
4839 | /* Need to ensure that 'name' is not a duplicate. | |
4840 | */ | |
fd01b88c | 4841 | struct mddev *mddev2; |
efeb53c0 N |
4842 | spin_lock(&all_mddevs_lock); |
4843 | ||
4844 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4845 | if (mddev2->gendisk && | |
4846 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4847 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4848 | goto abort; |
efeb53c0 N |
4849 | } |
4850 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4851 | } |
8b765398 | 4852 | |
0909dc44 | 4853 | error = -ENOMEM; |
8b765398 | 4854 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4855 | if (!mddev->queue) |
4856 | goto abort; | |
409c57f3 N |
4857 | mddev->queue->queuedata = mddev; |
4858 | ||
409c57f3 | 4859 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 4860 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 4861 | |
1da177e4 LT |
4862 | disk = alloc_disk(1 << shift); |
4863 | if (!disk) { | |
8b765398 N |
4864 | blk_cleanup_queue(mddev->queue); |
4865 | mddev->queue = NULL; | |
0909dc44 | 4866 | goto abort; |
1da177e4 | 4867 | } |
efeb53c0 | 4868 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4869 | disk->first_minor = unit << shift; |
efeb53c0 N |
4870 | if (name) |
4871 | strcpy(disk->disk_name, name); | |
4872 | else if (partitioned) | |
1da177e4 | 4873 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4874 | else |
1da177e4 | 4875 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4876 | disk->fops = &md_fops; |
4877 | disk->private_data = mddev; | |
4878 | disk->queue = mddev->queue; | |
b0140891 | 4879 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4880 | /* Allow extended partitions. This makes the |
d3374825 | 4881 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4882 | * remove it now. |
4883 | */ | |
4884 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4885 | mddev->gendisk = disk; |
b0140891 N |
4886 | /* As soon as we call add_disk(), another thread could get |
4887 | * through to md_open, so make sure it doesn't get too far | |
4888 | */ | |
4889 | mutex_lock(&mddev->open_mutex); | |
4890 | add_disk(disk); | |
4891 | ||
ed9e1982 TH |
4892 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4893 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4894 | if (error) { |
4895 | /* This isn't possible, but as kobject_init_and_add is marked | |
4896 | * __must_check, we must do something with the result | |
4897 | */ | |
5e55e2f5 N |
4898 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4899 | disk->disk_name); | |
0909dc44 N |
4900 | error = 0; |
4901 | } | |
00bcb4ac N |
4902 | if (mddev->kobj.sd && |
4903 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4904 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4905 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4906 | abort: |
4907 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4908 | if (!error && mddev->kobj.sd) { |
3830c62f | 4909 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4910 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4911 | } |
d3374825 | 4912 | mddev_put(mddev); |
0909dc44 | 4913 | return error; |
efeb53c0 N |
4914 | } |
4915 | ||
4916 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4917 | { | |
4918 | md_alloc(dev, NULL); | |
1da177e4 LT |
4919 | return NULL; |
4920 | } | |
4921 | ||
efeb53c0 N |
4922 | static int add_named_array(const char *val, struct kernel_param *kp) |
4923 | { | |
4924 | /* val must be "md_*" where * is not all digits. | |
4925 | * We allocate an array with a large free minor number, and | |
4926 | * set the name to val. val must not already be an active name. | |
4927 | */ | |
4928 | int len = strlen(val); | |
4929 | char buf[DISK_NAME_LEN]; | |
4930 | ||
4931 | while (len && val[len-1] == '\n') | |
4932 | len--; | |
4933 | if (len >= DISK_NAME_LEN) | |
4934 | return -E2BIG; | |
4935 | strlcpy(buf, val, len+1); | |
4936 | if (strncmp(buf, "md_", 3) != 0) | |
4937 | return -EINVAL; | |
4938 | return md_alloc(0, buf); | |
4939 | } | |
4940 | ||
1da177e4 LT |
4941 | static void md_safemode_timeout(unsigned long data) |
4942 | { | |
fd01b88c | 4943 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 4944 | |
0fd62b86 NB |
4945 | if (!atomic_read(&mddev->writes_pending)) { |
4946 | mddev->safemode = 1; | |
4947 | if (mddev->external) | |
00bcb4ac | 4948 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4949 | } |
1da177e4 LT |
4950 | md_wakeup_thread(mddev->thread); |
4951 | } | |
4952 | ||
6ff8d8ec | 4953 | static int start_dirty_degraded; |
1da177e4 | 4954 | |
fd01b88c | 4955 | int md_run(struct mddev *mddev) |
1da177e4 | 4956 | { |
2604b703 | 4957 | int err; |
3cb03002 | 4958 | struct md_rdev *rdev; |
84fc4b56 | 4959 | struct md_personality *pers; |
1da177e4 | 4960 | |
a757e64c N |
4961 | if (list_empty(&mddev->disks)) |
4962 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4963 | return -EINVAL; |
1da177e4 LT |
4964 | |
4965 | if (mddev->pers) | |
4966 | return -EBUSY; | |
bb4f1e9d N |
4967 | /* Cannot run until previous stop completes properly */ |
4968 | if (mddev->sysfs_active) | |
4969 | return -EBUSY; | |
b6eb127d | 4970 | |
1da177e4 LT |
4971 | /* |
4972 | * Analyze all RAID superblock(s) | |
4973 | */ | |
1ec4a939 N |
4974 | if (!mddev->raid_disks) { |
4975 | if (!mddev->persistent) | |
4976 | return -EINVAL; | |
a757e64c | 4977 | analyze_sbs(mddev); |
1ec4a939 | 4978 | } |
1da177e4 | 4979 | |
d9d166c2 N |
4980 | if (mddev->level != LEVEL_NONE) |
4981 | request_module("md-level-%d", mddev->level); | |
4982 | else if (mddev->clevel[0]) | |
4983 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4984 | |
4985 | /* | |
4986 | * Drop all container device buffers, from now on | |
4987 | * the only valid external interface is through the md | |
4988 | * device. | |
1da177e4 | 4989 | */ |
dafb20fa | 4990 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 4991 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4992 | continue; |
4993 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4994 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4995 | |
4996 | /* perform some consistency tests on the device. | |
4997 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4998 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4999 | */ |
a6ff7e08 JB |
5000 | if (rdev->meta_bdev) { |
5001 | /* Nothing to check */; | |
5002 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5003 | if (mddev->dev_sectors && |
5004 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5005 | > rdev->sb_start) { |
f0d76d70 N |
5006 | printk("md: %s: data overlaps metadata\n", |
5007 | mdname(mddev)); | |
5008 | return -EINVAL; | |
5009 | } | |
5010 | } else { | |
0f420358 | 5011 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
5012 | > rdev->data_offset) { |
5013 | printk("md: %s: metadata overlaps data\n", | |
5014 | mdname(mddev)); | |
5015 | return -EINVAL; | |
5016 | } | |
5017 | } | |
00bcb4ac | 5018 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5019 | } |
5020 | ||
a167f663 | 5021 | if (mddev->bio_set == NULL) |
395c72a7 | 5022 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
a167f663 | 5023 | |
1da177e4 | 5024 | spin_lock(&pers_lock); |
d9d166c2 | 5025 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5026 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5027 | spin_unlock(&pers_lock); |
d9d166c2 N |
5028 | if (mddev->level != LEVEL_NONE) |
5029 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
5030 | mddev->level); | |
5031 | else | |
5032 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
5033 | mddev->clevel); | |
1da177e4 LT |
5034 | return -EINVAL; |
5035 | } | |
2604b703 | 5036 | mddev->pers = pers; |
1da177e4 | 5037 | spin_unlock(&pers_lock); |
34817e8c N |
5038 | if (mddev->level != pers->level) { |
5039 | mddev->level = pers->level; | |
5040 | mddev->new_level = pers->level; | |
5041 | } | |
d9d166c2 | 5042 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5043 | |
f6705578 | 5044 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5045 | pers->start_reshape == NULL) { |
f6705578 N |
5046 | /* This personality cannot handle reshaping... */ |
5047 | mddev->pers = NULL; | |
5048 | module_put(pers->owner); | |
5049 | return -EINVAL; | |
5050 | } | |
5051 | ||
7dd5e7c3 N |
5052 | if (pers->sync_request) { |
5053 | /* Warn if this is a potentially silly | |
5054 | * configuration. | |
5055 | */ | |
5056 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5057 | struct md_rdev *rdev2; |
7dd5e7c3 | 5058 | int warned = 0; |
159ec1fc | 5059 | |
dafb20fa N |
5060 | rdev_for_each(rdev, mddev) |
5061 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5062 | if (rdev < rdev2 && |
5063 | rdev->bdev->bd_contains == | |
5064 | rdev2->bdev->bd_contains) { | |
5065 | printk(KERN_WARNING | |
5066 | "%s: WARNING: %s appears to be" | |
5067 | " on the same physical disk as" | |
5068 | " %s.\n", | |
5069 | mdname(mddev), | |
5070 | bdevname(rdev->bdev,b), | |
5071 | bdevname(rdev2->bdev,b2)); | |
5072 | warned = 1; | |
5073 | } | |
5074 | } | |
159ec1fc | 5075 | |
7dd5e7c3 N |
5076 | if (warned) |
5077 | printk(KERN_WARNING | |
5078 | "True protection against single-disk" | |
5079 | " failure might be compromised.\n"); | |
5080 | } | |
5081 | ||
657390d2 | 5082 | mddev->recovery = 0; |
58c0fed4 AN |
5083 | /* may be over-ridden by personality */ |
5084 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5085 | ||
6ff8d8ec | 5086 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5087 | |
0f9552b5 | 5088 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5089 | mddev->ro = 2; /* read-only, but switch on first write */ |
5090 | ||
b15c2e57 | 5091 | err = mddev->pers->run(mddev); |
13e53df3 AN |
5092 | if (err) |
5093 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
5094 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
5095 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
5096 | " but 'external_size' not in effect?\n", __func__); | |
5097 | printk(KERN_ERR | |
5098 | "md: invalid array_size %llu > default size %llu\n", | |
5099 | (unsigned long long)mddev->array_sectors / 2, | |
5100 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
5101 | err = -EINVAL; | |
5102 | mddev->pers->stop(mddev); | |
5103 | } | |
ef99bf48 N |
5104 | if (err == 0 && mddev->pers->sync_request && |
5105 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { | |
b15c2e57 N |
5106 | err = bitmap_create(mddev); |
5107 | if (err) { | |
5108 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
5109 | mdname(mddev), err); | |
5110 | mddev->pers->stop(mddev); | |
5111 | } | |
5112 | } | |
1da177e4 | 5113 | if (err) { |
1da177e4 LT |
5114 | module_put(mddev->pers->owner); |
5115 | mddev->pers = NULL; | |
32a7627c N |
5116 | bitmap_destroy(mddev); |
5117 | return err; | |
1da177e4 | 5118 | } |
5e55e2f5 | 5119 | if (mddev->pers->sync_request) { |
00bcb4ac N |
5120 | if (mddev->kobj.sd && |
5121 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
5122 | printk(KERN_WARNING |
5123 | "md: cannot register extra attributes for %s\n", | |
5124 | mdname(mddev)); | |
00bcb4ac | 5125 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5126 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5127 | mddev->ro = 0; |
5128 | ||
1da177e4 | 5129 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
5130 | atomic_set(&mddev->max_corr_read_errors, |
5131 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
5132 | mddev->safemode = 0; |
5133 | mddev->safemode_timer.function = md_safemode_timeout; | |
5134 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 5135 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 5136 | mddev->in_sync = 1; |
0ca69886 N |
5137 | smp_wmb(); |
5138 | mddev->ready = 1; | |
dafb20fa | 5139 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5140 | if (rdev->raid_disk >= 0) |
5141 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5142 | /* failure here is OK */; |
1da177e4 LT |
5143 | |
5144 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5145 | ||
850b2b42 N |
5146 | if (mddev->flags) |
5147 | md_update_sb(mddev, 0); | |
1da177e4 | 5148 | |
d7603b7e | 5149 | md_new_event(mddev); |
00bcb4ac N |
5150 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5151 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5152 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5153 | return 0; |
5154 | } | |
390ee602 | 5155 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5156 | |
fd01b88c | 5157 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5158 | { |
5159 | int err; | |
5160 | ||
5161 | err = md_run(mddev); | |
5162 | if (err) | |
5163 | goto out; | |
69e51b44 N |
5164 | err = bitmap_load(mddev); |
5165 | if (err) { | |
5166 | bitmap_destroy(mddev); | |
5167 | goto out; | |
5168 | } | |
0fd018af JB |
5169 | |
5170 | md_wakeup_thread(mddev->thread); | |
5171 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5172 | ||
fe60b014 N |
5173 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5174 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5175 | mddev->changed = 1; |
fe60b014 N |
5176 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5177 | out: | |
5178 | return err; | |
5179 | } | |
5180 | ||
fd01b88c | 5181 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5182 | { |
5183 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 5184 | |
80fab1d7 | 5185 | /* Complain if it has no devices */ |
1da177e4 | 5186 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5187 | return -ENXIO; |
5188 | if (!mddev->pers) | |
5189 | return -EINVAL; | |
5190 | if (!mddev->ro) | |
5191 | return -EBUSY; | |
5192 | mddev->safemode = 0; | |
5193 | mddev->ro = 0; | |
5194 | set_disk_ro(disk, 0); | |
5195 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
5196 | mdname(mddev)); | |
5197 | /* Kick recovery or resync if necessary */ | |
5198 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5199 | md_wakeup_thread(mddev->thread); | |
5200 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5201 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5202 | return 0; |
1da177e4 LT |
5203 | } |
5204 | ||
acc55e22 N |
5205 | /* similar to deny_write_access, but accounts for our holding a reference |
5206 | * to the file ourselves */ | |
5207 | static int deny_bitmap_write_access(struct file * file) | |
5208 | { | |
5209 | struct inode *inode = file->f_mapping->host; | |
5210 | ||
5211 | spin_lock(&inode->i_lock); | |
5212 | if (atomic_read(&inode->i_writecount) > 1) { | |
5213 | spin_unlock(&inode->i_lock); | |
5214 | return -ETXTBSY; | |
5215 | } | |
5216 | atomic_set(&inode->i_writecount, -1); | |
5217 | spin_unlock(&inode->i_lock); | |
5218 | ||
5219 | return 0; | |
5220 | } | |
5221 | ||
43a70507 | 5222 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
5223 | { |
5224 | struct inode *inode = file->f_mapping->host; | |
5225 | ||
5226 | spin_lock(&inode->i_lock); | |
5227 | atomic_set(&inode->i_writecount, 1); | |
5228 | spin_unlock(&inode->i_lock); | |
5229 | } | |
5230 | ||
fd01b88c | 5231 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5232 | { |
5233 | mddev->array_sectors = 0; | |
5234 | mddev->external_size = 0; | |
5235 | mddev->dev_sectors = 0; | |
5236 | mddev->raid_disks = 0; | |
5237 | mddev->recovery_cp = 0; | |
5238 | mddev->resync_min = 0; | |
5239 | mddev->resync_max = MaxSector; | |
5240 | mddev->reshape_position = MaxSector; | |
5241 | mddev->external = 0; | |
5242 | mddev->persistent = 0; | |
5243 | mddev->level = LEVEL_NONE; | |
5244 | mddev->clevel[0] = 0; | |
5245 | mddev->flags = 0; | |
5246 | mddev->ro = 0; | |
5247 | mddev->metadata_type[0] = 0; | |
5248 | mddev->chunk_sectors = 0; | |
5249 | mddev->ctime = mddev->utime = 0; | |
5250 | mddev->layout = 0; | |
5251 | mddev->max_disks = 0; | |
5252 | mddev->events = 0; | |
a8707c08 | 5253 | mddev->can_decrease_events = 0; |
6177b472 | 5254 | mddev->delta_disks = 0; |
2c810cdd | 5255 | mddev->reshape_backwards = 0; |
6177b472 N |
5256 | mddev->new_level = LEVEL_NONE; |
5257 | mddev->new_layout = 0; | |
5258 | mddev->new_chunk_sectors = 0; | |
5259 | mddev->curr_resync = 0; | |
7f7583d4 | 5260 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5261 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5262 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5263 | mddev->recovery = 0; | |
5264 | mddev->in_sync = 0; | |
f0b4f7e2 | 5265 | mddev->changed = 0; |
6177b472 | 5266 | mddev->degraded = 0; |
6177b472 | 5267 | mddev->safemode = 0; |
050b6615 | 5268 | mddev->merge_check_needed = 0; |
6177b472 N |
5269 | mddev->bitmap_info.offset = 0; |
5270 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5271 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5272 | mddev->bitmap_info.chunksize = 0; |
5273 | mddev->bitmap_info.daemon_sleep = 0; | |
5274 | mddev->bitmap_info.max_write_behind = 0; | |
5275 | } | |
5276 | ||
fd01b88c | 5277 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 N |
5278 | { |
5279 | if (mddev->sync_thread) { | |
5280 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5281 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 5282 | reap_sync_thread(mddev); |
a047e125 N |
5283 | } |
5284 | ||
5285 | del_timer_sync(&mddev->safemode_timer); | |
5286 | ||
5287 | bitmap_flush(mddev); | |
5288 | md_super_wait(mddev); | |
5289 | ||
5290 | if (!mddev->in_sync || mddev->flags) { | |
5291 | /* mark array as shutdown cleanly */ | |
5292 | mddev->in_sync = 1; | |
5293 | md_update_sb(mddev, 1); | |
5294 | } | |
5295 | } | |
defad61a | 5296 | |
fd01b88c | 5297 | void md_stop_writes(struct mddev *mddev) |
defad61a N |
5298 | { |
5299 | mddev_lock(mddev); | |
5300 | __md_stop_writes(mddev); | |
5301 | mddev_unlock(mddev); | |
5302 | } | |
390ee602 | 5303 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5304 | |
5eff3c43 | 5305 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5306 | { |
0ca69886 | 5307 | mddev->ready = 0; |
6177b472 N |
5308 | mddev->pers->stop(mddev); |
5309 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
5310 | mddev->to_remove = &md_redundancy_group; | |
5311 | module_put(mddev->pers->owner); | |
5312 | mddev->pers = NULL; | |
cca9cf90 | 5313 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5314 | } |
5eff3c43 N |
5315 | |
5316 | void md_stop(struct mddev *mddev) | |
5317 | { | |
5318 | /* stop the array and free an attached data structures. | |
5319 | * This is called from dm-raid | |
5320 | */ | |
5321 | __md_stop(mddev); | |
5322 | bitmap_destroy(mddev); | |
5323 | if (mddev->bio_set) | |
5324 | bioset_free(mddev->bio_set); | |
5325 | } | |
5326 | ||
390ee602 | 5327 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5328 | |
a05b7ea0 | 5329 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5330 | { |
5331 | int err = 0; | |
5332 | mutex_lock(&mddev->open_mutex); | |
a05b7ea0 | 5333 | if (atomic_read(&mddev->openers) > !!bdev) { |
a4bd82d0 N |
5334 | printk("md: %s still in use.\n",mdname(mddev)); |
5335 | err = -EBUSY; | |
5336 | goto out; | |
5337 | } | |
a05b7ea0 N |
5338 | if (bdev) |
5339 | sync_blockdev(bdev); | |
a4bd82d0 | 5340 | if (mddev->pers) { |
defad61a | 5341 | __md_stop_writes(mddev); |
a4bd82d0 N |
5342 | |
5343 | err = -ENXIO; | |
5344 | if (mddev->ro==1) | |
5345 | goto out; | |
5346 | mddev->ro = 1; | |
5347 | set_disk_ro(mddev->gendisk, 1); | |
5348 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 5349 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
a4bd82d0 N |
5350 | err = 0; |
5351 | } | |
5352 | out: | |
5353 | mutex_unlock(&mddev->open_mutex); | |
5354 | return err; | |
5355 | } | |
5356 | ||
9e653b63 N |
5357 | /* mode: |
5358 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5359 | * 2 - stop but do not disassemble array |
5360 | */ | |
a05b7ea0 N |
5361 | static int do_md_stop(struct mddev * mddev, int mode, |
5362 | struct block_device *bdev) | |
1da177e4 | 5363 | { |
1da177e4 | 5364 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5365 | struct md_rdev *rdev; |
1da177e4 | 5366 | |
c8c00a69 | 5367 | mutex_lock(&mddev->open_mutex); |
a05b7ea0 | 5368 | if (atomic_read(&mddev->openers) > !!bdev || |
bb4f1e9d | 5369 | mddev->sysfs_active) { |
df5b20cf | 5370 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 N |
5371 | mutex_unlock(&mddev->open_mutex); |
5372 | return -EBUSY; | |
5373 | } | |
a05b7ea0 N |
5374 | if (bdev) |
5375 | /* It is possible IO was issued on some other | |
5376 | * open file which was closed before we took ->open_mutex. | |
5377 | * As that was not the last close __blkdev_put will not | |
5378 | * have called sync_blockdev, so we must. | |
5379 | */ | |
5380 | sync_blockdev(bdev); | |
1da177e4 | 5381 | |
6e17b027 | 5382 | if (mddev->pers) { |
a4bd82d0 N |
5383 | if (mddev->ro) |
5384 | set_disk_ro(disk, 0); | |
409c57f3 | 5385 | |
defad61a | 5386 | __md_stop_writes(mddev); |
5eff3c43 | 5387 | __md_stop(mddev); |
a4bd82d0 | 5388 | mddev->queue->merge_bvec_fn = NULL; |
a4bd82d0 | 5389 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 5390 | |
a4bd82d0 | 5391 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5392 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5393 | |
dafb20fa | 5394 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5395 | if (rdev->raid_disk >= 0) |
5396 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5397 | |
a4bd82d0 | 5398 | set_capacity(disk, 0); |
6e17b027 | 5399 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5400 | mddev->changed = 1; |
a4bd82d0 | 5401 | revalidate_disk(disk); |
0d4ca600 | 5402 | |
a4bd82d0 N |
5403 | if (mddev->ro) |
5404 | mddev->ro = 0; | |
6e17b027 N |
5405 | } else |
5406 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5407 | /* |
5408 | * Free resources if final stop | |
5409 | */ | |
9e653b63 | 5410 | if (mode == 0) { |
1da177e4 LT |
5411 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
5412 | ||
978f946b | 5413 | bitmap_destroy(mddev); |
c3d9714e N |
5414 | if (mddev->bitmap_info.file) { |
5415 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5416 | fput(mddev->bitmap_info.file); | |
5417 | mddev->bitmap_info.file = NULL; | |
978f946b | 5418 | } |
c3d9714e | 5419 | mddev->bitmap_info.offset = 0; |
978f946b | 5420 | |
1da177e4 LT |
5421 | export_array(mddev); |
5422 | ||
6177b472 | 5423 | md_clean(mddev); |
934d9c23 | 5424 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
5425 | if (mddev->hold_active == UNTIL_STOP) |
5426 | mddev->hold_active = 0; | |
a4bd82d0 | 5427 | } |
3f9d99c1 | 5428 | blk_integrity_unregister(disk); |
d7603b7e | 5429 | md_new_event(mddev); |
00bcb4ac | 5430 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5431 | return 0; |
1da177e4 LT |
5432 | } |
5433 | ||
fdee8ae4 | 5434 | #ifndef MODULE |
fd01b88c | 5435 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5436 | { |
3cb03002 | 5437 | struct md_rdev *rdev; |
1da177e4 LT |
5438 | int err; |
5439 | ||
a757e64c | 5440 | if (list_empty(&mddev->disks)) |
1da177e4 | 5441 | return; |
1da177e4 LT |
5442 | |
5443 | printk(KERN_INFO "md: running: "); | |
5444 | ||
dafb20fa | 5445 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
5446 | char b[BDEVNAME_SIZE]; |
5447 | printk("<%s>", bdevname(rdev->bdev,b)); | |
5448 | } | |
5449 | printk("\n"); | |
5450 | ||
d710e138 | 5451 | err = do_md_run(mddev); |
1da177e4 LT |
5452 | if (err) { |
5453 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
a05b7ea0 | 5454 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5455 | } |
5456 | } | |
5457 | ||
5458 | /* | |
5459 | * lets try to run arrays based on all disks that have arrived | |
5460 | * until now. (those are in pending_raid_disks) | |
5461 | * | |
5462 | * the method: pick the first pending disk, collect all disks with | |
5463 | * the same UUID, remove all from the pending list and put them into | |
5464 | * the 'same_array' list. Then order this list based on superblock | |
5465 | * update time (freshest comes first), kick out 'old' disks and | |
5466 | * compare superblocks. If everything's fine then run it. | |
5467 | * | |
5468 | * If "unit" is allocated, then bump its reference count | |
5469 | */ | |
5470 | static void autorun_devices(int part) | |
5471 | { | |
3cb03002 | 5472 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5473 | struct mddev *mddev; |
1da177e4 LT |
5474 | char b[BDEVNAME_SIZE]; |
5475 | ||
5476 | printk(KERN_INFO "md: autorun ...\n"); | |
5477 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 5478 | int unit; |
1da177e4 | 5479 | dev_t dev; |
ad01c9e3 | 5480 | LIST_HEAD(candidates); |
1da177e4 | 5481 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5482 | struct md_rdev, same_set); |
1da177e4 LT |
5483 | |
5484 | printk(KERN_INFO "md: considering %s ...\n", | |
5485 | bdevname(rdev0->bdev,b)); | |
5486 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 5487 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
5488 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
5489 | printk(KERN_INFO "md: adding %s ...\n", | |
5490 | bdevname(rdev->bdev,b)); | |
5491 | list_move(&rdev->same_set, &candidates); | |
5492 | } | |
5493 | /* | |
5494 | * now we have a set of devices, with all of them having | |
5495 | * mostly sane superblocks. It's time to allocate the | |
5496 | * mddev. | |
5497 | */ | |
e8703fe1 N |
5498 | if (part) { |
5499 | dev = MKDEV(mdp_major, | |
5500 | rdev0->preferred_minor << MdpMinorShift); | |
5501 | unit = MINOR(dev) >> MdpMinorShift; | |
5502 | } else { | |
5503 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5504 | unit = MINOR(dev); | |
5505 | } | |
5506 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
5507 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
5508 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
5509 | break; | |
5510 | } | |
1da177e4 LT |
5511 | |
5512 | md_probe(dev, NULL, NULL); | |
5513 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5514 | if (!mddev || !mddev->gendisk) { |
5515 | if (mddev) | |
5516 | mddev_put(mddev); | |
5517 | printk(KERN_ERR | |
1da177e4 LT |
5518 | "md: cannot allocate memory for md drive.\n"); |
5519 | break; | |
5520 | } | |
5521 | if (mddev_lock(mddev)) | |
5522 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
5523 | mdname(mddev)); | |
5524 | else if (mddev->raid_disks || mddev->major_version | |
5525 | || !list_empty(&mddev->disks)) { | |
5526 | printk(KERN_WARNING | |
5527 | "md: %s already running, cannot run %s\n", | |
5528 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
5529 | mddev_unlock(mddev); | |
5530 | } else { | |
5531 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 5532 | mddev->persistent = 1; |
159ec1fc | 5533 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5534 | list_del_init(&rdev->same_set); |
5535 | if (bind_rdev_to_array(rdev, mddev)) | |
5536 | export_rdev(rdev); | |
5537 | } | |
5538 | autorun_array(mddev); | |
5539 | mddev_unlock(mddev); | |
5540 | } | |
5541 | /* on success, candidates will be empty, on error | |
5542 | * it won't... | |
5543 | */ | |
159ec1fc | 5544 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5545 | list_del_init(&rdev->same_set); |
1da177e4 | 5546 | export_rdev(rdev); |
4b80991c | 5547 | } |
1da177e4 LT |
5548 | mddev_put(mddev); |
5549 | } | |
5550 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5551 | } | |
fdee8ae4 | 5552 | #endif /* !MODULE */ |
1da177e4 | 5553 | |
1da177e4 LT |
5554 | static int get_version(void __user * arg) |
5555 | { | |
5556 | mdu_version_t ver; | |
5557 | ||
5558 | ver.major = MD_MAJOR_VERSION; | |
5559 | ver.minor = MD_MINOR_VERSION; | |
5560 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5561 | ||
5562 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5563 | return -EFAULT; | |
5564 | ||
5565 | return 0; | |
5566 | } | |
5567 | ||
fd01b88c | 5568 | static int get_array_info(struct mddev * mddev, void __user * arg) |
1da177e4 LT |
5569 | { |
5570 | mdu_array_info_t info; | |
a9f326eb | 5571 | int nr,working,insync,failed,spare; |
3cb03002 | 5572 | struct md_rdev *rdev; |
1da177e4 | 5573 | |
1ca69c4b N |
5574 | nr = working = insync = failed = spare = 0; |
5575 | rcu_read_lock(); | |
5576 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5577 | nr++; |
b2d444d7 | 5578 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5579 | failed++; |
5580 | else { | |
5581 | working++; | |
b2d444d7 | 5582 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 5583 | insync++; |
1da177e4 LT |
5584 | else |
5585 | spare++; | |
5586 | } | |
5587 | } | |
1ca69c4b | 5588 | rcu_read_unlock(); |
1da177e4 LT |
5589 | |
5590 | info.major_version = mddev->major_version; | |
5591 | info.minor_version = mddev->minor_version; | |
5592 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5593 | info.ctime = mddev->ctime; | |
5594 | info.level = mddev->level; | |
58c0fed4 AN |
5595 | info.size = mddev->dev_sectors / 2; |
5596 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5597 | info.size = -1; |
1da177e4 LT |
5598 | info.nr_disks = nr; |
5599 | info.raid_disks = mddev->raid_disks; | |
5600 | info.md_minor = mddev->md_minor; | |
5601 | info.not_persistent= !mddev->persistent; | |
5602 | ||
5603 | info.utime = mddev->utime; | |
5604 | info.state = 0; | |
5605 | if (mddev->in_sync) | |
5606 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5607 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5608 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5609 | info.active_disks = insync; |
1da177e4 LT |
5610 | info.working_disks = working; |
5611 | info.failed_disks = failed; | |
5612 | info.spare_disks = spare; | |
5613 | ||
5614 | info.layout = mddev->layout; | |
9d8f0363 | 5615 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5616 | |
5617 | if (copy_to_user(arg, &info, sizeof(info))) | |
5618 | return -EFAULT; | |
5619 | ||
5620 | return 0; | |
5621 | } | |
5622 | ||
fd01b88c | 5623 | static int get_bitmap_file(struct mddev * mddev, void __user * arg) |
32a7627c N |
5624 | { |
5625 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5626 | char *ptr, *buf = NULL; | |
5627 | int err = -ENOMEM; | |
5628 | ||
b5470dc5 DW |
5629 | if (md_allow_write(mddev)) |
5630 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
5631 | else | |
5632 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 5633 | |
32a7627c N |
5634 | if (!file) |
5635 | goto out; | |
5636 | ||
5637 | /* bitmap disabled, zero the first byte and copy out */ | |
1ec885cd | 5638 | if (!mddev->bitmap || !mddev->bitmap->storage.file) { |
32a7627c N |
5639 | file->pathname[0] = '\0'; |
5640 | goto copy_out; | |
5641 | } | |
5642 | ||
5643 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5644 | if (!buf) | |
5645 | goto out; | |
5646 | ||
1ec885cd N |
5647 | ptr = d_path(&mddev->bitmap->storage.file->f_path, |
5648 | buf, sizeof(file->pathname)); | |
6bcfd601 | 5649 | if (IS_ERR(ptr)) |
32a7627c N |
5650 | goto out; |
5651 | ||
5652 | strcpy(file->pathname, ptr); | |
5653 | ||
5654 | copy_out: | |
5655 | err = 0; | |
5656 | if (copy_to_user(arg, file, sizeof(*file))) | |
5657 | err = -EFAULT; | |
5658 | out: | |
5659 | kfree(buf); | |
5660 | kfree(file); | |
5661 | return err; | |
5662 | } | |
5663 | ||
fd01b88c | 5664 | static int get_disk_info(struct mddev * mddev, void __user * arg) |
1da177e4 LT |
5665 | { |
5666 | mdu_disk_info_t info; | |
3cb03002 | 5667 | struct md_rdev *rdev; |
1da177e4 LT |
5668 | |
5669 | if (copy_from_user(&info, arg, sizeof(info))) | |
5670 | return -EFAULT; | |
5671 | ||
1ca69c4b N |
5672 | rcu_read_lock(); |
5673 | rdev = find_rdev_nr_rcu(mddev, info.number); | |
1da177e4 LT |
5674 | if (rdev) { |
5675 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5676 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5677 | info.raid_disk = rdev->raid_disk; | |
5678 | info.state = 0; | |
b2d444d7 | 5679 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5680 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5681 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5682 | info.state |= (1<<MD_DISK_ACTIVE); |
5683 | info.state |= (1<<MD_DISK_SYNC); | |
5684 | } | |
8ddf9efe N |
5685 | if (test_bit(WriteMostly, &rdev->flags)) |
5686 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5687 | } else { |
5688 | info.major = info.minor = 0; | |
5689 | info.raid_disk = -1; | |
5690 | info.state = (1<<MD_DISK_REMOVED); | |
5691 | } | |
1ca69c4b | 5692 | rcu_read_unlock(); |
1da177e4 LT |
5693 | |
5694 | if (copy_to_user(arg, &info, sizeof(info))) | |
5695 | return -EFAULT; | |
5696 | ||
5697 | return 0; | |
5698 | } | |
5699 | ||
fd01b88c | 5700 | static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info) |
1da177e4 LT |
5701 | { |
5702 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5703 | struct md_rdev *rdev; |
1da177e4 LT |
5704 | dev_t dev = MKDEV(info->major,info->minor); |
5705 | ||
5706 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5707 | return -EOVERFLOW; | |
5708 | ||
5709 | if (!mddev->raid_disks) { | |
5710 | int err; | |
5711 | /* expecting a device which has a superblock */ | |
5712 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5713 | if (IS_ERR(rdev)) { | |
5714 | printk(KERN_WARNING | |
5715 | "md: md_import_device returned %ld\n", | |
5716 | PTR_ERR(rdev)); | |
5717 | return PTR_ERR(rdev); | |
5718 | } | |
5719 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
5720 | struct md_rdev *rdev0 |
5721 | = list_entry(mddev->disks.next, | |
5722 | struct md_rdev, same_set); | |
a9f326eb | 5723 | err = super_types[mddev->major_version] |
1da177e4 LT |
5724 | .load_super(rdev, rdev0, mddev->minor_version); |
5725 | if (err < 0) { | |
5726 | printk(KERN_WARNING | |
5727 | "md: %s has different UUID to %s\n", | |
5728 | bdevname(rdev->bdev,b), | |
5729 | bdevname(rdev0->bdev,b2)); | |
5730 | export_rdev(rdev); | |
5731 | return -EINVAL; | |
5732 | } | |
5733 | } | |
5734 | err = bind_rdev_to_array(rdev, mddev); | |
5735 | if (err) | |
5736 | export_rdev(rdev); | |
5737 | return err; | |
5738 | } | |
5739 | ||
5740 | /* | |
5741 | * add_new_disk can be used once the array is assembled | |
5742 | * to add "hot spares". They must already have a superblock | |
5743 | * written | |
5744 | */ | |
5745 | if (mddev->pers) { | |
5746 | int err; | |
5747 | if (!mddev->pers->hot_add_disk) { | |
5748 | printk(KERN_WARNING | |
5749 | "%s: personality does not support diskops!\n", | |
5750 | mdname(mddev)); | |
5751 | return -EINVAL; | |
5752 | } | |
7b1e35f6 N |
5753 | if (mddev->persistent) |
5754 | rdev = md_import_device(dev, mddev->major_version, | |
5755 | mddev->minor_version); | |
5756 | else | |
5757 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5758 | if (IS_ERR(rdev)) { |
5759 | printk(KERN_WARNING | |
5760 | "md: md_import_device returned %ld\n", | |
5761 | PTR_ERR(rdev)); | |
5762 | return PTR_ERR(rdev); | |
5763 | } | |
1a855a06 | 5764 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5765 | if (!mddev->persistent) { |
5766 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5767 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5768 | rdev->raid_disk = info->raid_disk; |
bf572541 N |
5769 | set_bit(In_sync, &rdev->flags); |
5770 | } else | |
41158c7e N |
5771 | rdev->raid_disk = -1; |
5772 | } else | |
5773 | super_types[mddev->major_version]. | |
5774 | validate_super(mddev, rdev); | |
bedd86b7 | 5775 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 5776 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
5777 | /* This was a hot-add request, but events doesn't |
5778 | * match, so reject it. | |
5779 | */ | |
5780 | export_rdev(rdev); | |
5781 | return -EINVAL; | |
5782 | } | |
5783 | ||
1a855a06 N |
5784 | if (test_bit(In_sync, &rdev->flags)) |
5785 | rdev->saved_raid_disk = rdev->raid_disk; | |
5786 | else | |
5787 | rdev->saved_raid_disk = -1; | |
41158c7e | 5788 | |
b2d444d7 | 5789 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5790 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5791 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5792 | else |
5793 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5794 | |
1da177e4 LT |
5795 | rdev->raid_disk = -1; |
5796 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5797 | if (!err && !mddev->pers->hot_remove_disk) { |
5798 | /* If there is hot_add_disk but no hot_remove_disk | |
5799 | * then added disks for geometry changes, | |
5800 | * and should be added immediately. | |
5801 | */ | |
5802 | super_types[mddev->major_version]. | |
5803 | validate_super(mddev, rdev); | |
5804 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5805 | if (err) | |
5806 | unbind_rdev_from_array(rdev); | |
5807 | } | |
1da177e4 LT |
5808 | if (err) |
5809 | export_rdev(rdev); | |
52664732 | 5810 | else |
00bcb4ac | 5811 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5812 | |
17571284 | 5813 | md_update_sb(mddev, 1); |
72a23c21 NB |
5814 | if (mddev->degraded) |
5815 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5816 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
9864c005 | 5817 | if (!err) |
5818 | md_new_event(mddev); | |
005eca5e | 5819 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5820 | return err; |
5821 | } | |
5822 | ||
5823 | /* otherwise, add_new_disk is only allowed | |
5824 | * for major_version==0 superblocks | |
5825 | */ | |
5826 | if (mddev->major_version != 0) { | |
5827 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5828 | mdname(mddev)); | |
5829 | return -EINVAL; | |
5830 | } | |
5831 | ||
5832 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5833 | int err; | |
d710e138 | 5834 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5835 | if (IS_ERR(rdev)) { |
5836 | printk(KERN_WARNING | |
5837 | "md: error, md_import_device() returned %ld\n", | |
5838 | PTR_ERR(rdev)); | |
5839 | return PTR_ERR(rdev); | |
5840 | } | |
5841 | rdev->desc_nr = info->number; | |
5842 | if (info->raid_disk < mddev->raid_disks) | |
5843 | rdev->raid_disk = info->raid_disk; | |
5844 | else | |
5845 | rdev->raid_disk = -1; | |
5846 | ||
1da177e4 | 5847 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5848 | if (info->state & (1<<MD_DISK_SYNC)) |
5849 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5850 | |
8ddf9efe N |
5851 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5852 | set_bit(WriteMostly, &rdev->flags); | |
5853 | ||
1da177e4 LT |
5854 | if (!mddev->persistent) { |
5855 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5856 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5857 | } else | |
57b2caa3 | 5858 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5859 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5860 | |
2bf071bf N |
5861 | err = bind_rdev_to_array(rdev, mddev); |
5862 | if (err) { | |
5863 | export_rdev(rdev); | |
5864 | return err; | |
5865 | } | |
1da177e4 LT |
5866 | } |
5867 | ||
5868 | return 0; | |
5869 | } | |
5870 | ||
fd01b88c | 5871 | static int hot_remove_disk(struct mddev * mddev, dev_t dev) |
1da177e4 LT |
5872 | { |
5873 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 5874 | struct md_rdev *rdev; |
1da177e4 | 5875 | |
1da177e4 LT |
5876 | rdev = find_rdev(mddev, dev); |
5877 | if (!rdev) | |
5878 | return -ENXIO; | |
5879 | ||
5880 | if (rdev->raid_disk >= 0) | |
5881 | goto busy; | |
5882 | ||
5883 | kick_rdev_from_array(rdev); | |
850b2b42 | 5884 | md_update_sb(mddev, 1); |
d7603b7e | 5885 | md_new_event(mddev); |
1da177e4 LT |
5886 | |
5887 | return 0; | |
5888 | busy: | |
fdefa4d8 | 5889 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5890 | bdevname(rdev->bdev,b), mdname(mddev)); |
5891 | return -EBUSY; | |
5892 | } | |
5893 | ||
fd01b88c | 5894 | static int hot_add_disk(struct mddev * mddev, dev_t dev) |
1da177e4 LT |
5895 | { |
5896 | char b[BDEVNAME_SIZE]; | |
5897 | int err; | |
3cb03002 | 5898 | struct md_rdev *rdev; |
1da177e4 LT |
5899 | |
5900 | if (!mddev->pers) | |
5901 | return -ENODEV; | |
5902 | ||
5903 | if (mddev->major_version != 0) { | |
5904 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5905 | " version-0 superblocks.\n", | |
5906 | mdname(mddev)); | |
5907 | return -EINVAL; | |
5908 | } | |
5909 | if (!mddev->pers->hot_add_disk) { | |
5910 | printk(KERN_WARNING | |
5911 | "%s: personality does not support diskops!\n", | |
5912 | mdname(mddev)); | |
5913 | return -EINVAL; | |
5914 | } | |
5915 | ||
d710e138 | 5916 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5917 | if (IS_ERR(rdev)) { |
5918 | printk(KERN_WARNING | |
5919 | "md: error, md_import_device() returned %ld\n", | |
5920 | PTR_ERR(rdev)); | |
5921 | return -EINVAL; | |
5922 | } | |
5923 | ||
5924 | if (mddev->persistent) | |
57b2caa3 | 5925 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5926 | else |
77304d2a | 5927 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5928 | |
8190e754 | 5929 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5930 | |
b2d444d7 | 5931 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5932 | printk(KERN_WARNING |
5933 | "md: can not hot-add faulty %s disk to %s!\n", | |
5934 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5935 | err = -EINVAL; | |
5936 | goto abort_export; | |
5937 | } | |
b2d444d7 | 5938 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5939 | rdev->desc_nr = -1; |
5842730d | 5940 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5941 | err = bind_rdev_to_array(rdev, mddev); |
5942 | if (err) | |
5943 | goto abort_export; | |
1da177e4 LT |
5944 | |
5945 | /* | |
5946 | * The rest should better be atomic, we can have disk failures | |
5947 | * noticed in interrupt contexts ... | |
5948 | */ | |
5949 | ||
1da177e4 LT |
5950 | rdev->raid_disk = -1; |
5951 | ||
850b2b42 | 5952 | md_update_sb(mddev, 1); |
1da177e4 LT |
5953 | |
5954 | /* | |
5955 | * Kick recovery, maybe this spare has to be added to the | |
5956 | * array immediately. | |
5957 | */ | |
5958 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5959 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5960 | md_new_event(mddev); |
1da177e4 LT |
5961 | return 0; |
5962 | ||
1da177e4 LT |
5963 | abort_export: |
5964 | export_rdev(rdev); | |
5965 | return err; | |
5966 | } | |
5967 | ||
fd01b88c | 5968 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c N |
5969 | { |
5970 | int err; | |
5971 | ||
36fa3063 N |
5972 | if (mddev->pers) { |
5973 | if (!mddev->pers->quiesce) | |
5974 | return -EBUSY; | |
5975 | if (mddev->recovery || mddev->sync_thread) | |
5976 | return -EBUSY; | |
5977 | /* we should be able to change the bitmap.. */ | |
5978 | } | |
32a7627c | 5979 | |
32a7627c | 5980 | |
36fa3063 N |
5981 | if (fd >= 0) { |
5982 | if (mddev->bitmap) | |
5983 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5984 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5985 | |
c3d9714e | 5986 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5987 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5988 | mdname(mddev)); | |
5989 | return -EBADF; | |
5990 | } | |
5991 | ||
c3d9714e | 5992 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
5993 | if (err) { |
5994 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5995 | mdname(mddev)); | |
c3d9714e N |
5996 | fput(mddev->bitmap_info.file); |
5997 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5998 | return err; |
5999 | } | |
c3d9714e | 6000 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6001 | } else if (mddev->bitmap == NULL) |
6002 | return -ENOENT; /* cannot remove what isn't there */ | |
6003 | err = 0; | |
6004 | if (mddev->pers) { | |
6005 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6006 | if (fd >= 0) { |
36fa3063 | 6007 | err = bitmap_create(mddev); |
69e51b44 N |
6008 | if (!err) |
6009 | err = bitmap_load(mddev); | |
6010 | } | |
d7375ab3 | 6011 | if (fd < 0 || err) { |
36fa3063 | 6012 | bitmap_destroy(mddev); |
d7375ab3 N |
6013 | fd = -1; /* make sure to put the file */ |
6014 | } | |
36fa3063 | 6015 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6016 | } |
6017 | if (fd < 0) { | |
c3d9714e N |
6018 | if (mddev->bitmap_info.file) { |
6019 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
6020 | fput(mddev->bitmap_info.file); | |
acc55e22 | 6021 | } |
c3d9714e | 6022 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
6023 | } |
6024 | ||
32a7627c N |
6025 | return err; |
6026 | } | |
6027 | ||
1da177e4 LT |
6028 | /* |
6029 | * set_array_info is used two different ways | |
6030 | * The original usage is when creating a new array. | |
6031 | * In this usage, raid_disks is > 0 and it together with | |
6032 | * level, size, not_persistent,layout,chunksize determine the | |
6033 | * shape of the array. | |
6034 | * This will always create an array with a type-0.90.0 superblock. | |
6035 | * The newer usage is when assembling an array. | |
6036 | * In this case raid_disks will be 0, and the major_version field is | |
6037 | * use to determine which style super-blocks are to be found on the devices. | |
6038 | * The minor and patch _version numbers are also kept incase the | |
6039 | * super_block handler wishes to interpret them. | |
6040 | */ | |
fd01b88c | 6041 | static int set_array_info(struct mddev * mddev, mdu_array_info_t *info) |
1da177e4 LT |
6042 | { |
6043 | ||
6044 | if (info->raid_disks == 0) { | |
6045 | /* just setting version number for superblock loading */ | |
6046 | if (info->major_version < 0 || | |
50511da3 | 6047 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6048 | super_types[info->major_version].name == NULL) { |
6049 | /* maybe try to auto-load a module? */ | |
6050 | printk(KERN_INFO | |
6051 | "md: superblock version %d not known\n", | |
6052 | info->major_version); | |
6053 | return -EINVAL; | |
6054 | } | |
6055 | mddev->major_version = info->major_version; | |
6056 | mddev->minor_version = info->minor_version; | |
6057 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6058 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6059 | /* ensure mddev_put doesn't delete this now that there |
6060 | * is some minimal configuration. | |
6061 | */ | |
6062 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
6063 | return 0; |
6064 | } | |
6065 | mddev->major_version = MD_MAJOR_VERSION; | |
6066 | mddev->minor_version = MD_MINOR_VERSION; | |
6067 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
6068 | mddev->ctime = get_seconds(); | |
6069 | ||
6070 | mddev->level = info->level; | |
17115e03 | 6071 | mddev->clevel[0] = 0; |
58c0fed4 | 6072 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6073 | mddev->raid_disks = info->raid_disks; |
6074 | /* don't set md_minor, it is determined by which /dev/md* was | |
6075 | * openned | |
6076 | */ | |
6077 | if (info->state & (1<<MD_SB_CLEAN)) | |
6078 | mddev->recovery_cp = MaxSector; | |
6079 | else | |
6080 | mddev->recovery_cp = 0; | |
6081 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6082 | mddev->external = 0; |
1da177e4 LT |
6083 | |
6084 | mddev->layout = info->layout; | |
9d8f0363 | 6085 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
6086 | |
6087 | mddev->max_disks = MD_SB_DISKS; | |
6088 | ||
e691063a N |
6089 | if (mddev->persistent) |
6090 | mddev->flags = 0; | |
850b2b42 | 6091 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6092 | |
c3d9714e | 6093 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6094 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6095 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6096 | |
f6705578 N |
6097 | mddev->reshape_position = MaxSector; |
6098 | ||
1da177e4 LT |
6099 | /* |
6100 | * Generate a 128 bit UUID | |
6101 | */ | |
6102 | get_random_bytes(mddev->uuid, 16); | |
6103 | ||
f6705578 | 6104 | mddev->new_level = mddev->level; |
664e7c41 | 6105 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6106 | mddev->new_layout = mddev->layout; |
6107 | mddev->delta_disks = 0; | |
2c810cdd | 6108 | mddev->reshape_backwards = 0; |
f6705578 | 6109 | |
1da177e4 LT |
6110 | return 0; |
6111 | } | |
6112 | ||
fd01b88c | 6113 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6114 | { |
b522adcd DW |
6115 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6116 | ||
6117 | if (mddev->external_size) | |
6118 | return; | |
6119 | ||
1f403624 DW |
6120 | mddev->array_sectors = array_sectors; |
6121 | } | |
6122 | EXPORT_SYMBOL(md_set_array_sectors); | |
6123 | ||
fd01b88c | 6124 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6125 | { |
3cb03002 | 6126 | struct md_rdev *rdev; |
a35b0d69 | 6127 | int rv; |
d71f9f88 | 6128 | int fit = (num_sectors == 0); |
a35b0d69 N |
6129 | |
6130 | if (mddev->pers->resize == NULL) | |
6131 | return -EINVAL; | |
d71f9f88 AN |
6132 | /* The "num_sectors" is the number of sectors of each device that |
6133 | * is used. This can only make sense for arrays with redundancy. | |
6134 | * linear and raid0 always use whatever space is available. We can only | |
6135 | * consider changing this number if no resync or reconstruction is | |
6136 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6137 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6138 | * of each device. If num_sectors is zero, we find the largest size |
6139 | * that fits. | |
a35b0d69 N |
6140 | */ |
6141 | if (mddev->sync_thread) | |
6142 | return -EBUSY; | |
a4a6125a | 6143 | |
dafb20fa | 6144 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6145 | sector_t avail = rdev->sectors; |
01ab5662 | 6146 | |
d71f9f88 AN |
6147 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6148 | num_sectors = avail; | |
6149 | if (avail < num_sectors) | |
a35b0d69 N |
6150 | return -ENOSPC; |
6151 | } | |
d71f9f88 | 6152 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
6153 | if (!rv) |
6154 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
6155 | return rv; |
6156 | } | |
6157 | ||
fd01b88c | 6158 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6159 | { |
6160 | int rv; | |
c6563a8c | 6161 | struct md_rdev *rdev; |
da943b99 | 6162 | /* change the number of raid disks */ |
63c70c4f | 6163 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
6164 | return -EINVAL; |
6165 | if (raid_disks <= 0 || | |
233fca36 | 6166 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6167 | return -EINVAL; |
63c70c4f | 6168 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 6169 | return -EBUSY; |
c6563a8c N |
6170 | |
6171 | rdev_for_each(rdev, mddev) { | |
6172 | if (mddev->raid_disks < raid_disks && | |
6173 | rdev->data_offset < rdev->new_data_offset) | |
6174 | return -EINVAL; | |
6175 | if (mddev->raid_disks > raid_disks && | |
6176 | rdev->data_offset > rdev->new_data_offset) | |
6177 | return -EINVAL; | |
6178 | } | |
6179 | ||
63c70c4f | 6180 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6181 | if (mddev->delta_disks < 0) |
6182 | mddev->reshape_backwards = 1; | |
6183 | else if (mddev->delta_disks > 0) | |
6184 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6185 | |
6186 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6187 | if (rv < 0) { |
de171cb9 | 6188 | mddev->delta_disks = 0; |
2c810cdd N |
6189 | mddev->reshape_backwards = 0; |
6190 | } | |
da943b99 N |
6191 | return rv; |
6192 | } | |
6193 | ||
6194 | ||
1da177e4 LT |
6195 | /* |
6196 | * update_array_info is used to change the configuration of an | |
6197 | * on-line array. | |
6198 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6199 | * fields in the info are checked against the array. | |
6200 | * Any differences that cannot be handled will cause an error. | |
6201 | * Normally, only one change can be managed at a time. | |
6202 | */ | |
fd01b88c | 6203 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6204 | { |
6205 | int rv = 0; | |
6206 | int cnt = 0; | |
36fa3063 N |
6207 | int state = 0; |
6208 | ||
6209 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6210 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6211 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6212 | |
6213 | if (mddev->major_version != info->major_version || | |
6214 | mddev->minor_version != info->minor_version || | |
6215 | /* mddev->patch_version != info->patch_version || */ | |
6216 | mddev->ctime != info->ctime || | |
6217 | mddev->level != info->level || | |
6218 | /* mddev->layout != info->layout || */ | |
6219 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 6220 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6221 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6222 | ((state^info->state) & 0xfffffe00) | |
6223 | ) | |
1da177e4 LT |
6224 | return -EINVAL; |
6225 | /* Check there is only one change */ | |
58c0fed4 AN |
6226 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6227 | cnt++; | |
6228 | if (mddev->raid_disks != info->raid_disks) | |
6229 | cnt++; | |
6230 | if (mddev->layout != info->layout) | |
6231 | cnt++; | |
6232 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6233 | cnt++; | |
6234 | if (cnt == 0) | |
6235 | return 0; | |
6236 | if (cnt > 1) | |
6237 | return -EINVAL; | |
1da177e4 LT |
6238 | |
6239 | if (mddev->layout != info->layout) { | |
6240 | /* Change layout | |
6241 | * we don't need to do anything at the md level, the | |
6242 | * personality will take care of it all. | |
6243 | */ | |
50ac168a | 6244 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6245 | return -EINVAL; |
597a711b N |
6246 | else { |
6247 | mddev->new_layout = info->layout; | |
50ac168a | 6248 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6249 | if (rv) |
6250 | mddev->new_layout = mddev->layout; | |
6251 | return rv; | |
6252 | } | |
1da177e4 | 6253 | } |
58c0fed4 | 6254 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6255 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6256 | |
da943b99 N |
6257 | if (mddev->raid_disks != info->raid_disks) |
6258 | rv = update_raid_disks(mddev, info->raid_disks); | |
6259 | ||
36fa3063 N |
6260 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
6261 | if (mddev->pers->quiesce == NULL) | |
6262 | return -EINVAL; | |
6263 | if (mddev->recovery || mddev->sync_thread) | |
6264 | return -EBUSY; | |
6265 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
6266 | /* add the bitmap */ | |
6267 | if (mddev->bitmap) | |
6268 | return -EEXIST; | |
c3d9714e | 6269 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 6270 | return -EINVAL; |
c3d9714e N |
6271 | mddev->bitmap_info.offset = |
6272 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6273 | mddev->bitmap_info.space = |
6274 | mddev->bitmap_info.default_space; | |
36fa3063 N |
6275 | mddev->pers->quiesce(mddev, 1); |
6276 | rv = bitmap_create(mddev); | |
69e51b44 N |
6277 | if (!rv) |
6278 | rv = bitmap_load(mddev); | |
36fa3063 N |
6279 | if (rv) |
6280 | bitmap_destroy(mddev); | |
6281 | mddev->pers->quiesce(mddev, 0); | |
6282 | } else { | |
6283 | /* remove the bitmap */ | |
6284 | if (!mddev->bitmap) | |
6285 | return -ENOENT; | |
1ec885cd | 6286 | if (mddev->bitmap->storage.file) |
36fa3063 N |
6287 | return -EINVAL; |
6288 | mddev->pers->quiesce(mddev, 1); | |
6289 | bitmap_destroy(mddev); | |
6290 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6291 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6292 | } |
6293 | } | |
850b2b42 | 6294 | md_update_sb(mddev, 1); |
1da177e4 LT |
6295 | return rv; |
6296 | } | |
6297 | ||
fd01b88c | 6298 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6299 | { |
3cb03002 | 6300 | struct md_rdev *rdev; |
1ca69c4b | 6301 | int err = 0; |
1da177e4 LT |
6302 | |
6303 | if (mddev->pers == NULL) | |
6304 | return -ENODEV; | |
6305 | ||
1ca69c4b N |
6306 | rcu_read_lock(); |
6307 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6308 | if (!rdev) |
1ca69c4b N |
6309 | err = -ENODEV; |
6310 | else { | |
6311 | md_error(mddev, rdev); | |
6312 | if (!test_bit(Faulty, &rdev->flags)) | |
6313 | err = -EBUSY; | |
6314 | } | |
6315 | rcu_read_unlock(); | |
6316 | return err; | |
1da177e4 LT |
6317 | } |
6318 | ||
2f9618ce AN |
6319 | /* |
6320 | * We have a problem here : there is no easy way to give a CHS | |
6321 | * virtual geometry. We currently pretend that we have a 2 heads | |
6322 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6323 | * dosfs just mad... ;-) | |
6324 | */ | |
a885c8c4 CH |
6325 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6326 | { | |
fd01b88c | 6327 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6328 | |
6329 | geo->heads = 2; | |
6330 | geo->sectors = 4; | |
49ce6cea | 6331 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6332 | return 0; |
6333 | } | |
6334 | ||
a39907fa | 6335 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6336 | unsigned int cmd, unsigned long arg) |
6337 | { | |
6338 | int err = 0; | |
6339 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6340 | struct mddev *mddev = NULL; |
e2218350 | 6341 | int ro; |
1da177e4 | 6342 | |
506c9e44 N |
6343 | switch (cmd) { |
6344 | case RAID_VERSION: | |
6345 | case GET_ARRAY_INFO: | |
6346 | case GET_DISK_INFO: | |
6347 | break; | |
6348 | default: | |
6349 | if (!capable(CAP_SYS_ADMIN)) | |
6350 | return -EACCES; | |
6351 | } | |
1da177e4 LT |
6352 | |
6353 | /* | |
6354 | * Commands dealing with the RAID driver but not any | |
6355 | * particular array: | |
6356 | */ | |
c02c0aeb N |
6357 | switch (cmd) { |
6358 | case RAID_VERSION: | |
6359 | err = get_version(argp); | |
6360 | goto done; | |
1da177e4 | 6361 | |
c02c0aeb N |
6362 | case PRINT_RAID_DEBUG: |
6363 | err = 0; | |
6364 | md_print_devices(); | |
6365 | goto done; | |
1da177e4 LT |
6366 | |
6367 | #ifndef MODULE | |
c02c0aeb N |
6368 | case RAID_AUTORUN: |
6369 | err = 0; | |
6370 | autostart_arrays(arg); | |
6371 | goto done; | |
1da177e4 | 6372 | #endif |
c02c0aeb | 6373 | default:; |
1da177e4 LT |
6374 | } |
6375 | ||
6376 | /* | |
6377 | * Commands creating/starting a new array: | |
6378 | */ | |
6379 | ||
a39907fa | 6380 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6381 | |
6382 | if (!mddev) { | |
6383 | BUG(); | |
6384 | goto abort; | |
6385 | } | |
6386 | ||
1ca69c4b N |
6387 | /* Some actions do not requires the mutex */ |
6388 | switch (cmd) { | |
6389 | case GET_ARRAY_INFO: | |
6390 | if (!mddev->raid_disks && !mddev->external) | |
6391 | err = -ENODEV; | |
6392 | else | |
6393 | err = get_array_info(mddev, argp); | |
6394 | goto abort; | |
6395 | ||
6396 | case GET_DISK_INFO: | |
6397 | if (!mddev->raid_disks && !mddev->external) | |
6398 | err = -ENODEV; | |
6399 | else | |
6400 | err = get_disk_info(mddev, argp); | |
6401 | goto abort; | |
6402 | ||
6403 | case SET_DISK_FAULTY: | |
6404 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
6405 | goto abort; | |
6406 | } | |
6407 | ||
a7a3f08d N |
6408 | if (cmd == ADD_NEW_DISK) |
6409 | /* need to ensure md_delayed_delete() has completed */ | |
6410 | flush_workqueue(md_misc_wq); | |
6411 | ||
1da177e4 LT |
6412 | err = mddev_lock(mddev); |
6413 | if (err) { | |
6414 | printk(KERN_INFO | |
6415 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
6416 | err, cmd); | |
6417 | goto abort; | |
6418 | } | |
6419 | ||
c02c0aeb N |
6420 | if (cmd == SET_ARRAY_INFO) { |
6421 | mdu_array_info_t info; | |
6422 | if (!arg) | |
6423 | memset(&info, 0, sizeof(info)); | |
6424 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6425 | err = -EFAULT; | |
6426 | goto abort_unlock; | |
6427 | } | |
6428 | if (mddev->pers) { | |
6429 | err = update_array_info(mddev, &info); | |
6430 | if (err) { | |
6431 | printk(KERN_WARNING "md: couldn't update" | |
6432 | " array info. %d\n", err); | |
6433 | goto abort_unlock; | |
1da177e4 LT |
6434 | } |
6435 | goto done_unlock; | |
c02c0aeb N |
6436 | } |
6437 | if (!list_empty(&mddev->disks)) { | |
6438 | printk(KERN_WARNING | |
6439 | "md: array %s already has disks!\n", | |
6440 | mdname(mddev)); | |
6441 | err = -EBUSY; | |
6442 | goto abort_unlock; | |
6443 | } | |
6444 | if (mddev->raid_disks) { | |
6445 | printk(KERN_WARNING | |
6446 | "md: array %s already initialised!\n", | |
6447 | mdname(mddev)); | |
6448 | err = -EBUSY; | |
6449 | goto abort_unlock; | |
6450 | } | |
6451 | err = set_array_info(mddev, &info); | |
6452 | if (err) { | |
6453 | printk(KERN_WARNING "md: couldn't set" | |
6454 | " array info. %d\n", err); | |
6455 | goto abort_unlock; | |
6456 | } | |
6457 | goto done_unlock; | |
1da177e4 LT |
6458 | } |
6459 | ||
6460 | /* | |
6461 | * Commands querying/configuring an existing array: | |
6462 | */ | |
32a7627c | 6463 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6464 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6465 | if ((!mddev->raid_disks && !mddev->external) |
6466 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6467 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6468 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
6469 | err = -ENODEV; |
6470 | goto abort_unlock; | |
6471 | } | |
6472 | ||
6473 | /* | |
6474 | * Commands even a read-only array can execute: | |
6475 | */ | |
c02c0aeb N |
6476 | switch (cmd) { |
6477 | case GET_BITMAP_FILE: | |
6478 | err = get_bitmap_file(mddev, argp); | |
6479 | goto done_unlock; | |
32a7627c | 6480 | |
c02c0aeb N |
6481 | case RESTART_ARRAY_RW: |
6482 | err = restart_array(mddev); | |
6483 | goto done_unlock; | |
1da177e4 | 6484 | |
c02c0aeb N |
6485 | case STOP_ARRAY: |
6486 | err = do_md_stop(mddev, 0, bdev); | |
6487 | goto done_unlock; | |
1da177e4 | 6488 | |
c02c0aeb N |
6489 | case STOP_ARRAY_RO: |
6490 | err = md_set_readonly(mddev, bdev); | |
6491 | goto done_unlock; | |
1da177e4 | 6492 | |
c02c0aeb N |
6493 | case BLKROSET: |
6494 | if (get_user(ro, (int __user *)(arg))) { | |
6495 | err = -EFAULT; | |
6496 | goto done_unlock; | |
6497 | } | |
6498 | err = -EINVAL; | |
e2218350 | 6499 | |
c02c0aeb N |
6500 | /* if the bdev is going readonly the value of mddev->ro |
6501 | * does not matter, no writes are coming | |
6502 | */ | |
6503 | if (ro) | |
6504 | goto done_unlock; | |
e2218350 | 6505 | |
c02c0aeb N |
6506 | /* are we are already prepared for writes? */ |
6507 | if (mddev->ro != 1) | |
6508 | goto done_unlock; | |
e2218350 | 6509 | |
c02c0aeb N |
6510 | /* transitioning to readauto need only happen for |
6511 | * arrays that call md_write_start | |
6512 | */ | |
6513 | if (mddev->pers) { | |
6514 | err = restart_array(mddev); | |
6515 | if (err == 0) { | |
6516 | mddev->ro = 2; | |
6517 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6518 | } |
c02c0aeb N |
6519 | } |
6520 | goto done_unlock; | |
1da177e4 LT |
6521 | } |
6522 | ||
6523 | /* | |
6524 | * The remaining ioctls are changing the state of the | |
f91de92e N |
6525 | * superblock, so we do not allow them on read-only arrays. |
6526 | * However non-MD ioctls (e.g. get-size) will still come through | |
6527 | * here and hit the 'default' below, so only disallow | |
6528 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 6529 | */ |
bb57fc64 | 6530 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
6531 | if (mddev->ro == 2) { |
6532 | mddev->ro = 0; | |
00bcb4ac | 6533 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 NB |
6534 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
6535 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
6536 | } else { |
6537 | err = -EROFS; | |
6538 | goto abort_unlock; | |
6539 | } | |
1da177e4 LT |
6540 | } |
6541 | ||
c02c0aeb N |
6542 | switch (cmd) { |
6543 | case ADD_NEW_DISK: | |
1da177e4 | 6544 | { |
c02c0aeb N |
6545 | mdu_disk_info_t info; |
6546 | if (copy_from_user(&info, argp, sizeof(info))) | |
6547 | err = -EFAULT; | |
6548 | else | |
6549 | err = add_new_disk(mddev, &info); | |
6550 | goto done_unlock; | |
6551 | } | |
1da177e4 | 6552 | |
c02c0aeb N |
6553 | case HOT_REMOVE_DISK: |
6554 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
6555 | goto done_unlock; | |
1da177e4 | 6556 | |
c02c0aeb N |
6557 | case HOT_ADD_DISK: |
6558 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
6559 | goto done_unlock; | |
1da177e4 | 6560 | |
c02c0aeb N |
6561 | case RUN_ARRAY: |
6562 | err = do_md_run(mddev); | |
6563 | goto done_unlock; | |
1da177e4 | 6564 | |
c02c0aeb N |
6565 | case SET_BITMAP_FILE: |
6566 | err = set_bitmap_file(mddev, (int)arg); | |
6567 | goto done_unlock; | |
32a7627c | 6568 | |
c02c0aeb N |
6569 | default: |
6570 | err = -EINVAL; | |
6571 | goto abort_unlock; | |
1da177e4 LT |
6572 | } |
6573 | ||
6574 | done_unlock: | |
6575 | abort_unlock: | |
d3374825 N |
6576 | if (mddev->hold_active == UNTIL_IOCTL && |
6577 | err != -EINVAL) | |
6578 | mddev->hold_active = 0; | |
1da177e4 LT |
6579 | mddev_unlock(mddev); |
6580 | ||
6581 | return err; | |
6582 | done: | |
6583 | if (err) | |
6584 | MD_BUG(); | |
6585 | abort: | |
6586 | return err; | |
6587 | } | |
aa98aa31 AB |
6588 | #ifdef CONFIG_COMPAT |
6589 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6590 | unsigned int cmd, unsigned long arg) | |
6591 | { | |
6592 | switch (cmd) { | |
6593 | case HOT_REMOVE_DISK: | |
6594 | case HOT_ADD_DISK: | |
6595 | case SET_DISK_FAULTY: | |
6596 | case SET_BITMAP_FILE: | |
6597 | /* These take in integer arg, do not convert */ | |
6598 | break; | |
6599 | default: | |
6600 | arg = (unsigned long)compat_ptr(arg); | |
6601 | break; | |
6602 | } | |
6603 | ||
6604 | return md_ioctl(bdev, mode, cmd, arg); | |
6605 | } | |
6606 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6607 | |
a39907fa | 6608 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6609 | { |
6610 | /* | |
6611 | * Succeed if we can lock the mddev, which confirms that | |
6612 | * it isn't being stopped right now. | |
6613 | */ | |
fd01b88c | 6614 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6615 | int err; |
6616 | ||
0c098220 YL |
6617 | if (!mddev) |
6618 | return -ENODEV; | |
6619 | ||
d3374825 N |
6620 | if (mddev->gendisk != bdev->bd_disk) { |
6621 | /* we are racing with mddev_put which is discarding this | |
6622 | * bd_disk. | |
6623 | */ | |
6624 | mddev_put(mddev); | |
6625 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6626 | flush_workqueue(md_misc_wq); |
d3374825 N |
6627 | /* Then retry the open from the top */ |
6628 | return -ERESTARTSYS; | |
6629 | } | |
6630 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6631 | ||
c8c00a69 | 6632 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6633 | goto out; |
6634 | ||
6635 | err = 0; | |
f2ea68cf | 6636 | atomic_inc(&mddev->openers); |
c8c00a69 | 6637 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6638 | |
f0b4f7e2 | 6639 | check_disk_change(bdev); |
1da177e4 LT |
6640 | out: |
6641 | return err; | |
6642 | } | |
6643 | ||
a39907fa | 6644 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6645 | { |
fd01b88c | 6646 | struct mddev *mddev = disk->private_data; |
1da177e4 | 6647 | |
52e5f9d1 | 6648 | BUG_ON(!mddev); |
f2ea68cf | 6649 | atomic_dec(&mddev->openers); |
1da177e4 LT |
6650 | mddev_put(mddev); |
6651 | ||
6652 | return 0; | |
6653 | } | |
f0b4f7e2 N |
6654 | |
6655 | static int md_media_changed(struct gendisk *disk) | |
6656 | { | |
fd01b88c | 6657 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6658 | |
6659 | return mddev->changed; | |
6660 | } | |
6661 | ||
6662 | static int md_revalidate(struct gendisk *disk) | |
6663 | { | |
fd01b88c | 6664 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6665 | |
6666 | mddev->changed = 0; | |
6667 | return 0; | |
6668 | } | |
83d5cde4 | 6669 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6670 | { |
6671 | .owner = THIS_MODULE, | |
a39907fa AV |
6672 | .open = md_open, |
6673 | .release = md_release, | |
b492b852 | 6674 | .ioctl = md_ioctl, |
aa98aa31 AB |
6675 | #ifdef CONFIG_COMPAT |
6676 | .compat_ioctl = md_compat_ioctl, | |
6677 | #endif | |
a885c8c4 | 6678 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6679 | .media_changed = md_media_changed, |
6680 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6681 | }; |
6682 | ||
75c96f85 | 6683 | static int md_thread(void * arg) |
1da177e4 | 6684 | { |
2b8bf345 | 6685 | struct md_thread *thread = arg; |
1da177e4 | 6686 | |
1da177e4 LT |
6687 | /* |
6688 | * md_thread is a 'system-thread', it's priority should be very | |
6689 | * high. We avoid resource deadlocks individually in each | |
6690 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6691 | * the very same RT priority as kswapd, thus we will never get | |
6692 | * into a priority inversion deadlock. | |
6693 | * | |
6694 | * we definitely have to have equal or higher priority than | |
6695 | * bdflush, otherwise bdflush will deadlock if there are too | |
6696 | * many dirty RAID5 blocks. | |
6697 | */ | |
1da177e4 | 6698 | |
6985c43f | 6699 | allow_signal(SIGKILL); |
a6fb0934 | 6700 | while (!kthread_should_stop()) { |
1da177e4 | 6701 | |
93588e22 N |
6702 | /* We need to wait INTERRUPTIBLE so that |
6703 | * we don't add to the load-average. | |
6704 | * That means we need to be sure no signals are | |
6705 | * pending | |
6706 | */ | |
6707 | if (signal_pending(current)) | |
6708 | flush_signals(current); | |
6709 | ||
6710 | wait_event_interruptible_timeout | |
6711 | (thread->wqueue, | |
6712 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6713 | || kthread_should_stop(), | |
6714 | thread->timeout); | |
1da177e4 | 6715 | |
6c987910 N |
6716 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6717 | if (!kthread_should_stop()) | |
4ed8731d | 6718 | thread->run(thread); |
1da177e4 | 6719 | } |
a6fb0934 | 6720 | |
1da177e4 LT |
6721 | return 0; |
6722 | } | |
6723 | ||
2b8bf345 | 6724 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
6725 | { |
6726 | if (thread) { | |
36a4e1fe | 6727 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
6728 | set_bit(THREAD_WAKEUP, &thread->flags); |
6729 | wake_up(&thread->wqueue); | |
6730 | } | |
6731 | } | |
6732 | ||
4ed8731d SL |
6733 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
6734 | struct mddev *mddev, const char *name) | |
1da177e4 | 6735 | { |
2b8bf345 | 6736 | struct md_thread *thread; |
1da177e4 | 6737 | |
2b8bf345 | 6738 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
6739 | if (!thread) |
6740 | return NULL; | |
6741 | ||
1da177e4 LT |
6742 | init_waitqueue_head(&thread->wqueue); |
6743 | ||
1da177e4 LT |
6744 | thread->run = run; |
6745 | thread->mddev = mddev; | |
32a7627c | 6746 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6747 | thread->tsk = kthread_run(md_thread, thread, |
6748 | "%s_%s", | |
6749 | mdname(thread->mddev), | |
0232605d | 6750 | name); |
a6fb0934 | 6751 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6752 | kfree(thread); |
6753 | return NULL; | |
6754 | } | |
1da177e4 LT |
6755 | return thread; |
6756 | } | |
6757 | ||
2b8bf345 | 6758 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 6759 | { |
2b8bf345 | 6760 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
6761 | if (!thread) |
6762 | return; | |
36a4e1fe | 6763 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
6764 | /* Locking ensures that mddev_unlock does not wake_up a |
6765 | * non-existent thread | |
6766 | */ | |
6767 | spin_lock(&pers_lock); | |
6768 | *threadp = NULL; | |
6769 | spin_unlock(&pers_lock); | |
a6fb0934 N |
6770 | |
6771 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6772 | kfree(thread); |
6773 | } | |
6774 | ||
fd01b88c | 6775 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
6776 | { |
6777 | if (!mddev) { | |
6778 | MD_BUG(); | |
6779 | return; | |
6780 | } | |
6781 | ||
b2d444d7 | 6782 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6783 | return; |
6bfe0b49 | 6784 | |
de393cde | 6785 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
6786 | return; |
6787 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6788 | if (mddev->degraded) |
6789 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6790 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6791 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6792 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6793 | md_wakeup_thread(mddev->thread); | |
768a418d | 6794 | if (mddev->event_work.func) |
e804ac78 | 6795 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6796 | md_new_event_inintr(mddev); |
1da177e4 LT |
6797 | } |
6798 | ||
6799 | /* seq_file implementation /proc/mdstat */ | |
6800 | ||
6801 | static void status_unused(struct seq_file *seq) | |
6802 | { | |
6803 | int i = 0; | |
3cb03002 | 6804 | struct md_rdev *rdev; |
1da177e4 LT |
6805 | |
6806 | seq_printf(seq, "unused devices: "); | |
6807 | ||
159ec1fc | 6808 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6809 | char b[BDEVNAME_SIZE]; |
6810 | i++; | |
6811 | seq_printf(seq, "%s ", | |
6812 | bdevname(rdev->bdev,b)); | |
6813 | } | |
6814 | if (!i) | |
6815 | seq_printf(seq, "<none>"); | |
6816 | ||
6817 | seq_printf(seq, "\n"); | |
6818 | } | |
6819 | ||
6820 | ||
fd01b88c | 6821 | static void status_resync(struct seq_file *seq, struct mddev * mddev) |
1da177e4 | 6822 | { |
dd71cf6b N |
6823 | sector_t max_sectors, resync, res; |
6824 | unsigned long dt, db; | |
6825 | sector_t rt; | |
4588b42e N |
6826 | int scale; |
6827 | unsigned int per_milli; | |
1da177e4 | 6828 | |
72f36d59 N |
6829 | if (mddev->curr_resync <= 3) |
6830 | resync = 0; | |
6831 | else | |
6832 | resync = mddev->curr_resync | |
6833 | - atomic_read(&mddev->recovery_active); | |
1da177e4 | 6834 | |
c804cdec N |
6835 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
6836 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 6837 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6838 | else |
dd71cf6b | 6839 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6840 | |
6841 | /* | |
6842 | * Should not happen. | |
6843 | */ | |
dd71cf6b | 6844 | if (!max_sectors) { |
1da177e4 LT |
6845 | MD_BUG(); |
6846 | return; | |
6847 | } | |
4588b42e | 6848 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6849 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6850 | * u32, as those are the requirements for sector_div. |
6851 | * Thus 'scale' must be at least 10 | |
6852 | */ | |
6853 | scale = 10; | |
6854 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6855 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6856 | scale++; |
6857 | } | |
6858 | res = (resync>>scale)*1000; | |
dd71cf6b | 6859 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6860 | |
6861 | per_milli = res; | |
1da177e4 | 6862 | { |
4588b42e | 6863 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6864 | seq_printf(seq, "["); |
6865 | for (i = 0; i < x; i++) | |
6866 | seq_printf(seq, "="); | |
6867 | seq_printf(seq, ">"); | |
6868 | for (i = 0; i < y; i++) | |
6869 | seq_printf(seq, "."); | |
6870 | seq_printf(seq, "] "); | |
6871 | } | |
4588b42e | 6872 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6873 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6874 | "reshape" : | |
61df9d91 N |
6875 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6876 | "check" : | |
6877 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6878 | "resync" : "recovery"))), | |
6879 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6880 | (unsigned long long) resync/2, |
6881 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6882 | |
6883 | /* | |
1da177e4 LT |
6884 | * dt: time from mark until now |
6885 | * db: blocks written from mark until now | |
6886 | * rt: remaining time | |
dd71cf6b N |
6887 | * |
6888 | * rt is a sector_t, so could be 32bit or 64bit. | |
6889 | * So we divide before multiply in case it is 32bit and close | |
6890 | * to the limit. | |
25985edc | 6891 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
6892 | * near the end of resync when the number of remaining sectors |
6893 | * is close to 'db'. | |
6894 | * We then divide rt by 32 after multiplying by db to compensate. | |
6895 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6896 | */ |
6897 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6898 | if (!dt) dt++; | |
ff4e8d9a N |
6899 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6900 | - mddev->resync_mark_cnt; | |
1da177e4 | 6901 | |
dd71cf6b N |
6902 | rt = max_sectors - resync; /* number of remaining sectors */ |
6903 | sector_div(rt, db/32+1); | |
6904 | rt *= dt; | |
6905 | rt >>= 5; | |
6906 | ||
6907 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6908 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6909 | |
ff4e8d9a | 6910 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6911 | } |
6912 | ||
6913 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6914 | { | |
6915 | struct list_head *tmp; | |
6916 | loff_t l = *pos; | |
fd01b88c | 6917 | struct mddev *mddev; |
1da177e4 LT |
6918 | |
6919 | if (l >= 0x10000) | |
6920 | return NULL; | |
6921 | if (!l--) | |
6922 | /* header */ | |
6923 | return (void*)1; | |
6924 | ||
6925 | spin_lock(&all_mddevs_lock); | |
6926 | list_for_each(tmp,&all_mddevs) | |
6927 | if (!l--) { | |
fd01b88c | 6928 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
6929 | mddev_get(mddev); |
6930 | spin_unlock(&all_mddevs_lock); | |
6931 | return mddev; | |
6932 | } | |
6933 | spin_unlock(&all_mddevs_lock); | |
6934 | if (!l--) | |
6935 | return (void*)2;/* tail */ | |
6936 | return NULL; | |
6937 | } | |
6938 | ||
6939 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6940 | { | |
6941 | struct list_head *tmp; | |
fd01b88c | 6942 | struct mddev *next_mddev, *mddev = v; |
1da177e4 LT |
6943 | |
6944 | ++*pos; | |
6945 | if (v == (void*)2) | |
6946 | return NULL; | |
6947 | ||
6948 | spin_lock(&all_mddevs_lock); | |
6949 | if (v == (void*)1) | |
6950 | tmp = all_mddevs.next; | |
6951 | else | |
6952 | tmp = mddev->all_mddevs.next; | |
6953 | if (tmp != &all_mddevs) | |
fd01b88c | 6954 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
6955 | else { |
6956 | next_mddev = (void*)2; | |
6957 | *pos = 0x10000; | |
6958 | } | |
6959 | spin_unlock(&all_mddevs_lock); | |
6960 | ||
6961 | if (v != (void*)1) | |
6962 | mddev_put(mddev); | |
6963 | return next_mddev; | |
6964 | ||
6965 | } | |
6966 | ||
6967 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6968 | { | |
fd01b88c | 6969 | struct mddev *mddev = v; |
1da177e4 LT |
6970 | |
6971 | if (mddev && v != (void*)1 && v != (void*)2) | |
6972 | mddev_put(mddev); | |
6973 | } | |
6974 | ||
6975 | static int md_seq_show(struct seq_file *seq, void *v) | |
6976 | { | |
fd01b88c | 6977 | struct mddev *mddev = v; |
dd8ac336 | 6978 | sector_t sectors; |
3cb03002 | 6979 | struct md_rdev *rdev; |
1da177e4 LT |
6980 | |
6981 | if (v == (void*)1) { | |
84fc4b56 | 6982 | struct md_personality *pers; |
1da177e4 LT |
6983 | seq_printf(seq, "Personalities : "); |
6984 | spin_lock(&pers_lock); | |
2604b703 N |
6985 | list_for_each_entry(pers, &pers_list, list) |
6986 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6987 | |
6988 | spin_unlock(&pers_lock); | |
6989 | seq_printf(seq, "\n"); | |
f1514638 | 6990 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
6991 | return 0; |
6992 | } | |
6993 | if (v == (void*)2) { | |
6994 | status_unused(seq); | |
6995 | return 0; | |
6996 | } | |
6997 | ||
5dc5cf7d | 6998 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6999 | return -EINTR; |
5dc5cf7d | 7000 | |
1da177e4 LT |
7001 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7002 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7003 | mddev->pers ? "" : "in"); | |
7004 | if (mddev->pers) { | |
f91de92e | 7005 | if (mddev->ro==1) |
1da177e4 | 7006 | seq_printf(seq, " (read-only)"); |
f91de92e | 7007 | if (mddev->ro==2) |
52720ae7 | 7008 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7009 | seq_printf(seq, " %s", mddev->pers->name); |
7010 | } | |
7011 | ||
dd8ac336 | 7012 | sectors = 0; |
dafb20fa | 7013 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
7014 | char b[BDEVNAME_SIZE]; |
7015 | seq_printf(seq, " %s[%d]", | |
7016 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7017 | if (test_bit(WriteMostly, &rdev->flags)) |
7018 | seq_printf(seq, "(W)"); | |
b2d444d7 | 7019 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7020 | seq_printf(seq, "(F)"); |
7021 | continue; | |
2d78f8c4 N |
7022 | } |
7023 | if (rdev->raid_disk < 0) | |
b325a32e | 7024 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7025 | if (test_bit(Replacement, &rdev->flags)) |
7026 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7027 | sectors += rdev->sectors; |
1da177e4 LT |
7028 | } |
7029 | ||
7030 | if (!list_empty(&mddev->disks)) { | |
7031 | if (mddev->pers) | |
7032 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7033 | (unsigned long long) |
7034 | mddev->array_sectors / 2); | |
1da177e4 LT |
7035 | else |
7036 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7037 | (unsigned long long)sectors / 2); |
1da177e4 | 7038 | } |
1cd6bf19 N |
7039 | if (mddev->persistent) { |
7040 | if (mddev->major_version != 0 || | |
7041 | mddev->minor_version != 90) { | |
7042 | seq_printf(seq," super %d.%d", | |
7043 | mddev->major_version, | |
7044 | mddev->minor_version); | |
7045 | } | |
e691063a N |
7046 | } else if (mddev->external) |
7047 | seq_printf(seq, " super external:%s", | |
7048 | mddev->metadata_type); | |
7049 | else | |
1cd6bf19 | 7050 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7051 | |
7052 | if (mddev->pers) { | |
d710e138 | 7053 | mddev->pers->status(seq, mddev); |
1da177e4 | 7054 | seq_printf(seq, "\n "); |
8e1b39d6 N |
7055 | if (mddev->pers->sync_request) { |
7056 | if (mddev->curr_resync > 2) { | |
d710e138 | 7057 | status_resync(seq, mddev); |
8e1b39d6 | 7058 | seq_printf(seq, "\n "); |
72f36d59 | 7059 | } else if (mddev->curr_resync >= 1) |
8e1b39d6 N |
7060 | seq_printf(seq, "\tresync=DELAYED\n "); |
7061 | else if (mddev->recovery_cp < MaxSector) | |
7062 | seq_printf(seq, "\tresync=PENDING\n "); | |
7063 | } | |
32a7627c N |
7064 | } else |
7065 | seq_printf(seq, "\n "); | |
7066 | ||
57148964 | 7067 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7068 | |
7069 | seq_printf(seq, "\n"); | |
7070 | } | |
7071 | mddev_unlock(mddev); | |
7072 | ||
7073 | return 0; | |
7074 | } | |
7075 | ||
110518bc | 7076 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7077 | .start = md_seq_start, |
7078 | .next = md_seq_next, | |
7079 | .stop = md_seq_stop, | |
7080 | .show = md_seq_show, | |
7081 | }; | |
7082 | ||
7083 | static int md_seq_open(struct inode *inode, struct file *file) | |
7084 | { | |
f1514638 | 7085 | struct seq_file *seq; |
1da177e4 LT |
7086 | int error; |
7087 | ||
7088 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7089 | if (error) |
f1514638 KS |
7090 | return error; |
7091 | ||
7092 | seq = file->private_data; | |
7093 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7094 | return error; |
7095 | } | |
7096 | ||
d7603b7e N |
7097 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7098 | { | |
f1514638 | 7099 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7100 | int mask; |
7101 | ||
7102 | poll_wait(filp, &md_event_waiters, wait); | |
7103 | ||
7104 | /* always allow read */ | |
7105 | mask = POLLIN | POLLRDNORM; | |
7106 | ||
f1514638 | 7107 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7108 | mask |= POLLERR | POLLPRI; |
7109 | return mask; | |
7110 | } | |
7111 | ||
fa027c2a | 7112 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7113 | .owner = THIS_MODULE, |
1da177e4 LT |
7114 | .open = md_seq_open, |
7115 | .read = seq_read, | |
7116 | .llseek = seq_lseek, | |
c3f94b40 | 7117 | .release = seq_release_private, |
d7603b7e | 7118 | .poll = mdstat_poll, |
1da177e4 LT |
7119 | }; |
7120 | ||
84fc4b56 | 7121 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7122 | { |
1da177e4 | 7123 | spin_lock(&pers_lock); |
2604b703 N |
7124 | list_add_tail(&p->list, &pers_list); |
7125 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
7126 | spin_unlock(&pers_lock); |
7127 | return 0; | |
7128 | } | |
7129 | ||
84fc4b56 | 7130 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7131 | { |
2604b703 | 7132 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 7133 | spin_lock(&pers_lock); |
2604b703 | 7134 | list_del_init(&p->list); |
1da177e4 LT |
7135 | spin_unlock(&pers_lock); |
7136 | return 0; | |
7137 | } | |
7138 | ||
fd01b88c | 7139 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7140 | { |
3cb03002 | 7141 | struct md_rdev * rdev; |
1da177e4 | 7142 | int idle; |
eea1bf38 | 7143 | int curr_events; |
1da177e4 LT |
7144 | |
7145 | idle = 1; | |
4b80991c N |
7146 | rcu_read_lock(); |
7147 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7148 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7149 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7150 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7151 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7152 | /* sync IO will cause sync_io to increase before the disk_stats |
7153 | * as sync_io is counted when a request starts, and | |
7154 | * disk_stats is counted when it completes. | |
7155 | * So resync activity will cause curr_events to be smaller than | |
7156 | * when there was no such activity. | |
7157 | * non-sync IO will cause disk_stat to increase without | |
7158 | * increasing sync_io so curr_events will (eventually) | |
7159 | * be larger than it was before. Once it becomes | |
7160 | * substantially larger, the test below will cause | |
7161 | * the array to appear non-idle, and resync will slow | |
7162 | * down. | |
7163 | * If there is a lot of outstanding resync activity when | |
7164 | * we set last_event to curr_events, then all that activity | |
7165 | * completing might cause the array to appear non-idle | |
7166 | * and resync will be slowed down even though there might | |
7167 | * not have been non-resync activity. This will only | |
7168 | * happen once though. 'last_events' will soon reflect | |
7169 | * the state where there is little or no outstanding | |
7170 | * resync requests, and further resync activity will | |
7171 | * always make curr_events less than last_events. | |
c0e48521 | 7172 | * |
1da177e4 | 7173 | */ |
eea1bf38 | 7174 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7175 | rdev->last_events = curr_events; |
7176 | idle = 0; | |
7177 | } | |
7178 | } | |
4b80991c | 7179 | rcu_read_unlock(); |
1da177e4 LT |
7180 | return idle; |
7181 | } | |
7182 | ||
fd01b88c | 7183 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7184 | { |
7185 | /* another "blocks" (512byte) blocks have been synced */ | |
7186 | atomic_sub(blocks, &mddev->recovery_active); | |
7187 | wake_up(&mddev->recovery_wait); | |
7188 | if (!ok) { | |
dfc70645 | 7189 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7190 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7191 | md_wakeup_thread(mddev->thread); |
7192 | // stop recovery, signal do_sync .... | |
7193 | } | |
7194 | } | |
7195 | ||
7196 | ||
06d91a5f N |
7197 | /* md_write_start(mddev, bi) |
7198 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7199 | * in superblock) before writing, schedule a superblock update |
7200 | * and wait for it to complete. | |
06d91a5f | 7201 | */ |
fd01b88c | 7202 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7203 | { |
0fd62b86 | 7204 | int did_change = 0; |
06d91a5f | 7205 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7206 | return; |
06d91a5f | 7207 | |
f91de92e N |
7208 | BUG_ON(mddev->ro == 1); |
7209 | if (mddev->ro == 2) { | |
7210 | /* need to switch to read/write */ | |
7211 | mddev->ro = 0; | |
7212 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7213 | md_wakeup_thread(mddev->thread); | |
25156198 | 7214 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7215 | did_change = 1; |
f91de92e | 7216 | } |
06d91a5f | 7217 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7218 | if (mddev->safemode == 1) |
7219 | mddev->safemode = 0; | |
06d91a5f | 7220 | if (mddev->in_sync) { |
a9701a30 | 7221 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
7222 | if (mddev->in_sync) { |
7223 | mddev->in_sync = 0; | |
850b2b42 | 7224 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 7225 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 7226 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7227 | did_change = 1; |
3d310eb7 | 7228 | } |
a9701a30 | 7229 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 7230 | } |
0fd62b86 | 7231 | if (did_change) |
00bcb4ac | 7232 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7233 | wait_event(mddev->sb_wait, |
09a44cc1 | 7234 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 LT |
7235 | } |
7236 | ||
fd01b88c | 7237 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7238 | { |
7239 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7240 | if (mddev->safemode == 2) | |
7241 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7242 | else if (mddev->safemode_delay) |
1da177e4 LT |
7243 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7244 | } | |
7245 | } | |
7246 | ||
2a2275d6 N |
7247 | /* md_allow_write(mddev) |
7248 | * Calling this ensures that the array is marked 'active' so that writes | |
7249 | * may proceed without blocking. It is important to call this before | |
7250 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7251 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
7252 | * |
7253 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
7254 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 7255 | */ |
fd01b88c | 7256 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7257 | { |
7258 | if (!mddev->pers) | |
b5470dc5 | 7259 | return 0; |
2a2275d6 | 7260 | if (mddev->ro) |
b5470dc5 | 7261 | return 0; |
1a0fd497 | 7262 | if (!mddev->pers->sync_request) |
b5470dc5 | 7263 | return 0; |
2a2275d6 N |
7264 | |
7265 | spin_lock_irq(&mddev->write_lock); | |
7266 | if (mddev->in_sync) { | |
7267 | mddev->in_sync = 0; | |
7268 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 7269 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
7270 | if (mddev->safemode_delay && |
7271 | mddev->safemode == 0) | |
7272 | mddev->safemode = 1; | |
7273 | spin_unlock_irq(&mddev->write_lock); | |
7274 | md_update_sb(mddev, 0); | |
00bcb4ac | 7275 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
7276 | } else |
7277 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 | 7278 | |
070dc6dd | 7279 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
7280 | return -EAGAIN; |
7281 | else | |
7282 | return 0; | |
2a2275d6 N |
7283 | } |
7284 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7285 | ||
1da177e4 LT |
7286 | #define SYNC_MARKS 10 |
7287 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7288 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7289 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7290 | { |
4ed8731d | 7291 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7292 | struct mddev *mddev2; |
1da177e4 LT |
7293 | unsigned int currspeed = 0, |
7294 | window; | |
57afd89f | 7295 | sector_t max_sectors,j, io_sectors; |
1da177e4 | 7296 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7297 | unsigned long update_time; |
1da177e4 LT |
7298 | sector_t mark_cnt[SYNC_MARKS]; |
7299 | int last_mark,m; | |
7300 | struct list_head *tmp; | |
7301 | sector_t last_check; | |
57afd89f | 7302 | int skipped = 0; |
3cb03002 | 7303 | struct md_rdev *rdev; |
61df9d91 | 7304 | char *desc; |
7c2c57c9 | 7305 | struct blk_plug plug; |
1da177e4 LT |
7306 | |
7307 | /* just incase thread restarts... */ | |
7308 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7309 | return; | |
5fd6c1dc N |
7310 | if (mddev->ro) /* never try to sync a read-only array */ |
7311 | return; | |
1da177e4 | 7312 | |
61df9d91 N |
7313 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
7314 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
7315 | desc = "data-check"; | |
7316 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7317 | desc = "requested-resync"; | |
7318 | else | |
7319 | desc = "resync"; | |
7320 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7321 | desc = "reshape"; | |
7322 | else | |
7323 | desc = "recovery"; | |
7324 | ||
1da177e4 LT |
7325 | /* we overload curr_resync somewhat here. |
7326 | * 0 == not engaged in resync at all | |
7327 | * 2 == checking that there is no conflict with another sync | |
7328 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7329 | * commense | |
7330 | * other == active in resync - this many blocks | |
7331 | * | |
7332 | * Before starting a resync we must have set curr_resync to | |
7333 | * 2, and then checked that every "conflicting" array has curr_resync | |
7334 | * less than ours. When we find one that is the same or higher | |
7335 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7336 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7337 | * This will mean we have to start checking from the beginning again. | |
7338 | * | |
7339 | */ | |
7340 | ||
7341 | do { | |
7342 | mddev->curr_resync = 2; | |
7343 | ||
7344 | try_again: | |
404e4b43 | 7345 | if (kthread_should_stop()) |
6985c43f | 7346 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
404e4b43 N |
7347 | |
7348 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
1da177e4 | 7349 | goto skip; |
29ac4aa3 | 7350 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7351 | if (mddev2 == mddev) |
7352 | continue; | |
90b08710 BS |
7353 | if (!mddev->parallel_resync |
7354 | && mddev2->curr_resync | |
7355 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7356 | DEFINE_WAIT(wq); |
7357 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7358 | /* arbitrarily yield */ | |
7359 | mddev->curr_resync = 1; | |
7360 | wake_up(&resync_wait); | |
7361 | } | |
7362 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7363 | /* no need to wait here, we can wait the next | |
7364 | * time 'round when curr_resync == 2 | |
7365 | */ | |
7366 | continue; | |
9744197c N |
7367 | /* We need to wait 'interruptible' so as not to |
7368 | * contribute to the load average, and not to | |
7369 | * be caught by 'softlockup' | |
7370 | */ | |
7371 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 7372 | if (!kthread_should_stop() && |
8712e553 | 7373 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
7374 | printk(KERN_INFO "md: delaying %s of %s" |
7375 | " until %s has finished (they" | |
1da177e4 | 7376 | " share one or more physical units)\n", |
61df9d91 | 7377 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 7378 | mddev_put(mddev2); |
9744197c N |
7379 | if (signal_pending(current)) |
7380 | flush_signals(current); | |
1da177e4 LT |
7381 | schedule(); |
7382 | finish_wait(&resync_wait, &wq); | |
7383 | goto try_again; | |
7384 | } | |
7385 | finish_wait(&resync_wait, &wq); | |
7386 | } | |
7387 | } | |
7388 | } while (mddev->curr_resync < 2); | |
7389 | ||
5fd6c1dc | 7390 | j = 0; |
9d88883e | 7391 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7392 | /* resync follows the size requested by the personality, |
57afd89f | 7393 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7394 | */ |
7395 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7396 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7397 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7398 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7399 | j = mddev->resync_min; | |
7400 | else if (!mddev->bitmap) | |
5fd6c1dc | 7401 | j = mddev->recovery_cp; |
5e96ee65 | 7402 | |
ccfcc3c1 | 7403 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7404 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7405 | else { |
1da177e4 | 7406 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7407 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7408 | j = MaxSector; |
4e59ca7d | 7409 | rcu_read_lock(); |
dafb20fa | 7410 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc N |
7411 | if (rdev->raid_disk >= 0 && |
7412 | !test_bit(Faulty, &rdev->flags) && | |
7413 | !test_bit(In_sync, &rdev->flags) && | |
7414 | rdev->recovery_offset < j) | |
7415 | j = rdev->recovery_offset; | |
4e59ca7d | 7416 | rcu_read_unlock(); |
5fd6c1dc | 7417 | } |
1da177e4 | 7418 | |
61df9d91 N |
7419 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
7420 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
7421 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 7422 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
7423 | "(but not more than %d KB/sec) for %s.\n", |
7424 | speed_max(mddev), desc); | |
1da177e4 | 7425 | |
eea1bf38 | 7426 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 7427 | |
57afd89f | 7428 | io_sectors = 0; |
1da177e4 LT |
7429 | for (m = 0; m < SYNC_MARKS; m++) { |
7430 | mark[m] = jiffies; | |
57afd89f | 7431 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
7432 | } |
7433 | last_mark = 0; | |
7434 | mddev->resync_mark = mark[last_mark]; | |
7435 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
7436 | ||
7437 | /* | |
7438 | * Tune reconstruction: | |
7439 | */ | |
7440 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
7441 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
7442 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
7443 | |
7444 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
7445 | last_check = 0; |
7446 | ||
7447 | if (j>2) { | |
7448 | printk(KERN_INFO | |
61df9d91 N |
7449 | "md: resuming %s of %s from checkpoint.\n", |
7450 | desc, mdname(mddev)); | |
1da177e4 | 7451 | mddev->curr_resync = j; |
72f36d59 N |
7452 | } else |
7453 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 7454 | mddev->curr_resync_completed = j; |
72f36d59 N |
7455 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
7456 | md_new_event(mddev); | |
54f89341 | 7457 | update_time = jiffies; |
1da177e4 | 7458 | |
7c2c57c9 | 7459 | blk_start_plug(&plug); |
1da177e4 | 7460 | while (j < max_sectors) { |
57afd89f | 7461 | sector_t sectors; |
1da177e4 | 7462 | |
57afd89f | 7463 | skipped = 0; |
97e4f42d | 7464 | |
7a91ee1f N |
7465 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7466 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
7467 | (mddev->curr_resync - mddev->curr_resync_completed) | |
7468 | > (max_sectors >> 4)) || | |
54f89341 | 7469 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f N |
7470 | (j - mddev->curr_resync_completed)*2 |
7471 | >= mddev->resync_max - mddev->curr_resync_completed | |
7472 | )) { | |
97e4f42d | 7473 | /* time to update curr_resync_completed */ |
97e4f42d N |
7474 | wait_event(mddev->recovery_wait, |
7475 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 7476 | mddev->curr_resync_completed = j; |
35d78c66 | 7477 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
7478 | j > mddev->recovery_cp) | |
7479 | mddev->recovery_cp = j; | |
54f89341 | 7480 | update_time = jiffies; |
070dc6dd | 7481 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 7482 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 7483 | } |
acb180b0 | 7484 | |
e62e58a5 N |
7485 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
7486 | /* As this condition is controlled by user-space, | |
7487 | * we can block indefinitely, so use '_interruptible' | |
7488 | * to avoid triggering warnings. | |
7489 | */ | |
7490 | flush_signals(current); /* just in case */ | |
7491 | wait_event_interruptible(mddev->recovery_wait, | |
7492 | mddev->resync_max > j | |
7493 | || kthread_should_stop()); | |
7494 | } | |
acb180b0 N |
7495 | |
7496 | if (kthread_should_stop()) | |
7497 | goto interrupted; | |
7498 | ||
57afd89f | 7499 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 7500 | currspeed < speed_min(mddev)); |
57afd89f | 7501 | if (sectors == 0) { |
dfc70645 | 7502 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
7503 | goto out; |
7504 | } | |
57afd89f N |
7505 | |
7506 | if (!skipped) { /* actual IO requested */ | |
7507 | io_sectors += sectors; | |
7508 | atomic_add(sectors, &mddev->recovery_active); | |
7509 | } | |
7510 | ||
e875ecea N |
7511 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7512 | break; | |
7513 | ||
1da177e4 | 7514 | j += sectors; |
72f36d59 N |
7515 | if (j > 2) |
7516 | mddev->curr_resync = j; | |
ff4e8d9a | 7517 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 7518 | if (last_check == 0) |
e875ecea | 7519 | /* this is the earliest that rebuild will be |
d7603b7e N |
7520 | * visible in /proc/mdstat |
7521 | */ | |
7522 | md_new_event(mddev); | |
57afd89f N |
7523 | |
7524 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
7525 | continue; |
7526 | ||
57afd89f | 7527 | last_check = io_sectors; |
1da177e4 LT |
7528 | repeat: |
7529 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
7530 | /* step marks */ | |
7531 | int next = (last_mark+1) % SYNC_MARKS; | |
7532 | ||
7533 | mddev->resync_mark = mark[next]; | |
7534 | mddev->resync_mark_cnt = mark_cnt[next]; | |
7535 | mark[next] = jiffies; | |
57afd89f | 7536 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
7537 | last_mark = next; |
7538 | } | |
7539 | ||
7540 | ||
c6207277 N |
7541 | if (kthread_should_stop()) |
7542 | goto interrupted; | |
7543 | ||
1da177e4 LT |
7544 | |
7545 | /* | |
7546 | * this loop exits only if either when we are slower than | |
7547 | * the 'hard' speed limit, or the system was IO-idle for | |
7548 | * a jiffy. | |
7549 | * the system might be non-idle CPU-wise, but we only care | |
7550 | * about not overloading the IO subsystem. (things like an | |
7551 | * e2fsck being done on the RAID array should execute fast) | |
7552 | */ | |
1da177e4 LT |
7553 | cond_resched(); |
7554 | ||
57afd89f N |
7555 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
7556 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 7557 | |
88202a0c N |
7558 | if (currspeed > speed_min(mddev)) { |
7559 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 7560 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 7561 | msleep(500); |
1da177e4 LT |
7562 | goto repeat; |
7563 | } | |
7564 | } | |
7565 | } | |
61df9d91 | 7566 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
7567 | /* |
7568 | * this also signals 'finished resyncing' to md_stop | |
7569 | */ | |
7570 | out: | |
7c2c57c9 | 7571 | blk_finish_plug(&plug); |
1da177e4 LT |
7572 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7573 | ||
7574 | /* tell personality that we are finished */ | |
57afd89f | 7575 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 7576 | |
dfc70645 | 7577 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7578 | mddev->curr_resync > 2) { |
7579 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7580 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7581 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7582 | printk(KERN_INFO | |
61df9d91 N |
7583 | "md: checkpointing %s of %s.\n", |
7584 | desc, mdname(mddev)); | |
0a19caab | 7585 | if (test_bit(MD_RECOVERY_ERROR, |
7586 | &mddev->recovery)) | |
7587 | mddev->recovery_cp = | |
7588 | mddev->curr_resync_completed; | |
7589 | else | |
7590 | mddev->recovery_cp = | |
7591 | mddev->curr_resync; | |
5fd6c1dc N |
7592 | } |
7593 | } else | |
7594 | mddev->recovery_cp = MaxSector; | |
7595 | } else { | |
7596 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7597 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 7598 | rcu_read_lock(); |
dafb20fa | 7599 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7600 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7601 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7602 | !test_bit(Faulty, &rdev->flags) && |
7603 | !test_bit(In_sync, &rdev->flags) && | |
7604 | rdev->recovery_offset < mddev->curr_resync) | |
7605 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7606 | rcu_read_unlock(); |
5fd6c1dc | 7607 | } |
1da177e4 | 7608 | } |
db91ff55 | 7609 | skip: |
17571284 | 7610 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7611 | |
c07b70ad N |
7612 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7613 | /* We completed so min/max setting can be forgotten if used. */ | |
7614 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7615 | mddev->resync_min = 0; | |
7616 | mddev->resync_max = MaxSector; | |
7617 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7618 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 LT |
7619 | mddev->curr_resync = 0; |
7620 | wake_up(&resync_wait); | |
7621 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
7622 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
7623 | return; |
7624 | ||
7625 | interrupted: | |
7626 | /* | |
7627 | * got a signal, exit. | |
7628 | */ | |
7629 | printk(KERN_INFO | |
7630 | "md: md_do_sync() got signal ... exiting\n"); | |
7631 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7632 | goto out; | |
7633 | ||
1da177e4 | 7634 | } |
29269553 | 7635 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 7636 | |
fd01b88c | 7637 | static int remove_and_add_spares(struct mddev *mddev) |
b4c4c7b8 | 7638 | { |
3cb03002 | 7639 | struct md_rdev *rdev; |
b4c4c7b8 | 7640 | int spares = 0; |
f2a371c5 | 7641 | int removed = 0; |
b4c4c7b8 | 7642 | |
dafb20fa | 7643 | rdev_for_each(rdev, mddev) |
b4c4c7b8 | 7644 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 7645 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7646 | (test_bit(Faulty, &rdev->flags) || |
7647 | ! test_bit(In_sync, &rdev->flags)) && | |
7648 | atomic_read(&rdev->nr_pending)==0) { | |
7649 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 7650 | mddev, rdev) == 0) { |
36fad858 | 7651 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 7652 | rdev->raid_disk = -1; |
f2a371c5 | 7653 | removed++; |
b4c4c7b8 N |
7654 | } |
7655 | } | |
f2a371c5 N |
7656 | if (removed) |
7657 | sysfs_notify(&mddev->kobj, NULL, | |
7658 | "degraded"); | |
7659 | ||
b4c4c7b8 | 7660 | |
dafb20fa | 7661 | rdev_for_each(rdev, mddev) { |
7bfec5f3 N |
7662 | if (rdev->raid_disk >= 0 && |
7663 | !test_bit(In_sync, &rdev->flags) && | |
7664 | !test_bit(Faulty, &rdev->flags)) | |
7665 | spares++; | |
7666 | if (rdev->raid_disk < 0 | |
7667 | && !test_bit(Faulty, &rdev->flags)) { | |
7668 | rdev->recovery_offset = 0; | |
7669 | if (mddev->pers-> | |
7670 | hot_add_disk(mddev, rdev) == 0) { | |
7671 | if (sysfs_link_rdev(mddev, rdev)) | |
7672 | /* failure here is OK */; | |
dfc70645 | 7673 | spares++; |
7bfec5f3 N |
7674 | md_new_event(mddev); |
7675 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 | 7676 | } |
dfc70645 | 7677 | } |
b4c4c7b8 | 7678 | } |
6dafab6b N |
7679 | if (removed) |
7680 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 N |
7681 | return spares; |
7682 | } | |
7ebc0be7 | 7683 | |
fd01b88c | 7684 | static void reap_sync_thread(struct mddev *mddev) |
7ebc0be7 | 7685 | { |
3cb03002 | 7686 | struct md_rdev *rdev; |
7ebc0be7 N |
7687 | |
7688 | /* resync has finished, collect result */ | |
01f96c0a | 7689 | md_unregister_thread(&mddev->sync_thread); |
7ebc0be7 N |
7690 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
7691 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
7692 | /* success...*/ | |
7693 | /* activate any spares */ | |
6dafab6b | 7694 | if (mddev->pers->spare_active(mddev)) { |
7ebc0be7 N |
7695 | sysfs_notify(&mddev->kobj, NULL, |
7696 | "degraded"); | |
6dafab6b N |
7697 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
7698 | } | |
7ebc0be7 N |
7699 | } |
7700 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
7701 | mddev->pers->finish_reshape) | |
7702 | mddev->pers->finish_reshape(mddev); | |
7ebc0be7 | 7703 | |
d70ed2e4 AW |
7704 | /* If array is no-longer degraded, then any saved_raid_disk |
7705 | * information must be scrapped. Also if any device is now | |
7706 | * In_sync we must scrape the saved_raid_disk for that device | |
7707 | * do the superblock for an incrementally recovered device | |
7708 | * written out. | |
7ebc0be7 | 7709 | */ |
dafb20fa | 7710 | rdev_for_each(rdev, mddev) |
d70ed2e4 AW |
7711 | if (!mddev->degraded || |
7712 | test_bit(In_sync, &rdev->flags)) | |
7ebc0be7 N |
7713 | rdev->saved_raid_disk = -1; |
7714 | ||
d70ed2e4 | 7715 | md_update_sb(mddev, 1); |
7ebc0be7 N |
7716 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
7717 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7718 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7719 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7720 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7721 | /* flag recovery needed just to double check */ | |
7722 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7723 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7724 | md_new_event(mddev); | |
768e587e JB |
7725 | if (mddev->event_work.func) |
7726 | queue_work(md_misc_wq, &mddev->event_work); | |
7ebc0be7 N |
7727 | } |
7728 | ||
1da177e4 LT |
7729 | /* |
7730 | * This routine is regularly called by all per-raid-array threads to | |
7731 | * deal with generic issues like resync and super-block update. | |
7732 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7733 | * need this as they never do any recovery or update the superblock. | |
7734 | * | |
7735 | * It does not do any resync itself, but rather "forks" off other threads | |
7736 | * to do that as needed. | |
7737 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7738 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7739 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7740 | * and wakeups up this thread which will reap the thread and finish up. |
7741 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7742 | * | |
7743 | * The overall approach is: | |
7744 | * 1/ if the superblock needs updating, update it. | |
7745 | * 2/ If a recovery thread is running, don't do anything else. | |
7746 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7747 | * 4/ If there are any faulty devices, remove them. | |
7748 | * 5/ If array is degraded, try to add spares devices | |
7749 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7750 | */ | |
fd01b88c | 7751 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 7752 | { |
68866e42 JB |
7753 | if (mddev->suspended) |
7754 | return; | |
7755 | ||
5f40402d | 7756 | if (mddev->bitmap) |
aa5cbd10 | 7757 | bitmap_daemon_work(mddev); |
1da177e4 | 7758 | |
fca4d848 | 7759 | if (signal_pending(current)) { |
31a59e34 | 7760 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7761 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7762 | mdname(mddev)); | |
7763 | mddev->safemode = 2; | |
7764 | } | |
7765 | flush_signals(current); | |
7766 | } | |
7767 | ||
c89a8eee N |
7768 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7769 | return; | |
1da177e4 | 7770 | if ( ! ( |
126925c0 | 7771 | (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7772 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7773 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7774 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7775 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7776 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7777 | )) |
7778 | return; | |
fca4d848 | 7779 | |
df5b89b3 | 7780 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7781 | int spares = 0; |
fca4d848 | 7782 | |
c89a8eee N |
7783 | if (mddev->ro) { |
7784 | /* Only thing we do on a ro array is remove | |
7785 | * failed devices. | |
7786 | */ | |
3cb03002 | 7787 | struct md_rdev *rdev; |
dafb20fa | 7788 | rdev_for_each(rdev, mddev) |
a8c42c7f N |
7789 | if (rdev->raid_disk >= 0 && |
7790 | !test_bit(Blocked, &rdev->flags) && | |
7791 | test_bit(Faulty, &rdev->flags) && | |
7792 | atomic_read(&rdev->nr_pending)==0) { | |
7793 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 7794 | mddev, rdev) == 0) { |
36fad858 | 7795 | sysfs_unlink_rdev(mddev, rdev); |
a8c42c7f N |
7796 | rdev->raid_disk = -1; |
7797 | } | |
7798 | } | |
c89a8eee N |
7799 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7800 | goto unlock; | |
7801 | } | |
7802 | ||
31a59e34 | 7803 | if (!mddev->external) { |
0fd62b86 | 7804 | int did_change = 0; |
31a59e34 N |
7805 | spin_lock_irq(&mddev->write_lock); |
7806 | if (mddev->safemode && | |
7807 | !atomic_read(&mddev->writes_pending) && | |
7808 | !mddev->in_sync && | |
7809 | mddev->recovery_cp == MaxSector) { | |
7810 | mddev->in_sync = 1; | |
0fd62b86 | 7811 | did_change = 1; |
070dc6dd | 7812 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7813 | } |
7814 | if (mddev->safemode == 1) | |
7815 | mddev->safemode = 0; | |
7816 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7817 | if (did_change) |
00bcb4ac | 7818 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7819 | } |
fca4d848 | 7820 | |
850b2b42 N |
7821 | if (mddev->flags) |
7822 | md_update_sb(mddev, 0); | |
06d91a5f | 7823 | |
1da177e4 LT |
7824 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7825 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7826 | /* resync/recovery still happening */ | |
7827 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7828 | goto unlock; | |
7829 | } | |
7830 | if (mddev->sync_thread) { | |
7ebc0be7 | 7831 | reap_sync_thread(mddev); |
1da177e4 LT |
7832 | goto unlock; |
7833 | } | |
72a23c21 NB |
7834 | /* Set RUNNING before clearing NEEDED to avoid |
7835 | * any transients in the value of "sync_action". | |
7836 | */ | |
72f36d59 | 7837 | mddev->curr_resync_completed = 0; |
72a23c21 | 7838 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
24dd469d N |
7839 | /* Clear some bits that don't mean anything, but |
7840 | * might be left set | |
7841 | */ | |
24dd469d N |
7842 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7843 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7844 | |
ed209584 N |
7845 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
7846 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
5fd6c1dc | 7847 | goto unlock; |
1da177e4 LT |
7848 | /* no recovery is running. |
7849 | * remove any failed drives, then | |
7850 | * add spares if possible. | |
72f36d59 | 7851 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
7852 | * the personality to fail the re-add. |
7853 | */ | |
1da177e4 | 7854 | |
b4c4c7b8 | 7855 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7856 | if (mddev->pers->check_reshape == NULL || |
7857 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7858 | /* Cannot proceed */ |
7859 | goto unlock; | |
7860 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7861 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 7862 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
7863 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7864 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7865 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7866 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7867 | } else if (mddev->recovery_cp < MaxSector) { |
7868 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7869 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7870 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7871 | /* nothing to be done ... */ | |
1da177e4 | 7872 | goto unlock; |
24dd469d | 7873 | |
1da177e4 | 7874 | if (mddev->pers->sync_request) { |
ef99bf48 | 7875 | if (spares) { |
a654b9d8 N |
7876 | /* We are adding a device or devices to an array |
7877 | * which has the bitmap stored on all devices. | |
7878 | * So make sure all bitmap pages get written | |
7879 | */ | |
7880 | bitmap_write_all(mddev->bitmap); | |
7881 | } | |
1da177e4 LT |
7882 | mddev->sync_thread = md_register_thread(md_do_sync, |
7883 | mddev, | |
0da3c619 | 7884 | "resync"); |
1da177e4 LT |
7885 | if (!mddev->sync_thread) { |
7886 | printk(KERN_ERR "%s: could not start resync" | |
7887 | " thread...\n", | |
7888 | mdname(mddev)); | |
7889 | /* leave the spares where they are, it shouldn't hurt */ | |
7ebc0be7 N |
7890 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
7891 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7892 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7893 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7894 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
d7603b7e | 7895 | } else |
1da177e4 | 7896 | md_wakeup_thread(mddev->sync_thread); |
00bcb4ac | 7897 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7898 | md_new_event(mddev); |
1da177e4 LT |
7899 | } |
7900 | unlock: | |
72a23c21 NB |
7901 | if (!mddev->sync_thread) { |
7902 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7903 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7904 | &mddev->recovery)) | |
0c3573f1 | 7905 | if (mddev->sysfs_action) |
00bcb4ac | 7906 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7907 | } |
1da177e4 LT |
7908 | mddev_unlock(mddev); |
7909 | } | |
7910 | } | |
7911 | ||
fd01b88c | 7912 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 7913 | { |
00bcb4ac | 7914 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 7915 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
7916 | !test_bit(Blocked, &rdev->flags) && |
7917 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
7918 | msecs_to_jiffies(5000)); |
7919 | rdev_dec_pending(rdev, mddev); | |
7920 | } | |
7921 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7922 | ||
c6563a8c N |
7923 | void md_finish_reshape(struct mddev *mddev) |
7924 | { | |
7925 | /* called be personality module when reshape completes. */ | |
7926 | struct md_rdev *rdev; | |
7927 | ||
7928 | rdev_for_each(rdev, mddev) { | |
7929 | if (rdev->data_offset > rdev->new_data_offset) | |
7930 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
7931 | else | |
7932 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
7933 | rdev->data_offset = rdev->new_data_offset; | |
7934 | } | |
7935 | } | |
7936 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 N |
7937 | |
7938 | /* Bad block management. | |
7939 | * We can record which blocks on each device are 'bad' and so just | |
7940 | * fail those blocks, or that stripe, rather than the whole device. | |
7941 | * Entries in the bad-block table are 64bits wide. This comprises: | |
7942 | * Length of bad-range, in sectors: 0-511 for lengths 1-512 | |
7943 | * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) | |
7944 | * A 'shift' can be set so that larger blocks are tracked and | |
7945 | * consequently larger devices can be covered. | |
7946 | * 'Acknowledged' flag - 1 bit. - the most significant bit. | |
7947 | * | |
7948 | * Locking of the bad-block table uses a seqlock so md_is_badblock | |
7949 | * might need to retry if it is very unlucky. | |
7950 | * We will sometimes want to check for bad blocks in a bi_end_io function, | |
7951 | * so we use the write_seqlock_irq variant. | |
7952 | * | |
7953 | * When looking for a bad block we specify a range and want to | |
7954 | * know if any block in the range is bad. So we binary-search | |
7955 | * to the last range that starts at-or-before the given endpoint, | |
7956 | * (or "before the sector after the target range") | |
7957 | * then see if it ends after the given start. | |
7958 | * We return | |
7959 | * 0 if there are no known bad blocks in the range | |
7960 | * 1 if there are known bad block which are all acknowledged | |
7961 | * -1 if there are bad blocks which have not yet been acknowledged in metadata. | |
7962 | * plus the start/length of the first bad section we overlap. | |
7963 | */ | |
7964 | int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, | |
7965 | sector_t *first_bad, int *bad_sectors) | |
7966 | { | |
7967 | int hi; | |
ab05613a | 7968 | int lo; |
2230dfe4 | 7969 | u64 *p = bb->page; |
ab05613a | 7970 | int rv; |
2230dfe4 N |
7971 | sector_t target = s + sectors; |
7972 | unsigned seq; | |
7973 | ||
7974 | if (bb->shift > 0) { | |
7975 | /* round the start down, and the end up */ | |
7976 | s >>= bb->shift; | |
7977 | target += (1<<bb->shift) - 1; | |
7978 | target >>= bb->shift; | |
7979 | sectors = target - s; | |
7980 | } | |
7981 | /* 'target' is now the first block after the bad range */ | |
7982 | ||
7983 | retry: | |
7984 | seq = read_seqbegin(&bb->lock); | |
ab05613a | 7985 | lo = 0; |
7986 | rv = 0; | |
2230dfe4 N |
7987 | hi = bb->count; |
7988 | ||
7989 | /* Binary search between lo and hi for 'target' | |
7990 | * i.e. for the last range that starts before 'target' | |
7991 | */ | |
7992 | /* INVARIANT: ranges before 'lo' and at-or-after 'hi' | |
7993 | * are known not to be the last range before target. | |
7994 | * VARIANT: hi-lo is the number of possible | |
7995 | * ranges, and decreases until it reaches 1 | |
7996 | */ | |
7997 | while (hi - lo > 1) { | |
7998 | int mid = (lo + hi) / 2; | |
7999 | sector_t a = BB_OFFSET(p[mid]); | |
8000 | if (a < target) | |
8001 | /* This could still be the one, earlier ranges | |
8002 | * could not. */ | |
8003 | lo = mid; | |
8004 | else | |
8005 | /* This and later ranges are definitely out. */ | |
8006 | hi = mid; | |
8007 | } | |
8008 | /* 'lo' might be the last that started before target, but 'hi' isn't */ | |
8009 | if (hi > lo) { | |
8010 | /* need to check all range that end after 's' to see if | |
8011 | * any are unacknowledged. | |
8012 | */ | |
8013 | while (lo >= 0 && | |
8014 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8015 | if (BB_OFFSET(p[lo]) < target) { | |
8016 | /* starts before the end, and finishes after | |
8017 | * the start, so they must overlap | |
8018 | */ | |
8019 | if (rv != -1 && BB_ACK(p[lo])) | |
8020 | rv = 1; | |
8021 | else | |
8022 | rv = -1; | |
8023 | *first_bad = BB_OFFSET(p[lo]); | |
8024 | *bad_sectors = BB_LEN(p[lo]); | |
8025 | } | |
8026 | lo--; | |
8027 | } | |
8028 | } | |
8029 | ||
8030 | if (read_seqretry(&bb->lock, seq)) | |
8031 | goto retry; | |
8032 | ||
8033 | return rv; | |
8034 | } | |
8035 | EXPORT_SYMBOL_GPL(md_is_badblock); | |
8036 | ||
8037 | /* | |
8038 | * Add a range of bad blocks to the table. | |
8039 | * This might extend the table, or might contract it | |
8040 | * if two adjacent ranges can be merged. | |
8041 | * We binary-search to find the 'insertion' point, then | |
8042 | * decide how best to handle it. | |
8043 | */ | |
8044 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, | |
8045 | int acknowledged) | |
8046 | { | |
8047 | u64 *p; | |
8048 | int lo, hi; | |
8049 | int rv = 1; | |
8050 | ||
8051 | if (bb->shift < 0) | |
8052 | /* badblocks are disabled */ | |
8053 | return 0; | |
8054 | ||
8055 | if (bb->shift) { | |
8056 | /* round the start down, and the end up */ | |
8057 | sector_t next = s + sectors; | |
8058 | s >>= bb->shift; | |
8059 | next += (1<<bb->shift) - 1; | |
8060 | next >>= bb->shift; | |
8061 | sectors = next - s; | |
8062 | } | |
8063 | ||
8064 | write_seqlock_irq(&bb->lock); | |
8065 | ||
8066 | p = bb->page; | |
8067 | lo = 0; | |
8068 | hi = bb->count; | |
8069 | /* Find the last range that starts at-or-before 's' */ | |
8070 | while (hi - lo > 1) { | |
8071 | int mid = (lo + hi) / 2; | |
8072 | sector_t a = BB_OFFSET(p[mid]); | |
8073 | if (a <= s) | |
8074 | lo = mid; | |
8075 | else | |
8076 | hi = mid; | |
8077 | } | |
8078 | if (hi > lo && BB_OFFSET(p[lo]) > s) | |
8079 | hi = lo; | |
8080 | ||
8081 | if (hi > lo) { | |
8082 | /* we found a range that might merge with the start | |
8083 | * of our new range | |
8084 | */ | |
8085 | sector_t a = BB_OFFSET(p[lo]); | |
8086 | sector_t e = a + BB_LEN(p[lo]); | |
8087 | int ack = BB_ACK(p[lo]); | |
8088 | if (e >= s) { | |
8089 | /* Yes, we can merge with a previous range */ | |
8090 | if (s == a && s + sectors >= e) | |
8091 | /* new range covers old */ | |
8092 | ack = acknowledged; | |
8093 | else | |
8094 | ack = ack && acknowledged; | |
8095 | ||
8096 | if (e < s + sectors) | |
8097 | e = s + sectors; | |
8098 | if (e - a <= BB_MAX_LEN) { | |
8099 | p[lo] = BB_MAKE(a, e-a, ack); | |
8100 | s = e; | |
8101 | } else { | |
8102 | /* does not all fit in one range, | |
8103 | * make p[lo] maximal | |
8104 | */ | |
8105 | if (BB_LEN(p[lo]) != BB_MAX_LEN) | |
8106 | p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8107 | s = a + BB_MAX_LEN; | |
8108 | } | |
8109 | sectors = e - s; | |
8110 | } | |
8111 | } | |
8112 | if (sectors && hi < bb->count) { | |
8113 | /* 'hi' points to the first range that starts after 's'. | |
8114 | * Maybe we can merge with the start of that range */ | |
8115 | sector_t a = BB_OFFSET(p[hi]); | |
8116 | sector_t e = a + BB_LEN(p[hi]); | |
8117 | int ack = BB_ACK(p[hi]); | |
8118 | if (a <= s + sectors) { | |
8119 | /* merging is possible */ | |
8120 | if (e <= s + sectors) { | |
8121 | /* full overlap */ | |
8122 | e = s + sectors; | |
8123 | ack = acknowledged; | |
8124 | } else | |
8125 | ack = ack && acknowledged; | |
8126 | ||
8127 | a = s; | |
8128 | if (e - a <= BB_MAX_LEN) { | |
8129 | p[hi] = BB_MAKE(a, e-a, ack); | |
8130 | s = e; | |
8131 | } else { | |
8132 | p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8133 | s = a + BB_MAX_LEN; | |
8134 | } | |
8135 | sectors = e - s; | |
8136 | lo = hi; | |
8137 | hi++; | |
8138 | } | |
8139 | } | |
8140 | if (sectors == 0 && hi < bb->count) { | |
8141 | /* we might be able to combine lo and hi */ | |
8142 | /* Note: 's' is at the end of 'lo' */ | |
8143 | sector_t a = BB_OFFSET(p[hi]); | |
8144 | int lolen = BB_LEN(p[lo]); | |
8145 | int hilen = BB_LEN(p[hi]); | |
8146 | int newlen = lolen + hilen - (s - a); | |
8147 | if (s >= a && newlen < BB_MAX_LEN) { | |
8148 | /* yes, we can combine them */ | |
8149 | int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); | |
8150 | p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); | |
8151 | memmove(p + hi, p + hi + 1, | |
8152 | (bb->count - hi - 1) * 8); | |
8153 | bb->count--; | |
8154 | } | |
8155 | } | |
8156 | while (sectors) { | |
8157 | /* didn't merge (it all). | |
8158 | * Need to add a range just before 'hi' */ | |
8159 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8160 | /* No room for more */ | |
8161 | rv = 0; | |
8162 | break; | |
8163 | } else { | |
8164 | int this_sectors = sectors; | |
8165 | memmove(p + hi + 1, p + hi, | |
8166 | (bb->count - hi) * 8); | |
8167 | bb->count++; | |
8168 | ||
8169 | if (this_sectors > BB_MAX_LEN) | |
8170 | this_sectors = BB_MAX_LEN; | |
8171 | p[hi] = BB_MAKE(s, this_sectors, acknowledged); | |
8172 | sectors -= this_sectors; | |
8173 | s += this_sectors; | |
8174 | } | |
8175 | } | |
8176 | ||
8177 | bb->changed = 1; | |
de393cde N |
8178 | if (!acknowledged) |
8179 | bb->unacked_exist = 1; | |
2230dfe4 N |
8180 | write_sequnlock_irq(&bb->lock); |
8181 | ||
8182 | return rv; | |
8183 | } | |
8184 | ||
3cb03002 | 8185 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8186 | int is_new) |
2230dfe4 | 8187 | { |
c6563a8c N |
8188 | int rv; |
8189 | if (is_new) | |
8190 | s += rdev->new_data_offset; | |
8191 | else | |
8192 | s += rdev->data_offset; | |
8193 | rv = md_set_badblocks(&rdev->badblocks, | |
8194 | s, sectors, 0); | |
2230dfe4 N |
8195 | if (rv) { |
8196 | /* Make sure they get written out promptly */ | |
8bd2f0a0 | 8197 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2230dfe4 N |
8198 | set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); |
8199 | md_wakeup_thread(rdev->mddev->thread); | |
8200 | } | |
8201 | return rv; | |
8202 | } | |
8203 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8204 | ||
8205 | /* | |
8206 | * Remove a range of bad blocks from the table. | |
8207 | * This may involve extending the table if we spilt a region, | |
8208 | * but it must not fail. So if the table becomes full, we just | |
8209 | * drop the remove request. | |
8210 | */ | |
8211 | static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) | |
8212 | { | |
8213 | u64 *p; | |
8214 | int lo, hi; | |
8215 | sector_t target = s + sectors; | |
8216 | int rv = 0; | |
8217 | ||
8218 | if (bb->shift > 0) { | |
8219 | /* When clearing we round the start up and the end down. | |
8220 | * This should not matter as the shift should align with | |
8221 | * the block size and no rounding should ever be needed. | |
8222 | * However it is better the think a block is bad when it | |
8223 | * isn't than to think a block is not bad when it is. | |
8224 | */ | |
8225 | s += (1<<bb->shift) - 1; | |
8226 | s >>= bb->shift; | |
8227 | target >>= bb->shift; | |
8228 | sectors = target - s; | |
8229 | } | |
8230 | ||
8231 | write_seqlock_irq(&bb->lock); | |
8232 | ||
8233 | p = bb->page; | |
8234 | lo = 0; | |
8235 | hi = bb->count; | |
8236 | /* Find the last range that starts before 'target' */ | |
8237 | while (hi - lo > 1) { | |
8238 | int mid = (lo + hi) / 2; | |
8239 | sector_t a = BB_OFFSET(p[mid]); | |
8240 | if (a < target) | |
8241 | lo = mid; | |
8242 | else | |
8243 | hi = mid; | |
8244 | } | |
8245 | if (hi > lo) { | |
8246 | /* p[lo] is the last range that could overlap the | |
8247 | * current range. Earlier ranges could also overlap, | |
8248 | * but only this one can overlap the end of the range. | |
8249 | */ | |
8250 | if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { | |
8251 | /* Partial overlap, leave the tail of this range */ | |
8252 | int ack = BB_ACK(p[lo]); | |
8253 | sector_t a = BB_OFFSET(p[lo]); | |
8254 | sector_t end = a + BB_LEN(p[lo]); | |
8255 | ||
8256 | if (a < s) { | |
8257 | /* we need to split this range */ | |
8258 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8259 | rv = 0; | |
8260 | goto out; | |
8261 | } | |
8262 | memmove(p+lo+1, p+lo, (bb->count - lo) * 8); | |
8263 | bb->count++; | |
8264 | p[lo] = BB_MAKE(a, s-a, ack); | |
8265 | lo++; | |
8266 | } | |
8267 | p[lo] = BB_MAKE(target, end - target, ack); | |
8268 | /* there is no longer an overlap */ | |
8269 | hi = lo; | |
8270 | lo--; | |
8271 | } | |
8272 | while (lo >= 0 && | |
8273 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8274 | /* This range does overlap */ | |
8275 | if (BB_OFFSET(p[lo]) < s) { | |
8276 | /* Keep the early parts of this range. */ | |
8277 | int ack = BB_ACK(p[lo]); | |
8278 | sector_t start = BB_OFFSET(p[lo]); | |
8279 | p[lo] = BB_MAKE(start, s - start, ack); | |
8280 | /* now low doesn't overlap, so.. */ | |
8281 | break; | |
8282 | } | |
8283 | lo--; | |
8284 | } | |
8285 | /* 'lo' is strictly before, 'hi' is strictly after, | |
8286 | * anything between needs to be discarded | |
8287 | */ | |
8288 | if (hi - lo > 1) { | |
8289 | memmove(p+lo+1, p+hi, (bb->count - hi) * 8); | |
8290 | bb->count -= (hi - lo - 1); | |
8291 | } | |
8292 | } | |
8293 | ||
8294 | bb->changed = 1; | |
8295 | out: | |
8296 | write_sequnlock_irq(&bb->lock); | |
8297 | return rv; | |
8298 | } | |
8299 | ||
c6563a8c N |
8300 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8301 | int is_new) | |
2230dfe4 | 8302 | { |
c6563a8c N |
8303 | if (is_new) |
8304 | s += rdev->new_data_offset; | |
8305 | else | |
8306 | s += rdev->data_offset; | |
2230dfe4 | 8307 | return md_clear_badblocks(&rdev->badblocks, |
c6563a8c | 8308 | s, sectors); |
2230dfe4 N |
8309 | } |
8310 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8311 | ||
8312 | /* | |
8313 | * Acknowledge all bad blocks in a list. | |
8314 | * This only succeeds if ->changed is clear. It is used by | |
8315 | * in-kernel metadata updates | |
8316 | */ | |
8317 | void md_ack_all_badblocks(struct badblocks *bb) | |
8318 | { | |
8319 | if (bb->page == NULL || bb->changed) | |
8320 | /* no point even trying */ | |
8321 | return; | |
8322 | write_seqlock_irq(&bb->lock); | |
8323 | ||
ecb178bb | 8324 | if (bb->changed == 0 && bb->unacked_exist) { |
2230dfe4 N |
8325 | u64 *p = bb->page; |
8326 | int i; | |
8327 | for (i = 0; i < bb->count ; i++) { | |
8328 | if (!BB_ACK(p[i])) { | |
8329 | sector_t start = BB_OFFSET(p[i]); | |
8330 | int len = BB_LEN(p[i]); | |
8331 | p[i] = BB_MAKE(start, len, 1); | |
8332 | } | |
8333 | } | |
de393cde | 8334 | bb->unacked_exist = 0; |
2230dfe4 N |
8335 | } |
8336 | write_sequnlock_irq(&bb->lock); | |
8337 | } | |
8338 | EXPORT_SYMBOL_GPL(md_ack_all_badblocks); | |
8339 | ||
16c791a5 N |
8340 | /* sysfs access to bad-blocks list. |
8341 | * We present two files. | |
8342 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
8343 | * are recorded as bad. The list is truncated to fit within | |
8344 | * the one-page limit of sysfs. | |
8345 | * Writing "sector length" to this file adds an acknowledged | |
8346 | * bad block list. | |
8347 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
8348 | * been acknowledged. Writing to this file adds bad blocks | |
8349 | * without acknowledging them. This is largely for testing. | |
8350 | */ | |
8351 | ||
8352 | static ssize_t | |
8353 | badblocks_show(struct badblocks *bb, char *page, int unack) | |
8354 | { | |
8355 | size_t len; | |
8356 | int i; | |
8357 | u64 *p = bb->page; | |
8358 | unsigned seq; | |
8359 | ||
8360 | if (bb->shift < 0) | |
8361 | return 0; | |
8362 | ||
8363 | retry: | |
8364 | seq = read_seqbegin(&bb->lock); | |
8365 | ||
8366 | len = 0; | |
8367 | i = 0; | |
8368 | ||
8369 | while (len < PAGE_SIZE && i < bb->count) { | |
8370 | sector_t s = BB_OFFSET(p[i]); | |
8371 | unsigned int length = BB_LEN(p[i]); | |
8372 | int ack = BB_ACK(p[i]); | |
8373 | i++; | |
8374 | ||
8375 | if (unack && ack) | |
8376 | continue; | |
8377 | ||
8378 | len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", | |
8379 | (unsigned long long)s << bb->shift, | |
8380 | length << bb->shift); | |
8381 | } | |
de393cde N |
8382 | if (unack && len == 0) |
8383 | bb->unacked_exist = 0; | |
16c791a5 N |
8384 | |
8385 | if (read_seqretry(&bb->lock, seq)) | |
8386 | goto retry; | |
8387 | ||
8388 | return len; | |
8389 | } | |
8390 | ||
8391 | #define DO_DEBUG 1 | |
8392 | ||
8393 | static ssize_t | |
8394 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) | |
8395 | { | |
8396 | unsigned long long sector; | |
8397 | int length; | |
8398 | char newline; | |
8399 | #ifdef DO_DEBUG | |
8400 | /* Allow clearing via sysfs *only* for testing/debugging. | |
8401 | * Normally only a successful write may clear a badblock | |
8402 | */ | |
8403 | int clear = 0; | |
8404 | if (page[0] == '-') { | |
8405 | clear = 1; | |
8406 | page++; | |
8407 | } | |
8408 | #endif /* DO_DEBUG */ | |
8409 | ||
8410 | switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { | |
8411 | case 3: | |
8412 | if (newline != '\n') | |
8413 | return -EINVAL; | |
8414 | case 2: | |
8415 | if (length <= 0) | |
8416 | return -EINVAL; | |
8417 | break; | |
8418 | default: | |
8419 | return -EINVAL; | |
8420 | } | |
8421 | ||
8422 | #ifdef DO_DEBUG | |
8423 | if (clear) { | |
8424 | md_clear_badblocks(bb, sector, length); | |
8425 | return len; | |
8426 | } | |
8427 | #endif /* DO_DEBUG */ | |
8428 | if (md_set_badblocks(bb, sector, length, !unack)) | |
8429 | return len; | |
8430 | else | |
8431 | return -ENOSPC; | |
8432 | } | |
8433 | ||
75c96f85 AB |
8434 | static int md_notify_reboot(struct notifier_block *this, |
8435 | unsigned long code, void *x) | |
1da177e4 LT |
8436 | { |
8437 | struct list_head *tmp; | |
fd01b88c | 8438 | struct mddev *mddev; |
2dba6a91 | 8439 | int need_delay = 0; |
1da177e4 | 8440 | |
c744a65c N |
8441 | for_each_mddev(mddev, tmp) { |
8442 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8443 | if (mddev->pers) |
8444 | __md_stop_writes(mddev); | |
c744a65c N |
8445 | mddev->safemode = 2; |
8446 | mddev_unlock(mddev); | |
2dba6a91 | 8447 | } |
c744a65c | 8448 | need_delay = 1; |
1da177e4 | 8449 | } |
c744a65c N |
8450 | /* |
8451 | * certain more exotic SCSI devices are known to be | |
8452 | * volatile wrt too early system reboots. While the | |
8453 | * right place to handle this issue is the given | |
8454 | * driver, we do want to have a safe RAID driver ... | |
8455 | */ | |
8456 | if (need_delay) | |
8457 | mdelay(1000*1); | |
8458 | ||
1da177e4 LT |
8459 | return NOTIFY_DONE; |
8460 | } | |
8461 | ||
75c96f85 | 8462 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8463 | .notifier_call = md_notify_reboot, |
8464 | .next = NULL, | |
8465 | .priority = INT_MAX, /* before any real devices */ | |
8466 | }; | |
8467 | ||
8468 | static void md_geninit(void) | |
8469 | { | |
36a4e1fe | 8470 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8471 | |
c7705f34 | 8472 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8473 | } |
8474 | ||
75c96f85 | 8475 | static int __init md_init(void) |
1da177e4 | 8476 | { |
e804ac78 TH |
8477 | int ret = -ENOMEM; |
8478 | ||
ada609ee | 8479 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8480 | if (!md_wq) |
8481 | goto err_wq; | |
8482 | ||
8483 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8484 | if (!md_misc_wq) | |
8485 | goto err_misc_wq; | |
8486 | ||
8487 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8488 | goto err_md; | |
8489 | ||
8490 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8491 | goto err_mdp; | |
8492 | mdp_major = ret; | |
8493 | ||
3dbd8c2e | 8494 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
8495 | md_probe, NULL, NULL); |
8496 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8497 | md_probe, NULL, NULL); |
8498 | ||
1da177e4 | 8499 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8500 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8501 | |
8502 | md_geninit(); | |
d710e138 | 8503 | return 0; |
1da177e4 | 8504 | |
e804ac78 TH |
8505 | err_mdp: |
8506 | unregister_blkdev(MD_MAJOR, "md"); | |
8507 | err_md: | |
8508 | destroy_workqueue(md_misc_wq); | |
8509 | err_misc_wq: | |
8510 | destroy_workqueue(md_wq); | |
8511 | err_wq: | |
8512 | return ret; | |
8513 | } | |
1da177e4 LT |
8514 | |
8515 | #ifndef MODULE | |
8516 | ||
8517 | /* | |
8518 | * Searches all registered partitions for autorun RAID arrays | |
8519 | * at boot time. | |
8520 | */ | |
4d936ec1 ME |
8521 | |
8522 | static LIST_HEAD(all_detected_devices); | |
8523 | struct detected_devices_node { | |
8524 | struct list_head list; | |
8525 | dev_t dev; | |
8526 | }; | |
1da177e4 LT |
8527 | |
8528 | void md_autodetect_dev(dev_t dev) | |
8529 | { | |
4d936ec1 ME |
8530 | struct detected_devices_node *node_detected_dev; |
8531 | ||
8532 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8533 | if (node_detected_dev) { | |
8534 | node_detected_dev->dev = dev; | |
8535 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
8536 | } else { | |
8537 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
8538 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
8539 | } | |
1da177e4 LT |
8540 | } |
8541 | ||
8542 | ||
8543 | static void autostart_arrays(int part) | |
8544 | { | |
3cb03002 | 8545 | struct md_rdev *rdev; |
4d936ec1 ME |
8546 | struct detected_devices_node *node_detected_dev; |
8547 | dev_t dev; | |
8548 | int i_scanned, i_passed; | |
1da177e4 | 8549 | |
4d936ec1 ME |
8550 | i_scanned = 0; |
8551 | i_passed = 0; | |
1da177e4 | 8552 | |
4d936ec1 | 8553 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8554 | |
4d936ec1 ME |
8555 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8556 | i_scanned++; | |
8557 | node_detected_dev = list_entry(all_detected_devices.next, | |
8558 | struct detected_devices_node, list); | |
8559 | list_del(&node_detected_dev->list); | |
8560 | dev = node_detected_dev->dev; | |
8561 | kfree(node_detected_dev); | |
df968c4e | 8562 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
8563 | if (IS_ERR(rdev)) |
8564 | continue; | |
8565 | ||
b2d444d7 | 8566 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
8567 | MD_BUG(); |
8568 | continue; | |
8569 | } | |
d0fae18f | 8570 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 8571 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 8572 | i_passed++; |
1da177e4 | 8573 | } |
4d936ec1 ME |
8574 | |
8575 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
8576 | i_scanned, i_passed); | |
1da177e4 LT |
8577 | |
8578 | autorun_devices(part); | |
8579 | } | |
8580 | ||
fdee8ae4 | 8581 | #endif /* !MODULE */ |
1da177e4 LT |
8582 | |
8583 | static __exit void md_exit(void) | |
8584 | { | |
fd01b88c | 8585 | struct mddev *mddev; |
1da177e4 | 8586 | struct list_head *tmp; |
8ab5e4c1 | 8587 | |
3dbd8c2e | 8588 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 8589 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 8590 | |
3dbd8c2e | 8591 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
8592 | unregister_blkdev(mdp_major, "mdp"); |
8593 | unregister_reboot_notifier(&md_notifier); | |
8594 | unregister_sysctl_table(raid_table_header); | |
8595 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 8596 | for_each_mddev(mddev, tmp) { |
1da177e4 | 8597 | export_array(mddev); |
d3374825 | 8598 | mddev->hold_active = 0; |
1da177e4 | 8599 | } |
e804ac78 TH |
8600 | destroy_workqueue(md_misc_wq); |
8601 | destroy_workqueue(md_wq); | |
1da177e4 LT |
8602 | } |
8603 | ||
685784aa | 8604 | subsys_initcall(md_init); |
1da177e4 LT |
8605 | module_exit(md_exit) |
8606 | ||
f91de92e N |
8607 | static int get_ro(char *buffer, struct kernel_param *kp) |
8608 | { | |
8609 | return sprintf(buffer, "%d", start_readonly); | |
8610 | } | |
8611 | static int set_ro(const char *val, struct kernel_param *kp) | |
8612 | { | |
8613 | char *e; | |
8614 | int num = simple_strtoul(val, &e, 10); | |
8615 | if (*val && (*e == '\0' || *e == '\n')) { | |
8616 | start_readonly = num; | |
4dbcdc75 | 8617 | return 0; |
f91de92e N |
8618 | } |
8619 | return -EINVAL; | |
8620 | } | |
8621 | ||
80ca3a44 N |
8622 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
8623 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 8624 | |
efeb53c0 | 8625 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 8626 | |
1da177e4 LT |
8627 | EXPORT_SYMBOL(register_md_personality); |
8628 | EXPORT_SYMBOL(unregister_md_personality); | |
8629 | EXPORT_SYMBOL(md_error); | |
8630 | EXPORT_SYMBOL(md_done_sync); | |
8631 | EXPORT_SYMBOL(md_write_start); | |
8632 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
8633 | EXPORT_SYMBOL(md_register_thread); |
8634 | EXPORT_SYMBOL(md_unregister_thread); | |
8635 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
8636 | EXPORT_SYMBOL(md_check_recovery); |
8637 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 8638 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 8639 | MODULE_ALIAS("md"); |
72008652 | 8640 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |