Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | md.c : Multiple Devices driver for Linux | |
3 | Copyright (C) 1998, 1999, 2000 Ingo Molnar | |
4 | ||
5 | completely rewritten, based on the MD driver code from Marc Zyngier | |
6 | ||
7 | Changes: | |
8 | ||
9 | - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar | |
10 | - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com> | |
11 | - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net> | |
12 | - kerneld support by Boris Tobotras <boris@xtalk.msk.su> | |
13 | - kmod support by: Cyrus Durgin | |
14 | - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com> | |
15 | - Devfs support by Richard Gooch <rgooch@atnf.csiro.au> | |
16 | ||
17 | - lots of fixes and improvements to the RAID1/RAID5 and generic | |
18 | RAID code (such as request based resynchronization): | |
19 | ||
20 | Neil Brown <neilb@cse.unsw.edu.au>. | |
21 | ||
32a7627c N |
22 | - persistent bitmap code |
23 | Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. | |
24 | ||
1da177e4 LT |
25 | This program is free software; you can redistribute it and/or modify |
26 | it under the terms of the GNU General Public License as published by | |
27 | the Free Software Foundation; either version 2, or (at your option) | |
28 | any later version. | |
29 | ||
30 | You should have received a copy of the GNU General Public License | |
31 | (for example /usr/src/linux/COPYING); if not, write to the Free | |
32 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
33 | */ | |
34 | ||
a6fb0934 | 35 | #include <linux/kthread.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
1da177e4 | 37 | #include <linux/sysctl.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
ff01bb48 | 39 | #include <linux/fs.h> |
d7603b7e | 40 | #include <linux/poll.h> |
16f17b39 | 41 | #include <linux/ctype.h> |
e7d2860b | 42 | #include <linux/string.h> |
fb4d8c76 N |
43 | #include <linux/hdreg.h> |
44 | #include <linux/proc_fs.h> | |
45 | #include <linux/random.h> | |
056075c7 | 46 | #include <linux/module.h> |
fb4d8c76 | 47 | #include <linux/reboot.h> |
32a7627c | 48 | #include <linux/file.h> |
aa98aa31 | 49 | #include <linux/compat.h> |
25570727 | 50 | #include <linux/delay.h> |
bff61975 N |
51 | #include <linux/raid/md_p.h> |
52 | #include <linux/raid/md_u.h> | |
5a0e3ad6 | 53 | #include <linux/slab.h> |
43b2e5d8 | 54 | #include "md.h" |
ef740c37 | 55 | #include "bitmap.h" |
1da177e4 | 56 | |
1da177e4 | 57 | #ifndef MODULE |
d710e138 | 58 | static void autostart_arrays(int part); |
1da177e4 LT |
59 | #endif |
60 | ||
01f96c0a N |
61 | /* pers_list is a list of registered personalities protected |
62 | * by pers_lock. | |
63 | * pers_lock does extra service to protect accesses to | |
64 | * mddev->thread when the mutex cannot be held. | |
65 | */ | |
2604b703 | 66 | static LIST_HEAD(pers_list); |
1da177e4 LT |
67 | static DEFINE_SPINLOCK(pers_lock); |
68 | ||
5e56341d AB |
69 | static void md_print_devices(void); |
70 | ||
90b08710 | 71 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
72 | static struct workqueue_struct *md_wq; |
73 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 74 | |
746d3207 N |
75 | static int remove_and_add_spares(struct mddev *mddev, |
76 | struct md_rdev *this); | |
77 | ||
5e56341d AB |
78 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } |
79 | ||
1e50915f RB |
80 | /* |
81 | * Default number of read corrections we'll attempt on an rdev | |
82 | * before ejecting it from the array. We divide the read error | |
83 | * count by 2 for every hour elapsed between read errors. | |
84 | */ | |
85 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
86 | /* |
87 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
88 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
89 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 90 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
91 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
92 | * speed limit - in case reconstruction slows down your system despite | |
93 | * idle IO detection. | |
94 | * | |
95 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 96 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
97 | */ |
98 | ||
99 | static int sysctl_speed_limit_min = 1000; | |
100 | static int sysctl_speed_limit_max = 200000; | |
fd01b88c | 101 | static inline int speed_min(struct mddev *mddev) |
88202a0c N |
102 | { |
103 | return mddev->sync_speed_min ? | |
104 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
105 | } | |
106 | ||
fd01b88c | 107 | static inline int speed_max(struct mddev *mddev) |
88202a0c N |
108 | { |
109 | return mddev->sync_speed_max ? | |
110 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
111 | } | |
1da177e4 LT |
112 | |
113 | static struct ctl_table_header *raid_table_header; | |
114 | ||
82592c38 | 115 | static struct ctl_table raid_table[] = { |
1da177e4 | 116 | { |
1da177e4 LT |
117 | .procname = "speed_limit_min", |
118 | .data = &sysctl_speed_limit_min, | |
119 | .maxlen = sizeof(int), | |
80ca3a44 | 120 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 121 | .proc_handler = proc_dointvec, |
1da177e4 LT |
122 | }, |
123 | { | |
1da177e4 LT |
124 | .procname = "speed_limit_max", |
125 | .data = &sysctl_speed_limit_max, | |
126 | .maxlen = sizeof(int), | |
80ca3a44 | 127 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 128 | .proc_handler = proc_dointvec, |
1da177e4 | 129 | }, |
894d2491 | 130 | { } |
1da177e4 LT |
131 | }; |
132 | ||
82592c38 | 133 | static struct ctl_table raid_dir_table[] = { |
1da177e4 | 134 | { |
1da177e4 LT |
135 | .procname = "raid", |
136 | .maxlen = 0, | |
80ca3a44 | 137 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
138 | .child = raid_table, |
139 | }, | |
894d2491 | 140 | { } |
1da177e4 LT |
141 | }; |
142 | ||
82592c38 | 143 | static struct ctl_table raid_root_table[] = { |
1da177e4 | 144 | { |
1da177e4 LT |
145 | .procname = "dev", |
146 | .maxlen = 0, | |
147 | .mode = 0555, | |
148 | .child = raid_dir_table, | |
149 | }, | |
894d2491 | 150 | { } |
1da177e4 LT |
151 | }; |
152 | ||
83d5cde4 | 153 | static const struct block_device_operations md_fops; |
1da177e4 | 154 | |
f91de92e N |
155 | static int start_readonly; |
156 | ||
a167f663 N |
157 | /* bio_clone_mddev |
158 | * like bio_clone, but with a local bio set | |
159 | */ | |
160 | ||
a167f663 | 161 | struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, |
fd01b88c | 162 | struct mddev *mddev) |
a167f663 N |
163 | { |
164 | struct bio *b; | |
a167f663 N |
165 | |
166 | if (!mddev || !mddev->bio_set) | |
167 | return bio_alloc(gfp_mask, nr_iovecs); | |
168 | ||
395c72a7 | 169 | b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set); |
a167f663 N |
170 | if (!b) |
171 | return NULL; | |
a167f663 N |
172 | return b; |
173 | } | |
174 | EXPORT_SYMBOL_GPL(bio_alloc_mddev); | |
175 | ||
176 | struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, | |
fd01b88c | 177 | struct mddev *mddev) |
a167f663 | 178 | { |
a167f663 N |
179 | if (!mddev || !mddev->bio_set) |
180 | return bio_clone(bio, gfp_mask); | |
181 | ||
bf800ef1 | 182 | return bio_clone_bioset(bio, gfp_mask, mddev->bio_set); |
a167f663 N |
183 | } |
184 | EXPORT_SYMBOL_GPL(bio_clone_mddev); | |
185 | ||
d7603b7e N |
186 | /* |
187 | * We have a system wide 'event count' that is incremented | |
188 | * on any 'interesting' event, and readers of /proc/mdstat | |
189 | * can use 'poll' or 'select' to find out when the event | |
190 | * count increases. | |
191 | * | |
192 | * Events are: | |
193 | * start array, stop array, error, add device, remove device, | |
194 | * start build, activate spare | |
195 | */ | |
2989ddbd | 196 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 197 | static atomic_t md_event_count; |
fd01b88c | 198 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
199 | { |
200 | atomic_inc(&md_event_count); | |
201 | wake_up(&md_event_waiters); | |
202 | } | |
29269553 | 203 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 204 | |
c331eb04 N |
205 | /* Alternate version that can be called from interrupts |
206 | * when calling sysfs_notify isn't needed. | |
207 | */ | |
fd01b88c | 208 | static void md_new_event_inintr(struct mddev *mddev) |
c331eb04 N |
209 | { |
210 | atomic_inc(&md_event_count); | |
211 | wake_up(&md_event_waiters); | |
212 | } | |
213 | ||
1da177e4 LT |
214 | /* |
215 | * Enables to iterate over all existing md arrays | |
216 | * all_mddevs_lock protects this list. | |
217 | */ | |
218 | static LIST_HEAD(all_mddevs); | |
219 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
220 | ||
221 | ||
222 | /* | |
223 | * iterates through all used mddevs in the system. | |
224 | * We take care to grab the all_mddevs_lock whenever navigating | |
225 | * the list, and to always hold a refcount when unlocked. | |
226 | * Any code which breaks out of this loop while own | |
227 | * a reference to the current mddev and must mddev_put it. | |
228 | */ | |
fd01b88c | 229 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 LT |
230 | \ |
231 | for (({ spin_lock(&all_mddevs_lock); \ | |
fd01b88c N |
232 | _tmp = all_mddevs.next; \ |
233 | _mddev = NULL;}); \ | |
234 | ({ if (_tmp != &all_mddevs) \ | |
235 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 236 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
237 | if (_mddev) mddev_put(_mddev); \ |
238 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
239 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 240 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 241 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
242 | ) |
243 | ||
244 | ||
409c57f3 N |
245 | /* Rather than calling directly into the personality make_request function, |
246 | * IO requests come here first so that we can check if the device is | |
247 | * being suspended pending a reconfiguration. | |
248 | * We hold a refcount over the call to ->make_request. By the time that | |
249 | * call has finished, the bio has been linked into some internal structure | |
250 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
251 | */ | |
5a7bbad2 | 252 | static void md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 253 | { |
49077326 | 254 | const int rw = bio_data_dir(bio); |
fd01b88c | 255 | struct mddev *mddev = q->queuedata; |
49077326 | 256 | int cpu; |
e91ece55 | 257 | unsigned int sectors; |
49077326 | 258 | |
0ca69886 N |
259 | if (mddev == NULL || mddev->pers == NULL |
260 | || !mddev->ready) { | |
409c57f3 | 261 | bio_io_error(bio); |
5a7bbad2 | 262 | return; |
409c57f3 | 263 | } |
bbfa57c0 SR |
264 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
265 | bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS); | |
266 | return; | |
267 | } | |
0ca69886 | 268 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 269 | rcu_read_lock(); |
e9c7469b | 270 | if (mddev->suspended) { |
409c57f3 N |
271 | DEFINE_WAIT(__wait); |
272 | for (;;) { | |
273 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
274 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 275 | if (!mddev->suspended) |
409c57f3 N |
276 | break; |
277 | rcu_read_unlock(); | |
278 | schedule(); | |
279 | rcu_read_lock(); | |
280 | } | |
281 | finish_wait(&mddev->sb_wait, &__wait); | |
282 | } | |
283 | atomic_inc(&mddev->active_io); | |
284 | rcu_read_unlock(); | |
49077326 | 285 | |
e91ece55 CM |
286 | /* |
287 | * save the sectors now since our bio can | |
288 | * go away inside make_request | |
289 | */ | |
290 | sectors = bio_sectors(bio); | |
5a7bbad2 | 291 | mddev->pers->make_request(mddev, bio); |
49077326 N |
292 | |
293 | cpu = part_stat_lock(); | |
294 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
e91ece55 | 295 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); |
49077326 N |
296 | part_stat_unlock(); |
297 | ||
409c57f3 N |
298 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
299 | wake_up(&mddev->sb_wait); | |
409c57f3 N |
300 | } |
301 | ||
9e35b99c N |
302 | /* mddev_suspend makes sure no new requests are submitted |
303 | * to the device, and that any requests that have been submitted | |
304 | * are completely handled. | |
305 | * Once ->stop is called and completes, the module will be completely | |
306 | * unused. | |
307 | */ | |
fd01b88c | 308 | void mddev_suspend(struct mddev *mddev) |
409c57f3 N |
309 | { |
310 | BUG_ON(mddev->suspended); | |
311 | mddev->suspended = 1; | |
312 | synchronize_rcu(); | |
313 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
314 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
315 | |
316 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 317 | } |
390ee602 | 318 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 319 | |
fd01b88c | 320 | void mddev_resume(struct mddev *mddev) |
409c57f3 N |
321 | { |
322 | mddev->suspended = 0; | |
323 | wake_up(&mddev->sb_wait); | |
324 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 325 | |
47525e59 | 326 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
327 | md_wakeup_thread(mddev->thread); |
328 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 329 | } |
390ee602 | 330 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 331 | |
fd01b88c | 332 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 N |
333 | { |
334 | return mddev->suspended; | |
335 | } | |
336 | EXPORT_SYMBOL(mddev_congested); | |
337 | ||
a2826aa9 | 338 | /* |
e9c7469b | 339 | * Generic flush handling for md |
a2826aa9 N |
340 | */ |
341 | ||
e9c7469b | 342 | static void md_end_flush(struct bio *bio, int err) |
a2826aa9 | 343 | { |
3cb03002 | 344 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 345 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
346 | |
347 | rdev_dec_pending(rdev, mddev); | |
348 | ||
349 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 350 | /* The pre-request flush has finished */ |
e804ac78 | 351 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
352 | } |
353 | bio_put(bio); | |
354 | } | |
355 | ||
a7a07e69 N |
356 | static void md_submit_flush_data(struct work_struct *ws); |
357 | ||
a035fc3e | 358 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 359 | { |
fd01b88c | 360 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 361 | struct md_rdev *rdev; |
a2826aa9 | 362 | |
a7a07e69 N |
363 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
364 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 365 | rcu_read_lock(); |
dafb20fa | 366 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
367 | if (rdev->raid_disk >= 0 && |
368 | !test_bit(Faulty, &rdev->flags)) { | |
369 | /* Take two references, one is dropped | |
370 | * when request finishes, one after | |
371 | * we reclaim rcu_read_lock | |
372 | */ | |
373 | struct bio *bi; | |
374 | atomic_inc(&rdev->nr_pending); | |
375 | atomic_inc(&rdev->nr_pending); | |
376 | rcu_read_unlock(); | |
b5e1b8ce | 377 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 378 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
379 | bi->bi_private = rdev; |
380 | bi->bi_bdev = rdev->bdev; | |
381 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 382 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
383 | rcu_read_lock(); |
384 | rdev_dec_pending(rdev, mddev); | |
385 | } | |
386 | rcu_read_unlock(); | |
a7a07e69 N |
387 | if (atomic_dec_and_test(&mddev->flush_pending)) |
388 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
389 | } |
390 | ||
e9c7469b | 391 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 392 | { |
fd01b88c | 393 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 394 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 395 | |
4f024f37 | 396 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 N |
397 | /* an empty barrier - all done */ |
398 | bio_endio(bio, 0); | |
399 | else { | |
e9c7469b | 400 | bio->bi_rw &= ~REQ_FLUSH; |
5a7bbad2 | 401 | mddev->pers->make_request(mddev, bio); |
a2826aa9 | 402 | } |
2b74e12e N |
403 | |
404 | mddev->flush_bio = NULL; | |
405 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
406 | } |
407 | ||
fd01b88c | 408 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 N |
409 | { |
410 | spin_lock_irq(&mddev->write_lock); | |
411 | wait_event_lock_irq(mddev->sb_wait, | |
e9c7469b | 412 | !mddev->flush_bio, |
eed8c02e | 413 | mddev->write_lock); |
e9c7469b | 414 | mddev->flush_bio = bio; |
a2826aa9 N |
415 | spin_unlock_irq(&mddev->write_lock); |
416 | ||
a035fc3e N |
417 | INIT_WORK(&mddev->flush_work, submit_flushes); |
418 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 419 | } |
e9c7469b | 420 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 421 | |
74018dc3 | 422 | void md_unplug(struct blk_plug_cb *cb, bool from_schedule) |
97658cdd | 423 | { |
9cbb1750 N |
424 | struct mddev *mddev = cb->data; |
425 | md_wakeup_thread(mddev->thread); | |
426 | kfree(cb); | |
97658cdd | 427 | } |
9cbb1750 | 428 | EXPORT_SYMBOL(md_unplug); |
2ac87401 | 429 | |
fd01b88c | 430 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
431 | { |
432 | atomic_inc(&mddev->active); | |
433 | return mddev; | |
434 | } | |
435 | ||
5fd3a17e | 436 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 437 | |
fd01b88c | 438 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 439 | { |
a167f663 N |
440 | struct bio_set *bs = NULL; |
441 | ||
1da177e4 LT |
442 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
443 | return; | |
d3374825 | 444 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
445 | mddev->ctime == 0 && !mddev->hold_active) { |
446 | /* Array is not configured at all, and not held active, | |
447 | * so destroy it */ | |
af8a2434 | 448 | list_del_init(&mddev->all_mddevs); |
a167f663 N |
449 | bs = mddev->bio_set; |
450 | mddev->bio_set = NULL; | |
d3374825 | 451 | if (mddev->gendisk) { |
e804ac78 TH |
452 | /* We did a probe so need to clean up. Call |
453 | * queue_work inside the spinlock so that | |
454 | * flush_workqueue() after mddev_find will | |
455 | * succeed in waiting for the work to be done. | |
d3374825 N |
456 | */ |
457 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 458 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
459 | } else |
460 | kfree(mddev); | |
461 | } | |
462 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
463 | if (bs) |
464 | bioset_free(bs); | |
1da177e4 LT |
465 | } |
466 | ||
fd01b88c | 467 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
468 | { |
469 | mutex_init(&mddev->open_mutex); | |
470 | mutex_init(&mddev->reconfig_mutex); | |
471 | mutex_init(&mddev->bitmap_info.mutex); | |
472 | INIT_LIST_HEAD(&mddev->disks); | |
473 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
474 | init_timer(&mddev->safemode_timer); | |
475 | atomic_set(&mddev->active, 1); | |
476 | atomic_set(&mddev->openers, 0); | |
477 | atomic_set(&mddev->active_io, 0); | |
478 | spin_lock_init(&mddev->write_lock); | |
479 | atomic_set(&mddev->flush_pending, 0); | |
480 | init_waitqueue_head(&mddev->sb_wait); | |
481 | init_waitqueue_head(&mddev->recovery_wait); | |
482 | mddev->reshape_position = MaxSector; | |
2c810cdd | 483 | mddev->reshape_backwards = 0; |
c4a39551 | 484 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
485 | mddev->resync_min = 0; |
486 | mddev->resync_max = MaxSector; | |
487 | mddev->level = LEVEL_NONE; | |
488 | } | |
390ee602 | 489 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 490 | |
fd01b88c | 491 | static struct mddev * mddev_find(dev_t unit) |
1da177e4 | 492 | { |
fd01b88c | 493 | struct mddev *mddev, *new = NULL; |
1da177e4 | 494 | |
8f5f02c4 N |
495 | if (unit && MAJOR(unit) != MD_MAJOR) |
496 | unit &= ~((1<<MdpMinorShift)-1); | |
497 | ||
1da177e4 LT |
498 | retry: |
499 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
500 | |
501 | if (unit) { | |
502 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
503 | if (mddev->unit == unit) { | |
504 | mddev_get(mddev); | |
505 | spin_unlock(&all_mddevs_lock); | |
506 | kfree(new); | |
507 | return mddev; | |
508 | } | |
509 | ||
510 | if (new) { | |
511 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 512 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
513 | new->hold_active = UNTIL_IOCTL; |
514 | return new; | |
1da177e4 | 515 | } |
efeb53c0 N |
516 | } else if (new) { |
517 | /* find an unused unit number */ | |
518 | static int next_minor = 512; | |
519 | int start = next_minor; | |
520 | int is_free = 0; | |
521 | int dev = 0; | |
522 | while (!is_free) { | |
523 | dev = MKDEV(MD_MAJOR, next_minor); | |
524 | next_minor++; | |
525 | if (next_minor > MINORMASK) | |
526 | next_minor = 0; | |
527 | if (next_minor == start) { | |
528 | /* Oh dear, all in use. */ | |
529 | spin_unlock(&all_mddevs_lock); | |
530 | kfree(new); | |
531 | return NULL; | |
532 | } | |
533 | ||
534 | is_free = 1; | |
535 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
536 | if (mddev->unit == dev) { | |
537 | is_free = 0; | |
538 | break; | |
539 | } | |
540 | } | |
541 | new->unit = dev; | |
542 | new->md_minor = MINOR(dev); | |
543 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
544 | list_add(&new->all_mddevs, &all_mddevs); |
545 | spin_unlock(&all_mddevs_lock); | |
546 | return new; | |
547 | } | |
548 | spin_unlock(&all_mddevs_lock); | |
549 | ||
9ffae0cf | 550 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
551 | if (!new) |
552 | return NULL; | |
553 | ||
1da177e4 LT |
554 | new->unit = unit; |
555 | if (MAJOR(unit) == MD_MAJOR) | |
556 | new->md_minor = MINOR(unit); | |
557 | else | |
558 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
559 | ||
fafd7fb0 | 560 | mddev_init(new); |
1da177e4 | 561 | |
1da177e4 LT |
562 | goto retry; |
563 | } | |
564 | ||
29f097c4 | 565 | static inline int __must_check mddev_lock(struct mddev * mddev) |
1da177e4 | 566 | { |
df5b89b3 | 567 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
568 | } |
569 | ||
29f097c4 N |
570 | /* Sometimes we need to take the lock in a situation where |
571 | * failure due to interrupts is not acceptable. | |
572 | */ | |
573 | static inline void mddev_lock_nointr(struct mddev * mddev) | |
574 | { | |
575 | mutex_lock(&mddev->reconfig_mutex); | |
576 | } | |
577 | ||
fd01b88c | 578 | static inline int mddev_is_locked(struct mddev *mddev) |
b522adcd DW |
579 | { |
580 | return mutex_is_locked(&mddev->reconfig_mutex); | |
581 | } | |
582 | ||
fd01b88c | 583 | static inline int mddev_trylock(struct mddev * mddev) |
1da177e4 | 584 | { |
df5b89b3 | 585 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
586 | } |
587 | ||
b6eb127d N |
588 | static struct attribute_group md_redundancy_group; |
589 | ||
fd01b88c | 590 | static void mddev_unlock(struct mddev * mddev) |
1da177e4 | 591 | { |
a64c876f | 592 | if (mddev->to_remove) { |
b6eb127d N |
593 | /* These cannot be removed under reconfig_mutex as |
594 | * an access to the files will try to take reconfig_mutex | |
595 | * while holding the file unremovable, which leads to | |
596 | * a deadlock. | |
bb4f1e9d N |
597 | * So hold set sysfs_active while the remove in happeing, |
598 | * and anything else which might set ->to_remove or my | |
599 | * otherwise change the sysfs namespace will fail with | |
600 | * -EBUSY if sysfs_active is still set. | |
601 | * We set sysfs_active under reconfig_mutex and elsewhere | |
602 | * test it under the same mutex to ensure its correct value | |
603 | * is seen. | |
b6eb127d | 604 | */ |
a64c876f N |
605 | struct attribute_group *to_remove = mddev->to_remove; |
606 | mddev->to_remove = NULL; | |
bb4f1e9d | 607 | mddev->sysfs_active = 1; |
b6eb127d N |
608 | mutex_unlock(&mddev->reconfig_mutex); |
609 | ||
00bcb4ac N |
610 | if (mddev->kobj.sd) { |
611 | if (to_remove != &md_redundancy_group) | |
612 | sysfs_remove_group(&mddev->kobj, to_remove); | |
613 | if (mddev->pers == NULL || | |
614 | mddev->pers->sync_request == NULL) { | |
615 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
616 | if (mddev->sysfs_action) | |
617 | sysfs_put(mddev->sysfs_action); | |
618 | mddev->sysfs_action = NULL; | |
619 | } | |
a64c876f | 620 | } |
bb4f1e9d | 621 | mddev->sysfs_active = 0; |
b6eb127d N |
622 | } else |
623 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 624 | |
751e67ca CD |
625 | /* As we've dropped the mutex we need a spinlock to |
626 | * make sure the thread doesn't disappear | |
01f96c0a N |
627 | */ |
628 | spin_lock(&pers_lock); | |
005eca5e | 629 | md_wakeup_thread(mddev->thread); |
01f96c0a | 630 | spin_unlock(&pers_lock); |
1da177e4 LT |
631 | } |
632 | ||
fd01b88c | 633 | static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr) |
1da177e4 | 634 | { |
3cb03002 | 635 | struct md_rdev *rdev; |
1da177e4 | 636 | |
dafb20fa | 637 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
638 | if (rdev->desc_nr == nr) |
639 | return rdev; | |
159ec1fc | 640 | |
1da177e4 LT |
641 | return NULL; |
642 | } | |
643 | ||
1ca69c4b N |
644 | static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr) |
645 | { | |
646 | struct md_rdev *rdev; | |
647 | ||
648 | rdev_for_each_rcu(rdev, mddev) | |
649 | if (rdev->desc_nr == nr) | |
650 | return rdev; | |
651 | ||
652 | return NULL; | |
653 | } | |
654 | ||
655 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 656 | { |
3cb03002 | 657 | struct md_rdev *rdev; |
1da177e4 | 658 | |
dafb20fa | 659 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
660 | if (rdev->bdev->bd_dev == dev) |
661 | return rdev; | |
159ec1fc | 662 | |
1da177e4 LT |
663 | return NULL; |
664 | } | |
665 | ||
1ca69c4b N |
666 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
667 | { | |
668 | struct md_rdev *rdev; | |
669 | ||
670 | rdev_for_each_rcu(rdev, mddev) | |
671 | if (rdev->bdev->bd_dev == dev) | |
672 | return rdev; | |
673 | ||
674 | return NULL; | |
675 | } | |
676 | ||
84fc4b56 | 677 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 678 | { |
84fc4b56 | 679 | struct md_personality *pers; |
d9d166c2 N |
680 | list_for_each_entry(pers, &pers_list, list) { |
681 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 682 | return pers; |
d9d166c2 N |
683 | if (strcmp(pers->name, clevel)==0) |
684 | return pers; | |
685 | } | |
2604b703 N |
686 | return NULL; |
687 | } | |
688 | ||
b73df2d3 | 689 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 690 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 691 | { |
57b2caa3 | 692 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 693 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
694 | } |
695 | ||
3cb03002 | 696 | static int alloc_disk_sb(struct md_rdev * rdev) |
1da177e4 LT |
697 | { |
698 | if (rdev->sb_page) | |
699 | MD_BUG(); | |
700 | ||
701 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
702 | if (!rdev->sb_page) { | |
703 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 704 | return -ENOMEM; |
1da177e4 LT |
705 | } |
706 | ||
707 | return 0; | |
708 | } | |
709 | ||
545c8795 | 710 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
711 | { |
712 | if (rdev->sb_page) { | |
2d1f3b5d | 713 | put_page(rdev->sb_page); |
1da177e4 LT |
714 | rdev->sb_loaded = 0; |
715 | rdev->sb_page = NULL; | |
0f420358 | 716 | rdev->sb_start = 0; |
dd8ac336 | 717 | rdev->sectors = 0; |
1da177e4 | 718 | } |
2699b672 N |
719 | if (rdev->bb_page) { |
720 | put_page(rdev->bb_page); | |
721 | rdev->bb_page = NULL; | |
722 | } | |
4fa2f327 N |
723 | kfree(rdev->badblocks.page); |
724 | rdev->badblocks.page = NULL; | |
1da177e4 | 725 | } |
545c8795 | 726 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 727 | |
6712ecf8 | 728 | static void super_written(struct bio *bio, int error) |
7bfa19f2 | 729 | { |
3cb03002 | 730 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 731 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 732 | |
3a0f5bbb N |
733 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
734 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
735 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
736 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 737 | md_error(mddev, rdev); |
3a0f5bbb | 738 | } |
7bfa19f2 | 739 | |
a9701a30 N |
740 | if (atomic_dec_and_test(&mddev->pending_writes)) |
741 | wake_up(&mddev->sb_wait); | |
f8b58edf | 742 | bio_put(bio); |
7bfa19f2 N |
743 | } |
744 | ||
fd01b88c | 745 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
746 | sector_t sector, int size, struct page *page) |
747 | { | |
748 | /* write first size bytes of page to sector of rdev | |
749 | * Increment mddev->pending_writes before returning | |
750 | * and decrement it on completion, waking up sb_wait | |
751 | * if zero is reached. | |
752 | * If an error occurred, call md_error | |
753 | */ | |
a167f663 | 754 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 755 | |
a6ff7e08 | 756 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
4f024f37 | 757 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
758 | bio_add_page(bio, page, size, 0); |
759 | bio->bi_private = rdev; | |
760 | bio->bi_end_io = super_written; | |
a9701a30 | 761 | |
7bfa19f2 | 762 | atomic_inc(&mddev->pending_writes); |
a5bf4df0 | 763 | submit_bio(WRITE_FLUSH_FUA, bio); |
a9701a30 N |
764 | } |
765 | ||
fd01b88c | 766 | void md_super_wait(struct mddev *mddev) |
a9701a30 | 767 | { |
e9c7469b | 768 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 769 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
7bfa19f2 N |
770 | } |
771 | ||
3cb03002 | 772 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
ccebd4c4 | 773 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 774 | { |
a167f663 | 775 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
776 | int ret; |
777 | ||
a6ff7e08 JB |
778 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
779 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 | 780 | if (metadata_op) |
4f024f37 | 781 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
782 | else if (rdev->mddev->reshape_position != MaxSector && |
783 | (rdev->mddev->reshape_backwards == | |
784 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 785 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 786 | else |
4f024f37 | 787 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 788 | bio_add_page(bio, page, size, 0); |
c170bbb4 | 789 | submit_bio_wait(rw, bio); |
1da177e4 LT |
790 | |
791 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
792 | bio_put(bio); | |
793 | return ret; | |
794 | } | |
a8745db2 | 795 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 796 | |
3cb03002 | 797 | static int read_disk_sb(struct md_rdev * rdev, int size) |
1da177e4 LT |
798 | { |
799 | char b[BDEVNAME_SIZE]; | |
800 | if (!rdev->sb_page) { | |
801 | MD_BUG(); | |
802 | return -EINVAL; | |
803 | } | |
804 | if (rdev->sb_loaded) | |
805 | return 0; | |
806 | ||
807 | ||
ccebd4c4 | 808 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
809 | goto fail; |
810 | rdev->sb_loaded = 1; | |
811 | return 0; | |
812 | ||
813 | fail: | |
814 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
815 | bdevname(rdev->bdev,b)); | |
816 | return -EINVAL; | |
817 | } | |
818 | ||
819 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
820 | { | |
05710466 AN |
821 | return sb1->set_uuid0 == sb2->set_uuid0 && |
822 | sb1->set_uuid1 == sb2->set_uuid1 && | |
823 | sb1->set_uuid2 == sb2->set_uuid2 && | |
824 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
825 | } |
826 | ||
1da177e4 LT |
827 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
828 | { | |
829 | int ret; | |
830 | mdp_super_t *tmp1, *tmp2; | |
831 | ||
832 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
833 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
834 | ||
835 | if (!tmp1 || !tmp2) { | |
836 | ret = 0; | |
35020f1a | 837 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
838 | goto abort; |
839 | } | |
840 | ||
841 | *tmp1 = *sb1; | |
842 | *tmp2 = *sb2; | |
843 | ||
844 | /* | |
845 | * nr_disks is not constant | |
846 | */ | |
847 | tmp1->nr_disks = 0; | |
848 | tmp2->nr_disks = 0; | |
849 | ||
ce0c8e05 | 850 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 851 | abort: |
990a8baf JJ |
852 | kfree(tmp1); |
853 | kfree(tmp2); | |
1da177e4 LT |
854 | return ret; |
855 | } | |
856 | ||
4d167f09 N |
857 | |
858 | static u32 md_csum_fold(u32 csum) | |
859 | { | |
860 | csum = (csum & 0xffff) + (csum >> 16); | |
861 | return (csum & 0xffff) + (csum >> 16); | |
862 | } | |
863 | ||
1da177e4 LT |
864 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
865 | { | |
4d167f09 N |
866 | u64 newcsum = 0; |
867 | u32 *sb32 = (u32*)sb; | |
868 | int i; | |
1da177e4 LT |
869 | unsigned int disk_csum, csum; |
870 | ||
871 | disk_csum = sb->sb_csum; | |
872 | sb->sb_csum = 0; | |
4d167f09 N |
873 | |
874 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
875 | newcsum += sb32[i]; | |
876 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
877 | ||
878 | ||
879 | #ifdef CONFIG_ALPHA | |
880 | /* This used to use csum_partial, which was wrong for several | |
881 | * reasons including that different results are returned on | |
882 | * different architectures. It isn't critical that we get exactly | |
883 | * the same return value as before (we always csum_fold before | |
884 | * testing, and that removes any differences). However as we | |
885 | * know that csum_partial always returned a 16bit value on | |
886 | * alphas, do a fold to maximise conformity to previous behaviour. | |
887 | */ | |
888 | sb->sb_csum = md_csum_fold(disk_csum); | |
889 | #else | |
1da177e4 | 890 | sb->sb_csum = disk_csum; |
4d167f09 | 891 | #endif |
1da177e4 LT |
892 | return csum; |
893 | } | |
894 | ||
895 | ||
896 | /* | |
897 | * Handle superblock details. | |
898 | * We want to be able to handle multiple superblock formats | |
899 | * so we have a common interface to them all, and an array of | |
900 | * different handlers. | |
901 | * We rely on user-space to write the initial superblock, and support | |
902 | * reading and updating of superblocks. | |
903 | * Interface methods are: | |
3cb03002 | 904 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
905 | * loads and validates a superblock on dev. |
906 | * if refdev != NULL, compare superblocks on both devices | |
907 | * Return: | |
908 | * 0 - dev has a superblock that is compatible with refdev | |
909 | * 1 - dev has a superblock that is compatible and newer than refdev | |
910 | * so dev should be used as the refdev in future | |
911 | * -EINVAL superblock incompatible or invalid | |
912 | * -othererror e.g. -EIO | |
913 | * | |
fd01b88c | 914 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
915 | * Verify that dev is acceptable into mddev. |
916 | * The first time, mddev->raid_disks will be 0, and data from | |
917 | * dev should be merged in. Subsequent calls check that dev | |
918 | * is new enough. Return 0 or -EINVAL | |
919 | * | |
fd01b88c | 920 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
921 | * Update the superblock for rdev with data in mddev |
922 | * This does not write to disc. | |
923 | * | |
924 | */ | |
925 | ||
926 | struct super_type { | |
0cd17fec CW |
927 | char *name; |
928 | struct module *owner; | |
c6563a8c N |
929 | int (*load_super)(struct md_rdev *rdev, |
930 | struct md_rdev *refdev, | |
0cd17fec | 931 | int minor_version); |
c6563a8c N |
932 | int (*validate_super)(struct mddev *mddev, |
933 | struct md_rdev *rdev); | |
934 | void (*sync_super)(struct mddev *mddev, | |
935 | struct md_rdev *rdev); | |
3cb03002 | 936 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 937 | sector_t num_sectors); |
c6563a8c N |
938 | int (*allow_new_offset)(struct md_rdev *rdev, |
939 | unsigned long long new_offset); | |
1da177e4 LT |
940 | }; |
941 | ||
0894cc30 AN |
942 | /* |
943 | * Check that the given mddev has no bitmap. | |
944 | * | |
945 | * This function is called from the run method of all personalities that do not | |
946 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
947 | * has a bitmap. Otherwise, it returns 0. | |
948 | * | |
949 | */ | |
fd01b88c | 950 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 951 | { |
c3d9714e | 952 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
953 | return 0; |
954 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
955 | mdname(mddev), mddev->pers->name); | |
956 | return 1; | |
957 | } | |
958 | EXPORT_SYMBOL(md_check_no_bitmap); | |
959 | ||
1da177e4 LT |
960 | /* |
961 | * load_super for 0.90.0 | |
962 | */ | |
3cb03002 | 963 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
964 | { |
965 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
966 | mdp_super_t *sb; | |
967 | int ret; | |
1da177e4 LT |
968 | |
969 | /* | |
0f420358 | 970 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
971 | * it's at the end of the disk. |
972 | * | |
973 | * It also happens to be a multiple of 4Kb. | |
974 | */ | |
57b2caa3 | 975 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 976 | |
0002b271 | 977 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
978 | if (ret) return ret; |
979 | ||
980 | ret = -EINVAL; | |
981 | ||
982 | bdevname(rdev->bdev, b); | |
65a06f06 | 983 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
984 | |
985 | if (sb->md_magic != MD_SB_MAGIC) { | |
986 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
987 | b); | |
988 | goto abort; | |
989 | } | |
990 | ||
991 | if (sb->major_version != 0 || | |
f6705578 N |
992 | sb->minor_version < 90 || |
993 | sb->minor_version > 91) { | |
1da177e4 LT |
994 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
995 | sb->major_version, sb->minor_version, | |
996 | b); | |
997 | goto abort; | |
998 | } | |
999 | ||
1000 | if (sb->raid_disks <= 0) | |
1001 | goto abort; | |
1002 | ||
4d167f09 | 1003 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
1004 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
1005 | b); | |
1006 | goto abort; | |
1007 | } | |
1008 | ||
1009 | rdev->preferred_minor = sb->md_minor; | |
1010 | rdev->data_offset = 0; | |
c6563a8c | 1011 | rdev->new_data_offset = 0; |
0002b271 | 1012 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1013 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1014 | |
1015 | if (sb->level == LEVEL_MULTIPATH) | |
1016 | rdev->desc_nr = -1; | |
1017 | else | |
1018 | rdev->desc_nr = sb->this_disk.number; | |
1019 | ||
9a7b2b0f | 1020 | if (!refdev) { |
1da177e4 | 1021 | ret = 1; |
9a7b2b0f | 1022 | } else { |
1da177e4 | 1023 | __u64 ev1, ev2; |
65a06f06 | 1024 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1025 | if (!uuid_equal(refsb, sb)) { |
1026 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1027 | b, bdevname(refdev->bdev,b2)); | |
1028 | goto abort; | |
1029 | } | |
1030 | if (!sb_equal(refsb, sb)) { | |
1031 | printk(KERN_WARNING "md: %s has same UUID" | |
1032 | " but different superblock to %s\n", | |
1033 | b, bdevname(refdev->bdev, b2)); | |
1034 | goto abort; | |
1035 | } | |
1036 | ev1 = md_event(sb); | |
1037 | ev2 = md_event(refsb); | |
1038 | if (ev1 > ev2) | |
1039 | ret = 1; | |
1040 | else | |
1041 | ret = 0; | |
1042 | } | |
8190e754 | 1043 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1044 | /* Limit to 4TB as metadata cannot record more than that. |
1045 | * (not needed for Linear and RAID0 as metadata doesn't | |
1046 | * record this size) | |
1047 | */ | |
1048 | if (rdev->sectors >= (2ULL << 32) && sb->level >= 1) | |
27a7b260 | 1049 | rdev->sectors = (2ULL << 32) - 2; |
1da177e4 | 1050 | |
27a7b260 | 1051 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1052 | /* "this cannot possibly happen" ... */ |
1053 | ret = -EINVAL; | |
1054 | ||
1da177e4 LT |
1055 | abort: |
1056 | return ret; | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * validate_super for 0.90.0 | |
1061 | */ | |
fd01b88c | 1062 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1063 | { |
1064 | mdp_disk_t *desc; | |
65a06f06 | 1065 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1066 | __u64 ev1 = md_event(sb); |
1da177e4 | 1067 | |
41158c7e | 1068 | rdev->raid_disk = -1; |
c5d79adb N |
1069 | clear_bit(Faulty, &rdev->flags); |
1070 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1071 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1072 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1073 | |
1da177e4 LT |
1074 | if (mddev->raid_disks == 0) { |
1075 | mddev->major_version = 0; | |
1076 | mddev->minor_version = sb->minor_version; | |
1077 | mddev->patch_version = sb->patch_version; | |
e691063a | 1078 | mddev->external = 0; |
9d8f0363 | 1079 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1080 | mddev->ctime = sb->ctime; |
1081 | mddev->utime = sb->utime; | |
1082 | mddev->level = sb->level; | |
d9d166c2 | 1083 | mddev->clevel[0] = 0; |
1da177e4 LT |
1084 | mddev->layout = sb->layout; |
1085 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1086 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1087 | mddev->events = ev1; |
c3d9714e | 1088 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1089 | mddev->bitmap_info.space = 0; |
1090 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1091 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1092 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1093 | mddev->reshape_backwards = 0; |
1da177e4 | 1094 | |
f6705578 N |
1095 | if (mddev->minor_version >= 91) { |
1096 | mddev->reshape_position = sb->reshape_position; | |
1097 | mddev->delta_disks = sb->delta_disks; | |
1098 | mddev->new_level = sb->new_level; | |
1099 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1100 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1101 | if (mddev->delta_disks < 0) |
1102 | mddev->reshape_backwards = 1; | |
f6705578 N |
1103 | } else { |
1104 | mddev->reshape_position = MaxSector; | |
1105 | mddev->delta_disks = 0; | |
1106 | mddev->new_level = mddev->level; | |
1107 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1108 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1109 | } |
1110 | ||
1da177e4 LT |
1111 | if (sb->state & (1<<MD_SB_CLEAN)) |
1112 | mddev->recovery_cp = MaxSector; | |
1113 | else { | |
1114 | if (sb->events_hi == sb->cp_events_hi && | |
1115 | sb->events_lo == sb->cp_events_lo) { | |
1116 | mddev->recovery_cp = sb->recovery_cp; | |
1117 | } else | |
1118 | mddev->recovery_cp = 0; | |
1119 | } | |
1120 | ||
1121 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1122 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1123 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1124 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1125 | ||
1126 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1127 | |
1128 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1129 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1130 | mddev->bitmap_info.offset = |
1131 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1132 | mddev->bitmap_info.space = |
c9ad020f | 1133 | mddev->bitmap_info.default_space; |
6409bb05 | 1134 | } |
a654b9d8 | 1135 | |
41158c7e | 1136 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1137 | /* Insist on good event counter while assembling, except |
1138 | * for spares (which don't need an event count) */ | |
1da177e4 | 1139 | ++ev1; |
be6800a7 N |
1140 | if (sb->disks[rdev->desc_nr].state & ( |
1141 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
1142 | if (ev1 < mddev->events) | |
1143 | return -EINVAL; | |
41158c7e N |
1144 | } else if (mddev->bitmap) { |
1145 | /* if adding to array with a bitmap, then we can accept an | |
1146 | * older device ... but not too old. | |
1147 | */ | |
41158c7e N |
1148 | if (ev1 < mddev->bitmap->events_cleared) |
1149 | return 0; | |
8313b8e5 N |
1150 | if (ev1 < mddev->events) |
1151 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1152 | } else { |
1153 | if (ev1 < mddev->events) | |
1154 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1155 | return 0; | |
1156 | } | |
41158c7e | 1157 | |
1da177e4 | 1158 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1159 | desc = sb->disks + rdev->desc_nr; |
1160 | ||
1161 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1162 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1163 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1164 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1165 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1166 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1167 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1168 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1169 | /* active but not in sync implies recovery up to | |
1170 | * reshape position. We don't know exactly where | |
1171 | * that is, so set to zero for now */ | |
1172 | if (mddev->minor_version >= 91) { | |
1173 | rdev->recovery_offset = 0; | |
1174 | rdev->raid_disk = desc->raid_disk; | |
1175 | } | |
1da177e4 | 1176 | } |
8ddf9efe N |
1177 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1178 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1179 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1180 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1181 | return 0; |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * sync_super for 0.90.0 | |
1186 | */ | |
fd01b88c | 1187 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1188 | { |
1189 | mdp_super_t *sb; | |
3cb03002 | 1190 | struct md_rdev *rdev2; |
1da177e4 | 1191 | int next_spare = mddev->raid_disks; |
19133a42 | 1192 | |
1da177e4 LT |
1193 | |
1194 | /* make rdev->sb match mddev data.. | |
1195 | * | |
1196 | * 1/ zero out disks | |
1197 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1198 | * 3/ any empty disks < next_spare become removed | |
1199 | * | |
1200 | * disks[0] gets initialised to REMOVED because | |
1201 | * we cannot be sure from other fields if it has | |
1202 | * been initialised or not. | |
1203 | */ | |
1204 | int i; | |
1205 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1206 | ||
61181565 N |
1207 | rdev->sb_size = MD_SB_BYTES; |
1208 | ||
65a06f06 | 1209 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1210 | |
1211 | memset(sb, 0, sizeof(*sb)); | |
1212 | ||
1213 | sb->md_magic = MD_SB_MAGIC; | |
1214 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1215 | sb->patch_version = mddev->patch_version; |
1216 | sb->gvalid_words = 0; /* ignored */ | |
1217 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1218 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1219 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1220 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1221 | ||
1222 | sb->ctime = mddev->ctime; | |
1223 | sb->level = mddev->level; | |
58c0fed4 | 1224 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1225 | sb->raid_disks = mddev->raid_disks; |
1226 | sb->md_minor = mddev->md_minor; | |
e691063a | 1227 | sb->not_persistent = 0; |
1da177e4 LT |
1228 | sb->utime = mddev->utime; |
1229 | sb->state = 0; | |
1230 | sb->events_hi = (mddev->events>>32); | |
1231 | sb->events_lo = (u32)mddev->events; | |
1232 | ||
f6705578 N |
1233 | if (mddev->reshape_position == MaxSector) |
1234 | sb->minor_version = 90; | |
1235 | else { | |
1236 | sb->minor_version = 91; | |
1237 | sb->reshape_position = mddev->reshape_position; | |
1238 | sb->new_level = mddev->new_level; | |
1239 | sb->delta_disks = mddev->delta_disks; | |
1240 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1241 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1242 | } |
1243 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1244 | if (mddev->in_sync) |
1245 | { | |
1246 | sb->recovery_cp = mddev->recovery_cp; | |
1247 | sb->cp_events_hi = (mddev->events>>32); | |
1248 | sb->cp_events_lo = (u32)mddev->events; | |
1249 | if (mddev->recovery_cp == MaxSector) | |
1250 | sb->state = (1<< MD_SB_CLEAN); | |
1251 | } else | |
1252 | sb->recovery_cp = 0; | |
1253 | ||
1254 | sb->layout = mddev->layout; | |
9d8f0363 | 1255 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1256 | |
c3d9714e | 1257 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1258 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1259 | ||
1da177e4 | 1260 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1261 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1262 | mdp_disk_t *d; |
86e6ffdd | 1263 | int desc_nr; |
0261cd9f N |
1264 | int is_active = test_bit(In_sync, &rdev2->flags); |
1265 | ||
1266 | if (rdev2->raid_disk >= 0 && | |
1267 | sb->minor_version >= 91) | |
1268 | /* we have nowhere to store the recovery_offset, | |
1269 | * but if it is not below the reshape_position, | |
1270 | * we can piggy-back on that. | |
1271 | */ | |
1272 | is_active = 1; | |
1273 | if (rdev2->raid_disk < 0 || | |
1274 | test_bit(Faulty, &rdev2->flags)) | |
1275 | is_active = 0; | |
1276 | if (is_active) | |
86e6ffdd | 1277 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1278 | else |
86e6ffdd | 1279 | desc_nr = next_spare++; |
19133a42 | 1280 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1281 | d = &sb->disks[rdev2->desc_nr]; |
1282 | nr_disks++; | |
1283 | d->number = rdev2->desc_nr; | |
1284 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1285 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1286 | if (is_active) |
1da177e4 LT |
1287 | d->raid_disk = rdev2->raid_disk; |
1288 | else | |
1289 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1290 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1291 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1292 | else if (is_active) { |
1da177e4 | 1293 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1294 | if (test_bit(In_sync, &rdev2->flags)) |
1295 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1296 | active++; |
1297 | working++; | |
1298 | } else { | |
1299 | d->state = 0; | |
1300 | spare++; | |
1301 | working++; | |
1302 | } | |
8ddf9efe N |
1303 | if (test_bit(WriteMostly, &rdev2->flags)) |
1304 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1305 | } |
1da177e4 LT |
1306 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1307 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1308 | mdp_disk_t *d = &sb->disks[i]; | |
1309 | if (d->state == 0 && d->number == 0) { | |
1310 | d->number = i; | |
1311 | d->raid_disk = i; | |
1312 | d->state = (1<<MD_DISK_REMOVED); | |
1313 | d->state |= (1<<MD_DISK_FAULTY); | |
1314 | failed++; | |
1315 | } | |
1316 | } | |
1317 | sb->nr_disks = nr_disks; | |
1318 | sb->active_disks = active; | |
1319 | sb->working_disks = working; | |
1320 | sb->failed_disks = failed; | |
1321 | sb->spare_disks = spare; | |
1322 | ||
1323 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1324 | sb->sb_csum = calc_sb_csum(sb); | |
1325 | } | |
1326 | ||
0cd17fec CW |
1327 | /* |
1328 | * rdev_size_change for 0.90.0 | |
1329 | */ | |
1330 | static unsigned long long | |
3cb03002 | 1331 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1332 | { |
58c0fed4 | 1333 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1334 | return 0; /* component must fit device */ |
c3d9714e | 1335 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1336 | return 0; /* can't move bitmap */ |
57b2caa3 | 1337 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1338 | if (!num_sectors || num_sectors > rdev->sb_start) |
1339 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1340 | /* Limit to 4TB as metadata cannot record more than that. |
1341 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1342 | */ | |
667a5313 | 1343 | if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1) |
27a7b260 | 1344 | num_sectors = (2ULL << 32) - 2; |
0f420358 | 1345 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1346 | rdev->sb_page); |
1347 | md_super_wait(rdev->mddev); | |
c26a44ed | 1348 | return num_sectors; |
0cd17fec CW |
1349 | } |
1350 | ||
c6563a8c N |
1351 | static int |
1352 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1353 | { | |
1354 | /* non-zero offset changes not possible with v0.90 */ | |
1355 | return new_offset == 0; | |
1356 | } | |
0cd17fec | 1357 | |
1da177e4 LT |
1358 | /* |
1359 | * version 1 superblock | |
1360 | */ | |
1361 | ||
1c05b4bc | 1362 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1363 | { |
1c05b4bc N |
1364 | __le32 disk_csum; |
1365 | u32 csum; | |
1da177e4 LT |
1366 | unsigned long long newcsum; |
1367 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1368 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1369 | |
1370 | disk_csum = sb->sb_csum; | |
1371 | sb->sb_csum = 0; | |
1372 | newcsum = 0; | |
1f3c9907 | 1373 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1374 | newcsum += le32_to_cpu(*isuper++); |
1375 | ||
1376 | if (size == 2) | |
1c05b4bc | 1377 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1378 | |
1379 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1380 | sb->sb_csum = disk_csum; | |
1381 | return cpu_to_le32(csum); | |
1382 | } | |
1383 | ||
2699b672 N |
1384 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, |
1385 | int acknowledged); | |
3cb03002 | 1386 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1387 | { |
1388 | struct mdp_superblock_1 *sb; | |
1389 | int ret; | |
0f420358 | 1390 | sector_t sb_start; |
c6563a8c | 1391 | sector_t sectors; |
1da177e4 | 1392 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1393 | int bmask; |
1da177e4 LT |
1394 | |
1395 | /* | |
0f420358 | 1396 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1397 | * It is always aligned to a 4K boundary and |
1398 | * depeding on minor_version, it can be: | |
1399 | * 0: At least 8K, but less than 12K, from end of device | |
1400 | * 1: At start of device | |
1401 | * 2: 4K from start of device. | |
1402 | */ | |
1403 | switch(minor_version) { | |
1404 | case 0: | |
77304d2a | 1405 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1406 | sb_start -= 8*2; |
1407 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1408 | break; |
1409 | case 1: | |
0f420358 | 1410 | sb_start = 0; |
1da177e4 LT |
1411 | break; |
1412 | case 2: | |
0f420358 | 1413 | sb_start = 8; |
1da177e4 LT |
1414 | break; |
1415 | default: | |
1416 | return -EINVAL; | |
1417 | } | |
0f420358 | 1418 | rdev->sb_start = sb_start; |
1da177e4 | 1419 | |
0002b271 N |
1420 | /* superblock is rarely larger than 1K, but it can be larger, |
1421 | * and it is safe to read 4k, so we do that | |
1422 | */ | |
1423 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1424 | if (ret) return ret; |
1425 | ||
1426 | ||
65a06f06 | 1427 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1428 | |
1429 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1430 | sb->major_version != cpu_to_le32(1) || | |
1431 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1432 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1433 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1434 | return -EINVAL; |
1435 | ||
1436 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1437 | printk("md: invalid superblock checksum on %s\n", | |
1438 | bdevname(rdev->bdev,b)); | |
1439 | return -EINVAL; | |
1440 | } | |
1441 | if (le64_to_cpu(sb->data_size) < 10) { | |
1442 | printk("md: data_size too small on %s\n", | |
1443 | bdevname(rdev->bdev,b)); | |
1444 | return -EINVAL; | |
1445 | } | |
c6563a8c N |
1446 | if (sb->pad0 || |
1447 | sb->pad3[0] || | |
1448 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1449 | /* Some padding is non-zero, might be a new feature */ | |
1450 | return -EINVAL; | |
e11e93fa | 1451 | |
1da177e4 LT |
1452 | rdev->preferred_minor = 0xffff; |
1453 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1454 | rdev->new_data_offset = rdev->data_offset; |
1455 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1456 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1457 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1458 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1459 | |
0002b271 | 1460 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1461 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1462 | if (rdev->sb_size & bmask) |
a1801f85 N |
1463 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1464 | ||
1465 | if (minor_version | |
0f420358 | 1466 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1467 | return -EINVAL; |
c6563a8c N |
1468 | if (minor_version |
1469 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1470 | return -EINVAL; | |
0002b271 | 1471 | |
31b65a0d N |
1472 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1473 | rdev->desc_nr = -1; | |
1474 | else | |
1475 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1476 | ||
2699b672 N |
1477 | if (!rdev->bb_page) { |
1478 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1479 | if (!rdev->bb_page) | |
1480 | return -ENOMEM; | |
1481 | } | |
1482 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1483 | rdev->badblocks.count == 0) { | |
1484 | /* need to load the bad block list. | |
1485 | * Currently we limit it to one page. | |
1486 | */ | |
1487 | s32 offset; | |
1488 | sector_t bb_sector; | |
1489 | u64 *bbp; | |
1490 | int i; | |
1491 | int sectors = le16_to_cpu(sb->bblog_size); | |
1492 | if (sectors > (PAGE_SIZE / 512)) | |
1493 | return -EINVAL; | |
1494 | offset = le32_to_cpu(sb->bblog_offset); | |
1495 | if (offset == 0) | |
1496 | return -EINVAL; | |
1497 | bb_sector = (long long)offset; | |
1498 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
1499 | rdev->bb_page, READ, true)) | |
1500 | return -EIO; | |
1501 | bbp = (u64 *)page_address(rdev->bb_page); | |
1502 | rdev->badblocks.shift = sb->bblog_shift; | |
1503 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1504 | u64 bb = le64_to_cpu(*bbp); | |
1505 | int count = bb & (0x3ff); | |
1506 | u64 sector = bb >> 10; | |
1507 | sector <<= sb->bblog_shift; | |
1508 | count <<= sb->bblog_shift; | |
1509 | if (bb + 1 == 0) | |
1510 | break; | |
1511 | if (md_set_badblocks(&rdev->badblocks, | |
1512 | sector, count, 1) == 0) | |
1513 | return -EINVAL; | |
1514 | } | |
486adf72 N |
1515 | } else if (sb->bblog_offset != 0) |
1516 | rdev->badblocks.shift = 0; | |
2699b672 | 1517 | |
9a7b2b0f | 1518 | if (!refdev) { |
8ed75463 | 1519 | ret = 1; |
9a7b2b0f | 1520 | } else { |
1da177e4 | 1521 | __u64 ev1, ev2; |
65a06f06 | 1522 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1523 | |
1524 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1525 | sb->level != refsb->level || | |
1526 | sb->layout != refsb->layout || | |
1527 | sb->chunksize != refsb->chunksize) { | |
1528 | printk(KERN_WARNING "md: %s has strangely different" | |
1529 | " superblock to %s\n", | |
1530 | bdevname(rdev->bdev,b), | |
1531 | bdevname(refdev->bdev,b2)); | |
1532 | return -EINVAL; | |
1533 | } | |
1534 | ev1 = le64_to_cpu(sb->events); | |
1535 | ev2 = le64_to_cpu(refsb->events); | |
1536 | ||
1537 | if (ev1 > ev2) | |
8ed75463 N |
1538 | ret = 1; |
1539 | else | |
1540 | ret = 0; | |
1da177e4 | 1541 | } |
c6563a8c N |
1542 | if (minor_version) { |
1543 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1544 | sectors -= rdev->data_offset; | |
1545 | } else | |
1546 | sectors = rdev->sb_start; | |
1547 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1548 | return -EINVAL; |
dd8ac336 | 1549 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1550 | return ret; |
1da177e4 LT |
1551 | } |
1552 | ||
fd01b88c | 1553 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1554 | { |
65a06f06 | 1555 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1556 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1557 | |
41158c7e | 1558 | rdev->raid_disk = -1; |
c5d79adb N |
1559 | clear_bit(Faulty, &rdev->flags); |
1560 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1561 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1562 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1563 | |
1da177e4 LT |
1564 | if (mddev->raid_disks == 0) { |
1565 | mddev->major_version = 1; | |
1566 | mddev->patch_version = 0; | |
e691063a | 1567 | mddev->external = 0; |
9d8f0363 | 1568 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1569 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1570 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1571 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1572 | mddev->clevel[0] = 0; |
1da177e4 LT |
1573 | mddev->layout = le32_to_cpu(sb->layout); |
1574 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1575 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1576 | mddev->events = ev1; |
c3d9714e | 1577 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1578 | mddev->bitmap_info.space = 0; |
1579 | /* Default location for bitmap is 1K after superblock | |
1580 | * using 3K - total of 4K | |
1581 | */ | |
c3d9714e | 1582 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1583 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1584 | mddev->reshape_backwards = 0; |
1585 | ||
1da177e4 LT |
1586 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1587 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1588 | ||
1589 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1590 | |
71c0805c | 1591 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1592 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1593 | mddev->bitmap_info.offset = |
1594 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1595 | /* Metadata doesn't record how much space is available. |
1596 | * For 1.0, we assume we can use up to the superblock | |
1597 | * if before, else to 4K beyond superblock. | |
1598 | * For others, assume no change is possible. | |
1599 | */ | |
1600 | if (mddev->minor_version > 0) | |
1601 | mddev->bitmap_info.space = 0; | |
1602 | else if (mddev->bitmap_info.offset > 0) | |
1603 | mddev->bitmap_info.space = | |
1604 | 8 - mddev->bitmap_info.offset; | |
1605 | else | |
1606 | mddev->bitmap_info.space = | |
1607 | -mddev->bitmap_info.offset; | |
1608 | } | |
e11e93fa | 1609 | |
f6705578 N |
1610 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1611 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1612 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1613 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1614 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1615 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1616 | if (mddev->delta_disks < 0 || |
1617 | (mddev->delta_disks == 0 && | |
1618 | (le32_to_cpu(sb->feature_map) | |
1619 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1620 | mddev->reshape_backwards = 1; | |
f6705578 N |
1621 | } else { |
1622 | mddev->reshape_position = MaxSector; | |
1623 | mddev->delta_disks = 0; | |
1624 | mddev->new_level = mddev->level; | |
1625 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1626 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1627 | } |
1628 | ||
41158c7e | 1629 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1630 | /* Insist of good event counter while assembling, except for |
1631 | * spares (which don't need an event count) */ | |
1da177e4 | 1632 | ++ev1; |
be6800a7 N |
1633 | if (rdev->desc_nr >= 0 && |
1634 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1635 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1636 | if (ev1 < mddev->events) | |
1637 | return -EINVAL; | |
41158c7e N |
1638 | } else if (mddev->bitmap) { |
1639 | /* If adding to array with a bitmap, then we can accept an | |
1640 | * older device, but not too old. | |
1641 | */ | |
41158c7e N |
1642 | if (ev1 < mddev->bitmap->events_cleared) |
1643 | return 0; | |
8313b8e5 N |
1644 | if (ev1 < mddev->events) |
1645 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1646 | } else { |
1647 | if (ev1 < mddev->events) | |
1648 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1649 | return 0; | |
1650 | } | |
1da177e4 LT |
1651 | if (mddev->level != LEVEL_MULTIPATH) { |
1652 | int role; | |
3673f305 N |
1653 | if (rdev->desc_nr < 0 || |
1654 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1655 | role = 0xffff; | |
1656 | rdev->desc_nr = -1; | |
1657 | } else | |
1658 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1659 | switch(role) { |
1660 | case 0xffff: /* spare */ | |
1da177e4 LT |
1661 | break; |
1662 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1663 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1664 | break; |
1665 | default: | |
f466722c | 1666 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1667 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1668 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1669 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1670 | if (!(le32_to_cpu(sb->feature_map) & |
1671 | MD_FEATURE_RECOVERY_BITMAP)) | |
1672 | rdev->saved_raid_disk = -1; | |
1673 | } else | |
5fd6c1dc | 1674 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1675 | rdev->raid_disk = role; |
1676 | break; | |
1677 | } | |
8ddf9efe N |
1678 | if (sb->devflags & WriteMostly1) |
1679 | set_bit(WriteMostly, &rdev->flags); | |
2d78f8c4 N |
1680 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1681 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1682 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1683 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1684 | |
1da177e4 LT |
1685 | return 0; |
1686 | } | |
1687 | ||
fd01b88c | 1688 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1689 | { |
1690 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1691 | struct md_rdev *rdev2; |
1da177e4 LT |
1692 | int max_dev, i; |
1693 | /* make rdev->sb match mddev and rdev data. */ | |
1694 | ||
65a06f06 | 1695 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1696 | |
1697 | sb->feature_map = 0; | |
1698 | sb->pad0 = 0; | |
5fd6c1dc | 1699 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1700 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1701 | ||
1702 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1703 | sb->events = cpu_to_le64(mddev->events); | |
1704 | if (mddev->in_sync) | |
1705 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1706 | else | |
1707 | sb->resync_offset = cpu_to_le64(0); | |
1708 | ||
1c05b4bc | 1709 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1710 | |
f0ca340c | 1711 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1712 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1713 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1714 | sb->level = cpu_to_le32(mddev->level); |
1715 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1716 | |
aeb9b211 N |
1717 | if (test_bit(WriteMostly, &rdev->flags)) |
1718 | sb->devflags |= WriteMostly1; | |
1719 | else | |
1720 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1721 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1722 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1723 | |
c3d9714e N |
1724 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1725 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1726 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1727 | } |
5fd6c1dc N |
1728 | |
1729 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1730 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1731 | sb->feature_map |= |
1732 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1733 | sb->recovery_offset = | |
1734 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1735 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1736 | sb->feature_map |= | |
1737 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1738 | } |
2d78f8c4 N |
1739 | if (test_bit(Replacement, &rdev->flags)) |
1740 | sb->feature_map |= | |
1741 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1742 | |
f6705578 N |
1743 | if (mddev->reshape_position != MaxSector) { |
1744 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1745 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1746 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1747 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1748 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1749 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1750 | if (mddev->delta_disks == 0 && |
1751 | mddev->reshape_backwards) | |
1752 | sb->feature_map | |
1753 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1754 | if (rdev->new_data_offset != rdev->data_offset) { |
1755 | sb->feature_map | |
1756 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1757 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1758 | - rdev->data_offset)); | |
1759 | } | |
f6705578 | 1760 | } |
a654b9d8 | 1761 | |
2699b672 N |
1762 | if (rdev->badblocks.count == 0) |
1763 | /* Nothing to do for bad blocks*/ ; | |
1764 | else if (sb->bblog_offset == 0) | |
1765 | /* Cannot record bad blocks on this device */ | |
1766 | md_error(mddev, rdev); | |
1767 | else { | |
1768 | struct badblocks *bb = &rdev->badblocks; | |
1769 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1770 | u64 *p = bb->page; | |
1771 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1772 | if (bb->changed) { | |
1773 | unsigned seq; | |
1774 | ||
1775 | retry: | |
1776 | seq = read_seqbegin(&bb->lock); | |
1777 | ||
1778 | memset(bbp, 0xff, PAGE_SIZE); | |
1779 | ||
1780 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1781 | u64 internal_bb = p[i]; |
2699b672 N |
1782 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1783 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1784 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1785 | } |
d0962936 | 1786 | bb->changed = 0; |
2699b672 N |
1787 | if (read_seqretry(&bb->lock, seq)) |
1788 | goto retry; | |
1789 | ||
1790 | bb->sector = (rdev->sb_start + | |
1791 | (int)le32_to_cpu(sb->bblog_offset)); | |
1792 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1793 | } |
1794 | } | |
1795 | ||
1da177e4 | 1796 | max_dev = 0; |
dafb20fa | 1797 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1798 | if (rdev2->desc_nr+1 > max_dev) |
1799 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1800 | |
70471daf N |
1801 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1802 | int bmask; | |
a778b73f | 1803 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1804 | rdev->sb_size = max_dev * 2 + 256; |
1805 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1806 | if (rdev->sb_size & bmask) | |
1807 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1808 | } else |
1809 | max_dev = le32_to_cpu(sb->max_dev); | |
1810 | ||
1da177e4 LT |
1811 | for (i=0; i<max_dev;i++) |
1812 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1813 | ||
dafb20fa | 1814 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1815 | i = rdev2->desc_nr; |
b2d444d7 | 1816 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1817 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1818 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1819 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1820 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1821 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1822 | else |
1823 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1824 | } | |
1825 | ||
1da177e4 LT |
1826 | sb->sb_csum = calc_sb_1_csum(sb); |
1827 | } | |
1828 | ||
0cd17fec | 1829 | static unsigned long long |
3cb03002 | 1830 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1831 | { |
1832 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1833 | sector_t max_sectors; |
58c0fed4 | 1834 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1835 | return 0; /* component must fit device */ |
c6563a8c N |
1836 | if (rdev->data_offset != rdev->new_data_offset) |
1837 | return 0; /* too confusing */ | |
0f420358 | 1838 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1839 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1840 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1841 | max_sectors -= rdev->data_offset; |
1842 | if (!num_sectors || num_sectors > max_sectors) | |
1843 | num_sectors = max_sectors; | |
c3d9714e | 1844 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1845 | /* minor version 0 with bitmap we can't move */ |
1846 | return 0; | |
1847 | } else { | |
1848 | /* minor version 0; superblock after data */ | |
0f420358 | 1849 | sector_t sb_start; |
77304d2a | 1850 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1851 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1852 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1853 | if (!num_sectors || num_sectors > max_sectors) |
1854 | num_sectors = max_sectors; | |
0f420358 | 1855 | rdev->sb_start = sb_start; |
0cd17fec | 1856 | } |
65a06f06 | 1857 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1858 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1859 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1860 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1861 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1862 | rdev->sb_page); |
1863 | md_super_wait(rdev->mddev); | |
c26a44ed | 1864 | return num_sectors; |
c6563a8c N |
1865 | |
1866 | } | |
1867 | ||
1868 | static int | |
1869 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1870 | unsigned long long new_offset) | |
1871 | { | |
1872 | /* All necessary checks on new >= old have been done */ | |
1873 | struct bitmap *bitmap; | |
1874 | if (new_offset >= rdev->data_offset) | |
1875 | return 1; | |
1876 | ||
1877 | /* with 1.0 metadata, there is no metadata to tread on | |
1878 | * so we can always move back */ | |
1879 | if (rdev->mddev->minor_version == 0) | |
1880 | return 1; | |
1881 | ||
1882 | /* otherwise we must be sure not to step on | |
1883 | * any metadata, so stay: | |
1884 | * 36K beyond start of superblock | |
1885 | * beyond end of badblocks | |
1886 | * beyond write-intent bitmap | |
1887 | */ | |
1888 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1889 | return 0; | |
1890 | bitmap = rdev->mddev->bitmap; | |
1891 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1892 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1893 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1894 | return 0; |
1895 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1896 | return 0; | |
1897 | ||
1898 | return 1; | |
0cd17fec | 1899 | } |
1da177e4 | 1900 | |
75c96f85 | 1901 | static struct super_type super_types[] = { |
1da177e4 LT |
1902 | [0] = { |
1903 | .name = "0.90.0", | |
1904 | .owner = THIS_MODULE, | |
0cd17fec CW |
1905 | .load_super = super_90_load, |
1906 | .validate_super = super_90_validate, | |
1907 | .sync_super = super_90_sync, | |
1908 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1909 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1910 | }, |
1911 | [1] = { | |
1912 | .name = "md-1", | |
1913 | .owner = THIS_MODULE, | |
0cd17fec CW |
1914 | .load_super = super_1_load, |
1915 | .validate_super = super_1_validate, | |
1916 | .sync_super = super_1_sync, | |
1917 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1918 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1919 | }, |
1920 | }; | |
1da177e4 | 1921 | |
fd01b88c | 1922 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1923 | { |
1924 | if (mddev->sync_super) { | |
1925 | mddev->sync_super(mddev, rdev); | |
1926 | return; | |
1927 | } | |
1928 | ||
1929 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1930 | ||
1931 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1932 | } | |
1933 | ||
fd01b88c | 1934 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1935 | { |
3cb03002 | 1936 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1937 | |
4b80991c N |
1938 | rcu_read_lock(); |
1939 | rdev_for_each_rcu(rdev, mddev1) | |
1940 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1941 | if (rdev->bdev->bd_contains == |
4b80991c N |
1942 | rdev2->bdev->bd_contains) { |
1943 | rcu_read_unlock(); | |
7dd5e7c3 | 1944 | return 1; |
4b80991c N |
1945 | } |
1946 | rcu_read_unlock(); | |
1da177e4 LT |
1947 | return 0; |
1948 | } | |
1949 | ||
1950 | static LIST_HEAD(pending_raid_disks); | |
1951 | ||
ac5e7113 AN |
1952 | /* |
1953 | * Try to register data integrity profile for an mddev | |
1954 | * | |
1955 | * This is called when an array is started and after a disk has been kicked | |
1956 | * from the array. It only succeeds if all working and active component devices | |
1957 | * are integrity capable with matching profiles. | |
1958 | */ | |
fd01b88c | 1959 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 1960 | { |
3cb03002 | 1961 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
1962 | |
1963 | if (list_empty(&mddev->disks)) | |
1964 | return 0; /* nothing to do */ | |
629acb6a JB |
1965 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
1966 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 1967 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
1968 | /* skip spares and non-functional disks */ |
1969 | if (test_bit(Faulty, &rdev->flags)) | |
1970 | continue; | |
1971 | if (rdev->raid_disk < 0) | |
1972 | continue; | |
ac5e7113 AN |
1973 | if (!reference) { |
1974 | /* Use the first rdev as the reference */ | |
1975 | reference = rdev; | |
1976 | continue; | |
1977 | } | |
1978 | /* does this rdev's profile match the reference profile? */ | |
1979 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1980 | rdev->bdev->bd_disk) < 0) | |
1981 | return -EINVAL; | |
1982 | } | |
89078d57 MP |
1983 | if (!reference || !bdev_get_integrity(reference->bdev)) |
1984 | return 0; | |
ac5e7113 AN |
1985 | /* |
1986 | * All component devices are integrity capable and have matching | |
1987 | * profiles, register the common profile for the md device. | |
1988 | */ | |
1989 | if (blk_integrity_register(mddev->gendisk, | |
1990 | bdev_get_integrity(reference->bdev)) != 0) { | |
1991 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1992 | mdname(mddev)); | |
1993 | return -EINVAL; | |
1994 | } | |
a91a2785 MP |
1995 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
1996 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
1997 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
1998 | mdname(mddev)); | |
1999 | return -EINVAL; | |
2000 | } | |
ac5e7113 AN |
2001 | return 0; |
2002 | } | |
2003 | EXPORT_SYMBOL(md_integrity_register); | |
2004 | ||
2005 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
fd01b88c | 2006 | void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) |
3f9d99c1 | 2007 | { |
2863b9eb JB |
2008 | struct blk_integrity *bi_rdev; |
2009 | struct blk_integrity *bi_mddev; | |
2010 | ||
2011 | if (!mddev->gendisk) | |
2012 | return; | |
2013 | ||
2014 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2015 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2016 | |
ac5e7113 | 2017 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 2018 | return; |
ac5e7113 | 2019 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 2020 | return; |
ac5e7113 AN |
2021 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
2022 | rdev->bdev->bd_disk) >= 0) | |
2023 | return; | |
2024 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
2025 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 2026 | } |
ac5e7113 | 2027 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2028 | |
fd01b88c | 2029 | static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev) |
1da177e4 | 2030 | { |
7dd5e7c3 | 2031 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2032 | struct kobject *ko; |
1edf80d3 | 2033 | char *s; |
5e55e2f5 | 2034 | int err; |
1da177e4 LT |
2035 | |
2036 | if (rdev->mddev) { | |
2037 | MD_BUG(); | |
2038 | return -EINVAL; | |
2039 | } | |
11e2ede0 DW |
2040 | |
2041 | /* prevent duplicates */ | |
2042 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2043 | return -EEXIST; | |
2044 | ||
dd8ac336 AN |
2045 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
2046 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
2047 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2048 | if (mddev->pers) { |
2049 | /* Cannot change size, so fail | |
2050 | * If mddev->level <= 0, then we don't care | |
2051 | * about aligning sizes (e.g. linear) | |
2052 | */ | |
2053 | if (mddev->level > 0) | |
2054 | return -ENOSPC; | |
2055 | } else | |
dd8ac336 | 2056 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2057 | } |
1da177e4 LT |
2058 | |
2059 | /* Verify rdev->desc_nr is unique. | |
2060 | * If it is -1, assign a free number, else | |
2061 | * check number is not in use | |
2062 | */ | |
2063 | if (rdev->desc_nr < 0) { | |
2064 | int choice = 0; | |
2065 | if (mddev->pers) choice = mddev->raid_disks; | |
2066 | while (find_rdev_nr(mddev, choice)) | |
2067 | choice++; | |
2068 | rdev->desc_nr = choice; | |
2069 | } else { | |
2070 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
2071 | return -EBUSY; | |
2072 | } | |
de01dfad N |
2073 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
2074 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
2075 | mdname(mddev), mddev->max_disks); | |
2076 | return -EBUSY; | |
2077 | } | |
19133a42 | 2078 | bdevname(rdev->bdev,b); |
649316b2 | 2079 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 2080 | *s = '!'; |
649316b2 | 2081 | |
1da177e4 | 2082 | rdev->mddev = mddev; |
19133a42 | 2083 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 2084 | |
b2d6db58 | 2085 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2086 | goto fail; |
86e6ffdd | 2087 | |
0762b8bd | 2088 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2089 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2090 | /* failure here is OK */; | |
2091 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2092 | |
4b80991c | 2093 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2094 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2095 | |
2096 | /* May as well allow recovery to be retried once */ | |
5389042f | 2097 | mddev->recovery_disabled++; |
3f9d99c1 | 2098 | |
1da177e4 | 2099 | return 0; |
5e55e2f5 N |
2100 | |
2101 | fail: | |
2102 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
2103 | b, mdname(mddev)); | |
2104 | return err; | |
1da177e4 LT |
2105 | } |
2106 | ||
177a99b2 | 2107 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2108 | { |
3cb03002 | 2109 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2110 | kobject_del(&rdev->kobj); |
177a99b2 | 2111 | kobject_put(&rdev->kobj); |
5792a285 N |
2112 | } |
2113 | ||
3cb03002 | 2114 | static void unbind_rdev_from_array(struct md_rdev * rdev) |
1da177e4 LT |
2115 | { |
2116 | char b[BDEVNAME_SIZE]; | |
2117 | if (!rdev->mddev) { | |
2118 | MD_BUG(); | |
2119 | return; | |
2120 | } | |
49731baa | 2121 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2122 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
2123 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
2124 | rdev->mddev = NULL; | |
86e6ffdd | 2125 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2126 | sysfs_put(rdev->sysfs_state); |
2127 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2128 | rdev->badblocks.count = 0; |
5792a285 | 2129 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2130 | * writing to 'remove' to "dev/state". We also need |
2131 | * to delay it due to rcu usage. | |
5792a285 | 2132 | */ |
4b80991c | 2133 | synchronize_rcu(); |
177a99b2 N |
2134 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2135 | kobject_get(&rdev->kobj); | |
e804ac78 | 2136 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2137 | } |
2138 | ||
2139 | /* | |
2140 | * prevent the device from being mounted, repartitioned or | |
2141 | * otherwise reused by a RAID array (or any other kernel | |
2142 | * subsystem), by bd_claiming the device. | |
2143 | */ | |
3cb03002 | 2144 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2145 | { |
2146 | int err = 0; | |
2147 | struct block_device *bdev; | |
2148 | char b[BDEVNAME_SIZE]; | |
2149 | ||
d4d77629 | 2150 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2151 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 LT |
2152 | if (IS_ERR(bdev)) { |
2153 | printk(KERN_ERR "md: could not open %s.\n", | |
2154 | __bdevname(dev, b)); | |
2155 | return PTR_ERR(bdev); | |
2156 | } | |
1da177e4 LT |
2157 | rdev->bdev = bdev; |
2158 | return err; | |
2159 | } | |
2160 | ||
3cb03002 | 2161 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2162 | { |
2163 | struct block_device *bdev = rdev->bdev; | |
2164 | rdev->bdev = NULL; | |
2165 | if (!bdev) | |
2166 | MD_BUG(); | |
e525fd89 | 2167 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2168 | } |
2169 | ||
2170 | void md_autodetect_dev(dev_t dev); | |
2171 | ||
3cb03002 | 2172 | static void export_rdev(struct md_rdev * rdev) |
1da177e4 LT |
2173 | { |
2174 | char b[BDEVNAME_SIZE]; | |
2175 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
2176 | bdevname(rdev->bdev,b)); | |
2177 | if (rdev->mddev) | |
2178 | MD_BUG(); | |
545c8795 | 2179 | md_rdev_clear(rdev); |
1da177e4 | 2180 | #ifndef MODULE |
d0fae18f N |
2181 | if (test_bit(AutoDetected, &rdev->flags)) |
2182 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2183 | #endif |
2184 | unlock_rdev(rdev); | |
86e6ffdd | 2185 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2186 | } |
2187 | ||
3cb03002 | 2188 | static void kick_rdev_from_array(struct md_rdev * rdev) |
1da177e4 LT |
2189 | { |
2190 | unbind_rdev_from_array(rdev); | |
2191 | export_rdev(rdev); | |
2192 | } | |
2193 | ||
fd01b88c | 2194 | static void export_array(struct mddev *mddev) |
1da177e4 | 2195 | { |
3cb03002 | 2196 | struct md_rdev *rdev, *tmp; |
1da177e4 | 2197 | |
dafb20fa | 2198 | rdev_for_each_safe(rdev, tmp, mddev) { |
1da177e4 LT |
2199 | if (!rdev->mddev) { |
2200 | MD_BUG(); | |
2201 | continue; | |
2202 | } | |
2203 | kick_rdev_from_array(rdev); | |
2204 | } | |
2205 | if (!list_empty(&mddev->disks)) | |
2206 | MD_BUG(); | |
2207 | mddev->raid_disks = 0; | |
2208 | mddev->major_version = 0; | |
2209 | } | |
2210 | ||
2211 | static void print_desc(mdp_disk_t *desc) | |
2212 | { | |
2213 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
2214 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
2215 | } | |
2216 | ||
cd2ac932 | 2217 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
2218 | { |
2219 | int i; | |
2220 | ||
2221 | printk(KERN_INFO | |
2222 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
2223 | sb->major_version, sb->minor_version, sb->patch_version, | |
2224 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
2225 | sb->ctime); | |
2226 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
2227 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
2228 | sb->md_minor, sb->layout, sb->chunk_size); | |
2229 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
2230 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
2231 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
2232 | sb->failed_disks, sb->spare_disks, | |
2233 | sb->sb_csum, (unsigned long)sb->events_lo); | |
2234 | ||
2235 | printk(KERN_INFO); | |
2236 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2237 | mdp_disk_t *desc; | |
2238 | ||
2239 | desc = sb->disks + i; | |
2240 | if (desc->number || desc->major || desc->minor || | |
2241 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
2242 | printk(" D %2d: ", i); | |
2243 | print_desc(desc); | |
2244 | } | |
2245 | } | |
2246 | printk(KERN_INFO "md: THIS: "); | |
2247 | print_desc(&sb->this_disk); | |
cd2ac932 | 2248 | } |
1da177e4 | 2249 | |
cd2ac932 CR |
2250 | static void print_sb_1(struct mdp_superblock_1 *sb) |
2251 | { | |
2252 | __u8 *uuid; | |
2253 | ||
2254 | uuid = sb->set_uuid; | |
ad361c98 | 2255 | printk(KERN_INFO |
7b75c2f8 | 2256 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 2257 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
2258 | le32_to_cpu(sb->major_version), |
2259 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 2260 | uuid, |
cd2ac932 CR |
2261 | sb->set_name, |
2262 | (unsigned long long)le64_to_cpu(sb->ctime) | |
2263 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
2264 | ||
2265 | uuid = sb->device_uuid; | |
ad361c98 JP |
2266 | printk(KERN_INFO |
2267 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 2268 | " RO:%llu\n" |
7b75c2f8 | 2269 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
2270 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
2271 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
2272 | le32_to_cpu(sb->level), |
2273 | (unsigned long long)le64_to_cpu(sb->size), | |
2274 | le32_to_cpu(sb->raid_disks), | |
2275 | le32_to_cpu(sb->layout), | |
2276 | le32_to_cpu(sb->chunksize), | |
2277 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2278 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2279 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2280 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2281 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2282 | uuid, |
cd2ac932 CR |
2283 | sb->devflags, |
2284 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2285 | (unsigned long long)le64_to_cpu(sb->events), | |
2286 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2287 | le32_to_cpu(sb->sb_csum), | |
2288 | le32_to_cpu(sb->max_dev) | |
2289 | ); | |
1da177e4 LT |
2290 | } |
2291 | ||
3cb03002 | 2292 | static void print_rdev(struct md_rdev *rdev, int major_version) |
1da177e4 LT |
2293 | { |
2294 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2295 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2296 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2297 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2298 | rdev->desc_nr); | |
1da177e4 | 2299 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2300 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2301 | switch (major_version) { | |
2302 | case 0: | |
65a06f06 | 2303 | print_sb_90(page_address(rdev->sb_page)); |
cd2ac932 CR |
2304 | break; |
2305 | case 1: | |
65a06f06 | 2306 | print_sb_1(page_address(rdev->sb_page)); |
cd2ac932 CR |
2307 | break; |
2308 | } | |
1da177e4 LT |
2309 | } else |
2310 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2311 | } | |
2312 | ||
5e56341d | 2313 | static void md_print_devices(void) |
1da177e4 | 2314 | { |
159ec1fc | 2315 | struct list_head *tmp; |
3cb03002 | 2316 | struct md_rdev *rdev; |
fd01b88c | 2317 | struct mddev *mddev; |
1da177e4 LT |
2318 | char b[BDEVNAME_SIZE]; |
2319 | ||
2320 | printk("\n"); | |
2321 | printk("md: **********************************\n"); | |
2322 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2323 | printk("md: **********************************\n"); | |
29ac4aa3 | 2324 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2325 | |
32a7627c N |
2326 | if (mddev->bitmap) |
2327 | bitmap_print_sb(mddev->bitmap); | |
2328 | else | |
2329 | printk("%s: ", mdname(mddev)); | |
dafb20fa | 2330 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
2331 | printk("<%s>", bdevname(rdev->bdev,b)); |
2332 | printk("\n"); | |
2333 | ||
dafb20fa | 2334 | rdev_for_each(rdev, mddev) |
cd2ac932 | 2335 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2336 | } |
2337 | printk("md: **********************************\n"); | |
2338 | printk("\n"); | |
2339 | } | |
2340 | ||
2341 | ||
fd01b88c | 2342 | static void sync_sbs(struct mddev * mddev, int nospares) |
1da177e4 | 2343 | { |
42543769 N |
2344 | /* Update each superblock (in-memory image), but |
2345 | * if we are allowed to, skip spares which already | |
2346 | * have the right event counter, or have one earlier | |
2347 | * (which would mean they aren't being marked as dirty | |
2348 | * with the rest of the array) | |
2349 | */ | |
3cb03002 | 2350 | struct md_rdev *rdev; |
dafb20fa | 2351 | rdev_for_each(rdev, mddev) { |
42543769 N |
2352 | if (rdev->sb_events == mddev->events || |
2353 | (nospares && | |
2354 | rdev->raid_disk < 0 && | |
42543769 N |
2355 | rdev->sb_events+1 == mddev->events)) { |
2356 | /* Don't update this superblock */ | |
2357 | rdev->sb_loaded = 2; | |
2358 | } else { | |
076f968b | 2359 | sync_super(mddev, rdev); |
42543769 N |
2360 | rdev->sb_loaded = 1; |
2361 | } | |
1da177e4 LT |
2362 | } |
2363 | } | |
2364 | ||
fd01b88c | 2365 | static void md_update_sb(struct mddev * mddev, int force_change) |
1da177e4 | 2366 | { |
3cb03002 | 2367 | struct md_rdev *rdev; |
06d91a5f | 2368 | int sync_req; |
42543769 | 2369 | int nospares = 0; |
2699b672 | 2370 | int any_badblocks_changed = 0; |
1da177e4 | 2371 | |
d87f064f N |
2372 | if (mddev->ro) { |
2373 | if (force_change) | |
2374 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2375 | return; | |
2376 | } | |
1da177e4 | 2377 | repeat: |
3a3a5ddb | 2378 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2379 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2380 | if (rdev->raid_disk >= 0 && |
2381 | mddev->delta_disks >= 0 && | |
2382 | !test_bit(In_sync, &rdev->flags) && | |
2383 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2384 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2385 | ||
2386 | } | |
bd52b746 | 2387 | if (!mddev->persistent) { |
070dc6dd | 2388 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2389 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
de393cde | 2390 | if (!mddev->external) { |
d97a41dc | 2391 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
dafb20fa | 2392 | rdev_for_each(rdev, mddev) { |
de393cde | 2393 | if (rdev->badblocks.changed) { |
d0962936 | 2394 | rdev->badblocks.changed = 0; |
de393cde N |
2395 | md_ack_all_badblocks(&rdev->badblocks); |
2396 | md_error(mddev, rdev); | |
2397 | } | |
2398 | clear_bit(Blocked, &rdev->flags); | |
2399 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2400 | wake_up(&rdev->blocked_wait); | |
2401 | } | |
2402 | } | |
3a3a5ddb N |
2403 | wake_up(&mddev->sb_wait); |
2404 | return; | |
2405 | } | |
2406 | ||
a9701a30 | 2407 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2408 | |
3a3a5ddb N |
2409 | mddev->utime = get_seconds(); |
2410 | ||
850b2b42 N |
2411 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2412 | force_change = 1; | |
2413 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2414 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2415 | * though if events isn't the right even/odd, we will have to do | |
2416 | * spares after all | |
2417 | */ | |
2418 | nospares = 1; | |
2419 | if (force_change) | |
2420 | nospares = 0; | |
2421 | if (mddev->degraded) | |
84692195 N |
2422 | /* If the array is degraded, then skipping spares is both |
2423 | * dangerous and fairly pointless. | |
2424 | * Dangerous because a device that was removed from the array | |
2425 | * might have a event_count that still looks up-to-date, | |
2426 | * so it can be re-added without a resync. | |
2427 | * Pointless because if there are any spares to skip, | |
2428 | * then a recovery will happen and soon that array won't | |
2429 | * be degraded any more and the spare can go back to sleep then. | |
2430 | */ | |
850b2b42 | 2431 | nospares = 0; |
84692195 | 2432 | |
06d91a5f | 2433 | sync_req = mddev->in_sync; |
42543769 N |
2434 | |
2435 | /* If this is just a dirty<->clean transition, and the array is clean | |
2436 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2437 | if (nospares |
42543769 | 2438 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2439 | && mddev->can_decrease_events |
2440 | && mddev->events != 1) { | |
42543769 | 2441 | mddev->events--; |
a8707c08 N |
2442 | mddev->can_decrease_events = 0; |
2443 | } else { | |
42543769 N |
2444 | /* otherwise we have to go forward and ... */ |
2445 | mddev->events ++; | |
a8707c08 | 2446 | mddev->can_decrease_events = nospares; |
42543769 | 2447 | } |
1da177e4 LT |
2448 | |
2449 | if (!mddev->events) { | |
2450 | /* | |
2451 | * oops, this 64-bit counter should never wrap. | |
2452 | * Either we are in around ~1 trillion A.C., assuming | |
2453 | * 1 reboot per second, or we have a bug: | |
2454 | */ | |
2455 | MD_BUG(); | |
2456 | mddev->events --; | |
2457 | } | |
2699b672 | 2458 | |
dafb20fa | 2459 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2460 | if (rdev->badblocks.changed) |
2461 | any_badblocks_changed++; | |
de393cde N |
2462 | if (test_bit(Faulty, &rdev->flags)) |
2463 | set_bit(FaultRecorded, &rdev->flags); | |
2464 | } | |
2699b672 | 2465 | |
e691063a | 2466 | sync_sbs(mddev, nospares); |
a9701a30 | 2467 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 | 2468 | |
36a4e1fe N |
2469 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2470 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2471 | |
4ad13663 | 2472 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2473 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2474 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2475 | |
42543769 N |
2476 | if (rdev->sb_loaded != 1) |
2477 | continue; /* no noise on spare devices */ | |
1da177e4 | 2478 | |
f466722c | 2479 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2480 | md_super_write(mddev,rdev, |
0f420358 | 2481 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2482 | rdev->sb_page); |
36a4e1fe N |
2483 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2484 | bdevname(rdev->bdev, b), | |
2485 | (unsigned long long)rdev->sb_start); | |
42543769 | 2486 | rdev->sb_events = mddev->events; |
2699b672 N |
2487 | if (rdev->badblocks.size) { |
2488 | md_super_write(mddev, rdev, | |
2489 | rdev->badblocks.sector, | |
2490 | rdev->badblocks.size << 9, | |
2491 | rdev->bb_page); | |
2492 | rdev->badblocks.size = 0; | |
2493 | } | |
7bfa19f2 | 2494 | |
f466722c | 2495 | } else |
36a4e1fe N |
2496 | pr_debug("md: %s (skipping faulty)\n", |
2497 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2498 | |
7bfa19f2 | 2499 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2500 | /* only need to write one superblock... */ |
2501 | break; | |
2502 | } | |
a9701a30 | 2503 | md_super_wait(mddev); |
850b2b42 | 2504 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2505 | |
a9701a30 | 2506 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2507 | if (mddev->in_sync != sync_req || |
2508 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2509 | /* have to write it out again */ |
a9701a30 | 2510 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2511 | goto repeat; |
2512 | } | |
850b2b42 | 2513 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2514 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2515 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2516 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2517 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2518 | |
dafb20fa | 2519 | rdev_for_each(rdev, mddev) { |
de393cde N |
2520 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2521 | clear_bit(Blocked, &rdev->flags); | |
2522 | ||
2523 | if (any_badblocks_changed) | |
2699b672 | 2524 | md_ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2525 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2526 | wake_up(&rdev->blocked_wait); | |
2527 | } | |
1da177e4 LT |
2528 | } |
2529 | ||
7f6ce769 | 2530 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2531 | * We want to accept with case. For this we use cmd_match. |
2532 | */ | |
2533 | static int cmd_match(const char *cmd, const char *str) | |
2534 | { | |
2535 | /* See if cmd, written into a sysfs file, matches | |
2536 | * str. They must either be the same, or cmd can | |
2537 | * have a trailing newline | |
2538 | */ | |
2539 | while (*cmd && *str && *cmd == *str) { | |
2540 | cmd++; | |
2541 | str++; | |
2542 | } | |
2543 | if (*cmd == '\n') | |
2544 | cmd++; | |
2545 | if (*str || *cmd) | |
2546 | return 0; | |
2547 | return 1; | |
2548 | } | |
2549 | ||
86e6ffdd N |
2550 | struct rdev_sysfs_entry { |
2551 | struct attribute attr; | |
3cb03002 N |
2552 | ssize_t (*show)(struct md_rdev *, char *); |
2553 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2554 | }; |
2555 | ||
2556 | static ssize_t | |
3cb03002 | 2557 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd N |
2558 | { |
2559 | char *sep = ""; | |
20a49ff6 | 2560 | size_t len = 0; |
86e6ffdd | 2561 | |
de393cde N |
2562 | if (test_bit(Faulty, &rdev->flags) || |
2563 | rdev->badblocks.unacked_exist) { | |
86e6ffdd N |
2564 | len+= sprintf(page+len, "%sfaulty",sep); |
2565 | sep = ","; | |
2566 | } | |
b2d444d7 | 2567 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2568 | len += sprintf(page+len, "%sin_sync",sep); |
2569 | sep = ","; | |
2570 | } | |
f655675b N |
2571 | if (test_bit(WriteMostly, &rdev->flags)) { |
2572 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2573 | sep = ","; | |
2574 | } | |
de393cde | 2575 | if (test_bit(Blocked, &rdev->flags) || |
52c64152 N |
2576 | (rdev->badblocks.unacked_exist |
2577 | && !test_bit(Faulty, &rdev->flags))) { | |
6bfe0b49 DW |
2578 | len += sprintf(page+len, "%sblocked", sep); |
2579 | sep = ","; | |
2580 | } | |
b2d444d7 N |
2581 | if (!test_bit(Faulty, &rdev->flags) && |
2582 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2583 | len += sprintf(page+len, "%sspare", sep); |
2584 | sep = ","; | |
2585 | } | |
d7a9d443 N |
2586 | if (test_bit(WriteErrorSeen, &rdev->flags)) { |
2587 | len += sprintf(page+len, "%swrite_error", sep); | |
2588 | sep = ","; | |
2589 | } | |
2d78f8c4 N |
2590 | if (test_bit(WantReplacement, &rdev->flags)) { |
2591 | len += sprintf(page+len, "%swant_replacement", sep); | |
2592 | sep = ","; | |
2593 | } | |
2594 | if (test_bit(Replacement, &rdev->flags)) { | |
2595 | len += sprintf(page+len, "%sreplacement", sep); | |
2596 | sep = ","; | |
2597 | } | |
2598 | ||
86e6ffdd N |
2599 | return len+sprintf(page+len, "\n"); |
2600 | } | |
2601 | ||
45dc2de1 | 2602 | static ssize_t |
3cb03002 | 2603 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2604 | { |
2605 | /* can write | |
de393cde | 2606 | * faulty - simulates an error |
45dc2de1 | 2607 | * remove - disconnects the device |
f655675b N |
2608 | * writemostly - sets write_mostly |
2609 | * -writemostly - clears write_mostly | |
de393cde N |
2610 | * blocked - sets the Blocked flags |
2611 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2612 | * insync - sets Insync providing device isn't active |
f466722c N |
2613 | * -insync - clear Insync for a device with a slot assigned, |
2614 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2615 | * write_error - sets WriteErrorSeen |
2616 | * -write_error - clears WriteErrorSeen | |
45dc2de1 N |
2617 | */ |
2618 | int err = -EINVAL; | |
2619 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2620 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2621 | if (test_bit(Faulty, &rdev->flags)) |
2622 | err = 0; | |
2623 | else | |
2624 | err = -EBUSY; | |
45dc2de1 N |
2625 | } else if (cmd_match(buf, "remove")) { |
2626 | if (rdev->raid_disk >= 0) | |
2627 | err = -EBUSY; | |
2628 | else { | |
fd01b88c | 2629 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2630 | kick_rdev_from_array(rdev); |
3f9d7b0d N |
2631 | if (mddev->pers) |
2632 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2633 | md_new_event(mddev); |
2634 | err = 0; | |
2635 | } | |
f655675b N |
2636 | } else if (cmd_match(buf, "writemostly")) { |
2637 | set_bit(WriteMostly, &rdev->flags); | |
2638 | err = 0; | |
2639 | } else if (cmd_match(buf, "-writemostly")) { | |
2640 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2641 | err = 0; |
2642 | } else if (cmd_match(buf, "blocked")) { | |
2643 | set_bit(Blocked, &rdev->flags); | |
2644 | err = 0; | |
2645 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2646 | if (!test_bit(Faulty, &rdev->flags) && |
7da64a0a | 2647 | rdev->badblocks.unacked_exist) { |
de393cde N |
2648 | /* metadata handler doesn't understand badblocks, |
2649 | * so we need to fail the device | |
2650 | */ | |
2651 | md_error(rdev->mddev, rdev); | |
2652 | } | |
6bfe0b49 | 2653 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2654 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2655 | wake_up(&rdev->blocked_wait); |
2656 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2657 | md_wakeup_thread(rdev->mddev->thread); | |
2658 | ||
6d56e278 N |
2659 | err = 0; |
2660 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2661 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2662 | err = 0; |
f466722c N |
2663 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) { |
2664 | clear_bit(In_sync, &rdev->flags); | |
2665 | rdev->saved_raid_disk = rdev->raid_disk; | |
2666 | rdev->raid_disk = -1; | |
2667 | err = 0; | |
d7a9d443 N |
2668 | } else if (cmd_match(buf, "write_error")) { |
2669 | set_bit(WriteErrorSeen, &rdev->flags); | |
2670 | err = 0; | |
2671 | } else if (cmd_match(buf, "-write_error")) { | |
2672 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2673 | err = 0; | |
2d78f8c4 N |
2674 | } else if (cmd_match(buf, "want_replacement")) { |
2675 | /* Any non-spare device that is not a replacement can | |
2676 | * become want_replacement at any time, but we then need to | |
2677 | * check if recovery is needed. | |
2678 | */ | |
2679 | if (rdev->raid_disk >= 0 && | |
2680 | !test_bit(Replacement, &rdev->flags)) | |
2681 | set_bit(WantReplacement, &rdev->flags); | |
2682 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2683 | md_wakeup_thread(rdev->mddev->thread); | |
2684 | err = 0; | |
2685 | } else if (cmd_match(buf, "-want_replacement")) { | |
2686 | /* Clearing 'want_replacement' is always allowed. | |
2687 | * Once replacements starts it is too late though. | |
2688 | */ | |
2689 | err = 0; | |
2690 | clear_bit(WantReplacement, &rdev->flags); | |
2691 | } else if (cmd_match(buf, "replacement")) { | |
2692 | /* Can only set a device as a replacement when array has not | |
2693 | * yet been started. Once running, replacement is automatic | |
2694 | * from spares, or by assigning 'slot'. | |
2695 | */ | |
2696 | if (rdev->mddev->pers) | |
2697 | err = -EBUSY; | |
2698 | else { | |
2699 | set_bit(Replacement, &rdev->flags); | |
2700 | err = 0; | |
2701 | } | |
2702 | } else if (cmd_match(buf, "-replacement")) { | |
2703 | /* Similarly, can only clear Replacement before start */ | |
2704 | if (rdev->mddev->pers) | |
2705 | err = -EBUSY; | |
2706 | else { | |
2707 | clear_bit(Replacement, &rdev->flags); | |
2708 | err = 0; | |
2709 | } | |
45dc2de1 | 2710 | } |
00bcb4ac N |
2711 | if (!err) |
2712 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2713 | return err ? err : len; |
2714 | } | |
80ca3a44 N |
2715 | static struct rdev_sysfs_entry rdev_state = |
2716 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2717 | |
4dbcdc75 | 2718 | static ssize_t |
3cb03002 | 2719 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2720 | { |
2721 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2722 | } | |
2723 | ||
2724 | static ssize_t | |
3cb03002 | 2725 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 N |
2726 | { |
2727 | char *e; | |
2728 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2729 | if (*buf && (*e == 0 || *e == '\n')) { | |
2730 | atomic_set(&rdev->corrected_errors, n); | |
2731 | return len; | |
2732 | } | |
2733 | return -EINVAL; | |
2734 | } | |
2735 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2736 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2737 | |
014236d2 | 2738 | static ssize_t |
3cb03002 | 2739 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 N |
2740 | { |
2741 | if (rdev->raid_disk < 0) | |
2742 | return sprintf(page, "none\n"); | |
2743 | else | |
2744 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2745 | } | |
2746 | ||
2747 | static ssize_t | |
3cb03002 | 2748 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 N |
2749 | { |
2750 | char *e; | |
c303da6d | 2751 | int err; |
014236d2 N |
2752 | int slot = simple_strtoul(buf, &e, 10); |
2753 | if (strncmp(buf, "none", 4)==0) | |
2754 | slot = -1; | |
2755 | else if (e==buf || (*e && *e!= '\n')) | |
2756 | return -EINVAL; | |
6c2fce2e | 2757 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2758 | /* Setting 'slot' on an active array requires also |
2759 | * updating the 'rd%d' link, and communicating | |
2760 | * with the personality with ->hot_*_disk. | |
2761 | * For now we only support removing | |
2762 | * failed/spare devices. This normally happens automatically, | |
2763 | * but not when the metadata is externally managed. | |
2764 | */ | |
c303da6d N |
2765 | if (rdev->raid_disk == -1) |
2766 | return -EEXIST; | |
2767 | /* personality does all needed checks */ | |
01393f3d | 2768 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2769 | return -EINVAL; |
746d3207 N |
2770 | clear_bit(Blocked, &rdev->flags); |
2771 | remove_and_add_spares(rdev->mddev, rdev); | |
2772 | if (rdev->raid_disk >= 0) | |
2773 | return -EBUSY; | |
c303da6d N |
2774 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2775 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2776 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2777 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2778 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2779 | */ |
2780 | ||
2781 | if (rdev->raid_disk != -1) | |
2782 | return -EBUSY; | |
2783 | ||
c6751b2b N |
2784 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2785 | return -EBUSY; | |
2786 | ||
6c2fce2e NB |
2787 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2788 | return -EINVAL; | |
2789 | ||
ba1b41b6 N |
2790 | if (slot >= rdev->mddev->raid_disks && |
2791 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2792 | return -ENOSPC; | |
2793 | ||
6c2fce2e NB |
2794 | rdev->raid_disk = slot; |
2795 | if (test_bit(In_sync, &rdev->flags)) | |
2796 | rdev->saved_raid_disk = slot; | |
2797 | else | |
2798 | rdev->saved_raid_disk = -1; | |
d30519fc | 2799 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2800 | clear_bit(Bitmap_sync, &rdev->flags); |
6c2fce2e NB |
2801 | err = rdev->mddev->pers-> |
2802 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2803 | if (err) { |
6c2fce2e | 2804 | rdev->raid_disk = -1; |
6c2fce2e | 2805 | return err; |
52664732 | 2806 | } else |
00bcb4ac | 2807 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
36fad858 | 2808 | if (sysfs_link_rdev(rdev->mddev, rdev)) |
00bcb4ac | 2809 | /* failure here is OK */; |
6c2fce2e | 2810 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2811 | } else { |
ba1b41b6 N |
2812 | if (slot >= rdev->mddev->raid_disks && |
2813 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2814 | return -ENOSPC; |
2815 | rdev->raid_disk = slot; | |
2816 | /* assume it is working */ | |
c5d79adb N |
2817 | clear_bit(Faulty, &rdev->flags); |
2818 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2819 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2820 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2821 | } |
014236d2 N |
2822 | return len; |
2823 | } | |
2824 | ||
2825 | ||
2826 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2827 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2828 | |
93c8cad0 | 2829 | static ssize_t |
3cb03002 | 2830 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2831 | { |
6961ece4 | 2832 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2833 | } |
2834 | ||
2835 | static ssize_t | |
3cb03002 | 2836 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2837 | { |
c6563a8c | 2838 | unsigned long long offset; |
b29bebd6 | 2839 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2840 | return -EINVAL; |
8ed0a521 | 2841 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2842 | return -EBUSY; |
dd8ac336 | 2843 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2844 | /* Must set offset before size, so overlap checks |
2845 | * can be sane */ | |
2846 | return -EBUSY; | |
93c8cad0 | 2847 | rdev->data_offset = offset; |
25f7fd47 | 2848 | rdev->new_data_offset = offset; |
93c8cad0 N |
2849 | return len; |
2850 | } | |
2851 | ||
2852 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2853 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2854 | |
c6563a8c N |
2855 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2856 | { | |
2857 | return sprintf(page, "%llu\n", | |
2858 | (unsigned long long)rdev->new_data_offset); | |
2859 | } | |
2860 | ||
2861 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2862 | const char *buf, size_t len) | |
2863 | { | |
2864 | unsigned long long new_offset; | |
2865 | struct mddev *mddev = rdev->mddev; | |
2866 | ||
b29bebd6 | 2867 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2868 | return -EINVAL; |
2869 | ||
2870 | if (mddev->sync_thread) | |
2871 | return -EBUSY; | |
2872 | if (new_offset == rdev->data_offset) | |
2873 | /* reset is always permitted */ | |
2874 | ; | |
2875 | else if (new_offset > rdev->data_offset) { | |
2876 | /* must not push array size beyond rdev_sectors */ | |
2877 | if (new_offset - rdev->data_offset | |
2878 | + mddev->dev_sectors > rdev->sectors) | |
2879 | return -E2BIG; | |
2880 | } | |
2881 | /* Metadata worries about other space details. */ | |
2882 | ||
2883 | /* decreasing the offset is inconsistent with a backwards | |
2884 | * reshape. | |
2885 | */ | |
2886 | if (new_offset < rdev->data_offset && | |
2887 | mddev->reshape_backwards) | |
2888 | return -EINVAL; | |
2889 | /* Increasing offset is inconsistent with forwards | |
2890 | * reshape. reshape_direction should be set to | |
2891 | * 'backwards' first. | |
2892 | */ | |
2893 | if (new_offset > rdev->data_offset && | |
2894 | !mddev->reshape_backwards) | |
2895 | return -EINVAL; | |
2896 | ||
2897 | if (mddev->pers && mddev->persistent && | |
2898 | !super_types[mddev->major_version] | |
2899 | .allow_new_offset(rdev, new_offset)) | |
2900 | return -E2BIG; | |
2901 | rdev->new_data_offset = new_offset; | |
2902 | if (new_offset > rdev->data_offset) | |
2903 | mddev->reshape_backwards = 1; | |
2904 | else if (new_offset < rdev->data_offset) | |
2905 | mddev->reshape_backwards = 0; | |
2906 | ||
2907 | return len; | |
2908 | } | |
2909 | static struct rdev_sysfs_entry rdev_new_offset = | |
2910 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2911 | ||
83303b61 | 2912 | static ssize_t |
3cb03002 | 2913 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2914 | { |
dd8ac336 | 2915 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2916 | } |
2917 | ||
c5d79adb N |
2918 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2919 | { | |
2920 | /* check if two start/length pairs overlap */ | |
2921 | if (s1+l1 <= s2) | |
2922 | return 0; | |
2923 | if (s2+l2 <= s1) | |
2924 | return 0; | |
2925 | return 1; | |
2926 | } | |
2927 | ||
b522adcd DW |
2928 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2929 | { | |
2930 | unsigned long long blocks; | |
2931 | sector_t new; | |
2932 | ||
b29bebd6 | 2933 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
2934 | return -EINVAL; |
2935 | ||
2936 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2937 | return -EINVAL; /* sector conversion overflow */ | |
2938 | ||
2939 | new = blocks * 2; | |
2940 | if (new != blocks * 2) | |
2941 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2942 | ||
2943 | *sectors = new; | |
2944 | return 0; | |
2945 | } | |
2946 | ||
83303b61 | 2947 | static ssize_t |
3cb03002 | 2948 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2949 | { |
fd01b88c | 2950 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2951 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2952 | sector_t sectors; |
27c529bb | 2953 | |
b522adcd | 2954 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2955 | return -EINVAL; |
c6563a8c N |
2956 | if (rdev->data_offset != rdev->new_data_offset) |
2957 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2958 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2959 | if (my_mddev->persistent) { |
dd8ac336 AN |
2960 | sectors = super_types[my_mddev->major_version]. |
2961 | rdev_size_change(rdev, sectors); | |
2962 | if (!sectors) | |
0cd17fec | 2963 | return -EBUSY; |
dd8ac336 | 2964 | } else if (!sectors) |
77304d2a | 2965 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2966 | rdev->data_offset; |
a6468539 N |
2967 | if (!my_mddev->pers->resize) |
2968 | /* Cannot change size for RAID0 or Linear etc */ | |
2969 | return -EINVAL; | |
0cd17fec | 2970 | } |
dd8ac336 | 2971 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2972 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2973 | |
dd8ac336 AN |
2974 | rdev->sectors = sectors; |
2975 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2976 | /* need to check that all other rdevs with the same ->bdev |
2977 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2978 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2979 | * we have to change it back, we will have the lock again. |
2980 | */ | |
fd01b88c | 2981 | struct mddev *mddev; |
c5d79adb | 2982 | int overlap = 0; |
159ec1fc | 2983 | struct list_head *tmp; |
c5d79adb | 2984 | |
27c529bb | 2985 | mddev_unlock(my_mddev); |
29ac4aa3 | 2986 | for_each_mddev(mddev, tmp) { |
3cb03002 | 2987 | struct md_rdev *rdev2; |
c5d79adb | 2988 | |
29f097c4 | 2989 | mddev_lock_nointr(mddev); |
dafb20fa | 2990 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
2991 | if (rdev->bdev == rdev2->bdev && |
2992 | rdev != rdev2 && | |
2993 | overlaps(rdev->data_offset, rdev->sectors, | |
2994 | rdev2->data_offset, | |
2995 | rdev2->sectors)) { | |
c5d79adb N |
2996 | overlap = 1; |
2997 | break; | |
2998 | } | |
2999 | mddev_unlock(mddev); | |
3000 | if (overlap) { | |
3001 | mddev_put(mddev); | |
3002 | break; | |
3003 | } | |
3004 | } | |
29f097c4 | 3005 | mddev_lock_nointr(my_mddev); |
c5d79adb N |
3006 | if (overlap) { |
3007 | /* Someone else could have slipped in a size | |
3008 | * change here, but doing so is just silly. | |
dd8ac336 | 3009 | * We put oldsectors back because we *know* it is |
c5d79adb N |
3010 | * safe, and trust userspace not to race with |
3011 | * itself | |
3012 | */ | |
dd8ac336 | 3013 | rdev->sectors = oldsectors; |
c5d79adb N |
3014 | return -EBUSY; |
3015 | } | |
3016 | } | |
83303b61 N |
3017 | return len; |
3018 | } | |
3019 | ||
3020 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 3021 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 3022 | |
06e3c817 | 3023 | |
3cb03002 | 3024 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
3025 | { |
3026 | unsigned long long recovery_start = rdev->recovery_offset; | |
3027 | ||
3028 | if (test_bit(In_sync, &rdev->flags) || | |
3029 | recovery_start == MaxSector) | |
3030 | return sprintf(page, "none\n"); | |
3031 | ||
3032 | return sprintf(page, "%llu\n", recovery_start); | |
3033 | } | |
3034 | ||
3cb03002 | 3035 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
3036 | { |
3037 | unsigned long long recovery_start; | |
3038 | ||
3039 | if (cmd_match(buf, "none")) | |
3040 | recovery_start = MaxSector; | |
b29bebd6 | 3041 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
3042 | return -EINVAL; |
3043 | ||
3044 | if (rdev->mddev->pers && | |
3045 | rdev->raid_disk >= 0) | |
3046 | return -EBUSY; | |
3047 | ||
3048 | rdev->recovery_offset = recovery_start; | |
3049 | if (recovery_start == MaxSector) | |
3050 | set_bit(In_sync, &rdev->flags); | |
3051 | else | |
3052 | clear_bit(In_sync, &rdev->flags); | |
3053 | return len; | |
3054 | } | |
3055 | ||
3056 | static struct rdev_sysfs_entry rdev_recovery_start = | |
3057 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
3058 | ||
16c791a5 N |
3059 | |
3060 | static ssize_t | |
3061 | badblocks_show(struct badblocks *bb, char *page, int unack); | |
3062 | static ssize_t | |
3063 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); | |
3064 | ||
3cb03002 | 3065 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3066 | { |
3067 | return badblocks_show(&rdev->badblocks, page, 0); | |
3068 | } | |
3cb03002 | 3069 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 3070 | { |
de393cde N |
3071 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
3072 | /* Maybe that ack was all we needed */ | |
3073 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
3074 | wake_up(&rdev->blocked_wait); | |
3075 | return rv; | |
16c791a5 N |
3076 | } |
3077 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
3078 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
3079 | ||
3080 | ||
3cb03002 | 3081 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3082 | { |
3083 | return badblocks_show(&rdev->badblocks, page, 1); | |
3084 | } | |
3cb03002 | 3085 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
3086 | { |
3087 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
3088 | } | |
3089 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
3090 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
3091 | ||
86e6ffdd N |
3092 | static struct attribute *rdev_default_attrs[] = { |
3093 | &rdev_state.attr, | |
4dbcdc75 | 3094 | &rdev_errors.attr, |
014236d2 | 3095 | &rdev_slot.attr, |
93c8cad0 | 3096 | &rdev_offset.attr, |
c6563a8c | 3097 | &rdev_new_offset.attr, |
83303b61 | 3098 | &rdev_size.attr, |
06e3c817 | 3099 | &rdev_recovery_start.attr, |
16c791a5 N |
3100 | &rdev_bad_blocks.attr, |
3101 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
3102 | NULL, |
3103 | }; | |
3104 | static ssize_t | |
3105 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3106 | { | |
3107 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3108 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
fd01b88c | 3109 | struct mddev *mddev = rdev->mddev; |
27c529bb | 3110 | ssize_t rv; |
86e6ffdd N |
3111 | |
3112 | if (!entry->show) | |
3113 | return -EIO; | |
27c529bb N |
3114 | |
3115 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
3116 | if (!rv) { | |
3117 | if (rdev->mddev == NULL) | |
3118 | rv = -EBUSY; | |
3119 | else | |
3120 | rv = entry->show(rdev, page); | |
3121 | mddev_unlock(mddev); | |
3122 | } | |
3123 | return rv; | |
86e6ffdd N |
3124 | } |
3125 | ||
3126 | static ssize_t | |
3127 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3128 | const char *page, size_t length) | |
3129 | { | |
3130 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3131 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3132 | ssize_t rv; |
fd01b88c | 3133 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3134 | |
3135 | if (!entry->store) | |
3136 | return -EIO; | |
67463acb N |
3137 | if (!capable(CAP_SYS_ADMIN)) |
3138 | return -EACCES; | |
27c529bb | 3139 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3140 | if (!rv) { |
27c529bb N |
3141 | if (rdev->mddev == NULL) |
3142 | rv = -EBUSY; | |
3143 | else | |
3144 | rv = entry->store(rdev, page, length); | |
6a51830e | 3145 | mddev_unlock(mddev); |
ca388059 N |
3146 | } |
3147 | return rv; | |
86e6ffdd N |
3148 | } |
3149 | ||
3150 | static void rdev_free(struct kobject *ko) | |
3151 | { | |
3cb03002 | 3152 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3153 | kfree(rdev); |
3154 | } | |
52cf25d0 | 3155 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3156 | .show = rdev_attr_show, |
3157 | .store = rdev_attr_store, | |
3158 | }; | |
3159 | static struct kobj_type rdev_ktype = { | |
3160 | .release = rdev_free, | |
3161 | .sysfs_ops = &rdev_sysfs_ops, | |
3162 | .default_attrs = rdev_default_attrs, | |
3163 | }; | |
3164 | ||
3cb03002 | 3165 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3166 | { |
3167 | rdev->desc_nr = -1; | |
3168 | rdev->saved_raid_disk = -1; | |
3169 | rdev->raid_disk = -1; | |
3170 | rdev->flags = 0; | |
3171 | rdev->data_offset = 0; | |
c6563a8c | 3172 | rdev->new_data_offset = 0; |
e8bb9a83 N |
3173 | rdev->sb_events = 0; |
3174 | rdev->last_read_error.tv_sec = 0; | |
3175 | rdev->last_read_error.tv_nsec = 0; | |
2699b672 N |
3176 | rdev->sb_loaded = 0; |
3177 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3178 | atomic_set(&rdev->nr_pending, 0); |
3179 | atomic_set(&rdev->read_errors, 0); | |
3180 | atomic_set(&rdev->corrected_errors, 0); | |
3181 | ||
3182 | INIT_LIST_HEAD(&rdev->same_set); | |
3183 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3184 | |
3185 | /* Add space to store bad block list. | |
3186 | * This reserves the space even on arrays where it cannot | |
3187 | * be used - I wonder if that matters | |
3188 | */ | |
3189 | rdev->badblocks.count = 0; | |
486adf72 | 3190 | rdev->badblocks.shift = -1; /* disabled until explicitly enabled */ |
2230dfe4 N |
3191 | rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3192 | seqlock_init(&rdev->badblocks.lock); | |
3193 | if (rdev->badblocks.page == NULL) | |
3194 | return -ENOMEM; | |
3195 | ||
3196 | return 0; | |
e8bb9a83 N |
3197 | } |
3198 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3199 | /* |
3200 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3201 | * | |
3202 | * mark the device faulty if: | |
3203 | * | |
3204 | * - the device is nonexistent (zero size) | |
3205 | * - the device has no valid superblock | |
3206 | * | |
3207 | * a faulty rdev _never_ has rdev->sb set. | |
3208 | */ | |
3cb03002 | 3209 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3210 | { |
3211 | char b[BDEVNAME_SIZE]; | |
3212 | int err; | |
3cb03002 | 3213 | struct md_rdev *rdev; |
1da177e4 LT |
3214 | sector_t size; |
3215 | ||
9ffae0cf | 3216 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
3217 | if (!rdev) { |
3218 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
3219 | return ERR_PTR(-ENOMEM); | |
3220 | } | |
1da177e4 | 3221 | |
2230dfe4 N |
3222 | err = md_rdev_init(rdev); |
3223 | if (err) | |
3224 | goto abort_free; | |
3225 | err = alloc_disk_sb(rdev); | |
3226 | if (err) | |
1da177e4 LT |
3227 | goto abort_free; |
3228 | ||
c5d79adb | 3229 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3230 | if (err) |
3231 | goto abort_free; | |
3232 | ||
f9cb074b | 3233 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3234 | |
77304d2a | 3235 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 LT |
3236 | if (!size) { |
3237 | printk(KERN_WARNING | |
3238 | "md: %s has zero or unknown size, marking faulty!\n", | |
3239 | bdevname(rdev->bdev,b)); | |
3240 | err = -EINVAL; | |
3241 | goto abort_free; | |
3242 | } | |
3243 | ||
3244 | if (super_format >= 0) { | |
3245 | err = super_types[super_format]. | |
3246 | load_super(rdev, NULL, super_minor); | |
3247 | if (err == -EINVAL) { | |
df968c4e N |
3248 | printk(KERN_WARNING |
3249 | "md: %s does not have a valid v%d.%d " | |
3250 | "superblock, not importing!\n", | |
3251 | bdevname(rdev->bdev,b), | |
3252 | super_format, super_minor); | |
1da177e4 LT |
3253 | goto abort_free; |
3254 | } | |
3255 | if (err < 0) { | |
3256 | printk(KERN_WARNING | |
3257 | "md: could not read %s's sb, not importing!\n", | |
3258 | bdevname(rdev->bdev,b)); | |
3259 | goto abort_free; | |
3260 | } | |
3261 | } | |
6bfe0b49 | 3262 | |
1da177e4 LT |
3263 | return rdev; |
3264 | ||
3265 | abort_free: | |
2699b672 N |
3266 | if (rdev->bdev) |
3267 | unlock_rdev(rdev); | |
545c8795 | 3268 | md_rdev_clear(rdev); |
1da177e4 LT |
3269 | kfree(rdev); |
3270 | return ERR_PTR(err); | |
3271 | } | |
3272 | ||
3273 | /* | |
3274 | * Check a full RAID array for plausibility | |
3275 | */ | |
3276 | ||
3277 | ||
fd01b88c | 3278 | static void analyze_sbs(struct mddev * mddev) |
1da177e4 LT |
3279 | { |
3280 | int i; | |
3cb03002 | 3281 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3282 | char b[BDEVNAME_SIZE]; |
3283 | ||
3284 | freshest = NULL; | |
dafb20fa | 3285 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3286 | switch (super_types[mddev->major_version]. |
3287 | load_super(rdev, freshest, mddev->minor_version)) { | |
3288 | case 1: | |
3289 | freshest = rdev; | |
3290 | break; | |
3291 | case 0: | |
3292 | break; | |
3293 | default: | |
3294 | printk( KERN_ERR \ | |
3295 | "md: fatal superblock inconsistency in %s" | |
3296 | " -- removing from array\n", | |
3297 | bdevname(rdev->bdev,b)); | |
3298 | kick_rdev_from_array(rdev); | |
3299 | } | |
3300 | ||
3301 | ||
3302 | super_types[mddev->major_version]. | |
3303 | validate_super(mddev, freshest); | |
3304 | ||
3305 | i = 0; | |
dafb20fa | 3306 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3307 | if (mddev->max_disks && |
3308 | (rdev->desc_nr >= mddev->max_disks || | |
3309 | i > mddev->max_disks)) { | |
de01dfad N |
3310 | printk(KERN_WARNING |
3311 | "md: %s: %s: only %d devices permitted\n", | |
3312 | mdname(mddev), bdevname(rdev->bdev, b), | |
3313 | mddev->max_disks); | |
3314 | kick_rdev_from_array(rdev); | |
3315 | continue; | |
3316 | } | |
1da177e4 LT |
3317 | if (rdev != freshest) |
3318 | if (super_types[mddev->major_version]. | |
3319 | validate_super(mddev, rdev)) { | |
3320 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
3321 | " from array!\n", | |
3322 | bdevname(rdev->bdev,b)); | |
3323 | kick_rdev_from_array(rdev); | |
3324 | continue; | |
3325 | } | |
3326 | if (mddev->level == LEVEL_MULTIPATH) { | |
3327 | rdev->desc_nr = i++; | |
3328 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3329 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 3330 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
3331 | rdev->raid_disk = -1; |
3332 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3333 | } |
3334 | } | |
1da177e4 LT |
3335 | } |
3336 | ||
72e02075 N |
3337 | /* Read a fixed-point number. |
3338 | * Numbers in sysfs attributes should be in "standard" units where | |
3339 | * possible, so time should be in seconds. | |
3340 | * However we internally use a a much smaller unit such as | |
3341 | * milliseconds or jiffies. | |
3342 | * This function takes a decimal number with a possible fractional | |
3343 | * component, and produces an integer which is the result of | |
3344 | * multiplying that number by 10^'scale'. | |
3345 | * all without any floating-point arithmetic. | |
3346 | */ | |
3347 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3348 | { | |
3349 | unsigned long result = 0; | |
3350 | long decimals = -1; | |
3351 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3352 | if (*cp == '.') | |
3353 | decimals = 0; | |
3354 | else if (decimals < scale) { | |
3355 | unsigned int value; | |
3356 | value = *cp - '0'; | |
3357 | result = result * 10 + value; | |
3358 | if (decimals >= 0) | |
3359 | decimals++; | |
3360 | } | |
3361 | cp++; | |
3362 | } | |
3363 | if (*cp == '\n') | |
3364 | cp++; | |
3365 | if (*cp) | |
3366 | return -EINVAL; | |
3367 | if (decimals < 0) | |
3368 | decimals = 0; | |
3369 | while (decimals < scale) { | |
3370 | result *= 10; | |
3371 | decimals ++; | |
3372 | } | |
3373 | *res = result; | |
3374 | return 0; | |
3375 | } | |
3376 | ||
3377 | ||
19052c0e N |
3378 | static void md_safemode_timeout(unsigned long data); |
3379 | ||
16f17b39 | 3380 | static ssize_t |
fd01b88c | 3381 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3382 | { |
3383 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3384 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3385 | } | |
3386 | static ssize_t | |
fd01b88c | 3387 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3388 | { |
16f17b39 | 3389 | unsigned long msec; |
97ce0a7f | 3390 | |
72e02075 | 3391 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3392 | return -EINVAL; |
16f17b39 N |
3393 | if (msec == 0) |
3394 | mddev->safemode_delay = 0; | |
3395 | else { | |
19052c0e | 3396 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
3397 | mddev->safemode_delay = (msec*HZ)/1000; |
3398 | if (mddev->safemode_delay == 0) | |
3399 | mddev->safemode_delay = 1; | |
275c51c4 | 3400 | if (mddev->safemode_delay < old_delay || old_delay == 0) |
19052c0e | 3401 | md_safemode_timeout((unsigned long)mddev); |
16f17b39 N |
3402 | } |
3403 | return len; | |
3404 | } | |
3405 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3406 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3407 | |
eae1701f | 3408 | static ssize_t |
fd01b88c | 3409 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3410 | { |
84fc4b56 | 3411 | struct md_personality *p = mddev->pers; |
d9d166c2 | 3412 | if (p) |
eae1701f | 3413 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
3414 | else if (mddev->clevel[0]) |
3415 | return sprintf(page, "%s\n", mddev->clevel); | |
3416 | else if (mddev->level != LEVEL_NONE) | |
3417 | return sprintf(page, "%d\n", mddev->level); | |
3418 | else | |
3419 | return 0; | |
eae1701f N |
3420 | } |
3421 | ||
d9d166c2 | 3422 | static ssize_t |
fd01b88c | 3423 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3424 | { |
f2859af6 | 3425 | char clevel[16]; |
20a49ff6 | 3426 | ssize_t rv = len; |
84fc4b56 | 3427 | struct md_personality *pers; |
f2859af6 | 3428 | long level; |
245f46c2 | 3429 | void *priv; |
3cb03002 | 3430 | struct md_rdev *rdev; |
245f46c2 N |
3431 | |
3432 | if (mddev->pers == NULL) { | |
3433 | if (len == 0) | |
3434 | return 0; | |
3435 | if (len >= sizeof(mddev->clevel)) | |
3436 | return -ENOSPC; | |
3437 | strncpy(mddev->clevel, buf, len); | |
3438 | if (mddev->clevel[len-1] == '\n') | |
3439 | len--; | |
3440 | mddev->clevel[len] = 0; | |
3441 | mddev->level = LEVEL_NONE; | |
3442 | return rv; | |
3443 | } | |
bd8839e0 N |
3444 | if (mddev->ro) |
3445 | return -EROFS; | |
245f46c2 N |
3446 | |
3447 | /* request to change the personality. Need to ensure: | |
3448 | * - array is not engaged in resync/recovery/reshape | |
3449 | * - old personality can be suspended | |
3450 | * - new personality will access other array. | |
3451 | */ | |
3452 | ||
bb4f1e9d N |
3453 | if (mddev->sync_thread || |
3454 | mddev->reshape_position != MaxSector || | |
3455 | mddev->sysfs_active) | |
d9d166c2 | 3456 | return -EBUSY; |
245f46c2 N |
3457 | |
3458 | if (!mddev->pers->quiesce) { | |
3459 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3460 | mdname(mddev), mddev->pers->name); | |
3461 | return -EINVAL; | |
3462 | } | |
3463 | ||
3464 | /* Now find the new personality */ | |
f2859af6 | 3465 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3466 | return -EINVAL; |
f2859af6 DW |
3467 | strncpy(clevel, buf, len); |
3468 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3469 | len--; |
f2859af6 | 3470 | clevel[len] = 0; |
b29bebd6 | 3471 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3472 | level = LEVEL_NONE; |
245f46c2 | 3473 | |
f2859af6 DW |
3474 | if (request_module("md-%s", clevel) != 0) |
3475 | request_module("md-level-%s", clevel); | |
245f46c2 | 3476 | spin_lock(&pers_lock); |
f2859af6 | 3477 | pers = find_pers(level, clevel); |
245f46c2 N |
3478 | if (!pers || !try_module_get(pers->owner)) { |
3479 | spin_unlock(&pers_lock); | |
f2859af6 | 3480 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3481 | return -EINVAL; |
3482 | } | |
3483 | spin_unlock(&pers_lock); | |
3484 | ||
3485 | if (pers == mddev->pers) { | |
3486 | /* Nothing to do! */ | |
3487 | module_put(pers->owner); | |
3488 | return rv; | |
3489 | } | |
3490 | if (!pers->takeover) { | |
3491 | module_put(pers->owner); | |
3492 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3493 | mdname(mddev), clevel); |
245f46c2 N |
3494 | return -EINVAL; |
3495 | } | |
3496 | ||
dafb20fa | 3497 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3498 | rdev->new_raid_disk = rdev->raid_disk; |
3499 | ||
245f46c2 N |
3500 | /* ->takeover must set new_* and/or delta_disks |
3501 | * if it succeeds, and may set them when it fails. | |
3502 | */ | |
3503 | priv = pers->takeover(mddev); | |
3504 | if (IS_ERR(priv)) { | |
3505 | mddev->new_level = mddev->level; | |
3506 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3507 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3508 | mddev->raid_disks -= mddev->delta_disks; |
3509 | mddev->delta_disks = 0; | |
2c810cdd | 3510 | mddev->reshape_backwards = 0; |
245f46c2 N |
3511 | module_put(pers->owner); |
3512 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3513 | mdname(mddev), clevel); |
245f46c2 N |
3514 | return PTR_ERR(priv); |
3515 | } | |
3516 | ||
3517 | /* Looks like we have a winner */ | |
3518 | mddev_suspend(mddev); | |
3519 | mddev->pers->stop(mddev); | |
a64c876f N |
3520 | |
3521 | if (mddev->pers->sync_request == NULL && | |
3522 | pers->sync_request != NULL) { | |
3523 | /* need to add the md_redundancy_group */ | |
3524 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3525 | printk(KERN_WARNING | |
3526 | "md: cannot register extra attributes for %s\n", | |
3527 | mdname(mddev)); | |
388975cc | 3528 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
a64c876f N |
3529 | } |
3530 | if (mddev->pers->sync_request != NULL && | |
3531 | pers->sync_request == NULL) { | |
3532 | /* need to remove the md_redundancy_group */ | |
3533 | if (mddev->to_remove == NULL) | |
3534 | mddev->to_remove = &md_redundancy_group; | |
3535 | } | |
3536 | ||
54071b38 TM |
3537 | if (mddev->pers->sync_request == NULL && |
3538 | mddev->external) { | |
3539 | /* We are converting from a no-redundancy array | |
3540 | * to a redundancy array and metadata is managed | |
3541 | * externally so we need to be sure that writes | |
3542 | * won't block due to a need to transition | |
3543 | * clean->dirty | |
3544 | * until external management is started. | |
3545 | */ | |
3546 | mddev->in_sync = 0; | |
3547 | mddev->safemode_delay = 0; | |
3548 | mddev->safemode = 0; | |
3549 | } | |
3550 | ||
dafb20fa | 3551 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3552 | if (rdev->raid_disk < 0) |
3553 | continue; | |
bf2cb0da | 3554 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3555 | rdev->new_raid_disk = -1; |
3556 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3557 | continue; | |
36fad858 | 3558 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3559 | } |
dafb20fa | 3560 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3561 | if (rdev->raid_disk < 0) |
3562 | continue; | |
3563 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3564 | continue; | |
3565 | rdev->raid_disk = rdev->new_raid_disk; | |
3566 | if (rdev->raid_disk < 0) | |
3a981b03 | 3567 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3568 | else { |
36fad858 NK |
3569 | if (sysfs_link_rdev(mddev, rdev)) |
3570 | printk(KERN_WARNING "md: cannot register rd%d" | |
3571 | " for %s after level change\n", | |
3572 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3573 | } |
e93f68a1 N |
3574 | } |
3575 | ||
3576 | module_put(mddev->pers->owner); | |
245f46c2 N |
3577 | mddev->pers = pers; |
3578 | mddev->private = priv; | |
3579 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3580 | mddev->level = mddev->new_level; | |
3581 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3582 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3583 | mddev->delta_disks = 0; |
2c810cdd | 3584 | mddev->reshape_backwards = 0; |
fee68723 | 3585 | mddev->degraded = 0; |
9af204cf TM |
3586 | if (mddev->pers->sync_request == NULL) { |
3587 | /* this is now an array without redundancy, so | |
3588 | * it must always be in_sync | |
3589 | */ | |
3590 | mddev->in_sync = 1; | |
3591 | del_timer_sync(&mddev->safemode_timer); | |
3592 | } | |
02e5f5c0 | 3593 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3594 | pers->run(mddev); |
245f46c2 | 3595 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
47525e59 | 3596 | mddev_resume(mddev); |
830778a1 N |
3597 | if (!mddev->thread) |
3598 | md_update_sb(mddev, 1); | |
5cac7861 | 3599 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3600 | md_new_event(mddev); |
d9d166c2 N |
3601 | return rv; |
3602 | } | |
3603 | ||
3604 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3605 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3606 | |
d4dbd025 N |
3607 | |
3608 | static ssize_t | |
fd01b88c | 3609 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3610 | { |
3611 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3612 | if (mddev->reshape_position != MaxSector && |
3613 | mddev->layout != mddev->new_layout) | |
3614 | return sprintf(page, "%d (%d)\n", | |
3615 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3616 | return sprintf(page, "%d\n", mddev->layout); |
3617 | } | |
3618 | ||
3619 | static ssize_t | |
fd01b88c | 3620 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 N |
3621 | { |
3622 | char *e; | |
3623 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3624 | |
3625 | if (!*buf || (*e && *e != '\n')) | |
3626 | return -EINVAL; | |
3627 | ||
b3546035 N |
3628 | if (mddev->pers) { |
3629 | int err; | |
50ac168a | 3630 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3631 | return -EBUSY; |
bd8839e0 N |
3632 | if (mddev->ro) |
3633 | return -EROFS; | |
597a711b | 3634 | mddev->new_layout = n; |
50ac168a | 3635 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3636 | if (err) { |
3637 | mddev->new_layout = mddev->layout; | |
b3546035 | 3638 | return err; |
597a711b | 3639 | } |
b3546035 | 3640 | } else { |
08a02ecd | 3641 | mddev->new_layout = n; |
b3546035 N |
3642 | if (mddev->reshape_position == MaxSector) |
3643 | mddev->layout = n; | |
3644 | } | |
d4dbd025 N |
3645 | return len; |
3646 | } | |
3647 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3648 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3649 | |
3650 | ||
eae1701f | 3651 | static ssize_t |
fd01b88c | 3652 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3653 | { |
bb636547 N |
3654 | if (mddev->raid_disks == 0) |
3655 | return 0; | |
08a02ecd N |
3656 | if (mddev->reshape_position != MaxSector && |
3657 | mddev->delta_disks != 0) | |
3658 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3659 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3660 | return sprintf(page, "%d\n", mddev->raid_disks); |
3661 | } | |
3662 | ||
fd01b88c | 3663 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3664 | |
3665 | static ssize_t | |
fd01b88c | 3666 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3667 | { |
da943b99 N |
3668 | char *e; |
3669 | int rv = 0; | |
3670 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3671 | ||
3672 | if (!*buf || (*e && *e != '\n')) | |
3673 | return -EINVAL; | |
3674 | ||
3675 | if (mddev->pers) | |
3676 | rv = update_raid_disks(mddev, n); | |
08a02ecd | 3677 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3678 | struct md_rdev *rdev; |
08a02ecd | 3679 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c N |
3680 | |
3681 | rdev_for_each(rdev, mddev) { | |
3682 | if (olddisks < n && | |
3683 | rdev->data_offset < rdev->new_data_offset) | |
3684 | return -EINVAL; | |
3685 | if (olddisks > n && | |
3686 | rdev->data_offset > rdev->new_data_offset) | |
3687 | return -EINVAL; | |
3688 | } | |
08a02ecd N |
3689 | mddev->delta_disks = n - olddisks; |
3690 | mddev->raid_disks = n; | |
2c810cdd | 3691 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3692 | } else |
da943b99 N |
3693 | mddev->raid_disks = n; |
3694 | return rv ? rv : len; | |
3695 | } | |
3696 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3697 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3698 | |
3b34380a | 3699 | static ssize_t |
fd01b88c | 3700 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3701 | { |
08a02ecd | 3702 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3703 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3704 | return sprintf(page, "%d (%d)\n", | |
3705 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3706 | mddev->chunk_sectors << 9); |
3707 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3708 | } |
3709 | ||
3710 | static ssize_t | |
fd01b88c | 3711 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3712 | { |
3b34380a N |
3713 | char *e; |
3714 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3715 | ||
3b34380a N |
3716 | if (!*buf || (*e && *e != '\n')) |
3717 | return -EINVAL; | |
3718 | ||
b3546035 N |
3719 | if (mddev->pers) { |
3720 | int err; | |
50ac168a | 3721 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3722 | return -EBUSY; |
bd8839e0 N |
3723 | if (mddev->ro) |
3724 | return -EROFS; | |
597a711b | 3725 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3726 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3727 | if (err) { |
3728 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3729 | return err; |
597a711b | 3730 | } |
b3546035 | 3731 | } else { |
664e7c41 | 3732 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3733 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3734 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3735 | } |
3b34380a N |
3736 | return len; |
3737 | } | |
3738 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3739 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3740 | |
a94213b1 | 3741 | static ssize_t |
fd01b88c | 3742 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3743 | { |
d1a7c503 N |
3744 | if (mddev->recovery_cp == MaxSector) |
3745 | return sprintf(page, "none\n"); | |
a94213b1 N |
3746 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3747 | } | |
3748 | ||
3749 | static ssize_t | |
fd01b88c | 3750 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3751 | { |
a94213b1 N |
3752 | char *e; |
3753 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3754 | ||
b098636c | 3755 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
a94213b1 | 3756 | return -EBUSY; |
06e3c817 DW |
3757 | if (cmd_match(buf, "none")) |
3758 | n = MaxSector; | |
3759 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3760 | return -EINVAL; |
3761 | ||
3762 | mddev->recovery_cp = n; | |
db07d85e N |
3763 | if (mddev->pers) |
3764 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
a94213b1 N |
3765 | return len; |
3766 | } | |
3767 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3768 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3769 | |
9e653b63 N |
3770 | /* |
3771 | * The array state can be: | |
3772 | * | |
3773 | * clear | |
3774 | * No devices, no size, no level | |
3775 | * Equivalent to STOP_ARRAY ioctl | |
3776 | * inactive | |
3777 | * May have some settings, but array is not active | |
3778 | * all IO results in error | |
3779 | * When written, doesn't tear down array, but just stops it | |
3780 | * suspended (not supported yet) | |
3781 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3782 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3783 | * readonly |
3784 | * no resync can happen. no superblocks get written. | |
3785 | * write requests fail | |
3786 | * read-auto | |
3787 | * like readonly, but behaves like 'clean' on a write request. | |
3788 | * | |
3789 | * clean - no pending writes, but otherwise active. | |
3790 | * When written to inactive array, starts without resync | |
3791 | * If a write request arrives then | |
3792 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3793 | * if not known, block and switch to write-pending | |
3794 | * If written to an active array that has pending writes, then fails. | |
3795 | * active | |
3796 | * fully active: IO and resync can be happening. | |
3797 | * When written to inactive array, starts with resync | |
3798 | * | |
3799 | * write-pending | |
3800 | * clean, but writes are blocked waiting for 'active' to be written. | |
3801 | * | |
3802 | * active-idle | |
3803 | * like active, but no writes have been seen for a while (100msec). | |
3804 | * | |
3805 | */ | |
3806 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3807 | write_pending, active_idle, bad_word}; | |
05381954 | 3808 | static char *array_states[] = { |
9e653b63 N |
3809 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3810 | "write-pending", "active-idle", NULL }; | |
3811 | ||
3812 | static int match_word(const char *word, char **list) | |
3813 | { | |
3814 | int n; | |
3815 | for (n=0; list[n]; n++) | |
3816 | if (cmd_match(word, list[n])) | |
3817 | break; | |
3818 | return n; | |
3819 | } | |
3820 | ||
3821 | static ssize_t | |
fd01b88c | 3822 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3823 | { |
3824 | enum array_state st = inactive; | |
3825 | ||
3826 | if (mddev->pers) | |
3827 | switch(mddev->ro) { | |
3828 | case 1: | |
3829 | st = readonly; | |
3830 | break; | |
3831 | case 2: | |
3832 | st = read_auto; | |
3833 | break; | |
3834 | case 0: | |
3835 | if (mddev->in_sync) | |
3836 | st = clean; | |
070dc6dd | 3837 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3838 | st = write_pending; |
9e653b63 N |
3839 | else if (mddev->safemode) |
3840 | st = active_idle; | |
3841 | else | |
3842 | st = active; | |
3843 | } | |
3844 | else { | |
3845 | if (list_empty(&mddev->disks) && | |
3846 | mddev->raid_disks == 0 && | |
58c0fed4 | 3847 | mddev->dev_sectors == 0) |
9e653b63 N |
3848 | st = clear; |
3849 | else | |
3850 | st = inactive; | |
3851 | } | |
3852 | return sprintf(page, "%s\n", array_states[st]); | |
3853 | } | |
3854 | ||
a05b7ea0 N |
3855 | static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev); |
3856 | static int md_set_readonly(struct mddev * mddev, struct block_device *bdev); | |
fd01b88c N |
3857 | static int do_md_run(struct mddev * mddev); |
3858 | static int restart_array(struct mddev *mddev); | |
9e653b63 N |
3859 | |
3860 | static ssize_t | |
fd01b88c | 3861 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 N |
3862 | { |
3863 | int err = -EINVAL; | |
3864 | enum array_state st = match_word(buf, array_states); | |
3865 | switch(st) { | |
3866 | case bad_word: | |
3867 | break; | |
3868 | case clear: | |
3869 | /* stopping an active array */ | |
a05b7ea0 | 3870 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3871 | break; |
3872 | case inactive: | |
3873 | /* stopping an active array */ | |
90cf195d | 3874 | if (mddev->pers) |
a05b7ea0 | 3875 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3876 | else |
e691063a | 3877 | err = 0; /* already inactive */ |
9e653b63 N |
3878 | break; |
3879 | case suspended: | |
3880 | break; /* not supported yet */ | |
3881 | case readonly: | |
3882 | if (mddev->pers) | |
a05b7ea0 | 3883 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3884 | else { |
3885 | mddev->ro = 1; | |
648b629e | 3886 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3887 | err = do_md_run(mddev); |
3888 | } | |
3889 | break; | |
3890 | case read_auto: | |
9e653b63 | 3891 | if (mddev->pers) { |
80268ee9 | 3892 | if (mddev->ro == 0) |
a05b7ea0 | 3893 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3894 | else if (mddev->ro == 1) |
648b629e N |
3895 | err = restart_array(mddev); |
3896 | if (err == 0) { | |
3897 | mddev->ro = 2; | |
3898 | set_disk_ro(mddev->gendisk, 0); | |
3899 | } | |
9e653b63 N |
3900 | } else { |
3901 | mddev->ro = 2; | |
3902 | err = do_md_run(mddev); | |
3903 | } | |
3904 | break; | |
3905 | case clean: | |
3906 | if (mddev->pers) { | |
3907 | restart_array(mddev); | |
3908 | spin_lock_irq(&mddev->write_lock); | |
3909 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3910 | if (mddev->in_sync == 0) { |
3911 | mddev->in_sync = 1; | |
31a59e34 N |
3912 | if (mddev->safemode == 1) |
3913 | mddev->safemode = 0; | |
070dc6dd | 3914 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3915 | } |
3916 | err = 0; | |
3917 | } else | |
3918 | err = -EBUSY; | |
9e653b63 | 3919 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3920 | } else |
3921 | err = -EINVAL; | |
9e653b63 N |
3922 | break; |
3923 | case active: | |
3924 | if (mddev->pers) { | |
3925 | restart_array(mddev); | |
070dc6dd | 3926 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3927 | wake_up(&mddev->sb_wait); |
3928 | err = 0; | |
3929 | } else { | |
3930 | mddev->ro = 0; | |
648b629e | 3931 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3932 | err = do_md_run(mddev); |
3933 | } | |
3934 | break; | |
3935 | case write_pending: | |
3936 | case active_idle: | |
3937 | /* these cannot be set */ | |
3938 | break; | |
3939 | } | |
3940 | if (err) | |
3941 | return err; | |
0fd62b86 | 3942 | else { |
1d23f178 N |
3943 | if (mddev->hold_active == UNTIL_IOCTL) |
3944 | mddev->hold_active = 0; | |
00bcb4ac | 3945 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3946 | return len; |
0fd62b86 | 3947 | } |
9e653b63 | 3948 | } |
80ca3a44 N |
3949 | static struct md_sysfs_entry md_array_state = |
3950 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3951 | |
1e50915f | 3952 | static ssize_t |
fd01b88c | 3953 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
3954 | return sprintf(page, "%d\n", |
3955 | atomic_read(&mddev->max_corr_read_errors)); | |
3956 | } | |
3957 | ||
3958 | static ssize_t | |
fd01b88c | 3959 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f RB |
3960 | { |
3961 | char *e; | |
3962 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3963 | ||
3964 | if (*buf && (*e == 0 || *e == '\n')) { | |
3965 | atomic_set(&mddev->max_corr_read_errors, n); | |
3966 | return len; | |
3967 | } | |
3968 | return -EINVAL; | |
3969 | } | |
3970 | ||
3971 | static struct md_sysfs_entry max_corr_read_errors = | |
3972 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3973 | max_corrected_read_errors_store); | |
3974 | ||
6d7ff738 | 3975 | static ssize_t |
fd01b88c | 3976 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
3977 | { |
3978 | return -EINVAL; | |
3979 | } | |
3980 | ||
3981 | static ssize_t | |
fd01b88c | 3982 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
3983 | { |
3984 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3985 | /* The new device is added to the array. | |
3986 | * If the array has a persistent superblock, we read the | |
3987 | * superblock to initialise info and check validity. | |
3988 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3989 | * which mainly checks size. | |
3990 | */ | |
3991 | char *e; | |
3992 | int major = simple_strtoul(buf, &e, 10); | |
3993 | int minor; | |
3994 | dev_t dev; | |
3cb03002 | 3995 | struct md_rdev *rdev; |
6d7ff738 N |
3996 | int err; |
3997 | ||
3998 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3999 | return -EINVAL; | |
4000 | minor = simple_strtoul(e+1, &e, 10); | |
4001 | if (*e && *e != '\n') | |
4002 | return -EINVAL; | |
4003 | dev = MKDEV(major, minor); | |
4004 | if (major != MAJOR(dev) || | |
4005 | minor != MINOR(dev)) | |
4006 | return -EOVERFLOW; | |
4007 | ||
4008 | ||
4009 | if (mddev->persistent) { | |
4010 | rdev = md_import_device(dev, mddev->major_version, | |
4011 | mddev->minor_version); | |
4012 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4013 | struct md_rdev *rdev0 |
4014 | = list_entry(mddev->disks.next, | |
4015 | struct md_rdev, same_set); | |
6d7ff738 N |
4016 | err = super_types[mddev->major_version] |
4017 | .load_super(rdev, rdev0, mddev->minor_version); | |
4018 | if (err < 0) | |
4019 | goto out; | |
4020 | } | |
c5d79adb N |
4021 | } else if (mddev->external) |
4022 | rdev = md_import_device(dev, -2, -1); | |
4023 | else | |
6d7ff738 N |
4024 | rdev = md_import_device(dev, -1, -1); |
4025 | ||
4026 | if (IS_ERR(rdev)) | |
4027 | return PTR_ERR(rdev); | |
4028 | err = bind_rdev_to_array(rdev, mddev); | |
4029 | out: | |
4030 | if (err) | |
4031 | export_rdev(rdev); | |
4032 | return err ? err : len; | |
4033 | } | |
4034 | ||
4035 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4036 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4037 | |
9b1d1dac | 4038 | static ssize_t |
fd01b88c | 4039 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4040 | { |
4041 | char *end; | |
4042 | unsigned long chunk, end_chunk; | |
4043 | ||
4044 | if (!mddev->bitmap) | |
4045 | goto out; | |
4046 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4047 | while (*buf) { | |
4048 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4049 | if (buf == end) break; | |
4050 | if (*end == '-') { /* range */ | |
4051 | buf = end + 1; | |
4052 | end_chunk = simple_strtoul(buf, &end, 0); | |
4053 | if (buf == end) break; | |
4054 | } | |
4055 | if (*end && !isspace(*end)) break; | |
4056 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4057 | buf = skip_spaces(end); |
9b1d1dac PC |
4058 | } |
4059 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4060 | out: | |
4061 | return len; | |
4062 | } | |
4063 | ||
4064 | static struct md_sysfs_entry md_bitmap = | |
4065 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4066 | ||
a35b0d69 | 4067 | static ssize_t |
fd01b88c | 4068 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4069 | { |
58c0fed4 AN |
4070 | return sprintf(page, "%llu\n", |
4071 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4072 | } |
4073 | ||
fd01b88c | 4074 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4075 | |
4076 | static ssize_t | |
fd01b88c | 4077 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4078 | { |
4079 | /* If array is inactive, we can reduce the component size, but | |
4080 | * not increase it (except from 0). | |
4081 | * If array is active, we can try an on-line resize | |
4082 | */ | |
b522adcd DW |
4083 | sector_t sectors; |
4084 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4085 | |
58c0fed4 AN |
4086 | if (err < 0) |
4087 | return err; | |
a35b0d69 | 4088 | if (mddev->pers) { |
58c0fed4 | 4089 | err = update_size(mddev, sectors); |
850b2b42 | 4090 | md_update_sb(mddev, 1); |
a35b0d69 | 4091 | } else { |
58c0fed4 AN |
4092 | if (mddev->dev_sectors == 0 || |
4093 | mddev->dev_sectors > sectors) | |
4094 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4095 | else |
4096 | err = -ENOSPC; | |
4097 | } | |
4098 | return err ? err : len; | |
4099 | } | |
4100 | ||
4101 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4102 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4103 | |
8bb93aac | 4104 | |
83f0d77a | 4105 | /* Metadata version. |
e691063a N |
4106 | * This is one of |
4107 | * 'none' for arrays with no metadata (good luck...) | |
4108 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4109 | * or N.M for internally known formats |
4110 | */ | |
4111 | static ssize_t | |
fd01b88c | 4112 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4113 | { |
4114 | if (mddev->persistent) | |
4115 | return sprintf(page, "%d.%d\n", | |
4116 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4117 | else if (mddev->external) |
4118 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4119 | else |
4120 | return sprintf(page, "none\n"); | |
4121 | } | |
4122 | ||
4123 | static ssize_t | |
fd01b88c | 4124 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4125 | { |
4126 | int major, minor; | |
4127 | char *e; | |
ea43ddd8 N |
4128 | /* Changing the details of 'external' metadata is |
4129 | * always permitted. Otherwise there must be | |
4130 | * no devices attached to the array. | |
4131 | */ | |
4132 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
4133 | ; | |
4134 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
4135 | return -EBUSY; |
4136 | ||
4137 | if (cmd_match(buf, "none")) { | |
4138 | mddev->persistent = 0; | |
e691063a N |
4139 | mddev->external = 0; |
4140 | mddev->major_version = 0; | |
4141 | mddev->minor_version = 90; | |
4142 | return len; | |
4143 | } | |
4144 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4145 | size_t namelen = len-9; |
e691063a N |
4146 | if (namelen >= sizeof(mddev->metadata_type)) |
4147 | namelen = sizeof(mddev->metadata_type)-1; | |
4148 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4149 | mddev->metadata_type[namelen] = 0; | |
4150 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4151 | mddev->metadata_type[--namelen] = 0; | |
4152 | mddev->persistent = 0; | |
4153 | mddev->external = 1; | |
8bb93aac N |
4154 | mddev->major_version = 0; |
4155 | mddev->minor_version = 90; | |
4156 | return len; | |
4157 | } | |
4158 | major = simple_strtoul(buf, &e, 10); | |
4159 | if (e==buf || *e != '.') | |
4160 | return -EINVAL; | |
4161 | buf = e+1; | |
4162 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4163 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 4164 | return -EINVAL; |
50511da3 | 4165 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
4166 | return -ENOENT; |
4167 | mddev->major_version = major; | |
4168 | mddev->minor_version = minor; | |
4169 | mddev->persistent = 1; | |
e691063a | 4170 | mddev->external = 0; |
8bb93aac N |
4171 | return len; |
4172 | } | |
4173 | ||
4174 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 4175 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4176 | |
24dd469d | 4177 | static ssize_t |
fd01b88c | 4178 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4179 | { |
7eec314d | 4180 | char *type = "idle"; |
b6a9ce68 N |
4181 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
4182 | type = "frozen"; | |
4183 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 4184 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
4185 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4186 | type = "reshape"; | |
4187 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
4188 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
4189 | type = "resync"; | |
4190 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
4191 | type = "check"; | |
4192 | else | |
4193 | type = "repair"; | |
72a23c21 | 4194 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
4195 | type = "recover"; |
4196 | } | |
4197 | return sprintf(page, "%s\n", type); | |
4198 | } | |
4199 | ||
4200 | static ssize_t | |
fd01b88c | 4201 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4202 | { |
7eec314d N |
4203 | if (!mddev->pers || !mddev->pers->sync_request) |
4204 | return -EINVAL; | |
4205 | ||
b6a9ce68 N |
4206 | if (cmd_match(page, "frozen")) |
4207 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4208 | else | |
4209 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4210 | ||
4211 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
4212 | if (mddev->sync_thread) { |
4213 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
a91d5ac0 | 4214 | md_reap_sync_thread(mddev); |
7eec314d | 4215 | } |
03c902e1 N |
4216 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
4217 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 4218 | return -EBUSY; |
72a23c21 NB |
4219 | else if (cmd_match(page, "resync")) |
4220 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4221 | else if (cmd_match(page, "recover")) { | |
4222 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 4223 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 4224 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4225 | int err; |
4226 | if (mddev->pers->start_reshape == NULL) | |
4227 | return -EINVAL; | |
4228 | err = mddev->pers->start_reshape(mddev); | |
4229 | if (err) | |
4230 | return err; | |
a99ac971 | 4231 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4232 | } else { |
bce74dac | 4233 | if (cmd_match(page, "check")) |
7eec314d | 4234 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4235 | else if (!cmd_match(page, "repair")) |
7eec314d N |
4236 | return -EINVAL; |
4237 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
4238 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4239 | } |
48c26ddc N |
4240 | if (mddev->ro == 2) { |
4241 | /* A write to sync_action is enough to justify | |
4242 | * canceling read-auto mode | |
4243 | */ | |
4244 | mddev->ro = 0; | |
4245 | md_wakeup_thread(mddev->sync_thread); | |
4246 | } | |
03c902e1 | 4247 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4248 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4249 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4250 | return len; |
4251 | } | |
4252 | ||
c4a39551 JB |
4253 | static struct md_sysfs_entry md_scan_mode = |
4254 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
4255 | ||
4256 | static ssize_t | |
4257 | last_sync_action_show(struct mddev *mddev, char *page) | |
4258 | { | |
4259 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4260 | } | |
4261 | ||
4262 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4263 | ||
9d88883e | 4264 | static ssize_t |
fd01b88c | 4265 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4266 | { |
4267 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4268 | (unsigned long long) |
4269 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4270 | } |
4271 | ||
80ca3a44 | 4272 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4273 | |
88202a0c | 4274 | static ssize_t |
fd01b88c | 4275 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4276 | { |
4277 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4278 | mddev->sync_speed_min ? "local": "system"); | |
4279 | } | |
4280 | ||
4281 | static ssize_t | |
fd01b88c | 4282 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4283 | { |
4284 | int min; | |
4285 | char *e; | |
4286 | if (strncmp(buf, "system", 6)==0) { | |
4287 | mddev->sync_speed_min = 0; | |
4288 | return len; | |
4289 | } | |
4290 | min = simple_strtoul(buf, &e, 10); | |
4291 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
4292 | return -EINVAL; | |
4293 | mddev->sync_speed_min = min; | |
4294 | return len; | |
4295 | } | |
4296 | ||
4297 | static struct md_sysfs_entry md_sync_min = | |
4298 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4299 | ||
4300 | static ssize_t | |
fd01b88c | 4301 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4302 | { |
4303 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4304 | mddev->sync_speed_max ? "local": "system"); | |
4305 | } | |
4306 | ||
4307 | static ssize_t | |
fd01b88c | 4308 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4309 | { |
4310 | int max; | |
4311 | char *e; | |
4312 | if (strncmp(buf, "system", 6)==0) { | |
4313 | mddev->sync_speed_max = 0; | |
4314 | return len; | |
4315 | } | |
4316 | max = simple_strtoul(buf, &e, 10); | |
4317 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
4318 | return -EINVAL; | |
4319 | mddev->sync_speed_max = max; | |
4320 | return len; | |
4321 | } | |
4322 | ||
4323 | static struct md_sysfs_entry md_sync_max = | |
4324 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4325 | ||
d7f3d291 | 4326 | static ssize_t |
fd01b88c | 4327 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4328 | { |
4329 | return sprintf(page, "%d\n", mddev->degraded); | |
4330 | } | |
4331 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4332 | |
90b08710 | 4333 | static ssize_t |
fd01b88c | 4334 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4335 | { |
4336 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4337 | } | |
4338 | ||
4339 | static ssize_t | |
fd01b88c | 4340 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4341 | { |
4342 | long n; | |
4343 | ||
b29bebd6 | 4344 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4345 | return -EINVAL; |
4346 | ||
4347 | if (n != 0 && n != 1) | |
4348 | return -EINVAL; | |
4349 | ||
4350 | mddev->parallel_resync = n; | |
4351 | ||
4352 | if (mddev->sync_thread) | |
4353 | wake_up(&resync_wait); | |
4354 | ||
4355 | return len; | |
4356 | } | |
4357 | ||
4358 | /* force parallel resync, even with shared block devices */ | |
4359 | static struct md_sysfs_entry md_sync_force_parallel = | |
4360 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4361 | sync_force_parallel_show, sync_force_parallel_store); | |
4362 | ||
88202a0c | 4363 | static ssize_t |
fd01b88c | 4364 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4365 | { |
4366 | unsigned long resync, dt, db; | |
d1a7c503 N |
4367 | if (mddev->curr_resync == 0) |
4368 | return sprintf(page, "none\n"); | |
9687a60c AN |
4369 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4370 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4371 | if (!dt) dt++; |
9687a60c AN |
4372 | db = resync - mddev->resync_mark_cnt; |
4373 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4374 | } |
4375 | ||
80ca3a44 | 4376 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4377 | |
4378 | static ssize_t | |
fd01b88c | 4379 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4380 | { |
13ae864b | 4381 | unsigned long long max_sectors, resync; |
88202a0c | 4382 | |
acb180b0 N |
4383 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4384 | return sprintf(page, "none\n"); | |
4385 | ||
72f36d59 N |
4386 | if (mddev->curr_resync == 1 || |
4387 | mddev->curr_resync == 2) | |
4388 | return sprintf(page, "delayed\n"); | |
4389 | ||
c804cdec N |
4390 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4391 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4392 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4393 | else |
58c0fed4 | 4394 | max_sectors = mddev->dev_sectors; |
88202a0c | 4395 | |
acb180b0 | 4396 | resync = mddev->curr_resync_completed; |
13ae864b | 4397 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4398 | } |
4399 | ||
80ca3a44 | 4400 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 4401 | |
5e96ee65 | 4402 | static ssize_t |
fd01b88c | 4403 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4404 | { |
4405 | return sprintf(page, "%llu\n", | |
4406 | (unsigned long long)mddev->resync_min); | |
4407 | } | |
4408 | static ssize_t | |
fd01b88c | 4409 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4410 | { |
4411 | unsigned long long min; | |
b29bebd6 | 4412 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 NB |
4413 | return -EINVAL; |
4414 | if (min > mddev->resync_max) | |
4415 | return -EINVAL; | |
4416 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4417 | return -EBUSY; | |
4418 | ||
4419 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4420 | if (mddev->chunk_sectors) { |
2ac06c33 | 4421 | sector_t temp = min; |
9d8f0363 | 4422 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
4423 | return -EINVAL; |
4424 | } | |
4425 | mddev->resync_min = min; | |
4426 | ||
4427 | return len; | |
4428 | } | |
4429 | ||
4430 | static struct md_sysfs_entry md_min_sync = | |
4431 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4432 | ||
c6207277 | 4433 | static ssize_t |
fd01b88c | 4434 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4435 | { |
4436 | if (mddev->resync_max == MaxSector) | |
4437 | return sprintf(page, "max\n"); | |
4438 | else | |
4439 | return sprintf(page, "%llu\n", | |
4440 | (unsigned long long)mddev->resync_max); | |
4441 | } | |
4442 | static ssize_t | |
fd01b88c | 4443 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 N |
4444 | { |
4445 | if (strncmp(buf, "max", 3) == 0) | |
4446 | mddev->resync_max = MaxSector; | |
4447 | else { | |
5e96ee65 | 4448 | unsigned long long max; |
b29bebd6 | 4449 | if (kstrtoull(buf, 10, &max)) |
5e96ee65 NB |
4450 | return -EINVAL; |
4451 | if (max < mddev->resync_min) | |
c6207277 N |
4452 | return -EINVAL; |
4453 | if (max < mddev->resync_max && | |
4d484a4a | 4454 | mddev->ro == 0 && |
c6207277 N |
4455 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4456 | return -EBUSY; | |
4457 | ||
4458 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4459 | if (mddev->chunk_sectors) { |
2ac06c33 | 4460 | sector_t temp = max; |
9d8f0363 | 4461 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
4462 | return -EINVAL; |
4463 | } | |
4464 | mddev->resync_max = max; | |
4465 | } | |
4466 | wake_up(&mddev->recovery_wait); | |
4467 | return len; | |
4468 | } | |
4469 | ||
4470 | static struct md_sysfs_entry md_max_sync = | |
4471 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4472 | ||
e464eafd | 4473 | static ssize_t |
fd01b88c | 4474 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4475 | { |
4476 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4477 | } | |
4478 | ||
4479 | static ssize_t | |
fd01b88c | 4480 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4481 | { |
4482 | char *e; | |
4483 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4484 | unsigned long long old = mddev->suspend_lo; |
e464eafd | 4485 | |
b8d966ef N |
4486 | if (mddev->pers == NULL || |
4487 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4488 | return -EINVAL; |
4489 | if (buf == e || (*e && *e != '\n')) | |
4490 | return -EINVAL; | |
23ddff37 N |
4491 | |
4492 | mddev->suspend_lo = new; | |
4493 | if (new >= old) | |
4494 | /* Shrinking suspended region */ | |
e464eafd | 4495 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4496 | else { |
4497 | /* Expanding suspended region - need to wait */ | |
4498 | mddev->pers->quiesce(mddev, 1); | |
4499 | mddev->pers->quiesce(mddev, 0); | |
4500 | } | |
4501 | return len; | |
e464eafd N |
4502 | } |
4503 | static struct md_sysfs_entry md_suspend_lo = | |
4504 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4505 | ||
4506 | ||
4507 | static ssize_t | |
fd01b88c | 4508 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4509 | { |
4510 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4511 | } | |
4512 | ||
4513 | static ssize_t | |
fd01b88c | 4514 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4515 | { |
4516 | char *e; | |
4517 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4518 | unsigned long long old = mddev->suspend_hi; |
e464eafd | 4519 | |
b8d966ef N |
4520 | if (mddev->pers == NULL || |
4521 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4522 | return -EINVAL; |
4523 | if (buf == e || (*e && *e != '\n')) | |
4524 | return -EINVAL; | |
23ddff37 N |
4525 | |
4526 | mddev->suspend_hi = new; | |
4527 | if (new <= old) | |
4528 | /* Shrinking suspended region */ | |
4529 | mddev->pers->quiesce(mddev, 2); | |
4530 | else { | |
4531 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4532 | mddev->pers->quiesce(mddev, 1); |
4533 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 N |
4534 | } |
4535 | return len; | |
e464eafd N |
4536 | } |
4537 | static struct md_sysfs_entry md_suspend_hi = | |
4538 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4539 | ||
08a02ecd | 4540 | static ssize_t |
fd01b88c | 4541 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4542 | { |
4543 | if (mddev->reshape_position != MaxSector) | |
4544 | return sprintf(page, "%llu\n", | |
4545 | (unsigned long long)mddev->reshape_position); | |
4546 | strcpy(page, "none\n"); | |
4547 | return 5; | |
4548 | } | |
4549 | ||
4550 | static ssize_t | |
fd01b88c | 4551 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4552 | { |
c6563a8c | 4553 | struct md_rdev *rdev; |
08a02ecd N |
4554 | char *e; |
4555 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4556 | if (mddev->pers) | |
4557 | return -EBUSY; | |
4558 | if (buf == e || (*e && *e != '\n')) | |
4559 | return -EINVAL; | |
4560 | mddev->reshape_position = new; | |
4561 | mddev->delta_disks = 0; | |
2c810cdd | 4562 | mddev->reshape_backwards = 0; |
08a02ecd N |
4563 | mddev->new_level = mddev->level; |
4564 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4565 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4566 | rdev_for_each(rdev, mddev) |
4567 | rdev->new_data_offset = rdev->data_offset; | |
08a02ecd N |
4568 | return len; |
4569 | } | |
4570 | ||
4571 | static struct md_sysfs_entry md_reshape_position = | |
4572 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4573 | reshape_position_store); | |
4574 | ||
2c810cdd N |
4575 | static ssize_t |
4576 | reshape_direction_show(struct mddev *mddev, char *page) | |
4577 | { | |
4578 | return sprintf(page, "%s\n", | |
4579 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4580 | } | |
4581 | ||
4582 | static ssize_t | |
4583 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4584 | { | |
4585 | int backwards = 0; | |
4586 | if (cmd_match(buf, "forwards")) | |
4587 | backwards = 0; | |
4588 | else if (cmd_match(buf, "backwards")) | |
4589 | backwards = 1; | |
4590 | else | |
4591 | return -EINVAL; | |
4592 | if (mddev->reshape_backwards == backwards) | |
4593 | return len; | |
4594 | ||
4595 | /* check if we are allowed to change */ | |
4596 | if (mddev->delta_disks) | |
4597 | return -EBUSY; | |
4598 | ||
4599 | if (mddev->persistent && | |
4600 | mddev->major_version == 0) | |
4601 | return -EINVAL; | |
4602 | ||
4603 | mddev->reshape_backwards = backwards; | |
4604 | return len; | |
4605 | } | |
4606 | ||
4607 | static struct md_sysfs_entry md_reshape_direction = | |
4608 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4609 | reshape_direction_store); | |
4610 | ||
b522adcd | 4611 | static ssize_t |
fd01b88c | 4612 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4613 | { |
4614 | if (mddev->external_size) | |
4615 | return sprintf(page, "%llu\n", | |
4616 | (unsigned long long)mddev->array_sectors/2); | |
4617 | else | |
4618 | return sprintf(page, "default\n"); | |
4619 | } | |
4620 | ||
4621 | static ssize_t | |
fd01b88c | 4622 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4623 | { |
4624 | sector_t sectors; | |
4625 | ||
4626 | if (strncmp(buf, "default", 7) == 0) { | |
4627 | if (mddev->pers) | |
4628 | sectors = mddev->pers->size(mddev, 0, 0); | |
4629 | else | |
4630 | sectors = mddev->array_sectors; | |
4631 | ||
4632 | mddev->external_size = 0; | |
4633 | } else { | |
4634 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4635 | return -EINVAL; | |
4636 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4637 | return -E2BIG; |
b522adcd DW |
4638 | |
4639 | mddev->external_size = 1; | |
4640 | } | |
4641 | ||
4642 | mddev->array_sectors = sectors; | |
cbe6ef1d N |
4643 | if (mddev->pers) { |
4644 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 4645 | revalidate_disk(mddev->gendisk); |
cbe6ef1d | 4646 | } |
b522adcd DW |
4647 | return len; |
4648 | } | |
4649 | ||
4650 | static struct md_sysfs_entry md_array_size = | |
4651 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4652 | array_size_store); | |
e464eafd | 4653 | |
eae1701f N |
4654 | static struct attribute *md_default_attrs[] = { |
4655 | &md_level.attr, | |
d4dbd025 | 4656 | &md_layout.attr, |
eae1701f | 4657 | &md_raid_disks.attr, |
3b34380a | 4658 | &md_chunk_size.attr, |
a35b0d69 | 4659 | &md_size.attr, |
a94213b1 | 4660 | &md_resync_start.attr, |
8bb93aac | 4661 | &md_metadata.attr, |
6d7ff738 | 4662 | &md_new_device.attr, |
16f17b39 | 4663 | &md_safe_delay.attr, |
9e653b63 | 4664 | &md_array_state.attr, |
08a02ecd | 4665 | &md_reshape_position.attr, |
2c810cdd | 4666 | &md_reshape_direction.attr, |
b522adcd | 4667 | &md_array_size.attr, |
1e50915f | 4668 | &max_corr_read_errors.attr, |
411036fa N |
4669 | NULL, |
4670 | }; | |
4671 | ||
4672 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4673 | &md_scan_mode.attr, |
c4a39551 | 4674 | &md_last_scan_mode.attr, |
9d88883e | 4675 | &md_mismatches.attr, |
88202a0c N |
4676 | &md_sync_min.attr, |
4677 | &md_sync_max.attr, | |
4678 | &md_sync_speed.attr, | |
90b08710 | 4679 | &md_sync_force_parallel.attr, |
88202a0c | 4680 | &md_sync_completed.attr, |
5e96ee65 | 4681 | &md_min_sync.attr, |
c6207277 | 4682 | &md_max_sync.attr, |
e464eafd N |
4683 | &md_suspend_lo.attr, |
4684 | &md_suspend_hi.attr, | |
9b1d1dac | 4685 | &md_bitmap.attr, |
d7f3d291 | 4686 | &md_degraded.attr, |
eae1701f N |
4687 | NULL, |
4688 | }; | |
411036fa N |
4689 | static struct attribute_group md_redundancy_group = { |
4690 | .name = NULL, | |
4691 | .attrs = md_redundancy_attrs, | |
4692 | }; | |
4693 | ||
eae1701f N |
4694 | |
4695 | static ssize_t | |
4696 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4697 | { | |
4698 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4699 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4700 | ssize_t rv; |
eae1701f N |
4701 | |
4702 | if (!entry->show) | |
4703 | return -EIO; | |
af8a2434 N |
4704 | spin_lock(&all_mddevs_lock); |
4705 | if (list_empty(&mddev->all_mddevs)) { | |
4706 | spin_unlock(&all_mddevs_lock); | |
4707 | return -EBUSY; | |
4708 | } | |
4709 | mddev_get(mddev); | |
4710 | spin_unlock(&all_mddevs_lock); | |
4711 | ||
5dc5cf7d IM |
4712 | rv = mddev_lock(mddev); |
4713 | if (!rv) { | |
4714 | rv = entry->show(mddev, page); | |
4715 | mddev_unlock(mddev); | |
4716 | } | |
af8a2434 | 4717 | mddev_put(mddev); |
96de1e66 | 4718 | return rv; |
eae1701f N |
4719 | } |
4720 | ||
4721 | static ssize_t | |
4722 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4723 | const char *page, size_t length) | |
4724 | { | |
4725 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4726 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4727 | ssize_t rv; |
eae1701f N |
4728 | |
4729 | if (!entry->store) | |
4730 | return -EIO; | |
67463acb N |
4731 | if (!capable(CAP_SYS_ADMIN)) |
4732 | return -EACCES; | |
af8a2434 N |
4733 | spin_lock(&all_mddevs_lock); |
4734 | if (list_empty(&mddev->all_mddevs)) { | |
4735 | spin_unlock(&all_mddevs_lock); | |
4736 | return -EBUSY; | |
4737 | } | |
4738 | mddev_get(mddev); | |
4739 | spin_unlock(&all_mddevs_lock); | |
a7a3f08d N |
4740 | if (entry->store == new_dev_store) |
4741 | flush_workqueue(md_misc_wq); | |
5dc5cf7d IM |
4742 | rv = mddev_lock(mddev); |
4743 | if (!rv) { | |
4744 | rv = entry->store(mddev, page, length); | |
4745 | mddev_unlock(mddev); | |
4746 | } | |
af8a2434 | 4747 | mddev_put(mddev); |
96de1e66 | 4748 | return rv; |
eae1701f N |
4749 | } |
4750 | ||
4751 | static void md_free(struct kobject *ko) | |
4752 | { | |
fd01b88c | 4753 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4754 | |
4755 | if (mddev->sysfs_state) | |
4756 | sysfs_put(mddev->sysfs_state); | |
4757 | ||
4758 | if (mddev->gendisk) { | |
4759 | del_gendisk(mddev->gendisk); | |
4760 | put_disk(mddev->gendisk); | |
4761 | } | |
4762 | if (mddev->queue) | |
4763 | blk_cleanup_queue(mddev->queue); | |
4764 | ||
eae1701f N |
4765 | kfree(mddev); |
4766 | } | |
4767 | ||
52cf25d0 | 4768 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4769 | .show = md_attr_show, |
4770 | .store = md_attr_store, | |
4771 | }; | |
4772 | static struct kobj_type md_ktype = { | |
4773 | .release = md_free, | |
4774 | .sysfs_ops = &md_sysfs_ops, | |
4775 | .default_attrs = md_default_attrs, | |
4776 | }; | |
4777 | ||
1da177e4 LT |
4778 | int mdp_major = 0; |
4779 | ||
5fd3a17e DW |
4780 | static void mddev_delayed_delete(struct work_struct *ws) |
4781 | { | |
fd01b88c | 4782 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4783 | |
43a70507 | 4784 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4785 | kobject_del(&mddev->kobj); |
4786 | kobject_put(&mddev->kobj); | |
4787 | } | |
4788 | ||
efeb53c0 | 4789 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4790 | { |
48c9c27b | 4791 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 4792 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 4793 | struct gendisk *disk; |
efeb53c0 N |
4794 | int partitioned; |
4795 | int shift; | |
4796 | int unit; | |
3830c62f | 4797 | int error; |
1da177e4 LT |
4798 | |
4799 | if (!mddev) | |
efeb53c0 N |
4800 | return -ENODEV; |
4801 | ||
4802 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4803 | shift = partitioned ? MdpMinorShift : 0; | |
4804 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4805 | |
e804ac78 TH |
4806 | /* wait for any previous instance of this device to be |
4807 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4808 | */ |
e804ac78 | 4809 | flush_workqueue(md_misc_wq); |
d3374825 | 4810 | |
48c9c27b | 4811 | mutex_lock(&disks_mutex); |
0909dc44 N |
4812 | error = -EEXIST; |
4813 | if (mddev->gendisk) | |
4814 | goto abort; | |
efeb53c0 N |
4815 | |
4816 | if (name) { | |
4817 | /* Need to ensure that 'name' is not a duplicate. | |
4818 | */ | |
fd01b88c | 4819 | struct mddev *mddev2; |
efeb53c0 N |
4820 | spin_lock(&all_mddevs_lock); |
4821 | ||
4822 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4823 | if (mddev2->gendisk && | |
4824 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4825 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4826 | goto abort; |
efeb53c0 N |
4827 | } |
4828 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4829 | } |
8b765398 | 4830 | |
0909dc44 | 4831 | error = -ENOMEM; |
8b765398 | 4832 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4833 | if (!mddev->queue) |
4834 | goto abort; | |
409c57f3 N |
4835 | mddev->queue->queuedata = mddev; |
4836 | ||
409c57f3 | 4837 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 4838 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 4839 | |
1da177e4 LT |
4840 | disk = alloc_disk(1 << shift); |
4841 | if (!disk) { | |
8b765398 N |
4842 | blk_cleanup_queue(mddev->queue); |
4843 | mddev->queue = NULL; | |
0909dc44 | 4844 | goto abort; |
1da177e4 | 4845 | } |
efeb53c0 | 4846 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4847 | disk->first_minor = unit << shift; |
efeb53c0 N |
4848 | if (name) |
4849 | strcpy(disk->disk_name, name); | |
4850 | else if (partitioned) | |
1da177e4 | 4851 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4852 | else |
1da177e4 | 4853 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4854 | disk->fops = &md_fops; |
4855 | disk->private_data = mddev; | |
4856 | disk->queue = mddev->queue; | |
b0140891 | 4857 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4858 | /* Allow extended partitions. This makes the |
d3374825 | 4859 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4860 | * remove it now. |
4861 | */ | |
4862 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4863 | mddev->gendisk = disk; |
b0140891 N |
4864 | /* As soon as we call add_disk(), another thread could get |
4865 | * through to md_open, so make sure it doesn't get too far | |
4866 | */ | |
4867 | mutex_lock(&mddev->open_mutex); | |
4868 | add_disk(disk); | |
4869 | ||
ed9e1982 TH |
4870 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4871 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4872 | if (error) { |
4873 | /* This isn't possible, but as kobject_init_and_add is marked | |
4874 | * __must_check, we must do something with the result | |
4875 | */ | |
5e55e2f5 N |
4876 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4877 | disk->disk_name); | |
0909dc44 N |
4878 | error = 0; |
4879 | } | |
00bcb4ac N |
4880 | if (mddev->kobj.sd && |
4881 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4882 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4883 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4884 | abort: |
4885 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4886 | if (!error && mddev->kobj.sd) { |
3830c62f | 4887 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4888 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4889 | } |
d3374825 | 4890 | mddev_put(mddev); |
0909dc44 | 4891 | return error; |
efeb53c0 N |
4892 | } |
4893 | ||
4894 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4895 | { | |
4896 | md_alloc(dev, NULL); | |
1da177e4 LT |
4897 | return NULL; |
4898 | } | |
4899 | ||
efeb53c0 N |
4900 | static int add_named_array(const char *val, struct kernel_param *kp) |
4901 | { | |
4902 | /* val must be "md_*" where * is not all digits. | |
4903 | * We allocate an array with a large free minor number, and | |
4904 | * set the name to val. val must not already be an active name. | |
4905 | */ | |
4906 | int len = strlen(val); | |
4907 | char buf[DISK_NAME_LEN]; | |
4908 | ||
4909 | while (len && val[len-1] == '\n') | |
4910 | len--; | |
4911 | if (len >= DISK_NAME_LEN) | |
4912 | return -E2BIG; | |
4913 | strlcpy(buf, val, len+1); | |
4914 | if (strncmp(buf, "md_", 3) != 0) | |
4915 | return -EINVAL; | |
4916 | return md_alloc(0, buf); | |
4917 | } | |
4918 | ||
1da177e4 LT |
4919 | static void md_safemode_timeout(unsigned long data) |
4920 | { | |
fd01b88c | 4921 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 4922 | |
0fd62b86 NB |
4923 | if (!atomic_read(&mddev->writes_pending)) { |
4924 | mddev->safemode = 1; | |
4925 | if (mddev->external) | |
00bcb4ac | 4926 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4927 | } |
1da177e4 LT |
4928 | md_wakeup_thread(mddev->thread); |
4929 | } | |
4930 | ||
6ff8d8ec | 4931 | static int start_dirty_degraded; |
1da177e4 | 4932 | |
fd01b88c | 4933 | int md_run(struct mddev *mddev) |
1da177e4 | 4934 | { |
2604b703 | 4935 | int err; |
3cb03002 | 4936 | struct md_rdev *rdev; |
84fc4b56 | 4937 | struct md_personality *pers; |
1da177e4 | 4938 | |
a757e64c N |
4939 | if (list_empty(&mddev->disks)) |
4940 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4941 | return -EINVAL; |
1da177e4 LT |
4942 | |
4943 | if (mddev->pers) | |
4944 | return -EBUSY; | |
bb4f1e9d N |
4945 | /* Cannot run until previous stop completes properly */ |
4946 | if (mddev->sysfs_active) | |
4947 | return -EBUSY; | |
b6eb127d | 4948 | |
1da177e4 LT |
4949 | /* |
4950 | * Analyze all RAID superblock(s) | |
4951 | */ | |
1ec4a939 N |
4952 | if (!mddev->raid_disks) { |
4953 | if (!mddev->persistent) | |
4954 | return -EINVAL; | |
a757e64c | 4955 | analyze_sbs(mddev); |
1ec4a939 | 4956 | } |
1da177e4 | 4957 | |
d9d166c2 N |
4958 | if (mddev->level != LEVEL_NONE) |
4959 | request_module("md-level-%d", mddev->level); | |
4960 | else if (mddev->clevel[0]) | |
4961 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4962 | |
4963 | /* | |
4964 | * Drop all container device buffers, from now on | |
4965 | * the only valid external interface is through the md | |
4966 | * device. | |
1da177e4 | 4967 | */ |
dafb20fa | 4968 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 4969 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4970 | continue; |
4971 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4972 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4973 | |
4974 | /* perform some consistency tests on the device. | |
4975 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4976 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4977 | */ |
a6ff7e08 JB |
4978 | if (rdev->meta_bdev) { |
4979 | /* Nothing to check */; | |
4980 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
4981 | if (mddev->dev_sectors && |
4982 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4983 | > rdev->sb_start) { |
f0d76d70 N |
4984 | printk("md: %s: data overlaps metadata\n", |
4985 | mdname(mddev)); | |
4986 | return -EINVAL; | |
4987 | } | |
4988 | } else { | |
0f420358 | 4989 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4990 | > rdev->data_offset) { |
4991 | printk("md: %s: metadata overlaps data\n", | |
4992 | mdname(mddev)); | |
4993 | return -EINVAL; | |
4994 | } | |
4995 | } | |
00bcb4ac | 4996 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4997 | } |
4998 | ||
a167f663 | 4999 | if (mddev->bio_set == NULL) |
395c72a7 | 5000 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
a167f663 | 5001 | |
1da177e4 | 5002 | spin_lock(&pers_lock); |
d9d166c2 | 5003 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5004 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5005 | spin_unlock(&pers_lock); |
d9d166c2 N |
5006 | if (mddev->level != LEVEL_NONE) |
5007 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
5008 | mddev->level); | |
5009 | else | |
5010 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
5011 | mddev->clevel); | |
1da177e4 LT |
5012 | return -EINVAL; |
5013 | } | |
2604b703 | 5014 | mddev->pers = pers; |
1da177e4 | 5015 | spin_unlock(&pers_lock); |
34817e8c N |
5016 | if (mddev->level != pers->level) { |
5017 | mddev->level = pers->level; | |
5018 | mddev->new_level = pers->level; | |
5019 | } | |
d9d166c2 | 5020 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5021 | |
f6705578 | 5022 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5023 | pers->start_reshape == NULL) { |
f6705578 N |
5024 | /* This personality cannot handle reshaping... */ |
5025 | mddev->pers = NULL; | |
5026 | module_put(pers->owner); | |
5027 | return -EINVAL; | |
5028 | } | |
5029 | ||
7dd5e7c3 N |
5030 | if (pers->sync_request) { |
5031 | /* Warn if this is a potentially silly | |
5032 | * configuration. | |
5033 | */ | |
5034 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5035 | struct md_rdev *rdev2; |
7dd5e7c3 | 5036 | int warned = 0; |
159ec1fc | 5037 | |
dafb20fa N |
5038 | rdev_for_each(rdev, mddev) |
5039 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5040 | if (rdev < rdev2 && |
5041 | rdev->bdev->bd_contains == | |
5042 | rdev2->bdev->bd_contains) { | |
5043 | printk(KERN_WARNING | |
5044 | "%s: WARNING: %s appears to be" | |
5045 | " on the same physical disk as" | |
5046 | " %s.\n", | |
5047 | mdname(mddev), | |
5048 | bdevname(rdev->bdev,b), | |
5049 | bdevname(rdev2->bdev,b2)); | |
5050 | warned = 1; | |
5051 | } | |
5052 | } | |
159ec1fc | 5053 | |
7dd5e7c3 N |
5054 | if (warned) |
5055 | printk(KERN_WARNING | |
5056 | "True protection against single-disk" | |
5057 | " failure might be compromised.\n"); | |
5058 | } | |
5059 | ||
657390d2 | 5060 | mddev->recovery = 0; |
58c0fed4 AN |
5061 | /* may be over-ridden by personality */ |
5062 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5063 | ||
6ff8d8ec | 5064 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5065 | |
0f9552b5 | 5066 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5067 | mddev->ro = 2; /* read-only, but switch on first write */ |
5068 | ||
b15c2e57 | 5069 | err = mddev->pers->run(mddev); |
13e53df3 AN |
5070 | if (err) |
5071 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
5072 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
5073 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
5074 | " but 'external_size' not in effect?\n", __func__); | |
5075 | printk(KERN_ERR | |
5076 | "md: invalid array_size %llu > default size %llu\n", | |
5077 | (unsigned long long)mddev->array_sectors / 2, | |
5078 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
5079 | err = -EINVAL; | |
5080 | mddev->pers->stop(mddev); | |
5081 | } | |
ef99bf48 N |
5082 | if (err == 0 && mddev->pers->sync_request && |
5083 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { | |
b15c2e57 N |
5084 | err = bitmap_create(mddev); |
5085 | if (err) { | |
5086 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
5087 | mdname(mddev), err); | |
5088 | mddev->pers->stop(mddev); | |
5089 | } | |
5090 | } | |
1da177e4 | 5091 | if (err) { |
1da177e4 LT |
5092 | module_put(mddev->pers->owner); |
5093 | mddev->pers = NULL; | |
32a7627c N |
5094 | bitmap_destroy(mddev); |
5095 | return err; | |
1da177e4 | 5096 | } |
5e55e2f5 | 5097 | if (mddev->pers->sync_request) { |
00bcb4ac N |
5098 | if (mddev->kobj.sd && |
5099 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
5100 | printk(KERN_WARNING |
5101 | "md: cannot register extra attributes for %s\n", | |
5102 | mdname(mddev)); | |
00bcb4ac | 5103 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5104 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5105 | mddev->ro = 0; |
5106 | ||
1da177e4 | 5107 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
5108 | atomic_set(&mddev->max_corr_read_errors, |
5109 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
5110 | mddev->safemode = 0; |
5111 | mddev->safemode_timer.function = md_safemode_timeout; | |
5112 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 5113 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 5114 | mddev->in_sync = 1; |
0ca69886 N |
5115 | smp_wmb(); |
5116 | mddev->ready = 1; | |
dafb20fa | 5117 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5118 | if (rdev->raid_disk >= 0) |
5119 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5120 | /* failure here is OK */; |
1da177e4 LT |
5121 | |
5122 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5123 | ||
7a0a5355 | 5124 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 5125 | md_update_sb(mddev, 0); |
1da177e4 | 5126 | |
d7603b7e | 5127 | md_new_event(mddev); |
00bcb4ac N |
5128 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5129 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5130 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5131 | return 0; |
5132 | } | |
390ee602 | 5133 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5134 | |
fd01b88c | 5135 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5136 | { |
5137 | int err; | |
5138 | ||
5139 | err = md_run(mddev); | |
5140 | if (err) | |
5141 | goto out; | |
69e51b44 N |
5142 | err = bitmap_load(mddev); |
5143 | if (err) { | |
5144 | bitmap_destroy(mddev); | |
5145 | goto out; | |
5146 | } | |
0fd018af JB |
5147 | |
5148 | md_wakeup_thread(mddev->thread); | |
5149 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5150 | ||
fe60b014 N |
5151 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5152 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5153 | mddev->changed = 1; |
fe60b014 N |
5154 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5155 | out: | |
5156 | return err; | |
5157 | } | |
5158 | ||
fd01b88c | 5159 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5160 | { |
5161 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 5162 | |
80fab1d7 | 5163 | /* Complain if it has no devices */ |
1da177e4 | 5164 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5165 | return -ENXIO; |
5166 | if (!mddev->pers) | |
5167 | return -EINVAL; | |
5168 | if (!mddev->ro) | |
5169 | return -EBUSY; | |
5170 | mddev->safemode = 0; | |
5171 | mddev->ro = 0; | |
5172 | set_disk_ro(disk, 0); | |
5173 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
5174 | mdname(mddev)); | |
5175 | /* Kick recovery or resync if necessary */ | |
5176 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5177 | md_wakeup_thread(mddev->thread); | |
5178 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5179 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5180 | return 0; |
1da177e4 LT |
5181 | } |
5182 | ||
fd01b88c | 5183 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5184 | { |
5185 | mddev->array_sectors = 0; | |
5186 | mddev->external_size = 0; | |
5187 | mddev->dev_sectors = 0; | |
5188 | mddev->raid_disks = 0; | |
5189 | mddev->recovery_cp = 0; | |
5190 | mddev->resync_min = 0; | |
5191 | mddev->resync_max = MaxSector; | |
5192 | mddev->reshape_position = MaxSector; | |
5193 | mddev->external = 0; | |
5194 | mddev->persistent = 0; | |
5195 | mddev->level = LEVEL_NONE; | |
5196 | mddev->clevel[0] = 0; | |
5197 | mddev->flags = 0; | |
5198 | mddev->ro = 0; | |
5199 | mddev->metadata_type[0] = 0; | |
5200 | mddev->chunk_sectors = 0; | |
5201 | mddev->ctime = mddev->utime = 0; | |
5202 | mddev->layout = 0; | |
5203 | mddev->max_disks = 0; | |
5204 | mddev->events = 0; | |
a8707c08 | 5205 | mddev->can_decrease_events = 0; |
6177b472 | 5206 | mddev->delta_disks = 0; |
2c810cdd | 5207 | mddev->reshape_backwards = 0; |
6177b472 N |
5208 | mddev->new_level = LEVEL_NONE; |
5209 | mddev->new_layout = 0; | |
5210 | mddev->new_chunk_sectors = 0; | |
5211 | mddev->curr_resync = 0; | |
7f7583d4 | 5212 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5213 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5214 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5215 | mddev->recovery = 0; | |
5216 | mddev->in_sync = 0; | |
f0b4f7e2 | 5217 | mddev->changed = 0; |
6177b472 | 5218 | mddev->degraded = 0; |
6177b472 | 5219 | mddev->safemode = 0; |
050b6615 | 5220 | mddev->merge_check_needed = 0; |
6177b472 N |
5221 | mddev->bitmap_info.offset = 0; |
5222 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5223 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5224 | mddev->bitmap_info.chunksize = 0; |
5225 | mddev->bitmap_info.daemon_sleep = 0; | |
5226 | mddev->bitmap_info.max_write_behind = 0; | |
5227 | } | |
5228 | ||
fd01b88c | 5229 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5230 | { |
6b6204ee | 5231 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
a047e125 | 5232 | if (mddev->sync_thread) { |
a047e125 | 5233 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5234 | md_reap_sync_thread(mddev); |
a047e125 N |
5235 | } |
5236 | ||
5237 | del_timer_sync(&mddev->safemode_timer); | |
5238 | ||
5239 | bitmap_flush(mddev); | |
5240 | md_super_wait(mddev); | |
5241 | ||
b6d428c6 | 5242 | if (mddev->ro == 0 && |
7a0a5355 | 5243 | (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) { |
a047e125 N |
5244 | /* mark array as shutdown cleanly */ |
5245 | mddev->in_sync = 1; | |
5246 | md_update_sb(mddev, 1); | |
5247 | } | |
5248 | } | |
defad61a | 5249 | |
fd01b88c | 5250 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5251 | { |
29f097c4 | 5252 | mddev_lock_nointr(mddev); |
defad61a N |
5253 | __md_stop_writes(mddev); |
5254 | mddev_unlock(mddev); | |
5255 | } | |
390ee602 | 5256 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5257 | |
5eff3c43 | 5258 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5259 | { |
0ca69886 | 5260 | mddev->ready = 0; |
6177b472 N |
5261 | mddev->pers->stop(mddev); |
5262 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
5263 | mddev->to_remove = &md_redundancy_group; | |
5264 | module_put(mddev->pers->owner); | |
5265 | mddev->pers = NULL; | |
cca9cf90 | 5266 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5267 | } |
5eff3c43 N |
5268 | |
5269 | void md_stop(struct mddev *mddev) | |
5270 | { | |
5271 | /* stop the array and free an attached data structures. | |
5272 | * This is called from dm-raid | |
5273 | */ | |
5274 | __md_stop(mddev); | |
5275 | bitmap_destroy(mddev); | |
5276 | if (mddev->bio_set) | |
5277 | bioset_free(mddev->bio_set); | |
5278 | } | |
5279 | ||
390ee602 | 5280 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5281 | |
a05b7ea0 | 5282 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5283 | { |
5284 | int err = 0; | |
30b8feb7 N |
5285 | int did_freeze = 0; |
5286 | ||
5287 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5288 | did_freeze = 1; | |
5289 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5290 | md_wakeup_thread(mddev->thread); | |
5291 | } | |
5292 | if (mddev->sync_thread) { | |
5293 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5294 | /* Thread might be blocked waiting for metadata update | |
5295 | * which will now never happen */ | |
5296 | wake_up_process(mddev->sync_thread->tsk); | |
5297 | } | |
5298 | mddev_unlock(mddev); | |
5299 | wait_event(resync_wait, mddev->sync_thread == NULL); | |
5300 | mddev_lock_nointr(mddev); | |
5301 | ||
a4bd82d0 | 5302 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5303 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5304 | mddev->sync_thread || |
5305 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { | |
a4bd82d0 | 5306 | printk("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5307 | if (did_freeze) { |
5308 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5309 | md_wakeup_thread(mddev->thread); | |
5310 | } | |
a4bd82d0 N |
5311 | err = -EBUSY; |
5312 | goto out; | |
5313 | } | |
5314 | if (mddev->pers) { | |
defad61a | 5315 | __md_stop_writes(mddev); |
a4bd82d0 N |
5316 | |
5317 | err = -ENXIO; | |
5318 | if (mddev->ro==1) | |
5319 | goto out; | |
5320 | mddev->ro = 1; | |
5321 | set_disk_ro(mddev->gendisk, 1); | |
5322 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 5323 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5324 | err = 0; |
a4bd82d0 N |
5325 | } |
5326 | out: | |
5327 | mutex_unlock(&mddev->open_mutex); | |
5328 | return err; | |
5329 | } | |
5330 | ||
9e653b63 N |
5331 | /* mode: |
5332 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5333 | * 2 - stop but do not disassemble array |
5334 | */ | |
a05b7ea0 N |
5335 | static int do_md_stop(struct mddev * mddev, int mode, |
5336 | struct block_device *bdev) | |
1da177e4 | 5337 | { |
1da177e4 | 5338 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5339 | struct md_rdev *rdev; |
30b8feb7 N |
5340 | int did_freeze = 0; |
5341 | ||
5342 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5343 | did_freeze = 1; | |
5344 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5345 | md_wakeup_thread(mddev->thread); | |
5346 | } | |
5347 | if (mddev->sync_thread) { | |
5348 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5349 | /* Thread might be blocked waiting for metadata update | |
5350 | * which will now never happen */ | |
5351 | wake_up_process(mddev->sync_thread->tsk); | |
5352 | } | |
5353 | mddev_unlock(mddev); | |
5354 | wait_event(resync_wait, mddev->sync_thread == NULL); | |
5355 | mddev_lock_nointr(mddev); | |
1da177e4 | 5356 | |
c8c00a69 | 5357 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5358 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5359 | mddev->sysfs_active || |
5360 | mddev->sync_thread || | |
5361 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { | |
df5b20cf | 5362 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5363 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5364 | if (did_freeze) { |
5365 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5366 | md_wakeup_thread(mddev->thread); | |
5367 | } | |
260fa034 N |
5368 | return -EBUSY; |
5369 | } | |
6e17b027 | 5370 | if (mddev->pers) { |
a4bd82d0 N |
5371 | if (mddev->ro) |
5372 | set_disk_ro(disk, 0); | |
409c57f3 | 5373 | |
defad61a | 5374 | __md_stop_writes(mddev); |
5eff3c43 | 5375 | __md_stop(mddev); |
a4bd82d0 | 5376 | mddev->queue->merge_bvec_fn = NULL; |
a4bd82d0 | 5377 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 5378 | |
a4bd82d0 | 5379 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5380 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5381 | |
dafb20fa | 5382 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5383 | if (rdev->raid_disk >= 0) |
5384 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5385 | |
a4bd82d0 | 5386 | set_capacity(disk, 0); |
6e17b027 | 5387 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5388 | mddev->changed = 1; |
a4bd82d0 | 5389 | revalidate_disk(disk); |
0d4ca600 | 5390 | |
a4bd82d0 N |
5391 | if (mddev->ro) |
5392 | mddev->ro = 0; | |
6e17b027 N |
5393 | } else |
5394 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5395 | /* |
5396 | * Free resources if final stop | |
5397 | */ | |
9e653b63 | 5398 | if (mode == 0) { |
1da177e4 LT |
5399 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
5400 | ||
978f946b | 5401 | bitmap_destroy(mddev); |
c3d9714e | 5402 | if (mddev->bitmap_info.file) { |
c3d9714e N |
5403 | fput(mddev->bitmap_info.file); |
5404 | mddev->bitmap_info.file = NULL; | |
978f946b | 5405 | } |
c3d9714e | 5406 | mddev->bitmap_info.offset = 0; |
978f946b | 5407 | |
1da177e4 LT |
5408 | export_array(mddev); |
5409 | ||
6177b472 | 5410 | md_clean(mddev); |
934d9c23 | 5411 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
5412 | if (mddev->hold_active == UNTIL_STOP) |
5413 | mddev->hold_active = 0; | |
a4bd82d0 | 5414 | } |
3f9d99c1 | 5415 | blk_integrity_unregister(disk); |
d7603b7e | 5416 | md_new_event(mddev); |
00bcb4ac | 5417 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5418 | return 0; |
1da177e4 LT |
5419 | } |
5420 | ||
fdee8ae4 | 5421 | #ifndef MODULE |
fd01b88c | 5422 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5423 | { |
3cb03002 | 5424 | struct md_rdev *rdev; |
1da177e4 LT |
5425 | int err; |
5426 | ||
a757e64c | 5427 | if (list_empty(&mddev->disks)) |
1da177e4 | 5428 | return; |
1da177e4 LT |
5429 | |
5430 | printk(KERN_INFO "md: running: "); | |
5431 | ||
dafb20fa | 5432 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
5433 | char b[BDEVNAME_SIZE]; |
5434 | printk("<%s>", bdevname(rdev->bdev,b)); | |
5435 | } | |
5436 | printk("\n"); | |
5437 | ||
d710e138 | 5438 | err = do_md_run(mddev); |
1da177e4 LT |
5439 | if (err) { |
5440 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
a05b7ea0 | 5441 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5442 | } |
5443 | } | |
5444 | ||
5445 | /* | |
5446 | * lets try to run arrays based on all disks that have arrived | |
5447 | * until now. (those are in pending_raid_disks) | |
5448 | * | |
5449 | * the method: pick the first pending disk, collect all disks with | |
5450 | * the same UUID, remove all from the pending list and put them into | |
5451 | * the 'same_array' list. Then order this list based on superblock | |
5452 | * update time (freshest comes first), kick out 'old' disks and | |
5453 | * compare superblocks. If everything's fine then run it. | |
5454 | * | |
5455 | * If "unit" is allocated, then bump its reference count | |
5456 | */ | |
5457 | static void autorun_devices(int part) | |
5458 | { | |
3cb03002 | 5459 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5460 | struct mddev *mddev; |
1da177e4 LT |
5461 | char b[BDEVNAME_SIZE]; |
5462 | ||
5463 | printk(KERN_INFO "md: autorun ...\n"); | |
5464 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 5465 | int unit; |
1da177e4 | 5466 | dev_t dev; |
ad01c9e3 | 5467 | LIST_HEAD(candidates); |
1da177e4 | 5468 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5469 | struct md_rdev, same_set); |
1da177e4 LT |
5470 | |
5471 | printk(KERN_INFO "md: considering %s ...\n", | |
5472 | bdevname(rdev0->bdev,b)); | |
5473 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 5474 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
5475 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
5476 | printk(KERN_INFO "md: adding %s ...\n", | |
5477 | bdevname(rdev->bdev,b)); | |
5478 | list_move(&rdev->same_set, &candidates); | |
5479 | } | |
5480 | /* | |
5481 | * now we have a set of devices, with all of them having | |
5482 | * mostly sane superblocks. It's time to allocate the | |
5483 | * mddev. | |
5484 | */ | |
e8703fe1 N |
5485 | if (part) { |
5486 | dev = MKDEV(mdp_major, | |
5487 | rdev0->preferred_minor << MdpMinorShift); | |
5488 | unit = MINOR(dev) >> MdpMinorShift; | |
5489 | } else { | |
5490 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5491 | unit = MINOR(dev); | |
5492 | } | |
5493 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
5494 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
5495 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
5496 | break; | |
5497 | } | |
1da177e4 LT |
5498 | |
5499 | md_probe(dev, NULL, NULL); | |
5500 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5501 | if (!mddev || !mddev->gendisk) { |
5502 | if (mddev) | |
5503 | mddev_put(mddev); | |
5504 | printk(KERN_ERR | |
1da177e4 LT |
5505 | "md: cannot allocate memory for md drive.\n"); |
5506 | break; | |
5507 | } | |
5508 | if (mddev_lock(mddev)) | |
5509 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
5510 | mdname(mddev)); | |
5511 | else if (mddev->raid_disks || mddev->major_version | |
5512 | || !list_empty(&mddev->disks)) { | |
5513 | printk(KERN_WARNING | |
5514 | "md: %s already running, cannot run %s\n", | |
5515 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
5516 | mddev_unlock(mddev); | |
5517 | } else { | |
5518 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 5519 | mddev->persistent = 1; |
159ec1fc | 5520 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5521 | list_del_init(&rdev->same_set); |
5522 | if (bind_rdev_to_array(rdev, mddev)) | |
5523 | export_rdev(rdev); | |
5524 | } | |
5525 | autorun_array(mddev); | |
5526 | mddev_unlock(mddev); | |
5527 | } | |
5528 | /* on success, candidates will be empty, on error | |
5529 | * it won't... | |
5530 | */ | |
159ec1fc | 5531 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5532 | list_del_init(&rdev->same_set); |
1da177e4 | 5533 | export_rdev(rdev); |
4b80991c | 5534 | } |
1da177e4 LT |
5535 | mddev_put(mddev); |
5536 | } | |
5537 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5538 | } | |
fdee8ae4 | 5539 | #endif /* !MODULE */ |
1da177e4 | 5540 | |
1da177e4 LT |
5541 | static int get_version(void __user * arg) |
5542 | { | |
5543 | mdu_version_t ver; | |
5544 | ||
5545 | ver.major = MD_MAJOR_VERSION; | |
5546 | ver.minor = MD_MINOR_VERSION; | |
5547 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5548 | ||
5549 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5550 | return -EFAULT; | |
5551 | ||
5552 | return 0; | |
5553 | } | |
5554 | ||
fd01b88c | 5555 | static int get_array_info(struct mddev * mddev, void __user * arg) |
1da177e4 LT |
5556 | { |
5557 | mdu_array_info_t info; | |
a9f326eb | 5558 | int nr,working,insync,failed,spare; |
3cb03002 | 5559 | struct md_rdev *rdev; |
1da177e4 | 5560 | |
1ca69c4b N |
5561 | nr = working = insync = failed = spare = 0; |
5562 | rcu_read_lock(); | |
5563 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5564 | nr++; |
b2d444d7 | 5565 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5566 | failed++; |
5567 | else { | |
5568 | working++; | |
b2d444d7 | 5569 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 5570 | insync++; |
1da177e4 LT |
5571 | else |
5572 | spare++; | |
5573 | } | |
5574 | } | |
1ca69c4b | 5575 | rcu_read_unlock(); |
1da177e4 LT |
5576 | |
5577 | info.major_version = mddev->major_version; | |
5578 | info.minor_version = mddev->minor_version; | |
5579 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5580 | info.ctime = mddev->ctime; | |
5581 | info.level = mddev->level; | |
58c0fed4 AN |
5582 | info.size = mddev->dev_sectors / 2; |
5583 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5584 | info.size = -1; |
1da177e4 LT |
5585 | info.nr_disks = nr; |
5586 | info.raid_disks = mddev->raid_disks; | |
5587 | info.md_minor = mddev->md_minor; | |
5588 | info.not_persistent= !mddev->persistent; | |
5589 | ||
5590 | info.utime = mddev->utime; | |
5591 | info.state = 0; | |
5592 | if (mddev->in_sync) | |
5593 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5594 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 5595 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5596 | info.active_disks = insync; |
1da177e4 LT |
5597 | info.working_disks = working; |
5598 | info.failed_disks = failed; | |
5599 | info.spare_disks = spare; | |
5600 | ||
5601 | info.layout = mddev->layout; | |
9d8f0363 | 5602 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5603 | |
5604 | if (copy_to_user(arg, &info, sizeof(info))) | |
5605 | return -EFAULT; | |
5606 | ||
5607 | return 0; | |
5608 | } | |
5609 | ||
fd01b88c | 5610 | static int get_bitmap_file(struct mddev * mddev, void __user * arg) |
32a7627c N |
5611 | { |
5612 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5613 | char *ptr, *buf = NULL; | |
5614 | int err = -ENOMEM; | |
5615 | ||
60559da4 | 5616 | file = kmalloc(sizeof(*file), GFP_NOIO); |
2a2275d6 | 5617 | |
32a7627c N |
5618 | if (!file) |
5619 | goto out; | |
5620 | ||
5621 | /* bitmap disabled, zero the first byte and copy out */ | |
1ec885cd | 5622 | if (!mddev->bitmap || !mddev->bitmap->storage.file) { |
32a7627c N |
5623 | file->pathname[0] = '\0'; |
5624 | goto copy_out; | |
5625 | } | |
5626 | ||
5627 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5628 | if (!buf) | |
5629 | goto out; | |
5630 | ||
1ec885cd N |
5631 | ptr = d_path(&mddev->bitmap->storage.file->f_path, |
5632 | buf, sizeof(file->pathname)); | |
6bcfd601 | 5633 | if (IS_ERR(ptr)) |
32a7627c N |
5634 | goto out; |
5635 | ||
5636 | strcpy(file->pathname, ptr); | |
5637 | ||
5638 | copy_out: | |
5639 | err = 0; | |
5640 | if (copy_to_user(arg, file, sizeof(*file))) | |
5641 | err = -EFAULT; | |
5642 | out: | |
5643 | kfree(buf); | |
5644 | kfree(file); | |
5645 | return err; | |
5646 | } | |
5647 | ||
fd01b88c | 5648 | static int get_disk_info(struct mddev * mddev, void __user * arg) |
1da177e4 LT |
5649 | { |
5650 | mdu_disk_info_t info; | |
3cb03002 | 5651 | struct md_rdev *rdev; |
1da177e4 LT |
5652 | |
5653 | if (copy_from_user(&info, arg, sizeof(info))) | |
5654 | return -EFAULT; | |
5655 | ||
1ca69c4b N |
5656 | rcu_read_lock(); |
5657 | rdev = find_rdev_nr_rcu(mddev, info.number); | |
1da177e4 LT |
5658 | if (rdev) { |
5659 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5660 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5661 | info.raid_disk = rdev->raid_disk; | |
5662 | info.state = 0; | |
b2d444d7 | 5663 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5664 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5665 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5666 | info.state |= (1<<MD_DISK_ACTIVE); |
5667 | info.state |= (1<<MD_DISK_SYNC); | |
5668 | } | |
8ddf9efe N |
5669 | if (test_bit(WriteMostly, &rdev->flags)) |
5670 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5671 | } else { |
5672 | info.major = info.minor = 0; | |
5673 | info.raid_disk = -1; | |
5674 | info.state = (1<<MD_DISK_REMOVED); | |
5675 | } | |
1ca69c4b | 5676 | rcu_read_unlock(); |
1da177e4 LT |
5677 | |
5678 | if (copy_to_user(arg, &info, sizeof(info))) | |
5679 | return -EFAULT; | |
5680 | ||
5681 | return 0; | |
5682 | } | |
5683 | ||
fd01b88c | 5684 | static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info) |
1da177e4 LT |
5685 | { |
5686 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5687 | struct md_rdev *rdev; |
1da177e4 LT |
5688 | dev_t dev = MKDEV(info->major,info->minor); |
5689 | ||
5690 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5691 | return -EOVERFLOW; | |
5692 | ||
5693 | if (!mddev->raid_disks) { | |
5694 | int err; | |
5695 | /* expecting a device which has a superblock */ | |
5696 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5697 | if (IS_ERR(rdev)) { | |
5698 | printk(KERN_WARNING | |
5699 | "md: md_import_device returned %ld\n", | |
5700 | PTR_ERR(rdev)); | |
5701 | return PTR_ERR(rdev); | |
5702 | } | |
5703 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
5704 | struct md_rdev *rdev0 |
5705 | = list_entry(mddev->disks.next, | |
5706 | struct md_rdev, same_set); | |
a9f326eb | 5707 | err = super_types[mddev->major_version] |
1da177e4 LT |
5708 | .load_super(rdev, rdev0, mddev->minor_version); |
5709 | if (err < 0) { | |
5710 | printk(KERN_WARNING | |
5711 | "md: %s has different UUID to %s\n", | |
5712 | bdevname(rdev->bdev,b), | |
5713 | bdevname(rdev0->bdev,b2)); | |
5714 | export_rdev(rdev); | |
5715 | return -EINVAL; | |
5716 | } | |
5717 | } | |
5718 | err = bind_rdev_to_array(rdev, mddev); | |
5719 | if (err) | |
5720 | export_rdev(rdev); | |
5721 | return err; | |
5722 | } | |
5723 | ||
5724 | /* | |
5725 | * add_new_disk can be used once the array is assembled | |
5726 | * to add "hot spares". They must already have a superblock | |
5727 | * written | |
5728 | */ | |
5729 | if (mddev->pers) { | |
5730 | int err; | |
5731 | if (!mddev->pers->hot_add_disk) { | |
5732 | printk(KERN_WARNING | |
5733 | "%s: personality does not support diskops!\n", | |
5734 | mdname(mddev)); | |
5735 | return -EINVAL; | |
5736 | } | |
7b1e35f6 N |
5737 | if (mddev->persistent) |
5738 | rdev = md_import_device(dev, mddev->major_version, | |
5739 | mddev->minor_version); | |
5740 | else | |
5741 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5742 | if (IS_ERR(rdev)) { |
5743 | printk(KERN_WARNING | |
5744 | "md: md_import_device returned %ld\n", | |
5745 | PTR_ERR(rdev)); | |
5746 | return PTR_ERR(rdev); | |
5747 | } | |
1a855a06 | 5748 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5749 | if (!mddev->persistent) { |
5750 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5751 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5752 | rdev->raid_disk = info->raid_disk; |
bf572541 | 5753 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 5754 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 5755 | } else |
41158c7e | 5756 | rdev->raid_disk = -1; |
f466722c | 5757 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
5758 | } else |
5759 | super_types[mddev->major_version]. | |
5760 | validate_super(mddev, rdev); | |
bedd86b7 | 5761 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 5762 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
5763 | /* This was a hot-add request, but events doesn't |
5764 | * match, so reject it. | |
5765 | */ | |
5766 | export_rdev(rdev); | |
5767 | return -EINVAL; | |
5768 | } | |
5769 | ||
b2d444d7 | 5770 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5771 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5772 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5773 | else |
5774 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5775 | |
1da177e4 LT |
5776 | rdev->raid_disk = -1; |
5777 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5778 | if (!err && !mddev->pers->hot_remove_disk) { |
5779 | /* If there is hot_add_disk but no hot_remove_disk | |
5780 | * then added disks for geometry changes, | |
5781 | * and should be added immediately. | |
5782 | */ | |
5783 | super_types[mddev->major_version]. | |
5784 | validate_super(mddev, rdev); | |
5785 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5786 | if (err) | |
5787 | unbind_rdev_from_array(rdev); | |
5788 | } | |
1da177e4 LT |
5789 | if (err) |
5790 | export_rdev(rdev); | |
52664732 | 5791 | else |
00bcb4ac | 5792 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5793 | |
7ceb17e8 | 5794 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
72a23c21 NB |
5795 | if (mddev->degraded) |
5796 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5797 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
9864c005 | 5798 | if (!err) |
5799 | md_new_event(mddev); | |
005eca5e | 5800 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5801 | return err; |
5802 | } | |
5803 | ||
5804 | /* otherwise, add_new_disk is only allowed | |
5805 | * for major_version==0 superblocks | |
5806 | */ | |
5807 | if (mddev->major_version != 0) { | |
5808 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5809 | mdname(mddev)); | |
5810 | return -EINVAL; | |
5811 | } | |
5812 | ||
5813 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5814 | int err; | |
d710e138 | 5815 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5816 | if (IS_ERR(rdev)) { |
5817 | printk(KERN_WARNING | |
5818 | "md: error, md_import_device() returned %ld\n", | |
5819 | PTR_ERR(rdev)); | |
5820 | return PTR_ERR(rdev); | |
5821 | } | |
5822 | rdev->desc_nr = info->number; | |
5823 | if (info->raid_disk < mddev->raid_disks) | |
5824 | rdev->raid_disk = info->raid_disk; | |
5825 | else | |
5826 | rdev->raid_disk = -1; | |
5827 | ||
1da177e4 | 5828 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5829 | if (info->state & (1<<MD_DISK_SYNC)) |
5830 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5831 | |
8ddf9efe N |
5832 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5833 | set_bit(WriteMostly, &rdev->flags); | |
5834 | ||
1da177e4 LT |
5835 | if (!mddev->persistent) { |
5836 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5837 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5838 | } else | |
57b2caa3 | 5839 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5840 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5841 | |
2bf071bf N |
5842 | err = bind_rdev_to_array(rdev, mddev); |
5843 | if (err) { | |
5844 | export_rdev(rdev); | |
5845 | return err; | |
5846 | } | |
1da177e4 LT |
5847 | } |
5848 | ||
5849 | return 0; | |
5850 | } | |
5851 | ||
fd01b88c | 5852 | static int hot_remove_disk(struct mddev * mddev, dev_t dev) |
1da177e4 LT |
5853 | { |
5854 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 5855 | struct md_rdev *rdev; |
1da177e4 | 5856 | |
1da177e4 LT |
5857 | rdev = find_rdev(mddev, dev); |
5858 | if (!rdev) | |
5859 | return -ENXIO; | |
5860 | ||
3ea8929d N |
5861 | clear_bit(Blocked, &rdev->flags); |
5862 | remove_and_add_spares(mddev, rdev); | |
5863 | ||
1da177e4 LT |
5864 | if (rdev->raid_disk >= 0) |
5865 | goto busy; | |
5866 | ||
5867 | kick_rdev_from_array(rdev); | |
850b2b42 | 5868 | md_update_sb(mddev, 1); |
d7603b7e | 5869 | md_new_event(mddev); |
1da177e4 LT |
5870 | |
5871 | return 0; | |
5872 | busy: | |
fdefa4d8 | 5873 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5874 | bdevname(rdev->bdev,b), mdname(mddev)); |
5875 | return -EBUSY; | |
5876 | } | |
5877 | ||
fd01b88c | 5878 | static int hot_add_disk(struct mddev * mddev, dev_t dev) |
1da177e4 LT |
5879 | { |
5880 | char b[BDEVNAME_SIZE]; | |
5881 | int err; | |
3cb03002 | 5882 | struct md_rdev *rdev; |
1da177e4 LT |
5883 | |
5884 | if (!mddev->pers) | |
5885 | return -ENODEV; | |
5886 | ||
5887 | if (mddev->major_version != 0) { | |
5888 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5889 | " version-0 superblocks.\n", | |
5890 | mdname(mddev)); | |
5891 | return -EINVAL; | |
5892 | } | |
5893 | if (!mddev->pers->hot_add_disk) { | |
5894 | printk(KERN_WARNING | |
5895 | "%s: personality does not support diskops!\n", | |
5896 | mdname(mddev)); | |
5897 | return -EINVAL; | |
5898 | } | |
5899 | ||
d710e138 | 5900 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5901 | if (IS_ERR(rdev)) { |
5902 | printk(KERN_WARNING | |
5903 | "md: error, md_import_device() returned %ld\n", | |
5904 | PTR_ERR(rdev)); | |
5905 | return -EINVAL; | |
5906 | } | |
5907 | ||
5908 | if (mddev->persistent) | |
57b2caa3 | 5909 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5910 | else |
77304d2a | 5911 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5912 | |
8190e754 | 5913 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5914 | |
b2d444d7 | 5915 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5916 | printk(KERN_WARNING |
5917 | "md: can not hot-add faulty %s disk to %s!\n", | |
5918 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5919 | err = -EINVAL; | |
5920 | goto abort_export; | |
5921 | } | |
b2d444d7 | 5922 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5923 | rdev->desc_nr = -1; |
5842730d | 5924 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5925 | err = bind_rdev_to_array(rdev, mddev); |
5926 | if (err) | |
5927 | goto abort_export; | |
1da177e4 LT |
5928 | |
5929 | /* | |
5930 | * The rest should better be atomic, we can have disk failures | |
5931 | * noticed in interrupt contexts ... | |
5932 | */ | |
5933 | ||
1da177e4 LT |
5934 | rdev->raid_disk = -1; |
5935 | ||
850b2b42 | 5936 | md_update_sb(mddev, 1); |
1da177e4 LT |
5937 | |
5938 | /* | |
5939 | * Kick recovery, maybe this spare has to be added to the | |
5940 | * array immediately. | |
5941 | */ | |
5942 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5943 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5944 | md_new_event(mddev); |
1da177e4 LT |
5945 | return 0; |
5946 | ||
1da177e4 LT |
5947 | abort_export: |
5948 | export_rdev(rdev); | |
5949 | return err; | |
5950 | } | |
5951 | ||
fd01b88c | 5952 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 5953 | { |
035328c2 | 5954 | int err = 0; |
32a7627c | 5955 | |
36fa3063 | 5956 | if (mddev->pers) { |
d66b1b39 | 5957 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
5958 | return -EBUSY; |
5959 | if (mddev->recovery || mddev->sync_thread) | |
5960 | return -EBUSY; | |
5961 | /* we should be able to change the bitmap.. */ | |
5962 | } | |
32a7627c | 5963 | |
32a7627c | 5964 | |
36fa3063 | 5965 | if (fd >= 0) { |
035328c2 | 5966 | struct inode *inode; |
36fa3063 N |
5967 | if (mddev->bitmap) |
5968 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5969 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5970 | |
c3d9714e | 5971 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5972 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5973 | mdname(mddev)); | |
5974 | return -EBADF; | |
5975 | } | |
5976 | ||
035328c2 N |
5977 | inode = mddev->bitmap_info.file->f_mapping->host; |
5978 | if (!S_ISREG(inode->i_mode)) { | |
5979 | printk(KERN_ERR "%s: error: bitmap file must be a regular file\n", | |
5980 | mdname(mddev)); | |
5981 | err = -EBADF; | |
5982 | } else if (!(mddev->bitmap_info.file->f_mode & FMODE_WRITE)) { | |
5983 | printk(KERN_ERR "%s: error: bitmap file must open for write\n", | |
5984 | mdname(mddev)); | |
5985 | err = -EBADF; | |
5986 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
36fa3063 N |
5987 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", |
5988 | mdname(mddev)); | |
035328c2 N |
5989 | err = -EBUSY; |
5990 | } | |
5991 | if (err) { | |
c3d9714e N |
5992 | fput(mddev->bitmap_info.file); |
5993 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5994 | return err; |
5995 | } | |
c3d9714e | 5996 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5997 | } else if (mddev->bitmap == NULL) |
5998 | return -ENOENT; /* cannot remove what isn't there */ | |
5999 | err = 0; | |
6000 | if (mddev->pers) { | |
6001 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6002 | if (fd >= 0) { |
36fa3063 | 6003 | err = bitmap_create(mddev); |
69e51b44 N |
6004 | if (!err) |
6005 | err = bitmap_load(mddev); | |
6006 | } | |
d7375ab3 | 6007 | if (fd < 0 || err) { |
36fa3063 | 6008 | bitmap_destroy(mddev); |
d7375ab3 N |
6009 | fd = -1; /* make sure to put the file */ |
6010 | } | |
36fa3063 | 6011 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6012 | } |
6013 | if (fd < 0) { | |
035328c2 | 6014 | if (mddev->bitmap_info.file) |
c3d9714e | 6015 | fput(mddev->bitmap_info.file); |
c3d9714e | 6016 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
6017 | } |
6018 | ||
32a7627c N |
6019 | return err; |
6020 | } | |
6021 | ||
1da177e4 LT |
6022 | /* |
6023 | * set_array_info is used two different ways | |
6024 | * The original usage is when creating a new array. | |
6025 | * In this usage, raid_disks is > 0 and it together with | |
6026 | * level, size, not_persistent,layout,chunksize determine the | |
6027 | * shape of the array. | |
6028 | * This will always create an array with a type-0.90.0 superblock. | |
6029 | * The newer usage is when assembling an array. | |
6030 | * In this case raid_disks will be 0, and the major_version field is | |
6031 | * use to determine which style super-blocks are to be found on the devices. | |
6032 | * The minor and patch _version numbers are also kept incase the | |
6033 | * super_block handler wishes to interpret them. | |
6034 | */ | |
fd01b88c | 6035 | static int set_array_info(struct mddev * mddev, mdu_array_info_t *info) |
1da177e4 LT |
6036 | { |
6037 | ||
6038 | if (info->raid_disks == 0) { | |
6039 | /* just setting version number for superblock loading */ | |
6040 | if (info->major_version < 0 || | |
50511da3 | 6041 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6042 | super_types[info->major_version].name == NULL) { |
6043 | /* maybe try to auto-load a module? */ | |
6044 | printk(KERN_INFO | |
6045 | "md: superblock version %d not known\n", | |
6046 | info->major_version); | |
6047 | return -EINVAL; | |
6048 | } | |
6049 | mddev->major_version = info->major_version; | |
6050 | mddev->minor_version = info->minor_version; | |
6051 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6052 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6053 | /* ensure mddev_put doesn't delete this now that there |
6054 | * is some minimal configuration. | |
6055 | */ | |
6056 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
6057 | return 0; |
6058 | } | |
6059 | mddev->major_version = MD_MAJOR_VERSION; | |
6060 | mddev->minor_version = MD_MINOR_VERSION; | |
6061 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
6062 | mddev->ctime = get_seconds(); | |
6063 | ||
6064 | mddev->level = info->level; | |
17115e03 | 6065 | mddev->clevel[0] = 0; |
58c0fed4 | 6066 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6067 | mddev->raid_disks = info->raid_disks; |
6068 | /* don't set md_minor, it is determined by which /dev/md* was | |
6069 | * openned | |
6070 | */ | |
6071 | if (info->state & (1<<MD_SB_CLEAN)) | |
6072 | mddev->recovery_cp = MaxSector; | |
6073 | else | |
6074 | mddev->recovery_cp = 0; | |
6075 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6076 | mddev->external = 0; |
1da177e4 LT |
6077 | |
6078 | mddev->layout = info->layout; | |
9d8f0363 | 6079 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
6080 | |
6081 | mddev->max_disks = MD_SB_DISKS; | |
6082 | ||
e691063a N |
6083 | if (mddev->persistent) |
6084 | mddev->flags = 0; | |
850b2b42 | 6085 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6086 | |
c3d9714e | 6087 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6088 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6089 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6090 | |
f6705578 N |
6091 | mddev->reshape_position = MaxSector; |
6092 | ||
1da177e4 LT |
6093 | /* |
6094 | * Generate a 128 bit UUID | |
6095 | */ | |
6096 | get_random_bytes(mddev->uuid, 16); | |
6097 | ||
f6705578 | 6098 | mddev->new_level = mddev->level; |
664e7c41 | 6099 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6100 | mddev->new_layout = mddev->layout; |
6101 | mddev->delta_disks = 0; | |
2c810cdd | 6102 | mddev->reshape_backwards = 0; |
f6705578 | 6103 | |
1da177e4 LT |
6104 | return 0; |
6105 | } | |
6106 | ||
fd01b88c | 6107 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6108 | { |
b522adcd DW |
6109 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6110 | ||
6111 | if (mddev->external_size) | |
6112 | return; | |
6113 | ||
1f403624 DW |
6114 | mddev->array_sectors = array_sectors; |
6115 | } | |
6116 | EXPORT_SYMBOL(md_set_array_sectors); | |
6117 | ||
fd01b88c | 6118 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6119 | { |
3cb03002 | 6120 | struct md_rdev *rdev; |
a35b0d69 | 6121 | int rv; |
d71f9f88 | 6122 | int fit = (num_sectors == 0); |
a35b0d69 N |
6123 | |
6124 | if (mddev->pers->resize == NULL) | |
6125 | return -EINVAL; | |
d71f9f88 AN |
6126 | /* The "num_sectors" is the number of sectors of each device that |
6127 | * is used. This can only make sense for arrays with redundancy. | |
6128 | * linear and raid0 always use whatever space is available. We can only | |
6129 | * consider changing this number if no resync or reconstruction is | |
6130 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6131 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6132 | * of each device. If num_sectors is zero, we find the largest size |
6133 | * that fits. | |
a35b0d69 N |
6134 | */ |
6135 | if (mddev->sync_thread) | |
6136 | return -EBUSY; | |
bd8839e0 N |
6137 | if (mddev->ro) |
6138 | return -EROFS; | |
a4a6125a | 6139 | |
dafb20fa | 6140 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6141 | sector_t avail = rdev->sectors; |
01ab5662 | 6142 | |
d71f9f88 AN |
6143 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6144 | num_sectors = avail; | |
6145 | if (avail < num_sectors) | |
a35b0d69 N |
6146 | return -ENOSPC; |
6147 | } | |
d71f9f88 | 6148 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
6149 | if (!rv) |
6150 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
6151 | return rv; |
6152 | } | |
6153 | ||
fd01b88c | 6154 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6155 | { |
6156 | int rv; | |
c6563a8c | 6157 | struct md_rdev *rdev; |
da943b99 | 6158 | /* change the number of raid disks */ |
63c70c4f | 6159 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6160 | return -EINVAL; |
bd8839e0 N |
6161 | if (mddev->ro) |
6162 | return -EROFS; | |
da943b99 | 6163 | if (raid_disks <= 0 || |
233fca36 | 6164 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6165 | return -EINVAL; |
63c70c4f | 6166 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 6167 | return -EBUSY; |
c6563a8c N |
6168 | |
6169 | rdev_for_each(rdev, mddev) { | |
6170 | if (mddev->raid_disks < raid_disks && | |
6171 | rdev->data_offset < rdev->new_data_offset) | |
6172 | return -EINVAL; | |
6173 | if (mddev->raid_disks > raid_disks && | |
6174 | rdev->data_offset > rdev->new_data_offset) | |
6175 | return -EINVAL; | |
6176 | } | |
6177 | ||
63c70c4f | 6178 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6179 | if (mddev->delta_disks < 0) |
6180 | mddev->reshape_backwards = 1; | |
6181 | else if (mddev->delta_disks > 0) | |
6182 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6183 | |
6184 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6185 | if (rv < 0) { |
de171cb9 | 6186 | mddev->delta_disks = 0; |
2c810cdd N |
6187 | mddev->reshape_backwards = 0; |
6188 | } | |
da943b99 N |
6189 | return rv; |
6190 | } | |
6191 | ||
6192 | ||
1da177e4 LT |
6193 | /* |
6194 | * update_array_info is used to change the configuration of an | |
6195 | * on-line array. | |
6196 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6197 | * fields in the info are checked against the array. | |
6198 | * Any differences that cannot be handled will cause an error. | |
6199 | * Normally, only one change can be managed at a time. | |
6200 | */ | |
fd01b88c | 6201 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6202 | { |
6203 | int rv = 0; | |
6204 | int cnt = 0; | |
36fa3063 N |
6205 | int state = 0; |
6206 | ||
6207 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6208 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6209 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6210 | |
6211 | if (mddev->major_version != info->major_version || | |
6212 | mddev->minor_version != info->minor_version || | |
6213 | /* mddev->patch_version != info->patch_version || */ | |
6214 | mddev->ctime != info->ctime || | |
6215 | mddev->level != info->level || | |
6216 | /* mddev->layout != info->layout || */ | |
6217 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 6218 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6219 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6220 | ((state^info->state) & 0xfffffe00) | |
6221 | ) | |
1da177e4 LT |
6222 | return -EINVAL; |
6223 | /* Check there is only one change */ | |
58c0fed4 AN |
6224 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6225 | cnt++; | |
6226 | if (mddev->raid_disks != info->raid_disks) | |
6227 | cnt++; | |
6228 | if (mddev->layout != info->layout) | |
6229 | cnt++; | |
6230 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6231 | cnt++; | |
6232 | if (cnt == 0) | |
6233 | return 0; | |
6234 | if (cnt > 1) | |
6235 | return -EINVAL; | |
1da177e4 LT |
6236 | |
6237 | if (mddev->layout != info->layout) { | |
6238 | /* Change layout | |
6239 | * we don't need to do anything at the md level, the | |
6240 | * personality will take care of it all. | |
6241 | */ | |
50ac168a | 6242 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6243 | return -EINVAL; |
597a711b N |
6244 | else { |
6245 | mddev->new_layout = info->layout; | |
50ac168a | 6246 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6247 | if (rv) |
6248 | mddev->new_layout = mddev->layout; | |
6249 | return rv; | |
6250 | } | |
1da177e4 | 6251 | } |
58c0fed4 | 6252 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6253 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6254 | |
da943b99 N |
6255 | if (mddev->raid_disks != info->raid_disks) |
6256 | rv = update_raid_disks(mddev, info->raid_disks); | |
6257 | ||
36fa3063 | 6258 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
d66b1b39 | 6259 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) |
36fa3063 N |
6260 | return -EINVAL; |
6261 | if (mddev->recovery || mddev->sync_thread) | |
6262 | return -EBUSY; | |
6263 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
6264 | /* add the bitmap */ | |
6265 | if (mddev->bitmap) | |
6266 | return -EEXIST; | |
c3d9714e | 6267 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 6268 | return -EINVAL; |
c3d9714e N |
6269 | mddev->bitmap_info.offset = |
6270 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6271 | mddev->bitmap_info.space = |
6272 | mddev->bitmap_info.default_space; | |
36fa3063 N |
6273 | mddev->pers->quiesce(mddev, 1); |
6274 | rv = bitmap_create(mddev); | |
69e51b44 N |
6275 | if (!rv) |
6276 | rv = bitmap_load(mddev); | |
36fa3063 N |
6277 | if (rv) |
6278 | bitmap_destroy(mddev); | |
6279 | mddev->pers->quiesce(mddev, 0); | |
6280 | } else { | |
6281 | /* remove the bitmap */ | |
6282 | if (!mddev->bitmap) | |
6283 | return -ENOENT; | |
1ec885cd | 6284 | if (mddev->bitmap->storage.file) |
36fa3063 N |
6285 | return -EINVAL; |
6286 | mddev->pers->quiesce(mddev, 1); | |
6287 | bitmap_destroy(mddev); | |
6288 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6289 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6290 | } |
6291 | } | |
850b2b42 | 6292 | md_update_sb(mddev, 1); |
1da177e4 LT |
6293 | return rv; |
6294 | } | |
6295 | ||
fd01b88c | 6296 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6297 | { |
3cb03002 | 6298 | struct md_rdev *rdev; |
1ca69c4b | 6299 | int err = 0; |
1da177e4 LT |
6300 | |
6301 | if (mddev->pers == NULL) | |
6302 | return -ENODEV; | |
6303 | ||
1ca69c4b N |
6304 | rcu_read_lock(); |
6305 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6306 | if (!rdev) |
1ca69c4b N |
6307 | err = -ENODEV; |
6308 | else { | |
6309 | md_error(mddev, rdev); | |
6310 | if (!test_bit(Faulty, &rdev->flags)) | |
6311 | err = -EBUSY; | |
6312 | } | |
6313 | rcu_read_unlock(); | |
6314 | return err; | |
1da177e4 LT |
6315 | } |
6316 | ||
2f9618ce AN |
6317 | /* |
6318 | * We have a problem here : there is no easy way to give a CHS | |
6319 | * virtual geometry. We currently pretend that we have a 2 heads | |
6320 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6321 | * dosfs just mad... ;-) | |
6322 | */ | |
a885c8c4 CH |
6323 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6324 | { | |
fd01b88c | 6325 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6326 | |
6327 | geo->heads = 2; | |
6328 | geo->sectors = 4; | |
49ce6cea | 6329 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6330 | return 0; |
6331 | } | |
6332 | ||
cb335f88 NS |
6333 | static inline bool md_ioctl_valid(unsigned int cmd) |
6334 | { | |
6335 | switch (cmd) { | |
6336 | case ADD_NEW_DISK: | |
6337 | case BLKROSET: | |
6338 | case GET_ARRAY_INFO: | |
6339 | case GET_BITMAP_FILE: | |
6340 | case GET_DISK_INFO: | |
6341 | case HOT_ADD_DISK: | |
6342 | case HOT_REMOVE_DISK: | |
6343 | case PRINT_RAID_DEBUG: | |
6344 | case RAID_AUTORUN: | |
6345 | case RAID_VERSION: | |
6346 | case RESTART_ARRAY_RW: | |
6347 | case RUN_ARRAY: | |
6348 | case SET_ARRAY_INFO: | |
6349 | case SET_BITMAP_FILE: | |
6350 | case SET_DISK_FAULTY: | |
6351 | case STOP_ARRAY: | |
6352 | case STOP_ARRAY_RO: | |
6353 | return true; | |
6354 | default: | |
6355 | return false; | |
6356 | } | |
6357 | } | |
6358 | ||
a39907fa | 6359 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6360 | unsigned int cmd, unsigned long arg) |
6361 | { | |
6362 | int err = 0; | |
6363 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6364 | struct mddev *mddev = NULL; |
e2218350 | 6365 | int ro; |
1da177e4 | 6366 | |
cb335f88 NS |
6367 | if (!md_ioctl_valid(cmd)) |
6368 | return -ENOTTY; | |
6369 | ||
506c9e44 N |
6370 | switch (cmd) { |
6371 | case RAID_VERSION: | |
6372 | case GET_ARRAY_INFO: | |
6373 | case GET_DISK_INFO: | |
6374 | break; | |
6375 | default: | |
6376 | if (!capable(CAP_SYS_ADMIN)) | |
6377 | return -EACCES; | |
6378 | } | |
1da177e4 LT |
6379 | |
6380 | /* | |
6381 | * Commands dealing with the RAID driver but not any | |
6382 | * particular array: | |
6383 | */ | |
c02c0aeb N |
6384 | switch (cmd) { |
6385 | case RAID_VERSION: | |
6386 | err = get_version(argp); | |
6387 | goto done; | |
1da177e4 | 6388 | |
c02c0aeb N |
6389 | case PRINT_RAID_DEBUG: |
6390 | err = 0; | |
6391 | md_print_devices(); | |
6392 | goto done; | |
1da177e4 LT |
6393 | |
6394 | #ifndef MODULE | |
c02c0aeb N |
6395 | case RAID_AUTORUN: |
6396 | err = 0; | |
6397 | autostart_arrays(arg); | |
6398 | goto done; | |
1da177e4 | 6399 | #endif |
c02c0aeb | 6400 | default:; |
1da177e4 LT |
6401 | } |
6402 | ||
6403 | /* | |
6404 | * Commands creating/starting a new array: | |
6405 | */ | |
6406 | ||
a39907fa | 6407 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6408 | |
6409 | if (!mddev) { | |
6410 | BUG(); | |
6411 | goto abort; | |
6412 | } | |
6413 | ||
1ca69c4b N |
6414 | /* Some actions do not requires the mutex */ |
6415 | switch (cmd) { | |
6416 | case GET_ARRAY_INFO: | |
6417 | if (!mddev->raid_disks && !mddev->external) | |
6418 | err = -ENODEV; | |
6419 | else | |
6420 | err = get_array_info(mddev, argp); | |
6421 | goto abort; | |
6422 | ||
6423 | case GET_DISK_INFO: | |
6424 | if (!mddev->raid_disks && !mddev->external) | |
6425 | err = -ENODEV; | |
6426 | else | |
6427 | err = get_disk_info(mddev, argp); | |
6428 | goto abort; | |
6429 | ||
6430 | case SET_DISK_FAULTY: | |
6431 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
6432 | goto abort; | |
6433 | } | |
6434 | ||
a7a3f08d N |
6435 | if (cmd == ADD_NEW_DISK) |
6436 | /* need to ensure md_delayed_delete() has completed */ | |
6437 | flush_workqueue(md_misc_wq); | |
6438 | ||
90f5f7ad HR |
6439 | if (cmd == HOT_REMOVE_DISK) |
6440 | /* need to ensure recovery thread has run */ | |
6441 | wait_event_interruptible_timeout(mddev->sb_wait, | |
6442 | !test_bit(MD_RECOVERY_NEEDED, | |
6443 | &mddev->flags), | |
6444 | msecs_to_jiffies(5000)); | |
260fa034 N |
6445 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
6446 | /* Need to flush page cache, and ensure no-one else opens | |
6447 | * and writes | |
6448 | */ | |
6449 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 6450 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
6451 | mutex_unlock(&mddev->open_mutex); |
6452 | err = -EBUSY; | |
6453 | goto abort; | |
6454 | } | |
6455 | set_bit(MD_STILL_CLOSED, &mddev->flags); | |
6456 | mutex_unlock(&mddev->open_mutex); | |
6457 | sync_blockdev(bdev); | |
6458 | } | |
1da177e4 LT |
6459 | err = mddev_lock(mddev); |
6460 | if (err) { | |
6461 | printk(KERN_INFO | |
6462 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
6463 | err, cmd); | |
6464 | goto abort; | |
6465 | } | |
6466 | ||
c02c0aeb N |
6467 | if (cmd == SET_ARRAY_INFO) { |
6468 | mdu_array_info_t info; | |
6469 | if (!arg) | |
6470 | memset(&info, 0, sizeof(info)); | |
6471 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6472 | err = -EFAULT; | |
6473 | goto abort_unlock; | |
6474 | } | |
6475 | if (mddev->pers) { | |
6476 | err = update_array_info(mddev, &info); | |
6477 | if (err) { | |
6478 | printk(KERN_WARNING "md: couldn't update" | |
6479 | " array info. %d\n", err); | |
6480 | goto abort_unlock; | |
1da177e4 LT |
6481 | } |
6482 | goto done_unlock; | |
c02c0aeb N |
6483 | } |
6484 | if (!list_empty(&mddev->disks)) { | |
6485 | printk(KERN_WARNING | |
6486 | "md: array %s already has disks!\n", | |
6487 | mdname(mddev)); | |
6488 | err = -EBUSY; | |
6489 | goto abort_unlock; | |
6490 | } | |
6491 | if (mddev->raid_disks) { | |
6492 | printk(KERN_WARNING | |
6493 | "md: array %s already initialised!\n", | |
6494 | mdname(mddev)); | |
6495 | err = -EBUSY; | |
6496 | goto abort_unlock; | |
6497 | } | |
6498 | err = set_array_info(mddev, &info); | |
6499 | if (err) { | |
6500 | printk(KERN_WARNING "md: couldn't set" | |
6501 | " array info. %d\n", err); | |
6502 | goto abort_unlock; | |
6503 | } | |
6504 | goto done_unlock; | |
1da177e4 LT |
6505 | } |
6506 | ||
6507 | /* | |
6508 | * Commands querying/configuring an existing array: | |
6509 | */ | |
32a7627c | 6510 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6511 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6512 | if ((!mddev->raid_disks && !mddev->external) |
6513 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6514 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6515 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
6516 | err = -ENODEV; |
6517 | goto abort_unlock; | |
6518 | } | |
6519 | ||
6520 | /* | |
6521 | * Commands even a read-only array can execute: | |
6522 | */ | |
c02c0aeb N |
6523 | switch (cmd) { |
6524 | case GET_BITMAP_FILE: | |
6525 | err = get_bitmap_file(mddev, argp); | |
6526 | goto done_unlock; | |
32a7627c | 6527 | |
c02c0aeb N |
6528 | case RESTART_ARRAY_RW: |
6529 | err = restart_array(mddev); | |
6530 | goto done_unlock; | |
1da177e4 | 6531 | |
c02c0aeb N |
6532 | case STOP_ARRAY: |
6533 | err = do_md_stop(mddev, 0, bdev); | |
6534 | goto done_unlock; | |
1da177e4 | 6535 | |
c02c0aeb N |
6536 | case STOP_ARRAY_RO: |
6537 | err = md_set_readonly(mddev, bdev); | |
6538 | goto done_unlock; | |
1da177e4 | 6539 | |
3ea8929d N |
6540 | case HOT_REMOVE_DISK: |
6541 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
6542 | goto done_unlock; | |
6543 | ||
7ceb17e8 N |
6544 | case ADD_NEW_DISK: |
6545 | /* We can support ADD_NEW_DISK on read-only arrays | |
6546 | * on if we are re-adding a preexisting device. | |
6547 | * So require mddev->pers and MD_DISK_SYNC. | |
6548 | */ | |
6549 | if (mddev->pers) { | |
6550 | mdu_disk_info_t info; | |
6551 | if (copy_from_user(&info, argp, sizeof(info))) | |
6552 | err = -EFAULT; | |
6553 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
6554 | /* Need to clear read-only for this */ | |
6555 | break; | |
6556 | else | |
6557 | err = add_new_disk(mddev, &info); | |
6558 | goto done_unlock; | |
6559 | } | |
6560 | break; | |
6561 | ||
c02c0aeb N |
6562 | case BLKROSET: |
6563 | if (get_user(ro, (int __user *)(arg))) { | |
6564 | err = -EFAULT; | |
6565 | goto done_unlock; | |
6566 | } | |
6567 | err = -EINVAL; | |
e2218350 | 6568 | |
c02c0aeb N |
6569 | /* if the bdev is going readonly the value of mddev->ro |
6570 | * does not matter, no writes are coming | |
6571 | */ | |
6572 | if (ro) | |
6573 | goto done_unlock; | |
e2218350 | 6574 | |
c02c0aeb N |
6575 | /* are we are already prepared for writes? */ |
6576 | if (mddev->ro != 1) | |
6577 | goto done_unlock; | |
e2218350 | 6578 | |
c02c0aeb N |
6579 | /* transitioning to readauto need only happen for |
6580 | * arrays that call md_write_start | |
6581 | */ | |
6582 | if (mddev->pers) { | |
6583 | err = restart_array(mddev); | |
6584 | if (err == 0) { | |
6585 | mddev->ro = 2; | |
6586 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6587 | } |
c02c0aeb N |
6588 | } |
6589 | goto done_unlock; | |
1da177e4 LT |
6590 | } |
6591 | ||
6592 | /* | |
6593 | * The remaining ioctls are changing the state of the | |
f91de92e N |
6594 | * superblock, so we do not allow them on read-only arrays. |
6595 | * However non-MD ioctls (e.g. get-size) will still come through | |
6596 | * here and hit the 'default' below, so only disallow | |
6597 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 6598 | */ |
bb57fc64 | 6599 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
6600 | if (mddev->ro == 2) { |
6601 | mddev->ro = 0; | |
00bcb4ac | 6602 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 6603 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
6604 | /* mddev_unlock will wake thread */ |
6605 | /* If a device failed while we were read-only, we | |
6606 | * need to make sure the metadata is updated now. | |
6607 | */ | |
6608 | if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
6609 | mddev_unlock(mddev); | |
6610 | wait_event(mddev->sb_wait, | |
6611 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) && | |
6612 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
29f097c4 | 6613 | mddev_lock_nointr(mddev); |
f3378b48 | 6614 | } |
f91de92e N |
6615 | } else { |
6616 | err = -EROFS; | |
6617 | goto abort_unlock; | |
6618 | } | |
1da177e4 LT |
6619 | } |
6620 | ||
c02c0aeb N |
6621 | switch (cmd) { |
6622 | case ADD_NEW_DISK: | |
1da177e4 | 6623 | { |
c02c0aeb N |
6624 | mdu_disk_info_t info; |
6625 | if (copy_from_user(&info, argp, sizeof(info))) | |
6626 | err = -EFAULT; | |
6627 | else | |
6628 | err = add_new_disk(mddev, &info); | |
6629 | goto done_unlock; | |
6630 | } | |
1da177e4 | 6631 | |
c02c0aeb N |
6632 | case HOT_ADD_DISK: |
6633 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
6634 | goto done_unlock; | |
1da177e4 | 6635 | |
c02c0aeb N |
6636 | case RUN_ARRAY: |
6637 | err = do_md_run(mddev); | |
6638 | goto done_unlock; | |
1da177e4 | 6639 | |
c02c0aeb N |
6640 | case SET_BITMAP_FILE: |
6641 | err = set_bitmap_file(mddev, (int)arg); | |
6642 | goto done_unlock; | |
32a7627c | 6643 | |
c02c0aeb N |
6644 | default: |
6645 | err = -EINVAL; | |
6646 | goto abort_unlock; | |
1da177e4 LT |
6647 | } |
6648 | ||
6649 | done_unlock: | |
6650 | abort_unlock: | |
d3374825 N |
6651 | if (mddev->hold_active == UNTIL_IOCTL && |
6652 | err != -EINVAL) | |
6653 | mddev->hold_active = 0; | |
1da177e4 LT |
6654 | mddev_unlock(mddev); |
6655 | ||
6656 | return err; | |
6657 | done: | |
6658 | if (err) | |
6659 | MD_BUG(); | |
6660 | abort: | |
6661 | return err; | |
6662 | } | |
aa98aa31 AB |
6663 | #ifdef CONFIG_COMPAT |
6664 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6665 | unsigned int cmd, unsigned long arg) | |
6666 | { | |
6667 | switch (cmd) { | |
6668 | case HOT_REMOVE_DISK: | |
6669 | case HOT_ADD_DISK: | |
6670 | case SET_DISK_FAULTY: | |
6671 | case SET_BITMAP_FILE: | |
6672 | /* These take in integer arg, do not convert */ | |
6673 | break; | |
6674 | default: | |
6675 | arg = (unsigned long)compat_ptr(arg); | |
6676 | break; | |
6677 | } | |
6678 | ||
6679 | return md_ioctl(bdev, mode, cmd, arg); | |
6680 | } | |
6681 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6682 | |
a39907fa | 6683 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6684 | { |
6685 | /* | |
6686 | * Succeed if we can lock the mddev, which confirms that | |
6687 | * it isn't being stopped right now. | |
6688 | */ | |
fd01b88c | 6689 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6690 | int err; |
6691 | ||
0c098220 YL |
6692 | if (!mddev) |
6693 | return -ENODEV; | |
6694 | ||
d3374825 N |
6695 | if (mddev->gendisk != bdev->bd_disk) { |
6696 | /* we are racing with mddev_put which is discarding this | |
6697 | * bd_disk. | |
6698 | */ | |
6699 | mddev_put(mddev); | |
6700 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6701 | flush_workqueue(md_misc_wq); |
d3374825 N |
6702 | /* Then retry the open from the top */ |
6703 | return -ERESTARTSYS; | |
6704 | } | |
6705 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6706 | ||
c8c00a69 | 6707 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6708 | goto out; |
6709 | ||
6710 | err = 0; | |
f2ea68cf | 6711 | atomic_inc(&mddev->openers); |
260fa034 | 6712 | clear_bit(MD_STILL_CLOSED, &mddev->flags); |
c8c00a69 | 6713 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6714 | |
f0b4f7e2 | 6715 | check_disk_change(bdev); |
1da177e4 LT |
6716 | out: |
6717 | return err; | |
6718 | } | |
6719 | ||
db2a144b | 6720 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6721 | { |
fd01b88c | 6722 | struct mddev *mddev = disk->private_data; |
1da177e4 | 6723 | |
52e5f9d1 | 6724 | BUG_ON(!mddev); |
f2ea68cf | 6725 | atomic_dec(&mddev->openers); |
1da177e4 | 6726 | mddev_put(mddev); |
1da177e4 | 6727 | } |
f0b4f7e2 N |
6728 | |
6729 | static int md_media_changed(struct gendisk *disk) | |
6730 | { | |
fd01b88c | 6731 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6732 | |
6733 | return mddev->changed; | |
6734 | } | |
6735 | ||
6736 | static int md_revalidate(struct gendisk *disk) | |
6737 | { | |
fd01b88c | 6738 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6739 | |
6740 | mddev->changed = 0; | |
6741 | return 0; | |
6742 | } | |
83d5cde4 | 6743 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6744 | { |
6745 | .owner = THIS_MODULE, | |
a39907fa AV |
6746 | .open = md_open, |
6747 | .release = md_release, | |
b492b852 | 6748 | .ioctl = md_ioctl, |
aa98aa31 AB |
6749 | #ifdef CONFIG_COMPAT |
6750 | .compat_ioctl = md_compat_ioctl, | |
6751 | #endif | |
a885c8c4 | 6752 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6753 | .media_changed = md_media_changed, |
6754 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6755 | }; |
6756 | ||
75c96f85 | 6757 | static int md_thread(void * arg) |
1da177e4 | 6758 | { |
2b8bf345 | 6759 | struct md_thread *thread = arg; |
1da177e4 | 6760 | |
1da177e4 LT |
6761 | /* |
6762 | * md_thread is a 'system-thread', it's priority should be very | |
6763 | * high. We avoid resource deadlocks individually in each | |
6764 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6765 | * the very same RT priority as kswapd, thus we will never get | |
6766 | * into a priority inversion deadlock. | |
6767 | * | |
6768 | * we definitely have to have equal or higher priority than | |
6769 | * bdflush, otherwise bdflush will deadlock if there are too | |
6770 | * many dirty RAID5 blocks. | |
6771 | */ | |
1da177e4 | 6772 | |
6985c43f | 6773 | allow_signal(SIGKILL); |
a6fb0934 | 6774 | while (!kthread_should_stop()) { |
1da177e4 | 6775 | |
93588e22 N |
6776 | /* We need to wait INTERRUPTIBLE so that |
6777 | * we don't add to the load-average. | |
6778 | * That means we need to be sure no signals are | |
6779 | * pending | |
6780 | */ | |
6781 | if (signal_pending(current)) | |
6782 | flush_signals(current); | |
6783 | ||
6784 | wait_event_interruptible_timeout | |
6785 | (thread->wqueue, | |
6786 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6787 | || kthread_should_stop(), | |
6788 | thread->timeout); | |
1da177e4 | 6789 | |
6c987910 N |
6790 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6791 | if (!kthread_should_stop()) | |
4ed8731d | 6792 | thread->run(thread); |
1da177e4 | 6793 | } |
a6fb0934 | 6794 | |
1da177e4 LT |
6795 | return 0; |
6796 | } | |
6797 | ||
2b8bf345 | 6798 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
6799 | { |
6800 | if (thread) { | |
36a4e1fe | 6801 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
6802 | set_bit(THREAD_WAKEUP, &thread->flags); |
6803 | wake_up(&thread->wqueue); | |
6804 | } | |
6805 | } | |
6806 | ||
4ed8731d SL |
6807 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
6808 | struct mddev *mddev, const char *name) | |
1da177e4 | 6809 | { |
2b8bf345 | 6810 | struct md_thread *thread; |
1da177e4 | 6811 | |
2b8bf345 | 6812 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
6813 | if (!thread) |
6814 | return NULL; | |
6815 | ||
1da177e4 LT |
6816 | init_waitqueue_head(&thread->wqueue); |
6817 | ||
1da177e4 LT |
6818 | thread->run = run; |
6819 | thread->mddev = mddev; | |
32a7627c | 6820 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6821 | thread->tsk = kthread_run(md_thread, thread, |
6822 | "%s_%s", | |
6823 | mdname(thread->mddev), | |
0232605d | 6824 | name); |
a6fb0934 | 6825 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6826 | kfree(thread); |
6827 | return NULL; | |
6828 | } | |
1da177e4 LT |
6829 | return thread; |
6830 | } | |
6831 | ||
2b8bf345 | 6832 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 6833 | { |
2b8bf345 | 6834 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
6835 | if (!thread) |
6836 | return; | |
36a4e1fe | 6837 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
6838 | /* Locking ensures that mddev_unlock does not wake_up a |
6839 | * non-existent thread | |
6840 | */ | |
6841 | spin_lock(&pers_lock); | |
6842 | *threadp = NULL; | |
6843 | spin_unlock(&pers_lock); | |
a6fb0934 N |
6844 | |
6845 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6846 | kfree(thread); |
6847 | } | |
6848 | ||
fd01b88c | 6849 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
6850 | { |
6851 | if (!mddev) { | |
6852 | MD_BUG(); | |
6853 | return; | |
6854 | } | |
6855 | ||
b2d444d7 | 6856 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6857 | return; |
6bfe0b49 | 6858 | |
de393cde | 6859 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
6860 | return; |
6861 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6862 | if (mddev->degraded) |
6863 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6864 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6865 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6866 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6867 | md_wakeup_thread(mddev->thread); | |
768a418d | 6868 | if (mddev->event_work.func) |
e804ac78 | 6869 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6870 | md_new_event_inintr(mddev); |
1da177e4 LT |
6871 | } |
6872 | ||
6873 | /* seq_file implementation /proc/mdstat */ | |
6874 | ||
6875 | static void status_unused(struct seq_file *seq) | |
6876 | { | |
6877 | int i = 0; | |
3cb03002 | 6878 | struct md_rdev *rdev; |
1da177e4 LT |
6879 | |
6880 | seq_printf(seq, "unused devices: "); | |
6881 | ||
159ec1fc | 6882 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6883 | char b[BDEVNAME_SIZE]; |
6884 | i++; | |
6885 | seq_printf(seq, "%s ", | |
6886 | bdevname(rdev->bdev,b)); | |
6887 | } | |
6888 | if (!i) | |
6889 | seq_printf(seq, "<none>"); | |
6890 | ||
6891 | seq_printf(seq, "\n"); | |
6892 | } | |
6893 | ||
6894 | ||
fd01b88c | 6895 | static void status_resync(struct seq_file *seq, struct mddev * mddev) |
1da177e4 | 6896 | { |
dd71cf6b N |
6897 | sector_t max_sectors, resync, res; |
6898 | unsigned long dt, db; | |
6899 | sector_t rt; | |
4588b42e N |
6900 | int scale; |
6901 | unsigned int per_milli; | |
1da177e4 | 6902 | |
72f36d59 N |
6903 | if (mddev->curr_resync <= 3) |
6904 | resync = 0; | |
6905 | else | |
6906 | resync = mddev->curr_resync | |
6907 | - atomic_read(&mddev->recovery_active); | |
1da177e4 | 6908 | |
c804cdec N |
6909 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
6910 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 6911 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6912 | else |
dd71cf6b | 6913 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6914 | |
6915 | /* | |
6916 | * Should not happen. | |
6917 | */ | |
dd71cf6b | 6918 | if (!max_sectors) { |
1da177e4 LT |
6919 | MD_BUG(); |
6920 | return; | |
6921 | } | |
4588b42e | 6922 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6923 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6924 | * u32, as those are the requirements for sector_div. |
6925 | * Thus 'scale' must be at least 10 | |
6926 | */ | |
6927 | scale = 10; | |
6928 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6929 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6930 | scale++; |
6931 | } | |
6932 | res = (resync>>scale)*1000; | |
dd71cf6b | 6933 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6934 | |
6935 | per_milli = res; | |
1da177e4 | 6936 | { |
4588b42e | 6937 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6938 | seq_printf(seq, "["); |
6939 | for (i = 0; i < x; i++) | |
6940 | seq_printf(seq, "="); | |
6941 | seq_printf(seq, ">"); | |
6942 | for (i = 0; i < y; i++) | |
6943 | seq_printf(seq, "."); | |
6944 | seq_printf(seq, "] "); | |
6945 | } | |
4588b42e | 6946 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6947 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6948 | "reshape" : | |
61df9d91 N |
6949 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6950 | "check" : | |
6951 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6952 | "resync" : "recovery"))), | |
6953 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6954 | (unsigned long long) resync/2, |
6955 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6956 | |
6957 | /* | |
1da177e4 LT |
6958 | * dt: time from mark until now |
6959 | * db: blocks written from mark until now | |
6960 | * rt: remaining time | |
dd71cf6b N |
6961 | * |
6962 | * rt is a sector_t, so could be 32bit or 64bit. | |
6963 | * So we divide before multiply in case it is 32bit and close | |
6964 | * to the limit. | |
25985edc | 6965 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
6966 | * near the end of resync when the number of remaining sectors |
6967 | * is close to 'db'. | |
6968 | * We then divide rt by 32 after multiplying by db to compensate. | |
6969 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6970 | */ |
6971 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6972 | if (!dt) dt++; | |
ff4e8d9a N |
6973 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6974 | - mddev->resync_mark_cnt; | |
1da177e4 | 6975 | |
dd71cf6b N |
6976 | rt = max_sectors - resync; /* number of remaining sectors */ |
6977 | sector_div(rt, db/32+1); | |
6978 | rt *= dt; | |
6979 | rt >>= 5; | |
6980 | ||
6981 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6982 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6983 | |
ff4e8d9a | 6984 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6985 | } |
6986 | ||
6987 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6988 | { | |
6989 | struct list_head *tmp; | |
6990 | loff_t l = *pos; | |
fd01b88c | 6991 | struct mddev *mddev; |
1da177e4 LT |
6992 | |
6993 | if (l >= 0x10000) | |
6994 | return NULL; | |
6995 | if (!l--) | |
6996 | /* header */ | |
6997 | return (void*)1; | |
6998 | ||
6999 | spin_lock(&all_mddevs_lock); | |
7000 | list_for_each(tmp,&all_mddevs) | |
7001 | if (!l--) { | |
fd01b88c | 7002 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7003 | mddev_get(mddev); |
7004 | spin_unlock(&all_mddevs_lock); | |
7005 | return mddev; | |
7006 | } | |
7007 | spin_unlock(&all_mddevs_lock); | |
7008 | if (!l--) | |
7009 | return (void*)2;/* tail */ | |
7010 | return NULL; | |
7011 | } | |
7012 | ||
7013 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7014 | { | |
7015 | struct list_head *tmp; | |
fd01b88c | 7016 | struct mddev *next_mddev, *mddev = v; |
1da177e4 LT |
7017 | |
7018 | ++*pos; | |
7019 | if (v == (void*)2) | |
7020 | return NULL; | |
7021 | ||
7022 | spin_lock(&all_mddevs_lock); | |
7023 | if (v == (void*)1) | |
7024 | tmp = all_mddevs.next; | |
7025 | else | |
7026 | tmp = mddev->all_mddevs.next; | |
7027 | if (tmp != &all_mddevs) | |
fd01b88c | 7028 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7029 | else { |
7030 | next_mddev = (void*)2; | |
7031 | *pos = 0x10000; | |
7032 | } | |
7033 | spin_unlock(&all_mddevs_lock); | |
7034 | ||
7035 | if (v != (void*)1) | |
7036 | mddev_put(mddev); | |
7037 | return next_mddev; | |
7038 | ||
7039 | } | |
7040 | ||
7041 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7042 | { | |
fd01b88c | 7043 | struct mddev *mddev = v; |
1da177e4 LT |
7044 | |
7045 | if (mddev && v != (void*)1 && v != (void*)2) | |
7046 | mddev_put(mddev); | |
7047 | } | |
7048 | ||
7049 | static int md_seq_show(struct seq_file *seq, void *v) | |
7050 | { | |
fd01b88c | 7051 | struct mddev *mddev = v; |
dd8ac336 | 7052 | sector_t sectors; |
3cb03002 | 7053 | struct md_rdev *rdev; |
1da177e4 LT |
7054 | |
7055 | if (v == (void*)1) { | |
84fc4b56 | 7056 | struct md_personality *pers; |
1da177e4 LT |
7057 | seq_printf(seq, "Personalities : "); |
7058 | spin_lock(&pers_lock); | |
2604b703 N |
7059 | list_for_each_entry(pers, &pers_list, list) |
7060 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7061 | |
7062 | spin_unlock(&pers_lock); | |
7063 | seq_printf(seq, "\n"); | |
f1514638 | 7064 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7065 | return 0; |
7066 | } | |
7067 | if (v == (void*)2) { | |
7068 | status_unused(seq); | |
7069 | return 0; | |
7070 | } | |
7071 | ||
5dc5cf7d | 7072 | if (mddev_lock(mddev) < 0) |
1da177e4 | 7073 | return -EINTR; |
5dc5cf7d | 7074 | |
1da177e4 LT |
7075 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7076 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7077 | mddev->pers ? "" : "in"); | |
7078 | if (mddev->pers) { | |
f91de92e | 7079 | if (mddev->ro==1) |
1da177e4 | 7080 | seq_printf(seq, " (read-only)"); |
f91de92e | 7081 | if (mddev->ro==2) |
52720ae7 | 7082 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7083 | seq_printf(seq, " %s", mddev->pers->name); |
7084 | } | |
7085 | ||
dd8ac336 | 7086 | sectors = 0; |
dafb20fa | 7087 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
7088 | char b[BDEVNAME_SIZE]; |
7089 | seq_printf(seq, " %s[%d]", | |
7090 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7091 | if (test_bit(WriteMostly, &rdev->flags)) |
7092 | seq_printf(seq, "(W)"); | |
b2d444d7 | 7093 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7094 | seq_printf(seq, "(F)"); |
7095 | continue; | |
2d78f8c4 N |
7096 | } |
7097 | if (rdev->raid_disk < 0) | |
b325a32e | 7098 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7099 | if (test_bit(Replacement, &rdev->flags)) |
7100 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7101 | sectors += rdev->sectors; |
1da177e4 LT |
7102 | } |
7103 | ||
7104 | if (!list_empty(&mddev->disks)) { | |
7105 | if (mddev->pers) | |
7106 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7107 | (unsigned long long) |
7108 | mddev->array_sectors / 2); | |
1da177e4 LT |
7109 | else |
7110 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7111 | (unsigned long long)sectors / 2); |
1da177e4 | 7112 | } |
1cd6bf19 N |
7113 | if (mddev->persistent) { |
7114 | if (mddev->major_version != 0 || | |
7115 | mddev->minor_version != 90) { | |
7116 | seq_printf(seq," super %d.%d", | |
7117 | mddev->major_version, | |
7118 | mddev->minor_version); | |
7119 | } | |
e691063a N |
7120 | } else if (mddev->external) |
7121 | seq_printf(seq, " super external:%s", | |
7122 | mddev->metadata_type); | |
7123 | else | |
1cd6bf19 | 7124 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7125 | |
7126 | if (mddev->pers) { | |
d710e138 | 7127 | mddev->pers->status(seq, mddev); |
1da177e4 | 7128 | seq_printf(seq, "\n "); |
8e1b39d6 N |
7129 | if (mddev->pers->sync_request) { |
7130 | if (mddev->curr_resync > 2) { | |
d710e138 | 7131 | status_resync(seq, mddev); |
8e1b39d6 | 7132 | seq_printf(seq, "\n "); |
72f36d59 | 7133 | } else if (mddev->curr_resync >= 1) |
8e1b39d6 N |
7134 | seq_printf(seq, "\tresync=DELAYED\n "); |
7135 | else if (mddev->recovery_cp < MaxSector) | |
7136 | seq_printf(seq, "\tresync=PENDING\n "); | |
7137 | } | |
32a7627c N |
7138 | } else |
7139 | seq_printf(seq, "\n "); | |
7140 | ||
57148964 | 7141 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7142 | |
7143 | seq_printf(seq, "\n"); | |
7144 | } | |
7145 | mddev_unlock(mddev); | |
7146 | ||
7147 | return 0; | |
7148 | } | |
7149 | ||
110518bc | 7150 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7151 | .start = md_seq_start, |
7152 | .next = md_seq_next, | |
7153 | .stop = md_seq_stop, | |
7154 | .show = md_seq_show, | |
7155 | }; | |
7156 | ||
7157 | static int md_seq_open(struct inode *inode, struct file *file) | |
7158 | { | |
f1514638 | 7159 | struct seq_file *seq; |
1da177e4 LT |
7160 | int error; |
7161 | ||
7162 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7163 | if (error) |
f1514638 KS |
7164 | return error; |
7165 | ||
7166 | seq = file->private_data; | |
7167 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7168 | return error; |
7169 | } | |
7170 | ||
e2f23b60 | 7171 | static int md_unloading; |
d7603b7e N |
7172 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7173 | { | |
f1514638 | 7174 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7175 | int mask; |
7176 | ||
e2f23b60 N |
7177 | if (md_unloading) |
7178 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI;; | |
d7603b7e N |
7179 | poll_wait(filp, &md_event_waiters, wait); |
7180 | ||
7181 | /* always allow read */ | |
7182 | mask = POLLIN | POLLRDNORM; | |
7183 | ||
f1514638 | 7184 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7185 | mask |= POLLERR | POLLPRI; |
7186 | return mask; | |
7187 | } | |
7188 | ||
fa027c2a | 7189 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7190 | .owner = THIS_MODULE, |
1da177e4 LT |
7191 | .open = md_seq_open, |
7192 | .read = seq_read, | |
7193 | .llseek = seq_lseek, | |
c3f94b40 | 7194 | .release = seq_release_private, |
d7603b7e | 7195 | .poll = mdstat_poll, |
1da177e4 LT |
7196 | }; |
7197 | ||
84fc4b56 | 7198 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7199 | { |
1da177e4 | 7200 | spin_lock(&pers_lock); |
2604b703 N |
7201 | list_add_tail(&p->list, &pers_list); |
7202 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
7203 | spin_unlock(&pers_lock); |
7204 | return 0; | |
7205 | } | |
7206 | ||
84fc4b56 | 7207 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7208 | { |
2604b703 | 7209 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 7210 | spin_lock(&pers_lock); |
2604b703 | 7211 | list_del_init(&p->list); |
1da177e4 LT |
7212 | spin_unlock(&pers_lock); |
7213 | return 0; | |
7214 | } | |
7215 | ||
fd01b88c | 7216 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7217 | { |
3cb03002 | 7218 | struct md_rdev * rdev; |
1da177e4 | 7219 | int idle; |
eea1bf38 | 7220 | int curr_events; |
1da177e4 LT |
7221 | |
7222 | idle = 1; | |
4b80991c N |
7223 | rcu_read_lock(); |
7224 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7225 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7226 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7227 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7228 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7229 | /* sync IO will cause sync_io to increase before the disk_stats |
7230 | * as sync_io is counted when a request starts, and | |
7231 | * disk_stats is counted when it completes. | |
7232 | * So resync activity will cause curr_events to be smaller than | |
7233 | * when there was no such activity. | |
7234 | * non-sync IO will cause disk_stat to increase without | |
7235 | * increasing sync_io so curr_events will (eventually) | |
7236 | * be larger than it was before. Once it becomes | |
7237 | * substantially larger, the test below will cause | |
7238 | * the array to appear non-idle, and resync will slow | |
7239 | * down. | |
7240 | * If there is a lot of outstanding resync activity when | |
7241 | * we set last_event to curr_events, then all that activity | |
7242 | * completing might cause the array to appear non-idle | |
7243 | * and resync will be slowed down even though there might | |
7244 | * not have been non-resync activity. This will only | |
7245 | * happen once though. 'last_events' will soon reflect | |
7246 | * the state where there is little or no outstanding | |
7247 | * resync requests, and further resync activity will | |
7248 | * always make curr_events less than last_events. | |
c0e48521 | 7249 | * |
1da177e4 | 7250 | */ |
eea1bf38 | 7251 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7252 | rdev->last_events = curr_events; |
7253 | idle = 0; | |
7254 | } | |
7255 | } | |
4b80991c | 7256 | rcu_read_unlock(); |
1da177e4 LT |
7257 | return idle; |
7258 | } | |
7259 | ||
fd01b88c | 7260 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7261 | { |
7262 | /* another "blocks" (512byte) blocks have been synced */ | |
7263 | atomic_sub(blocks, &mddev->recovery_active); | |
7264 | wake_up(&mddev->recovery_wait); | |
7265 | if (!ok) { | |
dfc70645 | 7266 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7267 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7268 | md_wakeup_thread(mddev->thread); |
7269 | // stop recovery, signal do_sync .... | |
7270 | } | |
7271 | } | |
7272 | ||
7273 | ||
06d91a5f N |
7274 | /* md_write_start(mddev, bi) |
7275 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7276 | * in superblock) before writing, schedule a superblock update |
7277 | * and wait for it to complete. | |
06d91a5f | 7278 | */ |
fd01b88c | 7279 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7280 | { |
0fd62b86 | 7281 | int did_change = 0; |
06d91a5f | 7282 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7283 | return; |
06d91a5f | 7284 | |
f91de92e N |
7285 | BUG_ON(mddev->ro == 1); |
7286 | if (mddev->ro == 2) { | |
7287 | /* need to switch to read/write */ | |
7288 | mddev->ro = 0; | |
7289 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7290 | md_wakeup_thread(mddev->thread); | |
25156198 | 7291 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7292 | did_change = 1; |
f91de92e | 7293 | } |
06d91a5f | 7294 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7295 | if (mddev->safemode == 1) |
7296 | mddev->safemode = 0; | |
06d91a5f | 7297 | if (mddev->in_sync) { |
a9701a30 | 7298 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
7299 | if (mddev->in_sync) { |
7300 | mddev->in_sync = 0; | |
850b2b42 | 7301 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 7302 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 7303 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7304 | did_change = 1; |
3d310eb7 | 7305 | } |
a9701a30 | 7306 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 7307 | } |
0fd62b86 | 7308 | if (did_change) |
00bcb4ac | 7309 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7310 | wait_event(mddev->sb_wait, |
09a44cc1 | 7311 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 LT |
7312 | } |
7313 | ||
fd01b88c | 7314 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7315 | { |
7316 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7317 | if (mddev->safemode == 2) | |
7318 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7319 | else if (mddev->safemode_delay) |
1da177e4 LT |
7320 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7321 | } | |
7322 | } | |
7323 | ||
2a2275d6 N |
7324 | /* md_allow_write(mddev) |
7325 | * Calling this ensures that the array is marked 'active' so that writes | |
7326 | * may proceed without blocking. It is important to call this before | |
7327 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7328 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
7329 | * |
7330 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
7331 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 7332 | */ |
fd01b88c | 7333 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7334 | { |
7335 | if (!mddev->pers) | |
b5470dc5 | 7336 | return 0; |
2a2275d6 | 7337 | if (mddev->ro) |
b5470dc5 | 7338 | return 0; |
1a0fd497 | 7339 | if (!mddev->pers->sync_request) |
b5470dc5 | 7340 | return 0; |
2a2275d6 N |
7341 | |
7342 | spin_lock_irq(&mddev->write_lock); | |
7343 | if (mddev->in_sync) { | |
7344 | mddev->in_sync = 0; | |
7345 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 7346 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
7347 | if (mddev->safemode_delay && |
7348 | mddev->safemode == 0) | |
7349 | mddev->safemode = 1; | |
7350 | spin_unlock_irq(&mddev->write_lock); | |
7351 | md_update_sb(mddev, 0); | |
00bcb4ac | 7352 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
7353 | } else |
7354 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 | 7355 | |
070dc6dd | 7356 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
7357 | return -EAGAIN; |
7358 | else | |
7359 | return 0; | |
2a2275d6 N |
7360 | } |
7361 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7362 | ||
1da177e4 LT |
7363 | #define SYNC_MARKS 10 |
7364 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7365 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7366 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7367 | { |
4ed8731d | 7368 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7369 | struct mddev *mddev2; |
1da177e4 LT |
7370 | unsigned int currspeed = 0, |
7371 | window; | |
ac7e50a3 | 7372 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 7373 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7374 | unsigned long update_time; |
1da177e4 LT |
7375 | sector_t mark_cnt[SYNC_MARKS]; |
7376 | int last_mark,m; | |
7377 | struct list_head *tmp; | |
7378 | sector_t last_check; | |
57afd89f | 7379 | int skipped = 0; |
3cb03002 | 7380 | struct md_rdev *rdev; |
c4a39551 | 7381 | char *desc, *action = NULL; |
7c2c57c9 | 7382 | struct blk_plug plug; |
1da177e4 LT |
7383 | |
7384 | /* just incase thread restarts... */ | |
7385 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7386 | return; | |
3991b31e N |
7387 | if (mddev->ro) {/* never try to sync a read-only array */ |
7388 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 7389 | return; |
3991b31e | 7390 | } |
1da177e4 | 7391 | |
61df9d91 | 7392 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 7393 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 7394 | desc = "data-check"; |
c4a39551 JB |
7395 | action = "check"; |
7396 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 7397 | desc = "requested-resync"; |
c4a39551 JB |
7398 | action = "repair"; |
7399 | } else | |
61df9d91 N |
7400 | desc = "resync"; |
7401 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7402 | desc = "reshape"; | |
7403 | else | |
7404 | desc = "recovery"; | |
7405 | ||
c4a39551 JB |
7406 | mddev->last_sync_action = action ?: desc; |
7407 | ||
1da177e4 LT |
7408 | /* we overload curr_resync somewhat here. |
7409 | * 0 == not engaged in resync at all | |
7410 | * 2 == checking that there is no conflict with another sync | |
7411 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7412 | * commense | |
7413 | * other == active in resync - this many blocks | |
7414 | * | |
7415 | * Before starting a resync we must have set curr_resync to | |
7416 | * 2, and then checked that every "conflicting" array has curr_resync | |
7417 | * less than ours. When we find one that is the same or higher | |
7418 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7419 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7420 | * This will mean we have to start checking from the beginning again. | |
7421 | * | |
7422 | */ | |
7423 | ||
7424 | do { | |
7425 | mddev->curr_resync = 2; | |
7426 | ||
7427 | try_again: | |
404e4b43 | 7428 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 7429 | goto skip; |
29ac4aa3 | 7430 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7431 | if (mddev2 == mddev) |
7432 | continue; | |
90b08710 BS |
7433 | if (!mddev->parallel_resync |
7434 | && mddev2->curr_resync | |
7435 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7436 | DEFINE_WAIT(wq); |
7437 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7438 | /* arbitrarily yield */ | |
7439 | mddev->curr_resync = 1; | |
7440 | wake_up(&resync_wait); | |
7441 | } | |
7442 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7443 | /* no need to wait here, we can wait the next | |
7444 | * time 'round when curr_resync == 2 | |
7445 | */ | |
7446 | continue; | |
9744197c N |
7447 | /* We need to wait 'interruptible' so as not to |
7448 | * contribute to the load average, and not to | |
7449 | * be caught by 'softlockup' | |
7450 | */ | |
7451 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 7452 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 7453 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
7454 | printk(KERN_INFO "md: delaying %s of %s" |
7455 | " until %s has finished (they" | |
1da177e4 | 7456 | " share one or more physical units)\n", |
61df9d91 | 7457 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 7458 | mddev_put(mddev2); |
9744197c N |
7459 | if (signal_pending(current)) |
7460 | flush_signals(current); | |
1da177e4 LT |
7461 | schedule(); |
7462 | finish_wait(&resync_wait, &wq); | |
7463 | goto try_again; | |
7464 | } | |
7465 | finish_wait(&resync_wait, &wq); | |
7466 | } | |
7467 | } | |
7468 | } while (mddev->curr_resync < 2); | |
7469 | ||
5fd6c1dc | 7470 | j = 0; |
9d88883e | 7471 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7472 | /* resync follows the size requested by the personality, |
57afd89f | 7473 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7474 | */ |
7475 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7476 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7477 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7478 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7479 | j = mddev->resync_min; | |
7480 | else if (!mddev->bitmap) | |
5fd6c1dc | 7481 | j = mddev->recovery_cp; |
5e96ee65 | 7482 | |
ccfcc3c1 | 7483 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7484 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7485 | else { |
1da177e4 | 7486 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7487 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7488 | j = MaxSector; |
4e59ca7d | 7489 | rcu_read_lock(); |
dafb20fa | 7490 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc N |
7491 | if (rdev->raid_disk >= 0 && |
7492 | !test_bit(Faulty, &rdev->flags) && | |
7493 | !test_bit(In_sync, &rdev->flags) && | |
7494 | rdev->recovery_offset < j) | |
7495 | j = rdev->recovery_offset; | |
4e59ca7d | 7496 | rcu_read_unlock(); |
133d4527 N |
7497 | |
7498 | /* If there is a bitmap, we need to make sure all | |
7499 | * writes that started before we added a spare | |
7500 | * complete before we start doing a recovery. | |
7501 | * Otherwise the write might complete and (via | |
7502 | * bitmap_endwrite) set a bit in the bitmap after the | |
7503 | * recovery has checked that bit and skipped that | |
7504 | * region. | |
7505 | */ | |
7506 | if (mddev->bitmap) { | |
7507 | mddev->pers->quiesce(mddev, 1); | |
7508 | mddev->pers->quiesce(mddev, 0); | |
7509 | } | |
5fd6c1dc | 7510 | } |
1da177e4 | 7511 | |
61df9d91 N |
7512 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
7513 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
7514 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 7515 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
7516 | "(but not more than %d KB/sec) for %s.\n", |
7517 | speed_max(mddev), desc); | |
1da177e4 | 7518 | |
eea1bf38 | 7519 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 7520 | |
57afd89f | 7521 | io_sectors = 0; |
1da177e4 LT |
7522 | for (m = 0; m < SYNC_MARKS; m++) { |
7523 | mark[m] = jiffies; | |
57afd89f | 7524 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
7525 | } |
7526 | last_mark = 0; | |
7527 | mddev->resync_mark = mark[last_mark]; | |
7528 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
7529 | ||
7530 | /* | |
7531 | * Tune reconstruction: | |
7532 | */ | |
7533 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
7534 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
7535 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
7536 | |
7537 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
7538 | last_check = 0; |
7539 | ||
7540 | if (j>2) { | |
c91abf5a | 7541 | printk(KERN_INFO |
61df9d91 N |
7542 | "md: resuming %s of %s from checkpoint.\n", |
7543 | desc, mdname(mddev)); | |
1da177e4 | 7544 | mddev->curr_resync = j; |
72f36d59 N |
7545 | } else |
7546 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 7547 | mddev->curr_resync_completed = j; |
72f36d59 N |
7548 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
7549 | md_new_event(mddev); | |
54f89341 | 7550 | update_time = jiffies; |
1da177e4 | 7551 | |
7c2c57c9 | 7552 | blk_start_plug(&plug); |
1da177e4 | 7553 | while (j < max_sectors) { |
57afd89f | 7554 | sector_t sectors; |
1da177e4 | 7555 | |
57afd89f | 7556 | skipped = 0; |
97e4f42d | 7557 | |
7a91ee1f N |
7558 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7559 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
7560 | (mddev->curr_resync - mddev->curr_resync_completed) | |
7561 | > (max_sectors >> 4)) || | |
54f89341 | 7562 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f N |
7563 | (j - mddev->curr_resync_completed)*2 |
7564 | >= mddev->resync_max - mddev->curr_resync_completed | |
7565 | )) { | |
97e4f42d | 7566 | /* time to update curr_resync_completed */ |
97e4f42d N |
7567 | wait_event(mddev->recovery_wait, |
7568 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 7569 | mddev->curr_resync_completed = j; |
35d78c66 | 7570 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
7571 | j > mddev->recovery_cp) | |
7572 | mddev->recovery_cp = j; | |
54f89341 | 7573 | update_time = jiffies; |
070dc6dd | 7574 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 7575 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 7576 | } |
acb180b0 | 7577 | |
c91abf5a N |
7578 | while (j >= mddev->resync_max && |
7579 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
7580 | /* As this condition is controlled by user-space, |
7581 | * we can block indefinitely, so use '_interruptible' | |
7582 | * to avoid triggering warnings. | |
7583 | */ | |
7584 | flush_signals(current); /* just in case */ | |
7585 | wait_event_interruptible(mddev->recovery_wait, | |
7586 | mddev->resync_max > j | |
c91abf5a N |
7587 | || test_bit(MD_RECOVERY_INTR, |
7588 | &mddev->recovery)); | |
e62e58a5 | 7589 | } |
acb180b0 | 7590 | |
c91abf5a N |
7591 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7592 | break; | |
acb180b0 | 7593 | |
57afd89f | 7594 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 7595 | currspeed < speed_min(mddev)); |
57afd89f | 7596 | if (sectors == 0) { |
dfc70645 | 7597 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 7598 | break; |
1da177e4 | 7599 | } |
57afd89f N |
7600 | |
7601 | if (!skipped) { /* actual IO requested */ | |
7602 | io_sectors += sectors; | |
7603 | atomic_add(sectors, &mddev->recovery_active); | |
7604 | } | |
7605 | ||
e875ecea N |
7606 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7607 | break; | |
7608 | ||
1da177e4 | 7609 | j += sectors; |
72f36d59 N |
7610 | if (j > 2) |
7611 | mddev->curr_resync = j; | |
ff4e8d9a | 7612 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 7613 | if (last_check == 0) |
e875ecea | 7614 | /* this is the earliest that rebuild will be |
d7603b7e N |
7615 | * visible in /proc/mdstat |
7616 | */ | |
7617 | md_new_event(mddev); | |
57afd89f N |
7618 | |
7619 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
7620 | continue; |
7621 | ||
57afd89f | 7622 | last_check = io_sectors; |
1da177e4 LT |
7623 | repeat: |
7624 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
7625 | /* step marks */ | |
7626 | int next = (last_mark+1) % SYNC_MARKS; | |
7627 | ||
7628 | mddev->resync_mark = mark[next]; | |
7629 | mddev->resync_mark_cnt = mark_cnt[next]; | |
7630 | mark[next] = jiffies; | |
57afd89f | 7631 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
7632 | last_mark = next; |
7633 | } | |
7634 | ||
c91abf5a N |
7635 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7636 | break; | |
1da177e4 LT |
7637 | |
7638 | /* | |
7639 | * this loop exits only if either when we are slower than | |
7640 | * the 'hard' speed limit, or the system was IO-idle for | |
7641 | * a jiffy. | |
7642 | * the system might be non-idle CPU-wise, but we only care | |
7643 | * about not overloading the IO subsystem. (things like an | |
7644 | * e2fsck being done on the RAID array should execute fast) | |
7645 | */ | |
1da177e4 LT |
7646 | cond_resched(); |
7647 | ||
ac7e50a3 XN |
7648 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
7649 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 7650 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 7651 | |
88202a0c N |
7652 | if (currspeed > speed_min(mddev)) { |
7653 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 7654 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 7655 | msleep(500); |
1da177e4 LT |
7656 | goto repeat; |
7657 | } | |
7658 | } | |
7659 | } | |
c91abf5a N |
7660 | printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc, |
7661 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
7662 | ? "interrupted" : "done"); | |
1da177e4 LT |
7663 | /* |
7664 | * this also signals 'finished resyncing' to md_stop | |
7665 | */ | |
7c2c57c9 | 7666 | blk_finish_plug(&plug); |
1da177e4 LT |
7667 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7668 | ||
7669 | /* tell personality that we are finished */ | |
57afd89f | 7670 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 7671 | |
dfc70645 | 7672 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7673 | mddev->curr_resync > 2) { |
7674 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7675 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7676 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7677 | printk(KERN_INFO | |
61df9d91 N |
7678 | "md: checkpointing %s of %s.\n", |
7679 | desc, mdname(mddev)); | |
0a19caab | 7680 | if (test_bit(MD_RECOVERY_ERROR, |
7681 | &mddev->recovery)) | |
7682 | mddev->recovery_cp = | |
7683 | mddev->curr_resync_completed; | |
7684 | else | |
7685 | mddev->recovery_cp = | |
7686 | mddev->curr_resync; | |
5fd6c1dc N |
7687 | } |
7688 | } else | |
7689 | mddev->recovery_cp = MaxSector; | |
7690 | } else { | |
7691 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7692 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 7693 | rcu_read_lock(); |
dafb20fa | 7694 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7695 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7696 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7697 | !test_bit(Faulty, &rdev->flags) && |
7698 | !test_bit(In_sync, &rdev->flags) && | |
7699 | rdev->recovery_offset < mddev->curr_resync) | |
7700 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7701 | rcu_read_unlock(); |
5fd6c1dc | 7702 | } |
1da177e4 | 7703 | } |
db91ff55 | 7704 | skip: |
17571284 | 7705 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7706 | |
c07b70ad N |
7707 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7708 | /* We completed so min/max setting can be forgotten if used. */ | |
7709 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7710 | mddev->resync_min = 0; | |
7711 | mddev->resync_max = MaxSector; | |
7712 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7713 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 LT |
7714 | mddev->curr_resync = 0; |
7715 | wake_up(&resync_wait); | |
7716 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
7717 | md_wakeup_thread(mddev->thread); | |
c6207277 | 7718 | return; |
1da177e4 | 7719 | } |
29269553 | 7720 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 7721 | |
746d3207 N |
7722 | static int remove_and_add_spares(struct mddev *mddev, |
7723 | struct md_rdev *this) | |
b4c4c7b8 | 7724 | { |
3cb03002 | 7725 | struct md_rdev *rdev; |
b4c4c7b8 | 7726 | int spares = 0; |
f2a371c5 | 7727 | int removed = 0; |
b4c4c7b8 | 7728 | |
dafb20fa | 7729 | rdev_for_each(rdev, mddev) |
746d3207 N |
7730 | if ((this == NULL || rdev == this) && |
7731 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 7732 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7733 | (test_bit(Faulty, &rdev->flags) || |
7734 | ! test_bit(In_sync, &rdev->flags)) && | |
7735 | atomic_read(&rdev->nr_pending)==0) { | |
7736 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 7737 | mddev, rdev) == 0) { |
36fad858 | 7738 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 7739 | rdev->raid_disk = -1; |
f2a371c5 | 7740 | removed++; |
b4c4c7b8 N |
7741 | } |
7742 | } | |
90584fc9 JB |
7743 | if (removed && mddev->kobj.sd) |
7744 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 7745 | |
746d3207 N |
7746 | if (this) |
7747 | goto no_add; | |
7748 | ||
dafb20fa | 7749 | rdev_for_each(rdev, mddev) { |
7bfec5f3 N |
7750 | if (rdev->raid_disk >= 0 && |
7751 | !test_bit(In_sync, &rdev->flags) && | |
7752 | !test_bit(Faulty, &rdev->flags)) | |
7753 | spares++; | |
7ceb17e8 N |
7754 | if (rdev->raid_disk >= 0) |
7755 | continue; | |
7756 | if (test_bit(Faulty, &rdev->flags)) | |
7757 | continue; | |
7758 | if (mddev->ro && | |
8313b8e5 N |
7759 | ! (rdev->saved_raid_disk >= 0 && |
7760 | !test_bit(Bitmap_sync, &rdev->flags))) | |
7ceb17e8 N |
7761 | continue; |
7762 | ||
7eb41885 N |
7763 | if (rdev->saved_raid_disk < 0) |
7764 | rdev->recovery_offset = 0; | |
7ceb17e8 N |
7765 | if (mddev->pers-> |
7766 | hot_add_disk(mddev, rdev) == 0) { | |
7767 | if (sysfs_link_rdev(mddev, rdev)) | |
7768 | /* failure here is OK */; | |
7769 | spares++; | |
7770 | md_new_event(mddev); | |
7771 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
dfc70645 | 7772 | } |
b4c4c7b8 | 7773 | } |
746d3207 | 7774 | no_add: |
6dafab6b N |
7775 | if (removed) |
7776 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 N |
7777 | return spares; |
7778 | } | |
7ebc0be7 | 7779 | |
1da177e4 LT |
7780 | /* |
7781 | * This routine is regularly called by all per-raid-array threads to | |
7782 | * deal with generic issues like resync and super-block update. | |
7783 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7784 | * need this as they never do any recovery or update the superblock. | |
7785 | * | |
7786 | * It does not do any resync itself, but rather "forks" off other threads | |
7787 | * to do that as needed. | |
7788 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7789 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7790 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7791 | * and wakeups up this thread which will reap the thread and finish up. |
7792 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7793 | * | |
7794 | * The overall approach is: | |
7795 | * 1/ if the superblock needs updating, update it. | |
7796 | * 2/ If a recovery thread is running, don't do anything else. | |
7797 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7798 | * 4/ If there are any faulty devices, remove them. | |
7799 | * 5/ If array is degraded, try to add spares devices | |
7800 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7801 | */ | |
fd01b88c | 7802 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 7803 | { |
68866e42 JB |
7804 | if (mddev->suspended) |
7805 | return; | |
7806 | ||
5f40402d | 7807 | if (mddev->bitmap) |
aa5cbd10 | 7808 | bitmap_daemon_work(mddev); |
1da177e4 | 7809 | |
fca4d848 | 7810 | if (signal_pending(current)) { |
31a59e34 | 7811 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7812 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7813 | mdname(mddev)); | |
7814 | mddev->safemode = 2; | |
7815 | } | |
7816 | flush_signals(current); | |
7817 | } | |
7818 | ||
c89a8eee N |
7819 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7820 | return; | |
1da177e4 | 7821 | if ( ! ( |
142d44c3 | 7822 | (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7823 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7824 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7825 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7826 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7827 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7828 | )) |
7829 | return; | |
fca4d848 | 7830 | |
df5b89b3 | 7831 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7832 | int spares = 0; |
fca4d848 | 7833 | |
c89a8eee | 7834 | if (mddev->ro) { |
7ceb17e8 N |
7835 | /* On a read-only array we can: |
7836 | * - remove failed devices | |
7837 | * - add already-in_sync devices if the array itself | |
7838 | * is in-sync. | |
7839 | * As we only add devices that are already in-sync, | |
7840 | * we can activate the spares immediately. | |
c89a8eee | 7841 | */ |
7ceb17e8 | 7842 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
7843 | /* There is no thread, but we need to call |
7844 | * ->spare_active and clear saved_raid_disk | |
7845 | */ | |
2ac295a5 | 7846 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 N |
7847 | md_reap_sync_thread(mddev); |
7848 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
c89a8eee N |
7849 | goto unlock; |
7850 | } | |
7851 | ||
31a59e34 | 7852 | if (!mddev->external) { |
0fd62b86 | 7853 | int did_change = 0; |
31a59e34 N |
7854 | spin_lock_irq(&mddev->write_lock); |
7855 | if (mddev->safemode && | |
7856 | !atomic_read(&mddev->writes_pending) && | |
7857 | !mddev->in_sync && | |
7858 | mddev->recovery_cp == MaxSector) { | |
7859 | mddev->in_sync = 1; | |
0fd62b86 | 7860 | did_change = 1; |
070dc6dd | 7861 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7862 | } |
7863 | if (mddev->safemode == 1) | |
7864 | mddev->safemode = 0; | |
7865 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7866 | if (did_change) |
00bcb4ac | 7867 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7868 | } |
fca4d848 | 7869 | |
7a0a5355 | 7870 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 7871 | md_update_sb(mddev, 0); |
06d91a5f | 7872 | |
1da177e4 LT |
7873 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7874 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7875 | /* resync/recovery still happening */ | |
7876 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7877 | goto unlock; | |
7878 | } | |
7879 | if (mddev->sync_thread) { | |
a91d5ac0 | 7880 | md_reap_sync_thread(mddev); |
1da177e4 LT |
7881 | goto unlock; |
7882 | } | |
72a23c21 NB |
7883 | /* Set RUNNING before clearing NEEDED to avoid |
7884 | * any transients in the value of "sync_action". | |
7885 | */ | |
72f36d59 | 7886 | mddev->curr_resync_completed = 0; |
72a23c21 | 7887 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
24dd469d N |
7888 | /* Clear some bits that don't mean anything, but |
7889 | * might be left set | |
7890 | */ | |
24dd469d N |
7891 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7892 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7893 | |
ed209584 N |
7894 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
7895 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
5fd6c1dc | 7896 | goto unlock; |
1da177e4 LT |
7897 | /* no recovery is running. |
7898 | * remove any failed drives, then | |
7899 | * add spares if possible. | |
72f36d59 | 7900 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
7901 | * the personality to fail the re-add. |
7902 | */ | |
1da177e4 | 7903 | |
b4c4c7b8 | 7904 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7905 | if (mddev->pers->check_reshape == NULL || |
7906 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7907 | /* Cannot proceed */ |
7908 | goto unlock; | |
7909 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7910 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 7911 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
7912 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7913 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7914 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7915 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7916 | } else if (mddev->recovery_cp < MaxSector) { |
7917 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7918 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7919 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7920 | /* nothing to be done ... */ | |
1da177e4 | 7921 | goto unlock; |
24dd469d | 7922 | |
1da177e4 | 7923 | if (mddev->pers->sync_request) { |
ef99bf48 | 7924 | if (spares) { |
a654b9d8 N |
7925 | /* We are adding a device or devices to an array |
7926 | * which has the bitmap stored on all devices. | |
7927 | * So make sure all bitmap pages get written | |
7928 | */ | |
7929 | bitmap_write_all(mddev->bitmap); | |
7930 | } | |
1da177e4 LT |
7931 | mddev->sync_thread = md_register_thread(md_do_sync, |
7932 | mddev, | |
0da3c619 | 7933 | "resync"); |
1da177e4 LT |
7934 | if (!mddev->sync_thread) { |
7935 | printk(KERN_ERR "%s: could not start resync" | |
7936 | " thread...\n", | |
7937 | mdname(mddev)); | |
7938 | /* leave the spares where they are, it shouldn't hurt */ | |
7ebc0be7 N |
7939 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
7940 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7941 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7942 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7943 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
d7603b7e | 7944 | } else |
1da177e4 | 7945 | md_wakeup_thread(mddev->sync_thread); |
00bcb4ac | 7946 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7947 | md_new_event(mddev); |
1da177e4 LT |
7948 | } |
7949 | unlock: | |
90f5f7ad HR |
7950 | wake_up(&mddev->sb_wait); |
7951 | ||
72a23c21 NB |
7952 | if (!mddev->sync_thread) { |
7953 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7954 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7955 | &mddev->recovery)) | |
0c3573f1 | 7956 | if (mddev->sysfs_action) |
00bcb4ac | 7957 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7958 | } |
1da177e4 LT |
7959 | mddev_unlock(mddev); |
7960 | } | |
7961 | } | |
7962 | ||
a91d5ac0 JB |
7963 | void md_reap_sync_thread(struct mddev *mddev) |
7964 | { | |
7965 | struct md_rdev *rdev; | |
7966 | ||
7967 | /* resync has finished, collect result */ | |
7968 | md_unregister_thread(&mddev->sync_thread); | |
30b8feb7 | 7969 | wake_up(&resync_wait); |
a91d5ac0 JB |
7970 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
7971 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
7972 | /* success...*/ | |
7973 | /* activate any spares */ | |
7974 | if (mddev->pers->spare_active(mddev)) { | |
7975 | sysfs_notify(&mddev->kobj, NULL, | |
7976 | "degraded"); | |
7977 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
7978 | } | |
7979 | } | |
7980 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
7981 | mddev->pers->finish_reshape) | |
7982 | mddev->pers->finish_reshape(mddev); | |
7983 | ||
7984 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 7985 | * information must be scrapped. |
a91d5ac0 | 7986 | */ |
f466722c N |
7987 | if (!mddev->degraded) |
7988 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
7989 | rdev->saved_raid_disk = -1; |
7990 | ||
7991 | md_update_sb(mddev, 1); | |
7992 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7993 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7994 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7995 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7996 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7997 | /* flag recovery needed just to double check */ | |
7998 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7999 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8000 | md_new_event(mddev); | |
8001 | if (mddev->event_work.func) | |
8002 | queue_work(md_misc_wq, &mddev->event_work); | |
8003 | } | |
8004 | ||
fd01b88c | 8005 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8006 | { |
00bcb4ac | 8007 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8008 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8009 | !test_bit(Blocked, &rdev->flags) && |
8010 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8011 | msecs_to_jiffies(5000)); |
8012 | rdev_dec_pending(rdev, mddev); | |
8013 | } | |
8014 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8015 | ||
c6563a8c N |
8016 | void md_finish_reshape(struct mddev *mddev) |
8017 | { | |
8018 | /* called be personality module when reshape completes. */ | |
8019 | struct md_rdev *rdev; | |
8020 | ||
8021 | rdev_for_each(rdev, mddev) { | |
8022 | if (rdev->data_offset > rdev->new_data_offset) | |
8023 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8024 | else | |
8025 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8026 | rdev->data_offset = rdev->new_data_offset; | |
8027 | } | |
8028 | } | |
8029 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 N |
8030 | |
8031 | /* Bad block management. | |
8032 | * We can record which blocks on each device are 'bad' and so just | |
8033 | * fail those blocks, or that stripe, rather than the whole device. | |
8034 | * Entries in the bad-block table are 64bits wide. This comprises: | |
8035 | * Length of bad-range, in sectors: 0-511 for lengths 1-512 | |
8036 | * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) | |
8037 | * A 'shift' can be set so that larger blocks are tracked and | |
8038 | * consequently larger devices can be covered. | |
8039 | * 'Acknowledged' flag - 1 bit. - the most significant bit. | |
8040 | * | |
8041 | * Locking of the bad-block table uses a seqlock so md_is_badblock | |
8042 | * might need to retry if it is very unlucky. | |
8043 | * We will sometimes want to check for bad blocks in a bi_end_io function, | |
8044 | * so we use the write_seqlock_irq variant. | |
8045 | * | |
8046 | * When looking for a bad block we specify a range and want to | |
8047 | * know if any block in the range is bad. So we binary-search | |
8048 | * to the last range that starts at-or-before the given endpoint, | |
8049 | * (or "before the sector after the target range") | |
8050 | * then see if it ends after the given start. | |
8051 | * We return | |
8052 | * 0 if there are no known bad blocks in the range | |
8053 | * 1 if there are known bad block which are all acknowledged | |
8054 | * -1 if there are bad blocks which have not yet been acknowledged in metadata. | |
8055 | * plus the start/length of the first bad section we overlap. | |
8056 | */ | |
8057 | int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, | |
8058 | sector_t *first_bad, int *bad_sectors) | |
8059 | { | |
8060 | int hi; | |
ab05613a | 8061 | int lo; |
2230dfe4 | 8062 | u64 *p = bb->page; |
ab05613a | 8063 | int rv; |
2230dfe4 N |
8064 | sector_t target = s + sectors; |
8065 | unsigned seq; | |
8066 | ||
8067 | if (bb->shift > 0) { | |
8068 | /* round the start down, and the end up */ | |
8069 | s >>= bb->shift; | |
8070 | target += (1<<bb->shift) - 1; | |
8071 | target >>= bb->shift; | |
8072 | sectors = target - s; | |
8073 | } | |
8074 | /* 'target' is now the first block after the bad range */ | |
8075 | ||
8076 | retry: | |
8077 | seq = read_seqbegin(&bb->lock); | |
ab05613a | 8078 | lo = 0; |
8079 | rv = 0; | |
2230dfe4 N |
8080 | hi = bb->count; |
8081 | ||
8082 | /* Binary search between lo and hi for 'target' | |
8083 | * i.e. for the last range that starts before 'target' | |
8084 | */ | |
8085 | /* INVARIANT: ranges before 'lo' and at-or-after 'hi' | |
8086 | * are known not to be the last range before target. | |
8087 | * VARIANT: hi-lo is the number of possible | |
8088 | * ranges, and decreases until it reaches 1 | |
8089 | */ | |
8090 | while (hi - lo > 1) { | |
8091 | int mid = (lo + hi) / 2; | |
8092 | sector_t a = BB_OFFSET(p[mid]); | |
8093 | if (a < target) | |
8094 | /* This could still be the one, earlier ranges | |
8095 | * could not. */ | |
8096 | lo = mid; | |
8097 | else | |
8098 | /* This and later ranges are definitely out. */ | |
8099 | hi = mid; | |
8100 | } | |
8101 | /* 'lo' might be the last that started before target, but 'hi' isn't */ | |
8102 | if (hi > lo) { | |
8103 | /* need to check all range that end after 's' to see if | |
8104 | * any are unacknowledged. | |
8105 | */ | |
8106 | while (lo >= 0 && | |
8107 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8108 | if (BB_OFFSET(p[lo]) < target) { | |
8109 | /* starts before the end, and finishes after | |
8110 | * the start, so they must overlap | |
8111 | */ | |
8112 | if (rv != -1 && BB_ACK(p[lo])) | |
8113 | rv = 1; | |
8114 | else | |
8115 | rv = -1; | |
8116 | *first_bad = BB_OFFSET(p[lo]); | |
8117 | *bad_sectors = BB_LEN(p[lo]); | |
8118 | } | |
8119 | lo--; | |
8120 | } | |
8121 | } | |
8122 | ||
8123 | if (read_seqretry(&bb->lock, seq)) | |
8124 | goto retry; | |
8125 | ||
8126 | return rv; | |
8127 | } | |
8128 | EXPORT_SYMBOL_GPL(md_is_badblock); | |
8129 | ||
8130 | /* | |
8131 | * Add a range of bad blocks to the table. | |
8132 | * This might extend the table, or might contract it | |
8133 | * if two adjacent ranges can be merged. | |
8134 | * We binary-search to find the 'insertion' point, then | |
8135 | * decide how best to handle it. | |
8136 | */ | |
8137 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, | |
8138 | int acknowledged) | |
8139 | { | |
8140 | u64 *p; | |
8141 | int lo, hi; | |
8142 | int rv = 1; | |
905b0297 | 8143 | unsigned long flags; |
2230dfe4 N |
8144 | |
8145 | if (bb->shift < 0) | |
8146 | /* badblocks are disabled */ | |
8147 | return 0; | |
8148 | ||
8149 | if (bb->shift) { | |
8150 | /* round the start down, and the end up */ | |
8151 | sector_t next = s + sectors; | |
8152 | s >>= bb->shift; | |
8153 | next += (1<<bb->shift) - 1; | |
8154 | next >>= bb->shift; | |
8155 | sectors = next - s; | |
8156 | } | |
8157 | ||
905b0297 | 8158 | write_seqlock_irqsave(&bb->lock, flags); |
2230dfe4 N |
8159 | |
8160 | p = bb->page; | |
8161 | lo = 0; | |
8162 | hi = bb->count; | |
8163 | /* Find the last range that starts at-or-before 's' */ | |
8164 | while (hi - lo > 1) { | |
8165 | int mid = (lo + hi) / 2; | |
8166 | sector_t a = BB_OFFSET(p[mid]); | |
8167 | if (a <= s) | |
8168 | lo = mid; | |
8169 | else | |
8170 | hi = mid; | |
8171 | } | |
8172 | if (hi > lo && BB_OFFSET(p[lo]) > s) | |
8173 | hi = lo; | |
8174 | ||
8175 | if (hi > lo) { | |
8176 | /* we found a range that might merge with the start | |
8177 | * of our new range | |
8178 | */ | |
8179 | sector_t a = BB_OFFSET(p[lo]); | |
8180 | sector_t e = a + BB_LEN(p[lo]); | |
8181 | int ack = BB_ACK(p[lo]); | |
8182 | if (e >= s) { | |
8183 | /* Yes, we can merge with a previous range */ | |
8184 | if (s == a && s + sectors >= e) | |
8185 | /* new range covers old */ | |
8186 | ack = acknowledged; | |
8187 | else | |
8188 | ack = ack && acknowledged; | |
8189 | ||
8190 | if (e < s + sectors) | |
8191 | e = s + sectors; | |
8192 | if (e - a <= BB_MAX_LEN) { | |
8193 | p[lo] = BB_MAKE(a, e-a, ack); | |
8194 | s = e; | |
8195 | } else { | |
8196 | /* does not all fit in one range, | |
8197 | * make p[lo] maximal | |
8198 | */ | |
8199 | if (BB_LEN(p[lo]) != BB_MAX_LEN) | |
8200 | p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8201 | s = a + BB_MAX_LEN; | |
8202 | } | |
8203 | sectors = e - s; | |
8204 | } | |
8205 | } | |
8206 | if (sectors && hi < bb->count) { | |
8207 | /* 'hi' points to the first range that starts after 's'. | |
8208 | * Maybe we can merge with the start of that range */ | |
8209 | sector_t a = BB_OFFSET(p[hi]); | |
8210 | sector_t e = a + BB_LEN(p[hi]); | |
8211 | int ack = BB_ACK(p[hi]); | |
8212 | if (a <= s + sectors) { | |
8213 | /* merging is possible */ | |
8214 | if (e <= s + sectors) { | |
8215 | /* full overlap */ | |
8216 | e = s + sectors; | |
8217 | ack = acknowledged; | |
8218 | } else | |
8219 | ack = ack && acknowledged; | |
8220 | ||
8221 | a = s; | |
8222 | if (e - a <= BB_MAX_LEN) { | |
8223 | p[hi] = BB_MAKE(a, e-a, ack); | |
8224 | s = e; | |
8225 | } else { | |
8226 | p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8227 | s = a + BB_MAX_LEN; | |
8228 | } | |
8229 | sectors = e - s; | |
8230 | lo = hi; | |
8231 | hi++; | |
8232 | } | |
8233 | } | |
8234 | if (sectors == 0 && hi < bb->count) { | |
8235 | /* we might be able to combine lo and hi */ | |
8236 | /* Note: 's' is at the end of 'lo' */ | |
8237 | sector_t a = BB_OFFSET(p[hi]); | |
8238 | int lolen = BB_LEN(p[lo]); | |
8239 | int hilen = BB_LEN(p[hi]); | |
8240 | int newlen = lolen + hilen - (s - a); | |
8241 | if (s >= a && newlen < BB_MAX_LEN) { | |
8242 | /* yes, we can combine them */ | |
8243 | int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); | |
8244 | p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); | |
8245 | memmove(p + hi, p + hi + 1, | |
8246 | (bb->count - hi - 1) * 8); | |
8247 | bb->count--; | |
8248 | } | |
8249 | } | |
8250 | while (sectors) { | |
8251 | /* didn't merge (it all). | |
8252 | * Need to add a range just before 'hi' */ | |
8253 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8254 | /* No room for more */ | |
8255 | rv = 0; | |
8256 | break; | |
8257 | } else { | |
8258 | int this_sectors = sectors; | |
8259 | memmove(p + hi + 1, p + hi, | |
8260 | (bb->count - hi) * 8); | |
8261 | bb->count++; | |
8262 | ||
8263 | if (this_sectors > BB_MAX_LEN) | |
8264 | this_sectors = BB_MAX_LEN; | |
8265 | p[hi] = BB_MAKE(s, this_sectors, acknowledged); | |
8266 | sectors -= this_sectors; | |
8267 | s += this_sectors; | |
8268 | } | |
8269 | } | |
8270 | ||
8271 | bb->changed = 1; | |
de393cde N |
8272 | if (!acknowledged) |
8273 | bb->unacked_exist = 1; | |
905b0297 | 8274 | write_sequnlock_irqrestore(&bb->lock, flags); |
2230dfe4 N |
8275 | |
8276 | return rv; | |
8277 | } | |
8278 | ||
3cb03002 | 8279 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8280 | int is_new) |
2230dfe4 | 8281 | { |
c6563a8c N |
8282 | int rv; |
8283 | if (is_new) | |
8284 | s += rdev->new_data_offset; | |
8285 | else | |
8286 | s += rdev->data_offset; | |
8287 | rv = md_set_badblocks(&rdev->badblocks, | |
8288 | s, sectors, 0); | |
2230dfe4 N |
8289 | if (rv) { |
8290 | /* Make sure they get written out promptly */ | |
8bd2f0a0 | 8291 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2230dfe4 N |
8292 | set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); |
8293 | md_wakeup_thread(rdev->mddev->thread); | |
8294 | } | |
8295 | return rv; | |
8296 | } | |
8297 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8298 | ||
8299 | /* | |
8300 | * Remove a range of bad blocks from the table. | |
8301 | * This may involve extending the table if we spilt a region, | |
8302 | * but it must not fail. So if the table becomes full, we just | |
8303 | * drop the remove request. | |
8304 | */ | |
8305 | static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) | |
8306 | { | |
8307 | u64 *p; | |
8308 | int lo, hi; | |
8309 | sector_t target = s + sectors; | |
8310 | int rv = 0; | |
8311 | ||
8312 | if (bb->shift > 0) { | |
8313 | /* When clearing we round the start up and the end down. | |
8314 | * This should not matter as the shift should align with | |
8315 | * the block size and no rounding should ever be needed. | |
8316 | * However it is better the think a block is bad when it | |
8317 | * isn't than to think a block is not bad when it is. | |
8318 | */ | |
8319 | s += (1<<bb->shift) - 1; | |
8320 | s >>= bb->shift; | |
8321 | target >>= bb->shift; | |
8322 | sectors = target - s; | |
8323 | } | |
8324 | ||
8325 | write_seqlock_irq(&bb->lock); | |
8326 | ||
8327 | p = bb->page; | |
8328 | lo = 0; | |
8329 | hi = bb->count; | |
8330 | /* Find the last range that starts before 'target' */ | |
8331 | while (hi - lo > 1) { | |
8332 | int mid = (lo + hi) / 2; | |
8333 | sector_t a = BB_OFFSET(p[mid]); | |
8334 | if (a < target) | |
8335 | lo = mid; | |
8336 | else | |
8337 | hi = mid; | |
8338 | } | |
8339 | if (hi > lo) { | |
8340 | /* p[lo] is the last range that could overlap the | |
8341 | * current range. Earlier ranges could also overlap, | |
8342 | * but only this one can overlap the end of the range. | |
8343 | */ | |
8344 | if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { | |
8345 | /* Partial overlap, leave the tail of this range */ | |
8346 | int ack = BB_ACK(p[lo]); | |
8347 | sector_t a = BB_OFFSET(p[lo]); | |
8348 | sector_t end = a + BB_LEN(p[lo]); | |
8349 | ||
8350 | if (a < s) { | |
8351 | /* we need to split this range */ | |
8352 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8b32bf5e | 8353 | rv = -ENOSPC; |
2230dfe4 N |
8354 | goto out; |
8355 | } | |
8356 | memmove(p+lo+1, p+lo, (bb->count - lo) * 8); | |
8357 | bb->count++; | |
8358 | p[lo] = BB_MAKE(a, s-a, ack); | |
8359 | lo++; | |
8360 | } | |
8361 | p[lo] = BB_MAKE(target, end - target, ack); | |
8362 | /* there is no longer an overlap */ | |
8363 | hi = lo; | |
8364 | lo--; | |
8365 | } | |
8366 | while (lo >= 0 && | |
8367 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8368 | /* This range does overlap */ | |
8369 | if (BB_OFFSET(p[lo]) < s) { | |
8370 | /* Keep the early parts of this range. */ | |
8371 | int ack = BB_ACK(p[lo]); | |
8372 | sector_t start = BB_OFFSET(p[lo]); | |
8373 | p[lo] = BB_MAKE(start, s - start, ack); | |
8374 | /* now low doesn't overlap, so.. */ | |
8375 | break; | |
8376 | } | |
8377 | lo--; | |
8378 | } | |
8379 | /* 'lo' is strictly before, 'hi' is strictly after, | |
8380 | * anything between needs to be discarded | |
8381 | */ | |
8382 | if (hi - lo > 1) { | |
8383 | memmove(p+lo+1, p+hi, (bb->count - hi) * 8); | |
8384 | bb->count -= (hi - lo - 1); | |
8385 | } | |
8386 | } | |
8387 | ||
8388 | bb->changed = 1; | |
8389 | out: | |
8390 | write_sequnlock_irq(&bb->lock); | |
8391 | return rv; | |
8392 | } | |
8393 | ||
c6563a8c N |
8394 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8395 | int is_new) | |
2230dfe4 | 8396 | { |
c6563a8c N |
8397 | if (is_new) |
8398 | s += rdev->new_data_offset; | |
8399 | else | |
8400 | s += rdev->data_offset; | |
2230dfe4 | 8401 | return md_clear_badblocks(&rdev->badblocks, |
c6563a8c | 8402 | s, sectors); |
2230dfe4 N |
8403 | } |
8404 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8405 | ||
8406 | /* | |
8407 | * Acknowledge all bad blocks in a list. | |
8408 | * This only succeeds if ->changed is clear. It is used by | |
8409 | * in-kernel metadata updates | |
8410 | */ | |
8411 | void md_ack_all_badblocks(struct badblocks *bb) | |
8412 | { | |
8413 | if (bb->page == NULL || bb->changed) | |
8414 | /* no point even trying */ | |
8415 | return; | |
8416 | write_seqlock_irq(&bb->lock); | |
8417 | ||
ecb178bb | 8418 | if (bb->changed == 0 && bb->unacked_exist) { |
2230dfe4 N |
8419 | u64 *p = bb->page; |
8420 | int i; | |
8421 | for (i = 0; i < bb->count ; i++) { | |
8422 | if (!BB_ACK(p[i])) { | |
8423 | sector_t start = BB_OFFSET(p[i]); | |
8424 | int len = BB_LEN(p[i]); | |
8425 | p[i] = BB_MAKE(start, len, 1); | |
8426 | } | |
8427 | } | |
de393cde | 8428 | bb->unacked_exist = 0; |
2230dfe4 N |
8429 | } |
8430 | write_sequnlock_irq(&bb->lock); | |
8431 | } | |
8432 | EXPORT_SYMBOL_GPL(md_ack_all_badblocks); | |
8433 | ||
16c791a5 N |
8434 | /* sysfs access to bad-blocks list. |
8435 | * We present two files. | |
8436 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
8437 | * are recorded as bad. The list is truncated to fit within | |
8438 | * the one-page limit of sysfs. | |
8439 | * Writing "sector length" to this file adds an acknowledged | |
8440 | * bad block list. | |
8441 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
8442 | * been acknowledged. Writing to this file adds bad blocks | |
8443 | * without acknowledging them. This is largely for testing. | |
8444 | */ | |
8445 | ||
8446 | static ssize_t | |
8447 | badblocks_show(struct badblocks *bb, char *page, int unack) | |
8448 | { | |
8449 | size_t len; | |
8450 | int i; | |
8451 | u64 *p = bb->page; | |
8452 | unsigned seq; | |
8453 | ||
8454 | if (bb->shift < 0) | |
8455 | return 0; | |
8456 | ||
8457 | retry: | |
8458 | seq = read_seqbegin(&bb->lock); | |
8459 | ||
8460 | len = 0; | |
8461 | i = 0; | |
8462 | ||
8463 | while (len < PAGE_SIZE && i < bb->count) { | |
8464 | sector_t s = BB_OFFSET(p[i]); | |
8465 | unsigned int length = BB_LEN(p[i]); | |
8466 | int ack = BB_ACK(p[i]); | |
8467 | i++; | |
8468 | ||
8469 | if (unack && ack) | |
8470 | continue; | |
8471 | ||
8472 | len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", | |
8473 | (unsigned long long)s << bb->shift, | |
8474 | length << bb->shift); | |
8475 | } | |
de393cde N |
8476 | if (unack && len == 0) |
8477 | bb->unacked_exist = 0; | |
16c791a5 N |
8478 | |
8479 | if (read_seqretry(&bb->lock, seq)) | |
8480 | goto retry; | |
8481 | ||
8482 | return len; | |
8483 | } | |
8484 | ||
8485 | #define DO_DEBUG 1 | |
8486 | ||
8487 | static ssize_t | |
8488 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) | |
8489 | { | |
8490 | unsigned long long sector; | |
8491 | int length; | |
8492 | char newline; | |
8493 | #ifdef DO_DEBUG | |
8494 | /* Allow clearing via sysfs *only* for testing/debugging. | |
8495 | * Normally only a successful write may clear a badblock | |
8496 | */ | |
8497 | int clear = 0; | |
8498 | if (page[0] == '-') { | |
8499 | clear = 1; | |
8500 | page++; | |
8501 | } | |
8502 | #endif /* DO_DEBUG */ | |
8503 | ||
8504 | switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { | |
8505 | case 3: | |
8506 | if (newline != '\n') | |
8507 | return -EINVAL; | |
8508 | case 2: | |
8509 | if (length <= 0) | |
8510 | return -EINVAL; | |
8511 | break; | |
8512 | default: | |
8513 | return -EINVAL; | |
8514 | } | |
8515 | ||
8516 | #ifdef DO_DEBUG | |
8517 | if (clear) { | |
8518 | md_clear_badblocks(bb, sector, length); | |
8519 | return len; | |
8520 | } | |
8521 | #endif /* DO_DEBUG */ | |
8522 | if (md_set_badblocks(bb, sector, length, !unack)) | |
8523 | return len; | |
8524 | else | |
8525 | return -ENOSPC; | |
8526 | } | |
8527 | ||
75c96f85 AB |
8528 | static int md_notify_reboot(struct notifier_block *this, |
8529 | unsigned long code, void *x) | |
1da177e4 LT |
8530 | { |
8531 | struct list_head *tmp; | |
fd01b88c | 8532 | struct mddev *mddev; |
2dba6a91 | 8533 | int need_delay = 0; |
1da177e4 | 8534 | |
c744a65c N |
8535 | for_each_mddev(mddev, tmp) { |
8536 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8537 | if (mddev->pers) |
8538 | __md_stop_writes(mddev); | |
0f62fb22 N |
8539 | if (mddev->persistent) |
8540 | mddev->safemode = 2; | |
c744a65c | 8541 | mddev_unlock(mddev); |
2dba6a91 | 8542 | } |
c744a65c | 8543 | need_delay = 1; |
1da177e4 | 8544 | } |
c744a65c N |
8545 | /* |
8546 | * certain more exotic SCSI devices are known to be | |
8547 | * volatile wrt too early system reboots. While the | |
8548 | * right place to handle this issue is the given | |
8549 | * driver, we do want to have a safe RAID driver ... | |
8550 | */ | |
8551 | if (need_delay) | |
8552 | mdelay(1000*1); | |
8553 | ||
1da177e4 LT |
8554 | return NOTIFY_DONE; |
8555 | } | |
8556 | ||
75c96f85 | 8557 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8558 | .notifier_call = md_notify_reboot, |
8559 | .next = NULL, | |
8560 | .priority = INT_MAX, /* before any real devices */ | |
8561 | }; | |
8562 | ||
8563 | static void md_geninit(void) | |
8564 | { | |
36a4e1fe | 8565 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8566 | |
c7705f34 | 8567 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8568 | } |
8569 | ||
75c96f85 | 8570 | static int __init md_init(void) |
1da177e4 | 8571 | { |
e804ac78 TH |
8572 | int ret = -ENOMEM; |
8573 | ||
ada609ee | 8574 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8575 | if (!md_wq) |
8576 | goto err_wq; | |
8577 | ||
8578 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8579 | if (!md_misc_wq) | |
8580 | goto err_misc_wq; | |
8581 | ||
8582 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8583 | goto err_md; | |
8584 | ||
8585 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8586 | goto err_mdp; | |
8587 | mdp_major = ret; | |
8588 | ||
af5628f0 | 8589 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
8590 | md_probe, NULL, NULL); |
8591 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8592 | md_probe, NULL, NULL); |
8593 | ||
1da177e4 | 8594 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8595 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8596 | |
8597 | md_geninit(); | |
d710e138 | 8598 | return 0; |
1da177e4 | 8599 | |
e804ac78 TH |
8600 | err_mdp: |
8601 | unregister_blkdev(MD_MAJOR, "md"); | |
8602 | err_md: | |
8603 | destroy_workqueue(md_misc_wq); | |
8604 | err_misc_wq: | |
8605 | destroy_workqueue(md_wq); | |
8606 | err_wq: | |
8607 | return ret; | |
8608 | } | |
1da177e4 LT |
8609 | |
8610 | #ifndef MODULE | |
8611 | ||
8612 | /* | |
8613 | * Searches all registered partitions for autorun RAID arrays | |
8614 | * at boot time. | |
8615 | */ | |
4d936ec1 ME |
8616 | |
8617 | static LIST_HEAD(all_detected_devices); | |
8618 | struct detected_devices_node { | |
8619 | struct list_head list; | |
8620 | dev_t dev; | |
8621 | }; | |
1da177e4 LT |
8622 | |
8623 | void md_autodetect_dev(dev_t dev) | |
8624 | { | |
4d936ec1 ME |
8625 | struct detected_devices_node *node_detected_dev; |
8626 | ||
8627 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8628 | if (node_detected_dev) { | |
8629 | node_detected_dev->dev = dev; | |
8630 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
8631 | } else { | |
8632 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
8633 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
8634 | } | |
1da177e4 LT |
8635 | } |
8636 | ||
8637 | ||
8638 | static void autostart_arrays(int part) | |
8639 | { | |
3cb03002 | 8640 | struct md_rdev *rdev; |
4d936ec1 ME |
8641 | struct detected_devices_node *node_detected_dev; |
8642 | dev_t dev; | |
8643 | int i_scanned, i_passed; | |
1da177e4 | 8644 | |
4d936ec1 ME |
8645 | i_scanned = 0; |
8646 | i_passed = 0; | |
1da177e4 | 8647 | |
4d936ec1 | 8648 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8649 | |
4d936ec1 ME |
8650 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8651 | i_scanned++; | |
8652 | node_detected_dev = list_entry(all_detected_devices.next, | |
8653 | struct detected_devices_node, list); | |
8654 | list_del(&node_detected_dev->list); | |
8655 | dev = node_detected_dev->dev; | |
8656 | kfree(node_detected_dev); | |
df968c4e | 8657 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
8658 | if (IS_ERR(rdev)) |
8659 | continue; | |
8660 | ||
b2d444d7 | 8661 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
8662 | MD_BUG(); |
8663 | continue; | |
8664 | } | |
d0fae18f | 8665 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 8666 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 8667 | i_passed++; |
1da177e4 | 8668 | } |
4d936ec1 ME |
8669 | |
8670 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
8671 | i_scanned, i_passed); | |
1da177e4 LT |
8672 | |
8673 | autorun_devices(part); | |
8674 | } | |
8675 | ||
fdee8ae4 | 8676 | #endif /* !MODULE */ |
1da177e4 LT |
8677 | |
8678 | static __exit void md_exit(void) | |
8679 | { | |
fd01b88c | 8680 | struct mddev *mddev; |
1da177e4 | 8681 | struct list_head *tmp; |
e2f23b60 | 8682 | int delay = 1; |
8ab5e4c1 | 8683 | |
af5628f0 | 8684 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 8685 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 8686 | |
3dbd8c2e | 8687 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
8688 | unregister_blkdev(mdp_major, "mdp"); |
8689 | unregister_reboot_notifier(&md_notifier); | |
8690 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
8691 | |
8692 | /* We cannot unload the modules while some process is | |
8693 | * waiting for us in select() or poll() - wake them up | |
8694 | */ | |
8695 | md_unloading = 1; | |
8696 | while (waitqueue_active(&md_event_waiters)) { | |
8697 | /* not safe to leave yet */ | |
8698 | wake_up(&md_event_waiters); | |
8699 | msleep(delay); | |
8700 | delay += delay; | |
8701 | } | |
1da177e4 | 8702 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 8703 | |
29ac4aa3 | 8704 | for_each_mddev(mddev, tmp) { |
1da177e4 | 8705 | export_array(mddev); |
d3374825 | 8706 | mddev->hold_active = 0; |
1da177e4 | 8707 | } |
e804ac78 TH |
8708 | destroy_workqueue(md_misc_wq); |
8709 | destroy_workqueue(md_wq); | |
1da177e4 LT |
8710 | } |
8711 | ||
685784aa | 8712 | subsys_initcall(md_init); |
1da177e4 LT |
8713 | module_exit(md_exit) |
8714 | ||
f91de92e N |
8715 | static int get_ro(char *buffer, struct kernel_param *kp) |
8716 | { | |
8717 | return sprintf(buffer, "%d", start_readonly); | |
8718 | } | |
8719 | static int set_ro(const char *val, struct kernel_param *kp) | |
8720 | { | |
8721 | char *e; | |
8722 | int num = simple_strtoul(val, &e, 10); | |
8723 | if (*val && (*e == '\0' || *e == '\n')) { | |
8724 | start_readonly = num; | |
4dbcdc75 | 8725 | return 0; |
f91de92e N |
8726 | } |
8727 | return -EINVAL; | |
8728 | } | |
8729 | ||
80ca3a44 N |
8730 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
8731 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 8732 | |
efeb53c0 | 8733 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 8734 | |
1da177e4 LT |
8735 | EXPORT_SYMBOL(register_md_personality); |
8736 | EXPORT_SYMBOL(unregister_md_personality); | |
8737 | EXPORT_SYMBOL(md_error); | |
8738 | EXPORT_SYMBOL(md_done_sync); | |
8739 | EXPORT_SYMBOL(md_write_start); | |
8740 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
8741 | EXPORT_SYMBOL(md_register_thread); |
8742 | EXPORT_SYMBOL(md_unregister_thread); | |
8743 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 | 8744 | EXPORT_SYMBOL(md_check_recovery); |
a91d5ac0 | 8745 | EXPORT_SYMBOL(md_reap_sync_thread); |
1da177e4 | 8746 | MODULE_LICENSE("GPL"); |
0efb9e61 | 8747 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 8748 | MODULE_ALIAS("md"); |
72008652 | 8749 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |