Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | md.c : Multiple Devices driver for Linux | |
f72ffdd6 | 3 | Copyright (C) 1998, 1999, 2000 Ingo Molnar |
1da177e4 LT |
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 | ||
90b08710 | 69 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
70 | static struct workqueue_struct *md_wq; |
71 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 72 | |
746d3207 N |
73 | static int remove_and_add_spares(struct mddev *mddev, |
74 | struct md_rdev *this); | |
5aa61f42 | 75 | static void mddev_detach(struct mddev *mddev); |
746d3207 | 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 | ||
82592c38 | 112 | static struct ctl_table raid_table[] = { |
1da177e4 | 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 | ||
82592c38 | 130 | static struct ctl_table raid_dir_table[] = { |
1da177e4 | 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 | ||
82592c38 | 140 | static struct ctl_table raid_root_table[] = { |
1da177e4 | 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 | ||
d7603b7e N |
183 | /* |
184 | * We have a system wide 'event count' that is incremented | |
185 | * on any 'interesting' event, and readers of /proc/mdstat | |
186 | * can use 'poll' or 'select' to find out when the event | |
187 | * count increases. | |
188 | * | |
189 | * Events are: | |
190 | * start array, stop array, error, add device, remove device, | |
191 | * start build, activate spare | |
192 | */ | |
2989ddbd | 193 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 194 | static atomic_t md_event_count; |
fd01b88c | 195 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
196 | { |
197 | atomic_inc(&md_event_count); | |
198 | wake_up(&md_event_waiters); | |
199 | } | |
29269553 | 200 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 201 | |
c331eb04 N |
202 | /* Alternate version that can be called from interrupts |
203 | * when calling sysfs_notify isn't needed. | |
204 | */ | |
fd01b88c | 205 | static void md_new_event_inintr(struct mddev *mddev) |
c331eb04 N |
206 | { |
207 | atomic_inc(&md_event_count); | |
208 | wake_up(&md_event_waiters); | |
209 | } | |
210 | ||
1da177e4 LT |
211 | /* |
212 | * Enables to iterate over all existing md arrays | |
213 | * all_mddevs_lock protects this list. | |
214 | */ | |
215 | static LIST_HEAD(all_mddevs); | |
216 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
217 | ||
1da177e4 LT |
218 | /* |
219 | * iterates through all used mddevs in the system. | |
220 | * We take care to grab the all_mddevs_lock whenever navigating | |
221 | * the list, and to always hold a refcount when unlocked. | |
222 | * Any code which breaks out of this loop while own | |
223 | * a reference to the current mddev and must mddev_put it. | |
224 | */ | |
fd01b88c | 225 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 | 226 | \ |
f72ffdd6 | 227 | for (({ spin_lock(&all_mddevs_lock); \ |
fd01b88c N |
228 | _tmp = all_mddevs.next; \ |
229 | _mddev = NULL;}); \ | |
230 | ({ if (_tmp != &all_mddevs) \ | |
231 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 232 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
233 | if (_mddev) mddev_put(_mddev); \ |
234 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
235 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 236 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 237 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
238 | ) |
239 | ||
409c57f3 N |
240 | /* Rather than calling directly into the personality make_request function, |
241 | * IO requests come here first so that we can check if the device is | |
242 | * being suspended pending a reconfiguration. | |
243 | * We hold a refcount over the call to ->make_request. By the time that | |
244 | * call has finished, the bio has been linked into some internal structure | |
245 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
246 | */ | |
5a7bbad2 | 247 | static void md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 248 | { |
49077326 | 249 | const int rw = bio_data_dir(bio); |
fd01b88c | 250 | struct mddev *mddev = q->queuedata; |
e91ece55 | 251 | unsigned int sectors; |
49077326 | 252 | |
0ca69886 N |
253 | if (mddev == NULL || mddev->pers == NULL |
254 | || !mddev->ready) { | |
409c57f3 | 255 | bio_io_error(bio); |
5a7bbad2 | 256 | return; |
409c57f3 | 257 | } |
bbfa57c0 SR |
258 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
259 | bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS); | |
260 | return; | |
261 | } | |
0ca69886 | 262 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 263 | rcu_read_lock(); |
e9c7469b | 264 | if (mddev->suspended) { |
409c57f3 N |
265 | DEFINE_WAIT(__wait); |
266 | for (;;) { | |
267 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
268 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 269 | if (!mddev->suspended) |
409c57f3 N |
270 | break; |
271 | rcu_read_unlock(); | |
272 | schedule(); | |
273 | rcu_read_lock(); | |
274 | } | |
275 | finish_wait(&mddev->sb_wait, &__wait); | |
276 | } | |
277 | atomic_inc(&mddev->active_io); | |
278 | rcu_read_unlock(); | |
49077326 | 279 | |
e91ece55 CM |
280 | /* |
281 | * save the sectors now since our bio can | |
282 | * go away inside make_request | |
283 | */ | |
284 | sectors = bio_sectors(bio); | |
5a7bbad2 | 285 | mddev->pers->make_request(mddev, bio); |
49077326 | 286 | |
18c0b223 | 287 | generic_start_io_acct(rw, sectors, &mddev->gendisk->part0); |
49077326 | 288 | |
409c57f3 N |
289 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
290 | wake_up(&mddev->sb_wait); | |
409c57f3 N |
291 | } |
292 | ||
9e35b99c N |
293 | /* mddev_suspend makes sure no new requests are submitted |
294 | * to the device, and that any requests that have been submitted | |
295 | * are completely handled. | |
afa0f557 N |
296 | * Once mddev_detach() is called and completes, the module will be |
297 | * completely unused. | |
9e35b99c | 298 | */ |
fd01b88c | 299 | void mddev_suspend(struct mddev *mddev) |
409c57f3 N |
300 | { |
301 | BUG_ON(mddev->suspended); | |
302 | mddev->suspended = 1; | |
303 | synchronize_rcu(); | |
304 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
305 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
306 | |
307 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 308 | } |
390ee602 | 309 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 310 | |
fd01b88c | 311 | void mddev_resume(struct mddev *mddev) |
409c57f3 N |
312 | { |
313 | mddev->suspended = 0; | |
314 | wake_up(&mddev->sb_wait); | |
315 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 316 | |
47525e59 | 317 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
318 | md_wakeup_thread(mddev->thread); |
319 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 320 | } |
390ee602 | 321 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 322 | |
fd01b88c | 323 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 | 324 | { |
5c675f83 N |
325 | struct md_personality *pers = mddev->pers; |
326 | int ret = 0; | |
327 | ||
328 | rcu_read_lock(); | |
329 | if (mddev->suspended) | |
330 | ret = 1; | |
331 | else if (pers && pers->congested) | |
332 | ret = pers->congested(mddev, bits); | |
333 | rcu_read_unlock(); | |
334 | return ret; | |
335 | } | |
336 | EXPORT_SYMBOL_GPL(mddev_congested); | |
337 | static int md_congested(void *data, int bits) | |
338 | { | |
339 | struct mddev *mddev = data; | |
340 | return mddev_congested(mddev, bits); | |
3fa841d7 | 341 | } |
3fa841d7 | 342 | |
64590f45 N |
343 | static int md_mergeable_bvec(struct request_queue *q, |
344 | struct bvec_merge_data *bvm, | |
345 | struct bio_vec *biovec) | |
346 | { | |
347 | struct mddev *mddev = q->queuedata; | |
348 | int ret; | |
349 | rcu_read_lock(); | |
350 | if (mddev->suspended) { | |
351 | /* Must always allow one vec */ | |
352 | if (bvm->bi_size == 0) | |
353 | ret = biovec->bv_len; | |
354 | else | |
355 | ret = 0; | |
356 | } else { | |
357 | struct md_personality *pers = mddev->pers; | |
358 | if (pers && pers->mergeable_bvec) | |
359 | ret = pers->mergeable_bvec(mddev, bvm, biovec); | |
360 | else | |
361 | ret = biovec->bv_len; | |
362 | } | |
363 | rcu_read_unlock(); | |
364 | return ret; | |
365 | } | |
a2826aa9 | 366 | /* |
e9c7469b | 367 | * Generic flush handling for md |
a2826aa9 N |
368 | */ |
369 | ||
e9c7469b | 370 | static void md_end_flush(struct bio *bio, int err) |
a2826aa9 | 371 | { |
3cb03002 | 372 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 373 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
374 | |
375 | rdev_dec_pending(rdev, mddev); | |
376 | ||
377 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 378 | /* The pre-request flush has finished */ |
e804ac78 | 379 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
380 | } |
381 | bio_put(bio); | |
382 | } | |
383 | ||
a7a07e69 N |
384 | static void md_submit_flush_data(struct work_struct *ws); |
385 | ||
a035fc3e | 386 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 387 | { |
fd01b88c | 388 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 389 | struct md_rdev *rdev; |
a2826aa9 | 390 | |
a7a07e69 N |
391 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
392 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 393 | rcu_read_lock(); |
dafb20fa | 394 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
395 | if (rdev->raid_disk >= 0 && |
396 | !test_bit(Faulty, &rdev->flags)) { | |
397 | /* Take two references, one is dropped | |
398 | * when request finishes, one after | |
399 | * we reclaim rcu_read_lock | |
400 | */ | |
401 | struct bio *bi; | |
402 | atomic_inc(&rdev->nr_pending); | |
403 | atomic_inc(&rdev->nr_pending); | |
404 | rcu_read_unlock(); | |
b5e1b8ce | 405 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 406 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
407 | bi->bi_private = rdev; |
408 | bi->bi_bdev = rdev->bdev; | |
409 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 410 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
411 | rcu_read_lock(); |
412 | rdev_dec_pending(rdev, mddev); | |
413 | } | |
414 | rcu_read_unlock(); | |
a7a07e69 N |
415 | if (atomic_dec_and_test(&mddev->flush_pending)) |
416 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
417 | } |
418 | ||
e9c7469b | 419 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 420 | { |
fd01b88c | 421 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 422 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 423 | |
4f024f37 | 424 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 N |
425 | /* an empty barrier - all done */ |
426 | bio_endio(bio, 0); | |
427 | else { | |
e9c7469b | 428 | bio->bi_rw &= ~REQ_FLUSH; |
5a7bbad2 | 429 | mddev->pers->make_request(mddev, bio); |
a2826aa9 | 430 | } |
2b74e12e N |
431 | |
432 | mddev->flush_bio = NULL; | |
433 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
434 | } |
435 | ||
fd01b88c | 436 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 | 437 | { |
85572d7c | 438 | spin_lock_irq(&mddev->lock); |
a2826aa9 | 439 | wait_event_lock_irq(mddev->sb_wait, |
e9c7469b | 440 | !mddev->flush_bio, |
85572d7c | 441 | mddev->lock); |
e9c7469b | 442 | mddev->flush_bio = bio; |
85572d7c | 443 | spin_unlock_irq(&mddev->lock); |
a2826aa9 | 444 | |
a035fc3e N |
445 | INIT_WORK(&mddev->flush_work, submit_flushes); |
446 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 447 | } |
e9c7469b | 448 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 449 | |
74018dc3 | 450 | void md_unplug(struct blk_plug_cb *cb, bool from_schedule) |
97658cdd | 451 | { |
9cbb1750 N |
452 | struct mddev *mddev = cb->data; |
453 | md_wakeup_thread(mddev->thread); | |
454 | kfree(cb); | |
97658cdd | 455 | } |
9cbb1750 | 456 | EXPORT_SYMBOL(md_unplug); |
2ac87401 | 457 | |
fd01b88c | 458 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
459 | { |
460 | atomic_inc(&mddev->active); | |
461 | return mddev; | |
462 | } | |
463 | ||
5fd3a17e | 464 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 465 | |
fd01b88c | 466 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 467 | { |
a167f663 N |
468 | struct bio_set *bs = NULL; |
469 | ||
1da177e4 LT |
470 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
471 | return; | |
d3374825 | 472 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
473 | mddev->ctime == 0 && !mddev->hold_active) { |
474 | /* Array is not configured at all, and not held active, | |
475 | * so destroy it */ | |
af8a2434 | 476 | list_del_init(&mddev->all_mddevs); |
a167f663 N |
477 | bs = mddev->bio_set; |
478 | mddev->bio_set = NULL; | |
d3374825 | 479 | if (mddev->gendisk) { |
e804ac78 TH |
480 | /* We did a probe so need to clean up. Call |
481 | * queue_work inside the spinlock so that | |
482 | * flush_workqueue() after mddev_find will | |
483 | * succeed in waiting for the work to be done. | |
d3374825 N |
484 | */ |
485 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 486 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
487 | } else |
488 | kfree(mddev); | |
489 | } | |
490 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
491 | if (bs) |
492 | bioset_free(bs); | |
1da177e4 LT |
493 | } |
494 | ||
fd01b88c | 495 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
496 | { |
497 | mutex_init(&mddev->open_mutex); | |
498 | mutex_init(&mddev->reconfig_mutex); | |
499 | mutex_init(&mddev->bitmap_info.mutex); | |
500 | INIT_LIST_HEAD(&mddev->disks); | |
501 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
502 | init_timer(&mddev->safemode_timer); | |
503 | atomic_set(&mddev->active, 1); | |
504 | atomic_set(&mddev->openers, 0); | |
505 | atomic_set(&mddev->active_io, 0); | |
85572d7c | 506 | spin_lock_init(&mddev->lock); |
fafd7fb0 N |
507 | atomic_set(&mddev->flush_pending, 0); |
508 | init_waitqueue_head(&mddev->sb_wait); | |
509 | init_waitqueue_head(&mddev->recovery_wait); | |
510 | mddev->reshape_position = MaxSector; | |
2c810cdd | 511 | mddev->reshape_backwards = 0; |
c4a39551 | 512 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
513 | mddev->resync_min = 0; |
514 | mddev->resync_max = MaxSector; | |
515 | mddev->level = LEVEL_NONE; | |
516 | } | |
390ee602 | 517 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 518 | |
f72ffdd6 | 519 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 520 | { |
fd01b88c | 521 | struct mddev *mddev, *new = NULL; |
1da177e4 | 522 | |
8f5f02c4 N |
523 | if (unit && MAJOR(unit) != MD_MAJOR) |
524 | unit &= ~((1<<MdpMinorShift)-1); | |
525 | ||
1da177e4 LT |
526 | retry: |
527 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
528 | |
529 | if (unit) { | |
530 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
531 | if (mddev->unit == unit) { | |
532 | mddev_get(mddev); | |
533 | spin_unlock(&all_mddevs_lock); | |
534 | kfree(new); | |
535 | return mddev; | |
536 | } | |
537 | ||
538 | if (new) { | |
539 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 540 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
541 | new->hold_active = UNTIL_IOCTL; |
542 | return new; | |
1da177e4 | 543 | } |
efeb53c0 N |
544 | } else if (new) { |
545 | /* find an unused unit number */ | |
546 | static int next_minor = 512; | |
547 | int start = next_minor; | |
548 | int is_free = 0; | |
549 | int dev = 0; | |
550 | while (!is_free) { | |
551 | dev = MKDEV(MD_MAJOR, next_minor); | |
552 | next_minor++; | |
553 | if (next_minor > MINORMASK) | |
554 | next_minor = 0; | |
555 | if (next_minor == start) { | |
556 | /* Oh dear, all in use. */ | |
557 | spin_unlock(&all_mddevs_lock); | |
558 | kfree(new); | |
559 | return NULL; | |
560 | } | |
f72ffdd6 | 561 | |
efeb53c0 N |
562 | is_free = 1; |
563 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
564 | if (mddev->unit == dev) { | |
565 | is_free = 0; | |
566 | break; | |
567 | } | |
568 | } | |
569 | new->unit = dev; | |
570 | new->md_minor = MINOR(dev); | |
571 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
572 | list_add(&new->all_mddevs, &all_mddevs); |
573 | spin_unlock(&all_mddevs_lock); | |
574 | return new; | |
575 | } | |
576 | spin_unlock(&all_mddevs_lock); | |
577 | ||
9ffae0cf | 578 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
579 | if (!new) |
580 | return NULL; | |
581 | ||
1da177e4 LT |
582 | new->unit = unit; |
583 | if (MAJOR(unit) == MD_MAJOR) | |
584 | new->md_minor = MINOR(unit); | |
585 | else | |
586 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
587 | ||
fafd7fb0 | 588 | mddev_init(new); |
1da177e4 | 589 | |
1da177e4 LT |
590 | goto retry; |
591 | } | |
592 | ||
b6eb127d N |
593 | static struct attribute_group md_redundancy_group; |
594 | ||
5c47daf6 | 595 | void mddev_unlock(struct mddev *mddev) |
1da177e4 | 596 | { |
a64c876f | 597 | if (mddev->to_remove) { |
b6eb127d N |
598 | /* These cannot be removed under reconfig_mutex as |
599 | * an access to the files will try to take reconfig_mutex | |
600 | * while holding the file unremovable, which leads to | |
601 | * a deadlock. | |
bb4f1e9d N |
602 | * So hold set sysfs_active while the remove in happeing, |
603 | * and anything else which might set ->to_remove or my | |
604 | * otherwise change the sysfs namespace will fail with | |
605 | * -EBUSY if sysfs_active is still set. | |
606 | * We set sysfs_active under reconfig_mutex and elsewhere | |
607 | * test it under the same mutex to ensure its correct value | |
608 | * is seen. | |
b6eb127d | 609 | */ |
a64c876f N |
610 | struct attribute_group *to_remove = mddev->to_remove; |
611 | mddev->to_remove = NULL; | |
bb4f1e9d | 612 | mddev->sysfs_active = 1; |
b6eb127d N |
613 | mutex_unlock(&mddev->reconfig_mutex); |
614 | ||
00bcb4ac N |
615 | if (mddev->kobj.sd) { |
616 | if (to_remove != &md_redundancy_group) | |
617 | sysfs_remove_group(&mddev->kobj, to_remove); | |
618 | if (mddev->pers == NULL || | |
619 | mddev->pers->sync_request == NULL) { | |
620 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
621 | if (mddev->sysfs_action) | |
622 | sysfs_put(mddev->sysfs_action); | |
623 | mddev->sysfs_action = NULL; | |
624 | } | |
a64c876f | 625 | } |
bb4f1e9d | 626 | mddev->sysfs_active = 0; |
b6eb127d N |
627 | } else |
628 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 629 | |
751e67ca CD |
630 | /* As we've dropped the mutex we need a spinlock to |
631 | * make sure the thread doesn't disappear | |
01f96c0a N |
632 | */ |
633 | spin_lock(&pers_lock); | |
005eca5e | 634 | md_wakeup_thread(mddev->thread); |
01f96c0a | 635 | spin_unlock(&pers_lock); |
1da177e4 | 636 | } |
5c47daf6 | 637 | EXPORT_SYMBOL_GPL(mddev_unlock); |
1da177e4 | 638 | |
1ca69c4b N |
639 | static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr) |
640 | { | |
641 | struct md_rdev *rdev; | |
642 | ||
643 | rdev_for_each_rcu(rdev, mddev) | |
644 | if (rdev->desc_nr == nr) | |
645 | return rdev; | |
646 | ||
647 | return NULL; | |
648 | } | |
649 | ||
650 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 651 | { |
3cb03002 | 652 | struct md_rdev *rdev; |
1da177e4 | 653 | |
dafb20fa | 654 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
655 | if (rdev->bdev->bd_dev == dev) |
656 | return rdev; | |
159ec1fc | 657 | |
1da177e4 LT |
658 | return NULL; |
659 | } | |
660 | ||
1ca69c4b N |
661 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
662 | { | |
663 | struct md_rdev *rdev; | |
664 | ||
665 | rdev_for_each_rcu(rdev, mddev) | |
666 | if (rdev->bdev->bd_dev == dev) | |
667 | return rdev; | |
668 | ||
669 | return NULL; | |
670 | } | |
671 | ||
84fc4b56 | 672 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 673 | { |
84fc4b56 | 674 | struct md_personality *pers; |
d9d166c2 N |
675 | list_for_each_entry(pers, &pers_list, list) { |
676 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 677 | return pers; |
d9d166c2 N |
678 | if (strcmp(pers->name, clevel)==0) |
679 | return pers; | |
680 | } | |
2604b703 N |
681 | return NULL; |
682 | } | |
683 | ||
b73df2d3 | 684 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 685 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 686 | { |
57b2caa3 | 687 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 688 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
689 | } |
690 | ||
f72ffdd6 | 691 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 692 | { |
1da177e4 LT |
693 | rdev->sb_page = alloc_page(GFP_KERNEL); |
694 | if (!rdev->sb_page) { | |
695 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 696 | return -ENOMEM; |
1da177e4 LT |
697 | } |
698 | ||
699 | return 0; | |
700 | } | |
701 | ||
545c8795 | 702 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
703 | { |
704 | if (rdev->sb_page) { | |
2d1f3b5d | 705 | put_page(rdev->sb_page); |
1da177e4 LT |
706 | rdev->sb_loaded = 0; |
707 | rdev->sb_page = NULL; | |
0f420358 | 708 | rdev->sb_start = 0; |
dd8ac336 | 709 | rdev->sectors = 0; |
1da177e4 | 710 | } |
2699b672 N |
711 | if (rdev->bb_page) { |
712 | put_page(rdev->bb_page); | |
713 | rdev->bb_page = NULL; | |
714 | } | |
4fa2f327 N |
715 | kfree(rdev->badblocks.page); |
716 | rdev->badblocks.page = NULL; | |
1da177e4 | 717 | } |
545c8795 | 718 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 719 | |
6712ecf8 | 720 | static void super_written(struct bio *bio, int error) |
7bfa19f2 | 721 | { |
3cb03002 | 722 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 723 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 724 | |
3a0f5bbb N |
725 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
726 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
727 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
728 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 729 | md_error(mddev, rdev); |
3a0f5bbb | 730 | } |
7bfa19f2 | 731 | |
a9701a30 N |
732 | if (atomic_dec_and_test(&mddev->pending_writes)) |
733 | wake_up(&mddev->sb_wait); | |
f8b58edf | 734 | bio_put(bio); |
7bfa19f2 N |
735 | } |
736 | ||
fd01b88c | 737 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
738 | sector_t sector, int size, struct page *page) |
739 | { | |
740 | /* write first size bytes of page to sector of rdev | |
741 | * Increment mddev->pending_writes before returning | |
742 | * and decrement it on completion, waking up sb_wait | |
743 | * if zero is reached. | |
744 | * If an error occurred, call md_error | |
745 | */ | |
a167f663 | 746 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 747 | |
a6ff7e08 | 748 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
4f024f37 | 749 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
750 | bio_add_page(bio, page, size, 0); |
751 | bio->bi_private = rdev; | |
752 | bio->bi_end_io = super_written; | |
a9701a30 | 753 | |
7bfa19f2 | 754 | atomic_inc(&mddev->pending_writes); |
a5bf4df0 | 755 | submit_bio(WRITE_FLUSH_FUA, bio); |
a9701a30 N |
756 | } |
757 | ||
fd01b88c | 758 | void md_super_wait(struct mddev *mddev) |
a9701a30 | 759 | { |
e9c7469b | 760 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 761 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
7bfa19f2 N |
762 | } |
763 | ||
3cb03002 | 764 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
ccebd4c4 | 765 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 766 | { |
a167f663 | 767 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
768 | int ret; |
769 | ||
a6ff7e08 JB |
770 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
771 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 | 772 | if (metadata_op) |
4f024f37 | 773 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
774 | else if (rdev->mddev->reshape_position != MaxSector && |
775 | (rdev->mddev->reshape_backwards == | |
776 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 777 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 778 | else |
4f024f37 | 779 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 780 | bio_add_page(bio, page, size, 0); |
c170bbb4 | 781 | submit_bio_wait(rw, bio); |
1da177e4 LT |
782 | |
783 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
784 | bio_put(bio); | |
785 | return ret; | |
786 | } | |
a8745db2 | 787 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 788 | |
f72ffdd6 | 789 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
790 | { |
791 | char b[BDEVNAME_SIZE]; | |
403df478 | 792 | |
1da177e4 LT |
793 | if (rdev->sb_loaded) |
794 | return 0; | |
795 | ||
ccebd4c4 | 796 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
797 | goto fail; |
798 | rdev->sb_loaded = 1; | |
799 | return 0; | |
800 | ||
801 | fail: | |
802 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
803 | bdevname(rdev->bdev,b)); | |
804 | return -EINVAL; | |
805 | } | |
806 | ||
807 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
808 | { | |
f72ffdd6 | 809 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
810 | sb1->set_uuid1 == sb2->set_uuid1 && |
811 | sb1->set_uuid2 == sb2->set_uuid2 && | |
812 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
813 | } |
814 | ||
1da177e4 LT |
815 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
816 | { | |
817 | int ret; | |
818 | mdp_super_t *tmp1, *tmp2; | |
819 | ||
820 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
821 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
822 | ||
823 | if (!tmp1 || !tmp2) { | |
824 | ret = 0; | |
35020f1a | 825 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
826 | goto abort; |
827 | } | |
828 | ||
829 | *tmp1 = *sb1; | |
830 | *tmp2 = *sb2; | |
831 | ||
832 | /* | |
833 | * nr_disks is not constant | |
834 | */ | |
835 | tmp1->nr_disks = 0; | |
836 | tmp2->nr_disks = 0; | |
837 | ||
ce0c8e05 | 838 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 839 | abort: |
990a8baf JJ |
840 | kfree(tmp1); |
841 | kfree(tmp2); | |
1da177e4 LT |
842 | return ret; |
843 | } | |
844 | ||
4d167f09 N |
845 | static u32 md_csum_fold(u32 csum) |
846 | { | |
847 | csum = (csum & 0xffff) + (csum >> 16); | |
848 | return (csum & 0xffff) + (csum >> 16); | |
849 | } | |
850 | ||
f72ffdd6 | 851 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 852 | { |
4d167f09 N |
853 | u64 newcsum = 0; |
854 | u32 *sb32 = (u32*)sb; | |
855 | int i; | |
1da177e4 LT |
856 | unsigned int disk_csum, csum; |
857 | ||
858 | disk_csum = sb->sb_csum; | |
859 | sb->sb_csum = 0; | |
4d167f09 N |
860 | |
861 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
862 | newcsum += sb32[i]; | |
863 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
864 | ||
4d167f09 N |
865 | #ifdef CONFIG_ALPHA |
866 | /* This used to use csum_partial, which was wrong for several | |
867 | * reasons including that different results are returned on | |
868 | * different architectures. It isn't critical that we get exactly | |
869 | * the same return value as before (we always csum_fold before | |
870 | * testing, and that removes any differences). However as we | |
871 | * know that csum_partial always returned a 16bit value on | |
872 | * alphas, do a fold to maximise conformity to previous behaviour. | |
873 | */ | |
874 | sb->sb_csum = md_csum_fold(disk_csum); | |
875 | #else | |
1da177e4 | 876 | sb->sb_csum = disk_csum; |
4d167f09 | 877 | #endif |
1da177e4 LT |
878 | return csum; |
879 | } | |
880 | ||
1da177e4 LT |
881 | /* |
882 | * Handle superblock details. | |
883 | * We want to be able to handle multiple superblock formats | |
884 | * so we have a common interface to them all, and an array of | |
885 | * different handlers. | |
886 | * We rely on user-space to write the initial superblock, and support | |
887 | * reading and updating of superblocks. | |
888 | * Interface methods are: | |
3cb03002 | 889 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
890 | * loads and validates a superblock on dev. |
891 | * if refdev != NULL, compare superblocks on both devices | |
892 | * Return: | |
893 | * 0 - dev has a superblock that is compatible with refdev | |
894 | * 1 - dev has a superblock that is compatible and newer than refdev | |
895 | * so dev should be used as the refdev in future | |
896 | * -EINVAL superblock incompatible or invalid | |
897 | * -othererror e.g. -EIO | |
898 | * | |
fd01b88c | 899 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
900 | * Verify that dev is acceptable into mddev. |
901 | * The first time, mddev->raid_disks will be 0, and data from | |
902 | * dev should be merged in. Subsequent calls check that dev | |
903 | * is new enough. Return 0 or -EINVAL | |
904 | * | |
fd01b88c | 905 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
906 | * Update the superblock for rdev with data in mddev |
907 | * This does not write to disc. | |
908 | * | |
909 | */ | |
910 | ||
911 | struct super_type { | |
0cd17fec CW |
912 | char *name; |
913 | struct module *owner; | |
c6563a8c N |
914 | int (*load_super)(struct md_rdev *rdev, |
915 | struct md_rdev *refdev, | |
0cd17fec | 916 | int minor_version); |
c6563a8c N |
917 | int (*validate_super)(struct mddev *mddev, |
918 | struct md_rdev *rdev); | |
919 | void (*sync_super)(struct mddev *mddev, | |
920 | struct md_rdev *rdev); | |
3cb03002 | 921 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 922 | sector_t num_sectors); |
c6563a8c N |
923 | int (*allow_new_offset)(struct md_rdev *rdev, |
924 | unsigned long long new_offset); | |
1da177e4 LT |
925 | }; |
926 | ||
0894cc30 AN |
927 | /* |
928 | * Check that the given mddev has no bitmap. | |
929 | * | |
930 | * This function is called from the run method of all personalities that do not | |
931 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
932 | * has a bitmap. Otherwise, it returns 0. | |
933 | * | |
934 | */ | |
fd01b88c | 935 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 936 | { |
c3d9714e | 937 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
938 | return 0; |
939 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
940 | mdname(mddev), mddev->pers->name); | |
941 | return 1; | |
942 | } | |
943 | EXPORT_SYMBOL(md_check_no_bitmap); | |
944 | ||
1da177e4 | 945 | /* |
f72ffdd6 | 946 | * load_super for 0.90.0 |
1da177e4 | 947 | */ |
3cb03002 | 948 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
949 | { |
950 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
951 | mdp_super_t *sb; | |
952 | int ret; | |
1da177e4 LT |
953 | |
954 | /* | |
0f420358 | 955 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
956 | * it's at the end of the disk. |
957 | * | |
958 | * It also happens to be a multiple of 4Kb. | |
959 | */ | |
57b2caa3 | 960 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 961 | |
0002b271 | 962 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
963 | if (ret) return ret; |
964 | ||
965 | ret = -EINVAL; | |
966 | ||
967 | bdevname(rdev->bdev, b); | |
65a06f06 | 968 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
969 | |
970 | if (sb->md_magic != MD_SB_MAGIC) { | |
971 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
972 | b); | |
973 | goto abort; | |
974 | } | |
975 | ||
976 | if (sb->major_version != 0 || | |
f6705578 N |
977 | sb->minor_version < 90 || |
978 | sb->minor_version > 91) { | |
1da177e4 LT |
979 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
980 | sb->major_version, sb->minor_version, | |
981 | b); | |
982 | goto abort; | |
983 | } | |
984 | ||
985 | if (sb->raid_disks <= 0) | |
986 | goto abort; | |
987 | ||
4d167f09 | 988 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
989 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
990 | b); | |
991 | goto abort; | |
992 | } | |
993 | ||
994 | rdev->preferred_minor = sb->md_minor; | |
995 | rdev->data_offset = 0; | |
c6563a8c | 996 | rdev->new_data_offset = 0; |
0002b271 | 997 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 998 | rdev->badblocks.shift = -1; |
1da177e4 LT |
999 | |
1000 | if (sb->level == LEVEL_MULTIPATH) | |
1001 | rdev->desc_nr = -1; | |
1002 | else | |
1003 | rdev->desc_nr = sb->this_disk.number; | |
1004 | ||
9a7b2b0f | 1005 | if (!refdev) { |
1da177e4 | 1006 | ret = 1; |
9a7b2b0f | 1007 | } else { |
1da177e4 | 1008 | __u64 ev1, ev2; |
65a06f06 | 1009 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1010 | if (!uuid_equal(refsb, sb)) { |
1011 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1012 | b, bdevname(refdev->bdev,b2)); | |
1013 | goto abort; | |
1014 | } | |
1015 | if (!sb_equal(refsb, sb)) { | |
1016 | printk(KERN_WARNING "md: %s has same UUID" | |
1017 | " but different superblock to %s\n", | |
1018 | b, bdevname(refdev->bdev, b2)); | |
1019 | goto abort; | |
1020 | } | |
1021 | ev1 = md_event(sb); | |
1022 | ev2 = md_event(refsb); | |
1023 | if (ev1 > ev2) | |
1024 | ret = 1; | |
f72ffdd6 | 1025 | else |
1da177e4 LT |
1026 | ret = 0; |
1027 | } | |
8190e754 | 1028 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1029 | /* Limit to 4TB as metadata cannot record more than that. |
1030 | * (not needed for Linear and RAID0 as metadata doesn't | |
1031 | * record this size) | |
1032 | */ | |
1033 | if (rdev->sectors >= (2ULL << 32) && sb->level >= 1) | |
27a7b260 | 1034 | rdev->sectors = (2ULL << 32) - 2; |
1da177e4 | 1035 | |
27a7b260 | 1036 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1037 | /* "this cannot possibly happen" ... */ |
1038 | ret = -EINVAL; | |
1039 | ||
1da177e4 LT |
1040 | abort: |
1041 | return ret; | |
1042 | } | |
1043 | ||
1044 | /* | |
1045 | * validate_super for 0.90.0 | |
1046 | */ | |
fd01b88c | 1047 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1048 | { |
1049 | mdp_disk_t *desc; | |
65a06f06 | 1050 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1051 | __u64 ev1 = md_event(sb); |
1da177e4 | 1052 | |
41158c7e | 1053 | rdev->raid_disk = -1; |
c5d79adb N |
1054 | clear_bit(Faulty, &rdev->flags); |
1055 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1056 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1057 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1058 | |
1da177e4 LT |
1059 | if (mddev->raid_disks == 0) { |
1060 | mddev->major_version = 0; | |
1061 | mddev->minor_version = sb->minor_version; | |
1062 | mddev->patch_version = sb->patch_version; | |
e691063a | 1063 | mddev->external = 0; |
9d8f0363 | 1064 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1065 | mddev->ctime = sb->ctime; |
1066 | mddev->utime = sb->utime; | |
1067 | mddev->level = sb->level; | |
d9d166c2 | 1068 | mddev->clevel[0] = 0; |
1da177e4 LT |
1069 | mddev->layout = sb->layout; |
1070 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1071 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1072 | mddev->events = ev1; |
c3d9714e | 1073 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1074 | mddev->bitmap_info.space = 0; |
1075 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1076 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1077 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1078 | mddev->reshape_backwards = 0; |
1da177e4 | 1079 | |
f6705578 N |
1080 | if (mddev->minor_version >= 91) { |
1081 | mddev->reshape_position = sb->reshape_position; | |
1082 | mddev->delta_disks = sb->delta_disks; | |
1083 | mddev->new_level = sb->new_level; | |
1084 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1085 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1086 | if (mddev->delta_disks < 0) |
1087 | mddev->reshape_backwards = 1; | |
f6705578 N |
1088 | } else { |
1089 | mddev->reshape_position = MaxSector; | |
1090 | mddev->delta_disks = 0; | |
1091 | mddev->new_level = mddev->level; | |
1092 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1093 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1094 | } |
1095 | ||
1da177e4 LT |
1096 | if (sb->state & (1<<MD_SB_CLEAN)) |
1097 | mddev->recovery_cp = MaxSector; | |
1098 | else { | |
f72ffdd6 | 1099 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1100 | sb->events_lo == sb->cp_events_lo) { |
1101 | mddev->recovery_cp = sb->recovery_cp; | |
1102 | } else | |
1103 | mddev->recovery_cp = 0; | |
1104 | } | |
1105 | ||
1106 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1107 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1108 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1109 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1110 | ||
1111 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1112 | |
1113 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1114 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1115 | mddev->bitmap_info.offset = |
1116 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1117 | mddev->bitmap_info.space = |
c9ad020f | 1118 | mddev->bitmap_info.default_space; |
6409bb05 | 1119 | } |
a654b9d8 | 1120 | |
41158c7e | 1121 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1122 | /* Insist on good event counter while assembling, except |
1123 | * for spares (which don't need an event count) */ | |
1da177e4 | 1124 | ++ev1; |
be6800a7 N |
1125 | if (sb->disks[rdev->desc_nr].state & ( |
1126 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1127 | if (ev1 < mddev->events) |
be6800a7 | 1128 | return -EINVAL; |
41158c7e N |
1129 | } else if (mddev->bitmap) { |
1130 | /* if adding to array with a bitmap, then we can accept an | |
1131 | * older device ... but not too old. | |
1132 | */ | |
41158c7e N |
1133 | if (ev1 < mddev->bitmap->events_cleared) |
1134 | return 0; | |
8313b8e5 N |
1135 | if (ev1 < mddev->events) |
1136 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1137 | } else { |
1138 | if (ev1 < mddev->events) | |
1139 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1140 | return 0; | |
1141 | } | |
41158c7e | 1142 | |
1da177e4 | 1143 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1144 | desc = sb->disks + rdev->desc_nr; |
1145 | ||
1146 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1147 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1148 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1149 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1150 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1151 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1152 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1153 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1154 | /* active but not in sync implies recovery up to | |
1155 | * reshape position. We don't know exactly where | |
1156 | * that is, so set to zero for now */ | |
1157 | if (mddev->minor_version >= 91) { | |
1158 | rdev->recovery_offset = 0; | |
1159 | rdev->raid_disk = desc->raid_disk; | |
1160 | } | |
1da177e4 | 1161 | } |
8ddf9efe N |
1162 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1163 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1164 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1165 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1166 | return 0; |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * sync_super for 0.90.0 | |
1171 | */ | |
fd01b88c | 1172 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1173 | { |
1174 | mdp_super_t *sb; | |
3cb03002 | 1175 | struct md_rdev *rdev2; |
1da177e4 | 1176 | int next_spare = mddev->raid_disks; |
19133a42 | 1177 | |
1da177e4 LT |
1178 | /* make rdev->sb match mddev data.. |
1179 | * | |
1180 | * 1/ zero out disks | |
1181 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1182 | * 3/ any empty disks < next_spare become removed | |
1183 | * | |
1184 | * disks[0] gets initialised to REMOVED because | |
1185 | * we cannot be sure from other fields if it has | |
1186 | * been initialised or not. | |
1187 | */ | |
1188 | int i; | |
1189 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1190 | ||
61181565 N |
1191 | rdev->sb_size = MD_SB_BYTES; |
1192 | ||
65a06f06 | 1193 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1194 | |
1195 | memset(sb, 0, sizeof(*sb)); | |
1196 | ||
1197 | sb->md_magic = MD_SB_MAGIC; | |
1198 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1199 | sb->patch_version = mddev->patch_version; |
1200 | sb->gvalid_words = 0; /* ignored */ | |
1201 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1202 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1203 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1204 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1205 | ||
1206 | sb->ctime = mddev->ctime; | |
1207 | sb->level = mddev->level; | |
58c0fed4 | 1208 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1209 | sb->raid_disks = mddev->raid_disks; |
1210 | sb->md_minor = mddev->md_minor; | |
e691063a | 1211 | sb->not_persistent = 0; |
1da177e4 LT |
1212 | sb->utime = mddev->utime; |
1213 | sb->state = 0; | |
1214 | sb->events_hi = (mddev->events>>32); | |
1215 | sb->events_lo = (u32)mddev->events; | |
1216 | ||
f6705578 N |
1217 | if (mddev->reshape_position == MaxSector) |
1218 | sb->minor_version = 90; | |
1219 | else { | |
1220 | sb->minor_version = 91; | |
1221 | sb->reshape_position = mddev->reshape_position; | |
1222 | sb->new_level = mddev->new_level; | |
1223 | sb->delta_disks = mddev->delta_disks; | |
1224 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1225 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1226 | } |
1227 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1228 | if (mddev->in_sync) |
1229 | { | |
1230 | sb->recovery_cp = mddev->recovery_cp; | |
1231 | sb->cp_events_hi = (mddev->events>>32); | |
1232 | sb->cp_events_lo = (u32)mddev->events; | |
1233 | if (mddev->recovery_cp == MaxSector) | |
1234 | sb->state = (1<< MD_SB_CLEAN); | |
1235 | } else | |
1236 | sb->recovery_cp = 0; | |
1237 | ||
1238 | sb->layout = mddev->layout; | |
9d8f0363 | 1239 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1240 | |
c3d9714e | 1241 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1242 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1243 | ||
1da177e4 | 1244 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1245 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1246 | mdp_disk_t *d; |
86e6ffdd | 1247 | int desc_nr; |
0261cd9f N |
1248 | int is_active = test_bit(In_sync, &rdev2->flags); |
1249 | ||
1250 | if (rdev2->raid_disk >= 0 && | |
1251 | sb->minor_version >= 91) | |
1252 | /* we have nowhere to store the recovery_offset, | |
1253 | * but if it is not below the reshape_position, | |
1254 | * we can piggy-back on that. | |
1255 | */ | |
1256 | is_active = 1; | |
1257 | if (rdev2->raid_disk < 0 || | |
1258 | test_bit(Faulty, &rdev2->flags)) | |
1259 | is_active = 0; | |
1260 | if (is_active) | |
86e6ffdd | 1261 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1262 | else |
86e6ffdd | 1263 | desc_nr = next_spare++; |
19133a42 | 1264 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1265 | d = &sb->disks[rdev2->desc_nr]; |
1266 | nr_disks++; | |
1267 | d->number = rdev2->desc_nr; | |
1268 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1269 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1270 | if (is_active) |
1da177e4 LT |
1271 | d->raid_disk = rdev2->raid_disk; |
1272 | else | |
1273 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1274 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1275 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1276 | else if (is_active) { |
1da177e4 | 1277 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1278 | if (test_bit(In_sync, &rdev2->flags)) |
1279 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1280 | active++; |
1281 | working++; | |
1282 | } else { | |
1283 | d->state = 0; | |
1284 | spare++; | |
1285 | working++; | |
1286 | } | |
8ddf9efe N |
1287 | if (test_bit(WriteMostly, &rdev2->flags)) |
1288 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1289 | } |
1da177e4 LT |
1290 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1291 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1292 | mdp_disk_t *d = &sb->disks[i]; | |
1293 | if (d->state == 0 && d->number == 0) { | |
1294 | d->number = i; | |
1295 | d->raid_disk = i; | |
1296 | d->state = (1<<MD_DISK_REMOVED); | |
1297 | d->state |= (1<<MD_DISK_FAULTY); | |
1298 | failed++; | |
1299 | } | |
1300 | } | |
1301 | sb->nr_disks = nr_disks; | |
1302 | sb->active_disks = active; | |
1303 | sb->working_disks = working; | |
1304 | sb->failed_disks = failed; | |
1305 | sb->spare_disks = spare; | |
1306 | ||
1307 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1308 | sb->sb_csum = calc_sb_csum(sb); | |
1309 | } | |
1310 | ||
0cd17fec CW |
1311 | /* |
1312 | * rdev_size_change for 0.90.0 | |
1313 | */ | |
1314 | static unsigned long long | |
3cb03002 | 1315 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1316 | { |
58c0fed4 | 1317 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1318 | return 0; /* component must fit device */ |
c3d9714e | 1319 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1320 | return 0; /* can't move bitmap */ |
57b2caa3 | 1321 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1322 | if (!num_sectors || num_sectors > rdev->sb_start) |
1323 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1324 | /* Limit to 4TB as metadata cannot record more than that. |
1325 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1326 | */ | |
667a5313 | 1327 | if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1) |
27a7b260 | 1328 | num_sectors = (2ULL << 32) - 2; |
0f420358 | 1329 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1330 | rdev->sb_page); |
1331 | md_super_wait(rdev->mddev); | |
c26a44ed | 1332 | return num_sectors; |
0cd17fec CW |
1333 | } |
1334 | ||
c6563a8c N |
1335 | static int |
1336 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1337 | { | |
1338 | /* non-zero offset changes not possible with v0.90 */ | |
1339 | return new_offset == 0; | |
1340 | } | |
0cd17fec | 1341 | |
1da177e4 LT |
1342 | /* |
1343 | * version 1 superblock | |
1344 | */ | |
1345 | ||
f72ffdd6 | 1346 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1347 | { |
1c05b4bc N |
1348 | __le32 disk_csum; |
1349 | u32 csum; | |
1da177e4 LT |
1350 | unsigned long long newcsum; |
1351 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1352 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1353 | |
1354 | disk_csum = sb->sb_csum; | |
1355 | sb->sb_csum = 0; | |
1356 | newcsum = 0; | |
1f3c9907 | 1357 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1358 | newcsum += le32_to_cpu(*isuper++); |
1359 | ||
1360 | if (size == 2) | |
1c05b4bc | 1361 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1362 | |
1363 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1364 | sb->sb_csum = disk_csum; | |
1365 | return cpu_to_le32(csum); | |
1366 | } | |
1367 | ||
2699b672 N |
1368 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, |
1369 | int acknowledged); | |
3cb03002 | 1370 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1371 | { |
1372 | struct mdp_superblock_1 *sb; | |
1373 | int ret; | |
0f420358 | 1374 | sector_t sb_start; |
c6563a8c | 1375 | sector_t sectors; |
1da177e4 | 1376 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1377 | int bmask; |
1da177e4 LT |
1378 | |
1379 | /* | |
0f420358 | 1380 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1381 | * It is always aligned to a 4K boundary and |
1382 | * depeding on minor_version, it can be: | |
1383 | * 0: At least 8K, but less than 12K, from end of device | |
1384 | * 1: At start of device | |
1385 | * 2: 4K from start of device. | |
1386 | */ | |
1387 | switch(minor_version) { | |
1388 | case 0: | |
77304d2a | 1389 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1390 | sb_start -= 8*2; |
1391 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1392 | break; |
1393 | case 1: | |
0f420358 | 1394 | sb_start = 0; |
1da177e4 LT |
1395 | break; |
1396 | case 2: | |
0f420358 | 1397 | sb_start = 8; |
1da177e4 LT |
1398 | break; |
1399 | default: | |
1400 | return -EINVAL; | |
1401 | } | |
0f420358 | 1402 | rdev->sb_start = sb_start; |
1da177e4 | 1403 | |
0002b271 N |
1404 | /* superblock is rarely larger than 1K, but it can be larger, |
1405 | * and it is safe to read 4k, so we do that | |
1406 | */ | |
1407 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1408 | if (ret) return ret; |
1409 | ||
65a06f06 | 1410 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1411 | |
1412 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1413 | sb->major_version != cpu_to_le32(1) || | |
1414 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1415 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1416 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1417 | return -EINVAL; |
1418 | ||
1419 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1420 | printk("md: invalid superblock checksum on %s\n", | |
1421 | bdevname(rdev->bdev,b)); | |
1422 | return -EINVAL; | |
1423 | } | |
1424 | if (le64_to_cpu(sb->data_size) < 10) { | |
1425 | printk("md: data_size too small on %s\n", | |
1426 | bdevname(rdev->bdev,b)); | |
1427 | return -EINVAL; | |
1428 | } | |
c6563a8c N |
1429 | if (sb->pad0 || |
1430 | sb->pad3[0] || | |
1431 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1432 | /* Some padding is non-zero, might be a new feature */ | |
1433 | return -EINVAL; | |
e11e93fa | 1434 | |
1da177e4 LT |
1435 | rdev->preferred_minor = 0xffff; |
1436 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1437 | rdev->new_data_offset = rdev->data_offset; |
1438 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1439 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1440 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1441 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1442 | |
0002b271 | 1443 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1444 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1445 | if (rdev->sb_size & bmask) |
a1801f85 N |
1446 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1447 | ||
1448 | if (minor_version | |
0f420358 | 1449 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1450 | return -EINVAL; |
c6563a8c N |
1451 | if (minor_version |
1452 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1453 | return -EINVAL; | |
0002b271 | 1454 | |
31b65a0d N |
1455 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1456 | rdev->desc_nr = -1; | |
1457 | else | |
1458 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1459 | ||
2699b672 N |
1460 | if (!rdev->bb_page) { |
1461 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1462 | if (!rdev->bb_page) | |
1463 | return -ENOMEM; | |
1464 | } | |
1465 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1466 | rdev->badblocks.count == 0) { | |
1467 | /* need to load the bad block list. | |
1468 | * Currently we limit it to one page. | |
1469 | */ | |
1470 | s32 offset; | |
1471 | sector_t bb_sector; | |
1472 | u64 *bbp; | |
1473 | int i; | |
1474 | int sectors = le16_to_cpu(sb->bblog_size); | |
1475 | if (sectors > (PAGE_SIZE / 512)) | |
1476 | return -EINVAL; | |
1477 | offset = le32_to_cpu(sb->bblog_offset); | |
1478 | if (offset == 0) | |
1479 | return -EINVAL; | |
1480 | bb_sector = (long long)offset; | |
1481 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
1482 | rdev->bb_page, READ, true)) | |
1483 | return -EIO; | |
1484 | bbp = (u64 *)page_address(rdev->bb_page); | |
1485 | rdev->badblocks.shift = sb->bblog_shift; | |
1486 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1487 | u64 bb = le64_to_cpu(*bbp); | |
1488 | int count = bb & (0x3ff); | |
1489 | u64 sector = bb >> 10; | |
1490 | sector <<= sb->bblog_shift; | |
1491 | count <<= sb->bblog_shift; | |
1492 | if (bb + 1 == 0) | |
1493 | break; | |
1494 | if (md_set_badblocks(&rdev->badblocks, | |
1495 | sector, count, 1) == 0) | |
1496 | return -EINVAL; | |
1497 | } | |
486adf72 N |
1498 | } else if (sb->bblog_offset != 0) |
1499 | rdev->badblocks.shift = 0; | |
2699b672 | 1500 | |
9a7b2b0f | 1501 | if (!refdev) { |
8ed75463 | 1502 | ret = 1; |
9a7b2b0f | 1503 | } else { |
1da177e4 | 1504 | __u64 ev1, ev2; |
65a06f06 | 1505 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1506 | |
1507 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1508 | sb->level != refsb->level || | |
1509 | sb->layout != refsb->layout || | |
1510 | sb->chunksize != refsb->chunksize) { | |
1511 | printk(KERN_WARNING "md: %s has strangely different" | |
1512 | " superblock to %s\n", | |
1513 | bdevname(rdev->bdev,b), | |
1514 | bdevname(refdev->bdev,b2)); | |
1515 | return -EINVAL; | |
1516 | } | |
1517 | ev1 = le64_to_cpu(sb->events); | |
1518 | ev2 = le64_to_cpu(refsb->events); | |
1519 | ||
1520 | if (ev1 > ev2) | |
8ed75463 N |
1521 | ret = 1; |
1522 | else | |
1523 | ret = 0; | |
1da177e4 | 1524 | } |
c6563a8c N |
1525 | if (minor_version) { |
1526 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1527 | sectors -= rdev->data_offset; | |
1528 | } else | |
1529 | sectors = rdev->sb_start; | |
1530 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1531 | return -EINVAL; |
dd8ac336 | 1532 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1533 | return ret; |
1da177e4 LT |
1534 | } |
1535 | ||
fd01b88c | 1536 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1537 | { |
65a06f06 | 1538 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1539 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1540 | |
41158c7e | 1541 | rdev->raid_disk = -1; |
c5d79adb N |
1542 | clear_bit(Faulty, &rdev->flags); |
1543 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1544 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1545 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1546 | |
1da177e4 LT |
1547 | if (mddev->raid_disks == 0) { |
1548 | mddev->major_version = 1; | |
1549 | mddev->patch_version = 0; | |
e691063a | 1550 | mddev->external = 0; |
9d8f0363 | 1551 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1552 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1553 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1554 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1555 | mddev->clevel[0] = 0; |
1da177e4 LT |
1556 | mddev->layout = le32_to_cpu(sb->layout); |
1557 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1558 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1559 | mddev->events = ev1; |
c3d9714e | 1560 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1561 | mddev->bitmap_info.space = 0; |
1562 | /* Default location for bitmap is 1K after superblock | |
1563 | * using 3K - total of 4K | |
1564 | */ | |
c3d9714e | 1565 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1566 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1567 | mddev->reshape_backwards = 0; |
1568 | ||
1da177e4 LT |
1569 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1570 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1571 | ||
1572 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1573 | |
71c0805c | 1574 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1575 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1576 | mddev->bitmap_info.offset = |
1577 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1578 | /* Metadata doesn't record how much space is available. |
1579 | * For 1.0, we assume we can use up to the superblock | |
1580 | * if before, else to 4K beyond superblock. | |
1581 | * For others, assume no change is possible. | |
1582 | */ | |
1583 | if (mddev->minor_version > 0) | |
1584 | mddev->bitmap_info.space = 0; | |
1585 | else if (mddev->bitmap_info.offset > 0) | |
1586 | mddev->bitmap_info.space = | |
1587 | 8 - mddev->bitmap_info.offset; | |
1588 | else | |
1589 | mddev->bitmap_info.space = | |
1590 | -mddev->bitmap_info.offset; | |
1591 | } | |
e11e93fa | 1592 | |
f6705578 N |
1593 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1594 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1595 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1596 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1597 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1598 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1599 | if (mddev->delta_disks < 0 || |
1600 | (mddev->delta_disks == 0 && | |
1601 | (le32_to_cpu(sb->feature_map) | |
1602 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1603 | mddev->reshape_backwards = 1; | |
f6705578 N |
1604 | } else { |
1605 | mddev->reshape_position = MaxSector; | |
1606 | mddev->delta_disks = 0; | |
1607 | mddev->new_level = mddev->level; | |
1608 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1609 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1610 | } |
1611 | ||
41158c7e | 1612 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1613 | /* Insist of good event counter while assembling, except for |
1614 | * spares (which don't need an event count) */ | |
1da177e4 | 1615 | ++ev1; |
be6800a7 N |
1616 | if (rdev->desc_nr >= 0 && |
1617 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1618 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1619 | if (ev1 < mddev->events) | |
1620 | return -EINVAL; | |
41158c7e N |
1621 | } else if (mddev->bitmap) { |
1622 | /* If adding to array with a bitmap, then we can accept an | |
1623 | * older device, but not too old. | |
1624 | */ | |
41158c7e N |
1625 | if (ev1 < mddev->bitmap->events_cleared) |
1626 | return 0; | |
8313b8e5 N |
1627 | if (ev1 < mddev->events) |
1628 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1629 | } else { |
1630 | if (ev1 < mddev->events) | |
1631 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1632 | return 0; | |
1633 | } | |
1da177e4 LT |
1634 | if (mddev->level != LEVEL_MULTIPATH) { |
1635 | int role; | |
3673f305 N |
1636 | if (rdev->desc_nr < 0 || |
1637 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1638 | role = 0xffff; | |
1639 | rdev->desc_nr = -1; | |
1640 | } else | |
1641 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1642 | switch(role) { |
1643 | case 0xffff: /* spare */ | |
1da177e4 LT |
1644 | break; |
1645 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1646 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1647 | break; |
1648 | default: | |
f466722c | 1649 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1650 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1651 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1652 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1653 | if (!(le32_to_cpu(sb->feature_map) & |
1654 | MD_FEATURE_RECOVERY_BITMAP)) | |
1655 | rdev->saved_raid_disk = -1; | |
1656 | } else | |
5fd6c1dc | 1657 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1658 | rdev->raid_disk = role; |
1659 | break; | |
1660 | } | |
8ddf9efe N |
1661 | if (sb->devflags & WriteMostly1) |
1662 | set_bit(WriteMostly, &rdev->flags); | |
2d78f8c4 N |
1663 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1664 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1665 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1666 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1667 | |
1da177e4 LT |
1668 | return 0; |
1669 | } | |
1670 | ||
fd01b88c | 1671 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1672 | { |
1673 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1674 | struct md_rdev *rdev2; |
1da177e4 LT |
1675 | int max_dev, i; |
1676 | /* make rdev->sb match mddev and rdev data. */ | |
1677 | ||
65a06f06 | 1678 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1679 | |
1680 | sb->feature_map = 0; | |
1681 | sb->pad0 = 0; | |
5fd6c1dc | 1682 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1683 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1684 | ||
1685 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1686 | sb->events = cpu_to_le64(mddev->events); | |
1687 | if (mddev->in_sync) | |
1688 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1689 | else | |
1690 | sb->resync_offset = cpu_to_le64(0); | |
1691 | ||
1c05b4bc | 1692 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1693 | |
f0ca340c | 1694 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1695 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1696 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1697 | sb->level = cpu_to_le32(mddev->level); |
1698 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1699 | |
aeb9b211 N |
1700 | if (test_bit(WriteMostly, &rdev->flags)) |
1701 | sb->devflags |= WriteMostly1; | |
1702 | else | |
1703 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1704 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1705 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1706 | |
c3d9714e N |
1707 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1708 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1709 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1710 | } |
5fd6c1dc N |
1711 | |
1712 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1713 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1714 | sb->feature_map |= |
1715 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1716 | sb->recovery_offset = | |
1717 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1718 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1719 | sb->feature_map |= | |
1720 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1721 | } |
2d78f8c4 N |
1722 | if (test_bit(Replacement, &rdev->flags)) |
1723 | sb->feature_map |= | |
1724 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1725 | |
f6705578 N |
1726 | if (mddev->reshape_position != MaxSector) { |
1727 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1728 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1729 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1730 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1731 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1732 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1733 | if (mddev->delta_disks == 0 && |
1734 | mddev->reshape_backwards) | |
1735 | sb->feature_map | |
1736 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1737 | if (rdev->new_data_offset != rdev->data_offset) { |
1738 | sb->feature_map | |
1739 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1740 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1741 | - rdev->data_offset)); | |
1742 | } | |
f6705578 | 1743 | } |
a654b9d8 | 1744 | |
2699b672 N |
1745 | if (rdev->badblocks.count == 0) |
1746 | /* Nothing to do for bad blocks*/ ; | |
1747 | else if (sb->bblog_offset == 0) | |
1748 | /* Cannot record bad blocks on this device */ | |
1749 | md_error(mddev, rdev); | |
1750 | else { | |
1751 | struct badblocks *bb = &rdev->badblocks; | |
1752 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1753 | u64 *p = bb->page; | |
1754 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1755 | if (bb->changed) { | |
1756 | unsigned seq; | |
1757 | ||
1758 | retry: | |
1759 | seq = read_seqbegin(&bb->lock); | |
1760 | ||
1761 | memset(bbp, 0xff, PAGE_SIZE); | |
1762 | ||
1763 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1764 | u64 internal_bb = p[i]; |
2699b672 N |
1765 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1766 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1767 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1768 | } |
d0962936 | 1769 | bb->changed = 0; |
2699b672 N |
1770 | if (read_seqretry(&bb->lock, seq)) |
1771 | goto retry; | |
1772 | ||
1773 | bb->sector = (rdev->sb_start + | |
1774 | (int)le32_to_cpu(sb->bblog_offset)); | |
1775 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1776 | } |
1777 | } | |
1778 | ||
1da177e4 | 1779 | max_dev = 0; |
dafb20fa | 1780 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1781 | if (rdev2->desc_nr+1 > max_dev) |
1782 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1783 | |
70471daf N |
1784 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1785 | int bmask; | |
a778b73f | 1786 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1787 | rdev->sb_size = max_dev * 2 + 256; |
1788 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1789 | if (rdev->sb_size & bmask) | |
1790 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1791 | } else |
1792 | max_dev = le32_to_cpu(sb->max_dev); | |
1793 | ||
1da177e4 LT |
1794 | for (i=0; i<max_dev;i++) |
1795 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
f72ffdd6 | 1796 | |
dafb20fa | 1797 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1798 | i = rdev2->desc_nr; |
b2d444d7 | 1799 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1800 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1801 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1802 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1803 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1804 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1805 | else |
1806 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1807 | } | |
1808 | ||
1da177e4 LT |
1809 | sb->sb_csum = calc_sb_1_csum(sb); |
1810 | } | |
1811 | ||
0cd17fec | 1812 | static unsigned long long |
3cb03002 | 1813 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1814 | { |
1815 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1816 | sector_t max_sectors; |
58c0fed4 | 1817 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1818 | return 0; /* component must fit device */ |
c6563a8c N |
1819 | if (rdev->data_offset != rdev->new_data_offset) |
1820 | return 0; /* too confusing */ | |
0f420358 | 1821 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1822 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1823 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1824 | max_sectors -= rdev->data_offset; |
1825 | if (!num_sectors || num_sectors > max_sectors) | |
1826 | num_sectors = max_sectors; | |
c3d9714e | 1827 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1828 | /* minor version 0 with bitmap we can't move */ |
1829 | return 0; | |
1830 | } else { | |
1831 | /* minor version 0; superblock after data */ | |
0f420358 | 1832 | sector_t sb_start; |
77304d2a | 1833 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1834 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1835 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1836 | if (!num_sectors || num_sectors > max_sectors) |
1837 | num_sectors = max_sectors; | |
0f420358 | 1838 | rdev->sb_start = sb_start; |
0cd17fec | 1839 | } |
65a06f06 | 1840 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1841 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1842 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1843 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1844 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1845 | rdev->sb_page); |
1846 | md_super_wait(rdev->mddev); | |
c26a44ed | 1847 | return num_sectors; |
c6563a8c N |
1848 | |
1849 | } | |
1850 | ||
1851 | static int | |
1852 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1853 | unsigned long long new_offset) | |
1854 | { | |
1855 | /* All necessary checks on new >= old have been done */ | |
1856 | struct bitmap *bitmap; | |
1857 | if (new_offset >= rdev->data_offset) | |
1858 | return 1; | |
1859 | ||
1860 | /* with 1.0 metadata, there is no metadata to tread on | |
1861 | * so we can always move back */ | |
1862 | if (rdev->mddev->minor_version == 0) | |
1863 | return 1; | |
1864 | ||
1865 | /* otherwise we must be sure not to step on | |
1866 | * any metadata, so stay: | |
1867 | * 36K beyond start of superblock | |
1868 | * beyond end of badblocks | |
1869 | * beyond write-intent bitmap | |
1870 | */ | |
1871 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1872 | return 0; | |
1873 | bitmap = rdev->mddev->bitmap; | |
1874 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1875 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1876 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1877 | return 0; |
1878 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1879 | return 0; | |
1880 | ||
1881 | return 1; | |
0cd17fec | 1882 | } |
1da177e4 | 1883 | |
75c96f85 | 1884 | static struct super_type super_types[] = { |
1da177e4 LT |
1885 | [0] = { |
1886 | .name = "0.90.0", | |
1887 | .owner = THIS_MODULE, | |
0cd17fec CW |
1888 | .load_super = super_90_load, |
1889 | .validate_super = super_90_validate, | |
1890 | .sync_super = super_90_sync, | |
1891 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1892 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1893 | }, |
1894 | [1] = { | |
1895 | .name = "md-1", | |
1896 | .owner = THIS_MODULE, | |
0cd17fec CW |
1897 | .load_super = super_1_load, |
1898 | .validate_super = super_1_validate, | |
1899 | .sync_super = super_1_sync, | |
1900 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1901 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1902 | }, |
1903 | }; | |
1da177e4 | 1904 | |
fd01b88c | 1905 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1906 | { |
1907 | if (mddev->sync_super) { | |
1908 | mddev->sync_super(mddev, rdev); | |
1909 | return; | |
1910 | } | |
1911 | ||
1912 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1913 | ||
1914 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1915 | } | |
1916 | ||
fd01b88c | 1917 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1918 | { |
3cb03002 | 1919 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1920 | |
4b80991c N |
1921 | rcu_read_lock(); |
1922 | rdev_for_each_rcu(rdev, mddev1) | |
1923 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1924 | if (rdev->bdev->bd_contains == |
4b80991c N |
1925 | rdev2->bdev->bd_contains) { |
1926 | rcu_read_unlock(); | |
7dd5e7c3 | 1927 | return 1; |
4b80991c N |
1928 | } |
1929 | rcu_read_unlock(); | |
1da177e4 LT |
1930 | return 0; |
1931 | } | |
1932 | ||
1933 | static LIST_HEAD(pending_raid_disks); | |
1934 | ||
ac5e7113 AN |
1935 | /* |
1936 | * Try to register data integrity profile for an mddev | |
1937 | * | |
1938 | * This is called when an array is started and after a disk has been kicked | |
1939 | * from the array. It only succeeds if all working and active component devices | |
1940 | * are integrity capable with matching profiles. | |
1941 | */ | |
fd01b88c | 1942 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 1943 | { |
3cb03002 | 1944 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
1945 | |
1946 | if (list_empty(&mddev->disks)) | |
1947 | return 0; /* nothing to do */ | |
629acb6a JB |
1948 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
1949 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 1950 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
1951 | /* skip spares and non-functional disks */ |
1952 | if (test_bit(Faulty, &rdev->flags)) | |
1953 | continue; | |
1954 | if (rdev->raid_disk < 0) | |
1955 | continue; | |
ac5e7113 AN |
1956 | if (!reference) { |
1957 | /* Use the first rdev as the reference */ | |
1958 | reference = rdev; | |
1959 | continue; | |
1960 | } | |
1961 | /* does this rdev's profile match the reference profile? */ | |
1962 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1963 | rdev->bdev->bd_disk) < 0) | |
1964 | return -EINVAL; | |
1965 | } | |
89078d57 MP |
1966 | if (!reference || !bdev_get_integrity(reference->bdev)) |
1967 | return 0; | |
ac5e7113 AN |
1968 | /* |
1969 | * All component devices are integrity capable and have matching | |
1970 | * profiles, register the common profile for the md device. | |
1971 | */ | |
1972 | if (blk_integrity_register(mddev->gendisk, | |
1973 | bdev_get_integrity(reference->bdev)) != 0) { | |
1974 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1975 | mdname(mddev)); | |
1976 | return -EINVAL; | |
1977 | } | |
a91a2785 MP |
1978 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
1979 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
1980 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
1981 | mdname(mddev)); | |
1982 | return -EINVAL; | |
1983 | } | |
ac5e7113 AN |
1984 | return 0; |
1985 | } | |
1986 | EXPORT_SYMBOL(md_integrity_register); | |
1987 | ||
1988 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
fd01b88c | 1989 | void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) |
3f9d99c1 | 1990 | { |
2863b9eb JB |
1991 | struct blk_integrity *bi_rdev; |
1992 | struct blk_integrity *bi_mddev; | |
1993 | ||
1994 | if (!mddev->gendisk) | |
1995 | return; | |
1996 | ||
1997 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
1998 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 1999 | |
ac5e7113 | 2000 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 2001 | return; |
ac5e7113 | 2002 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 2003 | return; |
ac5e7113 AN |
2004 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
2005 | rdev->bdev->bd_disk) >= 0) | |
2006 | return; | |
2007 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
2008 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 2009 | } |
ac5e7113 | 2010 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2011 | |
f72ffdd6 | 2012 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2013 | { |
7dd5e7c3 | 2014 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2015 | struct kobject *ko; |
1edf80d3 | 2016 | char *s; |
5e55e2f5 | 2017 | int err; |
1da177e4 | 2018 | |
11e2ede0 DW |
2019 | /* prevent duplicates */ |
2020 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2021 | return -EEXIST; | |
2022 | ||
dd8ac336 AN |
2023 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
2024 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
2025 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2026 | if (mddev->pers) { |
2027 | /* Cannot change size, so fail | |
2028 | * If mddev->level <= 0, then we don't care | |
2029 | * about aligning sizes (e.g. linear) | |
2030 | */ | |
2031 | if (mddev->level > 0) | |
2032 | return -ENOSPC; | |
2033 | } else | |
dd8ac336 | 2034 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2035 | } |
1da177e4 LT |
2036 | |
2037 | /* Verify rdev->desc_nr is unique. | |
2038 | * If it is -1, assign a free number, else | |
2039 | * check number is not in use | |
2040 | */ | |
4878e9eb | 2041 | rcu_read_lock(); |
1da177e4 LT |
2042 | if (rdev->desc_nr < 0) { |
2043 | int choice = 0; | |
4878e9eb N |
2044 | if (mddev->pers) |
2045 | choice = mddev->raid_disks; | |
2046 | while (find_rdev_nr_rcu(mddev, choice)) | |
1da177e4 LT |
2047 | choice++; |
2048 | rdev->desc_nr = choice; | |
2049 | } else { | |
4878e9eb N |
2050 | if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
2051 | rcu_read_unlock(); | |
1da177e4 | 2052 | return -EBUSY; |
4878e9eb | 2053 | } |
1da177e4 | 2054 | } |
4878e9eb | 2055 | rcu_read_unlock(); |
de01dfad N |
2056 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
2057 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
2058 | mdname(mddev), mddev->max_disks); | |
2059 | return -EBUSY; | |
2060 | } | |
19133a42 | 2061 | bdevname(rdev->bdev,b); |
649316b2 | 2062 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 2063 | *s = '!'; |
649316b2 | 2064 | |
1da177e4 | 2065 | rdev->mddev = mddev; |
19133a42 | 2066 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 2067 | |
b2d6db58 | 2068 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2069 | goto fail; |
86e6ffdd | 2070 | |
0762b8bd | 2071 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2072 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2073 | /* failure here is OK */; | |
2074 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2075 | |
4b80991c | 2076 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2077 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2078 | |
2079 | /* May as well allow recovery to be retried once */ | |
5389042f | 2080 | mddev->recovery_disabled++; |
3f9d99c1 | 2081 | |
1da177e4 | 2082 | return 0; |
5e55e2f5 N |
2083 | |
2084 | fail: | |
2085 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
2086 | b, mdname(mddev)); | |
2087 | return err; | |
1da177e4 LT |
2088 | } |
2089 | ||
177a99b2 | 2090 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2091 | { |
3cb03002 | 2092 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2093 | kobject_del(&rdev->kobj); |
177a99b2 | 2094 | kobject_put(&rdev->kobj); |
5792a285 N |
2095 | } |
2096 | ||
f72ffdd6 | 2097 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2098 | { |
2099 | char b[BDEVNAME_SIZE]; | |
403df478 | 2100 | |
49731baa | 2101 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2102 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
2103 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
2104 | rdev->mddev = NULL; | |
86e6ffdd | 2105 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2106 | sysfs_put(rdev->sysfs_state); |
2107 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2108 | rdev->badblocks.count = 0; |
5792a285 | 2109 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2110 | * writing to 'remove' to "dev/state". We also need |
2111 | * to delay it due to rcu usage. | |
5792a285 | 2112 | */ |
4b80991c | 2113 | synchronize_rcu(); |
177a99b2 N |
2114 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2115 | kobject_get(&rdev->kobj); | |
e804ac78 | 2116 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2117 | } |
2118 | ||
2119 | /* | |
2120 | * prevent the device from being mounted, repartitioned or | |
2121 | * otherwise reused by a RAID array (or any other kernel | |
2122 | * subsystem), by bd_claiming the device. | |
2123 | */ | |
3cb03002 | 2124 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2125 | { |
2126 | int err = 0; | |
2127 | struct block_device *bdev; | |
2128 | char b[BDEVNAME_SIZE]; | |
2129 | ||
d4d77629 | 2130 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2131 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 LT |
2132 | if (IS_ERR(bdev)) { |
2133 | printk(KERN_ERR "md: could not open %s.\n", | |
2134 | __bdevname(dev, b)); | |
2135 | return PTR_ERR(bdev); | |
2136 | } | |
1da177e4 LT |
2137 | rdev->bdev = bdev; |
2138 | return err; | |
2139 | } | |
2140 | ||
3cb03002 | 2141 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2142 | { |
2143 | struct block_device *bdev = rdev->bdev; | |
2144 | rdev->bdev = NULL; | |
e525fd89 | 2145 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2146 | } |
2147 | ||
2148 | void md_autodetect_dev(dev_t dev); | |
2149 | ||
f72ffdd6 | 2150 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2151 | { |
2152 | char b[BDEVNAME_SIZE]; | |
403df478 | 2153 | |
1da177e4 LT |
2154 | printk(KERN_INFO "md: export_rdev(%s)\n", |
2155 | bdevname(rdev->bdev,b)); | |
545c8795 | 2156 | md_rdev_clear(rdev); |
1da177e4 | 2157 | #ifndef MODULE |
d0fae18f N |
2158 | if (test_bit(AutoDetected, &rdev->flags)) |
2159 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2160 | #endif |
2161 | unlock_rdev(rdev); | |
86e6ffdd | 2162 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2163 | } |
2164 | ||
f72ffdd6 | 2165 | static void kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2166 | { |
2167 | unbind_rdev_from_array(rdev); | |
2168 | export_rdev(rdev); | |
2169 | } | |
2170 | ||
fd01b88c | 2171 | static void export_array(struct mddev *mddev) |
1da177e4 | 2172 | { |
0638bb0e | 2173 | struct md_rdev *rdev; |
1da177e4 | 2174 | |
0638bb0e N |
2175 | while (!list_empty(&mddev->disks)) { |
2176 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2177 | same_set); | |
1da177e4 LT |
2178 | kick_rdev_from_array(rdev); |
2179 | } | |
1da177e4 LT |
2180 | mddev->raid_disks = 0; |
2181 | mddev->major_version = 0; | |
2182 | } | |
2183 | ||
f72ffdd6 | 2184 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2185 | { |
42543769 N |
2186 | /* Update each superblock (in-memory image), but |
2187 | * if we are allowed to, skip spares which already | |
2188 | * have the right event counter, or have one earlier | |
2189 | * (which would mean they aren't being marked as dirty | |
2190 | * with the rest of the array) | |
2191 | */ | |
3cb03002 | 2192 | struct md_rdev *rdev; |
dafb20fa | 2193 | rdev_for_each(rdev, mddev) { |
42543769 N |
2194 | if (rdev->sb_events == mddev->events || |
2195 | (nospares && | |
2196 | rdev->raid_disk < 0 && | |
42543769 N |
2197 | rdev->sb_events+1 == mddev->events)) { |
2198 | /* Don't update this superblock */ | |
2199 | rdev->sb_loaded = 2; | |
2200 | } else { | |
076f968b | 2201 | sync_super(mddev, rdev); |
42543769 N |
2202 | rdev->sb_loaded = 1; |
2203 | } | |
1da177e4 LT |
2204 | } |
2205 | } | |
2206 | ||
f72ffdd6 | 2207 | static void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2208 | { |
3cb03002 | 2209 | struct md_rdev *rdev; |
06d91a5f | 2210 | int sync_req; |
42543769 | 2211 | int nospares = 0; |
2699b672 | 2212 | int any_badblocks_changed = 0; |
1da177e4 | 2213 | |
d87f064f N |
2214 | if (mddev->ro) { |
2215 | if (force_change) | |
2216 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2217 | return; | |
2218 | } | |
1da177e4 | 2219 | repeat: |
3a3a5ddb | 2220 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2221 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2222 | if (rdev->raid_disk >= 0 && |
2223 | mddev->delta_disks >= 0 && | |
2224 | !test_bit(In_sync, &rdev->flags) && | |
2225 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2226 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2227 | ||
f72ffdd6 | 2228 | } |
bd52b746 | 2229 | if (!mddev->persistent) { |
070dc6dd | 2230 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2231 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
de393cde | 2232 | if (!mddev->external) { |
d97a41dc | 2233 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
dafb20fa | 2234 | rdev_for_each(rdev, mddev) { |
de393cde | 2235 | if (rdev->badblocks.changed) { |
d0962936 | 2236 | rdev->badblocks.changed = 0; |
de393cde N |
2237 | md_ack_all_badblocks(&rdev->badblocks); |
2238 | md_error(mddev, rdev); | |
2239 | } | |
2240 | clear_bit(Blocked, &rdev->flags); | |
2241 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2242 | wake_up(&rdev->blocked_wait); | |
2243 | } | |
2244 | } | |
3a3a5ddb N |
2245 | wake_up(&mddev->sb_wait); |
2246 | return; | |
2247 | } | |
2248 | ||
85572d7c | 2249 | spin_lock(&mddev->lock); |
84692195 | 2250 | |
3a3a5ddb N |
2251 | mddev->utime = get_seconds(); |
2252 | ||
850b2b42 N |
2253 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2254 | force_change = 1; | |
2255 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2256 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2257 | * though if events isn't the right even/odd, we will have to do | |
2258 | * spares after all | |
2259 | */ | |
2260 | nospares = 1; | |
2261 | if (force_change) | |
2262 | nospares = 0; | |
2263 | if (mddev->degraded) | |
84692195 N |
2264 | /* If the array is degraded, then skipping spares is both |
2265 | * dangerous and fairly pointless. | |
2266 | * Dangerous because a device that was removed from the array | |
2267 | * might have a event_count that still looks up-to-date, | |
2268 | * so it can be re-added without a resync. | |
2269 | * Pointless because if there are any spares to skip, | |
2270 | * then a recovery will happen and soon that array won't | |
2271 | * be degraded any more and the spare can go back to sleep then. | |
2272 | */ | |
850b2b42 | 2273 | nospares = 0; |
84692195 | 2274 | |
06d91a5f | 2275 | sync_req = mddev->in_sync; |
42543769 N |
2276 | |
2277 | /* If this is just a dirty<->clean transition, and the array is clean | |
2278 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2279 | if (nospares |
42543769 | 2280 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2281 | && mddev->can_decrease_events |
2282 | && mddev->events != 1) { | |
42543769 | 2283 | mddev->events--; |
a8707c08 N |
2284 | mddev->can_decrease_events = 0; |
2285 | } else { | |
42543769 N |
2286 | /* otherwise we have to go forward and ... */ |
2287 | mddev->events ++; | |
a8707c08 | 2288 | mddev->can_decrease_events = nospares; |
42543769 | 2289 | } |
1da177e4 | 2290 | |
403df478 N |
2291 | /* |
2292 | * This 64-bit counter should never wrap. | |
2293 | * Either we are in around ~1 trillion A.C., assuming | |
2294 | * 1 reboot per second, or we have a bug... | |
2295 | */ | |
2296 | WARN_ON(mddev->events == 0); | |
2699b672 | 2297 | |
dafb20fa | 2298 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2299 | if (rdev->badblocks.changed) |
2300 | any_badblocks_changed++; | |
de393cde N |
2301 | if (test_bit(Faulty, &rdev->flags)) |
2302 | set_bit(FaultRecorded, &rdev->flags); | |
2303 | } | |
2699b672 | 2304 | |
e691063a | 2305 | sync_sbs(mddev, nospares); |
85572d7c | 2306 | spin_unlock(&mddev->lock); |
1da177e4 | 2307 | |
36a4e1fe N |
2308 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2309 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2310 | |
4ad13663 | 2311 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2312 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2313 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2314 | |
42543769 N |
2315 | if (rdev->sb_loaded != 1) |
2316 | continue; /* no noise on spare devices */ | |
1da177e4 | 2317 | |
f466722c | 2318 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2319 | md_super_write(mddev,rdev, |
0f420358 | 2320 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2321 | rdev->sb_page); |
36a4e1fe N |
2322 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2323 | bdevname(rdev->bdev, b), | |
2324 | (unsigned long long)rdev->sb_start); | |
42543769 | 2325 | rdev->sb_events = mddev->events; |
2699b672 N |
2326 | if (rdev->badblocks.size) { |
2327 | md_super_write(mddev, rdev, | |
2328 | rdev->badblocks.sector, | |
2329 | rdev->badblocks.size << 9, | |
2330 | rdev->bb_page); | |
2331 | rdev->badblocks.size = 0; | |
2332 | } | |
7bfa19f2 | 2333 | |
f466722c | 2334 | } else |
36a4e1fe N |
2335 | pr_debug("md: %s (skipping faulty)\n", |
2336 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2337 | |
7bfa19f2 | 2338 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2339 | /* only need to write one superblock... */ |
2340 | break; | |
2341 | } | |
a9701a30 | 2342 | md_super_wait(mddev); |
850b2b42 | 2343 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2344 | |
85572d7c | 2345 | spin_lock(&mddev->lock); |
850b2b42 N |
2346 | if (mddev->in_sync != sync_req || |
2347 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2348 | /* have to write it out again */ |
85572d7c | 2349 | spin_unlock(&mddev->lock); |
06d91a5f N |
2350 | goto repeat; |
2351 | } | |
850b2b42 | 2352 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
85572d7c | 2353 | spin_unlock(&mddev->lock); |
3d310eb7 | 2354 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2355 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2356 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2357 | |
dafb20fa | 2358 | rdev_for_each(rdev, mddev) { |
de393cde N |
2359 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2360 | clear_bit(Blocked, &rdev->flags); | |
2361 | ||
2362 | if (any_badblocks_changed) | |
2699b672 | 2363 | md_ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2364 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2365 | wake_up(&rdev->blocked_wait); | |
2366 | } | |
1da177e4 LT |
2367 | } |
2368 | ||
7f6ce769 | 2369 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2370 | * We want to accept with case. For this we use cmd_match. |
2371 | */ | |
2372 | static int cmd_match(const char *cmd, const char *str) | |
2373 | { | |
2374 | /* See if cmd, written into a sysfs file, matches | |
2375 | * str. They must either be the same, or cmd can | |
2376 | * have a trailing newline | |
2377 | */ | |
2378 | while (*cmd && *str && *cmd == *str) { | |
2379 | cmd++; | |
2380 | str++; | |
2381 | } | |
2382 | if (*cmd == '\n') | |
2383 | cmd++; | |
2384 | if (*str || *cmd) | |
2385 | return 0; | |
2386 | return 1; | |
2387 | } | |
2388 | ||
86e6ffdd N |
2389 | struct rdev_sysfs_entry { |
2390 | struct attribute attr; | |
3cb03002 N |
2391 | ssize_t (*show)(struct md_rdev *, char *); |
2392 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2393 | }; |
2394 | ||
2395 | static ssize_t | |
3cb03002 | 2396 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd N |
2397 | { |
2398 | char *sep = ""; | |
20a49ff6 | 2399 | size_t len = 0; |
758bfc8a | 2400 | unsigned long flags = ACCESS_ONCE(rdev->flags); |
86e6ffdd | 2401 | |
758bfc8a | 2402 | if (test_bit(Faulty, &flags) || |
de393cde | 2403 | rdev->badblocks.unacked_exist) { |
86e6ffdd N |
2404 | len+= sprintf(page+len, "%sfaulty",sep); |
2405 | sep = ","; | |
2406 | } | |
758bfc8a | 2407 | if (test_bit(In_sync, &flags)) { |
86e6ffdd N |
2408 | len += sprintf(page+len, "%sin_sync",sep); |
2409 | sep = ","; | |
2410 | } | |
758bfc8a | 2411 | if (test_bit(WriteMostly, &flags)) { |
f655675b N |
2412 | len += sprintf(page+len, "%swrite_mostly",sep); |
2413 | sep = ","; | |
2414 | } | |
758bfc8a | 2415 | if (test_bit(Blocked, &flags) || |
52c64152 | 2416 | (rdev->badblocks.unacked_exist |
758bfc8a | 2417 | && !test_bit(Faulty, &flags))) { |
6bfe0b49 DW |
2418 | len += sprintf(page+len, "%sblocked", sep); |
2419 | sep = ","; | |
2420 | } | |
758bfc8a N |
2421 | if (!test_bit(Faulty, &flags) && |
2422 | !test_bit(In_sync, &flags)) { | |
86e6ffdd N |
2423 | len += sprintf(page+len, "%sspare", sep); |
2424 | sep = ","; | |
2425 | } | |
758bfc8a | 2426 | if (test_bit(WriteErrorSeen, &flags)) { |
d7a9d443 N |
2427 | len += sprintf(page+len, "%swrite_error", sep); |
2428 | sep = ","; | |
2429 | } | |
758bfc8a | 2430 | if (test_bit(WantReplacement, &flags)) { |
2d78f8c4 N |
2431 | len += sprintf(page+len, "%swant_replacement", sep); |
2432 | sep = ","; | |
2433 | } | |
758bfc8a | 2434 | if (test_bit(Replacement, &flags)) { |
2d78f8c4 N |
2435 | len += sprintf(page+len, "%sreplacement", sep); |
2436 | sep = ","; | |
2437 | } | |
2438 | ||
86e6ffdd N |
2439 | return len+sprintf(page+len, "\n"); |
2440 | } | |
2441 | ||
45dc2de1 | 2442 | static ssize_t |
3cb03002 | 2443 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2444 | { |
2445 | /* can write | |
de393cde | 2446 | * faulty - simulates an error |
45dc2de1 | 2447 | * remove - disconnects the device |
f655675b N |
2448 | * writemostly - sets write_mostly |
2449 | * -writemostly - clears write_mostly | |
de393cde N |
2450 | * blocked - sets the Blocked flags |
2451 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2452 | * insync - sets Insync providing device isn't active |
f466722c N |
2453 | * -insync - clear Insync for a device with a slot assigned, |
2454 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2455 | * write_error - sets WriteErrorSeen |
2456 | * -write_error - clears WriteErrorSeen | |
45dc2de1 N |
2457 | */ |
2458 | int err = -EINVAL; | |
2459 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2460 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2461 | if (test_bit(Faulty, &rdev->flags)) |
2462 | err = 0; | |
2463 | else | |
2464 | err = -EBUSY; | |
45dc2de1 N |
2465 | } else if (cmd_match(buf, "remove")) { |
2466 | if (rdev->raid_disk >= 0) | |
2467 | err = -EBUSY; | |
2468 | else { | |
fd01b88c | 2469 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2470 | kick_rdev_from_array(rdev); |
3f9d7b0d N |
2471 | if (mddev->pers) |
2472 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2473 | md_new_event(mddev); |
2474 | err = 0; | |
2475 | } | |
f655675b N |
2476 | } else if (cmd_match(buf, "writemostly")) { |
2477 | set_bit(WriteMostly, &rdev->flags); | |
2478 | err = 0; | |
2479 | } else if (cmd_match(buf, "-writemostly")) { | |
2480 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2481 | err = 0; |
2482 | } else if (cmd_match(buf, "blocked")) { | |
2483 | set_bit(Blocked, &rdev->flags); | |
2484 | err = 0; | |
2485 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2486 | if (!test_bit(Faulty, &rdev->flags) && |
7da64a0a | 2487 | rdev->badblocks.unacked_exist) { |
de393cde N |
2488 | /* metadata handler doesn't understand badblocks, |
2489 | * so we need to fail the device | |
2490 | */ | |
2491 | md_error(rdev->mddev, rdev); | |
2492 | } | |
6bfe0b49 | 2493 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2494 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2495 | wake_up(&rdev->blocked_wait); |
2496 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2497 | md_wakeup_thread(rdev->mddev->thread); | |
2498 | ||
6d56e278 N |
2499 | err = 0; |
2500 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2501 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2502 | err = 0; |
f466722c | 2503 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) { |
e1960f8c N |
2504 | if (rdev->mddev->pers == NULL) { |
2505 | clear_bit(In_sync, &rdev->flags); | |
2506 | rdev->saved_raid_disk = rdev->raid_disk; | |
2507 | rdev->raid_disk = -1; | |
2508 | err = 0; | |
2509 | } | |
d7a9d443 N |
2510 | } else if (cmd_match(buf, "write_error")) { |
2511 | set_bit(WriteErrorSeen, &rdev->flags); | |
2512 | err = 0; | |
2513 | } else if (cmd_match(buf, "-write_error")) { | |
2514 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2515 | err = 0; | |
2d78f8c4 N |
2516 | } else if (cmd_match(buf, "want_replacement")) { |
2517 | /* Any non-spare device that is not a replacement can | |
2518 | * become want_replacement at any time, but we then need to | |
2519 | * check if recovery is needed. | |
2520 | */ | |
2521 | if (rdev->raid_disk >= 0 && | |
2522 | !test_bit(Replacement, &rdev->flags)) | |
2523 | set_bit(WantReplacement, &rdev->flags); | |
2524 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2525 | md_wakeup_thread(rdev->mddev->thread); | |
2526 | err = 0; | |
2527 | } else if (cmd_match(buf, "-want_replacement")) { | |
2528 | /* Clearing 'want_replacement' is always allowed. | |
2529 | * Once replacements starts it is too late though. | |
2530 | */ | |
2531 | err = 0; | |
2532 | clear_bit(WantReplacement, &rdev->flags); | |
2533 | } else if (cmd_match(buf, "replacement")) { | |
2534 | /* Can only set a device as a replacement when array has not | |
2535 | * yet been started. Once running, replacement is automatic | |
2536 | * from spares, or by assigning 'slot'. | |
2537 | */ | |
2538 | if (rdev->mddev->pers) | |
2539 | err = -EBUSY; | |
2540 | else { | |
2541 | set_bit(Replacement, &rdev->flags); | |
2542 | err = 0; | |
2543 | } | |
2544 | } else if (cmd_match(buf, "-replacement")) { | |
2545 | /* Similarly, can only clear Replacement before start */ | |
2546 | if (rdev->mddev->pers) | |
2547 | err = -EBUSY; | |
2548 | else { | |
2549 | clear_bit(Replacement, &rdev->flags); | |
2550 | err = 0; | |
2551 | } | |
45dc2de1 | 2552 | } |
00bcb4ac N |
2553 | if (!err) |
2554 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2555 | return err ? err : len; |
2556 | } | |
80ca3a44 N |
2557 | static struct rdev_sysfs_entry rdev_state = |
2558 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2559 | |
4dbcdc75 | 2560 | static ssize_t |
3cb03002 | 2561 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2562 | { |
2563 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2564 | } | |
2565 | ||
2566 | static ssize_t | |
3cb03002 | 2567 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 N |
2568 | { |
2569 | char *e; | |
2570 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2571 | if (*buf && (*e == 0 || *e == '\n')) { | |
2572 | atomic_set(&rdev->corrected_errors, n); | |
2573 | return len; | |
2574 | } | |
2575 | return -EINVAL; | |
2576 | } | |
2577 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2578 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2579 | |
014236d2 | 2580 | static ssize_t |
3cb03002 | 2581 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 N |
2582 | { |
2583 | if (rdev->raid_disk < 0) | |
2584 | return sprintf(page, "none\n"); | |
2585 | else | |
2586 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2587 | } | |
2588 | ||
2589 | static ssize_t | |
3cb03002 | 2590 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 N |
2591 | { |
2592 | char *e; | |
c303da6d | 2593 | int err; |
014236d2 N |
2594 | int slot = simple_strtoul(buf, &e, 10); |
2595 | if (strncmp(buf, "none", 4)==0) | |
2596 | slot = -1; | |
2597 | else if (e==buf || (*e && *e!= '\n')) | |
2598 | return -EINVAL; | |
6c2fce2e | 2599 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2600 | /* Setting 'slot' on an active array requires also |
2601 | * updating the 'rd%d' link, and communicating | |
2602 | * with the personality with ->hot_*_disk. | |
2603 | * For now we only support removing | |
2604 | * failed/spare devices. This normally happens automatically, | |
2605 | * but not when the metadata is externally managed. | |
2606 | */ | |
c303da6d N |
2607 | if (rdev->raid_disk == -1) |
2608 | return -EEXIST; | |
2609 | /* personality does all needed checks */ | |
01393f3d | 2610 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2611 | return -EINVAL; |
746d3207 N |
2612 | clear_bit(Blocked, &rdev->flags); |
2613 | remove_and_add_spares(rdev->mddev, rdev); | |
2614 | if (rdev->raid_disk >= 0) | |
2615 | return -EBUSY; | |
c303da6d N |
2616 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2617 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2618 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2619 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2620 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2621 | */ |
2622 | ||
2623 | if (rdev->raid_disk != -1) | |
2624 | return -EBUSY; | |
2625 | ||
c6751b2b N |
2626 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2627 | return -EBUSY; | |
2628 | ||
6c2fce2e NB |
2629 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2630 | return -EINVAL; | |
2631 | ||
ba1b41b6 N |
2632 | if (slot >= rdev->mddev->raid_disks && |
2633 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2634 | return -ENOSPC; | |
2635 | ||
6c2fce2e NB |
2636 | rdev->raid_disk = slot; |
2637 | if (test_bit(In_sync, &rdev->flags)) | |
2638 | rdev->saved_raid_disk = slot; | |
2639 | else | |
2640 | rdev->saved_raid_disk = -1; | |
d30519fc | 2641 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2642 | clear_bit(Bitmap_sync, &rdev->flags); |
6c2fce2e NB |
2643 | err = rdev->mddev->pers-> |
2644 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2645 | if (err) { |
6c2fce2e | 2646 | rdev->raid_disk = -1; |
6c2fce2e | 2647 | return err; |
52664732 | 2648 | } else |
00bcb4ac | 2649 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
36fad858 | 2650 | if (sysfs_link_rdev(rdev->mddev, rdev)) |
00bcb4ac | 2651 | /* failure here is OK */; |
6c2fce2e | 2652 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2653 | } else { |
ba1b41b6 N |
2654 | if (slot >= rdev->mddev->raid_disks && |
2655 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2656 | return -ENOSPC; |
2657 | rdev->raid_disk = slot; | |
2658 | /* assume it is working */ | |
c5d79adb N |
2659 | clear_bit(Faulty, &rdev->flags); |
2660 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2661 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2662 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2663 | } |
014236d2 N |
2664 | return len; |
2665 | } | |
2666 | ||
014236d2 | 2667 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2668 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2669 | |
93c8cad0 | 2670 | static ssize_t |
3cb03002 | 2671 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2672 | { |
6961ece4 | 2673 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2674 | } |
2675 | ||
2676 | static ssize_t | |
3cb03002 | 2677 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2678 | { |
c6563a8c | 2679 | unsigned long long offset; |
b29bebd6 | 2680 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2681 | return -EINVAL; |
8ed0a521 | 2682 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2683 | return -EBUSY; |
dd8ac336 | 2684 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2685 | /* Must set offset before size, so overlap checks |
2686 | * can be sane */ | |
2687 | return -EBUSY; | |
93c8cad0 | 2688 | rdev->data_offset = offset; |
25f7fd47 | 2689 | rdev->new_data_offset = offset; |
93c8cad0 N |
2690 | return len; |
2691 | } | |
2692 | ||
2693 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2694 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2695 | |
c6563a8c N |
2696 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2697 | { | |
2698 | return sprintf(page, "%llu\n", | |
2699 | (unsigned long long)rdev->new_data_offset); | |
2700 | } | |
2701 | ||
2702 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2703 | const char *buf, size_t len) | |
2704 | { | |
2705 | unsigned long long new_offset; | |
2706 | struct mddev *mddev = rdev->mddev; | |
2707 | ||
b29bebd6 | 2708 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2709 | return -EINVAL; |
2710 | ||
f851b60d N |
2711 | if (mddev->sync_thread || |
2712 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
2713 | return -EBUSY; |
2714 | if (new_offset == rdev->data_offset) | |
2715 | /* reset is always permitted */ | |
2716 | ; | |
2717 | else if (new_offset > rdev->data_offset) { | |
2718 | /* must not push array size beyond rdev_sectors */ | |
2719 | if (new_offset - rdev->data_offset | |
2720 | + mddev->dev_sectors > rdev->sectors) | |
2721 | return -E2BIG; | |
2722 | } | |
2723 | /* Metadata worries about other space details. */ | |
2724 | ||
2725 | /* decreasing the offset is inconsistent with a backwards | |
2726 | * reshape. | |
2727 | */ | |
2728 | if (new_offset < rdev->data_offset && | |
2729 | mddev->reshape_backwards) | |
2730 | return -EINVAL; | |
2731 | /* Increasing offset is inconsistent with forwards | |
2732 | * reshape. reshape_direction should be set to | |
2733 | * 'backwards' first. | |
2734 | */ | |
2735 | if (new_offset > rdev->data_offset && | |
2736 | !mddev->reshape_backwards) | |
2737 | return -EINVAL; | |
2738 | ||
2739 | if (mddev->pers && mddev->persistent && | |
2740 | !super_types[mddev->major_version] | |
2741 | .allow_new_offset(rdev, new_offset)) | |
2742 | return -E2BIG; | |
2743 | rdev->new_data_offset = new_offset; | |
2744 | if (new_offset > rdev->data_offset) | |
2745 | mddev->reshape_backwards = 1; | |
2746 | else if (new_offset < rdev->data_offset) | |
2747 | mddev->reshape_backwards = 0; | |
2748 | ||
2749 | return len; | |
2750 | } | |
2751 | static struct rdev_sysfs_entry rdev_new_offset = | |
2752 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2753 | ||
83303b61 | 2754 | static ssize_t |
3cb03002 | 2755 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2756 | { |
dd8ac336 | 2757 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2758 | } |
2759 | ||
c5d79adb N |
2760 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2761 | { | |
2762 | /* check if two start/length pairs overlap */ | |
2763 | if (s1+l1 <= s2) | |
2764 | return 0; | |
2765 | if (s2+l2 <= s1) | |
2766 | return 0; | |
2767 | return 1; | |
2768 | } | |
2769 | ||
b522adcd DW |
2770 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2771 | { | |
2772 | unsigned long long blocks; | |
2773 | sector_t new; | |
2774 | ||
b29bebd6 | 2775 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
2776 | return -EINVAL; |
2777 | ||
2778 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2779 | return -EINVAL; /* sector conversion overflow */ | |
2780 | ||
2781 | new = blocks * 2; | |
2782 | if (new != blocks * 2) | |
2783 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2784 | ||
2785 | *sectors = new; | |
2786 | return 0; | |
2787 | } | |
2788 | ||
83303b61 | 2789 | static ssize_t |
3cb03002 | 2790 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2791 | { |
fd01b88c | 2792 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2793 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2794 | sector_t sectors; |
27c529bb | 2795 | |
b522adcd | 2796 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2797 | return -EINVAL; |
c6563a8c N |
2798 | if (rdev->data_offset != rdev->new_data_offset) |
2799 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2800 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2801 | if (my_mddev->persistent) { |
dd8ac336 AN |
2802 | sectors = super_types[my_mddev->major_version]. |
2803 | rdev_size_change(rdev, sectors); | |
2804 | if (!sectors) | |
0cd17fec | 2805 | return -EBUSY; |
dd8ac336 | 2806 | } else if (!sectors) |
77304d2a | 2807 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2808 | rdev->data_offset; |
a6468539 N |
2809 | if (!my_mddev->pers->resize) |
2810 | /* Cannot change size for RAID0 or Linear etc */ | |
2811 | return -EINVAL; | |
0cd17fec | 2812 | } |
dd8ac336 | 2813 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2814 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2815 | |
dd8ac336 AN |
2816 | rdev->sectors = sectors; |
2817 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
2818 | /* Need to check that all other rdevs with the same |
2819 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
2820 | * the rdev lists safely. | |
2821 | * This check does not provide a hard guarantee, it | |
2822 | * just helps avoid dangerous mistakes. | |
c5d79adb | 2823 | */ |
fd01b88c | 2824 | struct mddev *mddev; |
c5d79adb | 2825 | int overlap = 0; |
159ec1fc | 2826 | struct list_head *tmp; |
c5d79adb | 2827 | |
8b1afc3d | 2828 | rcu_read_lock(); |
29ac4aa3 | 2829 | for_each_mddev(mddev, tmp) { |
3cb03002 | 2830 | struct md_rdev *rdev2; |
c5d79adb | 2831 | |
dafb20fa | 2832 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
2833 | if (rdev->bdev == rdev2->bdev && |
2834 | rdev != rdev2 && | |
2835 | overlaps(rdev->data_offset, rdev->sectors, | |
2836 | rdev2->data_offset, | |
2837 | rdev2->sectors)) { | |
c5d79adb N |
2838 | overlap = 1; |
2839 | break; | |
2840 | } | |
c5d79adb N |
2841 | if (overlap) { |
2842 | mddev_put(mddev); | |
2843 | break; | |
2844 | } | |
2845 | } | |
8b1afc3d | 2846 | rcu_read_unlock(); |
c5d79adb N |
2847 | if (overlap) { |
2848 | /* Someone else could have slipped in a size | |
2849 | * change here, but doing so is just silly. | |
dd8ac336 | 2850 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2851 | * safe, and trust userspace not to race with |
2852 | * itself | |
2853 | */ | |
dd8ac336 | 2854 | rdev->sectors = oldsectors; |
c5d79adb N |
2855 | return -EBUSY; |
2856 | } | |
2857 | } | |
83303b61 N |
2858 | return len; |
2859 | } | |
2860 | ||
2861 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2862 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2863 | |
3cb03002 | 2864 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
2865 | { |
2866 | unsigned long long recovery_start = rdev->recovery_offset; | |
2867 | ||
2868 | if (test_bit(In_sync, &rdev->flags) || | |
2869 | recovery_start == MaxSector) | |
2870 | return sprintf(page, "none\n"); | |
2871 | ||
2872 | return sprintf(page, "%llu\n", recovery_start); | |
2873 | } | |
2874 | ||
3cb03002 | 2875 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
2876 | { |
2877 | unsigned long long recovery_start; | |
2878 | ||
2879 | if (cmd_match(buf, "none")) | |
2880 | recovery_start = MaxSector; | |
b29bebd6 | 2881 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
2882 | return -EINVAL; |
2883 | ||
2884 | if (rdev->mddev->pers && | |
2885 | rdev->raid_disk >= 0) | |
2886 | return -EBUSY; | |
2887 | ||
2888 | rdev->recovery_offset = recovery_start; | |
2889 | if (recovery_start == MaxSector) | |
2890 | set_bit(In_sync, &rdev->flags); | |
2891 | else | |
2892 | clear_bit(In_sync, &rdev->flags); | |
2893 | return len; | |
2894 | } | |
2895 | ||
2896 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2897 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2898 | ||
16c791a5 N |
2899 | static ssize_t |
2900 | badblocks_show(struct badblocks *bb, char *page, int unack); | |
2901 | static ssize_t | |
2902 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); | |
2903 | ||
3cb03002 | 2904 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2905 | { |
2906 | return badblocks_show(&rdev->badblocks, page, 0); | |
2907 | } | |
3cb03002 | 2908 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 2909 | { |
de393cde N |
2910 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
2911 | /* Maybe that ack was all we needed */ | |
2912 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
2913 | wake_up(&rdev->blocked_wait); | |
2914 | return rv; | |
16c791a5 N |
2915 | } |
2916 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
2917 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
2918 | ||
3cb03002 | 2919 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2920 | { |
2921 | return badblocks_show(&rdev->badblocks, page, 1); | |
2922 | } | |
3cb03002 | 2923 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
2924 | { |
2925 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
2926 | } | |
2927 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
2928 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
2929 | ||
86e6ffdd N |
2930 | static struct attribute *rdev_default_attrs[] = { |
2931 | &rdev_state.attr, | |
4dbcdc75 | 2932 | &rdev_errors.attr, |
014236d2 | 2933 | &rdev_slot.attr, |
93c8cad0 | 2934 | &rdev_offset.attr, |
c6563a8c | 2935 | &rdev_new_offset.attr, |
83303b61 | 2936 | &rdev_size.attr, |
06e3c817 | 2937 | &rdev_recovery_start.attr, |
16c791a5 N |
2938 | &rdev_bad_blocks.attr, |
2939 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
2940 | NULL, |
2941 | }; | |
2942 | static ssize_t | |
2943 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2944 | { | |
2945 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 2946 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
2947 | |
2948 | if (!entry->show) | |
2949 | return -EIO; | |
758bfc8a N |
2950 | if (!rdev->mddev) |
2951 | return -EBUSY; | |
2952 | return entry->show(rdev, page); | |
86e6ffdd N |
2953 | } |
2954 | ||
2955 | static ssize_t | |
2956 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2957 | const char *page, size_t length) | |
2958 | { | |
2959 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 2960 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 2961 | ssize_t rv; |
fd01b88c | 2962 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
2963 | |
2964 | if (!entry->store) | |
2965 | return -EIO; | |
67463acb N |
2966 | if (!capable(CAP_SYS_ADMIN)) |
2967 | return -EACCES; | |
27c529bb | 2968 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2969 | if (!rv) { |
27c529bb N |
2970 | if (rdev->mddev == NULL) |
2971 | rv = -EBUSY; | |
2972 | else | |
2973 | rv = entry->store(rdev, page, length); | |
6a51830e | 2974 | mddev_unlock(mddev); |
ca388059 N |
2975 | } |
2976 | return rv; | |
86e6ffdd N |
2977 | } |
2978 | ||
2979 | static void rdev_free(struct kobject *ko) | |
2980 | { | |
3cb03002 | 2981 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
2982 | kfree(rdev); |
2983 | } | |
52cf25d0 | 2984 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
2985 | .show = rdev_attr_show, |
2986 | .store = rdev_attr_store, | |
2987 | }; | |
2988 | static struct kobj_type rdev_ktype = { | |
2989 | .release = rdev_free, | |
2990 | .sysfs_ops = &rdev_sysfs_ops, | |
2991 | .default_attrs = rdev_default_attrs, | |
2992 | }; | |
2993 | ||
3cb03002 | 2994 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
2995 | { |
2996 | rdev->desc_nr = -1; | |
2997 | rdev->saved_raid_disk = -1; | |
2998 | rdev->raid_disk = -1; | |
2999 | rdev->flags = 0; | |
3000 | rdev->data_offset = 0; | |
c6563a8c | 3001 | rdev->new_data_offset = 0; |
e8bb9a83 N |
3002 | rdev->sb_events = 0; |
3003 | rdev->last_read_error.tv_sec = 0; | |
3004 | rdev->last_read_error.tv_nsec = 0; | |
2699b672 N |
3005 | rdev->sb_loaded = 0; |
3006 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3007 | atomic_set(&rdev->nr_pending, 0); |
3008 | atomic_set(&rdev->read_errors, 0); | |
3009 | atomic_set(&rdev->corrected_errors, 0); | |
3010 | ||
3011 | INIT_LIST_HEAD(&rdev->same_set); | |
3012 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3013 | |
3014 | /* Add space to store bad block list. | |
3015 | * This reserves the space even on arrays where it cannot | |
3016 | * be used - I wonder if that matters | |
3017 | */ | |
3018 | rdev->badblocks.count = 0; | |
486adf72 | 3019 | rdev->badblocks.shift = -1; /* disabled until explicitly enabled */ |
2230dfe4 N |
3020 | rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3021 | seqlock_init(&rdev->badblocks.lock); | |
3022 | if (rdev->badblocks.page == NULL) | |
3023 | return -ENOMEM; | |
3024 | ||
3025 | return 0; | |
e8bb9a83 N |
3026 | } |
3027 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3028 | /* |
3029 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3030 | * | |
3031 | * mark the device faulty if: | |
3032 | * | |
3033 | * - the device is nonexistent (zero size) | |
3034 | * - the device has no valid superblock | |
3035 | * | |
3036 | * a faulty rdev _never_ has rdev->sb set. | |
3037 | */ | |
3cb03002 | 3038 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3039 | { |
3040 | char b[BDEVNAME_SIZE]; | |
3041 | int err; | |
3cb03002 | 3042 | struct md_rdev *rdev; |
1da177e4 LT |
3043 | sector_t size; |
3044 | ||
9ffae0cf | 3045 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
3046 | if (!rdev) { |
3047 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
3048 | return ERR_PTR(-ENOMEM); | |
3049 | } | |
1da177e4 | 3050 | |
2230dfe4 N |
3051 | err = md_rdev_init(rdev); |
3052 | if (err) | |
3053 | goto abort_free; | |
3054 | err = alloc_disk_sb(rdev); | |
3055 | if (err) | |
1da177e4 LT |
3056 | goto abort_free; |
3057 | ||
c5d79adb | 3058 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3059 | if (err) |
3060 | goto abort_free; | |
3061 | ||
f9cb074b | 3062 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3063 | |
77304d2a | 3064 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3065 | if (!size) { |
f72ffdd6 | 3066 | printk(KERN_WARNING |
1da177e4 LT |
3067 | "md: %s has zero or unknown size, marking faulty!\n", |
3068 | bdevname(rdev->bdev,b)); | |
3069 | err = -EINVAL; | |
3070 | goto abort_free; | |
3071 | } | |
3072 | ||
3073 | if (super_format >= 0) { | |
3074 | err = super_types[super_format]. | |
3075 | load_super(rdev, NULL, super_minor); | |
3076 | if (err == -EINVAL) { | |
df968c4e N |
3077 | printk(KERN_WARNING |
3078 | "md: %s does not have a valid v%d.%d " | |
3079 | "superblock, not importing!\n", | |
3080 | bdevname(rdev->bdev,b), | |
3081 | super_format, super_minor); | |
1da177e4 LT |
3082 | goto abort_free; |
3083 | } | |
3084 | if (err < 0) { | |
f72ffdd6 | 3085 | printk(KERN_WARNING |
1da177e4 LT |
3086 | "md: could not read %s's sb, not importing!\n", |
3087 | bdevname(rdev->bdev,b)); | |
3088 | goto abort_free; | |
3089 | } | |
3090 | } | |
6bfe0b49 | 3091 | |
1da177e4 LT |
3092 | return rdev; |
3093 | ||
3094 | abort_free: | |
2699b672 N |
3095 | if (rdev->bdev) |
3096 | unlock_rdev(rdev); | |
545c8795 | 3097 | md_rdev_clear(rdev); |
1da177e4 LT |
3098 | kfree(rdev); |
3099 | return ERR_PTR(err); | |
3100 | } | |
3101 | ||
3102 | /* | |
3103 | * Check a full RAID array for plausibility | |
3104 | */ | |
3105 | ||
f72ffdd6 | 3106 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3107 | { |
3108 | int i; | |
3cb03002 | 3109 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3110 | char b[BDEVNAME_SIZE]; |
3111 | ||
3112 | freshest = NULL; | |
dafb20fa | 3113 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3114 | switch (super_types[mddev->major_version]. |
3115 | load_super(rdev, freshest, mddev->minor_version)) { | |
3116 | case 1: | |
3117 | freshest = rdev; | |
3118 | break; | |
3119 | case 0: | |
3120 | break; | |
3121 | default: | |
3122 | printk( KERN_ERR \ | |
3123 | "md: fatal superblock inconsistency in %s" | |
f72ffdd6 | 3124 | " -- removing from array\n", |
1da177e4 LT |
3125 | bdevname(rdev->bdev,b)); |
3126 | kick_rdev_from_array(rdev); | |
3127 | } | |
3128 | ||
1da177e4 LT |
3129 | super_types[mddev->major_version]. |
3130 | validate_super(mddev, freshest); | |
3131 | ||
3132 | i = 0; | |
dafb20fa | 3133 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3134 | if (mddev->max_disks && |
3135 | (rdev->desc_nr >= mddev->max_disks || | |
3136 | i > mddev->max_disks)) { | |
de01dfad N |
3137 | printk(KERN_WARNING |
3138 | "md: %s: %s: only %d devices permitted\n", | |
3139 | mdname(mddev), bdevname(rdev->bdev, b), | |
3140 | mddev->max_disks); | |
3141 | kick_rdev_from_array(rdev); | |
3142 | continue; | |
3143 | } | |
1da177e4 LT |
3144 | if (rdev != freshest) |
3145 | if (super_types[mddev->major_version]. | |
3146 | validate_super(mddev, rdev)) { | |
3147 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
3148 | " from array!\n", | |
3149 | bdevname(rdev->bdev,b)); | |
3150 | kick_rdev_from_array(rdev); | |
3151 | continue; | |
3152 | } | |
3153 | if (mddev->level == LEVEL_MULTIPATH) { | |
3154 | rdev->desc_nr = i++; | |
3155 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3156 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 3157 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
3158 | rdev->raid_disk = -1; |
3159 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3160 | } |
3161 | } | |
1da177e4 LT |
3162 | } |
3163 | ||
72e02075 N |
3164 | /* Read a fixed-point number. |
3165 | * Numbers in sysfs attributes should be in "standard" units where | |
3166 | * possible, so time should be in seconds. | |
f72ffdd6 | 3167 | * However we internally use a a much smaller unit such as |
72e02075 N |
3168 | * milliseconds or jiffies. |
3169 | * This function takes a decimal number with a possible fractional | |
3170 | * component, and produces an integer which is the result of | |
3171 | * multiplying that number by 10^'scale'. | |
3172 | * all without any floating-point arithmetic. | |
3173 | */ | |
3174 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3175 | { | |
3176 | unsigned long result = 0; | |
3177 | long decimals = -1; | |
3178 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3179 | if (*cp == '.') | |
3180 | decimals = 0; | |
3181 | else if (decimals < scale) { | |
3182 | unsigned int value; | |
3183 | value = *cp - '0'; | |
3184 | result = result * 10 + value; | |
3185 | if (decimals >= 0) | |
3186 | decimals++; | |
3187 | } | |
3188 | cp++; | |
3189 | } | |
3190 | if (*cp == '\n') | |
3191 | cp++; | |
3192 | if (*cp) | |
3193 | return -EINVAL; | |
3194 | if (decimals < 0) | |
3195 | decimals = 0; | |
3196 | while (decimals < scale) { | |
3197 | result *= 10; | |
3198 | decimals ++; | |
3199 | } | |
3200 | *res = result; | |
3201 | return 0; | |
3202 | } | |
3203 | ||
19052c0e N |
3204 | static void md_safemode_timeout(unsigned long data); |
3205 | ||
16f17b39 | 3206 | static ssize_t |
fd01b88c | 3207 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3208 | { |
3209 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3210 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3211 | } | |
3212 | static ssize_t | |
fd01b88c | 3213 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3214 | { |
16f17b39 | 3215 | unsigned long msec; |
97ce0a7f | 3216 | |
72e02075 | 3217 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3218 | return -EINVAL; |
16f17b39 N |
3219 | if (msec == 0) |
3220 | mddev->safemode_delay = 0; | |
3221 | else { | |
19052c0e | 3222 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3223 | unsigned long new_delay = (msec*HZ)/1000; |
3224 | ||
3225 | if (new_delay == 0) | |
3226 | new_delay = 1; | |
3227 | mddev->safemode_delay = new_delay; | |
3228 | if (new_delay < old_delay || old_delay == 0) | |
3229 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3230 | } |
3231 | return len; | |
3232 | } | |
3233 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3234 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3235 | |
eae1701f | 3236 | static ssize_t |
fd01b88c | 3237 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3238 | { |
36d091f4 N |
3239 | struct md_personality *p; |
3240 | int ret; | |
3241 | spin_lock(&mddev->lock); | |
3242 | p = mddev->pers; | |
d9d166c2 | 3243 | if (p) |
36d091f4 | 3244 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3245 | else if (mddev->clevel[0]) |
36d091f4 | 3246 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3247 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3248 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3249 | else |
36d091f4 N |
3250 | ret = 0; |
3251 | spin_unlock(&mddev->lock); | |
3252 | return ret; | |
eae1701f N |
3253 | } |
3254 | ||
d9d166c2 | 3255 | static ssize_t |
fd01b88c | 3256 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3257 | { |
f2859af6 | 3258 | char clevel[16]; |
6791875e N |
3259 | ssize_t rv; |
3260 | size_t slen = len; | |
db721d32 | 3261 | struct md_personality *pers, *oldpers; |
f2859af6 | 3262 | long level; |
db721d32 | 3263 | void *priv, *oldpriv; |
3cb03002 | 3264 | struct md_rdev *rdev; |
245f46c2 | 3265 | |
6791875e N |
3266 | if (slen == 0 || slen >= sizeof(clevel)) |
3267 | return -EINVAL; | |
3268 | ||
3269 | rv = mddev_lock(mddev); | |
3270 | if (rv) | |
3271 | return rv; | |
3272 | ||
245f46c2 | 3273 | if (mddev->pers == NULL) { |
6791875e N |
3274 | strncpy(mddev->clevel, buf, slen); |
3275 | if (mddev->clevel[slen-1] == '\n') | |
3276 | slen--; | |
3277 | mddev->clevel[slen] = 0; | |
245f46c2 | 3278 | mddev->level = LEVEL_NONE; |
6791875e N |
3279 | rv = len; |
3280 | goto out_unlock; | |
245f46c2 | 3281 | } |
6791875e | 3282 | rv = -EROFS; |
bd8839e0 | 3283 | if (mddev->ro) |
6791875e | 3284 | goto out_unlock; |
245f46c2 N |
3285 | |
3286 | /* request to change the personality. Need to ensure: | |
3287 | * - array is not engaged in resync/recovery/reshape | |
3288 | * - old personality can be suspended | |
3289 | * - new personality will access other array. | |
3290 | */ | |
3291 | ||
6791875e | 3292 | rv = -EBUSY; |
bb4f1e9d | 3293 | if (mddev->sync_thread || |
f851b60d | 3294 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3295 | mddev->reshape_position != MaxSector || |
3296 | mddev->sysfs_active) | |
6791875e | 3297 | goto out_unlock; |
245f46c2 | 3298 | |
6791875e | 3299 | rv = -EINVAL; |
245f46c2 N |
3300 | if (!mddev->pers->quiesce) { |
3301 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3302 | mdname(mddev), mddev->pers->name); | |
6791875e | 3303 | goto out_unlock; |
245f46c2 N |
3304 | } |
3305 | ||
3306 | /* Now find the new personality */ | |
6791875e N |
3307 | strncpy(clevel, buf, slen); |
3308 | if (clevel[slen-1] == '\n') | |
3309 | slen--; | |
3310 | clevel[slen] = 0; | |
b29bebd6 | 3311 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3312 | level = LEVEL_NONE; |
245f46c2 | 3313 | |
f2859af6 DW |
3314 | if (request_module("md-%s", clevel) != 0) |
3315 | request_module("md-level-%s", clevel); | |
245f46c2 | 3316 | spin_lock(&pers_lock); |
f2859af6 | 3317 | pers = find_pers(level, clevel); |
245f46c2 N |
3318 | if (!pers || !try_module_get(pers->owner)) { |
3319 | spin_unlock(&pers_lock); | |
f2859af6 | 3320 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
6791875e N |
3321 | rv = -EINVAL; |
3322 | goto out_unlock; | |
245f46c2 N |
3323 | } |
3324 | spin_unlock(&pers_lock); | |
3325 | ||
3326 | if (pers == mddev->pers) { | |
3327 | /* Nothing to do! */ | |
3328 | module_put(pers->owner); | |
6791875e N |
3329 | rv = len; |
3330 | goto out_unlock; | |
245f46c2 N |
3331 | } |
3332 | if (!pers->takeover) { | |
3333 | module_put(pers->owner); | |
3334 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3335 | mdname(mddev), clevel); |
6791875e N |
3336 | rv = -EINVAL; |
3337 | goto out_unlock; | |
245f46c2 N |
3338 | } |
3339 | ||
dafb20fa | 3340 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3341 | rdev->new_raid_disk = rdev->raid_disk; |
3342 | ||
245f46c2 N |
3343 | /* ->takeover must set new_* and/or delta_disks |
3344 | * if it succeeds, and may set them when it fails. | |
3345 | */ | |
3346 | priv = pers->takeover(mddev); | |
3347 | if (IS_ERR(priv)) { | |
3348 | mddev->new_level = mddev->level; | |
3349 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3350 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3351 | mddev->raid_disks -= mddev->delta_disks; |
3352 | mddev->delta_disks = 0; | |
2c810cdd | 3353 | mddev->reshape_backwards = 0; |
245f46c2 N |
3354 | module_put(pers->owner); |
3355 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3356 | mdname(mddev), clevel); |
6791875e N |
3357 | rv = PTR_ERR(priv); |
3358 | goto out_unlock; | |
245f46c2 N |
3359 | } |
3360 | ||
3361 | /* Looks like we have a winner */ | |
3362 | mddev_suspend(mddev); | |
5aa61f42 | 3363 | mddev_detach(mddev); |
36d091f4 N |
3364 | |
3365 | spin_lock(&mddev->lock); | |
db721d32 N |
3366 | oldpers = mddev->pers; |
3367 | oldpriv = mddev->private; | |
3368 | mddev->pers = pers; | |
3369 | mddev->private = priv; | |
3370 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3371 | mddev->level = mddev->new_level; | |
3372 | mddev->layout = mddev->new_layout; | |
3373 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3374 | mddev->delta_disks = 0; | |
3375 | mddev->reshape_backwards = 0; | |
3376 | mddev->degraded = 0; | |
36d091f4 | 3377 | spin_unlock(&mddev->lock); |
db721d32 N |
3378 | |
3379 | if (oldpers->sync_request == NULL && | |
3380 | mddev->external) { | |
3381 | /* We are converting from a no-redundancy array | |
3382 | * to a redundancy array and metadata is managed | |
3383 | * externally so we need to be sure that writes | |
3384 | * won't block due to a need to transition | |
3385 | * clean->dirty | |
3386 | * until external management is started. | |
3387 | */ | |
3388 | mddev->in_sync = 0; | |
3389 | mddev->safemode_delay = 0; | |
3390 | mddev->safemode = 0; | |
3391 | } | |
f72ffdd6 | 3392 | |
db721d32 N |
3393 | oldpers->free(mddev, oldpriv); |
3394 | ||
3395 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3396 | pers->sync_request != NULL) { |
3397 | /* need to add the md_redundancy_group */ | |
3398 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3399 | printk(KERN_WARNING | |
3400 | "md: cannot register extra attributes for %s\n", | |
3401 | mdname(mddev)); | |
388975cc | 3402 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3403 | } |
db721d32 | 3404 | if (oldpers->sync_request != NULL && |
a64c876f N |
3405 | pers->sync_request == NULL) { |
3406 | /* need to remove the md_redundancy_group */ | |
3407 | if (mddev->to_remove == NULL) | |
3408 | mddev->to_remove = &md_redundancy_group; | |
3409 | } | |
3410 | ||
dafb20fa | 3411 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3412 | if (rdev->raid_disk < 0) |
3413 | continue; | |
bf2cb0da | 3414 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3415 | rdev->new_raid_disk = -1; |
3416 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3417 | continue; | |
36fad858 | 3418 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3419 | } |
dafb20fa | 3420 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3421 | if (rdev->raid_disk < 0) |
3422 | continue; | |
3423 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3424 | continue; | |
3425 | rdev->raid_disk = rdev->new_raid_disk; | |
3426 | if (rdev->raid_disk < 0) | |
3a981b03 | 3427 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3428 | else { |
36fad858 NK |
3429 | if (sysfs_link_rdev(mddev, rdev)) |
3430 | printk(KERN_WARNING "md: cannot register rd%d" | |
3431 | " for %s after level change\n", | |
3432 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3433 | } |
e93f68a1 N |
3434 | } |
3435 | ||
db721d32 | 3436 | if (pers->sync_request == NULL) { |
9af204cf TM |
3437 | /* this is now an array without redundancy, so |
3438 | * it must always be in_sync | |
3439 | */ | |
3440 | mddev->in_sync = 1; | |
3441 | del_timer_sync(&mddev->safemode_timer); | |
3442 | } | |
02e5f5c0 | 3443 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3444 | pers->run(mddev); |
245f46c2 | 3445 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
47525e59 | 3446 | mddev_resume(mddev); |
830778a1 N |
3447 | if (!mddev->thread) |
3448 | md_update_sb(mddev, 1); | |
5cac7861 | 3449 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3450 | md_new_event(mddev); |
6791875e N |
3451 | rv = len; |
3452 | out_unlock: | |
3453 | mddev_unlock(mddev); | |
d9d166c2 N |
3454 | return rv; |
3455 | } | |
3456 | ||
3457 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3458 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3459 | |
d4dbd025 | 3460 | static ssize_t |
fd01b88c | 3461 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3462 | { |
3463 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3464 | if (mddev->reshape_position != MaxSector && |
3465 | mddev->layout != mddev->new_layout) | |
3466 | return sprintf(page, "%d (%d)\n", | |
3467 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3468 | return sprintf(page, "%d\n", mddev->layout); |
3469 | } | |
3470 | ||
3471 | static ssize_t | |
fd01b88c | 3472 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 N |
3473 | { |
3474 | char *e; | |
3475 | unsigned long n = simple_strtoul(buf, &e, 10); | |
6791875e | 3476 | int err; |
d4dbd025 N |
3477 | |
3478 | if (!*buf || (*e && *e != '\n')) | |
3479 | return -EINVAL; | |
6791875e N |
3480 | err = mddev_lock(mddev); |
3481 | if (err) | |
3482 | return err; | |
d4dbd025 | 3483 | |
b3546035 | 3484 | if (mddev->pers) { |
50ac168a | 3485 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3486 | err = -EBUSY; |
3487 | else if (mddev->ro) | |
3488 | err = -EROFS; | |
3489 | else { | |
3490 | mddev->new_layout = n; | |
3491 | err = mddev->pers->check_reshape(mddev); | |
3492 | if (err) | |
3493 | mddev->new_layout = mddev->layout; | |
597a711b | 3494 | } |
b3546035 | 3495 | } else { |
08a02ecd | 3496 | mddev->new_layout = n; |
b3546035 N |
3497 | if (mddev->reshape_position == MaxSector) |
3498 | mddev->layout = n; | |
3499 | } | |
6791875e N |
3500 | mddev_unlock(mddev); |
3501 | return err ?: len; | |
d4dbd025 N |
3502 | } |
3503 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3504 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3505 | |
eae1701f | 3506 | static ssize_t |
fd01b88c | 3507 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3508 | { |
bb636547 N |
3509 | if (mddev->raid_disks == 0) |
3510 | return 0; | |
08a02ecd N |
3511 | if (mddev->reshape_position != MaxSector && |
3512 | mddev->delta_disks != 0) | |
3513 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3514 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3515 | return sprintf(page, "%d\n", mddev->raid_disks); |
3516 | } | |
3517 | ||
fd01b88c | 3518 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3519 | |
3520 | static ssize_t | |
fd01b88c | 3521 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3522 | { |
da943b99 | 3523 | char *e; |
6791875e | 3524 | int err; |
da943b99 N |
3525 | unsigned long n = simple_strtoul(buf, &e, 10); |
3526 | ||
3527 | if (!*buf || (*e && *e != '\n')) | |
3528 | return -EINVAL; | |
3529 | ||
6791875e N |
3530 | err = mddev_lock(mddev); |
3531 | if (err) | |
3532 | return err; | |
da943b99 | 3533 | if (mddev->pers) |
6791875e | 3534 | err = update_raid_disks(mddev, n); |
08a02ecd | 3535 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3536 | struct md_rdev *rdev; |
08a02ecd | 3537 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3538 | |
6791875e | 3539 | err = -EINVAL; |
c6563a8c N |
3540 | rdev_for_each(rdev, mddev) { |
3541 | if (olddisks < n && | |
3542 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3543 | goto out_unlock; |
c6563a8c N |
3544 | if (olddisks > n && |
3545 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3546 | goto out_unlock; |
c6563a8c | 3547 | } |
6791875e | 3548 | err = 0; |
08a02ecd N |
3549 | mddev->delta_disks = n - olddisks; |
3550 | mddev->raid_disks = n; | |
2c810cdd | 3551 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3552 | } else |
da943b99 | 3553 | mddev->raid_disks = n; |
6791875e N |
3554 | out_unlock: |
3555 | mddev_unlock(mddev); | |
3556 | return err ? err : len; | |
da943b99 N |
3557 | } |
3558 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3559 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3560 | |
3b34380a | 3561 | static ssize_t |
fd01b88c | 3562 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3563 | { |
08a02ecd | 3564 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3565 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3566 | return sprintf(page, "%d (%d)\n", | |
3567 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3568 | mddev->chunk_sectors << 9); |
3569 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3570 | } |
3571 | ||
3572 | static ssize_t | |
fd01b88c | 3573 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3574 | { |
6791875e | 3575 | int err; |
3b34380a N |
3576 | char *e; |
3577 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3578 | ||
3b34380a N |
3579 | if (!*buf || (*e && *e != '\n')) |
3580 | return -EINVAL; | |
3581 | ||
6791875e N |
3582 | err = mddev_lock(mddev); |
3583 | if (err) | |
3584 | return err; | |
b3546035 | 3585 | if (mddev->pers) { |
50ac168a | 3586 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3587 | err = -EBUSY; |
3588 | else if (mddev->ro) | |
3589 | err = -EROFS; | |
3590 | else { | |
3591 | mddev->new_chunk_sectors = n >> 9; | |
3592 | err = mddev->pers->check_reshape(mddev); | |
3593 | if (err) | |
3594 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 3595 | } |
b3546035 | 3596 | } else { |
664e7c41 | 3597 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3598 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3599 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3600 | } |
6791875e N |
3601 | mddev_unlock(mddev); |
3602 | return err ?: len; | |
3b34380a N |
3603 | } |
3604 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3605 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3606 | |
a94213b1 | 3607 | static ssize_t |
fd01b88c | 3608 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3609 | { |
d1a7c503 N |
3610 | if (mddev->recovery_cp == MaxSector) |
3611 | return sprintf(page, "none\n"); | |
a94213b1 N |
3612 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3613 | } | |
3614 | ||
3615 | static ssize_t | |
fd01b88c | 3616 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3617 | { |
6791875e | 3618 | int err; |
a94213b1 N |
3619 | char *e; |
3620 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3621 | ||
6791875e N |
3622 | err = mddev_lock(mddev); |
3623 | if (err) | |
3624 | return err; | |
b098636c | 3625 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e N |
3626 | err = -EBUSY; |
3627 | else if (cmd_match(buf, "none")) | |
06e3c817 DW |
3628 | n = MaxSector; |
3629 | else if (!*buf || (*e && *e != '\n')) | |
6791875e | 3630 | err = -EINVAL; |
a94213b1 | 3631 | |
6791875e N |
3632 | if (!err) { |
3633 | mddev->recovery_cp = n; | |
3634 | if (mddev->pers) | |
3635 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
3636 | } | |
3637 | mddev_unlock(mddev); | |
3638 | return err ?: len; | |
a94213b1 N |
3639 | } |
3640 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3641 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3642 | |
9e653b63 N |
3643 | /* |
3644 | * The array state can be: | |
3645 | * | |
3646 | * clear | |
3647 | * No devices, no size, no level | |
3648 | * Equivalent to STOP_ARRAY ioctl | |
3649 | * inactive | |
3650 | * May have some settings, but array is not active | |
3651 | * all IO results in error | |
3652 | * When written, doesn't tear down array, but just stops it | |
3653 | * suspended (not supported yet) | |
3654 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3655 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3656 | * readonly |
3657 | * no resync can happen. no superblocks get written. | |
3658 | * write requests fail | |
3659 | * read-auto | |
3660 | * like readonly, but behaves like 'clean' on a write request. | |
3661 | * | |
3662 | * clean - no pending writes, but otherwise active. | |
3663 | * When written to inactive array, starts without resync | |
3664 | * If a write request arrives then | |
3665 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3666 | * if not known, block and switch to write-pending | |
3667 | * If written to an active array that has pending writes, then fails. | |
3668 | * active | |
3669 | * fully active: IO and resync can be happening. | |
3670 | * When written to inactive array, starts with resync | |
3671 | * | |
3672 | * write-pending | |
3673 | * clean, but writes are blocked waiting for 'active' to be written. | |
3674 | * | |
3675 | * active-idle | |
3676 | * like active, but no writes have been seen for a while (100msec). | |
3677 | * | |
3678 | */ | |
3679 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3680 | write_pending, active_idle, bad_word}; | |
05381954 | 3681 | static char *array_states[] = { |
9e653b63 N |
3682 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3683 | "write-pending", "active-idle", NULL }; | |
3684 | ||
3685 | static int match_word(const char *word, char **list) | |
3686 | { | |
3687 | int n; | |
3688 | for (n=0; list[n]; n++) | |
3689 | if (cmd_match(word, list[n])) | |
3690 | break; | |
3691 | return n; | |
3692 | } | |
3693 | ||
3694 | static ssize_t | |
fd01b88c | 3695 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3696 | { |
3697 | enum array_state st = inactive; | |
3698 | ||
3699 | if (mddev->pers) | |
3700 | switch(mddev->ro) { | |
3701 | case 1: | |
3702 | st = readonly; | |
3703 | break; | |
3704 | case 2: | |
3705 | st = read_auto; | |
3706 | break; | |
3707 | case 0: | |
3708 | if (mddev->in_sync) | |
3709 | st = clean; | |
070dc6dd | 3710 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3711 | st = write_pending; |
9e653b63 N |
3712 | else if (mddev->safemode) |
3713 | st = active_idle; | |
3714 | else | |
3715 | st = active; | |
3716 | } | |
3717 | else { | |
3718 | if (list_empty(&mddev->disks) && | |
3719 | mddev->raid_disks == 0 && | |
58c0fed4 | 3720 | mddev->dev_sectors == 0) |
9e653b63 N |
3721 | st = clear; |
3722 | else | |
3723 | st = inactive; | |
3724 | } | |
3725 | return sprintf(page, "%s\n", array_states[st]); | |
3726 | } | |
3727 | ||
f72ffdd6 N |
3728 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
3729 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
3730 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 3731 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
3732 | |
3733 | static ssize_t | |
fd01b88c | 3734 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 3735 | { |
6791875e | 3736 | int err; |
9e653b63 | 3737 | enum array_state st = match_word(buf, array_states); |
6791875e N |
3738 | |
3739 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
3740 | /* don't take reconfig_mutex when toggling between | |
3741 | * clean and active | |
3742 | */ | |
3743 | spin_lock(&mddev->lock); | |
3744 | if (st == active) { | |
3745 | restart_array(mddev); | |
3746 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
3747 | wake_up(&mddev->sb_wait); | |
3748 | err = 0; | |
3749 | } else /* st == clean */ { | |
3750 | restart_array(mddev); | |
3751 | if (atomic_read(&mddev->writes_pending) == 0) { | |
3752 | if (mddev->in_sync == 0) { | |
3753 | mddev->in_sync = 1; | |
3754 | if (mddev->safemode == 1) | |
3755 | mddev->safemode = 0; | |
3756 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
3757 | } | |
3758 | err = 0; | |
3759 | } else | |
3760 | err = -EBUSY; | |
3761 | } | |
3762 | spin_unlock(&mddev->lock); | |
3763 | return err; | |
3764 | } | |
3765 | err = mddev_lock(mddev); | |
3766 | if (err) | |
3767 | return err; | |
3768 | err = -EINVAL; | |
9e653b63 N |
3769 | switch(st) { |
3770 | case bad_word: | |
3771 | break; | |
3772 | case clear: | |
3773 | /* stopping an active array */ | |
a05b7ea0 | 3774 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3775 | break; |
3776 | case inactive: | |
3777 | /* stopping an active array */ | |
90cf195d | 3778 | if (mddev->pers) |
a05b7ea0 | 3779 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3780 | else |
e691063a | 3781 | err = 0; /* already inactive */ |
9e653b63 N |
3782 | break; |
3783 | case suspended: | |
3784 | break; /* not supported yet */ | |
3785 | case readonly: | |
3786 | if (mddev->pers) | |
a05b7ea0 | 3787 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3788 | else { |
3789 | mddev->ro = 1; | |
648b629e | 3790 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3791 | err = do_md_run(mddev); |
3792 | } | |
3793 | break; | |
3794 | case read_auto: | |
9e653b63 | 3795 | if (mddev->pers) { |
80268ee9 | 3796 | if (mddev->ro == 0) |
a05b7ea0 | 3797 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3798 | else if (mddev->ro == 1) |
648b629e N |
3799 | err = restart_array(mddev); |
3800 | if (err == 0) { | |
3801 | mddev->ro = 2; | |
3802 | set_disk_ro(mddev->gendisk, 0); | |
3803 | } | |
9e653b63 N |
3804 | } else { |
3805 | mddev->ro = 2; | |
3806 | err = do_md_run(mddev); | |
3807 | } | |
3808 | break; | |
3809 | case clean: | |
3810 | if (mddev->pers) { | |
3811 | restart_array(mddev); | |
85572d7c | 3812 | spin_lock(&mddev->lock); |
9e653b63 | 3813 | if (atomic_read(&mddev->writes_pending) == 0) { |
e691063a N |
3814 | if (mddev->in_sync == 0) { |
3815 | mddev->in_sync = 1; | |
31a59e34 N |
3816 | if (mddev->safemode == 1) |
3817 | mddev->safemode = 0; | |
070dc6dd | 3818 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3819 | } |
3820 | err = 0; | |
3821 | } else | |
3822 | err = -EBUSY; | |
85572d7c | 3823 | spin_unlock(&mddev->lock); |
5bf29597 N |
3824 | } else |
3825 | err = -EINVAL; | |
9e653b63 N |
3826 | break; |
3827 | case active: | |
3828 | if (mddev->pers) { | |
3829 | restart_array(mddev); | |
070dc6dd | 3830 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3831 | wake_up(&mddev->sb_wait); |
3832 | err = 0; | |
3833 | } else { | |
3834 | mddev->ro = 0; | |
648b629e | 3835 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3836 | err = do_md_run(mddev); |
3837 | } | |
3838 | break; | |
3839 | case write_pending: | |
3840 | case active_idle: | |
3841 | /* these cannot be set */ | |
3842 | break; | |
3843 | } | |
6791875e N |
3844 | |
3845 | if (!err) { | |
1d23f178 N |
3846 | if (mddev->hold_active == UNTIL_IOCTL) |
3847 | mddev->hold_active = 0; | |
00bcb4ac | 3848 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 3849 | } |
6791875e N |
3850 | mddev_unlock(mddev); |
3851 | return err ?: len; | |
9e653b63 | 3852 | } |
80ca3a44 N |
3853 | static struct md_sysfs_entry md_array_state = |
3854 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3855 | |
1e50915f | 3856 | static ssize_t |
fd01b88c | 3857 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
3858 | return sprintf(page, "%d\n", |
3859 | atomic_read(&mddev->max_corr_read_errors)); | |
3860 | } | |
3861 | ||
3862 | static ssize_t | |
fd01b88c | 3863 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f RB |
3864 | { |
3865 | char *e; | |
3866 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3867 | ||
3868 | if (*buf && (*e == 0 || *e == '\n')) { | |
3869 | atomic_set(&mddev->max_corr_read_errors, n); | |
3870 | return len; | |
3871 | } | |
3872 | return -EINVAL; | |
3873 | } | |
3874 | ||
3875 | static struct md_sysfs_entry max_corr_read_errors = | |
3876 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3877 | max_corrected_read_errors_store); | |
3878 | ||
6d7ff738 | 3879 | static ssize_t |
fd01b88c | 3880 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
3881 | { |
3882 | return -EINVAL; | |
3883 | } | |
3884 | ||
3885 | static ssize_t | |
fd01b88c | 3886 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
3887 | { |
3888 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3889 | /* The new device is added to the array. | |
3890 | * If the array has a persistent superblock, we read the | |
3891 | * superblock to initialise info and check validity. | |
3892 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3893 | * which mainly checks size. | |
3894 | */ | |
3895 | char *e; | |
3896 | int major = simple_strtoul(buf, &e, 10); | |
3897 | int minor; | |
3898 | dev_t dev; | |
3cb03002 | 3899 | struct md_rdev *rdev; |
6d7ff738 N |
3900 | int err; |
3901 | ||
3902 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3903 | return -EINVAL; | |
3904 | minor = simple_strtoul(e+1, &e, 10); | |
3905 | if (*e && *e != '\n') | |
3906 | return -EINVAL; | |
3907 | dev = MKDEV(major, minor); | |
3908 | if (major != MAJOR(dev) || | |
3909 | minor != MINOR(dev)) | |
3910 | return -EOVERFLOW; | |
3911 | ||
6791875e N |
3912 | flush_workqueue(md_misc_wq); |
3913 | ||
3914 | err = mddev_lock(mddev); | |
3915 | if (err) | |
3916 | return err; | |
6d7ff738 N |
3917 | if (mddev->persistent) { |
3918 | rdev = md_import_device(dev, mddev->major_version, | |
3919 | mddev->minor_version); | |
3920 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
3921 | struct md_rdev *rdev0 |
3922 | = list_entry(mddev->disks.next, | |
3923 | struct md_rdev, same_set); | |
6d7ff738 N |
3924 | err = super_types[mddev->major_version] |
3925 | .load_super(rdev, rdev0, mddev->minor_version); | |
3926 | if (err < 0) | |
3927 | goto out; | |
3928 | } | |
c5d79adb N |
3929 | } else if (mddev->external) |
3930 | rdev = md_import_device(dev, -2, -1); | |
3931 | else | |
6d7ff738 N |
3932 | rdev = md_import_device(dev, -1, -1); |
3933 | ||
3934 | if (IS_ERR(rdev)) | |
3935 | return PTR_ERR(rdev); | |
3936 | err = bind_rdev_to_array(rdev, mddev); | |
3937 | out: | |
3938 | if (err) | |
3939 | export_rdev(rdev); | |
6791875e | 3940 | mddev_unlock(mddev); |
6d7ff738 N |
3941 | return err ? err : len; |
3942 | } | |
3943 | ||
3944 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3945 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3946 | |
9b1d1dac | 3947 | static ssize_t |
fd01b88c | 3948 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
3949 | { |
3950 | char *end; | |
3951 | unsigned long chunk, end_chunk; | |
6791875e | 3952 | int err; |
9b1d1dac | 3953 | |
6791875e N |
3954 | err = mddev_lock(mddev); |
3955 | if (err) | |
3956 | return err; | |
9b1d1dac PC |
3957 | if (!mddev->bitmap) |
3958 | goto out; | |
3959 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3960 | while (*buf) { | |
3961 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3962 | if (buf == end) break; | |
3963 | if (*end == '-') { /* range */ | |
3964 | buf = end + 1; | |
3965 | end_chunk = simple_strtoul(buf, &end, 0); | |
3966 | if (buf == end) break; | |
3967 | } | |
3968 | if (*end && !isspace(*end)) break; | |
3969 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3970 | buf = skip_spaces(end); |
9b1d1dac PC |
3971 | } |
3972 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3973 | out: | |
6791875e | 3974 | mddev_unlock(mddev); |
9b1d1dac PC |
3975 | return len; |
3976 | } | |
3977 | ||
3978 | static struct md_sysfs_entry md_bitmap = | |
3979 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3980 | ||
a35b0d69 | 3981 | static ssize_t |
fd01b88c | 3982 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 3983 | { |
58c0fed4 AN |
3984 | return sprintf(page, "%llu\n", |
3985 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3986 | } |
3987 | ||
fd01b88c | 3988 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
3989 | |
3990 | static ssize_t | |
fd01b88c | 3991 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
3992 | { |
3993 | /* If array is inactive, we can reduce the component size, but | |
3994 | * not increase it (except from 0). | |
3995 | * If array is active, we can try an on-line resize | |
3996 | */ | |
b522adcd DW |
3997 | sector_t sectors; |
3998 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3999 | |
58c0fed4 AN |
4000 | if (err < 0) |
4001 | return err; | |
6791875e N |
4002 | err = mddev_lock(mddev); |
4003 | if (err) | |
4004 | return err; | |
a35b0d69 | 4005 | if (mddev->pers) { |
58c0fed4 | 4006 | err = update_size(mddev, sectors); |
850b2b42 | 4007 | md_update_sb(mddev, 1); |
a35b0d69 | 4008 | } else { |
58c0fed4 AN |
4009 | if (mddev->dev_sectors == 0 || |
4010 | mddev->dev_sectors > sectors) | |
4011 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4012 | else |
4013 | err = -ENOSPC; | |
4014 | } | |
6791875e | 4015 | mddev_unlock(mddev); |
a35b0d69 N |
4016 | return err ? err : len; |
4017 | } | |
4018 | ||
4019 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4020 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4021 | |
83f0d77a | 4022 | /* Metadata version. |
e691063a N |
4023 | * This is one of |
4024 | * 'none' for arrays with no metadata (good luck...) | |
4025 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4026 | * or N.M for internally known formats |
4027 | */ | |
4028 | static ssize_t | |
fd01b88c | 4029 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4030 | { |
4031 | if (mddev->persistent) | |
4032 | return sprintf(page, "%d.%d\n", | |
4033 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4034 | else if (mddev->external) |
4035 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4036 | else |
4037 | return sprintf(page, "none\n"); | |
4038 | } | |
4039 | ||
4040 | static ssize_t | |
fd01b88c | 4041 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4042 | { |
4043 | int major, minor; | |
4044 | char *e; | |
6791875e | 4045 | int err; |
ea43ddd8 N |
4046 | /* Changing the details of 'external' metadata is |
4047 | * always permitted. Otherwise there must be | |
4048 | * no devices attached to the array. | |
4049 | */ | |
6791875e N |
4050 | |
4051 | err = mddev_lock(mddev); | |
4052 | if (err) | |
4053 | return err; | |
4054 | err = -EBUSY; | |
ea43ddd8 N |
4055 | if (mddev->external && strncmp(buf, "external:", 9) == 0) |
4056 | ; | |
4057 | else if (!list_empty(&mddev->disks)) | |
6791875e | 4058 | goto out_unlock; |
8bb93aac | 4059 | |
6791875e | 4060 | err = 0; |
8bb93aac N |
4061 | if (cmd_match(buf, "none")) { |
4062 | mddev->persistent = 0; | |
e691063a N |
4063 | mddev->external = 0; |
4064 | mddev->major_version = 0; | |
4065 | mddev->minor_version = 90; | |
6791875e | 4066 | goto out_unlock; |
e691063a N |
4067 | } |
4068 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4069 | size_t namelen = len-9; |
e691063a N |
4070 | if (namelen >= sizeof(mddev->metadata_type)) |
4071 | namelen = sizeof(mddev->metadata_type)-1; | |
4072 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4073 | mddev->metadata_type[namelen] = 0; | |
4074 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4075 | mddev->metadata_type[--namelen] = 0; | |
4076 | mddev->persistent = 0; | |
4077 | mddev->external = 1; | |
8bb93aac N |
4078 | mddev->major_version = 0; |
4079 | mddev->minor_version = 90; | |
6791875e | 4080 | goto out_unlock; |
8bb93aac N |
4081 | } |
4082 | major = simple_strtoul(buf, &e, 10); | |
6791875e | 4083 | err = -EINVAL; |
8bb93aac | 4084 | if (e==buf || *e != '.') |
6791875e | 4085 | goto out_unlock; |
8bb93aac N |
4086 | buf = e+1; |
4087 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4088 | if (e==buf || (*e && *e != '\n') ) |
6791875e N |
4089 | goto out_unlock; |
4090 | err = -ENOENT; | |
50511da3 | 4091 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
6791875e | 4092 | goto out_unlock; |
8bb93aac N |
4093 | mddev->major_version = major; |
4094 | mddev->minor_version = minor; | |
4095 | mddev->persistent = 1; | |
e691063a | 4096 | mddev->external = 0; |
6791875e N |
4097 | err = 0; |
4098 | out_unlock: | |
4099 | mddev_unlock(mddev); | |
4100 | return err ?: len; | |
8bb93aac N |
4101 | } |
4102 | ||
4103 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 4104 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4105 | |
24dd469d | 4106 | static ssize_t |
fd01b88c | 4107 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4108 | { |
7eec314d | 4109 | char *type = "idle"; |
b7b17c9b N |
4110 | unsigned long recovery = mddev->recovery; |
4111 | if (test_bit(MD_RECOVERY_FROZEN, &recovery)) | |
b6a9ce68 | 4112 | type = "frozen"; |
b7b17c9b N |
4113 | else if (test_bit(MD_RECOVERY_RUNNING, &recovery) || |
4114 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) { | |
4115 | if (test_bit(MD_RECOVERY_RESHAPE, &recovery)) | |
ccfcc3c1 | 4116 | type = "reshape"; |
b7b17c9b N |
4117 | else if (test_bit(MD_RECOVERY_SYNC, &recovery)) { |
4118 | if (!test_bit(MD_RECOVERY_REQUESTED, &recovery)) | |
24dd469d | 4119 | type = "resync"; |
b7b17c9b | 4120 | else if (test_bit(MD_RECOVERY_CHECK, &recovery)) |
24dd469d N |
4121 | type = "check"; |
4122 | else | |
4123 | type = "repair"; | |
b7b17c9b | 4124 | } else if (test_bit(MD_RECOVERY_RECOVER, &recovery)) |
24dd469d N |
4125 | type = "recover"; |
4126 | } | |
4127 | return sprintf(page, "%s\n", type); | |
4128 | } | |
4129 | ||
4130 | static ssize_t | |
fd01b88c | 4131 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4132 | { |
7eec314d N |
4133 | if (!mddev->pers || !mddev->pers->sync_request) |
4134 | return -EINVAL; | |
4135 | ||
b6a9ce68 N |
4136 | if (cmd_match(page, "frozen")) |
4137 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4138 | else | |
4139 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4140 | ||
4141 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
f851b60d | 4142 | flush_workqueue(md_misc_wq); |
7eec314d N |
4143 | if (mddev->sync_thread) { |
4144 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e N |
4145 | if (mddev_lock(mddev) == 0) { |
4146 | md_reap_sync_thread(mddev); | |
4147 | mddev_unlock(mddev); | |
4148 | } | |
7eec314d | 4149 | } |
03c902e1 N |
4150 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
4151 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 4152 | return -EBUSY; |
72a23c21 NB |
4153 | else if (cmd_match(page, "resync")) |
4154 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4155 | else if (cmd_match(page, "recover")) { | |
4156 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 4157 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 4158 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4159 | int err; |
4160 | if (mddev->pers->start_reshape == NULL) | |
4161 | return -EINVAL; | |
6791875e N |
4162 | err = mddev_lock(mddev); |
4163 | if (!err) { | |
4164 | err = mddev->pers->start_reshape(mddev); | |
4165 | mddev_unlock(mddev); | |
4166 | } | |
16484bf5 N |
4167 | if (err) |
4168 | return err; | |
a99ac971 | 4169 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4170 | } else { |
bce74dac | 4171 | if (cmd_match(page, "check")) |
7eec314d | 4172 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4173 | else if (!cmd_match(page, "repair")) |
7eec314d N |
4174 | return -EINVAL; |
4175 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
4176 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4177 | } |
48c26ddc N |
4178 | if (mddev->ro == 2) { |
4179 | /* A write to sync_action is enough to justify | |
4180 | * canceling read-auto mode | |
4181 | */ | |
4182 | mddev->ro = 0; | |
4183 | md_wakeup_thread(mddev->sync_thread); | |
4184 | } | |
03c902e1 | 4185 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4186 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4187 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4188 | return len; |
4189 | } | |
4190 | ||
c4a39551 JB |
4191 | static struct md_sysfs_entry md_scan_mode = |
4192 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
4193 | ||
4194 | static ssize_t | |
4195 | last_sync_action_show(struct mddev *mddev, char *page) | |
4196 | { | |
4197 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4198 | } | |
4199 | ||
4200 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4201 | ||
9d88883e | 4202 | static ssize_t |
fd01b88c | 4203 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4204 | { |
4205 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4206 | (unsigned long long) |
4207 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4208 | } |
4209 | ||
80ca3a44 | 4210 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4211 | |
88202a0c | 4212 | static ssize_t |
fd01b88c | 4213 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4214 | { |
4215 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4216 | mddev->sync_speed_min ? "local": "system"); | |
4217 | } | |
4218 | ||
4219 | static ssize_t | |
fd01b88c | 4220 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4221 | { |
4222 | int min; | |
4223 | char *e; | |
4224 | if (strncmp(buf, "system", 6)==0) { | |
4225 | mddev->sync_speed_min = 0; | |
4226 | return len; | |
4227 | } | |
4228 | min = simple_strtoul(buf, &e, 10); | |
4229 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
4230 | return -EINVAL; | |
4231 | mddev->sync_speed_min = min; | |
4232 | return len; | |
4233 | } | |
4234 | ||
4235 | static struct md_sysfs_entry md_sync_min = | |
4236 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4237 | ||
4238 | static ssize_t | |
fd01b88c | 4239 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4240 | { |
4241 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4242 | mddev->sync_speed_max ? "local": "system"); | |
4243 | } | |
4244 | ||
4245 | static ssize_t | |
fd01b88c | 4246 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4247 | { |
4248 | int max; | |
4249 | char *e; | |
4250 | if (strncmp(buf, "system", 6)==0) { | |
4251 | mddev->sync_speed_max = 0; | |
4252 | return len; | |
4253 | } | |
4254 | max = simple_strtoul(buf, &e, 10); | |
4255 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
4256 | return -EINVAL; | |
4257 | mddev->sync_speed_max = max; | |
4258 | return len; | |
4259 | } | |
4260 | ||
4261 | static struct md_sysfs_entry md_sync_max = | |
4262 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4263 | ||
d7f3d291 | 4264 | static ssize_t |
fd01b88c | 4265 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4266 | { |
4267 | return sprintf(page, "%d\n", mddev->degraded); | |
4268 | } | |
4269 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4270 | |
90b08710 | 4271 | static ssize_t |
fd01b88c | 4272 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4273 | { |
4274 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4275 | } | |
4276 | ||
4277 | static ssize_t | |
fd01b88c | 4278 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4279 | { |
4280 | long n; | |
4281 | ||
b29bebd6 | 4282 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4283 | return -EINVAL; |
4284 | ||
4285 | if (n != 0 && n != 1) | |
4286 | return -EINVAL; | |
4287 | ||
4288 | mddev->parallel_resync = n; | |
4289 | ||
4290 | if (mddev->sync_thread) | |
4291 | wake_up(&resync_wait); | |
4292 | ||
4293 | return len; | |
4294 | } | |
4295 | ||
4296 | /* force parallel resync, even with shared block devices */ | |
4297 | static struct md_sysfs_entry md_sync_force_parallel = | |
4298 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4299 | sync_force_parallel_show, sync_force_parallel_store); | |
4300 | ||
88202a0c | 4301 | static ssize_t |
fd01b88c | 4302 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4303 | { |
4304 | unsigned long resync, dt, db; | |
d1a7c503 N |
4305 | if (mddev->curr_resync == 0) |
4306 | return sprintf(page, "none\n"); | |
9687a60c AN |
4307 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4308 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4309 | if (!dt) dt++; |
9687a60c AN |
4310 | db = resync - mddev->resync_mark_cnt; |
4311 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4312 | } |
4313 | ||
80ca3a44 | 4314 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4315 | |
4316 | static ssize_t | |
fd01b88c | 4317 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4318 | { |
13ae864b | 4319 | unsigned long long max_sectors, resync; |
88202a0c | 4320 | |
acb180b0 N |
4321 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4322 | return sprintf(page, "none\n"); | |
4323 | ||
72f36d59 N |
4324 | if (mddev->curr_resync == 1 || |
4325 | mddev->curr_resync == 2) | |
4326 | return sprintf(page, "delayed\n"); | |
4327 | ||
c804cdec N |
4328 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4329 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4330 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4331 | else |
58c0fed4 | 4332 | max_sectors = mddev->dev_sectors; |
88202a0c | 4333 | |
acb180b0 | 4334 | resync = mddev->curr_resync_completed; |
13ae864b | 4335 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4336 | } |
4337 | ||
80ca3a44 | 4338 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 4339 | |
5e96ee65 | 4340 | static ssize_t |
fd01b88c | 4341 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4342 | { |
4343 | return sprintf(page, "%llu\n", | |
4344 | (unsigned long long)mddev->resync_min); | |
4345 | } | |
4346 | static ssize_t | |
fd01b88c | 4347 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4348 | { |
4349 | unsigned long long min; | |
23da422b N |
4350 | int err; |
4351 | int chunk; | |
4352 | ||
b29bebd6 | 4353 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4354 | return -EINVAL; |
23da422b N |
4355 | |
4356 | spin_lock(&mddev->lock); | |
4357 | err = -EINVAL; | |
5e96ee65 | 4358 | if (min > mddev->resync_max) |
23da422b N |
4359 | goto out_unlock; |
4360 | ||
4361 | err = -EBUSY; | |
5e96ee65 | 4362 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4363 | goto out_unlock; |
5e96ee65 NB |
4364 | |
4365 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4366 | chunk = mddev->chunk_sectors; |
4367 | if (chunk) { | |
2ac06c33 | 4368 | sector_t temp = min; |
23da422b N |
4369 | |
4370 | err = -EINVAL; | |
4371 | if (sector_div(temp, chunk)) | |
4372 | goto out_unlock; | |
5e96ee65 NB |
4373 | } |
4374 | mddev->resync_min = min; | |
23da422b | 4375 | err = 0; |
5e96ee65 | 4376 | |
23da422b N |
4377 | out_unlock: |
4378 | spin_unlock(&mddev->lock); | |
4379 | return err ?: len; | |
5e96ee65 NB |
4380 | } |
4381 | ||
4382 | static struct md_sysfs_entry md_min_sync = | |
4383 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4384 | ||
c6207277 | 4385 | static ssize_t |
fd01b88c | 4386 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4387 | { |
4388 | if (mddev->resync_max == MaxSector) | |
4389 | return sprintf(page, "max\n"); | |
4390 | else | |
4391 | return sprintf(page, "%llu\n", | |
4392 | (unsigned long long)mddev->resync_max); | |
4393 | } | |
4394 | static ssize_t | |
fd01b88c | 4395 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4396 | { |
23da422b N |
4397 | int err; |
4398 | spin_lock(&mddev->lock); | |
c6207277 N |
4399 | if (strncmp(buf, "max", 3) == 0) |
4400 | mddev->resync_max = MaxSector; | |
4401 | else { | |
5e96ee65 | 4402 | unsigned long long max; |
23da422b N |
4403 | int chunk; |
4404 | ||
4405 | err = -EINVAL; | |
b29bebd6 | 4406 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4407 | goto out_unlock; |
5e96ee65 | 4408 | if (max < mddev->resync_min) |
23da422b N |
4409 | goto out_unlock; |
4410 | ||
4411 | err = -EBUSY; | |
c6207277 | 4412 | if (max < mddev->resync_max && |
4d484a4a | 4413 | mddev->ro == 0 && |
c6207277 | 4414 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4415 | goto out_unlock; |
c6207277 N |
4416 | |
4417 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4418 | chunk = mddev->chunk_sectors; |
4419 | if (chunk) { | |
2ac06c33 | 4420 | sector_t temp = max; |
23da422b N |
4421 | |
4422 | err = -EINVAL; | |
4423 | if (sector_div(temp, chunk)) | |
4424 | goto out_unlock; | |
c6207277 N |
4425 | } |
4426 | mddev->resync_max = max; | |
4427 | } | |
4428 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4429 | err = 0; |
4430 | out_unlock: | |
4431 | spin_unlock(&mddev->lock); | |
4432 | return err ?: len; | |
c6207277 N |
4433 | } |
4434 | ||
4435 | static struct md_sysfs_entry md_max_sync = | |
4436 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4437 | ||
e464eafd | 4438 | static ssize_t |
fd01b88c | 4439 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4440 | { |
4441 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4442 | } | |
4443 | ||
4444 | static ssize_t | |
fd01b88c | 4445 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4446 | { |
4447 | char *e; | |
4448 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
6791875e N |
4449 | unsigned long long old; |
4450 | int err; | |
e464eafd | 4451 | |
e464eafd N |
4452 | if (buf == e || (*e && *e != '\n')) |
4453 | return -EINVAL; | |
23ddff37 | 4454 | |
6791875e N |
4455 | err = mddev_lock(mddev); |
4456 | if (err) | |
4457 | return err; | |
4458 | err = -EINVAL; | |
4459 | if (mddev->pers == NULL || | |
4460 | mddev->pers->quiesce == NULL) | |
4461 | goto unlock; | |
4462 | old = mddev->suspend_lo; | |
23ddff37 N |
4463 | mddev->suspend_lo = new; |
4464 | if (new >= old) | |
4465 | /* Shrinking suspended region */ | |
e464eafd | 4466 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4467 | else { |
4468 | /* Expanding suspended region - need to wait */ | |
4469 | mddev->pers->quiesce(mddev, 1); | |
4470 | mddev->pers->quiesce(mddev, 0); | |
4471 | } | |
6791875e N |
4472 | err = 0; |
4473 | unlock: | |
4474 | mddev_unlock(mddev); | |
4475 | return err ?: len; | |
e464eafd N |
4476 | } |
4477 | static struct md_sysfs_entry md_suspend_lo = | |
4478 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4479 | ||
e464eafd | 4480 | static ssize_t |
fd01b88c | 4481 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4482 | { |
4483 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4484 | } | |
4485 | ||
4486 | static ssize_t | |
fd01b88c | 4487 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4488 | { |
4489 | char *e; | |
4490 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
6791875e N |
4491 | unsigned long long old; |
4492 | int err; | |
e464eafd | 4493 | |
e464eafd N |
4494 | if (buf == e || (*e && *e != '\n')) |
4495 | return -EINVAL; | |
23ddff37 | 4496 | |
6791875e N |
4497 | err = mddev_lock(mddev); |
4498 | if (err) | |
4499 | return err; | |
4500 | err = -EINVAL; | |
4501 | if (mddev->pers == NULL || | |
4502 | mddev->pers->quiesce == NULL) | |
4503 | goto unlock; | |
4504 | old = mddev->suspend_hi; | |
23ddff37 N |
4505 | mddev->suspend_hi = new; |
4506 | if (new <= old) | |
4507 | /* Shrinking suspended region */ | |
4508 | mddev->pers->quiesce(mddev, 2); | |
4509 | else { | |
4510 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4511 | mddev->pers->quiesce(mddev, 1); |
4512 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 | 4513 | } |
6791875e N |
4514 | err = 0; |
4515 | unlock: | |
4516 | mddev_unlock(mddev); | |
4517 | return err ?: len; | |
e464eafd N |
4518 | } |
4519 | static struct md_sysfs_entry md_suspend_hi = | |
4520 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4521 | ||
08a02ecd | 4522 | static ssize_t |
fd01b88c | 4523 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4524 | { |
4525 | if (mddev->reshape_position != MaxSector) | |
4526 | return sprintf(page, "%llu\n", | |
4527 | (unsigned long long)mddev->reshape_position); | |
4528 | strcpy(page, "none\n"); | |
4529 | return 5; | |
4530 | } | |
4531 | ||
4532 | static ssize_t | |
fd01b88c | 4533 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4534 | { |
c6563a8c | 4535 | struct md_rdev *rdev; |
08a02ecd | 4536 | char *e; |
6791875e | 4537 | int err; |
08a02ecd | 4538 | unsigned long long new = simple_strtoull(buf, &e, 10); |
6791875e | 4539 | |
08a02ecd N |
4540 | if (buf == e || (*e && *e != '\n')) |
4541 | return -EINVAL; | |
6791875e N |
4542 | err = mddev_lock(mddev); |
4543 | if (err) | |
4544 | return err; | |
4545 | err = -EBUSY; | |
4546 | if (mddev->pers) | |
4547 | goto unlock; | |
08a02ecd N |
4548 | mddev->reshape_position = new; |
4549 | mddev->delta_disks = 0; | |
2c810cdd | 4550 | mddev->reshape_backwards = 0; |
08a02ecd N |
4551 | mddev->new_level = mddev->level; |
4552 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4553 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4554 | rdev_for_each(rdev, mddev) |
4555 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4556 | err = 0; |
4557 | unlock: | |
4558 | mddev_unlock(mddev); | |
4559 | return err ?: len; | |
08a02ecd N |
4560 | } |
4561 | ||
4562 | static struct md_sysfs_entry md_reshape_position = | |
4563 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4564 | reshape_position_store); | |
4565 | ||
2c810cdd N |
4566 | static ssize_t |
4567 | reshape_direction_show(struct mddev *mddev, char *page) | |
4568 | { | |
4569 | return sprintf(page, "%s\n", | |
4570 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4571 | } | |
4572 | ||
4573 | static ssize_t | |
4574 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4575 | { | |
4576 | int backwards = 0; | |
6791875e N |
4577 | int err; |
4578 | ||
2c810cdd N |
4579 | if (cmd_match(buf, "forwards")) |
4580 | backwards = 0; | |
4581 | else if (cmd_match(buf, "backwards")) | |
4582 | backwards = 1; | |
4583 | else | |
4584 | return -EINVAL; | |
4585 | if (mddev->reshape_backwards == backwards) | |
4586 | return len; | |
4587 | ||
6791875e N |
4588 | err = mddev_lock(mddev); |
4589 | if (err) | |
4590 | return err; | |
2c810cdd N |
4591 | /* check if we are allowed to change */ |
4592 | if (mddev->delta_disks) | |
6791875e N |
4593 | err = -EBUSY; |
4594 | else if (mddev->persistent && | |
2c810cdd | 4595 | mddev->major_version == 0) |
6791875e N |
4596 | err = -EINVAL; |
4597 | else | |
4598 | mddev->reshape_backwards = backwards; | |
4599 | mddev_unlock(mddev); | |
4600 | return err ?: len; | |
2c810cdd N |
4601 | } |
4602 | ||
4603 | static struct md_sysfs_entry md_reshape_direction = | |
4604 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4605 | reshape_direction_store); | |
4606 | ||
b522adcd | 4607 | static ssize_t |
fd01b88c | 4608 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4609 | { |
4610 | if (mddev->external_size) | |
4611 | return sprintf(page, "%llu\n", | |
4612 | (unsigned long long)mddev->array_sectors/2); | |
4613 | else | |
4614 | return sprintf(page, "default\n"); | |
4615 | } | |
4616 | ||
4617 | static ssize_t | |
fd01b88c | 4618 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4619 | { |
4620 | sector_t sectors; | |
6791875e N |
4621 | int err; |
4622 | ||
4623 | err = mddev_lock(mddev); | |
4624 | if (err) | |
4625 | return err; | |
b522adcd DW |
4626 | |
4627 | if (strncmp(buf, "default", 7) == 0) { | |
4628 | if (mddev->pers) | |
4629 | sectors = mddev->pers->size(mddev, 0, 0); | |
4630 | else | |
4631 | sectors = mddev->array_sectors; | |
4632 | ||
4633 | mddev->external_size = 0; | |
4634 | } else { | |
4635 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
4636 | err = -EINVAL; |
4637 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
4638 | err = -E2BIG; | |
4639 | else | |
4640 | mddev->external_size = 1; | |
b522adcd DW |
4641 | } |
4642 | ||
6791875e N |
4643 | if (!err) { |
4644 | mddev->array_sectors = sectors; | |
4645 | if (mddev->pers) { | |
4646 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4647 | revalidate_disk(mddev->gendisk); | |
4648 | } | |
cbe6ef1d | 4649 | } |
6791875e N |
4650 | mddev_unlock(mddev); |
4651 | return err ?: len; | |
b522adcd DW |
4652 | } |
4653 | ||
4654 | static struct md_sysfs_entry md_array_size = | |
4655 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4656 | array_size_store); | |
e464eafd | 4657 | |
eae1701f N |
4658 | static struct attribute *md_default_attrs[] = { |
4659 | &md_level.attr, | |
d4dbd025 | 4660 | &md_layout.attr, |
eae1701f | 4661 | &md_raid_disks.attr, |
3b34380a | 4662 | &md_chunk_size.attr, |
a35b0d69 | 4663 | &md_size.attr, |
a94213b1 | 4664 | &md_resync_start.attr, |
8bb93aac | 4665 | &md_metadata.attr, |
6d7ff738 | 4666 | &md_new_device.attr, |
16f17b39 | 4667 | &md_safe_delay.attr, |
9e653b63 | 4668 | &md_array_state.attr, |
08a02ecd | 4669 | &md_reshape_position.attr, |
2c810cdd | 4670 | &md_reshape_direction.attr, |
b522adcd | 4671 | &md_array_size.attr, |
1e50915f | 4672 | &max_corr_read_errors.attr, |
411036fa N |
4673 | NULL, |
4674 | }; | |
4675 | ||
4676 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4677 | &md_scan_mode.attr, |
c4a39551 | 4678 | &md_last_scan_mode.attr, |
9d88883e | 4679 | &md_mismatches.attr, |
88202a0c N |
4680 | &md_sync_min.attr, |
4681 | &md_sync_max.attr, | |
4682 | &md_sync_speed.attr, | |
90b08710 | 4683 | &md_sync_force_parallel.attr, |
88202a0c | 4684 | &md_sync_completed.attr, |
5e96ee65 | 4685 | &md_min_sync.attr, |
c6207277 | 4686 | &md_max_sync.attr, |
e464eafd N |
4687 | &md_suspend_lo.attr, |
4688 | &md_suspend_hi.attr, | |
9b1d1dac | 4689 | &md_bitmap.attr, |
d7f3d291 | 4690 | &md_degraded.attr, |
eae1701f N |
4691 | NULL, |
4692 | }; | |
411036fa N |
4693 | static struct attribute_group md_redundancy_group = { |
4694 | .name = NULL, | |
4695 | .attrs = md_redundancy_attrs, | |
4696 | }; | |
4697 | ||
eae1701f N |
4698 | static ssize_t |
4699 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4700 | { | |
4701 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4702 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4703 | ssize_t rv; |
eae1701f N |
4704 | |
4705 | if (!entry->show) | |
4706 | return -EIO; | |
af8a2434 N |
4707 | spin_lock(&all_mddevs_lock); |
4708 | if (list_empty(&mddev->all_mddevs)) { | |
4709 | spin_unlock(&all_mddevs_lock); | |
4710 | return -EBUSY; | |
4711 | } | |
4712 | mddev_get(mddev); | |
4713 | spin_unlock(&all_mddevs_lock); | |
4714 | ||
b7b17c9b | 4715 | rv = entry->show(mddev, page); |
af8a2434 | 4716 | mddev_put(mddev); |
96de1e66 | 4717 | return rv; |
eae1701f N |
4718 | } |
4719 | ||
4720 | static ssize_t | |
4721 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4722 | const char *page, size_t length) | |
4723 | { | |
4724 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4725 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4726 | ssize_t rv; |
eae1701f N |
4727 | |
4728 | if (!entry->store) | |
4729 | return -EIO; | |
67463acb N |
4730 | if (!capable(CAP_SYS_ADMIN)) |
4731 | return -EACCES; | |
af8a2434 N |
4732 | spin_lock(&all_mddevs_lock); |
4733 | if (list_empty(&mddev->all_mddevs)) { | |
4734 | spin_unlock(&all_mddevs_lock); | |
4735 | return -EBUSY; | |
4736 | } | |
4737 | mddev_get(mddev); | |
4738 | spin_unlock(&all_mddevs_lock); | |
6791875e | 4739 | rv = entry->store(mddev, page, length); |
af8a2434 | 4740 | mddev_put(mddev); |
96de1e66 | 4741 | return rv; |
eae1701f N |
4742 | } |
4743 | ||
4744 | static void md_free(struct kobject *ko) | |
4745 | { | |
fd01b88c | 4746 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4747 | |
4748 | if (mddev->sysfs_state) | |
4749 | sysfs_put(mddev->sysfs_state); | |
4750 | ||
4751 | if (mddev->gendisk) { | |
4752 | del_gendisk(mddev->gendisk); | |
4753 | put_disk(mddev->gendisk); | |
4754 | } | |
4755 | if (mddev->queue) | |
4756 | blk_cleanup_queue(mddev->queue); | |
4757 | ||
eae1701f N |
4758 | kfree(mddev); |
4759 | } | |
4760 | ||
52cf25d0 | 4761 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4762 | .show = md_attr_show, |
4763 | .store = md_attr_store, | |
4764 | }; | |
4765 | static struct kobj_type md_ktype = { | |
4766 | .release = md_free, | |
4767 | .sysfs_ops = &md_sysfs_ops, | |
4768 | .default_attrs = md_default_attrs, | |
4769 | }; | |
4770 | ||
1da177e4 LT |
4771 | int mdp_major = 0; |
4772 | ||
5fd3a17e DW |
4773 | static void mddev_delayed_delete(struct work_struct *ws) |
4774 | { | |
fd01b88c | 4775 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4776 | |
43a70507 | 4777 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4778 | kobject_del(&mddev->kobj); |
4779 | kobject_put(&mddev->kobj); | |
4780 | } | |
4781 | ||
efeb53c0 | 4782 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4783 | { |
48c9c27b | 4784 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 4785 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 4786 | struct gendisk *disk; |
efeb53c0 N |
4787 | int partitioned; |
4788 | int shift; | |
4789 | int unit; | |
3830c62f | 4790 | int error; |
1da177e4 LT |
4791 | |
4792 | if (!mddev) | |
efeb53c0 N |
4793 | return -ENODEV; |
4794 | ||
4795 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4796 | shift = partitioned ? MdpMinorShift : 0; | |
4797 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4798 | |
e804ac78 TH |
4799 | /* wait for any previous instance of this device to be |
4800 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4801 | */ |
e804ac78 | 4802 | flush_workqueue(md_misc_wq); |
d3374825 | 4803 | |
48c9c27b | 4804 | mutex_lock(&disks_mutex); |
0909dc44 N |
4805 | error = -EEXIST; |
4806 | if (mddev->gendisk) | |
4807 | goto abort; | |
efeb53c0 N |
4808 | |
4809 | if (name) { | |
4810 | /* Need to ensure that 'name' is not a duplicate. | |
4811 | */ | |
fd01b88c | 4812 | struct mddev *mddev2; |
efeb53c0 N |
4813 | spin_lock(&all_mddevs_lock); |
4814 | ||
4815 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4816 | if (mddev2->gendisk && | |
4817 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4818 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4819 | goto abort; |
efeb53c0 N |
4820 | } |
4821 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4822 | } |
8b765398 | 4823 | |
0909dc44 | 4824 | error = -ENOMEM; |
8b765398 | 4825 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4826 | if (!mddev->queue) |
4827 | goto abort; | |
409c57f3 N |
4828 | mddev->queue->queuedata = mddev; |
4829 | ||
409c57f3 | 4830 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 4831 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 4832 | |
1da177e4 LT |
4833 | disk = alloc_disk(1 << shift); |
4834 | if (!disk) { | |
8b765398 N |
4835 | blk_cleanup_queue(mddev->queue); |
4836 | mddev->queue = NULL; | |
0909dc44 | 4837 | goto abort; |
1da177e4 | 4838 | } |
efeb53c0 | 4839 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4840 | disk->first_minor = unit << shift; |
efeb53c0 N |
4841 | if (name) |
4842 | strcpy(disk->disk_name, name); | |
4843 | else if (partitioned) | |
1da177e4 | 4844 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4845 | else |
1da177e4 | 4846 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4847 | disk->fops = &md_fops; |
4848 | disk->private_data = mddev; | |
4849 | disk->queue = mddev->queue; | |
b0140891 | 4850 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4851 | /* Allow extended partitions. This makes the |
d3374825 | 4852 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4853 | * remove it now. |
4854 | */ | |
4855 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4856 | mddev->gendisk = disk; |
b0140891 N |
4857 | /* As soon as we call add_disk(), another thread could get |
4858 | * through to md_open, so make sure it doesn't get too far | |
4859 | */ | |
4860 | mutex_lock(&mddev->open_mutex); | |
4861 | add_disk(disk); | |
4862 | ||
ed9e1982 TH |
4863 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4864 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4865 | if (error) { |
4866 | /* This isn't possible, but as kobject_init_and_add is marked | |
4867 | * __must_check, we must do something with the result | |
4868 | */ | |
5e55e2f5 N |
4869 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4870 | disk->disk_name); | |
0909dc44 N |
4871 | error = 0; |
4872 | } | |
00bcb4ac N |
4873 | if (mddev->kobj.sd && |
4874 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4875 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4876 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4877 | abort: |
4878 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4879 | if (!error && mddev->kobj.sd) { |
3830c62f | 4880 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4881 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4882 | } |
d3374825 | 4883 | mddev_put(mddev); |
0909dc44 | 4884 | return error; |
efeb53c0 N |
4885 | } |
4886 | ||
4887 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4888 | { | |
4889 | md_alloc(dev, NULL); | |
1da177e4 LT |
4890 | return NULL; |
4891 | } | |
4892 | ||
efeb53c0 N |
4893 | static int add_named_array(const char *val, struct kernel_param *kp) |
4894 | { | |
4895 | /* val must be "md_*" where * is not all digits. | |
4896 | * We allocate an array with a large free minor number, and | |
4897 | * set the name to val. val must not already be an active name. | |
4898 | */ | |
4899 | int len = strlen(val); | |
4900 | char buf[DISK_NAME_LEN]; | |
4901 | ||
4902 | while (len && val[len-1] == '\n') | |
4903 | len--; | |
4904 | if (len >= DISK_NAME_LEN) | |
4905 | return -E2BIG; | |
4906 | strlcpy(buf, val, len+1); | |
4907 | if (strncmp(buf, "md_", 3) != 0) | |
4908 | return -EINVAL; | |
4909 | return md_alloc(0, buf); | |
4910 | } | |
4911 | ||
1da177e4 LT |
4912 | static void md_safemode_timeout(unsigned long data) |
4913 | { | |
fd01b88c | 4914 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 4915 | |
0fd62b86 NB |
4916 | if (!atomic_read(&mddev->writes_pending)) { |
4917 | mddev->safemode = 1; | |
4918 | if (mddev->external) | |
00bcb4ac | 4919 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4920 | } |
1da177e4 LT |
4921 | md_wakeup_thread(mddev->thread); |
4922 | } | |
4923 | ||
6ff8d8ec | 4924 | static int start_dirty_degraded; |
1da177e4 | 4925 | |
fd01b88c | 4926 | int md_run(struct mddev *mddev) |
1da177e4 | 4927 | { |
2604b703 | 4928 | int err; |
3cb03002 | 4929 | struct md_rdev *rdev; |
84fc4b56 | 4930 | struct md_personality *pers; |
1da177e4 | 4931 | |
a757e64c N |
4932 | if (list_empty(&mddev->disks)) |
4933 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4934 | return -EINVAL; |
1da177e4 LT |
4935 | |
4936 | if (mddev->pers) | |
4937 | return -EBUSY; | |
bb4f1e9d N |
4938 | /* Cannot run until previous stop completes properly */ |
4939 | if (mddev->sysfs_active) | |
4940 | return -EBUSY; | |
b6eb127d | 4941 | |
1da177e4 LT |
4942 | /* |
4943 | * Analyze all RAID superblock(s) | |
4944 | */ | |
1ec4a939 N |
4945 | if (!mddev->raid_disks) { |
4946 | if (!mddev->persistent) | |
4947 | return -EINVAL; | |
a757e64c | 4948 | analyze_sbs(mddev); |
1ec4a939 | 4949 | } |
1da177e4 | 4950 | |
d9d166c2 N |
4951 | if (mddev->level != LEVEL_NONE) |
4952 | request_module("md-level-%d", mddev->level); | |
4953 | else if (mddev->clevel[0]) | |
4954 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4955 | |
4956 | /* | |
4957 | * Drop all container device buffers, from now on | |
4958 | * the only valid external interface is through the md | |
4959 | * device. | |
1da177e4 | 4960 | */ |
dafb20fa | 4961 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 4962 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4963 | continue; |
4964 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4965 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4966 | |
4967 | /* perform some consistency tests on the device. | |
4968 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4969 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4970 | */ |
a6ff7e08 JB |
4971 | if (rdev->meta_bdev) { |
4972 | /* Nothing to check */; | |
4973 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
4974 | if (mddev->dev_sectors && |
4975 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4976 | > rdev->sb_start) { |
f0d76d70 N |
4977 | printk("md: %s: data overlaps metadata\n", |
4978 | mdname(mddev)); | |
4979 | return -EINVAL; | |
4980 | } | |
4981 | } else { | |
0f420358 | 4982 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4983 | > rdev->data_offset) { |
4984 | printk("md: %s: metadata overlaps data\n", | |
4985 | mdname(mddev)); | |
4986 | return -EINVAL; | |
4987 | } | |
4988 | } | |
00bcb4ac | 4989 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4990 | } |
4991 | ||
a167f663 | 4992 | if (mddev->bio_set == NULL) |
395c72a7 | 4993 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
a167f663 | 4994 | |
1da177e4 | 4995 | spin_lock(&pers_lock); |
d9d166c2 | 4996 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4997 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4998 | spin_unlock(&pers_lock); |
d9d166c2 N |
4999 | if (mddev->level != LEVEL_NONE) |
5000 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
5001 | mddev->level); | |
5002 | else | |
5003 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
5004 | mddev->clevel); | |
1da177e4 LT |
5005 | return -EINVAL; |
5006 | } | |
1da177e4 | 5007 | spin_unlock(&pers_lock); |
34817e8c N |
5008 | if (mddev->level != pers->level) { |
5009 | mddev->level = pers->level; | |
5010 | mddev->new_level = pers->level; | |
5011 | } | |
d9d166c2 | 5012 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5013 | |
f6705578 | 5014 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5015 | pers->start_reshape == NULL) { |
f6705578 | 5016 | /* This personality cannot handle reshaping... */ |
f6705578 N |
5017 | module_put(pers->owner); |
5018 | return -EINVAL; | |
5019 | } | |
5020 | ||
7dd5e7c3 N |
5021 | if (pers->sync_request) { |
5022 | /* Warn if this is a potentially silly | |
5023 | * configuration. | |
5024 | */ | |
5025 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5026 | struct md_rdev *rdev2; |
7dd5e7c3 | 5027 | int warned = 0; |
159ec1fc | 5028 | |
dafb20fa N |
5029 | rdev_for_each(rdev, mddev) |
5030 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5031 | if (rdev < rdev2 && |
5032 | rdev->bdev->bd_contains == | |
5033 | rdev2->bdev->bd_contains) { | |
5034 | printk(KERN_WARNING | |
5035 | "%s: WARNING: %s appears to be" | |
5036 | " on the same physical disk as" | |
5037 | " %s.\n", | |
5038 | mdname(mddev), | |
5039 | bdevname(rdev->bdev,b), | |
5040 | bdevname(rdev2->bdev,b2)); | |
5041 | warned = 1; | |
5042 | } | |
5043 | } | |
159ec1fc | 5044 | |
7dd5e7c3 N |
5045 | if (warned) |
5046 | printk(KERN_WARNING | |
5047 | "True protection against single-disk" | |
5048 | " failure might be compromised.\n"); | |
5049 | } | |
5050 | ||
657390d2 | 5051 | mddev->recovery = 0; |
58c0fed4 AN |
5052 | /* may be over-ridden by personality */ |
5053 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5054 | ||
6ff8d8ec | 5055 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5056 | |
0f9552b5 | 5057 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5058 | mddev->ro = 2; /* read-only, but switch on first write */ |
5059 | ||
36d091f4 | 5060 | err = pers->run(mddev); |
13e53df3 AN |
5061 | if (err) |
5062 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
36d091f4 | 5063 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
b522adcd DW |
5064 | WARN_ONCE(!mddev->external_size, "%s: default size too small," |
5065 | " but 'external_size' not in effect?\n", __func__); | |
5066 | printk(KERN_ERR | |
5067 | "md: invalid array_size %llu > default size %llu\n", | |
5068 | (unsigned long long)mddev->array_sectors / 2, | |
36d091f4 | 5069 | (unsigned long long)pers->size(mddev, 0, 0) / 2); |
b522adcd | 5070 | err = -EINVAL; |
b522adcd | 5071 | } |
36d091f4 | 5072 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5073 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
b15c2e57 | 5074 | err = bitmap_create(mddev); |
5aa61f42 | 5075 | if (err) |
b15c2e57 N |
5076 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", |
5077 | mdname(mddev), err); | |
b15c2e57 | 5078 | } |
1da177e4 | 5079 | if (err) { |
5aa61f42 | 5080 | mddev_detach(mddev); |
36d091f4 N |
5081 | pers->free(mddev, mddev->private); |
5082 | module_put(pers->owner); | |
32a7627c N |
5083 | bitmap_destroy(mddev); |
5084 | return err; | |
1da177e4 | 5085 | } |
5c675f83 N |
5086 | if (mddev->queue) { |
5087 | mddev->queue->backing_dev_info.congested_data = mddev; | |
5088 | mddev->queue->backing_dev_info.congested_fn = md_congested; | |
64590f45 | 5089 | blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec); |
5c675f83 | 5090 | } |
36d091f4 | 5091 | if (pers->sync_request) { |
00bcb4ac N |
5092 | if (mddev->kobj.sd && |
5093 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
5094 | printk(KERN_WARNING |
5095 | "md: cannot register extra attributes for %s\n", | |
5096 | mdname(mddev)); | |
00bcb4ac | 5097 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5098 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5099 | mddev->ro = 0; |
5100 | ||
f72ffdd6 | 5101 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
5102 | atomic_set(&mddev->max_corr_read_errors, |
5103 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
5104 | mddev->safemode = 0; |
5105 | mddev->safemode_timer.function = md_safemode_timeout; | |
5106 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 5107 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 5108 | mddev->in_sync = 1; |
0ca69886 | 5109 | smp_wmb(); |
36d091f4 N |
5110 | spin_lock(&mddev->lock); |
5111 | mddev->pers = pers; | |
0ca69886 | 5112 | mddev->ready = 1; |
36d091f4 | 5113 | spin_unlock(&mddev->lock); |
dafb20fa | 5114 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5115 | if (rdev->raid_disk >= 0) |
5116 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5117 | /* failure here is OK */; |
f72ffdd6 | 5118 | |
1da177e4 | 5119 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5120 | |
7a0a5355 | 5121 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 5122 | md_update_sb(mddev, 0); |
1da177e4 | 5123 | |
d7603b7e | 5124 | md_new_event(mddev); |
00bcb4ac N |
5125 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5126 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5127 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5128 | return 0; |
5129 | } | |
390ee602 | 5130 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5131 | |
fd01b88c | 5132 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5133 | { |
5134 | int err; | |
5135 | ||
5136 | err = md_run(mddev); | |
5137 | if (err) | |
5138 | goto out; | |
69e51b44 N |
5139 | err = bitmap_load(mddev); |
5140 | if (err) { | |
5141 | bitmap_destroy(mddev); | |
5142 | goto out; | |
5143 | } | |
0fd018af JB |
5144 | |
5145 | md_wakeup_thread(mddev->thread); | |
5146 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5147 | ||
fe60b014 N |
5148 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5149 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5150 | mddev->changed = 1; |
fe60b014 N |
5151 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5152 | out: | |
5153 | return err; | |
5154 | } | |
5155 | ||
fd01b88c | 5156 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5157 | { |
5158 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 5159 | |
80fab1d7 | 5160 | /* Complain if it has no devices */ |
1da177e4 | 5161 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5162 | return -ENXIO; |
5163 | if (!mddev->pers) | |
5164 | return -EINVAL; | |
5165 | if (!mddev->ro) | |
5166 | return -EBUSY; | |
5167 | mddev->safemode = 0; | |
5168 | mddev->ro = 0; | |
5169 | set_disk_ro(disk, 0); | |
5170 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
5171 | mdname(mddev)); | |
5172 | /* Kick recovery or resync if necessary */ | |
5173 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5174 | md_wakeup_thread(mddev->thread); | |
5175 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5176 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5177 | return 0; |
1da177e4 LT |
5178 | } |
5179 | ||
fd01b88c | 5180 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5181 | { |
5182 | mddev->array_sectors = 0; | |
5183 | mddev->external_size = 0; | |
5184 | mddev->dev_sectors = 0; | |
5185 | mddev->raid_disks = 0; | |
5186 | mddev->recovery_cp = 0; | |
5187 | mddev->resync_min = 0; | |
5188 | mddev->resync_max = MaxSector; | |
5189 | mddev->reshape_position = MaxSector; | |
5190 | mddev->external = 0; | |
5191 | mddev->persistent = 0; | |
5192 | mddev->level = LEVEL_NONE; | |
5193 | mddev->clevel[0] = 0; | |
5194 | mddev->flags = 0; | |
5195 | mddev->ro = 0; | |
5196 | mddev->metadata_type[0] = 0; | |
5197 | mddev->chunk_sectors = 0; | |
5198 | mddev->ctime = mddev->utime = 0; | |
5199 | mddev->layout = 0; | |
5200 | mddev->max_disks = 0; | |
5201 | mddev->events = 0; | |
a8707c08 | 5202 | mddev->can_decrease_events = 0; |
6177b472 | 5203 | mddev->delta_disks = 0; |
2c810cdd | 5204 | mddev->reshape_backwards = 0; |
6177b472 N |
5205 | mddev->new_level = LEVEL_NONE; |
5206 | mddev->new_layout = 0; | |
5207 | mddev->new_chunk_sectors = 0; | |
5208 | mddev->curr_resync = 0; | |
7f7583d4 | 5209 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5210 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5211 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5212 | mddev->recovery = 0; | |
5213 | mddev->in_sync = 0; | |
f0b4f7e2 | 5214 | mddev->changed = 0; |
6177b472 | 5215 | mddev->degraded = 0; |
6177b472 | 5216 | mddev->safemode = 0; |
050b6615 | 5217 | mddev->merge_check_needed = 0; |
6177b472 N |
5218 | mddev->bitmap_info.offset = 0; |
5219 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5220 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5221 | mddev->bitmap_info.chunksize = 0; |
5222 | mddev->bitmap_info.daemon_sleep = 0; | |
5223 | mddev->bitmap_info.max_write_behind = 0; | |
5224 | } | |
5225 | ||
fd01b88c | 5226 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5227 | { |
6b6204ee | 5228 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5229 | flush_workqueue(md_misc_wq); |
a047e125 | 5230 | if (mddev->sync_thread) { |
a047e125 | 5231 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5232 | md_reap_sync_thread(mddev); |
a047e125 N |
5233 | } |
5234 | ||
5235 | del_timer_sync(&mddev->safemode_timer); | |
5236 | ||
5237 | bitmap_flush(mddev); | |
5238 | md_super_wait(mddev); | |
5239 | ||
b6d428c6 | 5240 | if (mddev->ro == 0 && |
7a0a5355 | 5241 | (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) { |
a047e125 N |
5242 | /* mark array as shutdown cleanly */ |
5243 | mddev->in_sync = 1; | |
5244 | md_update_sb(mddev, 1); | |
5245 | } | |
5246 | } | |
defad61a | 5247 | |
fd01b88c | 5248 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5249 | { |
29f097c4 | 5250 | mddev_lock_nointr(mddev); |
defad61a N |
5251 | __md_stop_writes(mddev); |
5252 | mddev_unlock(mddev); | |
5253 | } | |
390ee602 | 5254 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5255 | |
5aa61f42 N |
5256 | static void mddev_detach(struct mddev *mddev) |
5257 | { | |
5258 | struct bitmap *bitmap = mddev->bitmap; | |
5259 | /* wait for behind writes to complete */ | |
5260 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { | |
5261 | printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n", | |
5262 | mdname(mddev)); | |
5263 | /* need to kick something here to make sure I/O goes? */ | |
5264 | wait_event(bitmap->behind_wait, | |
5265 | atomic_read(&bitmap->behind_writes) == 0); | |
5266 | } | |
36d091f4 | 5267 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5268 | mddev->pers->quiesce(mddev, 1); |
5269 | mddev->pers->quiesce(mddev, 0); | |
5270 | } | |
5271 | md_unregister_thread(&mddev->thread); | |
5272 | if (mddev->queue) | |
5273 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5274 | } | |
5275 | ||
5eff3c43 | 5276 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5277 | { |
36d091f4 | 5278 | struct md_personality *pers = mddev->pers; |
5aa61f42 | 5279 | mddev_detach(mddev); |
36d091f4 N |
5280 | spin_lock(&mddev->lock); |
5281 | mddev->ready = 0; | |
6177b472 | 5282 | mddev->pers = NULL; |
36d091f4 N |
5283 | spin_unlock(&mddev->lock); |
5284 | pers->free(mddev, mddev->private); | |
5285 | if (pers->sync_request && mddev->to_remove == NULL) | |
5286 | mddev->to_remove = &md_redundancy_group; | |
5287 | module_put(pers->owner); | |
cca9cf90 | 5288 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5289 | } |
5eff3c43 N |
5290 | |
5291 | void md_stop(struct mddev *mddev) | |
5292 | { | |
5293 | /* stop the array and free an attached data structures. | |
5294 | * This is called from dm-raid | |
5295 | */ | |
5296 | __md_stop(mddev); | |
5297 | bitmap_destroy(mddev); | |
5298 | if (mddev->bio_set) | |
5299 | bioset_free(mddev->bio_set); | |
5300 | } | |
5301 | ||
390ee602 | 5302 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5303 | |
a05b7ea0 | 5304 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5305 | { |
5306 | int err = 0; | |
30b8feb7 N |
5307 | int did_freeze = 0; |
5308 | ||
5309 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5310 | did_freeze = 1; | |
5311 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5312 | md_wakeup_thread(mddev->thread); | |
5313 | } | |
f851b60d | 5314 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5315 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5316 | if (mddev->sync_thread) |
30b8feb7 N |
5317 | /* Thread might be blocked waiting for metadata update |
5318 | * which will now never happen */ | |
5319 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5320 | |
30b8feb7 | 5321 | mddev_unlock(mddev); |
f851b60d N |
5322 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5323 | &mddev->recovery)); | |
30b8feb7 N |
5324 | mddev_lock_nointr(mddev); |
5325 | ||
a4bd82d0 | 5326 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5327 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5328 | mddev->sync_thread || |
f851b60d | 5329 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
30b8feb7 | 5330 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { |
a4bd82d0 | 5331 | printk("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5332 | if (did_freeze) { |
5333 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5334 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5335 | md_wakeup_thread(mddev->thread); |
5336 | } | |
a4bd82d0 N |
5337 | err = -EBUSY; |
5338 | goto out; | |
5339 | } | |
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); | |
45eaf45d N |
5349 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5350 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5351 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5352 | err = 0; |
a4bd82d0 N |
5353 | } |
5354 | out: | |
5355 | mutex_unlock(&mddev->open_mutex); | |
5356 | return err; | |
5357 | } | |
5358 | ||
9e653b63 N |
5359 | /* mode: |
5360 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5361 | * 2 - stop but do not disassemble array |
5362 | */ | |
f72ffdd6 | 5363 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5364 | struct block_device *bdev) |
1da177e4 | 5365 | { |
1da177e4 | 5366 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5367 | struct md_rdev *rdev; |
30b8feb7 N |
5368 | int did_freeze = 0; |
5369 | ||
5370 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5371 | did_freeze = 1; | |
5372 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5373 | md_wakeup_thread(mddev->thread); | |
5374 | } | |
f851b60d | 5375 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5376 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5377 | if (mddev->sync_thread) |
30b8feb7 N |
5378 | /* Thread might be blocked waiting for metadata update |
5379 | * which will now never happen */ | |
5380 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5381 | |
30b8feb7 | 5382 | mddev_unlock(mddev); |
f851b60d N |
5383 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
5384 | !test_bit(MD_RECOVERY_RUNNING, | |
5385 | &mddev->recovery))); | |
30b8feb7 | 5386 | mddev_lock_nointr(mddev); |
1da177e4 | 5387 | |
c8c00a69 | 5388 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5389 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5390 | mddev->sysfs_active || |
5391 | mddev->sync_thread || | |
f851b60d | 5392 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
30b8feb7 | 5393 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { |
df5b20cf | 5394 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5395 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5396 | if (did_freeze) { |
5397 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5398 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5399 | md_wakeup_thread(mddev->thread); |
5400 | } | |
260fa034 N |
5401 | return -EBUSY; |
5402 | } | |
6e17b027 | 5403 | if (mddev->pers) { |
a4bd82d0 N |
5404 | if (mddev->ro) |
5405 | set_disk_ro(disk, 0); | |
409c57f3 | 5406 | |
defad61a | 5407 | __md_stop_writes(mddev); |
5eff3c43 | 5408 | __md_stop(mddev); |
a4bd82d0 | 5409 | mddev->queue->merge_bvec_fn = NULL; |
a4bd82d0 | 5410 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 5411 | |
a4bd82d0 | 5412 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5413 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5414 | |
dafb20fa | 5415 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5416 | if (rdev->raid_disk >= 0) |
5417 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5418 | |
a4bd82d0 | 5419 | set_capacity(disk, 0); |
6e17b027 | 5420 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5421 | mddev->changed = 1; |
a4bd82d0 | 5422 | revalidate_disk(disk); |
0d4ca600 | 5423 | |
a4bd82d0 N |
5424 | if (mddev->ro) |
5425 | mddev->ro = 0; | |
6e17b027 N |
5426 | } else |
5427 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5428 | /* |
5429 | * Free resources if final stop | |
5430 | */ | |
9e653b63 | 5431 | if (mode == 0) { |
1da177e4 LT |
5432 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
5433 | ||
978f946b | 5434 | bitmap_destroy(mddev); |
c3d9714e | 5435 | if (mddev->bitmap_info.file) { |
4af1a041 N |
5436 | struct file *f = mddev->bitmap_info.file; |
5437 | spin_lock(&mddev->lock); | |
c3d9714e | 5438 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
5439 | spin_unlock(&mddev->lock); |
5440 | fput(f); | |
978f946b | 5441 | } |
c3d9714e | 5442 | mddev->bitmap_info.offset = 0; |
978f946b | 5443 | |
1da177e4 LT |
5444 | export_array(mddev); |
5445 | ||
6177b472 | 5446 | md_clean(mddev); |
934d9c23 | 5447 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
5448 | if (mddev->hold_active == UNTIL_STOP) |
5449 | mddev->hold_active = 0; | |
a4bd82d0 | 5450 | } |
3f9d99c1 | 5451 | blk_integrity_unregister(disk); |
d7603b7e | 5452 | md_new_event(mddev); |
00bcb4ac | 5453 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5454 | return 0; |
1da177e4 LT |
5455 | } |
5456 | ||
fdee8ae4 | 5457 | #ifndef MODULE |
fd01b88c | 5458 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5459 | { |
3cb03002 | 5460 | struct md_rdev *rdev; |
1da177e4 LT |
5461 | int err; |
5462 | ||
a757e64c | 5463 | if (list_empty(&mddev->disks)) |
1da177e4 | 5464 | return; |
1da177e4 LT |
5465 | |
5466 | printk(KERN_INFO "md: running: "); | |
5467 | ||
dafb20fa | 5468 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
5469 | char b[BDEVNAME_SIZE]; |
5470 | printk("<%s>", bdevname(rdev->bdev,b)); | |
5471 | } | |
5472 | printk("\n"); | |
5473 | ||
d710e138 | 5474 | err = do_md_run(mddev); |
1da177e4 LT |
5475 | if (err) { |
5476 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
a05b7ea0 | 5477 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5478 | } |
5479 | } | |
5480 | ||
5481 | /* | |
5482 | * lets try to run arrays based on all disks that have arrived | |
5483 | * until now. (those are in pending_raid_disks) | |
5484 | * | |
5485 | * the method: pick the first pending disk, collect all disks with | |
5486 | * the same UUID, remove all from the pending list and put them into | |
5487 | * the 'same_array' list. Then order this list based on superblock | |
5488 | * update time (freshest comes first), kick out 'old' disks and | |
5489 | * compare superblocks. If everything's fine then run it. | |
5490 | * | |
5491 | * If "unit" is allocated, then bump its reference count | |
5492 | */ | |
5493 | static void autorun_devices(int part) | |
5494 | { | |
3cb03002 | 5495 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5496 | struct mddev *mddev; |
1da177e4 LT |
5497 | char b[BDEVNAME_SIZE]; |
5498 | ||
5499 | printk(KERN_INFO "md: autorun ...\n"); | |
5500 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 5501 | int unit; |
1da177e4 | 5502 | dev_t dev; |
ad01c9e3 | 5503 | LIST_HEAD(candidates); |
1da177e4 | 5504 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5505 | struct md_rdev, same_set); |
1da177e4 LT |
5506 | |
5507 | printk(KERN_INFO "md: considering %s ...\n", | |
5508 | bdevname(rdev0->bdev,b)); | |
5509 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 5510 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
5511 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
5512 | printk(KERN_INFO "md: adding %s ...\n", | |
5513 | bdevname(rdev->bdev,b)); | |
5514 | list_move(&rdev->same_set, &candidates); | |
5515 | } | |
5516 | /* | |
5517 | * now we have a set of devices, with all of them having | |
5518 | * mostly sane superblocks. It's time to allocate the | |
5519 | * mddev. | |
5520 | */ | |
e8703fe1 N |
5521 | if (part) { |
5522 | dev = MKDEV(mdp_major, | |
5523 | rdev0->preferred_minor << MdpMinorShift); | |
5524 | unit = MINOR(dev) >> MdpMinorShift; | |
5525 | } else { | |
5526 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5527 | unit = MINOR(dev); | |
5528 | } | |
5529 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
5530 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
5531 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
5532 | break; | |
5533 | } | |
1da177e4 LT |
5534 | |
5535 | md_probe(dev, NULL, NULL); | |
5536 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5537 | if (!mddev || !mddev->gendisk) { |
5538 | if (mddev) | |
5539 | mddev_put(mddev); | |
5540 | printk(KERN_ERR | |
1da177e4 LT |
5541 | "md: cannot allocate memory for md drive.\n"); |
5542 | break; | |
5543 | } | |
f72ffdd6 | 5544 | if (mddev_lock(mddev)) |
1da177e4 LT |
5545 | printk(KERN_WARNING "md: %s locked, cannot run\n", |
5546 | mdname(mddev)); | |
5547 | else if (mddev->raid_disks || mddev->major_version | |
5548 | || !list_empty(&mddev->disks)) { | |
f72ffdd6 | 5549 | printk(KERN_WARNING |
1da177e4 LT |
5550 | "md: %s already running, cannot run %s\n", |
5551 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
5552 | mddev_unlock(mddev); | |
5553 | } else { | |
5554 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 5555 | mddev->persistent = 1; |
159ec1fc | 5556 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5557 | list_del_init(&rdev->same_set); |
5558 | if (bind_rdev_to_array(rdev, mddev)) | |
5559 | export_rdev(rdev); | |
5560 | } | |
5561 | autorun_array(mddev); | |
5562 | mddev_unlock(mddev); | |
5563 | } | |
5564 | /* on success, candidates will be empty, on error | |
5565 | * it won't... | |
5566 | */ | |
159ec1fc | 5567 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5568 | list_del_init(&rdev->same_set); |
1da177e4 | 5569 | export_rdev(rdev); |
4b80991c | 5570 | } |
1da177e4 LT |
5571 | mddev_put(mddev); |
5572 | } | |
5573 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5574 | } | |
fdee8ae4 | 5575 | #endif /* !MODULE */ |
1da177e4 | 5576 | |
f72ffdd6 | 5577 | static int get_version(void __user *arg) |
1da177e4 LT |
5578 | { |
5579 | mdu_version_t ver; | |
5580 | ||
5581 | ver.major = MD_MAJOR_VERSION; | |
5582 | ver.minor = MD_MINOR_VERSION; | |
5583 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5584 | ||
5585 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5586 | return -EFAULT; | |
5587 | ||
5588 | return 0; | |
5589 | } | |
5590 | ||
f72ffdd6 | 5591 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
5592 | { |
5593 | mdu_array_info_t info; | |
a9f326eb | 5594 | int nr,working,insync,failed,spare; |
3cb03002 | 5595 | struct md_rdev *rdev; |
1da177e4 | 5596 | |
1ca69c4b N |
5597 | nr = working = insync = failed = spare = 0; |
5598 | rcu_read_lock(); | |
5599 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5600 | nr++; |
b2d444d7 | 5601 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5602 | failed++; |
5603 | else { | |
5604 | working++; | |
b2d444d7 | 5605 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 5606 | insync++; |
1da177e4 LT |
5607 | else |
5608 | spare++; | |
5609 | } | |
5610 | } | |
1ca69c4b | 5611 | rcu_read_unlock(); |
1da177e4 LT |
5612 | |
5613 | info.major_version = mddev->major_version; | |
5614 | info.minor_version = mddev->minor_version; | |
5615 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5616 | info.ctime = mddev->ctime; | |
5617 | info.level = mddev->level; | |
58c0fed4 AN |
5618 | info.size = mddev->dev_sectors / 2; |
5619 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5620 | info.size = -1; |
1da177e4 LT |
5621 | info.nr_disks = nr; |
5622 | info.raid_disks = mddev->raid_disks; | |
5623 | info.md_minor = mddev->md_minor; | |
5624 | info.not_persistent= !mddev->persistent; | |
5625 | ||
5626 | info.utime = mddev->utime; | |
5627 | info.state = 0; | |
5628 | if (mddev->in_sync) | |
5629 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5630 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 5631 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5632 | info.active_disks = insync; |
1da177e4 LT |
5633 | info.working_disks = working; |
5634 | info.failed_disks = failed; | |
5635 | info.spare_disks = spare; | |
5636 | ||
5637 | info.layout = mddev->layout; | |
9d8f0363 | 5638 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5639 | |
5640 | if (copy_to_user(arg, &info, sizeof(info))) | |
5641 | return -EFAULT; | |
5642 | ||
5643 | return 0; | |
5644 | } | |
5645 | ||
f72ffdd6 | 5646 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
5647 | { |
5648 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 5649 | char *ptr; |
4af1a041 | 5650 | int err; |
32a7627c | 5651 | |
60559da4 | 5652 | file = kmalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 5653 | if (!file) |
4af1a041 | 5654 | return -ENOMEM; |
32a7627c | 5655 | |
4af1a041 N |
5656 | err = 0; |
5657 | spin_lock(&mddev->lock); | |
32a7627c | 5658 | /* bitmap disabled, zero the first byte and copy out */ |
4af1a041 | 5659 | if (!mddev->bitmap_info.file) |
32a7627c | 5660 | file->pathname[0] = '\0'; |
4af1a041 N |
5661 | else if ((ptr = d_path(&mddev->bitmap_info.file->f_path, |
5662 | file->pathname, sizeof(file->pathname))), | |
5663 | IS_ERR(ptr)) | |
5664 | err = PTR_ERR(ptr); | |
5665 | else | |
5666 | memmove(file->pathname, ptr, | |
5667 | sizeof(file->pathname)-(ptr-file->pathname)); | |
5668 | spin_unlock(&mddev->lock); | |
32a7627c | 5669 | |
4af1a041 N |
5670 | if (err == 0 && |
5671 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 5672 | err = -EFAULT; |
4af1a041 | 5673 | |
32a7627c N |
5674 | kfree(file); |
5675 | return err; | |
5676 | } | |
5677 | ||
f72ffdd6 | 5678 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
5679 | { |
5680 | mdu_disk_info_t info; | |
3cb03002 | 5681 | struct md_rdev *rdev; |
1da177e4 LT |
5682 | |
5683 | if (copy_from_user(&info, arg, sizeof(info))) | |
5684 | return -EFAULT; | |
5685 | ||
1ca69c4b N |
5686 | rcu_read_lock(); |
5687 | rdev = find_rdev_nr_rcu(mddev, info.number); | |
1da177e4 LT |
5688 | if (rdev) { |
5689 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5690 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5691 | info.raid_disk = rdev->raid_disk; | |
5692 | info.state = 0; | |
b2d444d7 | 5693 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5694 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5695 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5696 | info.state |= (1<<MD_DISK_ACTIVE); |
5697 | info.state |= (1<<MD_DISK_SYNC); | |
5698 | } | |
8ddf9efe N |
5699 | if (test_bit(WriteMostly, &rdev->flags)) |
5700 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5701 | } else { |
5702 | info.major = info.minor = 0; | |
5703 | info.raid_disk = -1; | |
5704 | info.state = (1<<MD_DISK_REMOVED); | |
5705 | } | |
1ca69c4b | 5706 | rcu_read_unlock(); |
1da177e4 LT |
5707 | |
5708 | if (copy_to_user(arg, &info, sizeof(info))) | |
5709 | return -EFAULT; | |
5710 | ||
5711 | return 0; | |
5712 | } | |
5713 | ||
f72ffdd6 | 5714 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
5715 | { |
5716 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5717 | struct md_rdev *rdev; |
1da177e4 LT |
5718 | dev_t dev = MKDEV(info->major,info->minor); |
5719 | ||
5720 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5721 | return -EOVERFLOW; | |
5722 | ||
5723 | if (!mddev->raid_disks) { | |
5724 | int err; | |
5725 | /* expecting a device which has a superblock */ | |
5726 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5727 | if (IS_ERR(rdev)) { | |
f72ffdd6 | 5728 | printk(KERN_WARNING |
1da177e4 LT |
5729 | "md: md_import_device returned %ld\n", |
5730 | PTR_ERR(rdev)); | |
5731 | return PTR_ERR(rdev); | |
5732 | } | |
5733 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
5734 | struct md_rdev *rdev0 |
5735 | = list_entry(mddev->disks.next, | |
5736 | struct md_rdev, same_set); | |
a9f326eb | 5737 | err = super_types[mddev->major_version] |
1da177e4 LT |
5738 | .load_super(rdev, rdev0, mddev->minor_version); |
5739 | if (err < 0) { | |
f72ffdd6 | 5740 | printk(KERN_WARNING |
1da177e4 | 5741 | "md: %s has different UUID to %s\n", |
f72ffdd6 | 5742 | bdevname(rdev->bdev,b), |
1da177e4 LT |
5743 | bdevname(rdev0->bdev,b2)); |
5744 | export_rdev(rdev); | |
5745 | return -EINVAL; | |
5746 | } | |
5747 | } | |
5748 | err = bind_rdev_to_array(rdev, mddev); | |
5749 | if (err) | |
5750 | export_rdev(rdev); | |
5751 | return err; | |
5752 | } | |
5753 | ||
5754 | /* | |
5755 | * add_new_disk can be used once the array is assembled | |
5756 | * to add "hot spares". They must already have a superblock | |
5757 | * written | |
5758 | */ | |
5759 | if (mddev->pers) { | |
5760 | int err; | |
5761 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 5762 | printk(KERN_WARNING |
1da177e4 LT |
5763 | "%s: personality does not support diskops!\n", |
5764 | mdname(mddev)); | |
5765 | return -EINVAL; | |
5766 | } | |
7b1e35f6 N |
5767 | if (mddev->persistent) |
5768 | rdev = md_import_device(dev, mddev->major_version, | |
5769 | mddev->minor_version); | |
5770 | else | |
5771 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 5772 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5773 | printk(KERN_WARNING |
1da177e4 LT |
5774 | "md: md_import_device returned %ld\n", |
5775 | PTR_ERR(rdev)); | |
5776 | return PTR_ERR(rdev); | |
5777 | } | |
1a855a06 | 5778 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5779 | if (!mddev->persistent) { |
5780 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5781 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5782 | rdev->raid_disk = info->raid_disk; |
bf572541 | 5783 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 5784 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 5785 | } else |
41158c7e | 5786 | rdev->raid_disk = -1; |
f466722c | 5787 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
5788 | } else |
5789 | super_types[mddev->major_version]. | |
5790 | validate_super(mddev, rdev); | |
bedd86b7 | 5791 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 5792 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
5793 | /* This was a hot-add request, but events doesn't |
5794 | * match, so reject it. | |
5795 | */ | |
5796 | export_rdev(rdev); | |
5797 | return -EINVAL; | |
5798 | } | |
5799 | ||
b2d444d7 | 5800 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5801 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5802 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5803 | else |
5804 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5805 | |
1da177e4 LT |
5806 | rdev->raid_disk = -1; |
5807 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5808 | if (!err && !mddev->pers->hot_remove_disk) { |
5809 | /* If there is hot_add_disk but no hot_remove_disk | |
5810 | * then added disks for geometry changes, | |
5811 | * and should be added immediately. | |
5812 | */ | |
5813 | super_types[mddev->major_version]. | |
5814 | validate_super(mddev, rdev); | |
5815 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5816 | if (err) | |
5817 | unbind_rdev_from_array(rdev); | |
5818 | } | |
1da177e4 LT |
5819 | if (err) |
5820 | export_rdev(rdev); | |
52664732 | 5821 | else |
00bcb4ac | 5822 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5823 | |
7ceb17e8 | 5824 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
72a23c21 NB |
5825 | if (mddev->degraded) |
5826 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5827 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
9864c005 | 5828 | if (!err) |
5829 | md_new_event(mddev); | |
005eca5e | 5830 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5831 | return err; |
5832 | } | |
5833 | ||
5834 | /* otherwise, add_new_disk is only allowed | |
5835 | * for major_version==0 superblocks | |
5836 | */ | |
5837 | if (mddev->major_version != 0) { | |
5838 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5839 | mdname(mddev)); | |
5840 | return -EINVAL; | |
5841 | } | |
5842 | ||
5843 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5844 | int err; | |
d710e138 | 5845 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 5846 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5847 | printk(KERN_WARNING |
1da177e4 LT |
5848 | "md: error, md_import_device() returned %ld\n", |
5849 | PTR_ERR(rdev)); | |
5850 | return PTR_ERR(rdev); | |
5851 | } | |
5852 | rdev->desc_nr = info->number; | |
5853 | if (info->raid_disk < mddev->raid_disks) | |
5854 | rdev->raid_disk = info->raid_disk; | |
5855 | else | |
5856 | rdev->raid_disk = -1; | |
5857 | ||
1da177e4 | 5858 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5859 | if (info->state & (1<<MD_DISK_SYNC)) |
5860 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5861 | |
8ddf9efe N |
5862 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5863 | set_bit(WriteMostly, &rdev->flags); | |
5864 | ||
1da177e4 LT |
5865 | if (!mddev->persistent) { |
5866 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5867 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5868 | } else | |
57b2caa3 | 5869 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5870 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5871 | |
2bf071bf N |
5872 | err = bind_rdev_to_array(rdev, mddev); |
5873 | if (err) { | |
5874 | export_rdev(rdev); | |
5875 | return err; | |
5876 | } | |
1da177e4 LT |
5877 | } |
5878 | ||
5879 | return 0; | |
5880 | } | |
5881 | ||
f72ffdd6 | 5882 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
5883 | { |
5884 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 5885 | struct md_rdev *rdev; |
1da177e4 | 5886 | |
1da177e4 LT |
5887 | rdev = find_rdev(mddev, dev); |
5888 | if (!rdev) | |
5889 | return -ENXIO; | |
5890 | ||
3ea8929d N |
5891 | clear_bit(Blocked, &rdev->flags); |
5892 | remove_and_add_spares(mddev, rdev); | |
5893 | ||
1da177e4 LT |
5894 | if (rdev->raid_disk >= 0) |
5895 | goto busy; | |
5896 | ||
5897 | kick_rdev_from_array(rdev); | |
850b2b42 | 5898 | md_update_sb(mddev, 1); |
d7603b7e | 5899 | md_new_event(mddev); |
1da177e4 LT |
5900 | |
5901 | return 0; | |
5902 | busy: | |
fdefa4d8 | 5903 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5904 | bdevname(rdev->bdev,b), mdname(mddev)); |
5905 | return -EBUSY; | |
5906 | } | |
5907 | ||
f72ffdd6 | 5908 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
5909 | { |
5910 | char b[BDEVNAME_SIZE]; | |
5911 | int err; | |
3cb03002 | 5912 | struct md_rdev *rdev; |
1da177e4 LT |
5913 | |
5914 | if (!mddev->pers) | |
5915 | return -ENODEV; | |
5916 | ||
5917 | if (mddev->major_version != 0) { | |
5918 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5919 | " version-0 superblocks.\n", | |
5920 | mdname(mddev)); | |
5921 | return -EINVAL; | |
5922 | } | |
5923 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 5924 | printk(KERN_WARNING |
1da177e4 LT |
5925 | "%s: personality does not support diskops!\n", |
5926 | mdname(mddev)); | |
5927 | return -EINVAL; | |
5928 | } | |
5929 | ||
d710e138 | 5930 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 5931 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5932 | printk(KERN_WARNING |
1da177e4 LT |
5933 | "md: error, md_import_device() returned %ld\n", |
5934 | PTR_ERR(rdev)); | |
5935 | return -EINVAL; | |
5936 | } | |
5937 | ||
5938 | if (mddev->persistent) | |
57b2caa3 | 5939 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5940 | else |
77304d2a | 5941 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5942 | |
8190e754 | 5943 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5944 | |
b2d444d7 | 5945 | if (test_bit(Faulty, &rdev->flags)) { |
f72ffdd6 | 5946 | printk(KERN_WARNING |
1da177e4 LT |
5947 | "md: can not hot-add faulty %s disk to %s!\n", |
5948 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5949 | err = -EINVAL; | |
5950 | goto abort_export; | |
5951 | } | |
b2d444d7 | 5952 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5953 | rdev->desc_nr = -1; |
5842730d | 5954 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5955 | err = bind_rdev_to_array(rdev, mddev); |
5956 | if (err) | |
5957 | goto abort_export; | |
1da177e4 LT |
5958 | |
5959 | /* | |
5960 | * The rest should better be atomic, we can have disk failures | |
5961 | * noticed in interrupt contexts ... | |
5962 | */ | |
5963 | ||
1da177e4 LT |
5964 | rdev->raid_disk = -1; |
5965 | ||
850b2b42 | 5966 | md_update_sb(mddev, 1); |
1da177e4 LT |
5967 | |
5968 | /* | |
5969 | * Kick recovery, maybe this spare has to be added to the | |
5970 | * array immediately. | |
5971 | */ | |
5972 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5973 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5974 | md_new_event(mddev); |
1da177e4 LT |
5975 | return 0; |
5976 | ||
1da177e4 LT |
5977 | abort_export: |
5978 | export_rdev(rdev); | |
5979 | return err; | |
5980 | } | |
5981 | ||
fd01b88c | 5982 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 5983 | { |
035328c2 | 5984 | int err = 0; |
32a7627c | 5985 | |
36fa3063 | 5986 | if (mddev->pers) { |
d66b1b39 | 5987 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
5988 | return -EBUSY; |
5989 | if (mddev->recovery || mddev->sync_thread) | |
5990 | return -EBUSY; | |
5991 | /* we should be able to change the bitmap.. */ | |
5992 | } | |
32a7627c | 5993 | |
36fa3063 | 5994 | if (fd >= 0) { |
035328c2 | 5995 | struct inode *inode; |
1e594bb2 N |
5996 | struct file *f; |
5997 | ||
5998 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 5999 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6000 | f = fget(fd); |
32a7627c | 6001 | |
1e594bb2 | 6002 | if (f == NULL) { |
36fa3063 N |
6003 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
6004 | mdname(mddev)); | |
6005 | return -EBADF; | |
6006 | } | |
6007 | ||
1e594bb2 | 6008 | inode = f->f_mapping->host; |
035328c2 N |
6009 | if (!S_ISREG(inode->i_mode)) { |
6010 | printk(KERN_ERR "%s: error: bitmap file must be a regular file\n", | |
6011 | mdname(mddev)); | |
6012 | err = -EBADF; | |
1e594bb2 | 6013 | } else if (!(f->f_mode & FMODE_WRITE)) { |
035328c2 N |
6014 | printk(KERN_ERR "%s: error: bitmap file must open for write\n", |
6015 | mdname(mddev)); | |
6016 | err = -EBADF; | |
6017 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
36fa3063 N |
6018 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", |
6019 | mdname(mddev)); | |
035328c2 N |
6020 | err = -EBUSY; |
6021 | } | |
6022 | if (err) { | |
1e594bb2 | 6023 | fput(f); |
36fa3063 N |
6024 | return err; |
6025 | } | |
1e594bb2 | 6026 | mddev->bitmap_info.file = f; |
c3d9714e | 6027 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6028 | } else if (mddev->bitmap == NULL) |
6029 | return -ENOENT; /* cannot remove what isn't there */ | |
6030 | err = 0; | |
6031 | if (mddev->pers) { | |
6032 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6033 | if (fd >= 0) { |
36fa3063 | 6034 | err = bitmap_create(mddev); |
69e51b44 N |
6035 | if (!err) |
6036 | err = bitmap_load(mddev); | |
6037 | } | |
d7375ab3 | 6038 | if (fd < 0 || err) { |
36fa3063 | 6039 | bitmap_destroy(mddev); |
d7375ab3 N |
6040 | fd = -1; /* make sure to put the file */ |
6041 | } | |
36fa3063 | 6042 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6043 | } |
6044 | if (fd < 0) { | |
4af1a041 N |
6045 | struct file *f = mddev->bitmap_info.file; |
6046 | if (f) { | |
6047 | spin_lock(&mddev->lock); | |
6048 | mddev->bitmap_info.file = NULL; | |
6049 | spin_unlock(&mddev->lock); | |
6050 | fput(f); | |
6051 | } | |
36fa3063 N |
6052 | } |
6053 | ||
32a7627c N |
6054 | return err; |
6055 | } | |
6056 | ||
1da177e4 LT |
6057 | /* |
6058 | * set_array_info is used two different ways | |
6059 | * The original usage is when creating a new array. | |
6060 | * In this usage, raid_disks is > 0 and it together with | |
6061 | * level, size, not_persistent,layout,chunksize determine the | |
6062 | * shape of the array. | |
6063 | * This will always create an array with a type-0.90.0 superblock. | |
6064 | * The newer usage is when assembling an array. | |
6065 | * In this case raid_disks will be 0, and the major_version field is | |
6066 | * use to determine which style super-blocks are to be found on the devices. | |
6067 | * The minor and patch _version numbers are also kept incase the | |
6068 | * super_block handler wishes to interpret them. | |
6069 | */ | |
f72ffdd6 | 6070 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6071 | { |
6072 | ||
6073 | if (info->raid_disks == 0) { | |
6074 | /* just setting version number for superblock loading */ | |
6075 | if (info->major_version < 0 || | |
50511da3 | 6076 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6077 | super_types[info->major_version].name == NULL) { |
6078 | /* maybe try to auto-load a module? */ | |
f72ffdd6 | 6079 | printk(KERN_INFO |
1da177e4 LT |
6080 | "md: superblock version %d not known\n", |
6081 | info->major_version); | |
6082 | return -EINVAL; | |
6083 | } | |
6084 | mddev->major_version = info->major_version; | |
6085 | mddev->minor_version = info->minor_version; | |
6086 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6087 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6088 | /* ensure mddev_put doesn't delete this now that there |
6089 | * is some minimal configuration. | |
6090 | */ | |
6091 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
6092 | return 0; |
6093 | } | |
6094 | mddev->major_version = MD_MAJOR_VERSION; | |
6095 | mddev->minor_version = MD_MINOR_VERSION; | |
6096 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
6097 | mddev->ctime = get_seconds(); | |
6098 | ||
6099 | mddev->level = info->level; | |
17115e03 | 6100 | mddev->clevel[0] = 0; |
58c0fed4 | 6101 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6102 | mddev->raid_disks = info->raid_disks; |
6103 | /* don't set md_minor, it is determined by which /dev/md* was | |
6104 | * openned | |
6105 | */ | |
6106 | if (info->state & (1<<MD_SB_CLEAN)) | |
6107 | mddev->recovery_cp = MaxSector; | |
6108 | else | |
6109 | mddev->recovery_cp = 0; | |
6110 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6111 | mddev->external = 0; |
1da177e4 LT |
6112 | |
6113 | mddev->layout = info->layout; | |
9d8f0363 | 6114 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
6115 | |
6116 | mddev->max_disks = MD_SB_DISKS; | |
6117 | ||
e691063a N |
6118 | if (mddev->persistent) |
6119 | mddev->flags = 0; | |
850b2b42 | 6120 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6121 | |
c3d9714e | 6122 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6123 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6124 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6125 | |
f6705578 N |
6126 | mddev->reshape_position = MaxSector; |
6127 | ||
1da177e4 LT |
6128 | /* |
6129 | * Generate a 128 bit UUID | |
6130 | */ | |
6131 | get_random_bytes(mddev->uuid, 16); | |
6132 | ||
f6705578 | 6133 | mddev->new_level = mddev->level; |
664e7c41 | 6134 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6135 | mddev->new_layout = mddev->layout; |
6136 | mddev->delta_disks = 0; | |
2c810cdd | 6137 | mddev->reshape_backwards = 0; |
f6705578 | 6138 | |
1da177e4 LT |
6139 | return 0; |
6140 | } | |
6141 | ||
fd01b88c | 6142 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6143 | { |
b522adcd DW |
6144 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6145 | ||
6146 | if (mddev->external_size) | |
6147 | return; | |
6148 | ||
1f403624 DW |
6149 | mddev->array_sectors = array_sectors; |
6150 | } | |
6151 | EXPORT_SYMBOL(md_set_array_sectors); | |
6152 | ||
fd01b88c | 6153 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6154 | { |
3cb03002 | 6155 | struct md_rdev *rdev; |
a35b0d69 | 6156 | int rv; |
d71f9f88 | 6157 | int fit = (num_sectors == 0); |
a35b0d69 N |
6158 | |
6159 | if (mddev->pers->resize == NULL) | |
6160 | return -EINVAL; | |
d71f9f88 AN |
6161 | /* The "num_sectors" is the number of sectors of each device that |
6162 | * is used. This can only make sense for arrays with redundancy. | |
6163 | * linear and raid0 always use whatever space is available. We can only | |
6164 | * consider changing this number if no resync or reconstruction is | |
6165 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6166 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6167 | * of each device. If num_sectors is zero, we find the largest size |
6168 | * that fits. | |
a35b0d69 | 6169 | */ |
f851b60d N |
6170 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6171 | mddev->sync_thread) | |
a35b0d69 | 6172 | return -EBUSY; |
bd8839e0 N |
6173 | if (mddev->ro) |
6174 | return -EROFS; | |
a4a6125a | 6175 | |
dafb20fa | 6176 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6177 | sector_t avail = rdev->sectors; |
01ab5662 | 6178 | |
d71f9f88 AN |
6179 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6180 | num_sectors = avail; | |
6181 | if (avail < num_sectors) | |
a35b0d69 N |
6182 | return -ENOSPC; |
6183 | } | |
d71f9f88 | 6184 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
6185 | if (!rv) |
6186 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
6187 | return rv; |
6188 | } | |
6189 | ||
fd01b88c | 6190 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6191 | { |
6192 | int rv; | |
c6563a8c | 6193 | struct md_rdev *rdev; |
da943b99 | 6194 | /* change the number of raid disks */ |
63c70c4f | 6195 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6196 | return -EINVAL; |
bd8839e0 N |
6197 | if (mddev->ro) |
6198 | return -EROFS; | |
da943b99 | 6199 | if (raid_disks <= 0 || |
233fca36 | 6200 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6201 | return -EINVAL; |
f851b60d N |
6202 | if (mddev->sync_thread || |
6203 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6204 | mddev->reshape_position != MaxSector) | |
da943b99 | 6205 | return -EBUSY; |
c6563a8c N |
6206 | |
6207 | rdev_for_each(rdev, mddev) { | |
6208 | if (mddev->raid_disks < raid_disks && | |
6209 | rdev->data_offset < rdev->new_data_offset) | |
6210 | return -EINVAL; | |
6211 | if (mddev->raid_disks > raid_disks && | |
6212 | rdev->data_offset > rdev->new_data_offset) | |
6213 | return -EINVAL; | |
6214 | } | |
6215 | ||
63c70c4f | 6216 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6217 | if (mddev->delta_disks < 0) |
6218 | mddev->reshape_backwards = 1; | |
6219 | else if (mddev->delta_disks > 0) | |
6220 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6221 | |
6222 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6223 | if (rv < 0) { |
de171cb9 | 6224 | mddev->delta_disks = 0; |
2c810cdd N |
6225 | mddev->reshape_backwards = 0; |
6226 | } | |
da943b99 N |
6227 | return rv; |
6228 | } | |
6229 | ||
1da177e4 LT |
6230 | /* |
6231 | * update_array_info is used to change the configuration of an | |
6232 | * on-line array. | |
6233 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6234 | * fields in the info are checked against the array. | |
6235 | * Any differences that cannot be handled will cause an error. | |
6236 | * Normally, only one change can be managed at a time. | |
6237 | */ | |
fd01b88c | 6238 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6239 | { |
6240 | int rv = 0; | |
6241 | int cnt = 0; | |
36fa3063 N |
6242 | int state = 0; |
6243 | ||
6244 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6245 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6246 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6247 | |
6248 | if (mddev->major_version != info->major_version || | |
6249 | mddev->minor_version != info->minor_version || | |
6250 | /* mddev->patch_version != info->patch_version || */ | |
6251 | mddev->ctime != info->ctime || | |
6252 | mddev->level != info->level || | |
6253 | /* mddev->layout != info->layout || */ | |
6254 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 6255 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6256 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6257 | ((state^info->state) & 0xfffffe00) | |
6258 | ) | |
1da177e4 LT |
6259 | return -EINVAL; |
6260 | /* Check there is only one change */ | |
58c0fed4 AN |
6261 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6262 | cnt++; | |
6263 | if (mddev->raid_disks != info->raid_disks) | |
6264 | cnt++; | |
6265 | if (mddev->layout != info->layout) | |
6266 | cnt++; | |
6267 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6268 | cnt++; | |
6269 | if (cnt == 0) | |
6270 | return 0; | |
6271 | if (cnt > 1) | |
6272 | return -EINVAL; | |
1da177e4 LT |
6273 | |
6274 | if (mddev->layout != info->layout) { | |
6275 | /* Change layout | |
6276 | * we don't need to do anything at the md level, the | |
6277 | * personality will take care of it all. | |
6278 | */ | |
50ac168a | 6279 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6280 | return -EINVAL; |
597a711b N |
6281 | else { |
6282 | mddev->new_layout = info->layout; | |
50ac168a | 6283 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6284 | if (rv) |
6285 | mddev->new_layout = mddev->layout; | |
6286 | return rv; | |
6287 | } | |
1da177e4 | 6288 | } |
58c0fed4 | 6289 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6290 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6291 | |
da943b99 N |
6292 | if (mddev->raid_disks != info->raid_disks) |
6293 | rv = update_raid_disks(mddev, info->raid_disks); | |
6294 | ||
36fa3063 | 6295 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
d66b1b39 | 6296 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) |
36fa3063 N |
6297 | return -EINVAL; |
6298 | if (mddev->recovery || mddev->sync_thread) | |
6299 | return -EBUSY; | |
6300 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
6301 | /* add the bitmap */ | |
6302 | if (mddev->bitmap) | |
6303 | return -EEXIST; | |
c3d9714e | 6304 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 6305 | return -EINVAL; |
c3d9714e N |
6306 | mddev->bitmap_info.offset = |
6307 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6308 | mddev->bitmap_info.space = |
6309 | mddev->bitmap_info.default_space; | |
36fa3063 N |
6310 | mddev->pers->quiesce(mddev, 1); |
6311 | rv = bitmap_create(mddev); | |
69e51b44 N |
6312 | if (!rv) |
6313 | rv = bitmap_load(mddev); | |
36fa3063 N |
6314 | if (rv) |
6315 | bitmap_destroy(mddev); | |
6316 | mddev->pers->quiesce(mddev, 0); | |
6317 | } else { | |
6318 | /* remove the bitmap */ | |
6319 | if (!mddev->bitmap) | |
6320 | return -ENOENT; | |
1ec885cd | 6321 | if (mddev->bitmap->storage.file) |
36fa3063 N |
6322 | return -EINVAL; |
6323 | mddev->pers->quiesce(mddev, 1); | |
6324 | bitmap_destroy(mddev); | |
6325 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6326 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6327 | } |
6328 | } | |
850b2b42 | 6329 | md_update_sb(mddev, 1); |
1da177e4 LT |
6330 | return rv; |
6331 | } | |
6332 | ||
fd01b88c | 6333 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6334 | { |
3cb03002 | 6335 | struct md_rdev *rdev; |
1ca69c4b | 6336 | int err = 0; |
1da177e4 LT |
6337 | |
6338 | if (mddev->pers == NULL) | |
6339 | return -ENODEV; | |
6340 | ||
1ca69c4b N |
6341 | rcu_read_lock(); |
6342 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6343 | if (!rdev) |
1ca69c4b N |
6344 | err = -ENODEV; |
6345 | else { | |
6346 | md_error(mddev, rdev); | |
6347 | if (!test_bit(Faulty, &rdev->flags)) | |
6348 | err = -EBUSY; | |
6349 | } | |
6350 | rcu_read_unlock(); | |
6351 | return err; | |
1da177e4 LT |
6352 | } |
6353 | ||
2f9618ce AN |
6354 | /* |
6355 | * We have a problem here : there is no easy way to give a CHS | |
6356 | * virtual geometry. We currently pretend that we have a 2 heads | |
6357 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6358 | * dosfs just mad... ;-) | |
6359 | */ | |
a885c8c4 CH |
6360 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6361 | { | |
fd01b88c | 6362 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6363 | |
6364 | geo->heads = 2; | |
6365 | geo->sectors = 4; | |
49ce6cea | 6366 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6367 | return 0; |
6368 | } | |
6369 | ||
cb335f88 NS |
6370 | static inline bool md_ioctl_valid(unsigned int cmd) |
6371 | { | |
6372 | switch (cmd) { | |
6373 | case ADD_NEW_DISK: | |
6374 | case BLKROSET: | |
6375 | case GET_ARRAY_INFO: | |
6376 | case GET_BITMAP_FILE: | |
6377 | case GET_DISK_INFO: | |
6378 | case HOT_ADD_DISK: | |
6379 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
6380 | case RAID_AUTORUN: |
6381 | case RAID_VERSION: | |
6382 | case RESTART_ARRAY_RW: | |
6383 | case RUN_ARRAY: | |
6384 | case SET_ARRAY_INFO: | |
6385 | case SET_BITMAP_FILE: | |
6386 | case SET_DISK_FAULTY: | |
6387 | case STOP_ARRAY: | |
6388 | case STOP_ARRAY_RO: | |
6389 | return true; | |
6390 | default: | |
6391 | return false; | |
6392 | } | |
6393 | } | |
6394 | ||
a39907fa | 6395 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6396 | unsigned int cmd, unsigned long arg) |
6397 | { | |
6398 | int err = 0; | |
6399 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6400 | struct mddev *mddev = NULL; |
e2218350 | 6401 | int ro; |
1da177e4 | 6402 | |
cb335f88 NS |
6403 | if (!md_ioctl_valid(cmd)) |
6404 | return -ENOTTY; | |
6405 | ||
506c9e44 N |
6406 | switch (cmd) { |
6407 | case RAID_VERSION: | |
6408 | case GET_ARRAY_INFO: | |
6409 | case GET_DISK_INFO: | |
6410 | break; | |
6411 | default: | |
6412 | if (!capable(CAP_SYS_ADMIN)) | |
6413 | return -EACCES; | |
6414 | } | |
1da177e4 LT |
6415 | |
6416 | /* | |
6417 | * Commands dealing with the RAID driver but not any | |
6418 | * particular array: | |
6419 | */ | |
c02c0aeb N |
6420 | switch (cmd) { |
6421 | case RAID_VERSION: | |
6422 | err = get_version(argp); | |
3adc28d8 | 6423 | goto out; |
1da177e4 | 6424 | |
1da177e4 | 6425 | #ifndef MODULE |
c02c0aeb N |
6426 | case RAID_AUTORUN: |
6427 | err = 0; | |
6428 | autostart_arrays(arg); | |
3adc28d8 | 6429 | goto out; |
1da177e4 | 6430 | #endif |
c02c0aeb | 6431 | default:; |
1da177e4 LT |
6432 | } |
6433 | ||
6434 | /* | |
6435 | * Commands creating/starting a new array: | |
6436 | */ | |
6437 | ||
a39907fa | 6438 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6439 | |
6440 | if (!mddev) { | |
6441 | BUG(); | |
3adc28d8 | 6442 | goto out; |
1da177e4 LT |
6443 | } |
6444 | ||
1ca69c4b N |
6445 | /* Some actions do not requires the mutex */ |
6446 | switch (cmd) { | |
6447 | case GET_ARRAY_INFO: | |
6448 | if (!mddev->raid_disks && !mddev->external) | |
6449 | err = -ENODEV; | |
6450 | else | |
6451 | err = get_array_info(mddev, argp); | |
3adc28d8 | 6452 | goto out; |
1ca69c4b N |
6453 | |
6454 | case GET_DISK_INFO: | |
6455 | if (!mddev->raid_disks && !mddev->external) | |
6456 | err = -ENODEV; | |
6457 | else | |
6458 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 6459 | goto out; |
1ca69c4b N |
6460 | |
6461 | case SET_DISK_FAULTY: | |
6462 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6463 | goto out; |
4af1a041 N |
6464 | |
6465 | case GET_BITMAP_FILE: | |
6466 | err = get_bitmap_file(mddev, argp); | |
6467 | goto out; | |
6468 | ||
1ca69c4b N |
6469 | } |
6470 | ||
a7a3f08d N |
6471 | if (cmd == ADD_NEW_DISK) |
6472 | /* need to ensure md_delayed_delete() has completed */ | |
6473 | flush_workqueue(md_misc_wq); | |
6474 | ||
90f5f7ad HR |
6475 | if (cmd == HOT_REMOVE_DISK) |
6476 | /* need to ensure recovery thread has run */ | |
6477 | wait_event_interruptible_timeout(mddev->sb_wait, | |
6478 | !test_bit(MD_RECOVERY_NEEDED, | |
6479 | &mddev->flags), | |
6480 | msecs_to_jiffies(5000)); | |
260fa034 N |
6481 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
6482 | /* Need to flush page cache, and ensure no-one else opens | |
6483 | * and writes | |
6484 | */ | |
6485 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 6486 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
6487 | mutex_unlock(&mddev->open_mutex); |
6488 | err = -EBUSY; | |
3adc28d8 | 6489 | goto out; |
260fa034 N |
6490 | } |
6491 | set_bit(MD_STILL_CLOSED, &mddev->flags); | |
6492 | mutex_unlock(&mddev->open_mutex); | |
6493 | sync_blockdev(bdev); | |
6494 | } | |
1da177e4 LT |
6495 | err = mddev_lock(mddev); |
6496 | if (err) { | |
f72ffdd6 | 6497 | printk(KERN_INFO |
1da177e4 LT |
6498 | "md: ioctl lock interrupted, reason %d, cmd %d\n", |
6499 | err, cmd); | |
3adc28d8 | 6500 | goto out; |
1da177e4 LT |
6501 | } |
6502 | ||
c02c0aeb N |
6503 | if (cmd == SET_ARRAY_INFO) { |
6504 | mdu_array_info_t info; | |
6505 | if (!arg) | |
6506 | memset(&info, 0, sizeof(info)); | |
6507 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6508 | err = -EFAULT; | |
3adc28d8 | 6509 | goto unlock; |
c02c0aeb N |
6510 | } |
6511 | if (mddev->pers) { | |
6512 | err = update_array_info(mddev, &info); | |
6513 | if (err) { | |
6514 | printk(KERN_WARNING "md: couldn't update" | |
6515 | " array info. %d\n", err); | |
3adc28d8 | 6516 | goto unlock; |
1da177e4 | 6517 | } |
3adc28d8 | 6518 | goto unlock; |
c02c0aeb N |
6519 | } |
6520 | if (!list_empty(&mddev->disks)) { | |
6521 | printk(KERN_WARNING | |
6522 | "md: array %s already has disks!\n", | |
6523 | mdname(mddev)); | |
6524 | err = -EBUSY; | |
3adc28d8 | 6525 | goto unlock; |
c02c0aeb N |
6526 | } |
6527 | if (mddev->raid_disks) { | |
6528 | printk(KERN_WARNING | |
6529 | "md: array %s already initialised!\n", | |
6530 | mdname(mddev)); | |
6531 | err = -EBUSY; | |
3adc28d8 | 6532 | goto unlock; |
c02c0aeb N |
6533 | } |
6534 | err = set_array_info(mddev, &info); | |
6535 | if (err) { | |
6536 | printk(KERN_WARNING "md: couldn't set" | |
6537 | " array info. %d\n", err); | |
3adc28d8 | 6538 | goto unlock; |
c02c0aeb | 6539 | } |
3adc28d8 | 6540 | goto unlock; |
1da177e4 LT |
6541 | } |
6542 | ||
6543 | /* | |
6544 | * Commands querying/configuring an existing array: | |
6545 | */ | |
32a7627c | 6546 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6547 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6548 | if ((!mddev->raid_disks && !mddev->external) |
6549 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6550 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6551 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 6552 | err = -ENODEV; |
3adc28d8 | 6553 | goto unlock; |
1da177e4 LT |
6554 | } |
6555 | ||
6556 | /* | |
6557 | * Commands even a read-only array can execute: | |
6558 | */ | |
c02c0aeb | 6559 | switch (cmd) { |
c02c0aeb N |
6560 | case RESTART_ARRAY_RW: |
6561 | err = restart_array(mddev); | |
3adc28d8 | 6562 | goto unlock; |
1da177e4 | 6563 | |
c02c0aeb N |
6564 | case STOP_ARRAY: |
6565 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 6566 | goto unlock; |
1da177e4 | 6567 | |
c02c0aeb N |
6568 | case STOP_ARRAY_RO: |
6569 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 6570 | goto unlock; |
1da177e4 | 6571 | |
3ea8929d N |
6572 | case HOT_REMOVE_DISK: |
6573 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6574 | goto unlock; |
3ea8929d | 6575 | |
7ceb17e8 N |
6576 | case ADD_NEW_DISK: |
6577 | /* We can support ADD_NEW_DISK on read-only arrays | |
6578 | * on if we are re-adding a preexisting device. | |
6579 | * So require mddev->pers and MD_DISK_SYNC. | |
6580 | */ | |
6581 | if (mddev->pers) { | |
6582 | mdu_disk_info_t info; | |
6583 | if (copy_from_user(&info, argp, sizeof(info))) | |
6584 | err = -EFAULT; | |
6585 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
6586 | /* Need to clear read-only for this */ | |
6587 | break; | |
6588 | else | |
6589 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6590 | goto unlock; |
7ceb17e8 N |
6591 | } |
6592 | break; | |
6593 | ||
c02c0aeb N |
6594 | case BLKROSET: |
6595 | if (get_user(ro, (int __user *)(arg))) { | |
6596 | err = -EFAULT; | |
3adc28d8 | 6597 | goto unlock; |
c02c0aeb N |
6598 | } |
6599 | err = -EINVAL; | |
e2218350 | 6600 | |
c02c0aeb N |
6601 | /* if the bdev is going readonly the value of mddev->ro |
6602 | * does not matter, no writes are coming | |
6603 | */ | |
6604 | if (ro) | |
3adc28d8 | 6605 | goto unlock; |
e2218350 | 6606 | |
c02c0aeb N |
6607 | /* are we are already prepared for writes? */ |
6608 | if (mddev->ro != 1) | |
3adc28d8 | 6609 | goto unlock; |
e2218350 | 6610 | |
c02c0aeb N |
6611 | /* transitioning to readauto need only happen for |
6612 | * arrays that call md_write_start | |
6613 | */ | |
6614 | if (mddev->pers) { | |
6615 | err = restart_array(mddev); | |
6616 | if (err == 0) { | |
6617 | mddev->ro = 2; | |
6618 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6619 | } |
c02c0aeb | 6620 | } |
3adc28d8 | 6621 | goto unlock; |
1da177e4 LT |
6622 | } |
6623 | ||
6624 | /* | |
6625 | * The remaining ioctls are changing the state of the | |
f91de92e | 6626 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 6627 | */ |
326eb17d | 6628 | if (mddev->ro && mddev->pers) { |
f91de92e N |
6629 | if (mddev->ro == 2) { |
6630 | mddev->ro = 0; | |
00bcb4ac | 6631 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 6632 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
6633 | /* mddev_unlock will wake thread */ |
6634 | /* If a device failed while we were read-only, we | |
6635 | * need to make sure the metadata is updated now. | |
6636 | */ | |
6637 | if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
6638 | mddev_unlock(mddev); | |
6639 | wait_event(mddev->sb_wait, | |
6640 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) && | |
6641 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
29f097c4 | 6642 | mddev_lock_nointr(mddev); |
f3378b48 | 6643 | } |
f91de92e N |
6644 | } else { |
6645 | err = -EROFS; | |
3adc28d8 | 6646 | goto unlock; |
f91de92e | 6647 | } |
1da177e4 LT |
6648 | } |
6649 | ||
c02c0aeb N |
6650 | switch (cmd) { |
6651 | case ADD_NEW_DISK: | |
1da177e4 | 6652 | { |
c02c0aeb N |
6653 | mdu_disk_info_t info; |
6654 | if (copy_from_user(&info, argp, sizeof(info))) | |
6655 | err = -EFAULT; | |
6656 | else | |
6657 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6658 | goto unlock; |
c02c0aeb | 6659 | } |
1da177e4 | 6660 | |
c02c0aeb N |
6661 | case HOT_ADD_DISK: |
6662 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6663 | goto unlock; |
1da177e4 | 6664 | |
c02c0aeb N |
6665 | case RUN_ARRAY: |
6666 | err = do_md_run(mddev); | |
3adc28d8 | 6667 | goto unlock; |
1da177e4 | 6668 | |
c02c0aeb N |
6669 | case SET_BITMAP_FILE: |
6670 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 6671 | goto unlock; |
32a7627c | 6672 | |
c02c0aeb N |
6673 | default: |
6674 | err = -EINVAL; | |
3adc28d8 | 6675 | goto unlock; |
1da177e4 LT |
6676 | } |
6677 | ||
3adc28d8 | 6678 | unlock: |
d3374825 N |
6679 | if (mddev->hold_active == UNTIL_IOCTL && |
6680 | err != -EINVAL) | |
6681 | mddev->hold_active = 0; | |
1da177e4 | 6682 | mddev_unlock(mddev); |
3adc28d8 | 6683 | out: |
1da177e4 LT |
6684 | return err; |
6685 | } | |
aa98aa31 AB |
6686 | #ifdef CONFIG_COMPAT |
6687 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6688 | unsigned int cmd, unsigned long arg) | |
6689 | { | |
6690 | switch (cmd) { | |
6691 | case HOT_REMOVE_DISK: | |
6692 | case HOT_ADD_DISK: | |
6693 | case SET_DISK_FAULTY: | |
6694 | case SET_BITMAP_FILE: | |
6695 | /* These take in integer arg, do not convert */ | |
6696 | break; | |
6697 | default: | |
6698 | arg = (unsigned long)compat_ptr(arg); | |
6699 | break; | |
6700 | } | |
6701 | ||
6702 | return md_ioctl(bdev, mode, cmd, arg); | |
6703 | } | |
6704 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6705 | |
a39907fa | 6706 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6707 | { |
6708 | /* | |
6709 | * Succeed if we can lock the mddev, which confirms that | |
6710 | * it isn't being stopped right now. | |
6711 | */ | |
fd01b88c | 6712 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6713 | int err; |
6714 | ||
0c098220 YL |
6715 | if (!mddev) |
6716 | return -ENODEV; | |
6717 | ||
d3374825 N |
6718 | if (mddev->gendisk != bdev->bd_disk) { |
6719 | /* we are racing with mddev_put which is discarding this | |
6720 | * bd_disk. | |
6721 | */ | |
6722 | mddev_put(mddev); | |
6723 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6724 | flush_workqueue(md_misc_wq); |
d3374825 N |
6725 | /* Then retry the open from the top */ |
6726 | return -ERESTARTSYS; | |
6727 | } | |
6728 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6729 | ||
c8c00a69 | 6730 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6731 | goto out; |
6732 | ||
6733 | err = 0; | |
f2ea68cf | 6734 | atomic_inc(&mddev->openers); |
260fa034 | 6735 | clear_bit(MD_STILL_CLOSED, &mddev->flags); |
c8c00a69 | 6736 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6737 | |
f0b4f7e2 | 6738 | check_disk_change(bdev); |
1da177e4 LT |
6739 | out: |
6740 | return err; | |
6741 | } | |
6742 | ||
db2a144b | 6743 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6744 | { |
f72ffdd6 | 6745 | struct mddev *mddev = disk->private_data; |
1da177e4 | 6746 | |
52e5f9d1 | 6747 | BUG_ON(!mddev); |
f2ea68cf | 6748 | atomic_dec(&mddev->openers); |
1da177e4 | 6749 | mddev_put(mddev); |
1da177e4 | 6750 | } |
f0b4f7e2 N |
6751 | |
6752 | static int md_media_changed(struct gendisk *disk) | |
6753 | { | |
fd01b88c | 6754 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6755 | |
6756 | return mddev->changed; | |
6757 | } | |
6758 | ||
6759 | static int md_revalidate(struct gendisk *disk) | |
6760 | { | |
fd01b88c | 6761 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6762 | |
6763 | mddev->changed = 0; | |
6764 | return 0; | |
6765 | } | |
83d5cde4 | 6766 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6767 | { |
6768 | .owner = THIS_MODULE, | |
a39907fa AV |
6769 | .open = md_open, |
6770 | .release = md_release, | |
b492b852 | 6771 | .ioctl = md_ioctl, |
aa98aa31 AB |
6772 | #ifdef CONFIG_COMPAT |
6773 | .compat_ioctl = md_compat_ioctl, | |
6774 | #endif | |
a885c8c4 | 6775 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6776 | .media_changed = md_media_changed, |
6777 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6778 | }; |
6779 | ||
f72ffdd6 | 6780 | static int md_thread(void *arg) |
1da177e4 | 6781 | { |
2b8bf345 | 6782 | struct md_thread *thread = arg; |
1da177e4 | 6783 | |
1da177e4 LT |
6784 | /* |
6785 | * md_thread is a 'system-thread', it's priority should be very | |
6786 | * high. We avoid resource deadlocks individually in each | |
6787 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6788 | * the very same RT priority as kswapd, thus we will never get | |
6789 | * into a priority inversion deadlock. | |
6790 | * | |
6791 | * we definitely have to have equal or higher priority than | |
6792 | * bdflush, otherwise bdflush will deadlock if there are too | |
6793 | * many dirty RAID5 blocks. | |
6794 | */ | |
1da177e4 | 6795 | |
6985c43f | 6796 | allow_signal(SIGKILL); |
a6fb0934 | 6797 | while (!kthread_should_stop()) { |
1da177e4 | 6798 | |
93588e22 N |
6799 | /* We need to wait INTERRUPTIBLE so that |
6800 | * we don't add to the load-average. | |
6801 | * That means we need to be sure no signals are | |
6802 | * pending | |
6803 | */ | |
6804 | if (signal_pending(current)) | |
6805 | flush_signals(current); | |
6806 | ||
6807 | wait_event_interruptible_timeout | |
6808 | (thread->wqueue, | |
6809 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6810 | || kthread_should_stop(), | |
6811 | thread->timeout); | |
1da177e4 | 6812 | |
6c987910 N |
6813 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6814 | if (!kthread_should_stop()) | |
4ed8731d | 6815 | thread->run(thread); |
1da177e4 | 6816 | } |
a6fb0934 | 6817 | |
1da177e4 LT |
6818 | return 0; |
6819 | } | |
6820 | ||
2b8bf345 | 6821 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
6822 | { |
6823 | if (thread) { | |
36a4e1fe | 6824 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
6825 | set_bit(THREAD_WAKEUP, &thread->flags); |
6826 | wake_up(&thread->wqueue); | |
6827 | } | |
6828 | } | |
6c144d31 | 6829 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 6830 | |
4ed8731d SL |
6831 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
6832 | struct mddev *mddev, const char *name) | |
1da177e4 | 6833 | { |
2b8bf345 | 6834 | struct md_thread *thread; |
1da177e4 | 6835 | |
2b8bf345 | 6836 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
6837 | if (!thread) |
6838 | return NULL; | |
6839 | ||
1da177e4 LT |
6840 | init_waitqueue_head(&thread->wqueue); |
6841 | ||
1da177e4 LT |
6842 | thread->run = run; |
6843 | thread->mddev = mddev; | |
32a7627c | 6844 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6845 | thread->tsk = kthread_run(md_thread, thread, |
6846 | "%s_%s", | |
6847 | mdname(thread->mddev), | |
0232605d | 6848 | name); |
a6fb0934 | 6849 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6850 | kfree(thread); |
6851 | return NULL; | |
6852 | } | |
1da177e4 LT |
6853 | return thread; |
6854 | } | |
6c144d31 | 6855 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 6856 | |
2b8bf345 | 6857 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 6858 | { |
2b8bf345 | 6859 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
6860 | if (!thread) |
6861 | return; | |
36a4e1fe | 6862 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
6863 | /* Locking ensures that mddev_unlock does not wake_up a |
6864 | * non-existent thread | |
6865 | */ | |
6866 | spin_lock(&pers_lock); | |
6867 | *threadp = NULL; | |
6868 | spin_unlock(&pers_lock); | |
a6fb0934 N |
6869 | |
6870 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6871 | kfree(thread); |
6872 | } | |
6c144d31 | 6873 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 6874 | |
fd01b88c | 6875 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 6876 | { |
b2d444d7 | 6877 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6878 | return; |
6bfe0b49 | 6879 | |
de393cde | 6880 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
6881 | return; |
6882 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6883 | if (mddev->degraded) |
6884 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6885 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6886 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6887 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6888 | md_wakeup_thread(mddev->thread); | |
768a418d | 6889 | if (mddev->event_work.func) |
e804ac78 | 6890 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6891 | md_new_event_inintr(mddev); |
1da177e4 | 6892 | } |
6c144d31 | 6893 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
6894 | |
6895 | /* seq_file implementation /proc/mdstat */ | |
6896 | ||
6897 | static void status_unused(struct seq_file *seq) | |
6898 | { | |
6899 | int i = 0; | |
3cb03002 | 6900 | struct md_rdev *rdev; |
1da177e4 LT |
6901 | |
6902 | seq_printf(seq, "unused devices: "); | |
6903 | ||
159ec1fc | 6904 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6905 | char b[BDEVNAME_SIZE]; |
6906 | i++; | |
6907 | seq_printf(seq, "%s ", | |
6908 | bdevname(rdev->bdev,b)); | |
6909 | } | |
6910 | if (!i) | |
6911 | seq_printf(seq, "<none>"); | |
6912 | ||
6913 | seq_printf(seq, "\n"); | |
6914 | } | |
6915 | ||
f72ffdd6 | 6916 | static void status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 6917 | { |
dd71cf6b N |
6918 | sector_t max_sectors, resync, res; |
6919 | unsigned long dt, db; | |
6920 | sector_t rt; | |
4588b42e N |
6921 | int scale; |
6922 | unsigned int per_milli; | |
1da177e4 | 6923 | |
72f36d59 N |
6924 | if (mddev->curr_resync <= 3) |
6925 | resync = 0; | |
6926 | else | |
6927 | resync = mddev->curr_resync | |
6928 | - atomic_read(&mddev->recovery_active); | |
1da177e4 | 6929 | |
c804cdec N |
6930 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
6931 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 6932 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6933 | else |
dd71cf6b | 6934 | max_sectors = mddev->dev_sectors; |
1da177e4 | 6935 | |
403df478 | 6936 | WARN_ON(max_sectors == 0); |
4588b42e | 6937 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6938 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6939 | * u32, as those are the requirements for sector_div. |
6940 | * Thus 'scale' must be at least 10 | |
6941 | */ | |
6942 | scale = 10; | |
6943 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6944 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6945 | scale++; |
6946 | } | |
6947 | res = (resync>>scale)*1000; | |
dd71cf6b | 6948 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6949 | |
6950 | per_milli = res; | |
1da177e4 | 6951 | { |
4588b42e | 6952 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6953 | seq_printf(seq, "["); |
6954 | for (i = 0; i < x; i++) | |
6955 | seq_printf(seq, "="); | |
6956 | seq_printf(seq, ">"); | |
6957 | for (i = 0; i < y; i++) | |
6958 | seq_printf(seq, "."); | |
6959 | seq_printf(seq, "] "); | |
6960 | } | |
4588b42e | 6961 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6962 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6963 | "reshape" : | |
61df9d91 N |
6964 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6965 | "check" : | |
6966 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6967 | "resync" : "recovery"))), | |
6968 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6969 | (unsigned long long) resync/2, |
6970 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6971 | |
6972 | /* | |
1da177e4 LT |
6973 | * dt: time from mark until now |
6974 | * db: blocks written from mark until now | |
6975 | * rt: remaining time | |
dd71cf6b N |
6976 | * |
6977 | * rt is a sector_t, so could be 32bit or 64bit. | |
6978 | * So we divide before multiply in case it is 32bit and close | |
6979 | * to the limit. | |
25985edc | 6980 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
6981 | * near the end of resync when the number of remaining sectors |
6982 | * is close to 'db'. | |
6983 | * We then divide rt by 32 after multiplying by db to compensate. | |
6984 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6985 | */ |
6986 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6987 | if (!dt) dt++; | |
ff4e8d9a N |
6988 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6989 | - mddev->resync_mark_cnt; | |
1da177e4 | 6990 | |
dd71cf6b N |
6991 | rt = max_sectors - resync; /* number of remaining sectors */ |
6992 | sector_div(rt, db/32+1); | |
6993 | rt *= dt; | |
6994 | rt >>= 5; | |
6995 | ||
6996 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6997 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6998 | |
ff4e8d9a | 6999 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
7000 | } |
7001 | ||
7002 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7003 | { | |
7004 | struct list_head *tmp; | |
7005 | loff_t l = *pos; | |
fd01b88c | 7006 | struct mddev *mddev; |
1da177e4 LT |
7007 | |
7008 | if (l >= 0x10000) | |
7009 | return NULL; | |
7010 | if (!l--) | |
7011 | /* header */ | |
7012 | return (void*)1; | |
7013 | ||
7014 | spin_lock(&all_mddevs_lock); | |
7015 | list_for_each(tmp,&all_mddevs) | |
7016 | if (!l--) { | |
fd01b88c | 7017 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7018 | mddev_get(mddev); |
7019 | spin_unlock(&all_mddevs_lock); | |
7020 | return mddev; | |
7021 | } | |
7022 | spin_unlock(&all_mddevs_lock); | |
7023 | if (!l--) | |
7024 | return (void*)2;/* tail */ | |
7025 | return NULL; | |
7026 | } | |
7027 | ||
7028 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7029 | { | |
7030 | struct list_head *tmp; | |
fd01b88c | 7031 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7032 | |
1da177e4 LT |
7033 | ++*pos; |
7034 | if (v == (void*)2) | |
7035 | return NULL; | |
7036 | ||
7037 | spin_lock(&all_mddevs_lock); | |
7038 | if (v == (void*)1) | |
7039 | tmp = all_mddevs.next; | |
7040 | else | |
7041 | tmp = mddev->all_mddevs.next; | |
7042 | if (tmp != &all_mddevs) | |
fd01b88c | 7043 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7044 | else { |
7045 | next_mddev = (void*)2; | |
7046 | *pos = 0x10000; | |
f72ffdd6 | 7047 | } |
1da177e4 LT |
7048 | spin_unlock(&all_mddevs_lock); |
7049 | ||
7050 | if (v != (void*)1) | |
7051 | mddev_put(mddev); | |
7052 | return next_mddev; | |
7053 | ||
7054 | } | |
7055 | ||
7056 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7057 | { | |
fd01b88c | 7058 | struct mddev *mddev = v; |
1da177e4 LT |
7059 | |
7060 | if (mddev && v != (void*)1 && v != (void*)2) | |
7061 | mddev_put(mddev); | |
7062 | } | |
7063 | ||
7064 | static int md_seq_show(struct seq_file *seq, void *v) | |
7065 | { | |
fd01b88c | 7066 | struct mddev *mddev = v; |
dd8ac336 | 7067 | sector_t sectors; |
3cb03002 | 7068 | struct md_rdev *rdev; |
1da177e4 LT |
7069 | |
7070 | if (v == (void*)1) { | |
84fc4b56 | 7071 | struct md_personality *pers; |
1da177e4 LT |
7072 | seq_printf(seq, "Personalities : "); |
7073 | spin_lock(&pers_lock); | |
2604b703 N |
7074 | list_for_each_entry(pers, &pers_list, list) |
7075 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7076 | |
7077 | spin_unlock(&pers_lock); | |
7078 | seq_printf(seq, "\n"); | |
f1514638 | 7079 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7080 | return 0; |
7081 | } | |
7082 | if (v == (void*)2) { | |
7083 | status_unused(seq); | |
7084 | return 0; | |
7085 | } | |
7086 | ||
36d091f4 | 7087 | spin_lock(&mddev->lock); |
1da177e4 LT |
7088 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7089 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7090 | mddev->pers ? "" : "in"); | |
7091 | if (mddev->pers) { | |
f91de92e | 7092 | if (mddev->ro==1) |
1da177e4 | 7093 | seq_printf(seq, " (read-only)"); |
f91de92e | 7094 | if (mddev->ro==2) |
52720ae7 | 7095 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7096 | seq_printf(seq, " %s", mddev->pers->name); |
7097 | } | |
7098 | ||
dd8ac336 | 7099 | sectors = 0; |
f97fcad3 N |
7100 | rcu_read_lock(); |
7101 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7102 | char b[BDEVNAME_SIZE]; |
7103 | seq_printf(seq, " %s[%d]", | |
7104 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7105 | if (test_bit(WriteMostly, &rdev->flags)) |
7106 | seq_printf(seq, "(W)"); | |
b2d444d7 | 7107 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7108 | seq_printf(seq, "(F)"); |
7109 | continue; | |
2d78f8c4 N |
7110 | } |
7111 | if (rdev->raid_disk < 0) | |
b325a32e | 7112 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7113 | if (test_bit(Replacement, &rdev->flags)) |
7114 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7115 | sectors += rdev->sectors; |
1da177e4 | 7116 | } |
f97fcad3 | 7117 | rcu_read_unlock(); |
1da177e4 LT |
7118 | |
7119 | if (!list_empty(&mddev->disks)) { | |
7120 | if (mddev->pers) | |
7121 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7122 | (unsigned long long) |
7123 | mddev->array_sectors / 2); | |
1da177e4 LT |
7124 | else |
7125 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7126 | (unsigned long long)sectors / 2); |
1da177e4 | 7127 | } |
1cd6bf19 N |
7128 | if (mddev->persistent) { |
7129 | if (mddev->major_version != 0 || | |
7130 | mddev->minor_version != 90) { | |
7131 | seq_printf(seq," super %d.%d", | |
7132 | mddev->major_version, | |
7133 | mddev->minor_version); | |
7134 | } | |
e691063a N |
7135 | } else if (mddev->external) |
7136 | seq_printf(seq, " super external:%s", | |
7137 | mddev->metadata_type); | |
7138 | else | |
1cd6bf19 | 7139 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7140 | |
7141 | if (mddev->pers) { | |
d710e138 | 7142 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7143 | seq_printf(seq, "\n "); |
8e1b39d6 N |
7144 | if (mddev->pers->sync_request) { |
7145 | if (mddev->curr_resync > 2) { | |
d710e138 | 7146 | status_resync(seq, mddev); |
8e1b39d6 | 7147 | seq_printf(seq, "\n "); |
72f36d59 | 7148 | } else if (mddev->curr_resync >= 1) |
8e1b39d6 N |
7149 | seq_printf(seq, "\tresync=DELAYED\n "); |
7150 | else if (mddev->recovery_cp < MaxSector) | |
7151 | seq_printf(seq, "\tresync=PENDING\n "); | |
7152 | } | |
32a7627c N |
7153 | } else |
7154 | seq_printf(seq, "\n "); | |
7155 | ||
57148964 | 7156 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7157 | |
7158 | seq_printf(seq, "\n"); | |
7159 | } | |
36d091f4 | 7160 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7161 | |
1da177e4 LT |
7162 | return 0; |
7163 | } | |
7164 | ||
110518bc | 7165 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7166 | .start = md_seq_start, |
7167 | .next = md_seq_next, | |
7168 | .stop = md_seq_stop, | |
7169 | .show = md_seq_show, | |
7170 | }; | |
7171 | ||
7172 | static int md_seq_open(struct inode *inode, struct file *file) | |
7173 | { | |
f1514638 | 7174 | struct seq_file *seq; |
1da177e4 LT |
7175 | int error; |
7176 | ||
7177 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7178 | if (error) |
f1514638 KS |
7179 | return error; |
7180 | ||
7181 | seq = file->private_data; | |
7182 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7183 | return error; |
7184 | } | |
7185 | ||
e2f23b60 | 7186 | static int md_unloading; |
d7603b7e N |
7187 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7188 | { | |
f1514638 | 7189 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7190 | int mask; |
7191 | ||
e2f23b60 | 7192 | if (md_unloading) |
7d7e64f2 | 7193 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI; |
d7603b7e N |
7194 | poll_wait(filp, &md_event_waiters, wait); |
7195 | ||
7196 | /* always allow read */ | |
7197 | mask = POLLIN | POLLRDNORM; | |
7198 | ||
f1514638 | 7199 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7200 | mask |= POLLERR | POLLPRI; |
7201 | return mask; | |
7202 | } | |
7203 | ||
fa027c2a | 7204 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7205 | .owner = THIS_MODULE, |
1da177e4 LT |
7206 | .open = md_seq_open, |
7207 | .read = seq_read, | |
7208 | .llseek = seq_lseek, | |
c3f94b40 | 7209 | .release = seq_release_private, |
d7603b7e | 7210 | .poll = mdstat_poll, |
1da177e4 LT |
7211 | }; |
7212 | ||
84fc4b56 | 7213 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7214 | { |
50bd3774 CY |
7215 | printk(KERN_INFO "md: %s personality registered for level %d\n", |
7216 | p->name, p->level); | |
1da177e4 | 7217 | spin_lock(&pers_lock); |
2604b703 | 7218 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7219 | spin_unlock(&pers_lock); |
7220 | return 0; | |
7221 | } | |
6c144d31 | 7222 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7223 | |
84fc4b56 | 7224 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7225 | { |
2604b703 | 7226 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 7227 | spin_lock(&pers_lock); |
2604b703 | 7228 | list_del_init(&p->list); |
1da177e4 LT |
7229 | spin_unlock(&pers_lock); |
7230 | return 0; | |
7231 | } | |
6c144d31 | 7232 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7233 | |
fd01b88c | 7234 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7235 | { |
f72ffdd6 | 7236 | struct md_rdev *rdev; |
1da177e4 | 7237 | int idle; |
eea1bf38 | 7238 | int curr_events; |
1da177e4 LT |
7239 | |
7240 | idle = 1; | |
4b80991c N |
7241 | rcu_read_lock(); |
7242 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7243 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7244 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7245 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7246 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7247 | /* sync IO will cause sync_io to increase before the disk_stats |
7248 | * as sync_io is counted when a request starts, and | |
7249 | * disk_stats is counted when it completes. | |
7250 | * So resync activity will cause curr_events to be smaller than | |
7251 | * when there was no such activity. | |
7252 | * non-sync IO will cause disk_stat to increase without | |
7253 | * increasing sync_io so curr_events will (eventually) | |
7254 | * be larger than it was before. Once it becomes | |
7255 | * substantially larger, the test below will cause | |
7256 | * the array to appear non-idle, and resync will slow | |
7257 | * down. | |
7258 | * If there is a lot of outstanding resync activity when | |
7259 | * we set last_event to curr_events, then all that activity | |
7260 | * completing might cause the array to appear non-idle | |
7261 | * and resync will be slowed down even though there might | |
7262 | * not have been non-resync activity. This will only | |
7263 | * happen once though. 'last_events' will soon reflect | |
7264 | * the state where there is little or no outstanding | |
7265 | * resync requests, and further resync activity will | |
7266 | * always make curr_events less than last_events. | |
c0e48521 | 7267 | * |
1da177e4 | 7268 | */ |
eea1bf38 | 7269 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7270 | rdev->last_events = curr_events; |
7271 | idle = 0; | |
7272 | } | |
7273 | } | |
4b80991c | 7274 | rcu_read_unlock(); |
1da177e4 LT |
7275 | return idle; |
7276 | } | |
7277 | ||
fd01b88c | 7278 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7279 | { |
7280 | /* another "blocks" (512byte) blocks have been synced */ | |
7281 | atomic_sub(blocks, &mddev->recovery_active); | |
7282 | wake_up(&mddev->recovery_wait); | |
7283 | if (!ok) { | |
dfc70645 | 7284 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7285 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7286 | md_wakeup_thread(mddev->thread); |
7287 | // stop recovery, signal do_sync .... | |
7288 | } | |
7289 | } | |
6c144d31 | 7290 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 7291 | |
06d91a5f N |
7292 | /* md_write_start(mddev, bi) |
7293 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7294 | * in superblock) before writing, schedule a superblock update |
7295 | * and wait for it to complete. | |
06d91a5f | 7296 | */ |
fd01b88c | 7297 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7298 | { |
0fd62b86 | 7299 | int did_change = 0; |
06d91a5f | 7300 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7301 | return; |
06d91a5f | 7302 | |
f91de92e N |
7303 | BUG_ON(mddev->ro == 1); |
7304 | if (mddev->ro == 2) { | |
7305 | /* need to switch to read/write */ | |
7306 | mddev->ro = 0; | |
7307 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7308 | md_wakeup_thread(mddev->thread); | |
25156198 | 7309 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7310 | did_change = 1; |
f91de92e | 7311 | } |
06d91a5f | 7312 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7313 | if (mddev->safemode == 1) |
7314 | mddev->safemode = 0; | |
06d91a5f | 7315 | if (mddev->in_sync) { |
85572d7c | 7316 | spin_lock(&mddev->lock); |
3d310eb7 N |
7317 | if (mddev->in_sync) { |
7318 | mddev->in_sync = 0; | |
850b2b42 | 7319 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 7320 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 7321 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7322 | did_change = 1; |
3d310eb7 | 7323 | } |
85572d7c | 7324 | spin_unlock(&mddev->lock); |
06d91a5f | 7325 | } |
0fd62b86 | 7326 | if (did_change) |
00bcb4ac | 7327 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7328 | wait_event(mddev->sb_wait, |
09a44cc1 | 7329 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 | 7330 | } |
6c144d31 | 7331 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 7332 | |
fd01b88c | 7333 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7334 | { |
7335 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7336 | if (mddev->safemode == 2) | |
7337 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7338 | else if (mddev->safemode_delay) |
1da177e4 LT |
7339 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7340 | } | |
7341 | } | |
6c144d31 | 7342 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 7343 | |
2a2275d6 N |
7344 | /* md_allow_write(mddev) |
7345 | * Calling this ensures that the array is marked 'active' so that writes | |
7346 | * may proceed without blocking. It is important to call this before | |
7347 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7348 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
7349 | * |
7350 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
7351 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 7352 | */ |
fd01b88c | 7353 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7354 | { |
7355 | if (!mddev->pers) | |
b5470dc5 | 7356 | return 0; |
2a2275d6 | 7357 | if (mddev->ro) |
b5470dc5 | 7358 | return 0; |
1a0fd497 | 7359 | if (!mddev->pers->sync_request) |
b5470dc5 | 7360 | return 0; |
2a2275d6 | 7361 | |
85572d7c | 7362 | spin_lock(&mddev->lock); |
2a2275d6 N |
7363 | if (mddev->in_sync) { |
7364 | mddev->in_sync = 0; | |
7365 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 7366 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
7367 | if (mddev->safemode_delay && |
7368 | mddev->safemode == 0) | |
7369 | mddev->safemode = 1; | |
85572d7c | 7370 | spin_unlock(&mddev->lock); |
2a2275d6 | 7371 | md_update_sb(mddev, 0); |
00bcb4ac | 7372 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 | 7373 | } else |
85572d7c | 7374 | spin_unlock(&mddev->lock); |
b5470dc5 | 7375 | |
070dc6dd | 7376 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
7377 | return -EAGAIN; |
7378 | else | |
7379 | return 0; | |
2a2275d6 N |
7380 | } |
7381 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7382 | ||
1da177e4 LT |
7383 | #define SYNC_MARKS 10 |
7384 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7385 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7386 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7387 | { |
4ed8731d | 7388 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7389 | struct mddev *mddev2; |
1da177e4 LT |
7390 | unsigned int currspeed = 0, |
7391 | window; | |
ac7e50a3 | 7392 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 7393 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7394 | unsigned long update_time; |
1da177e4 LT |
7395 | sector_t mark_cnt[SYNC_MARKS]; |
7396 | int last_mark,m; | |
7397 | struct list_head *tmp; | |
7398 | sector_t last_check; | |
57afd89f | 7399 | int skipped = 0; |
3cb03002 | 7400 | struct md_rdev *rdev; |
c4a39551 | 7401 | char *desc, *action = NULL; |
7c2c57c9 | 7402 | struct blk_plug plug; |
1da177e4 LT |
7403 | |
7404 | /* just incase thread restarts... */ | |
7405 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7406 | return; | |
3991b31e N |
7407 | if (mddev->ro) {/* never try to sync a read-only array */ |
7408 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 7409 | return; |
3991b31e | 7410 | } |
1da177e4 | 7411 | |
61df9d91 | 7412 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 7413 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 7414 | desc = "data-check"; |
c4a39551 JB |
7415 | action = "check"; |
7416 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 7417 | desc = "requested-resync"; |
c4a39551 JB |
7418 | action = "repair"; |
7419 | } else | |
61df9d91 N |
7420 | desc = "resync"; |
7421 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7422 | desc = "reshape"; | |
7423 | else | |
7424 | desc = "recovery"; | |
7425 | ||
c4a39551 JB |
7426 | mddev->last_sync_action = action ?: desc; |
7427 | ||
1da177e4 LT |
7428 | /* we overload curr_resync somewhat here. |
7429 | * 0 == not engaged in resync at all | |
7430 | * 2 == checking that there is no conflict with another sync | |
7431 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7432 | * commense | |
7433 | * other == active in resync - this many blocks | |
7434 | * | |
7435 | * Before starting a resync we must have set curr_resync to | |
7436 | * 2, and then checked that every "conflicting" array has curr_resync | |
7437 | * less than ours. When we find one that is the same or higher | |
7438 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7439 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7440 | * This will mean we have to start checking from the beginning again. | |
7441 | * | |
7442 | */ | |
7443 | ||
7444 | do { | |
7445 | mddev->curr_resync = 2; | |
7446 | ||
7447 | try_again: | |
404e4b43 | 7448 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 7449 | goto skip; |
29ac4aa3 | 7450 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7451 | if (mddev2 == mddev) |
7452 | continue; | |
90b08710 BS |
7453 | if (!mddev->parallel_resync |
7454 | && mddev2->curr_resync | |
7455 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7456 | DEFINE_WAIT(wq); |
7457 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7458 | /* arbitrarily yield */ | |
7459 | mddev->curr_resync = 1; | |
7460 | wake_up(&resync_wait); | |
7461 | } | |
7462 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7463 | /* no need to wait here, we can wait the next | |
7464 | * time 'round when curr_resync == 2 | |
7465 | */ | |
7466 | continue; | |
9744197c N |
7467 | /* We need to wait 'interruptible' so as not to |
7468 | * contribute to the load average, and not to | |
7469 | * be caught by 'softlockup' | |
7470 | */ | |
7471 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 7472 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 7473 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
7474 | printk(KERN_INFO "md: delaying %s of %s" |
7475 | " until %s has finished (they" | |
1da177e4 | 7476 | " share one or more physical units)\n", |
61df9d91 | 7477 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 7478 | mddev_put(mddev2); |
9744197c N |
7479 | if (signal_pending(current)) |
7480 | flush_signals(current); | |
1da177e4 LT |
7481 | schedule(); |
7482 | finish_wait(&resync_wait, &wq); | |
7483 | goto try_again; | |
7484 | } | |
7485 | finish_wait(&resync_wait, &wq); | |
7486 | } | |
7487 | } | |
7488 | } while (mddev->curr_resync < 2); | |
7489 | ||
5fd6c1dc | 7490 | j = 0; |
9d88883e | 7491 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7492 | /* resync follows the size requested by the personality, |
57afd89f | 7493 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7494 | */ |
7495 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7496 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7497 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7498 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7499 | j = mddev->resync_min; | |
7500 | else if (!mddev->bitmap) | |
5fd6c1dc | 7501 | j = mddev->recovery_cp; |
5e96ee65 | 7502 | |
ccfcc3c1 | 7503 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7504 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7505 | else { |
1da177e4 | 7506 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7507 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7508 | j = MaxSector; |
4e59ca7d | 7509 | rcu_read_lock(); |
dafb20fa | 7510 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc N |
7511 | if (rdev->raid_disk >= 0 && |
7512 | !test_bit(Faulty, &rdev->flags) && | |
7513 | !test_bit(In_sync, &rdev->flags) && | |
7514 | rdev->recovery_offset < j) | |
7515 | j = rdev->recovery_offset; | |
4e59ca7d | 7516 | rcu_read_unlock(); |
133d4527 N |
7517 | |
7518 | /* If there is a bitmap, we need to make sure all | |
7519 | * writes that started before we added a spare | |
7520 | * complete before we start doing a recovery. | |
7521 | * Otherwise the write might complete and (via | |
7522 | * bitmap_endwrite) set a bit in the bitmap after the | |
7523 | * recovery has checked that bit and skipped that | |
7524 | * region. | |
7525 | */ | |
7526 | if (mddev->bitmap) { | |
7527 | mddev->pers->quiesce(mddev, 1); | |
7528 | mddev->pers->quiesce(mddev, 0); | |
7529 | } | |
5fd6c1dc | 7530 | } |
1da177e4 | 7531 | |
61df9d91 N |
7532 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
7533 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
7534 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 7535 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
7536 | "(but not more than %d KB/sec) for %s.\n", |
7537 | speed_max(mddev), desc); | |
1da177e4 | 7538 | |
eea1bf38 | 7539 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 7540 | |
57afd89f | 7541 | io_sectors = 0; |
1da177e4 LT |
7542 | for (m = 0; m < SYNC_MARKS; m++) { |
7543 | mark[m] = jiffies; | |
57afd89f | 7544 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
7545 | } |
7546 | last_mark = 0; | |
7547 | mddev->resync_mark = mark[last_mark]; | |
7548 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
7549 | ||
7550 | /* | |
7551 | * Tune reconstruction: | |
7552 | */ | |
7553 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
7554 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
7555 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
7556 | |
7557 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
7558 | last_check = 0; |
7559 | ||
7560 | if (j>2) { | |
c91abf5a | 7561 | printk(KERN_INFO |
61df9d91 N |
7562 | "md: resuming %s of %s from checkpoint.\n", |
7563 | desc, mdname(mddev)); | |
1da177e4 | 7564 | mddev->curr_resync = j; |
72f36d59 N |
7565 | } else |
7566 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 7567 | mddev->curr_resync_completed = j; |
72f36d59 N |
7568 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
7569 | md_new_event(mddev); | |
54f89341 | 7570 | update_time = jiffies; |
1da177e4 | 7571 | |
7c2c57c9 | 7572 | blk_start_plug(&plug); |
1da177e4 | 7573 | while (j < max_sectors) { |
57afd89f | 7574 | sector_t sectors; |
1da177e4 | 7575 | |
57afd89f | 7576 | skipped = 0; |
97e4f42d | 7577 | |
7a91ee1f N |
7578 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7579 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
7580 | (mddev->curr_resync - mddev->curr_resync_completed) | |
7581 | > (max_sectors >> 4)) || | |
54f89341 | 7582 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f N |
7583 | (j - mddev->curr_resync_completed)*2 |
7584 | >= mddev->resync_max - mddev->curr_resync_completed | |
7585 | )) { | |
97e4f42d | 7586 | /* time to update curr_resync_completed */ |
97e4f42d N |
7587 | wait_event(mddev->recovery_wait, |
7588 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 7589 | mddev->curr_resync_completed = j; |
35d78c66 | 7590 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
7591 | j > mddev->recovery_cp) | |
7592 | mddev->recovery_cp = j; | |
54f89341 | 7593 | update_time = jiffies; |
070dc6dd | 7594 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 7595 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 7596 | } |
acb180b0 | 7597 | |
c91abf5a N |
7598 | while (j >= mddev->resync_max && |
7599 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
7600 | /* As this condition is controlled by user-space, |
7601 | * we can block indefinitely, so use '_interruptible' | |
7602 | * to avoid triggering warnings. | |
7603 | */ | |
7604 | flush_signals(current); /* just in case */ | |
7605 | wait_event_interruptible(mddev->recovery_wait, | |
7606 | mddev->resync_max > j | |
c91abf5a N |
7607 | || test_bit(MD_RECOVERY_INTR, |
7608 | &mddev->recovery)); | |
e62e58a5 | 7609 | } |
acb180b0 | 7610 | |
c91abf5a N |
7611 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7612 | break; | |
acb180b0 | 7613 | |
57afd89f | 7614 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 7615 | currspeed < speed_min(mddev)); |
57afd89f | 7616 | if (sectors == 0) { |
dfc70645 | 7617 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 7618 | break; |
1da177e4 | 7619 | } |
57afd89f N |
7620 | |
7621 | if (!skipped) { /* actual IO requested */ | |
7622 | io_sectors += sectors; | |
7623 | atomic_add(sectors, &mddev->recovery_active); | |
7624 | } | |
7625 | ||
e875ecea N |
7626 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7627 | break; | |
7628 | ||
1da177e4 | 7629 | j += sectors; |
72f36d59 N |
7630 | if (j > 2) |
7631 | mddev->curr_resync = j; | |
ff4e8d9a | 7632 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 7633 | if (last_check == 0) |
e875ecea | 7634 | /* this is the earliest that rebuild will be |
d7603b7e N |
7635 | * visible in /proc/mdstat |
7636 | */ | |
7637 | md_new_event(mddev); | |
57afd89f N |
7638 | |
7639 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
7640 | continue; |
7641 | ||
57afd89f | 7642 | last_check = io_sectors; |
1da177e4 LT |
7643 | repeat: |
7644 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
7645 | /* step marks */ | |
7646 | int next = (last_mark+1) % SYNC_MARKS; | |
7647 | ||
7648 | mddev->resync_mark = mark[next]; | |
7649 | mddev->resync_mark_cnt = mark_cnt[next]; | |
7650 | mark[next] = jiffies; | |
57afd89f | 7651 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
7652 | last_mark = next; |
7653 | } | |
7654 | ||
c91abf5a N |
7655 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7656 | break; | |
1da177e4 LT |
7657 | |
7658 | /* | |
7659 | * this loop exits only if either when we are slower than | |
7660 | * the 'hard' speed limit, or the system was IO-idle for | |
7661 | * a jiffy. | |
7662 | * the system might be non-idle CPU-wise, but we only care | |
7663 | * about not overloading the IO subsystem. (things like an | |
7664 | * e2fsck being done on the RAID array should execute fast) | |
7665 | */ | |
1da177e4 LT |
7666 | cond_resched(); |
7667 | ||
ac7e50a3 XN |
7668 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
7669 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 7670 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 7671 | |
88202a0c N |
7672 | if (currspeed > speed_min(mddev)) { |
7673 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 7674 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 7675 | msleep(500); |
1da177e4 LT |
7676 | goto repeat; |
7677 | } | |
7678 | } | |
7679 | } | |
c91abf5a N |
7680 | printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc, |
7681 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
7682 | ? "interrupted" : "done"); | |
1da177e4 LT |
7683 | /* |
7684 | * this also signals 'finished resyncing' to md_stop | |
7685 | */ | |
7c2c57c9 | 7686 | blk_finish_plug(&plug); |
1da177e4 LT |
7687 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7688 | ||
7689 | /* tell personality that we are finished */ | |
57afd89f | 7690 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 7691 | |
dfc70645 | 7692 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7693 | mddev->curr_resync > 2) { |
7694 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7695 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7696 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7697 | printk(KERN_INFO | |
61df9d91 N |
7698 | "md: checkpointing %s of %s.\n", |
7699 | desc, mdname(mddev)); | |
0a19caab | 7700 | if (test_bit(MD_RECOVERY_ERROR, |
7701 | &mddev->recovery)) | |
7702 | mddev->recovery_cp = | |
7703 | mddev->curr_resync_completed; | |
7704 | else | |
7705 | mddev->recovery_cp = | |
7706 | mddev->curr_resync; | |
5fd6c1dc N |
7707 | } |
7708 | } else | |
7709 | mddev->recovery_cp = MaxSector; | |
7710 | } else { | |
7711 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7712 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 7713 | rcu_read_lock(); |
dafb20fa | 7714 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7715 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7716 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7717 | !test_bit(Faulty, &rdev->flags) && |
7718 | !test_bit(In_sync, &rdev->flags) && | |
7719 | rdev->recovery_offset < mddev->curr_resync) | |
7720 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7721 | rcu_read_unlock(); |
5fd6c1dc | 7722 | } |
1da177e4 | 7723 | } |
db91ff55 | 7724 | skip: |
17571284 | 7725 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7726 | |
23da422b | 7727 | spin_lock(&mddev->lock); |
c07b70ad N |
7728 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7729 | /* We completed so min/max setting can be forgotten if used. */ | |
7730 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7731 | mddev->resync_min = 0; | |
7732 | mddev->resync_max = MaxSector; | |
7733 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7734 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 | 7735 | mddev->curr_resync = 0; |
23da422b N |
7736 | spin_unlock(&mddev->lock); |
7737 | ||
1da177e4 LT |
7738 | wake_up(&resync_wait); |
7739 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
7740 | md_wakeup_thread(mddev->thread); | |
c6207277 | 7741 | return; |
1da177e4 | 7742 | } |
29269553 | 7743 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 7744 | |
746d3207 N |
7745 | static int remove_and_add_spares(struct mddev *mddev, |
7746 | struct md_rdev *this) | |
b4c4c7b8 | 7747 | { |
3cb03002 | 7748 | struct md_rdev *rdev; |
b4c4c7b8 | 7749 | int spares = 0; |
f2a371c5 | 7750 | int removed = 0; |
b4c4c7b8 | 7751 | |
dafb20fa | 7752 | rdev_for_each(rdev, mddev) |
746d3207 N |
7753 | if ((this == NULL || rdev == this) && |
7754 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 7755 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7756 | (test_bit(Faulty, &rdev->flags) || |
7757 | ! test_bit(In_sync, &rdev->flags)) && | |
7758 | atomic_read(&rdev->nr_pending)==0) { | |
7759 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 7760 | mddev, rdev) == 0) { |
36fad858 | 7761 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 7762 | rdev->raid_disk = -1; |
f2a371c5 | 7763 | removed++; |
b4c4c7b8 N |
7764 | } |
7765 | } | |
90584fc9 JB |
7766 | if (removed && mddev->kobj.sd) |
7767 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 7768 | |
746d3207 N |
7769 | if (this) |
7770 | goto no_add; | |
7771 | ||
dafb20fa | 7772 | rdev_for_each(rdev, mddev) { |
7bfec5f3 N |
7773 | if (rdev->raid_disk >= 0 && |
7774 | !test_bit(In_sync, &rdev->flags) && | |
7775 | !test_bit(Faulty, &rdev->flags)) | |
7776 | spares++; | |
7ceb17e8 N |
7777 | if (rdev->raid_disk >= 0) |
7778 | continue; | |
7779 | if (test_bit(Faulty, &rdev->flags)) | |
7780 | continue; | |
7781 | if (mddev->ro && | |
8313b8e5 N |
7782 | ! (rdev->saved_raid_disk >= 0 && |
7783 | !test_bit(Bitmap_sync, &rdev->flags))) | |
7ceb17e8 N |
7784 | continue; |
7785 | ||
7eb41885 N |
7786 | if (rdev->saved_raid_disk < 0) |
7787 | rdev->recovery_offset = 0; | |
7ceb17e8 N |
7788 | if (mddev->pers-> |
7789 | hot_add_disk(mddev, rdev) == 0) { | |
7790 | if (sysfs_link_rdev(mddev, rdev)) | |
7791 | /* failure here is OK */; | |
7792 | spares++; | |
7793 | md_new_event(mddev); | |
7794 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
dfc70645 | 7795 | } |
b4c4c7b8 | 7796 | } |
746d3207 | 7797 | no_add: |
6dafab6b N |
7798 | if (removed) |
7799 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 N |
7800 | return spares; |
7801 | } | |
7ebc0be7 | 7802 | |
ac05f256 N |
7803 | static void md_start_sync(struct work_struct *ws) |
7804 | { | |
7805 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
7806 | ||
7807 | mddev->sync_thread = md_register_thread(md_do_sync, | |
7808 | mddev, | |
7809 | "resync"); | |
7810 | if (!mddev->sync_thread) { | |
7811 | printk(KERN_ERR "%s: could not start resync" | |
7812 | " thread...\n", | |
7813 | mdname(mddev)); | |
7814 | /* leave the spares where they are, it shouldn't hurt */ | |
7815 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7816 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7817 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7818 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7819 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 7820 | wake_up(&resync_wait); |
ac05f256 N |
7821 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
7822 | &mddev->recovery)) | |
7823 | if (mddev->sysfs_action) | |
7824 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7825 | } else | |
7826 | md_wakeup_thread(mddev->sync_thread); | |
7827 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7828 | md_new_event(mddev); | |
7829 | } | |
7830 | ||
1da177e4 LT |
7831 | /* |
7832 | * This routine is regularly called by all per-raid-array threads to | |
7833 | * deal with generic issues like resync and super-block update. | |
7834 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7835 | * need this as they never do any recovery or update the superblock. | |
7836 | * | |
7837 | * It does not do any resync itself, but rather "forks" off other threads | |
7838 | * to do that as needed. | |
7839 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7840 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7841 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7842 | * and wakeups up this thread which will reap the thread and finish up. |
7843 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7844 | * | |
7845 | * The overall approach is: | |
7846 | * 1/ if the superblock needs updating, update it. | |
7847 | * 2/ If a recovery thread is running, don't do anything else. | |
7848 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7849 | * 4/ If there are any faulty devices, remove them. | |
7850 | * 5/ If array is degraded, try to add spares devices | |
7851 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7852 | */ | |
fd01b88c | 7853 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 7854 | { |
68866e42 JB |
7855 | if (mddev->suspended) |
7856 | return; | |
7857 | ||
5f40402d | 7858 | if (mddev->bitmap) |
aa5cbd10 | 7859 | bitmap_daemon_work(mddev); |
1da177e4 | 7860 | |
fca4d848 | 7861 | if (signal_pending(current)) { |
31a59e34 | 7862 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7863 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7864 | mdname(mddev)); | |
7865 | mddev->safemode = 2; | |
7866 | } | |
7867 | flush_signals(current); | |
7868 | } | |
7869 | ||
c89a8eee N |
7870 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7871 | return; | |
1da177e4 | 7872 | if ( ! ( |
142d44c3 | 7873 | (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7874 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7875 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7876 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7877 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7878 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7879 | )) |
7880 | return; | |
fca4d848 | 7881 | |
df5b89b3 | 7882 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7883 | int spares = 0; |
fca4d848 | 7884 | |
c89a8eee | 7885 | if (mddev->ro) { |
7ceb17e8 N |
7886 | /* On a read-only array we can: |
7887 | * - remove failed devices | |
7888 | * - add already-in_sync devices if the array itself | |
7889 | * is in-sync. | |
7890 | * As we only add devices that are already in-sync, | |
7891 | * we can activate the spares immediately. | |
c89a8eee | 7892 | */ |
7ceb17e8 | 7893 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
7894 | /* There is no thread, but we need to call |
7895 | * ->spare_active and clear saved_raid_disk | |
7896 | */ | |
2ac295a5 | 7897 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 N |
7898 | md_reap_sync_thread(mddev); |
7899 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
c89a8eee N |
7900 | goto unlock; |
7901 | } | |
7902 | ||
31a59e34 | 7903 | if (!mddev->external) { |
0fd62b86 | 7904 | int did_change = 0; |
85572d7c | 7905 | spin_lock(&mddev->lock); |
31a59e34 N |
7906 | if (mddev->safemode && |
7907 | !atomic_read(&mddev->writes_pending) && | |
7908 | !mddev->in_sync && | |
7909 | mddev->recovery_cp == MaxSector) { | |
7910 | mddev->in_sync = 1; | |
0fd62b86 | 7911 | did_change = 1; |
070dc6dd | 7912 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7913 | } |
7914 | if (mddev->safemode == 1) | |
7915 | mddev->safemode = 0; | |
85572d7c | 7916 | spin_unlock(&mddev->lock); |
0fd62b86 | 7917 | if (did_change) |
00bcb4ac | 7918 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7919 | } |
fca4d848 | 7920 | |
7a0a5355 | 7921 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 7922 | md_update_sb(mddev, 0); |
06d91a5f | 7923 | |
1da177e4 LT |
7924 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7925 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7926 | /* resync/recovery still happening */ | |
7927 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7928 | goto unlock; | |
7929 | } | |
7930 | if (mddev->sync_thread) { | |
a91d5ac0 | 7931 | md_reap_sync_thread(mddev); |
1da177e4 LT |
7932 | goto unlock; |
7933 | } | |
72a23c21 NB |
7934 | /* Set RUNNING before clearing NEEDED to avoid |
7935 | * any transients in the value of "sync_action". | |
7936 | */ | |
72f36d59 | 7937 | mddev->curr_resync_completed = 0; |
23da422b | 7938 | spin_lock(&mddev->lock); |
72a23c21 | 7939 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 7940 | spin_unlock(&mddev->lock); |
24dd469d N |
7941 | /* Clear some bits that don't mean anything, but |
7942 | * might be left set | |
7943 | */ | |
24dd469d N |
7944 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7945 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7946 | |
ed209584 N |
7947 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
7948 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 7949 | goto not_running; |
1da177e4 LT |
7950 | /* no recovery is running. |
7951 | * remove any failed drives, then | |
7952 | * add spares if possible. | |
72f36d59 | 7953 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
7954 | * the personality to fail the re-add. |
7955 | */ | |
1da177e4 | 7956 | |
b4c4c7b8 | 7957 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7958 | if (mddev->pers->check_reshape == NULL || |
7959 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 7960 | /* Cannot proceed */ |
ac05f256 | 7961 | goto not_running; |
b4c4c7b8 | 7962 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 7963 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 7964 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
7965 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7966 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7967 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7968 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7969 | } else if (mddev->recovery_cp < MaxSector) { |
7970 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7971 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7972 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7973 | /* nothing to be done ... */ | |
ac05f256 | 7974 | goto not_running; |
24dd469d | 7975 | |
1da177e4 | 7976 | if (mddev->pers->sync_request) { |
ef99bf48 | 7977 | if (spares) { |
a654b9d8 N |
7978 | /* We are adding a device or devices to an array |
7979 | * which has the bitmap stored on all devices. | |
7980 | * So make sure all bitmap pages get written | |
7981 | */ | |
7982 | bitmap_write_all(mddev->bitmap); | |
7983 | } | |
ac05f256 N |
7984 | INIT_WORK(&mddev->del_work, md_start_sync); |
7985 | queue_work(md_misc_wq, &mddev->del_work); | |
7986 | goto unlock; | |
1da177e4 | 7987 | } |
ac05f256 | 7988 | not_running: |
72a23c21 NB |
7989 | if (!mddev->sync_thread) { |
7990 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 7991 | wake_up(&resync_wait); |
72a23c21 NB |
7992 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
7993 | &mddev->recovery)) | |
0c3573f1 | 7994 | if (mddev->sysfs_action) |
00bcb4ac | 7995 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7996 | } |
ac05f256 N |
7997 | unlock: |
7998 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
7999 | mddev_unlock(mddev); |
8000 | } | |
8001 | } | |
6c144d31 | 8002 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8003 | |
a91d5ac0 JB |
8004 | void md_reap_sync_thread(struct mddev *mddev) |
8005 | { | |
8006 | struct md_rdev *rdev; | |
8007 | ||
8008 | /* resync has finished, collect result */ | |
8009 | md_unregister_thread(&mddev->sync_thread); | |
8010 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8011 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8012 | /* success...*/ | |
8013 | /* activate any spares */ | |
8014 | if (mddev->pers->spare_active(mddev)) { | |
8015 | sysfs_notify(&mddev->kobj, NULL, | |
8016 | "degraded"); | |
8017 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
8018 | } | |
8019 | } | |
8020 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
8021 | mddev->pers->finish_reshape) | |
8022 | mddev->pers->finish_reshape(mddev); | |
8023 | ||
8024 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8025 | * information must be scrapped. |
a91d5ac0 | 8026 | */ |
f466722c N |
8027 | if (!mddev->degraded) |
8028 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8029 | rdev->saved_raid_disk = -1; |
8030 | ||
8031 | md_update_sb(mddev, 1); | |
8032 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
8033 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8034 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8035 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8036 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8037 | wake_up(&resync_wait); |
a91d5ac0 JB |
8038 | /* flag recovery needed just to double check */ |
8039 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8040 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8041 | md_new_event(mddev); | |
8042 | if (mddev->event_work.func) | |
8043 | queue_work(md_misc_wq, &mddev->event_work); | |
8044 | } | |
6c144d31 | 8045 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 8046 | |
fd01b88c | 8047 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8048 | { |
00bcb4ac | 8049 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8050 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8051 | !test_bit(Blocked, &rdev->flags) && |
8052 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8053 | msecs_to_jiffies(5000)); |
8054 | rdev_dec_pending(rdev, mddev); | |
8055 | } | |
8056 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8057 | ||
c6563a8c N |
8058 | void md_finish_reshape(struct mddev *mddev) |
8059 | { | |
8060 | /* called be personality module when reshape completes. */ | |
8061 | struct md_rdev *rdev; | |
8062 | ||
8063 | rdev_for_each(rdev, mddev) { | |
8064 | if (rdev->data_offset > rdev->new_data_offset) | |
8065 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8066 | else | |
8067 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8068 | rdev->data_offset = rdev->new_data_offset; | |
8069 | } | |
8070 | } | |
8071 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 N |
8072 | |
8073 | /* Bad block management. | |
8074 | * We can record which blocks on each device are 'bad' and so just | |
8075 | * fail those blocks, or that stripe, rather than the whole device. | |
8076 | * Entries in the bad-block table are 64bits wide. This comprises: | |
8077 | * Length of bad-range, in sectors: 0-511 for lengths 1-512 | |
8078 | * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) | |
8079 | * A 'shift' can be set so that larger blocks are tracked and | |
8080 | * consequently larger devices can be covered. | |
8081 | * 'Acknowledged' flag - 1 bit. - the most significant bit. | |
8082 | * | |
8083 | * Locking of the bad-block table uses a seqlock so md_is_badblock | |
8084 | * might need to retry if it is very unlucky. | |
8085 | * We will sometimes want to check for bad blocks in a bi_end_io function, | |
8086 | * so we use the write_seqlock_irq variant. | |
8087 | * | |
8088 | * When looking for a bad block we specify a range and want to | |
8089 | * know if any block in the range is bad. So we binary-search | |
8090 | * to the last range that starts at-or-before the given endpoint, | |
8091 | * (or "before the sector after the target range") | |
8092 | * then see if it ends after the given start. | |
8093 | * We return | |
8094 | * 0 if there are no known bad blocks in the range | |
8095 | * 1 if there are known bad block which are all acknowledged | |
8096 | * -1 if there are bad blocks which have not yet been acknowledged in metadata. | |
8097 | * plus the start/length of the first bad section we overlap. | |
8098 | */ | |
8099 | int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, | |
8100 | sector_t *first_bad, int *bad_sectors) | |
8101 | { | |
8102 | int hi; | |
ab05613a | 8103 | int lo; |
2230dfe4 | 8104 | u64 *p = bb->page; |
ab05613a | 8105 | int rv; |
2230dfe4 N |
8106 | sector_t target = s + sectors; |
8107 | unsigned seq; | |
8108 | ||
8109 | if (bb->shift > 0) { | |
8110 | /* round the start down, and the end up */ | |
8111 | s >>= bb->shift; | |
8112 | target += (1<<bb->shift) - 1; | |
8113 | target >>= bb->shift; | |
8114 | sectors = target - s; | |
8115 | } | |
8116 | /* 'target' is now the first block after the bad range */ | |
8117 | ||
8118 | retry: | |
8119 | seq = read_seqbegin(&bb->lock); | |
ab05613a | 8120 | lo = 0; |
8121 | rv = 0; | |
2230dfe4 N |
8122 | hi = bb->count; |
8123 | ||
8124 | /* Binary search between lo and hi for 'target' | |
8125 | * i.e. for the last range that starts before 'target' | |
8126 | */ | |
8127 | /* INVARIANT: ranges before 'lo' and at-or-after 'hi' | |
8128 | * are known not to be the last range before target. | |
8129 | * VARIANT: hi-lo is the number of possible | |
8130 | * ranges, and decreases until it reaches 1 | |
8131 | */ | |
8132 | while (hi - lo > 1) { | |
8133 | int mid = (lo + hi) / 2; | |
8134 | sector_t a = BB_OFFSET(p[mid]); | |
8135 | if (a < target) | |
8136 | /* This could still be the one, earlier ranges | |
8137 | * could not. */ | |
8138 | lo = mid; | |
8139 | else | |
8140 | /* This and later ranges are definitely out. */ | |
8141 | hi = mid; | |
8142 | } | |
8143 | /* 'lo' might be the last that started before target, but 'hi' isn't */ | |
8144 | if (hi > lo) { | |
8145 | /* need to check all range that end after 's' to see if | |
8146 | * any are unacknowledged. | |
8147 | */ | |
8148 | while (lo >= 0 && | |
8149 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8150 | if (BB_OFFSET(p[lo]) < target) { | |
8151 | /* starts before the end, and finishes after | |
8152 | * the start, so they must overlap | |
8153 | */ | |
8154 | if (rv != -1 && BB_ACK(p[lo])) | |
8155 | rv = 1; | |
8156 | else | |
8157 | rv = -1; | |
8158 | *first_bad = BB_OFFSET(p[lo]); | |
8159 | *bad_sectors = BB_LEN(p[lo]); | |
8160 | } | |
8161 | lo--; | |
8162 | } | |
8163 | } | |
8164 | ||
8165 | if (read_seqretry(&bb->lock, seq)) | |
8166 | goto retry; | |
8167 | ||
8168 | return rv; | |
8169 | } | |
8170 | EXPORT_SYMBOL_GPL(md_is_badblock); | |
8171 | ||
8172 | /* | |
8173 | * Add a range of bad blocks to the table. | |
8174 | * This might extend the table, or might contract it | |
8175 | * if two adjacent ranges can be merged. | |
8176 | * We binary-search to find the 'insertion' point, then | |
8177 | * decide how best to handle it. | |
8178 | */ | |
8179 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, | |
8180 | int acknowledged) | |
8181 | { | |
8182 | u64 *p; | |
8183 | int lo, hi; | |
8184 | int rv = 1; | |
905b0297 | 8185 | unsigned long flags; |
2230dfe4 N |
8186 | |
8187 | if (bb->shift < 0) | |
8188 | /* badblocks are disabled */ | |
8189 | return 0; | |
8190 | ||
8191 | if (bb->shift) { | |
8192 | /* round the start down, and the end up */ | |
8193 | sector_t next = s + sectors; | |
8194 | s >>= bb->shift; | |
8195 | next += (1<<bb->shift) - 1; | |
8196 | next >>= bb->shift; | |
8197 | sectors = next - s; | |
8198 | } | |
8199 | ||
905b0297 | 8200 | write_seqlock_irqsave(&bb->lock, flags); |
2230dfe4 N |
8201 | |
8202 | p = bb->page; | |
8203 | lo = 0; | |
8204 | hi = bb->count; | |
8205 | /* Find the last range that starts at-or-before 's' */ | |
8206 | while (hi - lo > 1) { | |
8207 | int mid = (lo + hi) / 2; | |
8208 | sector_t a = BB_OFFSET(p[mid]); | |
8209 | if (a <= s) | |
8210 | lo = mid; | |
8211 | else | |
8212 | hi = mid; | |
8213 | } | |
8214 | if (hi > lo && BB_OFFSET(p[lo]) > s) | |
8215 | hi = lo; | |
8216 | ||
8217 | if (hi > lo) { | |
8218 | /* we found a range that might merge with the start | |
8219 | * of our new range | |
8220 | */ | |
8221 | sector_t a = BB_OFFSET(p[lo]); | |
8222 | sector_t e = a + BB_LEN(p[lo]); | |
8223 | int ack = BB_ACK(p[lo]); | |
8224 | if (e >= s) { | |
8225 | /* Yes, we can merge with a previous range */ | |
8226 | if (s == a && s + sectors >= e) | |
8227 | /* new range covers old */ | |
8228 | ack = acknowledged; | |
8229 | else | |
8230 | ack = ack && acknowledged; | |
8231 | ||
8232 | if (e < s + sectors) | |
8233 | e = s + sectors; | |
8234 | if (e - a <= BB_MAX_LEN) { | |
8235 | p[lo] = BB_MAKE(a, e-a, ack); | |
8236 | s = e; | |
8237 | } else { | |
8238 | /* does not all fit in one range, | |
8239 | * make p[lo] maximal | |
8240 | */ | |
8241 | if (BB_LEN(p[lo]) != BB_MAX_LEN) | |
8242 | p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8243 | s = a + BB_MAX_LEN; | |
8244 | } | |
8245 | sectors = e - s; | |
8246 | } | |
8247 | } | |
8248 | if (sectors && hi < bb->count) { | |
8249 | /* 'hi' points to the first range that starts after 's'. | |
8250 | * Maybe we can merge with the start of that range */ | |
8251 | sector_t a = BB_OFFSET(p[hi]); | |
8252 | sector_t e = a + BB_LEN(p[hi]); | |
8253 | int ack = BB_ACK(p[hi]); | |
8254 | if (a <= s + sectors) { | |
8255 | /* merging is possible */ | |
8256 | if (e <= s + sectors) { | |
8257 | /* full overlap */ | |
8258 | e = s + sectors; | |
8259 | ack = acknowledged; | |
8260 | } else | |
8261 | ack = ack && acknowledged; | |
8262 | ||
8263 | a = s; | |
8264 | if (e - a <= BB_MAX_LEN) { | |
8265 | p[hi] = BB_MAKE(a, e-a, ack); | |
8266 | s = e; | |
8267 | } else { | |
8268 | p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8269 | s = a + BB_MAX_LEN; | |
8270 | } | |
8271 | sectors = e - s; | |
8272 | lo = hi; | |
8273 | hi++; | |
8274 | } | |
8275 | } | |
8276 | if (sectors == 0 && hi < bb->count) { | |
8277 | /* we might be able to combine lo and hi */ | |
8278 | /* Note: 's' is at the end of 'lo' */ | |
8279 | sector_t a = BB_OFFSET(p[hi]); | |
8280 | int lolen = BB_LEN(p[lo]); | |
8281 | int hilen = BB_LEN(p[hi]); | |
8282 | int newlen = lolen + hilen - (s - a); | |
8283 | if (s >= a && newlen < BB_MAX_LEN) { | |
8284 | /* yes, we can combine them */ | |
8285 | int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); | |
8286 | p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); | |
8287 | memmove(p + hi, p + hi + 1, | |
8288 | (bb->count - hi - 1) * 8); | |
8289 | bb->count--; | |
8290 | } | |
8291 | } | |
8292 | while (sectors) { | |
8293 | /* didn't merge (it all). | |
8294 | * Need to add a range just before 'hi' */ | |
8295 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8296 | /* No room for more */ | |
8297 | rv = 0; | |
8298 | break; | |
8299 | } else { | |
8300 | int this_sectors = sectors; | |
8301 | memmove(p + hi + 1, p + hi, | |
8302 | (bb->count - hi) * 8); | |
8303 | bb->count++; | |
8304 | ||
8305 | if (this_sectors > BB_MAX_LEN) | |
8306 | this_sectors = BB_MAX_LEN; | |
8307 | p[hi] = BB_MAKE(s, this_sectors, acknowledged); | |
8308 | sectors -= this_sectors; | |
8309 | s += this_sectors; | |
8310 | } | |
8311 | } | |
8312 | ||
8313 | bb->changed = 1; | |
de393cde N |
8314 | if (!acknowledged) |
8315 | bb->unacked_exist = 1; | |
905b0297 | 8316 | write_sequnlock_irqrestore(&bb->lock, flags); |
2230dfe4 N |
8317 | |
8318 | return rv; | |
8319 | } | |
8320 | ||
3cb03002 | 8321 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8322 | int is_new) |
2230dfe4 | 8323 | { |
c6563a8c N |
8324 | int rv; |
8325 | if (is_new) | |
8326 | s += rdev->new_data_offset; | |
8327 | else | |
8328 | s += rdev->data_offset; | |
8329 | rv = md_set_badblocks(&rdev->badblocks, | |
8330 | s, sectors, 0); | |
2230dfe4 N |
8331 | if (rv) { |
8332 | /* Make sure they get written out promptly */ | |
8bd2f0a0 | 8333 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2230dfe4 N |
8334 | set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); |
8335 | md_wakeup_thread(rdev->mddev->thread); | |
8336 | } | |
8337 | return rv; | |
8338 | } | |
8339 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8340 | ||
8341 | /* | |
8342 | * Remove a range of bad blocks from the table. | |
8343 | * This may involve extending the table if we spilt a region, | |
8344 | * but it must not fail. So if the table becomes full, we just | |
8345 | * drop the remove request. | |
8346 | */ | |
8347 | static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) | |
8348 | { | |
8349 | u64 *p; | |
8350 | int lo, hi; | |
8351 | sector_t target = s + sectors; | |
8352 | int rv = 0; | |
8353 | ||
8354 | if (bb->shift > 0) { | |
8355 | /* When clearing we round the start up and the end down. | |
8356 | * This should not matter as the shift should align with | |
8357 | * the block size and no rounding should ever be needed. | |
8358 | * However it is better the think a block is bad when it | |
8359 | * isn't than to think a block is not bad when it is. | |
8360 | */ | |
8361 | s += (1<<bb->shift) - 1; | |
8362 | s >>= bb->shift; | |
8363 | target >>= bb->shift; | |
8364 | sectors = target - s; | |
8365 | } | |
8366 | ||
8367 | write_seqlock_irq(&bb->lock); | |
8368 | ||
8369 | p = bb->page; | |
8370 | lo = 0; | |
8371 | hi = bb->count; | |
8372 | /* Find the last range that starts before 'target' */ | |
8373 | while (hi - lo > 1) { | |
8374 | int mid = (lo + hi) / 2; | |
8375 | sector_t a = BB_OFFSET(p[mid]); | |
8376 | if (a < target) | |
8377 | lo = mid; | |
8378 | else | |
8379 | hi = mid; | |
8380 | } | |
8381 | if (hi > lo) { | |
8382 | /* p[lo] is the last range that could overlap the | |
8383 | * current range. Earlier ranges could also overlap, | |
8384 | * but only this one can overlap the end of the range. | |
8385 | */ | |
8386 | if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { | |
8387 | /* Partial overlap, leave the tail of this range */ | |
8388 | int ack = BB_ACK(p[lo]); | |
8389 | sector_t a = BB_OFFSET(p[lo]); | |
8390 | sector_t end = a + BB_LEN(p[lo]); | |
8391 | ||
8392 | if (a < s) { | |
8393 | /* we need to split this range */ | |
8394 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8b32bf5e | 8395 | rv = -ENOSPC; |
2230dfe4 N |
8396 | goto out; |
8397 | } | |
8398 | memmove(p+lo+1, p+lo, (bb->count - lo) * 8); | |
8399 | bb->count++; | |
8400 | p[lo] = BB_MAKE(a, s-a, ack); | |
8401 | lo++; | |
8402 | } | |
8403 | p[lo] = BB_MAKE(target, end - target, ack); | |
8404 | /* there is no longer an overlap */ | |
8405 | hi = lo; | |
8406 | lo--; | |
8407 | } | |
8408 | while (lo >= 0 && | |
8409 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8410 | /* This range does overlap */ | |
8411 | if (BB_OFFSET(p[lo]) < s) { | |
8412 | /* Keep the early parts of this range. */ | |
8413 | int ack = BB_ACK(p[lo]); | |
8414 | sector_t start = BB_OFFSET(p[lo]); | |
8415 | p[lo] = BB_MAKE(start, s - start, ack); | |
8416 | /* now low doesn't overlap, so.. */ | |
8417 | break; | |
8418 | } | |
8419 | lo--; | |
8420 | } | |
8421 | /* 'lo' is strictly before, 'hi' is strictly after, | |
8422 | * anything between needs to be discarded | |
8423 | */ | |
8424 | if (hi - lo > 1) { | |
8425 | memmove(p+lo+1, p+hi, (bb->count - hi) * 8); | |
8426 | bb->count -= (hi - lo - 1); | |
8427 | } | |
8428 | } | |
8429 | ||
8430 | bb->changed = 1; | |
8431 | out: | |
8432 | write_sequnlock_irq(&bb->lock); | |
8433 | return rv; | |
8434 | } | |
8435 | ||
c6563a8c N |
8436 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8437 | int is_new) | |
2230dfe4 | 8438 | { |
c6563a8c N |
8439 | if (is_new) |
8440 | s += rdev->new_data_offset; | |
8441 | else | |
8442 | s += rdev->data_offset; | |
2230dfe4 | 8443 | return md_clear_badblocks(&rdev->badblocks, |
c6563a8c | 8444 | s, sectors); |
2230dfe4 N |
8445 | } |
8446 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8447 | ||
8448 | /* | |
8449 | * Acknowledge all bad blocks in a list. | |
8450 | * This only succeeds if ->changed is clear. It is used by | |
8451 | * in-kernel metadata updates | |
8452 | */ | |
8453 | void md_ack_all_badblocks(struct badblocks *bb) | |
8454 | { | |
8455 | if (bb->page == NULL || bb->changed) | |
8456 | /* no point even trying */ | |
8457 | return; | |
8458 | write_seqlock_irq(&bb->lock); | |
8459 | ||
ecb178bb | 8460 | if (bb->changed == 0 && bb->unacked_exist) { |
2230dfe4 N |
8461 | u64 *p = bb->page; |
8462 | int i; | |
8463 | for (i = 0; i < bb->count ; i++) { | |
8464 | if (!BB_ACK(p[i])) { | |
8465 | sector_t start = BB_OFFSET(p[i]); | |
8466 | int len = BB_LEN(p[i]); | |
8467 | p[i] = BB_MAKE(start, len, 1); | |
8468 | } | |
8469 | } | |
de393cde | 8470 | bb->unacked_exist = 0; |
2230dfe4 N |
8471 | } |
8472 | write_sequnlock_irq(&bb->lock); | |
8473 | } | |
8474 | EXPORT_SYMBOL_GPL(md_ack_all_badblocks); | |
8475 | ||
16c791a5 N |
8476 | /* sysfs access to bad-blocks list. |
8477 | * We present two files. | |
8478 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
8479 | * are recorded as bad. The list is truncated to fit within | |
8480 | * the one-page limit of sysfs. | |
8481 | * Writing "sector length" to this file adds an acknowledged | |
8482 | * bad block list. | |
8483 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
8484 | * been acknowledged. Writing to this file adds bad blocks | |
8485 | * without acknowledging them. This is largely for testing. | |
8486 | */ | |
8487 | ||
8488 | static ssize_t | |
8489 | badblocks_show(struct badblocks *bb, char *page, int unack) | |
8490 | { | |
8491 | size_t len; | |
8492 | int i; | |
8493 | u64 *p = bb->page; | |
8494 | unsigned seq; | |
8495 | ||
8496 | if (bb->shift < 0) | |
8497 | return 0; | |
8498 | ||
8499 | retry: | |
8500 | seq = read_seqbegin(&bb->lock); | |
8501 | ||
8502 | len = 0; | |
8503 | i = 0; | |
8504 | ||
8505 | while (len < PAGE_SIZE && i < bb->count) { | |
8506 | sector_t s = BB_OFFSET(p[i]); | |
8507 | unsigned int length = BB_LEN(p[i]); | |
8508 | int ack = BB_ACK(p[i]); | |
8509 | i++; | |
8510 | ||
8511 | if (unack && ack) | |
8512 | continue; | |
8513 | ||
8514 | len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", | |
8515 | (unsigned long long)s << bb->shift, | |
8516 | length << bb->shift); | |
8517 | } | |
de393cde N |
8518 | if (unack && len == 0) |
8519 | bb->unacked_exist = 0; | |
16c791a5 N |
8520 | |
8521 | if (read_seqretry(&bb->lock, seq)) | |
8522 | goto retry; | |
8523 | ||
8524 | return len; | |
8525 | } | |
8526 | ||
8527 | #define DO_DEBUG 1 | |
8528 | ||
8529 | static ssize_t | |
8530 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) | |
8531 | { | |
8532 | unsigned long long sector; | |
8533 | int length; | |
8534 | char newline; | |
8535 | #ifdef DO_DEBUG | |
8536 | /* Allow clearing via sysfs *only* for testing/debugging. | |
8537 | * Normally only a successful write may clear a badblock | |
8538 | */ | |
8539 | int clear = 0; | |
8540 | if (page[0] == '-') { | |
8541 | clear = 1; | |
8542 | page++; | |
8543 | } | |
8544 | #endif /* DO_DEBUG */ | |
8545 | ||
8546 | switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { | |
8547 | case 3: | |
8548 | if (newline != '\n') | |
8549 | return -EINVAL; | |
8550 | case 2: | |
8551 | if (length <= 0) | |
8552 | return -EINVAL; | |
8553 | break; | |
8554 | default: | |
8555 | return -EINVAL; | |
8556 | } | |
8557 | ||
8558 | #ifdef DO_DEBUG | |
8559 | if (clear) { | |
8560 | md_clear_badblocks(bb, sector, length); | |
8561 | return len; | |
8562 | } | |
8563 | #endif /* DO_DEBUG */ | |
8564 | if (md_set_badblocks(bb, sector, length, !unack)) | |
8565 | return len; | |
8566 | else | |
8567 | return -ENOSPC; | |
8568 | } | |
8569 | ||
75c96f85 AB |
8570 | static int md_notify_reboot(struct notifier_block *this, |
8571 | unsigned long code, void *x) | |
1da177e4 LT |
8572 | { |
8573 | struct list_head *tmp; | |
fd01b88c | 8574 | struct mddev *mddev; |
2dba6a91 | 8575 | int need_delay = 0; |
1da177e4 | 8576 | |
c744a65c N |
8577 | for_each_mddev(mddev, tmp) { |
8578 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8579 | if (mddev->pers) |
8580 | __md_stop_writes(mddev); | |
0f62fb22 N |
8581 | if (mddev->persistent) |
8582 | mddev->safemode = 2; | |
c744a65c | 8583 | mddev_unlock(mddev); |
2dba6a91 | 8584 | } |
c744a65c | 8585 | need_delay = 1; |
1da177e4 | 8586 | } |
c744a65c N |
8587 | /* |
8588 | * certain more exotic SCSI devices are known to be | |
8589 | * volatile wrt too early system reboots. While the | |
8590 | * right place to handle this issue is the given | |
8591 | * driver, we do want to have a safe RAID driver ... | |
8592 | */ | |
8593 | if (need_delay) | |
8594 | mdelay(1000*1); | |
8595 | ||
1da177e4 LT |
8596 | return NOTIFY_DONE; |
8597 | } | |
8598 | ||
75c96f85 | 8599 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8600 | .notifier_call = md_notify_reboot, |
8601 | .next = NULL, | |
8602 | .priority = INT_MAX, /* before any real devices */ | |
8603 | }; | |
8604 | ||
8605 | static void md_geninit(void) | |
8606 | { | |
36a4e1fe | 8607 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8608 | |
c7705f34 | 8609 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8610 | } |
8611 | ||
75c96f85 | 8612 | static int __init md_init(void) |
1da177e4 | 8613 | { |
e804ac78 TH |
8614 | int ret = -ENOMEM; |
8615 | ||
ada609ee | 8616 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8617 | if (!md_wq) |
8618 | goto err_wq; | |
8619 | ||
8620 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8621 | if (!md_misc_wq) | |
8622 | goto err_misc_wq; | |
8623 | ||
8624 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8625 | goto err_md; | |
8626 | ||
8627 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8628 | goto err_mdp; | |
8629 | mdp_major = ret; | |
8630 | ||
af5628f0 | 8631 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
8632 | md_probe, NULL, NULL); |
8633 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8634 | md_probe, NULL, NULL); |
8635 | ||
1da177e4 | 8636 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8637 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8638 | |
8639 | md_geninit(); | |
d710e138 | 8640 | return 0; |
1da177e4 | 8641 | |
e804ac78 TH |
8642 | err_mdp: |
8643 | unregister_blkdev(MD_MAJOR, "md"); | |
8644 | err_md: | |
8645 | destroy_workqueue(md_misc_wq); | |
8646 | err_misc_wq: | |
8647 | destroy_workqueue(md_wq); | |
8648 | err_wq: | |
8649 | return ret; | |
8650 | } | |
1da177e4 LT |
8651 | |
8652 | #ifndef MODULE | |
8653 | ||
8654 | /* | |
8655 | * Searches all registered partitions for autorun RAID arrays | |
8656 | * at boot time. | |
8657 | */ | |
4d936ec1 ME |
8658 | |
8659 | static LIST_HEAD(all_detected_devices); | |
8660 | struct detected_devices_node { | |
8661 | struct list_head list; | |
8662 | dev_t dev; | |
8663 | }; | |
1da177e4 LT |
8664 | |
8665 | void md_autodetect_dev(dev_t dev) | |
8666 | { | |
4d936ec1 ME |
8667 | struct detected_devices_node *node_detected_dev; |
8668 | ||
8669 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8670 | if (node_detected_dev) { | |
8671 | node_detected_dev->dev = dev; | |
8672 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
8673 | } else { | |
8674 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
8675 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
8676 | } | |
1da177e4 LT |
8677 | } |
8678 | ||
1da177e4 LT |
8679 | static void autostart_arrays(int part) |
8680 | { | |
3cb03002 | 8681 | struct md_rdev *rdev; |
4d936ec1 ME |
8682 | struct detected_devices_node *node_detected_dev; |
8683 | dev_t dev; | |
8684 | int i_scanned, i_passed; | |
1da177e4 | 8685 | |
4d936ec1 ME |
8686 | i_scanned = 0; |
8687 | i_passed = 0; | |
1da177e4 | 8688 | |
4d936ec1 | 8689 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8690 | |
4d936ec1 ME |
8691 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8692 | i_scanned++; | |
8693 | node_detected_dev = list_entry(all_detected_devices.next, | |
8694 | struct detected_devices_node, list); | |
8695 | list_del(&node_detected_dev->list); | |
8696 | dev = node_detected_dev->dev; | |
8697 | kfree(node_detected_dev); | |
df968c4e | 8698 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
8699 | if (IS_ERR(rdev)) |
8700 | continue; | |
8701 | ||
403df478 | 8702 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 8703 | continue; |
403df478 | 8704 | |
d0fae18f | 8705 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 8706 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 8707 | i_passed++; |
1da177e4 | 8708 | } |
4d936ec1 ME |
8709 | |
8710 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
8711 | i_scanned, i_passed); | |
1da177e4 LT |
8712 | |
8713 | autorun_devices(part); | |
8714 | } | |
8715 | ||
fdee8ae4 | 8716 | #endif /* !MODULE */ |
1da177e4 LT |
8717 | |
8718 | static __exit void md_exit(void) | |
8719 | { | |
fd01b88c | 8720 | struct mddev *mddev; |
1da177e4 | 8721 | struct list_head *tmp; |
e2f23b60 | 8722 | int delay = 1; |
8ab5e4c1 | 8723 | |
af5628f0 | 8724 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 8725 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 8726 | |
3dbd8c2e | 8727 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
8728 | unregister_blkdev(mdp_major, "mdp"); |
8729 | unregister_reboot_notifier(&md_notifier); | |
8730 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
8731 | |
8732 | /* We cannot unload the modules while some process is | |
8733 | * waiting for us in select() or poll() - wake them up | |
8734 | */ | |
8735 | md_unloading = 1; | |
8736 | while (waitqueue_active(&md_event_waiters)) { | |
8737 | /* not safe to leave yet */ | |
8738 | wake_up(&md_event_waiters); | |
8739 | msleep(delay); | |
8740 | delay += delay; | |
8741 | } | |
1da177e4 | 8742 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 8743 | |
29ac4aa3 | 8744 | for_each_mddev(mddev, tmp) { |
1da177e4 | 8745 | export_array(mddev); |
d3374825 | 8746 | mddev->hold_active = 0; |
1da177e4 | 8747 | } |
e804ac78 TH |
8748 | destroy_workqueue(md_misc_wq); |
8749 | destroy_workqueue(md_wq); | |
1da177e4 LT |
8750 | } |
8751 | ||
685784aa | 8752 | subsys_initcall(md_init); |
1da177e4 LT |
8753 | module_exit(md_exit) |
8754 | ||
f91de92e N |
8755 | static int get_ro(char *buffer, struct kernel_param *kp) |
8756 | { | |
8757 | return sprintf(buffer, "%d", start_readonly); | |
8758 | } | |
8759 | static int set_ro(const char *val, struct kernel_param *kp) | |
8760 | { | |
8761 | char *e; | |
8762 | int num = simple_strtoul(val, &e, 10); | |
8763 | if (*val && (*e == '\0' || *e == '\n')) { | |
8764 | start_readonly = num; | |
4dbcdc75 | 8765 | return 0; |
f91de92e N |
8766 | } |
8767 | return -EINVAL; | |
8768 | } | |
8769 | ||
80ca3a44 N |
8770 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
8771 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 8772 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 8773 | |
1da177e4 | 8774 | MODULE_LICENSE("GPL"); |
0efb9e61 | 8775 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 8776 | MODULE_ALIAS("md"); |
72008652 | 8777 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |