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 | |
e9c7469b | 396 | if (bio->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; |
7bfa19f2 N |
757 | bio->bi_sector = sector; |
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 */ |
a9701a30 N |
769 | DEFINE_WAIT(wq); |
770 | for(;;) { | |
771 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
772 | if (atomic_read(&mddev->pending_writes)==0) | |
773 | break; | |
a9701a30 N |
774 | schedule(); |
775 | } | |
776 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
777 | } |
778 | ||
6712ecf8 | 779 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 780 | { |
1da177e4 | 781 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
782 | } |
783 | ||
3cb03002 | 784 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
ccebd4c4 | 785 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 786 | { |
a167f663 | 787 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
788 | struct completion event; |
789 | int ret; | |
790 | ||
721a9602 | 791 | rw |= REQ_SYNC; |
1da177e4 | 792 | |
a6ff7e08 JB |
793 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
794 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 JB |
795 | if (metadata_op) |
796 | bio->bi_sector = sector + rdev->sb_start; | |
1fdd6fc9 N |
797 | else if (rdev->mddev->reshape_position != MaxSector && |
798 | (rdev->mddev->reshape_backwards == | |
799 | (sector >= rdev->mddev->reshape_position))) | |
800 | bio->bi_sector = sector + rdev->new_data_offset; | |
ccebd4c4 JB |
801 | else |
802 | bio->bi_sector = sector + rdev->data_offset; | |
1da177e4 LT |
803 | bio_add_page(bio, page, size, 0); |
804 | init_completion(&event); | |
805 | bio->bi_private = &event; | |
806 | bio->bi_end_io = bi_complete; | |
807 | submit_bio(rw, bio); | |
808 | wait_for_completion(&event); | |
809 | ||
810 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
811 | bio_put(bio); | |
812 | return ret; | |
813 | } | |
a8745db2 | 814 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 815 | |
3cb03002 | 816 | static int read_disk_sb(struct md_rdev * rdev, int size) |
1da177e4 LT |
817 | { |
818 | char b[BDEVNAME_SIZE]; | |
819 | if (!rdev->sb_page) { | |
820 | MD_BUG(); | |
821 | return -EINVAL; | |
822 | } | |
823 | if (rdev->sb_loaded) | |
824 | return 0; | |
825 | ||
826 | ||
ccebd4c4 | 827 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
828 | goto fail; |
829 | rdev->sb_loaded = 1; | |
830 | return 0; | |
831 | ||
832 | fail: | |
833 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
834 | bdevname(rdev->bdev,b)); | |
835 | return -EINVAL; | |
836 | } | |
837 | ||
838 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
839 | { | |
05710466 AN |
840 | return sb1->set_uuid0 == sb2->set_uuid0 && |
841 | sb1->set_uuid1 == sb2->set_uuid1 && | |
842 | sb1->set_uuid2 == sb2->set_uuid2 && | |
843 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
844 | } |
845 | ||
1da177e4 LT |
846 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
847 | { | |
848 | int ret; | |
849 | mdp_super_t *tmp1, *tmp2; | |
850 | ||
851 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
852 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
853 | ||
854 | if (!tmp1 || !tmp2) { | |
855 | ret = 0; | |
35020f1a | 856 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
857 | goto abort; |
858 | } | |
859 | ||
860 | *tmp1 = *sb1; | |
861 | *tmp2 = *sb2; | |
862 | ||
863 | /* | |
864 | * nr_disks is not constant | |
865 | */ | |
866 | tmp1->nr_disks = 0; | |
867 | tmp2->nr_disks = 0; | |
868 | ||
ce0c8e05 | 869 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 870 | abort: |
990a8baf JJ |
871 | kfree(tmp1); |
872 | kfree(tmp2); | |
1da177e4 LT |
873 | return ret; |
874 | } | |
875 | ||
4d167f09 N |
876 | |
877 | static u32 md_csum_fold(u32 csum) | |
878 | { | |
879 | csum = (csum & 0xffff) + (csum >> 16); | |
880 | return (csum & 0xffff) + (csum >> 16); | |
881 | } | |
882 | ||
1da177e4 LT |
883 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
884 | { | |
4d167f09 N |
885 | u64 newcsum = 0; |
886 | u32 *sb32 = (u32*)sb; | |
887 | int i; | |
1da177e4 LT |
888 | unsigned int disk_csum, csum; |
889 | ||
890 | disk_csum = sb->sb_csum; | |
891 | sb->sb_csum = 0; | |
4d167f09 N |
892 | |
893 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
894 | newcsum += sb32[i]; | |
895 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
896 | ||
897 | ||
898 | #ifdef CONFIG_ALPHA | |
899 | /* This used to use csum_partial, which was wrong for several | |
900 | * reasons including that different results are returned on | |
901 | * different architectures. It isn't critical that we get exactly | |
902 | * the same return value as before (we always csum_fold before | |
903 | * testing, and that removes any differences). However as we | |
904 | * know that csum_partial always returned a 16bit value on | |
905 | * alphas, do a fold to maximise conformity to previous behaviour. | |
906 | */ | |
907 | sb->sb_csum = md_csum_fold(disk_csum); | |
908 | #else | |
1da177e4 | 909 | sb->sb_csum = disk_csum; |
4d167f09 | 910 | #endif |
1da177e4 LT |
911 | return csum; |
912 | } | |
913 | ||
914 | ||
915 | /* | |
916 | * Handle superblock details. | |
917 | * We want to be able to handle multiple superblock formats | |
918 | * so we have a common interface to them all, and an array of | |
919 | * different handlers. | |
920 | * We rely on user-space to write the initial superblock, and support | |
921 | * reading and updating of superblocks. | |
922 | * Interface methods are: | |
3cb03002 | 923 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
924 | * loads and validates a superblock on dev. |
925 | * if refdev != NULL, compare superblocks on both devices | |
926 | * Return: | |
927 | * 0 - dev has a superblock that is compatible with refdev | |
928 | * 1 - dev has a superblock that is compatible and newer than refdev | |
929 | * so dev should be used as the refdev in future | |
930 | * -EINVAL superblock incompatible or invalid | |
931 | * -othererror e.g. -EIO | |
932 | * | |
fd01b88c | 933 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
934 | * Verify that dev is acceptable into mddev. |
935 | * The first time, mddev->raid_disks will be 0, and data from | |
936 | * dev should be merged in. Subsequent calls check that dev | |
937 | * is new enough. Return 0 or -EINVAL | |
938 | * | |
fd01b88c | 939 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
940 | * Update the superblock for rdev with data in mddev |
941 | * This does not write to disc. | |
942 | * | |
943 | */ | |
944 | ||
945 | struct super_type { | |
0cd17fec CW |
946 | char *name; |
947 | struct module *owner; | |
c6563a8c N |
948 | int (*load_super)(struct md_rdev *rdev, |
949 | struct md_rdev *refdev, | |
0cd17fec | 950 | int minor_version); |
c6563a8c N |
951 | int (*validate_super)(struct mddev *mddev, |
952 | struct md_rdev *rdev); | |
953 | void (*sync_super)(struct mddev *mddev, | |
954 | struct md_rdev *rdev); | |
3cb03002 | 955 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 956 | sector_t num_sectors); |
c6563a8c N |
957 | int (*allow_new_offset)(struct md_rdev *rdev, |
958 | unsigned long long new_offset); | |
1da177e4 LT |
959 | }; |
960 | ||
0894cc30 AN |
961 | /* |
962 | * Check that the given mddev has no bitmap. | |
963 | * | |
964 | * This function is called from the run method of all personalities that do not | |
965 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
966 | * has a bitmap. Otherwise, it returns 0. | |
967 | * | |
968 | */ | |
fd01b88c | 969 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 970 | { |
c3d9714e | 971 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
972 | return 0; |
973 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
974 | mdname(mddev), mddev->pers->name); | |
975 | return 1; | |
976 | } | |
977 | EXPORT_SYMBOL(md_check_no_bitmap); | |
978 | ||
1da177e4 LT |
979 | /* |
980 | * load_super for 0.90.0 | |
981 | */ | |
3cb03002 | 982 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
983 | { |
984 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
985 | mdp_super_t *sb; | |
986 | int ret; | |
1da177e4 LT |
987 | |
988 | /* | |
0f420358 | 989 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
990 | * it's at the end of the disk. |
991 | * | |
992 | * It also happens to be a multiple of 4Kb. | |
993 | */ | |
57b2caa3 | 994 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 995 | |
0002b271 | 996 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
997 | if (ret) return ret; |
998 | ||
999 | ret = -EINVAL; | |
1000 | ||
1001 | bdevname(rdev->bdev, b); | |
65a06f06 | 1002 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1003 | |
1004 | if (sb->md_magic != MD_SB_MAGIC) { | |
1005 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
1006 | b); | |
1007 | goto abort; | |
1008 | } | |
1009 | ||
1010 | if (sb->major_version != 0 || | |
f6705578 N |
1011 | sb->minor_version < 90 || |
1012 | sb->minor_version > 91) { | |
1da177e4 LT |
1013 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
1014 | sb->major_version, sb->minor_version, | |
1015 | b); | |
1016 | goto abort; | |
1017 | } | |
1018 | ||
1019 | if (sb->raid_disks <= 0) | |
1020 | goto abort; | |
1021 | ||
4d167f09 | 1022 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
1023 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
1024 | b); | |
1025 | goto abort; | |
1026 | } | |
1027 | ||
1028 | rdev->preferred_minor = sb->md_minor; | |
1029 | rdev->data_offset = 0; | |
c6563a8c | 1030 | rdev->new_data_offset = 0; |
0002b271 | 1031 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1032 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1033 | |
1034 | if (sb->level == LEVEL_MULTIPATH) | |
1035 | rdev->desc_nr = -1; | |
1036 | else | |
1037 | rdev->desc_nr = sb->this_disk.number; | |
1038 | ||
9a7b2b0f | 1039 | if (!refdev) { |
1da177e4 | 1040 | ret = 1; |
9a7b2b0f | 1041 | } else { |
1da177e4 | 1042 | __u64 ev1, ev2; |
65a06f06 | 1043 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1044 | if (!uuid_equal(refsb, sb)) { |
1045 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1046 | b, bdevname(refdev->bdev,b2)); | |
1047 | goto abort; | |
1048 | } | |
1049 | if (!sb_equal(refsb, sb)) { | |
1050 | printk(KERN_WARNING "md: %s has same UUID" | |
1051 | " but different superblock to %s\n", | |
1052 | b, bdevname(refdev->bdev, b2)); | |
1053 | goto abort; | |
1054 | } | |
1055 | ev1 = md_event(sb); | |
1056 | ev2 = md_event(refsb); | |
1057 | if (ev1 > ev2) | |
1058 | ret = 1; | |
1059 | else | |
1060 | ret = 0; | |
1061 | } | |
8190e754 | 1062 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1063 | /* Limit to 4TB as metadata cannot record more than that. |
1064 | * (not needed for Linear and RAID0 as metadata doesn't | |
1065 | * record this size) | |
1066 | */ | |
1067 | if (rdev->sectors >= (2ULL << 32) && sb->level >= 1) | |
27a7b260 | 1068 | rdev->sectors = (2ULL << 32) - 2; |
1da177e4 | 1069 | |
27a7b260 | 1070 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1071 | /* "this cannot possibly happen" ... */ |
1072 | ret = -EINVAL; | |
1073 | ||
1da177e4 LT |
1074 | abort: |
1075 | return ret; | |
1076 | } | |
1077 | ||
1078 | /* | |
1079 | * validate_super for 0.90.0 | |
1080 | */ | |
fd01b88c | 1081 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1082 | { |
1083 | mdp_disk_t *desc; | |
65a06f06 | 1084 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1085 | __u64 ev1 = md_event(sb); |
1da177e4 | 1086 | |
41158c7e | 1087 | rdev->raid_disk = -1; |
c5d79adb N |
1088 | clear_bit(Faulty, &rdev->flags); |
1089 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1090 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1091 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1092 | |
1da177e4 LT |
1093 | if (mddev->raid_disks == 0) { |
1094 | mddev->major_version = 0; | |
1095 | mddev->minor_version = sb->minor_version; | |
1096 | mddev->patch_version = sb->patch_version; | |
e691063a | 1097 | mddev->external = 0; |
9d8f0363 | 1098 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1099 | mddev->ctime = sb->ctime; |
1100 | mddev->utime = sb->utime; | |
1101 | mddev->level = sb->level; | |
d9d166c2 | 1102 | mddev->clevel[0] = 0; |
1da177e4 LT |
1103 | mddev->layout = sb->layout; |
1104 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1105 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1106 | mddev->events = ev1; |
c3d9714e | 1107 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1108 | mddev->bitmap_info.space = 0; |
1109 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1110 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1111 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1112 | mddev->reshape_backwards = 0; |
1da177e4 | 1113 | |
f6705578 N |
1114 | if (mddev->minor_version >= 91) { |
1115 | mddev->reshape_position = sb->reshape_position; | |
1116 | mddev->delta_disks = sb->delta_disks; | |
1117 | mddev->new_level = sb->new_level; | |
1118 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1119 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1120 | if (mddev->delta_disks < 0) |
1121 | mddev->reshape_backwards = 1; | |
f6705578 N |
1122 | } else { |
1123 | mddev->reshape_position = MaxSector; | |
1124 | mddev->delta_disks = 0; | |
1125 | mddev->new_level = mddev->level; | |
1126 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1127 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1128 | } |
1129 | ||
1da177e4 LT |
1130 | if (sb->state & (1<<MD_SB_CLEAN)) |
1131 | mddev->recovery_cp = MaxSector; | |
1132 | else { | |
1133 | if (sb->events_hi == sb->cp_events_hi && | |
1134 | sb->events_lo == sb->cp_events_lo) { | |
1135 | mddev->recovery_cp = sb->recovery_cp; | |
1136 | } else | |
1137 | mddev->recovery_cp = 0; | |
1138 | } | |
1139 | ||
1140 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1141 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1142 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1143 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1144 | ||
1145 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1146 | |
1147 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1148 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1149 | mddev->bitmap_info.offset = |
1150 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1151 | mddev->bitmap_info.space = |
c9ad020f | 1152 | mddev->bitmap_info.default_space; |
6409bb05 | 1153 | } |
a654b9d8 | 1154 | |
41158c7e | 1155 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1156 | /* Insist on good event counter while assembling, except |
1157 | * for spares (which don't need an event count) */ | |
1da177e4 | 1158 | ++ev1; |
be6800a7 N |
1159 | if (sb->disks[rdev->desc_nr].state & ( |
1160 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
1161 | if (ev1 < mddev->events) | |
1162 | return -EINVAL; | |
41158c7e N |
1163 | } else if (mddev->bitmap) { |
1164 | /* if adding to array with a bitmap, then we can accept an | |
1165 | * older device ... but not too old. | |
1166 | */ | |
41158c7e N |
1167 | if (ev1 < mddev->bitmap->events_cleared) |
1168 | return 0; | |
8313b8e5 N |
1169 | if (ev1 < mddev->events) |
1170 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1171 | } else { |
1172 | if (ev1 < mddev->events) | |
1173 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1174 | return 0; | |
1175 | } | |
41158c7e | 1176 | |
1da177e4 | 1177 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1178 | desc = sb->disks + rdev->desc_nr; |
1179 | ||
1180 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1181 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1182 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1183 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1184 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1185 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
1186 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1187 | /* active but not in sync implies recovery up to | |
1188 | * reshape position. We don't know exactly where | |
1189 | * that is, so set to zero for now */ | |
1190 | if (mddev->minor_version >= 91) { | |
1191 | rdev->recovery_offset = 0; | |
1192 | rdev->raid_disk = desc->raid_disk; | |
1193 | } | |
1da177e4 | 1194 | } |
8ddf9efe N |
1195 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1196 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1197 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1198 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1199 | return 0; |
1200 | } | |
1201 | ||
1202 | /* | |
1203 | * sync_super for 0.90.0 | |
1204 | */ | |
fd01b88c | 1205 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1206 | { |
1207 | mdp_super_t *sb; | |
3cb03002 | 1208 | struct md_rdev *rdev2; |
1da177e4 | 1209 | int next_spare = mddev->raid_disks; |
19133a42 | 1210 | |
1da177e4 LT |
1211 | |
1212 | /* make rdev->sb match mddev data.. | |
1213 | * | |
1214 | * 1/ zero out disks | |
1215 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1216 | * 3/ any empty disks < next_spare become removed | |
1217 | * | |
1218 | * disks[0] gets initialised to REMOVED because | |
1219 | * we cannot be sure from other fields if it has | |
1220 | * been initialised or not. | |
1221 | */ | |
1222 | int i; | |
1223 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1224 | ||
61181565 N |
1225 | rdev->sb_size = MD_SB_BYTES; |
1226 | ||
65a06f06 | 1227 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1228 | |
1229 | memset(sb, 0, sizeof(*sb)); | |
1230 | ||
1231 | sb->md_magic = MD_SB_MAGIC; | |
1232 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1233 | sb->patch_version = mddev->patch_version; |
1234 | sb->gvalid_words = 0; /* ignored */ | |
1235 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1236 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1237 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1238 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1239 | ||
1240 | sb->ctime = mddev->ctime; | |
1241 | sb->level = mddev->level; | |
58c0fed4 | 1242 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1243 | sb->raid_disks = mddev->raid_disks; |
1244 | sb->md_minor = mddev->md_minor; | |
e691063a | 1245 | sb->not_persistent = 0; |
1da177e4 LT |
1246 | sb->utime = mddev->utime; |
1247 | sb->state = 0; | |
1248 | sb->events_hi = (mddev->events>>32); | |
1249 | sb->events_lo = (u32)mddev->events; | |
1250 | ||
f6705578 N |
1251 | if (mddev->reshape_position == MaxSector) |
1252 | sb->minor_version = 90; | |
1253 | else { | |
1254 | sb->minor_version = 91; | |
1255 | sb->reshape_position = mddev->reshape_position; | |
1256 | sb->new_level = mddev->new_level; | |
1257 | sb->delta_disks = mddev->delta_disks; | |
1258 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1259 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1260 | } |
1261 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1262 | if (mddev->in_sync) |
1263 | { | |
1264 | sb->recovery_cp = mddev->recovery_cp; | |
1265 | sb->cp_events_hi = (mddev->events>>32); | |
1266 | sb->cp_events_lo = (u32)mddev->events; | |
1267 | if (mddev->recovery_cp == MaxSector) | |
1268 | sb->state = (1<< MD_SB_CLEAN); | |
1269 | } else | |
1270 | sb->recovery_cp = 0; | |
1271 | ||
1272 | sb->layout = mddev->layout; | |
9d8f0363 | 1273 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1274 | |
c3d9714e | 1275 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1276 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1277 | ||
1da177e4 | 1278 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1279 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1280 | mdp_disk_t *d; |
86e6ffdd | 1281 | int desc_nr; |
0261cd9f N |
1282 | int is_active = test_bit(In_sync, &rdev2->flags); |
1283 | ||
1284 | if (rdev2->raid_disk >= 0 && | |
1285 | sb->minor_version >= 91) | |
1286 | /* we have nowhere to store the recovery_offset, | |
1287 | * but if it is not below the reshape_position, | |
1288 | * we can piggy-back on that. | |
1289 | */ | |
1290 | is_active = 1; | |
1291 | if (rdev2->raid_disk < 0 || | |
1292 | test_bit(Faulty, &rdev2->flags)) | |
1293 | is_active = 0; | |
1294 | if (is_active) | |
86e6ffdd | 1295 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1296 | else |
86e6ffdd | 1297 | desc_nr = next_spare++; |
19133a42 | 1298 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1299 | d = &sb->disks[rdev2->desc_nr]; |
1300 | nr_disks++; | |
1301 | d->number = rdev2->desc_nr; | |
1302 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1303 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1304 | if (is_active) |
1da177e4 LT |
1305 | d->raid_disk = rdev2->raid_disk; |
1306 | else | |
1307 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1308 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1309 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1310 | else if (is_active) { |
1da177e4 | 1311 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1312 | if (test_bit(In_sync, &rdev2->flags)) |
1313 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1314 | active++; |
1315 | working++; | |
1316 | } else { | |
1317 | d->state = 0; | |
1318 | spare++; | |
1319 | working++; | |
1320 | } | |
8ddf9efe N |
1321 | if (test_bit(WriteMostly, &rdev2->flags)) |
1322 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1323 | } |
1da177e4 LT |
1324 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1325 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1326 | mdp_disk_t *d = &sb->disks[i]; | |
1327 | if (d->state == 0 && d->number == 0) { | |
1328 | d->number = i; | |
1329 | d->raid_disk = i; | |
1330 | d->state = (1<<MD_DISK_REMOVED); | |
1331 | d->state |= (1<<MD_DISK_FAULTY); | |
1332 | failed++; | |
1333 | } | |
1334 | } | |
1335 | sb->nr_disks = nr_disks; | |
1336 | sb->active_disks = active; | |
1337 | sb->working_disks = working; | |
1338 | sb->failed_disks = failed; | |
1339 | sb->spare_disks = spare; | |
1340 | ||
1341 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1342 | sb->sb_csum = calc_sb_csum(sb); | |
1343 | } | |
1344 | ||
0cd17fec CW |
1345 | /* |
1346 | * rdev_size_change for 0.90.0 | |
1347 | */ | |
1348 | static unsigned long long | |
3cb03002 | 1349 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1350 | { |
58c0fed4 | 1351 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1352 | return 0; /* component must fit device */ |
c3d9714e | 1353 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1354 | return 0; /* can't move bitmap */ |
57b2caa3 | 1355 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1356 | if (!num_sectors || num_sectors > rdev->sb_start) |
1357 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1358 | /* Limit to 4TB as metadata cannot record more than that. |
1359 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1360 | */ | |
667a5313 | 1361 | if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1) |
27a7b260 | 1362 | num_sectors = (2ULL << 32) - 2; |
0f420358 | 1363 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1364 | rdev->sb_page); |
1365 | md_super_wait(rdev->mddev); | |
c26a44ed | 1366 | return num_sectors; |
0cd17fec CW |
1367 | } |
1368 | ||
c6563a8c N |
1369 | static int |
1370 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1371 | { | |
1372 | /* non-zero offset changes not possible with v0.90 */ | |
1373 | return new_offset == 0; | |
1374 | } | |
0cd17fec | 1375 | |
1da177e4 LT |
1376 | /* |
1377 | * version 1 superblock | |
1378 | */ | |
1379 | ||
1c05b4bc | 1380 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1381 | { |
1c05b4bc N |
1382 | __le32 disk_csum; |
1383 | u32 csum; | |
1da177e4 LT |
1384 | unsigned long long newcsum; |
1385 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1386 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1387 | |
1388 | disk_csum = sb->sb_csum; | |
1389 | sb->sb_csum = 0; | |
1390 | newcsum = 0; | |
1f3c9907 | 1391 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1392 | newcsum += le32_to_cpu(*isuper++); |
1393 | ||
1394 | if (size == 2) | |
1c05b4bc | 1395 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1396 | |
1397 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1398 | sb->sb_csum = disk_csum; | |
1399 | return cpu_to_le32(csum); | |
1400 | } | |
1401 | ||
2699b672 N |
1402 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, |
1403 | int acknowledged); | |
3cb03002 | 1404 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1405 | { |
1406 | struct mdp_superblock_1 *sb; | |
1407 | int ret; | |
0f420358 | 1408 | sector_t sb_start; |
c6563a8c | 1409 | sector_t sectors; |
1da177e4 | 1410 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1411 | int bmask; |
1da177e4 LT |
1412 | |
1413 | /* | |
0f420358 | 1414 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1415 | * It is always aligned to a 4K boundary and |
1416 | * depeding on minor_version, it can be: | |
1417 | * 0: At least 8K, but less than 12K, from end of device | |
1418 | * 1: At start of device | |
1419 | * 2: 4K from start of device. | |
1420 | */ | |
1421 | switch(minor_version) { | |
1422 | case 0: | |
77304d2a | 1423 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1424 | sb_start -= 8*2; |
1425 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1426 | break; |
1427 | case 1: | |
0f420358 | 1428 | sb_start = 0; |
1da177e4 LT |
1429 | break; |
1430 | case 2: | |
0f420358 | 1431 | sb_start = 8; |
1da177e4 LT |
1432 | break; |
1433 | default: | |
1434 | return -EINVAL; | |
1435 | } | |
0f420358 | 1436 | rdev->sb_start = sb_start; |
1da177e4 | 1437 | |
0002b271 N |
1438 | /* superblock is rarely larger than 1K, but it can be larger, |
1439 | * and it is safe to read 4k, so we do that | |
1440 | */ | |
1441 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1442 | if (ret) return ret; |
1443 | ||
1444 | ||
65a06f06 | 1445 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1446 | |
1447 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1448 | sb->major_version != cpu_to_le32(1) || | |
1449 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1450 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1451 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1452 | return -EINVAL; |
1453 | ||
1454 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1455 | printk("md: invalid superblock checksum on %s\n", | |
1456 | bdevname(rdev->bdev,b)); | |
1457 | return -EINVAL; | |
1458 | } | |
1459 | if (le64_to_cpu(sb->data_size) < 10) { | |
1460 | printk("md: data_size too small on %s\n", | |
1461 | bdevname(rdev->bdev,b)); | |
1462 | return -EINVAL; | |
1463 | } | |
c6563a8c N |
1464 | if (sb->pad0 || |
1465 | sb->pad3[0] || | |
1466 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1467 | /* Some padding is non-zero, might be a new feature */ | |
1468 | return -EINVAL; | |
e11e93fa | 1469 | |
1da177e4 LT |
1470 | rdev->preferred_minor = 0xffff; |
1471 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1472 | rdev->new_data_offset = rdev->data_offset; |
1473 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1474 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1475 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1476 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1477 | |
0002b271 | 1478 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1479 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1480 | if (rdev->sb_size & bmask) |
a1801f85 N |
1481 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1482 | ||
1483 | if (minor_version | |
0f420358 | 1484 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1485 | return -EINVAL; |
c6563a8c N |
1486 | if (minor_version |
1487 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1488 | return -EINVAL; | |
0002b271 | 1489 | |
31b65a0d N |
1490 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1491 | rdev->desc_nr = -1; | |
1492 | else | |
1493 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1494 | ||
2699b672 N |
1495 | if (!rdev->bb_page) { |
1496 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1497 | if (!rdev->bb_page) | |
1498 | return -ENOMEM; | |
1499 | } | |
1500 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1501 | rdev->badblocks.count == 0) { | |
1502 | /* need to load the bad block list. | |
1503 | * Currently we limit it to one page. | |
1504 | */ | |
1505 | s32 offset; | |
1506 | sector_t bb_sector; | |
1507 | u64 *bbp; | |
1508 | int i; | |
1509 | int sectors = le16_to_cpu(sb->bblog_size); | |
1510 | if (sectors > (PAGE_SIZE / 512)) | |
1511 | return -EINVAL; | |
1512 | offset = le32_to_cpu(sb->bblog_offset); | |
1513 | if (offset == 0) | |
1514 | return -EINVAL; | |
1515 | bb_sector = (long long)offset; | |
1516 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
1517 | rdev->bb_page, READ, true)) | |
1518 | return -EIO; | |
1519 | bbp = (u64 *)page_address(rdev->bb_page); | |
1520 | rdev->badblocks.shift = sb->bblog_shift; | |
1521 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1522 | u64 bb = le64_to_cpu(*bbp); | |
1523 | int count = bb & (0x3ff); | |
1524 | u64 sector = bb >> 10; | |
1525 | sector <<= sb->bblog_shift; | |
1526 | count <<= sb->bblog_shift; | |
1527 | if (bb + 1 == 0) | |
1528 | break; | |
1529 | if (md_set_badblocks(&rdev->badblocks, | |
1530 | sector, count, 1) == 0) | |
1531 | return -EINVAL; | |
1532 | } | |
486adf72 N |
1533 | } else if (sb->bblog_offset != 0) |
1534 | rdev->badblocks.shift = 0; | |
2699b672 | 1535 | |
9a7b2b0f | 1536 | if (!refdev) { |
8ed75463 | 1537 | ret = 1; |
9a7b2b0f | 1538 | } else { |
1da177e4 | 1539 | __u64 ev1, ev2; |
65a06f06 | 1540 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1541 | |
1542 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1543 | sb->level != refsb->level || | |
1544 | sb->layout != refsb->layout || | |
1545 | sb->chunksize != refsb->chunksize) { | |
1546 | printk(KERN_WARNING "md: %s has strangely different" | |
1547 | " superblock to %s\n", | |
1548 | bdevname(rdev->bdev,b), | |
1549 | bdevname(refdev->bdev,b2)); | |
1550 | return -EINVAL; | |
1551 | } | |
1552 | ev1 = le64_to_cpu(sb->events); | |
1553 | ev2 = le64_to_cpu(refsb->events); | |
1554 | ||
1555 | if (ev1 > ev2) | |
8ed75463 N |
1556 | ret = 1; |
1557 | else | |
1558 | ret = 0; | |
1da177e4 | 1559 | } |
c6563a8c N |
1560 | if (minor_version) { |
1561 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1562 | sectors -= rdev->data_offset; | |
1563 | } else | |
1564 | sectors = rdev->sb_start; | |
1565 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1566 | return -EINVAL; |
dd8ac336 | 1567 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1568 | return ret; |
1da177e4 LT |
1569 | } |
1570 | ||
fd01b88c | 1571 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1572 | { |
65a06f06 | 1573 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1574 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1575 | |
41158c7e | 1576 | rdev->raid_disk = -1; |
c5d79adb N |
1577 | clear_bit(Faulty, &rdev->flags); |
1578 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1579 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1580 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1581 | |
1da177e4 LT |
1582 | if (mddev->raid_disks == 0) { |
1583 | mddev->major_version = 1; | |
1584 | mddev->patch_version = 0; | |
e691063a | 1585 | mddev->external = 0; |
9d8f0363 | 1586 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1587 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1588 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1589 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1590 | mddev->clevel[0] = 0; |
1da177e4 LT |
1591 | mddev->layout = le32_to_cpu(sb->layout); |
1592 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1593 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1594 | mddev->events = ev1; |
c3d9714e | 1595 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1596 | mddev->bitmap_info.space = 0; |
1597 | /* Default location for bitmap is 1K after superblock | |
1598 | * using 3K - total of 4K | |
1599 | */ | |
c3d9714e | 1600 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1601 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1602 | mddev->reshape_backwards = 0; |
1603 | ||
1da177e4 LT |
1604 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1605 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1606 | ||
1607 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1608 | |
71c0805c | 1609 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1610 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1611 | mddev->bitmap_info.offset = |
1612 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1613 | /* Metadata doesn't record how much space is available. |
1614 | * For 1.0, we assume we can use up to the superblock | |
1615 | * if before, else to 4K beyond superblock. | |
1616 | * For others, assume no change is possible. | |
1617 | */ | |
1618 | if (mddev->minor_version > 0) | |
1619 | mddev->bitmap_info.space = 0; | |
1620 | else if (mddev->bitmap_info.offset > 0) | |
1621 | mddev->bitmap_info.space = | |
1622 | 8 - mddev->bitmap_info.offset; | |
1623 | else | |
1624 | mddev->bitmap_info.space = | |
1625 | -mddev->bitmap_info.offset; | |
1626 | } | |
e11e93fa | 1627 | |
f6705578 N |
1628 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1629 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1630 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1631 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1632 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1633 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1634 | if (mddev->delta_disks < 0 || |
1635 | (mddev->delta_disks == 0 && | |
1636 | (le32_to_cpu(sb->feature_map) | |
1637 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1638 | mddev->reshape_backwards = 1; | |
f6705578 N |
1639 | } else { |
1640 | mddev->reshape_position = MaxSector; | |
1641 | mddev->delta_disks = 0; | |
1642 | mddev->new_level = mddev->level; | |
1643 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1644 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1645 | } |
1646 | ||
41158c7e | 1647 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1648 | /* Insist of good event counter while assembling, except for |
1649 | * spares (which don't need an event count) */ | |
1da177e4 | 1650 | ++ev1; |
be6800a7 N |
1651 | if (rdev->desc_nr >= 0 && |
1652 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1653 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1654 | if (ev1 < mddev->events) | |
1655 | return -EINVAL; | |
41158c7e N |
1656 | } else if (mddev->bitmap) { |
1657 | /* If adding to array with a bitmap, then we can accept an | |
1658 | * older device, but not too old. | |
1659 | */ | |
41158c7e N |
1660 | if (ev1 < mddev->bitmap->events_cleared) |
1661 | return 0; | |
8313b8e5 N |
1662 | if (ev1 < mddev->events) |
1663 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1664 | } else { |
1665 | if (ev1 < mddev->events) | |
1666 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1667 | return 0; | |
1668 | } | |
1da177e4 LT |
1669 | if (mddev->level != LEVEL_MULTIPATH) { |
1670 | int role; | |
3673f305 N |
1671 | if (rdev->desc_nr < 0 || |
1672 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1673 | role = 0xffff; | |
1674 | rdev->desc_nr = -1; | |
1675 | } else | |
1676 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1677 | switch(role) { |
1678 | case 0xffff: /* spare */ | |
1da177e4 LT |
1679 | break; |
1680 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1681 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1682 | break; |
1683 | default: | |
5fd6c1dc N |
1684 | if ((le32_to_cpu(sb->feature_map) & |
1685 | MD_FEATURE_RECOVERY_OFFSET)) | |
1686 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1687 | else | |
1688 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1689 | rdev->raid_disk = role; |
1690 | break; | |
1691 | } | |
8ddf9efe N |
1692 | if (sb->devflags & WriteMostly1) |
1693 | set_bit(WriteMostly, &rdev->flags); | |
2d78f8c4 N |
1694 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1695 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1696 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1697 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1698 | |
1da177e4 LT |
1699 | return 0; |
1700 | } | |
1701 | ||
fd01b88c | 1702 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1703 | { |
1704 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1705 | struct md_rdev *rdev2; |
1da177e4 LT |
1706 | int max_dev, i; |
1707 | /* make rdev->sb match mddev and rdev data. */ | |
1708 | ||
65a06f06 | 1709 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1710 | |
1711 | sb->feature_map = 0; | |
1712 | sb->pad0 = 0; | |
5fd6c1dc | 1713 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1714 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1715 | ||
1716 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1717 | sb->events = cpu_to_le64(mddev->events); | |
1718 | if (mddev->in_sync) | |
1719 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1720 | else | |
1721 | sb->resync_offset = cpu_to_le64(0); | |
1722 | ||
1c05b4bc | 1723 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1724 | |
f0ca340c | 1725 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1726 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1727 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1728 | sb->level = cpu_to_le32(mddev->level); |
1729 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1730 | |
aeb9b211 N |
1731 | if (test_bit(WriteMostly, &rdev->flags)) |
1732 | sb->devflags |= WriteMostly1; | |
1733 | else | |
1734 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1735 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1736 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1737 | |
c3d9714e N |
1738 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1739 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1740 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1741 | } |
5fd6c1dc N |
1742 | |
1743 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1744 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1745 | sb->feature_map |= |
1746 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1747 | sb->recovery_offset = | |
1748 | cpu_to_le64(rdev->recovery_offset); | |
5fd6c1dc | 1749 | } |
2d78f8c4 N |
1750 | if (test_bit(Replacement, &rdev->flags)) |
1751 | sb->feature_map |= | |
1752 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1753 | |
f6705578 N |
1754 | if (mddev->reshape_position != MaxSector) { |
1755 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1756 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1757 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1758 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1759 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1760 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1761 | if (mddev->delta_disks == 0 && |
1762 | mddev->reshape_backwards) | |
1763 | sb->feature_map | |
1764 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1765 | if (rdev->new_data_offset != rdev->data_offset) { |
1766 | sb->feature_map | |
1767 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1768 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1769 | - rdev->data_offset)); | |
1770 | } | |
f6705578 | 1771 | } |
a654b9d8 | 1772 | |
2699b672 N |
1773 | if (rdev->badblocks.count == 0) |
1774 | /* Nothing to do for bad blocks*/ ; | |
1775 | else if (sb->bblog_offset == 0) | |
1776 | /* Cannot record bad blocks on this device */ | |
1777 | md_error(mddev, rdev); | |
1778 | else { | |
1779 | struct badblocks *bb = &rdev->badblocks; | |
1780 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1781 | u64 *p = bb->page; | |
1782 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1783 | if (bb->changed) { | |
1784 | unsigned seq; | |
1785 | ||
1786 | retry: | |
1787 | seq = read_seqbegin(&bb->lock); | |
1788 | ||
1789 | memset(bbp, 0xff, PAGE_SIZE); | |
1790 | ||
1791 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1792 | u64 internal_bb = p[i]; |
2699b672 N |
1793 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1794 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1795 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1796 | } |
d0962936 | 1797 | bb->changed = 0; |
2699b672 N |
1798 | if (read_seqretry(&bb->lock, seq)) |
1799 | goto retry; | |
1800 | ||
1801 | bb->sector = (rdev->sb_start + | |
1802 | (int)le32_to_cpu(sb->bblog_offset)); | |
1803 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1804 | } |
1805 | } | |
1806 | ||
1da177e4 | 1807 | max_dev = 0; |
dafb20fa | 1808 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1809 | if (rdev2->desc_nr+1 > max_dev) |
1810 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1811 | |
70471daf N |
1812 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1813 | int bmask; | |
a778b73f | 1814 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1815 | rdev->sb_size = max_dev * 2 + 256; |
1816 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1817 | if (rdev->sb_size & bmask) | |
1818 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1819 | } else |
1820 | max_dev = le32_to_cpu(sb->max_dev); | |
1821 | ||
1da177e4 LT |
1822 | for (i=0; i<max_dev;i++) |
1823 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1824 | ||
dafb20fa | 1825 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1826 | i = rdev2->desc_nr; |
b2d444d7 | 1827 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1828 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1829 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1830 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1831 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1832 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1833 | else |
1834 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1835 | } | |
1836 | ||
1da177e4 LT |
1837 | sb->sb_csum = calc_sb_1_csum(sb); |
1838 | } | |
1839 | ||
0cd17fec | 1840 | static unsigned long long |
3cb03002 | 1841 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1842 | { |
1843 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1844 | sector_t max_sectors; |
58c0fed4 | 1845 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1846 | return 0; /* component must fit device */ |
c6563a8c N |
1847 | if (rdev->data_offset != rdev->new_data_offset) |
1848 | return 0; /* too confusing */ | |
0f420358 | 1849 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1850 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1851 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1852 | max_sectors -= rdev->data_offset; |
1853 | if (!num_sectors || num_sectors > max_sectors) | |
1854 | num_sectors = max_sectors; | |
c3d9714e | 1855 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1856 | /* minor version 0 with bitmap we can't move */ |
1857 | return 0; | |
1858 | } else { | |
1859 | /* minor version 0; superblock after data */ | |
0f420358 | 1860 | sector_t sb_start; |
77304d2a | 1861 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1862 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1863 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1864 | if (!num_sectors || num_sectors > max_sectors) |
1865 | num_sectors = max_sectors; | |
0f420358 | 1866 | rdev->sb_start = sb_start; |
0cd17fec | 1867 | } |
65a06f06 | 1868 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1869 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1870 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1871 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1872 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1873 | rdev->sb_page); |
1874 | md_super_wait(rdev->mddev); | |
c26a44ed | 1875 | return num_sectors; |
c6563a8c N |
1876 | |
1877 | } | |
1878 | ||
1879 | static int | |
1880 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1881 | unsigned long long new_offset) | |
1882 | { | |
1883 | /* All necessary checks on new >= old have been done */ | |
1884 | struct bitmap *bitmap; | |
1885 | if (new_offset >= rdev->data_offset) | |
1886 | return 1; | |
1887 | ||
1888 | /* with 1.0 metadata, there is no metadata to tread on | |
1889 | * so we can always move back */ | |
1890 | if (rdev->mddev->minor_version == 0) | |
1891 | return 1; | |
1892 | ||
1893 | /* otherwise we must be sure not to step on | |
1894 | * any metadata, so stay: | |
1895 | * 36K beyond start of superblock | |
1896 | * beyond end of badblocks | |
1897 | * beyond write-intent bitmap | |
1898 | */ | |
1899 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1900 | return 0; | |
1901 | bitmap = rdev->mddev->bitmap; | |
1902 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1903 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1904 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1905 | return 0; |
1906 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1907 | return 0; | |
1908 | ||
1909 | return 1; | |
0cd17fec | 1910 | } |
1da177e4 | 1911 | |
75c96f85 | 1912 | static struct super_type super_types[] = { |
1da177e4 LT |
1913 | [0] = { |
1914 | .name = "0.90.0", | |
1915 | .owner = THIS_MODULE, | |
0cd17fec CW |
1916 | .load_super = super_90_load, |
1917 | .validate_super = super_90_validate, | |
1918 | .sync_super = super_90_sync, | |
1919 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1920 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1921 | }, |
1922 | [1] = { | |
1923 | .name = "md-1", | |
1924 | .owner = THIS_MODULE, | |
0cd17fec CW |
1925 | .load_super = super_1_load, |
1926 | .validate_super = super_1_validate, | |
1927 | .sync_super = super_1_sync, | |
1928 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1929 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1930 | }, |
1931 | }; | |
1da177e4 | 1932 | |
fd01b88c | 1933 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1934 | { |
1935 | if (mddev->sync_super) { | |
1936 | mddev->sync_super(mddev, rdev); | |
1937 | return; | |
1938 | } | |
1939 | ||
1940 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1941 | ||
1942 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1943 | } | |
1944 | ||
fd01b88c | 1945 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1946 | { |
3cb03002 | 1947 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1948 | |
4b80991c N |
1949 | rcu_read_lock(); |
1950 | rdev_for_each_rcu(rdev, mddev1) | |
1951 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1952 | if (rdev->bdev->bd_contains == |
4b80991c N |
1953 | rdev2->bdev->bd_contains) { |
1954 | rcu_read_unlock(); | |
7dd5e7c3 | 1955 | return 1; |
4b80991c N |
1956 | } |
1957 | rcu_read_unlock(); | |
1da177e4 LT |
1958 | return 0; |
1959 | } | |
1960 | ||
1961 | static LIST_HEAD(pending_raid_disks); | |
1962 | ||
ac5e7113 AN |
1963 | /* |
1964 | * Try to register data integrity profile for an mddev | |
1965 | * | |
1966 | * This is called when an array is started and after a disk has been kicked | |
1967 | * from the array. It only succeeds if all working and active component devices | |
1968 | * are integrity capable with matching profiles. | |
1969 | */ | |
fd01b88c | 1970 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 1971 | { |
3cb03002 | 1972 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
1973 | |
1974 | if (list_empty(&mddev->disks)) | |
1975 | return 0; /* nothing to do */ | |
629acb6a JB |
1976 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
1977 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 1978 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
1979 | /* skip spares and non-functional disks */ |
1980 | if (test_bit(Faulty, &rdev->flags)) | |
1981 | continue; | |
1982 | if (rdev->raid_disk < 0) | |
1983 | continue; | |
ac5e7113 AN |
1984 | if (!reference) { |
1985 | /* Use the first rdev as the reference */ | |
1986 | reference = rdev; | |
1987 | continue; | |
1988 | } | |
1989 | /* does this rdev's profile match the reference profile? */ | |
1990 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1991 | rdev->bdev->bd_disk) < 0) | |
1992 | return -EINVAL; | |
1993 | } | |
89078d57 MP |
1994 | if (!reference || !bdev_get_integrity(reference->bdev)) |
1995 | return 0; | |
ac5e7113 AN |
1996 | /* |
1997 | * All component devices are integrity capable and have matching | |
1998 | * profiles, register the common profile for the md device. | |
1999 | */ | |
2000 | if (blk_integrity_register(mddev->gendisk, | |
2001 | bdev_get_integrity(reference->bdev)) != 0) { | |
2002 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
2003 | mdname(mddev)); | |
2004 | return -EINVAL; | |
2005 | } | |
a91a2785 MP |
2006 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
2007 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
2008 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
2009 | mdname(mddev)); | |
2010 | return -EINVAL; | |
2011 | } | |
ac5e7113 AN |
2012 | return 0; |
2013 | } | |
2014 | EXPORT_SYMBOL(md_integrity_register); | |
2015 | ||
2016 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
fd01b88c | 2017 | void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) |
3f9d99c1 | 2018 | { |
2863b9eb JB |
2019 | struct blk_integrity *bi_rdev; |
2020 | struct blk_integrity *bi_mddev; | |
2021 | ||
2022 | if (!mddev->gendisk) | |
2023 | return; | |
2024 | ||
2025 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2026 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2027 | |
ac5e7113 | 2028 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 2029 | return; |
ac5e7113 | 2030 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 2031 | return; |
ac5e7113 AN |
2032 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
2033 | rdev->bdev->bd_disk) >= 0) | |
2034 | return; | |
2035 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
2036 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 2037 | } |
ac5e7113 | 2038 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2039 | |
fd01b88c | 2040 | static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev) |
1da177e4 | 2041 | { |
7dd5e7c3 | 2042 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2043 | struct kobject *ko; |
1edf80d3 | 2044 | char *s; |
5e55e2f5 | 2045 | int err; |
1da177e4 LT |
2046 | |
2047 | if (rdev->mddev) { | |
2048 | MD_BUG(); | |
2049 | return -EINVAL; | |
2050 | } | |
11e2ede0 DW |
2051 | |
2052 | /* prevent duplicates */ | |
2053 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2054 | return -EEXIST; | |
2055 | ||
dd8ac336 AN |
2056 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
2057 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
2058 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2059 | if (mddev->pers) { |
2060 | /* Cannot change size, so fail | |
2061 | * If mddev->level <= 0, then we don't care | |
2062 | * about aligning sizes (e.g. linear) | |
2063 | */ | |
2064 | if (mddev->level > 0) | |
2065 | return -ENOSPC; | |
2066 | } else | |
dd8ac336 | 2067 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2068 | } |
1da177e4 LT |
2069 | |
2070 | /* Verify rdev->desc_nr is unique. | |
2071 | * If it is -1, assign a free number, else | |
2072 | * check number is not in use | |
2073 | */ | |
2074 | if (rdev->desc_nr < 0) { | |
2075 | int choice = 0; | |
2076 | if (mddev->pers) choice = mddev->raid_disks; | |
2077 | while (find_rdev_nr(mddev, choice)) | |
2078 | choice++; | |
2079 | rdev->desc_nr = choice; | |
2080 | } else { | |
2081 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
2082 | return -EBUSY; | |
2083 | } | |
de01dfad N |
2084 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
2085 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
2086 | mdname(mddev), mddev->max_disks); | |
2087 | return -EBUSY; | |
2088 | } | |
19133a42 | 2089 | bdevname(rdev->bdev,b); |
649316b2 | 2090 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 2091 | *s = '!'; |
649316b2 | 2092 | |
1da177e4 | 2093 | rdev->mddev = mddev; |
19133a42 | 2094 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 2095 | |
b2d6db58 | 2096 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2097 | goto fail; |
86e6ffdd | 2098 | |
0762b8bd | 2099 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2100 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2101 | /* failure here is OK */; | |
2102 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2103 | |
4b80991c | 2104 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2105 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2106 | |
2107 | /* May as well allow recovery to be retried once */ | |
5389042f | 2108 | mddev->recovery_disabled++; |
3f9d99c1 | 2109 | |
1da177e4 | 2110 | return 0; |
5e55e2f5 N |
2111 | |
2112 | fail: | |
2113 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
2114 | b, mdname(mddev)); | |
2115 | return err; | |
1da177e4 LT |
2116 | } |
2117 | ||
177a99b2 | 2118 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2119 | { |
3cb03002 | 2120 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2121 | kobject_del(&rdev->kobj); |
177a99b2 | 2122 | kobject_put(&rdev->kobj); |
5792a285 N |
2123 | } |
2124 | ||
3cb03002 | 2125 | static void unbind_rdev_from_array(struct md_rdev * rdev) |
1da177e4 LT |
2126 | { |
2127 | char b[BDEVNAME_SIZE]; | |
2128 | if (!rdev->mddev) { | |
2129 | MD_BUG(); | |
2130 | return; | |
2131 | } | |
49731baa | 2132 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2133 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
2134 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
2135 | rdev->mddev = NULL; | |
86e6ffdd | 2136 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2137 | sysfs_put(rdev->sysfs_state); |
2138 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2139 | rdev->badblocks.count = 0; |
5792a285 | 2140 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2141 | * writing to 'remove' to "dev/state". We also need |
2142 | * to delay it due to rcu usage. | |
5792a285 | 2143 | */ |
4b80991c | 2144 | synchronize_rcu(); |
177a99b2 N |
2145 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2146 | kobject_get(&rdev->kobj); | |
e804ac78 | 2147 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2148 | } |
2149 | ||
2150 | /* | |
2151 | * prevent the device from being mounted, repartitioned or | |
2152 | * otherwise reused by a RAID array (or any other kernel | |
2153 | * subsystem), by bd_claiming the device. | |
2154 | */ | |
3cb03002 | 2155 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2156 | { |
2157 | int err = 0; | |
2158 | struct block_device *bdev; | |
2159 | char b[BDEVNAME_SIZE]; | |
2160 | ||
d4d77629 | 2161 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2162 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 LT |
2163 | if (IS_ERR(bdev)) { |
2164 | printk(KERN_ERR "md: could not open %s.\n", | |
2165 | __bdevname(dev, b)); | |
2166 | return PTR_ERR(bdev); | |
2167 | } | |
1da177e4 LT |
2168 | rdev->bdev = bdev; |
2169 | return err; | |
2170 | } | |
2171 | ||
3cb03002 | 2172 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2173 | { |
2174 | struct block_device *bdev = rdev->bdev; | |
2175 | rdev->bdev = NULL; | |
2176 | if (!bdev) | |
2177 | MD_BUG(); | |
e525fd89 | 2178 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2179 | } |
2180 | ||
2181 | void md_autodetect_dev(dev_t dev); | |
2182 | ||
3cb03002 | 2183 | static void export_rdev(struct md_rdev * rdev) |
1da177e4 LT |
2184 | { |
2185 | char b[BDEVNAME_SIZE]; | |
2186 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
2187 | bdevname(rdev->bdev,b)); | |
2188 | if (rdev->mddev) | |
2189 | MD_BUG(); | |
545c8795 | 2190 | md_rdev_clear(rdev); |
1da177e4 | 2191 | #ifndef MODULE |
d0fae18f N |
2192 | if (test_bit(AutoDetected, &rdev->flags)) |
2193 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2194 | #endif |
2195 | unlock_rdev(rdev); | |
86e6ffdd | 2196 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2197 | } |
2198 | ||
3cb03002 | 2199 | static void kick_rdev_from_array(struct md_rdev * rdev) |
1da177e4 LT |
2200 | { |
2201 | unbind_rdev_from_array(rdev); | |
2202 | export_rdev(rdev); | |
2203 | } | |
2204 | ||
fd01b88c | 2205 | static void export_array(struct mddev *mddev) |
1da177e4 | 2206 | { |
3cb03002 | 2207 | struct md_rdev *rdev, *tmp; |
1da177e4 | 2208 | |
dafb20fa | 2209 | rdev_for_each_safe(rdev, tmp, mddev) { |
1da177e4 LT |
2210 | if (!rdev->mddev) { |
2211 | MD_BUG(); | |
2212 | continue; | |
2213 | } | |
2214 | kick_rdev_from_array(rdev); | |
2215 | } | |
2216 | if (!list_empty(&mddev->disks)) | |
2217 | MD_BUG(); | |
2218 | mddev->raid_disks = 0; | |
2219 | mddev->major_version = 0; | |
2220 | } | |
2221 | ||
2222 | static void print_desc(mdp_disk_t *desc) | |
2223 | { | |
2224 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
2225 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
2226 | } | |
2227 | ||
cd2ac932 | 2228 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
2229 | { |
2230 | int i; | |
2231 | ||
2232 | printk(KERN_INFO | |
2233 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
2234 | sb->major_version, sb->minor_version, sb->patch_version, | |
2235 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
2236 | sb->ctime); | |
2237 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
2238 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
2239 | sb->md_minor, sb->layout, sb->chunk_size); | |
2240 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
2241 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
2242 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
2243 | sb->failed_disks, sb->spare_disks, | |
2244 | sb->sb_csum, (unsigned long)sb->events_lo); | |
2245 | ||
2246 | printk(KERN_INFO); | |
2247 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2248 | mdp_disk_t *desc; | |
2249 | ||
2250 | desc = sb->disks + i; | |
2251 | if (desc->number || desc->major || desc->minor || | |
2252 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
2253 | printk(" D %2d: ", i); | |
2254 | print_desc(desc); | |
2255 | } | |
2256 | } | |
2257 | printk(KERN_INFO "md: THIS: "); | |
2258 | print_desc(&sb->this_disk); | |
cd2ac932 | 2259 | } |
1da177e4 | 2260 | |
cd2ac932 CR |
2261 | static void print_sb_1(struct mdp_superblock_1 *sb) |
2262 | { | |
2263 | __u8 *uuid; | |
2264 | ||
2265 | uuid = sb->set_uuid; | |
ad361c98 | 2266 | printk(KERN_INFO |
7b75c2f8 | 2267 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 2268 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
2269 | le32_to_cpu(sb->major_version), |
2270 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 2271 | uuid, |
cd2ac932 CR |
2272 | sb->set_name, |
2273 | (unsigned long long)le64_to_cpu(sb->ctime) | |
2274 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
2275 | ||
2276 | uuid = sb->device_uuid; | |
ad361c98 JP |
2277 | printk(KERN_INFO |
2278 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 2279 | " RO:%llu\n" |
7b75c2f8 | 2280 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
2281 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
2282 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
2283 | le32_to_cpu(sb->level), |
2284 | (unsigned long long)le64_to_cpu(sb->size), | |
2285 | le32_to_cpu(sb->raid_disks), | |
2286 | le32_to_cpu(sb->layout), | |
2287 | le32_to_cpu(sb->chunksize), | |
2288 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2289 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2290 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2291 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2292 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2293 | uuid, |
cd2ac932 CR |
2294 | sb->devflags, |
2295 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2296 | (unsigned long long)le64_to_cpu(sb->events), | |
2297 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2298 | le32_to_cpu(sb->sb_csum), | |
2299 | le32_to_cpu(sb->max_dev) | |
2300 | ); | |
1da177e4 LT |
2301 | } |
2302 | ||
3cb03002 | 2303 | static void print_rdev(struct md_rdev *rdev, int major_version) |
1da177e4 LT |
2304 | { |
2305 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2306 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2307 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2308 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2309 | rdev->desc_nr); | |
1da177e4 | 2310 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2311 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2312 | switch (major_version) { | |
2313 | case 0: | |
65a06f06 | 2314 | print_sb_90(page_address(rdev->sb_page)); |
cd2ac932 CR |
2315 | break; |
2316 | case 1: | |
65a06f06 | 2317 | print_sb_1(page_address(rdev->sb_page)); |
cd2ac932 CR |
2318 | break; |
2319 | } | |
1da177e4 LT |
2320 | } else |
2321 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2322 | } | |
2323 | ||
5e56341d | 2324 | static void md_print_devices(void) |
1da177e4 | 2325 | { |
159ec1fc | 2326 | struct list_head *tmp; |
3cb03002 | 2327 | struct md_rdev *rdev; |
fd01b88c | 2328 | struct mddev *mddev; |
1da177e4 LT |
2329 | char b[BDEVNAME_SIZE]; |
2330 | ||
2331 | printk("\n"); | |
2332 | printk("md: **********************************\n"); | |
2333 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2334 | printk("md: **********************************\n"); | |
29ac4aa3 | 2335 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2336 | |
32a7627c N |
2337 | if (mddev->bitmap) |
2338 | bitmap_print_sb(mddev->bitmap); | |
2339 | else | |
2340 | printk("%s: ", mdname(mddev)); | |
dafb20fa | 2341 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
2342 | printk("<%s>", bdevname(rdev->bdev,b)); |
2343 | printk("\n"); | |
2344 | ||
dafb20fa | 2345 | rdev_for_each(rdev, mddev) |
cd2ac932 | 2346 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2347 | } |
2348 | printk("md: **********************************\n"); | |
2349 | printk("\n"); | |
2350 | } | |
2351 | ||
2352 | ||
fd01b88c | 2353 | static void sync_sbs(struct mddev * mddev, int nospares) |
1da177e4 | 2354 | { |
42543769 N |
2355 | /* Update each superblock (in-memory image), but |
2356 | * if we are allowed to, skip spares which already | |
2357 | * have the right event counter, or have one earlier | |
2358 | * (which would mean they aren't being marked as dirty | |
2359 | * with the rest of the array) | |
2360 | */ | |
3cb03002 | 2361 | struct md_rdev *rdev; |
dafb20fa | 2362 | rdev_for_each(rdev, mddev) { |
42543769 N |
2363 | if (rdev->sb_events == mddev->events || |
2364 | (nospares && | |
2365 | rdev->raid_disk < 0 && | |
42543769 N |
2366 | rdev->sb_events+1 == mddev->events)) { |
2367 | /* Don't update this superblock */ | |
2368 | rdev->sb_loaded = 2; | |
2369 | } else { | |
076f968b | 2370 | sync_super(mddev, rdev); |
42543769 N |
2371 | rdev->sb_loaded = 1; |
2372 | } | |
1da177e4 LT |
2373 | } |
2374 | } | |
2375 | ||
fd01b88c | 2376 | static void md_update_sb(struct mddev * mddev, int force_change) |
1da177e4 | 2377 | { |
3cb03002 | 2378 | struct md_rdev *rdev; |
06d91a5f | 2379 | int sync_req; |
42543769 | 2380 | int nospares = 0; |
2699b672 | 2381 | int any_badblocks_changed = 0; |
1da177e4 | 2382 | |
d87f064f N |
2383 | if (mddev->ro) { |
2384 | if (force_change) | |
2385 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2386 | return; | |
2387 | } | |
1da177e4 | 2388 | repeat: |
3a3a5ddb | 2389 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2390 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2391 | if (rdev->raid_disk >= 0 && |
2392 | mddev->delta_disks >= 0 && | |
2393 | !test_bit(In_sync, &rdev->flags) && | |
2394 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2395 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2396 | ||
2397 | } | |
bd52b746 | 2398 | if (!mddev->persistent) { |
070dc6dd | 2399 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2400 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
de393cde | 2401 | if (!mddev->external) { |
d97a41dc | 2402 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
dafb20fa | 2403 | rdev_for_each(rdev, mddev) { |
de393cde | 2404 | if (rdev->badblocks.changed) { |
d0962936 | 2405 | rdev->badblocks.changed = 0; |
de393cde N |
2406 | md_ack_all_badblocks(&rdev->badblocks); |
2407 | md_error(mddev, rdev); | |
2408 | } | |
2409 | clear_bit(Blocked, &rdev->flags); | |
2410 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2411 | wake_up(&rdev->blocked_wait); | |
2412 | } | |
2413 | } | |
3a3a5ddb N |
2414 | wake_up(&mddev->sb_wait); |
2415 | return; | |
2416 | } | |
2417 | ||
a9701a30 | 2418 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2419 | |
3a3a5ddb N |
2420 | mddev->utime = get_seconds(); |
2421 | ||
850b2b42 N |
2422 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2423 | force_change = 1; | |
2424 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2425 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2426 | * though if events isn't the right even/odd, we will have to do | |
2427 | * spares after all | |
2428 | */ | |
2429 | nospares = 1; | |
2430 | if (force_change) | |
2431 | nospares = 0; | |
2432 | if (mddev->degraded) | |
84692195 N |
2433 | /* If the array is degraded, then skipping spares is both |
2434 | * dangerous and fairly pointless. | |
2435 | * Dangerous because a device that was removed from the array | |
2436 | * might have a event_count that still looks up-to-date, | |
2437 | * so it can be re-added without a resync. | |
2438 | * Pointless because if there are any spares to skip, | |
2439 | * then a recovery will happen and soon that array won't | |
2440 | * be degraded any more and the spare can go back to sleep then. | |
2441 | */ | |
850b2b42 | 2442 | nospares = 0; |
84692195 | 2443 | |
06d91a5f | 2444 | sync_req = mddev->in_sync; |
42543769 N |
2445 | |
2446 | /* If this is just a dirty<->clean transition, and the array is clean | |
2447 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2448 | if (nospares |
42543769 | 2449 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2450 | && mddev->can_decrease_events |
2451 | && mddev->events != 1) { | |
42543769 | 2452 | mddev->events--; |
a8707c08 N |
2453 | mddev->can_decrease_events = 0; |
2454 | } else { | |
42543769 N |
2455 | /* otherwise we have to go forward and ... */ |
2456 | mddev->events ++; | |
a8707c08 | 2457 | mddev->can_decrease_events = nospares; |
42543769 | 2458 | } |
1da177e4 LT |
2459 | |
2460 | if (!mddev->events) { | |
2461 | /* | |
2462 | * oops, this 64-bit counter should never wrap. | |
2463 | * Either we are in around ~1 trillion A.C., assuming | |
2464 | * 1 reboot per second, or we have a bug: | |
2465 | */ | |
2466 | MD_BUG(); | |
2467 | mddev->events --; | |
2468 | } | |
2699b672 | 2469 | |
dafb20fa | 2470 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2471 | if (rdev->badblocks.changed) |
2472 | any_badblocks_changed++; | |
de393cde N |
2473 | if (test_bit(Faulty, &rdev->flags)) |
2474 | set_bit(FaultRecorded, &rdev->flags); | |
2475 | } | |
2699b672 | 2476 | |
e691063a | 2477 | sync_sbs(mddev, nospares); |
a9701a30 | 2478 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 | 2479 | |
36a4e1fe N |
2480 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2481 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2482 | |
4ad13663 | 2483 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2484 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2485 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2486 | |
42543769 N |
2487 | if (rdev->sb_loaded != 1) |
2488 | continue; /* no noise on spare devices */ | |
1da177e4 | 2489 | |
d70ed2e4 AW |
2490 | if (!test_bit(Faulty, &rdev->flags) && |
2491 | rdev->saved_raid_disk == -1) { | |
7bfa19f2 | 2492 | md_super_write(mddev,rdev, |
0f420358 | 2493 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2494 | rdev->sb_page); |
36a4e1fe N |
2495 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2496 | bdevname(rdev->bdev, b), | |
2497 | (unsigned long long)rdev->sb_start); | |
42543769 | 2498 | rdev->sb_events = mddev->events; |
2699b672 N |
2499 | if (rdev->badblocks.size) { |
2500 | md_super_write(mddev, rdev, | |
2501 | rdev->badblocks.sector, | |
2502 | rdev->badblocks.size << 9, | |
2503 | rdev->bb_page); | |
2504 | rdev->badblocks.size = 0; | |
2505 | } | |
7bfa19f2 | 2506 | |
d70ed2e4 | 2507 | } else if (test_bit(Faulty, &rdev->flags)) |
36a4e1fe N |
2508 | pr_debug("md: %s (skipping faulty)\n", |
2509 | bdevname(rdev->bdev, b)); | |
d70ed2e4 AW |
2510 | else |
2511 | pr_debug("(skipping incremental s/r "); | |
2512 | ||
7bfa19f2 | 2513 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2514 | /* only need to write one superblock... */ |
2515 | break; | |
2516 | } | |
a9701a30 | 2517 | md_super_wait(mddev); |
850b2b42 | 2518 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2519 | |
a9701a30 | 2520 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2521 | if (mddev->in_sync != sync_req || |
2522 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2523 | /* have to write it out again */ |
a9701a30 | 2524 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2525 | goto repeat; |
2526 | } | |
850b2b42 | 2527 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2528 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2529 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2530 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2531 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2532 | |
dafb20fa | 2533 | rdev_for_each(rdev, mddev) { |
de393cde N |
2534 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2535 | clear_bit(Blocked, &rdev->flags); | |
2536 | ||
2537 | if (any_badblocks_changed) | |
2699b672 | 2538 | md_ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2539 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2540 | wake_up(&rdev->blocked_wait); | |
2541 | } | |
1da177e4 LT |
2542 | } |
2543 | ||
7f6ce769 | 2544 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2545 | * We want to accept with case. For this we use cmd_match. |
2546 | */ | |
2547 | static int cmd_match(const char *cmd, const char *str) | |
2548 | { | |
2549 | /* See if cmd, written into a sysfs file, matches | |
2550 | * str. They must either be the same, or cmd can | |
2551 | * have a trailing newline | |
2552 | */ | |
2553 | while (*cmd && *str && *cmd == *str) { | |
2554 | cmd++; | |
2555 | str++; | |
2556 | } | |
2557 | if (*cmd == '\n') | |
2558 | cmd++; | |
2559 | if (*str || *cmd) | |
2560 | return 0; | |
2561 | return 1; | |
2562 | } | |
2563 | ||
86e6ffdd N |
2564 | struct rdev_sysfs_entry { |
2565 | struct attribute attr; | |
3cb03002 N |
2566 | ssize_t (*show)(struct md_rdev *, char *); |
2567 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2568 | }; |
2569 | ||
2570 | static ssize_t | |
3cb03002 | 2571 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd N |
2572 | { |
2573 | char *sep = ""; | |
20a49ff6 | 2574 | size_t len = 0; |
86e6ffdd | 2575 | |
de393cde N |
2576 | if (test_bit(Faulty, &rdev->flags) || |
2577 | rdev->badblocks.unacked_exist) { | |
86e6ffdd N |
2578 | len+= sprintf(page+len, "%sfaulty",sep); |
2579 | sep = ","; | |
2580 | } | |
b2d444d7 | 2581 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2582 | len += sprintf(page+len, "%sin_sync",sep); |
2583 | sep = ","; | |
2584 | } | |
f655675b N |
2585 | if (test_bit(WriteMostly, &rdev->flags)) { |
2586 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2587 | sep = ","; | |
2588 | } | |
de393cde | 2589 | if (test_bit(Blocked, &rdev->flags) || |
52c64152 N |
2590 | (rdev->badblocks.unacked_exist |
2591 | && !test_bit(Faulty, &rdev->flags))) { | |
6bfe0b49 DW |
2592 | len += sprintf(page+len, "%sblocked", sep); |
2593 | sep = ","; | |
2594 | } | |
b2d444d7 N |
2595 | if (!test_bit(Faulty, &rdev->flags) && |
2596 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2597 | len += sprintf(page+len, "%sspare", sep); |
2598 | sep = ","; | |
2599 | } | |
d7a9d443 N |
2600 | if (test_bit(WriteErrorSeen, &rdev->flags)) { |
2601 | len += sprintf(page+len, "%swrite_error", sep); | |
2602 | sep = ","; | |
2603 | } | |
2d78f8c4 N |
2604 | if (test_bit(WantReplacement, &rdev->flags)) { |
2605 | len += sprintf(page+len, "%swant_replacement", sep); | |
2606 | sep = ","; | |
2607 | } | |
2608 | if (test_bit(Replacement, &rdev->flags)) { | |
2609 | len += sprintf(page+len, "%sreplacement", sep); | |
2610 | sep = ","; | |
2611 | } | |
2612 | ||
86e6ffdd N |
2613 | return len+sprintf(page+len, "\n"); |
2614 | } | |
2615 | ||
45dc2de1 | 2616 | static ssize_t |
3cb03002 | 2617 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2618 | { |
2619 | /* can write | |
de393cde | 2620 | * faulty - simulates an error |
45dc2de1 | 2621 | * remove - disconnects the device |
f655675b N |
2622 | * writemostly - sets write_mostly |
2623 | * -writemostly - clears write_mostly | |
de393cde N |
2624 | * blocked - sets the Blocked flags |
2625 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2626 | * insync - sets Insync providing device isn't active |
d7a9d443 N |
2627 | * write_error - sets WriteErrorSeen |
2628 | * -write_error - clears WriteErrorSeen | |
45dc2de1 N |
2629 | */ |
2630 | int err = -EINVAL; | |
2631 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2632 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2633 | if (test_bit(Faulty, &rdev->flags)) |
2634 | err = 0; | |
2635 | else | |
2636 | err = -EBUSY; | |
45dc2de1 N |
2637 | } else if (cmd_match(buf, "remove")) { |
2638 | if (rdev->raid_disk >= 0) | |
2639 | err = -EBUSY; | |
2640 | else { | |
fd01b88c | 2641 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2642 | kick_rdev_from_array(rdev); |
3f9d7b0d N |
2643 | if (mddev->pers) |
2644 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2645 | md_new_event(mddev); |
2646 | err = 0; | |
2647 | } | |
f655675b N |
2648 | } else if (cmd_match(buf, "writemostly")) { |
2649 | set_bit(WriteMostly, &rdev->flags); | |
2650 | err = 0; | |
2651 | } else if (cmd_match(buf, "-writemostly")) { | |
2652 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2653 | err = 0; |
2654 | } else if (cmd_match(buf, "blocked")) { | |
2655 | set_bit(Blocked, &rdev->flags); | |
2656 | err = 0; | |
2657 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2658 | if (!test_bit(Faulty, &rdev->flags) && |
7da64a0a | 2659 | rdev->badblocks.unacked_exist) { |
de393cde N |
2660 | /* metadata handler doesn't understand badblocks, |
2661 | * so we need to fail the device | |
2662 | */ | |
2663 | md_error(rdev->mddev, rdev); | |
2664 | } | |
6bfe0b49 | 2665 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2666 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2667 | wake_up(&rdev->blocked_wait); |
2668 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2669 | md_wakeup_thread(rdev->mddev->thread); | |
2670 | ||
6d56e278 N |
2671 | err = 0; |
2672 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2673 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2674 | err = 0; |
d7a9d443 N |
2675 | } else if (cmd_match(buf, "write_error")) { |
2676 | set_bit(WriteErrorSeen, &rdev->flags); | |
2677 | err = 0; | |
2678 | } else if (cmd_match(buf, "-write_error")) { | |
2679 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2680 | err = 0; | |
2d78f8c4 N |
2681 | } else if (cmd_match(buf, "want_replacement")) { |
2682 | /* Any non-spare device that is not a replacement can | |
2683 | * become want_replacement at any time, but we then need to | |
2684 | * check if recovery is needed. | |
2685 | */ | |
2686 | if (rdev->raid_disk >= 0 && | |
2687 | !test_bit(Replacement, &rdev->flags)) | |
2688 | set_bit(WantReplacement, &rdev->flags); | |
2689 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2690 | md_wakeup_thread(rdev->mddev->thread); | |
2691 | err = 0; | |
2692 | } else if (cmd_match(buf, "-want_replacement")) { | |
2693 | /* Clearing 'want_replacement' is always allowed. | |
2694 | * Once replacements starts it is too late though. | |
2695 | */ | |
2696 | err = 0; | |
2697 | clear_bit(WantReplacement, &rdev->flags); | |
2698 | } else if (cmd_match(buf, "replacement")) { | |
2699 | /* Can only set a device as a replacement when array has not | |
2700 | * yet been started. Once running, replacement is automatic | |
2701 | * from spares, or by assigning 'slot'. | |
2702 | */ | |
2703 | if (rdev->mddev->pers) | |
2704 | err = -EBUSY; | |
2705 | else { | |
2706 | set_bit(Replacement, &rdev->flags); | |
2707 | err = 0; | |
2708 | } | |
2709 | } else if (cmd_match(buf, "-replacement")) { | |
2710 | /* Similarly, can only clear Replacement before start */ | |
2711 | if (rdev->mddev->pers) | |
2712 | err = -EBUSY; | |
2713 | else { | |
2714 | clear_bit(Replacement, &rdev->flags); | |
2715 | err = 0; | |
2716 | } | |
45dc2de1 | 2717 | } |
00bcb4ac N |
2718 | if (!err) |
2719 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2720 | return err ? err : len; |
2721 | } | |
80ca3a44 N |
2722 | static struct rdev_sysfs_entry rdev_state = |
2723 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2724 | |
4dbcdc75 | 2725 | static ssize_t |
3cb03002 | 2726 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2727 | { |
2728 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2729 | } | |
2730 | ||
2731 | static ssize_t | |
3cb03002 | 2732 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 N |
2733 | { |
2734 | char *e; | |
2735 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2736 | if (*buf && (*e == 0 || *e == '\n')) { | |
2737 | atomic_set(&rdev->corrected_errors, n); | |
2738 | return len; | |
2739 | } | |
2740 | return -EINVAL; | |
2741 | } | |
2742 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2743 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2744 | |
014236d2 | 2745 | static ssize_t |
3cb03002 | 2746 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 N |
2747 | { |
2748 | if (rdev->raid_disk < 0) | |
2749 | return sprintf(page, "none\n"); | |
2750 | else | |
2751 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2752 | } | |
2753 | ||
2754 | static ssize_t | |
3cb03002 | 2755 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 N |
2756 | { |
2757 | char *e; | |
c303da6d | 2758 | int err; |
014236d2 N |
2759 | int slot = simple_strtoul(buf, &e, 10); |
2760 | if (strncmp(buf, "none", 4)==0) | |
2761 | slot = -1; | |
2762 | else if (e==buf || (*e && *e!= '\n')) | |
2763 | return -EINVAL; | |
6c2fce2e | 2764 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2765 | /* Setting 'slot' on an active array requires also |
2766 | * updating the 'rd%d' link, and communicating | |
2767 | * with the personality with ->hot_*_disk. | |
2768 | * For now we only support removing | |
2769 | * failed/spare devices. This normally happens automatically, | |
2770 | * but not when the metadata is externally managed. | |
2771 | */ | |
c303da6d N |
2772 | if (rdev->raid_disk == -1) |
2773 | return -EEXIST; | |
2774 | /* personality does all needed checks */ | |
01393f3d | 2775 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2776 | return -EINVAL; |
746d3207 N |
2777 | clear_bit(Blocked, &rdev->flags); |
2778 | remove_and_add_spares(rdev->mddev, rdev); | |
2779 | if (rdev->raid_disk >= 0) | |
2780 | return -EBUSY; | |
c303da6d N |
2781 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2782 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2783 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2784 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2785 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2786 | */ |
2787 | ||
2788 | if (rdev->raid_disk != -1) | |
2789 | return -EBUSY; | |
2790 | ||
c6751b2b N |
2791 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2792 | return -EBUSY; | |
2793 | ||
6c2fce2e NB |
2794 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2795 | return -EINVAL; | |
2796 | ||
ba1b41b6 N |
2797 | if (slot >= rdev->mddev->raid_disks && |
2798 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2799 | return -ENOSPC; | |
2800 | ||
6c2fce2e NB |
2801 | rdev->raid_disk = slot; |
2802 | if (test_bit(In_sync, &rdev->flags)) | |
2803 | rdev->saved_raid_disk = slot; | |
2804 | else | |
2805 | rdev->saved_raid_disk = -1; | |
d30519fc | 2806 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2807 | clear_bit(Bitmap_sync, &rdev->flags); |
6c2fce2e NB |
2808 | err = rdev->mddev->pers-> |
2809 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2810 | if (err) { |
6c2fce2e | 2811 | rdev->raid_disk = -1; |
6c2fce2e | 2812 | return err; |
52664732 | 2813 | } else |
00bcb4ac | 2814 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
36fad858 | 2815 | if (sysfs_link_rdev(rdev->mddev, rdev)) |
00bcb4ac | 2816 | /* failure here is OK */; |
6c2fce2e | 2817 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2818 | } else { |
ba1b41b6 N |
2819 | if (slot >= rdev->mddev->raid_disks && |
2820 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2821 | return -ENOSPC; |
2822 | rdev->raid_disk = slot; | |
2823 | /* assume it is working */ | |
c5d79adb N |
2824 | clear_bit(Faulty, &rdev->flags); |
2825 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2826 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2827 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2828 | } |
014236d2 N |
2829 | return len; |
2830 | } | |
2831 | ||
2832 | ||
2833 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2834 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2835 | |
93c8cad0 | 2836 | static ssize_t |
3cb03002 | 2837 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2838 | { |
6961ece4 | 2839 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2840 | } |
2841 | ||
2842 | static ssize_t | |
3cb03002 | 2843 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2844 | { |
c6563a8c | 2845 | unsigned long long offset; |
b29bebd6 | 2846 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2847 | return -EINVAL; |
8ed0a521 | 2848 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2849 | return -EBUSY; |
dd8ac336 | 2850 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2851 | /* Must set offset before size, so overlap checks |
2852 | * can be sane */ | |
2853 | return -EBUSY; | |
93c8cad0 | 2854 | rdev->data_offset = offset; |
25f7fd47 | 2855 | rdev->new_data_offset = offset; |
93c8cad0 N |
2856 | return len; |
2857 | } | |
2858 | ||
2859 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2860 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2861 | |
c6563a8c N |
2862 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2863 | { | |
2864 | return sprintf(page, "%llu\n", | |
2865 | (unsigned long long)rdev->new_data_offset); | |
2866 | } | |
2867 | ||
2868 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2869 | const char *buf, size_t len) | |
2870 | { | |
2871 | unsigned long long new_offset; | |
2872 | struct mddev *mddev = rdev->mddev; | |
2873 | ||
b29bebd6 | 2874 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2875 | return -EINVAL; |
2876 | ||
2877 | if (mddev->sync_thread) | |
2878 | return -EBUSY; | |
2879 | if (new_offset == rdev->data_offset) | |
2880 | /* reset is always permitted */ | |
2881 | ; | |
2882 | else if (new_offset > rdev->data_offset) { | |
2883 | /* must not push array size beyond rdev_sectors */ | |
2884 | if (new_offset - rdev->data_offset | |
2885 | + mddev->dev_sectors > rdev->sectors) | |
2886 | return -E2BIG; | |
2887 | } | |
2888 | /* Metadata worries about other space details. */ | |
2889 | ||
2890 | /* decreasing the offset is inconsistent with a backwards | |
2891 | * reshape. | |
2892 | */ | |
2893 | if (new_offset < rdev->data_offset && | |
2894 | mddev->reshape_backwards) | |
2895 | return -EINVAL; | |
2896 | /* Increasing offset is inconsistent with forwards | |
2897 | * reshape. reshape_direction should be set to | |
2898 | * 'backwards' first. | |
2899 | */ | |
2900 | if (new_offset > rdev->data_offset && | |
2901 | !mddev->reshape_backwards) | |
2902 | return -EINVAL; | |
2903 | ||
2904 | if (mddev->pers && mddev->persistent && | |
2905 | !super_types[mddev->major_version] | |
2906 | .allow_new_offset(rdev, new_offset)) | |
2907 | return -E2BIG; | |
2908 | rdev->new_data_offset = new_offset; | |
2909 | if (new_offset > rdev->data_offset) | |
2910 | mddev->reshape_backwards = 1; | |
2911 | else if (new_offset < rdev->data_offset) | |
2912 | mddev->reshape_backwards = 0; | |
2913 | ||
2914 | return len; | |
2915 | } | |
2916 | static struct rdev_sysfs_entry rdev_new_offset = | |
2917 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2918 | ||
83303b61 | 2919 | static ssize_t |
3cb03002 | 2920 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2921 | { |
dd8ac336 | 2922 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2923 | } |
2924 | ||
c5d79adb N |
2925 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2926 | { | |
2927 | /* check if two start/length pairs overlap */ | |
2928 | if (s1+l1 <= s2) | |
2929 | return 0; | |
2930 | if (s2+l2 <= s1) | |
2931 | return 0; | |
2932 | return 1; | |
2933 | } | |
2934 | ||
b522adcd DW |
2935 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2936 | { | |
2937 | unsigned long long blocks; | |
2938 | sector_t new; | |
2939 | ||
b29bebd6 | 2940 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
2941 | return -EINVAL; |
2942 | ||
2943 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2944 | return -EINVAL; /* sector conversion overflow */ | |
2945 | ||
2946 | new = blocks * 2; | |
2947 | if (new != blocks * 2) | |
2948 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2949 | ||
2950 | *sectors = new; | |
2951 | return 0; | |
2952 | } | |
2953 | ||
83303b61 | 2954 | static ssize_t |
3cb03002 | 2955 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2956 | { |
fd01b88c | 2957 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2958 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2959 | sector_t sectors; |
27c529bb | 2960 | |
b522adcd | 2961 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2962 | return -EINVAL; |
c6563a8c N |
2963 | if (rdev->data_offset != rdev->new_data_offset) |
2964 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2965 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2966 | if (my_mddev->persistent) { |
dd8ac336 AN |
2967 | sectors = super_types[my_mddev->major_version]. |
2968 | rdev_size_change(rdev, sectors); | |
2969 | if (!sectors) | |
0cd17fec | 2970 | return -EBUSY; |
dd8ac336 | 2971 | } else if (!sectors) |
77304d2a | 2972 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2973 | rdev->data_offset; |
a6468539 N |
2974 | if (!my_mddev->pers->resize) |
2975 | /* Cannot change size for RAID0 or Linear etc */ | |
2976 | return -EINVAL; | |
0cd17fec | 2977 | } |
dd8ac336 | 2978 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2979 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2980 | |
dd8ac336 AN |
2981 | rdev->sectors = sectors; |
2982 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2983 | /* need to check that all other rdevs with the same ->bdev |
2984 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2985 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2986 | * we have to change it back, we will have the lock again. |
2987 | */ | |
fd01b88c | 2988 | struct mddev *mddev; |
c5d79adb | 2989 | int overlap = 0; |
159ec1fc | 2990 | struct list_head *tmp; |
c5d79adb | 2991 | |
27c529bb | 2992 | mddev_unlock(my_mddev); |
29ac4aa3 | 2993 | for_each_mddev(mddev, tmp) { |
3cb03002 | 2994 | struct md_rdev *rdev2; |
c5d79adb | 2995 | |
29f097c4 | 2996 | mddev_lock_nointr(mddev); |
dafb20fa | 2997 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
2998 | if (rdev->bdev == rdev2->bdev && |
2999 | rdev != rdev2 && | |
3000 | overlaps(rdev->data_offset, rdev->sectors, | |
3001 | rdev2->data_offset, | |
3002 | rdev2->sectors)) { | |
c5d79adb N |
3003 | overlap = 1; |
3004 | break; | |
3005 | } | |
3006 | mddev_unlock(mddev); | |
3007 | if (overlap) { | |
3008 | mddev_put(mddev); | |
3009 | break; | |
3010 | } | |
3011 | } | |
29f097c4 | 3012 | mddev_lock_nointr(my_mddev); |
c5d79adb N |
3013 | if (overlap) { |
3014 | /* Someone else could have slipped in a size | |
3015 | * change here, but doing so is just silly. | |
dd8ac336 | 3016 | * We put oldsectors back because we *know* it is |
c5d79adb N |
3017 | * safe, and trust userspace not to race with |
3018 | * itself | |
3019 | */ | |
dd8ac336 | 3020 | rdev->sectors = oldsectors; |
c5d79adb N |
3021 | return -EBUSY; |
3022 | } | |
3023 | } | |
83303b61 N |
3024 | return len; |
3025 | } | |
3026 | ||
3027 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 3028 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 3029 | |
06e3c817 | 3030 | |
3cb03002 | 3031 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
3032 | { |
3033 | unsigned long long recovery_start = rdev->recovery_offset; | |
3034 | ||
3035 | if (test_bit(In_sync, &rdev->flags) || | |
3036 | recovery_start == MaxSector) | |
3037 | return sprintf(page, "none\n"); | |
3038 | ||
3039 | return sprintf(page, "%llu\n", recovery_start); | |
3040 | } | |
3041 | ||
3cb03002 | 3042 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
3043 | { |
3044 | unsigned long long recovery_start; | |
3045 | ||
3046 | if (cmd_match(buf, "none")) | |
3047 | recovery_start = MaxSector; | |
b29bebd6 | 3048 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
3049 | return -EINVAL; |
3050 | ||
3051 | if (rdev->mddev->pers && | |
3052 | rdev->raid_disk >= 0) | |
3053 | return -EBUSY; | |
3054 | ||
3055 | rdev->recovery_offset = recovery_start; | |
3056 | if (recovery_start == MaxSector) | |
3057 | set_bit(In_sync, &rdev->flags); | |
3058 | else | |
3059 | clear_bit(In_sync, &rdev->flags); | |
3060 | return len; | |
3061 | } | |
3062 | ||
3063 | static struct rdev_sysfs_entry rdev_recovery_start = | |
3064 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
3065 | ||
16c791a5 N |
3066 | |
3067 | static ssize_t | |
3068 | badblocks_show(struct badblocks *bb, char *page, int unack); | |
3069 | static ssize_t | |
3070 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); | |
3071 | ||
3cb03002 | 3072 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3073 | { |
3074 | return badblocks_show(&rdev->badblocks, page, 0); | |
3075 | } | |
3cb03002 | 3076 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 3077 | { |
de393cde N |
3078 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
3079 | /* Maybe that ack was all we needed */ | |
3080 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
3081 | wake_up(&rdev->blocked_wait); | |
3082 | return rv; | |
16c791a5 N |
3083 | } |
3084 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
3085 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
3086 | ||
3087 | ||
3cb03002 | 3088 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3089 | { |
3090 | return badblocks_show(&rdev->badblocks, page, 1); | |
3091 | } | |
3cb03002 | 3092 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
3093 | { |
3094 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
3095 | } | |
3096 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
3097 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
3098 | ||
86e6ffdd N |
3099 | static struct attribute *rdev_default_attrs[] = { |
3100 | &rdev_state.attr, | |
4dbcdc75 | 3101 | &rdev_errors.attr, |
014236d2 | 3102 | &rdev_slot.attr, |
93c8cad0 | 3103 | &rdev_offset.attr, |
c6563a8c | 3104 | &rdev_new_offset.attr, |
83303b61 | 3105 | &rdev_size.attr, |
06e3c817 | 3106 | &rdev_recovery_start.attr, |
16c791a5 N |
3107 | &rdev_bad_blocks.attr, |
3108 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
3109 | NULL, |
3110 | }; | |
3111 | static ssize_t | |
3112 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3113 | { | |
3114 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3115 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
fd01b88c | 3116 | struct mddev *mddev = rdev->mddev; |
27c529bb | 3117 | ssize_t rv; |
86e6ffdd N |
3118 | |
3119 | if (!entry->show) | |
3120 | return -EIO; | |
27c529bb N |
3121 | |
3122 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
3123 | if (!rv) { | |
3124 | if (rdev->mddev == NULL) | |
3125 | rv = -EBUSY; | |
3126 | else | |
3127 | rv = entry->show(rdev, page); | |
3128 | mddev_unlock(mddev); | |
3129 | } | |
3130 | return rv; | |
86e6ffdd N |
3131 | } |
3132 | ||
3133 | static ssize_t | |
3134 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3135 | const char *page, size_t length) | |
3136 | { | |
3137 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3138 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3139 | ssize_t rv; |
fd01b88c | 3140 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3141 | |
3142 | if (!entry->store) | |
3143 | return -EIO; | |
67463acb N |
3144 | if (!capable(CAP_SYS_ADMIN)) |
3145 | return -EACCES; | |
27c529bb | 3146 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3147 | if (!rv) { |
27c529bb N |
3148 | if (rdev->mddev == NULL) |
3149 | rv = -EBUSY; | |
3150 | else | |
3151 | rv = entry->store(rdev, page, length); | |
6a51830e | 3152 | mddev_unlock(mddev); |
ca388059 N |
3153 | } |
3154 | return rv; | |
86e6ffdd N |
3155 | } |
3156 | ||
3157 | static void rdev_free(struct kobject *ko) | |
3158 | { | |
3cb03002 | 3159 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3160 | kfree(rdev); |
3161 | } | |
52cf25d0 | 3162 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3163 | .show = rdev_attr_show, |
3164 | .store = rdev_attr_store, | |
3165 | }; | |
3166 | static struct kobj_type rdev_ktype = { | |
3167 | .release = rdev_free, | |
3168 | .sysfs_ops = &rdev_sysfs_ops, | |
3169 | .default_attrs = rdev_default_attrs, | |
3170 | }; | |
3171 | ||
3cb03002 | 3172 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3173 | { |
3174 | rdev->desc_nr = -1; | |
3175 | rdev->saved_raid_disk = -1; | |
3176 | rdev->raid_disk = -1; | |
3177 | rdev->flags = 0; | |
3178 | rdev->data_offset = 0; | |
c6563a8c | 3179 | rdev->new_data_offset = 0; |
e8bb9a83 N |
3180 | rdev->sb_events = 0; |
3181 | rdev->last_read_error.tv_sec = 0; | |
3182 | rdev->last_read_error.tv_nsec = 0; | |
2699b672 N |
3183 | rdev->sb_loaded = 0; |
3184 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3185 | atomic_set(&rdev->nr_pending, 0); |
3186 | atomic_set(&rdev->read_errors, 0); | |
3187 | atomic_set(&rdev->corrected_errors, 0); | |
3188 | ||
3189 | INIT_LIST_HEAD(&rdev->same_set); | |
3190 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3191 | |
3192 | /* Add space to store bad block list. | |
3193 | * This reserves the space even on arrays where it cannot | |
3194 | * be used - I wonder if that matters | |
3195 | */ | |
3196 | rdev->badblocks.count = 0; | |
486adf72 | 3197 | rdev->badblocks.shift = -1; /* disabled until explicitly enabled */ |
2230dfe4 N |
3198 | rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3199 | seqlock_init(&rdev->badblocks.lock); | |
3200 | if (rdev->badblocks.page == NULL) | |
3201 | return -ENOMEM; | |
3202 | ||
3203 | return 0; | |
e8bb9a83 N |
3204 | } |
3205 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3206 | /* |
3207 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3208 | * | |
3209 | * mark the device faulty if: | |
3210 | * | |
3211 | * - the device is nonexistent (zero size) | |
3212 | * - the device has no valid superblock | |
3213 | * | |
3214 | * a faulty rdev _never_ has rdev->sb set. | |
3215 | */ | |
3cb03002 | 3216 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3217 | { |
3218 | char b[BDEVNAME_SIZE]; | |
3219 | int err; | |
3cb03002 | 3220 | struct md_rdev *rdev; |
1da177e4 LT |
3221 | sector_t size; |
3222 | ||
9ffae0cf | 3223 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
3224 | if (!rdev) { |
3225 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
3226 | return ERR_PTR(-ENOMEM); | |
3227 | } | |
1da177e4 | 3228 | |
2230dfe4 N |
3229 | err = md_rdev_init(rdev); |
3230 | if (err) | |
3231 | goto abort_free; | |
3232 | err = alloc_disk_sb(rdev); | |
3233 | if (err) | |
1da177e4 LT |
3234 | goto abort_free; |
3235 | ||
c5d79adb | 3236 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3237 | if (err) |
3238 | goto abort_free; | |
3239 | ||
f9cb074b | 3240 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3241 | |
77304d2a | 3242 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 LT |
3243 | if (!size) { |
3244 | printk(KERN_WARNING | |
3245 | "md: %s has zero or unknown size, marking faulty!\n", | |
3246 | bdevname(rdev->bdev,b)); | |
3247 | err = -EINVAL; | |
3248 | goto abort_free; | |
3249 | } | |
3250 | ||
3251 | if (super_format >= 0) { | |
3252 | err = super_types[super_format]. | |
3253 | load_super(rdev, NULL, super_minor); | |
3254 | if (err == -EINVAL) { | |
df968c4e N |
3255 | printk(KERN_WARNING |
3256 | "md: %s does not have a valid v%d.%d " | |
3257 | "superblock, not importing!\n", | |
3258 | bdevname(rdev->bdev,b), | |
3259 | super_format, super_minor); | |
1da177e4 LT |
3260 | goto abort_free; |
3261 | } | |
3262 | if (err < 0) { | |
3263 | printk(KERN_WARNING | |
3264 | "md: could not read %s's sb, not importing!\n", | |
3265 | bdevname(rdev->bdev,b)); | |
3266 | goto abort_free; | |
3267 | } | |
3268 | } | |
6bfe0b49 | 3269 | |
1da177e4 LT |
3270 | return rdev; |
3271 | ||
3272 | abort_free: | |
2699b672 N |
3273 | if (rdev->bdev) |
3274 | unlock_rdev(rdev); | |
545c8795 | 3275 | md_rdev_clear(rdev); |
1da177e4 LT |
3276 | kfree(rdev); |
3277 | return ERR_PTR(err); | |
3278 | } | |
3279 | ||
3280 | /* | |
3281 | * Check a full RAID array for plausibility | |
3282 | */ | |
3283 | ||
3284 | ||
fd01b88c | 3285 | static void analyze_sbs(struct mddev * mddev) |
1da177e4 LT |
3286 | { |
3287 | int i; | |
3cb03002 | 3288 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3289 | char b[BDEVNAME_SIZE]; |
3290 | ||
3291 | freshest = NULL; | |
dafb20fa | 3292 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3293 | switch (super_types[mddev->major_version]. |
3294 | load_super(rdev, freshest, mddev->minor_version)) { | |
3295 | case 1: | |
3296 | freshest = rdev; | |
3297 | break; | |
3298 | case 0: | |
3299 | break; | |
3300 | default: | |
3301 | printk( KERN_ERR \ | |
3302 | "md: fatal superblock inconsistency in %s" | |
3303 | " -- removing from array\n", | |
3304 | bdevname(rdev->bdev,b)); | |
3305 | kick_rdev_from_array(rdev); | |
3306 | } | |
3307 | ||
3308 | ||
3309 | super_types[mddev->major_version]. | |
3310 | validate_super(mddev, freshest); | |
3311 | ||
3312 | i = 0; | |
dafb20fa | 3313 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3314 | if (mddev->max_disks && |
3315 | (rdev->desc_nr >= mddev->max_disks || | |
3316 | i > mddev->max_disks)) { | |
de01dfad N |
3317 | printk(KERN_WARNING |
3318 | "md: %s: %s: only %d devices permitted\n", | |
3319 | mdname(mddev), bdevname(rdev->bdev, b), | |
3320 | mddev->max_disks); | |
3321 | kick_rdev_from_array(rdev); | |
3322 | continue; | |
3323 | } | |
1da177e4 LT |
3324 | if (rdev != freshest) |
3325 | if (super_types[mddev->major_version]. | |
3326 | validate_super(mddev, rdev)) { | |
3327 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
3328 | " from array!\n", | |
3329 | bdevname(rdev->bdev,b)); | |
3330 | kick_rdev_from_array(rdev); | |
3331 | continue; | |
3332 | } | |
3333 | if (mddev->level == LEVEL_MULTIPATH) { | |
3334 | rdev->desc_nr = i++; | |
3335 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3336 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 3337 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
3338 | rdev->raid_disk = -1; |
3339 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3340 | } |
3341 | } | |
1da177e4 LT |
3342 | } |
3343 | ||
72e02075 N |
3344 | /* Read a fixed-point number. |
3345 | * Numbers in sysfs attributes should be in "standard" units where | |
3346 | * possible, so time should be in seconds. | |
3347 | * However we internally use a a much smaller unit such as | |
3348 | * milliseconds or jiffies. | |
3349 | * This function takes a decimal number with a possible fractional | |
3350 | * component, and produces an integer which is the result of | |
3351 | * multiplying that number by 10^'scale'. | |
3352 | * all without any floating-point arithmetic. | |
3353 | */ | |
3354 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3355 | { | |
3356 | unsigned long result = 0; | |
3357 | long decimals = -1; | |
3358 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3359 | if (*cp == '.') | |
3360 | decimals = 0; | |
3361 | else if (decimals < scale) { | |
3362 | unsigned int value; | |
3363 | value = *cp - '0'; | |
3364 | result = result * 10 + value; | |
3365 | if (decimals >= 0) | |
3366 | decimals++; | |
3367 | } | |
3368 | cp++; | |
3369 | } | |
3370 | if (*cp == '\n') | |
3371 | cp++; | |
3372 | if (*cp) | |
3373 | return -EINVAL; | |
3374 | if (decimals < 0) | |
3375 | decimals = 0; | |
3376 | while (decimals < scale) { | |
3377 | result *= 10; | |
3378 | decimals ++; | |
3379 | } | |
3380 | *res = result; | |
3381 | return 0; | |
3382 | } | |
3383 | ||
3384 | ||
19052c0e N |
3385 | static void md_safemode_timeout(unsigned long data); |
3386 | ||
16f17b39 | 3387 | static ssize_t |
fd01b88c | 3388 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3389 | { |
3390 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3391 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3392 | } | |
3393 | static ssize_t | |
fd01b88c | 3394 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3395 | { |
16f17b39 | 3396 | unsigned long msec; |
97ce0a7f | 3397 | |
72e02075 | 3398 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3399 | return -EINVAL; |
16f17b39 N |
3400 | if (msec == 0) |
3401 | mddev->safemode_delay = 0; | |
3402 | else { | |
19052c0e | 3403 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
3404 | mddev->safemode_delay = (msec*HZ)/1000; |
3405 | if (mddev->safemode_delay == 0) | |
3406 | mddev->safemode_delay = 1; | |
275c51c4 | 3407 | if (mddev->safemode_delay < old_delay || old_delay == 0) |
19052c0e | 3408 | md_safemode_timeout((unsigned long)mddev); |
16f17b39 N |
3409 | } |
3410 | return len; | |
3411 | } | |
3412 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3413 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3414 | |
eae1701f | 3415 | static ssize_t |
fd01b88c | 3416 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3417 | { |
84fc4b56 | 3418 | struct md_personality *p = mddev->pers; |
d9d166c2 | 3419 | if (p) |
eae1701f | 3420 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
3421 | else if (mddev->clevel[0]) |
3422 | return sprintf(page, "%s\n", mddev->clevel); | |
3423 | else if (mddev->level != LEVEL_NONE) | |
3424 | return sprintf(page, "%d\n", mddev->level); | |
3425 | else | |
3426 | return 0; | |
eae1701f N |
3427 | } |
3428 | ||
d9d166c2 | 3429 | static ssize_t |
fd01b88c | 3430 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3431 | { |
f2859af6 | 3432 | char clevel[16]; |
20a49ff6 | 3433 | ssize_t rv = len; |
84fc4b56 | 3434 | struct md_personality *pers; |
f2859af6 | 3435 | long level; |
245f46c2 | 3436 | void *priv; |
3cb03002 | 3437 | struct md_rdev *rdev; |
245f46c2 N |
3438 | |
3439 | if (mddev->pers == NULL) { | |
3440 | if (len == 0) | |
3441 | return 0; | |
3442 | if (len >= sizeof(mddev->clevel)) | |
3443 | return -ENOSPC; | |
3444 | strncpy(mddev->clevel, buf, len); | |
3445 | if (mddev->clevel[len-1] == '\n') | |
3446 | len--; | |
3447 | mddev->clevel[len] = 0; | |
3448 | mddev->level = LEVEL_NONE; | |
3449 | return rv; | |
3450 | } | |
3451 | ||
3452 | /* request to change the personality. Need to ensure: | |
3453 | * - array is not engaged in resync/recovery/reshape | |
3454 | * - old personality can be suspended | |
3455 | * - new personality will access other array. | |
3456 | */ | |
3457 | ||
bb4f1e9d N |
3458 | if (mddev->sync_thread || |
3459 | mddev->reshape_position != MaxSector || | |
3460 | mddev->sysfs_active) | |
d9d166c2 | 3461 | return -EBUSY; |
245f46c2 N |
3462 | |
3463 | if (!mddev->pers->quiesce) { | |
3464 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3465 | mdname(mddev), mddev->pers->name); | |
3466 | return -EINVAL; | |
3467 | } | |
3468 | ||
3469 | /* Now find the new personality */ | |
f2859af6 | 3470 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3471 | return -EINVAL; |
f2859af6 DW |
3472 | strncpy(clevel, buf, len); |
3473 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3474 | len--; |
f2859af6 | 3475 | clevel[len] = 0; |
b29bebd6 | 3476 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3477 | level = LEVEL_NONE; |
245f46c2 | 3478 | |
f2859af6 DW |
3479 | if (request_module("md-%s", clevel) != 0) |
3480 | request_module("md-level-%s", clevel); | |
245f46c2 | 3481 | spin_lock(&pers_lock); |
f2859af6 | 3482 | pers = find_pers(level, clevel); |
245f46c2 N |
3483 | if (!pers || !try_module_get(pers->owner)) { |
3484 | spin_unlock(&pers_lock); | |
f2859af6 | 3485 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3486 | return -EINVAL; |
3487 | } | |
3488 | spin_unlock(&pers_lock); | |
3489 | ||
3490 | if (pers == mddev->pers) { | |
3491 | /* Nothing to do! */ | |
3492 | module_put(pers->owner); | |
3493 | return rv; | |
3494 | } | |
3495 | if (!pers->takeover) { | |
3496 | module_put(pers->owner); | |
3497 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3498 | mdname(mddev), clevel); |
245f46c2 N |
3499 | return -EINVAL; |
3500 | } | |
3501 | ||
dafb20fa | 3502 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3503 | rdev->new_raid_disk = rdev->raid_disk; |
3504 | ||
245f46c2 N |
3505 | /* ->takeover must set new_* and/or delta_disks |
3506 | * if it succeeds, and may set them when it fails. | |
3507 | */ | |
3508 | priv = pers->takeover(mddev); | |
3509 | if (IS_ERR(priv)) { | |
3510 | mddev->new_level = mddev->level; | |
3511 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3512 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3513 | mddev->raid_disks -= mddev->delta_disks; |
3514 | mddev->delta_disks = 0; | |
2c810cdd | 3515 | mddev->reshape_backwards = 0; |
245f46c2 N |
3516 | module_put(pers->owner); |
3517 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3518 | mdname(mddev), clevel); |
245f46c2 N |
3519 | return PTR_ERR(priv); |
3520 | } | |
3521 | ||
3522 | /* Looks like we have a winner */ | |
3523 | mddev_suspend(mddev); | |
3524 | mddev->pers->stop(mddev); | |
a64c876f N |
3525 | |
3526 | if (mddev->pers->sync_request == NULL && | |
3527 | pers->sync_request != NULL) { | |
3528 | /* need to add the md_redundancy_group */ | |
3529 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3530 | printk(KERN_WARNING | |
3531 | "md: cannot register extra attributes for %s\n", | |
3532 | mdname(mddev)); | |
388975cc | 3533 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
a64c876f N |
3534 | } |
3535 | if (mddev->pers->sync_request != NULL && | |
3536 | pers->sync_request == NULL) { | |
3537 | /* need to remove the md_redundancy_group */ | |
3538 | if (mddev->to_remove == NULL) | |
3539 | mddev->to_remove = &md_redundancy_group; | |
3540 | } | |
3541 | ||
54071b38 TM |
3542 | if (mddev->pers->sync_request == NULL && |
3543 | mddev->external) { | |
3544 | /* We are converting from a no-redundancy array | |
3545 | * to a redundancy array and metadata is managed | |
3546 | * externally so we need to be sure that writes | |
3547 | * won't block due to a need to transition | |
3548 | * clean->dirty | |
3549 | * until external management is started. | |
3550 | */ | |
3551 | mddev->in_sync = 0; | |
3552 | mddev->safemode_delay = 0; | |
3553 | mddev->safemode = 0; | |
3554 | } | |
3555 | ||
dafb20fa | 3556 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3557 | if (rdev->raid_disk < 0) |
3558 | continue; | |
bf2cb0da | 3559 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3560 | rdev->new_raid_disk = -1; |
3561 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3562 | continue; | |
36fad858 | 3563 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3564 | } |
dafb20fa | 3565 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3566 | if (rdev->raid_disk < 0) |
3567 | continue; | |
3568 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3569 | continue; | |
3570 | rdev->raid_disk = rdev->new_raid_disk; | |
3571 | if (rdev->raid_disk < 0) | |
3a981b03 | 3572 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3573 | else { |
36fad858 NK |
3574 | if (sysfs_link_rdev(mddev, rdev)) |
3575 | printk(KERN_WARNING "md: cannot register rd%d" | |
3576 | " for %s after level change\n", | |
3577 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3578 | } |
e93f68a1 N |
3579 | } |
3580 | ||
3581 | module_put(mddev->pers->owner); | |
245f46c2 N |
3582 | mddev->pers = pers; |
3583 | mddev->private = priv; | |
3584 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3585 | mddev->level = mddev->new_level; | |
3586 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3587 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3588 | mddev->delta_disks = 0; |
2c810cdd | 3589 | mddev->reshape_backwards = 0; |
fee68723 | 3590 | mddev->degraded = 0; |
9af204cf TM |
3591 | if (mddev->pers->sync_request == NULL) { |
3592 | /* this is now an array without redundancy, so | |
3593 | * it must always be in_sync | |
3594 | */ | |
3595 | mddev->in_sync = 1; | |
3596 | del_timer_sync(&mddev->safemode_timer); | |
3597 | } | |
02e5f5c0 | 3598 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3599 | pers->run(mddev); |
245f46c2 | 3600 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
47525e59 | 3601 | mddev_resume(mddev); |
5cac7861 | 3602 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3603 | md_new_event(mddev); |
d9d166c2 N |
3604 | return rv; |
3605 | } | |
3606 | ||
3607 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3608 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3609 | |
d4dbd025 N |
3610 | |
3611 | static ssize_t | |
fd01b88c | 3612 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3613 | { |
3614 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3615 | if (mddev->reshape_position != MaxSector && |
3616 | mddev->layout != mddev->new_layout) | |
3617 | return sprintf(page, "%d (%d)\n", | |
3618 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3619 | return sprintf(page, "%d\n", mddev->layout); |
3620 | } | |
3621 | ||
3622 | static ssize_t | |
fd01b88c | 3623 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 N |
3624 | { |
3625 | char *e; | |
3626 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3627 | |
3628 | if (!*buf || (*e && *e != '\n')) | |
3629 | return -EINVAL; | |
3630 | ||
b3546035 N |
3631 | if (mddev->pers) { |
3632 | int err; | |
50ac168a | 3633 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3634 | return -EBUSY; |
597a711b | 3635 | mddev->new_layout = n; |
50ac168a | 3636 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3637 | if (err) { |
3638 | mddev->new_layout = mddev->layout; | |
b3546035 | 3639 | return err; |
597a711b | 3640 | } |
b3546035 | 3641 | } else { |
08a02ecd | 3642 | mddev->new_layout = n; |
b3546035 N |
3643 | if (mddev->reshape_position == MaxSector) |
3644 | mddev->layout = n; | |
3645 | } | |
d4dbd025 N |
3646 | return len; |
3647 | } | |
3648 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3649 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3650 | |
3651 | ||
eae1701f | 3652 | static ssize_t |
fd01b88c | 3653 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3654 | { |
bb636547 N |
3655 | if (mddev->raid_disks == 0) |
3656 | return 0; | |
08a02ecd N |
3657 | if (mddev->reshape_position != MaxSector && |
3658 | mddev->delta_disks != 0) | |
3659 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3660 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3661 | return sprintf(page, "%d\n", mddev->raid_disks); |
3662 | } | |
3663 | ||
fd01b88c | 3664 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3665 | |
3666 | static ssize_t | |
fd01b88c | 3667 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3668 | { |
da943b99 N |
3669 | char *e; |
3670 | int rv = 0; | |
3671 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3672 | ||
3673 | if (!*buf || (*e && *e != '\n')) | |
3674 | return -EINVAL; | |
3675 | ||
3676 | if (mddev->pers) | |
3677 | rv = update_raid_disks(mddev, n); | |
08a02ecd | 3678 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3679 | struct md_rdev *rdev; |
08a02ecd | 3680 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c N |
3681 | |
3682 | rdev_for_each(rdev, mddev) { | |
3683 | if (olddisks < n && | |
3684 | rdev->data_offset < rdev->new_data_offset) | |
3685 | return -EINVAL; | |
3686 | if (olddisks > n && | |
3687 | rdev->data_offset > rdev->new_data_offset) | |
3688 | return -EINVAL; | |
3689 | } | |
08a02ecd N |
3690 | mddev->delta_disks = n - olddisks; |
3691 | mddev->raid_disks = n; | |
2c810cdd | 3692 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3693 | } else |
da943b99 N |
3694 | mddev->raid_disks = n; |
3695 | return rv ? rv : len; | |
3696 | } | |
3697 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3698 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3699 | |
3b34380a | 3700 | static ssize_t |
fd01b88c | 3701 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3702 | { |
08a02ecd | 3703 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3704 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3705 | return sprintf(page, "%d (%d)\n", | |
3706 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3707 | mddev->chunk_sectors << 9); |
3708 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3709 | } |
3710 | ||
3711 | static ssize_t | |
fd01b88c | 3712 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3713 | { |
3b34380a N |
3714 | char *e; |
3715 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3716 | ||
3b34380a N |
3717 | if (!*buf || (*e && *e != '\n')) |
3718 | return -EINVAL; | |
3719 | ||
b3546035 N |
3720 | if (mddev->pers) { |
3721 | int err; | |
50ac168a | 3722 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3723 | return -EBUSY; |
597a711b | 3724 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3725 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3726 | if (err) { |
3727 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3728 | return err; |
597a711b | 3729 | } |
b3546035 | 3730 | } else { |
664e7c41 | 3731 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3732 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3733 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3734 | } |
3b34380a N |
3735 | return len; |
3736 | } | |
3737 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3738 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3739 | |
a94213b1 | 3740 | static ssize_t |
fd01b88c | 3741 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3742 | { |
d1a7c503 N |
3743 | if (mddev->recovery_cp == MaxSector) |
3744 | return sprintf(page, "none\n"); | |
a94213b1 N |
3745 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3746 | } | |
3747 | ||
3748 | static ssize_t | |
fd01b88c | 3749 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3750 | { |
a94213b1 N |
3751 | char *e; |
3752 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3753 | ||
b098636c | 3754 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
a94213b1 | 3755 | return -EBUSY; |
06e3c817 DW |
3756 | if (cmd_match(buf, "none")) |
3757 | n = MaxSector; | |
3758 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3759 | return -EINVAL; |
3760 | ||
3761 | mddev->recovery_cp = n; | |
db07d85e N |
3762 | if (mddev->pers) |
3763 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
a94213b1 N |
3764 | return len; |
3765 | } | |
3766 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3767 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3768 | |
9e653b63 N |
3769 | /* |
3770 | * The array state can be: | |
3771 | * | |
3772 | * clear | |
3773 | * No devices, no size, no level | |
3774 | * Equivalent to STOP_ARRAY ioctl | |
3775 | * inactive | |
3776 | * May have some settings, but array is not active | |
3777 | * all IO results in error | |
3778 | * When written, doesn't tear down array, but just stops it | |
3779 | * suspended (not supported yet) | |
3780 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3781 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3782 | * readonly |
3783 | * no resync can happen. no superblocks get written. | |
3784 | * write requests fail | |
3785 | * read-auto | |
3786 | * like readonly, but behaves like 'clean' on a write request. | |
3787 | * | |
3788 | * clean - no pending writes, but otherwise active. | |
3789 | * When written to inactive array, starts without resync | |
3790 | * If a write request arrives then | |
3791 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3792 | * if not known, block and switch to write-pending | |
3793 | * If written to an active array that has pending writes, then fails. | |
3794 | * active | |
3795 | * fully active: IO and resync can be happening. | |
3796 | * When written to inactive array, starts with resync | |
3797 | * | |
3798 | * write-pending | |
3799 | * clean, but writes are blocked waiting for 'active' to be written. | |
3800 | * | |
3801 | * active-idle | |
3802 | * like active, but no writes have been seen for a while (100msec). | |
3803 | * | |
3804 | */ | |
3805 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3806 | write_pending, active_idle, bad_word}; | |
05381954 | 3807 | static char *array_states[] = { |
9e653b63 N |
3808 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3809 | "write-pending", "active-idle", NULL }; | |
3810 | ||
3811 | static int match_word(const char *word, char **list) | |
3812 | { | |
3813 | int n; | |
3814 | for (n=0; list[n]; n++) | |
3815 | if (cmd_match(word, list[n])) | |
3816 | break; | |
3817 | return n; | |
3818 | } | |
3819 | ||
3820 | static ssize_t | |
fd01b88c | 3821 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3822 | { |
3823 | enum array_state st = inactive; | |
3824 | ||
3825 | if (mddev->pers) | |
3826 | switch(mddev->ro) { | |
3827 | case 1: | |
3828 | st = readonly; | |
3829 | break; | |
3830 | case 2: | |
3831 | st = read_auto; | |
3832 | break; | |
3833 | case 0: | |
3834 | if (mddev->in_sync) | |
3835 | st = clean; | |
070dc6dd | 3836 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3837 | st = write_pending; |
9e653b63 N |
3838 | else if (mddev->safemode) |
3839 | st = active_idle; | |
3840 | else | |
3841 | st = active; | |
3842 | } | |
3843 | else { | |
3844 | if (list_empty(&mddev->disks) && | |
3845 | mddev->raid_disks == 0 && | |
58c0fed4 | 3846 | mddev->dev_sectors == 0) |
9e653b63 N |
3847 | st = clear; |
3848 | else | |
3849 | st = inactive; | |
3850 | } | |
3851 | return sprintf(page, "%s\n", array_states[st]); | |
3852 | } | |
3853 | ||
a05b7ea0 N |
3854 | static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev); |
3855 | static int md_set_readonly(struct mddev * mddev, struct block_device *bdev); | |
fd01b88c N |
3856 | static int do_md_run(struct mddev * mddev); |
3857 | static int restart_array(struct mddev *mddev); | |
9e653b63 N |
3858 | |
3859 | static ssize_t | |
fd01b88c | 3860 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 N |
3861 | { |
3862 | int err = -EINVAL; | |
3863 | enum array_state st = match_word(buf, array_states); | |
3864 | switch(st) { | |
3865 | case bad_word: | |
3866 | break; | |
3867 | case clear: | |
3868 | /* stopping an active array */ | |
a05b7ea0 | 3869 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3870 | break; |
3871 | case inactive: | |
3872 | /* stopping an active array */ | |
90cf195d | 3873 | if (mddev->pers) |
a05b7ea0 | 3874 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3875 | else |
e691063a | 3876 | err = 0; /* already inactive */ |
9e653b63 N |
3877 | break; |
3878 | case suspended: | |
3879 | break; /* not supported yet */ | |
3880 | case readonly: | |
3881 | if (mddev->pers) | |
a05b7ea0 | 3882 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3883 | else { |
3884 | mddev->ro = 1; | |
648b629e | 3885 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3886 | err = do_md_run(mddev); |
3887 | } | |
3888 | break; | |
3889 | case read_auto: | |
9e653b63 | 3890 | if (mddev->pers) { |
80268ee9 | 3891 | if (mddev->ro == 0) |
a05b7ea0 | 3892 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3893 | else if (mddev->ro == 1) |
648b629e N |
3894 | err = restart_array(mddev); |
3895 | if (err == 0) { | |
3896 | mddev->ro = 2; | |
3897 | set_disk_ro(mddev->gendisk, 0); | |
3898 | } | |
9e653b63 N |
3899 | } else { |
3900 | mddev->ro = 2; | |
3901 | err = do_md_run(mddev); | |
3902 | } | |
3903 | break; | |
3904 | case clean: | |
3905 | if (mddev->pers) { | |
3906 | restart_array(mddev); | |
3907 | spin_lock_irq(&mddev->write_lock); | |
3908 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3909 | if (mddev->in_sync == 0) { |
3910 | mddev->in_sync = 1; | |
31a59e34 N |
3911 | if (mddev->safemode == 1) |
3912 | mddev->safemode = 0; | |
070dc6dd | 3913 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3914 | } |
3915 | err = 0; | |
3916 | } else | |
3917 | err = -EBUSY; | |
9e653b63 | 3918 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3919 | } else |
3920 | err = -EINVAL; | |
9e653b63 N |
3921 | break; |
3922 | case active: | |
3923 | if (mddev->pers) { | |
3924 | restart_array(mddev); | |
070dc6dd | 3925 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3926 | wake_up(&mddev->sb_wait); |
3927 | err = 0; | |
3928 | } else { | |
3929 | mddev->ro = 0; | |
648b629e | 3930 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3931 | err = do_md_run(mddev); |
3932 | } | |
3933 | break; | |
3934 | case write_pending: | |
3935 | case active_idle: | |
3936 | /* these cannot be set */ | |
3937 | break; | |
3938 | } | |
3939 | if (err) | |
3940 | return err; | |
0fd62b86 | 3941 | else { |
1d23f178 N |
3942 | if (mddev->hold_active == UNTIL_IOCTL) |
3943 | mddev->hold_active = 0; | |
00bcb4ac | 3944 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3945 | return len; |
0fd62b86 | 3946 | } |
9e653b63 | 3947 | } |
80ca3a44 N |
3948 | static struct md_sysfs_entry md_array_state = |
3949 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3950 | |
1e50915f | 3951 | static ssize_t |
fd01b88c | 3952 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
3953 | return sprintf(page, "%d\n", |
3954 | atomic_read(&mddev->max_corr_read_errors)); | |
3955 | } | |
3956 | ||
3957 | static ssize_t | |
fd01b88c | 3958 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f RB |
3959 | { |
3960 | char *e; | |
3961 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3962 | ||
3963 | if (*buf && (*e == 0 || *e == '\n')) { | |
3964 | atomic_set(&mddev->max_corr_read_errors, n); | |
3965 | return len; | |
3966 | } | |
3967 | return -EINVAL; | |
3968 | } | |
3969 | ||
3970 | static struct md_sysfs_entry max_corr_read_errors = | |
3971 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3972 | max_corrected_read_errors_store); | |
3973 | ||
6d7ff738 | 3974 | static ssize_t |
fd01b88c | 3975 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
3976 | { |
3977 | return -EINVAL; | |
3978 | } | |
3979 | ||
3980 | static ssize_t | |
fd01b88c | 3981 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
3982 | { |
3983 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3984 | /* The new device is added to the array. | |
3985 | * If the array has a persistent superblock, we read the | |
3986 | * superblock to initialise info and check validity. | |
3987 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3988 | * which mainly checks size. | |
3989 | */ | |
3990 | char *e; | |
3991 | int major = simple_strtoul(buf, &e, 10); | |
3992 | int minor; | |
3993 | dev_t dev; | |
3cb03002 | 3994 | struct md_rdev *rdev; |
6d7ff738 N |
3995 | int err; |
3996 | ||
3997 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3998 | return -EINVAL; | |
3999 | minor = simple_strtoul(e+1, &e, 10); | |
4000 | if (*e && *e != '\n') | |
4001 | return -EINVAL; | |
4002 | dev = MKDEV(major, minor); | |
4003 | if (major != MAJOR(dev) || | |
4004 | minor != MINOR(dev)) | |
4005 | return -EOVERFLOW; | |
4006 | ||
4007 | ||
4008 | if (mddev->persistent) { | |
4009 | rdev = md_import_device(dev, mddev->major_version, | |
4010 | mddev->minor_version); | |
4011 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4012 | struct md_rdev *rdev0 |
4013 | = list_entry(mddev->disks.next, | |
4014 | struct md_rdev, same_set); | |
6d7ff738 N |
4015 | err = super_types[mddev->major_version] |
4016 | .load_super(rdev, rdev0, mddev->minor_version); | |
4017 | if (err < 0) | |
4018 | goto out; | |
4019 | } | |
c5d79adb N |
4020 | } else if (mddev->external) |
4021 | rdev = md_import_device(dev, -2, -1); | |
4022 | else | |
6d7ff738 N |
4023 | rdev = md_import_device(dev, -1, -1); |
4024 | ||
4025 | if (IS_ERR(rdev)) | |
4026 | return PTR_ERR(rdev); | |
4027 | err = bind_rdev_to_array(rdev, mddev); | |
4028 | out: | |
4029 | if (err) | |
4030 | export_rdev(rdev); | |
4031 | return err ? err : len; | |
4032 | } | |
4033 | ||
4034 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4035 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4036 | |
9b1d1dac | 4037 | static ssize_t |
fd01b88c | 4038 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4039 | { |
4040 | char *end; | |
4041 | unsigned long chunk, end_chunk; | |
4042 | ||
4043 | if (!mddev->bitmap) | |
4044 | goto out; | |
4045 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4046 | while (*buf) { | |
4047 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4048 | if (buf == end) break; | |
4049 | if (*end == '-') { /* range */ | |
4050 | buf = end + 1; | |
4051 | end_chunk = simple_strtoul(buf, &end, 0); | |
4052 | if (buf == end) break; | |
4053 | } | |
4054 | if (*end && !isspace(*end)) break; | |
4055 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4056 | buf = skip_spaces(end); |
9b1d1dac PC |
4057 | } |
4058 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4059 | out: | |
4060 | return len; | |
4061 | } | |
4062 | ||
4063 | static struct md_sysfs_entry md_bitmap = | |
4064 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4065 | ||
a35b0d69 | 4066 | static ssize_t |
fd01b88c | 4067 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4068 | { |
58c0fed4 AN |
4069 | return sprintf(page, "%llu\n", |
4070 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4071 | } |
4072 | ||
fd01b88c | 4073 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4074 | |
4075 | static ssize_t | |
fd01b88c | 4076 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4077 | { |
4078 | /* If array is inactive, we can reduce the component size, but | |
4079 | * not increase it (except from 0). | |
4080 | * If array is active, we can try an on-line resize | |
4081 | */ | |
b522adcd DW |
4082 | sector_t sectors; |
4083 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4084 | |
58c0fed4 AN |
4085 | if (err < 0) |
4086 | return err; | |
a35b0d69 | 4087 | if (mddev->pers) { |
58c0fed4 | 4088 | err = update_size(mddev, sectors); |
850b2b42 | 4089 | md_update_sb(mddev, 1); |
a35b0d69 | 4090 | } else { |
58c0fed4 AN |
4091 | if (mddev->dev_sectors == 0 || |
4092 | mddev->dev_sectors > sectors) | |
4093 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4094 | else |
4095 | err = -ENOSPC; | |
4096 | } | |
4097 | return err ? err : len; | |
4098 | } | |
4099 | ||
4100 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4101 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4102 | |
8bb93aac | 4103 | |
83f0d77a | 4104 | /* Metadata version. |
e691063a N |
4105 | * This is one of |
4106 | * 'none' for arrays with no metadata (good luck...) | |
4107 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4108 | * or N.M for internally known formats |
4109 | */ | |
4110 | static ssize_t | |
fd01b88c | 4111 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4112 | { |
4113 | if (mddev->persistent) | |
4114 | return sprintf(page, "%d.%d\n", | |
4115 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4116 | else if (mddev->external) |
4117 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4118 | else |
4119 | return sprintf(page, "none\n"); | |
4120 | } | |
4121 | ||
4122 | static ssize_t | |
fd01b88c | 4123 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4124 | { |
4125 | int major, minor; | |
4126 | char *e; | |
ea43ddd8 N |
4127 | /* Changing the details of 'external' metadata is |
4128 | * always permitted. Otherwise there must be | |
4129 | * no devices attached to the array. | |
4130 | */ | |
4131 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
4132 | ; | |
4133 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
4134 | return -EBUSY; |
4135 | ||
4136 | if (cmd_match(buf, "none")) { | |
4137 | mddev->persistent = 0; | |
e691063a N |
4138 | mddev->external = 0; |
4139 | mddev->major_version = 0; | |
4140 | mddev->minor_version = 90; | |
4141 | return len; | |
4142 | } | |
4143 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4144 | size_t namelen = len-9; |
e691063a N |
4145 | if (namelen >= sizeof(mddev->metadata_type)) |
4146 | namelen = sizeof(mddev->metadata_type)-1; | |
4147 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4148 | mddev->metadata_type[namelen] = 0; | |
4149 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4150 | mddev->metadata_type[--namelen] = 0; | |
4151 | mddev->persistent = 0; | |
4152 | mddev->external = 1; | |
8bb93aac N |
4153 | mddev->major_version = 0; |
4154 | mddev->minor_version = 90; | |
4155 | return len; | |
4156 | } | |
4157 | major = simple_strtoul(buf, &e, 10); | |
4158 | if (e==buf || *e != '.') | |
4159 | return -EINVAL; | |
4160 | buf = e+1; | |
4161 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4162 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 4163 | return -EINVAL; |
50511da3 | 4164 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
4165 | return -ENOENT; |
4166 | mddev->major_version = major; | |
4167 | mddev->minor_version = minor; | |
4168 | mddev->persistent = 1; | |
e691063a | 4169 | mddev->external = 0; |
8bb93aac N |
4170 | return len; |
4171 | } | |
4172 | ||
4173 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 4174 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4175 | |
24dd469d | 4176 | static ssize_t |
fd01b88c | 4177 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4178 | { |
7eec314d | 4179 | char *type = "idle"; |
b6a9ce68 N |
4180 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
4181 | type = "frozen"; | |
4182 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 4183 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
4184 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4185 | type = "reshape"; | |
4186 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
4187 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
4188 | type = "resync"; | |
4189 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
4190 | type = "check"; | |
4191 | else | |
4192 | type = "repair"; | |
72a23c21 | 4193 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
4194 | type = "recover"; |
4195 | } | |
4196 | return sprintf(page, "%s\n", type); | |
4197 | } | |
4198 | ||
4199 | static ssize_t | |
fd01b88c | 4200 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4201 | { |
7eec314d N |
4202 | if (!mddev->pers || !mddev->pers->sync_request) |
4203 | return -EINVAL; | |
4204 | ||
b6a9ce68 N |
4205 | if (cmd_match(page, "frozen")) |
4206 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4207 | else | |
4208 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4209 | ||
4210 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
4211 | if (mddev->sync_thread) { |
4212 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
a91d5ac0 | 4213 | md_reap_sync_thread(mddev); |
7eec314d | 4214 | } |
03c902e1 N |
4215 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
4216 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 4217 | return -EBUSY; |
72a23c21 NB |
4218 | else if (cmd_match(page, "resync")) |
4219 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4220 | else if (cmd_match(page, "recover")) { | |
4221 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 4222 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 4223 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4224 | int err; |
4225 | if (mddev->pers->start_reshape == NULL) | |
4226 | return -EINVAL; | |
4227 | err = mddev->pers->start_reshape(mddev); | |
4228 | if (err) | |
4229 | return err; | |
a99ac971 | 4230 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4231 | } else { |
bce74dac | 4232 | if (cmd_match(page, "check")) |
7eec314d | 4233 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4234 | else if (!cmd_match(page, "repair")) |
7eec314d N |
4235 | return -EINVAL; |
4236 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
4237 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4238 | } |
48c26ddc N |
4239 | if (mddev->ro == 2) { |
4240 | /* A write to sync_action is enough to justify | |
4241 | * canceling read-auto mode | |
4242 | */ | |
4243 | mddev->ro = 0; | |
4244 | md_wakeup_thread(mddev->sync_thread); | |
4245 | } | |
03c902e1 | 4246 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4247 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4248 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4249 | return len; |
4250 | } | |
4251 | ||
c4a39551 JB |
4252 | static struct md_sysfs_entry md_scan_mode = |
4253 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
4254 | ||
4255 | static ssize_t | |
4256 | last_sync_action_show(struct mddev *mddev, char *page) | |
4257 | { | |
4258 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4259 | } | |
4260 | ||
4261 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4262 | ||
9d88883e | 4263 | static ssize_t |
fd01b88c | 4264 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4265 | { |
4266 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4267 | (unsigned long long) |
4268 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4269 | } |
4270 | ||
80ca3a44 | 4271 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4272 | |
88202a0c | 4273 | static ssize_t |
fd01b88c | 4274 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4275 | { |
4276 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4277 | mddev->sync_speed_min ? "local": "system"); | |
4278 | } | |
4279 | ||
4280 | static ssize_t | |
fd01b88c | 4281 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4282 | { |
4283 | int min; | |
4284 | char *e; | |
4285 | if (strncmp(buf, "system", 6)==0) { | |
4286 | mddev->sync_speed_min = 0; | |
4287 | return len; | |
4288 | } | |
4289 | min = simple_strtoul(buf, &e, 10); | |
4290 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
4291 | return -EINVAL; | |
4292 | mddev->sync_speed_min = min; | |
4293 | return len; | |
4294 | } | |
4295 | ||
4296 | static struct md_sysfs_entry md_sync_min = | |
4297 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4298 | ||
4299 | static ssize_t | |
fd01b88c | 4300 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4301 | { |
4302 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4303 | mddev->sync_speed_max ? "local": "system"); | |
4304 | } | |
4305 | ||
4306 | static ssize_t | |
fd01b88c | 4307 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4308 | { |
4309 | int max; | |
4310 | char *e; | |
4311 | if (strncmp(buf, "system", 6)==0) { | |
4312 | mddev->sync_speed_max = 0; | |
4313 | return len; | |
4314 | } | |
4315 | max = simple_strtoul(buf, &e, 10); | |
4316 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
4317 | return -EINVAL; | |
4318 | mddev->sync_speed_max = max; | |
4319 | return len; | |
4320 | } | |
4321 | ||
4322 | static struct md_sysfs_entry md_sync_max = | |
4323 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4324 | ||
d7f3d291 | 4325 | static ssize_t |
fd01b88c | 4326 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4327 | { |
4328 | return sprintf(page, "%d\n", mddev->degraded); | |
4329 | } | |
4330 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4331 | |
90b08710 | 4332 | static ssize_t |
fd01b88c | 4333 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4334 | { |
4335 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4336 | } | |
4337 | ||
4338 | static ssize_t | |
fd01b88c | 4339 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4340 | { |
4341 | long n; | |
4342 | ||
b29bebd6 | 4343 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4344 | return -EINVAL; |
4345 | ||
4346 | if (n != 0 && n != 1) | |
4347 | return -EINVAL; | |
4348 | ||
4349 | mddev->parallel_resync = n; | |
4350 | ||
4351 | if (mddev->sync_thread) | |
4352 | wake_up(&resync_wait); | |
4353 | ||
4354 | return len; | |
4355 | } | |
4356 | ||
4357 | /* force parallel resync, even with shared block devices */ | |
4358 | static struct md_sysfs_entry md_sync_force_parallel = | |
4359 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4360 | sync_force_parallel_show, sync_force_parallel_store); | |
4361 | ||
88202a0c | 4362 | static ssize_t |
fd01b88c | 4363 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4364 | { |
4365 | unsigned long resync, dt, db; | |
d1a7c503 N |
4366 | if (mddev->curr_resync == 0) |
4367 | return sprintf(page, "none\n"); | |
9687a60c AN |
4368 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4369 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4370 | if (!dt) dt++; |
9687a60c AN |
4371 | db = resync - mddev->resync_mark_cnt; |
4372 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4373 | } |
4374 | ||
80ca3a44 | 4375 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4376 | |
4377 | static ssize_t | |
fd01b88c | 4378 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4379 | { |
13ae864b | 4380 | unsigned long long max_sectors, resync; |
88202a0c | 4381 | |
acb180b0 N |
4382 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4383 | return sprintf(page, "none\n"); | |
4384 | ||
72f36d59 N |
4385 | if (mddev->curr_resync == 1 || |
4386 | mddev->curr_resync == 2) | |
4387 | return sprintf(page, "delayed\n"); | |
4388 | ||
c804cdec N |
4389 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4390 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4391 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4392 | else |
58c0fed4 | 4393 | max_sectors = mddev->dev_sectors; |
88202a0c | 4394 | |
acb180b0 | 4395 | resync = mddev->curr_resync_completed; |
13ae864b | 4396 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4397 | } |
4398 | ||
80ca3a44 | 4399 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 4400 | |
5e96ee65 | 4401 | static ssize_t |
fd01b88c | 4402 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4403 | { |
4404 | return sprintf(page, "%llu\n", | |
4405 | (unsigned long long)mddev->resync_min); | |
4406 | } | |
4407 | static ssize_t | |
fd01b88c | 4408 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4409 | { |
4410 | unsigned long long min; | |
b29bebd6 | 4411 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 NB |
4412 | return -EINVAL; |
4413 | if (min > mddev->resync_max) | |
4414 | return -EINVAL; | |
4415 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4416 | return -EBUSY; | |
4417 | ||
4418 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4419 | if (mddev->chunk_sectors) { |
2ac06c33 | 4420 | sector_t temp = min; |
9d8f0363 | 4421 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
4422 | return -EINVAL; |
4423 | } | |
4424 | mddev->resync_min = min; | |
4425 | ||
4426 | return len; | |
4427 | } | |
4428 | ||
4429 | static struct md_sysfs_entry md_min_sync = | |
4430 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4431 | ||
c6207277 | 4432 | static ssize_t |
fd01b88c | 4433 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4434 | { |
4435 | if (mddev->resync_max == MaxSector) | |
4436 | return sprintf(page, "max\n"); | |
4437 | else | |
4438 | return sprintf(page, "%llu\n", | |
4439 | (unsigned long long)mddev->resync_max); | |
4440 | } | |
4441 | static ssize_t | |
fd01b88c | 4442 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 N |
4443 | { |
4444 | if (strncmp(buf, "max", 3) == 0) | |
4445 | mddev->resync_max = MaxSector; | |
4446 | else { | |
5e96ee65 | 4447 | unsigned long long max; |
b29bebd6 | 4448 | if (kstrtoull(buf, 10, &max)) |
5e96ee65 NB |
4449 | return -EINVAL; |
4450 | if (max < mddev->resync_min) | |
c6207277 N |
4451 | return -EINVAL; |
4452 | if (max < mddev->resync_max && | |
4d484a4a | 4453 | mddev->ro == 0 && |
c6207277 N |
4454 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4455 | return -EBUSY; | |
4456 | ||
4457 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4458 | if (mddev->chunk_sectors) { |
2ac06c33 | 4459 | sector_t temp = max; |
9d8f0363 | 4460 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
4461 | return -EINVAL; |
4462 | } | |
4463 | mddev->resync_max = max; | |
4464 | } | |
4465 | wake_up(&mddev->recovery_wait); | |
4466 | return len; | |
4467 | } | |
4468 | ||
4469 | static struct md_sysfs_entry md_max_sync = | |
4470 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4471 | ||
e464eafd | 4472 | static ssize_t |
fd01b88c | 4473 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4474 | { |
4475 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4476 | } | |
4477 | ||
4478 | static ssize_t | |
fd01b88c | 4479 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4480 | { |
4481 | char *e; | |
4482 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4483 | unsigned long long old = mddev->suspend_lo; |
e464eafd | 4484 | |
b8d966ef N |
4485 | if (mddev->pers == NULL || |
4486 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4487 | return -EINVAL; |
4488 | if (buf == e || (*e && *e != '\n')) | |
4489 | return -EINVAL; | |
23ddff37 N |
4490 | |
4491 | mddev->suspend_lo = new; | |
4492 | if (new >= old) | |
4493 | /* Shrinking suspended region */ | |
e464eafd | 4494 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4495 | else { |
4496 | /* Expanding suspended region - need to wait */ | |
4497 | mddev->pers->quiesce(mddev, 1); | |
4498 | mddev->pers->quiesce(mddev, 0); | |
4499 | } | |
4500 | return len; | |
e464eafd N |
4501 | } |
4502 | static struct md_sysfs_entry md_suspend_lo = | |
4503 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4504 | ||
4505 | ||
4506 | static ssize_t | |
fd01b88c | 4507 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4508 | { |
4509 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4510 | } | |
4511 | ||
4512 | static ssize_t | |
fd01b88c | 4513 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4514 | { |
4515 | char *e; | |
4516 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4517 | unsigned long long old = mddev->suspend_hi; |
e464eafd | 4518 | |
b8d966ef N |
4519 | if (mddev->pers == NULL || |
4520 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4521 | return -EINVAL; |
4522 | if (buf == e || (*e && *e != '\n')) | |
4523 | return -EINVAL; | |
23ddff37 N |
4524 | |
4525 | mddev->suspend_hi = new; | |
4526 | if (new <= old) | |
4527 | /* Shrinking suspended region */ | |
4528 | mddev->pers->quiesce(mddev, 2); | |
4529 | else { | |
4530 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4531 | mddev->pers->quiesce(mddev, 1); |
4532 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 N |
4533 | } |
4534 | return len; | |
e464eafd N |
4535 | } |
4536 | static struct md_sysfs_entry md_suspend_hi = | |
4537 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4538 | ||
08a02ecd | 4539 | static ssize_t |
fd01b88c | 4540 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4541 | { |
4542 | if (mddev->reshape_position != MaxSector) | |
4543 | return sprintf(page, "%llu\n", | |
4544 | (unsigned long long)mddev->reshape_position); | |
4545 | strcpy(page, "none\n"); | |
4546 | return 5; | |
4547 | } | |
4548 | ||
4549 | static ssize_t | |
fd01b88c | 4550 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4551 | { |
c6563a8c | 4552 | struct md_rdev *rdev; |
08a02ecd N |
4553 | char *e; |
4554 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4555 | if (mddev->pers) | |
4556 | return -EBUSY; | |
4557 | if (buf == e || (*e && *e != '\n')) | |
4558 | return -EINVAL; | |
4559 | mddev->reshape_position = new; | |
4560 | mddev->delta_disks = 0; | |
2c810cdd | 4561 | mddev->reshape_backwards = 0; |
08a02ecd N |
4562 | mddev->new_level = mddev->level; |
4563 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4564 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4565 | rdev_for_each(rdev, mddev) |
4566 | rdev->new_data_offset = rdev->data_offset; | |
08a02ecd N |
4567 | return len; |
4568 | } | |
4569 | ||
4570 | static struct md_sysfs_entry md_reshape_position = | |
4571 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4572 | reshape_position_store); | |
4573 | ||
2c810cdd N |
4574 | static ssize_t |
4575 | reshape_direction_show(struct mddev *mddev, char *page) | |
4576 | { | |
4577 | return sprintf(page, "%s\n", | |
4578 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4579 | } | |
4580 | ||
4581 | static ssize_t | |
4582 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4583 | { | |
4584 | int backwards = 0; | |
4585 | if (cmd_match(buf, "forwards")) | |
4586 | backwards = 0; | |
4587 | else if (cmd_match(buf, "backwards")) | |
4588 | backwards = 1; | |
4589 | else | |
4590 | return -EINVAL; | |
4591 | if (mddev->reshape_backwards == backwards) | |
4592 | return len; | |
4593 | ||
4594 | /* check if we are allowed to change */ | |
4595 | if (mddev->delta_disks) | |
4596 | return -EBUSY; | |
4597 | ||
4598 | if (mddev->persistent && | |
4599 | mddev->major_version == 0) | |
4600 | return -EINVAL; | |
4601 | ||
4602 | mddev->reshape_backwards = backwards; | |
4603 | return len; | |
4604 | } | |
4605 | ||
4606 | static struct md_sysfs_entry md_reshape_direction = | |
4607 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4608 | reshape_direction_store); | |
4609 | ||
b522adcd | 4610 | static ssize_t |
fd01b88c | 4611 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4612 | { |
4613 | if (mddev->external_size) | |
4614 | return sprintf(page, "%llu\n", | |
4615 | (unsigned long long)mddev->array_sectors/2); | |
4616 | else | |
4617 | return sprintf(page, "default\n"); | |
4618 | } | |
4619 | ||
4620 | static ssize_t | |
fd01b88c | 4621 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4622 | { |
4623 | sector_t sectors; | |
4624 | ||
4625 | if (strncmp(buf, "default", 7) == 0) { | |
4626 | if (mddev->pers) | |
4627 | sectors = mddev->pers->size(mddev, 0, 0); | |
4628 | else | |
4629 | sectors = mddev->array_sectors; | |
4630 | ||
4631 | mddev->external_size = 0; | |
4632 | } else { | |
4633 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4634 | return -EINVAL; | |
4635 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4636 | return -E2BIG; |
b522adcd DW |
4637 | |
4638 | mddev->external_size = 1; | |
4639 | } | |
4640 | ||
4641 | mddev->array_sectors = sectors; | |
cbe6ef1d N |
4642 | if (mddev->pers) { |
4643 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 4644 | revalidate_disk(mddev->gendisk); |
cbe6ef1d | 4645 | } |
b522adcd DW |
4646 | return len; |
4647 | } | |
4648 | ||
4649 | static struct md_sysfs_entry md_array_size = | |
4650 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4651 | array_size_store); | |
e464eafd | 4652 | |
eae1701f N |
4653 | static struct attribute *md_default_attrs[] = { |
4654 | &md_level.attr, | |
d4dbd025 | 4655 | &md_layout.attr, |
eae1701f | 4656 | &md_raid_disks.attr, |
3b34380a | 4657 | &md_chunk_size.attr, |
a35b0d69 | 4658 | &md_size.attr, |
a94213b1 | 4659 | &md_resync_start.attr, |
8bb93aac | 4660 | &md_metadata.attr, |
6d7ff738 | 4661 | &md_new_device.attr, |
16f17b39 | 4662 | &md_safe_delay.attr, |
9e653b63 | 4663 | &md_array_state.attr, |
08a02ecd | 4664 | &md_reshape_position.attr, |
2c810cdd | 4665 | &md_reshape_direction.attr, |
b522adcd | 4666 | &md_array_size.attr, |
1e50915f | 4667 | &max_corr_read_errors.attr, |
411036fa N |
4668 | NULL, |
4669 | }; | |
4670 | ||
4671 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4672 | &md_scan_mode.attr, |
c4a39551 | 4673 | &md_last_scan_mode.attr, |
9d88883e | 4674 | &md_mismatches.attr, |
88202a0c N |
4675 | &md_sync_min.attr, |
4676 | &md_sync_max.attr, | |
4677 | &md_sync_speed.attr, | |
90b08710 | 4678 | &md_sync_force_parallel.attr, |
88202a0c | 4679 | &md_sync_completed.attr, |
5e96ee65 | 4680 | &md_min_sync.attr, |
c6207277 | 4681 | &md_max_sync.attr, |
e464eafd N |
4682 | &md_suspend_lo.attr, |
4683 | &md_suspend_hi.attr, | |
9b1d1dac | 4684 | &md_bitmap.attr, |
d7f3d291 | 4685 | &md_degraded.attr, |
eae1701f N |
4686 | NULL, |
4687 | }; | |
411036fa N |
4688 | static struct attribute_group md_redundancy_group = { |
4689 | .name = NULL, | |
4690 | .attrs = md_redundancy_attrs, | |
4691 | }; | |
4692 | ||
eae1701f N |
4693 | |
4694 | static ssize_t | |
4695 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4696 | { | |
4697 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4698 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4699 | ssize_t rv; |
eae1701f N |
4700 | |
4701 | if (!entry->show) | |
4702 | return -EIO; | |
af8a2434 N |
4703 | spin_lock(&all_mddevs_lock); |
4704 | if (list_empty(&mddev->all_mddevs)) { | |
4705 | spin_unlock(&all_mddevs_lock); | |
4706 | return -EBUSY; | |
4707 | } | |
4708 | mddev_get(mddev); | |
4709 | spin_unlock(&all_mddevs_lock); | |
4710 | ||
5dc5cf7d IM |
4711 | rv = mddev_lock(mddev); |
4712 | if (!rv) { | |
4713 | rv = entry->show(mddev, page); | |
4714 | mddev_unlock(mddev); | |
4715 | } | |
af8a2434 | 4716 | mddev_put(mddev); |
96de1e66 | 4717 | return rv; |
eae1701f N |
4718 | } |
4719 | ||
4720 | static ssize_t | |
4721 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4722 | const char *page, size_t length) | |
4723 | { | |
4724 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4725 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4726 | ssize_t rv; |
eae1701f N |
4727 | |
4728 | if (!entry->store) | |
4729 | return -EIO; | |
67463acb N |
4730 | if (!capable(CAP_SYS_ADMIN)) |
4731 | return -EACCES; | |
af8a2434 N |
4732 | spin_lock(&all_mddevs_lock); |
4733 | if (list_empty(&mddev->all_mddevs)) { | |
4734 | spin_unlock(&all_mddevs_lock); | |
4735 | return -EBUSY; | |
4736 | } | |
4737 | mddev_get(mddev); | |
4738 | spin_unlock(&all_mddevs_lock); | |
a7a3f08d N |
4739 | if (entry->store == new_dev_store) |
4740 | flush_workqueue(md_misc_wq); | |
5dc5cf7d IM |
4741 | rv = mddev_lock(mddev); |
4742 | if (!rv) { | |
4743 | rv = entry->store(mddev, page, length); | |
4744 | mddev_unlock(mddev); | |
4745 | } | |
af8a2434 | 4746 | mddev_put(mddev); |
96de1e66 | 4747 | return rv; |
eae1701f N |
4748 | } |
4749 | ||
4750 | static void md_free(struct kobject *ko) | |
4751 | { | |
fd01b88c | 4752 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4753 | |
4754 | if (mddev->sysfs_state) | |
4755 | sysfs_put(mddev->sysfs_state); | |
4756 | ||
4757 | if (mddev->gendisk) { | |
4758 | del_gendisk(mddev->gendisk); | |
4759 | put_disk(mddev->gendisk); | |
4760 | } | |
4761 | if (mddev->queue) | |
4762 | blk_cleanup_queue(mddev->queue); | |
4763 | ||
eae1701f N |
4764 | kfree(mddev); |
4765 | } | |
4766 | ||
52cf25d0 | 4767 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4768 | .show = md_attr_show, |
4769 | .store = md_attr_store, | |
4770 | }; | |
4771 | static struct kobj_type md_ktype = { | |
4772 | .release = md_free, | |
4773 | .sysfs_ops = &md_sysfs_ops, | |
4774 | .default_attrs = md_default_attrs, | |
4775 | }; | |
4776 | ||
1da177e4 LT |
4777 | int mdp_major = 0; |
4778 | ||
5fd3a17e DW |
4779 | static void mddev_delayed_delete(struct work_struct *ws) |
4780 | { | |
fd01b88c | 4781 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4782 | |
43a70507 | 4783 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4784 | kobject_del(&mddev->kobj); |
4785 | kobject_put(&mddev->kobj); | |
4786 | } | |
4787 | ||
efeb53c0 | 4788 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4789 | { |
48c9c27b | 4790 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 4791 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 4792 | struct gendisk *disk; |
efeb53c0 N |
4793 | int partitioned; |
4794 | int shift; | |
4795 | int unit; | |
3830c62f | 4796 | int error; |
1da177e4 LT |
4797 | |
4798 | if (!mddev) | |
efeb53c0 N |
4799 | return -ENODEV; |
4800 | ||
4801 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4802 | shift = partitioned ? MdpMinorShift : 0; | |
4803 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4804 | |
e804ac78 TH |
4805 | /* wait for any previous instance of this device to be |
4806 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4807 | */ |
e804ac78 | 4808 | flush_workqueue(md_misc_wq); |
d3374825 | 4809 | |
48c9c27b | 4810 | mutex_lock(&disks_mutex); |
0909dc44 N |
4811 | error = -EEXIST; |
4812 | if (mddev->gendisk) | |
4813 | goto abort; | |
efeb53c0 N |
4814 | |
4815 | if (name) { | |
4816 | /* Need to ensure that 'name' is not a duplicate. | |
4817 | */ | |
fd01b88c | 4818 | struct mddev *mddev2; |
efeb53c0 N |
4819 | spin_lock(&all_mddevs_lock); |
4820 | ||
4821 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4822 | if (mddev2->gendisk && | |
4823 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4824 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4825 | goto abort; |
efeb53c0 N |
4826 | } |
4827 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4828 | } |
8b765398 | 4829 | |
0909dc44 | 4830 | error = -ENOMEM; |
8b765398 | 4831 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4832 | if (!mddev->queue) |
4833 | goto abort; | |
409c57f3 N |
4834 | mddev->queue->queuedata = mddev; |
4835 | ||
409c57f3 | 4836 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 4837 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 4838 | |
1da177e4 LT |
4839 | disk = alloc_disk(1 << shift); |
4840 | if (!disk) { | |
8b765398 N |
4841 | blk_cleanup_queue(mddev->queue); |
4842 | mddev->queue = NULL; | |
0909dc44 | 4843 | goto abort; |
1da177e4 | 4844 | } |
efeb53c0 | 4845 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4846 | disk->first_minor = unit << shift; |
efeb53c0 N |
4847 | if (name) |
4848 | strcpy(disk->disk_name, name); | |
4849 | else if (partitioned) | |
1da177e4 | 4850 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4851 | else |
1da177e4 | 4852 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4853 | disk->fops = &md_fops; |
4854 | disk->private_data = mddev; | |
4855 | disk->queue = mddev->queue; | |
b0140891 | 4856 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4857 | /* Allow extended partitions. This makes the |
d3374825 | 4858 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4859 | * remove it now. |
4860 | */ | |
4861 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4862 | mddev->gendisk = disk; |
b0140891 N |
4863 | /* As soon as we call add_disk(), another thread could get |
4864 | * through to md_open, so make sure it doesn't get too far | |
4865 | */ | |
4866 | mutex_lock(&mddev->open_mutex); | |
4867 | add_disk(disk); | |
4868 | ||
ed9e1982 TH |
4869 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4870 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4871 | if (error) { |
4872 | /* This isn't possible, but as kobject_init_and_add is marked | |
4873 | * __must_check, we must do something with the result | |
4874 | */ | |
5e55e2f5 N |
4875 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4876 | disk->disk_name); | |
0909dc44 N |
4877 | error = 0; |
4878 | } | |
00bcb4ac N |
4879 | if (mddev->kobj.sd && |
4880 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4881 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4882 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4883 | abort: |
4884 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4885 | if (!error && mddev->kobj.sd) { |
3830c62f | 4886 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4887 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4888 | } |
d3374825 | 4889 | mddev_put(mddev); |
0909dc44 | 4890 | return error; |
efeb53c0 N |
4891 | } |
4892 | ||
4893 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4894 | { | |
4895 | md_alloc(dev, NULL); | |
1da177e4 LT |
4896 | return NULL; |
4897 | } | |
4898 | ||
efeb53c0 N |
4899 | static int add_named_array(const char *val, struct kernel_param *kp) |
4900 | { | |
4901 | /* val must be "md_*" where * is not all digits. | |
4902 | * We allocate an array with a large free minor number, and | |
4903 | * set the name to val. val must not already be an active name. | |
4904 | */ | |
4905 | int len = strlen(val); | |
4906 | char buf[DISK_NAME_LEN]; | |
4907 | ||
4908 | while (len && val[len-1] == '\n') | |
4909 | len--; | |
4910 | if (len >= DISK_NAME_LEN) | |
4911 | return -E2BIG; | |
4912 | strlcpy(buf, val, len+1); | |
4913 | if (strncmp(buf, "md_", 3) != 0) | |
4914 | return -EINVAL; | |
4915 | return md_alloc(0, buf); | |
4916 | } | |
4917 | ||
1da177e4 LT |
4918 | static void md_safemode_timeout(unsigned long data) |
4919 | { | |
fd01b88c | 4920 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 4921 | |
0fd62b86 NB |
4922 | if (!atomic_read(&mddev->writes_pending)) { |
4923 | mddev->safemode = 1; | |
4924 | if (mddev->external) | |
00bcb4ac | 4925 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4926 | } |
1da177e4 LT |
4927 | md_wakeup_thread(mddev->thread); |
4928 | } | |
4929 | ||
6ff8d8ec | 4930 | static int start_dirty_degraded; |
1da177e4 | 4931 | |
fd01b88c | 4932 | int md_run(struct mddev *mddev) |
1da177e4 | 4933 | { |
2604b703 | 4934 | int err; |
3cb03002 | 4935 | struct md_rdev *rdev; |
84fc4b56 | 4936 | struct md_personality *pers; |
1da177e4 | 4937 | |
a757e64c N |
4938 | if (list_empty(&mddev->disks)) |
4939 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4940 | return -EINVAL; |
1da177e4 LT |
4941 | |
4942 | if (mddev->pers) | |
4943 | return -EBUSY; | |
bb4f1e9d N |
4944 | /* Cannot run until previous stop completes properly */ |
4945 | if (mddev->sysfs_active) | |
4946 | return -EBUSY; | |
b6eb127d | 4947 | |
1da177e4 LT |
4948 | /* |
4949 | * Analyze all RAID superblock(s) | |
4950 | */ | |
1ec4a939 N |
4951 | if (!mddev->raid_disks) { |
4952 | if (!mddev->persistent) | |
4953 | return -EINVAL; | |
a757e64c | 4954 | analyze_sbs(mddev); |
1ec4a939 | 4955 | } |
1da177e4 | 4956 | |
d9d166c2 N |
4957 | if (mddev->level != LEVEL_NONE) |
4958 | request_module("md-level-%d", mddev->level); | |
4959 | else if (mddev->clevel[0]) | |
4960 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4961 | |
4962 | /* | |
4963 | * Drop all container device buffers, from now on | |
4964 | * the only valid external interface is through the md | |
4965 | * device. | |
1da177e4 | 4966 | */ |
dafb20fa | 4967 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 4968 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4969 | continue; |
4970 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4971 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4972 | |
4973 | /* perform some consistency tests on the device. | |
4974 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4975 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4976 | */ |
a6ff7e08 JB |
4977 | if (rdev->meta_bdev) { |
4978 | /* Nothing to check */; | |
4979 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
4980 | if (mddev->dev_sectors && |
4981 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4982 | > rdev->sb_start) { |
f0d76d70 N |
4983 | printk("md: %s: data overlaps metadata\n", |
4984 | mdname(mddev)); | |
4985 | return -EINVAL; | |
4986 | } | |
4987 | } else { | |
0f420358 | 4988 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4989 | > rdev->data_offset) { |
4990 | printk("md: %s: metadata overlaps data\n", | |
4991 | mdname(mddev)); | |
4992 | return -EINVAL; | |
4993 | } | |
4994 | } | |
00bcb4ac | 4995 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4996 | } |
4997 | ||
a167f663 | 4998 | if (mddev->bio_set == NULL) |
395c72a7 | 4999 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
a167f663 | 5000 | |
1da177e4 | 5001 | spin_lock(&pers_lock); |
d9d166c2 | 5002 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5003 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5004 | spin_unlock(&pers_lock); |
d9d166c2 N |
5005 | if (mddev->level != LEVEL_NONE) |
5006 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
5007 | mddev->level); | |
5008 | else | |
5009 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
5010 | mddev->clevel); | |
1da177e4 LT |
5011 | return -EINVAL; |
5012 | } | |
2604b703 | 5013 | mddev->pers = pers; |
1da177e4 | 5014 | spin_unlock(&pers_lock); |
34817e8c N |
5015 | if (mddev->level != pers->level) { |
5016 | mddev->level = pers->level; | |
5017 | mddev->new_level = pers->level; | |
5018 | } | |
d9d166c2 | 5019 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5020 | |
f6705578 | 5021 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5022 | pers->start_reshape == NULL) { |
f6705578 N |
5023 | /* This personality cannot handle reshaping... */ |
5024 | mddev->pers = NULL; | |
5025 | module_put(pers->owner); | |
5026 | return -EINVAL; | |
5027 | } | |
5028 | ||
7dd5e7c3 N |
5029 | if (pers->sync_request) { |
5030 | /* Warn if this is a potentially silly | |
5031 | * configuration. | |
5032 | */ | |
5033 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5034 | struct md_rdev *rdev2; |
7dd5e7c3 | 5035 | int warned = 0; |
159ec1fc | 5036 | |
dafb20fa N |
5037 | rdev_for_each(rdev, mddev) |
5038 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5039 | if (rdev < rdev2 && |
5040 | rdev->bdev->bd_contains == | |
5041 | rdev2->bdev->bd_contains) { | |
5042 | printk(KERN_WARNING | |
5043 | "%s: WARNING: %s appears to be" | |
5044 | " on the same physical disk as" | |
5045 | " %s.\n", | |
5046 | mdname(mddev), | |
5047 | bdevname(rdev->bdev,b), | |
5048 | bdevname(rdev2->bdev,b2)); | |
5049 | warned = 1; | |
5050 | } | |
5051 | } | |
159ec1fc | 5052 | |
7dd5e7c3 N |
5053 | if (warned) |
5054 | printk(KERN_WARNING | |
5055 | "True protection against single-disk" | |
5056 | " failure might be compromised.\n"); | |
5057 | } | |
5058 | ||
657390d2 | 5059 | mddev->recovery = 0; |
58c0fed4 AN |
5060 | /* may be over-ridden by personality */ |
5061 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5062 | ||
6ff8d8ec | 5063 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5064 | |
0f9552b5 | 5065 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5066 | mddev->ro = 2; /* read-only, but switch on first write */ |
5067 | ||
b15c2e57 | 5068 | err = mddev->pers->run(mddev); |
13e53df3 AN |
5069 | if (err) |
5070 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
5071 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
5072 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
5073 | " but 'external_size' not in effect?\n", __func__); | |
5074 | printk(KERN_ERR | |
5075 | "md: invalid array_size %llu > default size %llu\n", | |
5076 | (unsigned long long)mddev->array_sectors / 2, | |
5077 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
5078 | err = -EINVAL; | |
5079 | mddev->pers->stop(mddev); | |
5080 | } | |
ef99bf48 N |
5081 | if (err == 0 && mddev->pers->sync_request && |
5082 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { | |
b15c2e57 N |
5083 | err = bitmap_create(mddev); |
5084 | if (err) { | |
5085 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
5086 | mdname(mddev), err); | |
5087 | mddev->pers->stop(mddev); | |
5088 | } | |
5089 | } | |
1da177e4 | 5090 | if (err) { |
1da177e4 LT |
5091 | module_put(mddev->pers->owner); |
5092 | mddev->pers = NULL; | |
32a7627c N |
5093 | bitmap_destroy(mddev); |
5094 | return err; | |
1da177e4 | 5095 | } |
5e55e2f5 | 5096 | if (mddev->pers->sync_request) { |
00bcb4ac N |
5097 | if (mddev->kobj.sd && |
5098 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
5099 | printk(KERN_WARNING |
5100 | "md: cannot register extra attributes for %s\n", | |
5101 | mdname(mddev)); | |
00bcb4ac | 5102 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5103 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5104 | mddev->ro = 0; |
5105 | ||
1da177e4 | 5106 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
5107 | atomic_set(&mddev->max_corr_read_errors, |
5108 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
5109 | mddev->safemode = 0; |
5110 | mddev->safemode_timer.function = md_safemode_timeout; | |
5111 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 5112 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 5113 | mddev->in_sync = 1; |
0ca69886 N |
5114 | smp_wmb(); |
5115 | mddev->ready = 1; | |
dafb20fa | 5116 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5117 | if (rdev->raid_disk >= 0) |
5118 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5119 | /* failure here is OK */; |
1da177e4 LT |
5120 | |
5121 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5122 | ||
7a0a5355 | 5123 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 5124 | md_update_sb(mddev, 0); |
1da177e4 | 5125 | |
d7603b7e | 5126 | md_new_event(mddev); |
00bcb4ac N |
5127 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5128 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5129 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5130 | return 0; |
5131 | } | |
390ee602 | 5132 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5133 | |
fd01b88c | 5134 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5135 | { |
5136 | int err; | |
5137 | ||
5138 | err = md_run(mddev); | |
5139 | if (err) | |
5140 | goto out; | |
69e51b44 N |
5141 | err = bitmap_load(mddev); |
5142 | if (err) { | |
5143 | bitmap_destroy(mddev); | |
5144 | goto out; | |
5145 | } | |
0fd018af JB |
5146 | |
5147 | md_wakeup_thread(mddev->thread); | |
5148 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5149 | ||
fe60b014 N |
5150 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5151 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5152 | mddev->changed = 1; |
fe60b014 N |
5153 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5154 | out: | |
5155 | return err; | |
5156 | } | |
5157 | ||
fd01b88c | 5158 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5159 | { |
5160 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 5161 | |
80fab1d7 | 5162 | /* Complain if it has no devices */ |
1da177e4 | 5163 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5164 | return -ENXIO; |
5165 | if (!mddev->pers) | |
5166 | return -EINVAL; | |
5167 | if (!mddev->ro) | |
5168 | return -EBUSY; | |
5169 | mddev->safemode = 0; | |
5170 | mddev->ro = 0; | |
5171 | set_disk_ro(disk, 0); | |
5172 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
5173 | mdname(mddev)); | |
5174 | /* Kick recovery or resync if necessary */ | |
5175 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5176 | md_wakeup_thread(mddev->thread); | |
5177 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5178 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5179 | return 0; |
1da177e4 LT |
5180 | } |
5181 | ||
acc55e22 N |
5182 | /* similar to deny_write_access, but accounts for our holding a reference |
5183 | * to the file ourselves */ | |
5184 | static int deny_bitmap_write_access(struct file * file) | |
5185 | { | |
5186 | struct inode *inode = file->f_mapping->host; | |
5187 | ||
5188 | spin_lock(&inode->i_lock); | |
5189 | if (atomic_read(&inode->i_writecount) > 1) { | |
5190 | spin_unlock(&inode->i_lock); | |
5191 | return -ETXTBSY; | |
5192 | } | |
5193 | atomic_set(&inode->i_writecount, -1); | |
5194 | spin_unlock(&inode->i_lock); | |
5195 | ||
5196 | return 0; | |
5197 | } | |
5198 | ||
43a70507 | 5199 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
5200 | { |
5201 | struct inode *inode = file->f_mapping->host; | |
5202 | ||
5203 | spin_lock(&inode->i_lock); | |
5204 | atomic_set(&inode->i_writecount, 1); | |
5205 | spin_unlock(&inode->i_lock); | |
5206 | } | |
5207 | ||
fd01b88c | 5208 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5209 | { |
5210 | mddev->array_sectors = 0; | |
5211 | mddev->external_size = 0; | |
5212 | mddev->dev_sectors = 0; | |
5213 | mddev->raid_disks = 0; | |
5214 | mddev->recovery_cp = 0; | |
5215 | mddev->resync_min = 0; | |
5216 | mddev->resync_max = MaxSector; | |
5217 | mddev->reshape_position = MaxSector; | |
5218 | mddev->external = 0; | |
5219 | mddev->persistent = 0; | |
5220 | mddev->level = LEVEL_NONE; | |
5221 | mddev->clevel[0] = 0; | |
5222 | mddev->flags = 0; | |
5223 | mddev->ro = 0; | |
5224 | mddev->metadata_type[0] = 0; | |
5225 | mddev->chunk_sectors = 0; | |
5226 | mddev->ctime = mddev->utime = 0; | |
5227 | mddev->layout = 0; | |
5228 | mddev->max_disks = 0; | |
5229 | mddev->events = 0; | |
a8707c08 | 5230 | mddev->can_decrease_events = 0; |
6177b472 | 5231 | mddev->delta_disks = 0; |
2c810cdd | 5232 | mddev->reshape_backwards = 0; |
6177b472 N |
5233 | mddev->new_level = LEVEL_NONE; |
5234 | mddev->new_layout = 0; | |
5235 | mddev->new_chunk_sectors = 0; | |
5236 | mddev->curr_resync = 0; | |
7f7583d4 | 5237 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5238 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5239 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5240 | mddev->recovery = 0; | |
5241 | mddev->in_sync = 0; | |
f0b4f7e2 | 5242 | mddev->changed = 0; |
6177b472 | 5243 | mddev->degraded = 0; |
6177b472 | 5244 | mddev->safemode = 0; |
050b6615 | 5245 | mddev->merge_check_needed = 0; |
6177b472 N |
5246 | mddev->bitmap_info.offset = 0; |
5247 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5248 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5249 | mddev->bitmap_info.chunksize = 0; |
5250 | mddev->bitmap_info.daemon_sleep = 0; | |
5251 | mddev->bitmap_info.max_write_behind = 0; | |
5252 | } | |
5253 | ||
fd01b88c | 5254 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5255 | { |
6b6204ee | 5256 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
a047e125 | 5257 | if (mddev->sync_thread) { |
a047e125 | 5258 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5259 | md_reap_sync_thread(mddev); |
a047e125 N |
5260 | } |
5261 | ||
5262 | del_timer_sync(&mddev->safemode_timer); | |
5263 | ||
5264 | bitmap_flush(mddev); | |
5265 | md_super_wait(mddev); | |
5266 | ||
b6d428c6 | 5267 | if (mddev->ro == 0 && |
7a0a5355 | 5268 | (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) { |
a047e125 N |
5269 | /* mark array as shutdown cleanly */ |
5270 | mddev->in_sync = 1; | |
5271 | md_update_sb(mddev, 1); | |
5272 | } | |
5273 | } | |
defad61a | 5274 | |
fd01b88c | 5275 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5276 | { |
29f097c4 | 5277 | mddev_lock_nointr(mddev); |
defad61a N |
5278 | __md_stop_writes(mddev); |
5279 | mddev_unlock(mddev); | |
5280 | } | |
390ee602 | 5281 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5282 | |
5eff3c43 | 5283 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5284 | { |
0ca69886 | 5285 | mddev->ready = 0; |
6177b472 N |
5286 | mddev->pers->stop(mddev); |
5287 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
5288 | mddev->to_remove = &md_redundancy_group; | |
5289 | module_put(mddev->pers->owner); | |
5290 | mddev->pers = NULL; | |
cca9cf90 | 5291 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5292 | } |
5eff3c43 N |
5293 | |
5294 | void md_stop(struct mddev *mddev) | |
5295 | { | |
5296 | /* stop the array and free an attached data structures. | |
5297 | * This is called from dm-raid | |
5298 | */ | |
5299 | __md_stop(mddev); | |
5300 | bitmap_destroy(mddev); | |
5301 | if (mddev->bio_set) | |
5302 | bioset_free(mddev->bio_set); | |
5303 | } | |
5304 | ||
390ee602 | 5305 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5306 | |
a05b7ea0 | 5307 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5308 | { |
5309 | int err = 0; | |
30b8feb7 N |
5310 | int did_freeze = 0; |
5311 | ||
5312 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5313 | did_freeze = 1; | |
5314 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5315 | md_wakeup_thread(mddev->thread); | |
5316 | } | |
5317 | if (mddev->sync_thread) { | |
5318 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5319 | /* Thread might be blocked waiting for metadata update | |
5320 | * which will now never happen */ | |
5321 | wake_up_process(mddev->sync_thread->tsk); | |
5322 | } | |
5323 | mddev_unlock(mddev); | |
5324 | wait_event(resync_wait, mddev->sync_thread == NULL); | |
5325 | mddev_lock_nointr(mddev); | |
5326 | ||
a4bd82d0 | 5327 | mutex_lock(&mddev->open_mutex); |
30b8feb7 N |
5328 | if (atomic_read(&mddev->openers) > !!bdev || |
5329 | mddev->sync_thread || | |
5330 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { | |
a4bd82d0 | 5331 | printk("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5332 | if (did_freeze) { |
5333 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5334 | md_wakeup_thread(mddev->thread); | |
5335 | } | |
a4bd82d0 N |
5336 | err = -EBUSY; |
5337 | goto out; | |
5338 | } | |
5339 | if (mddev->pers) { | |
defad61a | 5340 | __md_stop_writes(mddev); |
a4bd82d0 N |
5341 | |
5342 | err = -ENXIO; | |
5343 | if (mddev->ro==1) | |
5344 | goto out; | |
5345 | mddev->ro = 1; | |
5346 | set_disk_ro(mddev->gendisk, 1); | |
5347 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 5348 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5349 | err = 0; |
a4bd82d0 N |
5350 | } |
5351 | out: | |
5352 | mutex_unlock(&mddev->open_mutex); | |
5353 | return err; | |
5354 | } | |
5355 | ||
9e653b63 N |
5356 | /* mode: |
5357 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5358 | * 2 - stop but do not disassemble array |
5359 | */ | |
a05b7ea0 N |
5360 | static int do_md_stop(struct mddev * mddev, int mode, |
5361 | struct block_device *bdev) | |
1da177e4 | 5362 | { |
1da177e4 | 5363 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5364 | struct md_rdev *rdev; |
30b8feb7 N |
5365 | int did_freeze = 0; |
5366 | ||
5367 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5368 | did_freeze = 1; | |
5369 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5370 | md_wakeup_thread(mddev->thread); | |
5371 | } | |
5372 | if (mddev->sync_thread) { | |
5373 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5374 | /* Thread might be blocked waiting for metadata update | |
5375 | * which will now never happen */ | |
5376 | wake_up_process(mddev->sync_thread->tsk); | |
5377 | } | |
5378 | mddev_unlock(mddev); | |
5379 | wait_event(resync_wait, mddev->sync_thread == NULL); | |
5380 | mddev_lock_nointr(mddev); | |
1da177e4 | 5381 | |
c8c00a69 | 5382 | mutex_lock(&mddev->open_mutex); |
a05b7ea0 | 5383 | if (atomic_read(&mddev->openers) > !!bdev || |
30b8feb7 N |
5384 | mddev->sysfs_active || |
5385 | mddev->sync_thread || | |
5386 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { | |
df5b20cf | 5387 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5388 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5389 | if (did_freeze) { |
5390 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5391 | md_wakeup_thread(mddev->thread); | |
5392 | } | |
260fa034 N |
5393 | return -EBUSY; |
5394 | } | |
6e17b027 | 5395 | if (mddev->pers) { |
a4bd82d0 N |
5396 | if (mddev->ro) |
5397 | set_disk_ro(disk, 0); | |
409c57f3 | 5398 | |
defad61a | 5399 | __md_stop_writes(mddev); |
5eff3c43 | 5400 | __md_stop(mddev); |
a4bd82d0 | 5401 | mddev->queue->merge_bvec_fn = NULL; |
a4bd82d0 | 5402 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 5403 | |
a4bd82d0 | 5404 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5405 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5406 | |
dafb20fa | 5407 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5408 | if (rdev->raid_disk >= 0) |
5409 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5410 | |
a4bd82d0 | 5411 | set_capacity(disk, 0); |
6e17b027 | 5412 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5413 | mddev->changed = 1; |
a4bd82d0 | 5414 | revalidate_disk(disk); |
0d4ca600 | 5415 | |
a4bd82d0 N |
5416 | if (mddev->ro) |
5417 | mddev->ro = 0; | |
6e17b027 N |
5418 | } else |
5419 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5420 | /* |
5421 | * Free resources if final stop | |
5422 | */ | |
9e653b63 | 5423 | if (mode == 0) { |
1da177e4 LT |
5424 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
5425 | ||
978f946b | 5426 | bitmap_destroy(mddev); |
c3d9714e N |
5427 | if (mddev->bitmap_info.file) { |
5428 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5429 | fput(mddev->bitmap_info.file); | |
5430 | mddev->bitmap_info.file = NULL; | |
978f946b | 5431 | } |
c3d9714e | 5432 | mddev->bitmap_info.offset = 0; |
978f946b | 5433 | |
1da177e4 LT |
5434 | export_array(mddev); |
5435 | ||
6177b472 | 5436 | md_clean(mddev); |
934d9c23 | 5437 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
5438 | if (mddev->hold_active == UNTIL_STOP) |
5439 | mddev->hold_active = 0; | |
a4bd82d0 | 5440 | } |
3f9d99c1 | 5441 | blk_integrity_unregister(disk); |
d7603b7e | 5442 | md_new_event(mddev); |
00bcb4ac | 5443 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5444 | return 0; |
1da177e4 LT |
5445 | } |
5446 | ||
fdee8ae4 | 5447 | #ifndef MODULE |
fd01b88c | 5448 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5449 | { |
3cb03002 | 5450 | struct md_rdev *rdev; |
1da177e4 LT |
5451 | int err; |
5452 | ||
a757e64c | 5453 | if (list_empty(&mddev->disks)) |
1da177e4 | 5454 | return; |
1da177e4 LT |
5455 | |
5456 | printk(KERN_INFO "md: running: "); | |
5457 | ||
dafb20fa | 5458 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
5459 | char b[BDEVNAME_SIZE]; |
5460 | printk("<%s>", bdevname(rdev->bdev,b)); | |
5461 | } | |
5462 | printk("\n"); | |
5463 | ||
d710e138 | 5464 | err = do_md_run(mddev); |
1da177e4 LT |
5465 | if (err) { |
5466 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
a05b7ea0 | 5467 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5468 | } |
5469 | } | |
5470 | ||
5471 | /* | |
5472 | * lets try to run arrays based on all disks that have arrived | |
5473 | * until now. (those are in pending_raid_disks) | |
5474 | * | |
5475 | * the method: pick the first pending disk, collect all disks with | |
5476 | * the same UUID, remove all from the pending list and put them into | |
5477 | * the 'same_array' list. Then order this list based on superblock | |
5478 | * update time (freshest comes first), kick out 'old' disks and | |
5479 | * compare superblocks. If everything's fine then run it. | |
5480 | * | |
5481 | * If "unit" is allocated, then bump its reference count | |
5482 | */ | |
5483 | static void autorun_devices(int part) | |
5484 | { | |
3cb03002 | 5485 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5486 | struct mddev *mddev; |
1da177e4 LT |
5487 | char b[BDEVNAME_SIZE]; |
5488 | ||
5489 | printk(KERN_INFO "md: autorun ...\n"); | |
5490 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 5491 | int unit; |
1da177e4 | 5492 | dev_t dev; |
ad01c9e3 | 5493 | LIST_HEAD(candidates); |
1da177e4 | 5494 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5495 | struct md_rdev, same_set); |
1da177e4 LT |
5496 | |
5497 | printk(KERN_INFO "md: considering %s ...\n", | |
5498 | bdevname(rdev0->bdev,b)); | |
5499 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 5500 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
5501 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
5502 | printk(KERN_INFO "md: adding %s ...\n", | |
5503 | bdevname(rdev->bdev,b)); | |
5504 | list_move(&rdev->same_set, &candidates); | |
5505 | } | |
5506 | /* | |
5507 | * now we have a set of devices, with all of them having | |
5508 | * mostly sane superblocks. It's time to allocate the | |
5509 | * mddev. | |
5510 | */ | |
e8703fe1 N |
5511 | if (part) { |
5512 | dev = MKDEV(mdp_major, | |
5513 | rdev0->preferred_minor << MdpMinorShift); | |
5514 | unit = MINOR(dev) >> MdpMinorShift; | |
5515 | } else { | |
5516 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5517 | unit = MINOR(dev); | |
5518 | } | |
5519 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
5520 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
5521 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
5522 | break; | |
5523 | } | |
1da177e4 LT |
5524 | |
5525 | md_probe(dev, NULL, NULL); | |
5526 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5527 | if (!mddev || !mddev->gendisk) { |
5528 | if (mddev) | |
5529 | mddev_put(mddev); | |
5530 | printk(KERN_ERR | |
1da177e4 LT |
5531 | "md: cannot allocate memory for md drive.\n"); |
5532 | break; | |
5533 | } | |
5534 | if (mddev_lock(mddev)) | |
5535 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
5536 | mdname(mddev)); | |
5537 | else if (mddev->raid_disks || mddev->major_version | |
5538 | || !list_empty(&mddev->disks)) { | |
5539 | printk(KERN_WARNING | |
5540 | "md: %s already running, cannot run %s\n", | |
5541 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
5542 | mddev_unlock(mddev); | |
5543 | } else { | |
5544 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 5545 | mddev->persistent = 1; |
159ec1fc | 5546 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5547 | list_del_init(&rdev->same_set); |
5548 | if (bind_rdev_to_array(rdev, mddev)) | |
5549 | export_rdev(rdev); | |
5550 | } | |
5551 | autorun_array(mddev); | |
5552 | mddev_unlock(mddev); | |
5553 | } | |
5554 | /* on success, candidates will be empty, on error | |
5555 | * it won't... | |
5556 | */ | |
159ec1fc | 5557 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5558 | list_del_init(&rdev->same_set); |
1da177e4 | 5559 | export_rdev(rdev); |
4b80991c | 5560 | } |
1da177e4 LT |
5561 | mddev_put(mddev); |
5562 | } | |
5563 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5564 | } | |
fdee8ae4 | 5565 | #endif /* !MODULE */ |
1da177e4 | 5566 | |
1da177e4 LT |
5567 | static int get_version(void __user * arg) |
5568 | { | |
5569 | mdu_version_t ver; | |
5570 | ||
5571 | ver.major = MD_MAJOR_VERSION; | |
5572 | ver.minor = MD_MINOR_VERSION; | |
5573 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5574 | ||
5575 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5576 | return -EFAULT; | |
5577 | ||
5578 | return 0; | |
5579 | } | |
5580 | ||
fd01b88c | 5581 | static int get_array_info(struct mddev * mddev, void __user * arg) |
1da177e4 LT |
5582 | { |
5583 | mdu_array_info_t info; | |
a9f326eb | 5584 | int nr,working,insync,failed,spare; |
3cb03002 | 5585 | struct md_rdev *rdev; |
1da177e4 | 5586 | |
1ca69c4b N |
5587 | nr = working = insync = failed = spare = 0; |
5588 | rcu_read_lock(); | |
5589 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5590 | nr++; |
b2d444d7 | 5591 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5592 | failed++; |
5593 | else { | |
5594 | working++; | |
b2d444d7 | 5595 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 5596 | insync++; |
1da177e4 LT |
5597 | else |
5598 | spare++; | |
5599 | } | |
5600 | } | |
1ca69c4b | 5601 | rcu_read_unlock(); |
1da177e4 LT |
5602 | |
5603 | info.major_version = mddev->major_version; | |
5604 | info.minor_version = mddev->minor_version; | |
5605 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5606 | info.ctime = mddev->ctime; | |
5607 | info.level = mddev->level; | |
58c0fed4 AN |
5608 | info.size = mddev->dev_sectors / 2; |
5609 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5610 | info.size = -1; |
1da177e4 LT |
5611 | info.nr_disks = nr; |
5612 | info.raid_disks = mddev->raid_disks; | |
5613 | info.md_minor = mddev->md_minor; | |
5614 | info.not_persistent= !mddev->persistent; | |
5615 | ||
5616 | info.utime = mddev->utime; | |
5617 | info.state = 0; | |
5618 | if (mddev->in_sync) | |
5619 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5620 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5621 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5622 | info.active_disks = insync; |
1da177e4 LT |
5623 | info.working_disks = working; |
5624 | info.failed_disks = failed; | |
5625 | info.spare_disks = spare; | |
5626 | ||
5627 | info.layout = mddev->layout; | |
9d8f0363 | 5628 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5629 | |
5630 | if (copy_to_user(arg, &info, sizeof(info))) | |
5631 | return -EFAULT; | |
5632 | ||
5633 | return 0; | |
5634 | } | |
5635 | ||
fd01b88c | 5636 | static int get_bitmap_file(struct mddev * mddev, void __user * arg) |
32a7627c N |
5637 | { |
5638 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5639 | char *ptr, *buf = NULL; | |
5640 | int err = -ENOMEM; | |
5641 | ||
60559da4 | 5642 | file = kmalloc(sizeof(*file), GFP_NOIO); |
2a2275d6 | 5643 | |
32a7627c N |
5644 | if (!file) |
5645 | goto out; | |
5646 | ||
5647 | /* bitmap disabled, zero the first byte and copy out */ | |
1ec885cd | 5648 | if (!mddev->bitmap || !mddev->bitmap->storage.file) { |
32a7627c N |
5649 | file->pathname[0] = '\0'; |
5650 | goto copy_out; | |
5651 | } | |
5652 | ||
5653 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5654 | if (!buf) | |
5655 | goto out; | |
5656 | ||
1ec885cd N |
5657 | ptr = d_path(&mddev->bitmap->storage.file->f_path, |
5658 | buf, sizeof(file->pathname)); | |
6bcfd601 | 5659 | if (IS_ERR(ptr)) |
32a7627c N |
5660 | goto out; |
5661 | ||
5662 | strcpy(file->pathname, ptr); | |
5663 | ||
5664 | copy_out: | |
5665 | err = 0; | |
5666 | if (copy_to_user(arg, file, sizeof(*file))) | |
5667 | err = -EFAULT; | |
5668 | out: | |
5669 | kfree(buf); | |
5670 | kfree(file); | |
5671 | return err; | |
5672 | } | |
5673 | ||
fd01b88c | 5674 | static int get_disk_info(struct mddev * mddev, void __user * arg) |
1da177e4 LT |
5675 | { |
5676 | mdu_disk_info_t info; | |
3cb03002 | 5677 | struct md_rdev *rdev; |
1da177e4 LT |
5678 | |
5679 | if (copy_from_user(&info, arg, sizeof(info))) | |
5680 | return -EFAULT; | |
5681 | ||
1ca69c4b N |
5682 | rcu_read_lock(); |
5683 | rdev = find_rdev_nr_rcu(mddev, info.number); | |
1da177e4 LT |
5684 | if (rdev) { |
5685 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5686 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5687 | info.raid_disk = rdev->raid_disk; | |
5688 | info.state = 0; | |
b2d444d7 | 5689 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5690 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5691 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5692 | info.state |= (1<<MD_DISK_ACTIVE); |
5693 | info.state |= (1<<MD_DISK_SYNC); | |
5694 | } | |
8ddf9efe N |
5695 | if (test_bit(WriteMostly, &rdev->flags)) |
5696 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5697 | } else { |
5698 | info.major = info.minor = 0; | |
5699 | info.raid_disk = -1; | |
5700 | info.state = (1<<MD_DISK_REMOVED); | |
5701 | } | |
1ca69c4b | 5702 | rcu_read_unlock(); |
1da177e4 LT |
5703 | |
5704 | if (copy_to_user(arg, &info, sizeof(info))) | |
5705 | return -EFAULT; | |
5706 | ||
5707 | return 0; | |
5708 | } | |
5709 | ||
fd01b88c | 5710 | static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info) |
1da177e4 LT |
5711 | { |
5712 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5713 | struct md_rdev *rdev; |
1da177e4 LT |
5714 | dev_t dev = MKDEV(info->major,info->minor); |
5715 | ||
5716 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5717 | return -EOVERFLOW; | |
5718 | ||
5719 | if (!mddev->raid_disks) { | |
5720 | int err; | |
5721 | /* expecting a device which has a superblock */ | |
5722 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5723 | if (IS_ERR(rdev)) { | |
5724 | printk(KERN_WARNING | |
5725 | "md: md_import_device returned %ld\n", | |
5726 | PTR_ERR(rdev)); | |
5727 | return PTR_ERR(rdev); | |
5728 | } | |
5729 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
5730 | struct md_rdev *rdev0 |
5731 | = list_entry(mddev->disks.next, | |
5732 | struct md_rdev, same_set); | |
a9f326eb | 5733 | err = super_types[mddev->major_version] |
1da177e4 LT |
5734 | .load_super(rdev, rdev0, mddev->minor_version); |
5735 | if (err < 0) { | |
5736 | printk(KERN_WARNING | |
5737 | "md: %s has different UUID to %s\n", | |
5738 | bdevname(rdev->bdev,b), | |
5739 | bdevname(rdev0->bdev,b2)); | |
5740 | export_rdev(rdev); | |
5741 | return -EINVAL; | |
5742 | } | |
5743 | } | |
5744 | err = bind_rdev_to_array(rdev, mddev); | |
5745 | if (err) | |
5746 | export_rdev(rdev); | |
5747 | return err; | |
5748 | } | |
5749 | ||
5750 | /* | |
5751 | * add_new_disk can be used once the array is assembled | |
5752 | * to add "hot spares". They must already have a superblock | |
5753 | * written | |
5754 | */ | |
5755 | if (mddev->pers) { | |
5756 | int err; | |
5757 | if (!mddev->pers->hot_add_disk) { | |
5758 | printk(KERN_WARNING | |
5759 | "%s: personality does not support diskops!\n", | |
5760 | mdname(mddev)); | |
5761 | return -EINVAL; | |
5762 | } | |
7b1e35f6 N |
5763 | if (mddev->persistent) |
5764 | rdev = md_import_device(dev, mddev->major_version, | |
5765 | mddev->minor_version); | |
5766 | else | |
5767 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5768 | if (IS_ERR(rdev)) { |
5769 | printk(KERN_WARNING | |
5770 | "md: md_import_device returned %ld\n", | |
5771 | PTR_ERR(rdev)); | |
5772 | return PTR_ERR(rdev); | |
5773 | } | |
1a855a06 | 5774 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5775 | if (!mddev->persistent) { |
5776 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5777 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5778 | rdev->raid_disk = info->raid_disk; |
bf572541 | 5779 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 5780 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 5781 | } else |
41158c7e N |
5782 | rdev->raid_disk = -1; |
5783 | } else | |
5784 | super_types[mddev->major_version]. | |
5785 | validate_super(mddev, rdev); | |
bedd86b7 | 5786 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 5787 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
5788 | /* This was a hot-add request, but events doesn't |
5789 | * match, so reject it. | |
5790 | */ | |
5791 | export_rdev(rdev); | |
5792 | return -EINVAL; | |
5793 | } | |
5794 | ||
1a855a06 N |
5795 | if (test_bit(In_sync, &rdev->flags)) |
5796 | rdev->saved_raid_disk = rdev->raid_disk; | |
5797 | else | |
5798 | rdev->saved_raid_disk = -1; | |
41158c7e | 5799 | |
b2d444d7 | 5800 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5801 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5802 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5803 | else |
5804 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5805 | |
1da177e4 LT |
5806 | rdev->raid_disk = -1; |
5807 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5808 | if (!err && !mddev->pers->hot_remove_disk) { |
5809 | /* If there is hot_add_disk but no hot_remove_disk | |
5810 | * then added disks for geometry changes, | |
5811 | * and should be added immediately. | |
5812 | */ | |
5813 | super_types[mddev->major_version]. | |
5814 | validate_super(mddev, rdev); | |
5815 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5816 | if (err) | |
5817 | unbind_rdev_from_array(rdev); | |
5818 | } | |
1da177e4 LT |
5819 | if (err) |
5820 | export_rdev(rdev); | |
52664732 | 5821 | else |
00bcb4ac | 5822 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5823 | |
7ceb17e8 | 5824 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
72a23c21 NB |
5825 | if (mddev->degraded) |
5826 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5827 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
9864c005 | 5828 | if (!err) |
5829 | md_new_event(mddev); | |
005eca5e | 5830 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5831 | return err; |
5832 | } | |
5833 | ||
5834 | /* otherwise, add_new_disk is only allowed | |
5835 | * for major_version==0 superblocks | |
5836 | */ | |
5837 | if (mddev->major_version != 0) { | |
5838 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5839 | mdname(mddev)); | |
5840 | return -EINVAL; | |
5841 | } | |
5842 | ||
5843 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5844 | int err; | |
d710e138 | 5845 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5846 | if (IS_ERR(rdev)) { |
5847 | printk(KERN_WARNING | |
5848 | "md: error, md_import_device() returned %ld\n", | |
5849 | PTR_ERR(rdev)); | |
5850 | return PTR_ERR(rdev); | |
5851 | } | |
5852 | rdev->desc_nr = info->number; | |
5853 | if (info->raid_disk < mddev->raid_disks) | |
5854 | rdev->raid_disk = info->raid_disk; | |
5855 | else | |
5856 | rdev->raid_disk = -1; | |
5857 | ||
1da177e4 | 5858 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5859 | if (info->state & (1<<MD_DISK_SYNC)) |
5860 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5861 | |
8ddf9efe N |
5862 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5863 | set_bit(WriteMostly, &rdev->flags); | |
5864 | ||
1da177e4 LT |
5865 | if (!mddev->persistent) { |
5866 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5867 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5868 | } else | |
57b2caa3 | 5869 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5870 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5871 | |
2bf071bf N |
5872 | err = bind_rdev_to_array(rdev, mddev); |
5873 | if (err) { | |
5874 | export_rdev(rdev); | |
5875 | return err; | |
5876 | } | |
1da177e4 LT |
5877 | } |
5878 | ||
5879 | return 0; | |
5880 | } | |
5881 | ||
fd01b88c | 5882 | static int hot_remove_disk(struct mddev * mddev, dev_t dev) |
1da177e4 LT |
5883 | { |
5884 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 5885 | struct md_rdev *rdev; |
1da177e4 | 5886 | |
1da177e4 LT |
5887 | rdev = find_rdev(mddev, dev); |
5888 | if (!rdev) | |
5889 | return -ENXIO; | |
5890 | ||
3ea8929d N |
5891 | clear_bit(Blocked, &rdev->flags); |
5892 | remove_and_add_spares(mddev, rdev); | |
5893 | ||
1da177e4 LT |
5894 | if (rdev->raid_disk >= 0) |
5895 | goto busy; | |
5896 | ||
5897 | kick_rdev_from_array(rdev); | |
850b2b42 | 5898 | md_update_sb(mddev, 1); |
d7603b7e | 5899 | md_new_event(mddev); |
1da177e4 LT |
5900 | |
5901 | return 0; | |
5902 | busy: | |
fdefa4d8 | 5903 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5904 | bdevname(rdev->bdev,b), mdname(mddev)); |
5905 | return -EBUSY; | |
5906 | } | |
5907 | ||
fd01b88c | 5908 | static int hot_add_disk(struct mddev * mddev, dev_t dev) |
1da177e4 LT |
5909 | { |
5910 | char b[BDEVNAME_SIZE]; | |
5911 | int err; | |
3cb03002 | 5912 | struct md_rdev *rdev; |
1da177e4 LT |
5913 | |
5914 | if (!mddev->pers) | |
5915 | return -ENODEV; | |
5916 | ||
5917 | if (mddev->major_version != 0) { | |
5918 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5919 | " version-0 superblocks.\n", | |
5920 | mdname(mddev)); | |
5921 | return -EINVAL; | |
5922 | } | |
5923 | if (!mddev->pers->hot_add_disk) { | |
5924 | printk(KERN_WARNING | |
5925 | "%s: personality does not support diskops!\n", | |
5926 | mdname(mddev)); | |
5927 | return -EINVAL; | |
5928 | } | |
5929 | ||
d710e138 | 5930 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5931 | if (IS_ERR(rdev)) { |
5932 | printk(KERN_WARNING | |
5933 | "md: error, md_import_device() returned %ld\n", | |
5934 | PTR_ERR(rdev)); | |
5935 | return -EINVAL; | |
5936 | } | |
5937 | ||
5938 | if (mddev->persistent) | |
57b2caa3 | 5939 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5940 | else |
77304d2a | 5941 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5942 | |
8190e754 | 5943 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5944 | |
b2d444d7 | 5945 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5946 | printk(KERN_WARNING |
5947 | "md: can not hot-add faulty %s disk to %s!\n", | |
5948 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5949 | err = -EINVAL; | |
5950 | goto abort_export; | |
5951 | } | |
b2d444d7 | 5952 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5953 | rdev->desc_nr = -1; |
5842730d | 5954 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5955 | err = bind_rdev_to_array(rdev, mddev); |
5956 | if (err) | |
5957 | goto abort_export; | |
1da177e4 LT |
5958 | |
5959 | /* | |
5960 | * The rest should better be atomic, we can have disk failures | |
5961 | * noticed in interrupt contexts ... | |
5962 | */ | |
5963 | ||
1da177e4 LT |
5964 | rdev->raid_disk = -1; |
5965 | ||
850b2b42 | 5966 | md_update_sb(mddev, 1); |
1da177e4 LT |
5967 | |
5968 | /* | |
5969 | * Kick recovery, maybe this spare has to be added to the | |
5970 | * array immediately. | |
5971 | */ | |
5972 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5973 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5974 | md_new_event(mddev); |
1da177e4 LT |
5975 | return 0; |
5976 | ||
1da177e4 LT |
5977 | abort_export: |
5978 | export_rdev(rdev); | |
5979 | return err; | |
5980 | } | |
5981 | ||
fd01b88c | 5982 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c N |
5983 | { |
5984 | int err; | |
5985 | ||
36fa3063 N |
5986 | if (mddev->pers) { |
5987 | if (!mddev->pers->quiesce) | |
5988 | return -EBUSY; | |
5989 | if (mddev->recovery || mddev->sync_thread) | |
5990 | return -EBUSY; | |
5991 | /* we should be able to change the bitmap.. */ | |
5992 | } | |
32a7627c | 5993 | |
32a7627c | 5994 | |
36fa3063 N |
5995 | if (fd >= 0) { |
5996 | if (mddev->bitmap) | |
5997 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5998 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5999 | |
c3d9714e | 6000 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
6001 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
6002 | mdname(mddev)); | |
6003 | return -EBADF; | |
6004 | } | |
6005 | ||
c3d9714e | 6006 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
6007 | if (err) { |
6008 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
6009 | mdname(mddev)); | |
c3d9714e N |
6010 | fput(mddev->bitmap_info.file); |
6011 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
6012 | return err; |
6013 | } | |
c3d9714e | 6014 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6015 | } else if (mddev->bitmap == NULL) |
6016 | return -ENOENT; /* cannot remove what isn't there */ | |
6017 | err = 0; | |
6018 | if (mddev->pers) { | |
6019 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6020 | if (fd >= 0) { |
36fa3063 | 6021 | err = bitmap_create(mddev); |
69e51b44 N |
6022 | if (!err) |
6023 | err = bitmap_load(mddev); | |
6024 | } | |
d7375ab3 | 6025 | if (fd < 0 || err) { |
36fa3063 | 6026 | bitmap_destroy(mddev); |
d7375ab3 N |
6027 | fd = -1; /* make sure to put the file */ |
6028 | } | |
36fa3063 | 6029 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6030 | } |
6031 | if (fd < 0) { | |
c3d9714e N |
6032 | if (mddev->bitmap_info.file) { |
6033 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
6034 | fput(mddev->bitmap_info.file); | |
acc55e22 | 6035 | } |
c3d9714e | 6036 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
6037 | } |
6038 | ||
32a7627c N |
6039 | return err; |
6040 | } | |
6041 | ||
1da177e4 LT |
6042 | /* |
6043 | * set_array_info is used two different ways | |
6044 | * The original usage is when creating a new array. | |
6045 | * In this usage, raid_disks is > 0 and it together with | |
6046 | * level, size, not_persistent,layout,chunksize determine the | |
6047 | * shape of the array. | |
6048 | * This will always create an array with a type-0.90.0 superblock. | |
6049 | * The newer usage is when assembling an array. | |
6050 | * In this case raid_disks will be 0, and the major_version field is | |
6051 | * use to determine which style super-blocks are to be found on the devices. | |
6052 | * The minor and patch _version numbers are also kept incase the | |
6053 | * super_block handler wishes to interpret them. | |
6054 | */ | |
fd01b88c | 6055 | static int set_array_info(struct mddev * mddev, mdu_array_info_t *info) |
1da177e4 LT |
6056 | { |
6057 | ||
6058 | if (info->raid_disks == 0) { | |
6059 | /* just setting version number for superblock loading */ | |
6060 | if (info->major_version < 0 || | |
50511da3 | 6061 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6062 | super_types[info->major_version].name == NULL) { |
6063 | /* maybe try to auto-load a module? */ | |
6064 | printk(KERN_INFO | |
6065 | "md: superblock version %d not known\n", | |
6066 | info->major_version); | |
6067 | return -EINVAL; | |
6068 | } | |
6069 | mddev->major_version = info->major_version; | |
6070 | mddev->minor_version = info->minor_version; | |
6071 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6072 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6073 | /* ensure mddev_put doesn't delete this now that there |
6074 | * is some minimal configuration. | |
6075 | */ | |
6076 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
6077 | return 0; |
6078 | } | |
6079 | mddev->major_version = MD_MAJOR_VERSION; | |
6080 | mddev->minor_version = MD_MINOR_VERSION; | |
6081 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
6082 | mddev->ctime = get_seconds(); | |
6083 | ||
6084 | mddev->level = info->level; | |
17115e03 | 6085 | mddev->clevel[0] = 0; |
58c0fed4 | 6086 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6087 | mddev->raid_disks = info->raid_disks; |
6088 | /* don't set md_minor, it is determined by which /dev/md* was | |
6089 | * openned | |
6090 | */ | |
6091 | if (info->state & (1<<MD_SB_CLEAN)) | |
6092 | mddev->recovery_cp = MaxSector; | |
6093 | else | |
6094 | mddev->recovery_cp = 0; | |
6095 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6096 | mddev->external = 0; |
1da177e4 LT |
6097 | |
6098 | mddev->layout = info->layout; | |
9d8f0363 | 6099 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
6100 | |
6101 | mddev->max_disks = MD_SB_DISKS; | |
6102 | ||
e691063a N |
6103 | if (mddev->persistent) |
6104 | mddev->flags = 0; | |
850b2b42 | 6105 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6106 | |
c3d9714e | 6107 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6108 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6109 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6110 | |
f6705578 N |
6111 | mddev->reshape_position = MaxSector; |
6112 | ||
1da177e4 LT |
6113 | /* |
6114 | * Generate a 128 bit UUID | |
6115 | */ | |
6116 | get_random_bytes(mddev->uuid, 16); | |
6117 | ||
f6705578 | 6118 | mddev->new_level = mddev->level; |
664e7c41 | 6119 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6120 | mddev->new_layout = mddev->layout; |
6121 | mddev->delta_disks = 0; | |
2c810cdd | 6122 | mddev->reshape_backwards = 0; |
f6705578 | 6123 | |
1da177e4 LT |
6124 | return 0; |
6125 | } | |
6126 | ||
fd01b88c | 6127 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6128 | { |
b522adcd DW |
6129 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6130 | ||
6131 | if (mddev->external_size) | |
6132 | return; | |
6133 | ||
1f403624 DW |
6134 | mddev->array_sectors = array_sectors; |
6135 | } | |
6136 | EXPORT_SYMBOL(md_set_array_sectors); | |
6137 | ||
fd01b88c | 6138 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6139 | { |
3cb03002 | 6140 | struct md_rdev *rdev; |
a35b0d69 | 6141 | int rv; |
d71f9f88 | 6142 | int fit = (num_sectors == 0); |
a35b0d69 N |
6143 | |
6144 | if (mddev->pers->resize == NULL) | |
6145 | return -EINVAL; | |
d71f9f88 AN |
6146 | /* The "num_sectors" is the number of sectors of each device that |
6147 | * is used. This can only make sense for arrays with redundancy. | |
6148 | * linear and raid0 always use whatever space is available. We can only | |
6149 | * consider changing this number if no resync or reconstruction is | |
6150 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6151 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6152 | * of each device. If num_sectors is zero, we find the largest size |
6153 | * that fits. | |
a35b0d69 N |
6154 | */ |
6155 | if (mddev->sync_thread) | |
6156 | return -EBUSY; | |
a4a6125a | 6157 | |
dafb20fa | 6158 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6159 | sector_t avail = rdev->sectors; |
01ab5662 | 6160 | |
d71f9f88 AN |
6161 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6162 | num_sectors = avail; | |
6163 | if (avail < num_sectors) | |
a35b0d69 N |
6164 | return -ENOSPC; |
6165 | } | |
d71f9f88 | 6166 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
6167 | if (!rv) |
6168 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
6169 | return rv; |
6170 | } | |
6171 | ||
fd01b88c | 6172 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6173 | { |
6174 | int rv; | |
c6563a8c | 6175 | struct md_rdev *rdev; |
da943b99 | 6176 | /* change the number of raid disks */ |
63c70c4f | 6177 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
6178 | return -EINVAL; |
6179 | if (raid_disks <= 0 || | |
233fca36 | 6180 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6181 | return -EINVAL; |
63c70c4f | 6182 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 6183 | return -EBUSY; |
c6563a8c N |
6184 | |
6185 | rdev_for_each(rdev, mddev) { | |
6186 | if (mddev->raid_disks < raid_disks && | |
6187 | rdev->data_offset < rdev->new_data_offset) | |
6188 | return -EINVAL; | |
6189 | if (mddev->raid_disks > raid_disks && | |
6190 | rdev->data_offset > rdev->new_data_offset) | |
6191 | return -EINVAL; | |
6192 | } | |
6193 | ||
63c70c4f | 6194 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6195 | if (mddev->delta_disks < 0) |
6196 | mddev->reshape_backwards = 1; | |
6197 | else if (mddev->delta_disks > 0) | |
6198 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6199 | |
6200 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6201 | if (rv < 0) { |
de171cb9 | 6202 | mddev->delta_disks = 0; |
2c810cdd N |
6203 | mddev->reshape_backwards = 0; |
6204 | } | |
da943b99 N |
6205 | return rv; |
6206 | } | |
6207 | ||
6208 | ||
1da177e4 LT |
6209 | /* |
6210 | * update_array_info is used to change the configuration of an | |
6211 | * on-line array. | |
6212 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6213 | * fields in the info are checked against the array. | |
6214 | * Any differences that cannot be handled will cause an error. | |
6215 | * Normally, only one change can be managed at a time. | |
6216 | */ | |
fd01b88c | 6217 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6218 | { |
6219 | int rv = 0; | |
6220 | int cnt = 0; | |
36fa3063 N |
6221 | int state = 0; |
6222 | ||
6223 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6224 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6225 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6226 | |
6227 | if (mddev->major_version != info->major_version || | |
6228 | mddev->minor_version != info->minor_version || | |
6229 | /* mddev->patch_version != info->patch_version || */ | |
6230 | mddev->ctime != info->ctime || | |
6231 | mddev->level != info->level || | |
6232 | /* mddev->layout != info->layout || */ | |
6233 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 6234 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6235 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6236 | ((state^info->state) & 0xfffffe00) | |
6237 | ) | |
1da177e4 LT |
6238 | return -EINVAL; |
6239 | /* Check there is only one change */ | |
58c0fed4 AN |
6240 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6241 | cnt++; | |
6242 | if (mddev->raid_disks != info->raid_disks) | |
6243 | cnt++; | |
6244 | if (mddev->layout != info->layout) | |
6245 | cnt++; | |
6246 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6247 | cnt++; | |
6248 | if (cnt == 0) | |
6249 | return 0; | |
6250 | if (cnt > 1) | |
6251 | return -EINVAL; | |
1da177e4 LT |
6252 | |
6253 | if (mddev->layout != info->layout) { | |
6254 | /* Change layout | |
6255 | * we don't need to do anything at the md level, the | |
6256 | * personality will take care of it all. | |
6257 | */ | |
50ac168a | 6258 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6259 | return -EINVAL; |
597a711b N |
6260 | else { |
6261 | mddev->new_layout = info->layout; | |
50ac168a | 6262 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6263 | if (rv) |
6264 | mddev->new_layout = mddev->layout; | |
6265 | return rv; | |
6266 | } | |
1da177e4 | 6267 | } |
58c0fed4 | 6268 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6269 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6270 | |
da943b99 N |
6271 | if (mddev->raid_disks != info->raid_disks) |
6272 | rv = update_raid_disks(mddev, info->raid_disks); | |
6273 | ||
36fa3063 N |
6274 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
6275 | if (mddev->pers->quiesce == NULL) | |
6276 | return -EINVAL; | |
6277 | if (mddev->recovery || mddev->sync_thread) | |
6278 | return -EBUSY; | |
6279 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
6280 | /* add the bitmap */ | |
6281 | if (mddev->bitmap) | |
6282 | return -EEXIST; | |
c3d9714e | 6283 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 6284 | return -EINVAL; |
c3d9714e N |
6285 | mddev->bitmap_info.offset = |
6286 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6287 | mddev->bitmap_info.space = |
6288 | mddev->bitmap_info.default_space; | |
36fa3063 N |
6289 | mddev->pers->quiesce(mddev, 1); |
6290 | rv = bitmap_create(mddev); | |
69e51b44 N |
6291 | if (!rv) |
6292 | rv = bitmap_load(mddev); | |
36fa3063 N |
6293 | if (rv) |
6294 | bitmap_destroy(mddev); | |
6295 | mddev->pers->quiesce(mddev, 0); | |
6296 | } else { | |
6297 | /* remove the bitmap */ | |
6298 | if (!mddev->bitmap) | |
6299 | return -ENOENT; | |
1ec885cd | 6300 | if (mddev->bitmap->storage.file) |
36fa3063 N |
6301 | return -EINVAL; |
6302 | mddev->pers->quiesce(mddev, 1); | |
6303 | bitmap_destroy(mddev); | |
6304 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6305 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6306 | } |
6307 | } | |
850b2b42 | 6308 | md_update_sb(mddev, 1); |
1da177e4 LT |
6309 | return rv; |
6310 | } | |
6311 | ||
fd01b88c | 6312 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6313 | { |
3cb03002 | 6314 | struct md_rdev *rdev; |
1ca69c4b | 6315 | int err = 0; |
1da177e4 LT |
6316 | |
6317 | if (mddev->pers == NULL) | |
6318 | return -ENODEV; | |
6319 | ||
1ca69c4b N |
6320 | rcu_read_lock(); |
6321 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6322 | if (!rdev) |
1ca69c4b N |
6323 | err = -ENODEV; |
6324 | else { | |
6325 | md_error(mddev, rdev); | |
6326 | if (!test_bit(Faulty, &rdev->flags)) | |
6327 | err = -EBUSY; | |
6328 | } | |
6329 | rcu_read_unlock(); | |
6330 | return err; | |
1da177e4 LT |
6331 | } |
6332 | ||
2f9618ce AN |
6333 | /* |
6334 | * We have a problem here : there is no easy way to give a CHS | |
6335 | * virtual geometry. We currently pretend that we have a 2 heads | |
6336 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6337 | * dosfs just mad... ;-) | |
6338 | */ | |
a885c8c4 CH |
6339 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6340 | { | |
fd01b88c | 6341 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6342 | |
6343 | geo->heads = 2; | |
6344 | geo->sectors = 4; | |
49ce6cea | 6345 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6346 | return 0; |
6347 | } | |
6348 | ||
a39907fa | 6349 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6350 | unsigned int cmd, unsigned long arg) |
6351 | { | |
6352 | int err = 0; | |
6353 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6354 | struct mddev *mddev = NULL; |
e2218350 | 6355 | int ro; |
1da177e4 | 6356 | |
506c9e44 N |
6357 | switch (cmd) { |
6358 | case RAID_VERSION: | |
6359 | case GET_ARRAY_INFO: | |
6360 | case GET_DISK_INFO: | |
6361 | break; | |
6362 | default: | |
6363 | if (!capable(CAP_SYS_ADMIN)) | |
6364 | return -EACCES; | |
6365 | } | |
1da177e4 LT |
6366 | |
6367 | /* | |
6368 | * Commands dealing with the RAID driver but not any | |
6369 | * particular array: | |
6370 | */ | |
c02c0aeb N |
6371 | switch (cmd) { |
6372 | case RAID_VERSION: | |
6373 | err = get_version(argp); | |
6374 | goto done; | |
1da177e4 | 6375 | |
c02c0aeb N |
6376 | case PRINT_RAID_DEBUG: |
6377 | err = 0; | |
6378 | md_print_devices(); | |
6379 | goto done; | |
1da177e4 LT |
6380 | |
6381 | #ifndef MODULE | |
c02c0aeb N |
6382 | case RAID_AUTORUN: |
6383 | err = 0; | |
6384 | autostart_arrays(arg); | |
6385 | goto done; | |
1da177e4 | 6386 | #endif |
c02c0aeb | 6387 | default:; |
1da177e4 LT |
6388 | } |
6389 | ||
6390 | /* | |
6391 | * Commands creating/starting a new array: | |
6392 | */ | |
6393 | ||
a39907fa | 6394 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6395 | |
6396 | if (!mddev) { | |
6397 | BUG(); | |
6398 | goto abort; | |
6399 | } | |
6400 | ||
1ca69c4b N |
6401 | /* Some actions do not requires the mutex */ |
6402 | switch (cmd) { | |
6403 | case GET_ARRAY_INFO: | |
6404 | if (!mddev->raid_disks && !mddev->external) | |
6405 | err = -ENODEV; | |
6406 | else | |
6407 | err = get_array_info(mddev, argp); | |
6408 | goto abort; | |
6409 | ||
6410 | case GET_DISK_INFO: | |
6411 | if (!mddev->raid_disks && !mddev->external) | |
6412 | err = -ENODEV; | |
6413 | else | |
6414 | err = get_disk_info(mddev, argp); | |
6415 | goto abort; | |
6416 | ||
6417 | case SET_DISK_FAULTY: | |
6418 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
6419 | goto abort; | |
6420 | } | |
6421 | ||
a7a3f08d N |
6422 | if (cmd == ADD_NEW_DISK) |
6423 | /* need to ensure md_delayed_delete() has completed */ | |
6424 | flush_workqueue(md_misc_wq); | |
6425 | ||
90f5f7ad HR |
6426 | if (cmd == HOT_REMOVE_DISK) |
6427 | /* need to ensure recovery thread has run */ | |
6428 | wait_event_interruptible_timeout(mddev->sb_wait, | |
6429 | !test_bit(MD_RECOVERY_NEEDED, | |
6430 | &mddev->flags), | |
6431 | msecs_to_jiffies(5000)); | |
260fa034 N |
6432 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
6433 | /* Need to flush page cache, and ensure no-one else opens | |
6434 | * and writes | |
6435 | */ | |
6436 | mutex_lock(&mddev->open_mutex); | |
6437 | if (atomic_read(&mddev->openers) > 1) { | |
6438 | mutex_unlock(&mddev->open_mutex); | |
6439 | err = -EBUSY; | |
6440 | goto abort; | |
6441 | } | |
6442 | set_bit(MD_STILL_CLOSED, &mddev->flags); | |
6443 | mutex_unlock(&mddev->open_mutex); | |
6444 | sync_blockdev(bdev); | |
6445 | } | |
1da177e4 LT |
6446 | err = mddev_lock(mddev); |
6447 | if (err) { | |
6448 | printk(KERN_INFO | |
6449 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
6450 | err, cmd); | |
6451 | goto abort; | |
6452 | } | |
6453 | ||
c02c0aeb N |
6454 | if (cmd == SET_ARRAY_INFO) { |
6455 | mdu_array_info_t info; | |
6456 | if (!arg) | |
6457 | memset(&info, 0, sizeof(info)); | |
6458 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6459 | err = -EFAULT; | |
6460 | goto abort_unlock; | |
6461 | } | |
6462 | if (mddev->pers) { | |
6463 | err = update_array_info(mddev, &info); | |
6464 | if (err) { | |
6465 | printk(KERN_WARNING "md: couldn't update" | |
6466 | " array info. %d\n", err); | |
6467 | goto abort_unlock; | |
1da177e4 LT |
6468 | } |
6469 | goto done_unlock; | |
c02c0aeb N |
6470 | } |
6471 | if (!list_empty(&mddev->disks)) { | |
6472 | printk(KERN_WARNING | |
6473 | "md: array %s already has disks!\n", | |
6474 | mdname(mddev)); | |
6475 | err = -EBUSY; | |
6476 | goto abort_unlock; | |
6477 | } | |
6478 | if (mddev->raid_disks) { | |
6479 | printk(KERN_WARNING | |
6480 | "md: array %s already initialised!\n", | |
6481 | mdname(mddev)); | |
6482 | err = -EBUSY; | |
6483 | goto abort_unlock; | |
6484 | } | |
6485 | err = set_array_info(mddev, &info); | |
6486 | if (err) { | |
6487 | printk(KERN_WARNING "md: couldn't set" | |
6488 | " array info. %d\n", err); | |
6489 | goto abort_unlock; | |
6490 | } | |
6491 | goto done_unlock; | |
1da177e4 LT |
6492 | } |
6493 | ||
6494 | /* | |
6495 | * Commands querying/configuring an existing array: | |
6496 | */ | |
32a7627c | 6497 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6498 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6499 | if ((!mddev->raid_disks && !mddev->external) |
6500 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6501 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6502 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
6503 | err = -ENODEV; |
6504 | goto abort_unlock; | |
6505 | } | |
6506 | ||
6507 | /* | |
6508 | * Commands even a read-only array can execute: | |
6509 | */ | |
c02c0aeb N |
6510 | switch (cmd) { |
6511 | case GET_BITMAP_FILE: | |
6512 | err = get_bitmap_file(mddev, argp); | |
6513 | goto done_unlock; | |
32a7627c | 6514 | |
c02c0aeb N |
6515 | case RESTART_ARRAY_RW: |
6516 | err = restart_array(mddev); | |
6517 | goto done_unlock; | |
1da177e4 | 6518 | |
c02c0aeb N |
6519 | case STOP_ARRAY: |
6520 | err = do_md_stop(mddev, 0, bdev); | |
6521 | goto done_unlock; | |
1da177e4 | 6522 | |
c02c0aeb N |
6523 | case STOP_ARRAY_RO: |
6524 | err = md_set_readonly(mddev, bdev); | |
6525 | goto done_unlock; | |
1da177e4 | 6526 | |
3ea8929d N |
6527 | case HOT_REMOVE_DISK: |
6528 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
6529 | goto done_unlock; | |
6530 | ||
7ceb17e8 N |
6531 | case ADD_NEW_DISK: |
6532 | /* We can support ADD_NEW_DISK on read-only arrays | |
6533 | * on if we are re-adding a preexisting device. | |
6534 | * So require mddev->pers and MD_DISK_SYNC. | |
6535 | */ | |
6536 | if (mddev->pers) { | |
6537 | mdu_disk_info_t info; | |
6538 | if (copy_from_user(&info, argp, sizeof(info))) | |
6539 | err = -EFAULT; | |
6540 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
6541 | /* Need to clear read-only for this */ | |
6542 | break; | |
6543 | else | |
6544 | err = add_new_disk(mddev, &info); | |
6545 | goto done_unlock; | |
6546 | } | |
6547 | break; | |
6548 | ||
c02c0aeb N |
6549 | case BLKROSET: |
6550 | if (get_user(ro, (int __user *)(arg))) { | |
6551 | err = -EFAULT; | |
6552 | goto done_unlock; | |
6553 | } | |
6554 | err = -EINVAL; | |
e2218350 | 6555 | |
c02c0aeb N |
6556 | /* if the bdev is going readonly the value of mddev->ro |
6557 | * does not matter, no writes are coming | |
6558 | */ | |
6559 | if (ro) | |
6560 | goto done_unlock; | |
e2218350 | 6561 | |
c02c0aeb N |
6562 | /* are we are already prepared for writes? */ |
6563 | if (mddev->ro != 1) | |
6564 | goto done_unlock; | |
e2218350 | 6565 | |
c02c0aeb N |
6566 | /* transitioning to readauto need only happen for |
6567 | * arrays that call md_write_start | |
6568 | */ | |
6569 | if (mddev->pers) { | |
6570 | err = restart_array(mddev); | |
6571 | if (err == 0) { | |
6572 | mddev->ro = 2; | |
6573 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6574 | } |
c02c0aeb N |
6575 | } |
6576 | goto done_unlock; | |
1da177e4 LT |
6577 | } |
6578 | ||
6579 | /* | |
6580 | * The remaining ioctls are changing the state of the | |
f91de92e N |
6581 | * superblock, so we do not allow them on read-only arrays. |
6582 | * However non-MD ioctls (e.g. get-size) will still come through | |
6583 | * here and hit the 'default' below, so only disallow | |
6584 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 6585 | */ |
bb57fc64 | 6586 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
6587 | if (mddev->ro == 2) { |
6588 | mddev->ro = 0; | |
00bcb4ac | 6589 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 6590 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
6591 | /* mddev_unlock will wake thread */ |
6592 | /* If a device failed while we were read-only, we | |
6593 | * need to make sure the metadata is updated now. | |
6594 | */ | |
6595 | if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
6596 | mddev_unlock(mddev); | |
6597 | wait_event(mddev->sb_wait, | |
6598 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) && | |
6599 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
29f097c4 | 6600 | mddev_lock_nointr(mddev); |
f3378b48 | 6601 | } |
f91de92e N |
6602 | } else { |
6603 | err = -EROFS; | |
6604 | goto abort_unlock; | |
6605 | } | |
1da177e4 LT |
6606 | } |
6607 | ||
c02c0aeb N |
6608 | switch (cmd) { |
6609 | case ADD_NEW_DISK: | |
1da177e4 | 6610 | { |
c02c0aeb N |
6611 | mdu_disk_info_t info; |
6612 | if (copy_from_user(&info, argp, sizeof(info))) | |
6613 | err = -EFAULT; | |
6614 | else | |
6615 | err = add_new_disk(mddev, &info); | |
6616 | goto done_unlock; | |
6617 | } | |
1da177e4 | 6618 | |
c02c0aeb N |
6619 | case HOT_ADD_DISK: |
6620 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
6621 | goto done_unlock; | |
1da177e4 | 6622 | |
c02c0aeb N |
6623 | case RUN_ARRAY: |
6624 | err = do_md_run(mddev); | |
6625 | goto done_unlock; | |
1da177e4 | 6626 | |
c02c0aeb N |
6627 | case SET_BITMAP_FILE: |
6628 | err = set_bitmap_file(mddev, (int)arg); | |
6629 | goto done_unlock; | |
32a7627c | 6630 | |
c02c0aeb N |
6631 | default: |
6632 | err = -EINVAL; | |
6633 | goto abort_unlock; | |
1da177e4 LT |
6634 | } |
6635 | ||
6636 | done_unlock: | |
6637 | abort_unlock: | |
d3374825 N |
6638 | if (mddev->hold_active == UNTIL_IOCTL && |
6639 | err != -EINVAL) | |
6640 | mddev->hold_active = 0; | |
1da177e4 LT |
6641 | mddev_unlock(mddev); |
6642 | ||
6643 | return err; | |
6644 | done: | |
6645 | if (err) | |
6646 | MD_BUG(); | |
6647 | abort: | |
6648 | return err; | |
6649 | } | |
aa98aa31 AB |
6650 | #ifdef CONFIG_COMPAT |
6651 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6652 | unsigned int cmd, unsigned long arg) | |
6653 | { | |
6654 | switch (cmd) { | |
6655 | case HOT_REMOVE_DISK: | |
6656 | case HOT_ADD_DISK: | |
6657 | case SET_DISK_FAULTY: | |
6658 | case SET_BITMAP_FILE: | |
6659 | /* These take in integer arg, do not convert */ | |
6660 | break; | |
6661 | default: | |
6662 | arg = (unsigned long)compat_ptr(arg); | |
6663 | break; | |
6664 | } | |
6665 | ||
6666 | return md_ioctl(bdev, mode, cmd, arg); | |
6667 | } | |
6668 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6669 | |
a39907fa | 6670 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6671 | { |
6672 | /* | |
6673 | * Succeed if we can lock the mddev, which confirms that | |
6674 | * it isn't being stopped right now. | |
6675 | */ | |
fd01b88c | 6676 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6677 | int err; |
6678 | ||
0c098220 YL |
6679 | if (!mddev) |
6680 | return -ENODEV; | |
6681 | ||
d3374825 N |
6682 | if (mddev->gendisk != bdev->bd_disk) { |
6683 | /* we are racing with mddev_put which is discarding this | |
6684 | * bd_disk. | |
6685 | */ | |
6686 | mddev_put(mddev); | |
6687 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6688 | flush_workqueue(md_misc_wq); |
d3374825 N |
6689 | /* Then retry the open from the top */ |
6690 | return -ERESTARTSYS; | |
6691 | } | |
6692 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6693 | ||
c8c00a69 | 6694 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6695 | goto out; |
6696 | ||
6697 | err = 0; | |
f2ea68cf | 6698 | atomic_inc(&mddev->openers); |
260fa034 | 6699 | clear_bit(MD_STILL_CLOSED, &mddev->flags); |
c8c00a69 | 6700 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6701 | |
f0b4f7e2 | 6702 | check_disk_change(bdev); |
1da177e4 LT |
6703 | out: |
6704 | return err; | |
6705 | } | |
6706 | ||
db2a144b | 6707 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6708 | { |
fd01b88c | 6709 | struct mddev *mddev = disk->private_data; |
1da177e4 | 6710 | |
52e5f9d1 | 6711 | BUG_ON(!mddev); |
f2ea68cf | 6712 | atomic_dec(&mddev->openers); |
1da177e4 | 6713 | mddev_put(mddev); |
1da177e4 | 6714 | } |
f0b4f7e2 N |
6715 | |
6716 | static int md_media_changed(struct gendisk *disk) | |
6717 | { | |
fd01b88c | 6718 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6719 | |
6720 | return mddev->changed; | |
6721 | } | |
6722 | ||
6723 | static int md_revalidate(struct gendisk *disk) | |
6724 | { | |
fd01b88c | 6725 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6726 | |
6727 | mddev->changed = 0; | |
6728 | return 0; | |
6729 | } | |
83d5cde4 | 6730 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6731 | { |
6732 | .owner = THIS_MODULE, | |
a39907fa AV |
6733 | .open = md_open, |
6734 | .release = md_release, | |
b492b852 | 6735 | .ioctl = md_ioctl, |
aa98aa31 AB |
6736 | #ifdef CONFIG_COMPAT |
6737 | .compat_ioctl = md_compat_ioctl, | |
6738 | #endif | |
a885c8c4 | 6739 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6740 | .media_changed = md_media_changed, |
6741 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6742 | }; |
6743 | ||
75c96f85 | 6744 | static int md_thread(void * arg) |
1da177e4 | 6745 | { |
2b8bf345 | 6746 | struct md_thread *thread = arg; |
1da177e4 | 6747 | |
1da177e4 LT |
6748 | /* |
6749 | * md_thread is a 'system-thread', it's priority should be very | |
6750 | * high. We avoid resource deadlocks individually in each | |
6751 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6752 | * the very same RT priority as kswapd, thus we will never get | |
6753 | * into a priority inversion deadlock. | |
6754 | * | |
6755 | * we definitely have to have equal or higher priority than | |
6756 | * bdflush, otherwise bdflush will deadlock if there are too | |
6757 | * many dirty RAID5 blocks. | |
6758 | */ | |
1da177e4 | 6759 | |
6985c43f | 6760 | allow_signal(SIGKILL); |
a6fb0934 | 6761 | while (!kthread_should_stop()) { |
1da177e4 | 6762 | |
93588e22 N |
6763 | /* We need to wait INTERRUPTIBLE so that |
6764 | * we don't add to the load-average. | |
6765 | * That means we need to be sure no signals are | |
6766 | * pending | |
6767 | */ | |
6768 | if (signal_pending(current)) | |
6769 | flush_signals(current); | |
6770 | ||
6771 | wait_event_interruptible_timeout | |
6772 | (thread->wqueue, | |
6773 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6774 | || kthread_should_stop(), | |
6775 | thread->timeout); | |
1da177e4 | 6776 | |
6c987910 N |
6777 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6778 | if (!kthread_should_stop()) | |
4ed8731d | 6779 | thread->run(thread); |
1da177e4 | 6780 | } |
a6fb0934 | 6781 | |
1da177e4 LT |
6782 | return 0; |
6783 | } | |
6784 | ||
2b8bf345 | 6785 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
6786 | { |
6787 | if (thread) { | |
36a4e1fe | 6788 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
6789 | set_bit(THREAD_WAKEUP, &thread->flags); |
6790 | wake_up(&thread->wqueue); | |
6791 | } | |
6792 | } | |
6793 | ||
4ed8731d SL |
6794 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
6795 | struct mddev *mddev, const char *name) | |
1da177e4 | 6796 | { |
2b8bf345 | 6797 | struct md_thread *thread; |
1da177e4 | 6798 | |
2b8bf345 | 6799 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
6800 | if (!thread) |
6801 | return NULL; | |
6802 | ||
1da177e4 LT |
6803 | init_waitqueue_head(&thread->wqueue); |
6804 | ||
1da177e4 LT |
6805 | thread->run = run; |
6806 | thread->mddev = mddev; | |
32a7627c | 6807 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6808 | thread->tsk = kthread_run(md_thread, thread, |
6809 | "%s_%s", | |
6810 | mdname(thread->mddev), | |
0232605d | 6811 | name); |
a6fb0934 | 6812 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6813 | kfree(thread); |
6814 | return NULL; | |
6815 | } | |
1da177e4 LT |
6816 | return thread; |
6817 | } | |
6818 | ||
2b8bf345 | 6819 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 6820 | { |
2b8bf345 | 6821 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
6822 | if (!thread) |
6823 | return; | |
36a4e1fe | 6824 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
6825 | /* Locking ensures that mddev_unlock does not wake_up a |
6826 | * non-existent thread | |
6827 | */ | |
6828 | spin_lock(&pers_lock); | |
6829 | *threadp = NULL; | |
6830 | spin_unlock(&pers_lock); | |
a6fb0934 N |
6831 | |
6832 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6833 | kfree(thread); |
6834 | } | |
6835 | ||
fd01b88c | 6836 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
6837 | { |
6838 | if (!mddev) { | |
6839 | MD_BUG(); | |
6840 | return; | |
6841 | } | |
6842 | ||
b2d444d7 | 6843 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6844 | return; |
6bfe0b49 | 6845 | |
de393cde | 6846 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
6847 | return; |
6848 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6849 | if (mddev->degraded) |
6850 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6851 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6852 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6853 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6854 | md_wakeup_thread(mddev->thread); | |
768a418d | 6855 | if (mddev->event_work.func) |
e804ac78 | 6856 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6857 | md_new_event_inintr(mddev); |
1da177e4 LT |
6858 | } |
6859 | ||
6860 | /* seq_file implementation /proc/mdstat */ | |
6861 | ||
6862 | static void status_unused(struct seq_file *seq) | |
6863 | { | |
6864 | int i = 0; | |
3cb03002 | 6865 | struct md_rdev *rdev; |
1da177e4 LT |
6866 | |
6867 | seq_printf(seq, "unused devices: "); | |
6868 | ||
159ec1fc | 6869 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6870 | char b[BDEVNAME_SIZE]; |
6871 | i++; | |
6872 | seq_printf(seq, "%s ", | |
6873 | bdevname(rdev->bdev,b)); | |
6874 | } | |
6875 | if (!i) | |
6876 | seq_printf(seq, "<none>"); | |
6877 | ||
6878 | seq_printf(seq, "\n"); | |
6879 | } | |
6880 | ||
6881 | ||
fd01b88c | 6882 | static void status_resync(struct seq_file *seq, struct mddev * mddev) |
1da177e4 | 6883 | { |
dd71cf6b N |
6884 | sector_t max_sectors, resync, res; |
6885 | unsigned long dt, db; | |
6886 | sector_t rt; | |
4588b42e N |
6887 | int scale; |
6888 | unsigned int per_milli; | |
1da177e4 | 6889 | |
72f36d59 N |
6890 | if (mddev->curr_resync <= 3) |
6891 | resync = 0; | |
6892 | else | |
6893 | resync = mddev->curr_resync | |
6894 | - atomic_read(&mddev->recovery_active); | |
1da177e4 | 6895 | |
c804cdec N |
6896 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
6897 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 6898 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6899 | else |
dd71cf6b | 6900 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6901 | |
6902 | /* | |
6903 | * Should not happen. | |
6904 | */ | |
dd71cf6b | 6905 | if (!max_sectors) { |
1da177e4 LT |
6906 | MD_BUG(); |
6907 | return; | |
6908 | } | |
4588b42e | 6909 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6910 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6911 | * u32, as those are the requirements for sector_div. |
6912 | * Thus 'scale' must be at least 10 | |
6913 | */ | |
6914 | scale = 10; | |
6915 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6916 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6917 | scale++; |
6918 | } | |
6919 | res = (resync>>scale)*1000; | |
dd71cf6b | 6920 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6921 | |
6922 | per_milli = res; | |
1da177e4 | 6923 | { |
4588b42e | 6924 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6925 | seq_printf(seq, "["); |
6926 | for (i = 0; i < x; i++) | |
6927 | seq_printf(seq, "="); | |
6928 | seq_printf(seq, ">"); | |
6929 | for (i = 0; i < y; i++) | |
6930 | seq_printf(seq, "."); | |
6931 | seq_printf(seq, "] "); | |
6932 | } | |
4588b42e | 6933 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6934 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6935 | "reshape" : | |
61df9d91 N |
6936 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6937 | "check" : | |
6938 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6939 | "resync" : "recovery"))), | |
6940 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6941 | (unsigned long long) resync/2, |
6942 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6943 | |
6944 | /* | |
1da177e4 LT |
6945 | * dt: time from mark until now |
6946 | * db: blocks written from mark until now | |
6947 | * rt: remaining time | |
dd71cf6b N |
6948 | * |
6949 | * rt is a sector_t, so could be 32bit or 64bit. | |
6950 | * So we divide before multiply in case it is 32bit and close | |
6951 | * to the limit. | |
25985edc | 6952 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
6953 | * near the end of resync when the number of remaining sectors |
6954 | * is close to 'db'. | |
6955 | * We then divide rt by 32 after multiplying by db to compensate. | |
6956 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6957 | */ |
6958 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6959 | if (!dt) dt++; | |
ff4e8d9a N |
6960 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6961 | - mddev->resync_mark_cnt; | |
1da177e4 | 6962 | |
dd71cf6b N |
6963 | rt = max_sectors - resync; /* number of remaining sectors */ |
6964 | sector_div(rt, db/32+1); | |
6965 | rt *= dt; | |
6966 | rt >>= 5; | |
6967 | ||
6968 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6969 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6970 | |
ff4e8d9a | 6971 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6972 | } |
6973 | ||
6974 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6975 | { | |
6976 | struct list_head *tmp; | |
6977 | loff_t l = *pos; | |
fd01b88c | 6978 | struct mddev *mddev; |
1da177e4 LT |
6979 | |
6980 | if (l >= 0x10000) | |
6981 | return NULL; | |
6982 | if (!l--) | |
6983 | /* header */ | |
6984 | return (void*)1; | |
6985 | ||
6986 | spin_lock(&all_mddevs_lock); | |
6987 | list_for_each(tmp,&all_mddevs) | |
6988 | if (!l--) { | |
fd01b88c | 6989 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
6990 | mddev_get(mddev); |
6991 | spin_unlock(&all_mddevs_lock); | |
6992 | return mddev; | |
6993 | } | |
6994 | spin_unlock(&all_mddevs_lock); | |
6995 | if (!l--) | |
6996 | return (void*)2;/* tail */ | |
6997 | return NULL; | |
6998 | } | |
6999 | ||
7000 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7001 | { | |
7002 | struct list_head *tmp; | |
fd01b88c | 7003 | struct mddev *next_mddev, *mddev = v; |
1da177e4 LT |
7004 | |
7005 | ++*pos; | |
7006 | if (v == (void*)2) | |
7007 | return NULL; | |
7008 | ||
7009 | spin_lock(&all_mddevs_lock); | |
7010 | if (v == (void*)1) | |
7011 | tmp = all_mddevs.next; | |
7012 | else | |
7013 | tmp = mddev->all_mddevs.next; | |
7014 | if (tmp != &all_mddevs) | |
fd01b88c | 7015 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7016 | else { |
7017 | next_mddev = (void*)2; | |
7018 | *pos = 0x10000; | |
7019 | } | |
7020 | spin_unlock(&all_mddevs_lock); | |
7021 | ||
7022 | if (v != (void*)1) | |
7023 | mddev_put(mddev); | |
7024 | return next_mddev; | |
7025 | ||
7026 | } | |
7027 | ||
7028 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7029 | { | |
fd01b88c | 7030 | struct mddev *mddev = v; |
1da177e4 LT |
7031 | |
7032 | if (mddev && v != (void*)1 && v != (void*)2) | |
7033 | mddev_put(mddev); | |
7034 | } | |
7035 | ||
7036 | static int md_seq_show(struct seq_file *seq, void *v) | |
7037 | { | |
fd01b88c | 7038 | struct mddev *mddev = v; |
dd8ac336 | 7039 | sector_t sectors; |
3cb03002 | 7040 | struct md_rdev *rdev; |
1da177e4 LT |
7041 | |
7042 | if (v == (void*)1) { | |
84fc4b56 | 7043 | struct md_personality *pers; |
1da177e4 LT |
7044 | seq_printf(seq, "Personalities : "); |
7045 | spin_lock(&pers_lock); | |
2604b703 N |
7046 | list_for_each_entry(pers, &pers_list, list) |
7047 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7048 | |
7049 | spin_unlock(&pers_lock); | |
7050 | seq_printf(seq, "\n"); | |
f1514638 | 7051 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7052 | return 0; |
7053 | } | |
7054 | if (v == (void*)2) { | |
7055 | status_unused(seq); | |
7056 | return 0; | |
7057 | } | |
7058 | ||
5dc5cf7d | 7059 | if (mddev_lock(mddev) < 0) |
1da177e4 | 7060 | return -EINTR; |
5dc5cf7d | 7061 | |
1da177e4 LT |
7062 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7063 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7064 | mddev->pers ? "" : "in"); | |
7065 | if (mddev->pers) { | |
f91de92e | 7066 | if (mddev->ro==1) |
1da177e4 | 7067 | seq_printf(seq, " (read-only)"); |
f91de92e | 7068 | if (mddev->ro==2) |
52720ae7 | 7069 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7070 | seq_printf(seq, " %s", mddev->pers->name); |
7071 | } | |
7072 | ||
dd8ac336 | 7073 | sectors = 0; |
dafb20fa | 7074 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
7075 | char b[BDEVNAME_SIZE]; |
7076 | seq_printf(seq, " %s[%d]", | |
7077 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7078 | if (test_bit(WriteMostly, &rdev->flags)) |
7079 | seq_printf(seq, "(W)"); | |
b2d444d7 | 7080 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7081 | seq_printf(seq, "(F)"); |
7082 | continue; | |
2d78f8c4 N |
7083 | } |
7084 | if (rdev->raid_disk < 0) | |
b325a32e | 7085 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7086 | if (test_bit(Replacement, &rdev->flags)) |
7087 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7088 | sectors += rdev->sectors; |
1da177e4 LT |
7089 | } |
7090 | ||
7091 | if (!list_empty(&mddev->disks)) { | |
7092 | if (mddev->pers) | |
7093 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7094 | (unsigned long long) |
7095 | mddev->array_sectors / 2); | |
1da177e4 LT |
7096 | else |
7097 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7098 | (unsigned long long)sectors / 2); |
1da177e4 | 7099 | } |
1cd6bf19 N |
7100 | if (mddev->persistent) { |
7101 | if (mddev->major_version != 0 || | |
7102 | mddev->minor_version != 90) { | |
7103 | seq_printf(seq," super %d.%d", | |
7104 | mddev->major_version, | |
7105 | mddev->minor_version); | |
7106 | } | |
e691063a N |
7107 | } else if (mddev->external) |
7108 | seq_printf(seq, " super external:%s", | |
7109 | mddev->metadata_type); | |
7110 | else | |
1cd6bf19 | 7111 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7112 | |
7113 | if (mddev->pers) { | |
d710e138 | 7114 | mddev->pers->status(seq, mddev); |
1da177e4 | 7115 | seq_printf(seq, "\n "); |
8e1b39d6 N |
7116 | if (mddev->pers->sync_request) { |
7117 | if (mddev->curr_resync > 2) { | |
d710e138 | 7118 | status_resync(seq, mddev); |
8e1b39d6 | 7119 | seq_printf(seq, "\n "); |
72f36d59 | 7120 | } else if (mddev->curr_resync >= 1) |
8e1b39d6 N |
7121 | seq_printf(seq, "\tresync=DELAYED\n "); |
7122 | else if (mddev->recovery_cp < MaxSector) | |
7123 | seq_printf(seq, "\tresync=PENDING\n "); | |
7124 | } | |
32a7627c N |
7125 | } else |
7126 | seq_printf(seq, "\n "); | |
7127 | ||
57148964 | 7128 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7129 | |
7130 | seq_printf(seq, "\n"); | |
7131 | } | |
7132 | mddev_unlock(mddev); | |
7133 | ||
7134 | return 0; | |
7135 | } | |
7136 | ||
110518bc | 7137 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7138 | .start = md_seq_start, |
7139 | .next = md_seq_next, | |
7140 | .stop = md_seq_stop, | |
7141 | .show = md_seq_show, | |
7142 | }; | |
7143 | ||
7144 | static int md_seq_open(struct inode *inode, struct file *file) | |
7145 | { | |
f1514638 | 7146 | struct seq_file *seq; |
1da177e4 LT |
7147 | int error; |
7148 | ||
7149 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7150 | if (error) |
f1514638 KS |
7151 | return error; |
7152 | ||
7153 | seq = file->private_data; | |
7154 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7155 | return error; |
7156 | } | |
7157 | ||
d7603b7e N |
7158 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7159 | { | |
f1514638 | 7160 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7161 | int mask; |
7162 | ||
7163 | poll_wait(filp, &md_event_waiters, wait); | |
7164 | ||
7165 | /* always allow read */ | |
7166 | mask = POLLIN | POLLRDNORM; | |
7167 | ||
f1514638 | 7168 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7169 | mask |= POLLERR | POLLPRI; |
7170 | return mask; | |
7171 | } | |
7172 | ||
fa027c2a | 7173 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7174 | .owner = THIS_MODULE, |
1da177e4 LT |
7175 | .open = md_seq_open, |
7176 | .read = seq_read, | |
7177 | .llseek = seq_lseek, | |
c3f94b40 | 7178 | .release = seq_release_private, |
d7603b7e | 7179 | .poll = mdstat_poll, |
1da177e4 LT |
7180 | }; |
7181 | ||
84fc4b56 | 7182 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7183 | { |
1da177e4 | 7184 | spin_lock(&pers_lock); |
2604b703 N |
7185 | list_add_tail(&p->list, &pers_list); |
7186 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
7187 | spin_unlock(&pers_lock); |
7188 | return 0; | |
7189 | } | |
7190 | ||
84fc4b56 | 7191 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7192 | { |
2604b703 | 7193 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 7194 | spin_lock(&pers_lock); |
2604b703 | 7195 | list_del_init(&p->list); |
1da177e4 LT |
7196 | spin_unlock(&pers_lock); |
7197 | return 0; | |
7198 | } | |
7199 | ||
fd01b88c | 7200 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7201 | { |
3cb03002 | 7202 | struct md_rdev * rdev; |
1da177e4 | 7203 | int idle; |
eea1bf38 | 7204 | int curr_events; |
1da177e4 LT |
7205 | |
7206 | idle = 1; | |
4b80991c N |
7207 | rcu_read_lock(); |
7208 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7209 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7210 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7211 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7212 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7213 | /* sync IO will cause sync_io to increase before the disk_stats |
7214 | * as sync_io is counted when a request starts, and | |
7215 | * disk_stats is counted when it completes. | |
7216 | * So resync activity will cause curr_events to be smaller than | |
7217 | * when there was no such activity. | |
7218 | * non-sync IO will cause disk_stat to increase without | |
7219 | * increasing sync_io so curr_events will (eventually) | |
7220 | * be larger than it was before. Once it becomes | |
7221 | * substantially larger, the test below will cause | |
7222 | * the array to appear non-idle, and resync will slow | |
7223 | * down. | |
7224 | * If there is a lot of outstanding resync activity when | |
7225 | * we set last_event to curr_events, then all that activity | |
7226 | * completing might cause the array to appear non-idle | |
7227 | * and resync will be slowed down even though there might | |
7228 | * not have been non-resync activity. This will only | |
7229 | * happen once though. 'last_events' will soon reflect | |
7230 | * the state where there is little or no outstanding | |
7231 | * resync requests, and further resync activity will | |
7232 | * always make curr_events less than last_events. | |
c0e48521 | 7233 | * |
1da177e4 | 7234 | */ |
eea1bf38 | 7235 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7236 | rdev->last_events = curr_events; |
7237 | idle = 0; | |
7238 | } | |
7239 | } | |
4b80991c | 7240 | rcu_read_unlock(); |
1da177e4 LT |
7241 | return idle; |
7242 | } | |
7243 | ||
fd01b88c | 7244 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7245 | { |
7246 | /* another "blocks" (512byte) blocks have been synced */ | |
7247 | atomic_sub(blocks, &mddev->recovery_active); | |
7248 | wake_up(&mddev->recovery_wait); | |
7249 | if (!ok) { | |
dfc70645 | 7250 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7251 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7252 | md_wakeup_thread(mddev->thread); |
7253 | // stop recovery, signal do_sync .... | |
7254 | } | |
7255 | } | |
7256 | ||
7257 | ||
06d91a5f N |
7258 | /* md_write_start(mddev, bi) |
7259 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7260 | * in superblock) before writing, schedule a superblock update |
7261 | * and wait for it to complete. | |
06d91a5f | 7262 | */ |
fd01b88c | 7263 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7264 | { |
0fd62b86 | 7265 | int did_change = 0; |
06d91a5f | 7266 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7267 | return; |
06d91a5f | 7268 | |
f91de92e N |
7269 | BUG_ON(mddev->ro == 1); |
7270 | if (mddev->ro == 2) { | |
7271 | /* need to switch to read/write */ | |
7272 | mddev->ro = 0; | |
7273 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7274 | md_wakeup_thread(mddev->thread); | |
25156198 | 7275 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7276 | did_change = 1; |
f91de92e | 7277 | } |
06d91a5f | 7278 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7279 | if (mddev->safemode == 1) |
7280 | mddev->safemode = 0; | |
06d91a5f | 7281 | if (mddev->in_sync) { |
a9701a30 | 7282 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
7283 | if (mddev->in_sync) { |
7284 | mddev->in_sync = 0; | |
850b2b42 | 7285 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 7286 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 7287 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7288 | did_change = 1; |
3d310eb7 | 7289 | } |
a9701a30 | 7290 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 7291 | } |
0fd62b86 | 7292 | if (did_change) |
00bcb4ac | 7293 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7294 | wait_event(mddev->sb_wait, |
09a44cc1 | 7295 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 LT |
7296 | } |
7297 | ||
fd01b88c | 7298 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7299 | { |
7300 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7301 | if (mddev->safemode == 2) | |
7302 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7303 | else if (mddev->safemode_delay) |
1da177e4 LT |
7304 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7305 | } | |
7306 | } | |
7307 | ||
2a2275d6 N |
7308 | /* md_allow_write(mddev) |
7309 | * Calling this ensures that the array is marked 'active' so that writes | |
7310 | * may proceed without blocking. It is important to call this before | |
7311 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7312 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
7313 | * |
7314 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
7315 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 7316 | */ |
fd01b88c | 7317 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7318 | { |
7319 | if (!mddev->pers) | |
b5470dc5 | 7320 | return 0; |
2a2275d6 | 7321 | if (mddev->ro) |
b5470dc5 | 7322 | return 0; |
1a0fd497 | 7323 | if (!mddev->pers->sync_request) |
b5470dc5 | 7324 | return 0; |
2a2275d6 N |
7325 | |
7326 | spin_lock_irq(&mddev->write_lock); | |
7327 | if (mddev->in_sync) { | |
7328 | mddev->in_sync = 0; | |
7329 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 7330 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
7331 | if (mddev->safemode_delay && |
7332 | mddev->safemode == 0) | |
7333 | mddev->safemode = 1; | |
7334 | spin_unlock_irq(&mddev->write_lock); | |
7335 | md_update_sb(mddev, 0); | |
00bcb4ac | 7336 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
7337 | } else |
7338 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 | 7339 | |
070dc6dd | 7340 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
7341 | return -EAGAIN; |
7342 | else | |
7343 | return 0; | |
2a2275d6 N |
7344 | } |
7345 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7346 | ||
1da177e4 LT |
7347 | #define SYNC_MARKS 10 |
7348 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7349 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7350 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7351 | { |
4ed8731d | 7352 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7353 | struct mddev *mddev2; |
1da177e4 LT |
7354 | unsigned int currspeed = 0, |
7355 | window; | |
57afd89f | 7356 | sector_t max_sectors,j, io_sectors; |
1da177e4 | 7357 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7358 | unsigned long update_time; |
1da177e4 LT |
7359 | sector_t mark_cnt[SYNC_MARKS]; |
7360 | int last_mark,m; | |
7361 | struct list_head *tmp; | |
7362 | sector_t last_check; | |
57afd89f | 7363 | int skipped = 0; |
3cb03002 | 7364 | struct md_rdev *rdev; |
c4a39551 | 7365 | char *desc, *action = NULL; |
7c2c57c9 | 7366 | struct blk_plug plug; |
1da177e4 LT |
7367 | |
7368 | /* just incase thread restarts... */ | |
7369 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7370 | return; | |
5fd6c1dc N |
7371 | if (mddev->ro) /* never try to sync a read-only array */ |
7372 | return; | |
1da177e4 | 7373 | |
61df9d91 | 7374 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 7375 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 7376 | desc = "data-check"; |
c4a39551 JB |
7377 | action = "check"; |
7378 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 7379 | desc = "requested-resync"; |
c4a39551 JB |
7380 | action = "repair"; |
7381 | } else | |
61df9d91 N |
7382 | desc = "resync"; |
7383 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7384 | desc = "reshape"; | |
7385 | else | |
7386 | desc = "recovery"; | |
7387 | ||
c4a39551 JB |
7388 | mddev->last_sync_action = action ?: desc; |
7389 | ||
1da177e4 LT |
7390 | /* we overload curr_resync somewhat here. |
7391 | * 0 == not engaged in resync at all | |
7392 | * 2 == checking that there is no conflict with another sync | |
7393 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7394 | * commense | |
7395 | * other == active in resync - this many blocks | |
7396 | * | |
7397 | * Before starting a resync we must have set curr_resync to | |
7398 | * 2, and then checked that every "conflicting" array has curr_resync | |
7399 | * less than ours. When we find one that is the same or higher | |
7400 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7401 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7402 | * This will mean we have to start checking from the beginning again. | |
7403 | * | |
7404 | */ | |
7405 | ||
7406 | do { | |
7407 | mddev->curr_resync = 2; | |
7408 | ||
7409 | try_again: | |
404e4b43 | 7410 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 7411 | goto skip; |
29ac4aa3 | 7412 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7413 | if (mddev2 == mddev) |
7414 | continue; | |
90b08710 BS |
7415 | if (!mddev->parallel_resync |
7416 | && mddev2->curr_resync | |
7417 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7418 | DEFINE_WAIT(wq); |
7419 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7420 | /* arbitrarily yield */ | |
7421 | mddev->curr_resync = 1; | |
7422 | wake_up(&resync_wait); | |
7423 | } | |
7424 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7425 | /* no need to wait here, we can wait the next | |
7426 | * time 'round when curr_resync == 2 | |
7427 | */ | |
7428 | continue; | |
9744197c N |
7429 | /* We need to wait 'interruptible' so as not to |
7430 | * contribute to the load average, and not to | |
7431 | * be caught by 'softlockup' | |
7432 | */ | |
7433 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 7434 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 7435 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
7436 | printk(KERN_INFO "md: delaying %s of %s" |
7437 | " until %s has finished (they" | |
1da177e4 | 7438 | " share one or more physical units)\n", |
61df9d91 | 7439 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 7440 | mddev_put(mddev2); |
9744197c N |
7441 | if (signal_pending(current)) |
7442 | flush_signals(current); | |
1da177e4 LT |
7443 | schedule(); |
7444 | finish_wait(&resync_wait, &wq); | |
7445 | goto try_again; | |
7446 | } | |
7447 | finish_wait(&resync_wait, &wq); | |
7448 | } | |
7449 | } | |
7450 | } while (mddev->curr_resync < 2); | |
7451 | ||
5fd6c1dc | 7452 | j = 0; |
9d88883e | 7453 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7454 | /* resync follows the size requested by the personality, |
57afd89f | 7455 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7456 | */ |
7457 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7458 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7459 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7460 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7461 | j = mddev->resync_min; | |
7462 | else if (!mddev->bitmap) | |
5fd6c1dc | 7463 | j = mddev->recovery_cp; |
5e96ee65 | 7464 | |
ccfcc3c1 | 7465 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7466 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7467 | else { |
1da177e4 | 7468 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7469 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7470 | j = MaxSector; |
4e59ca7d | 7471 | rcu_read_lock(); |
dafb20fa | 7472 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc N |
7473 | if (rdev->raid_disk >= 0 && |
7474 | !test_bit(Faulty, &rdev->flags) && | |
7475 | !test_bit(In_sync, &rdev->flags) && | |
7476 | rdev->recovery_offset < j) | |
7477 | j = rdev->recovery_offset; | |
4e59ca7d | 7478 | rcu_read_unlock(); |
5fd6c1dc | 7479 | } |
1da177e4 | 7480 | |
61df9d91 N |
7481 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
7482 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
7483 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 7484 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
7485 | "(but not more than %d KB/sec) for %s.\n", |
7486 | speed_max(mddev), desc); | |
1da177e4 | 7487 | |
eea1bf38 | 7488 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 7489 | |
57afd89f | 7490 | io_sectors = 0; |
1da177e4 LT |
7491 | for (m = 0; m < SYNC_MARKS; m++) { |
7492 | mark[m] = jiffies; | |
57afd89f | 7493 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
7494 | } |
7495 | last_mark = 0; | |
7496 | mddev->resync_mark = mark[last_mark]; | |
7497 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
7498 | ||
7499 | /* | |
7500 | * Tune reconstruction: | |
7501 | */ | |
7502 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
7503 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
7504 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
7505 | |
7506 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
7507 | last_check = 0; |
7508 | ||
7509 | if (j>2) { | |
c91abf5a | 7510 | printk(KERN_INFO |
61df9d91 N |
7511 | "md: resuming %s of %s from checkpoint.\n", |
7512 | desc, mdname(mddev)); | |
1da177e4 | 7513 | mddev->curr_resync = j; |
72f36d59 N |
7514 | } else |
7515 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 7516 | mddev->curr_resync_completed = j; |
72f36d59 N |
7517 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
7518 | md_new_event(mddev); | |
54f89341 | 7519 | update_time = jiffies; |
1da177e4 | 7520 | |
7c2c57c9 | 7521 | blk_start_plug(&plug); |
1da177e4 | 7522 | while (j < max_sectors) { |
57afd89f | 7523 | sector_t sectors; |
1da177e4 | 7524 | |
57afd89f | 7525 | skipped = 0; |
97e4f42d | 7526 | |
7a91ee1f N |
7527 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7528 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
7529 | (mddev->curr_resync - mddev->curr_resync_completed) | |
7530 | > (max_sectors >> 4)) || | |
54f89341 | 7531 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f N |
7532 | (j - mddev->curr_resync_completed)*2 |
7533 | >= mddev->resync_max - mddev->curr_resync_completed | |
7534 | )) { | |
97e4f42d | 7535 | /* time to update curr_resync_completed */ |
97e4f42d N |
7536 | wait_event(mddev->recovery_wait, |
7537 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 7538 | mddev->curr_resync_completed = j; |
35d78c66 | 7539 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
7540 | j > mddev->recovery_cp) | |
7541 | mddev->recovery_cp = j; | |
54f89341 | 7542 | update_time = jiffies; |
070dc6dd | 7543 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 7544 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 7545 | } |
acb180b0 | 7546 | |
c91abf5a N |
7547 | while (j >= mddev->resync_max && |
7548 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
7549 | /* As this condition is controlled by user-space, |
7550 | * we can block indefinitely, so use '_interruptible' | |
7551 | * to avoid triggering warnings. | |
7552 | */ | |
7553 | flush_signals(current); /* just in case */ | |
7554 | wait_event_interruptible(mddev->recovery_wait, | |
7555 | mddev->resync_max > j | |
c91abf5a N |
7556 | || test_bit(MD_RECOVERY_INTR, |
7557 | &mddev->recovery)); | |
e62e58a5 | 7558 | } |
acb180b0 | 7559 | |
c91abf5a N |
7560 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7561 | break; | |
acb180b0 | 7562 | |
57afd89f | 7563 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 7564 | currspeed < speed_min(mddev)); |
57afd89f | 7565 | if (sectors == 0) { |
dfc70645 | 7566 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 7567 | break; |
1da177e4 | 7568 | } |
57afd89f N |
7569 | |
7570 | if (!skipped) { /* actual IO requested */ | |
7571 | io_sectors += sectors; | |
7572 | atomic_add(sectors, &mddev->recovery_active); | |
7573 | } | |
7574 | ||
e875ecea N |
7575 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7576 | break; | |
7577 | ||
1da177e4 | 7578 | j += sectors; |
72f36d59 N |
7579 | if (j > 2) |
7580 | mddev->curr_resync = j; | |
ff4e8d9a | 7581 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 7582 | if (last_check == 0) |
e875ecea | 7583 | /* this is the earliest that rebuild will be |
d7603b7e N |
7584 | * visible in /proc/mdstat |
7585 | */ | |
7586 | md_new_event(mddev); | |
57afd89f N |
7587 | |
7588 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
7589 | continue; |
7590 | ||
57afd89f | 7591 | last_check = io_sectors; |
1da177e4 LT |
7592 | repeat: |
7593 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
7594 | /* step marks */ | |
7595 | int next = (last_mark+1) % SYNC_MARKS; | |
7596 | ||
7597 | mddev->resync_mark = mark[next]; | |
7598 | mddev->resync_mark_cnt = mark_cnt[next]; | |
7599 | mark[next] = jiffies; | |
57afd89f | 7600 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
7601 | last_mark = next; |
7602 | } | |
7603 | ||
c91abf5a N |
7604 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7605 | break; | |
1da177e4 LT |
7606 | |
7607 | /* | |
7608 | * this loop exits only if either when we are slower than | |
7609 | * the 'hard' speed limit, or the system was IO-idle for | |
7610 | * a jiffy. | |
7611 | * the system might be non-idle CPU-wise, but we only care | |
7612 | * about not overloading the IO subsystem. (things like an | |
7613 | * e2fsck being done on the RAID array should execute fast) | |
7614 | */ | |
1da177e4 LT |
7615 | cond_resched(); |
7616 | ||
57afd89f N |
7617 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
7618 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 7619 | |
88202a0c N |
7620 | if (currspeed > speed_min(mddev)) { |
7621 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 7622 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 7623 | msleep(500); |
1da177e4 LT |
7624 | goto repeat; |
7625 | } | |
7626 | } | |
7627 | } | |
c91abf5a N |
7628 | printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc, |
7629 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
7630 | ? "interrupted" : "done"); | |
1da177e4 LT |
7631 | /* |
7632 | * this also signals 'finished resyncing' to md_stop | |
7633 | */ | |
7c2c57c9 | 7634 | blk_finish_plug(&plug); |
1da177e4 LT |
7635 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7636 | ||
7637 | /* tell personality that we are finished */ | |
57afd89f | 7638 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 7639 | |
dfc70645 | 7640 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7641 | mddev->curr_resync > 2) { |
7642 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7643 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7644 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7645 | printk(KERN_INFO | |
61df9d91 N |
7646 | "md: checkpointing %s of %s.\n", |
7647 | desc, mdname(mddev)); | |
0a19caab | 7648 | if (test_bit(MD_RECOVERY_ERROR, |
7649 | &mddev->recovery)) | |
7650 | mddev->recovery_cp = | |
7651 | mddev->curr_resync_completed; | |
7652 | else | |
7653 | mddev->recovery_cp = | |
7654 | mddev->curr_resync; | |
5fd6c1dc N |
7655 | } |
7656 | } else | |
7657 | mddev->recovery_cp = MaxSector; | |
7658 | } else { | |
7659 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7660 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 7661 | rcu_read_lock(); |
dafb20fa | 7662 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7663 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7664 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7665 | !test_bit(Faulty, &rdev->flags) && |
7666 | !test_bit(In_sync, &rdev->flags) && | |
7667 | rdev->recovery_offset < mddev->curr_resync) | |
7668 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7669 | rcu_read_unlock(); |
5fd6c1dc | 7670 | } |
1da177e4 | 7671 | } |
db91ff55 | 7672 | skip: |
17571284 | 7673 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7674 | |
c07b70ad N |
7675 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7676 | /* We completed so min/max setting can be forgotten if used. */ | |
7677 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7678 | mddev->resync_min = 0; | |
7679 | mddev->resync_max = MaxSector; | |
7680 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7681 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 LT |
7682 | mddev->curr_resync = 0; |
7683 | wake_up(&resync_wait); | |
7684 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
7685 | md_wakeup_thread(mddev->thread); | |
c6207277 | 7686 | return; |
1da177e4 | 7687 | } |
29269553 | 7688 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 7689 | |
746d3207 N |
7690 | static int remove_and_add_spares(struct mddev *mddev, |
7691 | struct md_rdev *this) | |
b4c4c7b8 | 7692 | { |
3cb03002 | 7693 | struct md_rdev *rdev; |
b4c4c7b8 | 7694 | int spares = 0; |
f2a371c5 | 7695 | int removed = 0; |
b4c4c7b8 | 7696 | |
dafb20fa | 7697 | rdev_for_each(rdev, mddev) |
746d3207 N |
7698 | if ((this == NULL || rdev == this) && |
7699 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 7700 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7701 | (test_bit(Faulty, &rdev->flags) || |
7702 | ! test_bit(In_sync, &rdev->flags)) && | |
7703 | atomic_read(&rdev->nr_pending)==0) { | |
7704 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 7705 | mddev, rdev) == 0) { |
36fad858 | 7706 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 7707 | rdev->raid_disk = -1; |
f2a371c5 | 7708 | removed++; |
b4c4c7b8 N |
7709 | } |
7710 | } | |
90584fc9 JB |
7711 | if (removed && mddev->kobj.sd) |
7712 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 7713 | |
746d3207 N |
7714 | if (this) |
7715 | goto no_add; | |
7716 | ||
dafb20fa | 7717 | rdev_for_each(rdev, mddev) { |
7bfec5f3 N |
7718 | if (rdev->raid_disk >= 0 && |
7719 | !test_bit(In_sync, &rdev->flags) && | |
7720 | !test_bit(Faulty, &rdev->flags)) | |
7721 | spares++; | |
7ceb17e8 N |
7722 | if (rdev->raid_disk >= 0) |
7723 | continue; | |
7724 | if (test_bit(Faulty, &rdev->flags)) | |
7725 | continue; | |
7726 | if (mddev->ro && | |
8313b8e5 N |
7727 | ! (rdev->saved_raid_disk >= 0 && |
7728 | !test_bit(Bitmap_sync, &rdev->flags))) | |
7ceb17e8 N |
7729 | continue; |
7730 | ||
7731 | rdev->recovery_offset = 0; | |
7ceb17e8 N |
7732 | if (mddev->pers-> |
7733 | hot_add_disk(mddev, rdev) == 0) { | |
7734 | if (sysfs_link_rdev(mddev, rdev)) | |
7735 | /* failure here is OK */; | |
7736 | spares++; | |
7737 | md_new_event(mddev); | |
7738 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
dfc70645 | 7739 | } |
b4c4c7b8 | 7740 | } |
746d3207 | 7741 | no_add: |
6dafab6b N |
7742 | if (removed) |
7743 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 N |
7744 | return spares; |
7745 | } | |
7ebc0be7 | 7746 | |
1da177e4 LT |
7747 | /* |
7748 | * This routine is regularly called by all per-raid-array threads to | |
7749 | * deal with generic issues like resync and super-block update. | |
7750 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7751 | * need this as they never do any recovery or update the superblock. | |
7752 | * | |
7753 | * It does not do any resync itself, but rather "forks" off other threads | |
7754 | * to do that as needed. | |
7755 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7756 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7757 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7758 | * and wakeups up this thread which will reap the thread and finish up. |
7759 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7760 | * | |
7761 | * The overall approach is: | |
7762 | * 1/ if the superblock needs updating, update it. | |
7763 | * 2/ If a recovery thread is running, don't do anything else. | |
7764 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7765 | * 4/ If there are any faulty devices, remove them. | |
7766 | * 5/ If array is degraded, try to add spares devices | |
7767 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7768 | */ | |
fd01b88c | 7769 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 7770 | { |
68866e42 JB |
7771 | if (mddev->suspended) |
7772 | return; | |
7773 | ||
5f40402d | 7774 | if (mddev->bitmap) |
aa5cbd10 | 7775 | bitmap_daemon_work(mddev); |
1da177e4 | 7776 | |
fca4d848 | 7777 | if (signal_pending(current)) { |
31a59e34 | 7778 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7779 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7780 | mdname(mddev)); | |
7781 | mddev->safemode = 2; | |
7782 | } | |
7783 | flush_signals(current); | |
7784 | } | |
7785 | ||
c89a8eee N |
7786 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7787 | return; | |
1da177e4 | 7788 | if ( ! ( |
142d44c3 | 7789 | (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7790 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7791 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7792 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7793 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7794 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7795 | )) |
7796 | return; | |
fca4d848 | 7797 | |
df5b89b3 | 7798 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7799 | int spares = 0; |
fca4d848 | 7800 | |
c89a8eee | 7801 | if (mddev->ro) { |
7ceb17e8 N |
7802 | /* On a read-only array we can: |
7803 | * - remove failed devices | |
7804 | * - add already-in_sync devices if the array itself | |
7805 | * is in-sync. | |
7806 | * As we only add devices that are already in-sync, | |
7807 | * we can activate the spares immediately. | |
c89a8eee | 7808 | */ |
7ceb17e8 | 7809 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
7810 | /* There is no thread, but we need to call |
7811 | * ->spare_active and clear saved_raid_disk | |
7812 | */ | |
7813 | md_reap_sync_thread(mddev); | |
7814 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
c89a8eee N |
7815 | goto unlock; |
7816 | } | |
7817 | ||
31a59e34 | 7818 | if (!mddev->external) { |
0fd62b86 | 7819 | int did_change = 0; |
31a59e34 N |
7820 | spin_lock_irq(&mddev->write_lock); |
7821 | if (mddev->safemode && | |
7822 | !atomic_read(&mddev->writes_pending) && | |
7823 | !mddev->in_sync && | |
7824 | mddev->recovery_cp == MaxSector) { | |
7825 | mddev->in_sync = 1; | |
0fd62b86 | 7826 | did_change = 1; |
070dc6dd | 7827 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7828 | } |
7829 | if (mddev->safemode == 1) | |
7830 | mddev->safemode = 0; | |
7831 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7832 | if (did_change) |
00bcb4ac | 7833 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7834 | } |
fca4d848 | 7835 | |
7a0a5355 | 7836 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 7837 | md_update_sb(mddev, 0); |
06d91a5f | 7838 | |
1da177e4 LT |
7839 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7840 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7841 | /* resync/recovery still happening */ | |
7842 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7843 | goto unlock; | |
7844 | } | |
7845 | if (mddev->sync_thread) { | |
a91d5ac0 | 7846 | md_reap_sync_thread(mddev); |
1da177e4 LT |
7847 | goto unlock; |
7848 | } | |
72a23c21 NB |
7849 | /* Set RUNNING before clearing NEEDED to avoid |
7850 | * any transients in the value of "sync_action". | |
7851 | */ | |
72f36d59 | 7852 | mddev->curr_resync_completed = 0; |
72a23c21 | 7853 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
24dd469d N |
7854 | /* Clear some bits that don't mean anything, but |
7855 | * might be left set | |
7856 | */ | |
24dd469d N |
7857 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7858 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7859 | |
ed209584 N |
7860 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
7861 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
5fd6c1dc | 7862 | goto unlock; |
1da177e4 LT |
7863 | /* no recovery is running. |
7864 | * remove any failed drives, then | |
7865 | * add spares if possible. | |
72f36d59 | 7866 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
7867 | * the personality to fail the re-add. |
7868 | */ | |
1da177e4 | 7869 | |
b4c4c7b8 | 7870 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7871 | if (mddev->pers->check_reshape == NULL || |
7872 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7873 | /* Cannot proceed */ |
7874 | goto unlock; | |
7875 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7876 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 7877 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
7878 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7879 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7880 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7881 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7882 | } else if (mddev->recovery_cp < MaxSector) { |
7883 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7884 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7885 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7886 | /* nothing to be done ... */ | |
1da177e4 | 7887 | goto unlock; |
24dd469d | 7888 | |
1da177e4 | 7889 | if (mddev->pers->sync_request) { |
ef99bf48 | 7890 | if (spares) { |
a654b9d8 N |
7891 | /* We are adding a device or devices to an array |
7892 | * which has the bitmap stored on all devices. | |
7893 | * So make sure all bitmap pages get written | |
7894 | */ | |
7895 | bitmap_write_all(mddev->bitmap); | |
7896 | } | |
1da177e4 LT |
7897 | mddev->sync_thread = md_register_thread(md_do_sync, |
7898 | mddev, | |
0da3c619 | 7899 | "resync"); |
1da177e4 LT |
7900 | if (!mddev->sync_thread) { |
7901 | printk(KERN_ERR "%s: could not start resync" | |
7902 | " thread...\n", | |
7903 | mdname(mddev)); | |
7904 | /* leave the spares where they are, it shouldn't hurt */ | |
7ebc0be7 N |
7905 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
7906 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7907 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7908 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7909 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
d7603b7e | 7910 | } else |
1da177e4 | 7911 | md_wakeup_thread(mddev->sync_thread); |
00bcb4ac | 7912 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7913 | md_new_event(mddev); |
1da177e4 LT |
7914 | } |
7915 | unlock: | |
90f5f7ad HR |
7916 | wake_up(&mddev->sb_wait); |
7917 | ||
72a23c21 NB |
7918 | if (!mddev->sync_thread) { |
7919 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7920 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7921 | &mddev->recovery)) | |
0c3573f1 | 7922 | if (mddev->sysfs_action) |
00bcb4ac | 7923 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7924 | } |
1da177e4 LT |
7925 | mddev_unlock(mddev); |
7926 | } | |
7927 | } | |
7928 | ||
a91d5ac0 JB |
7929 | void md_reap_sync_thread(struct mddev *mddev) |
7930 | { | |
7931 | struct md_rdev *rdev; | |
7932 | ||
7933 | /* resync has finished, collect result */ | |
7934 | md_unregister_thread(&mddev->sync_thread); | |
30b8feb7 | 7935 | wake_up(&resync_wait); |
a91d5ac0 JB |
7936 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
7937 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
7938 | /* success...*/ | |
7939 | /* activate any spares */ | |
7940 | if (mddev->pers->spare_active(mddev)) { | |
7941 | sysfs_notify(&mddev->kobj, NULL, | |
7942 | "degraded"); | |
7943 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
7944 | } | |
7945 | } | |
7946 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
7947 | mddev->pers->finish_reshape) | |
7948 | mddev->pers->finish_reshape(mddev); | |
7949 | ||
7950 | /* If array is no-longer degraded, then any saved_raid_disk | |
7951 | * information must be scrapped. Also if any device is now | |
7952 | * In_sync we must scrape the saved_raid_disk for that device | |
7953 | * do the superblock for an incrementally recovered device | |
7954 | * written out. | |
7955 | */ | |
7956 | rdev_for_each(rdev, mddev) | |
7957 | if (!mddev->degraded || | |
7958 | test_bit(In_sync, &rdev->flags)) | |
7959 | rdev->saved_raid_disk = -1; | |
7960 | ||
7961 | md_update_sb(mddev, 1); | |
7962 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7963 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7964 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7965 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7966 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7967 | /* flag recovery needed just to double check */ | |
7968 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7969 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7970 | md_new_event(mddev); | |
7971 | if (mddev->event_work.func) | |
7972 | queue_work(md_misc_wq, &mddev->event_work); | |
7973 | } | |
7974 | ||
fd01b88c | 7975 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 7976 | { |
00bcb4ac | 7977 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 7978 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
7979 | !test_bit(Blocked, &rdev->flags) && |
7980 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
7981 | msecs_to_jiffies(5000)); |
7982 | rdev_dec_pending(rdev, mddev); | |
7983 | } | |
7984 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7985 | ||
c6563a8c N |
7986 | void md_finish_reshape(struct mddev *mddev) |
7987 | { | |
7988 | /* called be personality module when reshape completes. */ | |
7989 | struct md_rdev *rdev; | |
7990 | ||
7991 | rdev_for_each(rdev, mddev) { | |
7992 | if (rdev->data_offset > rdev->new_data_offset) | |
7993 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
7994 | else | |
7995 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
7996 | rdev->data_offset = rdev->new_data_offset; | |
7997 | } | |
7998 | } | |
7999 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 N |
8000 | |
8001 | /* Bad block management. | |
8002 | * We can record which blocks on each device are 'bad' and so just | |
8003 | * fail those blocks, or that stripe, rather than the whole device. | |
8004 | * Entries in the bad-block table are 64bits wide. This comprises: | |
8005 | * Length of bad-range, in sectors: 0-511 for lengths 1-512 | |
8006 | * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) | |
8007 | * A 'shift' can be set so that larger blocks are tracked and | |
8008 | * consequently larger devices can be covered. | |
8009 | * 'Acknowledged' flag - 1 bit. - the most significant bit. | |
8010 | * | |
8011 | * Locking of the bad-block table uses a seqlock so md_is_badblock | |
8012 | * might need to retry if it is very unlucky. | |
8013 | * We will sometimes want to check for bad blocks in a bi_end_io function, | |
8014 | * so we use the write_seqlock_irq variant. | |
8015 | * | |
8016 | * When looking for a bad block we specify a range and want to | |
8017 | * know if any block in the range is bad. So we binary-search | |
8018 | * to the last range that starts at-or-before the given endpoint, | |
8019 | * (or "before the sector after the target range") | |
8020 | * then see if it ends after the given start. | |
8021 | * We return | |
8022 | * 0 if there are no known bad blocks in the range | |
8023 | * 1 if there are known bad block which are all acknowledged | |
8024 | * -1 if there are bad blocks which have not yet been acknowledged in metadata. | |
8025 | * plus the start/length of the first bad section we overlap. | |
8026 | */ | |
8027 | int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, | |
8028 | sector_t *first_bad, int *bad_sectors) | |
8029 | { | |
8030 | int hi; | |
ab05613a | 8031 | int lo; |
2230dfe4 | 8032 | u64 *p = bb->page; |
ab05613a | 8033 | int rv; |
2230dfe4 N |
8034 | sector_t target = s + sectors; |
8035 | unsigned seq; | |
8036 | ||
8037 | if (bb->shift > 0) { | |
8038 | /* round the start down, and the end up */ | |
8039 | s >>= bb->shift; | |
8040 | target += (1<<bb->shift) - 1; | |
8041 | target >>= bb->shift; | |
8042 | sectors = target - s; | |
8043 | } | |
8044 | /* 'target' is now the first block after the bad range */ | |
8045 | ||
8046 | retry: | |
8047 | seq = read_seqbegin(&bb->lock); | |
ab05613a | 8048 | lo = 0; |
8049 | rv = 0; | |
2230dfe4 N |
8050 | hi = bb->count; |
8051 | ||
8052 | /* Binary search between lo and hi for 'target' | |
8053 | * i.e. for the last range that starts before 'target' | |
8054 | */ | |
8055 | /* INVARIANT: ranges before 'lo' and at-or-after 'hi' | |
8056 | * are known not to be the last range before target. | |
8057 | * VARIANT: hi-lo is the number of possible | |
8058 | * ranges, and decreases until it reaches 1 | |
8059 | */ | |
8060 | while (hi - lo > 1) { | |
8061 | int mid = (lo + hi) / 2; | |
8062 | sector_t a = BB_OFFSET(p[mid]); | |
8063 | if (a < target) | |
8064 | /* This could still be the one, earlier ranges | |
8065 | * could not. */ | |
8066 | lo = mid; | |
8067 | else | |
8068 | /* This and later ranges are definitely out. */ | |
8069 | hi = mid; | |
8070 | } | |
8071 | /* 'lo' might be the last that started before target, but 'hi' isn't */ | |
8072 | if (hi > lo) { | |
8073 | /* need to check all range that end after 's' to see if | |
8074 | * any are unacknowledged. | |
8075 | */ | |
8076 | while (lo >= 0 && | |
8077 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8078 | if (BB_OFFSET(p[lo]) < target) { | |
8079 | /* starts before the end, and finishes after | |
8080 | * the start, so they must overlap | |
8081 | */ | |
8082 | if (rv != -1 && BB_ACK(p[lo])) | |
8083 | rv = 1; | |
8084 | else | |
8085 | rv = -1; | |
8086 | *first_bad = BB_OFFSET(p[lo]); | |
8087 | *bad_sectors = BB_LEN(p[lo]); | |
8088 | } | |
8089 | lo--; | |
8090 | } | |
8091 | } | |
8092 | ||
8093 | if (read_seqretry(&bb->lock, seq)) | |
8094 | goto retry; | |
8095 | ||
8096 | return rv; | |
8097 | } | |
8098 | EXPORT_SYMBOL_GPL(md_is_badblock); | |
8099 | ||
8100 | /* | |
8101 | * Add a range of bad blocks to the table. | |
8102 | * This might extend the table, or might contract it | |
8103 | * if two adjacent ranges can be merged. | |
8104 | * We binary-search to find the 'insertion' point, then | |
8105 | * decide how best to handle it. | |
8106 | */ | |
8107 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, | |
8108 | int acknowledged) | |
8109 | { | |
8110 | u64 *p; | |
8111 | int lo, hi; | |
8112 | int rv = 1; | |
905b0297 | 8113 | unsigned long flags; |
2230dfe4 N |
8114 | |
8115 | if (bb->shift < 0) | |
8116 | /* badblocks are disabled */ | |
8117 | return 0; | |
8118 | ||
8119 | if (bb->shift) { | |
8120 | /* round the start down, and the end up */ | |
8121 | sector_t next = s + sectors; | |
8122 | s >>= bb->shift; | |
8123 | next += (1<<bb->shift) - 1; | |
8124 | next >>= bb->shift; | |
8125 | sectors = next - s; | |
8126 | } | |
8127 | ||
905b0297 | 8128 | write_seqlock_irqsave(&bb->lock, flags); |
2230dfe4 N |
8129 | |
8130 | p = bb->page; | |
8131 | lo = 0; | |
8132 | hi = bb->count; | |
8133 | /* Find the last range that starts at-or-before 's' */ | |
8134 | while (hi - lo > 1) { | |
8135 | int mid = (lo + hi) / 2; | |
8136 | sector_t a = BB_OFFSET(p[mid]); | |
8137 | if (a <= s) | |
8138 | lo = mid; | |
8139 | else | |
8140 | hi = mid; | |
8141 | } | |
8142 | if (hi > lo && BB_OFFSET(p[lo]) > s) | |
8143 | hi = lo; | |
8144 | ||
8145 | if (hi > lo) { | |
8146 | /* we found a range that might merge with the start | |
8147 | * of our new range | |
8148 | */ | |
8149 | sector_t a = BB_OFFSET(p[lo]); | |
8150 | sector_t e = a + BB_LEN(p[lo]); | |
8151 | int ack = BB_ACK(p[lo]); | |
8152 | if (e >= s) { | |
8153 | /* Yes, we can merge with a previous range */ | |
8154 | if (s == a && s + sectors >= e) | |
8155 | /* new range covers old */ | |
8156 | ack = acknowledged; | |
8157 | else | |
8158 | ack = ack && acknowledged; | |
8159 | ||
8160 | if (e < s + sectors) | |
8161 | e = s + sectors; | |
8162 | if (e - a <= BB_MAX_LEN) { | |
8163 | p[lo] = BB_MAKE(a, e-a, ack); | |
8164 | s = e; | |
8165 | } else { | |
8166 | /* does not all fit in one range, | |
8167 | * make p[lo] maximal | |
8168 | */ | |
8169 | if (BB_LEN(p[lo]) != BB_MAX_LEN) | |
8170 | p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8171 | s = a + BB_MAX_LEN; | |
8172 | } | |
8173 | sectors = e - s; | |
8174 | } | |
8175 | } | |
8176 | if (sectors && hi < bb->count) { | |
8177 | /* 'hi' points to the first range that starts after 's'. | |
8178 | * Maybe we can merge with the start of that range */ | |
8179 | sector_t a = BB_OFFSET(p[hi]); | |
8180 | sector_t e = a + BB_LEN(p[hi]); | |
8181 | int ack = BB_ACK(p[hi]); | |
8182 | if (a <= s + sectors) { | |
8183 | /* merging is possible */ | |
8184 | if (e <= s + sectors) { | |
8185 | /* full overlap */ | |
8186 | e = s + sectors; | |
8187 | ack = acknowledged; | |
8188 | } else | |
8189 | ack = ack && acknowledged; | |
8190 | ||
8191 | a = s; | |
8192 | if (e - a <= BB_MAX_LEN) { | |
8193 | p[hi] = BB_MAKE(a, e-a, ack); | |
8194 | s = e; | |
8195 | } else { | |
8196 | p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8197 | s = a + BB_MAX_LEN; | |
8198 | } | |
8199 | sectors = e - s; | |
8200 | lo = hi; | |
8201 | hi++; | |
8202 | } | |
8203 | } | |
8204 | if (sectors == 0 && hi < bb->count) { | |
8205 | /* we might be able to combine lo and hi */ | |
8206 | /* Note: 's' is at the end of 'lo' */ | |
8207 | sector_t a = BB_OFFSET(p[hi]); | |
8208 | int lolen = BB_LEN(p[lo]); | |
8209 | int hilen = BB_LEN(p[hi]); | |
8210 | int newlen = lolen + hilen - (s - a); | |
8211 | if (s >= a && newlen < BB_MAX_LEN) { | |
8212 | /* yes, we can combine them */ | |
8213 | int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); | |
8214 | p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); | |
8215 | memmove(p + hi, p + hi + 1, | |
8216 | (bb->count - hi - 1) * 8); | |
8217 | bb->count--; | |
8218 | } | |
8219 | } | |
8220 | while (sectors) { | |
8221 | /* didn't merge (it all). | |
8222 | * Need to add a range just before 'hi' */ | |
8223 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8224 | /* No room for more */ | |
8225 | rv = 0; | |
8226 | break; | |
8227 | } else { | |
8228 | int this_sectors = sectors; | |
8229 | memmove(p + hi + 1, p + hi, | |
8230 | (bb->count - hi) * 8); | |
8231 | bb->count++; | |
8232 | ||
8233 | if (this_sectors > BB_MAX_LEN) | |
8234 | this_sectors = BB_MAX_LEN; | |
8235 | p[hi] = BB_MAKE(s, this_sectors, acknowledged); | |
8236 | sectors -= this_sectors; | |
8237 | s += this_sectors; | |
8238 | } | |
8239 | } | |
8240 | ||
8241 | bb->changed = 1; | |
de393cde N |
8242 | if (!acknowledged) |
8243 | bb->unacked_exist = 1; | |
905b0297 | 8244 | write_sequnlock_irqrestore(&bb->lock, flags); |
2230dfe4 N |
8245 | |
8246 | return rv; | |
8247 | } | |
8248 | ||
3cb03002 | 8249 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8250 | int is_new) |
2230dfe4 | 8251 | { |
c6563a8c N |
8252 | int rv; |
8253 | if (is_new) | |
8254 | s += rdev->new_data_offset; | |
8255 | else | |
8256 | s += rdev->data_offset; | |
8257 | rv = md_set_badblocks(&rdev->badblocks, | |
8258 | s, sectors, 0); | |
2230dfe4 N |
8259 | if (rv) { |
8260 | /* Make sure they get written out promptly */ | |
8bd2f0a0 | 8261 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2230dfe4 N |
8262 | set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); |
8263 | md_wakeup_thread(rdev->mddev->thread); | |
8264 | } | |
8265 | return rv; | |
8266 | } | |
8267 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8268 | ||
8269 | /* | |
8270 | * Remove a range of bad blocks from the table. | |
8271 | * This may involve extending the table if we spilt a region, | |
8272 | * but it must not fail. So if the table becomes full, we just | |
8273 | * drop the remove request. | |
8274 | */ | |
8275 | static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) | |
8276 | { | |
8277 | u64 *p; | |
8278 | int lo, hi; | |
8279 | sector_t target = s + sectors; | |
8280 | int rv = 0; | |
8281 | ||
8282 | if (bb->shift > 0) { | |
8283 | /* When clearing we round the start up and the end down. | |
8284 | * This should not matter as the shift should align with | |
8285 | * the block size and no rounding should ever be needed. | |
8286 | * However it is better the think a block is bad when it | |
8287 | * isn't than to think a block is not bad when it is. | |
8288 | */ | |
8289 | s += (1<<bb->shift) - 1; | |
8290 | s >>= bb->shift; | |
8291 | target >>= bb->shift; | |
8292 | sectors = target - s; | |
8293 | } | |
8294 | ||
8295 | write_seqlock_irq(&bb->lock); | |
8296 | ||
8297 | p = bb->page; | |
8298 | lo = 0; | |
8299 | hi = bb->count; | |
8300 | /* Find the last range that starts before 'target' */ | |
8301 | while (hi - lo > 1) { | |
8302 | int mid = (lo + hi) / 2; | |
8303 | sector_t a = BB_OFFSET(p[mid]); | |
8304 | if (a < target) | |
8305 | lo = mid; | |
8306 | else | |
8307 | hi = mid; | |
8308 | } | |
8309 | if (hi > lo) { | |
8310 | /* p[lo] is the last range that could overlap the | |
8311 | * current range. Earlier ranges could also overlap, | |
8312 | * but only this one can overlap the end of the range. | |
8313 | */ | |
8314 | if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { | |
8315 | /* Partial overlap, leave the tail of this range */ | |
8316 | int ack = BB_ACK(p[lo]); | |
8317 | sector_t a = BB_OFFSET(p[lo]); | |
8318 | sector_t end = a + BB_LEN(p[lo]); | |
8319 | ||
8320 | if (a < s) { | |
8321 | /* we need to split this range */ | |
8322 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8323 | rv = 0; | |
8324 | goto out; | |
8325 | } | |
8326 | memmove(p+lo+1, p+lo, (bb->count - lo) * 8); | |
8327 | bb->count++; | |
8328 | p[lo] = BB_MAKE(a, s-a, ack); | |
8329 | lo++; | |
8330 | } | |
8331 | p[lo] = BB_MAKE(target, end - target, ack); | |
8332 | /* there is no longer an overlap */ | |
8333 | hi = lo; | |
8334 | lo--; | |
8335 | } | |
8336 | while (lo >= 0 && | |
8337 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8338 | /* This range does overlap */ | |
8339 | if (BB_OFFSET(p[lo]) < s) { | |
8340 | /* Keep the early parts of this range. */ | |
8341 | int ack = BB_ACK(p[lo]); | |
8342 | sector_t start = BB_OFFSET(p[lo]); | |
8343 | p[lo] = BB_MAKE(start, s - start, ack); | |
8344 | /* now low doesn't overlap, so.. */ | |
8345 | break; | |
8346 | } | |
8347 | lo--; | |
8348 | } | |
8349 | /* 'lo' is strictly before, 'hi' is strictly after, | |
8350 | * anything between needs to be discarded | |
8351 | */ | |
8352 | if (hi - lo > 1) { | |
8353 | memmove(p+lo+1, p+hi, (bb->count - hi) * 8); | |
8354 | bb->count -= (hi - lo - 1); | |
8355 | } | |
8356 | } | |
8357 | ||
8358 | bb->changed = 1; | |
8359 | out: | |
8360 | write_sequnlock_irq(&bb->lock); | |
8361 | return rv; | |
8362 | } | |
8363 | ||
c6563a8c N |
8364 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8365 | int is_new) | |
2230dfe4 | 8366 | { |
c6563a8c N |
8367 | if (is_new) |
8368 | s += rdev->new_data_offset; | |
8369 | else | |
8370 | s += rdev->data_offset; | |
2230dfe4 | 8371 | return md_clear_badblocks(&rdev->badblocks, |
c6563a8c | 8372 | s, sectors); |
2230dfe4 N |
8373 | } |
8374 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8375 | ||
8376 | /* | |
8377 | * Acknowledge all bad blocks in a list. | |
8378 | * This only succeeds if ->changed is clear. It is used by | |
8379 | * in-kernel metadata updates | |
8380 | */ | |
8381 | void md_ack_all_badblocks(struct badblocks *bb) | |
8382 | { | |
8383 | if (bb->page == NULL || bb->changed) | |
8384 | /* no point even trying */ | |
8385 | return; | |
8386 | write_seqlock_irq(&bb->lock); | |
8387 | ||
ecb178bb | 8388 | if (bb->changed == 0 && bb->unacked_exist) { |
2230dfe4 N |
8389 | u64 *p = bb->page; |
8390 | int i; | |
8391 | for (i = 0; i < bb->count ; i++) { | |
8392 | if (!BB_ACK(p[i])) { | |
8393 | sector_t start = BB_OFFSET(p[i]); | |
8394 | int len = BB_LEN(p[i]); | |
8395 | p[i] = BB_MAKE(start, len, 1); | |
8396 | } | |
8397 | } | |
de393cde | 8398 | bb->unacked_exist = 0; |
2230dfe4 N |
8399 | } |
8400 | write_sequnlock_irq(&bb->lock); | |
8401 | } | |
8402 | EXPORT_SYMBOL_GPL(md_ack_all_badblocks); | |
8403 | ||
16c791a5 N |
8404 | /* sysfs access to bad-blocks list. |
8405 | * We present two files. | |
8406 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
8407 | * are recorded as bad. The list is truncated to fit within | |
8408 | * the one-page limit of sysfs. | |
8409 | * Writing "sector length" to this file adds an acknowledged | |
8410 | * bad block list. | |
8411 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
8412 | * been acknowledged. Writing to this file adds bad blocks | |
8413 | * without acknowledging them. This is largely for testing. | |
8414 | */ | |
8415 | ||
8416 | static ssize_t | |
8417 | badblocks_show(struct badblocks *bb, char *page, int unack) | |
8418 | { | |
8419 | size_t len; | |
8420 | int i; | |
8421 | u64 *p = bb->page; | |
8422 | unsigned seq; | |
8423 | ||
8424 | if (bb->shift < 0) | |
8425 | return 0; | |
8426 | ||
8427 | retry: | |
8428 | seq = read_seqbegin(&bb->lock); | |
8429 | ||
8430 | len = 0; | |
8431 | i = 0; | |
8432 | ||
8433 | while (len < PAGE_SIZE && i < bb->count) { | |
8434 | sector_t s = BB_OFFSET(p[i]); | |
8435 | unsigned int length = BB_LEN(p[i]); | |
8436 | int ack = BB_ACK(p[i]); | |
8437 | i++; | |
8438 | ||
8439 | if (unack && ack) | |
8440 | continue; | |
8441 | ||
8442 | len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", | |
8443 | (unsigned long long)s << bb->shift, | |
8444 | length << bb->shift); | |
8445 | } | |
de393cde N |
8446 | if (unack && len == 0) |
8447 | bb->unacked_exist = 0; | |
16c791a5 N |
8448 | |
8449 | if (read_seqretry(&bb->lock, seq)) | |
8450 | goto retry; | |
8451 | ||
8452 | return len; | |
8453 | } | |
8454 | ||
8455 | #define DO_DEBUG 1 | |
8456 | ||
8457 | static ssize_t | |
8458 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) | |
8459 | { | |
8460 | unsigned long long sector; | |
8461 | int length; | |
8462 | char newline; | |
8463 | #ifdef DO_DEBUG | |
8464 | /* Allow clearing via sysfs *only* for testing/debugging. | |
8465 | * Normally only a successful write may clear a badblock | |
8466 | */ | |
8467 | int clear = 0; | |
8468 | if (page[0] == '-') { | |
8469 | clear = 1; | |
8470 | page++; | |
8471 | } | |
8472 | #endif /* DO_DEBUG */ | |
8473 | ||
8474 | switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { | |
8475 | case 3: | |
8476 | if (newline != '\n') | |
8477 | return -EINVAL; | |
8478 | case 2: | |
8479 | if (length <= 0) | |
8480 | return -EINVAL; | |
8481 | break; | |
8482 | default: | |
8483 | return -EINVAL; | |
8484 | } | |
8485 | ||
8486 | #ifdef DO_DEBUG | |
8487 | if (clear) { | |
8488 | md_clear_badblocks(bb, sector, length); | |
8489 | return len; | |
8490 | } | |
8491 | #endif /* DO_DEBUG */ | |
8492 | if (md_set_badblocks(bb, sector, length, !unack)) | |
8493 | return len; | |
8494 | else | |
8495 | return -ENOSPC; | |
8496 | } | |
8497 | ||
75c96f85 AB |
8498 | static int md_notify_reboot(struct notifier_block *this, |
8499 | unsigned long code, void *x) | |
1da177e4 LT |
8500 | { |
8501 | struct list_head *tmp; | |
fd01b88c | 8502 | struct mddev *mddev; |
2dba6a91 | 8503 | int need_delay = 0; |
1da177e4 | 8504 | |
c744a65c N |
8505 | for_each_mddev(mddev, tmp) { |
8506 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8507 | if (mddev->pers) |
8508 | __md_stop_writes(mddev); | |
c744a65c N |
8509 | mddev->safemode = 2; |
8510 | mddev_unlock(mddev); | |
2dba6a91 | 8511 | } |
c744a65c | 8512 | need_delay = 1; |
1da177e4 | 8513 | } |
c744a65c N |
8514 | /* |
8515 | * certain more exotic SCSI devices are known to be | |
8516 | * volatile wrt too early system reboots. While the | |
8517 | * right place to handle this issue is the given | |
8518 | * driver, we do want to have a safe RAID driver ... | |
8519 | */ | |
8520 | if (need_delay) | |
8521 | mdelay(1000*1); | |
8522 | ||
1da177e4 LT |
8523 | return NOTIFY_DONE; |
8524 | } | |
8525 | ||
75c96f85 | 8526 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8527 | .notifier_call = md_notify_reboot, |
8528 | .next = NULL, | |
8529 | .priority = INT_MAX, /* before any real devices */ | |
8530 | }; | |
8531 | ||
8532 | static void md_geninit(void) | |
8533 | { | |
36a4e1fe | 8534 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8535 | |
c7705f34 | 8536 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8537 | } |
8538 | ||
75c96f85 | 8539 | static int __init md_init(void) |
1da177e4 | 8540 | { |
e804ac78 TH |
8541 | int ret = -ENOMEM; |
8542 | ||
ada609ee | 8543 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8544 | if (!md_wq) |
8545 | goto err_wq; | |
8546 | ||
8547 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8548 | if (!md_misc_wq) | |
8549 | goto err_misc_wq; | |
8550 | ||
8551 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8552 | goto err_md; | |
8553 | ||
8554 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8555 | goto err_mdp; | |
8556 | mdp_major = ret; | |
8557 | ||
3dbd8c2e | 8558 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
8559 | md_probe, NULL, NULL); |
8560 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8561 | md_probe, NULL, NULL); |
8562 | ||
1da177e4 | 8563 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8564 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8565 | |
8566 | md_geninit(); | |
d710e138 | 8567 | return 0; |
1da177e4 | 8568 | |
e804ac78 TH |
8569 | err_mdp: |
8570 | unregister_blkdev(MD_MAJOR, "md"); | |
8571 | err_md: | |
8572 | destroy_workqueue(md_misc_wq); | |
8573 | err_misc_wq: | |
8574 | destroy_workqueue(md_wq); | |
8575 | err_wq: | |
8576 | return ret; | |
8577 | } | |
1da177e4 LT |
8578 | |
8579 | #ifndef MODULE | |
8580 | ||
8581 | /* | |
8582 | * Searches all registered partitions for autorun RAID arrays | |
8583 | * at boot time. | |
8584 | */ | |
4d936ec1 ME |
8585 | |
8586 | static LIST_HEAD(all_detected_devices); | |
8587 | struct detected_devices_node { | |
8588 | struct list_head list; | |
8589 | dev_t dev; | |
8590 | }; | |
1da177e4 LT |
8591 | |
8592 | void md_autodetect_dev(dev_t dev) | |
8593 | { | |
4d936ec1 ME |
8594 | struct detected_devices_node *node_detected_dev; |
8595 | ||
8596 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8597 | if (node_detected_dev) { | |
8598 | node_detected_dev->dev = dev; | |
8599 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
8600 | } else { | |
8601 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
8602 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
8603 | } | |
1da177e4 LT |
8604 | } |
8605 | ||
8606 | ||
8607 | static void autostart_arrays(int part) | |
8608 | { | |
3cb03002 | 8609 | struct md_rdev *rdev; |
4d936ec1 ME |
8610 | struct detected_devices_node *node_detected_dev; |
8611 | dev_t dev; | |
8612 | int i_scanned, i_passed; | |
1da177e4 | 8613 | |
4d936ec1 ME |
8614 | i_scanned = 0; |
8615 | i_passed = 0; | |
1da177e4 | 8616 | |
4d936ec1 | 8617 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8618 | |
4d936ec1 ME |
8619 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8620 | i_scanned++; | |
8621 | node_detected_dev = list_entry(all_detected_devices.next, | |
8622 | struct detected_devices_node, list); | |
8623 | list_del(&node_detected_dev->list); | |
8624 | dev = node_detected_dev->dev; | |
8625 | kfree(node_detected_dev); | |
df968c4e | 8626 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
8627 | if (IS_ERR(rdev)) |
8628 | continue; | |
8629 | ||
b2d444d7 | 8630 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
8631 | MD_BUG(); |
8632 | continue; | |
8633 | } | |
d0fae18f | 8634 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 8635 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 8636 | i_passed++; |
1da177e4 | 8637 | } |
4d936ec1 ME |
8638 | |
8639 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
8640 | i_scanned, i_passed); | |
1da177e4 LT |
8641 | |
8642 | autorun_devices(part); | |
8643 | } | |
8644 | ||
fdee8ae4 | 8645 | #endif /* !MODULE */ |
1da177e4 LT |
8646 | |
8647 | static __exit void md_exit(void) | |
8648 | { | |
fd01b88c | 8649 | struct mddev *mddev; |
1da177e4 | 8650 | struct list_head *tmp; |
8ab5e4c1 | 8651 | |
3dbd8c2e | 8652 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 8653 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 8654 | |
3dbd8c2e | 8655 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
8656 | unregister_blkdev(mdp_major, "mdp"); |
8657 | unregister_reboot_notifier(&md_notifier); | |
8658 | unregister_sysctl_table(raid_table_header); | |
8659 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 8660 | for_each_mddev(mddev, tmp) { |
1da177e4 | 8661 | export_array(mddev); |
d3374825 | 8662 | mddev->hold_active = 0; |
1da177e4 | 8663 | } |
e804ac78 TH |
8664 | destroy_workqueue(md_misc_wq); |
8665 | destroy_workqueue(md_wq); | |
1da177e4 LT |
8666 | } |
8667 | ||
685784aa | 8668 | subsys_initcall(md_init); |
1da177e4 LT |
8669 | module_exit(md_exit) |
8670 | ||
f91de92e N |
8671 | static int get_ro(char *buffer, struct kernel_param *kp) |
8672 | { | |
8673 | return sprintf(buffer, "%d", start_readonly); | |
8674 | } | |
8675 | static int set_ro(const char *val, struct kernel_param *kp) | |
8676 | { | |
8677 | char *e; | |
8678 | int num = simple_strtoul(val, &e, 10); | |
8679 | if (*val && (*e == '\0' || *e == '\n')) { | |
8680 | start_readonly = num; | |
4dbcdc75 | 8681 | return 0; |
f91de92e N |
8682 | } |
8683 | return -EINVAL; | |
8684 | } | |
8685 | ||
80ca3a44 N |
8686 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
8687 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 8688 | |
efeb53c0 | 8689 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 8690 | |
1da177e4 LT |
8691 | EXPORT_SYMBOL(register_md_personality); |
8692 | EXPORT_SYMBOL(unregister_md_personality); | |
8693 | EXPORT_SYMBOL(md_error); | |
8694 | EXPORT_SYMBOL(md_done_sync); | |
8695 | EXPORT_SYMBOL(md_write_start); | |
8696 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
8697 | EXPORT_SYMBOL(md_register_thread); |
8698 | EXPORT_SYMBOL(md_unregister_thread); | |
8699 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 | 8700 | EXPORT_SYMBOL(md_check_recovery); |
a91d5ac0 | 8701 | EXPORT_SYMBOL(md_reap_sync_thread); |
1da177e4 | 8702 | MODULE_LICENSE("GPL"); |
0efb9e61 | 8703 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 8704 | MODULE_ALIAS("md"); |
72008652 | 8705 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |