]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | md.c : Multiple Devices driver for Linux | |
f72ffdd6 | 3 | Copyright (C) 1998, 1999, 2000 Ingo Molnar |
1da177e4 LT |
4 | |
5 | completely rewritten, based on the MD driver code from Marc Zyngier | |
6 | ||
7 | Changes: | |
8 | ||
9 | - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar | |
10 | - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com> | |
11 | - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net> | |
12 | - kerneld support by Boris Tobotras <boris@xtalk.msk.su> | |
13 | - kmod support by: Cyrus Durgin | |
14 | - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com> | |
15 | - Devfs support by Richard Gooch <rgooch@atnf.csiro.au> | |
16 | ||
17 | - lots of fixes and improvements to the RAID1/RAID5 and generic | |
18 | RAID code (such as request based resynchronization): | |
19 | ||
20 | Neil Brown <neilb@cse.unsw.edu.au>. | |
21 | ||
32a7627c N |
22 | - persistent bitmap code |
23 | Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. | |
24 | ||
1da177e4 LT |
25 | This program is free software; you can redistribute it and/or modify |
26 | it under the terms of the GNU General Public License as published by | |
27 | the Free Software Foundation; either version 2, or (at your option) | |
28 | any later version. | |
29 | ||
30 | You should have received a copy of the GNU General Public License | |
31 | (for example /usr/src/linux/COPYING); if not, write to the Free | |
32 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
9d48739e N |
33 | |
34 | Errors, Warnings, etc. | |
35 | Please use: | |
36 | pr_crit() for error conditions that risk data loss | |
37 | pr_err() for error conditions that are unexpected, like an IO error | |
38 | or internal inconsistency | |
39 | pr_warn() for error conditions that could have been predicated, like | |
40 | adding a device to an array when it has incompatible metadata | |
41 | pr_info() for every interesting, very rare events, like an array starting | |
42 | or stopping, or resync starting or stopping | |
43 | pr_debug() for everything else. | |
44 | ||
1da177e4 LT |
45 | */ |
46 | ||
3f07c014 | 47 | #include <linux/sched/signal.h> |
a6fb0934 | 48 | #include <linux/kthread.h> |
bff61975 | 49 | #include <linux/blkdev.h> |
fc974ee2 | 50 | #include <linux/badblocks.h> |
1da177e4 | 51 | #include <linux/sysctl.h> |
bff61975 | 52 | #include <linux/seq_file.h> |
ff01bb48 | 53 | #include <linux/fs.h> |
d7603b7e | 54 | #include <linux/poll.h> |
16f17b39 | 55 | #include <linux/ctype.h> |
e7d2860b | 56 | #include <linux/string.h> |
fb4d8c76 N |
57 | #include <linux/hdreg.h> |
58 | #include <linux/proc_fs.h> | |
59 | #include <linux/random.h> | |
056075c7 | 60 | #include <linux/module.h> |
fb4d8c76 | 61 | #include <linux/reboot.h> |
32a7627c | 62 | #include <linux/file.h> |
aa98aa31 | 63 | #include <linux/compat.h> |
25570727 | 64 | #include <linux/delay.h> |
bff61975 N |
65 | #include <linux/raid/md_p.h> |
66 | #include <linux/raid/md_u.h> | |
5a0e3ad6 | 67 | #include <linux/slab.h> |
504634f6 | 68 | #include <trace/events/block.h> |
43b2e5d8 | 69 | #include "md.h" |
ef740c37 | 70 | #include "bitmap.h" |
edb39c9d | 71 | #include "md-cluster.h" |
1da177e4 | 72 | |
1da177e4 | 73 | #ifndef MODULE |
d710e138 | 74 | static void autostart_arrays(int part); |
1da177e4 LT |
75 | #endif |
76 | ||
01f96c0a N |
77 | /* pers_list is a list of registered personalities protected |
78 | * by pers_lock. | |
79 | * pers_lock does extra service to protect accesses to | |
80 | * mddev->thread when the mutex cannot be held. | |
81 | */ | |
2604b703 | 82 | static LIST_HEAD(pers_list); |
1da177e4 LT |
83 | static DEFINE_SPINLOCK(pers_lock); |
84 | ||
edb39c9d | 85 | struct md_cluster_operations *md_cluster_ops; |
589a1c49 | 86 | EXPORT_SYMBOL(md_cluster_ops); |
edb39c9d GR |
87 | struct module *md_cluster_mod; |
88 | EXPORT_SYMBOL(md_cluster_mod); | |
89 | ||
90b08710 | 90 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
91 | static struct workqueue_struct *md_wq; |
92 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 93 | |
746d3207 N |
94 | static int remove_and_add_spares(struct mddev *mddev, |
95 | struct md_rdev *this); | |
5aa61f42 | 96 | static void mddev_detach(struct mddev *mddev); |
746d3207 | 97 | |
1e50915f RB |
98 | /* |
99 | * Default number of read corrections we'll attempt on an rdev | |
100 | * before ejecting it from the array. We divide the read error | |
101 | * count by 2 for every hour elapsed between read errors. | |
102 | */ | |
103 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
104 | /* |
105 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
106 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
107 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 108 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
109 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
110 | * speed limit - in case reconstruction slows down your system despite | |
111 | * idle IO detection. | |
112 | * | |
113 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 114 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
115 | */ |
116 | ||
117 | static int sysctl_speed_limit_min = 1000; | |
118 | static int sysctl_speed_limit_max = 200000; | |
fd01b88c | 119 | static inline int speed_min(struct mddev *mddev) |
88202a0c N |
120 | { |
121 | return mddev->sync_speed_min ? | |
122 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
123 | } | |
124 | ||
fd01b88c | 125 | static inline int speed_max(struct mddev *mddev) |
88202a0c N |
126 | { |
127 | return mddev->sync_speed_max ? | |
128 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
129 | } | |
1da177e4 LT |
130 | |
131 | static struct ctl_table_header *raid_table_header; | |
132 | ||
82592c38 | 133 | static struct ctl_table raid_table[] = { |
1da177e4 | 134 | { |
1da177e4 LT |
135 | .procname = "speed_limit_min", |
136 | .data = &sysctl_speed_limit_min, | |
137 | .maxlen = sizeof(int), | |
80ca3a44 | 138 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 139 | .proc_handler = proc_dointvec, |
1da177e4 LT |
140 | }, |
141 | { | |
1da177e4 LT |
142 | .procname = "speed_limit_max", |
143 | .data = &sysctl_speed_limit_max, | |
144 | .maxlen = sizeof(int), | |
80ca3a44 | 145 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 146 | .proc_handler = proc_dointvec, |
1da177e4 | 147 | }, |
894d2491 | 148 | { } |
1da177e4 LT |
149 | }; |
150 | ||
82592c38 | 151 | static struct ctl_table raid_dir_table[] = { |
1da177e4 | 152 | { |
1da177e4 LT |
153 | .procname = "raid", |
154 | .maxlen = 0, | |
80ca3a44 | 155 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
156 | .child = raid_table, |
157 | }, | |
894d2491 | 158 | { } |
1da177e4 LT |
159 | }; |
160 | ||
82592c38 | 161 | static struct ctl_table raid_root_table[] = { |
1da177e4 | 162 | { |
1da177e4 LT |
163 | .procname = "dev", |
164 | .maxlen = 0, | |
165 | .mode = 0555, | |
166 | .child = raid_dir_table, | |
167 | }, | |
894d2491 | 168 | { } |
1da177e4 LT |
169 | }; |
170 | ||
83d5cde4 | 171 | static const struct block_device_operations md_fops; |
1da177e4 | 172 | |
f91de92e N |
173 | static int start_readonly; |
174 | ||
a167f663 N |
175 | /* bio_clone_mddev |
176 | * like bio_clone, but with a local bio set | |
177 | */ | |
178 | ||
a167f663 | 179 | struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, |
fd01b88c | 180 | struct mddev *mddev) |
a167f663 N |
181 | { |
182 | struct bio *b; | |
a167f663 N |
183 | |
184 | if (!mddev || !mddev->bio_set) | |
185 | return bio_alloc(gfp_mask, nr_iovecs); | |
186 | ||
395c72a7 | 187 | b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set); |
a167f663 N |
188 | if (!b) |
189 | return NULL; | |
a167f663 N |
190 | return b; |
191 | } | |
192 | EXPORT_SYMBOL_GPL(bio_alloc_mddev); | |
193 | ||
d7603b7e N |
194 | /* |
195 | * We have a system wide 'event count' that is incremented | |
196 | * on any 'interesting' event, and readers of /proc/mdstat | |
197 | * can use 'poll' or 'select' to find out when the event | |
198 | * count increases. | |
199 | * | |
200 | * Events are: | |
201 | * start array, stop array, error, add device, remove device, | |
202 | * start build, activate spare | |
203 | */ | |
2989ddbd | 204 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 205 | static atomic_t md_event_count; |
fd01b88c | 206 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
207 | { |
208 | atomic_inc(&md_event_count); | |
209 | wake_up(&md_event_waiters); | |
210 | } | |
29269553 | 211 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 212 | |
1da177e4 LT |
213 | /* |
214 | * Enables to iterate over all existing md arrays | |
215 | * all_mddevs_lock protects this list. | |
216 | */ | |
217 | static LIST_HEAD(all_mddevs); | |
218 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
219 | ||
1da177e4 LT |
220 | /* |
221 | * iterates through all used mddevs in the system. | |
222 | * We take care to grab the all_mddevs_lock whenever navigating | |
223 | * the list, and to always hold a refcount when unlocked. | |
224 | * Any code which breaks out of this loop while own | |
225 | * a reference to the current mddev and must mddev_put it. | |
226 | */ | |
fd01b88c | 227 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 | 228 | \ |
f72ffdd6 | 229 | for (({ spin_lock(&all_mddevs_lock); \ |
fd01b88c N |
230 | _tmp = all_mddevs.next; \ |
231 | _mddev = NULL;}); \ | |
232 | ({ if (_tmp != &all_mddevs) \ | |
233 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 234 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
235 | if (_mddev) mddev_put(_mddev); \ |
236 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
237 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 238 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 239 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
240 | ) |
241 | ||
409c57f3 N |
242 | /* Rather than calling directly into the personality make_request function, |
243 | * IO requests come here first so that we can check if the device is | |
244 | * being suspended pending a reconfiguration. | |
245 | * We hold a refcount over the call to ->make_request. By the time that | |
246 | * call has finished, the bio has been linked into some internal structure | |
247 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
248 | */ | |
dece1635 | 249 | static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 250 | { |
49077326 | 251 | const int rw = bio_data_dir(bio); |
fd01b88c | 252 | struct mddev *mddev = q->queuedata; |
e91ece55 | 253 | unsigned int sectors; |
74672d06 | 254 | int cpu; |
49077326 | 255 | |
54efd50b KO |
256 | blk_queue_split(q, &bio, q->bio_split); |
257 | ||
274d8cbd | 258 | if (mddev == NULL || mddev->pers == NULL) { |
409c57f3 | 259 | bio_io_error(bio); |
dece1635 | 260 | return BLK_QC_T_NONE; |
409c57f3 | 261 | } |
bbfa57c0 | 262 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
4246a0b6 CH |
263 | if (bio_sectors(bio) != 0) |
264 | bio->bi_error = -EROFS; | |
265 | bio_endio(bio); | |
dece1635 | 266 | return BLK_QC_T_NONE; |
bbfa57c0 | 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); | |
9c573de3 | 291 | /* bio could be mergeable after passing to underlayer */ |
1eff9d32 | 292 | bio->bi_opf &= ~REQ_NOMERGE; |
5a7bbad2 | 293 | mddev->pers->make_request(mddev, bio); |
49077326 | 294 | |
74672d06 GZ |
295 | cpu = part_stat_lock(); |
296 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
297 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); | |
298 | part_stat_unlock(); | |
49077326 | 299 | |
409c57f3 N |
300 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
301 | wake_up(&mddev->sb_wait); | |
dece1635 JA |
302 | |
303 | return BLK_QC_T_NONE; | |
409c57f3 N |
304 | } |
305 | ||
9e35b99c N |
306 | /* mddev_suspend makes sure no new requests are submitted |
307 | * to the device, and that any requests that have been submitted | |
308 | * are completely handled. | |
afa0f557 N |
309 | * Once mddev_detach() is called and completes, the module will be |
310 | * completely unused. | |
9e35b99c | 311 | */ |
fd01b88c | 312 | void mddev_suspend(struct mddev *mddev) |
409c57f3 | 313 | { |
092398dc | 314 | WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk); |
0dc10e50 MP |
315 | if (mddev->suspended++) |
316 | return; | |
409c57f3 N |
317 | synchronize_rcu(); |
318 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
319 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
320 | |
321 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 322 | } |
390ee602 | 323 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 324 | |
fd01b88c | 325 | void mddev_resume(struct mddev *mddev) |
409c57f3 | 326 | { |
0dc10e50 MP |
327 | if (--mddev->suspended) |
328 | return; | |
409c57f3 N |
329 | wake_up(&mddev->sb_wait); |
330 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 331 | |
47525e59 | 332 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
333 | md_wakeup_thread(mddev->thread); |
334 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 335 | } |
390ee602 | 336 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 337 | |
fd01b88c | 338 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 | 339 | { |
5c675f83 N |
340 | struct md_personality *pers = mddev->pers; |
341 | int ret = 0; | |
342 | ||
343 | rcu_read_lock(); | |
344 | if (mddev->suspended) | |
345 | ret = 1; | |
346 | else if (pers && pers->congested) | |
347 | ret = pers->congested(mddev, bits); | |
348 | rcu_read_unlock(); | |
349 | return ret; | |
350 | } | |
351 | EXPORT_SYMBOL_GPL(mddev_congested); | |
352 | static int md_congested(void *data, int bits) | |
353 | { | |
354 | struct mddev *mddev = data; | |
355 | return mddev_congested(mddev, bits); | |
3fa841d7 | 356 | } |
3fa841d7 | 357 | |
a2826aa9 | 358 | /* |
e9c7469b | 359 | * Generic flush handling for md |
a2826aa9 N |
360 | */ |
361 | ||
4246a0b6 | 362 | static void md_end_flush(struct bio *bio) |
a2826aa9 | 363 | { |
3cb03002 | 364 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 365 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
366 | |
367 | rdev_dec_pending(rdev, mddev); | |
368 | ||
369 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 370 | /* The pre-request flush has finished */ |
e804ac78 | 371 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
372 | } |
373 | bio_put(bio); | |
374 | } | |
375 | ||
a7a07e69 N |
376 | static void md_submit_flush_data(struct work_struct *ws); |
377 | ||
a035fc3e | 378 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 379 | { |
fd01b88c | 380 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 381 | struct md_rdev *rdev; |
a2826aa9 | 382 | |
a7a07e69 N |
383 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
384 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 385 | rcu_read_lock(); |
dafb20fa | 386 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
387 | if (rdev->raid_disk >= 0 && |
388 | !test_bit(Faulty, &rdev->flags)) { | |
389 | /* Take two references, one is dropped | |
390 | * when request finishes, one after | |
391 | * we reclaim rcu_read_lock | |
392 | */ | |
393 | struct bio *bi; | |
394 | atomic_inc(&rdev->nr_pending); | |
395 | atomic_inc(&rdev->nr_pending); | |
396 | rcu_read_unlock(); | |
b5e1b8ce | 397 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 398 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
399 | bi->bi_private = rdev; |
400 | bi->bi_bdev = rdev->bdev; | |
70fd7614 | 401 | bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
a2826aa9 | 402 | atomic_inc(&mddev->flush_pending); |
4e49ea4a | 403 | submit_bio(bi); |
a2826aa9 N |
404 | rcu_read_lock(); |
405 | rdev_dec_pending(rdev, mddev); | |
406 | } | |
407 | rcu_read_unlock(); | |
a7a07e69 N |
408 | if (atomic_dec_and_test(&mddev->flush_pending)) |
409 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
410 | } |
411 | ||
e9c7469b | 412 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 413 | { |
fd01b88c | 414 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 415 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 416 | |
4f024f37 | 417 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 | 418 | /* an empty barrier - all done */ |
4246a0b6 | 419 | bio_endio(bio); |
a2826aa9 | 420 | else { |
1eff9d32 | 421 | bio->bi_opf &= ~REQ_PREFLUSH; |
5a7bbad2 | 422 | mddev->pers->make_request(mddev, bio); |
a2826aa9 | 423 | } |
2b74e12e N |
424 | |
425 | mddev->flush_bio = NULL; | |
426 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
427 | } |
428 | ||
fd01b88c | 429 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 | 430 | { |
85572d7c | 431 | spin_lock_irq(&mddev->lock); |
a2826aa9 | 432 | wait_event_lock_irq(mddev->sb_wait, |
e9c7469b | 433 | !mddev->flush_bio, |
85572d7c | 434 | mddev->lock); |
e9c7469b | 435 | mddev->flush_bio = bio; |
85572d7c | 436 | spin_unlock_irq(&mddev->lock); |
a2826aa9 | 437 | |
a035fc3e N |
438 | INIT_WORK(&mddev->flush_work, submit_flushes); |
439 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 440 | } |
e9c7469b | 441 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 442 | |
fd01b88c | 443 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
444 | { |
445 | atomic_inc(&mddev->active); | |
446 | return mddev; | |
447 | } | |
448 | ||
5fd3a17e | 449 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 450 | |
fd01b88c | 451 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 452 | { |
a167f663 N |
453 | struct bio_set *bs = NULL; |
454 | ||
1da177e4 LT |
455 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
456 | return; | |
d3374825 | 457 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
458 | mddev->ctime == 0 && !mddev->hold_active) { |
459 | /* Array is not configured at all, and not held active, | |
460 | * so destroy it */ | |
af8a2434 | 461 | list_del_init(&mddev->all_mddevs); |
a167f663 N |
462 | bs = mddev->bio_set; |
463 | mddev->bio_set = NULL; | |
d3374825 | 464 | if (mddev->gendisk) { |
e804ac78 TH |
465 | /* We did a probe so need to clean up. Call |
466 | * queue_work inside the spinlock so that | |
467 | * flush_workqueue() after mddev_find will | |
468 | * succeed in waiting for the work to be done. | |
d3374825 N |
469 | */ |
470 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 471 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
472 | } else |
473 | kfree(mddev); | |
474 | } | |
475 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
476 | if (bs) |
477 | bioset_free(bs); | |
1da177e4 LT |
478 | } |
479 | ||
25b2edfa SL |
480 | static void md_safemode_timeout(unsigned long data); |
481 | ||
fd01b88c | 482 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
483 | { |
484 | mutex_init(&mddev->open_mutex); | |
485 | mutex_init(&mddev->reconfig_mutex); | |
486 | mutex_init(&mddev->bitmap_info.mutex); | |
487 | INIT_LIST_HEAD(&mddev->disks); | |
488 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
25b2edfa SL |
489 | setup_timer(&mddev->safemode_timer, md_safemode_timeout, |
490 | (unsigned long) mddev); | |
fafd7fb0 N |
491 | atomic_set(&mddev->active, 1); |
492 | atomic_set(&mddev->openers, 0); | |
493 | atomic_set(&mddev->active_io, 0); | |
85572d7c | 494 | spin_lock_init(&mddev->lock); |
fafd7fb0 N |
495 | atomic_set(&mddev->flush_pending, 0); |
496 | init_waitqueue_head(&mddev->sb_wait); | |
497 | init_waitqueue_head(&mddev->recovery_wait); | |
498 | mddev->reshape_position = MaxSector; | |
2c810cdd | 499 | mddev->reshape_backwards = 0; |
c4a39551 | 500 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
501 | mddev->resync_min = 0; |
502 | mddev->resync_max = MaxSector; | |
503 | mddev->level = LEVEL_NONE; | |
504 | } | |
390ee602 | 505 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 506 | |
f72ffdd6 | 507 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 508 | { |
fd01b88c | 509 | struct mddev *mddev, *new = NULL; |
1da177e4 | 510 | |
8f5f02c4 N |
511 | if (unit && MAJOR(unit) != MD_MAJOR) |
512 | unit &= ~((1<<MdpMinorShift)-1); | |
513 | ||
1da177e4 LT |
514 | retry: |
515 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
516 | |
517 | if (unit) { | |
518 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
519 | if (mddev->unit == unit) { | |
520 | mddev_get(mddev); | |
521 | spin_unlock(&all_mddevs_lock); | |
522 | kfree(new); | |
523 | return mddev; | |
524 | } | |
525 | ||
526 | if (new) { | |
527 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 528 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
529 | new->hold_active = UNTIL_IOCTL; |
530 | return new; | |
1da177e4 | 531 | } |
efeb53c0 N |
532 | } else if (new) { |
533 | /* find an unused unit number */ | |
534 | static int next_minor = 512; | |
535 | int start = next_minor; | |
536 | int is_free = 0; | |
537 | int dev = 0; | |
538 | while (!is_free) { | |
539 | dev = MKDEV(MD_MAJOR, next_minor); | |
540 | next_minor++; | |
541 | if (next_minor > MINORMASK) | |
542 | next_minor = 0; | |
543 | if (next_minor == start) { | |
544 | /* Oh dear, all in use. */ | |
545 | spin_unlock(&all_mddevs_lock); | |
546 | kfree(new); | |
547 | return NULL; | |
548 | } | |
f72ffdd6 | 549 | |
efeb53c0 N |
550 | is_free = 1; |
551 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
552 | if (mddev->unit == dev) { | |
553 | is_free = 0; | |
554 | break; | |
555 | } | |
556 | } | |
557 | new->unit = dev; | |
558 | new->md_minor = MINOR(dev); | |
559 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
560 | list_add(&new->all_mddevs, &all_mddevs); |
561 | spin_unlock(&all_mddevs_lock); | |
562 | return new; | |
563 | } | |
564 | spin_unlock(&all_mddevs_lock); | |
565 | ||
9ffae0cf | 566 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
567 | if (!new) |
568 | return NULL; | |
569 | ||
1da177e4 LT |
570 | new->unit = unit; |
571 | if (MAJOR(unit) == MD_MAJOR) | |
572 | new->md_minor = MINOR(unit); | |
573 | else | |
574 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
575 | ||
fafd7fb0 | 576 | mddev_init(new); |
1da177e4 | 577 | |
1da177e4 LT |
578 | goto retry; |
579 | } | |
580 | ||
b6eb127d N |
581 | static struct attribute_group md_redundancy_group; |
582 | ||
5c47daf6 | 583 | void mddev_unlock(struct mddev *mddev) |
1da177e4 | 584 | { |
a64c876f | 585 | if (mddev->to_remove) { |
b6eb127d N |
586 | /* These cannot be removed under reconfig_mutex as |
587 | * an access to the files will try to take reconfig_mutex | |
588 | * while holding the file unremovable, which leads to | |
589 | * a deadlock. | |
bb4f1e9d N |
590 | * So hold set sysfs_active while the remove in happeing, |
591 | * and anything else which might set ->to_remove or my | |
592 | * otherwise change the sysfs namespace will fail with | |
593 | * -EBUSY if sysfs_active is still set. | |
594 | * We set sysfs_active under reconfig_mutex and elsewhere | |
595 | * test it under the same mutex to ensure its correct value | |
596 | * is seen. | |
b6eb127d | 597 | */ |
a64c876f N |
598 | struct attribute_group *to_remove = mddev->to_remove; |
599 | mddev->to_remove = NULL; | |
bb4f1e9d | 600 | mddev->sysfs_active = 1; |
b6eb127d N |
601 | mutex_unlock(&mddev->reconfig_mutex); |
602 | ||
00bcb4ac N |
603 | if (mddev->kobj.sd) { |
604 | if (to_remove != &md_redundancy_group) | |
605 | sysfs_remove_group(&mddev->kobj, to_remove); | |
606 | if (mddev->pers == NULL || | |
607 | mddev->pers->sync_request == NULL) { | |
608 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
609 | if (mddev->sysfs_action) | |
610 | sysfs_put(mddev->sysfs_action); | |
611 | mddev->sysfs_action = NULL; | |
612 | } | |
a64c876f | 613 | } |
bb4f1e9d | 614 | mddev->sysfs_active = 0; |
b6eb127d N |
615 | } else |
616 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 617 | |
751e67ca CD |
618 | /* As we've dropped the mutex we need a spinlock to |
619 | * make sure the thread doesn't disappear | |
01f96c0a N |
620 | */ |
621 | spin_lock(&pers_lock); | |
005eca5e | 622 | md_wakeup_thread(mddev->thread); |
01f96c0a | 623 | spin_unlock(&pers_lock); |
1da177e4 | 624 | } |
5c47daf6 | 625 | EXPORT_SYMBOL_GPL(mddev_unlock); |
1da177e4 | 626 | |
57d051dc | 627 | struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr) |
1ca69c4b N |
628 | { |
629 | struct md_rdev *rdev; | |
630 | ||
631 | rdev_for_each_rcu(rdev, mddev) | |
632 | if (rdev->desc_nr == nr) | |
633 | return rdev; | |
634 | ||
635 | return NULL; | |
636 | } | |
57d051dc | 637 | EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu); |
1ca69c4b N |
638 | |
639 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 640 | { |
3cb03002 | 641 | struct md_rdev *rdev; |
1da177e4 | 642 | |
dafb20fa | 643 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
644 | if (rdev->bdev->bd_dev == dev) |
645 | return rdev; | |
159ec1fc | 646 | |
1da177e4 LT |
647 | return NULL; |
648 | } | |
649 | ||
1ca69c4b N |
650 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
651 | { | |
652 | struct md_rdev *rdev; | |
653 | ||
654 | rdev_for_each_rcu(rdev, mddev) | |
655 | if (rdev->bdev->bd_dev == dev) | |
656 | return rdev; | |
657 | ||
658 | return NULL; | |
659 | } | |
660 | ||
84fc4b56 | 661 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 662 | { |
84fc4b56 | 663 | struct md_personality *pers; |
d9d166c2 N |
664 | list_for_each_entry(pers, &pers_list, list) { |
665 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 666 | return pers; |
d9d166c2 N |
667 | if (strcmp(pers->name, clevel)==0) |
668 | return pers; | |
669 | } | |
2604b703 N |
670 | return NULL; |
671 | } | |
672 | ||
b73df2d3 | 673 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 674 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 675 | { |
57b2caa3 | 676 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 677 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
678 | } |
679 | ||
f72ffdd6 | 680 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 681 | { |
1da177e4 | 682 | rdev->sb_page = alloc_page(GFP_KERNEL); |
7f0f0d87 | 683 | if (!rdev->sb_page) |
ebc24337 | 684 | return -ENOMEM; |
1da177e4 LT |
685 | return 0; |
686 | } | |
687 | ||
545c8795 | 688 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
689 | { |
690 | if (rdev->sb_page) { | |
2d1f3b5d | 691 | put_page(rdev->sb_page); |
1da177e4 LT |
692 | rdev->sb_loaded = 0; |
693 | rdev->sb_page = NULL; | |
0f420358 | 694 | rdev->sb_start = 0; |
dd8ac336 | 695 | rdev->sectors = 0; |
1da177e4 | 696 | } |
2699b672 N |
697 | if (rdev->bb_page) { |
698 | put_page(rdev->bb_page); | |
699 | rdev->bb_page = NULL; | |
700 | } | |
d3b407fb | 701 | badblocks_exit(&rdev->badblocks); |
1da177e4 | 702 | } |
545c8795 | 703 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 704 | |
4246a0b6 | 705 | static void super_written(struct bio *bio) |
7bfa19f2 | 706 | { |
3cb03002 | 707 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 708 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 709 | |
4246a0b6 | 710 | if (bio->bi_error) { |
9d48739e | 711 | pr_err("md: super_written gets error=%d\n", bio->bi_error); |
a9701a30 | 712 | md_error(mddev, rdev); |
46533ff7 N |
713 | if (!test_bit(Faulty, &rdev->flags) |
714 | && (bio->bi_opf & MD_FAILFAST)) { | |
2953079c | 715 | set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags); |
46533ff7 N |
716 | set_bit(LastDev, &rdev->flags); |
717 | } | |
718 | } else | |
719 | clear_bit(LastDev, &rdev->flags); | |
7bfa19f2 | 720 | |
a9701a30 N |
721 | if (atomic_dec_and_test(&mddev->pending_writes)) |
722 | wake_up(&mddev->sb_wait); | |
ed3b98c7 | 723 | rdev_dec_pending(rdev, mddev); |
f8b58edf | 724 | bio_put(bio); |
7bfa19f2 N |
725 | } |
726 | ||
fd01b88c | 727 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
728 | sector_t sector, int size, struct page *page) |
729 | { | |
730 | /* write first size bytes of page to sector of rdev | |
731 | * Increment mddev->pending_writes before returning | |
732 | * and decrement it on completion, waking up sb_wait | |
733 | * if zero is reached. | |
734 | * If an error occurred, call md_error | |
735 | */ | |
46533ff7 N |
736 | struct bio *bio; |
737 | int ff = 0; | |
738 | ||
739 | if (test_bit(Faulty, &rdev->flags)) | |
740 | return; | |
741 | ||
742 | bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); | |
7bfa19f2 | 743 | |
ed3b98c7 SL |
744 | atomic_inc(&rdev->nr_pending); |
745 | ||
a6ff7e08 | 746 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
4f024f37 | 747 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
748 | bio_add_page(bio, page, size, 0); |
749 | bio->bi_private = rdev; | |
750 | bio->bi_end_io = super_written; | |
46533ff7 N |
751 | |
752 | if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) && | |
753 | test_bit(FailFast, &rdev->flags) && | |
754 | !test_bit(LastDev, &rdev->flags)) | |
755 | ff = MD_FAILFAST; | |
20737738 | 756 | bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_FUA | ff; |
a9701a30 | 757 | |
7bfa19f2 | 758 | atomic_inc(&mddev->pending_writes); |
4e49ea4a | 759 | submit_bio(bio); |
a9701a30 N |
760 | } |
761 | ||
46533ff7 | 762 | int md_super_wait(struct mddev *mddev) |
a9701a30 | 763 | { |
e9c7469b | 764 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 765 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
2953079c | 766 | if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags)) |
46533ff7 N |
767 | return -EAGAIN; |
768 | return 0; | |
7bfa19f2 N |
769 | } |
770 | ||
3cb03002 | 771 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
796a5cf0 | 772 | struct page *page, int op, int op_flags, bool metadata_op) |
1da177e4 | 773 | { |
a167f663 | 774 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
775 | int ret; |
776 | ||
a6ff7e08 JB |
777 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
778 | rdev->meta_bdev : rdev->bdev; | |
796a5cf0 | 779 | bio_set_op_attrs(bio, op, op_flags); |
ccebd4c4 | 780 | if (metadata_op) |
4f024f37 | 781 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
782 | else if (rdev->mddev->reshape_position != MaxSector && |
783 | (rdev->mddev->reshape_backwards == | |
784 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 785 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 786 | else |
4f024f37 | 787 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 788 | bio_add_page(bio, page, size, 0); |
4e49ea4a MC |
789 | |
790 | submit_bio_wait(bio); | |
1da177e4 | 791 | |
4246a0b6 | 792 | ret = !bio->bi_error; |
1da177e4 LT |
793 | bio_put(bio); |
794 | return ret; | |
795 | } | |
a8745db2 | 796 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 797 | |
f72ffdd6 | 798 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
799 | { |
800 | char b[BDEVNAME_SIZE]; | |
403df478 | 801 | |
1da177e4 LT |
802 | if (rdev->sb_loaded) |
803 | return 0; | |
804 | ||
796a5cf0 | 805 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) |
1da177e4 LT |
806 | goto fail; |
807 | rdev->sb_loaded = 1; | |
808 | return 0; | |
809 | ||
810 | fail: | |
9d48739e N |
811 | pr_err("md: disabled device %s, could not read superblock.\n", |
812 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
813 | return -EINVAL; |
814 | } | |
815 | ||
816 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
817 | { | |
f72ffdd6 | 818 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
819 | sb1->set_uuid1 == sb2->set_uuid1 && |
820 | sb1->set_uuid2 == sb2->set_uuid2 && | |
821 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
822 | } |
823 | ||
1da177e4 LT |
824 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
825 | { | |
826 | int ret; | |
827 | mdp_super_t *tmp1, *tmp2; | |
828 | ||
829 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
830 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
831 | ||
832 | if (!tmp1 || !tmp2) { | |
833 | ret = 0; | |
1da177e4 LT |
834 | goto abort; |
835 | } | |
836 | ||
837 | *tmp1 = *sb1; | |
838 | *tmp2 = *sb2; | |
839 | ||
840 | /* | |
841 | * nr_disks is not constant | |
842 | */ | |
843 | tmp1->nr_disks = 0; | |
844 | tmp2->nr_disks = 0; | |
845 | ||
ce0c8e05 | 846 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 847 | abort: |
990a8baf JJ |
848 | kfree(tmp1); |
849 | kfree(tmp2); | |
1da177e4 LT |
850 | return ret; |
851 | } | |
852 | ||
4d167f09 N |
853 | static u32 md_csum_fold(u32 csum) |
854 | { | |
855 | csum = (csum & 0xffff) + (csum >> 16); | |
856 | return (csum & 0xffff) + (csum >> 16); | |
857 | } | |
858 | ||
f72ffdd6 | 859 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 860 | { |
4d167f09 N |
861 | u64 newcsum = 0; |
862 | u32 *sb32 = (u32*)sb; | |
863 | int i; | |
1da177e4 LT |
864 | unsigned int disk_csum, csum; |
865 | ||
866 | disk_csum = sb->sb_csum; | |
867 | sb->sb_csum = 0; | |
4d167f09 N |
868 | |
869 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
870 | newcsum += sb32[i]; | |
871 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
872 | ||
4d167f09 N |
873 | #ifdef CONFIG_ALPHA |
874 | /* This used to use csum_partial, which was wrong for several | |
875 | * reasons including that different results are returned on | |
876 | * different architectures. It isn't critical that we get exactly | |
877 | * the same return value as before (we always csum_fold before | |
878 | * testing, and that removes any differences). However as we | |
879 | * know that csum_partial always returned a 16bit value on | |
880 | * alphas, do a fold to maximise conformity to previous behaviour. | |
881 | */ | |
882 | sb->sb_csum = md_csum_fold(disk_csum); | |
883 | #else | |
1da177e4 | 884 | sb->sb_csum = disk_csum; |
4d167f09 | 885 | #endif |
1da177e4 LT |
886 | return csum; |
887 | } | |
888 | ||
1da177e4 LT |
889 | /* |
890 | * Handle superblock details. | |
891 | * We want to be able to handle multiple superblock formats | |
892 | * so we have a common interface to them all, and an array of | |
893 | * different handlers. | |
894 | * We rely on user-space to write the initial superblock, and support | |
895 | * reading and updating of superblocks. | |
896 | * Interface methods are: | |
3cb03002 | 897 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
898 | * loads and validates a superblock on dev. |
899 | * if refdev != NULL, compare superblocks on both devices | |
900 | * Return: | |
901 | * 0 - dev has a superblock that is compatible with refdev | |
902 | * 1 - dev has a superblock that is compatible and newer than refdev | |
903 | * so dev should be used as the refdev in future | |
904 | * -EINVAL superblock incompatible or invalid | |
905 | * -othererror e.g. -EIO | |
906 | * | |
fd01b88c | 907 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
908 | * Verify that dev is acceptable into mddev. |
909 | * The first time, mddev->raid_disks will be 0, and data from | |
910 | * dev should be merged in. Subsequent calls check that dev | |
911 | * is new enough. Return 0 or -EINVAL | |
912 | * | |
fd01b88c | 913 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
914 | * Update the superblock for rdev with data in mddev |
915 | * This does not write to disc. | |
916 | * | |
917 | */ | |
918 | ||
919 | struct super_type { | |
0cd17fec CW |
920 | char *name; |
921 | struct module *owner; | |
c6563a8c N |
922 | int (*load_super)(struct md_rdev *rdev, |
923 | struct md_rdev *refdev, | |
0cd17fec | 924 | int minor_version); |
c6563a8c N |
925 | int (*validate_super)(struct mddev *mddev, |
926 | struct md_rdev *rdev); | |
927 | void (*sync_super)(struct mddev *mddev, | |
928 | struct md_rdev *rdev); | |
3cb03002 | 929 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 930 | sector_t num_sectors); |
c6563a8c N |
931 | int (*allow_new_offset)(struct md_rdev *rdev, |
932 | unsigned long long new_offset); | |
1da177e4 LT |
933 | }; |
934 | ||
0894cc30 AN |
935 | /* |
936 | * Check that the given mddev has no bitmap. | |
937 | * | |
938 | * This function is called from the run method of all personalities that do not | |
939 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
940 | * has a bitmap. Otherwise, it returns 0. | |
941 | * | |
942 | */ | |
fd01b88c | 943 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 944 | { |
c3d9714e | 945 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 | 946 | return 0; |
9d48739e | 947 | pr_warn("%s: bitmaps are not supported for %s\n", |
0894cc30 AN |
948 | mdname(mddev), mddev->pers->name); |
949 | return 1; | |
950 | } | |
951 | EXPORT_SYMBOL(md_check_no_bitmap); | |
952 | ||
1da177e4 | 953 | /* |
f72ffdd6 | 954 | * load_super for 0.90.0 |
1da177e4 | 955 | */ |
3cb03002 | 956 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
957 | { |
958 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
959 | mdp_super_t *sb; | |
960 | int ret; | |
1da177e4 LT |
961 | |
962 | /* | |
0f420358 | 963 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
964 | * it's at the end of the disk. |
965 | * | |
966 | * It also happens to be a multiple of 4Kb. | |
967 | */ | |
57b2caa3 | 968 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 969 | |
0002b271 | 970 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
9d48739e N |
971 | if (ret) |
972 | return ret; | |
1da177e4 LT |
973 | |
974 | ret = -EINVAL; | |
975 | ||
976 | bdevname(rdev->bdev, b); | |
65a06f06 | 977 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
978 | |
979 | if (sb->md_magic != MD_SB_MAGIC) { | |
9d48739e | 980 | pr_warn("md: invalid raid superblock magic on %s\n", b); |
1da177e4 LT |
981 | goto abort; |
982 | } | |
983 | ||
984 | if (sb->major_version != 0 || | |
f6705578 N |
985 | sb->minor_version < 90 || |
986 | sb->minor_version > 91) { | |
9d48739e N |
987 | pr_warn("Bad version number %d.%d on %s\n", |
988 | sb->major_version, sb->minor_version, b); | |
1da177e4 LT |
989 | goto abort; |
990 | } | |
991 | ||
992 | if (sb->raid_disks <= 0) | |
993 | goto abort; | |
994 | ||
4d167f09 | 995 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
9d48739e | 996 | pr_warn("md: invalid superblock checksum on %s\n", b); |
1da177e4 LT |
997 | goto abort; |
998 | } | |
999 | ||
1000 | rdev->preferred_minor = sb->md_minor; | |
1001 | rdev->data_offset = 0; | |
c6563a8c | 1002 | rdev->new_data_offset = 0; |
0002b271 | 1003 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1004 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1005 | |
1006 | if (sb->level == LEVEL_MULTIPATH) | |
1007 | rdev->desc_nr = -1; | |
1008 | else | |
1009 | rdev->desc_nr = sb->this_disk.number; | |
1010 | ||
9a7b2b0f | 1011 | if (!refdev) { |
1da177e4 | 1012 | ret = 1; |
9a7b2b0f | 1013 | } else { |
1da177e4 | 1014 | __u64 ev1, ev2; |
65a06f06 | 1015 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 | 1016 | if (!uuid_equal(refsb, sb)) { |
9d48739e | 1017 | pr_warn("md: %s has different UUID to %s\n", |
1da177e4 LT |
1018 | b, bdevname(refdev->bdev,b2)); |
1019 | goto abort; | |
1020 | } | |
1021 | if (!sb_equal(refsb, sb)) { | |
9d48739e N |
1022 | pr_warn("md: %s has same UUID but different superblock to %s\n", |
1023 | b, bdevname(refdev->bdev, b2)); | |
1da177e4 LT |
1024 | goto abort; |
1025 | } | |
1026 | ev1 = md_event(sb); | |
1027 | ev2 = md_event(refsb); | |
1028 | if (ev1 > ev2) | |
1029 | ret = 1; | |
f72ffdd6 | 1030 | else |
1da177e4 LT |
1031 | ret = 0; |
1032 | } | |
8190e754 | 1033 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1034 | /* Limit to 4TB as metadata cannot record more than that. |
1035 | * (not needed for Linear and RAID0 as metadata doesn't | |
1036 | * record this size) | |
1037 | */ | |
3312c951 AB |
1038 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) && |
1039 | sb->level >= 1) | |
1040 | rdev->sectors = (sector_t)(2ULL << 32) - 2; | |
1da177e4 | 1041 | |
27a7b260 | 1042 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1043 | /* "this cannot possibly happen" ... */ |
1044 | ret = -EINVAL; | |
1045 | ||
1da177e4 LT |
1046 | abort: |
1047 | return ret; | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * validate_super for 0.90.0 | |
1052 | */ | |
fd01b88c | 1053 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1054 | { |
1055 | mdp_disk_t *desc; | |
65a06f06 | 1056 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1057 | __u64 ev1 = md_event(sb); |
1da177e4 | 1058 | |
41158c7e | 1059 | rdev->raid_disk = -1; |
c5d79adb N |
1060 | clear_bit(Faulty, &rdev->flags); |
1061 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1062 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1063 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1064 | |
1da177e4 LT |
1065 | if (mddev->raid_disks == 0) { |
1066 | mddev->major_version = 0; | |
1067 | mddev->minor_version = sb->minor_version; | |
1068 | mddev->patch_version = sb->patch_version; | |
e691063a | 1069 | mddev->external = 0; |
9d8f0363 | 1070 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1071 | mddev->ctime = sb->ctime; |
1072 | mddev->utime = sb->utime; | |
1073 | mddev->level = sb->level; | |
d9d166c2 | 1074 | mddev->clevel[0] = 0; |
1da177e4 LT |
1075 | mddev->layout = sb->layout; |
1076 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1077 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1078 | mddev->events = ev1; |
c3d9714e | 1079 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1080 | mddev->bitmap_info.space = 0; |
1081 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1082 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1083 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1084 | mddev->reshape_backwards = 0; |
1da177e4 | 1085 | |
f6705578 N |
1086 | if (mddev->minor_version >= 91) { |
1087 | mddev->reshape_position = sb->reshape_position; | |
1088 | mddev->delta_disks = sb->delta_disks; | |
1089 | mddev->new_level = sb->new_level; | |
1090 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1091 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1092 | if (mddev->delta_disks < 0) |
1093 | mddev->reshape_backwards = 1; | |
f6705578 N |
1094 | } else { |
1095 | mddev->reshape_position = MaxSector; | |
1096 | mddev->delta_disks = 0; | |
1097 | mddev->new_level = mddev->level; | |
1098 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1099 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1100 | } |
1101 | ||
1da177e4 LT |
1102 | if (sb->state & (1<<MD_SB_CLEAN)) |
1103 | mddev->recovery_cp = MaxSector; | |
1104 | else { | |
f72ffdd6 | 1105 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1106 | sb->events_lo == sb->cp_events_lo) { |
1107 | mddev->recovery_cp = sb->recovery_cp; | |
1108 | } else | |
1109 | mddev->recovery_cp = 0; | |
1110 | } | |
1111 | ||
1112 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1113 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1114 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1115 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1116 | ||
1117 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1118 | |
1119 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1120 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1121 | mddev->bitmap_info.offset = |
1122 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1123 | mddev->bitmap_info.space = |
c9ad020f | 1124 | mddev->bitmap_info.default_space; |
6409bb05 | 1125 | } |
a654b9d8 | 1126 | |
41158c7e | 1127 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1128 | /* Insist on good event counter while assembling, except |
1129 | * for spares (which don't need an event count) */ | |
1da177e4 | 1130 | ++ev1; |
be6800a7 N |
1131 | if (sb->disks[rdev->desc_nr].state & ( |
1132 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1133 | if (ev1 < mddev->events) |
be6800a7 | 1134 | return -EINVAL; |
41158c7e N |
1135 | } else if (mddev->bitmap) { |
1136 | /* if adding to array with a bitmap, then we can accept an | |
1137 | * older device ... but not too old. | |
1138 | */ | |
41158c7e N |
1139 | if (ev1 < mddev->bitmap->events_cleared) |
1140 | return 0; | |
8313b8e5 N |
1141 | if (ev1 < mddev->events) |
1142 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1143 | } else { |
1144 | if (ev1 < mddev->events) | |
1145 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1146 | return 0; | |
1147 | } | |
41158c7e | 1148 | |
1da177e4 | 1149 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1150 | desc = sb->disks + rdev->desc_nr; |
1151 | ||
1152 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1153 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1154 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1155 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1156 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1157 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1158 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1159 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1160 | /* active but not in sync implies recovery up to | |
1161 | * reshape position. We don't know exactly where | |
1162 | * that is, so set to zero for now */ | |
1163 | if (mddev->minor_version >= 91) { | |
1164 | rdev->recovery_offset = 0; | |
1165 | rdev->raid_disk = desc->raid_disk; | |
1166 | } | |
1da177e4 | 1167 | } |
8ddf9efe N |
1168 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1169 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1170 | if (desc->state & (1<<MD_DISK_FAILFAST)) |
1171 | set_bit(FailFast, &rdev->flags); | |
41158c7e | 1172 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1173 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1174 | return 0; |
1175 | } | |
1176 | ||
1177 | /* | |
1178 | * sync_super for 0.90.0 | |
1179 | */ | |
fd01b88c | 1180 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1181 | { |
1182 | mdp_super_t *sb; | |
3cb03002 | 1183 | struct md_rdev *rdev2; |
1da177e4 | 1184 | int next_spare = mddev->raid_disks; |
19133a42 | 1185 | |
1da177e4 LT |
1186 | /* make rdev->sb match mddev data.. |
1187 | * | |
1188 | * 1/ zero out disks | |
1189 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1190 | * 3/ any empty disks < next_spare become removed | |
1191 | * | |
1192 | * disks[0] gets initialised to REMOVED because | |
1193 | * we cannot be sure from other fields if it has | |
1194 | * been initialised or not. | |
1195 | */ | |
1196 | int i; | |
1197 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1198 | ||
61181565 N |
1199 | rdev->sb_size = MD_SB_BYTES; |
1200 | ||
65a06f06 | 1201 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1202 | |
1203 | memset(sb, 0, sizeof(*sb)); | |
1204 | ||
1205 | sb->md_magic = MD_SB_MAGIC; | |
1206 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1207 | sb->patch_version = mddev->patch_version; |
1208 | sb->gvalid_words = 0; /* ignored */ | |
1209 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1210 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1211 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1212 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1213 | ||
9ebc6ef1 | 1214 | sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 1215 | sb->level = mddev->level; |
58c0fed4 | 1216 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1217 | sb->raid_disks = mddev->raid_disks; |
1218 | sb->md_minor = mddev->md_minor; | |
e691063a | 1219 | sb->not_persistent = 0; |
9ebc6ef1 | 1220 | sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
1221 | sb->state = 0; |
1222 | sb->events_hi = (mddev->events>>32); | |
1223 | sb->events_lo = (u32)mddev->events; | |
1224 | ||
f6705578 N |
1225 | if (mddev->reshape_position == MaxSector) |
1226 | sb->minor_version = 90; | |
1227 | else { | |
1228 | sb->minor_version = 91; | |
1229 | sb->reshape_position = mddev->reshape_position; | |
1230 | sb->new_level = mddev->new_level; | |
1231 | sb->delta_disks = mddev->delta_disks; | |
1232 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1233 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1234 | } |
1235 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1236 | if (mddev->in_sync) |
1237 | { | |
1238 | sb->recovery_cp = mddev->recovery_cp; | |
1239 | sb->cp_events_hi = (mddev->events>>32); | |
1240 | sb->cp_events_lo = (u32)mddev->events; | |
1241 | if (mddev->recovery_cp == MaxSector) | |
1242 | sb->state = (1<< MD_SB_CLEAN); | |
1243 | } else | |
1244 | sb->recovery_cp = 0; | |
1245 | ||
1246 | sb->layout = mddev->layout; | |
9d8f0363 | 1247 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1248 | |
c3d9714e | 1249 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1250 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1251 | ||
1da177e4 | 1252 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1253 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1254 | mdp_disk_t *d; |
86e6ffdd | 1255 | int desc_nr; |
0261cd9f N |
1256 | int is_active = test_bit(In_sync, &rdev2->flags); |
1257 | ||
1258 | if (rdev2->raid_disk >= 0 && | |
1259 | sb->minor_version >= 91) | |
1260 | /* we have nowhere to store the recovery_offset, | |
1261 | * but if it is not below the reshape_position, | |
1262 | * we can piggy-back on that. | |
1263 | */ | |
1264 | is_active = 1; | |
1265 | if (rdev2->raid_disk < 0 || | |
1266 | test_bit(Faulty, &rdev2->flags)) | |
1267 | is_active = 0; | |
1268 | if (is_active) | |
86e6ffdd | 1269 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1270 | else |
86e6ffdd | 1271 | desc_nr = next_spare++; |
19133a42 | 1272 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1273 | d = &sb->disks[rdev2->desc_nr]; |
1274 | nr_disks++; | |
1275 | d->number = rdev2->desc_nr; | |
1276 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1277 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1278 | if (is_active) |
1da177e4 LT |
1279 | d->raid_disk = rdev2->raid_disk; |
1280 | else | |
1281 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1282 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1283 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1284 | else if (is_active) { |
1da177e4 | 1285 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1286 | if (test_bit(In_sync, &rdev2->flags)) |
1287 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1288 | active++; |
1289 | working++; | |
1290 | } else { | |
1291 | d->state = 0; | |
1292 | spare++; | |
1293 | working++; | |
1294 | } | |
8ddf9efe N |
1295 | if (test_bit(WriteMostly, &rdev2->flags)) |
1296 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
1297 | if (test_bit(FailFast, &rdev2->flags)) |
1298 | d->state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 | 1299 | } |
1da177e4 LT |
1300 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1301 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1302 | mdp_disk_t *d = &sb->disks[i]; | |
1303 | if (d->state == 0 && d->number == 0) { | |
1304 | d->number = i; | |
1305 | d->raid_disk = i; | |
1306 | d->state = (1<<MD_DISK_REMOVED); | |
1307 | d->state |= (1<<MD_DISK_FAULTY); | |
1308 | failed++; | |
1309 | } | |
1310 | } | |
1311 | sb->nr_disks = nr_disks; | |
1312 | sb->active_disks = active; | |
1313 | sb->working_disks = working; | |
1314 | sb->failed_disks = failed; | |
1315 | sb->spare_disks = spare; | |
1316 | ||
1317 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1318 | sb->sb_csum = calc_sb_csum(sb); | |
1319 | } | |
1320 | ||
0cd17fec CW |
1321 | /* |
1322 | * rdev_size_change for 0.90.0 | |
1323 | */ | |
1324 | static unsigned long long | |
3cb03002 | 1325 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1326 | { |
58c0fed4 | 1327 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1328 | return 0; /* component must fit device */ |
c3d9714e | 1329 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1330 | return 0; /* can't move bitmap */ |
57b2caa3 | 1331 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1332 | if (!num_sectors || num_sectors > rdev->sb_start) |
1333 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1334 | /* Limit to 4TB as metadata cannot record more than that. |
1335 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1336 | */ | |
3312c951 AB |
1337 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) && |
1338 | rdev->mddev->level >= 1) | |
1339 | num_sectors = (sector_t)(2ULL << 32) - 2; | |
46533ff7 N |
1340 | do { |
1341 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
0cd17fec | 1342 | rdev->sb_page); |
46533ff7 | 1343 | } while (md_super_wait(rdev->mddev) < 0); |
c26a44ed | 1344 | return num_sectors; |
0cd17fec CW |
1345 | } |
1346 | ||
c6563a8c N |
1347 | static int |
1348 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1349 | { | |
1350 | /* non-zero offset changes not possible with v0.90 */ | |
1351 | return new_offset == 0; | |
1352 | } | |
0cd17fec | 1353 | |
1da177e4 LT |
1354 | /* |
1355 | * version 1 superblock | |
1356 | */ | |
1357 | ||
f72ffdd6 | 1358 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1359 | { |
1c05b4bc N |
1360 | __le32 disk_csum; |
1361 | u32 csum; | |
1da177e4 LT |
1362 | unsigned long long newcsum; |
1363 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1364 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1365 | |
1366 | disk_csum = sb->sb_csum; | |
1367 | sb->sb_csum = 0; | |
1368 | newcsum = 0; | |
1f3c9907 | 1369 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1370 | newcsum += le32_to_cpu(*isuper++); |
1371 | ||
1372 | if (size == 2) | |
1c05b4bc | 1373 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1374 | |
1375 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1376 | sb->sb_csum = disk_csum; | |
1377 | return cpu_to_le32(csum); | |
1378 | } | |
1379 | ||
3cb03002 | 1380 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1381 | { |
1382 | struct mdp_superblock_1 *sb; | |
1383 | int ret; | |
0f420358 | 1384 | sector_t sb_start; |
c6563a8c | 1385 | sector_t sectors; |
1da177e4 | 1386 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1387 | int bmask; |
1da177e4 LT |
1388 | |
1389 | /* | |
0f420358 | 1390 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1391 | * It is always aligned to a 4K boundary and |
1392 | * depeding on minor_version, it can be: | |
1393 | * 0: At least 8K, but less than 12K, from end of device | |
1394 | * 1: At start of device | |
1395 | * 2: 4K from start of device. | |
1396 | */ | |
1397 | switch(minor_version) { | |
1398 | case 0: | |
77304d2a | 1399 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1400 | sb_start -= 8*2; |
1401 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1402 | break; |
1403 | case 1: | |
0f420358 | 1404 | sb_start = 0; |
1da177e4 LT |
1405 | break; |
1406 | case 2: | |
0f420358 | 1407 | sb_start = 8; |
1da177e4 LT |
1408 | break; |
1409 | default: | |
1410 | return -EINVAL; | |
1411 | } | |
0f420358 | 1412 | rdev->sb_start = sb_start; |
1da177e4 | 1413 | |
0002b271 N |
1414 | /* superblock is rarely larger than 1K, but it can be larger, |
1415 | * and it is safe to read 4k, so we do that | |
1416 | */ | |
1417 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1418 | if (ret) return ret; |
1419 | ||
65a06f06 | 1420 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1421 | |
1422 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1423 | sb->major_version != cpu_to_le32(1) || | |
1424 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1425 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1426 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1427 | return -EINVAL; |
1428 | ||
1429 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
9d48739e | 1430 | pr_warn("md: invalid superblock checksum on %s\n", |
1da177e4 LT |
1431 | bdevname(rdev->bdev,b)); |
1432 | return -EINVAL; | |
1433 | } | |
1434 | if (le64_to_cpu(sb->data_size) < 10) { | |
9d48739e N |
1435 | pr_warn("md: data_size too small on %s\n", |
1436 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1437 | return -EINVAL; |
1438 | } | |
c6563a8c N |
1439 | if (sb->pad0 || |
1440 | sb->pad3[0] || | |
1441 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1442 | /* Some padding is non-zero, might be a new feature */ | |
1443 | return -EINVAL; | |
e11e93fa | 1444 | |
1da177e4 LT |
1445 | rdev->preferred_minor = 0xffff; |
1446 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1447 | rdev->new_data_offset = rdev->data_offset; |
1448 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1449 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1450 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1451 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1452 | |
0002b271 | 1453 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1454 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1455 | if (rdev->sb_size & bmask) |
a1801f85 N |
1456 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1457 | ||
1458 | if (minor_version | |
0f420358 | 1459 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1460 | return -EINVAL; |
c6563a8c N |
1461 | if (minor_version |
1462 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1463 | return -EINVAL; | |
0002b271 | 1464 | |
31b65a0d N |
1465 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1466 | rdev->desc_nr = -1; | |
1467 | else | |
1468 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1469 | ||
2699b672 N |
1470 | if (!rdev->bb_page) { |
1471 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1472 | if (!rdev->bb_page) | |
1473 | return -ENOMEM; | |
1474 | } | |
1475 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1476 | rdev->badblocks.count == 0) { | |
1477 | /* need to load the bad block list. | |
1478 | * Currently we limit it to one page. | |
1479 | */ | |
1480 | s32 offset; | |
1481 | sector_t bb_sector; | |
1482 | u64 *bbp; | |
1483 | int i; | |
1484 | int sectors = le16_to_cpu(sb->bblog_size); | |
1485 | if (sectors > (PAGE_SIZE / 512)) | |
1486 | return -EINVAL; | |
1487 | offset = le32_to_cpu(sb->bblog_offset); | |
1488 | if (offset == 0) | |
1489 | return -EINVAL; | |
1490 | bb_sector = (long long)offset; | |
1491 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
796a5cf0 | 1492 | rdev->bb_page, REQ_OP_READ, 0, true)) |
2699b672 N |
1493 | return -EIO; |
1494 | bbp = (u64 *)page_address(rdev->bb_page); | |
1495 | rdev->badblocks.shift = sb->bblog_shift; | |
1496 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1497 | u64 bb = le64_to_cpu(*bbp); | |
1498 | int count = bb & (0x3ff); | |
1499 | u64 sector = bb >> 10; | |
1500 | sector <<= sb->bblog_shift; | |
1501 | count <<= sb->bblog_shift; | |
1502 | if (bb + 1 == 0) | |
1503 | break; | |
fc974ee2 | 1504 | if (badblocks_set(&rdev->badblocks, sector, count, 1)) |
2699b672 N |
1505 | return -EINVAL; |
1506 | } | |
486adf72 N |
1507 | } else if (sb->bblog_offset != 0) |
1508 | rdev->badblocks.shift = 0; | |
2699b672 | 1509 | |
9a7b2b0f | 1510 | if (!refdev) { |
8ed75463 | 1511 | ret = 1; |
9a7b2b0f | 1512 | } else { |
1da177e4 | 1513 | __u64 ev1, ev2; |
65a06f06 | 1514 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1515 | |
1516 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1517 | sb->level != refsb->level || | |
1518 | sb->layout != refsb->layout || | |
1519 | sb->chunksize != refsb->chunksize) { | |
9d48739e | 1520 | pr_warn("md: %s has strangely different superblock to %s\n", |
1da177e4 LT |
1521 | bdevname(rdev->bdev,b), |
1522 | bdevname(refdev->bdev,b2)); | |
1523 | return -EINVAL; | |
1524 | } | |
1525 | ev1 = le64_to_cpu(sb->events); | |
1526 | ev2 = le64_to_cpu(refsb->events); | |
1527 | ||
1528 | if (ev1 > ev2) | |
8ed75463 N |
1529 | ret = 1; |
1530 | else | |
1531 | ret = 0; | |
1da177e4 | 1532 | } |
c6563a8c N |
1533 | if (minor_version) { |
1534 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1535 | sectors -= rdev->data_offset; | |
1536 | } else | |
1537 | sectors = rdev->sb_start; | |
1538 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1539 | return -EINVAL; |
dd8ac336 | 1540 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1541 | return ret; |
1da177e4 LT |
1542 | } |
1543 | ||
fd01b88c | 1544 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1545 | { |
65a06f06 | 1546 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1547 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1548 | |
41158c7e | 1549 | rdev->raid_disk = -1; |
c5d79adb N |
1550 | clear_bit(Faulty, &rdev->flags); |
1551 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1552 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1553 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1554 | |
1da177e4 LT |
1555 | if (mddev->raid_disks == 0) { |
1556 | mddev->major_version = 1; | |
1557 | mddev->patch_version = 0; | |
e691063a | 1558 | mddev->external = 0; |
9d8f0363 | 1559 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
9ebc6ef1 DD |
1560 | mddev->ctime = le64_to_cpu(sb->ctime); |
1561 | mddev->utime = le64_to_cpu(sb->utime); | |
1da177e4 | 1562 | mddev->level = le32_to_cpu(sb->level); |
d9d166c2 | 1563 | mddev->clevel[0] = 0; |
1da177e4 LT |
1564 | mddev->layout = le32_to_cpu(sb->layout); |
1565 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1566 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1567 | mddev->events = ev1; |
c3d9714e | 1568 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1569 | mddev->bitmap_info.space = 0; |
1570 | /* Default location for bitmap is 1K after superblock | |
1571 | * using 3K - total of 4K | |
1572 | */ | |
c3d9714e | 1573 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1574 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1575 | mddev->reshape_backwards = 0; |
1576 | ||
1da177e4 LT |
1577 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1578 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1579 | ||
1580 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1581 | |
71c0805c | 1582 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1583 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1584 | mddev->bitmap_info.offset = |
1585 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1586 | /* Metadata doesn't record how much space is available. |
1587 | * For 1.0, we assume we can use up to the superblock | |
1588 | * if before, else to 4K beyond superblock. | |
1589 | * For others, assume no change is possible. | |
1590 | */ | |
1591 | if (mddev->minor_version > 0) | |
1592 | mddev->bitmap_info.space = 0; | |
1593 | else if (mddev->bitmap_info.offset > 0) | |
1594 | mddev->bitmap_info.space = | |
1595 | 8 - mddev->bitmap_info.offset; | |
1596 | else | |
1597 | mddev->bitmap_info.space = | |
1598 | -mddev->bitmap_info.offset; | |
1599 | } | |
e11e93fa | 1600 | |
f6705578 N |
1601 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1602 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1603 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1604 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1605 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1606 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1607 | if (mddev->delta_disks < 0 || |
1608 | (mddev->delta_disks == 0 && | |
1609 | (le32_to_cpu(sb->feature_map) | |
1610 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1611 | mddev->reshape_backwards = 1; | |
f6705578 N |
1612 | } else { |
1613 | mddev->reshape_position = MaxSector; | |
1614 | mddev->delta_disks = 0; | |
1615 | mddev->new_level = mddev->level; | |
1616 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1617 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1618 | } |
1619 | ||
486b0f7b | 1620 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) |
a62ab49e | 1621 | set_bit(MD_HAS_JOURNAL, &mddev->flags); |
41158c7e | 1622 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1623 | /* Insist of good event counter while assembling, except for |
1624 | * spares (which don't need an event count) */ | |
1da177e4 | 1625 | ++ev1; |
be6800a7 N |
1626 | if (rdev->desc_nr >= 0 && |
1627 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
a3dfbdaa SL |
1628 | (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX || |
1629 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)) | |
be6800a7 N |
1630 | if (ev1 < mddev->events) |
1631 | return -EINVAL; | |
41158c7e N |
1632 | } else if (mddev->bitmap) { |
1633 | /* If adding to array with a bitmap, then we can accept an | |
1634 | * older device, but not too old. | |
1635 | */ | |
41158c7e N |
1636 | if (ev1 < mddev->bitmap->events_cleared) |
1637 | return 0; | |
8313b8e5 N |
1638 | if (ev1 < mddev->events) |
1639 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1640 | } else { |
1641 | if (ev1 < mddev->events) | |
1642 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1643 | return 0; | |
1644 | } | |
1da177e4 LT |
1645 | if (mddev->level != LEVEL_MULTIPATH) { |
1646 | int role; | |
3673f305 N |
1647 | if (rdev->desc_nr < 0 || |
1648 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
c4d4c91b | 1649 | role = MD_DISK_ROLE_SPARE; |
3673f305 N |
1650 | rdev->desc_nr = -1; |
1651 | } else | |
1652 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 | 1653 | switch(role) { |
c4d4c91b | 1654 | case MD_DISK_ROLE_SPARE: /* spare */ |
1da177e4 | 1655 | break; |
c4d4c91b | 1656 | case MD_DISK_ROLE_FAULTY: /* faulty */ |
b2d444d7 | 1657 | set_bit(Faulty, &rdev->flags); |
1da177e4 | 1658 | break; |
bac624f3 SL |
1659 | case MD_DISK_ROLE_JOURNAL: /* journal device */ |
1660 | if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) { | |
1661 | /* journal device without journal feature */ | |
9d48739e | 1662 | pr_warn("md: journal device provided without journal feature, ignoring the device\n"); |
bac624f3 SL |
1663 | return -EINVAL; |
1664 | } | |
1665 | set_bit(Journal, &rdev->flags); | |
3069aa8d | 1666 | rdev->journal_tail = le64_to_cpu(sb->journal_tail); |
9b15603d | 1667 | rdev->raid_disk = 0; |
bac624f3 | 1668 | break; |
1da177e4 | 1669 | default: |
f466722c | 1670 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1671 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1672 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1673 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1674 | if (!(le32_to_cpu(sb->feature_map) & |
1675 | MD_FEATURE_RECOVERY_BITMAP)) | |
1676 | rdev->saved_raid_disk = -1; | |
1677 | } else | |
5fd6c1dc | 1678 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1679 | rdev->raid_disk = role; |
1680 | break; | |
1681 | } | |
8ddf9efe N |
1682 | if (sb->devflags & WriteMostly1) |
1683 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1684 | if (sb->devflags & FailFast1) |
1685 | set_bit(FailFast, &rdev->flags); | |
2d78f8c4 N |
1686 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1687 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1688 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1689 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1690 | |
1da177e4 LT |
1691 | return 0; |
1692 | } | |
1693 | ||
fd01b88c | 1694 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1695 | { |
1696 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1697 | struct md_rdev *rdev2; |
1da177e4 LT |
1698 | int max_dev, i; |
1699 | /* make rdev->sb match mddev and rdev data. */ | |
1700 | ||
65a06f06 | 1701 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1702 | |
1703 | sb->feature_map = 0; | |
1704 | sb->pad0 = 0; | |
5fd6c1dc | 1705 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1706 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1707 | ||
1708 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1709 | sb->events = cpu_to_le64(mddev->events); | |
1710 | if (mddev->in_sync) | |
1711 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
bd18f646 SL |
1712 | else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags)) |
1713 | sb->resync_offset = cpu_to_le64(MaxSector); | |
1da177e4 LT |
1714 | else |
1715 | sb->resync_offset = cpu_to_le64(0); | |
1716 | ||
1c05b4bc | 1717 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1718 | |
f0ca340c | 1719 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1720 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1721 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1722 | sb->level = cpu_to_le32(mddev->level); |
1723 | sb->layout = cpu_to_le32(mddev->layout); | |
688834e6 N |
1724 | if (test_bit(FailFast, &rdev->flags)) |
1725 | sb->devflags |= FailFast1; | |
1726 | else | |
1727 | sb->devflags &= ~FailFast1; | |
f0ca340c | 1728 | |
aeb9b211 N |
1729 | if (test_bit(WriteMostly, &rdev->flags)) |
1730 | sb->devflags |= WriteMostly1; | |
1731 | else | |
1732 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1733 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1734 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1735 | |
c3d9714e N |
1736 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1737 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1738 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1739 | } |
5fd6c1dc | 1740 | |
f2076e7d | 1741 | if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) && |
97e4f42d | 1742 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1743 | sb->feature_map |= |
1744 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1745 | sb->recovery_offset = | |
1746 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1747 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1748 | sb->feature_map |= | |
1749 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1750 | } |
3069aa8d SL |
1751 | /* Note: recovery_offset and journal_tail share space */ |
1752 | if (test_bit(Journal, &rdev->flags)) | |
1753 | sb->journal_tail = cpu_to_le64(rdev->journal_tail); | |
2d78f8c4 N |
1754 | if (test_bit(Replacement, &rdev->flags)) |
1755 | sb->feature_map |= | |
1756 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1757 | |
f6705578 N |
1758 | if (mddev->reshape_position != MaxSector) { |
1759 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1760 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1761 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1762 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1763 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1764 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1765 | if (mddev->delta_disks == 0 && |
1766 | mddev->reshape_backwards) | |
1767 | sb->feature_map | |
1768 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1769 | if (rdev->new_data_offset != rdev->data_offset) { |
1770 | sb->feature_map | |
1771 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1772 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1773 | - rdev->data_offset)); | |
1774 | } | |
f6705578 | 1775 | } |
a654b9d8 | 1776 | |
3c462c88 GR |
1777 | if (mddev_is_clustered(mddev)) |
1778 | sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED); | |
1779 | ||
2699b672 N |
1780 | if (rdev->badblocks.count == 0) |
1781 | /* Nothing to do for bad blocks*/ ; | |
1782 | else if (sb->bblog_offset == 0) | |
1783 | /* Cannot record bad blocks on this device */ | |
1784 | md_error(mddev, rdev); | |
1785 | else { | |
1786 | struct badblocks *bb = &rdev->badblocks; | |
1787 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1788 | u64 *p = bb->page; | |
1789 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1790 | if (bb->changed) { | |
1791 | unsigned seq; | |
1792 | ||
1793 | retry: | |
1794 | seq = read_seqbegin(&bb->lock); | |
1795 | ||
1796 | memset(bbp, 0xff, PAGE_SIZE); | |
1797 | ||
1798 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1799 | u64 internal_bb = p[i]; |
2699b672 N |
1800 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1801 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1802 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1803 | } |
d0962936 | 1804 | bb->changed = 0; |
2699b672 N |
1805 | if (read_seqretry(&bb->lock, seq)) |
1806 | goto retry; | |
1807 | ||
1808 | bb->sector = (rdev->sb_start + | |
1809 | (int)le32_to_cpu(sb->bblog_offset)); | |
1810 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1811 | } |
1812 | } | |
1813 | ||
1da177e4 | 1814 | max_dev = 0; |
dafb20fa | 1815 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1816 | if (rdev2->desc_nr+1 > max_dev) |
1817 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1818 | |
70471daf N |
1819 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1820 | int bmask; | |
a778b73f | 1821 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1822 | rdev->sb_size = max_dev * 2 + 256; |
1823 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1824 | if (rdev->sb_size & bmask) | |
1825 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1826 | } else |
1827 | max_dev = le32_to_cpu(sb->max_dev); | |
1828 | ||
1da177e4 | 1829 | for (i=0; i<max_dev;i++) |
c4d4c91b | 1830 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY); |
f72ffdd6 | 1831 | |
a97b7896 SL |
1832 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) |
1833 | sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL); | |
f72ffdd6 | 1834 | |
dafb20fa | 1835 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1836 | i = rdev2->desc_nr; |
b2d444d7 | 1837 | if (test_bit(Faulty, &rdev2->flags)) |
c4d4c91b | 1838 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY); |
b2d444d7 | 1839 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1840 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
a97b7896 | 1841 | else if (test_bit(Journal, &rdev2->flags)) |
bac624f3 | 1842 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL); |
93be75ff | 1843 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1844 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 | 1845 | else |
c4d4c91b | 1846 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE); |
1da177e4 LT |
1847 | } |
1848 | ||
1da177e4 LT |
1849 | sb->sb_csum = calc_sb_1_csum(sb); |
1850 | } | |
1851 | ||
0cd17fec | 1852 | static unsigned long long |
3cb03002 | 1853 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1854 | { |
1855 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1856 | sector_t max_sectors; |
58c0fed4 | 1857 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1858 | return 0; /* component must fit device */ |
c6563a8c N |
1859 | if (rdev->data_offset != rdev->new_data_offset) |
1860 | return 0; /* too confusing */ | |
0f420358 | 1861 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1862 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1863 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1864 | max_sectors -= rdev->data_offset; |
1865 | if (!num_sectors || num_sectors > max_sectors) | |
1866 | num_sectors = max_sectors; | |
c3d9714e | 1867 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1868 | /* minor version 0 with bitmap we can't move */ |
1869 | return 0; | |
1870 | } else { | |
1871 | /* minor version 0; superblock after data */ | |
0f420358 | 1872 | sector_t sb_start; |
77304d2a | 1873 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1874 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1875 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1876 | if (!num_sectors || num_sectors > max_sectors) |
1877 | num_sectors = max_sectors; | |
0f420358 | 1878 | rdev->sb_start = sb_start; |
0cd17fec | 1879 | } |
65a06f06 | 1880 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1881 | sb->data_size = cpu_to_le64(num_sectors); |
3fb632e4 | 1882 | sb->super_offset = cpu_to_le64(rdev->sb_start); |
0cd17fec | 1883 | sb->sb_csum = calc_sb_1_csum(sb); |
46533ff7 N |
1884 | do { |
1885 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
1886 | rdev->sb_page); | |
1887 | } while (md_super_wait(rdev->mddev) < 0); | |
c26a44ed | 1888 | return num_sectors; |
c6563a8c N |
1889 | |
1890 | } | |
1891 | ||
1892 | static int | |
1893 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1894 | unsigned long long new_offset) | |
1895 | { | |
1896 | /* All necessary checks on new >= old have been done */ | |
1897 | struct bitmap *bitmap; | |
1898 | if (new_offset >= rdev->data_offset) | |
1899 | return 1; | |
1900 | ||
1901 | /* with 1.0 metadata, there is no metadata to tread on | |
1902 | * so we can always move back */ | |
1903 | if (rdev->mddev->minor_version == 0) | |
1904 | return 1; | |
1905 | ||
1906 | /* otherwise we must be sure not to step on | |
1907 | * any metadata, so stay: | |
1908 | * 36K beyond start of superblock | |
1909 | * beyond end of badblocks | |
1910 | * beyond write-intent bitmap | |
1911 | */ | |
1912 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1913 | return 0; | |
1914 | bitmap = rdev->mddev->bitmap; | |
1915 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1916 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1917 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1918 | return 0; |
1919 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1920 | return 0; | |
1921 | ||
1922 | return 1; | |
0cd17fec | 1923 | } |
1da177e4 | 1924 | |
75c96f85 | 1925 | static struct super_type super_types[] = { |
1da177e4 LT |
1926 | [0] = { |
1927 | .name = "0.90.0", | |
1928 | .owner = THIS_MODULE, | |
0cd17fec CW |
1929 | .load_super = super_90_load, |
1930 | .validate_super = super_90_validate, | |
1931 | .sync_super = super_90_sync, | |
1932 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1933 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1934 | }, |
1935 | [1] = { | |
1936 | .name = "md-1", | |
1937 | .owner = THIS_MODULE, | |
0cd17fec CW |
1938 | .load_super = super_1_load, |
1939 | .validate_super = super_1_validate, | |
1940 | .sync_super = super_1_sync, | |
1941 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1942 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1943 | }, |
1944 | }; | |
1da177e4 | 1945 | |
fd01b88c | 1946 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1947 | { |
1948 | if (mddev->sync_super) { | |
1949 | mddev->sync_super(mddev, rdev); | |
1950 | return; | |
1951 | } | |
1952 | ||
1953 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1954 | ||
1955 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1956 | } | |
1957 | ||
fd01b88c | 1958 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1959 | { |
3cb03002 | 1960 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1961 | |
4b80991c | 1962 | rcu_read_lock(); |
0b020e85 SL |
1963 | rdev_for_each_rcu(rdev, mddev1) { |
1964 | if (test_bit(Faulty, &rdev->flags) || | |
1965 | test_bit(Journal, &rdev->flags) || | |
1966 | rdev->raid_disk == -1) | |
1967 | continue; | |
1968 | rdev_for_each_rcu(rdev2, mddev2) { | |
1969 | if (test_bit(Faulty, &rdev2->flags) || | |
1970 | test_bit(Journal, &rdev2->flags) || | |
1971 | rdev2->raid_disk == -1) | |
1972 | continue; | |
7dd5e7c3 | 1973 | if (rdev->bdev->bd_contains == |
4b80991c N |
1974 | rdev2->bdev->bd_contains) { |
1975 | rcu_read_unlock(); | |
7dd5e7c3 | 1976 | return 1; |
4b80991c | 1977 | } |
0b020e85 SL |
1978 | } |
1979 | } | |
4b80991c | 1980 | rcu_read_unlock(); |
1da177e4 LT |
1981 | return 0; |
1982 | } | |
1983 | ||
1984 | static LIST_HEAD(pending_raid_disks); | |
1985 | ||
ac5e7113 AN |
1986 | /* |
1987 | * Try to register data integrity profile for an mddev | |
1988 | * | |
1989 | * This is called when an array is started and after a disk has been kicked | |
1990 | * from the array. It only succeeds if all working and active component devices | |
1991 | * are integrity capable with matching profiles. | |
1992 | */ | |
fd01b88c | 1993 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 1994 | { |
3cb03002 | 1995 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
1996 | |
1997 | if (list_empty(&mddev->disks)) | |
1998 | return 0; /* nothing to do */ | |
629acb6a JB |
1999 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
2000 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 2001 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
2002 | /* skip spares and non-functional disks */ |
2003 | if (test_bit(Faulty, &rdev->flags)) | |
2004 | continue; | |
2005 | if (rdev->raid_disk < 0) | |
2006 | continue; | |
ac5e7113 AN |
2007 | if (!reference) { |
2008 | /* Use the first rdev as the reference */ | |
2009 | reference = rdev; | |
2010 | continue; | |
2011 | } | |
2012 | /* does this rdev's profile match the reference profile? */ | |
2013 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
2014 | rdev->bdev->bd_disk) < 0) | |
2015 | return -EINVAL; | |
2016 | } | |
89078d57 MP |
2017 | if (!reference || !bdev_get_integrity(reference->bdev)) |
2018 | return 0; | |
ac5e7113 AN |
2019 | /* |
2020 | * All component devices are integrity capable and have matching | |
2021 | * profiles, register the common profile for the md device. | |
2022 | */ | |
25520d55 MP |
2023 | blk_integrity_register(mddev->gendisk, |
2024 | bdev_get_integrity(reference->bdev)); | |
2025 | ||
9d48739e | 2026 | pr_debug("md: data integrity enabled on %s\n", mdname(mddev)); |
a91a2785 | 2027 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { |
9d48739e | 2028 | pr_err("md: failed to create integrity pool for %s\n", |
a91a2785 MP |
2029 | mdname(mddev)); |
2030 | return -EINVAL; | |
2031 | } | |
ac5e7113 AN |
2032 | return 0; |
2033 | } | |
2034 | EXPORT_SYMBOL(md_integrity_register); | |
2035 | ||
1501efad DW |
2036 | /* |
2037 | * Attempt to add an rdev, but only if it is consistent with the current | |
2038 | * integrity profile | |
2039 | */ | |
2040 | int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) | |
3f9d99c1 | 2041 | { |
2863b9eb JB |
2042 | struct blk_integrity *bi_rdev; |
2043 | struct blk_integrity *bi_mddev; | |
1501efad | 2044 | char name[BDEVNAME_SIZE]; |
2863b9eb JB |
2045 | |
2046 | if (!mddev->gendisk) | |
1501efad | 2047 | return 0; |
2863b9eb JB |
2048 | |
2049 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2050 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2051 | |
ac5e7113 | 2052 | if (!bi_mddev) /* nothing to do */ |
1501efad DW |
2053 | return 0; |
2054 | ||
2055 | if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) { | |
9d48739e N |
2056 | pr_err("%s: incompatible integrity profile for %s\n", |
2057 | mdname(mddev), bdevname(rdev->bdev, name)); | |
1501efad DW |
2058 | return -ENXIO; |
2059 | } | |
2060 | ||
2061 | return 0; | |
3f9d99c1 | 2062 | } |
ac5e7113 | 2063 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2064 | |
f72ffdd6 | 2065 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2066 | { |
7dd5e7c3 | 2067 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2068 | struct kobject *ko; |
5e55e2f5 | 2069 | int err; |
1da177e4 | 2070 | |
11e2ede0 DW |
2071 | /* prevent duplicates */ |
2072 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2073 | return -EEXIST; | |
2074 | ||
dd8ac336 | 2075 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
f6b6ec5c SL |
2076 | if (!test_bit(Journal, &rdev->flags) && |
2077 | rdev->sectors && | |
2078 | (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2079 | if (mddev->pers) { |
2080 | /* Cannot change size, so fail | |
2081 | * If mddev->level <= 0, then we don't care | |
2082 | * about aligning sizes (e.g. linear) | |
2083 | */ | |
2084 | if (mddev->level > 0) | |
2085 | return -ENOSPC; | |
2086 | } else | |
dd8ac336 | 2087 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2088 | } |
1da177e4 LT |
2089 | |
2090 | /* Verify rdev->desc_nr is unique. | |
2091 | * If it is -1, assign a free number, else | |
2092 | * check number is not in use | |
2093 | */ | |
4878e9eb | 2094 | rcu_read_lock(); |
1da177e4 LT |
2095 | if (rdev->desc_nr < 0) { |
2096 | int choice = 0; | |
4878e9eb N |
2097 | if (mddev->pers) |
2098 | choice = mddev->raid_disks; | |
57d051dc | 2099 | while (md_find_rdev_nr_rcu(mddev, choice)) |
1da177e4 LT |
2100 | choice++; |
2101 | rdev->desc_nr = choice; | |
2102 | } else { | |
57d051dc | 2103 | if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
4878e9eb | 2104 | rcu_read_unlock(); |
1da177e4 | 2105 | return -EBUSY; |
4878e9eb | 2106 | } |
1da177e4 | 2107 | } |
4878e9eb | 2108 | rcu_read_unlock(); |
f6b6ec5c SL |
2109 | if (!test_bit(Journal, &rdev->flags) && |
2110 | mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { | |
9d48739e N |
2111 | pr_warn("md: %s: array is limited to %d devices\n", |
2112 | mdname(mddev), mddev->max_disks); | |
de01dfad N |
2113 | return -EBUSY; |
2114 | } | |
19133a42 | 2115 | bdevname(rdev->bdev,b); |
90a9befb | 2116 | strreplace(b, '/', '!'); |
649316b2 | 2117 | |
1da177e4 | 2118 | rdev->mddev = mddev; |
9d48739e | 2119 | pr_debug("md: bind<%s>\n", b); |
86e6ffdd | 2120 | |
b2d6db58 | 2121 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2122 | goto fail; |
86e6ffdd | 2123 | |
0762b8bd | 2124 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2125 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2126 | /* failure here is OK */; | |
2127 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2128 | |
4b80991c | 2129 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2130 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2131 | |
2132 | /* May as well allow recovery to be retried once */ | |
5389042f | 2133 | mddev->recovery_disabled++; |
3f9d99c1 | 2134 | |
1da177e4 | 2135 | return 0; |
5e55e2f5 N |
2136 | |
2137 | fail: | |
9d48739e N |
2138 | pr_warn("md: failed to register dev-%s for %s\n", |
2139 | b, mdname(mddev)); | |
5e55e2f5 | 2140 | return err; |
1da177e4 LT |
2141 | } |
2142 | ||
177a99b2 | 2143 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2144 | { |
3cb03002 | 2145 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2146 | kobject_del(&rdev->kobj); |
177a99b2 | 2147 | kobject_put(&rdev->kobj); |
5792a285 N |
2148 | } |
2149 | ||
f72ffdd6 | 2150 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2151 | { |
2152 | char b[BDEVNAME_SIZE]; | |
403df478 | 2153 | |
49731baa | 2154 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2155 | list_del_rcu(&rdev->same_set); |
9d48739e | 2156 | pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1da177e4 | 2157 | rdev->mddev = NULL; |
86e6ffdd | 2158 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2159 | sysfs_put(rdev->sysfs_state); |
2160 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2161 | rdev->badblocks.count = 0; |
5792a285 | 2162 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2163 | * writing to 'remove' to "dev/state". We also need |
2164 | * to delay it due to rcu usage. | |
5792a285 | 2165 | */ |
4b80991c | 2166 | synchronize_rcu(); |
177a99b2 N |
2167 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2168 | kobject_get(&rdev->kobj); | |
e804ac78 | 2169 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2170 | } |
2171 | ||
2172 | /* | |
2173 | * prevent the device from being mounted, repartitioned or | |
2174 | * otherwise reused by a RAID array (or any other kernel | |
2175 | * subsystem), by bd_claiming the device. | |
2176 | */ | |
3cb03002 | 2177 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2178 | { |
2179 | int err = 0; | |
2180 | struct block_device *bdev; | |
2181 | char b[BDEVNAME_SIZE]; | |
2182 | ||
d4d77629 | 2183 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2184 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 | 2185 | if (IS_ERR(bdev)) { |
9d48739e | 2186 | pr_warn("md: could not open %s.\n", __bdevname(dev, b)); |
1da177e4 LT |
2187 | return PTR_ERR(bdev); |
2188 | } | |
1da177e4 LT |
2189 | rdev->bdev = bdev; |
2190 | return err; | |
2191 | } | |
2192 | ||
3cb03002 | 2193 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2194 | { |
2195 | struct block_device *bdev = rdev->bdev; | |
2196 | rdev->bdev = NULL; | |
e525fd89 | 2197 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2198 | } |
2199 | ||
2200 | void md_autodetect_dev(dev_t dev); | |
2201 | ||
f72ffdd6 | 2202 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2203 | { |
2204 | char b[BDEVNAME_SIZE]; | |
403df478 | 2205 | |
9d48739e | 2206 | pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b)); |
545c8795 | 2207 | md_rdev_clear(rdev); |
1da177e4 | 2208 | #ifndef MODULE |
d0fae18f N |
2209 | if (test_bit(AutoDetected, &rdev->flags)) |
2210 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2211 | #endif |
2212 | unlock_rdev(rdev); | |
86e6ffdd | 2213 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2214 | } |
2215 | ||
fb56dfef | 2216 | void md_kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2217 | { |
2218 | unbind_rdev_from_array(rdev); | |
2219 | export_rdev(rdev); | |
2220 | } | |
fb56dfef | 2221 | EXPORT_SYMBOL_GPL(md_kick_rdev_from_array); |
1da177e4 | 2222 | |
fd01b88c | 2223 | static void export_array(struct mddev *mddev) |
1da177e4 | 2224 | { |
0638bb0e | 2225 | struct md_rdev *rdev; |
1da177e4 | 2226 | |
0638bb0e N |
2227 | while (!list_empty(&mddev->disks)) { |
2228 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2229 | same_set); | |
fb56dfef | 2230 | md_kick_rdev_from_array(rdev); |
1da177e4 | 2231 | } |
1da177e4 LT |
2232 | mddev->raid_disks = 0; |
2233 | mddev->major_version = 0; | |
2234 | } | |
2235 | ||
f72ffdd6 | 2236 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2237 | { |
42543769 N |
2238 | /* Update each superblock (in-memory image), but |
2239 | * if we are allowed to, skip spares which already | |
2240 | * have the right event counter, or have one earlier | |
2241 | * (which would mean they aren't being marked as dirty | |
2242 | * with the rest of the array) | |
2243 | */ | |
3cb03002 | 2244 | struct md_rdev *rdev; |
dafb20fa | 2245 | rdev_for_each(rdev, mddev) { |
42543769 N |
2246 | if (rdev->sb_events == mddev->events || |
2247 | (nospares && | |
2248 | rdev->raid_disk < 0 && | |
42543769 N |
2249 | rdev->sb_events+1 == mddev->events)) { |
2250 | /* Don't update this superblock */ | |
2251 | rdev->sb_loaded = 2; | |
2252 | } else { | |
076f968b | 2253 | sync_super(mddev, rdev); |
42543769 N |
2254 | rdev->sb_loaded = 1; |
2255 | } | |
1da177e4 LT |
2256 | } |
2257 | } | |
2258 | ||
2aa82191 GR |
2259 | static bool does_sb_need_changing(struct mddev *mddev) |
2260 | { | |
2261 | struct md_rdev *rdev; | |
2262 | struct mdp_superblock_1 *sb; | |
2263 | int role; | |
2264 | ||
2265 | /* Find a good rdev */ | |
2266 | rdev_for_each(rdev, mddev) | |
2267 | if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags)) | |
2268 | break; | |
2269 | ||
2270 | /* No good device found. */ | |
2271 | if (!rdev) | |
2272 | return false; | |
2273 | ||
2274 | sb = page_address(rdev->sb_page); | |
2275 | /* Check if a device has become faulty or a spare become active */ | |
2276 | rdev_for_each(rdev, mddev) { | |
2277 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
2278 | /* Device activated? */ | |
2279 | if (role == 0xffff && rdev->raid_disk >=0 && | |
2280 | !test_bit(Faulty, &rdev->flags)) | |
2281 | return true; | |
2282 | /* Device turned faulty? */ | |
2283 | if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd)) | |
2284 | return true; | |
2285 | } | |
2286 | ||
2287 | /* Check if any mddev parameters have changed */ | |
2288 | if ((mddev->dev_sectors != le64_to_cpu(sb->size)) || | |
2289 | (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) || | |
13459213 | 2290 | (mddev->layout != le32_to_cpu(sb->layout)) || |
2aa82191 GR |
2291 | (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) || |
2292 | (mddev->chunk_sectors != le32_to_cpu(sb->chunksize))) | |
2293 | return true; | |
2294 | ||
2295 | return false; | |
2296 | } | |
2297 | ||
1aee41f6 | 2298 | void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2299 | { |
3cb03002 | 2300 | struct md_rdev *rdev; |
06d91a5f | 2301 | int sync_req; |
42543769 | 2302 | int nospares = 0; |
2699b672 | 2303 | int any_badblocks_changed = 0; |
23b63f9f | 2304 | int ret = -1; |
1da177e4 | 2305 | |
d87f064f N |
2306 | if (mddev->ro) { |
2307 | if (force_change) | |
2953079c | 2308 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
d87f064f N |
2309 | return; |
2310 | } | |
2aa82191 | 2311 | |
2c97cf13 | 2312 | repeat: |
2aa82191 | 2313 | if (mddev_is_clustered(mddev)) { |
2953079c | 2314 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
2aa82191 | 2315 | force_change = 1; |
2953079c | 2316 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
85ad1d13 | 2317 | nospares = 1; |
23b63f9f | 2318 | ret = md_cluster_ops->metadata_update_start(mddev); |
2aa82191 GR |
2319 | /* Has someone else has updated the sb */ |
2320 | if (!does_sb_need_changing(mddev)) { | |
23b63f9f GJ |
2321 | if (ret == 0) |
2322 | md_cluster_ops->metadata_update_cancel(mddev); | |
2953079c SL |
2323 | bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2324 | BIT(MD_SB_CHANGE_DEVS) | | |
2325 | BIT(MD_SB_CHANGE_CLEAN)); | |
2aa82191 GR |
2326 | return; |
2327 | } | |
2328 | } | |
2c97cf13 | 2329 | |
3a3a5ddb | 2330 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2331 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2332 | if (rdev->raid_disk >= 0 && |
2333 | mddev->delta_disks >= 0 && | |
f2076e7d | 2334 | !test_bit(Journal, &rdev->flags) && |
3a3a5ddb N |
2335 | !test_bit(In_sync, &rdev->flags) && |
2336 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2337 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2338 | ||
f72ffdd6 | 2339 | } |
bd52b746 | 2340 | if (!mddev->persistent) { |
2953079c SL |
2341 | clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2342 | clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
de393cde | 2343 | if (!mddev->external) { |
2953079c | 2344 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
dafb20fa | 2345 | rdev_for_each(rdev, mddev) { |
de393cde | 2346 | if (rdev->badblocks.changed) { |
d0962936 | 2347 | rdev->badblocks.changed = 0; |
fc974ee2 | 2348 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2349 | md_error(mddev, rdev); |
2350 | } | |
2351 | clear_bit(Blocked, &rdev->flags); | |
2352 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2353 | wake_up(&rdev->blocked_wait); | |
2354 | } | |
2355 | } | |
3a3a5ddb N |
2356 | wake_up(&mddev->sb_wait); |
2357 | return; | |
2358 | } | |
2359 | ||
85572d7c | 2360 | spin_lock(&mddev->lock); |
84692195 | 2361 | |
9ebc6ef1 | 2362 | mddev->utime = ktime_get_real_seconds(); |
3a3a5ddb | 2363 | |
2953079c | 2364 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
850b2b42 | 2365 | force_change = 1; |
2953079c | 2366 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
850b2b42 N |
2367 | /* just a clean<-> dirty transition, possibly leave spares alone, |
2368 | * though if events isn't the right even/odd, we will have to do | |
2369 | * spares after all | |
2370 | */ | |
2371 | nospares = 1; | |
2372 | if (force_change) | |
2373 | nospares = 0; | |
2374 | if (mddev->degraded) | |
84692195 N |
2375 | /* If the array is degraded, then skipping spares is both |
2376 | * dangerous and fairly pointless. | |
2377 | * Dangerous because a device that was removed from the array | |
2378 | * might have a event_count that still looks up-to-date, | |
2379 | * so it can be re-added without a resync. | |
2380 | * Pointless because if there are any spares to skip, | |
2381 | * then a recovery will happen and soon that array won't | |
2382 | * be degraded any more and the spare can go back to sleep then. | |
2383 | */ | |
850b2b42 | 2384 | nospares = 0; |
84692195 | 2385 | |
06d91a5f | 2386 | sync_req = mddev->in_sync; |
42543769 N |
2387 | |
2388 | /* If this is just a dirty<->clean transition, and the array is clean | |
2389 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2390 | if (nospares |
42543769 | 2391 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2392 | && mddev->can_decrease_events |
2393 | && mddev->events != 1) { | |
42543769 | 2394 | mddev->events--; |
a8707c08 N |
2395 | mddev->can_decrease_events = 0; |
2396 | } else { | |
42543769 N |
2397 | /* otherwise we have to go forward and ... */ |
2398 | mddev->events ++; | |
a8707c08 | 2399 | mddev->can_decrease_events = nospares; |
42543769 | 2400 | } |
1da177e4 | 2401 | |
403df478 N |
2402 | /* |
2403 | * This 64-bit counter should never wrap. | |
2404 | * Either we are in around ~1 trillion A.C., assuming | |
2405 | * 1 reboot per second, or we have a bug... | |
2406 | */ | |
2407 | WARN_ON(mddev->events == 0); | |
2699b672 | 2408 | |
dafb20fa | 2409 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2410 | if (rdev->badblocks.changed) |
2411 | any_badblocks_changed++; | |
de393cde N |
2412 | if (test_bit(Faulty, &rdev->flags)) |
2413 | set_bit(FaultRecorded, &rdev->flags); | |
2414 | } | |
2699b672 | 2415 | |
e691063a | 2416 | sync_sbs(mddev, nospares); |
85572d7c | 2417 | spin_unlock(&mddev->lock); |
1da177e4 | 2418 | |
36a4e1fe N |
2419 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2420 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2421 | |
504634f6 SL |
2422 | if (mddev->queue) |
2423 | blk_add_trace_msg(mddev->queue, "md md_update_sb"); | |
46533ff7 | 2424 | rewrite: |
4ad13663 | 2425 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2426 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2427 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2428 | |
42543769 N |
2429 | if (rdev->sb_loaded != 1) |
2430 | continue; /* no noise on spare devices */ | |
1da177e4 | 2431 | |
f466722c | 2432 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2433 | md_super_write(mddev,rdev, |
0f420358 | 2434 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2435 | rdev->sb_page); |
36a4e1fe N |
2436 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2437 | bdevname(rdev->bdev, b), | |
2438 | (unsigned long long)rdev->sb_start); | |
42543769 | 2439 | rdev->sb_events = mddev->events; |
2699b672 N |
2440 | if (rdev->badblocks.size) { |
2441 | md_super_write(mddev, rdev, | |
2442 | rdev->badblocks.sector, | |
2443 | rdev->badblocks.size << 9, | |
2444 | rdev->bb_page); | |
2445 | rdev->badblocks.size = 0; | |
2446 | } | |
7bfa19f2 | 2447 | |
f466722c | 2448 | } else |
36a4e1fe N |
2449 | pr_debug("md: %s (skipping faulty)\n", |
2450 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2451 | |
7bfa19f2 | 2452 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2453 | /* only need to write one superblock... */ |
2454 | break; | |
2455 | } | |
46533ff7 N |
2456 | if (md_super_wait(mddev) < 0) |
2457 | goto rewrite; | |
2953079c | 2458 | /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2459 | |
2c97cf13 GJ |
2460 | if (mddev_is_clustered(mddev) && ret == 0) |
2461 | md_cluster_ops->metadata_update_finish(mddev); | |
2462 | ||
850b2b42 | 2463 | if (mddev->in_sync != sync_req || |
2953079c SL |
2464 | !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2465 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN))) | |
06d91a5f | 2466 | /* have to write it out again */ |
06d91a5f | 2467 | goto repeat; |
3d310eb7 | 2468 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2469 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2470 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2471 | |
dafb20fa | 2472 | rdev_for_each(rdev, mddev) { |
de393cde N |
2473 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2474 | clear_bit(Blocked, &rdev->flags); | |
2475 | ||
2476 | if (any_badblocks_changed) | |
fc974ee2 | 2477 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2478 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2479 | wake_up(&rdev->blocked_wait); | |
2480 | } | |
1da177e4 | 2481 | } |
1aee41f6 | 2482 | EXPORT_SYMBOL(md_update_sb); |
1da177e4 | 2483 | |
a6da4ef8 GR |
2484 | static int add_bound_rdev(struct md_rdev *rdev) |
2485 | { | |
2486 | struct mddev *mddev = rdev->mddev; | |
2487 | int err = 0; | |
87d4d916 | 2488 | bool add_journal = test_bit(Journal, &rdev->flags); |
a6da4ef8 | 2489 | |
87d4d916 | 2490 | if (!mddev->pers->hot_remove_disk || add_journal) { |
a6da4ef8 GR |
2491 | /* If there is hot_add_disk but no hot_remove_disk |
2492 | * then added disks for geometry changes, | |
2493 | * and should be added immediately. | |
2494 | */ | |
2495 | super_types[mddev->major_version]. | |
2496 | validate_super(mddev, rdev); | |
87d4d916 SL |
2497 | if (add_journal) |
2498 | mddev_suspend(mddev); | |
a6da4ef8 | 2499 | err = mddev->pers->hot_add_disk(mddev, rdev); |
87d4d916 SL |
2500 | if (add_journal) |
2501 | mddev_resume(mddev); | |
a6da4ef8 | 2502 | if (err) { |
db767672 | 2503 | md_kick_rdev_from_array(rdev); |
a6da4ef8 GR |
2504 | return err; |
2505 | } | |
2506 | } | |
2507 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2508 | ||
2953079c | 2509 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a6da4ef8 GR |
2510 | if (mddev->degraded) |
2511 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
2512 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2513 | md_new_event(mddev); | |
2514 | md_wakeup_thread(mddev->thread); | |
2515 | return 0; | |
2516 | } | |
1da177e4 | 2517 | |
7f6ce769 | 2518 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2519 | * We want to accept with case. For this we use cmd_match. |
2520 | */ | |
2521 | static int cmd_match(const char *cmd, const char *str) | |
2522 | { | |
2523 | /* See if cmd, written into a sysfs file, matches | |
2524 | * str. They must either be the same, or cmd can | |
2525 | * have a trailing newline | |
2526 | */ | |
2527 | while (*cmd && *str && *cmd == *str) { | |
2528 | cmd++; | |
2529 | str++; | |
2530 | } | |
2531 | if (*cmd == '\n') | |
2532 | cmd++; | |
2533 | if (*str || *cmd) | |
2534 | return 0; | |
2535 | return 1; | |
2536 | } | |
2537 | ||
86e6ffdd N |
2538 | struct rdev_sysfs_entry { |
2539 | struct attribute attr; | |
3cb03002 N |
2540 | ssize_t (*show)(struct md_rdev *, char *); |
2541 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2542 | }; |
2543 | ||
2544 | static ssize_t | |
3cb03002 | 2545 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd | 2546 | { |
35b785f7 | 2547 | char *sep = ","; |
20a49ff6 | 2548 | size_t len = 0; |
758bfc8a | 2549 | unsigned long flags = ACCESS_ONCE(rdev->flags); |
86e6ffdd | 2550 | |
758bfc8a | 2551 | if (test_bit(Faulty, &flags) || |
dcbcb486 TM |
2552 | (!test_bit(ExternalBbl, &flags) && |
2553 | rdev->badblocks.unacked_exist)) | |
35b785f7 TM |
2554 | len += sprintf(page+len, "faulty%s", sep); |
2555 | if (test_bit(In_sync, &flags)) | |
2556 | len += sprintf(page+len, "in_sync%s", sep); | |
2557 | if (test_bit(Journal, &flags)) | |
2558 | len += sprintf(page+len, "journal%s", sep); | |
2559 | if (test_bit(WriteMostly, &flags)) | |
2560 | len += sprintf(page+len, "write_mostly%s", sep); | |
758bfc8a | 2561 | if (test_bit(Blocked, &flags) || |
52c64152 | 2562 | (rdev->badblocks.unacked_exist |
35b785f7 TM |
2563 | && !test_bit(Faulty, &flags))) |
2564 | len += sprintf(page+len, "blocked%s", sep); | |
758bfc8a | 2565 | if (!test_bit(Faulty, &flags) && |
f2076e7d | 2566 | !test_bit(Journal, &flags) && |
35b785f7 TM |
2567 | !test_bit(In_sync, &flags)) |
2568 | len += sprintf(page+len, "spare%s", sep); | |
2569 | if (test_bit(WriteErrorSeen, &flags)) | |
2570 | len += sprintf(page+len, "write_error%s", sep); | |
2571 | if (test_bit(WantReplacement, &flags)) | |
2572 | len += sprintf(page+len, "want_replacement%s", sep); | |
2573 | if (test_bit(Replacement, &flags)) | |
2574 | len += sprintf(page+len, "replacement%s", sep); | |
2575 | if (test_bit(ExternalBbl, &flags)) | |
2576 | len += sprintf(page+len, "external_bbl%s", sep); | |
688834e6 N |
2577 | if (test_bit(FailFast, &flags)) |
2578 | len += sprintf(page+len, "failfast%s", sep); | |
35b785f7 TM |
2579 | |
2580 | if (len) | |
2581 | len -= strlen(sep); | |
2d78f8c4 | 2582 | |
86e6ffdd N |
2583 | return len+sprintf(page+len, "\n"); |
2584 | } | |
2585 | ||
45dc2de1 | 2586 | static ssize_t |
3cb03002 | 2587 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2588 | { |
2589 | /* can write | |
de393cde | 2590 | * faulty - simulates an error |
45dc2de1 | 2591 | * remove - disconnects the device |
f655675b N |
2592 | * writemostly - sets write_mostly |
2593 | * -writemostly - clears write_mostly | |
de393cde N |
2594 | * blocked - sets the Blocked flags |
2595 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2596 | * insync - sets Insync providing device isn't active |
f466722c N |
2597 | * -insync - clear Insync for a device with a slot assigned, |
2598 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2599 | * write_error - sets WriteErrorSeen |
2600 | * -write_error - clears WriteErrorSeen | |
688834e6 | 2601 | * {,-}failfast - set/clear FailFast |
45dc2de1 N |
2602 | */ |
2603 | int err = -EINVAL; | |
2604 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2605 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2606 | if (test_bit(Faulty, &rdev->flags)) |
2607 | err = 0; | |
2608 | else | |
2609 | err = -EBUSY; | |
45dc2de1 | 2610 | } else if (cmd_match(buf, "remove")) { |
5d881783 SL |
2611 | if (rdev->mddev->pers) { |
2612 | clear_bit(Blocked, &rdev->flags); | |
2613 | remove_and_add_spares(rdev->mddev, rdev); | |
2614 | } | |
45dc2de1 N |
2615 | if (rdev->raid_disk >= 0) |
2616 | err = -EBUSY; | |
2617 | else { | |
fd01b88c | 2618 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2619 | err = 0; |
a9720903 GJ |
2620 | if (mddev_is_clustered(mddev)) |
2621 | err = md_cluster_ops->remove_disk(mddev, rdev); | |
2622 | ||
2623 | if (err == 0) { | |
2624 | md_kick_rdev_from_array(rdev); | |
060b0689 | 2625 | if (mddev->pers) { |
2953079c | 2626 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
2627 | md_wakeup_thread(mddev->thread); |
2628 | } | |
a9720903 GJ |
2629 | md_new_event(mddev); |
2630 | } | |
45dc2de1 | 2631 | } |
f655675b N |
2632 | } else if (cmd_match(buf, "writemostly")) { |
2633 | set_bit(WriteMostly, &rdev->flags); | |
2634 | err = 0; | |
2635 | } else if (cmd_match(buf, "-writemostly")) { | |
2636 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2637 | err = 0; |
2638 | } else if (cmd_match(buf, "blocked")) { | |
2639 | set_bit(Blocked, &rdev->flags); | |
2640 | err = 0; | |
2641 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2642 | if (!test_bit(Faulty, &rdev->flags) && |
dcbcb486 | 2643 | !test_bit(ExternalBbl, &rdev->flags) && |
7da64a0a | 2644 | rdev->badblocks.unacked_exist) { |
de393cde N |
2645 | /* metadata handler doesn't understand badblocks, |
2646 | * so we need to fail the device | |
2647 | */ | |
2648 | md_error(rdev->mddev, rdev); | |
2649 | } | |
6bfe0b49 | 2650 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2651 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2652 | wake_up(&rdev->blocked_wait); |
2653 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2654 | md_wakeup_thread(rdev->mddev->thread); | |
2655 | ||
6d56e278 N |
2656 | err = 0; |
2657 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2658 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2659 | err = 0; |
688834e6 N |
2660 | } else if (cmd_match(buf, "failfast")) { |
2661 | set_bit(FailFast, &rdev->flags); | |
2662 | err = 0; | |
2663 | } else if (cmd_match(buf, "-failfast")) { | |
2664 | clear_bit(FailFast, &rdev->flags); | |
2665 | err = 0; | |
f2076e7d SL |
2666 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 && |
2667 | !test_bit(Journal, &rdev->flags)) { | |
e1960f8c N |
2668 | if (rdev->mddev->pers == NULL) { |
2669 | clear_bit(In_sync, &rdev->flags); | |
2670 | rdev->saved_raid_disk = rdev->raid_disk; | |
2671 | rdev->raid_disk = -1; | |
2672 | err = 0; | |
2673 | } | |
d7a9d443 N |
2674 | } else if (cmd_match(buf, "write_error")) { |
2675 | set_bit(WriteErrorSeen, &rdev->flags); | |
2676 | err = 0; | |
2677 | } else if (cmd_match(buf, "-write_error")) { | |
2678 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2679 | err = 0; | |
2d78f8c4 N |
2680 | } else if (cmd_match(buf, "want_replacement")) { |
2681 | /* Any non-spare device that is not a replacement can | |
2682 | * become want_replacement at any time, but we then need to | |
2683 | * check if recovery is needed. | |
2684 | */ | |
2685 | if (rdev->raid_disk >= 0 && | |
f2076e7d | 2686 | !test_bit(Journal, &rdev->flags) && |
2d78f8c4 N |
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 | } | |
a6da4ef8 GR |
2717 | } else if (cmd_match(buf, "re-add")) { |
2718 | if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) { | |
97f6cd39 GR |
2719 | /* clear_bit is performed _after_ all the devices |
2720 | * have their local Faulty bit cleared. If any writes | |
2721 | * happen in the meantime in the local node, they | |
2722 | * will land in the local bitmap, which will be synced | |
2723 | * by this node eventually | |
2724 | */ | |
2725 | if (!mddev_is_clustered(rdev->mddev) || | |
2726 | (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) { | |
2727 | clear_bit(Faulty, &rdev->flags); | |
2728 | err = add_bound_rdev(rdev); | |
2729 | } | |
a6da4ef8 GR |
2730 | } else |
2731 | err = -EBUSY; | |
35b785f7 TM |
2732 | } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) { |
2733 | set_bit(ExternalBbl, &rdev->flags); | |
2734 | rdev->badblocks.shift = 0; | |
2735 | err = 0; | |
2736 | } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) { | |
2737 | clear_bit(ExternalBbl, &rdev->flags); | |
2738 | err = 0; | |
45dc2de1 | 2739 | } |
00bcb4ac N |
2740 | if (!err) |
2741 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2742 | return err ? err : len; |
2743 | } | |
80ca3a44 | 2744 | static struct rdev_sysfs_entry rdev_state = |
750f199e | 2745 | __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store); |
86e6ffdd | 2746 | |
4dbcdc75 | 2747 | static ssize_t |
3cb03002 | 2748 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2749 | { |
2750 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2751 | } | |
2752 | ||
2753 | static ssize_t | |
3cb03002 | 2754 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 | 2755 | { |
4c9309c0 AD |
2756 | unsigned int n; |
2757 | int rv; | |
2758 | ||
2759 | rv = kstrtouint(buf, 10, &n); | |
2760 | if (rv < 0) | |
2761 | return rv; | |
2762 | atomic_set(&rdev->corrected_errors, n); | |
2763 | return len; | |
4dbcdc75 N |
2764 | } |
2765 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2766 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2767 | |
014236d2 | 2768 | static ssize_t |
3cb03002 | 2769 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 | 2770 | { |
f2076e7d SL |
2771 | if (test_bit(Journal, &rdev->flags)) |
2772 | return sprintf(page, "journal\n"); | |
2773 | else if (rdev->raid_disk < 0) | |
014236d2 N |
2774 | return sprintf(page, "none\n"); |
2775 | else | |
2776 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2777 | } | |
2778 | ||
2779 | static ssize_t | |
3cb03002 | 2780 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 | 2781 | { |
4c9309c0 | 2782 | int slot; |
c303da6d | 2783 | int err; |
4c9309c0 | 2784 | |
f2076e7d SL |
2785 | if (test_bit(Journal, &rdev->flags)) |
2786 | return -EBUSY; | |
014236d2 N |
2787 | if (strncmp(buf, "none", 4)==0) |
2788 | slot = -1; | |
4c9309c0 AD |
2789 | else { |
2790 | err = kstrtouint(buf, 10, (unsigned int *)&slot); | |
2791 | if (err < 0) | |
2792 | return err; | |
2793 | } | |
6c2fce2e | 2794 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2795 | /* Setting 'slot' on an active array requires also |
2796 | * updating the 'rd%d' link, and communicating | |
2797 | * with the personality with ->hot_*_disk. | |
2798 | * For now we only support removing | |
2799 | * failed/spare devices. This normally happens automatically, | |
2800 | * but not when the metadata is externally managed. | |
2801 | */ | |
c303da6d N |
2802 | if (rdev->raid_disk == -1) |
2803 | return -EEXIST; | |
2804 | /* personality does all needed checks */ | |
01393f3d | 2805 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2806 | return -EINVAL; |
746d3207 N |
2807 | clear_bit(Blocked, &rdev->flags); |
2808 | remove_and_add_spares(rdev->mddev, rdev); | |
2809 | if (rdev->raid_disk >= 0) | |
2810 | return -EBUSY; | |
c303da6d N |
2811 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2812 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2813 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2814 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2815 | * if we ever get bitmaps working here. |
6c2fce2e | 2816 | */ |
cb01c549 | 2817 | int err; |
6c2fce2e NB |
2818 | |
2819 | if (rdev->raid_disk != -1) | |
2820 | return -EBUSY; | |
2821 | ||
c6751b2b N |
2822 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2823 | return -EBUSY; | |
2824 | ||
6c2fce2e NB |
2825 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2826 | return -EINVAL; | |
2827 | ||
ba1b41b6 N |
2828 | if (slot >= rdev->mddev->raid_disks && |
2829 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2830 | return -ENOSPC; | |
2831 | ||
6c2fce2e NB |
2832 | rdev->raid_disk = slot; |
2833 | if (test_bit(In_sync, &rdev->flags)) | |
2834 | rdev->saved_raid_disk = slot; | |
2835 | else | |
2836 | rdev->saved_raid_disk = -1; | |
d30519fc | 2837 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2838 | clear_bit(Bitmap_sync, &rdev->flags); |
cb01c549 GR |
2839 | err = rdev->mddev->pers-> |
2840 | hot_add_disk(rdev->mddev, rdev); | |
2841 | if (err) { | |
2842 | rdev->raid_disk = -1; | |
2843 | return err; | |
2844 | } else | |
2845 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2846 | if (sysfs_link_rdev(rdev->mddev, rdev)) | |
2847 | /* failure here is OK */; | |
6c2fce2e | 2848 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2849 | } else { |
ba1b41b6 N |
2850 | if (slot >= rdev->mddev->raid_disks && |
2851 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2852 | return -ENOSPC; |
2853 | rdev->raid_disk = slot; | |
2854 | /* assume it is working */ | |
c5d79adb N |
2855 | clear_bit(Faulty, &rdev->flags); |
2856 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2857 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2858 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2859 | } |
014236d2 N |
2860 | return len; |
2861 | } | |
2862 | ||
014236d2 | 2863 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2864 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2865 | |
93c8cad0 | 2866 | static ssize_t |
3cb03002 | 2867 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2868 | { |
6961ece4 | 2869 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2870 | } |
2871 | ||
2872 | static ssize_t | |
3cb03002 | 2873 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2874 | { |
c6563a8c | 2875 | unsigned long long offset; |
b29bebd6 | 2876 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2877 | return -EINVAL; |
8ed0a521 | 2878 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2879 | return -EBUSY; |
dd8ac336 | 2880 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2881 | /* Must set offset before size, so overlap checks |
2882 | * can be sane */ | |
2883 | return -EBUSY; | |
93c8cad0 | 2884 | rdev->data_offset = offset; |
25f7fd47 | 2885 | rdev->new_data_offset = offset; |
93c8cad0 N |
2886 | return len; |
2887 | } | |
2888 | ||
2889 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2890 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2891 | |
c6563a8c N |
2892 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2893 | { | |
2894 | return sprintf(page, "%llu\n", | |
2895 | (unsigned long long)rdev->new_data_offset); | |
2896 | } | |
2897 | ||
2898 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2899 | const char *buf, size_t len) | |
2900 | { | |
2901 | unsigned long long new_offset; | |
2902 | struct mddev *mddev = rdev->mddev; | |
2903 | ||
b29bebd6 | 2904 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2905 | return -EINVAL; |
2906 | ||
f851b60d N |
2907 | if (mddev->sync_thread || |
2908 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
2909 | return -EBUSY; |
2910 | if (new_offset == rdev->data_offset) | |
2911 | /* reset is always permitted */ | |
2912 | ; | |
2913 | else if (new_offset > rdev->data_offset) { | |
2914 | /* must not push array size beyond rdev_sectors */ | |
2915 | if (new_offset - rdev->data_offset | |
2916 | + mddev->dev_sectors > rdev->sectors) | |
2917 | return -E2BIG; | |
2918 | } | |
2919 | /* Metadata worries about other space details. */ | |
2920 | ||
2921 | /* decreasing the offset is inconsistent with a backwards | |
2922 | * reshape. | |
2923 | */ | |
2924 | if (new_offset < rdev->data_offset && | |
2925 | mddev->reshape_backwards) | |
2926 | return -EINVAL; | |
2927 | /* Increasing offset is inconsistent with forwards | |
2928 | * reshape. reshape_direction should be set to | |
2929 | * 'backwards' first. | |
2930 | */ | |
2931 | if (new_offset > rdev->data_offset && | |
2932 | !mddev->reshape_backwards) | |
2933 | return -EINVAL; | |
2934 | ||
2935 | if (mddev->pers && mddev->persistent && | |
2936 | !super_types[mddev->major_version] | |
2937 | .allow_new_offset(rdev, new_offset)) | |
2938 | return -E2BIG; | |
2939 | rdev->new_data_offset = new_offset; | |
2940 | if (new_offset > rdev->data_offset) | |
2941 | mddev->reshape_backwards = 1; | |
2942 | else if (new_offset < rdev->data_offset) | |
2943 | mddev->reshape_backwards = 0; | |
2944 | ||
2945 | return len; | |
2946 | } | |
2947 | static struct rdev_sysfs_entry rdev_new_offset = | |
2948 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2949 | ||
83303b61 | 2950 | static ssize_t |
3cb03002 | 2951 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2952 | { |
dd8ac336 | 2953 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2954 | } |
2955 | ||
c5d79adb N |
2956 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2957 | { | |
2958 | /* check if two start/length pairs overlap */ | |
2959 | if (s1+l1 <= s2) | |
2960 | return 0; | |
2961 | if (s2+l2 <= s1) | |
2962 | return 0; | |
2963 | return 1; | |
2964 | } | |
2965 | ||
b522adcd DW |
2966 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2967 | { | |
2968 | unsigned long long blocks; | |
2969 | sector_t new; | |
2970 | ||
b29bebd6 | 2971 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
2972 | return -EINVAL; |
2973 | ||
2974 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2975 | return -EINVAL; /* sector conversion overflow */ | |
2976 | ||
2977 | new = blocks * 2; | |
2978 | if (new != blocks * 2) | |
2979 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2980 | ||
2981 | *sectors = new; | |
2982 | return 0; | |
2983 | } | |
2984 | ||
83303b61 | 2985 | static ssize_t |
3cb03002 | 2986 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2987 | { |
fd01b88c | 2988 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2989 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2990 | sector_t sectors; |
27c529bb | 2991 | |
f2076e7d SL |
2992 | if (test_bit(Journal, &rdev->flags)) |
2993 | return -EBUSY; | |
b522adcd | 2994 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2995 | return -EINVAL; |
c6563a8c N |
2996 | if (rdev->data_offset != rdev->new_data_offset) |
2997 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2998 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2999 | if (my_mddev->persistent) { |
dd8ac336 AN |
3000 | sectors = super_types[my_mddev->major_version]. |
3001 | rdev_size_change(rdev, sectors); | |
3002 | if (!sectors) | |
0cd17fec | 3003 | return -EBUSY; |
dd8ac336 | 3004 | } else if (!sectors) |
77304d2a | 3005 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 3006 | rdev->data_offset; |
a6468539 N |
3007 | if (!my_mddev->pers->resize) |
3008 | /* Cannot change size for RAID0 or Linear etc */ | |
3009 | return -EINVAL; | |
0cd17fec | 3010 | } |
dd8ac336 | 3011 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 3012 | return -EINVAL; /* component must fit device */ |
0cd17fec | 3013 | |
dd8ac336 AN |
3014 | rdev->sectors = sectors; |
3015 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
3016 | /* Need to check that all other rdevs with the same |
3017 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
3018 | * the rdev lists safely. | |
3019 | * This check does not provide a hard guarantee, it | |
3020 | * just helps avoid dangerous mistakes. | |
c5d79adb | 3021 | */ |
fd01b88c | 3022 | struct mddev *mddev; |
c5d79adb | 3023 | int overlap = 0; |
159ec1fc | 3024 | struct list_head *tmp; |
c5d79adb | 3025 | |
8b1afc3d | 3026 | rcu_read_lock(); |
29ac4aa3 | 3027 | for_each_mddev(mddev, tmp) { |
3cb03002 | 3028 | struct md_rdev *rdev2; |
c5d79adb | 3029 | |
dafb20fa | 3030 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
3031 | if (rdev->bdev == rdev2->bdev && |
3032 | rdev != rdev2 && | |
3033 | overlaps(rdev->data_offset, rdev->sectors, | |
3034 | rdev2->data_offset, | |
3035 | rdev2->sectors)) { | |
c5d79adb N |
3036 | overlap = 1; |
3037 | break; | |
3038 | } | |
c5d79adb N |
3039 | if (overlap) { |
3040 | mddev_put(mddev); | |
3041 | break; | |
3042 | } | |
3043 | } | |
8b1afc3d | 3044 | rcu_read_unlock(); |
c5d79adb N |
3045 | if (overlap) { |
3046 | /* Someone else could have slipped in a size | |
3047 | * change here, but doing so is just silly. | |
dd8ac336 | 3048 | * We put oldsectors back because we *know* it is |
c5d79adb N |
3049 | * safe, and trust userspace not to race with |
3050 | * itself | |
3051 | */ | |
dd8ac336 | 3052 | rdev->sectors = oldsectors; |
c5d79adb N |
3053 | return -EBUSY; |
3054 | } | |
3055 | } | |
83303b61 N |
3056 | return len; |
3057 | } | |
3058 | ||
3059 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 3060 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 3061 | |
3cb03002 | 3062 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
3063 | { |
3064 | unsigned long long recovery_start = rdev->recovery_offset; | |
3065 | ||
3066 | if (test_bit(In_sync, &rdev->flags) || | |
3067 | recovery_start == MaxSector) | |
3068 | return sprintf(page, "none\n"); | |
3069 | ||
3070 | return sprintf(page, "%llu\n", recovery_start); | |
3071 | } | |
3072 | ||
3cb03002 | 3073 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
3074 | { |
3075 | unsigned long long recovery_start; | |
3076 | ||
3077 | if (cmd_match(buf, "none")) | |
3078 | recovery_start = MaxSector; | |
b29bebd6 | 3079 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
3080 | return -EINVAL; |
3081 | ||
3082 | if (rdev->mddev->pers && | |
3083 | rdev->raid_disk >= 0) | |
3084 | return -EBUSY; | |
3085 | ||
3086 | rdev->recovery_offset = recovery_start; | |
3087 | if (recovery_start == MaxSector) | |
3088 | set_bit(In_sync, &rdev->flags); | |
3089 | else | |
3090 | clear_bit(In_sync, &rdev->flags); | |
3091 | return len; | |
3092 | } | |
3093 | ||
3094 | static struct rdev_sysfs_entry rdev_recovery_start = | |
3095 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
3096 | ||
fc974ee2 VV |
3097 | /* sysfs access to bad-blocks list. |
3098 | * We present two files. | |
3099 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
3100 | * are recorded as bad. The list is truncated to fit within | |
3101 | * the one-page limit of sysfs. | |
3102 | * Writing "sector length" to this file adds an acknowledged | |
3103 | * bad block list. | |
3104 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
3105 | * been acknowledged. Writing to this file adds bad blocks | |
3106 | * without acknowledging them. This is largely for testing. | |
3107 | */ | |
3cb03002 | 3108 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3109 | { |
3110 | return badblocks_show(&rdev->badblocks, page, 0); | |
3111 | } | |
3cb03002 | 3112 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 3113 | { |
de393cde N |
3114 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
3115 | /* Maybe that ack was all we needed */ | |
3116 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
3117 | wake_up(&rdev->blocked_wait); | |
3118 | return rv; | |
16c791a5 N |
3119 | } |
3120 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
3121 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
3122 | ||
3cb03002 | 3123 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3124 | { |
3125 | return badblocks_show(&rdev->badblocks, page, 1); | |
3126 | } | |
3cb03002 | 3127 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
3128 | { |
3129 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
3130 | } | |
3131 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
3132 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
3133 | ||
86e6ffdd N |
3134 | static struct attribute *rdev_default_attrs[] = { |
3135 | &rdev_state.attr, | |
4dbcdc75 | 3136 | &rdev_errors.attr, |
014236d2 | 3137 | &rdev_slot.attr, |
93c8cad0 | 3138 | &rdev_offset.attr, |
c6563a8c | 3139 | &rdev_new_offset.attr, |
83303b61 | 3140 | &rdev_size.attr, |
06e3c817 | 3141 | &rdev_recovery_start.attr, |
16c791a5 N |
3142 | &rdev_bad_blocks.attr, |
3143 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
3144 | NULL, |
3145 | }; | |
3146 | static ssize_t | |
3147 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3148 | { | |
3149 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3150 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
3151 | |
3152 | if (!entry->show) | |
3153 | return -EIO; | |
758bfc8a N |
3154 | if (!rdev->mddev) |
3155 | return -EBUSY; | |
3156 | return entry->show(rdev, page); | |
86e6ffdd N |
3157 | } |
3158 | ||
3159 | static ssize_t | |
3160 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3161 | const char *page, size_t length) | |
3162 | { | |
3163 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3164 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3165 | ssize_t rv; |
fd01b88c | 3166 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3167 | |
3168 | if (!entry->store) | |
3169 | return -EIO; | |
67463acb N |
3170 | if (!capable(CAP_SYS_ADMIN)) |
3171 | return -EACCES; | |
27c529bb | 3172 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3173 | if (!rv) { |
27c529bb N |
3174 | if (rdev->mddev == NULL) |
3175 | rv = -EBUSY; | |
3176 | else | |
3177 | rv = entry->store(rdev, page, length); | |
6a51830e | 3178 | mddev_unlock(mddev); |
ca388059 N |
3179 | } |
3180 | return rv; | |
86e6ffdd N |
3181 | } |
3182 | ||
3183 | static void rdev_free(struct kobject *ko) | |
3184 | { | |
3cb03002 | 3185 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3186 | kfree(rdev); |
3187 | } | |
52cf25d0 | 3188 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3189 | .show = rdev_attr_show, |
3190 | .store = rdev_attr_store, | |
3191 | }; | |
3192 | static struct kobj_type rdev_ktype = { | |
3193 | .release = rdev_free, | |
3194 | .sysfs_ops = &rdev_sysfs_ops, | |
3195 | .default_attrs = rdev_default_attrs, | |
3196 | }; | |
3197 | ||
3cb03002 | 3198 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3199 | { |
3200 | rdev->desc_nr = -1; | |
3201 | rdev->saved_raid_disk = -1; | |
3202 | rdev->raid_disk = -1; | |
3203 | rdev->flags = 0; | |
3204 | rdev->data_offset = 0; | |
c6563a8c | 3205 | rdev->new_data_offset = 0; |
e8bb9a83 | 3206 | rdev->sb_events = 0; |
0e3ef49e | 3207 | rdev->last_read_error = 0; |
2699b672 N |
3208 | rdev->sb_loaded = 0; |
3209 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3210 | atomic_set(&rdev->nr_pending, 0); |
3211 | atomic_set(&rdev->read_errors, 0); | |
3212 | atomic_set(&rdev->corrected_errors, 0); | |
3213 | ||
3214 | INIT_LIST_HEAD(&rdev->same_set); | |
3215 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3216 | |
3217 | /* Add space to store bad block list. | |
3218 | * This reserves the space even on arrays where it cannot | |
3219 | * be used - I wonder if that matters | |
3220 | */ | |
fc974ee2 | 3221 | return badblocks_init(&rdev->badblocks, 0); |
e8bb9a83 N |
3222 | } |
3223 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3224 | /* |
3225 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3226 | * | |
3227 | * mark the device faulty if: | |
3228 | * | |
3229 | * - the device is nonexistent (zero size) | |
3230 | * - the device has no valid superblock | |
3231 | * | |
3232 | * a faulty rdev _never_ has rdev->sb set. | |
3233 | */ | |
3cb03002 | 3234 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3235 | { |
3236 | char b[BDEVNAME_SIZE]; | |
3237 | int err; | |
3cb03002 | 3238 | struct md_rdev *rdev; |
1da177e4 LT |
3239 | sector_t size; |
3240 | ||
9ffae0cf | 3241 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
9d48739e | 3242 | if (!rdev) |
1da177e4 | 3243 | return ERR_PTR(-ENOMEM); |
1da177e4 | 3244 | |
2230dfe4 N |
3245 | err = md_rdev_init(rdev); |
3246 | if (err) | |
3247 | goto abort_free; | |
3248 | err = alloc_disk_sb(rdev); | |
3249 | if (err) | |
1da177e4 LT |
3250 | goto abort_free; |
3251 | ||
c5d79adb | 3252 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3253 | if (err) |
3254 | goto abort_free; | |
3255 | ||
f9cb074b | 3256 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3257 | |
77304d2a | 3258 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3259 | if (!size) { |
9d48739e | 3260 | pr_warn("md: %s has zero or unknown size, marking faulty!\n", |
1da177e4 LT |
3261 | bdevname(rdev->bdev,b)); |
3262 | err = -EINVAL; | |
3263 | goto abort_free; | |
3264 | } | |
3265 | ||
3266 | if (super_format >= 0) { | |
3267 | err = super_types[super_format]. | |
3268 | load_super(rdev, NULL, super_minor); | |
3269 | if (err == -EINVAL) { | |
9d48739e | 3270 | pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n", |
df968c4e | 3271 | bdevname(rdev->bdev,b), |
9d48739e | 3272 | super_format, super_minor); |
1da177e4 LT |
3273 | goto abort_free; |
3274 | } | |
3275 | if (err < 0) { | |
9d48739e | 3276 | pr_warn("md: could not read %s's sb, not importing!\n", |
1da177e4 LT |
3277 | bdevname(rdev->bdev,b)); |
3278 | goto abort_free; | |
3279 | } | |
3280 | } | |
6bfe0b49 | 3281 | |
1da177e4 LT |
3282 | return rdev; |
3283 | ||
3284 | abort_free: | |
2699b672 N |
3285 | if (rdev->bdev) |
3286 | unlock_rdev(rdev); | |
545c8795 | 3287 | md_rdev_clear(rdev); |
1da177e4 LT |
3288 | kfree(rdev); |
3289 | return ERR_PTR(err); | |
3290 | } | |
3291 | ||
3292 | /* | |
3293 | * Check a full RAID array for plausibility | |
3294 | */ | |
3295 | ||
f72ffdd6 | 3296 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3297 | { |
3298 | int i; | |
3cb03002 | 3299 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3300 | char b[BDEVNAME_SIZE]; |
3301 | ||
3302 | freshest = NULL; | |
dafb20fa | 3303 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3304 | switch (super_types[mddev->major_version]. |
3305 | load_super(rdev, freshest, mddev->minor_version)) { | |
3306 | case 1: | |
3307 | freshest = rdev; | |
3308 | break; | |
3309 | case 0: | |
3310 | break; | |
3311 | default: | |
9d48739e | 3312 | pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n", |
1da177e4 | 3313 | bdevname(rdev->bdev,b)); |
fb56dfef | 3314 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3315 | } |
3316 | ||
1da177e4 LT |
3317 | super_types[mddev->major_version]. |
3318 | validate_super(mddev, freshest); | |
3319 | ||
3320 | i = 0; | |
dafb20fa | 3321 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3322 | if (mddev->max_disks && |
3323 | (rdev->desc_nr >= mddev->max_disks || | |
3324 | i > mddev->max_disks)) { | |
9d48739e N |
3325 | pr_warn("md: %s: %s: only %d devices permitted\n", |
3326 | mdname(mddev), bdevname(rdev->bdev, b), | |
3327 | mddev->max_disks); | |
fb56dfef | 3328 | md_kick_rdev_from_array(rdev); |
de01dfad N |
3329 | continue; |
3330 | } | |
1aee41f6 | 3331 | if (rdev != freshest) { |
1da177e4 LT |
3332 | if (super_types[mddev->major_version]. |
3333 | validate_super(mddev, rdev)) { | |
9d48739e | 3334 | pr_warn("md: kicking non-fresh %s from array!\n", |
1da177e4 | 3335 | bdevname(rdev->bdev,b)); |
fb56dfef | 3336 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3337 | continue; |
3338 | } | |
1aee41f6 | 3339 | } |
1da177e4 LT |
3340 | if (mddev->level == LEVEL_MULTIPATH) { |
3341 | rdev->desc_nr = i++; | |
3342 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3343 | set_bit(In_sync, &rdev->flags); |
f2076e7d SL |
3344 | } else if (rdev->raid_disk >= |
3345 | (mddev->raid_disks - min(0, mddev->delta_disks)) && | |
3346 | !test_bit(Journal, &rdev->flags)) { | |
a778b73f N |
3347 | rdev->raid_disk = -1; |
3348 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3349 | } |
3350 | } | |
1da177e4 LT |
3351 | } |
3352 | ||
72e02075 N |
3353 | /* Read a fixed-point number. |
3354 | * Numbers in sysfs attributes should be in "standard" units where | |
3355 | * possible, so time should be in seconds. | |
f72ffdd6 | 3356 | * However we internally use a a much smaller unit such as |
72e02075 N |
3357 | * milliseconds or jiffies. |
3358 | * This function takes a decimal number with a possible fractional | |
3359 | * component, and produces an integer which is the result of | |
3360 | * multiplying that number by 10^'scale'. | |
3361 | * all without any floating-point arithmetic. | |
3362 | */ | |
3363 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3364 | { | |
3365 | unsigned long result = 0; | |
3366 | long decimals = -1; | |
3367 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3368 | if (*cp == '.') | |
3369 | decimals = 0; | |
3370 | else if (decimals < scale) { | |
3371 | unsigned int value; | |
3372 | value = *cp - '0'; | |
3373 | result = result * 10 + value; | |
3374 | if (decimals >= 0) | |
3375 | decimals++; | |
3376 | } | |
3377 | cp++; | |
3378 | } | |
3379 | if (*cp == '\n') | |
3380 | cp++; | |
3381 | if (*cp) | |
3382 | return -EINVAL; | |
3383 | if (decimals < 0) | |
3384 | decimals = 0; | |
3385 | while (decimals < scale) { | |
3386 | result *= 10; | |
3387 | decimals ++; | |
3388 | } | |
3389 | *res = result; | |
3390 | return 0; | |
3391 | } | |
3392 | ||
16f17b39 | 3393 | static ssize_t |
fd01b88c | 3394 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3395 | { |
3396 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3397 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3398 | } | |
3399 | static ssize_t | |
fd01b88c | 3400 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3401 | { |
16f17b39 | 3402 | unsigned long msec; |
97ce0a7f | 3403 | |
28c1b9fd | 3404 | if (mddev_is_clustered(mddev)) { |
9d48739e | 3405 | pr_warn("md: Safemode is disabled for clustered mode\n"); |
28c1b9fd GR |
3406 | return -EINVAL; |
3407 | } | |
3408 | ||
72e02075 | 3409 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3410 | return -EINVAL; |
16f17b39 N |
3411 | if (msec == 0) |
3412 | mddev->safemode_delay = 0; | |
3413 | else { | |
19052c0e | 3414 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3415 | unsigned long new_delay = (msec*HZ)/1000; |
3416 | ||
3417 | if (new_delay == 0) | |
3418 | new_delay = 1; | |
3419 | mddev->safemode_delay = new_delay; | |
3420 | if (new_delay < old_delay || old_delay == 0) | |
3421 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3422 | } |
3423 | return len; | |
3424 | } | |
3425 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3426 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3427 | |
eae1701f | 3428 | static ssize_t |
fd01b88c | 3429 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3430 | { |
36d091f4 N |
3431 | struct md_personality *p; |
3432 | int ret; | |
3433 | spin_lock(&mddev->lock); | |
3434 | p = mddev->pers; | |
d9d166c2 | 3435 | if (p) |
36d091f4 | 3436 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3437 | else if (mddev->clevel[0]) |
36d091f4 | 3438 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3439 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3440 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3441 | else |
36d091f4 N |
3442 | ret = 0; |
3443 | spin_unlock(&mddev->lock); | |
3444 | return ret; | |
eae1701f N |
3445 | } |
3446 | ||
d9d166c2 | 3447 | static ssize_t |
fd01b88c | 3448 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3449 | { |
f2859af6 | 3450 | char clevel[16]; |
6791875e N |
3451 | ssize_t rv; |
3452 | size_t slen = len; | |
db721d32 | 3453 | struct md_personality *pers, *oldpers; |
f2859af6 | 3454 | long level; |
db721d32 | 3455 | void *priv, *oldpriv; |
3cb03002 | 3456 | struct md_rdev *rdev; |
245f46c2 | 3457 | |
6791875e N |
3458 | if (slen == 0 || slen >= sizeof(clevel)) |
3459 | return -EINVAL; | |
3460 | ||
3461 | rv = mddev_lock(mddev); | |
3462 | if (rv) | |
3463 | return rv; | |
3464 | ||
245f46c2 | 3465 | if (mddev->pers == NULL) { |
6791875e N |
3466 | strncpy(mddev->clevel, buf, slen); |
3467 | if (mddev->clevel[slen-1] == '\n') | |
3468 | slen--; | |
3469 | mddev->clevel[slen] = 0; | |
245f46c2 | 3470 | mddev->level = LEVEL_NONE; |
6791875e N |
3471 | rv = len; |
3472 | goto out_unlock; | |
245f46c2 | 3473 | } |
6791875e | 3474 | rv = -EROFS; |
bd8839e0 | 3475 | if (mddev->ro) |
6791875e | 3476 | goto out_unlock; |
245f46c2 N |
3477 | |
3478 | /* request to change the personality. Need to ensure: | |
3479 | * - array is not engaged in resync/recovery/reshape | |
3480 | * - old personality can be suspended | |
3481 | * - new personality will access other array. | |
3482 | */ | |
3483 | ||
6791875e | 3484 | rv = -EBUSY; |
bb4f1e9d | 3485 | if (mddev->sync_thread || |
f851b60d | 3486 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3487 | mddev->reshape_position != MaxSector || |
3488 | mddev->sysfs_active) | |
6791875e | 3489 | goto out_unlock; |
245f46c2 | 3490 | |
6791875e | 3491 | rv = -EINVAL; |
245f46c2 | 3492 | if (!mddev->pers->quiesce) { |
9d48739e N |
3493 | pr_warn("md: %s: %s does not support online personality change\n", |
3494 | mdname(mddev), mddev->pers->name); | |
6791875e | 3495 | goto out_unlock; |
245f46c2 N |
3496 | } |
3497 | ||
3498 | /* Now find the new personality */ | |
6791875e N |
3499 | strncpy(clevel, buf, slen); |
3500 | if (clevel[slen-1] == '\n') | |
3501 | slen--; | |
3502 | clevel[slen] = 0; | |
b29bebd6 | 3503 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3504 | level = LEVEL_NONE; |
245f46c2 | 3505 | |
f2859af6 DW |
3506 | if (request_module("md-%s", clevel) != 0) |
3507 | request_module("md-level-%s", clevel); | |
245f46c2 | 3508 | spin_lock(&pers_lock); |
f2859af6 | 3509 | pers = find_pers(level, clevel); |
245f46c2 N |
3510 | if (!pers || !try_module_get(pers->owner)) { |
3511 | spin_unlock(&pers_lock); | |
9d48739e | 3512 | pr_warn("md: personality %s not loaded\n", clevel); |
6791875e N |
3513 | rv = -EINVAL; |
3514 | goto out_unlock; | |
245f46c2 N |
3515 | } |
3516 | spin_unlock(&pers_lock); | |
3517 | ||
3518 | if (pers == mddev->pers) { | |
3519 | /* Nothing to do! */ | |
3520 | module_put(pers->owner); | |
6791875e N |
3521 | rv = len; |
3522 | goto out_unlock; | |
245f46c2 N |
3523 | } |
3524 | if (!pers->takeover) { | |
3525 | module_put(pers->owner); | |
9d48739e N |
3526 | pr_warn("md: %s: %s does not support personality takeover\n", |
3527 | mdname(mddev), clevel); | |
6791875e N |
3528 | rv = -EINVAL; |
3529 | goto out_unlock; | |
245f46c2 N |
3530 | } |
3531 | ||
dafb20fa | 3532 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3533 | rdev->new_raid_disk = rdev->raid_disk; |
3534 | ||
245f46c2 N |
3535 | /* ->takeover must set new_* and/or delta_disks |
3536 | * if it succeeds, and may set them when it fails. | |
3537 | */ | |
3538 | priv = pers->takeover(mddev); | |
3539 | if (IS_ERR(priv)) { | |
3540 | mddev->new_level = mddev->level; | |
3541 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3542 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3543 | mddev->raid_disks -= mddev->delta_disks; |
3544 | mddev->delta_disks = 0; | |
2c810cdd | 3545 | mddev->reshape_backwards = 0; |
245f46c2 | 3546 | module_put(pers->owner); |
9d48739e N |
3547 | pr_warn("md: %s: %s would not accept array\n", |
3548 | mdname(mddev), clevel); | |
6791875e N |
3549 | rv = PTR_ERR(priv); |
3550 | goto out_unlock; | |
245f46c2 N |
3551 | } |
3552 | ||
3553 | /* Looks like we have a winner */ | |
3554 | mddev_suspend(mddev); | |
5aa61f42 | 3555 | mddev_detach(mddev); |
36d091f4 N |
3556 | |
3557 | spin_lock(&mddev->lock); | |
db721d32 N |
3558 | oldpers = mddev->pers; |
3559 | oldpriv = mddev->private; | |
3560 | mddev->pers = pers; | |
3561 | mddev->private = priv; | |
3562 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3563 | mddev->level = mddev->new_level; | |
3564 | mddev->layout = mddev->new_layout; | |
3565 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3566 | mddev->delta_disks = 0; | |
3567 | mddev->reshape_backwards = 0; | |
3568 | mddev->degraded = 0; | |
36d091f4 | 3569 | spin_unlock(&mddev->lock); |
db721d32 N |
3570 | |
3571 | if (oldpers->sync_request == NULL && | |
3572 | mddev->external) { | |
3573 | /* We are converting from a no-redundancy array | |
3574 | * to a redundancy array and metadata is managed | |
3575 | * externally so we need to be sure that writes | |
3576 | * won't block due to a need to transition | |
3577 | * clean->dirty | |
3578 | * until external management is started. | |
3579 | */ | |
3580 | mddev->in_sync = 0; | |
3581 | mddev->safemode_delay = 0; | |
3582 | mddev->safemode = 0; | |
3583 | } | |
f72ffdd6 | 3584 | |
db721d32 N |
3585 | oldpers->free(mddev, oldpriv); |
3586 | ||
3587 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3588 | pers->sync_request != NULL) { |
3589 | /* need to add the md_redundancy_group */ | |
3590 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
3591 | pr_warn("md: cannot register extra attributes for %s\n", |
3592 | mdname(mddev)); | |
388975cc | 3593 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3594 | } |
db721d32 | 3595 | if (oldpers->sync_request != NULL && |
a64c876f N |
3596 | pers->sync_request == NULL) { |
3597 | /* need to remove the md_redundancy_group */ | |
3598 | if (mddev->to_remove == NULL) | |
3599 | mddev->to_remove = &md_redundancy_group; | |
3600 | } | |
3601 | ||
4cb9da7d AO |
3602 | module_put(oldpers->owner); |
3603 | ||
dafb20fa | 3604 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3605 | if (rdev->raid_disk < 0) |
3606 | continue; | |
bf2cb0da | 3607 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3608 | rdev->new_raid_disk = -1; |
3609 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3610 | continue; | |
36fad858 | 3611 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3612 | } |
dafb20fa | 3613 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3614 | if (rdev->raid_disk < 0) |
3615 | continue; | |
3616 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3617 | continue; | |
3618 | rdev->raid_disk = rdev->new_raid_disk; | |
3619 | if (rdev->raid_disk < 0) | |
3a981b03 | 3620 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3621 | else { |
36fad858 | 3622 | if (sysfs_link_rdev(mddev, rdev)) |
9d48739e N |
3623 | pr_warn("md: cannot register rd%d for %s after level change\n", |
3624 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3625 | } |
e93f68a1 N |
3626 | } |
3627 | ||
db721d32 | 3628 | if (pers->sync_request == NULL) { |
9af204cf TM |
3629 | /* this is now an array without redundancy, so |
3630 | * it must always be in_sync | |
3631 | */ | |
3632 | mddev->in_sync = 1; | |
3633 | del_timer_sync(&mddev->safemode_timer); | |
3634 | } | |
02e5f5c0 | 3635 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3636 | pers->run(mddev); |
2953079c | 3637 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
47525e59 | 3638 | mddev_resume(mddev); |
830778a1 N |
3639 | if (!mddev->thread) |
3640 | md_update_sb(mddev, 1); | |
5cac7861 | 3641 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3642 | md_new_event(mddev); |
6791875e N |
3643 | rv = len; |
3644 | out_unlock: | |
3645 | mddev_unlock(mddev); | |
d9d166c2 N |
3646 | return rv; |
3647 | } | |
3648 | ||
3649 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3650 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3651 | |
d4dbd025 | 3652 | static ssize_t |
fd01b88c | 3653 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3654 | { |
3655 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3656 | if (mddev->reshape_position != MaxSector && |
3657 | mddev->layout != mddev->new_layout) | |
3658 | return sprintf(page, "%d (%d)\n", | |
3659 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3660 | return sprintf(page, "%d\n", mddev->layout); |
3661 | } | |
3662 | ||
3663 | static ssize_t | |
fd01b88c | 3664 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 | 3665 | { |
4c9309c0 | 3666 | unsigned int n; |
6791875e | 3667 | int err; |
d4dbd025 | 3668 | |
4c9309c0 AD |
3669 | err = kstrtouint(buf, 10, &n); |
3670 | if (err < 0) | |
3671 | return err; | |
6791875e N |
3672 | err = mddev_lock(mddev); |
3673 | if (err) | |
3674 | return err; | |
d4dbd025 | 3675 | |
b3546035 | 3676 | if (mddev->pers) { |
50ac168a | 3677 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3678 | err = -EBUSY; |
3679 | else if (mddev->ro) | |
3680 | err = -EROFS; | |
3681 | else { | |
3682 | mddev->new_layout = n; | |
3683 | err = mddev->pers->check_reshape(mddev); | |
3684 | if (err) | |
3685 | mddev->new_layout = mddev->layout; | |
597a711b | 3686 | } |
b3546035 | 3687 | } else { |
08a02ecd | 3688 | mddev->new_layout = n; |
b3546035 N |
3689 | if (mddev->reshape_position == MaxSector) |
3690 | mddev->layout = n; | |
3691 | } | |
6791875e N |
3692 | mddev_unlock(mddev); |
3693 | return err ?: len; | |
d4dbd025 N |
3694 | } |
3695 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3696 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3697 | |
eae1701f | 3698 | static ssize_t |
fd01b88c | 3699 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3700 | { |
bb636547 N |
3701 | if (mddev->raid_disks == 0) |
3702 | return 0; | |
08a02ecd N |
3703 | if (mddev->reshape_position != MaxSector && |
3704 | mddev->delta_disks != 0) | |
3705 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3706 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3707 | return sprintf(page, "%d\n", mddev->raid_disks); |
3708 | } | |
3709 | ||
fd01b88c | 3710 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3711 | |
3712 | static ssize_t | |
fd01b88c | 3713 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3714 | { |
4c9309c0 | 3715 | unsigned int n; |
6791875e | 3716 | int err; |
da943b99 | 3717 | |
4c9309c0 AD |
3718 | err = kstrtouint(buf, 10, &n); |
3719 | if (err < 0) | |
3720 | return err; | |
da943b99 | 3721 | |
6791875e N |
3722 | err = mddev_lock(mddev); |
3723 | if (err) | |
3724 | return err; | |
da943b99 | 3725 | if (mddev->pers) |
6791875e | 3726 | err = update_raid_disks(mddev, n); |
08a02ecd | 3727 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3728 | struct md_rdev *rdev; |
08a02ecd | 3729 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3730 | |
6791875e | 3731 | err = -EINVAL; |
c6563a8c N |
3732 | rdev_for_each(rdev, mddev) { |
3733 | if (olddisks < n && | |
3734 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3735 | goto out_unlock; |
c6563a8c N |
3736 | if (olddisks > n && |
3737 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3738 | goto out_unlock; |
c6563a8c | 3739 | } |
6791875e | 3740 | err = 0; |
08a02ecd N |
3741 | mddev->delta_disks = n - olddisks; |
3742 | mddev->raid_disks = n; | |
2c810cdd | 3743 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3744 | } else |
da943b99 | 3745 | mddev->raid_disks = n; |
6791875e N |
3746 | out_unlock: |
3747 | mddev_unlock(mddev); | |
3748 | return err ? err : len; | |
da943b99 N |
3749 | } |
3750 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3751 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3752 | |
3b34380a | 3753 | static ssize_t |
fd01b88c | 3754 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3755 | { |
08a02ecd | 3756 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3757 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3758 | return sprintf(page, "%d (%d)\n", | |
3759 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3760 | mddev->chunk_sectors << 9); |
3761 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3762 | } |
3763 | ||
3764 | static ssize_t | |
fd01b88c | 3765 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3766 | { |
4c9309c0 | 3767 | unsigned long n; |
6791875e | 3768 | int err; |
3b34380a | 3769 | |
4c9309c0 AD |
3770 | err = kstrtoul(buf, 10, &n); |
3771 | if (err < 0) | |
3772 | return err; | |
3b34380a | 3773 | |
6791875e N |
3774 | err = mddev_lock(mddev); |
3775 | if (err) | |
3776 | return err; | |
b3546035 | 3777 | if (mddev->pers) { |
50ac168a | 3778 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3779 | err = -EBUSY; |
3780 | else if (mddev->ro) | |
3781 | err = -EROFS; | |
3782 | else { | |
3783 | mddev->new_chunk_sectors = n >> 9; | |
3784 | err = mddev->pers->check_reshape(mddev); | |
3785 | if (err) | |
3786 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 3787 | } |
b3546035 | 3788 | } else { |
664e7c41 | 3789 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3790 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3791 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3792 | } |
6791875e N |
3793 | mddev_unlock(mddev); |
3794 | return err ?: len; | |
3b34380a N |
3795 | } |
3796 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3797 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3798 | |
a94213b1 | 3799 | static ssize_t |
fd01b88c | 3800 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3801 | { |
d1a7c503 N |
3802 | if (mddev->recovery_cp == MaxSector) |
3803 | return sprintf(page, "none\n"); | |
a94213b1 N |
3804 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3805 | } | |
3806 | ||
3807 | static ssize_t | |
fd01b88c | 3808 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3809 | { |
4c9309c0 | 3810 | unsigned long long n; |
6791875e | 3811 | int err; |
4c9309c0 AD |
3812 | |
3813 | if (cmd_match(buf, "none")) | |
3814 | n = MaxSector; | |
3815 | else { | |
3816 | err = kstrtoull(buf, 10, &n); | |
3817 | if (err < 0) | |
3818 | return err; | |
3819 | if (n != (sector_t)n) | |
3820 | return -EINVAL; | |
3821 | } | |
a94213b1 | 3822 | |
6791875e N |
3823 | err = mddev_lock(mddev); |
3824 | if (err) | |
3825 | return err; | |
b098636c | 3826 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e | 3827 | err = -EBUSY; |
a94213b1 | 3828 | |
6791875e N |
3829 | if (!err) { |
3830 | mddev->recovery_cp = n; | |
3831 | if (mddev->pers) | |
2953079c | 3832 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
6791875e N |
3833 | } |
3834 | mddev_unlock(mddev); | |
3835 | return err ?: len; | |
a94213b1 N |
3836 | } |
3837 | static struct md_sysfs_entry md_resync_start = | |
750f199e N |
3838 | __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR, |
3839 | resync_start_show, resync_start_store); | |
a94213b1 | 3840 | |
9e653b63 N |
3841 | /* |
3842 | * The array state can be: | |
3843 | * | |
3844 | * clear | |
3845 | * No devices, no size, no level | |
3846 | * Equivalent to STOP_ARRAY ioctl | |
3847 | * inactive | |
3848 | * May have some settings, but array is not active | |
3849 | * all IO results in error | |
3850 | * When written, doesn't tear down array, but just stops it | |
3851 | * suspended (not supported yet) | |
3852 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3853 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3854 | * readonly |
3855 | * no resync can happen. no superblocks get written. | |
3856 | * write requests fail | |
3857 | * read-auto | |
3858 | * like readonly, but behaves like 'clean' on a write request. | |
3859 | * | |
3860 | * clean - no pending writes, but otherwise active. | |
3861 | * When written to inactive array, starts without resync | |
3862 | * If a write request arrives then | |
3863 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3864 | * if not known, block and switch to write-pending | |
3865 | * If written to an active array that has pending writes, then fails. | |
3866 | * active | |
3867 | * fully active: IO and resync can be happening. | |
3868 | * When written to inactive array, starts with resync | |
3869 | * | |
3870 | * write-pending | |
3871 | * clean, but writes are blocked waiting for 'active' to be written. | |
3872 | * | |
3873 | * active-idle | |
3874 | * like active, but no writes have been seen for a while (100msec). | |
3875 | * | |
3876 | */ | |
3877 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3878 | write_pending, active_idle, bad_word}; | |
05381954 | 3879 | static char *array_states[] = { |
9e653b63 N |
3880 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3881 | "write-pending", "active-idle", NULL }; | |
3882 | ||
3883 | static int match_word(const char *word, char **list) | |
3884 | { | |
3885 | int n; | |
3886 | for (n=0; list[n]; n++) | |
3887 | if (cmd_match(word, list[n])) | |
3888 | break; | |
3889 | return n; | |
3890 | } | |
3891 | ||
3892 | static ssize_t | |
fd01b88c | 3893 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3894 | { |
3895 | enum array_state st = inactive; | |
3896 | ||
3897 | if (mddev->pers) | |
3898 | switch(mddev->ro) { | |
3899 | case 1: | |
3900 | st = readonly; | |
3901 | break; | |
3902 | case 2: | |
3903 | st = read_auto; | |
3904 | break; | |
3905 | case 0: | |
2953079c | 3906 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
e691063a | 3907 | st = write_pending; |
16f88949 TM |
3908 | else if (mddev->in_sync) |
3909 | st = clean; | |
9e653b63 N |
3910 | else if (mddev->safemode) |
3911 | st = active_idle; | |
3912 | else | |
3913 | st = active; | |
3914 | } | |
3915 | else { | |
3916 | if (list_empty(&mddev->disks) && | |
3917 | mddev->raid_disks == 0 && | |
58c0fed4 | 3918 | mddev->dev_sectors == 0) |
9e653b63 N |
3919 | st = clear; |
3920 | else | |
3921 | st = inactive; | |
3922 | } | |
3923 | return sprintf(page, "%s\n", array_states[st]); | |
3924 | } | |
3925 | ||
f72ffdd6 N |
3926 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
3927 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
3928 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 3929 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
3930 | |
3931 | static ssize_t | |
fd01b88c | 3932 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 3933 | { |
6791875e | 3934 | int err; |
9e653b63 | 3935 | enum array_state st = match_word(buf, array_states); |
6791875e N |
3936 | |
3937 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
3938 | /* don't take reconfig_mutex when toggling between | |
3939 | * clean and active | |
3940 | */ | |
3941 | spin_lock(&mddev->lock); | |
3942 | if (st == active) { | |
3943 | restart_array(mddev); | |
2953079c | 3944 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
91a6c4ad | 3945 | md_wakeup_thread(mddev->thread); |
6791875e N |
3946 | wake_up(&mddev->sb_wait); |
3947 | err = 0; | |
3948 | } else /* st == clean */ { | |
3949 | restart_array(mddev); | |
3950 | if (atomic_read(&mddev->writes_pending) == 0) { | |
3951 | if (mddev->in_sync == 0) { | |
3952 | mddev->in_sync = 1; | |
3953 | if (mddev->safemode == 1) | |
3954 | mddev->safemode = 0; | |
2953079c | 3955 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
6791875e N |
3956 | } |
3957 | err = 0; | |
3958 | } else | |
3959 | err = -EBUSY; | |
3960 | } | |
573275b5 TM |
3961 | if (!err) |
3962 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
6791875e | 3963 | spin_unlock(&mddev->lock); |
c008f1d3 | 3964 | return err ?: len; |
6791875e N |
3965 | } |
3966 | err = mddev_lock(mddev); | |
3967 | if (err) | |
3968 | return err; | |
3969 | err = -EINVAL; | |
9e653b63 N |
3970 | switch(st) { |
3971 | case bad_word: | |
3972 | break; | |
3973 | case clear: | |
3974 | /* stopping an active array */ | |
a05b7ea0 | 3975 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3976 | break; |
3977 | case inactive: | |
3978 | /* stopping an active array */ | |
90cf195d | 3979 | if (mddev->pers) |
a05b7ea0 | 3980 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3981 | else |
e691063a | 3982 | err = 0; /* already inactive */ |
9e653b63 N |
3983 | break; |
3984 | case suspended: | |
3985 | break; /* not supported yet */ | |
3986 | case readonly: | |
3987 | if (mddev->pers) | |
a05b7ea0 | 3988 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3989 | else { |
3990 | mddev->ro = 1; | |
648b629e | 3991 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3992 | err = do_md_run(mddev); |
3993 | } | |
3994 | break; | |
3995 | case read_auto: | |
9e653b63 | 3996 | if (mddev->pers) { |
80268ee9 | 3997 | if (mddev->ro == 0) |
a05b7ea0 | 3998 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3999 | else if (mddev->ro == 1) |
648b629e N |
4000 | err = restart_array(mddev); |
4001 | if (err == 0) { | |
4002 | mddev->ro = 2; | |
4003 | set_disk_ro(mddev->gendisk, 0); | |
4004 | } | |
9e653b63 N |
4005 | } else { |
4006 | mddev->ro = 2; | |
4007 | err = do_md_run(mddev); | |
4008 | } | |
4009 | break; | |
4010 | case clean: | |
4011 | if (mddev->pers) { | |
339421de SL |
4012 | err = restart_array(mddev); |
4013 | if (err) | |
4014 | break; | |
85572d7c | 4015 | spin_lock(&mddev->lock); |
9e653b63 | 4016 | if (atomic_read(&mddev->writes_pending) == 0) { |
e691063a N |
4017 | if (mddev->in_sync == 0) { |
4018 | mddev->in_sync = 1; | |
31a59e34 N |
4019 | if (mddev->safemode == 1) |
4020 | mddev->safemode = 0; | |
2953079c | 4021 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
e691063a N |
4022 | } |
4023 | err = 0; | |
4024 | } else | |
4025 | err = -EBUSY; | |
85572d7c | 4026 | spin_unlock(&mddev->lock); |
5bf29597 N |
4027 | } else |
4028 | err = -EINVAL; | |
9e653b63 N |
4029 | break; |
4030 | case active: | |
4031 | if (mddev->pers) { | |
339421de SL |
4032 | err = restart_array(mddev); |
4033 | if (err) | |
4034 | break; | |
2953079c | 4035 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
9e653b63 N |
4036 | wake_up(&mddev->sb_wait); |
4037 | err = 0; | |
4038 | } else { | |
4039 | mddev->ro = 0; | |
648b629e | 4040 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
4041 | err = do_md_run(mddev); |
4042 | } | |
4043 | break; | |
4044 | case write_pending: | |
4045 | case active_idle: | |
4046 | /* these cannot be set */ | |
4047 | break; | |
4048 | } | |
6791875e N |
4049 | |
4050 | if (!err) { | |
1d23f178 N |
4051 | if (mddev->hold_active == UNTIL_IOCTL) |
4052 | mddev->hold_active = 0; | |
00bcb4ac | 4053 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4054 | } |
6791875e N |
4055 | mddev_unlock(mddev); |
4056 | return err ?: len; | |
9e653b63 | 4057 | } |
80ca3a44 | 4058 | static struct md_sysfs_entry md_array_state = |
750f199e | 4059 | __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); |
9e653b63 | 4060 | |
1e50915f | 4061 | static ssize_t |
fd01b88c | 4062 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
4063 | return sprintf(page, "%d\n", |
4064 | atomic_read(&mddev->max_corr_read_errors)); | |
4065 | } | |
4066 | ||
4067 | static ssize_t | |
fd01b88c | 4068 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f | 4069 | { |
4c9309c0 AD |
4070 | unsigned int n; |
4071 | int rv; | |
1e50915f | 4072 | |
4c9309c0 AD |
4073 | rv = kstrtouint(buf, 10, &n); |
4074 | if (rv < 0) | |
4075 | return rv; | |
4076 | atomic_set(&mddev->max_corr_read_errors, n); | |
4077 | return len; | |
1e50915f RB |
4078 | } |
4079 | ||
4080 | static struct md_sysfs_entry max_corr_read_errors = | |
4081 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
4082 | max_corrected_read_errors_store); | |
4083 | ||
6d7ff738 | 4084 | static ssize_t |
fd01b88c | 4085 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
4086 | { |
4087 | return -EINVAL; | |
4088 | } | |
4089 | ||
4090 | static ssize_t | |
fd01b88c | 4091 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
4092 | { |
4093 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
4094 | /* The new device is added to the array. | |
4095 | * If the array has a persistent superblock, we read the | |
4096 | * superblock to initialise info and check validity. | |
4097 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
4098 | * which mainly checks size. | |
4099 | */ | |
4100 | char *e; | |
4101 | int major = simple_strtoul(buf, &e, 10); | |
4102 | int minor; | |
4103 | dev_t dev; | |
3cb03002 | 4104 | struct md_rdev *rdev; |
6d7ff738 N |
4105 | int err; |
4106 | ||
4107 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
4108 | return -EINVAL; | |
4109 | minor = simple_strtoul(e+1, &e, 10); | |
4110 | if (*e && *e != '\n') | |
4111 | return -EINVAL; | |
4112 | dev = MKDEV(major, minor); | |
4113 | if (major != MAJOR(dev) || | |
4114 | minor != MINOR(dev)) | |
4115 | return -EOVERFLOW; | |
4116 | ||
6791875e N |
4117 | flush_workqueue(md_misc_wq); |
4118 | ||
4119 | err = mddev_lock(mddev); | |
4120 | if (err) | |
4121 | return err; | |
6d7ff738 N |
4122 | if (mddev->persistent) { |
4123 | rdev = md_import_device(dev, mddev->major_version, | |
4124 | mddev->minor_version); | |
4125 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4126 | struct md_rdev *rdev0 |
4127 | = list_entry(mddev->disks.next, | |
4128 | struct md_rdev, same_set); | |
6d7ff738 N |
4129 | err = super_types[mddev->major_version] |
4130 | .load_super(rdev, rdev0, mddev->minor_version); | |
4131 | if (err < 0) | |
4132 | goto out; | |
4133 | } | |
c5d79adb N |
4134 | } else if (mddev->external) |
4135 | rdev = md_import_device(dev, -2, -1); | |
4136 | else | |
6d7ff738 N |
4137 | rdev = md_import_device(dev, -1, -1); |
4138 | ||
9a8c0fa8 N |
4139 | if (IS_ERR(rdev)) { |
4140 | mddev_unlock(mddev); | |
6d7ff738 | 4141 | return PTR_ERR(rdev); |
9a8c0fa8 | 4142 | } |
6d7ff738 N |
4143 | err = bind_rdev_to_array(rdev, mddev); |
4144 | out: | |
4145 | if (err) | |
4146 | export_rdev(rdev); | |
6791875e | 4147 | mddev_unlock(mddev); |
6d7ff738 N |
4148 | return err ? err : len; |
4149 | } | |
4150 | ||
4151 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4152 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4153 | |
9b1d1dac | 4154 | static ssize_t |
fd01b88c | 4155 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4156 | { |
4157 | char *end; | |
4158 | unsigned long chunk, end_chunk; | |
6791875e | 4159 | int err; |
9b1d1dac | 4160 | |
6791875e N |
4161 | err = mddev_lock(mddev); |
4162 | if (err) | |
4163 | return err; | |
9b1d1dac PC |
4164 | if (!mddev->bitmap) |
4165 | goto out; | |
4166 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4167 | while (*buf) { | |
4168 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4169 | if (buf == end) break; | |
4170 | if (*end == '-') { /* range */ | |
4171 | buf = end + 1; | |
4172 | end_chunk = simple_strtoul(buf, &end, 0); | |
4173 | if (buf == end) break; | |
4174 | } | |
4175 | if (*end && !isspace(*end)) break; | |
4176 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4177 | buf = skip_spaces(end); |
9b1d1dac PC |
4178 | } |
4179 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4180 | out: | |
6791875e | 4181 | mddev_unlock(mddev); |
9b1d1dac PC |
4182 | return len; |
4183 | } | |
4184 | ||
4185 | static struct md_sysfs_entry md_bitmap = | |
4186 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4187 | ||
a35b0d69 | 4188 | static ssize_t |
fd01b88c | 4189 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4190 | { |
58c0fed4 AN |
4191 | return sprintf(page, "%llu\n", |
4192 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4193 | } |
4194 | ||
fd01b88c | 4195 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4196 | |
4197 | static ssize_t | |
fd01b88c | 4198 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4199 | { |
4200 | /* If array is inactive, we can reduce the component size, but | |
4201 | * not increase it (except from 0). | |
4202 | * If array is active, we can try an on-line resize | |
4203 | */ | |
b522adcd DW |
4204 | sector_t sectors; |
4205 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4206 | |
58c0fed4 AN |
4207 | if (err < 0) |
4208 | return err; | |
6791875e N |
4209 | err = mddev_lock(mddev); |
4210 | if (err) | |
4211 | return err; | |
a35b0d69 | 4212 | if (mddev->pers) { |
58c0fed4 | 4213 | err = update_size(mddev, sectors); |
4ba1e788 XN |
4214 | if (err == 0) |
4215 | md_update_sb(mddev, 1); | |
a35b0d69 | 4216 | } else { |
58c0fed4 AN |
4217 | if (mddev->dev_sectors == 0 || |
4218 | mddev->dev_sectors > sectors) | |
4219 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4220 | else |
4221 | err = -ENOSPC; | |
4222 | } | |
6791875e | 4223 | mddev_unlock(mddev); |
a35b0d69 N |
4224 | return err ? err : len; |
4225 | } | |
4226 | ||
4227 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4228 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4229 | |
83f0d77a | 4230 | /* Metadata version. |
e691063a N |
4231 | * This is one of |
4232 | * 'none' for arrays with no metadata (good luck...) | |
4233 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4234 | * or N.M for internally known formats |
4235 | */ | |
4236 | static ssize_t | |
fd01b88c | 4237 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4238 | { |
4239 | if (mddev->persistent) | |
4240 | return sprintf(page, "%d.%d\n", | |
4241 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4242 | else if (mddev->external) |
4243 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4244 | else |
4245 | return sprintf(page, "none\n"); | |
4246 | } | |
4247 | ||
4248 | static ssize_t | |
fd01b88c | 4249 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4250 | { |
4251 | int major, minor; | |
4252 | char *e; | |
6791875e | 4253 | int err; |
ea43ddd8 N |
4254 | /* Changing the details of 'external' metadata is |
4255 | * always permitted. Otherwise there must be | |
4256 | * no devices attached to the array. | |
4257 | */ | |
6791875e N |
4258 | |
4259 | err = mddev_lock(mddev); | |
4260 | if (err) | |
4261 | return err; | |
4262 | err = -EBUSY; | |
ea43ddd8 N |
4263 | if (mddev->external && strncmp(buf, "external:", 9) == 0) |
4264 | ; | |
4265 | else if (!list_empty(&mddev->disks)) | |
6791875e | 4266 | goto out_unlock; |
8bb93aac | 4267 | |
6791875e | 4268 | err = 0; |
8bb93aac N |
4269 | if (cmd_match(buf, "none")) { |
4270 | mddev->persistent = 0; | |
e691063a N |
4271 | mddev->external = 0; |
4272 | mddev->major_version = 0; | |
4273 | mddev->minor_version = 90; | |
6791875e | 4274 | goto out_unlock; |
e691063a N |
4275 | } |
4276 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4277 | size_t namelen = len-9; |
e691063a N |
4278 | if (namelen >= sizeof(mddev->metadata_type)) |
4279 | namelen = sizeof(mddev->metadata_type)-1; | |
4280 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4281 | mddev->metadata_type[namelen] = 0; | |
4282 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4283 | mddev->metadata_type[--namelen] = 0; | |
4284 | mddev->persistent = 0; | |
4285 | mddev->external = 1; | |
8bb93aac N |
4286 | mddev->major_version = 0; |
4287 | mddev->minor_version = 90; | |
6791875e | 4288 | goto out_unlock; |
8bb93aac N |
4289 | } |
4290 | major = simple_strtoul(buf, &e, 10); | |
6791875e | 4291 | err = -EINVAL; |
8bb93aac | 4292 | if (e==buf || *e != '.') |
6791875e | 4293 | goto out_unlock; |
8bb93aac N |
4294 | buf = e+1; |
4295 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4296 | if (e==buf || (*e && *e != '\n') ) |
6791875e N |
4297 | goto out_unlock; |
4298 | err = -ENOENT; | |
50511da3 | 4299 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
6791875e | 4300 | goto out_unlock; |
8bb93aac N |
4301 | mddev->major_version = major; |
4302 | mddev->minor_version = minor; | |
4303 | mddev->persistent = 1; | |
e691063a | 4304 | mddev->external = 0; |
6791875e N |
4305 | err = 0; |
4306 | out_unlock: | |
4307 | mddev_unlock(mddev); | |
4308 | return err ?: len; | |
8bb93aac N |
4309 | } |
4310 | ||
4311 | static struct md_sysfs_entry md_metadata = | |
750f199e | 4312 | __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4313 | |
24dd469d | 4314 | static ssize_t |
fd01b88c | 4315 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4316 | { |
7eec314d | 4317 | char *type = "idle"; |
b7b17c9b N |
4318 | unsigned long recovery = mddev->recovery; |
4319 | if (test_bit(MD_RECOVERY_FROZEN, &recovery)) | |
b6a9ce68 | 4320 | type = "frozen"; |
b7b17c9b N |
4321 | else if (test_bit(MD_RECOVERY_RUNNING, &recovery) || |
4322 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) { | |
4323 | if (test_bit(MD_RECOVERY_RESHAPE, &recovery)) | |
ccfcc3c1 | 4324 | type = "reshape"; |
b7b17c9b N |
4325 | else if (test_bit(MD_RECOVERY_SYNC, &recovery)) { |
4326 | if (!test_bit(MD_RECOVERY_REQUESTED, &recovery)) | |
24dd469d | 4327 | type = "resync"; |
b7b17c9b | 4328 | else if (test_bit(MD_RECOVERY_CHECK, &recovery)) |
24dd469d N |
4329 | type = "check"; |
4330 | else | |
4331 | type = "repair"; | |
b7b17c9b | 4332 | } else if (test_bit(MD_RECOVERY_RECOVER, &recovery)) |
24dd469d | 4333 | type = "recover"; |
985ca973 N |
4334 | else if (mddev->reshape_position != MaxSector) |
4335 | type = "reshape"; | |
24dd469d N |
4336 | } |
4337 | return sprintf(page, "%s\n", type); | |
4338 | } | |
4339 | ||
4340 | static ssize_t | |
fd01b88c | 4341 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4342 | { |
7eec314d N |
4343 | if (!mddev->pers || !mddev->pers->sync_request) |
4344 | return -EINVAL; | |
4345 | ||
b6a9ce68 N |
4346 | |
4347 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
56ccc112 N |
4348 | if (cmd_match(page, "frozen")) |
4349 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4350 | else | |
4351 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
8e8e2518 N |
4352 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
4353 | mddev_lock(mddev) == 0) { | |
4354 | flush_workqueue(md_misc_wq); | |
4355 | if (mddev->sync_thread) { | |
4356 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e | 4357 | md_reap_sync_thread(mddev); |
6791875e | 4358 | } |
8e8e2518 | 4359 | mddev_unlock(mddev); |
7eec314d | 4360 | } |
312045ee | 4361 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
24dd469d | 4362 | return -EBUSY; |
72a23c21 | 4363 | else if (cmd_match(page, "resync")) |
56ccc112 | 4364 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4365 | else if (cmd_match(page, "recover")) { |
56ccc112 | 4366 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4367 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
72a23c21 | 4368 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4369 | int err; |
4370 | if (mddev->pers->start_reshape == NULL) | |
4371 | return -EINVAL; | |
6791875e N |
4372 | err = mddev_lock(mddev); |
4373 | if (!err) { | |
312045ee N |
4374 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4375 | err = -EBUSY; | |
4376 | else { | |
4377 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4378 | err = mddev->pers->start_reshape(mddev); | |
4379 | } | |
6791875e N |
4380 | mddev_unlock(mddev); |
4381 | } | |
16484bf5 N |
4382 | if (err) |
4383 | return err; | |
a99ac971 | 4384 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4385 | } else { |
bce74dac | 4386 | if (cmd_match(page, "check")) |
7eec314d | 4387 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4388 | else if (!cmd_match(page, "repair")) |
7eec314d | 4389 | return -EINVAL; |
56ccc112 | 4390 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
7eec314d N |
4391 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
4392 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4393 | } |
48c26ddc N |
4394 | if (mddev->ro == 2) { |
4395 | /* A write to sync_action is enough to justify | |
4396 | * canceling read-auto mode | |
4397 | */ | |
4398 | mddev->ro = 0; | |
4399 | md_wakeup_thread(mddev->sync_thread); | |
4400 | } | |
03c902e1 | 4401 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4402 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4403 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4404 | return len; |
4405 | } | |
4406 | ||
c4a39551 | 4407 | static struct md_sysfs_entry md_scan_mode = |
750f199e | 4408 | __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
c4a39551 JB |
4409 | |
4410 | static ssize_t | |
4411 | last_sync_action_show(struct mddev *mddev, char *page) | |
4412 | { | |
4413 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4414 | } | |
4415 | ||
4416 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4417 | ||
9d88883e | 4418 | static ssize_t |
fd01b88c | 4419 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4420 | { |
4421 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4422 | (unsigned long long) |
4423 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4424 | } |
4425 | ||
80ca3a44 | 4426 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4427 | |
88202a0c | 4428 | static ssize_t |
fd01b88c | 4429 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4430 | { |
4431 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4432 | mddev->sync_speed_min ? "local": "system"); | |
4433 | } | |
4434 | ||
4435 | static ssize_t | |
fd01b88c | 4436 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4437 | { |
4c9309c0 AD |
4438 | unsigned int min; |
4439 | int rv; | |
4440 | ||
88202a0c | 4441 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4442 | min = 0; |
4443 | } else { | |
4444 | rv = kstrtouint(buf, 10, &min); | |
4445 | if (rv < 0) | |
4446 | return rv; | |
4447 | if (min == 0) | |
4448 | return -EINVAL; | |
88202a0c | 4449 | } |
88202a0c N |
4450 | mddev->sync_speed_min = min; |
4451 | return len; | |
4452 | } | |
4453 | ||
4454 | static struct md_sysfs_entry md_sync_min = | |
4455 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4456 | ||
4457 | static ssize_t | |
fd01b88c | 4458 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4459 | { |
4460 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4461 | mddev->sync_speed_max ? "local": "system"); | |
4462 | } | |
4463 | ||
4464 | static ssize_t | |
fd01b88c | 4465 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4466 | { |
4c9309c0 AD |
4467 | unsigned int max; |
4468 | int rv; | |
4469 | ||
88202a0c | 4470 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4471 | max = 0; |
4472 | } else { | |
4473 | rv = kstrtouint(buf, 10, &max); | |
4474 | if (rv < 0) | |
4475 | return rv; | |
4476 | if (max == 0) | |
4477 | return -EINVAL; | |
88202a0c | 4478 | } |
88202a0c N |
4479 | mddev->sync_speed_max = max; |
4480 | return len; | |
4481 | } | |
4482 | ||
4483 | static struct md_sysfs_entry md_sync_max = | |
4484 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4485 | ||
d7f3d291 | 4486 | static ssize_t |
fd01b88c | 4487 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4488 | { |
4489 | return sprintf(page, "%d\n", mddev->degraded); | |
4490 | } | |
4491 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4492 | |
90b08710 | 4493 | static ssize_t |
fd01b88c | 4494 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4495 | { |
4496 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4497 | } | |
4498 | ||
4499 | static ssize_t | |
fd01b88c | 4500 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4501 | { |
4502 | long n; | |
4503 | ||
b29bebd6 | 4504 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4505 | return -EINVAL; |
4506 | ||
4507 | if (n != 0 && n != 1) | |
4508 | return -EINVAL; | |
4509 | ||
4510 | mddev->parallel_resync = n; | |
4511 | ||
4512 | if (mddev->sync_thread) | |
4513 | wake_up(&resync_wait); | |
4514 | ||
4515 | return len; | |
4516 | } | |
4517 | ||
4518 | /* force parallel resync, even with shared block devices */ | |
4519 | static struct md_sysfs_entry md_sync_force_parallel = | |
4520 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4521 | sync_force_parallel_show, sync_force_parallel_store); | |
4522 | ||
88202a0c | 4523 | static ssize_t |
fd01b88c | 4524 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4525 | { |
4526 | unsigned long resync, dt, db; | |
d1a7c503 N |
4527 | if (mddev->curr_resync == 0) |
4528 | return sprintf(page, "none\n"); | |
9687a60c AN |
4529 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4530 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4531 | if (!dt) dt++; |
9687a60c AN |
4532 | db = resync - mddev->resync_mark_cnt; |
4533 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4534 | } |
4535 | ||
80ca3a44 | 4536 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4537 | |
4538 | static ssize_t | |
fd01b88c | 4539 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4540 | { |
13ae864b | 4541 | unsigned long long max_sectors, resync; |
88202a0c | 4542 | |
acb180b0 N |
4543 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4544 | return sprintf(page, "none\n"); | |
4545 | ||
72f36d59 N |
4546 | if (mddev->curr_resync == 1 || |
4547 | mddev->curr_resync == 2) | |
4548 | return sprintf(page, "delayed\n"); | |
4549 | ||
c804cdec N |
4550 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4551 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4552 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4553 | else |
58c0fed4 | 4554 | max_sectors = mddev->dev_sectors; |
88202a0c | 4555 | |
acb180b0 | 4556 | resync = mddev->curr_resync_completed; |
13ae864b | 4557 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4558 | } |
4559 | ||
750f199e N |
4560 | static struct md_sysfs_entry md_sync_completed = |
4561 | __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL); | |
88202a0c | 4562 | |
5e96ee65 | 4563 | static ssize_t |
fd01b88c | 4564 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4565 | { |
4566 | return sprintf(page, "%llu\n", | |
4567 | (unsigned long long)mddev->resync_min); | |
4568 | } | |
4569 | static ssize_t | |
fd01b88c | 4570 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4571 | { |
4572 | unsigned long long min; | |
23da422b | 4573 | int err; |
23da422b | 4574 | |
b29bebd6 | 4575 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4576 | return -EINVAL; |
23da422b N |
4577 | |
4578 | spin_lock(&mddev->lock); | |
4579 | err = -EINVAL; | |
5e96ee65 | 4580 | if (min > mddev->resync_max) |
23da422b N |
4581 | goto out_unlock; |
4582 | ||
4583 | err = -EBUSY; | |
5e96ee65 | 4584 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4585 | goto out_unlock; |
5e96ee65 | 4586 | |
50c37b13 N |
4587 | /* Round down to multiple of 4K for safety */ |
4588 | mddev->resync_min = round_down(min, 8); | |
23da422b | 4589 | err = 0; |
5e96ee65 | 4590 | |
23da422b N |
4591 | out_unlock: |
4592 | spin_unlock(&mddev->lock); | |
4593 | return err ?: len; | |
5e96ee65 NB |
4594 | } |
4595 | ||
4596 | static struct md_sysfs_entry md_min_sync = | |
4597 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4598 | ||
c6207277 | 4599 | static ssize_t |
fd01b88c | 4600 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4601 | { |
4602 | if (mddev->resync_max == MaxSector) | |
4603 | return sprintf(page, "max\n"); | |
4604 | else | |
4605 | return sprintf(page, "%llu\n", | |
4606 | (unsigned long long)mddev->resync_max); | |
4607 | } | |
4608 | static ssize_t | |
fd01b88c | 4609 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4610 | { |
23da422b N |
4611 | int err; |
4612 | spin_lock(&mddev->lock); | |
c6207277 N |
4613 | if (strncmp(buf, "max", 3) == 0) |
4614 | mddev->resync_max = MaxSector; | |
4615 | else { | |
5e96ee65 | 4616 | unsigned long long max; |
23da422b N |
4617 | int chunk; |
4618 | ||
4619 | err = -EINVAL; | |
b29bebd6 | 4620 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4621 | goto out_unlock; |
5e96ee65 | 4622 | if (max < mddev->resync_min) |
23da422b N |
4623 | goto out_unlock; |
4624 | ||
4625 | err = -EBUSY; | |
c6207277 | 4626 | if (max < mddev->resync_max && |
4d484a4a | 4627 | mddev->ro == 0 && |
c6207277 | 4628 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4629 | goto out_unlock; |
c6207277 N |
4630 | |
4631 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4632 | chunk = mddev->chunk_sectors; |
4633 | if (chunk) { | |
2ac06c33 | 4634 | sector_t temp = max; |
23da422b N |
4635 | |
4636 | err = -EINVAL; | |
4637 | if (sector_div(temp, chunk)) | |
4638 | goto out_unlock; | |
c6207277 N |
4639 | } |
4640 | mddev->resync_max = max; | |
4641 | } | |
4642 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4643 | err = 0; |
4644 | out_unlock: | |
4645 | spin_unlock(&mddev->lock); | |
4646 | return err ?: len; | |
c6207277 N |
4647 | } |
4648 | ||
4649 | static struct md_sysfs_entry md_max_sync = | |
4650 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4651 | ||
e464eafd | 4652 | static ssize_t |
fd01b88c | 4653 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4654 | { |
4655 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4656 | } | |
4657 | ||
4658 | static ssize_t | |
fd01b88c | 4659 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4660 | { |
4c9309c0 | 4661 | unsigned long long old, new; |
6791875e | 4662 | int err; |
e464eafd | 4663 | |
4c9309c0 AD |
4664 | err = kstrtoull(buf, 10, &new); |
4665 | if (err < 0) | |
4666 | return err; | |
4667 | if (new != (sector_t)new) | |
e464eafd | 4668 | return -EINVAL; |
23ddff37 | 4669 | |
6791875e N |
4670 | err = mddev_lock(mddev); |
4671 | if (err) | |
4672 | return err; | |
4673 | err = -EINVAL; | |
4674 | if (mddev->pers == NULL || | |
4675 | mddev->pers->quiesce == NULL) | |
4676 | goto unlock; | |
4677 | old = mddev->suspend_lo; | |
23ddff37 N |
4678 | mddev->suspend_lo = new; |
4679 | if (new >= old) | |
4680 | /* Shrinking suspended region */ | |
e464eafd | 4681 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4682 | else { |
4683 | /* Expanding suspended region - need to wait */ | |
4684 | mddev->pers->quiesce(mddev, 1); | |
4685 | mddev->pers->quiesce(mddev, 0); | |
4686 | } | |
6791875e N |
4687 | err = 0; |
4688 | unlock: | |
4689 | mddev_unlock(mddev); | |
4690 | return err ?: len; | |
e464eafd N |
4691 | } |
4692 | static struct md_sysfs_entry md_suspend_lo = | |
4693 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4694 | ||
e464eafd | 4695 | static ssize_t |
fd01b88c | 4696 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4697 | { |
4698 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4699 | } | |
4700 | ||
4701 | static ssize_t | |
fd01b88c | 4702 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4703 | { |
4c9309c0 | 4704 | unsigned long long old, new; |
6791875e | 4705 | int err; |
e464eafd | 4706 | |
4c9309c0 AD |
4707 | err = kstrtoull(buf, 10, &new); |
4708 | if (err < 0) | |
4709 | return err; | |
4710 | if (new != (sector_t)new) | |
e464eafd | 4711 | return -EINVAL; |
23ddff37 | 4712 | |
6791875e N |
4713 | err = mddev_lock(mddev); |
4714 | if (err) | |
4715 | return err; | |
4716 | err = -EINVAL; | |
4717 | if (mddev->pers == NULL || | |
4718 | mddev->pers->quiesce == NULL) | |
4719 | goto unlock; | |
4720 | old = mddev->suspend_hi; | |
23ddff37 N |
4721 | mddev->suspend_hi = new; |
4722 | if (new <= old) | |
4723 | /* Shrinking suspended region */ | |
4724 | mddev->pers->quiesce(mddev, 2); | |
4725 | else { | |
4726 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4727 | mddev->pers->quiesce(mddev, 1); |
4728 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 | 4729 | } |
6791875e N |
4730 | err = 0; |
4731 | unlock: | |
4732 | mddev_unlock(mddev); | |
4733 | return err ?: len; | |
e464eafd N |
4734 | } |
4735 | static struct md_sysfs_entry md_suspend_hi = | |
4736 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4737 | ||
08a02ecd | 4738 | static ssize_t |
fd01b88c | 4739 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4740 | { |
4741 | if (mddev->reshape_position != MaxSector) | |
4742 | return sprintf(page, "%llu\n", | |
4743 | (unsigned long long)mddev->reshape_position); | |
4744 | strcpy(page, "none\n"); | |
4745 | return 5; | |
4746 | } | |
4747 | ||
4748 | static ssize_t | |
fd01b88c | 4749 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4750 | { |
c6563a8c | 4751 | struct md_rdev *rdev; |
4c9309c0 | 4752 | unsigned long long new; |
6791875e | 4753 | int err; |
6791875e | 4754 | |
4c9309c0 AD |
4755 | err = kstrtoull(buf, 10, &new); |
4756 | if (err < 0) | |
4757 | return err; | |
4758 | if (new != (sector_t)new) | |
08a02ecd | 4759 | return -EINVAL; |
6791875e N |
4760 | err = mddev_lock(mddev); |
4761 | if (err) | |
4762 | return err; | |
4763 | err = -EBUSY; | |
4764 | if (mddev->pers) | |
4765 | goto unlock; | |
08a02ecd N |
4766 | mddev->reshape_position = new; |
4767 | mddev->delta_disks = 0; | |
2c810cdd | 4768 | mddev->reshape_backwards = 0; |
08a02ecd N |
4769 | mddev->new_level = mddev->level; |
4770 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4771 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4772 | rdev_for_each(rdev, mddev) |
4773 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4774 | err = 0; |
4775 | unlock: | |
4776 | mddev_unlock(mddev); | |
4777 | return err ?: len; | |
08a02ecd N |
4778 | } |
4779 | ||
4780 | static struct md_sysfs_entry md_reshape_position = | |
4781 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4782 | reshape_position_store); | |
4783 | ||
2c810cdd N |
4784 | static ssize_t |
4785 | reshape_direction_show(struct mddev *mddev, char *page) | |
4786 | { | |
4787 | return sprintf(page, "%s\n", | |
4788 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4789 | } | |
4790 | ||
4791 | static ssize_t | |
4792 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4793 | { | |
4794 | int backwards = 0; | |
6791875e N |
4795 | int err; |
4796 | ||
2c810cdd N |
4797 | if (cmd_match(buf, "forwards")) |
4798 | backwards = 0; | |
4799 | else if (cmd_match(buf, "backwards")) | |
4800 | backwards = 1; | |
4801 | else | |
4802 | return -EINVAL; | |
4803 | if (mddev->reshape_backwards == backwards) | |
4804 | return len; | |
4805 | ||
6791875e N |
4806 | err = mddev_lock(mddev); |
4807 | if (err) | |
4808 | return err; | |
2c810cdd N |
4809 | /* check if we are allowed to change */ |
4810 | if (mddev->delta_disks) | |
6791875e N |
4811 | err = -EBUSY; |
4812 | else if (mddev->persistent && | |
2c810cdd | 4813 | mddev->major_version == 0) |
6791875e N |
4814 | err = -EINVAL; |
4815 | else | |
4816 | mddev->reshape_backwards = backwards; | |
4817 | mddev_unlock(mddev); | |
4818 | return err ?: len; | |
2c810cdd N |
4819 | } |
4820 | ||
4821 | static struct md_sysfs_entry md_reshape_direction = | |
4822 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4823 | reshape_direction_store); | |
4824 | ||
b522adcd | 4825 | static ssize_t |
fd01b88c | 4826 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4827 | { |
4828 | if (mddev->external_size) | |
4829 | return sprintf(page, "%llu\n", | |
4830 | (unsigned long long)mddev->array_sectors/2); | |
4831 | else | |
4832 | return sprintf(page, "default\n"); | |
4833 | } | |
4834 | ||
4835 | static ssize_t | |
fd01b88c | 4836 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4837 | { |
4838 | sector_t sectors; | |
6791875e N |
4839 | int err; |
4840 | ||
4841 | err = mddev_lock(mddev); | |
4842 | if (err) | |
4843 | return err; | |
b522adcd | 4844 | |
ab5a98b1 GJ |
4845 | /* cluster raid doesn't support change array_sectors */ |
4846 | if (mddev_is_clustered(mddev)) | |
4847 | return -EINVAL; | |
4848 | ||
b522adcd DW |
4849 | if (strncmp(buf, "default", 7) == 0) { |
4850 | if (mddev->pers) | |
4851 | sectors = mddev->pers->size(mddev, 0, 0); | |
4852 | else | |
4853 | sectors = mddev->array_sectors; | |
4854 | ||
4855 | mddev->external_size = 0; | |
4856 | } else { | |
4857 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
4858 | err = -EINVAL; |
4859 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
4860 | err = -E2BIG; | |
4861 | else | |
4862 | mddev->external_size = 1; | |
b522adcd DW |
4863 | } |
4864 | ||
6791875e N |
4865 | if (!err) { |
4866 | mddev->array_sectors = sectors; | |
4867 | if (mddev->pers) { | |
4868 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4869 | revalidate_disk(mddev->gendisk); | |
4870 | } | |
cbe6ef1d | 4871 | } |
6791875e N |
4872 | mddev_unlock(mddev); |
4873 | return err ?: len; | |
b522adcd DW |
4874 | } |
4875 | ||
4876 | static struct md_sysfs_entry md_array_size = | |
4877 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4878 | array_size_store); | |
e464eafd | 4879 | |
eae1701f N |
4880 | static struct attribute *md_default_attrs[] = { |
4881 | &md_level.attr, | |
d4dbd025 | 4882 | &md_layout.attr, |
eae1701f | 4883 | &md_raid_disks.attr, |
3b34380a | 4884 | &md_chunk_size.attr, |
a35b0d69 | 4885 | &md_size.attr, |
a94213b1 | 4886 | &md_resync_start.attr, |
8bb93aac | 4887 | &md_metadata.attr, |
6d7ff738 | 4888 | &md_new_device.attr, |
16f17b39 | 4889 | &md_safe_delay.attr, |
9e653b63 | 4890 | &md_array_state.attr, |
08a02ecd | 4891 | &md_reshape_position.attr, |
2c810cdd | 4892 | &md_reshape_direction.attr, |
b522adcd | 4893 | &md_array_size.attr, |
1e50915f | 4894 | &max_corr_read_errors.attr, |
411036fa N |
4895 | NULL, |
4896 | }; | |
4897 | ||
4898 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4899 | &md_scan_mode.attr, |
c4a39551 | 4900 | &md_last_scan_mode.attr, |
9d88883e | 4901 | &md_mismatches.attr, |
88202a0c N |
4902 | &md_sync_min.attr, |
4903 | &md_sync_max.attr, | |
4904 | &md_sync_speed.attr, | |
90b08710 | 4905 | &md_sync_force_parallel.attr, |
88202a0c | 4906 | &md_sync_completed.attr, |
5e96ee65 | 4907 | &md_min_sync.attr, |
c6207277 | 4908 | &md_max_sync.attr, |
e464eafd N |
4909 | &md_suspend_lo.attr, |
4910 | &md_suspend_hi.attr, | |
9b1d1dac | 4911 | &md_bitmap.attr, |
d7f3d291 | 4912 | &md_degraded.attr, |
eae1701f N |
4913 | NULL, |
4914 | }; | |
411036fa N |
4915 | static struct attribute_group md_redundancy_group = { |
4916 | .name = NULL, | |
4917 | .attrs = md_redundancy_attrs, | |
4918 | }; | |
4919 | ||
eae1701f N |
4920 | static ssize_t |
4921 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4922 | { | |
4923 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4924 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4925 | ssize_t rv; |
eae1701f N |
4926 | |
4927 | if (!entry->show) | |
4928 | return -EIO; | |
af8a2434 N |
4929 | spin_lock(&all_mddevs_lock); |
4930 | if (list_empty(&mddev->all_mddevs)) { | |
4931 | spin_unlock(&all_mddevs_lock); | |
4932 | return -EBUSY; | |
4933 | } | |
4934 | mddev_get(mddev); | |
4935 | spin_unlock(&all_mddevs_lock); | |
4936 | ||
b7b17c9b | 4937 | rv = entry->show(mddev, page); |
af8a2434 | 4938 | mddev_put(mddev); |
96de1e66 | 4939 | return rv; |
eae1701f N |
4940 | } |
4941 | ||
4942 | static ssize_t | |
4943 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4944 | const char *page, size_t length) | |
4945 | { | |
4946 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4947 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4948 | ssize_t rv; |
eae1701f N |
4949 | |
4950 | if (!entry->store) | |
4951 | return -EIO; | |
67463acb N |
4952 | if (!capable(CAP_SYS_ADMIN)) |
4953 | return -EACCES; | |
af8a2434 N |
4954 | spin_lock(&all_mddevs_lock); |
4955 | if (list_empty(&mddev->all_mddevs)) { | |
4956 | spin_unlock(&all_mddevs_lock); | |
4957 | return -EBUSY; | |
4958 | } | |
4959 | mddev_get(mddev); | |
4960 | spin_unlock(&all_mddevs_lock); | |
6791875e | 4961 | rv = entry->store(mddev, page, length); |
af8a2434 | 4962 | mddev_put(mddev); |
96de1e66 | 4963 | return rv; |
eae1701f N |
4964 | } |
4965 | ||
4966 | static void md_free(struct kobject *ko) | |
4967 | { | |
fd01b88c | 4968 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4969 | |
4970 | if (mddev->sysfs_state) | |
4971 | sysfs_put(mddev->sysfs_state); | |
4972 | ||
6cd18e71 N |
4973 | if (mddev->queue) |
4974 | blk_cleanup_queue(mddev->queue); | |
a21d1504 N |
4975 | if (mddev->gendisk) { |
4976 | del_gendisk(mddev->gendisk); | |
4977 | put_disk(mddev->gendisk); | |
4978 | } | |
a21d1504 | 4979 | |
eae1701f N |
4980 | kfree(mddev); |
4981 | } | |
4982 | ||
52cf25d0 | 4983 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4984 | .show = md_attr_show, |
4985 | .store = md_attr_store, | |
4986 | }; | |
4987 | static struct kobj_type md_ktype = { | |
4988 | .release = md_free, | |
4989 | .sysfs_ops = &md_sysfs_ops, | |
4990 | .default_attrs = md_default_attrs, | |
4991 | }; | |
4992 | ||
1da177e4 LT |
4993 | int mdp_major = 0; |
4994 | ||
5fd3a17e DW |
4995 | static void mddev_delayed_delete(struct work_struct *ws) |
4996 | { | |
fd01b88c | 4997 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4998 | |
43a70507 | 4999 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
5000 | kobject_del(&mddev->kobj); |
5001 | kobject_put(&mddev->kobj); | |
5002 | } | |
5003 | ||
efeb53c0 | 5004 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 5005 | { |
48c9c27b | 5006 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 5007 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 5008 | struct gendisk *disk; |
efeb53c0 N |
5009 | int partitioned; |
5010 | int shift; | |
5011 | int unit; | |
3830c62f | 5012 | int error; |
1da177e4 LT |
5013 | |
5014 | if (!mddev) | |
efeb53c0 N |
5015 | return -ENODEV; |
5016 | ||
5017 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
5018 | shift = partitioned ? MdpMinorShift : 0; | |
5019 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 5020 | |
e804ac78 TH |
5021 | /* wait for any previous instance of this device to be |
5022 | * completely removed (mddev_delayed_delete). | |
d3374825 | 5023 | */ |
e804ac78 | 5024 | flush_workqueue(md_misc_wq); |
d3374825 | 5025 | |
48c9c27b | 5026 | mutex_lock(&disks_mutex); |
0909dc44 N |
5027 | error = -EEXIST; |
5028 | if (mddev->gendisk) | |
5029 | goto abort; | |
efeb53c0 N |
5030 | |
5031 | if (name) { | |
5032 | /* Need to ensure that 'name' is not a duplicate. | |
5033 | */ | |
fd01b88c | 5034 | struct mddev *mddev2; |
efeb53c0 N |
5035 | spin_lock(&all_mddevs_lock); |
5036 | ||
5037 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
5038 | if (mddev2->gendisk && | |
5039 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
5040 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 5041 | goto abort; |
efeb53c0 N |
5042 | } |
5043 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 5044 | } |
8b765398 | 5045 | |
0909dc44 | 5046 | error = -ENOMEM; |
8b765398 | 5047 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
5048 | if (!mddev->queue) |
5049 | goto abort; | |
409c57f3 N |
5050 | mddev->queue->queuedata = mddev; |
5051 | ||
409c57f3 | 5052 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 5053 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 5054 | |
1da177e4 LT |
5055 | disk = alloc_disk(1 << shift); |
5056 | if (!disk) { | |
8b765398 N |
5057 | blk_cleanup_queue(mddev->queue); |
5058 | mddev->queue = NULL; | |
0909dc44 | 5059 | goto abort; |
1da177e4 | 5060 | } |
efeb53c0 | 5061 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 5062 | disk->first_minor = unit << shift; |
efeb53c0 N |
5063 | if (name) |
5064 | strcpy(disk->disk_name, name); | |
5065 | else if (partitioned) | |
1da177e4 | 5066 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 5067 | else |
1da177e4 | 5068 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
5069 | disk->fops = &md_fops; |
5070 | disk->private_data = mddev; | |
5071 | disk->queue = mddev->queue; | |
56883a7e | 5072 | blk_queue_write_cache(mddev->queue, true, true); |
92850bbd | 5073 | /* Allow extended partitions. This makes the |
d3374825 | 5074 | * 'mdp' device redundant, but we can't really |
92850bbd N |
5075 | * remove it now. |
5076 | */ | |
5077 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 5078 | mddev->gendisk = disk; |
b0140891 N |
5079 | /* As soon as we call add_disk(), another thread could get |
5080 | * through to md_open, so make sure it doesn't get too far | |
5081 | */ | |
5082 | mutex_lock(&mddev->open_mutex); | |
5083 | add_disk(disk); | |
5084 | ||
ed9e1982 TH |
5085 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
5086 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
5087 | if (error) { |
5088 | /* This isn't possible, but as kobject_init_and_add is marked | |
5089 | * __must_check, we must do something with the result | |
5090 | */ | |
9d48739e N |
5091 | pr_debug("md: cannot register %s/md - name in use\n", |
5092 | disk->disk_name); | |
0909dc44 N |
5093 | error = 0; |
5094 | } | |
00bcb4ac N |
5095 | if (mddev->kobj.sd && |
5096 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
9d48739e | 5097 | pr_debug("pointless warning\n"); |
b0140891 | 5098 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
5099 | abort: |
5100 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 5101 | if (!error && mddev->kobj.sd) { |
3830c62f | 5102 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 5103 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 5104 | } |
d3374825 | 5105 | mddev_put(mddev); |
0909dc44 | 5106 | return error; |
efeb53c0 N |
5107 | } |
5108 | ||
5109 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
5110 | { | |
5111 | md_alloc(dev, NULL); | |
1da177e4 LT |
5112 | return NULL; |
5113 | } | |
5114 | ||
efeb53c0 N |
5115 | static int add_named_array(const char *val, struct kernel_param *kp) |
5116 | { | |
5117 | /* val must be "md_*" where * is not all digits. | |
5118 | * We allocate an array with a large free minor number, and | |
5119 | * set the name to val. val must not already be an active name. | |
5120 | */ | |
5121 | int len = strlen(val); | |
5122 | char buf[DISK_NAME_LEN]; | |
5123 | ||
5124 | while (len && val[len-1] == '\n') | |
5125 | len--; | |
5126 | if (len >= DISK_NAME_LEN) | |
5127 | return -E2BIG; | |
5128 | strlcpy(buf, val, len+1); | |
5129 | if (strncmp(buf, "md_", 3) != 0) | |
5130 | return -EINVAL; | |
5131 | return md_alloc(0, buf); | |
5132 | } | |
5133 | ||
1da177e4 LT |
5134 | static void md_safemode_timeout(unsigned long data) |
5135 | { | |
fd01b88c | 5136 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 5137 | |
0fd62b86 NB |
5138 | if (!atomic_read(&mddev->writes_pending)) { |
5139 | mddev->safemode = 1; | |
5140 | if (mddev->external) | |
00bcb4ac | 5141 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 5142 | } |
1da177e4 LT |
5143 | md_wakeup_thread(mddev->thread); |
5144 | } | |
5145 | ||
6ff8d8ec | 5146 | static int start_dirty_degraded; |
1da177e4 | 5147 | |
fd01b88c | 5148 | int md_run(struct mddev *mddev) |
1da177e4 | 5149 | { |
2604b703 | 5150 | int err; |
3cb03002 | 5151 | struct md_rdev *rdev; |
84fc4b56 | 5152 | struct md_personality *pers; |
1da177e4 | 5153 | |
a757e64c N |
5154 | if (list_empty(&mddev->disks)) |
5155 | /* cannot run an array with no devices.. */ | |
1da177e4 | 5156 | return -EINVAL; |
1da177e4 LT |
5157 | |
5158 | if (mddev->pers) | |
5159 | return -EBUSY; | |
bb4f1e9d N |
5160 | /* Cannot run until previous stop completes properly */ |
5161 | if (mddev->sysfs_active) | |
5162 | return -EBUSY; | |
b6eb127d | 5163 | |
1da177e4 LT |
5164 | /* |
5165 | * Analyze all RAID superblock(s) | |
5166 | */ | |
1ec4a939 N |
5167 | if (!mddev->raid_disks) { |
5168 | if (!mddev->persistent) | |
5169 | return -EINVAL; | |
a757e64c | 5170 | analyze_sbs(mddev); |
1ec4a939 | 5171 | } |
1da177e4 | 5172 | |
d9d166c2 N |
5173 | if (mddev->level != LEVEL_NONE) |
5174 | request_module("md-level-%d", mddev->level); | |
5175 | else if (mddev->clevel[0]) | |
5176 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
5177 | |
5178 | /* | |
5179 | * Drop all container device buffers, from now on | |
5180 | * the only valid external interface is through the md | |
5181 | * device. | |
1da177e4 | 5182 | */ |
dafb20fa | 5183 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 5184 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5185 | continue; |
5186 | sync_blockdev(rdev->bdev); | |
f98393a6 | 5187 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
5188 | |
5189 | /* perform some consistency tests on the device. | |
5190 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 5191 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 5192 | */ |
a6ff7e08 JB |
5193 | if (rdev->meta_bdev) { |
5194 | /* Nothing to check */; | |
5195 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5196 | if (mddev->dev_sectors && |
5197 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5198 | > rdev->sb_start) { |
9d48739e N |
5199 | pr_warn("md: %s: data overlaps metadata\n", |
5200 | mdname(mddev)); | |
f0d76d70 N |
5201 | return -EINVAL; |
5202 | } | |
5203 | } else { | |
0f420358 | 5204 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 | 5205 | > rdev->data_offset) { |
9d48739e N |
5206 | pr_warn("md: %s: metadata overlaps data\n", |
5207 | mdname(mddev)); | |
f0d76d70 N |
5208 | return -EINVAL; |
5209 | } | |
5210 | } | |
00bcb4ac | 5211 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5212 | } |
5213 | ||
10273170 | 5214 | if (mddev->bio_set == NULL) { |
395c72a7 | 5215 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
10273170 ML |
5216 | if (!mddev->bio_set) |
5217 | return -ENOMEM; | |
5218 | } | |
a167f663 | 5219 | |
1da177e4 | 5220 | spin_lock(&pers_lock); |
d9d166c2 | 5221 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5222 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5223 | spin_unlock(&pers_lock); |
d9d166c2 | 5224 | if (mddev->level != LEVEL_NONE) |
9d48739e N |
5225 | pr_warn("md: personality for level %d is not loaded!\n", |
5226 | mddev->level); | |
d9d166c2 | 5227 | else |
9d48739e N |
5228 | pr_warn("md: personality for level %s is not loaded!\n", |
5229 | mddev->clevel); | |
1da177e4 LT |
5230 | return -EINVAL; |
5231 | } | |
1da177e4 | 5232 | spin_unlock(&pers_lock); |
34817e8c N |
5233 | if (mddev->level != pers->level) { |
5234 | mddev->level = pers->level; | |
5235 | mddev->new_level = pers->level; | |
5236 | } | |
d9d166c2 | 5237 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5238 | |
f6705578 | 5239 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5240 | pers->start_reshape == NULL) { |
f6705578 | 5241 | /* This personality cannot handle reshaping... */ |
f6705578 N |
5242 | module_put(pers->owner); |
5243 | return -EINVAL; | |
5244 | } | |
5245 | ||
7dd5e7c3 N |
5246 | if (pers->sync_request) { |
5247 | /* Warn if this is a potentially silly | |
5248 | * configuration. | |
5249 | */ | |
5250 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5251 | struct md_rdev *rdev2; |
7dd5e7c3 | 5252 | int warned = 0; |
159ec1fc | 5253 | |
dafb20fa N |
5254 | rdev_for_each(rdev, mddev) |
5255 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5256 | if (rdev < rdev2 && |
5257 | rdev->bdev->bd_contains == | |
5258 | rdev2->bdev->bd_contains) { | |
9d48739e N |
5259 | pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n", |
5260 | mdname(mddev), | |
5261 | bdevname(rdev->bdev,b), | |
5262 | bdevname(rdev2->bdev,b2)); | |
7dd5e7c3 N |
5263 | warned = 1; |
5264 | } | |
5265 | } | |
159ec1fc | 5266 | |
7dd5e7c3 | 5267 | if (warned) |
9d48739e | 5268 | pr_warn("True protection against single-disk failure might be compromised.\n"); |
7dd5e7c3 N |
5269 | } |
5270 | ||
657390d2 | 5271 | mddev->recovery = 0; |
58c0fed4 AN |
5272 | /* may be over-ridden by personality */ |
5273 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5274 | ||
6ff8d8ec | 5275 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5276 | |
0f9552b5 | 5277 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5278 | mddev->ro = 2; /* read-only, but switch on first write */ |
5279 | ||
a85dd7b8 SL |
5280 | /* |
5281 | * NOTE: some pers->run(), for example r5l_recovery_log(), wakes | |
5282 | * up mddev->thread. It is important to initialize critical | |
5283 | * resources for mddev->thread BEFORE calling pers->run(). | |
5284 | */ | |
36d091f4 | 5285 | err = pers->run(mddev); |
13e53df3 | 5286 | if (err) |
9d48739e | 5287 | pr_warn("md: pers->run() failed ...\n"); |
36d091f4 | 5288 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
9d48739e N |
5289 | WARN_ONCE(!mddev->external_size, |
5290 | "%s: default size too small, but 'external_size' not in effect?\n", | |
5291 | __func__); | |
5292 | pr_warn("md: invalid array_size %llu > default size %llu\n", | |
5293 | (unsigned long long)mddev->array_sectors / 2, | |
5294 | (unsigned long long)pers->size(mddev, 0, 0) / 2); | |
b522adcd | 5295 | err = -EINVAL; |
b522adcd | 5296 | } |
36d091f4 | 5297 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5298 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
f9209a32 GR |
5299 | struct bitmap *bitmap; |
5300 | ||
5301 | bitmap = bitmap_create(mddev, -1); | |
5302 | if (IS_ERR(bitmap)) { | |
5303 | err = PTR_ERR(bitmap); | |
9d48739e N |
5304 | pr_warn("%s: failed to create bitmap (%d)\n", |
5305 | mdname(mddev), err); | |
f9209a32 GR |
5306 | } else |
5307 | mddev->bitmap = bitmap; | |
5308 | ||
b15c2e57 | 5309 | } |
1da177e4 | 5310 | if (err) { |
5aa61f42 | 5311 | mddev_detach(mddev); |
0c35bd47 N |
5312 | if (mddev->private) |
5313 | pers->free(mddev, mddev->private); | |
bd691922 | 5314 | mddev->private = NULL; |
36d091f4 | 5315 | module_put(pers->owner); |
32a7627c N |
5316 | bitmap_destroy(mddev); |
5317 | return err; | |
1da177e4 | 5318 | } |
5c675f83 | 5319 | if (mddev->queue) { |
bb086a89 SL |
5320 | bool nonrot = true; |
5321 | ||
5322 | rdev_for_each(rdev, mddev) { | |
5323 | if (rdev->raid_disk >= 0 && | |
5324 | !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
5325 | nonrot = false; | |
5326 | break; | |
5327 | } | |
5328 | } | |
5329 | if (mddev->degraded) | |
5330 | nonrot = false; | |
5331 | if (nonrot) | |
5332 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue); | |
5333 | else | |
5334 | queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue); | |
dc3b17cc JK |
5335 | mddev->queue->backing_dev_info->congested_data = mddev; |
5336 | mddev->queue->backing_dev_info->congested_fn = md_congested; | |
5c675f83 | 5337 | } |
36d091f4 | 5338 | if (pers->sync_request) { |
00bcb4ac N |
5339 | if (mddev->kobj.sd && |
5340 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
5341 | pr_warn("md: cannot register extra attributes for %s\n", |
5342 | mdname(mddev)); | |
00bcb4ac | 5343 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5344 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5345 | mddev->ro = 0; |
5346 | ||
f72ffdd6 | 5347 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
5348 | atomic_set(&mddev->max_corr_read_errors, |
5349 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 | 5350 | mddev->safemode = 0; |
28c1b9fd GR |
5351 | if (mddev_is_clustered(mddev)) |
5352 | mddev->safemode_delay = 0; | |
5353 | else | |
5354 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ | |
1da177e4 | 5355 | mddev->in_sync = 1; |
0ca69886 | 5356 | smp_wmb(); |
36d091f4 N |
5357 | spin_lock(&mddev->lock); |
5358 | mddev->pers = pers; | |
36d091f4 | 5359 | spin_unlock(&mddev->lock); |
dafb20fa | 5360 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5361 | if (rdev->raid_disk >= 0) |
5362 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5363 | /* failure here is OK */; |
f72ffdd6 | 5364 | |
a4a3d26d N |
5365 | if (mddev->degraded && !mddev->ro) |
5366 | /* This ensures that recovering status is reported immediately | |
5367 | * via sysfs - until a lack of spares is confirmed. | |
5368 | */ | |
5369 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
1da177e4 | 5370 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5371 | |
2953079c | 5372 | if (mddev->sb_flags) |
850b2b42 | 5373 | md_update_sb(mddev, 0); |
1da177e4 | 5374 | |
d7603b7e | 5375 | md_new_event(mddev); |
00bcb4ac N |
5376 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5377 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5378 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5379 | return 0; |
5380 | } | |
390ee602 | 5381 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5382 | |
fd01b88c | 5383 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5384 | { |
5385 | int err; | |
5386 | ||
5387 | err = md_run(mddev); | |
5388 | if (err) | |
5389 | goto out; | |
69e51b44 N |
5390 | err = bitmap_load(mddev); |
5391 | if (err) { | |
5392 | bitmap_destroy(mddev); | |
5393 | goto out; | |
5394 | } | |
0fd018af | 5395 | |
28c1b9fd GR |
5396 | if (mddev_is_clustered(mddev)) |
5397 | md_allow_write(mddev); | |
5398 | ||
0fd018af JB |
5399 | md_wakeup_thread(mddev->thread); |
5400 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5401 | ||
fe60b014 N |
5402 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5403 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5404 | mddev->changed = 1; |
fe60b014 N |
5405 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5406 | out: | |
5407 | return err; | |
5408 | } | |
5409 | ||
fd01b88c | 5410 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5411 | { |
5412 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 5413 | |
80fab1d7 | 5414 | /* Complain if it has no devices */ |
1da177e4 | 5415 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5416 | return -ENXIO; |
5417 | if (!mddev->pers) | |
5418 | return -EINVAL; | |
5419 | if (!mddev->ro) | |
5420 | return -EBUSY; | |
339421de SL |
5421 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
5422 | struct md_rdev *rdev; | |
5423 | bool has_journal = false; | |
5424 | ||
5425 | rcu_read_lock(); | |
5426 | rdev_for_each_rcu(rdev, mddev) { | |
5427 | if (test_bit(Journal, &rdev->flags) && | |
5428 | !test_bit(Faulty, &rdev->flags)) { | |
5429 | has_journal = true; | |
5430 | break; | |
5431 | } | |
5432 | } | |
5433 | rcu_read_unlock(); | |
5434 | ||
5435 | /* Don't restart rw with journal missing/faulty */ | |
5436 | if (!has_journal) | |
5437 | return -EINVAL; | |
5438 | } | |
5439 | ||
80fab1d7 AN |
5440 | mddev->safemode = 0; |
5441 | mddev->ro = 0; | |
5442 | set_disk_ro(disk, 0); | |
9d48739e | 5443 | pr_debug("md: %s switched to read-write mode.\n", mdname(mddev)); |
80fab1d7 AN |
5444 | /* Kick recovery or resync if necessary */ |
5445 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5446 | md_wakeup_thread(mddev->thread); | |
5447 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5448 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5449 | return 0; |
1da177e4 LT |
5450 | } |
5451 | ||
fd01b88c | 5452 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5453 | { |
5454 | mddev->array_sectors = 0; | |
5455 | mddev->external_size = 0; | |
5456 | mddev->dev_sectors = 0; | |
5457 | mddev->raid_disks = 0; | |
5458 | mddev->recovery_cp = 0; | |
5459 | mddev->resync_min = 0; | |
5460 | mddev->resync_max = MaxSector; | |
5461 | mddev->reshape_position = MaxSector; | |
5462 | mddev->external = 0; | |
5463 | mddev->persistent = 0; | |
5464 | mddev->level = LEVEL_NONE; | |
5465 | mddev->clevel[0] = 0; | |
5466 | mddev->flags = 0; | |
2953079c | 5467 | mddev->sb_flags = 0; |
6177b472 N |
5468 | mddev->ro = 0; |
5469 | mddev->metadata_type[0] = 0; | |
5470 | mddev->chunk_sectors = 0; | |
5471 | mddev->ctime = mddev->utime = 0; | |
5472 | mddev->layout = 0; | |
5473 | mddev->max_disks = 0; | |
5474 | mddev->events = 0; | |
a8707c08 | 5475 | mddev->can_decrease_events = 0; |
6177b472 | 5476 | mddev->delta_disks = 0; |
2c810cdd | 5477 | mddev->reshape_backwards = 0; |
6177b472 N |
5478 | mddev->new_level = LEVEL_NONE; |
5479 | mddev->new_layout = 0; | |
5480 | mddev->new_chunk_sectors = 0; | |
5481 | mddev->curr_resync = 0; | |
7f7583d4 | 5482 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5483 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5484 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5485 | mddev->recovery = 0; | |
5486 | mddev->in_sync = 0; | |
f0b4f7e2 | 5487 | mddev->changed = 0; |
6177b472 | 5488 | mddev->degraded = 0; |
6177b472 | 5489 | mddev->safemode = 0; |
bd691922 | 5490 | mddev->private = NULL; |
c20c33f0 | 5491 | mddev->cluster_info = NULL; |
6177b472 N |
5492 | mddev->bitmap_info.offset = 0; |
5493 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5494 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5495 | mddev->bitmap_info.chunksize = 0; |
5496 | mddev->bitmap_info.daemon_sleep = 0; | |
5497 | mddev->bitmap_info.max_write_behind = 0; | |
c20c33f0 | 5498 | mddev->bitmap_info.nodes = 0; |
6177b472 N |
5499 | } |
5500 | ||
fd01b88c | 5501 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5502 | { |
6b6204ee | 5503 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5504 | flush_workqueue(md_misc_wq); |
a047e125 | 5505 | if (mddev->sync_thread) { |
a047e125 | 5506 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5507 | md_reap_sync_thread(mddev); |
a047e125 N |
5508 | } |
5509 | ||
5510 | del_timer_sync(&mddev->safemode_timer); | |
5511 | ||
034e33f5 SL |
5512 | if (mddev->pers && mddev->pers->quiesce) { |
5513 | mddev->pers->quiesce(mddev, 1); | |
5514 | mddev->pers->quiesce(mddev, 0); | |
5515 | } | |
a047e125 | 5516 | bitmap_flush(mddev); |
a047e125 | 5517 | |
b6d428c6 | 5518 | if (mddev->ro == 0 && |
28c1b9fd | 5519 | ((!mddev->in_sync && !mddev_is_clustered(mddev)) || |
2953079c | 5520 | mddev->sb_flags)) { |
a047e125 | 5521 | /* mark array as shutdown cleanly */ |
28c1b9fd GR |
5522 | if (!mddev_is_clustered(mddev)) |
5523 | mddev->in_sync = 1; | |
a047e125 N |
5524 | md_update_sb(mddev, 1); |
5525 | } | |
5526 | } | |
defad61a | 5527 | |
fd01b88c | 5528 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5529 | { |
29f097c4 | 5530 | mddev_lock_nointr(mddev); |
defad61a N |
5531 | __md_stop_writes(mddev); |
5532 | mddev_unlock(mddev); | |
5533 | } | |
390ee602 | 5534 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5535 | |
5aa61f42 N |
5536 | static void mddev_detach(struct mddev *mddev) |
5537 | { | |
5538 | struct bitmap *bitmap = mddev->bitmap; | |
5539 | /* wait for behind writes to complete */ | |
5540 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { | |
9d48739e N |
5541 | pr_debug("md:%s: behind writes in progress - waiting to stop.\n", |
5542 | mdname(mddev)); | |
5aa61f42 N |
5543 | /* need to kick something here to make sure I/O goes? */ |
5544 | wait_event(bitmap->behind_wait, | |
5545 | atomic_read(&bitmap->behind_writes) == 0); | |
5546 | } | |
36d091f4 | 5547 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5548 | mddev->pers->quiesce(mddev, 1); |
5549 | mddev->pers->quiesce(mddev, 0); | |
5550 | } | |
5551 | md_unregister_thread(&mddev->thread); | |
5552 | if (mddev->queue) | |
5553 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5554 | } | |
5555 | ||
5eff3c43 | 5556 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5557 | { |
36d091f4 | 5558 | struct md_personality *pers = mddev->pers; |
5aa61f42 | 5559 | mddev_detach(mddev); |
ee5d004f N |
5560 | /* Ensure ->event_work is done */ |
5561 | flush_workqueue(md_misc_wq); | |
36d091f4 | 5562 | spin_lock(&mddev->lock); |
6177b472 | 5563 | mddev->pers = NULL; |
36d091f4 N |
5564 | spin_unlock(&mddev->lock); |
5565 | pers->free(mddev, mddev->private); | |
bd691922 | 5566 | mddev->private = NULL; |
36d091f4 N |
5567 | if (pers->sync_request && mddev->to_remove == NULL) |
5568 | mddev->to_remove = &md_redundancy_group; | |
5569 | module_put(pers->owner); | |
cca9cf90 | 5570 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5571 | } |
5eff3c43 N |
5572 | |
5573 | void md_stop(struct mddev *mddev) | |
5574 | { | |
5575 | /* stop the array and free an attached data structures. | |
5576 | * This is called from dm-raid | |
5577 | */ | |
5578 | __md_stop(mddev); | |
5579 | bitmap_destroy(mddev); | |
5580 | if (mddev->bio_set) | |
5581 | bioset_free(mddev->bio_set); | |
5582 | } | |
5583 | ||
390ee602 | 5584 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5585 | |
a05b7ea0 | 5586 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5587 | { |
5588 | int err = 0; | |
30b8feb7 N |
5589 | int did_freeze = 0; |
5590 | ||
5591 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5592 | did_freeze = 1; | |
5593 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5594 | md_wakeup_thread(mddev->thread); | |
5595 | } | |
f851b60d | 5596 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5597 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5598 | if (mddev->sync_thread) |
30b8feb7 N |
5599 | /* Thread might be blocked waiting for metadata update |
5600 | * which will now never happen */ | |
5601 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5602 | |
2953079c | 5603 | if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
88724bfa | 5604 | return -EBUSY; |
30b8feb7 | 5605 | mddev_unlock(mddev); |
f851b60d N |
5606 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5607 | &mddev->recovery)); | |
88724bfa | 5608 | wait_event(mddev->sb_wait, |
2953079c | 5609 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
30b8feb7 N |
5610 | mddev_lock_nointr(mddev); |
5611 | ||
a4bd82d0 | 5612 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5613 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5614 | mddev->sync_thread || |
af8d8e6f | 5615 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5616 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5617 | if (did_freeze) { |
5618 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5619 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5620 | md_wakeup_thread(mddev->thread); |
5621 | } | |
a4bd82d0 N |
5622 | err = -EBUSY; |
5623 | goto out; | |
5624 | } | |
5625 | if (mddev->pers) { | |
defad61a | 5626 | __md_stop_writes(mddev); |
a4bd82d0 N |
5627 | |
5628 | err = -ENXIO; | |
5629 | if (mddev->ro==1) | |
5630 | goto out; | |
5631 | mddev->ro = 1; | |
5632 | set_disk_ro(mddev->gendisk, 1); | |
5633 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d N |
5634 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5635 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5636 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5637 | err = 0; |
a4bd82d0 N |
5638 | } |
5639 | out: | |
5640 | mutex_unlock(&mddev->open_mutex); | |
5641 | return err; | |
5642 | } | |
5643 | ||
9e653b63 N |
5644 | /* mode: |
5645 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5646 | * 2 - stop but do not disassemble array |
5647 | */ | |
f72ffdd6 | 5648 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5649 | struct block_device *bdev) |
1da177e4 | 5650 | { |
1da177e4 | 5651 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5652 | struct md_rdev *rdev; |
30b8feb7 N |
5653 | int did_freeze = 0; |
5654 | ||
5655 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5656 | did_freeze = 1; | |
5657 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5658 | md_wakeup_thread(mddev->thread); | |
5659 | } | |
f851b60d | 5660 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5661 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5662 | if (mddev->sync_thread) |
30b8feb7 N |
5663 | /* Thread might be blocked waiting for metadata update |
5664 | * which will now never happen */ | |
5665 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5666 | |
30b8feb7 | 5667 | mddev_unlock(mddev); |
f851b60d N |
5668 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
5669 | !test_bit(MD_RECOVERY_RUNNING, | |
5670 | &mddev->recovery))); | |
30b8feb7 | 5671 | mddev_lock_nointr(mddev); |
1da177e4 | 5672 | |
c8c00a69 | 5673 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5674 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5675 | mddev->sysfs_active || |
5676 | mddev->sync_thread || | |
af8d8e6f | 5677 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5678 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5679 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5680 | if (did_freeze) { |
5681 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5682 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5683 | md_wakeup_thread(mddev->thread); |
5684 | } | |
260fa034 N |
5685 | return -EBUSY; |
5686 | } | |
6e17b027 | 5687 | if (mddev->pers) { |
a4bd82d0 N |
5688 | if (mddev->ro) |
5689 | set_disk_ro(disk, 0); | |
409c57f3 | 5690 | |
defad61a | 5691 | __md_stop_writes(mddev); |
5eff3c43 | 5692 | __md_stop(mddev); |
dc3b17cc | 5693 | mddev->queue->backing_dev_info->congested_fn = NULL; |
6177b472 | 5694 | |
a4bd82d0 | 5695 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5696 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5697 | |
dafb20fa | 5698 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5699 | if (rdev->raid_disk >= 0) |
5700 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5701 | |
a4bd82d0 | 5702 | set_capacity(disk, 0); |
6e17b027 | 5703 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5704 | mddev->changed = 1; |
a4bd82d0 | 5705 | revalidate_disk(disk); |
0d4ca600 | 5706 | |
a4bd82d0 N |
5707 | if (mddev->ro) |
5708 | mddev->ro = 0; | |
6e17b027 N |
5709 | } else |
5710 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5711 | /* |
5712 | * Free resources if final stop | |
5713 | */ | |
9e653b63 | 5714 | if (mode == 0) { |
9d48739e | 5715 | pr_info("md: %s stopped.\n", mdname(mddev)); |
1da177e4 | 5716 | |
978f946b | 5717 | bitmap_destroy(mddev); |
c3d9714e | 5718 | if (mddev->bitmap_info.file) { |
4af1a041 N |
5719 | struct file *f = mddev->bitmap_info.file; |
5720 | spin_lock(&mddev->lock); | |
c3d9714e | 5721 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
5722 | spin_unlock(&mddev->lock); |
5723 | fput(f); | |
978f946b | 5724 | } |
c3d9714e | 5725 | mddev->bitmap_info.offset = 0; |
978f946b | 5726 | |
1da177e4 LT |
5727 | export_array(mddev); |
5728 | ||
6177b472 | 5729 | md_clean(mddev); |
efeb53c0 N |
5730 | if (mddev->hold_active == UNTIL_STOP) |
5731 | mddev->hold_active = 0; | |
a4bd82d0 | 5732 | } |
d7603b7e | 5733 | md_new_event(mddev); |
00bcb4ac | 5734 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5735 | return 0; |
1da177e4 LT |
5736 | } |
5737 | ||
fdee8ae4 | 5738 | #ifndef MODULE |
fd01b88c | 5739 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5740 | { |
3cb03002 | 5741 | struct md_rdev *rdev; |
1da177e4 LT |
5742 | int err; |
5743 | ||
a757e64c | 5744 | if (list_empty(&mddev->disks)) |
1da177e4 | 5745 | return; |
1da177e4 | 5746 | |
9d48739e | 5747 | pr_info("md: running: "); |
1da177e4 | 5748 | |
dafb20fa | 5749 | rdev_for_each(rdev, mddev) { |
1da177e4 | 5750 | char b[BDEVNAME_SIZE]; |
9d48739e | 5751 | pr_cont("<%s>", bdevname(rdev->bdev,b)); |
1da177e4 | 5752 | } |
9d48739e | 5753 | pr_cont("\n"); |
1da177e4 | 5754 | |
d710e138 | 5755 | err = do_md_run(mddev); |
1da177e4 | 5756 | if (err) { |
9d48739e | 5757 | pr_warn("md: do_md_run() returned %d\n", err); |
a05b7ea0 | 5758 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5759 | } |
5760 | } | |
5761 | ||
5762 | /* | |
5763 | * lets try to run arrays based on all disks that have arrived | |
5764 | * until now. (those are in pending_raid_disks) | |
5765 | * | |
5766 | * the method: pick the first pending disk, collect all disks with | |
5767 | * the same UUID, remove all from the pending list and put them into | |
5768 | * the 'same_array' list. Then order this list based on superblock | |
5769 | * update time (freshest comes first), kick out 'old' disks and | |
5770 | * compare superblocks. If everything's fine then run it. | |
5771 | * | |
5772 | * If "unit" is allocated, then bump its reference count | |
5773 | */ | |
5774 | static void autorun_devices(int part) | |
5775 | { | |
3cb03002 | 5776 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5777 | struct mddev *mddev; |
1da177e4 LT |
5778 | char b[BDEVNAME_SIZE]; |
5779 | ||
9d48739e | 5780 | pr_info("md: autorun ...\n"); |
1da177e4 | 5781 | while (!list_empty(&pending_raid_disks)) { |
e8703fe1 | 5782 | int unit; |
1da177e4 | 5783 | dev_t dev; |
ad01c9e3 | 5784 | LIST_HEAD(candidates); |
1da177e4 | 5785 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5786 | struct md_rdev, same_set); |
1da177e4 | 5787 | |
9d48739e | 5788 | pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b)); |
1da177e4 | 5789 | INIT_LIST_HEAD(&candidates); |
159ec1fc | 5790 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 | 5791 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
9d48739e N |
5792 | pr_debug("md: adding %s ...\n", |
5793 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
5794 | list_move(&rdev->same_set, &candidates); |
5795 | } | |
5796 | /* | |
5797 | * now we have a set of devices, with all of them having | |
5798 | * mostly sane superblocks. It's time to allocate the | |
5799 | * mddev. | |
5800 | */ | |
e8703fe1 N |
5801 | if (part) { |
5802 | dev = MKDEV(mdp_major, | |
5803 | rdev0->preferred_minor << MdpMinorShift); | |
5804 | unit = MINOR(dev) >> MdpMinorShift; | |
5805 | } else { | |
5806 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5807 | unit = MINOR(dev); | |
5808 | } | |
5809 | if (rdev0->preferred_minor != unit) { | |
9d48739e N |
5810 | pr_warn("md: unit number in %s is bad: %d\n", |
5811 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
1da177e4 LT |
5812 | break; |
5813 | } | |
1da177e4 LT |
5814 | |
5815 | md_probe(dev, NULL, NULL); | |
5816 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5817 | if (!mddev || !mddev->gendisk) { |
5818 | if (mddev) | |
5819 | mddev_put(mddev); | |
1da177e4 LT |
5820 | break; |
5821 | } | |
f72ffdd6 | 5822 | if (mddev_lock(mddev)) |
9d48739e | 5823 | pr_warn("md: %s locked, cannot run\n", mdname(mddev)); |
1da177e4 LT |
5824 | else if (mddev->raid_disks || mddev->major_version |
5825 | || !list_empty(&mddev->disks)) { | |
9d48739e | 5826 | pr_warn("md: %s already running, cannot run %s\n", |
1da177e4 LT |
5827 | mdname(mddev), bdevname(rdev0->bdev,b)); |
5828 | mddev_unlock(mddev); | |
5829 | } else { | |
9d48739e | 5830 | pr_debug("md: created %s\n", mdname(mddev)); |
1ec4a939 | 5831 | mddev->persistent = 1; |
159ec1fc | 5832 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5833 | list_del_init(&rdev->same_set); |
5834 | if (bind_rdev_to_array(rdev, mddev)) | |
5835 | export_rdev(rdev); | |
5836 | } | |
5837 | autorun_array(mddev); | |
5838 | mddev_unlock(mddev); | |
5839 | } | |
5840 | /* on success, candidates will be empty, on error | |
5841 | * it won't... | |
5842 | */ | |
159ec1fc | 5843 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5844 | list_del_init(&rdev->same_set); |
1da177e4 | 5845 | export_rdev(rdev); |
4b80991c | 5846 | } |
1da177e4 LT |
5847 | mddev_put(mddev); |
5848 | } | |
9d48739e | 5849 | pr_info("md: ... autorun DONE.\n"); |
1da177e4 | 5850 | } |
fdee8ae4 | 5851 | #endif /* !MODULE */ |
1da177e4 | 5852 | |
f72ffdd6 | 5853 | static int get_version(void __user *arg) |
1da177e4 LT |
5854 | { |
5855 | mdu_version_t ver; | |
5856 | ||
5857 | ver.major = MD_MAJOR_VERSION; | |
5858 | ver.minor = MD_MINOR_VERSION; | |
5859 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5860 | ||
5861 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5862 | return -EFAULT; | |
5863 | ||
5864 | return 0; | |
5865 | } | |
5866 | ||
f72ffdd6 | 5867 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
5868 | { |
5869 | mdu_array_info_t info; | |
a9f326eb | 5870 | int nr,working,insync,failed,spare; |
3cb03002 | 5871 | struct md_rdev *rdev; |
1da177e4 | 5872 | |
1ca69c4b N |
5873 | nr = working = insync = failed = spare = 0; |
5874 | rcu_read_lock(); | |
5875 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5876 | nr++; |
b2d444d7 | 5877 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5878 | failed++; |
5879 | else { | |
5880 | working++; | |
b2d444d7 | 5881 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 5882 | insync++; |
b347af81 SL |
5883 | else if (test_bit(Journal, &rdev->flags)) |
5884 | /* TODO: add journal count to md_u.h */ | |
5885 | ; | |
1da177e4 LT |
5886 | else |
5887 | spare++; | |
5888 | } | |
5889 | } | |
1ca69c4b | 5890 | rcu_read_unlock(); |
1da177e4 LT |
5891 | |
5892 | info.major_version = mddev->major_version; | |
5893 | info.minor_version = mddev->minor_version; | |
5894 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 5895 | info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 5896 | info.level = mddev->level; |
58c0fed4 AN |
5897 | info.size = mddev->dev_sectors / 2; |
5898 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5899 | info.size = -1; |
1da177e4 LT |
5900 | info.nr_disks = nr; |
5901 | info.raid_disks = mddev->raid_disks; | |
5902 | info.md_minor = mddev->md_minor; | |
5903 | info.not_persistent= !mddev->persistent; | |
5904 | ||
9ebc6ef1 | 5905 | info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
5906 | info.state = 0; |
5907 | if (mddev->in_sync) | |
5908 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5909 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 5910 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
ca8895d9 GR |
5911 | if (mddev_is_clustered(mddev)) |
5912 | info.state |= (1<<MD_SB_CLUSTERED); | |
a9f326eb | 5913 | info.active_disks = insync; |
1da177e4 LT |
5914 | info.working_disks = working; |
5915 | info.failed_disks = failed; | |
5916 | info.spare_disks = spare; | |
5917 | ||
5918 | info.layout = mddev->layout; | |
9d8f0363 | 5919 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5920 | |
5921 | if (copy_to_user(arg, &info, sizeof(info))) | |
5922 | return -EFAULT; | |
5923 | ||
5924 | return 0; | |
5925 | } | |
5926 | ||
f72ffdd6 | 5927 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
5928 | { |
5929 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 5930 | char *ptr; |
4af1a041 | 5931 | int err; |
32a7627c | 5932 | |
b6878d9e | 5933 | file = kzalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 5934 | if (!file) |
4af1a041 | 5935 | return -ENOMEM; |
32a7627c | 5936 | |
4af1a041 N |
5937 | err = 0; |
5938 | spin_lock(&mddev->lock); | |
25eafe1a BR |
5939 | /* bitmap enabled */ |
5940 | if (mddev->bitmap_info.file) { | |
5941 | ptr = file_path(mddev->bitmap_info.file, file->pathname, | |
5942 | sizeof(file->pathname)); | |
5943 | if (IS_ERR(ptr)) | |
5944 | err = PTR_ERR(ptr); | |
5945 | else | |
5946 | memmove(file->pathname, ptr, | |
5947 | sizeof(file->pathname)-(ptr-file->pathname)); | |
5948 | } | |
4af1a041 | 5949 | spin_unlock(&mddev->lock); |
32a7627c | 5950 | |
4af1a041 N |
5951 | if (err == 0 && |
5952 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 5953 | err = -EFAULT; |
4af1a041 | 5954 | |
32a7627c N |
5955 | kfree(file); |
5956 | return err; | |
5957 | } | |
5958 | ||
f72ffdd6 | 5959 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
5960 | { |
5961 | mdu_disk_info_t info; | |
3cb03002 | 5962 | struct md_rdev *rdev; |
1da177e4 LT |
5963 | |
5964 | if (copy_from_user(&info, arg, sizeof(info))) | |
5965 | return -EFAULT; | |
5966 | ||
1ca69c4b | 5967 | rcu_read_lock(); |
57d051dc | 5968 | rdev = md_find_rdev_nr_rcu(mddev, info.number); |
1da177e4 LT |
5969 | if (rdev) { |
5970 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5971 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5972 | info.raid_disk = rdev->raid_disk; | |
5973 | info.state = 0; | |
b2d444d7 | 5974 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5975 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5976 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5977 | info.state |= (1<<MD_DISK_ACTIVE); |
5978 | info.state |= (1<<MD_DISK_SYNC); | |
5979 | } | |
9efdca16 | 5980 | if (test_bit(Journal, &rdev->flags)) |
bac624f3 | 5981 | info.state |= (1<<MD_DISK_JOURNAL); |
8ddf9efe N |
5982 | if (test_bit(WriteMostly, &rdev->flags)) |
5983 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
5984 | if (test_bit(FailFast, &rdev->flags)) |
5985 | info.state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 LT |
5986 | } else { |
5987 | info.major = info.minor = 0; | |
5988 | info.raid_disk = -1; | |
5989 | info.state = (1<<MD_DISK_REMOVED); | |
5990 | } | |
1ca69c4b | 5991 | rcu_read_unlock(); |
1da177e4 LT |
5992 | |
5993 | if (copy_to_user(arg, &info, sizeof(info))) | |
5994 | return -EFAULT; | |
5995 | ||
5996 | return 0; | |
5997 | } | |
5998 | ||
f72ffdd6 | 5999 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
6000 | { |
6001 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 6002 | struct md_rdev *rdev; |
1da177e4 LT |
6003 | dev_t dev = MKDEV(info->major,info->minor); |
6004 | ||
1aee41f6 GR |
6005 | if (mddev_is_clustered(mddev) && |
6006 | !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) { | |
9d48739e N |
6007 | pr_warn("%s: Cannot add to clustered mddev.\n", |
6008 | mdname(mddev)); | |
1aee41f6 GR |
6009 | return -EINVAL; |
6010 | } | |
6011 | ||
1da177e4 LT |
6012 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) |
6013 | return -EOVERFLOW; | |
6014 | ||
6015 | if (!mddev->raid_disks) { | |
6016 | int err; | |
6017 | /* expecting a device which has a superblock */ | |
6018 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
6019 | if (IS_ERR(rdev)) { | |
9d48739e | 6020 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6021 | PTR_ERR(rdev)); |
6022 | return PTR_ERR(rdev); | |
6023 | } | |
6024 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
6025 | struct md_rdev *rdev0 |
6026 | = list_entry(mddev->disks.next, | |
6027 | struct md_rdev, same_set); | |
a9f326eb | 6028 | err = super_types[mddev->major_version] |
1da177e4 LT |
6029 | .load_super(rdev, rdev0, mddev->minor_version); |
6030 | if (err < 0) { | |
9d48739e | 6031 | pr_warn("md: %s has different UUID to %s\n", |
f72ffdd6 | 6032 | bdevname(rdev->bdev,b), |
1da177e4 LT |
6033 | bdevname(rdev0->bdev,b2)); |
6034 | export_rdev(rdev); | |
6035 | return -EINVAL; | |
6036 | } | |
6037 | } | |
6038 | err = bind_rdev_to_array(rdev, mddev); | |
6039 | if (err) | |
6040 | export_rdev(rdev); | |
6041 | return err; | |
6042 | } | |
6043 | ||
6044 | /* | |
6045 | * add_new_disk can be used once the array is assembled | |
6046 | * to add "hot spares". They must already have a superblock | |
6047 | * written | |
6048 | */ | |
6049 | if (mddev->pers) { | |
6050 | int err; | |
6051 | if (!mddev->pers->hot_add_disk) { | |
9d48739e N |
6052 | pr_warn("%s: personality does not support diskops!\n", |
6053 | mdname(mddev)); | |
1da177e4 LT |
6054 | return -EINVAL; |
6055 | } | |
7b1e35f6 N |
6056 | if (mddev->persistent) |
6057 | rdev = md_import_device(dev, mddev->major_version, | |
6058 | mddev->minor_version); | |
6059 | else | |
6060 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 6061 | if (IS_ERR(rdev)) { |
9d48739e | 6062 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6063 | PTR_ERR(rdev)); |
6064 | return PTR_ERR(rdev); | |
6065 | } | |
1a855a06 | 6066 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
6067 | if (!mddev->persistent) { |
6068 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 6069 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 6070 | rdev->raid_disk = info->raid_disk; |
bf572541 | 6071 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 6072 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 6073 | } else |
41158c7e | 6074 | rdev->raid_disk = -1; |
f466722c | 6075 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
6076 | } else |
6077 | super_types[mddev->major_version]. | |
6078 | validate_super(mddev, rdev); | |
bedd86b7 | 6079 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 6080 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
6081 | /* This was a hot-add request, but events doesn't |
6082 | * match, so reject it. | |
6083 | */ | |
6084 | export_rdev(rdev); | |
6085 | return -EINVAL; | |
6086 | } | |
6087 | ||
b2d444d7 | 6088 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
6089 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6090 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
6091 | else |
6092 | clear_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6093 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6094 | set_bit(FailFast, &rdev->flags); | |
6095 | else | |
6096 | clear_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6097 | |
f6b6ec5c SL |
6098 | if (info->state & (1<<MD_DISK_JOURNAL)) { |
6099 | struct md_rdev *rdev2; | |
6100 | bool has_journal = false; | |
6101 | ||
6102 | /* make sure no existing journal disk */ | |
6103 | rdev_for_each(rdev2, mddev) { | |
6104 | if (test_bit(Journal, &rdev2->flags)) { | |
6105 | has_journal = true; | |
6106 | break; | |
6107 | } | |
6108 | } | |
6109 | if (has_journal) { | |
6110 | export_rdev(rdev); | |
6111 | return -EBUSY; | |
6112 | } | |
bac624f3 | 6113 | set_bit(Journal, &rdev->flags); |
f6b6ec5c | 6114 | } |
1aee41f6 GR |
6115 | /* |
6116 | * check whether the device shows up in other nodes | |
6117 | */ | |
6118 | if (mddev_is_clustered(mddev)) { | |
dbb64f86 | 6119 | if (info->state & (1 << MD_DISK_CANDIDATE)) |
1aee41f6 | 6120 | set_bit(Candidate, &rdev->flags); |
dbb64f86 | 6121 | else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) { |
1aee41f6 | 6122 | /* --add initiated by this node */ |
dbb64f86 | 6123 | err = md_cluster_ops->add_new_disk(mddev, rdev); |
1aee41f6 | 6124 | if (err) { |
1aee41f6 GR |
6125 | export_rdev(rdev); |
6126 | return err; | |
6127 | } | |
6128 | } | |
6129 | } | |
6130 | ||
1da177e4 LT |
6131 | rdev->raid_disk = -1; |
6132 | err = bind_rdev_to_array(rdev, mddev); | |
dbb64f86 | 6133 | |
1da177e4 LT |
6134 | if (err) |
6135 | export_rdev(rdev); | |
dbb64f86 GR |
6136 | |
6137 | if (mddev_is_clustered(mddev)) { | |
e566aef1 GJ |
6138 | if (info->state & (1 << MD_DISK_CANDIDATE)) { |
6139 | if (!err) { | |
6140 | err = md_cluster_ops->new_disk_ack(mddev, | |
6141 | err == 0); | |
6142 | if (err) | |
6143 | md_kick_rdev_from_array(rdev); | |
6144 | } | |
6145 | } else { | |
dbb64f86 GR |
6146 | if (err) |
6147 | md_cluster_ops->add_new_disk_cancel(mddev); | |
6148 | else | |
6149 | err = add_bound_rdev(rdev); | |
6150 | } | |
6151 | ||
6152 | } else if (!err) | |
a6da4ef8 | 6153 | err = add_bound_rdev(rdev); |
dbb64f86 | 6154 | |
1da177e4 LT |
6155 | return err; |
6156 | } | |
6157 | ||
6158 | /* otherwise, add_new_disk is only allowed | |
6159 | * for major_version==0 superblocks | |
6160 | */ | |
6161 | if (mddev->major_version != 0) { | |
9d48739e | 6162 | pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev)); |
1da177e4 LT |
6163 | return -EINVAL; |
6164 | } | |
6165 | ||
6166 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
6167 | int err; | |
d710e138 | 6168 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6169 | if (IS_ERR(rdev)) { |
9d48739e | 6170 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6171 | PTR_ERR(rdev)); |
6172 | return PTR_ERR(rdev); | |
6173 | } | |
6174 | rdev->desc_nr = info->number; | |
6175 | if (info->raid_disk < mddev->raid_disks) | |
6176 | rdev->raid_disk = info->raid_disk; | |
6177 | else | |
6178 | rdev->raid_disk = -1; | |
6179 | ||
1da177e4 | 6180 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
6181 | if (info->state & (1<<MD_DISK_SYNC)) |
6182 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 6183 | |
8ddf9efe N |
6184 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6185 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6186 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6187 | set_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6188 | |
1da177e4 | 6189 | if (!mddev->persistent) { |
9d48739e | 6190 | pr_debug("md: nonpersistent superblock ...\n"); |
77304d2a MS |
6191 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
6192 | } else | |
57b2caa3 | 6193 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 6194 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6195 | |
2bf071bf N |
6196 | err = bind_rdev_to_array(rdev, mddev); |
6197 | if (err) { | |
6198 | export_rdev(rdev); | |
6199 | return err; | |
6200 | } | |
1da177e4 LT |
6201 | } |
6202 | ||
6203 | return 0; | |
6204 | } | |
6205 | ||
f72ffdd6 | 6206 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6207 | { |
6208 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 6209 | struct md_rdev *rdev; |
1da177e4 | 6210 | |
1da177e4 LT |
6211 | rdev = find_rdev(mddev, dev); |
6212 | if (!rdev) | |
6213 | return -ENXIO; | |
6214 | ||
2910ff17 GR |
6215 | if (rdev->raid_disk < 0) |
6216 | goto kick_rdev; | |
293467aa | 6217 | |
3ea8929d N |
6218 | clear_bit(Blocked, &rdev->flags); |
6219 | remove_and_add_spares(mddev, rdev); | |
6220 | ||
1da177e4 LT |
6221 | if (rdev->raid_disk >= 0) |
6222 | goto busy; | |
6223 | ||
2910ff17 | 6224 | kick_rdev: |
54a88392 | 6225 | if (mddev_is_clustered(mddev)) |
88bcfef7 GR |
6226 | md_cluster_ops->remove_disk(mddev, rdev); |
6227 | ||
fb56dfef | 6228 | md_kick_rdev_from_array(rdev); |
2953079c | 6229 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6230 | if (mddev->thread) |
6231 | md_wakeup_thread(mddev->thread); | |
6232 | else | |
6233 | md_update_sb(mddev, 1); | |
d7603b7e | 6234 | md_new_event(mddev); |
1da177e4 LT |
6235 | |
6236 | return 0; | |
6237 | busy: | |
9d48739e N |
6238 | pr_debug("md: cannot remove active disk %s from %s ...\n", |
6239 | bdevname(rdev->bdev,b), mdname(mddev)); | |
1da177e4 LT |
6240 | return -EBUSY; |
6241 | } | |
6242 | ||
f72ffdd6 | 6243 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6244 | { |
6245 | char b[BDEVNAME_SIZE]; | |
6246 | int err; | |
3cb03002 | 6247 | struct md_rdev *rdev; |
1da177e4 LT |
6248 | |
6249 | if (!mddev->pers) | |
6250 | return -ENODEV; | |
6251 | ||
6252 | if (mddev->major_version != 0) { | |
9d48739e | 6253 | pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n", |
1da177e4 LT |
6254 | mdname(mddev)); |
6255 | return -EINVAL; | |
6256 | } | |
6257 | if (!mddev->pers->hot_add_disk) { | |
9d48739e | 6258 | pr_warn("%s: personality does not support diskops!\n", |
1da177e4 LT |
6259 | mdname(mddev)); |
6260 | return -EINVAL; | |
6261 | } | |
6262 | ||
d710e138 | 6263 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6264 | if (IS_ERR(rdev)) { |
9d48739e | 6265 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6266 | PTR_ERR(rdev)); |
6267 | return -EINVAL; | |
6268 | } | |
6269 | ||
6270 | if (mddev->persistent) | |
57b2caa3 | 6271 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 6272 | else |
77304d2a | 6273 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 6274 | |
8190e754 | 6275 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6276 | |
b2d444d7 | 6277 | if (test_bit(Faulty, &rdev->flags)) { |
9d48739e | 6278 | pr_warn("md: can not hot-add faulty %s disk to %s!\n", |
1da177e4 LT |
6279 | bdevname(rdev->bdev,b), mdname(mddev)); |
6280 | err = -EINVAL; | |
6281 | goto abort_export; | |
6282 | } | |
293467aa | 6283 | |
b2d444d7 | 6284 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6285 | rdev->desc_nr = -1; |
5842730d | 6286 | rdev->saved_raid_disk = -1; |
2bf071bf N |
6287 | err = bind_rdev_to_array(rdev, mddev); |
6288 | if (err) | |
2aa82191 | 6289 | goto abort_export; |
1da177e4 LT |
6290 | |
6291 | /* | |
6292 | * The rest should better be atomic, we can have disk failures | |
6293 | * noticed in interrupt contexts ... | |
6294 | */ | |
6295 | ||
1da177e4 LT |
6296 | rdev->raid_disk = -1; |
6297 | ||
2953079c | 6298 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6299 | if (!mddev->thread) |
6300 | md_update_sb(mddev, 1); | |
1da177e4 LT |
6301 | /* |
6302 | * Kick recovery, maybe this spare has to be added to the | |
6303 | * array immediately. | |
6304 | */ | |
6305 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6306 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 6307 | md_new_event(mddev); |
1da177e4 LT |
6308 | return 0; |
6309 | ||
1da177e4 LT |
6310 | abort_export: |
6311 | export_rdev(rdev); | |
6312 | return err; | |
6313 | } | |
6314 | ||
fd01b88c | 6315 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 6316 | { |
035328c2 | 6317 | int err = 0; |
32a7627c | 6318 | |
36fa3063 | 6319 | if (mddev->pers) { |
d66b1b39 | 6320 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
6321 | return -EBUSY; |
6322 | if (mddev->recovery || mddev->sync_thread) | |
6323 | return -EBUSY; | |
6324 | /* we should be able to change the bitmap.. */ | |
6325 | } | |
32a7627c | 6326 | |
36fa3063 | 6327 | if (fd >= 0) { |
035328c2 | 6328 | struct inode *inode; |
1e594bb2 N |
6329 | struct file *f; |
6330 | ||
6331 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 6332 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6333 | f = fget(fd); |
32a7627c | 6334 | |
1e594bb2 | 6335 | if (f == NULL) { |
9d48739e N |
6336 | pr_warn("%s: error: failed to get bitmap file\n", |
6337 | mdname(mddev)); | |
36fa3063 N |
6338 | return -EBADF; |
6339 | } | |
6340 | ||
1e594bb2 | 6341 | inode = f->f_mapping->host; |
035328c2 | 6342 | if (!S_ISREG(inode->i_mode)) { |
9d48739e N |
6343 | pr_warn("%s: error: bitmap file must be a regular file\n", |
6344 | mdname(mddev)); | |
035328c2 | 6345 | err = -EBADF; |
1e594bb2 | 6346 | } else if (!(f->f_mode & FMODE_WRITE)) { |
9d48739e N |
6347 | pr_warn("%s: error: bitmap file must open for write\n", |
6348 | mdname(mddev)); | |
035328c2 N |
6349 | err = -EBADF; |
6350 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
9d48739e N |
6351 | pr_warn("%s: error: bitmap file is already in use\n", |
6352 | mdname(mddev)); | |
035328c2 N |
6353 | err = -EBUSY; |
6354 | } | |
6355 | if (err) { | |
1e594bb2 | 6356 | fput(f); |
36fa3063 N |
6357 | return err; |
6358 | } | |
1e594bb2 | 6359 | mddev->bitmap_info.file = f; |
c3d9714e | 6360 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6361 | } else if (mddev->bitmap == NULL) |
6362 | return -ENOENT; /* cannot remove what isn't there */ | |
6363 | err = 0; | |
6364 | if (mddev->pers) { | |
6365 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6366 | if (fd >= 0) { |
f9209a32 GR |
6367 | struct bitmap *bitmap; |
6368 | ||
6369 | bitmap = bitmap_create(mddev, -1); | |
6370 | if (!IS_ERR(bitmap)) { | |
6371 | mddev->bitmap = bitmap; | |
69e51b44 | 6372 | err = bitmap_load(mddev); |
ba599aca N |
6373 | } else |
6374 | err = PTR_ERR(bitmap); | |
69e51b44 | 6375 | } |
d7375ab3 | 6376 | if (fd < 0 || err) { |
36fa3063 | 6377 | bitmap_destroy(mddev); |
d7375ab3 N |
6378 | fd = -1; /* make sure to put the file */ |
6379 | } | |
36fa3063 | 6380 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6381 | } |
6382 | if (fd < 0) { | |
4af1a041 N |
6383 | struct file *f = mddev->bitmap_info.file; |
6384 | if (f) { | |
6385 | spin_lock(&mddev->lock); | |
6386 | mddev->bitmap_info.file = NULL; | |
6387 | spin_unlock(&mddev->lock); | |
6388 | fput(f); | |
6389 | } | |
36fa3063 N |
6390 | } |
6391 | ||
32a7627c N |
6392 | return err; |
6393 | } | |
6394 | ||
1da177e4 LT |
6395 | /* |
6396 | * set_array_info is used two different ways | |
6397 | * The original usage is when creating a new array. | |
6398 | * In this usage, raid_disks is > 0 and it together with | |
6399 | * level, size, not_persistent,layout,chunksize determine the | |
6400 | * shape of the array. | |
6401 | * This will always create an array with a type-0.90.0 superblock. | |
6402 | * The newer usage is when assembling an array. | |
6403 | * In this case raid_disks will be 0, and the major_version field is | |
6404 | * use to determine which style super-blocks are to be found on the devices. | |
6405 | * The minor and patch _version numbers are also kept incase the | |
6406 | * super_block handler wishes to interpret them. | |
6407 | */ | |
f72ffdd6 | 6408 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6409 | { |
6410 | ||
6411 | if (info->raid_disks == 0) { | |
6412 | /* just setting version number for superblock loading */ | |
6413 | if (info->major_version < 0 || | |
50511da3 | 6414 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6415 | super_types[info->major_version].name == NULL) { |
6416 | /* maybe try to auto-load a module? */ | |
9d48739e | 6417 | pr_warn("md: superblock version %d not known\n", |
1da177e4 LT |
6418 | info->major_version); |
6419 | return -EINVAL; | |
6420 | } | |
6421 | mddev->major_version = info->major_version; | |
6422 | mddev->minor_version = info->minor_version; | |
6423 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6424 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6425 | /* ensure mddev_put doesn't delete this now that there |
6426 | * is some minimal configuration. | |
6427 | */ | |
9ebc6ef1 | 6428 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6429 | return 0; |
6430 | } | |
6431 | mddev->major_version = MD_MAJOR_VERSION; | |
6432 | mddev->minor_version = MD_MINOR_VERSION; | |
6433 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6434 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6435 | |
6436 | mddev->level = info->level; | |
17115e03 | 6437 | mddev->clevel[0] = 0; |
58c0fed4 | 6438 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6439 | mddev->raid_disks = info->raid_disks; |
6440 | /* don't set md_minor, it is determined by which /dev/md* was | |
6441 | * openned | |
6442 | */ | |
6443 | if (info->state & (1<<MD_SB_CLEAN)) | |
6444 | mddev->recovery_cp = MaxSector; | |
6445 | else | |
6446 | mddev->recovery_cp = 0; | |
6447 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6448 | mddev->external = 0; |
1da177e4 LT |
6449 | |
6450 | mddev->layout = info->layout; | |
9d8f0363 | 6451 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 | 6452 | |
2953079c | 6453 | if (mddev->persistent) { |
1b3bae49 N |
6454 | mddev->max_disks = MD_SB_DISKS; |
6455 | mddev->flags = 0; | |
6456 | mddev->sb_flags = 0; | |
2953079c SL |
6457 | } |
6458 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
1da177e4 | 6459 | |
c3d9714e | 6460 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6461 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6462 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6463 | |
f6705578 N |
6464 | mddev->reshape_position = MaxSector; |
6465 | ||
1da177e4 LT |
6466 | /* |
6467 | * Generate a 128 bit UUID | |
6468 | */ | |
6469 | get_random_bytes(mddev->uuid, 16); | |
6470 | ||
f6705578 | 6471 | mddev->new_level = mddev->level; |
664e7c41 | 6472 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6473 | mddev->new_layout = mddev->layout; |
6474 | mddev->delta_disks = 0; | |
2c810cdd | 6475 | mddev->reshape_backwards = 0; |
f6705578 | 6476 | |
1da177e4 LT |
6477 | return 0; |
6478 | } | |
6479 | ||
fd01b88c | 6480 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6481 | { |
b522adcd DW |
6482 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6483 | ||
6484 | if (mddev->external_size) | |
6485 | return; | |
6486 | ||
1f403624 DW |
6487 | mddev->array_sectors = array_sectors; |
6488 | } | |
6489 | EXPORT_SYMBOL(md_set_array_sectors); | |
6490 | ||
fd01b88c | 6491 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6492 | { |
3cb03002 | 6493 | struct md_rdev *rdev; |
a35b0d69 | 6494 | int rv; |
d71f9f88 | 6495 | int fit = (num_sectors == 0); |
a35b0d69 | 6496 | |
ab5a98b1 GJ |
6497 | /* cluster raid doesn't support update size */ |
6498 | if (mddev_is_clustered(mddev)) | |
6499 | return -EINVAL; | |
6500 | ||
a35b0d69 N |
6501 | if (mddev->pers->resize == NULL) |
6502 | return -EINVAL; | |
d71f9f88 AN |
6503 | /* The "num_sectors" is the number of sectors of each device that |
6504 | * is used. This can only make sense for arrays with redundancy. | |
6505 | * linear and raid0 always use whatever space is available. We can only | |
6506 | * consider changing this number if no resync or reconstruction is | |
6507 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6508 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6509 | * of each device. If num_sectors is zero, we find the largest size |
6510 | * that fits. | |
a35b0d69 | 6511 | */ |
f851b60d N |
6512 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6513 | mddev->sync_thread) | |
a35b0d69 | 6514 | return -EBUSY; |
bd8839e0 N |
6515 | if (mddev->ro) |
6516 | return -EROFS; | |
a4a6125a | 6517 | |
dafb20fa | 6518 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6519 | sector_t avail = rdev->sectors; |
01ab5662 | 6520 | |
d71f9f88 AN |
6521 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6522 | num_sectors = avail; | |
6523 | if (avail < num_sectors) | |
a35b0d69 N |
6524 | return -ENOSPC; |
6525 | } | |
d71f9f88 | 6526 | rv = mddev->pers->resize(mddev, num_sectors); |
c9483634 GJ |
6527 | if (!rv) { |
6528 | if (mddev->queue) { | |
6529 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
6530 | revalidate_disk(mddev->gendisk); | |
6531 | } | |
6532 | } | |
a35b0d69 N |
6533 | return rv; |
6534 | } | |
6535 | ||
fd01b88c | 6536 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6537 | { |
6538 | int rv; | |
c6563a8c | 6539 | struct md_rdev *rdev; |
da943b99 | 6540 | /* change the number of raid disks */ |
63c70c4f | 6541 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6542 | return -EINVAL; |
bd8839e0 N |
6543 | if (mddev->ro) |
6544 | return -EROFS; | |
da943b99 | 6545 | if (raid_disks <= 0 || |
233fca36 | 6546 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6547 | return -EINVAL; |
f851b60d N |
6548 | if (mddev->sync_thread || |
6549 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6550 | mddev->reshape_position != MaxSector) | |
da943b99 | 6551 | return -EBUSY; |
c6563a8c N |
6552 | |
6553 | rdev_for_each(rdev, mddev) { | |
6554 | if (mddev->raid_disks < raid_disks && | |
6555 | rdev->data_offset < rdev->new_data_offset) | |
6556 | return -EINVAL; | |
6557 | if (mddev->raid_disks > raid_disks && | |
6558 | rdev->data_offset > rdev->new_data_offset) | |
6559 | return -EINVAL; | |
6560 | } | |
6561 | ||
63c70c4f | 6562 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6563 | if (mddev->delta_disks < 0) |
6564 | mddev->reshape_backwards = 1; | |
6565 | else if (mddev->delta_disks > 0) | |
6566 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6567 | |
6568 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6569 | if (rv < 0) { |
de171cb9 | 6570 | mddev->delta_disks = 0; |
2c810cdd N |
6571 | mddev->reshape_backwards = 0; |
6572 | } | |
da943b99 N |
6573 | return rv; |
6574 | } | |
6575 | ||
1da177e4 LT |
6576 | /* |
6577 | * update_array_info is used to change the configuration of an | |
6578 | * on-line array. | |
6579 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6580 | * fields in the info are checked against the array. | |
6581 | * Any differences that cannot be handled will cause an error. | |
6582 | * Normally, only one change can be managed at a time. | |
6583 | */ | |
fd01b88c | 6584 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6585 | { |
6586 | int rv = 0; | |
6587 | int cnt = 0; | |
36fa3063 N |
6588 | int state = 0; |
6589 | ||
6590 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6591 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6592 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6593 | |
6594 | if (mddev->major_version != info->major_version || | |
6595 | mddev->minor_version != info->minor_version || | |
6596 | /* mddev->patch_version != info->patch_version || */ | |
6597 | mddev->ctime != info->ctime || | |
6598 | mddev->level != info->level || | |
6599 | /* mddev->layout != info->layout || */ | |
4e023612 | 6600 | mddev->persistent != !info->not_persistent || |
9d8f0363 | 6601 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6602 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6603 | ((state^info->state) & 0xfffffe00) | |
6604 | ) | |
1da177e4 LT |
6605 | return -EINVAL; |
6606 | /* Check there is only one change */ | |
58c0fed4 AN |
6607 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6608 | cnt++; | |
6609 | if (mddev->raid_disks != info->raid_disks) | |
6610 | cnt++; | |
6611 | if (mddev->layout != info->layout) | |
6612 | cnt++; | |
6613 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6614 | cnt++; | |
6615 | if (cnt == 0) | |
6616 | return 0; | |
6617 | if (cnt > 1) | |
6618 | return -EINVAL; | |
1da177e4 LT |
6619 | |
6620 | if (mddev->layout != info->layout) { | |
6621 | /* Change layout | |
6622 | * we don't need to do anything at the md level, the | |
6623 | * personality will take care of it all. | |
6624 | */ | |
50ac168a | 6625 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6626 | return -EINVAL; |
597a711b N |
6627 | else { |
6628 | mddev->new_layout = info->layout; | |
50ac168a | 6629 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6630 | if (rv) |
6631 | mddev->new_layout = mddev->layout; | |
6632 | return rv; | |
6633 | } | |
1da177e4 | 6634 | } |
58c0fed4 | 6635 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6636 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6637 | |
da943b99 N |
6638 | if (mddev->raid_disks != info->raid_disks) |
6639 | rv = update_raid_disks(mddev, info->raid_disks); | |
6640 | ||
36fa3063 | 6641 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
293467aa GR |
6642 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) { |
6643 | rv = -EINVAL; | |
6644 | goto err; | |
6645 | } | |
6646 | if (mddev->recovery || mddev->sync_thread) { | |
6647 | rv = -EBUSY; | |
6648 | goto err; | |
6649 | } | |
36fa3063 | 6650 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { |
f9209a32 | 6651 | struct bitmap *bitmap; |
36fa3063 | 6652 | /* add the bitmap */ |
293467aa GR |
6653 | if (mddev->bitmap) { |
6654 | rv = -EEXIST; | |
6655 | goto err; | |
6656 | } | |
6657 | if (mddev->bitmap_info.default_offset == 0) { | |
6658 | rv = -EINVAL; | |
6659 | goto err; | |
6660 | } | |
c3d9714e N |
6661 | mddev->bitmap_info.offset = |
6662 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6663 | mddev->bitmap_info.space = |
6664 | mddev->bitmap_info.default_space; | |
36fa3063 | 6665 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
6666 | bitmap = bitmap_create(mddev, -1); |
6667 | if (!IS_ERR(bitmap)) { | |
6668 | mddev->bitmap = bitmap; | |
69e51b44 | 6669 | rv = bitmap_load(mddev); |
ba599aca N |
6670 | } else |
6671 | rv = PTR_ERR(bitmap); | |
36fa3063 N |
6672 | if (rv) |
6673 | bitmap_destroy(mddev); | |
6674 | mddev->pers->quiesce(mddev, 0); | |
6675 | } else { | |
6676 | /* remove the bitmap */ | |
293467aa GR |
6677 | if (!mddev->bitmap) { |
6678 | rv = -ENOENT; | |
6679 | goto err; | |
6680 | } | |
6681 | if (mddev->bitmap->storage.file) { | |
6682 | rv = -EINVAL; | |
6683 | goto err; | |
6684 | } | |
f6a2dc64 GJ |
6685 | if (mddev->bitmap_info.nodes) { |
6686 | /* hold PW on all the bitmap lock */ | |
6687 | if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) { | |
9d48739e | 6688 | pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n"); |
f6a2dc64 GJ |
6689 | rv = -EPERM; |
6690 | md_cluster_ops->unlock_all_bitmaps(mddev); | |
6691 | goto err; | |
6692 | } | |
6693 | ||
6694 | mddev->bitmap_info.nodes = 0; | |
6695 | md_cluster_ops->leave(mddev); | |
6696 | } | |
36fa3063 N |
6697 | mddev->pers->quiesce(mddev, 1); |
6698 | bitmap_destroy(mddev); | |
6699 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6700 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6701 | } |
6702 | } | |
850b2b42 | 6703 | md_update_sb(mddev, 1); |
293467aa GR |
6704 | return rv; |
6705 | err: | |
1da177e4 LT |
6706 | return rv; |
6707 | } | |
6708 | ||
fd01b88c | 6709 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6710 | { |
3cb03002 | 6711 | struct md_rdev *rdev; |
1ca69c4b | 6712 | int err = 0; |
1da177e4 LT |
6713 | |
6714 | if (mddev->pers == NULL) | |
6715 | return -ENODEV; | |
6716 | ||
1ca69c4b N |
6717 | rcu_read_lock(); |
6718 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6719 | if (!rdev) |
1ca69c4b N |
6720 | err = -ENODEV; |
6721 | else { | |
6722 | md_error(mddev, rdev); | |
6723 | if (!test_bit(Faulty, &rdev->flags)) | |
6724 | err = -EBUSY; | |
6725 | } | |
6726 | rcu_read_unlock(); | |
6727 | return err; | |
1da177e4 LT |
6728 | } |
6729 | ||
2f9618ce AN |
6730 | /* |
6731 | * We have a problem here : there is no easy way to give a CHS | |
6732 | * virtual geometry. We currently pretend that we have a 2 heads | |
6733 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6734 | * dosfs just mad... ;-) | |
6735 | */ | |
a885c8c4 CH |
6736 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6737 | { | |
fd01b88c | 6738 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6739 | |
6740 | geo->heads = 2; | |
6741 | geo->sectors = 4; | |
49ce6cea | 6742 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6743 | return 0; |
6744 | } | |
6745 | ||
cb335f88 NS |
6746 | static inline bool md_ioctl_valid(unsigned int cmd) |
6747 | { | |
6748 | switch (cmd) { | |
6749 | case ADD_NEW_DISK: | |
6750 | case BLKROSET: | |
6751 | case GET_ARRAY_INFO: | |
6752 | case GET_BITMAP_FILE: | |
6753 | case GET_DISK_INFO: | |
6754 | case HOT_ADD_DISK: | |
6755 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
6756 | case RAID_AUTORUN: |
6757 | case RAID_VERSION: | |
6758 | case RESTART_ARRAY_RW: | |
6759 | case RUN_ARRAY: | |
6760 | case SET_ARRAY_INFO: | |
6761 | case SET_BITMAP_FILE: | |
6762 | case SET_DISK_FAULTY: | |
6763 | case STOP_ARRAY: | |
6764 | case STOP_ARRAY_RO: | |
1aee41f6 | 6765 | case CLUSTERED_DISK_NACK: |
cb335f88 NS |
6766 | return true; |
6767 | default: | |
6768 | return false; | |
6769 | } | |
6770 | } | |
6771 | ||
a39907fa | 6772 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6773 | unsigned int cmd, unsigned long arg) |
6774 | { | |
6775 | int err = 0; | |
6776 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6777 | struct mddev *mddev = NULL; |
e2218350 | 6778 | int ro; |
1da177e4 | 6779 | |
cb335f88 NS |
6780 | if (!md_ioctl_valid(cmd)) |
6781 | return -ENOTTY; | |
6782 | ||
506c9e44 N |
6783 | switch (cmd) { |
6784 | case RAID_VERSION: | |
6785 | case GET_ARRAY_INFO: | |
6786 | case GET_DISK_INFO: | |
6787 | break; | |
6788 | default: | |
6789 | if (!capable(CAP_SYS_ADMIN)) | |
6790 | return -EACCES; | |
6791 | } | |
1da177e4 LT |
6792 | |
6793 | /* | |
6794 | * Commands dealing with the RAID driver but not any | |
6795 | * particular array: | |
6796 | */ | |
c02c0aeb N |
6797 | switch (cmd) { |
6798 | case RAID_VERSION: | |
6799 | err = get_version(argp); | |
3adc28d8 | 6800 | goto out; |
1da177e4 | 6801 | |
1da177e4 | 6802 | #ifndef MODULE |
c02c0aeb N |
6803 | case RAID_AUTORUN: |
6804 | err = 0; | |
6805 | autostart_arrays(arg); | |
3adc28d8 | 6806 | goto out; |
1da177e4 | 6807 | #endif |
c02c0aeb | 6808 | default:; |
1da177e4 LT |
6809 | } |
6810 | ||
6811 | /* | |
6812 | * Commands creating/starting a new array: | |
6813 | */ | |
6814 | ||
a39907fa | 6815 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6816 | |
6817 | if (!mddev) { | |
6818 | BUG(); | |
3adc28d8 | 6819 | goto out; |
1da177e4 LT |
6820 | } |
6821 | ||
1ca69c4b N |
6822 | /* Some actions do not requires the mutex */ |
6823 | switch (cmd) { | |
6824 | case GET_ARRAY_INFO: | |
6825 | if (!mddev->raid_disks && !mddev->external) | |
6826 | err = -ENODEV; | |
6827 | else | |
6828 | err = get_array_info(mddev, argp); | |
3adc28d8 | 6829 | goto out; |
1ca69c4b N |
6830 | |
6831 | case GET_DISK_INFO: | |
6832 | if (!mddev->raid_disks && !mddev->external) | |
6833 | err = -ENODEV; | |
6834 | else | |
6835 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 6836 | goto out; |
1ca69c4b N |
6837 | |
6838 | case SET_DISK_FAULTY: | |
6839 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6840 | goto out; |
4af1a041 N |
6841 | |
6842 | case GET_BITMAP_FILE: | |
6843 | err = get_bitmap_file(mddev, argp); | |
6844 | goto out; | |
6845 | ||
1ca69c4b N |
6846 | } |
6847 | ||
a7a3f08d N |
6848 | if (cmd == ADD_NEW_DISK) |
6849 | /* need to ensure md_delayed_delete() has completed */ | |
6850 | flush_workqueue(md_misc_wq); | |
6851 | ||
90f5f7ad HR |
6852 | if (cmd == HOT_REMOVE_DISK) |
6853 | /* need to ensure recovery thread has run */ | |
6854 | wait_event_interruptible_timeout(mddev->sb_wait, | |
6855 | !test_bit(MD_RECOVERY_NEEDED, | |
82a301cb | 6856 | &mddev->recovery), |
90f5f7ad | 6857 | msecs_to_jiffies(5000)); |
260fa034 N |
6858 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
6859 | /* Need to flush page cache, and ensure no-one else opens | |
6860 | * and writes | |
6861 | */ | |
6862 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 6863 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
6864 | mutex_unlock(&mddev->open_mutex); |
6865 | err = -EBUSY; | |
3adc28d8 | 6866 | goto out; |
260fa034 | 6867 | } |
af8d8e6f | 6868 | set_bit(MD_CLOSING, &mddev->flags); |
260fa034 N |
6869 | mutex_unlock(&mddev->open_mutex); |
6870 | sync_blockdev(bdev); | |
6871 | } | |
1da177e4 LT |
6872 | err = mddev_lock(mddev); |
6873 | if (err) { | |
9d48739e N |
6874 | pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n", |
6875 | err, cmd); | |
3adc28d8 | 6876 | goto out; |
1da177e4 LT |
6877 | } |
6878 | ||
c02c0aeb N |
6879 | if (cmd == SET_ARRAY_INFO) { |
6880 | mdu_array_info_t info; | |
6881 | if (!arg) | |
6882 | memset(&info, 0, sizeof(info)); | |
6883 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6884 | err = -EFAULT; | |
3adc28d8 | 6885 | goto unlock; |
c02c0aeb N |
6886 | } |
6887 | if (mddev->pers) { | |
6888 | err = update_array_info(mddev, &info); | |
6889 | if (err) { | |
9d48739e | 6890 | pr_warn("md: couldn't update array info. %d\n", err); |
3adc28d8 | 6891 | goto unlock; |
1da177e4 | 6892 | } |
3adc28d8 | 6893 | goto unlock; |
c02c0aeb N |
6894 | } |
6895 | if (!list_empty(&mddev->disks)) { | |
9d48739e | 6896 | pr_warn("md: array %s already has disks!\n", mdname(mddev)); |
c02c0aeb | 6897 | err = -EBUSY; |
3adc28d8 | 6898 | goto unlock; |
c02c0aeb N |
6899 | } |
6900 | if (mddev->raid_disks) { | |
9d48739e | 6901 | pr_warn("md: array %s already initialised!\n", mdname(mddev)); |
c02c0aeb | 6902 | err = -EBUSY; |
3adc28d8 | 6903 | goto unlock; |
c02c0aeb N |
6904 | } |
6905 | err = set_array_info(mddev, &info); | |
6906 | if (err) { | |
9d48739e | 6907 | pr_warn("md: couldn't set array info. %d\n", err); |
3adc28d8 | 6908 | goto unlock; |
c02c0aeb | 6909 | } |
3adc28d8 | 6910 | goto unlock; |
1da177e4 LT |
6911 | } |
6912 | ||
6913 | /* | |
6914 | * Commands querying/configuring an existing array: | |
6915 | */ | |
32a7627c | 6916 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6917 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6918 | if ((!mddev->raid_disks && !mddev->external) |
6919 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6920 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6921 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 6922 | err = -ENODEV; |
3adc28d8 | 6923 | goto unlock; |
1da177e4 LT |
6924 | } |
6925 | ||
6926 | /* | |
6927 | * Commands even a read-only array can execute: | |
6928 | */ | |
c02c0aeb | 6929 | switch (cmd) { |
c02c0aeb N |
6930 | case RESTART_ARRAY_RW: |
6931 | err = restart_array(mddev); | |
3adc28d8 | 6932 | goto unlock; |
1da177e4 | 6933 | |
c02c0aeb N |
6934 | case STOP_ARRAY: |
6935 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 6936 | goto unlock; |
1da177e4 | 6937 | |
c02c0aeb N |
6938 | case STOP_ARRAY_RO: |
6939 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 6940 | goto unlock; |
1da177e4 | 6941 | |
3ea8929d N |
6942 | case HOT_REMOVE_DISK: |
6943 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6944 | goto unlock; |
3ea8929d | 6945 | |
7ceb17e8 N |
6946 | case ADD_NEW_DISK: |
6947 | /* We can support ADD_NEW_DISK on read-only arrays | |
466ad292 | 6948 | * only if we are re-adding a preexisting device. |
7ceb17e8 N |
6949 | * So require mddev->pers and MD_DISK_SYNC. |
6950 | */ | |
6951 | if (mddev->pers) { | |
6952 | mdu_disk_info_t info; | |
6953 | if (copy_from_user(&info, argp, sizeof(info))) | |
6954 | err = -EFAULT; | |
6955 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
6956 | /* Need to clear read-only for this */ | |
6957 | break; | |
6958 | else | |
6959 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6960 | goto unlock; |
7ceb17e8 N |
6961 | } |
6962 | break; | |
6963 | ||
c02c0aeb N |
6964 | case BLKROSET: |
6965 | if (get_user(ro, (int __user *)(arg))) { | |
6966 | err = -EFAULT; | |
3adc28d8 | 6967 | goto unlock; |
c02c0aeb N |
6968 | } |
6969 | err = -EINVAL; | |
e2218350 | 6970 | |
c02c0aeb N |
6971 | /* if the bdev is going readonly the value of mddev->ro |
6972 | * does not matter, no writes are coming | |
6973 | */ | |
6974 | if (ro) | |
3adc28d8 | 6975 | goto unlock; |
e2218350 | 6976 | |
c02c0aeb N |
6977 | /* are we are already prepared for writes? */ |
6978 | if (mddev->ro != 1) | |
3adc28d8 | 6979 | goto unlock; |
e2218350 | 6980 | |
c02c0aeb N |
6981 | /* transitioning to readauto need only happen for |
6982 | * arrays that call md_write_start | |
6983 | */ | |
6984 | if (mddev->pers) { | |
6985 | err = restart_array(mddev); | |
6986 | if (err == 0) { | |
6987 | mddev->ro = 2; | |
6988 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6989 | } |
c02c0aeb | 6990 | } |
3adc28d8 | 6991 | goto unlock; |
1da177e4 LT |
6992 | } |
6993 | ||
6994 | /* | |
6995 | * The remaining ioctls are changing the state of the | |
f91de92e | 6996 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 6997 | */ |
326eb17d | 6998 | if (mddev->ro && mddev->pers) { |
f91de92e N |
6999 | if (mddev->ro == 2) { |
7000 | mddev->ro = 0; | |
00bcb4ac | 7001 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 7002 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
7003 | /* mddev_unlock will wake thread */ |
7004 | /* If a device failed while we were read-only, we | |
7005 | * need to make sure the metadata is updated now. | |
7006 | */ | |
2953079c | 7007 | if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) { |
f3378b48 N |
7008 | mddev_unlock(mddev); |
7009 | wait_event(mddev->sb_wait, | |
2953079c SL |
7010 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) && |
7011 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); | |
29f097c4 | 7012 | mddev_lock_nointr(mddev); |
f3378b48 | 7013 | } |
f91de92e N |
7014 | } else { |
7015 | err = -EROFS; | |
3adc28d8 | 7016 | goto unlock; |
f91de92e | 7017 | } |
1da177e4 LT |
7018 | } |
7019 | ||
c02c0aeb N |
7020 | switch (cmd) { |
7021 | case ADD_NEW_DISK: | |
1da177e4 | 7022 | { |
c02c0aeb N |
7023 | mdu_disk_info_t info; |
7024 | if (copy_from_user(&info, argp, sizeof(info))) | |
7025 | err = -EFAULT; | |
7026 | else | |
7027 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7028 | goto unlock; |
c02c0aeb | 7029 | } |
1da177e4 | 7030 | |
1aee41f6 GR |
7031 | case CLUSTERED_DISK_NACK: |
7032 | if (mddev_is_clustered(mddev)) | |
7033 | md_cluster_ops->new_disk_ack(mddev, false); | |
7034 | else | |
7035 | err = -EINVAL; | |
7036 | goto unlock; | |
7037 | ||
c02c0aeb N |
7038 | case HOT_ADD_DISK: |
7039 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7040 | goto unlock; |
1da177e4 | 7041 | |
c02c0aeb N |
7042 | case RUN_ARRAY: |
7043 | err = do_md_run(mddev); | |
3adc28d8 | 7044 | goto unlock; |
1da177e4 | 7045 | |
c02c0aeb N |
7046 | case SET_BITMAP_FILE: |
7047 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 7048 | goto unlock; |
32a7627c | 7049 | |
c02c0aeb N |
7050 | default: |
7051 | err = -EINVAL; | |
3adc28d8 | 7052 | goto unlock; |
1da177e4 LT |
7053 | } |
7054 | ||
3adc28d8 | 7055 | unlock: |
d3374825 N |
7056 | if (mddev->hold_active == UNTIL_IOCTL && |
7057 | err != -EINVAL) | |
7058 | mddev->hold_active = 0; | |
1da177e4 | 7059 | mddev_unlock(mddev); |
3adc28d8 | 7060 | out: |
1da177e4 LT |
7061 | return err; |
7062 | } | |
aa98aa31 AB |
7063 | #ifdef CONFIG_COMPAT |
7064 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
7065 | unsigned int cmd, unsigned long arg) | |
7066 | { | |
7067 | switch (cmd) { | |
7068 | case HOT_REMOVE_DISK: | |
7069 | case HOT_ADD_DISK: | |
7070 | case SET_DISK_FAULTY: | |
7071 | case SET_BITMAP_FILE: | |
7072 | /* These take in integer arg, do not convert */ | |
7073 | break; | |
7074 | default: | |
7075 | arg = (unsigned long)compat_ptr(arg); | |
7076 | break; | |
7077 | } | |
7078 | ||
7079 | return md_ioctl(bdev, mode, cmd, arg); | |
7080 | } | |
7081 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 7082 | |
a39907fa | 7083 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
7084 | { |
7085 | /* | |
7086 | * Succeed if we can lock the mddev, which confirms that | |
7087 | * it isn't being stopped right now. | |
7088 | */ | |
fd01b88c | 7089 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
7090 | int err; |
7091 | ||
0c098220 YL |
7092 | if (!mddev) |
7093 | return -ENODEV; | |
7094 | ||
d3374825 N |
7095 | if (mddev->gendisk != bdev->bd_disk) { |
7096 | /* we are racing with mddev_put which is discarding this | |
7097 | * bd_disk. | |
7098 | */ | |
7099 | mddev_put(mddev); | |
7100 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 7101 | flush_workqueue(md_misc_wq); |
d3374825 N |
7102 | /* Then retry the open from the top */ |
7103 | return -ERESTARTSYS; | |
7104 | } | |
7105 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
7106 | ||
c8c00a69 | 7107 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
7108 | goto out; |
7109 | ||
af8d8e6f GJ |
7110 | if (test_bit(MD_CLOSING, &mddev->flags)) { |
7111 | mutex_unlock(&mddev->open_mutex); | |
e2342ca8 N |
7112 | err = -ENODEV; |
7113 | goto out; | |
af8d8e6f GJ |
7114 | } |
7115 | ||
1da177e4 | 7116 | err = 0; |
f2ea68cf | 7117 | atomic_inc(&mddev->openers); |
c8c00a69 | 7118 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 7119 | |
f0b4f7e2 | 7120 | check_disk_change(bdev); |
1da177e4 | 7121 | out: |
e2342ca8 N |
7122 | if (err) |
7123 | mddev_put(mddev); | |
1da177e4 LT |
7124 | return err; |
7125 | } | |
7126 | ||
db2a144b | 7127 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 7128 | { |
f72ffdd6 | 7129 | struct mddev *mddev = disk->private_data; |
1da177e4 | 7130 | |
52e5f9d1 | 7131 | BUG_ON(!mddev); |
f2ea68cf | 7132 | atomic_dec(&mddev->openers); |
1da177e4 | 7133 | mddev_put(mddev); |
1da177e4 | 7134 | } |
f0b4f7e2 N |
7135 | |
7136 | static int md_media_changed(struct gendisk *disk) | |
7137 | { | |
fd01b88c | 7138 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7139 | |
7140 | return mddev->changed; | |
7141 | } | |
7142 | ||
7143 | static int md_revalidate(struct gendisk *disk) | |
7144 | { | |
fd01b88c | 7145 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7146 | |
7147 | mddev->changed = 0; | |
7148 | return 0; | |
7149 | } | |
83d5cde4 | 7150 | static const struct block_device_operations md_fops = |
1da177e4 LT |
7151 | { |
7152 | .owner = THIS_MODULE, | |
a39907fa AV |
7153 | .open = md_open, |
7154 | .release = md_release, | |
b492b852 | 7155 | .ioctl = md_ioctl, |
aa98aa31 AB |
7156 | #ifdef CONFIG_COMPAT |
7157 | .compat_ioctl = md_compat_ioctl, | |
7158 | #endif | |
a885c8c4 | 7159 | .getgeo = md_getgeo, |
f0b4f7e2 N |
7160 | .media_changed = md_media_changed, |
7161 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
7162 | }; |
7163 | ||
f72ffdd6 | 7164 | static int md_thread(void *arg) |
1da177e4 | 7165 | { |
2b8bf345 | 7166 | struct md_thread *thread = arg; |
1da177e4 | 7167 | |
1da177e4 LT |
7168 | /* |
7169 | * md_thread is a 'system-thread', it's priority should be very | |
7170 | * high. We avoid resource deadlocks individually in each | |
7171 | * raid personality. (RAID5 does preallocation) We also use RR and | |
7172 | * the very same RT priority as kswapd, thus we will never get | |
7173 | * into a priority inversion deadlock. | |
7174 | * | |
7175 | * we definitely have to have equal or higher priority than | |
7176 | * bdflush, otherwise bdflush will deadlock if there are too | |
7177 | * many dirty RAID5 blocks. | |
7178 | */ | |
1da177e4 | 7179 | |
6985c43f | 7180 | allow_signal(SIGKILL); |
a6fb0934 | 7181 | while (!kthread_should_stop()) { |
1da177e4 | 7182 | |
93588e22 N |
7183 | /* We need to wait INTERRUPTIBLE so that |
7184 | * we don't add to the load-average. | |
7185 | * That means we need to be sure no signals are | |
7186 | * pending | |
7187 | */ | |
7188 | if (signal_pending(current)) | |
7189 | flush_signals(current); | |
7190 | ||
7191 | wait_event_interruptible_timeout | |
7192 | (thread->wqueue, | |
7193 | test_bit(THREAD_WAKEUP, &thread->flags) | |
ce1ccd07 | 7194 | || kthread_should_stop() || kthread_should_park(), |
93588e22 | 7195 | thread->timeout); |
1da177e4 | 7196 | |
6c987910 | 7197 | clear_bit(THREAD_WAKEUP, &thread->flags); |
ce1ccd07 SL |
7198 | if (kthread_should_park()) |
7199 | kthread_parkme(); | |
6c987910 | 7200 | if (!kthread_should_stop()) |
4ed8731d | 7201 | thread->run(thread); |
1da177e4 | 7202 | } |
a6fb0934 | 7203 | |
1da177e4 LT |
7204 | return 0; |
7205 | } | |
7206 | ||
2b8bf345 | 7207 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
7208 | { |
7209 | if (thread) { | |
36a4e1fe | 7210 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
7211 | set_bit(THREAD_WAKEUP, &thread->flags); |
7212 | wake_up(&thread->wqueue); | |
7213 | } | |
7214 | } | |
6c144d31 | 7215 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 7216 | |
4ed8731d SL |
7217 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
7218 | struct mddev *mddev, const char *name) | |
1da177e4 | 7219 | { |
2b8bf345 | 7220 | struct md_thread *thread; |
1da177e4 | 7221 | |
2b8bf345 | 7222 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
7223 | if (!thread) |
7224 | return NULL; | |
7225 | ||
1da177e4 LT |
7226 | init_waitqueue_head(&thread->wqueue); |
7227 | ||
1da177e4 LT |
7228 | thread->run = run; |
7229 | thread->mddev = mddev; | |
32a7627c | 7230 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
7231 | thread->tsk = kthread_run(md_thread, thread, |
7232 | "%s_%s", | |
7233 | mdname(thread->mddev), | |
0232605d | 7234 | name); |
a6fb0934 | 7235 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
7236 | kfree(thread); |
7237 | return NULL; | |
7238 | } | |
1da177e4 LT |
7239 | return thread; |
7240 | } | |
6c144d31 | 7241 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 7242 | |
2b8bf345 | 7243 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 7244 | { |
2b8bf345 | 7245 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
7246 | if (!thread) |
7247 | return; | |
36a4e1fe | 7248 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
7249 | /* Locking ensures that mddev_unlock does not wake_up a |
7250 | * non-existent thread | |
7251 | */ | |
7252 | spin_lock(&pers_lock); | |
7253 | *threadp = NULL; | |
7254 | spin_unlock(&pers_lock); | |
a6fb0934 N |
7255 | |
7256 | kthread_stop(thread->tsk); | |
1da177e4 LT |
7257 | kfree(thread); |
7258 | } | |
6c144d31 | 7259 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 7260 | |
fd01b88c | 7261 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7262 | { |
b2d444d7 | 7263 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 7264 | return; |
6bfe0b49 | 7265 | |
de393cde | 7266 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
7267 | return; |
7268 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
7269 | if (mddev->degraded) |
7270 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 7271 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
7272 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7273 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7274 | md_wakeup_thread(mddev->thread); | |
768a418d | 7275 | if (mddev->event_work.func) |
e804ac78 | 7276 | queue_work(md_misc_wq, &mddev->event_work); |
bb9ef716 | 7277 | md_new_event(mddev); |
1da177e4 | 7278 | } |
6c144d31 | 7279 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
7280 | |
7281 | /* seq_file implementation /proc/mdstat */ | |
7282 | ||
7283 | static void status_unused(struct seq_file *seq) | |
7284 | { | |
7285 | int i = 0; | |
3cb03002 | 7286 | struct md_rdev *rdev; |
1da177e4 LT |
7287 | |
7288 | seq_printf(seq, "unused devices: "); | |
7289 | ||
159ec1fc | 7290 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
7291 | char b[BDEVNAME_SIZE]; |
7292 | i++; | |
7293 | seq_printf(seq, "%s ", | |
7294 | bdevname(rdev->bdev,b)); | |
7295 | } | |
7296 | if (!i) | |
7297 | seq_printf(seq, "<none>"); | |
7298 | ||
7299 | seq_printf(seq, "\n"); | |
7300 | } | |
7301 | ||
f7851be7 | 7302 | static int status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7303 | { |
dd71cf6b N |
7304 | sector_t max_sectors, resync, res; |
7305 | unsigned long dt, db; | |
7306 | sector_t rt; | |
4588b42e N |
7307 | int scale; |
7308 | unsigned int per_milli; | |
1da177e4 | 7309 | |
c804cdec N |
7310 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7311 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 7312 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 7313 | else |
dd71cf6b | 7314 | max_sectors = mddev->dev_sectors; |
1da177e4 | 7315 | |
f7851be7 N |
7316 | resync = mddev->curr_resync; |
7317 | if (resync <= 3) { | |
7318 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7319 | /* Still cleaning up */ | |
7320 | resync = max_sectors; | |
7321 | } else | |
7322 | resync -= atomic_read(&mddev->recovery_active); | |
7323 | ||
7324 | if (resync == 0) { | |
7325 | if (mddev->recovery_cp < MaxSector) { | |
7326 | seq_printf(seq, "\tresync=PENDING"); | |
7327 | return 1; | |
7328 | } | |
7329 | return 0; | |
7330 | } | |
7331 | if (resync < 3) { | |
7332 | seq_printf(seq, "\tresync=DELAYED"); | |
7333 | return 1; | |
7334 | } | |
7335 | ||
403df478 | 7336 | WARN_ON(max_sectors == 0); |
4588b42e | 7337 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 7338 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
7339 | * u32, as those are the requirements for sector_div. |
7340 | * Thus 'scale' must be at least 10 | |
7341 | */ | |
7342 | scale = 10; | |
7343 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 7344 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
7345 | scale++; |
7346 | } | |
7347 | res = (resync>>scale)*1000; | |
dd71cf6b | 7348 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
7349 | |
7350 | per_milli = res; | |
1da177e4 | 7351 | { |
4588b42e | 7352 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
7353 | seq_printf(seq, "["); |
7354 | for (i = 0; i < x; i++) | |
7355 | seq_printf(seq, "="); | |
7356 | seq_printf(seq, ">"); | |
7357 | for (i = 0; i < y; i++) | |
7358 | seq_printf(seq, "."); | |
7359 | seq_printf(seq, "] "); | |
7360 | } | |
4588b42e | 7361 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
7362 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
7363 | "reshape" : | |
61df9d91 N |
7364 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
7365 | "check" : | |
7366 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
7367 | "resync" : "recovery"))), | |
7368 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
7369 | (unsigned long long) resync/2, |
7370 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
7371 | |
7372 | /* | |
1da177e4 LT |
7373 | * dt: time from mark until now |
7374 | * db: blocks written from mark until now | |
7375 | * rt: remaining time | |
dd71cf6b N |
7376 | * |
7377 | * rt is a sector_t, so could be 32bit or 64bit. | |
7378 | * So we divide before multiply in case it is 32bit and close | |
7379 | * to the limit. | |
25985edc | 7380 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
7381 | * near the end of resync when the number of remaining sectors |
7382 | * is close to 'db'. | |
7383 | * We then divide rt by 32 after multiplying by db to compensate. | |
7384 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
7385 | */ |
7386 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
7387 | if (!dt) dt++; | |
ff4e8d9a N |
7388 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
7389 | - mddev->resync_mark_cnt; | |
1da177e4 | 7390 | |
dd71cf6b N |
7391 | rt = max_sectors - resync; /* number of remaining sectors */ |
7392 | sector_div(rt, db/32+1); | |
7393 | rt *= dt; | |
7394 | rt >>= 5; | |
7395 | ||
7396 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
7397 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 7398 | |
ff4e8d9a | 7399 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
f7851be7 | 7400 | return 1; |
1da177e4 LT |
7401 | } |
7402 | ||
7403 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7404 | { | |
7405 | struct list_head *tmp; | |
7406 | loff_t l = *pos; | |
fd01b88c | 7407 | struct mddev *mddev; |
1da177e4 LT |
7408 | |
7409 | if (l >= 0x10000) | |
7410 | return NULL; | |
7411 | if (!l--) | |
7412 | /* header */ | |
7413 | return (void*)1; | |
7414 | ||
7415 | spin_lock(&all_mddevs_lock); | |
7416 | list_for_each(tmp,&all_mddevs) | |
7417 | if (!l--) { | |
fd01b88c | 7418 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7419 | mddev_get(mddev); |
7420 | spin_unlock(&all_mddevs_lock); | |
7421 | return mddev; | |
7422 | } | |
7423 | spin_unlock(&all_mddevs_lock); | |
7424 | if (!l--) | |
7425 | return (void*)2;/* tail */ | |
7426 | return NULL; | |
7427 | } | |
7428 | ||
7429 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7430 | { | |
7431 | struct list_head *tmp; | |
fd01b88c | 7432 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7433 | |
1da177e4 LT |
7434 | ++*pos; |
7435 | if (v == (void*)2) | |
7436 | return NULL; | |
7437 | ||
7438 | spin_lock(&all_mddevs_lock); | |
7439 | if (v == (void*)1) | |
7440 | tmp = all_mddevs.next; | |
7441 | else | |
7442 | tmp = mddev->all_mddevs.next; | |
7443 | if (tmp != &all_mddevs) | |
fd01b88c | 7444 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7445 | else { |
7446 | next_mddev = (void*)2; | |
7447 | *pos = 0x10000; | |
f72ffdd6 | 7448 | } |
1da177e4 LT |
7449 | spin_unlock(&all_mddevs_lock); |
7450 | ||
7451 | if (v != (void*)1) | |
7452 | mddev_put(mddev); | |
7453 | return next_mddev; | |
7454 | ||
7455 | } | |
7456 | ||
7457 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7458 | { | |
fd01b88c | 7459 | struct mddev *mddev = v; |
1da177e4 LT |
7460 | |
7461 | if (mddev && v != (void*)1 && v != (void*)2) | |
7462 | mddev_put(mddev); | |
7463 | } | |
7464 | ||
7465 | static int md_seq_show(struct seq_file *seq, void *v) | |
7466 | { | |
fd01b88c | 7467 | struct mddev *mddev = v; |
dd8ac336 | 7468 | sector_t sectors; |
3cb03002 | 7469 | struct md_rdev *rdev; |
1da177e4 LT |
7470 | |
7471 | if (v == (void*)1) { | |
84fc4b56 | 7472 | struct md_personality *pers; |
1da177e4 LT |
7473 | seq_printf(seq, "Personalities : "); |
7474 | spin_lock(&pers_lock); | |
2604b703 N |
7475 | list_for_each_entry(pers, &pers_list, list) |
7476 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7477 | |
7478 | spin_unlock(&pers_lock); | |
7479 | seq_printf(seq, "\n"); | |
f1514638 | 7480 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7481 | return 0; |
7482 | } | |
7483 | if (v == (void*)2) { | |
7484 | status_unused(seq); | |
7485 | return 0; | |
7486 | } | |
7487 | ||
36d091f4 | 7488 | spin_lock(&mddev->lock); |
1da177e4 LT |
7489 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7490 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7491 | mddev->pers ? "" : "in"); | |
7492 | if (mddev->pers) { | |
f91de92e | 7493 | if (mddev->ro==1) |
1da177e4 | 7494 | seq_printf(seq, " (read-only)"); |
f91de92e | 7495 | if (mddev->ro==2) |
52720ae7 | 7496 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7497 | seq_printf(seq, " %s", mddev->pers->name); |
7498 | } | |
7499 | ||
dd8ac336 | 7500 | sectors = 0; |
f97fcad3 N |
7501 | rcu_read_lock(); |
7502 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7503 | char b[BDEVNAME_SIZE]; |
7504 | seq_printf(seq, " %s[%d]", | |
7505 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7506 | if (test_bit(WriteMostly, &rdev->flags)) |
7507 | seq_printf(seq, "(W)"); | |
9efdca16 SL |
7508 | if (test_bit(Journal, &rdev->flags)) |
7509 | seq_printf(seq, "(J)"); | |
b2d444d7 | 7510 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7511 | seq_printf(seq, "(F)"); |
7512 | continue; | |
2d78f8c4 N |
7513 | } |
7514 | if (rdev->raid_disk < 0) | |
b325a32e | 7515 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7516 | if (test_bit(Replacement, &rdev->flags)) |
7517 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7518 | sectors += rdev->sectors; |
1da177e4 | 7519 | } |
f97fcad3 | 7520 | rcu_read_unlock(); |
1da177e4 LT |
7521 | |
7522 | if (!list_empty(&mddev->disks)) { | |
7523 | if (mddev->pers) | |
7524 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7525 | (unsigned long long) |
7526 | mddev->array_sectors / 2); | |
1da177e4 LT |
7527 | else |
7528 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7529 | (unsigned long long)sectors / 2); |
1da177e4 | 7530 | } |
1cd6bf19 N |
7531 | if (mddev->persistent) { |
7532 | if (mddev->major_version != 0 || | |
7533 | mddev->minor_version != 90) { | |
7534 | seq_printf(seq," super %d.%d", | |
7535 | mddev->major_version, | |
7536 | mddev->minor_version); | |
7537 | } | |
e691063a N |
7538 | } else if (mddev->external) |
7539 | seq_printf(seq, " super external:%s", | |
7540 | mddev->metadata_type); | |
7541 | else | |
1cd6bf19 | 7542 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7543 | |
7544 | if (mddev->pers) { | |
d710e138 | 7545 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7546 | seq_printf(seq, "\n "); |
8e1b39d6 | 7547 | if (mddev->pers->sync_request) { |
f7851be7 | 7548 | if (status_resync(seq, mddev)) |
8e1b39d6 | 7549 | seq_printf(seq, "\n "); |
8e1b39d6 | 7550 | } |
32a7627c N |
7551 | } else |
7552 | seq_printf(seq, "\n "); | |
7553 | ||
57148964 | 7554 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7555 | |
7556 | seq_printf(seq, "\n"); | |
7557 | } | |
36d091f4 | 7558 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7559 | |
1da177e4 LT |
7560 | return 0; |
7561 | } | |
7562 | ||
110518bc | 7563 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7564 | .start = md_seq_start, |
7565 | .next = md_seq_next, | |
7566 | .stop = md_seq_stop, | |
7567 | .show = md_seq_show, | |
7568 | }; | |
7569 | ||
7570 | static int md_seq_open(struct inode *inode, struct file *file) | |
7571 | { | |
f1514638 | 7572 | struct seq_file *seq; |
1da177e4 LT |
7573 | int error; |
7574 | ||
7575 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7576 | if (error) |
f1514638 KS |
7577 | return error; |
7578 | ||
7579 | seq = file->private_data; | |
7580 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7581 | return error; |
7582 | } | |
7583 | ||
e2f23b60 | 7584 | static int md_unloading; |
d7603b7e N |
7585 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7586 | { | |
f1514638 | 7587 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7588 | int mask; |
7589 | ||
e2f23b60 | 7590 | if (md_unloading) |
7d7e64f2 | 7591 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI; |
d7603b7e N |
7592 | poll_wait(filp, &md_event_waiters, wait); |
7593 | ||
7594 | /* always allow read */ | |
7595 | mask = POLLIN | POLLRDNORM; | |
7596 | ||
f1514638 | 7597 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7598 | mask |= POLLERR | POLLPRI; |
7599 | return mask; | |
7600 | } | |
7601 | ||
fa027c2a | 7602 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7603 | .owner = THIS_MODULE, |
1da177e4 LT |
7604 | .open = md_seq_open, |
7605 | .read = seq_read, | |
7606 | .llseek = seq_lseek, | |
c3f94b40 | 7607 | .release = seq_release_private, |
d7603b7e | 7608 | .poll = mdstat_poll, |
1da177e4 LT |
7609 | }; |
7610 | ||
84fc4b56 | 7611 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7612 | { |
9d48739e N |
7613 | pr_debug("md: %s personality registered for level %d\n", |
7614 | p->name, p->level); | |
1da177e4 | 7615 | spin_lock(&pers_lock); |
2604b703 | 7616 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7617 | spin_unlock(&pers_lock); |
7618 | return 0; | |
7619 | } | |
6c144d31 | 7620 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7621 | |
84fc4b56 | 7622 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7623 | { |
9d48739e | 7624 | pr_debug("md: %s personality unregistered\n", p->name); |
1da177e4 | 7625 | spin_lock(&pers_lock); |
2604b703 | 7626 | list_del_init(&p->list); |
1da177e4 LT |
7627 | spin_unlock(&pers_lock); |
7628 | return 0; | |
7629 | } | |
6c144d31 | 7630 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7631 | |
6022e75b N |
7632 | int register_md_cluster_operations(struct md_cluster_operations *ops, |
7633 | struct module *module) | |
edb39c9d | 7634 | { |
6022e75b | 7635 | int ret = 0; |
edb39c9d | 7636 | spin_lock(&pers_lock); |
6022e75b N |
7637 | if (md_cluster_ops != NULL) |
7638 | ret = -EALREADY; | |
7639 | else { | |
7640 | md_cluster_ops = ops; | |
7641 | md_cluster_mod = module; | |
7642 | } | |
edb39c9d | 7643 | spin_unlock(&pers_lock); |
6022e75b | 7644 | return ret; |
edb39c9d GR |
7645 | } |
7646 | EXPORT_SYMBOL(register_md_cluster_operations); | |
7647 | ||
7648 | int unregister_md_cluster_operations(void) | |
7649 | { | |
7650 | spin_lock(&pers_lock); | |
7651 | md_cluster_ops = NULL; | |
7652 | spin_unlock(&pers_lock); | |
7653 | return 0; | |
7654 | } | |
7655 | EXPORT_SYMBOL(unregister_md_cluster_operations); | |
7656 | ||
7657 | int md_setup_cluster(struct mddev *mddev, int nodes) | |
7658 | { | |
47a7b0d8 GJ |
7659 | if (!md_cluster_ops) |
7660 | request_module("md-cluster"); | |
edb39c9d | 7661 | spin_lock(&pers_lock); |
47a7b0d8 | 7662 | /* ensure module won't be unloaded */ |
edb39c9d | 7663 | if (!md_cluster_ops || !try_module_get(md_cluster_mod)) { |
9d48739e | 7664 | pr_warn("can't find md-cluster module or get it's reference.\n"); |
edb39c9d GR |
7665 | spin_unlock(&pers_lock); |
7666 | return -ENOENT; | |
7667 | } | |
7668 | spin_unlock(&pers_lock); | |
7669 | ||
cf921cc1 | 7670 | return md_cluster_ops->join(mddev, nodes); |
edb39c9d GR |
7671 | } |
7672 | ||
7673 | void md_cluster_stop(struct mddev *mddev) | |
7674 | { | |
c4ce867f GR |
7675 | if (!md_cluster_ops) |
7676 | return; | |
edb39c9d GR |
7677 | md_cluster_ops->leave(mddev); |
7678 | module_put(md_cluster_mod); | |
7679 | } | |
7680 | ||
fd01b88c | 7681 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7682 | { |
f72ffdd6 | 7683 | struct md_rdev *rdev; |
1da177e4 | 7684 | int idle; |
eea1bf38 | 7685 | int curr_events; |
1da177e4 LT |
7686 | |
7687 | idle = 1; | |
4b80991c N |
7688 | rcu_read_lock(); |
7689 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7690 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7691 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7692 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7693 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7694 | /* sync IO will cause sync_io to increase before the disk_stats |
7695 | * as sync_io is counted when a request starts, and | |
7696 | * disk_stats is counted when it completes. | |
7697 | * So resync activity will cause curr_events to be smaller than | |
7698 | * when there was no such activity. | |
7699 | * non-sync IO will cause disk_stat to increase without | |
7700 | * increasing sync_io so curr_events will (eventually) | |
7701 | * be larger than it was before. Once it becomes | |
7702 | * substantially larger, the test below will cause | |
7703 | * the array to appear non-idle, and resync will slow | |
7704 | * down. | |
7705 | * If there is a lot of outstanding resync activity when | |
7706 | * we set last_event to curr_events, then all that activity | |
7707 | * completing might cause the array to appear non-idle | |
7708 | * and resync will be slowed down even though there might | |
7709 | * not have been non-resync activity. This will only | |
7710 | * happen once though. 'last_events' will soon reflect | |
7711 | * the state where there is little or no outstanding | |
7712 | * resync requests, and further resync activity will | |
7713 | * always make curr_events less than last_events. | |
c0e48521 | 7714 | * |
1da177e4 | 7715 | */ |
eea1bf38 | 7716 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7717 | rdev->last_events = curr_events; |
7718 | idle = 0; | |
7719 | } | |
7720 | } | |
4b80991c | 7721 | rcu_read_unlock(); |
1da177e4 LT |
7722 | return idle; |
7723 | } | |
7724 | ||
fd01b88c | 7725 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7726 | { |
7727 | /* another "blocks" (512byte) blocks have been synced */ | |
7728 | atomic_sub(blocks, &mddev->recovery_active); | |
7729 | wake_up(&mddev->recovery_wait); | |
7730 | if (!ok) { | |
dfc70645 | 7731 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7732 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7733 | md_wakeup_thread(mddev->thread); |
7734 | // stop recovery, signal do_sync .... | |
7735 | } | |
7736 | } | |
6c144d31 | 7737 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 7738 | |
06d91a5f N |
7739 | /* md_write_start(mddev, bi) |
7740 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7741 | * in superblock) before writing, schedule a superblock update |
7742 | * and wait for it to complete. | |
06d91a5f | 7743 | */ |
fd01b88c | 7744 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7745 | { |
0fd62b86 | 7746 | int did_change = 0; |
06d91a5f | 7747 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7748 | return; |
06d91a5f | 7749 | |
f91de92e N |
7750 | BUG_ON(mddev->ro == 1); |
7751 | if (mddev->ro == 2) { | |
7752 | /* need to switch to read/write */ | |
7753 | mddev->ro = 0; | |
7754 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7755 | md_wakeup_thread(mddev->thread); | |
25156198 | 7756 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7757 | did_change = 1; |
f91de92e | 7758 | } |
06d91a5f | 7759 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7760 | if (mddev->safemode == 1) |
7761 | mddev->safemode = 0; | |
06d91a5f | 7762 | if (mddev->in_sync) { |
85572d7c | 7763 | spin_lock(&mddev->lock); |
3d310eb7 N |
7764 | if (mddev->in_sync) { |
7765 | mddev->in_sync = 0; | |
2953079c SL |
7766 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
7767 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
3d310eb7 | 7768 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7769 | did_change = 1; |
3d310eb7 | 7770 | } |
85572d7c | 7771 | spin_unlock(&mddev->lock); |
06d91a5f | 7772 | } |
0fd62b86 | 7773 | if (did_change) |
00bcb4ac | 7774 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7775 | wait_event(mddev->sb_wait, |
2953079c | 7776 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
1da177e4 | 7777 | } |
6c144d31 | 7778 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 7779 | |
fd01b88c | 7780 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7781 | { |
7782 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7783 | if (mddev->safemode == 2) | |
7784 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7785 | else if (mddev->safemode_delay) |
1da177e4 LT |
7786 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7787 | } | |
7788 | } | |
6c144d31 | 7789 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 7790 | |
2a2275d6 N |
7791 | /* md_allow_write(mddev) |
7792 | * Calling this ensures that the array is marked 'active' so that writes | |
7793 | * may proceed without blocking. It is important to call this before | |
7794 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7795 | * Must be called with mddev_lock held. | |
b5470dc5 | 7796 | * |
2953079c | 7797 | * In the ->external case MD_SB_CHANGE_PENDING can not be cleared until mddev->lock |
b5470dc5 | 7798 | * is dropped, so return -EAGAIN after notifying userspace. |
2a2275d6 | 7799 | */ |
fd01b88c | 7800 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7801 | { |
7802 | if (!mddev->pers) | |
b5470dc5 | 7803 | return 0; |
2a2275d6 | 7804 | if (mddev->ro) |
b5470dc5 | 7805 | return 0; |
1a0fd497 | 7806 | if (!mddev->pers->sync_request) |
b5470dc5 | 7807 | return 0; |
2a2275d6 | 7808 | |
85572d7c | 7809 | spin_lock(&mddev->lock); |
2a2275d6 N |
7810 | if (mddev->in_sync) { |
7811 | mddev->in_sync = 0; | |
2953079c SL |
7812 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
7813 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
2a2275d6 N |
7814 | if (mddev->safemode_delay && |
7815 | mddev->safemode == 0) | |
7816 | mddev->safemode = 1; | |
85572d7c | 7817 | spin_unlock(&mddev->lock); |
2a2275d6 | 7818 | md_update_sb(mddev, 0); |
00bcb4ac | 7819 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 | 7820 | } else |
85572d7c | 7821 | spin_unlock(&mddev->lock); |
b5470dc5 | 7822 | |
2953079c | 7823 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
b5470dc5 DW |
7824 | return -EAGAIN; |
7825 | else | |
7826 | return 0; | |
2a2275d6 N |
7827 | } |
7828 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7829 | ||
1da177e4 LT |
7830 | #define SYNC_MARKS 10 |
7831 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7832 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7833 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7834 | { |
4ed8731d | 7835 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7836 | struct mddev *mddev2; |
1da177e4 LT |
7837 | unsigned int currspeed = 0, |
7838 | window; | |
ac7e50a3 | 7839 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 7840 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7841 | unsigned long update_time; |
1da177e4 LT |
7842 | sector_t mark_cnt[SYNC_MARKS]; |
7843 | int last_mark,m; | |
7844 | struct list_head *tmp; | |
7845 | sector_t last_check; | |
57afd89f | 7846 | int skipped = 0; |
3cb03002 | 7847 | struct md_rdev *rdev; |
c4a39551 | 7848 | char *desc, *action = NULL; |
7c2c57c9 | 7849 | struct blk_plug plug; |
41a9a0dc | 7850 | int ret; |
1da177e4 LT |
7851 | |
7852 | /* just incase thread restarts... */ | |
7853 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7854 | return; | |
3991b31e N |
7855 | if (mddev->ro) {/* never try to sync a read-only array */ |
7856 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 7857 | return; |
3991b31e | 7858 | } |
1da177e4 | 7859 | |
41a9a0dc GJ |
7860 | if (mddev_is_clustered(mddev)) { |
7861 | ret = md_cluster_ops->resync_start(mddev); | |
7862 | if (ret) | |
7863 | goto skip; | |
7864 | ||
bb8bf15b | 7865 | set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags); |
41a9a0dc GJ |
7866 | if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7867 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) || | |
7868 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) | |
7869 | && ((unsigned long long)mddev->curr_resync_completed | |
7870 | < (unsigned long long)mddev->resync_max_sectors)) | |
7871 | goto skip; | |
7872 | } | |
7873 | ||
61df9d91 | 7874 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 7875 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 7876 | desc = "data-check"; |
c4a39551 JB |
7877 | action = "check"; |
7878 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 7879 | desc = "requested-resync"; |
c4a39551 JB |
7880 | action = "repair"; |
7881 | } else | |
61df9d91 N |
7882 | desc = "resync"; |
7883 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7884 | desc = "reshape"; | |
7885 | else | |
7886 | desc = "recovery"; | |
7887 | ||
c4a39551 JB |
7888 | mddev->last_sync_action = action ?: desc; |
7889 | ||
1da177e4 LT |
7890 | /* we overload curr_resync somewhat here. |
7891 | * 0 == not engaged in resync at all | |
7892 | * 2 == checking that there is no conflict with another sync | |
7893 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7894 | * commense | |
7895 | * other == active in resync - this many blocks | |
7896 | * | |
7897 | * Before starting a resync we must have set curr_resync to | |
7898 | * 2, and then checked that every "conflicting" array has curr_resync | |
7899 | * less than ours. When we find one that is the same or higher | |
7900 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7901 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7902 | * This will mean we have to start checking from the beginning again. | |
7903 | * | |
7904 | */ | |
7905 | ||
7906 | do { | |
c622ca54 | 7907 | int mddev2_minor = -1; |
1da177e4 LT |
7908 | mddev->curr_resync = 2; |
7909 | ||
7910 | try_again: | |
404e4b43 | 7911 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 7912 | goto skip; |
29ac4aa3 | 7913 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7914 | if (mddev2 == mddev) |
7915 | continue; | |
90b08710 BS |
7916 | if (!mddev->parallel_resync |
7917 | && mddev2->curr_resync | |
7918 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7919 | DEFINE_WAIT(wq); |
7920 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7921 | /* arbitrarily yield */ | |
7922 | mddev->curr_resync = 1; | |
7923 | wake_up(&resync_wait); | |
7924 | } | |
7925 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7926 | /* no need to wait here, we can wait the next | |
7927 | * time 'round when curr_resync == 2 | |
7928 | */ | |
7929 | continue; | |
9744197c N |
7930 | /* We need to wait 'interruptible' so as not to |
7931 | * contribute to the load average, and not to | |
7932 | * be caught by 'softlockup' | |
7933 | */ | |
7934 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 7935 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 7936 | mddev2->curr_resync >= mddev->curr_resync) { |
c622ca54 AP |
7937 | if (mddev2_minor != mddev2->md_minor) { |
7938 | mddev2_minor = mddev2->md_minor; | |
9d48739e N |
7939 | pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n", |
7940 | desc, mdname(mddev), | |
7941 | mdname(mddev2)); | |
c622ca54 | 7942 | } |
1da177e4 | 7943 | mddev_put(mddev2); |
9744197c N |
7944 | if (signal_pending(current)) |
7945 | flush_signals(current); | |
1da177e4 LT |
7946 | schedule(); |
7947 | finish_wait(&resync_wait, &wq); | |
7948 | goto try_again; | |
7949 | } | |
7950 | finish_wait(&resync_wait, &wq); | |
7951 | } | |
7952 | } | |
7953 | } while (mddev->curr_resync < 2); | |
7954 | ||
5fd6c1dc | 7955 | j = 0; |
9d88883e | 7956 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7957 | /* resync follows the size requested by the personality, |
57afd89f | 7958 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7959 | */ |
7960 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7961 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7962 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7963 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7964 | j = mddev->resync_min; | |
7965 | else if (!mddev->bitmap) | |
5fd6c1dc | 7966 | j = mddev->recovery_cp; |
5e96ee65 | 7967 | |
ccfcc3c1 | 7968 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7969 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7970 | else { |
1da177e4 | 7971 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7972 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7973 | j = MaxSector; |
4e59ca7d | 7974 | rcu_read_lock(); |
dafb20fa | 7975 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7976 | if (rdev->raid_disk >= 0 && |
f2076e7d | 7977 | !test_bit(Journal, &rdev->flags) && |
5fd6c1dc N |
7978 | !test_bit(Faulty, &rdev->flags) && |
7979 | !test_bit(In_sync, &rdev->flags) && | |
7980 | rdev->recovery_offset < j) | |
7981 | j = rdev->recovery_offset; | |
4e59ca7d | 7982 | rcu_read_unlock(); |
133d4527 N |
7983 | |
7984 | /* If there is a bitmap, we need to make sure all | |
7985 | * writes that started before we added a spare | |
7986 | * complete before we start doing a recovery. | |
7987 | * Otherwise the write might complete and (via | |
7988 | * bitmap_endwrite) set a bit in the bitmap after the | |
7989 | * recovery has checked that bit and skipped that | |
7990 | * region. | |
7991 | */ | |
7992 | if (mddev->bitmap) { | |
7993 | mddev->pers->quiesce(mddev, 1); | |
7994 | mddev->pers->quiesce(mddev, 0); | |
7995 | } | |
5fd6c1dc | 7996 | } |
1da177e4 | 7997 | |
9d48739e N |
7998 | pr_info("md: %s of RAID array %s\n", desc, mdname(mddev)); |
7999 | pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev)); | |
8000 | pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n", | |
8001 | speed_max(mddev), desc); | |
1da177e4 | 8002 | |
eea1bf38 | 8003 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 8004 | |
57afd89f | 8005 | io_sectors = 0; |
1da177e4 LT |
8006 | for (m = 0; m < SYNC_MARKS; m++) { |
8007 | mark[m] = jiffies; | |
57afd89f | 8008 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
8009 | } |
8010 | last_mark = 0; | |
8011 | mddev->resync_mark = mark[last_mark]; | |
8012 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
8013 | ||
8014 | /* | |
8015 | * Tune reconstruction: | |
8016 | */ | |
8017 | window = 32*(PAGE_SIZE/512); | |
9d48739e N |
8018 | pr_debug("md: using %dk window, over a total of %lluk.\n", |
8019 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
8020 | |
8021 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
8022 | last_check = 0; |
8023 | ||
8024 | if (j>2) { | |
9d48739e N |
8025 | pr_debug("md: resuming %s of %s from checkpoint.\n", |
8026 | desc, mdname(mddev)); | |
1da177e4 | 8027 | mddev->curr_resync = j; |
72f36d59 N |
8028 | } else |
8029 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 8030 | mddev->curr_resync_completed = j; |
72f36d59 N |
8031 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
8032 | md_new_event(mddev); | |
54f89341 | 8033 | update_time = jiffies; |
1da177e4 | 8034 | |
7c2c57c9 | 8035 | blk_start_plug(&plug); |
1da177e4 | 8036 | while (j < max_sectors) { |
57afd89f | 8037 | sector_t sectors; |
1da177e4 | 8038 | |
57afd89f | 8039 | skipped = 0; |
97e4f42d | 8040 | |
7a91ee1f N |
8041 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8042 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
8043 | (mddev->curr_resync - mddev->curr_resync_completed) | |
8044 | > (max_sectors >> 4)) || | |
54f89341 | 8045 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f | 8046 | (j - mddev->curr_resync_completed)*2 |
c5e19d90 N |
8047 | >= mddev->resync_max - mddev->curr_resync_completed || |
8048 | mddev->curr_resync_completed > mddev->resync_max | |
7a91ee1f | 8049 | )) { |
97e4f42d | 8050 | /* time to update curr_resync_completed */ |
97e4f42d N |
8051 | wait_event(mddev->recovery_wait, |
8052 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 8053 | mddev->curr_resync_completed = j; |
35d78c66 | 8054 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
8055 | j > mddev->recovery_cp) | |
8056 | mddev->recovery_cp = j; | |
54f89341 | 8057 | update_time = jiffies; |
2953079c | 8058 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
acb180b0 | 8059 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 8060 | } |
acb180b0 | 8061 | |
c91abf5a N |
8062 | while (j >= mddev->resync_max && |
8063 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
8064 | /* As this condition is controlled by user-space, |
8065 | * we can block indefinitely, so use '_interruptible' | |
8066 | * to avoid triggering warnings. | |
8067 | */ | |
8068 | flush_signals(current); /* just in case */ | |
8069 | wait_event_interruptible(mddev->recovery_wait, | |
8070 | mddev->resync_max > j | |
c91abf5a N |
8071 | || test_bit(MD_RECOVERY_INTR, |
8072 | &mddev->recovery)); | |
e62e58a5 | 8073 | } |
acb180b0 | 8074 | |
c91abf5a N |
8075 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8076 | break; | |
acb180b0 | 8077 | |
09314799 | 8078 | sectors = mddev->pers->sync_request(mddev, j, &skipped); |
57afd89f | 8079 | if (sectors == 0) { |
dfc70645 | 8080 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 8081 | break; |
1da177e4 | 8082 | } |
57afd89f N |
8083 | |
8084 | if (!skipped) { /* actual IO requested */ | |
8085 | io_sectors += sectors; | |
8086 | atomic_add(sectors, &mddev->recovery_active); | |
8087 | } | |
8088 | ||
e875ecea N |
8089 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8090 | break; | |
8091 | ||
1da177e4 | 8092 | j += sectors; |
5ed1df2e N |
8093 | if (j > max_sectors) |
8094 | /* when skipping, extra large numbers can be returned. */ | |
8095 | j = max_sectors; | |
72f36d59 N |
8096 | if (j > 2) |
8097 | mddev->curr_resync = j; | |
ff4e8d9a | 8098 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 8099 | if (last_check == 0) |
e875ecea | 8100 | /* this is the earliest that rebuild will be |
d7603b7e N |
8101 | * visible in /proc/mdstat |
8102 | */ | |
8103 | md_new_event(mddev); | |
57afd89f N |
8104 | |
8105 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
8106 | continue; |
8107 | ||
57afd89f | 8108 | last_check = io_sectors; |
1da177e4 LT |
8109 | repeat: |
8110 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
8111 | /* step marks */ | |
8112 | int next = (last_mark+1) % SYNC_MARKS; | |
8113 | ||
8114 | mddev->resync_mark = mark[next]; | |
8115 | mddev->resync_mark_cnt = mark_cnt[next]; | |
8116 | mark[next] = jiffies; | |
57afd89f | 8117 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
8118 | last_mark = next; |
8119 | } | |
8120 | ||
c91abf5a N |
8121 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8122 | break; | |
1da177e4 LT |
8123 | |
8124 | /* | |
8125 | * this loop exits only if either when we are slower than | |
8126 | * the 'hard' speed limit, or the system was IO-idle for | |
8127 | * a jiffy. | |
8128 | * the system might be non-idle CPU-wise, but we only care | |
8129 | * about not overloading the IO subsystem. (things like an | |
8130 | * e2fsck being done on the RAID array should execute fast) | |
8131 | */ | |
1da177e4 LT |
8132 | cond_resched(); |
8133 | ||
ac7e50a3 XN |
8134 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
8135 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 8136 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 8137 | |
88202a0c | 8138 | if (currspeed > speed_min(mddev)) { |
ac8fa419 | 8139 | if (currspeed > speed_max(mddev)) { |
c0e48521 | 8140 | msleep(500); |
1da177e4 LT |
8141 | goto repeat; |
8142 | } | |
ac8fa419 N |
8143 | if (!is_mddev_idle(mddev, 0)) { |
8144 | /* | |
8145 | * Give other IO more of a chance. | |
8146 | * The faster the devices, the less we wait. | |
8147 | */ | |
8148 | wait_event(mddev->recovery_wait, | |
8149 | !atomic_read(&mddev->recovery_active)); | |
8150 | } | |
1da177e4 LT |
8151 | } |
8152 | } | |
9d48739e N |
8153 | pr_info("md: %s: %s %s.\n",mdname(mddev), desc, |
8154 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
8155 | ? "interrupted" : "done"); | |
1da177e4 LT |
8156 | /* |
8157 | * this also signals 'finished resyncing' to md_stop | |
8158 | */ | |
7c2c57c9 | 8159 | blk_finish_plug(&plug); |
1da177e4 LT |
8160 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
8161 | ||
5ed1df2e N |
8162 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8163 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
1217e1d1 | 8164 | mddev->curr_resync > 3) { |
5ed1df2e N |
8165 | mddev->curr_resync_completed = mddev->curr_resync; |
8166 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
8167 | } | |
09314799 | 8168 | mddev->pers->sync_request(mddev, max_sectors, &skipped); |
1da177e4 | 8169 | |
dfc70645 | 8170 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
1217e1d1 | 8171 | mddev->curr_resync > 3) { |
5fd6c1dc N |
8172 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
8173 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8174 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
9d48739e N |
8175 | pr_debug("md: checkpointing %s of %s.\n", |
8176 | desc, mdname(mddev)); | |
0a19caab | 8177 | if (test_bit(MD_RECOVERY_ERROR, |
8178 | &mddev->recovery)) | |
8179 | mddev->recovery_cp = | |
8180 | mddev->curr_resync_completed; | |
8181 | else | |
8182 | mddev->recovery_cp = | |
8183 | mddev->curr_resync; | |
5fd6c1dc N |
8184 | } |
8185 | } else | |
8186 | mddev->recovery_cp = MaxSector; | |
8187 | } else { | |
8188 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
8189 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 8190 | rcu_read_lock(); |
dafb20fa | 8191 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 8192 | if (rdev->raid_disk >= 0 && |
70fffd0b | 8193 | mddev->delta_disks >= 0 && |
f2076e7d | 8194 | !test_bit(Journal, &rdev->flags) && |
5fd6c1dc N |
8195 | !test_bit(Faulty, &rdev->flags) && |
8196 | !test_bit(In_sync, &rdev->flags) && | |
8197 | rdev->recovery_offset < mddev->curr_resync) | |
8198 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 8199 | rcu_read_unlock(); |
5fd6c1dc | 8200 | } |
1da177e4 | 8201 | } |
db91ff55 | 8202 | skip: |
bb8bf15b GJ |
8203 | /* set CHANGE_PENDING here since maybe another update is needed, |
8204 | * so other nodes are informed. It should be harmless for normal | |
8205 | * raid */ | |
2953079c SL |
8206 | set_mask_bits(&mddev->sb_flags, 0, |
8207 | BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS)); | |
c186b128 | 8208 | |
23da422b | 8209 | spin_lock(&mddev->lock); |
c07b70ad N |
8210 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
8211 | /* We completed so min/max setting can be forgotten if used. */ | |
8212 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8213 | mddev->resync_min = 0; | |
8214 | mddev->resync_max = MaxSector; | |
8215 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8216 | mddev->resync_min = mddev->curr_resync_completed; | |
f7851be7 | 8217 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); |
1da177e4 | 8218 | mddev->curr_resync = 0; |
23da422b N |
8219 | spin_unlock(&mddev->lock); |
8220 | ||
1da177e4 | 8221 | wake_up(&resync_wait); |
1da177e4 | 8222 | md_wakeup_thread(mddev->thread); |
c6207277 | 8223 | return; |
1da177e4 | 8224 | } |
29269553 | 8225 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 8226 | |
746d3207 N |
8227 | static int remove_and_add_spares(struct mddev *mddev, |
8228 | struct md_rdev *this) | |
b4c4c7b8 | 8229 | { |
3cb03002 | 8230 | struct md_rdev *rdev; |
b4c4c7b8 | 8231 | int spares = 0; |
f2a371c5 | 8232 | int removed = 0; |
d787be40 | 8233 | bool remove_some = false; |
b4c4c7b8 | 8234 | |
d787be40 N |
8235 | rdev_for_each(rdev, mddev) { |
8236 | if ((this == NULL || rdev == this) && | |
8237 | rdev->raid_disk >= 0 && | |
8238 | !test_bit(Blocked, &rdev->flags) && | |
8239 | test_bit(Faulty, &rdev->flags) && | |
8240 | atomic_read(&rdev->nr_pending)==0) { | |
8241 | /* Faulty non-Blocked devices with nr_pending == 0 | |
8242 | * never get nr_pending incremented, | |
8243 | * never get Faulty cleared, and never get Blocked set. | |
8244 | * So we can synchronize_rcu now rather than once per device | |
8245 | */ | |
8246 | remove_some = true; | |
8247 | set_bit(RemoveSynchronized, &rdev->flags); | |
8248 | } | |
8249 | } | |
8250 | ||
8251 | if (remove_some) | |
8252 | synchronize_rcu(); | |
8253 | rdev_for_each(rdev, mddev) { | |
746d3207 N |
8254 | if ((this == NULL || rdev == this) && |
8255 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 8256 | !test_bit(Blocked, &rdev->flags) && |
d787be40 | 8257 | ((test_bit(RemoveSynchronized, &rdev->flags) || |
f2076e7d SL |
8258 | (!test_bit(In_sync, &rdev->flags) && |
8259 | !test_bit(Journal, &rdev->flags))) && | |
d787be40 | 8260 | atomic_read(&rdev->nr_pending)==0)) { |
b4c4c7b8 | 8261 | if (mddev->pers->hot_remove_disk( |
b8321b68 | 8262 | mddev, rdev) == 0) { |
36fad858 | 8263 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 8264 | rdev->raid_disk = -1; |
f2a371c5 | 8265 | removed++; |
b4c4c7b8 N |
8266 | } |
8267 | } | |
d787be40 N |
8268 | if (remove_some && test_bit(RemoveSynchronized, &rdev->flags)) |
8269 | clear_bit(RemoveSynchronized, &rdev->flags); | |
8270 | } | |
8271 | ||
90584fc9 JB |
8272 | if (removed && mddev->kobj.sd) |
8273 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 8274 | |
2910ff17 | 8275 | if (this && removed) |
746d3207 N |
8276 | goto no_add; |
8277 | ||
dafb20fa | 8278 | rdev_for_each(rdev, mddev) { |
2910ff17 GR |
8279 | if (this && this != rdev) |
8280 | continue; | |
dbb64f86 GR |
8281 | if (test_bit(Candidate, &rdev->flags)) |
8282 | continue; | |
7bfec5f3 N |
8283 | if (rdev->raid_disk >= 0 && |
8284 | !test_bit(In_sync, &rdev->flags) && | |
f2076e7d | 8285 | !test_bit(Journal, &rdev->flags) && |
7bfec5f3 N |
8286 | !test_bit(Faulty, &rdev->flags)) |
8287 | spares++; | |
7ceb17e8 N |
8288 | if (rdev->raid_disk >= 0) |
8289 | continue; | |
8290 | if (test_bit(Faulty, &rdev->flags)) | |
8291 | continue; | |
f6b6ec5c SL |
8292 | if (!test_bit(Journal, &rdev->flags)) { |
8293 | if (mddev->ro && | |
8294 | ! (rdev->saved_raid_disk >= 0 && | |
8295 | !test_bit(Bitmap_sync, &rdev->flags))) | |
8296 | continue; | |
7ceb17e8 | 8297 | |
f6b6ec5c SL |
8298 | rdev->recovery_offset = 0; |
8299 | } | |
7ceb17e8 N |
8300 | if (mddev->pers-> |
8301 | hot_add_disk(mddev, rdev) == 0) { | |
8302 | if (sysfs_link_rdev(mddev, rdev)) | |
8303 | /* failure here is OK */; | |
f6b6ec5c SL |
8304 | if (!test_bit(Journal, &rdev->flags)) |
8305 | spares++; | |
7ceb17e8 | 8306 | md_new_event(mddev); |
2953079c | 8307 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
dfc70645 | 8308 | } |
b4c4c7b8 | 8309 | } |
746d3207 | 8310 | no_add: |
6dafab6b | 8311 | if (removed) |
2953079c | 8312 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
b4c4c7b8 N |
8313 | return spares; |
8314 | } | |
7ebc0be7 | 8315 | |
ac05f256 N |
8316 | static void md_start_sync(struct work_struct *ws) |
8317 | { | |
8318 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
c186b128 | 8319 | |
ac05f256 N |
8320 | mddev->sync_thread = md_register_thread(md_do_sync, |
8321 | mddev, | |
8322 | "resync"); | |
8323 | if (!mddev->sync_thread) { | |
9d48739e N |
8324 | pr_warn("%s: could not start resync thread...\n", |
8325 | mdname(mddev)); | |
ac05f256 N |
8326 | /* leave the spares where they are, it shouldn't hurt */ |
8327 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8328 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8329 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8330 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
8331 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8332 | wake_up(&resync_wait); |
ac05f256 N |
8333 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8334 | &mddev->recovery)) | |
8335 | if (mddev->sysfs_action) | |
8336 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8337 | } else | |
8338 | md_wakeup_thread(mddev->sync_thread); | |
8339 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8340 | md_new_event(mddev); | |
8341 | } | |
8342 | ||
1da177e4 LT |
8343 | /* |
8344 | * This routine is regularly called by all per-raid-array threads to | |
8345 | * deal with generic issues like resync and super-block update. | |
8346 | * Raid personalities that don't have a thread (linear/raid0) do not | |
8347 | * need this as they never do any recovery or update the superblock. | |
8348 | * | |
8349 | * It does not do any resync itself, but rather "forks" off other threads | |
8350 | * to do that as needed. | |
8351 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
8352 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 8353 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
8354 | * and wakeups up this thread which will reap the thread and finish up. |
8355 | * This thread also removes any faulty devices (with nr_pending == 0). | |
8356 | * | |
8357 | * The overall approach is: | |
8358 | * 1/ if the superblock needs updating, update it. | |
8359 | * 2/ If a recovery thread is running, don't do anything else. | |
8360 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
8361 | * 4/ If there are any faulty devices, remove them. | |
8362 | * 5/ If array is degraded, try to add spares devices | |
8363 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
8364 | */ | |
fd01b88c | 8365 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 8366 | { |
68866e42 JB |
8367 | if (mddev->suspended) |
8368 | return; | |
8369 | ||
5f40402d | 8370 | if (mddev->bitmap) |
aa5cbd10 | 8371 | bitmap_daemon_work(mddev); |
1da177e4 | 8372 | |
fca4d848 | 8373 | if (signal_pending(current)) { |
31a59e34 | 8374 | if (mddev->pers->sync_request && !mddev->external) { |
9d48739e N |
8375 | pr_debug("md: %s in immediate safe mode\n", |
8376 | mdname(mddev)); | |
fca4d848 N |
8377 | mddev->safemode = 2; |
8378 | } | |
8379 | flush_signals(current); | |
8380 | } | |
8381 | ||
c89a8eee N |
8382 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
8383 | return; | |
1da177e4 | 8384 | if ( ! ( |
2953079c | 8385 | (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) || |
1da177e4 | 8386 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 8387 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
15858fa5 | 8388 | test_bit(MD_RELOAD_SB, &mddev->flags) || |
31a59e34 | 8389 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
8390 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
8391 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
8392 | )) |
8393 | return; | |
fca4d848 | 8394 | |
df5b89b3 | 8395 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 8396 | int spares = 0; |
fca4d848 | 8397 | |
c89a8eee | 8398 | if (mddev->ro) { |
ab16bfc7 NB |
8399 | struct md_rdev *rdev; |
8400 | if (!mddev->external && mddev->in_sync) | |
8401 | /* 'Blocked' flag not needed as failed devices | |
8402 | * will be recorded if array switched to read/write. | |
8403 | * Leaving it set will prevent the device | |
8404 | * from being removed. | |
8405 | */ | |
8406 | rdev_for_each(rdev, mddev) | |
8407 | clear_bit(Blocked, &rdev->flags); | |
7ceb17e8 N |
8408 | /* On a read-only array we can: |
8409 | * - remove failed devices | |
8410 | * - add already-in_sync devices if the array itself | |
8411 | * is in-sync. | |
8412 | * As we only add devices that are already in-sync, | |
8413 | * we can activate the spares immediately. | |
c89a8eee | 8414 | */ |
7ceb17e8 | 8415 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
8416 | /* There is no thread, but we need to call |
8417 | * ->spare_active and clear saved_raid_disk | |
8418 | */ | |
2ac295a5 | 8419 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 | 8420 | md_reap_sync_thread(mddev); |
a4a3d26d | 8421 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
8313b8e5 | 8422 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2953079c | 8423 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
c89a8eee N |
8424 | goto unlock; |
8425 | } | |
8426 | ||
659b254f GJ |
8427 | if (mddev_is_clustered(mddev)) { |
8428 | struct md_rdev *rdev; | |
8429 | /* kick the device if another node issued a | |
8430 | * remove disk. | |
8431 | */ | |
8432 | rdev_for_each(rdev, mddev) { | |
8433 | if (test_and_clear_bit(ClusterRemove, &rdev->flags) && | |
8434 | rdev->raid_disk < 0) | |
8435 | md_kick_rdev_from_array(rdev); | |
8436 | } | |
15858fa5 GJ |
8437 | |
8438 | if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags)) | |
8439 | md_reload_sb(mddev, mddev->good_device_nr); | |
659b254f GJ |
8440 | } |
8441 | ||
31a59e34 | 8442 | if (!mddev->external) { |
0fd62b86 | 8443 | int did_change = 0; |
85572d7c | 8444 | spin_lock(&mddev->lock); |
31a59e34 N |
8445 | if (mddev->safemode && |
8446 | !atomic_read(&mddev->writes_pending) && | |
8447 | !mddev->in_sync && | |
8448 | mddev->recovery_cp == MaxSector) { | |
8449 | mddev->in_sync = 1; | |
0fd62b86 | 8450 | did_change = 1; |
2953079c | 8451 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
31a59e34 N |
8452 | } |
8453 | if (mddev->safemode == 1) | |
8454 | mddev->safemode = 0; | |
85572d7c | 8455 | spin_unlock(&mddev->lock); |
0fd62b86 | 8456 | if (did_change) |
00bcb4ac | 8457 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 8458 | } |
fca4d848 | 8459 | |
2953079c | 8460 | if (mddev->sb_flags) |
850b2b42 | 8461 | md_update_sb(mddev, 0); |
06d91a5f | 8462 | |
1da177e4 LT |
8463 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
8464 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
8465 | /* resync/recovery still happening */ | |
8466 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8467 | goto unlock; | |
8468 | } | |
8469 | if (mddev->sync_thread) { | |
a91d5ac0 | 8470 | md_reap_sync_thread(mddev); |
1da177e4 LT |
8471 | goto unlock; |
8472 | } | |
72a23c21 NB |
8473 | /* Set RUNNING before clearing NEEDED to avoid |
8474 | * any transients in the value of "sync_action". | |
8475 | */ | |
72f36d59 | 8476 | mddev->curr_resync_completed = 0; |
23da422b | 8477 | spin_lock(&mddev->lock); |
72a23c21 | 8478 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 8479 | spin_unlock(&mddev->lock); |
24dd469d N |
8480 | /* Clear some bits that don't mean anything, but |
8481 | * might be left set | |
8482 | */ | |
24dd469d N |
8483 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8484 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 8485 | |
ed209584 N |
8486 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
8487 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 8488 | goto not_running; |
1da177e4 LT |
8489 | /* no recovery is running. |
8490 | * remove any failed drives, then | |
8491 | * add spares if possible. | |
72f36d59 | 8492 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
8493 | * the personality to fail the re-add. |
8494 | */ | |
1da177e4 | 8495 | |
b4c4c7b8 | 8496 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
8497 | if (mddev->pers->check_reshape == NULL || |
8498 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 8499 | /* Cannot proceed */ |
ac05f256 | 8500 | goto not_running; |
b4c4c7b8 | 8501 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 8502 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 8503 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
8504 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8505 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 8506 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 8507 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8508 | } else if (mddev->recovery_cp < MaxSector) { |
8509 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 8510 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8511 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
8512 | /* nothing to be done ... */ | |
ac05f256 | 8513 | goto not_running; |
24dd469d | 8514 | |
1da177e4 | 8515 | if (mddev->pers->sync_request) { |
ef99bf48 | 8516 | if (spares) { |
a654b9d8 N |
8517 | /* We are adding a device or devices to an array |
8518 | * which has the bitmap stored on all devices. | |
8519 | * So make sure all bitmap pages get written | |
8520 | */ | |
8521 | bitmap_write_all(mddev->bitmap); | |
8522 | } | |
ac05f256 N |
8523 | INIT_WORK(&mddev->del_work, md_start_sync); |
8524 | queue_work(md_misc_wq, &mddev->del_work); | |
8525 | goto unlock; | |
1da177e4 | 8526 | } |
ac05f256 | 8527 | not_running: |
72a23c21 NB |
8528 | if (!mddev->sync_thread) { |
8529 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8530 | wake_up(&resync_wait); |
72a23c21 NB |
8531 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8532 | &mddev->recovery)) | |
0c3573f1 | 8533 | if (mddev->sysfs_action) |
00bcb4ac | 8534 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 8535 | } |
ac05f256 N |
8536 | unlock: |
8537 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
8538 | mddev_unlock(mddev); |
8539 | } | |
8540 | } | |
6c144d31 | 8541 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8542 | |
a91d5ac0 JB |
8543 | void md_reap_sync_thread(struct mddev *mddev) |
8544 | { | |
8545 | struct md_rdev *rdev; | |
8546 | ||
8547 | /* resync has finished, collect result */ | |
8548 | md_unregister_thread(&mddev->sync_thread); | |
8549 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8550 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8551 | /* success...*/ | |
8552 | /* activate any spares */ | |
8553 | if (mddev->pers->spare_active(mddev)) { | |
8554 | sysfs_notify(&mddev->kobj, NULL, | |
8555 | "degraded"); | |
2953079c | 8556 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a91d5ac0 JB |
8557 | } |
8558 | } | |
8559 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
8560 | mddev->pers->finish_reshape) | |
8561 | mddev->pers->finish_reshape(mddev); | |
8562 | ||
8563 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8564 | * information must be scrapped. |
a91d5ac0 | 8565 | */ |
f466722c N |
8566 | if (!mddev->degraded) |
8567 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8568 | rdev->saved_raid_disk = -1; |
8569 | ||
8570 | md_update_sb(mddev, 1); | |
2953079c | 8571 | /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can |
bb8bf15b GJ |
8572 | * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by |
8573 | * clustered raid */ | |
8574 | if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags)) | |
8575 | md_cluster_ops->resync_finish(mddev); | |
a91d5ac0 | 8576 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
ea358cd0 | 8577 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
a91d5ac0 JB |
8578 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8579 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8580 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8581 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8582 | wake_up(&resync_wait); |
a91d5ac0 JB |
8583 | /* flag recovery needed just to double check */ |
8584 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8585 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8586 | md_new_event(mddev); | |
8587 | if (mddev->event_work.func) | |
8588 | queue_work(md_misc_wq, &mddev->event_work); | |
8589 | } | |
6c144d31 | 8590 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 8591 | |
fd01b88c | 8592 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8593 | { |
00bcb4ac | 8594 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8595 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8596 | !test_bit(Blocked, &rdev->flags) && |
8597 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8598 | msecs_to_jiffies(5000)); |
8599 | rdev_dec_pending(rdev, mddev); | |
8600 | } | |
8601 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8602 | ||
c6563a8c N |
8603 | void md_finish_reshape(struct mddev *mddev) |
8604 | { | |
8605 | /* called be personality module when reshape completes. */ | |
8606 | struct md_rdev *rdev; | |
8607 | ||
8608 | rdev_for_each(rdev, mddev) { | |
8609 | if (rdev->data_offset > rdev->new_data_offset) | |
8610 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8611 | else | |
8612 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8613 | rdev->data_offset = rdev->new_data_offset; | |
8614 | } | |
8615 | } | |
8616 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 | 8617 | |
fc974ee2 | 8618 | /* Bad block management */ |
2230dfe4 | 8619 | |
fc974ee2 | 8620 | /* Returns 1 on success, 0 on failure */ |
3cb03002 | 8621 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8622 | int is_new) |
2230dfe4 | 8623 | { |
85ad1d13 | 8624 | struct mddev *mddev = rdev->mddev; |
c6563a8c N |
8625 | int rv; |
8626 | if (is_new) | |
8627 | s += rdev->new_data_offset; | |
8628 | else | |
8629 | s += rdev->data_offset; | |
fc974ee2 VV |
8630 | rv = badblocks_set(&rdev->badblocks, s, sectors, 0); |
8631 | if (rv == 0) { | |
2230dfe4 | 8632 | /* Make sure they get written out promptly */ |
35b785f7 TM |
8633 | if (test_bit(ExternalBbl, &rdev->flags)) |
8634 | sysfs_notify(&rdev->kobj, NULL, | |
8635 | "unacknowledged_bad_blocks"); | |
8bd2f0a0 | 8636 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2953079c SL |
8637 | set_mask_bits(&mddev->sb_flags, 0, |
8638 | BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING)); | |
2230dfe4 | 8639 | md_wakeup_thread(rdev->mddev->thread); |
fc974ee2 VV |
8640 | return 1; |
8641 | } else | |
8642 | return 0; | |
2230dfe4 N |
8643 | } |
8644 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8645 | ||
c6563a8c N |
8646 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8647 | int is_new) | |
2230dfe4 | 8648 | { |
35b785f7 | 8649 | int rv; |
c6563a8c N |
8650 | if (is_new) |
8651 | s += rdev->new_data_offset; | |
8652 | else | |
8653 | s += rdev->data_offset; | |
35b785f7 TM |
8654 | rv = badblocks_clear(&rdev->badblocks, s, sectors); |
8655 | if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags)) | |
8656 | sysfs_notify(&rdev->kobj, NULL, "bad_blocks"); | |
8657 | return rv; | |
2230dfe4 N |
8658 | } |
8659 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8660 | ||
75c96f85 AB |
8661 | static int md_notify_reboot(struct notifier_block *this, |
8662 | unsigned long code, void *x) | |
1da177e4 LT |
8663 | { |
8664 | struct list_head *tmp; | |
fd01b88c | 8665 | struct mddev *mddev; |
2dba6a91 | 8666 | int need_delay = 0; |
1da177e4 | 8667 | |
c744a65c N |
8668 | for_each_mddev(mddev, tmp) { |
8669 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8670 | if (mddev->pers) |
8671 | __md_stop_writes(mddev); | |
0f62fb22 N |
8672 | if (mddev->persistent) |
8673 | mddev->safemode = 2; | |
c744a65c | 8674 | mddev_unlock(mddev); |
2dba6a91 | 8675 | } |
c744a65c | 8676 | need_delay = 1; |
1da177e4 | 8677 | } |
c744a65c N |
8678 | /* |
8679 | * certain more exotic SCSI devices are known to be | |
8680 | * volatile wrt too early system reboots. While the | |
8681 | * right place to handle this issue is the given | |
8682 | * driver, we do want to have a safe RAID driver ... | |
8683 | */ | |
8684 | if (need_delay) | |
8685 | mdelay(1000*1); | |
8686 | ||
1da177e4 LT |
8687 | return NOTIFY_DONE; |
8688 | } | |
8689 | ||
75c96f85 | 8690 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8691 | .notifier_call = md_notify_reboot, |
8692 | .next = NULL, | |
8693 | .priority = INT_MAX, /* before any real devices */ | |
8694 | }; | |
8695 | ||
8696 | static void md_geninit(void) | |
8697 | { | |
36a4e1fe | 8698 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8699 | |
c7705f34 | 8700 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8701 | } |
8702 | ||
75c96f85 | 8703 | static int __init md_init(void) |
1da177e4 | 8704 | { |
e804ac78 TH |
8705 | int ret = -ENOMEM; |
8706 | ||
ada609ee | 8707 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8708 | if (!md_wq) |
8709 | goto err_wq; | |
8710 | ||
8711 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8712 | if (!md_misc_wq) | |
8713 | goto err_misc_wq; | |
8714 | ||
8715 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8716 | goto err_md; | |
8717 | ||
8718 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8719 | goto err_mdp; | |
8720 | mdp_major = ret; | |
8721 | ||
af5628f0 | 8722 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
8723 | md_probe, NULL, NULL); |
8724 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8725 | md_probe, NULL, NULL); |
8726 | ||
1da177e4 | 8727 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8728 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8729 | |
8730 | md_geninit(); | |
d710e138 | 8731 | return 0; |
1da177e4 | 8732 | |
e804ac78 TH |
8733 | err_mdp: |
8734 | unregister_blkdev(MD_MAJOR, "md"); | |
8735 | err_md: | |
8736 | destroy_workqueue(md_misc_wq); | |
8737 | err_misc_wq: | |
8738 | destroy_workqueue(md_wq); | |
8739 | err_wq: | |
8740 | return ret; | |
8741 | } | |
1da177e4 | 8742 | |
70bcecdb | 8743 | static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev) |
1d7e3e96 | 8744 | { |
70bcecdb GR |
8745 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
8746 | struct md_rdev *rdev2; | |
8747 | int role, ret; | |
8748 | char b[BDEVNAME_SIZE]; | |
1d7e3e96 | 8749 | |
70bcecdb GR |
8750 | /* Check for change of roles in the active devices */ |
8751 | rdev_for_each(rdev2, mddev) { | |
8752 | if (test_bit(Faulty, &rdev2->flags)) | |
8753 | continue; | |
8754 | ||
8755 | /* Check if the roles changed */ | |
8756 | role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]); | |
dbb64f86 GR |
8757 | |
8758 | if (test_bit(Candidate, &rdev2->flags)) { | |
8759 | if (role == 0xfffe) { | |
8760 | pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b)); | |
8761 | md_kick_rdev_from_array(rdev2); | |
8762 | continue; | |
8763 | } | |
8764 | else | |
8765 | clear_bit(Candidate, &rdev2->flags); | |
8766 | } | |
8767 | ||
70bcecdb GR |
8768 | if (role != rdev2->raid_disk) { |
8769 | /* got activated */ | |
8770 | if (rdev2->raid_disk == -1 && role != 0xffff) { | |
8771 | rdev2->saved_raid_disk = role; | |
8772 | ret = remove_and_add_spares(mddev, rdev2); | |
8773 | pr_info("Activated spare: %s\n", | |
9d48739e | 8774 | bdevname(rdev2->bdev,b)); |
a578183e GJ |
8775 | /* wakeup mddev->thread here, so array could |
8776 | * perform resync with the new activated disk */ | |
8777 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8778 | md_wakeup_thread(mddev->thread); | |
8779 | ||
70bcecdb GR |
8780 | } |
8781 | /* device faulty | |
8782 | * We just want to do the minimum to mark the disk | |
8783 | * as faulty. The recovery is performed by the | |
8784 | * one who initiated the error. | |
8785 | */ | |
8786 | if ((role == 0xfffe) || (role == 0xfffd)) { | |
8787 | md_error(mddev, rdev2); | |
8788 | clear_bit(Blocked, &rdev2->flags); | |
8789 | } | |
8790 | } | |
1d7e3e96 | 8791 | } |
70bcecdb | 8792 | |
28c1b9fd GR |
8793 | if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) |
8794 | update_raid_disks(mddev, le32_to_cpu(sb->raid_disks)); | |
70bcecdb GR |
8795 | |
8796 | /* Finally set the event to be up to date */ | |
8797 | mddev->events = le64_to_cpu(sb->events); | |
8798 | } | |
8799 | ||
8800 | static int read_rdev(struct mddev *mddev, struct md_rdev *rdev) | |
8801 | { | |
8802 | int err; | |
8803 | struct page *swapout = rdev->sb_page; | |
8804 | struct mdp_superblock_1 *sb; | |
8805 | ||
8806 | /* Store the sb page of the rdev in the swapout temporary | |
8807 | * variable in case we err in the future | |
8808 | */ | |
8809 | rdev->sb_page = NULL; | |
7f0f0d87 N |
8810 | err = alloc_disk_sb(rdev); |
8811 | if (err == 0) { | |
8812 | ClearPageUptodate(rdev->sb_page); | |
8813 | rdev->sb_loaded = 0; | |
8814 | err = super_types[mddev->major_version]. | |
8815 | load_super(rdev, NULL, mddev->minor_version); | |
8816 | } | |
70bcecdb GR |
8817 | if (err < 0) { |
8818 | pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n", | |
8819 | __func__, __LINE__, rdev->desc_nr, err); | |
7f0f0d87 N |
8820 | if (rdev->sb_page) |
8821 | put_page(rdev->sb_page); | |
70bcecdb GR |
8822 | rdev->sb_page = swapout; |
8823 | rdev->sb_loaded = 1; | |
8824 | return err; | |
1d7e3e96 GR |
8825 | } |
8826 | ||
70bcecdb GR |
8827 | sb = page_address(rdev->sb_page); |
8828 | /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET | |
8829 | * is not set | |
8830 | */ | |
8831 | ||
8832 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)) | |
8833 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
8834 | ||
8835 | /* The other node finished recovery, call spare_active to set | |
8836 | * device In_sync and mddev->degraded | |
8837 | */ | |
8838 | if (rdev->recovery_offset == MaxSector && | |
8839 | !test_bit(In_sync, &rdev->flags) && | |
8840 | mddev->pers->spare_active(mddev)) | |
8841 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
8842 | ||
8843 | put_page(swapout); | |
8844 | return 0; | |
8845 | } | |
8846 | ||
8847 | void md_reload_sb(struct mddev *mddev, int nr) | |
8848 | { | |
8849 | struct md_rdev *rdev; | |
8850 | int err; | |
8851 | ||
8852 | /* Find the rdev */ | |
8853 | rdev_for_each_rcu(rdev, mddev) { | |
8854 | if (rdev->desc_nr == nr) | |
8855 | break; | |
8856 | } | |
8857 | ||
8858 | if (!rdev || rdev->desc_nr != nr) { | |
8859 | pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr); | |
8860 | return; | |
8861 | } | |
8862 | ||
8863 | err = read_rdev(mddev, rdev); | |
8864 | if (err < 0) | |
8865 | return; | |
8866 | ||
8867 | check_sb_changes(mddev, rdev); | |
8868 | ||
8869 | /* Read all rdev's to update recovery_offset */ | |
8870 | rdev_for_each_rcu(rdev, mddev) | |
8871 | read_rdev(mddev, rdev); | |
1d7e3e96 GR |
8872 | } |
8873 | EXPORT_SYMBOL(md_reload_sb); | |
8874 | ||
1da177e4 LT |
8875 | #ifndef MODULE |
8876 | ||
8877 | /* | |
8878 | * Searches all registered partitions for autorun RAID arrays | |
8879 | * at boot time. | |
8880 | */ | |
4d936ec1 | 8881 | |
5b1f5bc3 | 8882 | static DEFINE_MUTEX(detected_devices_mutex); |
4d936ec1 ME |
8883 | static LIST_HEAD(all_detected_devices); |
8884 | struct detected_devices_node { | |
8885 | struct list_head list; | |
8886 | dev_t dev; | |
8887 | }; | |
1da177e4 LT |
8888 | |
8889 | void md_autodetect_dev(dev_t dev) | |
8890 | { | |
4d936ec1 ME |
8891 | struct detected_devices_node *node_detected_dev; |
8892 | ||
8893 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8894 | if (node_detected_dev) { | |
8895 | node_detected_dev->dev = dev; | |
5b1f5bc3 | 8896 | mutex_lock(&detected_devices_mutex); |
4d936ec1 | 8897 | list_add_tail(&node_detected_dev->list, &all_detected_devices); |
5b1f5bc3 | 8898 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 8899 | } |
1da177e4 LT |
8900 | } |
8901 | ||
1da177e4 LT |
8902 | static void autostart_arrays(int part) |
8903 | { | |
3cb03002 | 8904 | struct md_rdev *rdev; |
4d936ec1 ME |
8905 | struct detected_devices_node *node_detected_dev; |
8906 | dev_t dev; | |
8907 | int i_scanned, i_passed; | |
1da177e4 | 8908 | |
4d936ec1 ME |
8909 | i_scanned = 0; |
8910 | i_passed = 0; | |
1da177e4 | 8911 | |
9d48739e | 8912 | pr_info("md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8913 | |
5b1f5bc3 | 8914 | mutex_lock(&detected_devices_mutex); |
4d936ec1 ME |
8915 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8916 | i_scanned++; | |
8917 | node_detected_dev = list_entry(all_detected_devices.next, | |
8918 | struct detected_devices_node, list); | |
8919 | list_del(&node_detected_dev->list); | |
8920 | dev = node_detected_dev->dev; | |
8921 | kfree(node_detected_dev); | |
90bcf133 | 8922 | mutex_unlock(&detected_devices_mutex); |
df968c4e | 8923 | rdev = md_import_device(dev,0, 90); |
90bcf133 | 8924 | mutex_lock(&detected_devices_mutex); |
1da177e4 LT |
8925 | if (IS_ERR(rdev)) |
8926 | continue; | |
8927 | ||
403df478 | 8928 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 8929 | continue; |
403df478 | 8930 | |
d0fae18f | 8931 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 8932 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 8933 | i_passed++; |
1da177e4 | 8934 | } |
5b1f5bc3 | 8935 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 8936 | |
9d48739e | 8937 | pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed); |
1da177e4 LT |
8938 | |
8939 | autorun_devices(part); | |
8940 | } | |
8941 | ||
fdee8ae4 | 8942 | #endif /* !MODULE */ |
1da177e4 LT |
8943 | |
8944 | static __exit void md_exit(void) | |
8945 | { | |
fd01b88c | 8946 | struct mddev *mddev; |
1da177e4 | 8947 | struct list_head *tmp; |
e2f23b60 | 8948 | int delay = 1; |
8ab5e4c1 | 8949 | |
af5628f0 | 8950 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 8951 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 8952 | |
3dbd8c2e | 8953 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
8954 | unregister_blkdev(mdp_major, "mdp"); |
8955 | unregister_reboot_notifier(&md_notifier); | |
8956 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
8957 | |
8958 | /* We cannot unload the modules while some process is | |
8959 | * waiting for us in select() or poll() - wake them up | |
8960 | */ | |
8961 | md_unloading = 1; | |
8962 | while (waitqueue_active(&md_event_waiters)) { | |
8963 | /* not safe to leave yet */ | |
8964 | wake_up(&md_event_waiters); | |
8965 | msleep(delay); | |
8966 | delay += delay; | |
8967 | } | |
1da177e4 | 8968 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 8969 | |
29ac4aa3 | 8970 | for_each_mddev(mddev, tmp) { |
1da177e4 | 8971 | export_array(mddev); |
9356863c | 8972 | mddev->ctime = 0; |
d3374825 | 8973 | mddev->hold_active = 0; |
9356863c N |
8974 | /* |
8975 | * for_each_mddev() will call mddev_put() at the end of each | |
8976 | * iteration. As the mddev is now fully clear, this will | |
8977 | * schedule the mddev for destruction by a workqueue, and the | |
8978 | * destroy_workqueue() below will wait for that to complete. | |
8979 | */ | |
1da177e4 | 8980 | } |
e804ac78 TH |
8981 | destroy_workqueue(md_misc_wq); |
8982 | destroy_workqueue(md_wq); | |
1da177e4 LT |
8983 | } |
8984 | ||
685784aa | 8985 | subsys_initcall(md_init); |
1da177e4 LT |
8986 | module_exit(md_exit) |
8987 | ||
f91de92e N |
8988 | static int get_ro(char *buffer, struct kernel_param *kp) |
8989 | { | |
8990 | return sprintf(buffer, "%d", start_readonly); | |
8991 | } | |
8992 | static int set_ro(const char *val, struct kernel_param *kp) | |
8993 | { | |
4c9309c0 | 8994 | return kstrtouint(val, 10, (unsigned int *)&start_readonly); |
f91de92e N |
8995 | } |
8996 | ||
80ca3a44 N |
8997 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
8998 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 8999 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 9000 | |
1da177e4 | 9001 | MODULE_LICENSE("GPL"); |
0efb9e61 | 9002 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 9003 | MODULE_ALIAS("md"); |
72008652 | 9004 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |