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> |
4ad23a97 N |
68 | #include <linux/percpu-refcount.h> |
69 | ||
504634f6 | 70 | #include <trace/events/block.h> |
43b2e5d8 | 71 | #include "md.h" |
ef740c37 | 72 | #include "bitmap.h" |
edb39c9d | 73 | #include "md-cluster.h" |
1da177e4 | 74 | |
1da177e4 | 75 | #ifndef MODULE |
d710e138 | 76 | static void autostart_arrays(int part); |
1da177e4 LT |
77 | #endif |
78 | ||
01f96c0a N |
79 | /* pers_list is a list of registered personalities protected |
80 | * by pers_lock. | |
81 | * pers_lock does extra service to protect accesses to | |
82 | * mddev->thread when the mutex cannot be held. | |
83 | */ | |
2604b703 | 84 | static LIST_HEAD(pers_list); |
1da177e4 LT |
85 | static DEFINE_SPINLOCK(pers_lock); |
86 | ||
edb39c9d | 87 | struct md_cluster_operations *md_cluster_ops; |
589a1c49 | 88 | EXPORT_SYMBOL(md_cluster_ops); |
edb39c9d GR |
89 | struct module *md_cluster_mod; |
90 | EXPORT_SYMBOL(md_cluster_mod); | |
91 | ||
90b08710 | 92 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
93 | static struct workqueue_struct *md_wq; |
94 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 95 | |
746d3207 N |
96 | static int remove_and_add_spares(struct mddev *mddev, |
97 | struct md_rdev *this); | |
5aa61f42 | 98 | static void mddev_detach(struct mddev *mddev); |
746d3207 | 99 | |
1e50915f RB |
100 | /* |
101 | * Default number of read corrections we'll attempt on an rdev | |
102 | * before ejecting it from the array. We divide the read error | |
103 | * count by 2 for every hour elapsed between read errors. | |
104 | */ | |
105 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
106 | /* |
107 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
108 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
109 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 110 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
111 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
112 | * speed limit - in case reconstruction slows down your system despite | |
113 | * idle IO detection. | |
114 | * | |
115 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 116 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
117 | */ |
118 | ||
119 | static int sysctl_speed_limit_min = 1000; | |
120 | static int sysctl_speed_limit_max = 200000; | |
fd01b88c | 121 | static inline int speed_min(struct mddev *mddev) |
88202a0c N |
122 | { |
123 | return mddev->sync_speed_min ? | |
124 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
125 | } | |
126 | ||
fd01b88c | 127 | static inline int speed_max(struct mddev *mddev) |
88202a0c N |
128 | { |
129 | return mddev->sync_speed_max ? | |
130 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
131 | } | |
1da177e4 LT |
132 | |
133 | static struct ctl_table_header *raid_table_header; | |
134 | ||
82592c38 | 135 | static struct ctl_table raid_table[] = { |
1da177e4 | 136 | { |
1da177e4 LT |
137 | .procname = "speed_limit_min", |
138 | .data = &sysctl_speed_limit_min, | |
139 | .maxlen = sizeof(int), | |
80ca3a44 | 140 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 141 | .proc_handler = proc_dointvec, |
1da177e4 LT |
142 | }, |
143 | { | |
1da177e4 LT |
144 | .procname = "speed_limit_max", |
145 | .data = &sysctl_speed_limit_max, | |
146 | .maxlen = sizeof(int), | |
80ca3a44 | 147 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 148 | .proc_handler = proc_dointvec, |
1da177e4 | 149 | }, |
894d2491 | 150 | { } |
1da177e4 LT |
151 | }; |
152 | ||
82592c38 | 153 | static struct ctl_table raid_dir_table[] = { |
1da177e4 | 154 | { |
1da177e4 LT |
155 | .procname = "raid", |
156 | .maxlen = 0, | |
80ca3a44 | 157 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
158 | .child = raid_table, |
159 | }, | |
894d2491 | 160 | { } |
1da177e4 LT |
161 | }; |
162 | ||
82592c38 | 163 | static struct ctl_table raid_root_table[] = { |
1da177e4 | 164 | { |
1da177e4 LT |
165 | .procname = "dev", |
166 | .maxlen = 0, | |
167 | .mode = 0555, | |
168 | .child = raid_dir_table, | |
169 | }, | |
894d2491 | 170 | { } |
1da177e4 LT |
171 | }; |
172 | ||
83d5cde4 | 173 | static const struct block_device_operations md_fops; |
1da177e4 | 174 | |
f91de92e N |
175 | static int start_readonly; |
176 | ||
78b6350d N |
177 | /* |
178 | * The original mechanism for creating an md device is to create | |
179 | * a device node in /dev and to open it. This causes races with device-close. | |
180 | * The preferred method is to write to the "new_array" module parameter. | |
181 | * This can avoid races. | |
182 | * Setting create_on_open to false disables the original mechanism | |
183 | * so all the races disappear. | |
184 | */ | |
185 | static bool create_on_open = true; | |
186 | ||
a167f663 N |
187 | /* bio_clone_mddev |
188 | * like bio_clone, but with a local bio set | |
189 | */ | |
190 | ||
a167f663 | 191 | struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, |
fd01b88c | 192 | struct mddev *mddev) |
a167f663 N |
193 | { |
194 | struct bio *b; | |
a167f663 N |
195 | |
196 | if (!mddev || !mddev->bio_set) | |
197 | return bio_alloc(gfp_mask, nr_iovecs); | |
198 | ||
395c72a7 | 199 | b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set); |
a167f663 N |
200 | if (!b) |
201 | return NULL; | |
a167f663 N |
202 | return b; |
203 | } | |
204 | EXPORT_SYMBOL_GPL(bio_alloc_mddev); | |
205 | ||
d7603b7e N |
206 | /* |
207 | * We have a system wide 'event count' that is incremented | |
208 | * on any 'interesting' event, and readers of /proc/mdstat | |
209 | * can use 'poll' or 'select' to find out when the event | |
210 | * count increases. | |
211 | * | |
212 | * Events are: | |
213 | * start array, stop array, error, add device, remove device, | |
214 | * start build, activate spare | |
215 | */ | |
2989ddbd | 216 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 217 | static atomic_t md_event_count; |
fd01b88c | 218 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
219 | { |
220 | atomic_inc(&md_event_count); | |
221 | wake_up(&md_event_waiters); | |
222 | } | |
29269553 | 223 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 224 | |
1da177e4 LT |
225 | /* |
226 | * Enables to iterate over all existing md arrays | |
227 | * all_mddevs_lock protects this list. | |
228 | */ | |
229 | static LIST_HEAD(all_mddevs); | |
230 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
231 | ||
1da177e4 LT |
232 | /* |
233 | * iterates through all used mddevs in the system. | |
234 | * We take care to grab the all_mddevs_lock whenever navigating | |
235 | * the list, and to always hold a refcount when unlocked. | |
236 | * Any code which breaks out of this loop while own | |
237 | * a reference to the current mddev and must mddev_put it. | |
238 | */ | |
fd01b88c | 239 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 | 240 | \ |
f72ffdd6 | 241 | for (({ spin_lock(&all_mddevs_lock); \ |
fd01b88c N |
242 | _tmp = all_mddevs.next; \ |
243 | _mddev = NULL;}); \ | |
244 | ({ if (_tmp != &all_mddevs) \ | |
245 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 246 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
247 | if (_mddev) mddev_put(_mddev); \ |
248 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
249 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 250 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 251 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
252 | ) |
253 | ||
409c57f3 N |
254 | /* Rather than calling directly into the personality make_request function, |
255 | * IO requests come here first so that we can check if the device is | |
256 | * being suspended pending a reconfiguration. | |
257 | * We hold a refcount over the call to ->make_request. By the time that | |
258 | * call has finished, the bio has been linked into some internal structure | |
259 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
260 | */ | |
dece1635 | 261 | static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 262 | { |
49077326 | 263 | const int rw = bio_data_dir(bio); |
fd01b88c | 264 | struct mddev *mddev = q->queuedata; |
e91ece55 | 265 | unsigned int sectors; |
74672d06 | 266 | int cpu; |
49077326 | 267 | |
54efd50b KO |
268 | blk_queue_split(q, &bio, q->bio_split); |
269 | ||
274d8cbd | 270 | if (mddev == NULL || mddev->pers == NULL) { |
409c57f3 | 271 | bio_io_error(bio); |
dece1635 | 272 | return BLK_QC_T_NONE; |
409c57f3 | 273 | } |
bbfa57c0 | 274 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
4246a0b6 CH |
275 | if (bio_sectors(bio) != 0) |
276 | bio->bi_error = -EROFS; | |
277 | bio_endio(bio); | |
dece1635 | 278 | return BLK_QC_T_NONE; |
bbfa57c0 | 279 | } |
0ca69886 | 280 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 281 | rcu_read_lock(); |
e9c7469b | 282 | if (mddev->suspended) { |
409c57f3 N |
283 | DEFINE_WAIT(__wait); |
284 | for (;;) { | |
285 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
286 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 287 | if (!mddev->suspended) |
409c57f3 N |
288 | break; |
289 | rcu_read_unlock(); | |
290 | schedule(); | |
291 | rcu_read_lock(); | |
292 | } | |
293 | finish_wait(&mddev->sb_wait, &__wait); | |
294 | } | |
295 | atomic_inc(&mddev->active_io); | |
296 | rcu_read_unlock(); | |
49077326 | 297 | |
e91ece55 CM |
298 | /* |
299 | * save the sectors now since our bio can | |
300 | * go away inside make_request | |
301 | */ | |
302 | sectors = bio_sectors(bio); | |
9c573de3 | 303 | /* bio could be mergeable after passing to underlayer */ |
1eff9d32 | 304 | bio->bi_opf &= ~REQ_NOMERGE; |
5a7bbad2 | 305 | mddev->pers->make_request(mddev, bio); |
49077326 | 306 | |
74672d06 GZ |
307 | cpu = part_stat_lock(); |
308 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
309 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); | |
310 | part_stat_unlock(); | |
49077326 | 311 | |
409c57f3 N |
312 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
313 | wake_up(&mddev->sb_wait); | |
dece1635 JA |
314 | |
315 | return BLK_QC_T_NONE; | |
409c57f3 N |
316 | } |
317 | ||
9e35b99c N |
318 | /* mddev_suspend makes sure no new requests are submitted |
319 | * to the device, and that any requests that have been submitted | |
320 | * are completely handled. | |
afa0f557 N |
321 | * Once mddev_detach() is called and completes, the module will be |
322 | * completely unused. | |
9e35b99c | 323 | */ |
fd01b88c | 324 | void mddev_suspend(struct mddev *mddev) |
409c57f3 | 325 | { |
092398dc | 326 | WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk); |
0dc10e50 MP |
327 | if (mddev->suspended++) |
328 | return; | |
409c57f3 N |
329 | synchronize_rcu(); |
330 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
331 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
332 | |
333 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 334 | } |
390ee602 | 335 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 336 | |
fd01b88c | 337 | void mddev_resume(struct mddev *mddev) |
409c57f3 | 338 | { |
0dc10e50 MP |
339 | if (--mddev->suspended) |
340 | return; | |
409c57f3 N |
341 | wake_up(&mddev->sb_wait); |
342 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 343 | |
47525e59 | 344 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
345 | md_wakeup_thread(mddev->thread); |
346 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 347 | } |
390ee602 | 348 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 349 | |
fd01b88c | 350 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 | 351 | { |
5c675f83 N |
352 | struct md_personality *pers = mddev->pers; |
353 | int ret = 0; | |
354 | ||
355 | rcu_read_lock(); | |
356 | if (mddev->suspended) | |
357 | ret = 1; | |
358 | else if (pers && pers->congested) | |
359 | ret = pers->congested(mddev, bits); | |
360 | rcu_read_unlock(); | |
361 | return ret; | |
362 | } | |
363 | EXPORT_SYMBOL_GPL(mddev_congested); | |
364 | static int md_congested(void *data, int bits) | |
365 | { | |
366 | struct mddev *mddev = data; | |
367 | return mddev_congested(mddev, bits); | |
3fa841d7 | 368 | } |
3fa841d7 | 369 | |
a2826aa9 | 370 | /* |
e9c7469b | 371 | * Generic flush handling for md |
a2826aa9 N |
372 | */ |
373 | ||
4246a0b6 | 374 | static void md_end_flush(struct bio *bio) |
a2826aa9 | 375 | { |
3cb03002 | 376 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 377 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
378 | |
379 | rdev_dec_pending(rdev, mddev); | |
380 | ||
381 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 382 | /* The pre-request flush has finished */ |
e804ac78 | 383 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
384 | } |
385 | bio_put(bio); | |
386 | } | |
387 | ||
a7a07e69 N |
388 | static void md_submit_flush_data(struct work_struct *ws); |
389 | ||
a035fc3e | 390 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 391 | { |
fd01b88c | 392 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 393 | struct md_rdev *rdev; |
a2826aa9 | 394 | |
a7a07e69 N |
395 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
396 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 397 | rcu_read_lock(); |
dafb20fa | 398 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
399 | if (rdev->raid_disk >= 0 && |
400 | !test_bit(Faulty, &rdev->flags)) { | |
401 | /* Take two references, one is dropped | |
402 | * when request finishes, one after | |
403 | * we reclaim rcu_read_lock | |
404 | */ | |
405 | struct bio *bi; | |
406 | atomic_inc(&rdev->nr_pending); | |
407 | atomic_inc(&rdev->nr_pending); | |
408 | rcu_read_unlock(); | |
b5e1b8ce | 409 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 410 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
411 | bi->bi_private = rdev; |
412 | bi->bi_bdev = rdev->bdev; | |
70fd7614 | 413 | bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
a2826aa9 | 414 | atomic_inc(&mddev->flush_pending); |
4e49ea4a | 415 | submit_bio(bi); |
a2826aa9 N |
416 | rcu_read_lock(); |
417 | rdev_dec_pending(rdev, mddev); | |
418 | } | |
419 | rcu_read_unlock(); | |
a7a07e69 N |
420 | if (atomic_dec_and_test(&mddev->flush_pending)) |
421 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
422 | } |
423 | ||
e9c7469b | 424 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 425 | { |
fd01b88c | 426 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 427 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 428 | |
4f024f37 | 429 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 | 430 | /* an empty barrier - all done */ |
4246a0b6 | 431 | bio_endio(bio); |
a2826aa9 | 432 | else { |
1eff9d32 | 433 | bio->bi_opf &= ~REQ_PREFLUSH; |
5a7bbad2 | 434 | mddev->pers->make_request(mddev, bio); |
a2826aa9 | 435 | } |
2b74e12e N |
436 | |
437 | mddev->flush_bio = NULL; | |
438 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
439 | } |
440 | ||
fd01b88c | 441 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 | 442 | { |
85572d7c | 443 | spin_lock_irq(&mddev->lock); |
a2826aa9 | 444 | wait_event_lock_irq(mddev->sb_wait, |
e9c7469b | 445 | !mddev->flush_bio, |
85572d7c | 446 | mddev->lock); |
e9c7469b | 447 | mddev->flush_bio = bio; |
85572d7c | 448 | spin_unlock_irq(&mddev->lock); |
a2826aa9 | 449 | |
a035fc3e N |
450 | INIT_WORK(&mddev->flush_work, submit_flushes); |
451 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 452 | } |
e9c7469b | 453 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 454 | |
fd01b88c | 455 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
456 | { |
457 | atomic_inc(&mddev->active); | |
458 | return mddev; | |
459 | } | |
460 | ||
5fd3a17e | 461 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 462 | |
fd01b88c | 463 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 464 | { |
a167f663 N |
465 | struct bio_set *bs = NULL; |
466 | ||
1da177e4 LT |
467 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
468 | return; | |
d3374825 | 469 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
470 | mddev->ctime == 0 && !mddev->hold_active) { |
471 | /* Array is not configured at all, and not held active, | |
472 | * so destroy it */ | |
af8a2434 | 473 | list_del_init(&mddev->all_mddevs); |
a167f663 N |
474 | bs = mddev->bio_set; |
475 | mddev->bio_set = NULL; | |
d3374825 | 476 | if (mddev->gendisk) { |
e804ac78 TH |
477 | /* We did a probe so need to clean up. Call |
478 | * queue_work inside the spinlock so that | |
479 | * flush_workqueue() after mddev_find will | |
480 | * succeed in waiting for the work to be done. | |
d3374825 N |
481 | */ |
482 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 483 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
484 | } else |
485 | kfree(mddev); | |
486 | } | |
487 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
488 | if (bs) |
489 | bioset_free(bs); | |
1da177e4 LT |
490 | } |
491 | ||
25b2edfa SL |
492 | static void md_safemode_timeout(unsigned long data); |
493 | ||
fd01b88c | 494 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
495 | { |
496 | mutex_init(&mddev->open_mutex); | |
497 | mutex_init(&mddev->reconfig_mutex); | |
498 | mutex_init(&mddev->bitmap_info.mutex); | |
499 | INIT_LIST_HEAD(&mddev->disks); | |
500 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
25b2edfa SL |
501 | setup_timer(&mddev->safemode_timer, md_safemode_timeout, |
502 | (unsigned long) mddev); | |
fafd7fb0 N |
503 | atomic_set(&mddev->active, 1); |
504 | atomic_set(&mddev->openers, 0); | |
505 | atomic_set(&mddev->active_io, 0); | |
85572d7c | 506 | spin_lock_init(&mddev->lock); |
fafd7fb0 N |
507 | atomic_set(&mddev->flush_pending, 0); |
508 | init_waitqueue_head(&mddev->sb_wait); | |
509 | init_waitqueue_head(&mddev->recovery_wait); | |
510 | mddev->reshape_position = MaxSector; | |
2c810cdd | 511 | mddev->reshape_backwards = 0; |
c4a39551 | 512 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
513 | mddev->resync_min = 0; |
514 | mddev->resync_max = MaxSector; | |
515 | mddev->level = LEVEL_NONE; | |
516 | } | |
390ee602 | 517 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 518 | |
f72ffdd6 | 519 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 520 | { |
fd01b88c | 521 | struct mddev *mddev, *new = NULL; |
1da177e4 | 522 | |
8f5f02c4 N |
523 | if (unit && MAJOR(unit) != MD_MAJOR) |
524 | unit &= ~((1<<MdpMinorShift)-1); | |
525 | ||
1da177e4 LT |
526 | retry: |
527 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
528 | |
529 | if (unit) { | |
530 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
531 | if (mddev->unit == unit) { | |
532 | mddev_get(mddev); | |
533 | spin_unlock(&all_mddevs_lock); | |
534 | kfree(new); | |
535 | return mddev; | |
536 | } | |
537 | ||
538 | if (new) { | |
539 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 540 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
541 | new->hold_active = UNTIL_IOCTL; |
542 | return new; | |
1da177e4 | 543 | } |
efeb53c0 N |
544 | } else if (new) { |
545 | /* find an unused unit number */ | |
546 | static int next_minor = 512; | |
547 | int start = next_minor; | |
548 | int is_free = 0; | |
549 | int dev = 0; | |
550 | while (!is_free) { | |
551 | dev = MKDEV(MD_MAJOR, next_minor); | |
552 | next_minor++; | |
553 | if (next_minor > MINORMASK) | |
554 | next_minor = 0; | |
555 | if (next_minor == start) { | |
556 | /* Oh dear, all in use. */ | |
557 | spin_unlock(&all_mddevs_lock); | |
558 | kfree(new); | |
559 | return NULL; | |
560 | } | |
f72ffdd6 | 561 | |
efeb53c0 N |
562 | is_free = 1; |
563 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
564 | if (mddev->unit == dev) { | |
565 | is_free = 0; | |
566 | break; | |
567 | } | |
568 | } | |
569 | new->unit = dev; | |
570 | new->md_minor = MINOR(dev); | |
571 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
572 | list_add(&new->all_mddevs, &all_mddevs); |
573 | spin_unlock(&all_mddevs_lock); | |
574 | return new; | |
575 | } | |
576 | spin_unlock(&all_mddevs_lock); | |
577 | ||
9ffae0cf | 578 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
579 | if (!new) |
580 | return NULL; | |
581 | ||
1da177e4 LT |
582 | new->unit = unit; |
583 | if (MAJOR(unit) == MD_MAJOR) | |
584 | new->md_minor = MINOR(unit); | |
585 | else | |
586 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
587 | ||
fafd7fb0 | 588 | mddev_init(new); |
1da177e4 | 589 | |
1da177e4 LT |
590 | goto retry; |
591 | } | |
592 | ||
b6eb127d N |
593 | static struct attribute_group md_redundancy_group; |
594 | ||
5c47daf6 | 595 | void mddev_unlock(struct mddev *mddev) |
1da177e4 | 596 | { |
a64c876f | 597 | if (mddev->to_remove) { |
b6eb127d N |
598 | /* These cannot be removed under reconfig_mutex as |
599 | * an access to the files will try to take reconfig_mutex | |
600 | * while holding the file unremovable, which leads to | |
601 | * a deadlock. | |
bb4f1e9d N |
602 | * So hold set sysfs_active while the remove in happeing, |
603 | * and anything else which might set ->to_remove or my | |
604 | * otherwise change the sysfs namespace will fail with | |
605 | * -EBUSY if sysfs_active is still set. | |
606 | * We set sysfs_active under reconfig_mutex and elsewhere | |
607 | * test it under the same mutex to ensure its correct value | |
608 | * is seen. | |
b6eb127d | 609 | */ |
a64c876f N |
610 | struct attribute_group *to_remove = mddev->to_remove; |
611 | mddev->to_remove = NULL; | |
bb4f1e9d | 612 | mddev->sysfs_active = 1; |
b6eb127d N |
613 | mutex_unlock(&mddev->reconfig_mutex); |
614 | ||
00bcb4ac N |
615 | if (mddev->kobj.sd) { |
616 | if (to_remove != &md_redundancy_group) | |
617 | sysfs_remove_group(&mddev->kobj, to_remove); | |
618 | if (mddev->pers == NULL || | |
619 | mddev->pers->sync_request == NULL) { | |
620 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
621 | if (mddev->sysfs_action) | |
622 | sysfs_put(mddev->sysfs_action); | |
623 | mddev->sysfs_action = NULL; | |
624 | } | |
a64c876f | 625 | } |
bb4f1e9d | 626 | mddev->sysfs_active = 0; |
b6eb127d N |
627 | } else |
628 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 629 | |
751e67ca CD |
630 | /* As we've dropped the mutex we need a spinlock to |
631 | * make sure the thread doesn't disappear | |
01f96c0a N |
632 | */ |
633 | spin_lock(&pers_lock); | |
005eca5e | 634 | md_wakeup_thread(mddev->thread); |
01f96c0a | 635 | spin_unlock(&pers_lock); |
1da177e4 | 636 | } |
5c47daf6 | 637 | EXPORT_SYMBOL_GPL(mddev_unlock); |
1da177e4 | 638 | |
57d051dc | 639 | struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr) |
1ca69c4b N |
640 | { |
641 | struct md_rdev *rdev; | |
642 | ||
643 | rdev_for_each_rcu(rdev, mddev) | |
644 | if (rdev->desc_nr == nr) | |
645 | return rdev; | |
646 | ||
647 | return NULL; | |
648 | } | |
57d051dc | 649 | EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu); |
1ca69c4b N |
650 | |
651 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 652 | { |
3cb03002 | 653 | struct md_rdev *rdev; |
1da177e4 | 654 | |
dafb20fa | 655 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
656 | if (rdev->bdev->bd_dev == dev) |
657 | return rdev; | |
159ec1fc | 658 | |
1da177e4 LT |
659 | return NULL; |
660 | } | |
661 | ||
1ca69c4b N |
662 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
663 | { | |
664 | struct md_rdev *rdev; | |
665 | ||
666 | rdev_for_each_rcu(rdev, mddev) | |
667 | if (rdev->bdev->bd_dev == dev) | |
668 | return rdev; | |
669 | ||
670 | return NULL; | |
671 | } | |
672 | ||
84fc4b56 | 673 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 674 | { |
84fc4b56 | 675 | struct md_personality *pers; |
d9d166c2 N |
676 | list_for_each_entry(pers, &pers_list, list) { |
677 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 678 | return pers; |
d9d166c2 N |
679 | if (strcmp(pers->name, clevel)==0) |
680 | return pers; | |
681 | } | |
2604b703 N |
682 | return NULL; |
683 | } | |
684 | ||
b73df2d3 | 685 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 686 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 687 | { |
57b2caa3 | 688 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 689 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
690 | } |
691 | ||
f72ffdd6 | 692 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 693 | { |
1da177e4 | 694 | rdev->sb_page = alloc_page(GFP_KERNEL); |
7f0f0d87 | 695 | if (!rdev->sb_page) |
ebc24337 | 696 | return -ENOMEM; |
1da177e4 LT |
697 | return 0; |
698 | } | |
699 | ||
545c8795 | 700 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
701 | { |
702 | if (rdev->sb_page) { | |
2d1f3b5d | 703 | put_page(rdev->sb_page); |
1da177e4 LT |
704 | rdev->sb_loaded = 0; |
705 | rdev->sb_page = NULL; | |
0f420358 | 706 | rdev->sb_start = 0; |
dd8ac336 | 707 | rdev->sectors = 0; |
1da177e4 | 708 | } |
2699b672 N |
709 | if (rdev->bb_page) { |
710 | put_page(rdev->bb_page); | |
711 | rdev->bb_page = NULL; | |
712 | } | |
d3b407fb | 713 | badblocks_exit(&rdev->badblocks); |
1da177e4 | 714 | } |
545c8795 | 715 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 716 | |
4246a0b6 | 717 | static void super_written(struct bio *bio) |
7bfa19f2 | 718 | { |
3cb03002 | 719 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 720 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 721 | |
4246a0b6 | 722 | if (bio->bi_error) { |
9d48739e | 723 | pr_err("md: super_written gets error=%d\n", bio->bi_error); |
a9701a30 | 724 | md_error(mddev, rdev); |
46533ff7 N |
725 | if (!test_bit(Faulty, &rdev->flags) |
726 | && (bio->bi_opf & MD_FAILFAST)) { | |
2953079c | 727 | set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags); |
46533ff7 N |
728 | set_bit(LastDev, &rdev->flags); |
729 | } | |
730 | } else | |
731 | clear_bit(LastDev, &rdev->flags); | |
7bfa19f2 | 732 | |
a9701a30 N |
733 | if (atomic_dec_and_test(&mddev->pending_writes)) |
734 | wake_up(&mddev->sb_wait); | |
ed3b98c7 | 735 | rdev_dec_pending(rdev, mddev); |
f8b58edf | 736 | bio_put(bio); |
7bfa19f2 N |
737 | } |
738 | ||
fd01b88c | 739 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
740 | sector_t sector, int size, struct page *page) |
741 | { | |
742 | /* write first size bytes of page to sector of rdev | |
743 | * Increment mddev->pending_writes before returning | |
744 | * and decrement it on completion, waking up sb_wait | |
745 | * if zero is reached. | |
746 | * If an error occurred, call md_error | |
747 | */ | |
46533ff7 N |
748 | struct bio *bio; |
749 | int ff = 0; | |
750 | ||
751 | if (test_bit(Faulty, &rdev->flags)) | |
752 | return; | |
753 | ||
754 | bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); | |
7bfa19f2 | 755 | |
ed3b98c7 SL |
756 | atomic_inc(&rdev->nr_pending); |
757 | ||
a6ff7e08 | 758 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
4f024f37 | 759 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
760 | bio_add_page(bio, page, size, 0); |
761 | bio->bi_private = rdev; | |
762 | bio->bi_end_io = super_written; | |
46533ff7 N |
763 | |
764 | if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) && | |
765 | test_bit(FailFast, &rdev->flags) && | |
766 | !test_bit(LastDev, &rdev->flags)) | |
767 | ff = MD_FAILFAST; | |
20737738 | 768 | bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_FUA | ff; |
a9701a30 | 769 | |
7bfa19f2 | 770 | atomic_inc(&mddev->pending_writes); |
4e49ea4a | 771 | submit_bio(bio); |
a9701a30 N |
772 | } |
773 | ||
46533ff7 | 774 | int md_super_wait(struct mddev *mddev) |
a9701a30 | 775 | { |
e9c7469b | 776 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 777 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
2953079c | 778 | if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags)) |
46533ff7 N |
779 | return -EAGAIN; |
780 | return 0; | |
7bfa19f2 N |
781 | } |
782 | ||
3cb03002 | 783 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
796a5cf0 | 784 | struct page *page, int op, int op_flags, bool metadata_op) |
1da177e4 | 785 | { |
a167f663 | 786 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
787 | int ret; |
788 | ||
a6ff7e08 JB |
789 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
790 | rdev->meta_bdev : rdev->bdev; | |
796a5cf0 | 791 | bio_set_op_attrs(bio, op, op_flags); |
ccebd4c4 | 792 | if (metadata_op) |
4f024f37 | 793 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
794 | else if (rdev->mddev->reshape_position != MaxSector && |
795 | (rdev->mddev->reshape_backwards == | |
796 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 797 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 798 | else |
4f024f37 | 799 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 800 | bio_add_page(bio, page, size, 0); |
4e49ea4a MC |
801 | |
802 | submit_bio_wait(bio); | |
1da177e4 | 803 | |
4246a0b6 | 804 | ret = !bio->bi_error; |
1da177e4 LT |
805 | bio_put(bio); |
806 | return ret; | |
807 | } | |
a8745db2 | 808 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 809 | |
f72ffdd6 | 810 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
811 | { |
812 | char b[BDEVNAME_SIZE]; | |
403df478 | 813 | |
1da177e4 LT |
814 | if (rdev->sb_loaded) |
815 | return 0; | |
816 | ||
796a5cf0 | 817 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) |
1da177e4 LT |
818 | goto fail; |
819 | rdev->sb_loaded = 1; | |
820 | return 0; | |
821 | ||
822 | fail: | |
9d48739e N |
823 | pr_err("md: disabled device %s, could not read superblock.\n", |
824 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
825 | return -EINVAL; |
826 | } | |
827 | ||
828 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
829 | { | |
f72ffdd6 | 830 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
831 | sb1->set_uuid1 == sb2->set_uuid1 && |
832 | sb1->set_uuid2 == sb2->set_uuid2 && | |
833 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
834 | } |
835 | ||
1da177e4 LT |
836 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
837 | { | |
838 | int ret; | |
839 | mdp_super_t *tmp1, *tmp2; | |
840 | ||
841 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
842 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
843 | ||
844 | if (!tmp1 || !tmp2) { | |
845 | ret = 0; | |
1da177e4 LT |
846 | goto abort; |
847 | } | |
848 | ||
849 | *tmp1 = *sb1; | |
850 | *tmp2 = *sb2; | |
851 | ||
852 | /* | |
853 | * nr_disks is not constant | |
854 | */ | |
855 | tmp1->nr_disks = 0; | |
856 | tmp2->nr_disks = 0; | |
857 | ||
ce0c8e05 | 858 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 859 | abort: |
990a8baf JJ |
860 | kfree(tmp1); |
861 | kfree(tmp2); | |
1da177e4 LT |
862 | return ret; |
863 | } | |
864 | ||
4d167f09 N |
865 | static u32 md_csum_fold(u32 csum) |
866 | { | |
867 | csum = (csum & 0xffff) + (csum >> 16); | |
868 | return (csum & 0xffff) + (csum >> 16); | |
869 | } | |
870 | ||
f72ffdd6 | 871 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 872 | { |
4d167f09 N |
873 | u64 newcsum = 0; |
874 | u32 *sb32 = (u32*)sb; | |
875 | int i; | |
1da177e4 LT |
876 | unsigned int disk_csum, csum; |
877 | ||
878 | disk_csum = sb->sb_csum; | |
879 | sb->sb_csum = 0; | |
4d167f09 N |
880 | |
881 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
882 | newcsum += sb32[i]; | |
883 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
884 | ||
4d167f09 N |
885 | #ifdef CONFIG_ALPHA |
886 | /* This used to use csum_partial, which was wrong for several | |
887 | * reasons including that different results are returned on | |
888 | * different architectures. It isn't critical that we get exactly | |
889 | * the same return value as before (we always csum_fold before | |
890 | * testing, and that removes any differences). However as we | |
891 | * know that csum_partial always returned a 16bit value on | |
892 | * alphas, do a fold to maximise conformity to previous behaviour. | |
893 | */ | |
894 | sb->sb_csum = md_csum_fold(disk_csum); | |
895 | #else | |
1da177e4 | 896 | sb->sb_csum = disk_csum; |
4d167f09 | 897 | #endif |
1da177e4 LT |
898 | return csum; |
899 | } | |
900 | ||
1da177e4 LT |
901 | /* |
902 | * Handle superblock details. | |
903 | * We want to be able to handle multiple superblock formats | |
904 | * so we have a common interface to them all, and an array of | |
905 | * different handlers. | |
906 | * We rely on user-space to write the initial superblock, and support | |
907 | * reading and updating of superblocks. | |
908 | * Interface methods are: | |
3cb03002 | 909 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
910 | * loads and validates a superblock on dev. |
911 | * if refdev != NULL, compare superblocks on both devices | |
912 | * Return: | |
913 | * 0 - dev has a superblock that is compatible with refdev | |
914 | * 1 - dev has a superblock that is compatible and newer than refdev | |
915 | * so dev should be used as the refdev in future | |
916 | * -EINVAL superblock incompatible or invalid | |
917 | * -othererror e.g. -EIO | |
918 | * | |
fd01b88c | 919 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
920 | * Verify that dev is acceptable into mddev. |
921 | * The first time, mddev->raid_disks will be 0, and data from | |
922 | * dev should be merged in. Subsequent calls check that dev | |
923 | * is new enough. Return 0 or -EINVAL | |
924 | * | |
fd01b88c | 925 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
926 | * Update the superblock for rdev with data in mddev |
927 | * This does not write to disc. | |
928 | * | |
929 | */ | |
930 | ||
931 | struct super_type { | |
0cd17fec CW |
932 | char *name; |
933 | struct module *owner; | |
c6563a8c N |
934 | int (*load_super)(struct md_rdev *rdev, |
935 | struct md_rdev *refdev, | |
0cd17fec | 936 | int minor_version); |
c6563a8c N |
937 | int (*validate_super)(struct mddev *mddev, |
938 | struct md_rdev *rdev); | |
939 | void (*sync_super)(struct mddev *mddev, | |
940 | struct md_rdev *rdev); | |
3cb03002 | 941 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 942 | sector_t num_sectors); |
c6563a8c N |
943 | int (*allow_new_offset)(struct md_rdev *rdev, |
944 | unsigned long long new_offset); | |
1da177e4 LT |
945 | }; |
946 | ||
0894cc30 AN |
947 | /* |
948 | * Check that the given mddev has no bitmap. | |
949 | * | |
950 | * This function is called from the run method of all personalities that do not | |
951 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
952 | * has a bitmap. Otherwise, it returns 0. | |
953 | * | |
954 | */ | |
fd01b88c | 955 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 956 | { |
c3d9714e | 957 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 | 958 | return 0; |
9d48739e | 959 | pr_warn("%s: bitmaps are not supported for %s\n", |
0894cc30 AN |
960 | mdname(mddev), mddev->pers->name); |
961 | return 1; | |
962 | } | |
963 | EXPORT_SYMBOL(md_check_no_bitmap); | |
964 | ||
1da177e4 | 965 | /* |
f72ffdd6 | 966 | * load_super for 0.90.0 |
1da177e4 | 967 | */ |
3cb03002 | 968 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
969 | { |
970 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
971 | mdp_super_t *sb; | |
972 | int ret; | |
1da177e4 LT |
973 | |
974 | /* | |
0f420358 | 975 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
976 | * it's at the end of the disk. |
977 | * | |
978 | * It also happens to be a multiple of 4Kb. | |
979 | */ | |
57b2caa3 | 980 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 981 | |
0002b271 | 982 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
9d48739e N |
983 | if (ret) |
984 | return ret; | |
1da177e4 LT |
985 | |
986 | ret = -EINVAL; | |
987 | ||
988 | bdevname(rdev->bdev, b); | |
65a06f06 | 989 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
990 | |
991 | if (sb->md_magic != MD_SB_MAGIC) { | |
9d48739e | 992 | pr_warn("md: invalid raid superblock magic on %s\n", b); |
1da177e4 LT |
993 | goto abort; |
994 | } | |
995 | ||
996 | if (sb->major_version != 0 || | |
f6705578 N |
997 | sb->minor_version < 90 || |
998 | sb->minor_version > 91) { | |
9d48739e N |
999 | pr_warn("Bad version number %d.%d on %s\n", |
1000 | sb->major_version, sb->minor_version, b); | |
1da177e4 LT |
1001 | goto abort; |
1002 | } | |
1003 | ||
1004 | if (sb->raid_disks <= 0) | |
1005 | goto abort; | |
1006 | ||
4d167f09 | 1007 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
9d48739e | 1008 | pr_warn("md: invalid superblock checksum on %s\n", b); |
1da177e4 LT |
1009 | goto abort; |
1010 | } | |
1011 | ||
1012 | rdev->preferred_minor = sb->md_minor; | |
1013 | rdev->data_offset = 0; | |
c6563a8c | 1014 | rdev->new_data_offset = 0; |
0002b271 | 1015 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1016 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1017 | |
1018 | if (sb->level == LEVEL_MULTIPATH) | |
1019 | rdev->desc_nr = -1; | |
1020 | else | |
1021 | rdev->desc_nr = sb->this_disk.number; | |
1022 | ||
9a7b2b0f | 1023 | if (!refdev) { |
1da177e4 | 1024 | ret = 1; |
9a7b2b0f | 1025 | } else { |
1da177e4 | 1026 | __u64 ev1, ev2; |
65a06f06 | 1027 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 | 1028 | if (!uuid_equal(refsb, sb)) { |
9d48739e | 1029 | pr_warn("md: %s has different UUID to %s\n", |
1da177e4 LT |
1030 | b, bdevname(refdev->bdev,b2)); |
1031 | goto abort; | |
1032 | } | |
1033 | if (!sb_equal(refsb, sb)) { | |
9d48739e N |
1034 | pr_warn("md: %s has same UUID but different superblock to %s\n", |
1035 | b, bdevname(refdev->bdev, b2)); | |
1da177e4 LT |
1036 | goto abort; |
1037 | } | |
1038 | ev1 = md_event(sb); | |
1039 | ev2 = md_event(refsb); | |
1040 | if (ev1 > ev2) | |
1041 | ret = 1; | |
f72ffdd6 | 1042 | else |
1da177e4 LT |
1043 | ret = 0; |
1044 | } | |
8190e754 | 1045 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1046 | /* Limit to 4TB as metadata cannot record more than that. |
1047 | * (not needed for Linear and RAID0 as metadata doesn't | |
1048 | * record this size) | |
1049 | */ | |
3312c951 AB |
1050 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) && |
1051 | sb->level >= 1) | |
1052 | rdev->sectors = (sector_t)(2ULL << 32) - 2; | |
1da177e4 | 1053 | |
27a7b260 | 1054 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1055 | /* "this cannot possibly happen" ... */ |
1056 | ret = -EINVAL; | |
1057 | ||
1da177e4 LT |
1058 | abort: |
1059 | return ret; | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * validate_super for 0.90.0 | |
1064 | */ | |
fd01b88c | 1065 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1066 | { |
1067 | mdp_disk_t *desc; | |
65a06f06 | 1068 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1069 | __u64 ev1 = md_event(sb); |
1da177e4 | 1070 | |
41158c7e | 1071 | rdev->raid_disk = -1; |
c5d79adb N |
1072 | clear_bit(Faulty, &rdev->flags); |
1073 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1074 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1075 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1076 | |
1da177e4 LT |
1077 | if (mddev->raid_disks == 0) { |
1078 | mddev->major_version = 0; | |
1079 | mddev->minor_version = sb->minor_version; | |
1080 | mddev->patch_version = sb->patch_version; | |
e691063a | 1081 | mddev->external = 0; |
9d8f0363 | 1082 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1083 | mddev->ctime = sb->ctime; |
1084 | mddev->utime = sb->utime; | |
1085 | mddev->level = sb->level; | |
d9d166c2 | 1086 | mddev->clevel[0] = 0; |
1da177e4 LT |
1087 | mddev->layout = sb->layout; |
1088 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1089 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1090 | mddev->events = ev1; |
c3d9714e | 1091 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1092 | mddev->bitmap_info.space = 0; |
1093 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1094 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1095 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1096 | mddev->reshape_backwards = 0; |
1da177e4 | 1097 | |
f6705578 N |
1098 | if (mddev->minor_version >= 91) { |
1099 | mddev->reshape_position = sb->reshape_position; | |
1100 | mddev->delta_disks = sb->delta_disks; | |
1101 | mddev->new_level = sb->new_level; | |
1102 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1103 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1104 | if (mddev->delta_disks < 0) |
1105 | mddev->reshape_backwards = 1; | |
f6705578 N |
1106 | } else { |
1107 | mddev->reshape_position = MaxSector; | |
1108 | mddev->delta_disks = 0; | |
1109 | mddev->new_level = mddev->level; | |
1110 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1111 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1112 | } |
1113 | ||
1da177e4 LT |
1114 | if (sb->state & (1<<MD_SB_CLEAN)) |
1115 | mddev->recovery_cp = MaxSector; | |
1116 | else { | |
f72ffdd6 | 1117 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1118 | sb->events_lo == sb->cp_events_lo) { |
1119 | mddev->recovery_cp = sb->recovery_cp; | |
1120 | } else | |
1121 | mddev->recovery_cp = 0; | |
1122 | } | |
1123 | ||
1124 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1125 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1126 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1127 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1128 | ||
1129 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1130 | |
1131 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1132 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1133 | mddev->bitmap_info.offset = |
1134 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1135 | mddev->bitmap_info.space = |
c9ad020f | 1136 | mddev->bitmap_info.default_space; |
6409bb05 | 1137 | } |
a654b9d8 | 1138 | |
41158c7e | 1139 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1140 | /* Insist on good event counter while assembling, except |
1141 | * for spares (which don't need an event count) */ | |
1da177e4 | 1142 | ++ev1; |
be6800a7 N |
1143 | if (sb->disks[rdev->desc_nr].state & ( |
1144 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1145 | if (ev1 < mddev->events) |
be6800a7 | 1146 | return -EINVAL; |
41158c7e N |
1147 | } else if (mddev->bitmap) { |
1148 | /* if adding to array with a bitmap, then we can accept an | |
1149 | * older device ... but not too old. | |
1150 | */ | |
41158c7e N |
1151 | if (ev1 < mddev->bitmap->events_cleared) |
1152 | return 0; | |
8313b8e5 N |
1153 | if (ev1 < mddev->events) |
1154 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1155 | } else { |
1156 | if (ev1 < mddev->events) | |
1157 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1158 | return 0; | |
1159 | } | |
41158c7e | 1160 | |
1da177e4 | 1161 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1162 | desc = sb->disks + rdev->desc_nr; |
1163 | ||
1164 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1165 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1166 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1167 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1168 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1169 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1170 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1171 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1172 | /* active but not in sync implies recovery up to | |
1173 | * reshape position. We don't know exactly where | |
1174 | * that is, so set to zero for now */ | |
1175 | if (mddev->minor_version >= 91) { | |
1176 | rdev->recovery_offset = 0; | |
1177 | rdev->raid_disk = desc->raid_disk; | |
1178 | } | |
1da177e4 | 1179 | } |
8ddf9efe N |
1180 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1181 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1182 | if (desc->state & (1<<MD_DISK_FAILFAST)) |
1183 | set_bit(FailFast, &rdev->flags); | |
41158c7e | 1184 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1185 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1186 | return 0; |
1187 | } | |
1188 | ||
1189 | /* | |
1190 | * sync_super for 0.90.0 | |
1191 | */ | |
fd01b88c | 1192 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1193 | { |
1194 | mdp_super_t *sb; | |
3cb03002 | 1195 | struct md_rdev *rdev2; |
1da177e4 | 1196 | int next_spare = mddev->raid_disks; |
19133a42 | 1197 | |
1da177e4 LT |
1198 | /* make rdev->sb match mddev data.. |
1199 | * | |
1200 | * 1/ zero out disks | |
1201 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1202 | * 3/ any empty disks < next_spare become removed | |
1203 | * | |
1204 | * disks[0] gets initialised to REMOVED because | |
1205 | * we cannot be sure from other fields if it has | |
1206 | * been initialised or not. | |
1207 | */ | |
1208 | int i; | |
1209 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1210 | ||
61181565 N |
1211 | rdev->sb_size = MD_SB_BYTES; |
1212 | ||
65a06f06 | 1213 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1214 | |
1215 | memset(sb, 0, sizeof(*sb)); | |
1216 | ||
1217 | sb->md_magic = MD_SB_MAGIC; | |
1218 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1219 | sb->patch_version = mddev->patch_version; |
1220 | sb->gvalid_words = 0; /* ignored */ | |
1221 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1222 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1223 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1224 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1225 | ||
9ebc6ef1 | 1226 | sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 1227 | sb->level = mddev->level; |
58c0fed4 | 1228 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1229 | sb->raid_disks = mddev->raid_disks; |
1230 | sb->md_minor = mddev->md_minor; | |
e691063a | 1231 | sb->not_persistent = 0; |
9ebc6ef1 | 1232 | sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
1233 | sb->state = 0; |
1234 | sb->events_hi = (mddev->events>>32); | |
1235 | sb->events_lo = (u32)mddev->events; | |
1236 | ||
f6705578 N |
1237 | if (mddev->reshape_position == MaxSector) |
1238 | sb->minor_version = 90; | |
1239 | else { | |
1240 | sb->minor_version = 91; | |
1241 | sb->reshape_position = mddev->reshape_position; | |
1242 | sb->new_level = mddev->new_level; | |
1243 | sb->delta_disks = mddev->delta_disks; | |
1244 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1245 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1246 | } |
1247 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1248 | if (mddev->in_sync) |
1249 | { | |
1250 | sb->recovery_cp = mddev->recovery_cp; | |
1251 | sb->cp_events_hi = (mddev->events>>32); | |
1252 | sb->cp_events_lo = (u32)mddev->events; | |
1253 | if (mddev->recovery_cp == MaxSector) | |
1254 | sb->state = (1<< MD_SB_CLEAN); | |
1255 | } else | |
1256 | sb->recovery_cp = 0; | |
1257 | ||
1258 | sb->layout = mddev->layout; | |
9d8f0363 | 1259 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1260 | |
c3d9714e | 1261 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1262 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1263 | ||
1da177e4 | 1264 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1265 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1266 | mdp_disk_t *d; |
86e6ffdd | 1267 | int desc_nr; |
0261cd9f N |
1268 | int is_active = test_bit(In_sync, &rdev2->flags); |
1269 | ||
1270 | if (rdev2->raid_disk >= 0 && | |
1271 | sb->minor_version >= 91) | |
1272 | /* we have nowhere to store the recovery_offset, | |
1273 | * but if it is not below the reshape_position, | |
1274 | * we can piggy-back on that. | |
1275 | */ | |
1276 | is_active = 1; | |
1277 | if (rdev2->raid_disk < 0 || | |
1278 | test_bit(Faulty, &rdev2->flags)) | |
1279 | is_active = 0; | |
1280 | if (is_active) | |
86e6ffdd | 1281 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1282 | else |
86e6ffdd | 1283 | desc_nr = next_spare++; |
19133a42 | 1284 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1285 | d = &sb->disks[rdev2->desc_nr]; |
1286 | nr_disks++; | |
1287 | d->number = rdev2->desc_nr; | |
1288 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1289 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1290 | if (is_active) |
1da177e4 LT |
1291 | d->raid_disk = rdev2->raid_disk; |
1292 | else | |
1293 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1294 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1295 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1296 | else if (is_active) { |
1da177e4 | 1297 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1298 | if (test_bit(In_sync, &rdev2->flags)) |
1299 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1300 | active++; |
1301 | working++; | |
1302 | } else { | |
1303 | d->state = 0; | |
1304 | spare++; | |
1305 | working++; | |
1306 | } | |
8ddf9efe N |
1307 | if (test_bit(WriteMostly, &rdev2->flags)) |
1308 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
1309 | if (test_bit(FailFast, &rdev2->flags)) |
1310 | d->state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 | 1311 | } |
1da177e4 LT |
1312 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1313 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1314 | mdp_disk_t *d = &sb->disks[i]; | |
1315 | if (d->state == 0 && d->number == 0) { | |
1316 | d->number = i; | |
1317 | d->raid_disk = i; | |
1318 | d->state = (1<<MD_DISK_REMOVED); | |
1319 | d->state |= (1<<MD_DISK_FAULTY); | |
1320 | failed++; | |
1321 | } | |
1322 | } | |
1323 | sb->nr_disks = nr_disks; | |
1324 | sb->active_disks = active; | |
1325 | sb->working_disks = working; | |
1326 | sb->failed_disks = failed; | |
1327 | sb->spare_disks = spare; | |
1328 | ||
1329 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1330 | sb->sb_csum = calc_sb_csum(sb); | |
1331 | } | |
1332 | ||
0cd17fec CW |
1333 | /* |
1334 | * rdev_size_change for 0.90.0 | |
1335 | */ | |
1336 | static unsigned long long | |
3cb03002 | 1337 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1338 | { |
58c0fed4 | 1339 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1340 | return 0; /* component must fit device */ |
c3d9714e | 1341 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1342 | return 0; /* can't move bitmap */ |
57b2caa3 | 1343 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1344 | if (!num_sectors || num_sectors > rdev->sb_start) |
1345 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1346 | /* Limit to 4TB as metadata cannot record more than that. |
1347 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1348 | */ | |
3312c951 AB |
1349 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) && |
1350 | rdev->mddev->level >= 1) | |
1351 | num_sectors = (sector_t)(2ULL << 32) - 2; | |
46533ff7 N |
1352 | do { |
1353 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
0cd17fec | 1354 | rdev->sb_page); |
46533ff7 | 1355 | } while (md_super_wait(rdev->mddev) < 0); |
c26a44ed | 1356 | return num_sectors; |
0cd17fec CW |
1357 | } |
1358 | ||
c6563a8c N |
1359 | static int |
1360 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1361 | { | |
1362 | /* non-zero offset changes not possible with v0.90 */ | |
1363 | return new_offset == 0; | |
1364 | } | |
0cd17fec | 1365 | |
1da177e4 LT |
1366 | /* |
1367 | * version 1 superblock | |
1368 | */ | |
1369 | ||
f72ffdd6 | 1370 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1371 | { |
1c05b4bc N |
1372 | __le32 disk_csum; |
1373 | u32 csum; | |
1da177e4 LT |
1374 | unsigned long long newcsum; |
1375 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1376 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1377 | |
1378 | disk_csum = sb->sb_csum; | |
1379 | sb->sb_csum = 0; | |
1380 | newcsum = 0; | |
1f3c9907 | 1381 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1382 | newcsum += le32_to_cpu(*isuper++); |
1383 | ||
1384 | if (size == 2) | |
1c05b4bc | 1385 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1386 | |
1387 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1388 | sb->sb_csum = disk_csum; | |
1389 | return cpu_to_le32(csum); | |
1390 | } | |
1391 | ||
3cb03002 | 1392 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1393 | { |
1394 | struct mdp_superblock_1 *sb; | |
1395 | int ret; | |
0f420358 | 1396 | sector_t sb_start; |
c6563a8c | 1397 | sector_t sectors; |
1da177e4 | 1398 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1399 | int bmask; |
1da177e4 LT |
1400 | |
1401 | /* | |
0f420358 | 1402 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1403 | * It is always aligned to a 4K boundary and |
1404 | * depeding on minor_version, it can be: | |
1405 | * 0: At least 8K, but less than 12K, from end of device | |
1406 | * 1: At start of device | |
1407 | * 2: 4K from start of device. | |
1408 | */ | |
1409 | switch(minor_version) { | |
1410 | case 0: | |
77304d2a | 1411 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1412 | sb_start -= 8*2; |
1413 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1414 | break; |
1415 | case 1: | |
0f420358 | 1416 | sb_start = 0; |
1da177e4 LT |
1417 | break; |
1418 | case 2: | |
0f420358 | 1419 | sb_start = 8; |
1da177e4 LT |
1420 | break; |
1421 | default: | |
1422 | return -EINVAL; | |
1423 | } | |
0f420358 | 1424 | rdev->sb_start = sb_start; |
1da177e4 | 1425 | |
0002b271 N |
1426 | /* superblock is rarely larger than 1K, but it can be larger, |
1427 | * and it is safe to read 4k, so we do that | |
1428 | */ | |
1429 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1430 | if (ret) return ret; |
1431 | ||
65a06f06 | 1432 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1433 | |
1434 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1435 | sb->major_version != cpu_to_le32(1) || | |
1436 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1437 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1438 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1439 | return -EINVAL; |
1440 | ||
1441 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
9d48739e | 1442 | pr_warn("md: invalid superblock checksum on %s\n", |
1da177e4 LT |
1443 | bdevname(rdev->bdev,b)); |
1444 | return -EINVAL; | |
1445 | } | |
1446 | if (le64_to_cpu(sb->data_size) < 10) { | |
9d48739e N |
1447 | pr_warn("md: data_size too small on %s\n", |
1448 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1449 | return -EINVAL; |
1450 | } | |
c6563a8c N |
1451 | if (sb->pad0 || |
1452 | sb->pad3[0] || | |
1453 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1454 | /* Some padding is non-zero, might be a new feature */ | |
1455 | return -EINVAL; | |
e11e93fa | 1456 | |
1da177e4 LT |
1457 | rdev->preferred_minor = 0xffff; |
1458 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1459 | rdev->new_data_offset = rdev->data_offset; |
1460 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1461 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1462 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1463 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1464 | |
0002b271 | 1465 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1466 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1467 | if (rdev->sb_size & bmask) |
a1801f85 N |
1468 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1469 | ||
1470 | if (minor_version | |
0f420358 | 1471 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1472 | return -EINVAL; |
c6563a8c N |
1473 | if (minor_version |
1474 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1475 | return -EINVAL; | |
0002b271 | 1476 | |
31b65a0d N |
1477 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1478 | rdev->desc_nr = -1; | |
1479 | else | |
1480 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1481 | ||
2699b672 N |
1482 | if (!rdev->bb_page) { |
1483 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1484 | if (!rdev->bb_page) | |
1485 | return -ENOMEM; | |
1486 | } | |
1487 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1488 | rdev->badblocks.count == 0) { | |
1489 | /* need to load the bad block list. | |
1490 | * Currently we limit it to one page. | |
1491 | */ | |
1492 | s32 offset; | |
1493 | sector_t bb_sector; | |
1494 | u64 *bbp; | |
1495 | int i; | |
1496 | int sectors = le16_to_cpu(sb->bblog_size); | |
1497 | if (sectors > (PAGE_SIZE / 512)) | |
1498 | return -EINVAL; | |
1499 | offset = le32_to_cpu(sb->bblog_offset); | |
1500 | if (offset == 0) | |
1501 | return -EINVAL; | |
1502 | bb_sector = (long long)offset; | |
1503 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
796a5cf0 | 1504 | rdev->bb_page, REQ_OP_READ, 0, true)) |
2699b672 N |
1505 | return -EIO; |
1506 | bbp = (u64 *)page_address(rdev->bb_page); | |
1507 | rdev->badblocks.shift = sb->bblog_shift; | |
1508 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1509 | u64 bb = le64_to_cpu(*bbp); | |
1510 | int count = bb & (0x3ff); | |
1511 | u64 sector = bb >> 10; | |
1512 | sector <<= sb->bblog_shift; | |
1513 | count <<= sb->bblog_shift; | |
1514 | if (bb + 1 == 0) | |
1515 | break; | |
fc974ee2 | 1516 | if (badblocks_set(&rdev->badblocks, sector, count, 1)) |
2699b672 N |
1517 | return -EINVAL; |
1518 | } | |
486adf72 N |
1519 | } else if (sb->bblog_offset != 0) |
1520 | rdev->badblocks.shift = 0; | |
2699b672 | 1521 | |
ea0213e0 AP |
1522 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) { |
1523 | rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset); | |
1524 | rdev->ppl.size = le16_to_cpu(sb->ppl.size); | |
1525 | rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset; | |
1526 | } | |
1527 | ||
9a7b2b0f | 1528 | if (!refdev) { |
8ed75463 | 1529 | ret = 1; |
9a7b2b0f | 1530 | } else { |
1da177e4 | 1531 | __u64 ev1, ev2; |
65a06f06 | 1532 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1533 | |
1534 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1535 | sb->level != refsb->level || | |
1536 | sb->layout != refsb->layout || | |
1537 | sb->chunksize != refsb->chunksize) { | |
9d48739e | 1538 | pr_warn("md: %s has strangely different superblock to %s\n", |
1da177e4 LT |
1539 | bdevname(rdev->bdev,b), |
1540 | bdevname(refdev->bdev,b2)); | |
1541 | return -EINVAL; | |
1542 | } | |
1543 | ev1 = le64_to_cpu(sb->events); | |
1544 | ev2 = le64_to_cpu(refsb->events); | |
1545 | ||
1546 | if (ev1 > ev2) | |
8ed75463 N |
1547 | ret = 1; |
1548 | else | |
1549 | ret = 0; | |
1da177e4 | 1550 | } |
c6563a8c N |
1551 | if (minor_version) { |
1552 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1553 | sectors -= rdev->data_offset; | |
1554 | } else | |
1555 | sectors = rdev->sb_start; | |
1556 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1557 | return -EINVAL; |
dd8ac336 | 1558 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1559 | return ret; |
1da177e4 LT |
1560 | } |
1561 | ||
fd01b88c | 1562 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1563 | { |
65a06f06 | 1564 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1565 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1566 | |
41158c7e | 1567 | rdev->raid_disk = -1; |
c5d79adb N |
1568 | clear_bit(Faulty, &rdev->flags); |
1569 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1570 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1571 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1572 | |
1da177e4 LT |
1573 | if (mddev->raid_disks == 0) { |
1574 | mddev->major_version = 1; | |
1575 | mddev->patch_version = 0; | |
e691063a | 1576 | mddev->external = 0; |
9d8f0363 | 1577 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
9ebc6ef1 DD |
1578 | mddev->ctime = le64_to_cpu(sb->ctime); |
1579 | mddev->utime = le64_to_cpu(sb->utime); | |
1da177e4 | 1580 | mddev->level = le32_to_cpu(sb->level); |
d9d166c2 | 1581 | mddev->clevel[0] = 0; |
1da177e4 LT |
1582 | mddev->layout = le32_to_cpu(sb->layout); |
1583 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1584 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1585 | mddev->events = ev1; |
c3d9714e | 1586 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1587 | mddev->bitmap_info.space = 0; |
1588 | /* Default location for bitmap is 1K after superblock | |
1589 | * using 3K - total of 4K | |
1590 | */ | |
c3d9714e | 1591 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1592 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1593 | mddev->reshape_backwards = 0; |
1594 | ||
1da177e4 LT |
1595 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1596 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1597 | ||
1598 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1599 | |
71c0805c | 1600 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1601 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1602 | mddev->bitmap_info.offset = |
1603 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1604 | /* Metadata doesn't record how much space is available. |
1605 | * For 1.0, we assume we can use up to the superblock | |
1606 | * if before, else to 4K beyond superblock. | |
1607 | * For others, assume no change is possible. | |
1608 | */ | |
1609 | if (mddev->minor_version > 0) | |
1610 | mddev->bitmap_info.space = 0; | |
1611 | else if (mddev->bitmap_info.offset > 0) | |
1612 | mddev->bitmap_info.space = | |
1613 | 8 - mddev->bitmap_info.offset; | |
1614 | else | |
1615 | mddev->bitmap_info.space = | |
1616 | -mddev->bitmap_info.offset; | |
1617 | } | |
e11e93fa | 1618 | |
f6705578 N |
1619 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1620 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1621 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1622 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1623 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1624 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1625 | if (mddev->delta_disks < 0 || |
1626 | (mddev->delta_disks == 0 && | |
1627 | (le32_to_cpu(sb->feature_map) | |
1628 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1629 | mddev->reshape_backwards = 1; | |
f6705578 N |
1630 | } else { |
1631 | mddev->reshape_position = MaxSector; | |
1632 | mddev->delta_disks = 0; | |
1633 | mddev->new_level = mddev->level; | |
1634 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1635 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1636 | } |
1637 | ||
486b0f7b | 1638 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) |
a62ab49e | 1639 | set_bit(MD_HAS_JOURNAL, &mddev->flags); |
ea0213e0 AP |
1640 | |
1641 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) { | |
1642 | if (le32_to_cpu(sb->feature_map) & | |
1643 | (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL)) | |
1644 | return -EINVAL; | |
1645 | set_bit(MD_HAS_PPL, &mddev->flags); | |
1646 | } | |
41158c7e | 1647 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1648 | /* Insist of good event counter while assembling, except for |
1649 | * spares (which don't need an event count) */ | |
1da177e4 | 1650 | ++ev1; |
be6800a7 N |
1651 | if (rdev->desc_nr >= 0 && |
1652 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
a3dfbdaa SL |
1653 | (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX || |
1654 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)) | |
be6800a7 N |
1655 | if (ev1 < mddev->events) |
1656 | return -EINVAL; | |
41158c7e N |
1657 | } else if (mddev->bitmap) { |
1658 | /* If adding to array with a bitmap, then we can accept an | |
1659 | * older device, but not too old. | |
1660 | */ | |
41158c7e N |
1661 | if (ev1 < mddev->bitmap->events_cleared) |
1662 | return 0; | |
8313b8e5 N |
1663 | if (ev1 < mddev->events) |
1664 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1665 | } else { |
1666 | if (ev1 < mddev->events) | |
1667 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1668 | return 0; | |
1669 | } | |
1da177e4 LT |
1670 | if (mddev->level != LEVEL_MULTIPATH) { |
1671 | int role; | |
3673f305 N |
1672 | if (rdev->desc_nr < 0 || |
1673 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
c4d4c91b | 1674 | role = MD_DISK_ROLE_SPARE; |
3673f305 N |
1675 | rdev->desc_nr = -1; |
1676 | } else | |
1677 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 | 1678 | switch(role) { |
c4d4c91b | 1679 | case MD_DISK_ROLE_SPARE: /* spare */ |
1da177e4 | 1680 | break; |
c4d4c91b | 1681 | case MD_DISK_ROLE_FAULTY: /* faulty */ |
b2d444d7 | 1682 | set_bit(Faulty, &rdev->flags); |
1da177e4 | 1683 | break; |
bac624f3 SL |
1684 | case MD_DISK_ROLE_JOURNAL: /* journal device */ |
1685 | if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) { | |
1686 | /* journal device without journal feature */ | |
9d48739e | 1687 | pr_warn("md: journal device provided without journal feature, ignoring the device\n"); |
bac624f3 SL |
1688 | return -EINVAL; |
1689 | } | |
1690 | set_bit(Journal, &rdev->flags); | |
3069aa8d | 1691 | rdev->journal_tail = le64_to_cpu(sb->journal_tail); |
9b15603d | 1692 | rdev->raid_disk = 0; |
bac624f3 | 1693 | break; |
1da177e4 | 1694 | default: |
f466722c | 1695 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1696 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1697 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1698 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1699 | if (!(le32_to_cpu(sb->feature_map) & |
1700 | MD_FEATURE_RECOVERY_BITMAP)) | |
1701 | rdev->saved_raid_disk = -1; | |
1702 | } else | |
5fd6c1dc | 1703 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1704 | rdev->raid_disk = role; |
1705 | break; | |
1706 | } | |
8ddf9efe N |
1707 | if (sb->devflags & WriteMostly1) |
1708 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1709 | if (sb->devflags & FailFast1) |
1710 | set_bit(FailFast, &rdev->flags); | |
2d78f8c4 N |
1711 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1712 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1713 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1714 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1715 | |
1da177e4 LT |
1716 | return 0; |
1717 | } | |
1718 | ||
fd01b88c | 1719 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1720 | { |
1721 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1722 | struct md_rdev *rdev2; |
1da177e4 LT |
1723 | int max_dev, i; |
1724 | /* make rdev->sb match mddev and rdev data. */ | |
1725 | ||
65a06f06 | 1726 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1727 | |
1728 | sb->feature_map = 0; | |
1729 | sb->pad0 = 0; | |
5fd6c1dc | 1730 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1731 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1732 | ||
1733 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1734 | sb->events = cpu_to_le64(mddev->events); | |
1735 | if (mddev->in_sync) | |
1736 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
bd18f646 SL |
1737 | else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags)) |
1738 | sb->resync_offset = cpu_to_le64(MaxSector); | |
1da177e4 LT |
1739 | else |
1740 | sb->resync_offset = cpu_to_le64(0); | |
1741 | ||
1c05b4bc | 1742 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1743 | |
f0ca340c | 1744 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1745 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1746 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1747 | sb->level = cpu_to_le32(mddev->level); |
1748 | sb->layout = cpu_to_le32(mddev->layout); | |
688834e6 N |
1749 | if (test_bit(FailFast, &rdev->flags)) |
1750 | sb->devflags |= FailFast1; | |
1751 | else | |
1752 | sb->devflags &= ~FailFast1; | |
f0ca340c | 1753 | |
aeb9b211 N |
1754 | if (test_bit(WriteMostly, &rdev->flags)) |
1755 | sb->devflags |= WriteMostly1; | |
1756 | else | |
1757 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1758 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1759 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1760 | |
c3d9714e N |
1761 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1762 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1763 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1764 | } |
5fd6c1dc | 1765 | |
f2076e7d | 1766 | if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) && |
97e4f42d | 1767 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1768 | sb->feature_map |= |
1769 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1770 | sb->recovery_offset = | |
1771 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1772 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1773 | sb->feature_map |= | |
1774 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1775 | } |
3069aa8d SL |
1776 | /* Note: recovery_offset and journal_tail share space */ |
1777 | if (test_bit(Journal, &rdev->flags)) | |
1778 | sb->journal_tail = cpu_to_le64(rdev->journal_tail); | |
2d78f8c4 N |
1779 | if (test_bit(Replacement, &rdev->flags)) |
1780 | sb->feature_map |= | |
1781 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1782 | |
f6705578 N |
1783 | if (mddev->reshape_position != MaxSector) { |
1784 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1785 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1786 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1787 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1788 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1789 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1790 | if (mddev->delta_disks == 0 && |
1791 | mddev->reshape_backwards) | |
1792 | sb->feature_map | |
1793 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1794 | if (rdev->new_data_offset != rdev->data_offset) { |
1795 | sb->feature_map | |
1796 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1797 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1798 | - rdev->data_offset)); | |
1799 | } | |
f6705578 | 1800 | } |
a654b9d8 | 1801 | |
3c462c88 GR |
1802 | if (mddev_is_clustered(mddev)) |
1803 | sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED); | |
1804 | ||
2699b672 N |
1805 | if (rdev->badblocks.count == 0) |
1806 | /* Nothing to do for bad blocks*/ ; | |
1807 | else if (sb->bblog_offset == 0) | |
1808 | /* Cannot record bad blocks on this device */ | |
1809 | md_error(mddev, rdev); | |
1810 | else { | |
1811 | struct badblocks *bb = &rdev->badblocks; | |
1812 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1813 | u64 *p = bb->page; | |
1814 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1815 | if (bb->changed) { | |
1816 | unsigned seq; | |
1817 | ||
1818 | retry: | |
1819 | seq = read_seqbegin(&bb->lock); | |
1820 | ||
1821 | memset(bbp, 0xff, PAGE_SIZE); | |
1822 | ||
1823 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1824 | u64 internal_bb = p[i]; |
2699b672 N |
1825 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1826 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1827 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1828 | } |
d0962936 | 1829 | bb->changed = 0; |
2699b672 N |
1830 | if (read_seqretry(&bb->lock, seq)) |
1831 | goto retry; | |
1832 | ||
1833 | bb->sector = (rdev->sb_start + | |
1834 | (int)le32_to_cpu(sb->bblog_offset)); | |
1835 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1836 | } |
1837 | } | |
1838 | ||
1da177e4 | 1839 | max_dev = 0; |
dafb20fa | 1840 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1841 | if (rdev2->desc_nr+1 > max_dev) |
1842 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1843 | |
70471daf N |
1844 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1845 | int bmask; | |
a778b73f | 1846 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1847 | rdev->sb_size = max_dev * 2 + 256; |
1848 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1849 | if (rdev->sb_size & bmask) | |
1850 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1851 | } else |
1852 | max_dev = le32_to_cpu(sb->max_dev); | |
1853 | ||
1da177e4 | 1854 | for (i=0; i<max_dev;i++) |
c4d4c91b | 1855 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY); |
f72ffdd6 | 1856 | |
a97b7896 SL |
1857 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) |
1858 | sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL); | |
f72ffdd6 | 1859 | |
ea0213e0 AP |
1860 | if (test_bit(MD_HAS_PPL, &mddev->flags)) { |
1861 | sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL); | |
1862 | sb->ppl.offset = cpu_to_le16(rdev->ppl.offset); | |
1863 | sb->ppl.size = cpu_to_le16(rdev->ppl.size); | |
1864 | } | |
1865 | ||
dafb20fa | 1866 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1867 | i = rdev2->desc_nr; |
b2d444d7 | 1868 | if (test_bit(Faulty, &rdev2->flags)) |
c4d4c91b | 1869 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY); |
b2d444d7 | 1870 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1871 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
a97b7896 | 1872 | else if (test_bit(Journal, &rdev2->flags)) |
bac624f3 | 1873 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL); |
93be75ff | 1874 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1875 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 | 1876 | else |
c4d4c91b | 1877 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE); |
1da177e4 LT |
1878 | } |
1879 | ||
1da177e4 LT |
1880 | sb->sb_csum = calc_sb_1_csum(sb); |
1881 | } | |
1882 | ||
0cd17fec | 1883 | static unsigned long long |
3cb03002 | 1884 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1885 | { |
1886 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1887 | sector_t max_sectors; |
58c0fed4 | 1888 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1889 | return 0; /* component must fit device */ |
c6563a8c N |
1890 | if (rdev->data_offset != rdev->new_data_offset) |
1891 | return 0; /* too confusing */ | |
0f420358 | 1892 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1893 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1894 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1895 | max_sectors -= rdev->data_offset; |
1896 | if (!num_sectors || num_sectors > max_sectors) | |
1897 | num_sectors = max_sectors; | |
c3d9714e | 1898 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1899 | /* minor version 0 with bitmap we can't move */ |
1900 | return 0; | |
1901 | } else { | |
1902 | /* minor version 0; superblock after data */ | |
0f420358 | 1903 | sector_t sb_start; |
77304d2a | 1904 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1905 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1906 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1907 | if (!num_sectors || num_sectors > max_sectors) |
1908 | num_sectors = max_sectors; | |
0f420358 | 1909 | rdev->sb_start = sb_start; |
0cd17fec | 1910 | } |
65a06f06 | 1911 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1912 | sb->data_size = cpu_to_le64(num_sectors); |
3fb632e4 | 1913 | sb->super_offset = cpu_to_le64(rdev->sb_start); |
0cd17fec | 1914 | sb->sb_csum = calc_sb_1_csum(sb); |
46533ff7 N |
1915 | do { |
1916 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
1917 | rdev->sb_page); | |
1918 | } while (md_super_wait(rdev->mddev) < 0); | |
c26a44ed | 1919 | return num_sectors; |
c6563a8c N |
1920 | |
1921 | } | |
1922 | ||
1923 | static int | |
1924 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1925 | unsigned long long new_offset) | |
1926 | { | |
1927 | /* All necessary checks on new >= old have been done */ | |
1928 | struct bitmap *bitmap; | |
1929 | if (new_offset >= rdev->data_offset) | |
1930 | return 1; | |
1931 | ||
1932 | /* with 1.0 metadata, there is no metadata to tread on | |
1933 | * so we can always move back */ | |
1934 | if (rdev->mddev->minor_version == 0) | |
1935 | return 1; | |
1936 | ||
1937 | /* otherwise we must be sure not to step on | |
1938 | * any metadata, so stay: | |
1939 | * 36K beyond start of superblock | |
1940 | * beyond end of badblocks | |
1941 | * beyond write-intent bitmap | |
1942 | */ | |
1943 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1944 | return 0; | |
1945 | bitmap = rdev->mddev->bitmap; | |
1946 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1947 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1948 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1949 | return 0; |
1950 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1951 | return 0; | |
1952 | ||
1953 | return 1; | |
0cd17fec | 1954 | } |
1da177e4 | 1955 | |
75c96f85 | 1956 | static struct super_type super_types[] = { |
1da177e4 LT |
1957 | [0] = { |
1958 | .name = "0.90.0", | |
1959 | .owner = THIS_MODULE, | |
0cd17fec CW |
1960 | .load_super = super_90_load, |
1961 | .validate_super = super_90_validate, | |
1962 | .sync_super = super_90_sync, | |
1963 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1964 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1965 | }, |
1966 | [1] = { | |
1967 | .name = "md-1", | |
1968 | .owner = THIS_MODULE, | |
0cd17fec CW |
1969 | .load_super = super_1_load, |
1970 | .validate_super = super_1_validate, | |
1971 | .sync_super = super_1_sync, | |
1972 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1973 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1974 | }, |
1975 | }; | |
1da177e4 | 1976 | |
fd01b88c | 1977 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1978 | { |
1979 | if (mddev->sync_super) { | |
1980 | mddev->sync_super(mddev, rdev); | |
1981 | return; | |
1982 | } | |
1983 | ||
1984 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1985 | ||
1986 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1987 | } | |
1988 | ||
fd01b88c | 1989 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1990 | { |
3cb03002 | 1991 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1992 | |
4b80991c | 1993 | rcu_read_lock(); |
0b020e85 SL |
1994 | rdev_for_each_rcu(rdev, mddev1) { |
1995 | if (test_bit(Faulty, &rdev->flags) || | |
1996 | test_bit(Journal, &rdev->flags) || | |
1997 | rdev->raid_disk == -1) | |
1998 | continue; | |
1999 | rdev_for_each_rcu(rdev2, mddev2) { | |
2000 | if (test_bit(Faulty, &rdev2->flags) || | |
2001 | test_bit(Journal, &rdev2->flags) || | |
2002 | rdev2->raid_disk == -1) | |
2003 | continue; | |
7dd5e7c3 | 2004 | if (rdev->bdev->bd_contains == |
4b80991c N |
2005 | rdev2->bdev->bd_contains) { |
2006 | rcu_read_unlock(); | |
7dd5e7c3 | 2007 | return 1; |
4b80991c | 2008 | } |
0b020e85 SL |
2009 | } |
2010 | } | |
4b80991c | 2011 | rcu_read_unlock(); |
1da177e4 LT |
2012 | return 0; |
2013 | } | |
2014 | ||
2015 | static LIST_HEAD(pending_raid_disks); | |
2016 | ||
ac5e7113 AN |
2017 | /* |
2018 | * Try to register data integrity profile for an mddev | |
2019 | * | |
2020 | * This is called when an array is started and after a disk has been kicked | |
2021 | * from the array. It only succeeds if all working and active component devices | |
2022 | * are integrity capable with matching profiles. | |
2023 | */ | |
fd01b88c | 2024 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 2025 | { |
3cb03002 | 2026 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
2027 | |
2028 | if (list_empty(&mddev->disks)) | |
2029 | return 0; /* nothing to do */ | |
629acb6a JB |
2030 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
2031 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 2032 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
2033 | /* skip spares and non-functional disks */ |
2034 | if (test_bit(Faulty, &rdev->flags)) | |
2035 | continue; | |
2036 | if (rdev->raid_disk < 0) | |
2037 | continue; | |
ac5e7113 AN |
2038 | if (!reference) { |
2039 | /* Use the first rdev as the reference */ | |
2040 | reference = rdev; | |
2041 | continue; | |
2042 | } | |
2043 | /* does this rdev's profile match the reference profile? */ | |
2044 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
2045 | rdev->bdev->bd_disk) < 0) | |
2046 | return -EINVAL; | |
2047 | } | |
89078d57 MP |
2048 | if (!reference || !bdev_get_integrity(reference->bdev)) |
2049 | return 0; | |
ac5e7113 AN |
2050 | /* |
2051 | * All component devices are integrity capable and have matching | |
2052 | * profiles, register the common profile for the md device. | |
2053 | */ | |
25520d55 MP |
2054 | blk_integrity_register(mddev->gendisk, |
2055 | bdev_get_integrity(reference->bdev)); | |
2056 | ||
9d48739e | 2057 | pr_debug("md: data integrity enabled on %s\n", mdname(mddev)); |
a91a2785 | 2058 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { |
9d48739e | 2059 | pr_err("md: failed to create integrity pool for %s\n", |
a91a2785 MP |
2060 | mdname(mddev)); |
2061 | return -EINVAL; | |
2062 | } | |
ac5e7113 AN |
2063 | return 0; |
2064 | } | |
2065 | EXPORT_SYMBOL(md_integrity_register); | |
2066 | ||
1501efad DW |
2067 | /* |
2068 | * Attempt to add an rdev, but only if it is consistent with the current | |
2069 | * integrity profile | |
2070 | */ | |
2071 | int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) | |
3f9d99c1 | 2072 | { |
2863b9eb JB |
2073 | struct blk_integrity *bi_rdev; |
2074 | struct blk_integrity *bi_mddev; | |
1501efad | 2075 | char name[BDEVNAME_SIZE]; |
2863b9eb JB |
2076 | |
2077 | if (!mddev->gendisk) | |
1501efad | 2078 | return 0; |
2863b9eb JB |
2079 | |
2080 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2081 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2082 | |
ac5e7113 | 2083 | if (!bi_mddev) /* nothing to do */ |
1501efad DW |
2084 | return 0; |
2085 | ||
2086 | if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) { | |
9d48739e N |
2087 | pr_err("%s: incompatible integrity profile for %s\n", |
2088 | mdname(mddev), bdevname(rdev->bdev, name)); | |
1501efad DW |
2089 | return -ENXIO; |
2090 | } | |
2091 | ||
2092 | return 0; | |
3f9d99c1 | 2093 | } |
ac5e7113 | 2094 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2095 | |
f72ffdd6 | 2096 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2097 | { |
7dd5e7c3 | 2098 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2099 | struct kobject *ko; |
5e55e2f5 | 2100 | int err; |
1da177e4 | 2101 | |
11e2ede0 DW |
2102 | /* prevent duplicates */ |
2103 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2104 | return -EEXIST; | |
2105 | ||
97b20ef7 N |
2106 | if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) && |
2107 | mddev->pers) | |
2108 | return -EROFS; | |
2109 | ||
dd8ac336 | 2110 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
f6b6ec5c SL |
2111 | if (!test_bit(Journal, &rdev->flags) && |
2112 | rdev->sectors && | |
2113 | (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2114 | if (mddev->pers) { |
2115 | /* Cannot change size, so fail | |
2116 | * If mddev->level <= 0, then we don't care | |
2117 | * about aligning sizes (e.g. linear) | |
2118 | */ | |
2119 | if (mddev->level > 0) | |
2120 | return -ENOSPC; | |
2121 | } else | |
dd8ac336 | 2122 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2123 | } |
1da177e4 LT |
2124 | |
2125 | /* Verify rdev->desc_nr is unique. | |
2126 | * If it is -1, assign a free number, else | |
2127 | * check number is not in use | |
2128 | */ | |
4878e9eb | 2129 | rcu_read_lock(); |
1da177e4 LT |
2130 | if (rdev->desc_nr < 0) { |
2131 | int choice = 0; | |
4878e9eb N |
2132 | if (mddev->pers) |
2133 | choice = mddev->raid_disks; | |
57d051dc | 2134 | while (md_find_rdev_nr_rcu(mddev, choice)) |
1da177e4 LT |
2135 | choice++; |
2136 | rdev->desc_nr = choice; | |
2137 | } else { | |
57d051dc | 2138 | if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
4878e9eb | 2139 | rcu_read_unlock(); |
1da177e4 | 2140 | return -EBUSY; |
4878e9eb | 2141 | } |
1da177e4 | 2142 | } |
4878e9eb | 2143 | rcu_read_unlock(); |
f6b6ec5c SL |
2144 | if (!test_bit(Journal, &rdev->flags) && |
2145 | mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { | |
9d48739e N |
2146 | pr_warn("md: %s: array is limited to %d devices\n", |
2147 | mdname(mddev), mddev->max_disks); | |
de01dfad N |
2148 | return -EBUSY; |
2149 | } | |
19133a42 | 2150 | bdevname(rdev->bdev,b); |
90a9befb | 2151 | strreplace(b, '/', '!'); |
649316b2 | 2152 | |
1da177e4 | 2153 | rdev->mddev = mddev; |
9d48739e | 2154 | pr_debug("md: bind<%s>\n", b); |
86e6ffdd | 2155 | |
b2d6db58 | 2156 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2157 | goto fail; |
86e6ffdd | 2158 | |
0762b8bd | 2159 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2160 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2161 | /* failure here is OK */; | |
2162 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2163 | |
4b80991c | 2164 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2165 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2166 | |
2167 | /* May as well allow recovery to be retried once */ | |
5389042f | 2168 | mddev->recovery_disabled++; |
3f9d99c1 | 2169 | |
1da177e4 | 2170 | return 0; |
5e55e2f5 N |
2171 | |
2172 | fail: | |
9d48739e N |
2173 | pr_warn("md: failed to register dev-%s for %s\n", |
2174 | b, mdname(mddev)); | |
5e55e2f5 | 2175 | return err; |
1da177e4 LT |
2176 | } |
2177 | ||
177a99b2 | 2178 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2179 | { |
3cb03002 | 2180 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2181 | kobject_del(&rdev->kobj); |
177a99b2 | 2182 | kobject_put(&rdev->kobj); |
5792a285 N |
2183 | } |
2184 | ||
f72ffdd6 | 2185 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2186 | { |
2187 | char b[BDEVNAME_SIZE]; | |
403df478 | 2188 | |
49731baa | 2189 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2190 | list_del_rcu(&rdev->same_set); |
9d48739e | 2191 | pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1da177e4 | 2192 | rdev->mddev = NULL; |
86e6ffdd | 2193 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2194 | sysfs_put(rdev->sysfs_state); |
2195 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2196 | rdev->badblocks.count = 0; |
5792a285 | 2197 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2198 | * writing to 'remove' to "dev/state". We also need |
2199 | * to delay it due to rcu usage. | |
5792a285 | 2200 | */ |
4b80991c | 2201 | synchronize_rcu(); |
177a99b2 N |
2202 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2203 | kobject_get(&rdev->kobj); | |
e804ac78 | 2204 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2205 | } |
2206 | ||
2207 | /* | |
2208 | * prevent the device from being mounted, repartitioned or | |
2209 | * otherwise reused by a RAID array (or any other kernel | |
2210 | * subsystem), by bd_claiming the device. | |
2211 | */ | |
3cb03002 | 2212 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2213 | { |
2214 | int err = 0; | |
2215 | struct block_device *bdev; | |
2216 | char b[BDEVNAME_SIZE]; | |
2217 | ||
d4d77629 | 2218 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2219 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 | 2220 | if (IS_ERR(bdev)) { |
9d48739e | 2221 | pr_warn("md: could not open %s.\n", __bdevname(dev, b)); |
1da177e4 LT |
2222 | return PTR_ERR(bdev); |
2223 | } | |
1da177e4 LT |
2224 | rdev->bdev = bdev; |
2225 | return err; | |
2226 | } | |
2227 | ||
3cb03002 | 2228 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2229 | { |
2230 | struct block_device *bdev = rdev->bdev; | |
2231 | rdev->bdev = NULL; | |
e525fd89 | 2232 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2233 | } |
2234 | ||
2235 | void md_autodetect_dev(dev_t dev); | |
2236 | ||
f72ffdd6 | 2237 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2238 | { |
2239 | char b[BDEVNAME_SIZE]; | |
403df478 | 2240 | |
9d48739e | 2241 | pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b)); |
545c8795 | 2242 | md_rdev_clear(rdev); |
1da177e4 | 2243 | #ifndef MODULE |
d0fae18f N |
2244 | if (test_bit(AutoDetected, &rdev->flags)) |
2245 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2246 | #endif |
2247 | unlock_rdev(rdev); | |
86e6ffdd | 2248 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2249 | } |
2250 | ||
fb56dfef | 2251 | void md_kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2252 | { |
2253 | unbind_rdev_from_array(rdev); | |
2254 | export_rdev(rdev); | |
2255 | } | |
fb56dfef | 2256 | EXPORT_SYMBOL_GPL(md_kick_rdev_from_array); |
1da177e4 | 2257 | |
fd01b88c | 2258 | static void export_array(struct mddev *mddev) |
1da177e4 | 2259 | { |
0638bb0e | 2260 | struct md_rdev *rdev; |
1da177e4 | 2261 | |
0638bb0e N |
2262 | while (!list_empty(&mddev->disks)) { |
2263 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2264 | same_set); | |
fb56dfef | 2265 | md_kick_rdev_from_array(rdev); |
1da177e4 | 2266 | } |
1da177e4 LT |
2267 | mddev->raid_disks = 0; |
2268 | mddev->major_version = 0; | |
2269 | } | |
2270 | ||
6497709b N |
2271 | static bool set_in_sync(struct mddev *mddev) |
2272 | { | |
2273 | WARN_ON_ONCE(!spin_is_locked(&mddev->lock)); | |
4ad23a97 N |
2274 | if (!mddev->in_sync) { |
2275 | mddev->sync_checkers++; | |
2276 | spin_unlock(&mddev->lock); | |
2277 | percpu_ref_switch_to_atomic_sync(&mddev->writes_pending); | |
2278 | spin_lock(&mddev->lock); | |
2279 | if (!mddev->in_sync && | |
2280 | percpu_ref_is_zero(&mddev->writes_pending)) { | |
6497709b | 2281 | mddev->in_sync = 1; |
4ad23a97 N |
2282 | /* |
2283 | * Ensure ->in_sync is visible before we clear | |
2284 | * ->sync_checkers. | |
2285 | */ | |
55cc39f3 | 2286 | smp_mb(); |
6497709b N |
2287 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2288 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
2289 | } | |
4ad23a97 N |
2290 | if (--mddev->sync_checkers == 0) |
2291 | percpu_ref_switch_to_percpu(&mddev->writes_pending); | |
6497709b N |
2292 | } |
2293 | if (mddev->safemode == 1) | |
2294 | mddev->safemode = 0; | |
2295 | return mddev->in_sync; | |
2296 | } | |
2297 | ||
f72ffdd6 | 2298 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2299 | { |
42543769 N |
2300 | /* Update each superblock (in-memory image), but |
2301 | * if we are allowed to, skip spares which already | |
2302 | * have the right event counter, or have one earlier | |
2303 | * (which would mean they aren't being marked as dirty | |
2304 | * with the rest of the array) | |
2305 | */ | |
3cb03002 | 2306 | struct md_rdev *rdev; |
dafb20fa | 2307 | rdev_for_each(rdev, mddev) { |
42543769 N |
2308 | if (rdev->sb_events == mddev->events || |
2309 | (nospares && | |
2310 | rdev->raid_disk < 0 && | |
42543769 N |
2311 | rdev->sb_events+1 == mddev->events)) { |
2312 | /* Don't update this superblock */ | |
2313 | rdev->sb_loaded = 2; | |
2314 | } else { | |
076f968b | 2315 | sync_super(mddev, rdev); |
42543769 N |
2316 | rdev->sb_loaded = 1; |
2317 | } | |
1da177e4 LT |
2318 | } |
2319 | } | |
2320 | ||
2aa82191 GR |
2321 | static bool does_sb_need_changing(struct mddev *mddev) |
2322 | { | |
2323 | struct md_rdev *rdev; | |
2324 | struct mdp_superblock_1 *sb; | |
2325 | int role; | |
2326 | ||
2327 | /* Find a good rdev */ | |
2328 | rdev_for_each(rdev, mddev) | |
2329 | if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags)) | |
2330 | break; | |
2331 | ||
2332 | /* No good device found. */ | |
2333 | if (!rdev) | |
2334 | return false; | |
2335 | ||
2336 | sb = page_address(rdev->sb_page); | |
2337 | /* Check if a device has become faulty or a spare become active */ | |
2338 | rdev_for_each(rdev, mddev) { | |
2339 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
2340 | /* Device activated? */ | |
2341 | if (role == 0xffff && rdev->raid_disk >=0 && | |
2342 | !test_bit(Faulty, &rdev->flags)) | |
2343 | return true; | |
2344 | /* Device turned faulty? */ | |
2345 | if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd)) | |
2346 | return true; | |
2347 | } | |
2348 | ||
2349 | /* Check if any mddev parameters have changed */ | |
2350 | if ((mddev->dev_sectors != le64_to_cpu(sb->size)) || | |
2351 | (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) || | |
13459213 | 2352 | (mddev->layout != le32_to_cpu(sb->layout)) || |
2aa82191 GR |
2353 | (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) || |
2354 | (mddev->chunk_sectors != le32_to_cpu(sb->chunksize))) | |
2355 | return true; | |
2356 | ||
2357 | return false; | |
2358 | } | |
2359 | ||
1aee41f6 | 2360 | void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2361 | { |
3cb03002 | 2362 | struct md_rdev *rdev; |
06d91a5f | 2363 | int sync_req; |
42543769 | 2364 | int nospares = 0; |
2699b672 | 2365 | int any_badblocks_changed = 0; |
23b63f9f | 2366 | int ret = -1; |
1da177e4 | 2367 | |
d87f064f N |
2368 | if (mddev->ro) { |
2369 | if (force_change) | |
2953079c | 2370 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
d87f064f N |
2371 | return; |
2372 | } | |
2aa82191 | 2373 | |
2c97cf13 | 2374 | repeat: |
2aa82191 | 2375 | if (mddev_is_clustered(mddev)) { |
2953079c | 2376 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
2aa82191 | 2377 | force_change = 1; |
2953079c | 2378 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
85ad1d13 | 2379 | nospares = 1; |
23b63f9f | 2380 | ret = md_cluster_ops->metadata_update_start(mddev); |
2aa82191 GR |
2381 | /* Has someone else has updated the sb */ |
2382 | if (!does_sb_need_changing(mddev)) { | |
23b63f9f GJ |
2383 | if (ret == 0) |
2384 | md_cluster_ops->metadata_update_cancel(mddev); | |
2953079c SL |
2385 | bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2386 | BIT(MD_SB_CHANGE_DEVS) | | |
2387 | BIT(MD_SB_CHANGE_CLEAN)); | |
2aa82191 GR |
2388 | return; |
2389 | } | |
2390 | } | |
2c97cf13 | 2391 | |
3a3a5ddb | 2392 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2393 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2394 | if (rdev->raid_disk >= 0 && |
2395 | mddev->delta_disks >= 0 && | |
f2076e7d | 2396 | !test_bit(Journal, &rdev->flags) && |
3a3a5ddb N |
2397 | !test_bit(In_sync, &rdev->flags) && |
2398 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2399 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2400 | ||
f72ffdd6 | 2401 | } |
bd52b746 | 2402 | if (!mddev->persistent) { |
2953079c SL |
2403 | clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2404 | clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
de393cde | 2405 | if (!mddev->external) { |
2953079c | 2406 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
dafb20fa | 2407 | rdev_for_each(rdev, mddev) { |
de393cde | 2408 | if (rdev->badblocks.changed) { |
d0962936 | 2409 | rdev->badblocks.changed = 0; |
fc974ee2 | 2410 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2411 | md_error(mddev, rdev); |
2412 | } | |
2413 | clear_bit(Blocked, &rdev->flags); | |
2414 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2415 | wake_up(&rdev->blocked_wait); | |
2416 | } | |
2417 | } | |
3a3a5ddb N |
2418 | wake_up(&mddev->sb_wait); |
2419 | return; | |
2420 | } | |
2421 | ||
85572d7c | 2422 | spin_lock(&mddev->lock); |
84692195 | 2423 | |
9ebc6ef1 | 2424 | mddev->utime = ktime_get_real_seconds(); |
3a3a5ddb | 2425 | |
2953079c | 2426 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
850b2b42 | 2427 | force_change = 1; |
2953079c | 2428 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
850b2b42 N |
2429 | /* just a clean<-> dirty transition, possibly leave spares alone, |
2430 | * though if events isn't the right even/odd, we will have to do | |
2431 | * spares after all | |
2432 | */ | |
2433 | nospares = 1; | |
2434 | if (force_change) | |
2435 | nospares = 0; | |
2436 | if (mddev->degraded) | |
84692195 N |
2437 | /* If the array is degraded, then skipping spares is both |
2438 | * dangerous and fairly pointless. | |
2439 | * Dangerous because a device that was removed from the array | |
2440 | * might have a event_count that still looks up-to-date, | |
2441 | * so it can be re-added without a resync. | |
2442 | * Pointless because if there are any spares to skip, | |
2443 | * then a recovery will happen and soon that array won't | |
2444 | * be degraded any more and the spare can go back to sleep then. | |
2445 | */ | |
850b2b42 | 2446 | nospares = 0; |
84692195 | 2447 | |
06d91a5f | 2448 | sync_req = mddev->in_sync; |
42543769 N |
2449 | |
2450 | /* If this is just a dirty<->clean transition, and the array is clean | |
2451 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2452 | if (nospares |
42543769 | 2453 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2454 | && mddev->can_decrease_events |
2455 | && mddev->events != 1) { | |
42543769 | 2456 | mddev->events--; |
a8707c08 N |
2457 | mddev->can_decrease_events = 0; |
2458 | } else { | |
42543769 N |
2459 | /* otherwise we have to go forward and ... */ |
2460 | mddev->events ++; | |
a8707c08 | 2461 | mddev->can_decrease_events = nospares; |
42543769 | 2462 | } |
1da177e4 | 2463 | |
403df478 N |
2464 | /* |
2465 | * This 64-bit counter should never wrap. | |
2466 | * Either we are in around ~1 trillion A.C., assuming | |
2467 | * 1 reboot per second, or we have a bug... | |
2468 | */ | |
2469 | WARN_ON(mddev->events == 0); | |
2699b672 | 2470 | |
dafb20fa | 2471 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2472 | if (rdev->badblocks.changed) |
2473 | any_badblocks_changed++; | |
de393cde N |
2474 | if (test_bit(Faulty, &rdev->flags)) |
2475 | set_bit(FaultRecorded, &rdev->flags); | |
2476 | } | |
2699b672 | 2477 | |
e691063a | 2478 | sync_sbs(mddev, nospares); |
85572d7c | 2479 | spin_unlock(&mddev->lock); |
1da177e4 | 2480 | |
36a4e1fe N |
2481 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2482 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2483 | |
504634f6 SL |
2484 | if (mddev->queue) |
2485 | blk_add_trace_msg(mddev->queue, "md md_update_sb"); | |
46533ff7 | 2486 | rewrite: |
4ad13663 | 2487 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2488 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2489 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2490 | |
42543769 N |
2491 | if (rdev->sb_loaded != 1) |
2492 | continue; /* no noise on spare devices */ | |
1da177e4 | 2493 | |
f466722c | 2494 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2495 | md_super_write(mddev,rdev, |
0f420358 | 2496 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2497 | rdev->sb_page); |
36a4e1fe N |
2498 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2499 | bdevname(rdev->bdev, b), | |
2500 | (unsigned long long)rdev->sb_start); | |
42543769 | 2501 | rdev->sb_events = mddev->events; |
2699b672 N |
2502 | if (rdev->badblocks.size) { |
2503 | md_super_write(mddev, rdev, | |
2504 | rdev->badblocks.sector, | |
2505 | rdev->badblocks.size << 9, | |
2506 | rdev->bb_page); | |
2507 | rdev->badblocks.size = 0; | |
2508 | } | |
7bfa19f2 | 2509 | |
f466722c | 2510 | } else |
36a4e1fe N |
2511 | pr_debug("md: %s (skipping faulty)\n", |
2512 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2513 | |
7bfa19f2 | 2514 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2515 | /* only need to write one superblock... */ |
2516 | break; | |
2517 | } | |
46533ff7 N |
2518 | if (md_super_wait(mddev) < 0) |
2519 | goto rewrite; | |
2953079c | 2520 | /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2521 | |
2c97cf13 GJ |
2522 | if (mddev_is_clustered(mddev) && ret == 0) |
2523 | md_cluster_ops->metadata_update_finish(mddev); | |
2524 | ||
850b2b42 | 2525 | if (mddev->in_sync != sync_req || |
2953079c SL |
2526 | !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2527 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN))) | |
06d91a5f | 2528 | /* have to write it out again */ |
06d91a5f | 2529 | goto repeat; |
3d310eb7 | 2530 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2531 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2532 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2533 | |
dafb20fa | 2534 | rdev_for_each(rdev, mddev) { |
de393cde N |
2535 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2536 | clear_bit(Blocked, &rdev->flags); | |
2537 | ||
2538 | if (any_badblocks_changed) | |
fc974ee2 | 2539 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2540 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2541 | wake_up(&rdev->blocked_wait); | |
2542 | } | |
1da177e4 | 2543 | } |
1aee41f6 | 2544 | EXPORT_SYMBOL(md_update_sb); |
1da177e4 | 2545 | |
a6da4ef8 GR |
2546 | static int add_bound_rdev(struct md_rdev *rdev) |
2547 | { | |
2548 | struct mddev *mddev = rdev->mddev; | |
2549 | int err = 0; | |
87d4d916 | 2550 | bool add_journal = test_bit(Journal, &rdev->flags); |
a6da4ef8 | 2551 | |
87d4d916 | 2552 | if (!mddev->pers->hot_remove_disk || add_journal) { |
a6da4ef8 GR |
2553 | /* If there is hot_add_disk but no hot_remove_disk |
2554 | * then added disks for geometry changes, | |
2555 | * and should be added immediately. | |
2556 | */ | |
2557 | super_types[mddev->major_version]. | |
2558 | validate_super(mddev, rdev); | |
87d4d916 SL |
2559 | if (add_journal) |
2560 | mddev_suspend(mddev); | |
a6da4ef8 | 2561 | err = mddev->pers->hot_add_disk(mddev, rdev); |
87d4d916 SL |
2562 | if (add_journal) |
2563 | mddev_resume(mddev); | |
a6da4ef8 | 2564 | if (err) { |
db767672 | 2565 | md_kick_rdev_from_array(rdev); |
a6da4ef8 GR |
2566 | return err; |
2567 | } | |
2568 | } | |
2569 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2570 | ||
2953079c | 2571 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a6da4ef8 GR |
2572 | if (mddev->degraded) |
2573 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
2574 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2575 | md_new_event(mddev); | |
2576 | md_wakeup_thread(mddev->thread); | |
2577 | return 0; | |
2578 | } | |
1da177e4 | 2579 | |
7f6ce769 | 2580 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2581 | * We want to accept with case. For this we use cmd_match. |
2582 | */ | |
2583 | static int cmd_match(const char *cmd, const char *str) | |
2584 | { | |
2585 | /* See if cmd, written into a sysfs file, matches | |
2586 | * str. They must either be the same, or cmd can | |
2587 | * have a trailing newline | |
2588 | */ | |
2589 | while (*cmd && *str && *cmd == *str) { | |
2590 | cmd++; | |
2591 | str++; | |
2592 | } | |
2593 | if (*cmd == '\n') | |
2594 | cmd++; | |
2595 | if (*str || *cmd) | |
2596 | return 0; | |
2597 | return 1; | |
2598 | } | |
2599 | ||
86e6ffdd N |
2600 | struct rdev_sysfs_entry { |
2601 | struct attribute attr; | |
3cb03002 N |
2602 | ssize_t (*show)(struct md_rdev *, char *); |
2603 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2604 | }; |
2605 | ||
2606 | static ssize_t | |
3cb03002 | 2607 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd | 2608 | { |
35b785f7 | 2609 | char *sep = ","; |
20a49ff6 | 2610 | size_t len = 0; |
758bfc8a | 2611 | unsigned long flags = ACCESS_ONCE(rdev->flags); |
86e6ffdd | 2612 | |
758bfc8a | 2613 | if (test_bit(Faulty, &flags) || |
dcbcb486 TM |
2614 | (!test_bit(ExternalBbl, &flags) && |
2615 | rdev->badblocks.unacked_exist)) | |
35b785f7 TM |
2616 | len += sprintf(page+len, "faulty%s", sep); |
2617 | if (test_bit(In_sync, &flags)) | |
2618 | len += sprintf(page+len, "in_sync%s", sep); | |
2619 | if (test_bit(Journal, &flags)) | |
2620 | len += sprintf(page+len, "journal%s", sep); | |
2621 | if (test_bit(WriteMostly, &flags)) | |
2622 | len += sprintf(page+len, "write_mostly%s", sep); | |
758bfc8a | 2623 | if (test_bit(Blocked, &flags) || |
52c64152 | 2624 | (rdev->badblocks.unacked_exist |
35b785f7 TM |
2625 | && !test_bit(Faulty, &flags))) |
2626 | len += sprintf(page+len, "blocked%s", sep); | |
758bfc8a | 2627 | if (!test_bit(Faulty, &flags) && |
f2076e7d | 2628 | !test_bit(Journal, &flags) && |
35b785f7 TM |
2629 | !test_bit(In_sync, &flags)) |
2630 | len += sprintf(page+len, "spare%s", sep); | |
2631 | if (test_bit(WriteErrorSeen, &flags)) | |
2632 | len += sprintf(page+len, "write_error%s", sep); | |
2633 | if (test_bit(WantReplacement, &flags)) | |
2634 | len += sprintf(page+len, "want_replacement%s", sep); | |
2635 | if (test_bit(Replacement, &flags)) | |
2636 | len += sprintf(page+len, "replacement%s", sep); | |
2637 | if (test_bit(ExternalBbl, &flags)) | |
2638 | len += sprintf(page+len, "external_bbl%s", sep); | |
688834e6 N |
2639 | if (test_bit(FailFast, &flags)) |
2640 | len += sprintf(page+len, "failfast%s", sep); | |
35b785f7 TM |
2641 | |
2642 | if (len) | |
2643 | len -= strlen(sep); | |
2d78f8c4 | 2644 | |
86e6ffdd N |
2645 | return len+sprintf(page+len, "\n"); |
2646 | } | |
2647 | ||
45dc2de1 | 2648 | static ssize_t |
3cb03002 | 2649 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2650 | { |
2651 | /* can write | |
de393cde | 2652 | * faulty - simulates an error |
45dc2de1 | 2653 | * remove - disconnects the device |
f655675b N |
2654 | * writemostly - sets write_mostly |
2655 | * -writemostly - clears write_mostly | |
de393cde N |
2656 | * blocked - sets the Blocked flags |
2657 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2658 | * insync - sets Insync providing device isn't active |
f466722c N |
2659 | * -insync - clear Insync for a device with a slot assigned, |
2660 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2661 | * write_error - sets WriteErrorSeen |
2662 | * -write_error - clears WriteErrorSeen | |
688834e6 | 2663 | * {,-}failfast - set/clear FailFast |
45dc2de1 N |
2664 | */ |
2665 | int err = -EINVAL; | |
2666 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2667 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2668 | if (test_bit(Faulty, &rdev->flags)) |
2669 | err = 0; | |
2670 | else | |
2671 | err = -EBUSY; | |
45dc2de1 | 2672 | } else if (cmd_match(buf, "remove")) { |
5d881783 SL |
2673 | if (rdev->mddev->pers) { |
2674 | clear_bit(Blocked, &rdev->flags); | |
2675 | remove_and_add_spares(rdev->mddev, rdev); | |
2676 | } | |
45dc2de1 N |
2677 | if (rdev->raid_disk >= 0) |
2678 | err = -EBUSY; | |
2679 | else { | |
fd01b88c | 2680 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2681 | err = 0; |
a9720903 GJ |
2682 | if (mddev_is_clustered(mddev)) |
2683 | err = md_cluster_ops->remove_disk(mddev, rdev); | |
2684 | ||
2685 | if (err == 0) { | |
2686 | md_kick_rdev_from_array(rdev); | |
060b0689 | 2687 | if (mddev->pers) { |
2953079c | 2688 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
2689 | md_wakeup_thread(mddev->thread); |
2690 | } | |
a9720903 GJ |
2691 | md_new_event(mddev); |
2692 | } | |
45dc2de1 | 2693 | } |
f655675b N |
2694 | } else if (cmd_match(buf, "writemostly")) { |
2695 | set_bit(WriteMostly, &rdev->flags); | |
2696 | err = 0; | |
2697 | } else if (cmd_match(buf, "-writemostly")) { | |
2698 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2699 | err = 0; |
2700 | } else if (cmd_match(buf, "blocked")) { | |
2701 | set_bit(Blocked, &rdev->flags); | |
2702 | err = 0; | |
2703 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2704 | if (!test_bit(Faulty, &rdev->flags) && |
dcbcb486 | 2705 | !test_bit(ExternalBbl, &rdev->flags) && |
7da64a0a | 2706 | rdev->badblocks.unacked_exist) { |
de393cde N |
2707 | /* metadata handler doesn't understand badblocks, |
2708 | * so we need to fail the device | |
2709 | */ | |
2710 | md_error(rdev->mddev, rdev); | |
2711 | } | |
6bfe0b49 | 2712 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2713 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2714 | wake_up(&rdev->blocked_wait); |
2715 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2716 | md_wakeup_thread(rdev->mddev->thread); | |
2717 | ||
6d56e278 N |
2718 | err = 0; |
2719 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2720 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2721 | err = 0; |
688834e6 N |
2722 | } else if (cmd_match(buf, "failfast")) { |
2723 | set_bit(FailFast, &rdev->flags); | |
2724 | err = 0; | |
2725 | } else if (cmd_match(buf, "-failfast")) { | |
2726 | clear_bit(FailFast, &rdev->flags); | |
2727 | err = 0; | |
f2076e7d SL |
2728 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 && |
2729 | !test_bit(Journal, &rdev->flags)) { | |
e1960f8c N |
2730 | if (rdev->mddev->pers == NULL) { |
2731 | clear_bit(In_sync, &rdev->flags); | |
2732 | rdev->saved_raid_disk = rdev->raid_disk; | |
2733 | rdev->raid_disk = -1; | |
2734 | err = 0; | |
2735 | } | |
d7a9d443 N |
2736 | } else if (cmd_match(buf, "write_error")) { |
2737 | set_bit(WriteErrorSeen, &rdev->flags); | |
2738 | err = 0; | |
2739 | } else if (cmd_match(buf, "-write_error")) { | |
2740 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2741 | err = 0; | |
2d78f8c4 N |
2742 | } else if (cmd_match(buf, "want_replacement")) { |
2743 | /* Any non-spare device that is not a replacement can | |
2744 | * become want_replacement at any time, but we then need to | |
2745 | * check if recovery is needed. | |
2746 | */ | |
2747 | if (rdev->raid_disk >= 0 && | |
f2076e7d | 2748 | !test_bit(Journal, &rdev->flags) && |
2d78f8c4 N |
2749 | !test_bit(Replacement, &rdev->flags)) |
2750 | set_bit(WantReplacement, &rdev->flags); | |
2751 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2752 | md_wakeup_thread(rdev->mddev->thread); | |
2753 | err = 0; | |
2754 | } else if (cmd_match(buf, "-want_replacement")) { | |
2755 | /* Clearing 'want_replacement' is always allowed. | |
2756 | * Once replacements starts it is too late though. | |
2757 | */ | |
2758 | err = 0; | |
2759 | clear_bit(WantReplacement, &rdev->flags); | |
2760 | } else if (cmd_match(buf, "replacement")) { | |
2761 | /* Can only set a device as a replacement when array has not | |
2762 | * yet been started. Once running, replacement is automatic | |
2763 | * from spares, or by assigning 'slot'. | |
2764 | */ | |
2765 | if (rdev->mddev->pers) | |
2766 | err = -EBUSY; | |
2767 | else { | |
2768 | set_bit(Replacement, &rdev->flags); | |
2769 | err = 0; | |
2770 | } | |
2771 | } else if (cmd_match(buf, "-replacement")) { | |
2772 | /* Similarly, can only clear Replacement before start */ | |
2773 | if (rdev->mddev->pers) | |
2774 | err = -EBUSY; | |
2775 | else { | |
2776 | clear_bit(Replacement, &rdev->flags); | |
2777 | err = 0; | |
2778 | } | |
a6da4ef8 GR |
2779 | } else if (cmd_match(buf, "re-add")) { |
2780 | if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) { | |
97f6cd39 GR |
2781 | /* clear_bit is performed _after_ all the devices |
2782 | * have their local Faulty bit cleared. If any writes | |
2783 | * happen in the meantime in the local node, they | |
2784 | * will land in the local bitmap, which will be synced | |
2785 | * by this node eventually | |
2786 | */ | |
2787 | if (!mddev_is_clustered(rdev->mddev) || | |
2788 | (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) { | |
2789 | clear_bit(Faulty, &rdev->flags); | |
2790 | err = add_bound_rdev(rdev); | |
2791 | } | |
a6da4ef8 GR |
2792 | } else |
2793 | err = -EBUSY; | |
35b785f7 TM |
2794 | } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) { |
2795 | set_bit(ExternalBbl, &rdev->flags); | |
2796 | rdev->badblocks.shift = 0; | |
2797 | err = 0; | |
2798 | } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) { | |
2799 | clear_bit(ExternalBbl, &rdev->flags); | |
2800 | err = 0; | |
45dc2de1 | 2801 | } |
00bcb4ac N |
2802 | if (!err) |
2803 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2804 | return err ? err : len; |
2805 | } | |
80ca3a44 | 2806 | static struct rdev_sysfs_entry rdev_state = |
750f199e | 2807 | __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store); |
86e6ffdd | 2808 | |
4dbcdc75 | 2809 | static ssize_t |
3cb03002 | 2810 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2811 | { |
2812 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2813 | } | |
2814 | ||
2815 | static ssize_t | |
3cb03002 | 2816 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 | 2817 | { |
4c9309c0 AD |
2818 | unsigned int n; |
2819 | int rv; | |
2820 | ||
2821 | rv = kstrtouint(buf, 10, &n); | |
2822 | if (rv < 0) | |
2823 | return rv; | |
2824 | atomic_set(&rdev->corrected_errors, n); | |
2825 | return len; | |
4dbcdc75 N |
2826 | } |
2827 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2828 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2829 | |
014236d2 | 2830 | static ssize_t |
3cb03002 | 2831 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 | 2832 | { |
f2076e7d SL |
2833 | if (test_bit(Journal, &rdev->flags)) |
2834 | return sprintf(page, "journal\n"); | |
2835 | else if (rdev->raid_disk < 0) | |
014236d2 N |
2836 | return sprintf(page, "none\n"); |
2837 | else | |
2838 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2839 | } | |
2840 | ||
2841 | static ssize_t | |
3cb03002 | 2842 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 | 2843 | { |
4c9309c0 | 2844 | int slot; |
c303da6d | 2845 | int err; |
4c9309c0 | 2846 | |
f2076e7d SL |
2847 | if (test_bit(Journal, &rdev->flags)) |
2848 | return -EBUSY; | |
014236d2 N |
2849 | if (strncmp(buf, "none", 4)==0) |
2850 | slot = -1; | |
4c9309c0 AD |
2851 | else { |
2852 | err = kstrtouint(buf, 10, (unsigned int *)&slot); | |
2853 | if (err < 0) | |
2854 | return err; | |
2855 | } | |
6c2fce2e | 2856 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2857 | /* Setting 'slot' on an active array requires also |
2858 | * updating the 'rd%d' link, and communicating | |
2859 | * with the personality with ->hot_*_disk. | |
2860 | * For now we only support removing | |
2861 | * failed/spare devices. This normally happens automatically, | |
2862 | * but not when the metadata is externally managed. | |
2863 | */ | |
c303da6d N |
2864 | if (rdev->raid_disk == -1) |
2865 | return -EEXIST; | |
2866 | /* personality does all needed checks */ | |
01393f3d | 2867 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2868 | return -EINVAL; |
746d3207 N |
2869 | clear_bit(Blocked, &rdev->flags); |
2870 | remove_and_add_spares(rdev->mddev, rdev); | |
2871 | if (rdev->raid_disk >= 0) | |
2872 | return -EBUSY; | |
c303da6d N |
2873 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2874 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2875 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2876 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2877 | * if we ever get bitmaps working here. |
6c2fce2e | 2878 | */ |
cb01c549 | 2879 | int err; |
6c2fce2e NB |
2880 | |
2881 | if (rdev->raid_disk != -1) | |
2882 | return -EBUSY; | |
2883 | ||
c6751b2b N |
2884 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2885 | return -EBUSY; | |
2886 | ||
6c2fce2e NB |
2887 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2888 | return -EINVAL; | |
2889 | ||
ba1b41b6 N |
2890 | if (slot >= rdev->mddev->raid_disks && |
2891 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2892 | return -ENOSPC; | |
2893 | ||
6c2fce2e NB |
2894 | rdev->raid_disk = slot; |
2895 | if (test_bit(In_sync, &rdev->flags)) | |
2896 | rdev->saved_raid_disk = slot; | |
2897 | else | |
2898 | rdev->saved_raid_disk = -1; | |
d30519fc | 2899 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2900 | clear_bit(Bitmap_sync, &rdev->flags); |
cb01c549 GR |
2901 | err = rdev->mddev->pers-> |
2902 | hot_add_disk(rdev->mddev, rdev); | |
2903 | if (err) { | |
2904 | rdev->raid_disk = -1; | |
2905 | return err; | |
2906 | } else | |
2907 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2908 | if (sysfs_link_rdev(rdev->mddev, rdev)) | |
2909 | /* failure here is OK */; | |
6c2fce2e | 2910 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2911 | } else { |
ba1b41b6 N |
2912 | if (slot >= rdev->mddev->raid_disks && |
2913 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2914 | return -ENOSPC; |
2915 | rdev->raid_disk = slot; | |
2916 | /* assume it is working */ | |
c5d79adb N |
2917 | clear_bit(Faulty, &rdev->flags); |
2918 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2919 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2920 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2921 | } |
014236d2 N |
2922 | return len; |
2923 | } | |
2924 | ||
014236d2 | 2925 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2926 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2927 | |
93c8cad0 | 2928 | static ssize_t |
3cb03002 | 2929 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2930 | { |
6961ece4 | 2931 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2932 | } |
2933 | ||
2934 | static ssize_t | |
3cb03002 | 2935 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2936 | { |
c6563a8c | 2937 | unsigned long long offset; |
b29bebd6 | 2938 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2939 | return -EINVAL; |
8ed0a521 | 2940 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2941 | return -EBUSY; |
dd8ac336 | 2942 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2943 | /* Must set offset before size, so overlap checks |
2944 | * can be sane */ | |
2945 | return -EBUSY; | |
93c8cad0 | 2946 | rdev->data_offset = offset; |
25f7fd47 | 2947 | rdev->new_data_offset = offset; |
93c8cad0 N |
2948 | return len; |
2949 | } | |
2950 | ||
2951 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2952 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2953 | |
c6563a8c N |
2954 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2955 | { | |
2956 | return sprintf(page, "%llu\n", | |
2957 | (unsigned long long)rdev->new_data_offset); | |
2958 | } | |
2959 | ||
2960 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2961 | const char *buf, size_t len) | |
2962 | { | |
2963 | unsigned long long new_offset; | |
2964 | struct mddev *mddev = rdev->mddev; | |
2965 | ||
b29bebd6 | 2966 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2967 | return -EINVAL; |
2968 | ||
f851b60d N |
2969 | if (mddev->sync_thread || |
2970 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
2971 | return -EBUSY; |
2972 | if (new_offset == rdev->data_offset) | |
2973 | /* reset is always permitted */ | |
2974 | ; | |
2975 | else if (new_offset > rdev->data_offset) { | |
2976 | /* must not push array size beyond rdev_sectors */ | |
2977 | if (new_offset - rdev->data_offset | |
2978 | + mddev->dev_sectors > rdev->sectors) | |
2979 | return -E2BIG; | |
2980 | } | |
2981 | /* Metadata worries about other space details. */ | |
2982 | ||
2983 | /* decreasing the offset is inconsistent with a backwards | |
2984 | * reshape. | |
2985 | */ | |
2986 | if (new_offset < rdev->data_offset && | |
2987 | mddev->reshape_backwards) | |
2988 | return -EINVAL; | |
2989 | /* Increasing offset is inconsistent with forwards | |
2990 | * reshape. reshape_direction should be set to | |
2991 | * 'backwards' first. | |
2992 | */ | |
2993 | if (new_offset > rdev->data_offset && | |
2994 | !mddev->reshape_backwards) | |
2995 | return -EINVAL; | |
2996 | ||
2997 | if (mddev->pers && mddev->persistent && | |
2998 | !super_types[mddev->major_version] | |
2999 | .allow_new_offset(rdev, new_offset)) | |
3000 | return -E2BIG; | |
3001 | rdev->new_data_offset = new_offset; | |
3002 | if (new_offset > rdev->data_offset) | |
3003 | mddev->reshape_backwards = 1; | |
3004 | else if (new_offset < rdev->data_offset) | |
3005 | mddev->reshape_backwards = 0; | |
3006 | ||
3007 | return len; | |
3008 | } | |
3009 | static struct rdev_sysfs_entry rdev_new_offset = | |
3010 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
3011 | ||
83303b61 | 3012 | static ssize_t |
3cb03002 | 3013 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 3014 | { |
dd8ac336 | 3015 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
3016 | } |
3017 | ||
c5d79adb N |
3018 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
3019 | { | |
3020 | /* check if two start/length pairs overlap */ | |
3021 | if (s1+l1 <= s2) | |
3022 | return 0; | |
3023 | if (s2+l2 <= s1) | |
3024 | return 0; | |
3025 | return 1; | |
3026 | } | |
3027 | ||
b522adcd DW |
3028 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
3029 | { | |
3030 | unsigned long long blocks; | |
3031 | sector_t new; | |
3032 | ||
b29bebd6 | 3033 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
3034 | return -EINVAL; |
3035 | ||
3036 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
3037 | return -EINVAL; /* sector conversion overflow */ | |
3038 | ||
3039 | new = blocks * 2; | |
3040 | if (new != blocks * 2) | |
3041 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
3042 | ||
3043 | *sectors = new; | |
3044 | return 0; | |
3045 | } | |
3046 | ||
83303b61 | 3047 | static ssize_t |
3cb03002 | 3048 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 3049 | { |
fd01b88c | 3050 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 3051 | sector_t oldsectors = rdev->sectors; |
b522adcd | 3052 | sector_t sectors; |
27c529bb | 3053 | |
f2076e7d SL |
3054 | if (test_bit(Journal, &rdev->flags)) |
3055 | return -EBUSY; | |
b522adcd | 3056 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 3057 | return -EINVAL; |
c6563a8c N |
3058 | if (rdev->data_offset != rdev->new_data_offset) |
3059 | return -EINVAL; /* too confusing */ | |
0cd17fec | 3060 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 3061 | if (my_mddev->persistent) { |
dd8ac336 AN |
3062 | sectors = super_types[my_mddev->major_version]. |
3063 | rdev_size_change(rdev, sectors); | |
3064 | if (!sectors) | |
0cd17fec | 3065 | return -EBUSY; |
dd8ac336 | 3066 | } else if (!sectors) |
77304d2a | 3067 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 3068 | rdev->data_offset; |
a6468539 N |
3069 | if (!my_mddev->pers->resize) |
3070 | /* Cannot change size for RAID0 or Linear etc */ | |
3071 | return -EINVAL; | |
0cd17fec | 3072 | } |
dd8ac336 | 3073 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 3074 | return -EINVAL; /* component must fit device */ |
0cd17fec | 3075 | |
dd8ac336 AN |
3076 | rdev->sectors = sectors; |
3077 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
3078 | /* Need to check that all other rdevs with the same |
3079 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
3080 | * the rdev lists safely. | |
3081 | * This check does not provide a hard guarantee, it | |
3082 | * just helps avoid dangerous mistakes. | |
c5d79adb | 3083 | */ |
fd01b88c | 3084 | struct mddev *mddev; |
c5d79adb | 3085 | int overlap = 0; |
159ec1fc | 3086 | struct list_head *tmp; |
c5d79adb | 3087 | |
8b1afc3d | 3088 | rcu_read_lock(); |
29ac4aa3 | 3089 | for_each_mddev(mddev, tmp) { |
3cb03002 | 3090 | struct md_rdev *rdev2; |
c5d79adb | 3091 | |
dafb20fa | 3092 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
3093 | if (rdev->bdev == rdev2->bdev && |
3094 | rdev != rdev2 && | |
3095 | overlaps(rdev->data_offset, rdev->sectors, | |
3096 | rdev2->data_offset, | |
3097 | rdev2->sectors)) { | |
c5d79adb N |
3098 | overlap = 1; |
3099 | break; | |
3100 | } | |
c5d79adb N |
3101 | if (overlap) { |
3102 | mddev_put(mddev); | |
3103 | break; | |
3104 | } | |
3105 | } | |
8b1afc3d | 3106 | rcu_read_unlock(); |
c5d79adb N |
3107 | if (overlap) { |
3108 | /* Someone else could have slipped in a size | |
3109 | * change here, but doing so is just silly. | |
dd8ac336 | 3110 | * We put oldsectors back because we *know* it is |
c5d79adb N |
3111 | * safe, and trust userspace not to race with |
3112 | * itself | |
3113 | */ | |
dd8ac336 | 3114 | rdev->sectors = oldsectors; |
c5d79adb N |
3115 | return -EBUSY; |
3116 | } | |
3117 | } | |
83303b61 N |
3118 | return len; |
3119 | } | |
3120 | ||
3121 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 3122 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 3123 | |
3cb03002 | 3124 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
3125 | { |
3126 | unsigned long long recovery_start = rdev->recovery_offset; | |
3127 | ||
3128 | if (test_bit(In_sync, &rdev->flags) || | |
3129 | recovery_start == MaxSector) | |
3130 | return sprintf(page, "none\n"); | |
3131 | ||
3132 | return sprintf(page, "%llu\n", recovery_start); | |
3133 | } | |
3134 | ||
3cb03002 | 3135 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
3136 | { |
3137 | unsigned long long recovery_start; | |
3138 | ||
3139 | if (cmd_match(buf, "none")) | |
3140 | recovery_start = MaxSector; | |
b29bebd6 | 3141 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
3142 | return -EINVAL; |
3143 | ||
3144 | if (rdev->mddev->pers && | |
3145 | rdev->raid_disk >= 0) | |
3146 | return -EBUSY; | |
3147 | ||
3148 | rdev->recovery_offset = recovery_start; | |
3149 | if (recovery_start == MaxSector) | |
3150 | set_bit(In_sync, &rdev->flags); | |
3151 | else | |
3152 | clear_bit(In_sync, &rdev->flags); | |
3153 | return len; | |
3154 | } | |
3155 | ||
3156 | static struct rdev_sysfs_entry rdev_recovery_start = | |
3157 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
3158 | ||
fc974ee2 VV |
3159 | /* sysfs access to bad-blocks list. |
3160 | * We present two files. | |
3161 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
3162 | * are recorded as bad. The list is truncated to fit within | |
3163 | * the one-page limit of sysfs. | |
3164 | * Writing "sector length" to this file adds an acknowledged | |
3165 | * bad block list. | |
3166 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
3167 | * been acknowledged. Writing to this file adds bad blocks | |
3168 | * without acknowledging them. This is largely for testing. | |
3169 | */ | |
3cb03002 | 3170 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3171 | { |
3172 | return badblocks_show(&rdev->badblocks, page, 0); | |
3173 | } | |
3cb03002 | 3174 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 3175 | { |
de393cde N |
3176 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
3177 | /* Maybe that ack was all we needed */ | |
3178 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
3179 | wake_up(&rdev->blocked_wait); | |
3180 | return rv; | |
16c791a5 N |
3181 | } |
3182 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
3183 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
3184 | ||
3cb03002 | 3185 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3186 | { |
3187 | return badblocks_show(&rdev->badblocks, page, 1); | |
3188 | } | |
3cb03002 | 3189 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
3190 | { |
3191 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
3192 | } | |
3193 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
3194 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
3195 | ||
664aed04 AP |
3196 | static ssize_t |
3197 | ppl_sector_show(struct md_rdev *rdev, char *page) | |
3198 | { | |
3199 | return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector); | |
3200 | } | |
3201 | ||
3202 | static ssize_t | |
3203 | ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len) | |
3204 | { | |
3205 | unsigned long long sector; | |
3206 | ||
3207 | if (kstrtoull(buf, 10, §or) < 0) | |
3208 | return -EINVAL; | |
3209 | if (sector != (sector_t)sector) | |
3210 | return -EINVAL; | |
3211 | ||
3212 | if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) && | |
3213 | rdev->raid_disk >= 0) | |
3214 | return -EBUSY; | |
3215 | ||
3216 | if (rdev->mddev->persistent) { | |
3217 | if (rdev->mddev->major_version == 0) | |
3218 | return -EINVAL; | |
3219 | if ((sector > rdev->sb_start && | |
3220 | sector - rdev->sb_start > S16_MAX) || | |
3221 | (sector < rdev->sb_start && | |
3222 | rdev->sb_start - sector > -S16_MIN)) | |
3223 | return -EINVAL; | |
3224 | rdev->ppl.offset = sector - rdev->sb_start; | |
3225 | } else if (!rdev->mddev->external) { | |
3226 | return -EBUSY; | |
3227 | } | |
3228 | rdev->ppl.sector = sector; | |
3229 | return len; | |
3230 | } | |
3231 | ||
3232 | static struct rdev_sysfs_entry rdev_ppl_sector = | |
3233 | __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store); | |
3234 | ||
3235 | static ssize_t | |
3236 | ppl_size_show(struct md_rdev *rdev, char *page) | |
3237 | { | |
3238 | return sprintf(page, "%u\n", rdev->ppl.size); | |
3239 | } | |
3240 | ||
3241 | static ssize_t | |
3242 | ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len) | |
3243 | { | |
3244 | unsigned int size; | |
3245 | ||
3246 | if (kstrtouint(buf, 10, &size) < 0) | |
3247 | return -EINVAL; | |
3248 | ||
3249 | if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) && | |
3250 | rdev->raid_disk >= 0) | |
3251 | return -EBUSY; | |
3252 | ||
3253 | if (rdev->mddev->persistent) { | |
3254 | if (rdev->mddev->major_version == 0) | |
3255 | return -EINVAL; | |
3256 | if (size > U16_MAX) | |
3257 | return -EINVAL; | |
3258 | } else if (!rdev->mddev->external) { | |
3259 | return -EBUSY; | |
3260 | } | |
3261 | rdev->ppl.size = size; | |
3262 | return len; | |
3263 | } | |
3264 | ||
3265 | static struct rdev_sysfs_entry rdev_ppl_size = | |
3266 | __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store); | |
3267 | ||
86e6ffdd N |
3268 | static struct attribute *rdev_default_attrs[] = { |
3269 | &rdev_state.attr, | |
4dbcdc75 | 3270 | &rdev_errors.attr, |
014236d2 | 3271 | &rdev_slot.attr, |
93c8cad0 | 3272 | &rdev_offset.attr, |
c6563a8c | 3273 | &rdev_new_offset.attr, |
83303b61 | 3274 | &rdev_size.attr, |
06e3c817 | 3275 | &rdev_recovery_start.attr, |
16c791a5 N |
3276 | &rdev_bad_blocks.attr, |
3277 | &rdev_unack_bad_blocks.attr, | |
664aed04 AP |
3278 | &rdev_ppl_sector.attr, |
3279 | &rdev_ppl_size.attr, | |
86e6ffdd N |
3280 | NULL, |
3281 | }; | |
3282 | static ssize_t | |
3283 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3284 | { | |
3285 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3286 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
3287 | |
3288 | if (!entry->show) | |
3289 | return -EIO; | |
758bfc8a N |
3290 | if (!rdev->mddev) |
3291 | return -EBUSY; | |
3292 | return entry->show(rdev, page); | |
86e6ffdd N |
3293 | } |
3294 | ||
3295 | static ssize_t | |
3296 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3297 | const char *page, size_t length) | |
3298 | { | |
3299 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3300 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3301 | ssize_t rv; |
fd01b88c | 3302 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3303 | |
3304 | if (!entry->store) | |
3305 | return -EIO; | |
67463acb N |
3306 | if (!capable(CAP_SYS_ADMIN)) |
3307 | return -EACCES; | |
27c529bb | 3308 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3309 | if (!rv) { |
27c529bb N |
3310 | if (rdev->mddev == NULL) |
3311 | rv = -EBUSY; | |
3312 | else | |
3313 | rv = entry->store(rdev, page, length); | |
6a51830e | 3314 | mddev_unlock(mddev); |
ca388059 N |
3315 | } |
3316 | return rv; | |
86e6ffdd N |
3317 | } |
3318 | ||
3319 | static void rdev_free(struct kobject *ko) | |
3320 | { | |
3cb03002 | 3321 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3322 | kfree(rdev); |
3323 | } | |
52cf25d0 | 3324 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3325 | .show = rdev_attr_show, |
3326 | .store = rdev_attr_store, | |
3327 | }; | |
3328 | static struct kobj_type rdev_ktype = { | |
3329 | .release = rdev_free, | |
3330 | .sysfs_ops = &rdev_sysfs_ops, | |
3331 | .default_attrs = rdev_default_attrs, | |
3332 | }; | |
3333 | ||
3cb03002 | 3334 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3335 | { |
3336 | rdev->desc_nr = -1; | |
3337 | rdev->saved_raid_disk = -1; | |
3338 | rdev->raid_disk = -1; | |
3339 | rdev->flags = 0; | |
3340 | rdev->data_offset = 0; | |
c6563a8c | 3341 | rdev->new_data_offset = 0; |
e8bb9a83 | 3342 | rdev->sb_events = 0; |
0e3ef49e | 3343 | rdev->last_read_error = 0; |
2699b672 N |
3344 | rdev->sb_loaded = 0; |
3345 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3346 | atomic_set(&rdev->nr_pending, 0); |
3347 | atomic_set(&rdev->read_errors, 0); | |
3348 | atomic_set(&rdev->corrected_errors, 0); | |
3349 | ||
3350 | INIT_LIST_HEAD(&rdev->same_set); | |
3351 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3352 | |
3353 | /* Add space to store bad block list. | |
3354 | * This reserves the space even on arrays where it cannot | |
3355 | * be used - I wonder if that matters | |
3356 | */ | |
fc974ee2 | 3357 | return badblocks_init(&rdev->badblocks, 0); |
e8bb9a83 N |
3358 | } |
3359 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3360 | /* |
3361 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3362 | * | |
3363 | * mark the device faulty if: | |
3364 | * | |
3365 | * - the device is nonexistent (zero size) | |
3366 | * - the device has no valid superblock | |
3367 | * | |
3368 | * a faulty rdev _never_ has rdev->sb set. | |
3369 | */ | |
3cb03002 | 3370 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3371 | { |
3372 | char b[BDEVNAME_SIZE]; | |
3373 | int err; | |
3cb03002 | 3374 | struct md_rdev *rdev; |
1da177e4 LT |
3375 | sector_t size; |
3376 | ||
9ffae0cf | 3377 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
9d48739e | 3378 | if (!rdev) |
1da177e4 | 3379 | return ERR_PTR(-ENOMEM); |
1da177e4 | 3380 | |
2230dfe4 N |
3381 | err = md_rdev_init(rdev); |
3382 | if (err) | |
3383 | goto abort_free; | |
3384 | err = alloc_disk_sb(rdev); | |
3385 | if (err) | |
1da177e4 LT |
3386 | goto abort_free; |
3387 | ||
c5d79adb | 3388 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3389 | if (err) |
3390 | goto abort_free; | |
3391 | ||
f9cb074b | 3392 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3393 | |
77304d2a | 3394 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3395 | if (!size) { |
9d48739e | 3396 | pr_warn("md: %s has zero or unknown size, marking faulty!\n", |
1da177e4 LT |
3397 | bdevname(rdev->bdev,b)); |
3398 | err = -EINVAL; | |
3399 | goto abort_free; | |
3400 | } | |
3401 | ||
3402 | if (super_format >= 0) { | |
3403 | err = super_types[super_format]. | |
3404 | load_super(rdev, NULL, super_minor); | |
3405 | if (err == -EINVAL) { | |
9d48739e | 3406 | pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n", |
df968c4e | 3407 | bdevname(rdev->bdev,b), |
9d48739e | 3408 | super_format, super_minor); |
1da177e4 LT |
3409 | goto abort_free; |
3410 | } | |
3411 | if (err < 0) { | |
9d48739e | 3412 | pr_warn("md: could not read %s's sb, not importing!\n", |
1da177e4 LT |
3413 | bdevname(rdev->bdev,b)); |
3414 | goto abort_free; | |
3415 | } | |
3416 | } | |
6bfe0b49 | 3417 | |
1da177e4 LT |
3418 | return rdev; |
3419 | ||
3420 | abort_free: | |
2699b672 N |
3421 | if (rdev->bdev) |
3422 | unlock_rdev(rdev); | |
545c8795 | 3423 | md_rdev_clear(rdev); |
1da177e4 LT |
3424 | kfree(rdev); |
3425 | return ERR_PTR(err); | |
3426 | } | |
3427 | ||
3428 | /* | |
3429 | * Check a full RAID array for plausibility | |
3430 | */ | |
3431 | ||
f72ffdd6 | 3432 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3433 | { |
3434 | int i; | |
3cb03002 | 3435 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3436 | char b[BDEVNAME_SIZE]; |
3437 | ||
3438 | freshest = NULL; | |
dafb20fa | 3439 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3440 | switch (super_types[mddev->major_version]. |
3441 | load_super(rdev, freshest, mddev->minor_version)) { | |
3442 | case 1: | |
3443 | freshest = rdev; | |
3444 | break; | |
3445 | case 0: | |
3446 | break; | |
3447 | default: | |
9d48739e | 3448 | pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n", |
1da177e4 | 3449 | bdevname(rdev->bdev,b)); |
fb56dfef | 3450 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3451 | } |
3452 | ||
1da177e4 LT |
3453 | super_types[mddev->major_version]. |
3454 | validate_super(mddev, freshest); | |
3455 | ||
3456 | i = 0; | |
dafb20fa | 3457 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3458 | if (mddev->max_disks && |
3459 | (rdev->desc_nr >= mddev->max_disks || | |
3460 | i > mddev->max_disks)) { | |
9d48739e N |
3461 | pr_warn("md: %s: %s: only %d devices permitted\n", |
3462 | mdname(mddev), bdevname(rdev->bdev, b), | |
3463 | mddev->max_disks); | |
fb56dfef | 3464 | md_kick_rdev_from_array(rdev); |
de01dfad N |
3465 | continue; |
3466 | } | |
1aee41f6 | 3467 | if (rdev != freshest) { |
1da177e4 LT |
3468 | if (super_types[mddev->major_version]. |
3469 | validate_super(mddev, rdev)) { | |
9d48739e | 3470 | pr_warn("md: kicking non-fresh %s from array!\n", |
1da177e4 | 3471 | bdevname(rdev->bdev,b)); |
fb56dfef | 3472 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3473 | continue; |
3474 | } | |
1aee41f6 | 3475 | } |
1da177e4 LT |
3476 | if (mddev->level == LEVEL_MULTIPATH) { |
3477 | rdev->desc_nr = i++; | |
3478 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3479 | set_bit(In_sync, &rdev->flags); |
f2076e7d SL |
3480 | } else if (rdev->raid_disk >= |
3481 | (mddev->raid_disks - min(0, mddev->delta_disks)) && | |
3482 | !test_bit(Journal, &rdev->flags)) { | |
a778b73f N |
3483 | rdev->raid_disk = -1; |
3484 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3485 | } |
3486 | } | |
1da177e4 LT |
3487 | } |
3488 | ||
72e02075 N |
3489 | /* Read a fixed-point number. |
3490 | * Numbers in sysfs attributes should be in "standard" units where | |
3491 | * possible, so time should be in seconds. | |
f72ffdd6 | 3492 | * However we internally use a a much smaller unit such as |
72e02075 N |
3493 | * milliseconds or jiffies. |
3494 | * This function takes a decimal number with a possible fractional | |
3495 | * component, and produces an integer which is the result of | |
3496 | * multiplying that number by 10^'scale'. | |
3497 | * all without any floating-point arithmetic. | |
3498 | */ | |
3499 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3500 | { | |
3501 | unsigned long result = 0; | |
3502 | long decimals = -1; | |
3503 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3504 | if (*cp == '.') | |
3505 | decimals = 0; | |
3506 | else if (decimals < scale) { | |
3507 | unsigned int value; | |
3508 | value = *cp - '0'; | |
3509 | result = result * 10 + value; | |
3510 | if (decimals >= 0) | |
3511 | decimals++; | |
3512 | } | |
3513 | cp++; | |
3514 | } | |
3515 | if (*cp == '\n') | |
3516 | cp++; | |
3517 | if (*cp) | |
3518 | return -EINVAL; | |
3519 | if (decimals < 0) | |
3520 | decimals = 0; | |
3521 | while (decimals < scale) { | |
3522 | result *= 10; | |
3523 | decimals ++; | |
3524 | } | |
3525 | *res = result; | |
3526 | return 0; | |
3527 | } | |
3528 | ||
16f17b39 | 3529 | static ssize_t |
fd01b88c | 3530 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3531 | { |
3532 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3533 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3534 | } | |
3535 | static ssize_t | |
fd01b88c | 3536 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3537 | { |
16f17b39 | 3538 | unsigned long msec; |
97ce0a7f | 3539 | |
28c1b9fd | 3540 | if (mddev_is_clustered(mddev)) { |
9d48739e | 3541 | pr_warn("md: Safemode is disabled for clustered mode\n"); |
28c1b9fd GR |
3542 | return -EINVAL; |
3543 | } | |
3544 | ||
72e02075 | 3545 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3546 | return -EINVAL; |
16f17b39 N |
3547 | if (msec == 0) |
3548 | mddev->safemode_delay = 0; | |
3549 | else { | |
19052c0e | 3550 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3551 | unsigned long new_delay = (msec*HZ)/1000; |
3552 | ||
3553 | if (new_delay == 0) | |
3554 | new_delay = 1; | |
3555 | mddev->safemode_delay = new_delay; | |
3556 | if (new_delay < old_delay || old_delay == 0) | |
3557 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3558 | } |
3559 | return len; | |
3560 | } | |
3561 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3562 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3563 | |
eae1701f | 3564 | static ssize_t |
fd01b88c | 3565 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3566 | { |
36d091f4 N |
3567 | struct md_personality *p; |
3568 | int ret; | |
3569 | spin_lock(&mddev->lock); | |
3570 | p = mddev->pers; | |
d9d166c2 | 3571 | if (p) |
36d091f4 | 3572 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3573 | else if (mddev->clevel[0]) |
36d091f4 | 3574 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3575 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3576 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3577 | else |
36d091f4 N |
3578 | ret = 0; |
3579 | spin_unlock(&mddev->lock); | |
3580 | return ret; | |
eae1701f N |
3581 | } |
3582 | ||
d9d166c2 | 3583 | static ssize_t |
fd01b88c | 3584 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3585 | { |
f2859af6 | 3586 | char clevel[16]; |
6791875e N |
3587 | ssize_t rv; |
3588 | size_t slen = len; | |
db721d32 | 3589 | struct md_personality *pers, *oldpers; |
f2859af6 | 3590 | long level; |
db721d32 | 3591 | void *priv, *oldpriv; |
3cb03002 | 3592 | struct md_rdev *rdev; |
245f46c2 | 3593 | |
6791875e N |
3594 | if (slen == 0 || slen >= sizeof(clevel)) |
3595 | return -EINVAL; | |
3596 | ||
3597 | rv = mddev_lock(mddev); | |
3598 | if (rv) | |
3599 | return rv; | |
3600 | ||
245f46c2 | 3601 | if (mddev->pers == NULL) { |
6791875e N |
3602 | strncpy(mddev->clevel, buf, slen); |
3603 | if (mddev->clevel[slen-1] == '\n') | |
3604 | slen--; | |
3605 | mddev->clevel[slen] = 0; | |
245f46c2 | 3606 | mddev->level = LEVEL_NONE; |
6791875e N |
3607 | rv = len; |
3608 | goto out_unlock; | |
245f46c2 | 3609 | } |
6791875e | 3610 | rv = -EROFS; |
bd8839e0 | 3611 | if (mddev->ro) |
6791875e | 3612 | goto out_unlock; |
245f46c2 N |
3613 | |
3614 | /* request to change the personality. Need to ensure: | |
3615 | * - array is not engaged in resync/recovery/reshape | |
3616 | * - old personality can be suspended | |
3617 | * - new personality will access other array. | |
3618 | */ | |
3619 | ||
6791875e | 3620 | rv = -EBUSY; |
bb4f1e9d | 3621 | if (mddev->sync_thread || |
f851b60d | 3622 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3623 | mddev->reshape_position != MaxSector || |
3624 | mddev->sysfs_active) | |
6791875e | 3625 | goto out_unlock; |
245f46c2 | 3626 | |
6791875e | 3627 | rv = -EINVAL; |
245f46c2 | 3628 | if (!mddev->pers->quiesce) { |
9d48739e N |
3629 | pr_warn("md: %s: %s does not support online personality change\n", |
3630 | mdname(mddev), mddev->pers->name); | |
6791875e | 3631 | goto out_unlock; |
245f46c2 N |
3632 | } |
3633 | ||
3634 | /* Now find the new personality */ | |
6791875e N |
3635 | strncpy(clevel, buf, slen); |
3636 | if (clevel[slen-1] == '\n') | |
3637 | slen--; | |
3638 | clevel[slen] = 0; | |
b29bebd6 | 3639 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3640 | level = LEVEL_NONE; |
245f46c2 | 3641 | |
f2859af6 DW |
3642 | if (request_module("md-%s", clevel) != 0) |
3643 | request_module("md-level-%s", clevel); | |
245f46c2 | 3644 | spin_lock(&pers_lock); |
f2859af6 | 3645 | pers = find_pers(level, clevel); |
245f46c2 N |
3646 | if (!pers || !try_module_get(pers->owner)) { |
3647 | spin_unlock(&pers_lock); | |
9d48739e | 3648 | pr_warn("md: personality %s not loaded\n", clevel); |
6791875e N |
3649 | rv = -EINVAL; |
3650 | goto out_unlock; | |
245f46c2 N |
3651 | } |
3652 | spin_unlock(&pers_lock); | |
3653 | ||
3654 | if (pers == mddev->pers) { | |
3655 | /* Nothing to do! */ | |
3656 | module_put(pers->owner); | |
6791875e N |
3657 | rv = len; |
3658 | goto out_unlock; | |
245f46c2 N |
3659 | } |
3660 | if (!pers->takeover) { | |
3661 | module_put(pers->owner); | |
9d48739e N |
3662 | pr_warn("md: %s: %s does not support personality takeover\n", |
3663 | mdname(mddev), clevel); | |
6791875e N |
3664 | rv = -EINVAL; |
3665 | goto out_unlock; | |
245f46c2 N |
3666 | } |
3667 | ||
dafb20fa | 3668 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3669 | rdev->new_raid_disk = rdev->raid_disk; |
3670 | ||
245f46c2 N |
3671 | /* ->takeover must set new_* and/or delta_disks |
3672 | * if it succeeds, and may set them when it fails. | |
3673 | */ | |
3674 | priv = pers->takeover(mddev); | |
3675 | if (IS_ERR(priv)) { | |
3676 | mddev->new_level = mddev->level; | |
3677 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3678 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3679 | mddev->raid_disks -= mddev->delta_disks; |
3680 | mddev->delta_disks = 0; | |
2c810cdd | 3681 | mddev->reshape_backwards = 0; |
245f46c2 | 3682 | module_put(pers->owner); |
9d48739e N |
3683 | pr_warn("md: %s: %s would not accept array\n", |
3684 | mdname(mddev), clevel); | |
6791875e N |
3685 | rv = PTR_ERR(priv); |
3686 | goto out_unlock; | |
245f46c2 N |
3687 | } |
3688 | ||
3689 | /* Looks like we have a winner */ | |
3690 | mddev_suspend(mddev); | |
5aa61f42 | 3691 | mddev_detach(mddev); |
36d091f4 N |
3692 | |
3693 | spin_lock(&mddev->lock); | |
db721d32 N |
3694 | oldpers = mddev->pers; |
3695 | oldpriv = mddev->private; | |
3696 | mddev->pers = pers; | |
3697 | mddev->private = priv; | |
3698 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3699 | mddev->level = mddev->new_level; | |
3700 | mddev->layout = mddev->new_layout; | |
3701 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3702 | mddev->delta_disks = 0; | |
3703 | mddev->reshape_backwards = 0; | |
3704 | mddev->degraded = 0; | |
36d091f4 | 3705 | spin_unlock(&mddev->lock); |
db721d32 N |
3706 | |
3707 | if (oldpers->sync_request == NULL && | |
3708 | mddev->external) { | |
3709 | /* We are converting from a no-redundancy array | |
3710 | * to a redundancy array and metadata is managed | |
3711 | * externally so we need to be sure that writes | |
3712 | * won't block due to a need to transition | |
3713 | * clean->dirty | |
3714 | * until external management is started. | |
3715 | */ | |
3716 | mddev->in_sync = 0; | |
3717 | mddev->safemode_delay = 0; | |
3718 | mddev->safemode = 0; | |
3719 | } | |
f72ffdd6 | 3720 | |
db721d32 N |
3721 | oldpers->free(mddev, oldpriv); |
3722 | ||
3723 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3724 | pers->sync_request != NULL) { |
3725 | /* need to add the md_redundancy_group */ | |
3726 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
3727 | pr_warn("md: cannot register extra attributes for %s\n", |
3728 | mdname(mddev)); | |
388975cc | 3729 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3730 | } |
db721d32 | 3731 | if (oldpers->sync_request != NULL && |
a64c876f N |
3732 | pers->sync_request == NULL) { |
3733 | /* need to remove the md_redundancy_group */ | |
3734 | if (mddev->to_remove == NULL) | |
3735 | mddev->to_remove = &md_redundancy_group; | |
3736 | } | |
3737 | ||
4cb9da7d AO |
3738 | module_put(oldpers->owner); |
3739 | ||
dafb20fa | 3740 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3741 | if (rdev->raid_disk < 0) |
3742 | continue; | |
bf2cb0da | 3743 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3744 | rdev->new_raid_disk = -1; |
3745 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3746 | continue; | |
36fad858 | 3747 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3748 | } |
dafb20fa | 3749 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3750 | if (rdev->raid_disk < 0) |
3751 | continue; | |
3752 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3753 | continue; | |
3754 | rdev->raid_disk = rdev->new_raid_disk; | |
3755 | if (rdev->raid_disk < 0) | |
3a981b03 | 3756 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3757 | else { |
36fad858 | 3758 | if (sysfs_link_rdev(mddev, rdev)) |
9d48739e N |
3759 | pr_warn("md: cannot register rd%d for %s after level change\n", |
3760 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3761 | } |
e93f68a1 N |
3762 | } |
3763 | ||
db721d32 | 3764 | if (pers->sync_request == NULL) { |
9af204cf TM |
3765 | /* this is now an array without redundancy, so |
3766 | * it must always be in_sync | |
3767 | */ | |
3768 | mddev->in_sync = 1; | |
3769 | del_timer_sync(&mddev->safemode_timer); | |
3770 | } | |
02e5f5c0 | 3771 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3772 | pers->run(mddev); |
2953079c | 3773 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
47525e59 | 3774 | mddev_resume(mddev); |
830778a1 N |
3775 | if (!mddev->thread) |
3776 | md_update_sb(mddev, 1); | |
5cac7861 | 3777 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3778 | md_new_event(mddev); |
6791875e N |
3779 | rv = len; |
3780 | out_unlock: | |
3781 | mddev_unlock(mddev); | |
d9d166c2 N |
3782 | return rv; |
3783 | } | |
3784 | ||
3785 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3786 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3787 | |
d4dbd025 | 3788 | static ssize_t |
fd01b88c | 3789 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3790 | { |
3791 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3792 | if (mddev->reshape_position != MaxSector && |
3793 | mddev->layout != mddev->new_layout) | |
3794 | return sprintf(page, "%d (%d)\n", | |
3795 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3796 | return sprintf(page, "%d\n", mddev->layout); |
3797 | } | |
3798 | ||
3799 | static ssize_t | |
fd01b88c | 3800 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 | 3801 | { |
4c9309c0 | 3802 | unsigned int n; |
6791875e | 3803 | int err; |
d4dbd025 | 3804 | |
4c9309c0 AD |
3805 | err = kstrtouint(buf, 10, &n); |
3806 | if (err < 0) | |
3807 | return err; | |
6791875e N |
3808 | err = mddev_lock(mddev); |
3809 | if (err) | |
3810 | return err; | |
d4dbd025 | 3811 | |
b3546035 | 3812 | if (mddev->pers) { |
50ac168a | 3813 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3814 | err = -EBUSY; |
3815 | else if (mddev->ro) | |
3816 | err = -EROFS; | |
3817 | else { | |
3818 | mddev->new_layout = n; | |
3819 | err = mddev->pers->check_reshape(mddev); | |
3820 | if (err) | |
3821 | mddev->new_layout = mddev->layout; | |
597a711b | 3822 | } |
b3546035 | 3823 | } else { |
08a02ecd | 3824 | mddev->new_layout = n; |
b3546035 N |
3825 | if (mddev->reshape_position == MaxSector) |
3826 | mddev->layout = n; | |
3827 | } | |
6791875e N |
3828 | mddev_unlock(mddev); |
3829 | return err ?: len; | |
d4dbd025 N |
3830 | } |
3831 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3832 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3833 | |
eae1701f | 3834 | static ssize_t |
fd01b88c | 3835 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3836 | { |
bb636547 N |
3837 | if (mddev->raid_disks == 0) |
3838 | return 0; | |
08a02ecd N |
3839 | if (mddev->reshape_position != MaxSector && |
3840 | mddev->delta_disks != 0) | |
3841 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3842 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3843 | return sprintf(page, "%d\n", mddev->raid_disks); |
3844 | } | |
3845 | ||
fd01b88c | 3846 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3847 | |
3848 | static ssize_t | |
fd01b88c | 3849 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3850 | { |
4c9309c0 | 3851 | unsigned int n; |
6791875e | 3852 | int err; |
da943b99 | 3853 | |
4c9309c0 AD |
3854 | err = kstrtouint(buf, 10, &n); |
3855 | if (err < 0) | |
3856 | return err; | |
da943b99 | 3857 | |
6791875e N |
3858 | err = mddev_lock(mddev); |
3859 | if (err) | |
3860 | return err; | |
da943b99 | 3861 | if (mddev->pers) |
6791875e | 3862 | err = update_raid_disks(mddev, n); |
08a02ecd | 3863 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3864 | struct md_rdev *rdev; |
08a02ecd | 3865 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3866 | |
6791875e | 3867 | err = -EINVAL; |
c6563a8c N |
3868 | rdev_for_each(rdev, mddev) { |
3869 | if (olddisks < n && | |
3870 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3871 | goto out_unlock; |
c6563a8c N |
3872 | if (olddisks > n && |
3873 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3874 | goto out_unlock; |
c6563a8c | 3875 | } |
6791875e | 3876 | err = 0; |
08a02ecd N |
3877 | mddev->delta_disks = n - olddisks; |
3878 | mddev->raid_disks = n; | |
2c810cdd | 3879 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3880 | } else |
da943b99 | 3881 | mddev->raid_disks = n; |
6791875e N |
3882 | out_unlock: |
3883 | mddev_unlock(mddev); | |
3884 | return err ? err : len; | |
da943b99 N |
3885 | } |
3886 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3887 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3888 | |
3b34380a | 3889 | static ssize_t |
fd01b88c | 3890 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3891 | { |
08a02ecd | 3892 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3893 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3894 | return sprintf(page, "%d (%d)\n", | |
3895 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3896 | mddev->chunk_sectors << 9); |
3897 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3898 | } |
3899 | ||
3900 | static ssize_t | |
fd01b88c | 3901 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3902 | { |
4c9309c0 | 3903 | unsigned long n; |
6791875e | 3904 | int err; |
3b34380a | 3905 | |
4c9309c0 AD |
3906 | err = kstrtoul(buf, 10, &n); |
3907 | if (err < 0) | |
3908 | return err; | |
3b34380a | 3909 | |
6791875e N |
3910 | err = mddev_lock(mddev); |
3911 | if (err) | |
3912 | return err; | |
b3546035 | 3913 | if (mddev->pers) { |
50ac168a | 3914 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3915 | err = -EBUSY; |
3916 | else if (mddev->ro) | |
3917 | err = -EROFS; | |
3918 | else { | |
3919 | mddev->new_chunk_sectors = n >> 9; | |
3920 | err = mddev->pers->check_reshape(mddev); | |
3921 | if (err) | |
3922 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 3923 | } |
b3546035 | 3924 | } else { |
664e7c41 | 3925 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3926 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3927 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3928 | } |
6791875e N |
3929 | mddev_unlock(mddev); |
3930 | return err ?: len; | |
3b34380a N |
3931 | } |
3932 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3933 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3934 | |
a94213b1 | 3935 | static ssize_t |
fd01b88c | 3936 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3937 | { |
d1a7c503 N |
3938 | if (mddev->recovery_cp == MaxSector) |
3939 | return sprintf(page, "none\n"); | |
a94213b1 N |
3940 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3941 | } | |
3942 | ||
3943 | static ssize_t | |
fd01b88c | 3944 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3945 | { |
4c9309c0 | 3946 | unsigned long long n; |
6791875e | 3947 | int err; |
4c9309c0 AD |
3948 | |
3949 | if (cmd_match(buf, "none")) | |
3950 | n = MaxSector; | |
3951 | else { | |
3952 | err = kstrtoull(buf, 10, &n); | |
3953 | if (err < 0) | |
3954 | return err; | |
3955 | if (n != (sector_t)n) | |
3956 | return -EINVAL; | |
3957 | } | |
a94213b1 | 3958 | |
6791875e N |
3959 | err = mddev_lock(mddev); |
3960 | if (err) | |
3961 | return err; | |
b098636c | 3962 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e | 3963 | err = -EBUSY; |
a94213b1 | 3964 | |
6791875e N |
3965 | if (!err) { |
3966 | mddev->recovery_cp = n; | |
3967 | if (mddev->pers) | |
2953079c | 3968 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
6791875e N |
3969 | } |
3970 | mddev_unlock(mddev); | |
3971 | return err ?: len; | |
a94213b1 N |
3972 | } |
3973 | static struct md_sysfs_entry md_resync_start = | |
750f199e N |
3974 | __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR, |
3975 | resync_start_show, resync_start_store); | |
a94213b1 | 3976 | |
9e653b63 N |
3977 | /* |
3978 | * The array state can be: | |
3979 | * | |
3980 | * clear | |
3981 | * No devices, no size, no level | |
3982 | * Equivalent to STOP_ARRAY ioctl | |
3983 | * inactive | |
3984 | * May have some settings, but array is not active | |
3985 | * all IO results in error | |
3986 | * When written, doesn't tear down array, but just stops it | |
3987 | * suspended (not supported yet) | |
3988 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3989 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3990 | * readonly |
3991 | * no resync can happen. no superblocks get written. | |
3992 | * write requests fail | |
3993 | * read-auto | |
3994 | * like readonly, but behaves like 'clean' on a write request. | |
3995 | * | |
3996 | * clean - no pending writes, but otherwise active. | |
3997 | * When written to inactive array, starts without resync | |
3998 | * If a write request arrives then | |
3999 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
4000 | * if not known, block and switch to write-pending | |
4001 | * If written to an active array that has pending writes, then fails. | |
4002 | * active | |
4003 | * fully active: IO and resync can be happening. | |
4004 | * When written to inactive array, starts with resync | |
4005 | * | |
4006 | * write-pending | |
4007 | * clean, but writes are blocked waiting for 'active' to be written. | |
4008 | * | |
4009 | * active-idle | |
4010 | * like active, but no writes have been seen for a while (100msec). | |
4011 | * | |
4012 | */ | |
4013 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
4014 | write_pending, active_idle, bad_word}; | |
05381954 | 4015 | static char *array_states[] = { |
9e653b63 N |
4016 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
4017 | "write-pending", "active-idle", NULL }; | |
4018 | ||
4019 | static int match_word(const char *word, char **list) | |
4020 | { | |
4021 | int n; | |
4022 | for (n=0; list[n]; n++) | |
4023 | if (cmd_match(word, list[n])) | |
4024 | break; | |
4025 | return n; | |
4026 | } | |
4027 | ||
4028 | static ssize_t | |
fd01b88c | 4029 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
4030 | { |
4031 | enum array_state st = inactive; | |
4032 | ||
4033 | if (mddev->pers) | |
4034 | switch(mddev->ro) { | |
4035 | case 1: | |
4036 | st = readonly; | |
4037 | break; | |
4038 | case 2: | |
4039 | st = read_auto; | |
4040 | break; | |
4041 | case 0: | |
55cc39f3 | 4042 | spin_lock(&mddev->lock); |
2953079c | 4043 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
e691063a | 4044 | st = write_pending; |
16f88949 TM |
4045 | else if (mddev->in_sync) |
4046 | st = clean; | |
9e653b63 N |
4047 | else if (mddev->safemode) |
4048 | st = active_idle; | |
4049 | else | |
4050 | st = active; | |
55cc39f3 | 4051 | spin_unlock(&mddev->lock); |
9e653b63 N |
4052 | } |
4053 | else { | |
4054 | if (list_empty(&mddev->disks) && | |
4055 | mddev->raid_disks == 0 && | |
58c0fed4 | 4056 | mddev->dev_sectors == 0) |
9e653b63 N |
4057 | st = clear; |
4058 | else | |
4059 | st = inactive; | |
4060 | } | |
4061 | return sprintf(page, "%s\n", array_states[st]); | |
4062 | } | |
4063 | ||
f72ffdd6 N |
4064 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
4065 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
4066 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 4067 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
4068 | |
4069 | static ssize_t | |
fd01b88c | 4070 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 4071 | { |
6497709b | 4072 | int err = 0; |
9e653b63 | 4073 | enum array_state st = match_word(buf, array_states); |
6791875e N |
4074 | |
4075 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
4076 | /* don't take reconfig_mutex when toggling between | |
4077 | * clean and active | |
4078 | */ | |
4079 | spin_lock(&mddev->lock); | |
4080 | if (st == active) { | |
4081 | restart_array(mddev); | |
2953079c | 4082 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
91a6c4ad | 4083 | md_wakeup_thread(mddev->thread); |
6791875e | 4084 | wake_up(&mddev->sb_wait); |
6791875e N |
4085 | } else /* st == clean */ { |
4086 | restart_array(mddev); | |
6497709b | 4087 | if (!set_in_sync(mddev)) |
6791875e N |
4088 | err = -EBUSY; |
4089 | } | |
573275b5 TM |
4090 | if (!err) |
4091 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
6791875e | 4092 | spin_unlock(&mddev->lock); |
c008f1d3 | 4093 | return err ?: len; |
6791875e N |
4094 | } |
4095 | err = mddev_lock(mddev); | |
4096 | if (err) | |
4097 | return err; | |
4098 | err = -EINVAL; | |
9e653b63 N |
4099 | switch(st) { |
4100 | case bad_word: | |
4101 | break; | |
4102 | case clear: | |
4103 | /* stopping an active array */ | |
a05b7ea0 | 4104 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
4105 | break; |
4106 | case inactive: | |
4107 | /* stopping an active array */ | |
90cf195d | 4108 | if (mddev->pers) |
a05b7ea0 | 4109 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 4110 | else |
e691063a | 4111 | err = 0; /* already inactive */ |
9e653b63 N |
4112 | break; |
4113 | case suspended: | |
4114 | break; /* not supported yet */ | |
4115 | case readonly: | |
4116 | if (mddev->pers) | |
a05b7ea0 | 4117 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
4118 | else { |
4119 | mddev->ro = 1; | |
648b629e | 4120 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
4121 | err = do_md_run(mddev); |
4122 | } | |
4123 | break; | |
4124 | case read_auto: | |
9e653b63 | 4125 | if (mddev->pers) { |
80268ee9 | 4126 | if (mddev->ro == 0) |
a05b7ea0 | 4127 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 4128 | else if (mddev->ro == 1) |
648b629e N |
4129 | err = restart_array(mddev); |
4130 | if (err == 0) { | |
4131 | mddev->ro = 2; | |
4132 | set_disk_ro(mddev->gendisk, 0); | |
4133 | } | |
9e653b63 N |
4134 | } else { |
4135 | mddev->ro = 2; | |
4136 | err = do_md_run(mddev); | |
4137 | } | |
4138 | break; | |
4139 | case clean: | |
4140 | if (mddev->pers) { | |
339421de SL |
4141 | err = restart_array(mddev); |
4142 | if (err) | |
4143 | break; | |
85572d7c | 4144 | spin_lock(&mddev->lock); |
6497709b | 4145 | if (!set_in_sync(mddev)) |
e691063a | 4146 | err = -EBUSY; |
85572d7c | 4147 | spin_unlock(&mddev->lock); |
5bf29597 N |
4148 | } else |
4149 | err = -EINVAL; | |
9e653b63 N |
4150 | break; |
4151 | case active: | |
4152 | if (mddev->pers) { | |
339421de SL |
4153 | err = restart_array(mddev); |
4154 | if (err) | |
4155 | break; | |
2953079c | 4156 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
9e653b63 N |
4157 | wake_up(&mddev->sb_wait); |
4158 | err = 0; | |
4159 | } else { | |
4160 | mddev->ro = 0; | |
648b629e | 4161 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
4162 | err = do_md_run(mddev); |
4163 | } | |
4164 | break; | |
4165 | case write_pending: | |
4166 | case active_idle: | |
4167 | /* these cannot be set */ | |
4168 | break; | |
4169 | } | |
6791875e N |
4170 | |
4171 | if (!err) { | |
1d23f178 N |
4172 | if (mddev->hold_active == UNTIL_IOCTL) |
4173 | mddev->hold_active = 0; | |
00bcb4ac | 4174 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4175 | } |
6791875e N |
4176 | mddev_unlock(mddev); |
4177 | return err ?: len; | |
9e653b63 | 4178 | } |
80ca3a44 | 4179 | static struct md_sysfs_entry md_array_state = |
750f199e | 4180 | __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); |
9e653b63 | 4181 | |
1e50915f | 4182 | static ssize_t |
fd01b88c | 4183 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
4184 | return sprintf(page, "%d\n", |
4185 | atomic_read(&mddev->max_corr_read_errors)); | |
4186 | } | |
4187 | ||
4188 | static ssize_t | |
fd01b88c | 4189 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f | 4190 | { |
4c9309c0 AD |
4191 | unsigned int n; |
4192 | int rv; | |
1e50915f | 4193 | |
4c9309c0 AD |
4194 | rv = kstrtouint(buf, 10, &n); |
4195 | if (rv < 0) | |
4196 | return rv; | |
4197 | atomic_set(&mddev->max_corr_read_errors, n); | |
4198 | return len; | |
1e50915f RB |
4199 | } |
4200 | ||
4201 | static struct md_sysfs_entry max_corr_read_errors = | |
4202 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
4203 | max_corrected_read_errors_store); | |
4204 | ||
6d7ff738 | 4205 | static ssize_t |
fd01b88c | 4206 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
4207 | { |
4208 | return -EINVAL; | |
4209 | } | |
4210 | ||
4211 | static ssize_t | |
fd01b88c | 4212 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
4213 | { |
4214 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
4215 | /* The new device is added to the array. | |
4216 | * If the array has a persistent superblock, we read the | |
4217 | * superblock to initialise info and check validity. | |
4218 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
4219 | * which mainly checks size. | |
4220 | */ | |
4221 | char *e; | |
4222 | int major = simple_strtoul(buf, &e, 10); | |
4223 | int minor; | |
4224 | dev_t dev; | |
3cb03002 | 4225 | struct md_rdev *rdev; |
6d7ff738 N |
4226 | int err; |
4227 | ||
4228 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
4229 | return -EINVAL; | |
4230 | minor = simple_strtoul(e+1, &e, 10); | |
4231 | if (*e && *e != '\n') | |
4232 | return -EINVAL; | |
4233 | dev = MKDEV(major, minor); | |
4234 | if (major != MAJOR(dev) || | |
4235 | minor != MINOR(dev)) | |
4236 | return -EOVERFLOW; | |
4237 | ||
6791875e N |
4238 | flush_workqueue(md_misc_wq); |
4239 | ||
4240 | err = mddev_lock(mddev); | |
4241 | if (err) | |
4242 | return err; | |
6d7ff738 N |
4243 | if (mddev->persistent) { |
4244 | rdev = md_import_device(dev, mddev->major_version, | |
4245 | mddev->minor_version); | |
4246 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4247 | struct md_rdev *rdev0 |
4248 | = list_entry(mddev->disks.next, | |
4249 | struct md_rdev, same_set); | |
6d7ff738 N |
4250 | err = super_types[mddev->major_version] |
4251 | .load_super(rdev, rdev0, mddev->minor_version); | |
4252 | if (err < 0) | |
4253 | goto out; | |
4254 | } | |
c5d79adb N |
4255 | } else if (mddev->external) |
4256 | rdev = md_import_device(dev, -2, -1); | |
4257 | else | |
6d7ff738 N |
4258 | rdev = md_import_device(dev, -1, -1); |
4259 | ||
9a8c0fa8 N |
4260 | if (IS_ERR(rdev)) { |
4261 | mddev_unlock(mddev); | |
6d7ff738 | 4262 | return PTR_ERR(rdev); |
9a8c0fa8 | 4263 | } |
6d7ff738 N |
4264 | err = bind_rdev_to_array(rdev, mddev); |
4265 | out: | |
4266 | if (err) | |
4267 | export_rdev(rdev); | |
6791875e | 4268 | mddev_unlock(mddev); |
6d7ff738 N |
4269 | return err ? err : len; |
4270 | } | |
4271 | ||
4272 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4273 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4274 | |
9b1d1dac | 4275 | static ssize_t |
fd01b88c | 4276 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4277 | { |
4278 | char *end; | |
4279 | unsigned long chunk, end_chunk; | |
6791875e | 4280 | int err; |
9b1d1dac | 4281 | |
6791875e N |
4282 | err = mddev_lock(mddev); |
4283 | if (err) | |
4284 | return err; | |
9b1d1dac PC |
4285 | if (!mddev->bitmap) |
4286 | goto out; | |
4287 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4288 | while (*buf) { | |
4289 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4290 | if (buf == end) break; | |
4291 | if (*end == '-') { /* range */ | |
4292 | buf = end + 1; | |
4293 | end_chunk = simple_strtoul(buf, &end, 0); | |
4294 | if (buf == end) break; | |
4295 | } | |
4296 | if (*end && !isspace(*end)) break; | |
4297 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4298 | buf = skip_spaces(end); |
9b1d1dac PC |
4299 | } |
4300 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4301 | out: | |
6791875e | 4302 | mddev_unlock(mddev); |
9b1d1dac PC |
4303 | return len; |
4304 | } | |
4305 | ||
4306 | static struct md_sysfs_entry md_bitmap = | |
4307 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4308 | ||
a35b0d69 | 4309 | static ssize_t |
fd01b88c | 4310 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4311 | { |
58c0fed4 AN |
4312 | return sprintf(page, "%llu\n", |
4313 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4314 | } |
4315 | ||
fd01b88c | 4316 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4317 | |
4318 | static ssize_t | |
fd01b88c | 4319 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4320 | { |
4321 | /* If array is inactive, we can reduce the component size, but | |
4322 | * not increase it (except from 0). | |
4323 | * If array is active, we can try an on-line resize | |
4324 | */ | |
b522adcd DW |
4325 | sector_t sectors; |
4326 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4327 | |
58c0fed4 AN |
4328 | if (err < 0) |
4329 | return err; | |
6791875e N |
4330 | err = mddev_lock(mddev); |
4331 | if (err) | |
4332 | return err; | |
a35b0d69 | 4333 | if (mddev->pers) { |
58c0fed4 | 4334 | err = update_size(mddev, sectors); |
4ba1e788 XN |
4335 | if (err == 0) |
4336 | md_update_sb(mddev, 1); | |
a35b0d69 | 4337 | } else { |
58c0fed4 AN |
4338 | if (mddev->dev_sectors == 0 || |
4339 | mddev->dev_sectors > sectors) | |
4340 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4341 | else |
4342 | err = -ENOSPC; | |
4343 | } | |
6791875e | 4344 | mddev_unlock(mddev); |
a35b0d69 N |
4345 | return err ? err : len; |
4346 | } | |
4347 | ||
4348 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4349 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4350 | |
83f0d77a | 4351 | /* Metadata version. |
e691063a N |
4352 | * This is one of |
4353 | * 'none' for arrays with no metadata (good luck...) | |
4354 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4355 | * or N.M for internally known formats |
4356 | */ | |
4357 | static ssize_t | |
fd01b88c | 4358 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4359 | { |
4360 | if (mddev->persistent) | |
4361 | return sprintf(page, "%d.%d\n", | |
4362 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4363 | else if (mddev->external) |
4364 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4365 | else |
4366 | return sprintf(page, "none\n"); | |
4367 | } | |
4368 | ||
4369 | static ssize_t | |
fd01b88c | 4370 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4371 | { |
4372 | int major, minor; | |
4373 | char *e; | |
6791875e | 4374 | int err; |
ea43ddd8 N |
4375 | /* Changing the details of 'external' metadata is |
4376 | * always permitted. Otherwise there must be | |
4377 | * no devices attached to the array. | |
4378 | */ | |
6791875e N |
4379 | |
4380 | err = mddev_lock(mddev); | |
4381 | if (err) | |
4382 | return err; | |
4383 | err = -EBUSY; | |
ea43ddd8 N |
4384 | if (mddev->external && strncmp(buf, "external:", 9) == 0) |
4385 | ; | |
4386 | else if (!list_empty(&mddev->disks)) | |
6791875e | 4387 | goto out_unlock; |
8bb93aac | 4388 | |
6791875e | 4389 | err = 0; |
8bb93aac N |
4390 | if (cmd_match(buf, "none")) { |
4391 | mddev->persistent = 0; | |
e691063a N |
4392 | mddev->external = 0; |
4393 | mddev->major_version = 0; | |
4394 | mddev->minor_version = 90; | |
6791875e | 4395 | goto out_unlock; |
e691063a N |
4396 | } |
4397 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4398 | size_t namelen = len-9; |
e691063a N |
4399 | if (namelen >= sizeof(mddev->metadata_type)) |
4400 | namelen = sizeof(mddev->metadata_type)-1; | |
4401 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4402 | mddev->metadata_type[namelen] = 0; | |
4403 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4404 | mddev->metadata_type[--namelen] = 0; | |
4405 | mddev->persistent = 0; | |
4406 | mddev->external = 1; | |
8bb93aac N |
4407 | mddev->major_version = 0; |
4408 | mddev->minor_version = 90; | |
6791875e | 4409 | goto out_unlock; |
8bb93aac N |
4410 | } |
4411 | major = simple_strtoul(buf, &e, 10); | |
6791875e | 4412 | err = -EINVAL; |
8bb93aac | 4413 | if (e==buf || *e != '.') |
6791875e | 4414 | goto out_unlock; |
8bb93aac N |
4415 | buf = e+1; |
4416 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4417 | if (e==buf || (*e && *e != '\n') ) |
6791875e N |
4418 | goto out_unlock; |
4419 | err = -ENOENT; | |
50511da3 | 4420 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
6791875e | 4421 | goto out_unlock; |
8bb93aac N |
4422 | mddev->major_version = major; |
4423 | mddev->minor_version = minor; | |
4424 | mddev->persistent = 1; | |
e691063a | 4425 | mddev->external = 0; |
6791875e N |
4426 | err = 0; |
4427 | out_unlock: | |
4428 | mddev_unlock(mddev); | |
4429 | return err ?: len; | |
8bb93aac N |
4430 | } |
4431 | ||
4432 | static struct md_sysfs_entry md_metadata = | |
750f199e | 4433 | __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4434 | |
24dd469d | 4435 | static ssize_t |
fd01b88c | 4436 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4437 | { |
7eec314d | 4438 | char *type = "idle"; |
b7b17c9b N |
4439 | unsigned long recovery = mddev->recovery; |
4440 | if (test_bit(MD_RECOVERY_FROZEN, &recovery)) | |
b6a9ce68 | 4441 | type = "frozen"; |
b7b17c9b N |
4442 | else if (test_bit(MD_RECOVERY_RUNNING, &recovery) || |
4443 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) { | |
4444 | if (test_bit(MD_RECOVERY_RESHAPE, &recovery)) | |
ccfcc3c1 | 4445 | type = "reshape"; |
b7b17c9b N |
4446 | else if (test_bit(MD_RECOVERY_SYNC, &recovery)) { |
4447 | if (!test_bit(MD_RECOVERY_REQUESTED, &recovery)) | |
24dd469d | 4448 | type = "resync"; |
b7b17c9b | 4449 | else if (test_bit(MD_RECOVERY_CHECK, &recovery)) |
24dd469d N |
4450 | type = "check"; |
4451 | else | |
4452 | type = "repair"; | |
b7b17c9b | 4453 | } else if (test_bit(MD_RECOVERY_RECOVER, &recovery)) |
24dd469d | 4454 | type = "recover"; |
985ca973 N |
4455 | else if (mddev->reshape_position != MaxSector) |
4456 | type = "reshape"; | |
24dd469d N |
4457 | } |
4458 | return sprintf(page, "%s\n", type); | |
4459 | } | |
4460 | ||
4461 | static ssize_t | |
fd01b88c | 4462 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4463 | { |
7eec314d N |
4464 | if (!mddev->pers || !mddev->pers->sync_request) |
4465 | return -EINVAL; | |
4466 | ||
b6a9ce68 N |
4467 | |
4468 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
56ccc112 N |
4469 | if (cmd_match(page, "frozen")) |
4470 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4471 | else | |
4472 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
8e8e2518 N |
4473 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
4474 | mddev_lock(mddev) == 0) { | |
4475 | flush_workqueue(md_misc_wq); | |
4476 | if (mddev->sync_thread) { | |
4477 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e | 4478 | md_reap_sync_thread(mddev); |
6791875e | 4479 | } |
8e8e2518 | 4480 | mddev_unlock(mddev); |
7eec314d | 4481 | } |
312045ee | 4482 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
24dd469d | 4483 | return -EBUSY; |
72a23c21 | 4484 | else if (cmd_match(page, "resync")) |
56ccc112 | 4485 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4486 | else if (cmd_match(page, "recover")) { |
56ccc112 | 4487 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4488 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
72a23c21 | 4489 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4490 | int err; |
4491 | if (mddev->pers->start_reshape == NULL) | |
4492 | return -EINVAL; | |
6791875e N |
4493 | err = mddev_lock(mddev); |
4494 | if (!err) { | |
312045ee N |
4495 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4496 | err = -EBUSY; | |
4497 | else { | |
4498 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4499 | err = mddev->pers->start_reshape(mddev); | |
4500 | } | |
6791875e N |
4501 | mddev_unlock(mddev); |
4502 | } | |
16484bf5 N |
4503 | if (err) |
4504 | return err; | |
a99ac971 | 4505 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4506 | } else { |
bce74dac | 4507 | if (cmd_match(page, "check")) |
7eec314d | 4508 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4509 | else if (!cmd_match(page, "repair")) |
7eec314d | 4510 | return -EINVAL; |
56ccc112 | 4511 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
7eec314d N |
4512 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
4513 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4514 | } |
48c26ddc N |
4515 | if (mddev->ro == 2) { |
4516 | /* A write to sync_action is enough to justify | |
4517 | * canceling read-auto mode | |
4518 | */ | |
4519 | mddev->ro = 0; | |
4520 | md_wakeup_thread(mddev->sync_thread); | |
4521 | } | |
03c902e1 | 4522 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4523 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4524 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4525 | return len; |
4526 | } | |
4527 | ||
c4a39551 | 4528 | static struct md_sysfs_entry md_scan_mode = |
750f199e | 4529 | __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
c4a39551 JB |
4530 | |
4531 | static ssize_t | |
4532 | last_sync_action_show(struct mddev *mddev, char *page) | |
4533 | { | |
4534 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4535 | } | |
4536 | ||
4537 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4538 | ||
9d88883e | 4539 | static ssize_t |
fd01b88c | 4540 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4541 | { |
4542 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4543 | (unsigned long long) |
4544 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4545 | } |
4546 | ||
80ca3a44 | 4547 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4548 | |
88202a0c | 4549 | static ssize_t |
fd01b88c | 4550 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4551 | { |
4552 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4553 | mddev->sync_speed_min ? "local": "system"); | |
4554 | } | |
4555 | ||
4556 | static ssize_t | |
fd01b88c | 4557 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4558 | { |
4c9309c0 AD |
4559 | unsigned int min; |
4560 | int rv; | |
4561 | ||
88202a0c | 4562 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4563 | min = 0; |
4564 | } else { | |
4565 | rv = kstrtouint(buf, 10, &min); | |
4566 | if (rv < 0) | |
4567 | return rv; | |
4568 | if (min == 0) | |
4569 | return -EINVAL; | |
88202a0c | 4570 | } |
88202a0c N |
4571 | mddev->sync_speed_min = min; |
4572 | return len; | |
4573 | } | |
4574 | ||
4575 | static struct md_sysfs_entry md_sync_min = | |
4576 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4577 | ||
4578 | static ssize_t | |
fd01b88c | 4579 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4580 | { |
4581 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4582 | mddev->sync_speed_max ? "local": "system"); | |
4583 | } | |
4584 | ||
4585 | static ssize_t | |
fd01b88c | 4586 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4587 | { |
4c9309c0 AD |
4588 | unsigned int max; |
4589 | int rv; | |
4590 | ||
88202a0c | 4591 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4592 | max = 0; |
4593 | } else { | |
4594 | rv = kstrtouint(buf, 10, &max); | |
4595 | if (rv < 0) | |
4596 | return rv; | |
4597 | if (max == 0) | |
4598 | return -EINVAL; | |
88202a0c | 4599 | } |
88202a0c N |
4600 | mddev->sync_speed_max = max; |
4601 | return len; | |
4602 | } | |
4603 | ||
4604 | static struct md_sysfs_entry md_sync_max = | |
4605 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4606 | ||
d7f3d291 | 4607 | static ssize_t |
fd01b88c | 4608 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4609 | { |
4610 | return sprintf(page, "%d\n", mddev->degraded); | |
4611 | } | |
4612 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4613 | |
90b08710 | 4614 | static ssize_t |
fd01b88c | 4615 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4616 | { |
4617 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4618 | } | |
4619 | ||
4620 | static ssize_t | |
fd01b88c | 4621 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4622 | { |
4623 | long n; | |
4624 | ||
b29bebd6 | 4625 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4626 | return -EINVAL; |
4627 | ||
4628 | if (n != 0 && n != 1) | |
4629 | return -EINVAL; | |
4630 | ||
4631 | mddev->parallel_resync = n; | |
4632 | ||
4633 | if (mddev->sync_thread) | |
4634 | wake_up(&resync_wait); | |
4635 | ||
4636 | return len; | |
4637 | } | |
4638 | ||
4639 | /* force parallel resync, even with shared block devices */ | |
4640 | static struct md_sysfs_entry md_sync_force_parallel = | |
4641 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4642 | sync_force_parallel_show, sync_force_parallel_store); | |
4643 | ||
88202a0c | 4644 | static ssize_t |
fd01b88c | 4645 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4646 | { |
4647 | unsigned long resync, dt, db; | |
d1a7c503 N |
4648 | if (mddev->curr_resync == 0) |
4649 | return sprintf(page, "none\n"); | |
9687a60c AN |
4650 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4651 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4652 | if (!dt) dt++; |
9687a60c AN |
4653 | db = resync - mddev->resync_mark_cnt; |
4654 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4655 | } |
4656 | ||
80ca3a44 | 4657 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4658 | |
4659 | static ssize_t | |
fd01b88c | 4660 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4661 | { |
13ae864b | 4662 | unsigned long long max_sectors, resync; |
88202a0c | 4663 | |
acb180b0 N |
4664 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4665 | return sprintf(page, "none\n"); | |
4666 | ||
72f36d59 N |
4667 | if (mddev->curr_resync == 1 || |
4668 | mddev->curr_resync == 2) | |
4669 | return sprintf(page, "delayed\n"); | |
4670 | ||
c804cdec N |
4671 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4672 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4673 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4674 | else |
58c0fed4 | 4675 | max_sectors = mddev->dev_sectors; |
88202a0c | 4676 | |
acb180b0 | 4677 | resync = mddev->curr_resync_completed; |
13ae864b | 4678 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4679 | } |
4680 | ||
750f199e N |
4681 | static struct md_sysfs_entry md_sync_completed = |
4682 | __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL); | |
88202a0c | 4683 | |
5e96ee65 | 4684 | static ssize_t |
fd01b88c | 4685 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4686 | { |
4687 | return sprintf(page, "%llu\n", | |
4688 | (unsigned long long)mddev->resync_min); | |
4689 | } | |
4690 | static ssize_t | |
fd01b88c | 4691 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4692 | { |
4693 | unsigned long long min; | |
23da422b | 4694 | int err; |
23da422b | 4695 | |
b29bebd6 | 4696 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4697 | return -EINVAL; |
23da422b N |
4698 | |
4699 | spin_lock(&mddev->lock); | |
4700 | err = -EINVAL; | |
5e96ee65 | 4701 | if (min > mddev->resync_max) |
23da422b N |
4702 | goto out_unlock; |
4703 | ||
4704 | err = -EBUSY; | |
5e96ee65 | 4705 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4706 | goto out_unlock; |
5e96ee65 | 4707 | |
50c37b13 N |
4708 | /* Round down to multiple of 4K for safety */ |
4709 | mddev->resync_min = round_down(min, 8); | |
23da422b | 4710 | err = 0; |
5e96ee65 | 4711 | |
23da422b N |
4712 | out_unlock: |
4713 | spin_unlock(&mddev->lock); | |
4714 | return err ?: len; | |
5e96ee65 NB |
4715 | } |
4716 | ||
4717 | static struct md_sysfs_entry md_min_sync = | |
4718 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4719 | ||
c6207277 | 4720 | static ssize_t |
fd01b88c | 4721 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4722 | { |
4723 | if (mddev->resync_max == MaxSector) | |
4724 | return sprintf(page, "max\n"); | |
4725 | else | |
4726 | return sprintf(page, "%llu\n", | |
4727 | (unsigned long long)mddev->resync_max); | |
4728 | } | |
4729 | static ssize_t | |
fd01b88c | 4730 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4731 | { |
23da422b N |
4732 | int err; |
4733 | spin_lock(&mddev->lock); | |
c6207277 N |
4734 | if (strncmp(buf, "max", 3) == 0) |
4735 | mddev->resync_max = MaxSector; | |
4736 | else { | |
5e96ee65 | 4737 | unsigned long long max; |
23da422b N |
4738 | int chunk; |
4739 | ||
4740 | err = -EINVAL; | |
b29bebd6 | 4741 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4742 | goto out_unlock; |
5e96ee65 | 4743 | if (max < mddev->resync_min) |
23da422b N |
4744 | goto out_unlock; |
4745 | ||
4746 | err = -EBUSY; | |
c6207277 | 4747 | if (max < mddev->resync_max && |
4d484a4a | 4748 | mddev->ro == 0 && |
c6207277 | 4749 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4750 | goto out_unlock; |
c6207277 N |
4751 | |
4752 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4753 | chunk = mddev->chunk_sectors; |
4754 | if (chunk) { | |
2ac06c33 | 4755 | sector_t temp = max; |
23da422b N |
4756 | |
4757 | err = -EINVAL; | |
4758 | if (sector_div(temp, chunk)) | |
4759 | goto out_unlock; | |
c6207277 N |
4760 | } |
4761 | mddev->resync_max = max; | |
4762 | } | |
4763 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4764 | err = 0; |
4765 | out_unlock: | |
4766 | spin_unlock(&mddev->lock); | |
4767 | return err ?: len; | |
c6207277 N |
4768 | } |
4769 | ||
4770 | static struct md_sysfs_entry md_max_sync = | |
4771 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4772 | ||
e464eafd | 4773 | static ssize_t |
fd01b88c | 4774 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4775 | { |
4776 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4777 | } | |
4778 | ||
4779 | static ssize_t | |
fd01b88c | 4780 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4781 | { |
4c9309c0 | 4782 | unsigned long long old, new; |
6791875e | 4783 | int err; |
e464eafd | 4784 | |
4c9309c0 AD |
4785 | err = kstrtoull(buf, 10, &new); |
4786 | if (err < 0) | |
4787 | return err; | |
4788 | if (new != (sector_t)new) | |
e464eafd | 4789 | return -EINVAL; |
23ddff37 | 4790 | |
6791875e N |
4791 | err = mddev_lock(mddev); |
4792 | if (err) | |
4793 | return err; | |
4794 | err = -EINVAL; | |
4795 | if (mddev->pers == NULL || | |
4796 | mddev->pers->quiesce == NULL) | |
4797 | goto unlock; | |
4798 | old = mddev->suspend_lo; | |
23ddff37 N |
4799 | mddev->suspend_lo = new; |
4800 | if (new >= old) | |
4801 | /* Shrinking suspended region */ | |
e464eafd | 4802 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4803 | else { |
4804 | /* Expanding suspended region - need to wait */ | |
4805 | mddev->pers->quiesce(mddev, 1); | |
4806 | mddev->pers->quiesce(mddev, 0); | |
4807 | } | |
6791875e N |
4808 | err = 0; |
4809 | unlock: | |
4810 | mddev_unlock(mddev); | |
4811 | return err ?: len; | |
e464eafd N |
4812 | } |
4813 | static struct md_sysfs_entry md_suspend_lo = | |
4814 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4815 | ||
e464eafd | 4816 | static ssize_t |
fd01b88c | 4817 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4818 | { |
4819 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4820 | } | |
4821 | ||
4822 | static ssize_t | |
fd01b88c | 4823 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4824 | { |
4c9309c0 | 4825 | unsigned long long old, new; |
6791875e | 4826 | int err; |
e464eafd | 4827 | |
4c9309c0 AD |
4828 | err = kstrtoull(buf, 10, &new); |
4829 | if (err < 0) | |
4830 | return err; | |
4831 | if (new != (sector_t)new) | |
e464eafd | 4832 | return -EINVAL; |
23ddff37 | 4833 | |
6791875e N |
4834 | err = mddev_lock(mddev); |
4835 | if (err) | |
4836 | return err; | |
4837 | err = -EINVAL; | |
4838 | if (mddev->pers == NULL || | |
4839 | mddev->pers->quiesce == NULL) | |
4840 | goto unlock; | |
4841 | old = mddev->suspend_hi; | |
23ddff37 N |
4842 | mddev->suspend_hi = new; |
4843 | if (new <= old) | |
4844 | /* Shrinking suspended region */ | |
4845 | mddev->pers->quiesce(mddev, 2); | |
4846 | else { | |
4847 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4848 | mddev->pers->quiesce(mddev, 1); |
4849 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 | 4850 | } |
6791875e N |
4851 | err = 0; |
4852 | unlock: | |
4853 | mddev_unlock(mddev); | |
4854 | return err ?: len; | |
e464eafd N |
4855 | } |
4856 | static struct md_sysfs_entry md_suspend_hi = | |
4857 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4858 | ||
08a02ecd | 4859 | static ssize_t |
fd01b88c | 4860 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4861 | { |
4862 | if (mddev->reshape_position != MaxSector) | |
4863 | return sprintf(page, "%llu\n", | |
4864 | (unsigned long long)mddev->reshape_position); | |
4865 | strcpy(page, "none\n"); | |
4866 | return 5; | |
4867 | } | |
4868 | ||
4869 | static ssize_t | |
fd01b88c | 4870 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4871 | { |
c6563a8c | 4872 | struct md_rdev *rdev; |
4c9309c0 | 4873 | unsigned long long new; |
6791875e | 4874 | int err; |
6791875e | 4875 | |
4c9309c0 AD |
4876 | err = kstrtoull(buf, 10, &new); |
4877 | if (err < 0) | |
4878 | return err; | |
4879 | if (new != (sector_t)new) | |
08a02ecd | 4880 | return -EINVAL; |
6791875e N |
4881 | err = mddev_lock(mddev); |
4882 | if (err) | |
4883 | return err; | |
4884 | err = -EBUSY; | |
4885 | if (mddev->pers) | |
4886 | goto unlock; | |
08a02ecd N |
4887 | mddev->reshape_position = new; |
4888 | mddev->delta_disks = 0; | |
2c810cdd | 4889 | mddev->reshape_backwards = 0; |
08a02ecd N |
4890 | mddev->new_level = mddev->level; |
4891 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4892 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4893 | rdev_for_each(rdev, mddev) |
4894 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4895 | err = 0; |
4896 | unlock: | |
4897 | mddev_unlock(mddev); | |
4898 | return err ?: len; | |
08a02ecd N |
4899 | } |
4900 | ||
4901 | static struct md_sysfs_entry md_reshape_position = | |
4902 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4903 | reshape_position_store); | |
4904 | ||
2c810cdd N |
4905 | static ssize_t |
4906 | reshape_direction_show(struct mddev *mddev, char *page) | |
4907 | { | |
4908 | return sprintf(page, "%s\n", | |
4909 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4910 | } | |
4911 | ||
4912 | static ssize_t | |
4913 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4914 | { | |
4915 | int backwards = 0; | |
6791875e N |
4916 | int err; |
4917 | ||
2c810cdd N |
4918 | if (cmd_match(buf, "forwards")) |
4919 | backwards = 0; | |
4920 | else if (cmd_match(buf, "backwards")) | |
4921 | backwards = 1; | |
4922 | else | |
4923 | return -EINVAL; | |
4924 | if (mddev->reshape_backwards == backwards) | |
4925 | return len; | |
4926 | ||
6791875e N |
4927 | err = mddev_lock(mddev); |
4928 | if (err) | |
4929 | return err; | |
2c810cdd N |
4930 | /* check if we are allowed to change */ |
4931 | if (mddev->delta_disks) | |
6791875e N |
4932 | err = -EBUSY; |
4933 | else if (mddev->persistent && | |
2c810cdd | 4934 | mddev->major_version == 0) |
6791875e N |
4935 | err = -EINVAL; |
4936 | else | |
4937 | mddev->reshape_backwards = backwards; | |
4938 | mddev_unlock(mddev); | |
4939 | return err ?: len; | |
2c810cdd N |
4940 | } |
4941 | ||
4942 | static struct md_sysfs_entry md_reshape_direction = | |
4943 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4944 | reshape_direction_store); | |
4945 | ||
b522adcd | 4946 | static ssize_t |
fd01b88c | 4947 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4948 | { |
4949 | if (mddev->external_size) | |
4950 | return sprintf(page, "%llu\n", | |
4951 | (unsigned long long)mddev->array_sectors/2); | |
4952 | else | |
4953 | return sprintf(page, "default\n"); | |
4954 | } | |
4955 | ||
4956 | static ssize_t | |
fd01b88c | 4957 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4958 | { |
4959 | sector_t sectors; | |
6791875e N |
4960 | int err; |
4961 | ||
4962 | err = mddev_lock(mddev); | |
4963 | if (err) | |
4964 | return err; | |
b522adcd | 4965 | |
ab5a98b1 | 4966 | /* cluster raid doesn't support change array_sectors */ |
b670883b ZL |
4967 | if (mddev_is_clustered(mddev)) { |
4968 | mddev_unlock(mddev); | |
ab5a98b1 | 4969 | return -EINVAL; |
b670883b | 4970 | } |
ab5a98b1 | 4971 | |
b522adcd DW |
4972 | if (strncmp(buf, "default", 7) == 0) { |
4973 | if (mddev->pers) | |
4974 | sectors = mddev->pers->size(mddev, 0, 0); | |
4975 | else | |
4976 | sectors = mddev->array_sectors; | |
4977 | ||
4978 | mddev->external_size = 0; | |
4979 | } else { | |
4980 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
4981 | err = -EINVAL; |
4982 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
4983 | err = -E2BIG; | |
4984 | else | |
4985 | mddev->external_size = 1; | |
b522adcd DW |
4986 | } |
4987 | ||
6791875e N |
4988 | if (!err) { |
4989 | mddev->array_sectors = sectors; | |
4990 | if (mddev->pers) { | |
4991 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4992 | revalidate_disk(mddev->gendisk); | |
4993 | } | |
cbe6ef1d | 4994 | } |
6791875e N |
4995 | mddev_unlock(mddev); |
4996 | return err ?: len; | |
b522adcd DW |
4997 | } |
4998 | ||
4999 | static struct md_sysfs_entry md_array_size = | |
5000 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
5001 | array_size_store); | |
e464eafd | 5002 | |
664aed04 AP |
5003 | static ssize_t |
5004 | consistency_policy_show(struct mddev *mddev, char *page) | |
5005 | { | |
5006 | int ret; | |
5007 | ||
5008 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { | |
5009 | ret = sprintf(page, "journal\n"); | |
5010 | } else if (test_bit(MD_HAS_PPL, &mddev->flags)) { | |
5011 | ret = sprintf(page, "ppl\n"); | |
5012 | } else if (mddev->bitmap) { | |
5013 | ret = sprintf(page, "bitmap\n"); | |
5014 | } else if (mddev->pers) { | |
5015 | if (mddev->pers->sync_request) | |
5016 | ret = sprintf(page, "resync\n"); | |
5017 | else | |
5018 | ret = sprintf(page, "none\n"); | |
5019 | } else { | |
5020 | ret = sprintf(page, "unknown\n"); | |
5021 | } | |
5022 | ||
5023 | return ret; | |
5024 | } | |
5025 | ||
5026 | static ssize_t | |
5027 | consistency_policy_store(struct mddev *mddev, const char *buf, size_t len) | |
5028 | { | |
ba903a3e AP |
5029 | int err = 0; |
5030 | ||
664aed04 | 5031 | if (mddev->pers) { |
ba903a3e AP |
5032 | if (mddev->pers->change_consistency_policy) |
5033 | err = mddev->pers->change_consistency_policy(mddev, buf); | |
5034 | else | |
5035 | err = -EBUSY; | |
664aed04 AP |
5036 | } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) { |
5037 | set_bit(MD_HAS_PPL, &mddev->flags); | |
664aed04 | 5038 | } else { |
ba903a3e | 5039 | err = -EINVAL; |
664aed04 | 5040 | } |
ba903a3e AP |
5041 | |
5042 | return err ? err : len; | |
664aed04 AP |
5043 | } |
5044 | ||
5045 | static struct md_sysfs_entry md_consistency_policy = | |
5046 | __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show, | |
5047 | consistency_policy_store); | |
5048 | ||
eae1701f N |
5049 | static struct attribute *md_default_attrs[] = { |
5050 | &md_level.attr, | |
d4dbd025 | 5051 | &md_layout.attr, |
eae1701f | 5052 | &md_raid_disks.attr, |
3b34380a | 5053 | &md_chunk_size.attr, |
a35b0d69 | 5054 | &md_size.attr, |
a94213b1 | 5055 | &md_resync_start.attr, |
8bb93aac | 5056 | &md_metadata.attr, |
6d7ff738 | 5057 | &md_new_device.attr, |
16f17b39 | 5058 | &md_safe_delay.attr, |
9e653b63 | 5059 | &md_array_state.attr, |
08a02ecd | 5060 | &md_reshape_position.attr, |
2c810cdd | 5061 | &md_reshape_direction.attr, |
b522adcd | 5062 | &md_array_size.attr, |
1e50915f | 5063 | &max_corr_read_errors.attr, |
664aed04 | 5064 | &md_consistency_policy.attr, |
411036fa N |
5065 | NULL, |
5066 | }; | |
5067 | ||
5068 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 5069 | &md_scan_mode.attr, |
c4a39551 | 5070 | &md_last_scan_mode.attr, |
9d88883e | 5071 | &md_mismatches.attr, |
88202a0c N |
5072 | &md_sync_min.attr, |
5073 | &md_sync_max.attr, | |
5074 | &md_sync_speed.attr, | |
90b08710 | 5075 | &md_sync_force_parallel.attr, |
88202a0c | 5076 | &md_sync_completed.attr, |
5e96ee65 | 5077 | &md_min_sync.attr, |
c6207277 | 5078 | &md_max_sync.attr, |
e464eafd N |
5079 | &md_suspend_lo.attr, |
5080 | &md_suspend_hi.attr, | |
9b1d1dac | 5081 | &md_bitmap.attr, |
d7f3d291 | 5082 | &md_degraded.attr, |
eae1701f N |
5083 | NULL, |
5084 | }; | |
411036fa N |
5085 | static struct attribute_group md_redundancy_group = { |
5086 | .name = NULL, | |
5087 | .attrs = md_redundancy_attrs, | |
5088 | }; | |
5089 | ||
eae1701f N |
5090 | static ssize_t |
5091 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
5092 | { | |
5093 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 5094 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 5095 | ssize_t rv; |
eae1701f N |
5096 | |
5097 | if (!entry->show) | |
5098 | return -EIO; | |
af8a2434 N |
5099 | spin_lock(&all_mddevs_lock); |
5100 | if (list_empty(&mddev->all_mddevs)) { | |
5101 | spin_unlock(&all_mddevs_lock); | |
5102 | return -EBUSY; | |
5103 | } | |
5104 | mddev_get(mddev); | |
5105 | spin_unlock(&all_mddevs_lock); | |
5106 | ||
b7b17c9b | 5107 | rv = entry->show(mddev, page); |
af8a2434 | 5108 | mddev_put(mddev); |
96de1e66 | 5109 | return rv; |
eae1701f N |
5110 | } |
5111 | ||
5112 | static ssize_t | |
5113 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
5114 | const char *page, size_t length) | |
5115 | { | |
5116 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 5117 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 5118 | ssize_t rv; |
eae1701f N |
5119 | |
5120 | if (!entry->store) | |
5121 | return -EIO; | |
67463acb N |
5122 | if (!capable(CAP_SYS_ADMIN)) |
5123 | return -EACCES; | |
af8a2434 N |
5124 | spin_lock(&all_mddevs_lock); |
5125 | if (list_empty(&mddev->all_mddevs)) { | |
5126 | spin_unlock(&all_mddevs_lock); | |
5127 | return -EBUSY; | |
5128 | } | |
5129 | mddev_get(mddev); | |
5130 | spin_unlock(&all_mddevs_lock); | |
6791875e | 5131 | rv = entry->store(mddev, page, length); |
af8a2434 | 5132 | mddev_put(mddev); |
96de1e66 | 5133 | return rv; |
eae1701f N |
5134 | } |
5135 | ||
5136 | static void md_free(struct kobject *ko) | |
5137 | { | |
fd01b88c | 5138 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
5139 | |
5140 | if (mddev->sysfs_state) | |
5141 | sysfs_put(mddev->sysfs_state); | |
5142 | ||
6cd18e71 N |
5143 | if (mddev->queue) |
5144 | blk_cleanup_queue(mddev->queue); | |
a21d1504 N |
5145 | if (mddev->gendisk) { |
5146 | del_gendisk(mddev->gendisk); | |
5147 | put_disk(mddev->gendisk); | |
5148 | } | |
4ad23a97 | 5149 | percpu_ref_exit(&mddev->writes_pending); |
a21d1504 | 5150 | |
eae1701f N |
5151 | kfree(mddev); |
5152 | } | |
5153 | ||
52cf25d0 | 5154 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
5155 | .show = md_attr_show, |
5156 | .store = md_attr_store, | |
5157 | }; | |
5158 | static struct kobj_type md_ktype = { | |
5159 | .release = md_free, | |
5160 | .sysfs_ops = &md_sysfs_ops, | |
5161 | .default_attrs = md_default_attrs, | |
5162 | }; | |
5163 | ||
1da177e4 LT |
5164 | int mdp_major = 0; |
5165 | ||
5fd3a17e DW |
5166 | static void mddev_delayed_delete(struct work_struct *ws) |
5167 | { | |
fd01b88c | 5168 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 5169 | |
43a70507 | 5170 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
5171 | kobject_del(&mddev->kobj); |
5172 | kobject_put(&mddev->kobj); | |
5173 | } | |
5174 | ||
4ad23a97 N |
5175 | static void no_op(struct percpu_ref *r) {} |
5176 | ||
efeb53c0 | 5177 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 5178 | { |
039b7225 N |
5179 | /* |
5180 | * If dev is zero, name is the name of a device to allocate with | |
5181 | * an arbitrary minor number. It will be "md_???" | |
5182 | * If dev is non-zero it must be a device number with a MAJOR of | |
5183 | * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then | |
5184 | * the device is being created by opening a node in /dev. | |
5185 | * If "name" is not NULL, the device is being created by | |
5186 | * writing to /sys/module/md_mod/parameters/new_array. | |
5187 | */ | |
48c9c27b | 5188 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 5189 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 5190 | struct gendisk *disk; |
efeb53c0 N |
5191 | int partitioned; |
5192 | int shift; | |
5193 | int unit; | |
3830c62f | 5194 | int error; |
1da177e4 LT |
5195 | |
5196 | if (!mddev) | |
efeb53c0 N |
5197 | return -ENODEV; |
5198 | ||
5199 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
5200 | shift = partitioned ? MdpMinorShift : 0; | |
5201 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 5202 | |
e804ac78 TH |
5203 | /* wait for any previous instance of this device to be |
5204 | * completely removed (mddev_delayed_delete). | |
d3374825 | 5205 | */ |
e804ac78 | 5206 | flush_workqueue(md_misc_wq); |
d3374825 | 5207 | |
48c9c27b | 5208 | mutex_lock(&disks_mutex); |
0909dc44 N |
5209 | error = -EEXIST; |
5210 | if (mddev->gendisk) | |
5211 | goto abort; | |
efeb53c0 | 5212 | |
039b7225 | 5213 | if (name && !dev) { |
efeb53c0 N |
5214 | /* Need to ensure that 'name' is not a duplicate. |
5215 | */ | |
fd01b88c | 5216 | struct mddev *mddev2; |
efeb53c0 N |
5217 | spin_lock(&all_mddevs_lock); |
5218 | ||
5219 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
5220 | if (mddev2->gendisk && | |
5221 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
5222 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 5223 | goto abort; |
efeb53c0 N |
5224 | } |
5225 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 5226 | } |
039b7225 N |
5227 | if (name && dev) |
5228 | /* | |
5229 | * Creating /dev/mdNNN via "newarray", so adjust hold_active. | |
5230 | */ | |
5231 | mddev->hold_active = UNTIL_STOP; | |
8b765398 | 5232 | |
0909dc44 | 5233 | error = -ENOMEM; |
8b765398 | 5234 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
5235 | if (!mddev->queue) |
5236 | goto abort; | |
409c57f3 N |
5237 | mddev->queue->queuedata = mddev; |
5238 | ||
409c57f3 | 5239 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 5240 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 5241 | |
4ad23a97 N |
5242 | if (percpu_ref_init(&mddev->writes_pending, no_op, 0, GFP_KERNEL) < 0) |
5243 | goto abort; | |
5244 | /* We want to start with the refcount at zero */ | |
5245 | percpu_ref_put(&mddev->writes_pending); | |
1da177e4 LT |
5246 | disk = alloc_disk(1 << shift); |
5247 | if (!disk) { | |
8b765398 N |
5248 | blk_cleanup_queue(mddev->queue); |
5249 | mddev->queue = NULL; | |
0909dc44 | 5250 | goto abort; |
1da177e4 | 5251 | } |
efeb53c0 | 5252 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 5253 | disk->first_minor = unit << shift; |
efeb53c0 N |
5254 | if (name) |
5255 | strcpy(disk->disk_name, name); | |
5256 | else if (partitioned) | |
1da177e4 | 5257 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 5258 | else |
1da177e4 | 5259 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
5260 | disk->fops = &md_fops; |
5261 | disk->private_data = mddev; | |
5262 | disk->queue = mddev->queue; | |
56883a7e | 5263 | blk_queue_write_cache(mddev->queue, true, true); |
92850bbd | 5264 | /* Allow extended partitions. This makes the |
d3374825 | 5265 | * 'mdp' device redundant, but we can't really |
92850bbd N |
5266 | * remove it now. |
5267 | */ | |
5268 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 5269 | mddev->gendisk = disk; |
b0140891 N |
5270 | /* As soon as we call add_disk(), another thread could get |
5271 | * through to md_open, so make sure it doesn't get too far | |
5272 | */ | |
5273 | mutex_lock(&mddev->open_mutex); | |
5274 | add_disk(disk); | |
5275 | ||
ed9e1982 TH |
5276 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
5277 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
5278 | if (error) { |
5279 | /* This isn't possible, but as kobject_init_and_add is marked | |
5280 | * __must_check, we must do something with the result | |
5281 | */ | |
9d48739e N |
5282 | pr_debug("md: cannot register %s/md - name in use\n", |
5283 | disk->disk_name); | |
0909dc44 N |
5284 | error = 0; |
5285 | } | |
00bcb4ac N |
5286 | if (mddev->kobj.sd && |
5287 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
9d48739e | 5288 | pr_debug("pointless warning\n"); |
b0140891 | 5289 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
5290 | abort: |
5291 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 5292 | if (!error && mddev->kobj.sd) { |
3830c62f | 5293 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 5294 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 5295 | } |
d3374825 | 5296 | mddev_put(mddev); |
0909dc44 | 5297 | return error; |
efeb53c0 N |
5298 | } |
5299 | ||
5300 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
5301 | { | |
78b6350d N |
5302 | if (create_on_open) |
5303 | md_alloc(dev, NULL); | |
1da177e4 LT |
5304 | return NULL; |
5305 | } | |
5306 | ||
efeb53c0 N |
5307 | static int add_named_array(const char *val, struct kernel_param *kp) |
5308 | { | |
039b7225 N |
5309 | /* |
5310 | * val must be "md_*" or "mdNNN". | |
5311 | * For "md_*" we allocate an array with a large free minor number, and | |
efeb53c0 | 5312 | * set the name to val. val must not already be an active name. |
039b7225 N |
5313 | * For "mdNNN" we allocate an array with the minor number NNN |
5314 | * which must not already be in use. | |
efeb53c0 N |
5315 | */ |
5316 | int len = strlen(val); | |
5317 | char buf[DISK_NAME_LEN]; | |
039b7225 | 5318 | unsigned long devnum; |
efeb53c0 N |
5319 | |
5320 | while (len && val[len-1] == '\n') | |
5321 | len--; | |
5322 | if (len >= DISK_NAME_LEN) | |
5323 | return -E2BIG; | |
5324 | strlcpy(buf, val, len+1); | |
039b7225 N |
5325 | if (strncmp(buf, "md_", 3) == 0) |
5326 | return md_alloc(0, buf); | |
5327 | if (strncmp(buf, "md", 2) == 0 && | |
5328 | isdigit(buf[2]) && | |
5329 | kstrtoul(buf+2, 10, &devnum) == 0 && | |
5330 | devnum <= MINORMASK) | |
5331 | return md_alloc(MKDEV(MD_MAJOR, devnum), NULL); | |
5332 | ||
5333 | return -EINVAL; | |
efeb53c0 N |
5334 | } |
5335 | ||
1da177e4 LT |
5336 | static void md_safemode_timeout(unsigned long data) |
5337 | { | |
fd01b88c | 5338 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 5339 | |
4ad23a97 N |
5340 | mddev->safemode = 1; |
5341 | if (mddev->external) | |
5342 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
5343 | ||
1da177e4 LT |
5344 | md_wakeup_thread(mddev->thread); |
5345 | } | |
5346 | ||
6ff8d8ec | 5347 | static int start_dirty_degraded; |
1da177e4 | 5348 | |
fd01b88c | 5349 | int md_run(struct mddev *mddev) |
1da177e4 | 5350 | { |
2604b703 | 5351 | int err; |
3cb03002 | 5352 | struct md_rdev *rdev; |
84fc4b56 | 5353 | struct md_personality *pers; |
1da177e4 | 5354 | |
a757e64c N |
5355 | if (list_empty(&mddev->disks)) |
5356 | /* cannot run an array with no devices.. */ | |
1da177e4 | 5357 | return -EINVAL; |
1da177e4 LT |
5358 | |
5359 | if (mddev->pers) | |
5360 | return -EBUSY; | |
bb4f1e9d N |
5361 | /* Cannot run until previous stop completes properly */ |
5362 | if (mddev->sysfs_active) | |
5363 | return -EBUSY; | |
b6eb127d | 5364 | |
1da177e4 LT |
5365 | /* |
5366 | * Analyze all RAID superblock(s) | |
5367 | */ | |
1ec4a939 N |
5368 | if (!mddev->raid_disks) { |
5369 | if (!mddev->persistent) | |
5370 | return -EINVAL; | |
a757e64c | 5371 | analyze_sbs(mddev); |
1ec4a939 | 5372 | } |
1da177e4 | 5373 | |
d9d166c2 N |
5374 | if (mddev->level != LEVEL_NONE) |
5375 | request_module("md-level-%d", mddev->level); | |
5376 | else if (mddev->clevel[0]) | |
5377 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
5378 | |
5379 | /* | |
5380 | * Drop all container device buffers, from now on | |
5381 | * the only valid external interface is through the md | |
5382 | * device. | |
1da177e4 | 5383 | */ |
dafb20fa | 5384 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 5385 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5386 | continue; |
5387 | sync_blockdev(rdev->bdev); | |
f98393a6 | 5388 | invalidate_bdev(rdev->bdev); |
97b20ef7 N |
5389 | if (mddev->ro != 1 && |
5390 | (bdev_read_only(rdev->bdev) || | |
5391 | bdev_read_only(rdev->meta_bdev))) { | |
5392 | mddev->ro = 1; | |
5393 | if (mddev->gendisk) | |
5394 | set_disk_ro(mddev->gendisk, 1); | |
5395 | } | |
f0d76d70 N |
5396 | |
5397 | /* perform some consistency tests on the device. | |
5398 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 5399 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 5400 | */ |
a6ff7e08 JB |
5401 | if (rdev->meta_bdev) { |
5402 | /* Nothing to check */; | |
5403 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5404 | if (mddev->dev_sectors && |
5405 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5406 | > rdev->sb_start) { |
9d48739e N |
5407 | pr_warn("md: %s: data overlaps metadata\n", |
5408 | mdname(mddev)); | |
f0d76d70 N |
5409 | return -EINVAL; |
5410 | } | |
5411 | } else { | |
0f420358 | 5412 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 | 5413 | > rdev->data_offset) { |
9d48739e N |
5414 | pr_warn("md: %s: metadata overlaps data\n", |
5415 | mdname(mddev)); | |
f0d76d70 N |
5416 | return -EINVAL; |
5417 | } | |
5418 | } | |
00bcb4ac | 5419 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5420 | } |
5421 | ||
10273170 | 5422 | if (mddev->bio_set == NULL) { |
395c72a7 | 5423 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
10273170 ML |
5424 | if (!mddev->bio_set) |
5425 | return -ENOMEM; | |
5426 | } | |
a167f663 | 5427 | |
1da177e4 | 5428 | spin_lock(&pers_lock); |
d9d166c2 | 5429 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5430 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5431 | spin_unlock(&pers_lock); |
d9d166c2 | 5432 | if (mddev->level != LEVEL_NONE) |
9d48739e N |
5433 | pr_warn("md: personality for level %d is not loaded!\n", |
5434 | mddev->level); | |
d9d166c2 | 5435 | else |
9d48739e N |
5436 | pr_warn("md: personality for level %s is not loaded!\n", |
5437 | mddev->clevel); | |
1da177e4 LT |
5438 | return -EINVAL; |
5439 | } | |
1da177e4 | 5440 | spin_unlock(&pers_lock); |
34817e8c N |
5441 | if (mddev->level != pers->level) { |
5442 | mddev->level = pers->level; | |
5443 | mddev->new_level = pers->level; | |
5444 | } | |
d9d166c2 | 5445 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5446 | |
f6705578 | 5447 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5448 | pers->start_reshape == NULL) { |
f6705578 | 5449 | /* This personality cannot handle reshaping... */ |
f6705578 N |
5450 | module_put(pers->owner); |
5451 | return -EINVAL; | |
5452 | } | |
5453 | ||
7dd5e7c3 N |
5454 | if (pers->sync_request) { |
5455 | /* Warn if this is a potentially silly | |
5456 | * configuration. | |
5457 | */ | |
5458 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5459 | struct md_rdev *rdev2; |
7dd5e7c3 | 5460 | int warned = 0; |
159ec1fc | 5461 | |
dafb20fa N |
5462 | rdev_for_each(rdev, mddev) |
5463 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5464 | if (rdev < rdev2 && |
5465 | rdev->bdev->bd_contains == | |
5466 | rdev2->bdev->bd_contains) { | |
9d48739e N |
5467 | pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n", |
5468 | mdname(mddev), | |
5469 | bdevname(rdev->bdev,b), | |
5470 | bdevname(rdev2->bdev,b2)); | |
7dd5e7c3 N |
5471 | warned = 1; |
5472 | } | |
5473 | } | |
159ec1fc | 5474 | |
7dd5e7c3 | 5475 | if (warned) |
9d48739e | 5476 | pr_warn("True protection against single-disk failure might be compromised.\n"); |
7dd5e7c3 N |
5477 | } |
5478 | ||
657390d2 | 5479 | mddev->recovery = 0; |
58c0fed4 AN |
5480 | /* may be over-ridden by personality */ |
5481 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5482 | ||
6ff8d8ec | 5483 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5484 | |
0f9552b5 | 5485 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5486 | mddev->ro = 2; /* read-only, but switch on first write */ |
5487 | ||
a85dd7b8 SL |
5488 | /* |
5489 | * NOTE: some pers->run(), for example r5l_recovery_log(), wakes | |
5490 | * up mddev->thread. It is important to initialize critical | |
5491 | * resources for mddev->thread BEFORE calling pers->run(). | |
5492 | */ | |
36d091f4 | 5493 | err = pers->run(mddev); |
13e53df3 | 5494 | if (err) |
9d48739e | 5495 | pr_warn("md: pers->run() failed ...\n"); |
36d091f4 | 5496 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
9d48739e N |
5497 | WARN_ONCE(!mddev->external_size, |
5498 | "%s: default size too small, but 'external_size' not in effect?\n", | |
5499 | __func__); | |
5500 | pr_warn("md: invalid array_size %llu > default size %llu\n", | |
5501 | (unsigned long long)mddev->array_sectors / 2, | |
5502 | (unsigned long long)pers->size(mddev, 0, 0) / 2); | |
b522adcd | 5503 | err = -EINVAL; |
b522adcd | 5504 | } |
36d091f4 | 5505 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5506 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
f9209a32 GR |
5507 | struct bitmap *bitmap; |
5508 | ||
5509 | bitmap = bitmap_create(mddev, -1); | |
5510 | if (IS_ERR(bitmap)) { | |
5511 | err = PTR_ERR(bitmap); | |
9d48739e N |
5512 | pr_warn("%s: failed to create bitmap (%d)\n", |
5513 | mdname(mddev), err); | |
f9209a32 GR |
5514 | } else |
5515 | mddev->bitmap = bitmap; | |
5516 | ||
b15c2e57 | 5517 | } |
1da177e4 | 5518 | if (err) { |
5aa61f42 | 5519 | mddev_detach(mddev); |
0c35bd47 N |
5520 | if (mddev->private) |
5521 | pers->free(mddev, mddev->private); | |
bd691922 | 5522 | mddev->private = NULL; |
36d091f4 | 5523 | module_put(pers->owner); |
32a7627c N |
5524 | bitmap_destroy(mddev); |
5525 | return err; | |
1da177e4 | 5526 | } |
5c675f83 | 5527 | if (mddev->queue) { |
bb086a89 SL |
5528 | bool nonrot = true; |
5529 | ||
5530 | rdev_for_each(rdev, mddev) { | |
5531 | if (rdev->raid_disk >= 0 && | |
5532 | !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
5533 | nonrot = false; | |
5534 | break; | |
5535 | } | |
5536 | } | |
5537 | if (mddev->degraded) | |
5538 | nonrot = false; | |
5539 | if (nonrot) | |
5540 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue); | |
5541 | else | |
5542 | queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue); | |
dc3b17cc JK |
5543 | mddev->queue->backing_dev_info->congested_data = mddev; |
5544 | mddev->queue->backing_dev_info->congested_fn = md_congested; | |
5c675f83 | 5545 | } |
36d091f4 | 5546 | if (pers->sync_request) { |
00bcb4ac N |
5547 | if (mddev->kobj.sd && |
5548 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
5549 | pr_warn("md: cannot register extra attributes for %s\n", |
5550 | mdname(mddev)); | |
00bcb4ac | 5551 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5552 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5553 | mddev->ro = 0; |
5554 | ||
1e50915f RB |
5555 | atomic_set(&mddev->max_corr_read_errors, |
5556 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 | 5557 | mddev->safemode = 0; |
28c1b9fd GR |
5558 | if (mddev_is_clustered(mddev)) |
5559 | mddev->safemode_delay = 0; | |
5560 | else | |
5561 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ | |
1da177e4 | 5562 | mddev->in_sync = 1; |
0ca69886 | 5563 | smp_wmb(); |
36d091f4 N |
5564 | spin_lock(&mddev->lock); |
5565 | mddev->pers = pers; | |
36d091f4 | 5566 | spin_unlock(&mddev->lock); |
dafb20fa | 5567 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5568 | if (rdev->raid_disk >= 0) |
5569 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5570 | /* failure here is OK */; |
f72ffdd6 | 5571 | |
a4a3d26d N |
5572 | if (mddev->degraded && !mddev->ro) |
5573 | /* This ensures that recovering status is reported immediately | |
5574 | * via sysfs - until a lack of spares is confirmed. | |
5575 | */ | |
5576 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
1da177e4 | 5577 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5578 | |
2953079c | 5579 | if (mddev->sb_flags) |
850b2b42 | 5580 | md_update_sb(mddev, 0); |
1da177e4 | 5581 | |
d7603b7e | 5582 | md_new_event(mddev); |
00bcb4ac N |
5583 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5584 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5585 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5586 | return 0; |
5587 | } | |
390ee602 | 5588 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5589 | |
fd01b88c | 5590 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5591 | { |
5592 | int err; | |
5593 | ||
5594 | err = md_run(mddev); | |
5595 | if (err) | |
5596 | goto out; | |
69e51b44 N |
5597 | err = bitmap_load(mddev); |
5598 | if (err) { | |
5599 | bitmap_destroy(mddev); | |
5600 | goto out; | |
5601 | } | |
0fd018af | 5602 | |
28c1b9fd GR |
5603 | if (mddev_is_clustered(mddev)) |
5604 | md_allow_write(mddev); | |
5605 | ||
0fd018af JB |
5606 | md_wakeup_thread(mddev->thread); |
5607 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5608 | ||
fe60b014 N |
5609 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5610 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5611 | mddev->changed = 1; |
fe60b014 N |
5612 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5613 | out: | |
5614 | return err; | |
5615 | } | |
5616 | ||
fd01b88c | 5617 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5618 | { |
5619 | struct gendisk *disk = mddev->gendisk; | |
97b20ef7 N |
5620 | struct md_rdev *rdev; |
5621 | bool has_journal = false; | |
5622 | bool has_readonly = false; | |
1da177e4 | 5623 | |
80fab1d7 | 5624 | /* Complain if it has no devices */ |
1da177e4 | 5625 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5626 | return -ENXIO; |
5627 | if (!mddev->pers) | |
5628 | return -EINVAL; | |
5629 | if (!mddev->ro) | |
5630 | return -EBUSY; | |
339421de | 5631 | |
97b20ef7 N |
5632 | rcu_read_lock(); |
5633 | rdev_for_each_rcu(rdev, mddev) { | |
5634 | if (test_bit(Journal, &rdev->flags) && | |
5635 | !test_bit(Faulty, &rdev->flags)) | |
5636 | has_journal = true; | |
5637 | if (bdev_read_only(rdev->bdev)) | |
5638 | has_readonly = true; | |
5639 | } | |
5640 | rcu_read_unlock(); | |
5641 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal) | |
339421de | 5642 | /* Don't restart rw with journal missing/faulty */ |
339421de | 5643 | return -EINVAL; |
97b20ef7 N |
5644 | if (has_readonly) |
5645 | return -EROFS; | |
339421de | 5646 | |
80fab1d7 AN |
5647 | mddev->safemode = 0; |
5648 | mddev->ro = 0; | |
5649 | set_disk_ro(disk, 0); | |
9d48739e | 5650 | pr_debug("md: %s switched to read-write mode.\n", mdname(mddev)); |
80fab1d7 AN |
5651 | /* Kick recovery or resync if necessary */ |
5652 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5653 | md_wakeup_thread(mddev->thread); | |
5654 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5655 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5656 | return 0; |
1da177e4 LT |
5657 | } |
5658 | ||
fd01b88c | 5659 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5660 | { |
5661 | mddev->array_sectors = 0; | |
5662 | mddev->external_size = 0; | |
5663 | mddev->dev_sectors = 0; | |
5664 | mddev->raid_disks = 0; | |
5665 | mddev->recovery_cp = 0; | |
5666 | mddev->resync_min = 0; | |
5667 | mddev->resync_max = MaxSector; | |
5668 | mddev->reshape_position = MaxSector; | |
5669 | mddev->external = 0; | |
5670 | mddev->persistent = 0; | |
5671 | mddev->level = LEVEL_NONE; | |
5672 | mddev->clevel[0] = 0; | |
5673 | mddev->flags = 0; | |
2953079c | 5674 | mddev->sb_flags = 0; |
6177b472 N |
5675 | mddev->ro = 0; |
5676 | mddev->metadata_type[0] = 0; | |
5677 | mddev->chunk_sectors = 0; | |
5678 | mddev->ctime = mddev->utime = 0; | |
5679 | mddev->layout = 0; | |
5680 | mddev->max_disks = 0; | |
5681 | mddev->events = 0; | |
a8707c08 | 5682 | mddev->can_decrease_events = 0; |
6177b472 | 5683 | mddev->delta_disks = 0; |
2c810cdd | 5684 | mddev->reshape_backwards = 0; |
6177b472 N |
5685 | mddev->new_level = LEVEL_NONE; |
5686 | mddev->new_layout = 0; | |
5687 | mddev->new_chunk_sectors = 0; | |
5688 | mddev->curr_resync = 0; | |
7f7583d4 | 5689 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5690 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5691 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5692 | mddev->recovery = 0; | |
5693 | mddev->in_sync = 0; | |
f0b4f7e2 | 5694 | mddev->changed = 0; |
6177b472 | 5695 | mddev->degraded = 0; |
6177b472 | 5696 | mddev->safemode = 0; |
bd691922 | 5697 | mddev->private = NULL; |
c20c33f0 | 5698 | mddev->cluster_info = NULL; |
6177b472 N |
5699 | mddev->bitmap_info.offset = 0; |
5700 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5701 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5702 | mddev->bitmap_info.chunksize = 0; |
5703 | mddev->bitmap_info.daemon_sleep = 0; | |
5704 | mddev->bitmap_info.max_write_behind = 0; | |
c20c33f0 | 5705 | mddev->bitmap_info.nodes = 0; |
6177b472 N |
5706 | } |
5707 | ||
fd01b88c | 5708 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5709 | { |
6b6204ee | 5710 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5711 | flush_workqueue(md_misc_wq); |
a047e125 | 5712 | if (mddev->sync_thread) { |
a047e125 | 5713 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5714 | md_reap_sync_thread(mddev); |
a047e125 N |
5715 | } |
5716 | ||
5717 | del_timer_sync(&mddev->safemode_timer); | |
5718 | ||
034e33f5 SL |
5719 | if (mddev->pers && mddev->pers->quiesce) { |
5720 | mddev->pers->quiesce(mddev, 1); | |
5721 | mddev->pers->quiesce(mddev, 0); | |
5722 | } | |
a047e125 | 5723 | bitmap_flush(mddev); |
a047e125 | 5724 | |
b6d428c6 | 5725 | if (mddev->ro == 0 && |
28c1b9fd | 5726 | ((!mddev->in_sync && !mddev_is_clustered(mddev)) || |
2953079c | 5727 | mddev->sb_flags)) { |
a047e125 | 5728 | /* mark array as shutdown cleanly */ |
28c1b9fd GR |
5729 | if (!mddev_is_clustered(mddev)) |
5730 | mddev->in_sync = 1; | |
a047e125 N |
5731 | md_update_sb(mddev, 1); |
5732 | } | |
5733 | } | |
defad61a | 5734 | |
fd01b88c | 5735 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5736 | { |
29f097c4 | 5737 | mddev_lock_nointr(mddev); |
defad61a N |
5738 | __md_stop_writes(mddev); |
5739 | mddev_unlock(mddev); | |
5740 | } | |
390ee602 | 5741 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5742 | |
5aa61f42 N |
5743 | static void mddev_detach(struct mddev *mddev) |
5744 | { | |
48df498d | 5745 | bitmap_wait_behind_writes(mddev); |
36d091f4 | 5746 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5747 | mddev->pers->quiesce(mddev, 1); |
5748 | mddev->pers->quiesce(mddev, 0); | |
5749 | } | |
5750 | md_unregister_thread(&mddev->thread); | |
5751 | if (mddev->queue) | |
5752 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5753 | } | |
5754 | ||
5eff3c43 | 5755 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5756 | { |
36d091f4 | 5757 | struct md_personality *pers = mddev->pers; |
48df498d | 5758 | bitmap_destroy(mddev); |
5aa61f42 | 5759 | mddev_detach(mddev); |
ee5d004f N |
5760 | /* Ensure ->event_work is done */ |
5761 | flush_workqueue(md_misc_wq); | |
36d091f4 | 5762 | spin_lock(&mddev->lock); |
6177b472 | 5763 | mddev->pers = NULL; |
36d091f4 N |
5764 | spin_unlock(&mddev->lock); |
5765 | pers->free(mddev, mddev->private); | |
bd691922 | 5766 | mddev->private = NULL; |
36d091f4 N |
5767 | if (pers->sync_request && mddev->to_remove == NULL) |
5768 | mddev->to_remove = &md_redundancy_group; | |
5769 | module_put(pers->owner); | |
cca9cf90 | 5770 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5771 | } |
5eff3c43 N |
5772 | |
5773 | void md_stop(struct mddev *mddev) | |
5774 | { | |
5775 | /* stop the array and free an attached data structures. | |
5776 | * This is called from dm-raid | |
5777 | */ | |
5778 | __md_stop(mddev); | |
5eff3c43 N |
5779 | if (mddev->bio_set) |
5780 | bioset_free(mddev->bio_set); | |
5781 | } | |
5782 | ||
390ee602 | 5783 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5784 | |
a05b7ea0 | 5785 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5786 | { |
5787 | int err = 0; | |
30b8feb7 N |
5788 | int did_freeze = 0; |
5789 | ||
5790 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5791 | did_freeze = 1; | |
5792 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5793 | md_wakeup_thread(mddev->thread); | |
5794 | } | |
f851b60d | 5795 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5796 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5797 | if (mddev->sync_thread) |
30b8feb7 N |
5798 | /* Thread might be blocked waiting for metadata update |
5799 | * which will now never happen */ | |
5800 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5801 | |
2953079c | 5802 | if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
88724bfa | 5803 | return -EBUSY; |
30b8feb7 | 5804 | mddev_unlock(mddev); |
f851b60d N |
5805 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5806 | &mddev->recovery)); | |
88724bfa | 5807 | wait_event(mddev->sb_wait, |
2953079c | 5808 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
30b8feb7 N |
5809 | mddev_lock_nointr(mddev); |
5810 | ||
a4bd82d0 | 5811 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5812 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5813 | mddev->sync_thread || |
af8d8e6f | 5814 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5815 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5816 | if (did_freeze) { |
5817 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5818 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5819 | md_wakeup_thread(mddev->thread); |
5820 | } | |
a4bd82d0 N |
5821 | err = -EBUSY; |
5822 | goto out; | |
5823 | } | |
5824 | if (mddev->pers) { | |
defad61a | 5825 | __md_stop_writes(mddev); |
a4bd82d0 N |
5826 | |
5827 | err = -ENXIO; | |
5828 | if (mddev->ro==1) | |
5829 | goto out; | |
5830 | mddev->ro = 1; | |
5831 | set_disk_ro(mddev->gendisk, 1); | |
5832 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d N |
5833 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5834 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5835 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5836 | err = 0; |
a4bd82d0 N |
5837 | } |
5838 | out: | |
5839 | mutex_unlock(&mddev->open_mutex); | |
5840 | return err; | |
5841 | } | |
5842 | ||
9e653b63 N |
5843 | /* mode: |
5844 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5845 | * 2 - stop but do not disassemble array |
5846 | */ | |
f72ffdd6 | 5847 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5848 | struct block_device *bdev) |
1da177e4 | 5849 | { |
1da177e4 | 5850 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5851 | struct md_rdev *rdev; |
30b8feb7 N |
5852 | int did_freeze = 0; |
5853 | ||
5854 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5855 | did_freeze = 1; | |
5856 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5857 | md_wakeup_thread(mddev->thread); | |
5858 | } | |
f851b60d | 5859 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5860 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5861 | if (mddev->sync_thread) |
30b8feb7 N |
5862 | /* Thread might be blocked waiting for metadata update |
5863 | * which will now never happen */ | |
5864 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5865 | |
30b8feb7 | 5866 | mddev_unlock(mddev); |
f851b60d N |
5867 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
5868 | !test_bit(MD_RECOVERY_RUNNING, | |
5869 | &mddev->recovery))); | |
30b8feb7 | 5870 | mddev_lock_nointr(mddev); |
1da177e4 | 5871 | |
c8c00a69 | 5872 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5873 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5874 | mddev->sysfs_active || |
5875 | mddev->sync_thread || | |
af8d8e6f | 5876 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5877 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5878 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5879 | if (did_freeze) { |
5880 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5881 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5882 | md_wakeup_thread(mddev->thread); |
5883 | } | |
260fa034 N |
5884 | return -EBUSY; |
5885 | } | |
6e17b027 | 5886 | if (mddev->pers) { |
a4bd82d0 N |
5887 | if (mddev->ro) |
5888 | set_disk_ro(disk, 0); | |
409c57f3 | 5889 | |
defad61a | 5890 | __md_stop_writes(mddev); |
5eff3c43 | 5891 | __md_stop(mddev); |
dc3b17cc | 5892 | mddev->queue->backing_dev_info->congested_fn = NULL; |
6177b472 | 5893 | |
a4bd82d0 | 5894 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5895 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5896 | |
dafb20fa | 5897 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5898 | if (rdev->raid_disk >= 0) |
5899 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5900 | |
a4bd82d0 | 5901 | set_capacity(disk, 0); |
6e17b027 | 5902 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5903 | mddev->changed = 1; |
a4bd82d0 | 5904 | revalidate_disk(disk); |
0d4ca600 | 5905 | |
a4bd82d0 N |
5906 | if (mddev->ro) |
5907 | mddev->ro = 0; | |
6e17b027 N |
5908 | } else |
5909 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5910 | /* |
5911 | * Free resources if final stop | |
5912 | */ | |
9e653b63 | 5913 | if (mode == 0) { |
9d48739e | 5914 | pr_info("md: %s stopped.\n", mdname(mddev)); |
1da177e4 | 5915 | |
c3d9714e | 5916 | if (mddev->bitmap_info.file) { |
4af1a041 N |
5917 | struct file *f = mddev->bitmap_info.file; |
5918 | spin_lock(&mddev->lock); | |
c3d9714e | 5919 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
5920 | spin_unlock(&mddev->lock); |
5921 | fput(f); | |
978f946b | 5922 | } |
c3d9714e | 5923 | mddev->bitmap_info.offset = 0; |
978f946b | 5924 | |
1da177e4 LT |
5925 | export_array(mddev); |
5926 | ||
6177b472 | 5927 | md_clean(mddev); |
efeb53c0 N |
5928 | if (mddev->hold_active == UNTIL_STOP) |
5929 | mddev->hold_active = 0; | |
a4bd82d0 | 5930 | } |
d7603b7e | 5931 | md_new_event(mddev); |
00bcb4ac | 5932 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5933 | return 0; |
1da177e4 LT |
5934 | } |
5935 | ||
fdee8ae4 | 5936 | #ifndef MODULE |
fd01b88c | 5937 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5938 | { |
3cb03002 | 5939 | struct md_rdev *rdev; |
1da177e4 LT |
5940 | int err; |
5941 | ||
a757e64c | 5942 | if (list_empty(&mddev->disks)) |
1da177e4 | 5943 | return; |
1da177e4 | 5944 | |
9d48739e | 5945 | pr_info("md: running: "); |
1da177e4 | 5946 | |
dafb20fa | 5947 | rdev_for_each(rdev, mddev) { |
1da177e4 | 5948 | char b[BDEVNAME_SIZE]; |
9d48739e | 5949 | pr_cont("<%s>", bdevname(rdev->bdev,b)); |
1da177e4 | 5950 | } |
9d48739e | 5951 | pr_cont("\n"); |
1da177e4 | 5952 | |
d710e138 | 5953 | err = do_md_run(mddev); |
1da177e4 | 5954 | if (err) { |
9d48739e | 5955 | pr_warn("md: do_md_run() returned %d\n", err); |
a05b7ea0 | 5956 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5957 | } |
5958 | } | |
5959 | ||
5960 | /* | |
5961 | * lets try to run arrays based on all disks that have arrived | |
5962 | * until now. (those are in pending_raid_disks) | |
5963 | * | |
5964 | * the method: pick the first pending disk, collect all disks with | |
5965 | * the same UUID, remove all from the pending list and put them into | |
5966 | * the 'same_array' list. Then order this list based on superblock | |
5967 | * update time (freshest comes first), kick out 'old' disks and | |
5968 | * compare superblocks. If everything's fine then run it. | |
5969 | * | |
5970 | * If "unit" is allocated, then bump its reference count | |
5971 | */ | |
5972 | static void autorun_devices(int part) | |
5973 | { | |
3cb03002 | 5974 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5975 | struct mddev *mddev; |
1da177e4 LT |
5976 | char b[BDEVNAME_SIZE]; |
5977 | ||
9d48739e | 5978 | pr_info("md: autorun ...\n"); |
1da177e4 | 5979 | while (!list_empty(&pending_raid_disks)) { |
e8703fe1 | 5980 | int unit; |
1da177e4 | 5981 | dev_t dev; |
ad01c9e3 | 5982 | LIST_HEAD(candidates); |
1da177e4 | 5983 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5984 | struct md_rdev, same_set); |
1da177e4 | 5985 | |
9d48739e | 5986 | pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b)); |
1da177e4 | 5987 | INIT_LIST_HEAD(&candidates); |
159ec1fc | 5988 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 | 5989 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
9d48739e N |
5990 | pr_debug("md: adding %s ...\n", |
5991 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
5992 | list_move(&rdev->same_set, &candidates); |
5993 | } | |
5994 | /* | |
5995 | * now we have a set of devices, with all of them having | |
5996 | * mostly sane superblocks. It's time to allocate the | |
5997 | * mddev. | |
5998 | */ | |
e8703fe1 N |
5999 | if (part) { |
6000 | dev = MKDEV(mdp_major, | |
6001 | rdev0->preferred_minor << MdpMinorShift); | |
6002 | unit = MINOR(dev) >> MdpMinorShift; | |
6003 | } else { | |
6004 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
6005 | unit = MINOR(dev); | |
6006 | } | |
6007 | if (rdev0->preferred_minor != unit) { | |
9d48739e N |
6008 | pr_warn("md: unit number in %s is bad: %d\n", |
6009 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
1da177e4 LT |
6010 | break; |
6011 | } | |
1da177e4 LT |
6012 | |
6013 | md_probe(dev, NULL, NULL); | |
6014 | mddev = mddev_find(dev); | |
9bbbca3a NB |
6015 | if (!mddev || !mddev->gendisk) { |
6016 | if (mddev) | |
6017 | mddev_put(mddev); | |
1da177e4 LT |
6018 | break; |
6019 | } | |
f72ffdd6 | 6020 | if (mddev_lock(mddev)) |
9d48739e | 6021 | pr_warn("md: %s locked, cannot run\n", mdname(mddev)); |
1da177e4 LT |
6022 | else if (mddev->raid_disks || mddev->major_version |
6023 | || !list_empty(&mddev->disks)) { | |
9d48739e | 6024 | pr_warn("md: %s already running, cannot run %s\n", |
1da177e4 LT |
6025 | mdname(mddev), bdevname(rdev0->bdev,b)); |
6026 | mddev_unlock(mddev); | |
6027 | } else { | |
9d48739e | 6028 | pr_debug("md: created %s\n", mdname(mddev)); |
1ec4a939 | 6029 | mddev->persistent = 1; |
159ec1fc | 6030 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
6031 | list_del_init(&rdev->same_set); |
6032 | if (bind_rdev_to_array(rdev, mddev)) | |
6033 | export_rdev(rdev); | |
6034 | } | |
6035 | autorun_array(mddev); | |
6036 | mddev_unlock(mddev); | |
6037 | } | |
6038 | /* on success, candidates will be empty, on error | |
6039 | * it won't... | |
6040 | */ | |
159ec1fc | 6041 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 6042 | list_del_init(&rdev->same_set); |
1da177e4 | 6043 | export_rdev(rdev); |
4b80991c | 6044 | } |
1da177e4 LT |
6045 | mddev_put(mddev); |
6046 | } | |
9d48739e | 6047 | pr_info("md: ... autorun DONE.\n"); |
1da177e4 | 6048 | } |
fdee8ae4 | 6049 | #endif /* !MODULE */ |
1da177e4 | 6050 | |
f72ffdd6 | 6051 | static int get_version(void __user *arg) |
1da177e4 LT |
6052 | { |
6053 | mdu_version_t ver; | |
6054 | ||
6055 | ver.major = MD_MAJOR_VERSION; | |
6056 | ver.minor = MD_MINOR_VERSION; | |
6057 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
6058 | ||
6059 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
6060 | return -EFAULT; | |
6061 | ||
6062 | return 0; | |
6063 | } | |
6064 | ||
f72ffdd6 | 6065 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
6066 | { |
6067 | mdu_array_info_t info; | |
a9f326eb | 6068 | int nr,working,insync,failed,spare; |
3cb03002 | 6069 | struct md_rdev *rdev; |
1da177e4 | 6070 | |
1ca69c4b N |
6071 | nr = working = insync = failed = spare = 0; |
6072 | rcu_read_lock(); | |
6073 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6074 | nr++; |
b2d444d7 | 6075 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
6076 | failed++; |
6077 | else { | |
6078 | working++; | |
b2d444d7 | 6079 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 6080 | insync++; |
b347af81 SL |
6081 | else if (test_bit(Journal, &rdev->flags)) |
6082 | /* TODO: add journal count to md_u.h */ | |
6083 | ; | |
1da177e4 LT |
6084 | else |
6085 | spare++; | |
6086 | } | |
6087 | } | |
1ca69c4b | 6088 | rcu_read_unlock(); |
1da177e4 LT |
6089 | |
6090 | info.major_version = mddev->major_version; | |
6091 | info.minor_version = mddev->minor_version; | |
6092 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6093 | info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 6094 | info.level = mddev->level; |
58c0fed4 AN |
6095 | info.size = mddev->dev_sectors / 2; |
6096 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 6097 | info.size = -1; |
1da177e4 LT |
6098 | info.nr_disks = nr; |
6099 | info.raid_disks = mddev->raid_disks; | |
6100 | info.md_minor = mddev->md_minor; | |
6101 | info.not_persistent= !mddev->persistent; | |
6102 | ||
9ebc6ef1 | 6103 | info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
6104 | info.state = 0; |
6105 | if (mddev->in_sync) | |
6106 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 6107 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 6108 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
ca8895d9 GR |
6109 | if (mddev_is_clustered(mddev)) |
6110 | info.state |= (1<<MD_SB_CLUSTERED); | |
a9f326eb | 6111 | info.active_disks = insync; |
1da177e4 LT |
6112 | info.working_disks = working; |
6113 | info.failed_disks = failed; | |
6114 | info.spare_disks = spare; | |
6115 | ||
6116 | info.layout = mddev->layout; | |
9d8f0363 | 6117 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
6118 | |
6119 | if (copy_to_user(arg, &info, sizeof(info))) | |
6120 | return -EFAULT; | |
6121 | ||
6122 | return 0; | |
6123 | } | |
6124 | ||
f72ffdd6 | 6125 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
6126 | { |
6127 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 6128 | char *ptr; |
4af1a041 | 6129 | int err; |
32a7627c | 6130 | |
b6878d9e | 6131 | file = kzalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 6132 | if (!file) |
4af1a041 | 6133 | return -ENOMEM; |
32a7627c | 6134 | |
4af1a041 N |
6135 | err = 0; |
6136 | spin_lock(&mddev->lock); | |
25eafe1a BR |
6137 | /* bitmap enabled */ |
6138 | if (mddev->bitmap_info.file) { | |
6139 | ptr = file_path(mddev->bitmap_info.file, file->pathname, | |
6140 | sizeof(file->pathname)); | |
6141 | if (IS_ERR(ptr)) | |
6142 | err = PTR_ERR(ptr); | |
6143 | else | |
6144 | memmove(file->pathname, ptr, | |
6145 | sizeof(file->pathname)-(ptr-file->pathname)); | |
6146 | } | |
4af1a041 | 6147 | spin_unlock(&mddev->lock); |
32a7627c | 6148 | |
4af1a041 N |
6149 | if (err == 0 && |
6150 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 6151 | err = -EFAULT; |
4af1a041 | 6152 | |
32a7627c N |
6153 | kfree(file); |
6154 | return err; | |
6155 | } | |
6156 | ||
f72ffdd6 | 6157 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
6158 | { |
6159 | mdu_disk_info_t info; | |
3cb03002 | 6160 | struct md_rdev *rdev; |
1da177e4 LT |
6161 | |
6162 | if (copy_from_user(&info, arg, sizeof(info))) | |
6163 | return -EFAULT; | |
6164 | ||
1ca69c4b | 6165 | rcu_read_lock(); |
57d051dc | 6166 | rdev = md_find_rdev_nr_rcu(mddev, info.number); |
1da177e4 LT |
6167 | if (rdev) { |
6168 | info.major = MAJOR(rdev->bdev->bd_dev); | |
6169 | info.minor = MINOR(rdev->bdev->bd_dev); | |
6170 | info.raid_disk = rdev->raid_disk; | |
6171 | info.state = 0; | |
b2d444d7 | 6172 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6173 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 6174 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
6175 | info.state |= (1<<MD_DISK_ACTIVE); |
6176 | info.state |= (1<<MD_DISK_SYNC); | |
6177 | } | |
9efdca16 | 6178 | if (test_bit(Journal, &rdev->flags)) |
bac624f3 | 6179 | info.state |= (1<<MD_DISK_JOURNAL); |
8ddf9efe N |
6180 | if (test_bit(WriteMostly, &rdev->flags)) |
6181 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
6182 | if (test_bit(FailFast, &rdev->flags)) |
6183 | info.state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 LT |
6184 | } else { |
6185 | info.major = info.minor = 0; | |
6186 | info.raid_disk = -1; | |
6187 | info.state = (1<<MD_DISK_REMOVED); | |
6188 | } | |
1ca69c4b | 6189 | rcu_read_unlock(); |
1da177e4 LT |
6190 | |
6191 | if (copy_to_user(arg, &info, sizeof(info))) | |
6192 | return -EFAULT; | |
6193 | ||
6194 | return 0; | |
6195 | } | |
6196 | ||
f72ffdd6 | 6197 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
6198 | { |
6199 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 6200 | struct md_rdev *rdev; |
1da177e4 LT |
6201 | dev_t dev = MKDEV(info->major,info->minor); |
6202 | ||
1aee41f6 GR |
6203 | if (mddev_is_clustered(mddev) && |
6204 | !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) { | |
9d48739e N |
6205 | pr_warn("%s: Cannot add to clustered mddev.\n", |
6206 | mdname(mddev)); | |
1aee41f6 GR |
6207 | return -EINVAL; |
6208 | } | |
6209 | ||
1da177e4 LT |
6210 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) |
6211 | return -EOVERFLOW; | |
6212 | ||
6213 | if (!mddev->raid_disks) { | |
6214 | int err; | |
6215 | /* expecting a device which has a superblock */ | |
6216 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
6217 | if (IS_ERR(rdev)) { | |
9d48739e | 6218 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6219 | PTR_ERR(rdev)); |
6220 | return PTR_ERR(rdev); | |
6221 | } | |
6222 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
6223 | struct md_rdev *rdev0 |
6224 | = list_entry(mddev->disks.next, | |
6225 | struct md_rdev, same_set); | |
a9f326eb | 6226 | err = super_types[mddev->major_version] |
1da177e4 LT |
6227 | .load_super(rdev, rdev0, mddev->minor_version); |
6228 | if (err < 0) { | |
9d48739e | 6229 | pr_warn("md: %s has different UUID to %s\n", |
f72ffdd6 | 6230 | bdevname(rdev->bdev,b), |
1da177e4 LT |
6231 | bdevname(rdev0->bdev,b2)); |
6232 | export_rdev(rdev); | |
6233 | return -EINVAL; | |
6234 | } | |
6235 | } | |
6236 | err = bind_rdev_to_array(rdev, mddev); | |
6237 | if (err) | |
6238 | export_rdev(rdev); | |
6239 | return err; | |
6240 | } | |
6241 | ||
6242 | /* | |
6243 | * add_new_disk can be used once the array is assembled | |
6244 | * to add "hot spares". They must already have a superblock | |
6245 | * written | |
6246 | */ | |
6247 | if (mddev->pers) { | |
6248 | int err; | |
6249 | if (!mddev->pers->hot_add_disk) { | |
9d48739e N |
6250 | pr_warn("%s: personality does not support diskops!\n", |
6251 | mdname(mddev)); | |
1da177e4 LT |
6252 | return -EINVAL; |
6253 | } | |
7b1e35f6 N |
6254 | if (mddev->persistent) |
6255 | rdev = md_import_device(dev, mddev->major_version, | |
6256 | mddev->minor_version); | |
6257 | else | |
6258 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 6259 | if (IS_ERR(rdev)) { |
9d48739e | 6260 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6261 | PTR_ERR(rdev)); |
6262 | return PTR_ERR(rdev); | |
6263 | } | |
1a855a06 | 6264 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
6265 | if (!mddev->persistent) { |
6266 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 6267 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 6268 | rdev->raid_disk = info->raid_disk; |
bf572541 | 6269 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 6270 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 6271 | } else |
41158c7e | 6272 | rdev->raid_disk = -1; |
f466722c | 6273 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
6274 | } else |
6275 | super_types[mddev->major_version]. | |
6276 | validate_super(mddev, rdev); | |
bedd86b7 | 6277 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 6278 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
6279 | /* This was a hot-add request, but events doesn't |
6280 | * match, so reject it. | |
6281 | */ | |
6282 | export_rdev(rdev); | |
6283 | return -EINVAL; | |
6284 | } | |
6285 | ||
b2d444d7 | 6286 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
6287 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6288 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
6289 | else |
6290 | clear_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6291 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6292 | set_bit(FailFast, &rdev->flags); | |
6293 | else | |
6294 | clear_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6295 | |
f6b6ec5c SL |
6296 | if (info->state & (1<<MD_DISK_JOURNAL)) { |
6297 | struct md_rdev *rdev2; | |
6298 | bool has_journal = false; | |
6299 | ||
6300 | /* make sure no existing journal disk */ | |
6301 | rdev_for_each(rdev2, mddev) { | |
6302 | if (test_bit(Journal, &rdev2->flags)) { | |
6303 | has_journal = true; | |
6304 | break; | |
6305 | } | |
6306 | } | |
6307 | if (has_journal) { | |
6308 | export_rdev(rdev); | |
6309 | return -EBUSY; | |
6310 | } | |
bac624f3 | 6311 | set_bit(Journal, &rdev->flags); |
f6b6ec5c | 6312 | } |
1aee41f6 GR |
6313 | /* |
6314 | * check whether the device shows up in other nodes | |
6315 | */ | |
6316 | if (mddev_is_clustered(mddev)) { | |
dbb64f86 | 6317 | if (info->state & (1 << MD_DISK_CANDIDATE)) |
1aee41f6 | 6318 | set_bit(Candidate, &rdev->flags); |
dbb64f86 | 6319 | else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) { |
1aee41f6 | 6320 | /* --add initiated by this node */ |
dbb64f86 | 6321 | err = md_cluster_ops->add_new_disk(mddev, rdev); |
1aee41f6 | 6322 | if (err) { |
1aee41f6 GR |
6323 | export_rdev(rdev); |
6324 | return err; | |
6325 | } | |
6326 | } | |
6327 | } | |
6328 | ||
1da177e4 LT |
6329 | rdev->raid_disk = -1; |
6330 | err = bind_rdev_to_array(rdev, mddev); | |
dbb64f86 | 6331 | |
1da177e4 LT |
6332 | if (err) |
6333 | export_rdev(rdev); | |
dbb64f86 GR |
6334 | |
6335 | if (mddev_is_clustered(mddev)) { | |
e566aef1 GJ |
6336 | if (info->state & (1 << MD_DISK_CANDIDATE)) { |
6337 | if (!err) { | |
6338 | err = md_cluster_ops->new_disk_ack(mddev, | |
6339 | err == 0); | |
6340 | if (err) | |
6341 | md_kick_rdev_from_array(rdev); | |
6342 | } | |
6343 | } else { | |
dbb64f86 GR |
6344 | if (err) |
6345 | md_cluster_ops->add_new_disk_cancel(mddev); | |
6346 | else | |
6347 | err = add_bound_rdev(rdev); | |
6348 | } | |
6349 | ||
6350 | } else if (!err) | |
a6da4ef8 | 6351 | err = add_bound_rdev(rdev); |
dbb64f86 | 6352 | |
1da177e4 LT |
6353 | return err; |
6354 | } | |
6355 | ||
6356 | /* otherwise, add_new_disk is only allowed | |
6357 | * for major_version==0 superblocks | |
6358 | */ | |
6359 | if (mddev->major_version != 0) { | |
9d48739e | 6360 | pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev)); |
1da177e4 LT |
6361 | return -EINVAL; |
6362 | } | |
6363 | ||
6364 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
6365 | int err; | |
d710e138 | 6366 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6367 | if (IS_ERR(rdev)) { |
9d48739e | 6368 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6369 | PTR_ERR(rdev)); |
6370 | return PTR_ERR(rdev); | |
6371 | } | |
6372 | rdev->desc_nr = info->number; | |
6373 | if (info->raid_disk < mddev->raid_disks) | |
6374 | rdev->raid_disk = info->raid_disk; | |
6375 | else | |
6376 | rdev->raid_disk = -1; | |
6377 | ||
1da177e4 | 6378 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
6379 | if (info->state & (1<<MD_DISK_SYNC)) |
6380 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 6381 | |
8ddf9efe N |
6382 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6383 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6384 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6385 | set_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6386 | |
1da177e4 | 6387 | if (!mddev->persistent) { |
9d48739e | 6388 | pr_debug("md: nonpersistent superblock ...\n"); |
77304d2a MS |
6389 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
6390 | } else | |
57b2caa3 | 6391 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 6392 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6393 | |
2bf071bf N |
6394 | err = bind_rdev_to_array(rdev, mddev); |
6395 | if (err) { | |
6396 | export_rdev(rdev); | |
6397 | return err; | |
6398 | } | |
1da177e4 LT |
6399 | } |
6400 | ||
6401 | return 0; | |
6402 | } | |
6403 | ||
f72ffdd6 | 6404 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6405 | { |
6406 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 6407 | struct md_rdev *rdev; |
1da177e4 | 6408 | |
1da177e4 LT |
6409 | rdev = find_rdev(mddev, dev); |
6410 | if (!rdev) | |
6411 | return -ENXIO; | |
6412 | ||
2910ff17 GR |
6413 | if (rdev->raid_disk < 0) |
6414 | goto kick_rdev; | |
293467aa | 6415 | |
3ea8929d N |
6416 | clear_bit(Blocked, &rdev->flags); |
6417 | remove_and_add_spares(mddev, rdev); | |
6418 | ||
1da177e4 LT |
6419 | if (rdev->raid_disk >= 0) |
6420 | goto busy; | |
6421 | ||
2910ff17 | 6422 | kick_rdev: |
54a88392 | 6423 | if (mddev_is_clustered(mddev)) |
88bcfef7 GR |
6424 | md_cluster_ops->remove_disk(mddev, rdev); |
6425 | ||
fb56dfef | 6426 | md_kick_rdev_from_array(rdev); |
2953079c | 6427 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6428 | if (mddev->thread) |
6429 | md_wakeup_thread(mddev->thread); | |
6430 | else | |
6431 | md_update_sb(mddev, 1); | |
d7603b7e | 6432 | md_new_event(mddev); |
1da177e4 LT |
6433 | |
6434 | return 0; | |
6435 | busy: | |
9d48739e N |
6436 | pr_debug("md: cannot remove active disk %s from %s ...\n", |
6437 | bdevname(rdev->bdev,b), mdname(mddev)); | |
1da177e4 LT |
6438 | return -EBUSY; |
6439 | } | |
6440 | ||
f72ffdd6 | 6441 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6442 | { |
6443 | char b[BDEVNAME_SIZE]; | |
6444 | int err; | |
3cb03002 | 6445 | struct md_rdev *rdev; |
1da177e4 LT |
6446 | |
6447 | if (!mddev->pers) | |
6448 | return -ENODEV; | |
6449 | ||
6450 | if (mddev->major_version != 0) { | |
9d48739e | 6451 | pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n", |
1da177e4 LT |
6452 | mdname(mddev)); |
6453 | return -EINVAL; | |
6454 | } | |
6455 | if (!mddev->pers->hot_add_disk) { | |
9d48739e | 6456 | pr_warn("%s: personality does not support diskops!\n", |
1da177e4 LT |
6457 | mdname(mddev)); |
6458 | return -EINVAL; | |
6459 | } | |
6460 | ||
d710e138 | 6461 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6462 | if (IS_ERR(rdev)) { |
9d48739e | 6463 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6464 | PTR_ERR(rdev)); |
6465 | return -EINVAL; | |
6466 | } | |
6467 | ||
6468 | if (mddev->persistent) | |
57b2caa3 | 6469 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 6470 | else |
77304d2a | 6471 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 6472 | |
8190e754 | 6473 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6474 | |
b2d444d7 | 6475 | if (test_bit(Faulty, &rdev->flags)) { |
9d48739e | 6476 | pr_warn("md: can not hot-add faulty %s disk to %s!\n", |
1da177e4 LT |
6477 | bdevname(rdev->bdev,b), mdname(mddev)); |
6478 | err = -EINVAL; | |
6479 | goto abort_export; | |
6480 | } | |
293467aa | 6481 | |
b2d444d7 | 6482 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6483 | rdev->desc_nr = -1; |
5842730d | 6484 | rdev->saved_raid_disk = -1; |
2bf071bf N |
6485 | err = bind_rdev_to_array(rdev, mddev); |
6486 | if (err) | |
2aa82191 | 6487 | goto abort_export; |
1da177e4 LT |
6488 | |
6489 | /* | |
6490 | * The rest should better be atomic, we can have disk failures | |
6491 | * noticed in interrupt contexts ... | |
6492 | */ | |
6493 | ||
1da177e4 LT |
6494 | rdev->raid_disk = -1; |
6495 | ||
2953079c | 6496 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6497 | if (!mddev->thread) |
6498 | md_update_sb(mddev, 1); | |
1da177e4 LT |
6499 | /* |
6500 | * Kick recovery, maybe this spare has to be added to the | |
6501 | * array immediately. | |
6502 | */ | |
6503 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6504 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 6505 | md_new_event(mddev); |
1da177e4 LT |
6506 | return 0; |
6507 | ||
1da177e4 LT |
6508 | abort_export: |
6509 | export_rdev(rdev); | |
6510 | return err; | |
6511 | } | |
6512 | ||
fd01b88c | 6513 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 6514 | { |
035328c2 | 6515 | int err = 0; |
32a7627c | 6516 | |
36fa3063 | 6517 | if (mddev->pers) { |
d66b1b39 | 6518 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
6519 | return -EBUSY; |
6520 | if (mddev->recovery || mddev->sync_thread) | |
6521 | return -EBUSY; | |
6522 | /* we should be able to change the bitmap.. */ | |
6523 | } | |
32a7627c | 6524 | |
36fa3063 | 6525 | if (fd >= 0) { |
035328c2 | 6526 | struct inode *inode; |
1e594bb2 N |
6527 | struct file *f; |
6528 | ||
6529 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 6530 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6531 | f = fget(fd); |
32a7627c | 6532 | |
1e594bb2 | 6533 | if (f == NULL) { |
9d48739e N |
6534 | pr_warn("%s: error: failed to get bitmap file\n", |
6535 | mdname(mddev)); | |
36fa3063 N |
6536 | return -EBADF; |
6537 | } | |
6538 | ||
1e594bb2 | 6539 | inode = f->f_mapping->host; |
035328c2 | 6540 | if (!S_ISREG(inode->i_mode)) { |
9d48739e N |
6541 | pr_warn("%s: error: bitmap file must be a regular file\n", |
6542 | mdname(mddev)); | |
035328c2 | 6543 | err = -EBADF; |
1e594bb2 | 6544 | } else if (!(f->f_mode & FMODE_WRITE)) { |
9d48739e N |
6545 | pr_warn("%s: error: bitmap file must open for write\n", |
6546 | mdname(mddev)); | |
035328c2 N |
6547 | err = -EBADF; |
6548 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
9d48739e N |
6549 | pr_warn("%s: error: bitmap file is already in use\n", |
6550 | mdname(mddev)); | |
035328c2 N |
6551 | err = -EBUSY; |
6552 | } | |
6553 | if (err) { | |
1e594bb2 | 6554 | fput(f); |
36fa3063 N |
6555 | return err; |
6556 | } | |
1e594bb2 | 6557 | mddev->bitmap_info.file = f; |
c3d9714e | 6558 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6559 | } else if (mddev->bitmap == NULL) |
6560 | return -ENOENT; /* cannot remove what isn't there */ | |
6561 | err = 0; | |
6562 | if (mddev->pers) { | |
6563 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6564 | if (fd >= 0) { |
f9209a32 GR |
6565 | struct bitmap *bitmap; |
6566 | ||
6567 | bitmap = bitmap_create(mddev, -1); | |
6568 | if (!IS_ERR(bitmap)) { | |
6569 | mddev->bitmap = bitmap; | |
69e51b44 | 6570 | err = bitmap_load(mddev); |
ba599aca N |
6571 | } else |
6572 | err = PTR_ERR(bitmap); | |
69e51b44 | 6573 | } |
d7375ab3 | 6574 | if (fd < 0 || err) { |
36fa3063 | 6575 | bitmap_destroy(mddev); |
d7375ab3 N |
6576 | fd = -1; /* make sure to put the file */ |
6577 | } | |
36fa3063 | 6578 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6579 | } |
6580 | if (fd < 0) { | |
4af1a041 N |
6581 | struct file *f = mddev->bitmap_info.file; |
6582 | if (f) { | |
6583 | spin_lock(&mddev->lock); | |
6584 | mddev->bitmap_info.file = NULL; | |
6585 | spin_unlock(&mddev->lock); | |
6586 | fput(f); | |
6587 | } | |
36fa3063 N |
6588 | } |
6589 | ||
32a7627c N |
6590 | return err; |
6591 | } | |
6592 | ||
1da177e4 LT |
6593 | /* |
6594 | * set_array_info is used two different ways | |
6595 | * The original usage is when creating a new array. | |
6596 | * In this usage, raid_disks is > 0 and it together with | |
6597 | * level, size, not_persistent,layout,chunksize determine the | |
6598 | * shape of the array. | |
6599 | * This will always create an array with a type-0.90.0 superblock. | |
6600 | * The newer usage is when assembling an array. | |
6601 | * In this case raid_disks will be 0, and the major_version field is | |
6602 | * use to determine which style super-blocks are to be found on the devices. | |
6603 | * The minor and patch _version numbers are also kept incase the | |
6604 | * super_block handler wishes to interpret them. | |
6605 | */ | |
f72ffdd6 | 6606 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6607 | { |
6608 | ||
6609 | if (info->raid_disks == 0) { | |
6610 | /* just setting version number for superblock loading */ | |
6611 | if (info->major_version < 0 || | |
50511da3 | 6612 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6613 | super_types[info->major_version].name == NULL) { |
6614 | /* maybe try to auto-load a module? */ | |
9d48739e | 6615 | pr_warn("md: superblock version %d not known\n", |
1da177e4 LT |
6616 | info->major_version); |
6617 | return -EINVAL; | |
6618 | } | |
6619 | mddev->major_version = info->major_version; | |
6620 | mddev->minor_version = info->minor_version; | |
6621 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6622 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6623 | /* ensure mddev_put doesn't delete this now that there |
6624 | * is some minimal configuration. | |
6625 | */ | |
9ebc6ef1 | 6626 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6627 | return 0; |
6628 | } | |
6629 | mddev->major_version = MD_MAJOR_VERSION; | |
6630 | mddev->minor_version = MD_MINOR_VERSION; | |
6631 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6632 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6633 | |
6634 | mddev->level = info->level; | |
17115e03 | 6635 | mddev->clevel[0] = 0; |
58c0fed4 | 6636 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6637 | mddev->raid_disks = info->raid_disks; |
6638 | /* don't set md_minor, it is determined by which /dev/md* was | |
6639 | * openned | |
6640 | */ | |
6641 | if (info->state & (1<<MD_SB_CLEAN)) | |
6642 | mddev->recovery_cp = MaxSector; | |
6643 | else | |
6644 | mddev->recovery_cp = 0; | |
6645 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6646 | mddev->external = 0; |
1da177e4 LT |
6647 | |
6648 | mddev->layout = info->layout; | |
9d8f0363 | 6649 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 | 6650 | |
2953079c | 6651 | if (mddev->persistent) { |
1b3bae49 N |
6652 | mddev->max_disks = MD_SB_DISKS; |
6653 | mddev->flags = 0; | |
6654 | mddev->sb_flags = 0; | |
2953079c SL |
6655 | } |
6656 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
1da177e4 | 6657 | |
c3d9714e | 6658 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6659 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6660 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6661 | |
f6705578 N |
6662 | mddev->reshape_position = MaxSector; |
6663 | ||
1da177e4 LT |
6664 | /* |
6665 | * Generate a 128 bit UUID | |
6666 | */ | |
6667 | get_random_bytes(mddev->uuid, 16); | |
6668 | ||
f6705578 | 6669 | mddev->new_level = mddev->level; |
664e7c41 | 6670 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6671 | mddev->new_layout = mddev->layout; |
6672 | mddev->delta_disks = 0; | |
2c810cdd | 6673 | mddev->reshape_backwards = 0; |
f6705578 | 6674 | |
1da177e4 LT |
6675 | return 0; |
6676 | } | |
6677 | ||
fd01b88c | 6678 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6679 | { |
b522adcd DW |
6680 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6681 | ||
6682 | if (mddev->external_size) | |
6683 | return; | |
6684 | ||
1f403624 DW |
6685 | mddev->array_sectors = array_sectors; |
6686 | } | |
6687 | EXPORT_SYMBOL(md_set_array_sectors); | |
6688 | ||
fd01b88c | 6689 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6690 | { |
3cb03002 | 6691 | struct md_rdev *rdev; |
a35b0d69 | 6692 | int rv; |
d71f9f88 | 6693 | int fit = (num_sectors == 0); |
818da59f | 6694 | sector_t old_dev_sectors = mddev->dev_sectors; |
ab5a98b1 | 6695 | |
a35b0d69 N |
6696 | if (mddev->pers->resize == NULL) |
6697 | return -EINVAL; | |
d71f9f88 AN |
6698 | /* The "num_sectors" is the number of sectors of each device that |
6699 | * is used. This can only make sense for arrays with redundancy. | |
6700 | * linear and raid0 always use whatever space is available. We can only | |
6701 | * consider changing this number if no resync or reconstruction is | |
6702 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6703 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6704 | * of each device. If num_sectors is zero, we find the largest size |
6705 | * that fits. | |
a35b0d69 | 6706 | */ |
f851b60d N |
6707 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6708 | mddev->sync_thread) | |
a35b0d69 | 6709 | return -EBUSY; |
bd8839e0 N |
6710 | if (mddev->ro) |
6711 | return -EROFS; | |
a4a6125a | 6712 | |
dafb20fa | 6713 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6714 | sector_t avail = rdev->sectors; |
01ab5662 | 6715 | |
d71f9f88 AN |
6716 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6717 | num_sectors = avail; | |
6718 | if (avail < num_sectors) | |
a35b0d69 N |
6719 | return -ENOSPC; |
6720 | } | |
d71f9f88 | 6721 | rv = mddev->pers->resize(mddev, num_sectors); |
c9483634 | 6722 | if (!rv) { |
818da59f GJ |
6723 | if (mddev_is_clustered(mddev)) |
6724 | md_cluster_ops->update_size(mddev, old_dev_sectors); | |
6725 | else if (mddev->queue) { | |
c9483634 GJ |
6726 | set_capacity(mddev->gendisk, mddev->array_sectors); |
6727 | revalidate_disk(mddev->gendisk); | |
6728 | } | |
6729 | } | |
a35b0d69 N |
6730 | return rv; |
6731 | } | |
6732 | ||
fd01b88c | 6733 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6734 | { |
6735 | int rv; | |
c6563a8c | 6736 | struct md_rdev *rdev; |
da943b99 | 6737 | /* change the number of raid disks */ |
63c70c4f | 6738 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6739 | return -EINVAL; |
bd8839e0 N |
6740 | if (mddev->ro) |
6741 | return -EROFS; | |
da943b99 | 6742 | if (raid_disks <= 0 || |
233fca36 | 6743 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6744 | return -EINVAL; |
f851b60d N |
6745 | if (mddev->sync_thread || |
6746 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6747 | mddev->reshape_position != MaxSector) | |
da943b99 | 6748 | return -EBUSY; |
c6563a8c N |
6749 | |
6750 | rdev_for_each(rdev, mddev) { | |
6751 | if (mddev->raid_disks < raid_disks && | |
6752 | rdev->data_offset < rdev->new_data_offset) | |
6753 | return -EINVAL; | |
6754 | if (mddev->raid_disks > raid_disks && | |
6755 | rdev->data_offset > rdev->new_data_offset) | |
6756 | return -EINVAL; | |
6757 | } | |
6758 | ||
63c70c4f | 6759 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6760 | if (mddev->delta_disks < 0) |
6761 | mddev->reshape_backwards = 1; | |
6762 | else if (mddev->delta_disks > 0) | |
6763 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6764 | |
6765 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6766 | if (rv < 0) { |
de171cb9 | 6767 | mddev->delta_disks = 0; |
2c810cdd N |
6768 | mddev->reshape_backwards = 0; |
6769 | } | |
da943b99 N |
6770 | return rv; |
6771 | } | |
6772 | ||
1da177e4 LT |
6773 | /* |
6774 | * update_array_info is used to change the configuration of an | |
6775 | * on-line array. | |
6776 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6777 | * fields in the info are checked against the array. | |
6778 | * Any differences that cannot be handled will cause an error. | |
6779 | * Normally, only one change can be managed at a time. | |
6780 | */ | |
fd01b88c | 6781 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6782 | { |
6783 | int rv = 0; | |
6784 | int cnt = 0; | |
36fa3063 N |
6785 | int state = 0; |
6786 | ||
6787 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6788 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6789 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6790 | |
6791 | if (mddev->major_version != info->major_version || | |
6792 | mddev->minor_version != info->minor_version || | |
6793 | /* mddev->patch_version != info->patch_version || */ | |
6794 | mddev->ctime != info->ctime || | |
6795 | mddev->level != info->level || | |
6796 | /* mddev->layout != info->layout || */ | |
4e023612 | 6797 | mddev->persistent != !info->not_persistent || |
9d8f0363 | 6798 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6799 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6800 | ((state^info->state) & 0xfffffe00) | |
6801 | ) | |
1da177e4 LT |
6802 | return -EINVAL; |
6803 | /* Check there is only one change */ | |
58c0fed4 AN |
6804 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6805 | cnt++; | |
6806 | if (mddev->raid_disks != info->raid_disks) | |
6807 | cnt++; | |
6808 | if (mddev->layout != info->layout) | |
6809 | cnt++; | |
6810 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6811 | cnt++; | |
6812 | if (cnt == 0) | |
6813 | return 0; | |
6814 | if (cnt > 1) | |
6815 | return -EINVAL; | |
1da177e4 LT |
6816 | |
6817 | if (mddev->layout != info->layout) { | |
6818 | /* Change layout | |
6819 | * we don't need to do anything at the md level, the | |
6820 | * personality will take care of it all. | |
6821 | */ | |
50ac168a | 6822 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6823 | return -EINVAL; |
597a711b N |
6824 | else { |
6825 | mddev->new_layout = info->layout; | |
50ac168a | 6826 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6827 | if (rv) |
6828 | mddev->new_layout = mddev->layout; | |
6829 | return rv; | |
6830 | } | |
1da177e4 | 6831 | } |
58c0fed4 | 6832 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6833 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6834 | |
da943b99 N |
6835 | if (mddev->raid_disks != info->raid_disks) |
6836 | rv = update_raid_disks(mddev, info->raid_disks); | |
6837 | ||
36fa3063 | 6838 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
293467aa GR |
6839 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) { |
6840 | rv = -EINVAL; | |
6841 | goto err; | |
6842 | } | |
6843 | if (mddev->recovery || mddev->sync_thread) { | |
6844 | rv = -EBUSY; | |
6845 | goto err; | |
6846 | } | |
36fa3063 | 6847 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { |
f9209a32 | 6848 | struct bitmap *bitmap; |
36fa3063 | 6849 | /* add the bitmap */ |
293467aa GR |
6850 | if (mddev->bitmap) { |
6851 | rv = -EEXIST; | |
6852 | goto err; | |
6853 | } | |
6854 | if (mddev->bitmap_info.default_offset == 0) { | |
6855 | rv = -EINVAL; | |
6856 | goto err; | |
6857 | } | |
c3d9714e N |
6858 | mddev->bitmap_info.offset = |
6859 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6860 | mddev->bitmap_info.space = |
6861 | mddev->bitmap_info.default_space; | |
36fa3063 | 6862 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
6863 | bitmap = bitmap_create(mddev, -1); |
6864 | if (!IS_ERR(bitmap)) { | |
6865 | mddev->bitmap = bitmap; | |
69e51b44 | 6866 | rv = bitmap_load(mddev); |
ba599aca N |
6867 | } else |
6868 | rv = PTR_ERR(bitmap); | |
36fa3063 N |
6869 | if (rv) |
6870 | bitmap_destroy(mddev); | |
6871 | mddev->pers->quiesce(mddev, 0); | |
6872 | } else { | |
6873 | /* remove the bitmap */ | |
293467aa GR |
6874 | if (!mddev->bitmap) { |
6875 | rv = -ENOENT; | |
6876 | goto err; | |
6877 | } | |
6878 | if (mddev->bitmap->storage.file) { | |
6879 | rv = -EINVAL; | |
6880 | goto err; | |
6881 | } | |
f6a2dc64 GJ |
6882 | if (mddev->bitmap_info.nodes) { |
6883 | /* hold PW on all the bitmap lock */ | |
6884 | if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) { | |
9d48739e | 6885 | pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n"); |
f6a2dc64 GJ |
6886 | rv = -EPERM; |
6887 | md_cluster_ops->unlock_all_bitmaps(mddev); | |
6888 | goto err; | |
6889 | } | |
6890 | ||
6891 | mddev->bitmap_info.nodes = 0; | |
6892 | md_cluster_ops->leave(mddev); | |
6893 | } | |
36fa3063 N |
6894 | mddev->pers->quiesce(mddev, 1); |
6895 | bitmap_destroy(mddev); | |
6896 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6897 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6898 | } |
6899 | } | |
850b2b42 | 6900 | md_update_sb(mddev, 1); |
293467aa GR |
6901 | return rv; |
6902 | err: | |
1da177e4 LT |
6903 | return rv; |
6904 | } | |
6905 | ||
fd01b88c | 6906 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6907 | { |
3cb03002 | 6908 | struct md_rdev *rdev; |
1ca69c4b | 6909 | int err = 0; |
1da177e4 LT |
6910 | |
6911 | if (mddev->pers == NULL) | |
6912 | return -ENODEV; | |
6913 | ||
1ca69c4b N |
6914 | rcu_read_lock(); |
6915 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6916 | if (!rdev) |
1ca69c4b N |
6917 | err = -ENODEV; |
6918 | else { | |
6919 | md_error(mddev, rdev); | |
6920 | if (!test_bit(Faulty, &rdev->flags)) | |
6921 | err = -EBUSY; | |
6922 | } | |
6923 | rcu_read_unlock(); | |
6924 | return err; | |
1da177e4 LT |
6925 | } |
6926 | ||
2f9618ce AN |
6927 | /* |
6928 | * We have a problem here : there is no easy way to give a CHS | |
6929 | * virtual geometry. We currently pretend that we have a 2 heads | |
6930 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6931 | * dosfs just mad... ;-) | |
6932 | */ | |
a885c8c4 CH |
6933 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6934 | { | |
fd01b88c | 6935 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6936 | |
6937 | geo->heads = 2; | |
6938 | geo->sectors = 4; | |
49ce6cea | 6939 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6940 | return 0; |
6941 | } | |
6942 | ||
cb335f88 NS |
6943 | static inline bool md_ioctl_valid(unsigned int cmd) |
6944 | { | |
6945 | switch (cmd) { | |
6946 | case ADD_NEW_DISK: | |
6947 | case BLKROSET: | |
6948 | case GET_ARRAY_INFO: | |
6949 | case GET_BITMAP_FILE: | |
6950 | case GET_DISK_INFO: | |
6951 | case HOT_ADD_DISK: | |
6952 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
6953 | case RAID_AUTORUN: |
6954 | case RAID_VERSION: | |
6955 | case RESTART_ARRAY_RW: | |
6956 | case RUN_ARRAY: | |
6957 | case SET_ARRAY_INFO: | |
6958 | case SET_BITMAP_FILE: | |
6959 | case SET_DISK_FAULTY: | |
6960 | case STOP_ARRAY: | |
6961 | case STOP_ARRAY_RO: | |
1aee41f6 | 6962 | case CLUSTERED_DISK_NACK: |
cb335f88 NS |
6963 | return true; |
6964 | default: | |
6965 | return false; | |
6966 | } | |
6967 | } | |
6968 | ||
a39907fa | 6969 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6970 | unsigned int cmd, unsigned long arg) |
6971 | { | |
6972 | int err = 0; | |
6973 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6974 | struct mddev *mddev = NULL; |
e2218350 | 6975 | int ro; |
065e519e | 6976 | bool did_set_md_closing = false; |
1da177e4 | 6977 | |
cb335f88 NS |
6978 | if (!md_ioctl_valid(cmd)) |
6979 | return -ENOTTY; | |
6980 | ||
506c9e44 N |
6981 | switch (cmd) { |
6982 | case RAID_VERSION: | |
6983 | case GET_ARRAY_INFO: | |
6984 | case GET_DISK_INFO: | |
6985 | break; | |
6986 | default: | |
6987 | if (!capable(CAP_SYS_ADMIN)) | |
6988 | return -EACCES; | |
6989 | } | |
1da177e4 LT |
6990 | |
6991 | /* | |
6992 | * Commands dealing with the RAID driver but not any | |
6993 | * particular array: | |
6994 | */ | |
c02c0aeb N |
6995 | switch (cmd) { |
6996 | case RAID_VERSION: | |
6997 | err = get_version(argp); | |
3adc28d8 | 6998 | goto out; |
1da177e4 | 6999 | |
1da177e4 | 7000 | #ifndef MODULE |
c02c0aeb N |
7001 | case RAID_AUTORUN: |
7002 | err = 0; | |
7003 | autostart_arrays(arg); | |
3adc28d8 | 7004 | goto out; |
1da177e4 | 7005 | #endif |
c02c0aeb | 7006 | default:; |
1da177e4 LT |
7007 | } |
7008 | ||
7009 | /* | |
7010 | * Commands creating/starting a new array: | |
7011 | */ | |
7012 | ||
a39907fa | 7013 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
7014 | |
7015 | if (!mddev) { | |
7016 | BUG(); | |
3adc28d8 | 7017 | goto out; |
1da177e4 LT |
7018 | } |
7019 | ||
1ca69c4b N |
7020 | /* Some actions do not requires the mutex */ |
7021 | switch (cmd) { | |
7022 | case GET_ARRAY_INFO: | |
7023 | if (!mddev->raid_disks && !mddev->external) | |
7024 | err = -ENODEV; | |
7025 | else | |
7026 | err = get_array_info(mddev, argp); | |
3adc28d8 | 7027 | goto out; |
1ca69c4b N |
7028 | |
7029 | case GET_DISK_INFO: | |
7030 | if (!mddev->raid_disks && !mddev->external) | |
7031 | err = -ENODEV; | |
7032 | else | |
7033 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 7034 | goto out; |
1ca69c4b N |
7035 | |
7036 | case SET_DISK_FAULTY: | |
7037 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7038 | goto out; |
4af1a041 N |
7039 | |
7040 | case GET_BITMAP_FILE: | |
7041 | err = get_bitmap_file(mddev, argp); | |
7042 | goto out; | |
7043 | ||
1ca69c4b N |
7044 | } |
7045 | ||
a7a3f08d N |
7046 | if (cmd == ADD_NEW_DISK) |
7047 | /* need to ensure md_delayed_delete() has completed */ | |
7048 | flush_workqueue(md_misc_wq); | |
7049 | ||
90f5f7ad HR |
7050 | if (cmd == HOT_REMOVE_DISK) |
7051 | /* need to ensure recovery thread has run */ | |
7052 | wait_event_interruptible_timeout(mddev->sb_wait, | |
7053 | !test_bit(MD_RECOVERY_NEEDED, | |
82a301cb | 7054 | &mddev->recovery), |
90f5f7ad | 7055 | msecs_to_jiffies(5000)); |
260fa034 N |
7056 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
7057 | /* Need to flush page cache, and ensure no-one else opens | |
7058 | * and writes | |
7059 | */ | |
7060 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 7061 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
7062 | mutex_unlock(&mddev->open_mutex); |
7063 | err = -EBUSY; | |
3adc28d8 | 7064 | goto out; |
260fa034 | 7065 | } |
065e519e | 7066 | WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags)); |
af8d8e6f | 7067 | set_bit(MD_CLOSING, &mddev->flags); |
065e519e | 7068 | did_set_md_closing = true; |
260fa034 N |
7069 | mutex_unlock(&mddev->open_mutex); |
7070 | sync_blockdev(bdev); | |
7071 | } | |
1da177e4 LT |
7072 | err = mddev_lock(mddev); |
7073 | if (err) { | |
9d48739e N |
7074 | pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n", |
7075 | err, cmd); | |
3adc28d8 | 7076 | goto out; |
1da177e4 LT |
7077 | } |
7078 | ||
c02c0aeb N |
7079 | if (cmd == SET_ARRAY_INFO) { |
7080 | mdu_array_info_t info; | |
7081 | if (!arg) | |
7082 | memset(&info, 0, sizeof(info)); | |
7083 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
7084 | err = -EFAULT; | |
3adc28d8 | 7085 | goto unlock; |
c02c0aeb N |
7086 | } |
7087 | if (mddev->pers) { | |
7088 | err = update_array_info(mddev, &info); | |
7089 | if (err) { | |
9d48739e | 7090 | pr_warn("md: couldn't update array info. %d\n", err); |
3adc28d8 | 7091 | goto unlock; |
1da177e4 | 7092 | } |
3adc28d8 | 7093 | goto unlock; |
c02c0aeb N |
7094 | } |
7095 | if (!list_empty(&mddev->disks)) { | |
9d48739e | 7096 | pr_warn("md: array %s already has disks!\n", mdname(mddev)); |
c02c0aeb | 7097 | err = -EBUSY; |
3adc28d8 | 7098 | goto unlock; |
c02c0aeb N |
7099 | } |
7100 | if (mddev->raid_disks) { | |
9d48739e | 7101 | pr_warn("md: array %s already initialised!\n", mdname(mddev)); |
c02c0aeb | 7102 | err = -EBUSY; |
3adc28d8 | 7103 | goto unlock; |
c02c0aeb N |
7104 | } |
7105 | err = set_array_info(mddev, &info); | |
7106 | if (err) { | |
9d48739e | 7107 | pr_warn("md: couldn't set array info. %d\n", err); |
3adc28d8 | 7108 | goto unlock; |
c02c0aeb | 7109 | } |
3adc28d8 | 7110 | goto unlock; |
1da177e4 LT |
7111 | } |
7112 | ||
7113 | /* | |
7114 | * Commands querying/configuring an existing array: | |
7115 | */ | |
32a7627c | 7116 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 7117 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
7118 | if ((!mddev->raid_disks && !mddev->external) |
7119 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
7120 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
7121 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 7122 | err = -ENODEV; |
3adc28d8 | 7123 | goto unlock; |
1da177e4 LT |
7124 | } |
7125 | ||
7126 | /* | |
7127 | * Commands even a read-only array can execute: | |
7128 | */ | |
c02c0aeb | 7129 | switch (cmd) { |
c02c0aeb N |
7130 | case RESTART_ARRAY_RW: |
7131 | err = restart_array(mddev); | |
3adc28d8 | 7132 | goto unlock; |
1da177e4 | 7133 | |
c02c0aeb N |
7134 | case STOP_ARRAY: |
7135 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 7136 | goto unlock; |
1da177e4 | 7137 | |
c02c0aeb N |
7138 | case STOP_ARRAY_RO: |
7139 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 7140 | goto unlock; |
1da177e4 | 7141 | |
3ea8929d N |
7142 | case HOT_REMOVE_DISK: |
7143 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7144 | goto unlock; |
3ea8929d | 7145 | |
7ceb17e8 N |
7146 | case ADD_NEW_DISK: |
7147 | /* We can support ADD_NEW_DISK on read-only arrays | |
466ad292 | 7148 | * only if we are re-adding a preexisting device. |
7ceb17e8 N |
7149 | * So require mddev->pers and MD_DISK_SYNC. |
7150 | */ | |
7151 | if (mddev->pers) { | |
7152 | mdu_disk_info_t info; | |
7153 | if (copy_from_user(&info, argp, sizeof(info))) | |
7154 | err = -EFAULT; | |
7155 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
7156 | /* Need to clear read-only for this */ | |
7157 | break; | |
7158 | else | |
7159 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7160 | goto unlock; |
7ceb17e8 N |
7161 | } |
7162 | break; | |
7163 | ||
c02c0aeb N |
7164 | case BLKROSET: |
7165 | if (get_user(ro, (int __user *)(arg))) { | |
7166 | err = -EFAULT; | |
3adc28d8 | 7167 | goto unlock; |
c02c0aeb N |
7168 | } |
7169 | err = -EINVAL; | |
e2218350 | 7170 | |
c02c0aeb N |
7171 | /* if the bdev is going readonly the value of mddev->ro |
7172 | * does not matter, no writes are coming | |
7173 | */ | |
7174 | if (ro) | |
3adc28d8 | 7175 | goto unlock; |
e2218350 | 7176 | |
c02c0aeb N |
7177 | /* are we are already prepared for writes? */ |
7178 | if (mddev->ro != 1) | |
3adc28d8 | 7179 | goto unlock; |
e2218350 | 7180 | |
c02c0aeb N |
7181 | /* transitioning to readauto need only happen for |
7182 | * arrays that call md_write_start | |
7183 | */ | |
7184 | if (mddev->pers) { | |
7185 | err = restart_array(mddev); | |
7186 | if (err == 0) { | |
7187 | mddev->ro = 2; | |
7188 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 7189 | } |
c02c0aeb | 7190 | } |
3adc28d8 | 7191 | goto unlock; |
1da177e4 LT |
7192 | } |
7193 | ||
7194 | /* | |
7195 | * The remaining ioctls are changing the state of the | |
f91de92e | 7196 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 7197 | */ |
326eb17d | 7198 | if (mddev->ro && mddev->pers) { |
f91de92e N |
7199 | if (mddev->ro == 2) { |
7200 | mddev->ro = 0; | |
00bcb4ac | 7201 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 7202 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
7203 | /* mddev_unlock will wake thread */ |
7204 | /* If a device failed while we were read-only, we | |
7205 | * need to make sure the metadata is updated now. | |
7206 | */ | |
2953079c | 7207 | if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) { |
f3378b48 N |
7208 | mddev_unlock(mddev); |
7209 | wait_event(mddev->sb_wait, | |
2953079c SL |
7210 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) && |
7211 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); | |
29f097c4 | 7212 | mddev_lock_nointr(mddev); |
f3378b48 | 7213 | } |
f91de92e N |
7214 | } else { |
7215 | err = -EROFS; | |
3adc28d8 | 7216 | goto unlock; |
f91de92e | 7217 | } |
1da177e4 LT |
7218 | } |
7219 | ||
c02c0aeb N |
7220 | switch (cmd) { |
7221 | case ADD_NEW_DISK: | |
1da177e4 | 7222 | { |
c02c0aeb N |
7223 | mdu_disk_info_t info; |
7224 | if (copy_from_user(&info, argp, sizeof(info))) | |
7225 | err = -EFAULT; | |
7226 | else | |
7227 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7228 | goto unlock; |
c02c0aeb | 7229 | } |
1da177e4 | 7230 | |
1aee41f6 GR |
7231 | case CLUSTERED_DISK_NACK: |
7232 | if (mddev_is_clustered(mddev)) | |
7233 | md_cluster_ops->new_disk_ack(mddev, false); | |
7234 | else | |
7235 | err = -EINVAL; | |
7236 | goto unlock; | |
7237 | ||
c02c0aeb N |
7238 | case HOT_ADD_DISK: |
7239 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7240 | goto unlock; |
1da177e4 | 7241 | |
c02c0aeb N |
7242 | case RUN_ARRAY: |
7243 | err = do_md_run(mddev); | |
3adc28d8 | 7244 | goto unlock; |
1da177e4 | 7245 | |
c02c0aeb N |
7246 | case SET_BITMAP_FILE: |
7247 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 7248 | goto unlock; |
32a7627c | 7249 | |
c02c0aeb N |
7250 | default: |
7251 | err = -EINVAL; | |
3adc28d8 | 7252 | goto unlock; |
1da177e4 LT |
7253 | } |
7254 | ||
3adc28d8 | 7255 | unlock: |
d3374825 N |
7256 | if (mddev->hold_active == UNTIL_IOCTL && |
7257 | err != -EINVAL) | |
7258 | mddev->hold_active = 0; | |
1da177e4 | 7259 | mddev_unlock(mddev); |
3adc28d8 | 7260 | out: |
065e519e N |
7261 | if(did_set_md_closing) |
7262 | clear_bit(MD_CLOSING, &mddev->flags); | |
1da177e4 LT |
7263 | return err; |
7264 | } | |
aa98aa31 AB |
7265 | #ifdef CONFIG_COMPAT |
7266 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
7267 | unsigned int cmd, unsigned long arg) | |
7268 | { | |
7269 | switch (cmd) { | |
7270 | case HOT_REMOVE_DISK: | |
7271 | case HOT_ADD_DISK: | |
7272 | case SET_DISK_FAULTY: | |
7273 | case SET_BITMAP_FILE: | |
7274 | /* These take in integer arg, do not convert */ | |
7275 | break; | |
7276 | default: | |
7277 | arg = (unsigned long)compat_ptr(arg); | |
7278 | break; | |
7279 | } | |
7280 | ||
7281 | return md_ioctl(bdev, mode, cmd, arg); | |
7282 | } | |
7283 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 7284 | |
a39907fa | 7285 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
7286 | { |
7287 | /* | |
7288 | * Succeed if we can lock the mddev, which confirms that | |
7289 | * it isn't being stopped right now. | |
7290 | */ | |
fd01b88c | 7291 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
7292 | int err; |
7293 | ||
0c098220 YL |
7294 | if (!mddev) |
7295 | return -ENODEV; | |
7296 | ||
d3374825 N |
7297 | if (mddev->gendisk != bdev->bd_disk) { |
7298 | /* we are racing with mddev_put which is discarding this | |
7299 | * bd_disk. | |
7300 | */ | |
7301 | mddev_put(mddev); | |
7302 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 7303 | flush_workqueue(md_misc_wq); |
d3374825 N |
7304 | /* Then retry the open from the top */ |
7305 | return -ERESTARTSYS; | |
7306 | } | |
7307 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
7308 | ||
c8c00a69 | 7309 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
7310 | goto out; |
7311 | ||
af8d8e6f GJ |
7312 | if (test_bit(MD_CLOSING, &mddev->flags)) { |
7313 | mutex_unlock(&mddev->open_mutex); | |
e2342ca8 N |
7314 | err = -ENODEV; |
7315 | goto out; | |
af8d8e6f GJ |
7316 | } |
7317 | ||
1da177e4 | 7318 | err = 0; |
f2ea68cf | 7319 | atomic_inc(&mddev->openers); |
c8c00a69 | 7320 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 7321 | |
f0b4f7e2 | 7322 | check_disk_change(bdev); |
1da177e4 | 7323 | out: |
e2342ca8 N |
7324 | if (err) |
7325 | mddev_put(mddev); | |
1da177e4 LT |
7326 | return err; |
7327 | } | |
7328 | ||
db2a144b | 7329 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 7330 | { |
f72ffdd6 | 7331 | struct mddev *mddev = disk->private_data; |
1da177e4 | 7332 | |
52e5f9d1 | 7333 | BUG_ON(!mddev); |
f2ea68cf | 7334 | atomic_dec(&mddev->openers); |
1da177e4 | 7335 | mddev_put(mddev); |
1da177e4 | 7336 | } |
f0b4f7e2 N |
7337 | |
7338 | static int md_media_changed(struct gendisk *disk) | |
7339 | { | |
fd01b88c | 7340 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7341 | |
7342 | return mddev->changed; | |
7343 | } | |
7344 | ||
7345 | static int md_revalidate(struct gendisk *disk) | |
7346 | { | |
fd01b88c | 7347 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7348 | |
7349 | mddev->changed = 0; | |
7350 | return 0; | |
7351 | } | |
83d5cde4 | 7352 | static const struct block_device_operations md_fops = |
1da177e4 LT |
7353 | { |
7354 | .owner = THIS_MODULE, | |
a39907fa AV |
7355 | .open = md_open, |
7356 | .release = md_release, | |
b492b852 | 7357 | .ioctl = md_ioctl, |
aa98aa31 AB |
7358 | #ifdef CONFIG_COMPAT |
7359 | .compat_ioctl = md_compat_ioctl, | |
7360 | #endif | |
a885c8c4 | 7361 | .getgeo = md_getgeo, |
f0b4f7e2 N |
7362 | .media_changed = md_media_changed, |
7363 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
7364 | }; |
7365 | ||
f72ffdd6 | 7366 | static int md_thread(void *arg) |
1da177e4 | 7367 | { |
2b8bf345 | 7368 | struct md_thread *thread = arg; |
1da177e4 | 7369 | |
1da177e4 LT |
7370 | /* |
7371 | * md_thread is a 'system-thread', it's priority should be very | |
7372 | * high. We avoid resource deadlocks individually in each | |
7373 | * raid personality. (RAID5 does preallocation) We also use RR and | |
7374 | * the very same RT priority as kswapd, thus we will never get | |
7375 | * into a priority inversion deadlock. | |
7376 | * | |
7377 | * we definitely have to have equal or higher priority than | |
7378 | * bdflush, otherwise bdflush will deadlock if there are too | |
7379 | * many dirty RAID5 blocks. | |
7380 | */ | |
1da177e4 | 7381 | |
6985c43f | 7382 | allow_signal(SIGKILL); |
a6fb0934 | 7383 | while (!kthread_should_stop()) { |
1da177e4 | 7384 | |
93588e22 N |
7385 | /* We need to wait INTERRUPTIBLE so that |
7386 | * we don't add to the load-average. | |
7387 | * That means we need to be sure no signals are | |
7388 | * pending | |
7389 | */ | |
7390 | if (signal_pending(current)) | |
7391 | flush_signals(current); | |
7392 | ||
7393 | wait_event_interruptible_timeout | |
7394 | (thread->wqueue, | |
7395 | test_bit(THREAD_WAKEUP, &thread->flags) | |
ce1ccd07 | 7396 | || kthread_should_stop() || kthread_should_park(), |
93588e22 | 7397 | thread->timeout); |
1da177e4 | 7398 | |
6c987910 | 7399 | clear_bit(THREAD_WAKEUP, &thread->flags); |
ce1ccd07 SL |
7400 | if (kthread_should_park()) |
7401 | kthread_parkme(); | |
6c987910 | 7402 | if (!kthread_should_stop()) |
4ed8731d | 7403 | thread->run(thread); |
1da177e4 | 7404 | } |
a6fb0934 | 7405 | |
1da177e4 LT |
7406 | return 0; |
7407 | } | |
7408 | ||
2b8bf345 | 7409 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
7410 | { |
7411 | if (thread) { | |
36a4e1fe | 7412 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
4ad23a97 N |
7413 | if (!test_and_set_bit(THREAD_WAKEUP, &thread->flags)) |
7414 | wake_up(&thread->wqueue); | |
1da177e4 LT |
7415 | } |
7416 | } | |
6c144d31 | 7417 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 7418 | |
4ed8731d SL |
7419 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
7420 | struct mddev *mddev, const char *name) | |
1da177e4 | 7421 | { |
2b8bf345 | 7422 | struct md_thread *thread; |
1da177e4 | 7423 | |
2b8bf345 | 7424 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
7425 | if (!thread) |
7426 | return NULL; | |
7427 | ||
1da177e4 LT |
7428 | init_waitqueue_head(&thread->wqueue); |
7429 | ||
1da177e4 LT |
7430 | thread->run = run; |
7431 | thread->mddev = mddev; | |
32a7627c | 7432 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
7433 | thread->tsk = kthread_run(md_thread, thread, |
7434 | "%s_%s", | |
7435 | mdname(thread->mddev), | |
0232605d | 7436 | name); |
a6fb0934 | 7437 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
7438 | kfree(thread); |
7439 | return NULL; | |
7440 | } | |
1da177e4 LT |
7441 | return thread; |
7442 | } | |
6c144d31 | 7443 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 7444 | |
2b8bf345 | 7445 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 7446 | { |
2b8bf345 | 7447 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
7448 | if (!thread) |
7449 | return; | |
36a4e1fe | 7450 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
7451 | /* Locking ensures that mddev_unlock does not wake_up a |
7452 | * non-existent thread | |
7453 | */ | |
7454 | spin_lock(&pers_lock); | |
7455 | *threadp = NULL; | |
7456 | spin_unlock(&pers_lock); | |
a6fb0934 N |
7457 | |
7458 | kthread_stop(thread->tsk); | |
1da177e4 LT |
7459 | kfree(thread); |
7460 | } | |
6c144d31 | 7461 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 7462 | |
fd01b88c | 7463 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7464 | { |
b2d444d7 | 7465 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 7466 | return; |
6bfe0b49 | 7467 | |
de393cde | 7468 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
7469 | return; |
7470 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
7471 | if (mddev->degraded) |
7472 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 7473 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
7474 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7475 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7476 | md_wakeup_thread(mddev->thread); | |
768a418d | 7477 | if (mddev->event_work.func) |
e804ac78 | 7478 | queue_work(md_misc_wq, &mddev->event_work); |
bb9ef716 | 7479 | md_new_event(mddev); |
1da177e4 | 7480 | } |
6c144d31 | 7481 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
7482 | |
7483 | /* seq_file implementation /proc/mdstat */ | |
7484 | ||
7485 | static void status_unused(struct seq_file *seq) | |
7486 | { | |
7487 | int i = 0; | |
3cb03002 | 7488 | struct md_rdev *rdev; |
1da177e4 LT |
7489 | |
7490 | seq_printf(seq, "unused devices: "); | |
7491 | ||
159ec1fc | 7492 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
7493 | char b[BDEVNAME_SIZE]; |
7494 | i++; | |
7495 | seq_printf(seq, "%s ", | |
7496 | bdevname(rdev->bdev,b)); | |
7497 | } | |
7498 | if (!i) | |
7499 | seq_printf(seq, "<none>"); | |
7500 | ||
7501 | seq_printf(seq, "\n"); | |
7502 | } | |
7503 | ||
f7851be7 | 7504 | static int status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7505 | { |
dd71cf6b N |
7506 | sector_t max_sectors, resync, res; |
7507 | unsigned long dt, db; | |
7508 | sector_t rt; | |
4588b42e N |
7509 | int scale; |
7510 | unsigned int per_milli; | |
1da177e4 | 7511 | |
c804cdec N |
7512 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7513 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 7514 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 7515 | else |
dd71cf6b | 7516 | max_sectors = mddev->dev_sectors; |
1da177e4 | 7517 | |
f7851be7 N |
7518 | resync = mddev->curr_resync; |
7519 | if (resync <= 3) { | |
7520 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7521 | /* Still cleaning up */ | |
7522 | resync = max_sectors; | |
7523 | } else | |
7524 | resync -= atomic_read(&mddev->recovery_active); | |
7525 | ||
7526 | if (resync == 0) { | |
7527 | if (mddev->recovery_cp < MaxSector) { | |
7528 | seq_printf(seq, "\tresync=PENDING"); | |
7529 | return 1; | |
7530 | } | |
7531 | return 0; | |
7532 | } | |
7533 | if (resync < 3) { | |
7534 | seq_printf(seq, "\tresync=DELAYED"); | |
7535 | return 1; | |
7536 | } | |
7537 | ||
403df478 | 7538 | WARN_ON(max_sectors == 0); |
4588b42e | 7539 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 7540 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
7541 | * u32, as those are the requirements for sector_div. |
7542 | * Thus 'scale' must be at least 10 | |
7543 | */ | |
7544 | scale = 10; | |
7545 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 7546 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
7547 | scale++; |
7548 | } | |
7549 | res = (resync>>scale)*1000; | |
dd71cf6b | 7550 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
7551 | |
7552 | per_milli = res; | |
1da177e4 | 7553 | { |
4588b42e | 7554 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
7555 | seq_printf(seq, "["); |
7556 | for (i = 0; i < x; i++) | |
7557 | seq_printf(seq, "="); | |
7558 | seq_printf(seq, ">"); | |
7559 | for (i = 0; i < y; i++) | |
7560 | seq_printf(seq, "."); | |
7561 | seq_printf(seq, "] "); | |
7562 | } | |
4588b42e | 7563 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
7564 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
7565 | "reshape" : | |
61df9d91 N |
7566 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
7567 | "check" : | |
7568 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
7569 | "resync" : "recovery"))), | |
7570 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
7571 | (unsigned long long) resync/2, |
7572 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
7573 | |
7574 | /* | |
1da177e4 LT |
7575 | * dt: time from mark until now |
7576 | * db: blocks written from mark until now | |
7577 | * rt: remaining time | |
dd71cf6b N |
7578 | * |
7579 | * rt is a sector_t, so could be 32bit or 64bit. | |
7580 | * So we divide before multiply in case it is 32bit and close | |
7581 | * to the limit. | |
25985edc | 7582 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
7583 | * near the end of resync when the number of remaining sectors |
7584 | * is close to 'db'. | |
7585 | * We then divide rt by 32 after multiplying by db to compensate. | |
7586 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
7587 | */ |
7588 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
7589 | if (!dt) dt++; | |
ff4e8d9a N |
7590 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
7591 | - mddev->resync_mark_cnt; | |
1da177e4 | 7592 | |
dd71cf6b N |
7593 | rt = max_sectors - resync; /* number of remaining sectors */ |
7594 | sector_div(rt, db/32+1); | |
7595 | rt *= dt; | |
7596 | rt >>= 5; | |
7597 | ||
7598 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
7599 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 7600 | |
ff4e8d9a | 7601 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
f7851be7 | 7602 | return 1; |
1da177e4 LT |
7603 | } |
7604 | ||
7605 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7606 | { | |
7607 | struct list_head *tmp; | |
7608 | loff_t l = *pos; | |
fd01b88c | 7609 | struct mddev *mddev; |
1da177e4 LT |
7610 | |
7611 | if (l >= 0x10000) | |
7612 | return NULL; | |
7613 | if (!l--) | |
7614 | /* header */ | |
7615 | return (void*)1; | |
7616 | ||
7617 | spin_lock(&all_mddevs_lock); | |
7618 | list_for_each(tmp,&all_mddevs) | |
7619 | if (!l--) { | |
fd01b88c | 7620 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7621 | mddev_get(mddev); |
7622 | spin_unlock(&all_mddevs_lock); | |
7623 | return mddev; | |
7624 | } | |
7625 | spin_unlock(&all_mddevs_lock); | |
7626 | if (!l--) | |
7627 | return (void*)2;/* tail */ | |
7628 | return NULL; | |
7629 | } | |
7630 | ||
7631 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7632 | { | |
7633 | struct list_head *tmp; | |
fd01b88c | 7634 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7635 | |
1da177e4 LT |
7636 | ++*pos; |
7637 | if (v == (void*)2) | |
7638 | return NULL; | |
7639 | ||
7640 | spin_lock(&all_mddevs_lock); | |
7641 | if (v == (void*)1) | |
7642 | tmp = all_mddevs.next; | |
7643 | else | |
7644 | tmp = mddev->all_mddevs.next; | |
7645 | if (tmp != &all_mddevs) | |
fd01b88c | 7646 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7647 | else { |
7648 | next_mddev = (void*)2; | |
7649 | *pos = 0x10000; | |
f72ffdd6 | 7650 | } |
1da177e4 LT |
7651 | spin_unlock(&all_mddevs_lock); |
7652 | ||
7653 | if (v != (void*)1) | |
7654 | mddev_put(mddev); | |
7655 | return next_mddev; | |
7656 | ||
7657 | } | |
7658 | ||
7659 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7660 | { | |
fd01b88c | 7661 | struct mddev *mddev = v; |
1da177e4 LT |
7662 | |
7663 | if (mddev && v != (void*)1 && v != (void*)2) | |
7664 | mddev_put(mddev); | |
7665 | } | |
7666 | ||
7667 | static int md_seq_show(struct seq_file *seq, void *v) | |
7668 | { | |
fd01b88c | 7669 | struct mddev *mddev = v; |
dd8ac336 | 7670 | sector_t sectors; |
3cb03002 | 7671 | struct md_rdev *rdev; |
1da177e4 LT |
7672 | |
7673 | if (v == (void*)1) { | |
84fc4b56 | 7674 | struct md_personality *pers; |
1da177e4 LT |
7675 | seq_printf(seq, "Personalities : "); |
7676 | spin_lock(&pers_lock); | |
2604b703 N |
7677 | list_for_each_entry(pers, &pers_list, list) |
7678 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7679 | |
7680 | spin_unlock(&pers_lock); | |
7681 | seq_printf(seq, "\n"); | |
f1514638 | 7682 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7683 | return 0; |
7684 | } | |
7685 | if (v == (void*)2) { | |
7686 | status_unused(seq); | |
7687 | return 0; | |
7688 | } | |
7689 | ||
36d091f4 | 7690 | spin_lock(&mddev->lock); |
1da177e4 LT |
7691 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7692 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7693 | mddev->pers ? "" : "in"); | |
7694 | if (mddev->pers) { | |
f91de92e | 7695 | if (mddev->ro==1) |
1da177e4 | 7696 | seq_printf(seq, " (read-only)"); |
f91de92e | 7697 | if (mddev->ro==2) |
52720ae7 | 7698 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7699 | seq_printf(seq, " %s", mddev->pers->name); |
7700 | } | |
7701 | ||
dd8ac336 | 7702 | sectors = 0; |
f97fcad3 N |
7703 | rcu_read_lock(); |
7704 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7705 | char b[BDEVNAME_SIZE]; |
7706 | seq_printf(seq, " %s[%d]", | |
7707 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7708 | if (test_bit(WriteMostly, &rdev->flags)) |
7709 | seq_printf(seq, "(W)"); | |
9efdca16 SL |
7710 | if (test_bit(Journal, &rdev->flags)) |
7711 | seq_printf(seq, "(J)"); | |
b2d444d7 | 7712 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7713 | seq_printf(seq, "(F)"); |
7714 | continue; | |
2d78f8c4 N |
7715 | } |
7716 | if (rdev->raid_disk < 0) | |
b325a32e | 7717 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7718 | if (test_bit(Replacement, &rdev->flags)) |
7719 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7720 | sectors += rdev->sectors; |
1da177e4 | 7721 | } |
f97fcad3 | 7722 | rcu_read_unlock(); |
1da177e4 LT |
7723 | |
7724 | if (!list_empty(&mddev->disks)) { | |
7725 | if (mddev->pers) | |
7726 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7727 | (unsigned long long) |
7728 | mddev->array_sectors / 2); | |
1da177e4 LT |
7729 | else |
7730 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7731 | (unsigned long long)sectors / 2); |
1da177e4 | 7732 | } |
1cd6bf19 N |
7733 | if (mddev->persistent) { |
7734 | if (mddev->major_version != 0 || | |
7735 | mddev->minor_version != 90) { | |
7736 | seq_printf(seq," super %d.%d", | |
7737 | mddev->major_version, | |
7738 | mddev->minor_version); | |
7739 | } | |
e691063a N |
7740 | } else if (mddev->external) |
7741 | seq_printf(seq, " super external:%s", | |
7742 | mddev->metadata_type); | |
7743 | else | |
1cd6bf19 | 7744 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7745 | |
7746 | if (mddev->pers) { | |
d710e138 | 7747 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7748 | seq_printf(seq, "\n "); |
8e1b39d6 | 7749 | if (mddev->pers->sync_request) { |
f7851be7 | 7750 | if (status_resync(seq, mddev)) |
8e1b39d6 | 7751 | seq_printf(seq, "\n "); |
8e1b39d6 | 7752 | } |
32a7627c N |
7753 | } else |
7754 | seq_printf(seq, "\n "); | |
7755 | ||
57148964 | 7756 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7757 | |
7758 | seq_printf(seq, "\n"); | |
7759 | } | |
36d091f4 | 7760 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7761 | |
1da177e4 LT |
7762 | return 0; |
7763 | } | |
7764 | ||
110518bc | 7765 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7766 | .start = md_seq_start, |
7767 | .next = md_seq_next, | |
7768 | .stop = md_seq_stop, | |
7769 | .show = md_seq_show, | |
7770 | }; | |
7771 | ||
7772 | static int md_seq_open(struct inode *inode, struct file *file) | |
7773 | { | |
f1514638 | 7774 | struct seq_file *seq; |
1da177e4 LT |
7775 | int error; |
7776 | ||
7777 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7778 | if (error) |
f1514638 KS |
7779 | return error; |
7780 | ||
7781 | seq = file->private_data; | |
7782 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7783 | return error; |
7784 | } | |
7785 | ||
e2f23b60 | 7786 | static int md_unloading; |
d7603b7e N |
7787 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7788 | { | |
f1514638 | 7789 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7790 | int mask; |
7791 | ||
e2f23b60 | 7792 | if (md_unloading) |
7d7e64f2 | 7793 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI; |
d7603b7e N |
7794 | poll_wait(filp, &md_event_waiters, wait); |
7795 | ||
7796 | /* always allow read */ | |
7797 | mask = POLLIN | POLLRDNORM; | |
7798 | ||
f1514638 | 7799 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7800 | mask |= POLLERR | POLLPRI; |
7801 | return mask; | |
7802 | } | |
7803 | ||
fa027c2a | 7804 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7805 | .owner = THIS_MODULE, |
1da177e4 LT |
7806 | .open = md_seq_open, |
7807 | .read = seq_read, | |
7808 | .llseek = seq_lseek, | |
c3f94b40 | 7809 | .release = seq_release_private, |
d7603b7e | 7810 | .poll = mdstat_poll, |
1da177e4 LT |
7811 | }; |
7812 | ||
84fc4b56 | 7813 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7814 | { |
9d48739e N |
7815 | pr_debug("md: %s personality registered for level %d\n", |
7816 | p->name, p->level); | |
1da177e4 | 7817 | spin_lock(&pers_lock); |
2604b703 | 7818 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7819 | spin_unlock(&pers_lock); |
7820 | return 0; | |
7821 | } | |
6c144d31 | 7822 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7823 | |
84fc4b56 | 7824 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7825 | { |
9d48739e | 7826 | pr_debug("md: %s personality unregistered\n", p->name); |
1da177e4 | 7827 | spin_lock(&pers_lock); |
2604b703 | 7828 | list_del_init(&p->list); |
1da177e4 LT |
7829 | spin_unlock(&pers_lock); |
7830 | return 0; | |
7831 | } | |
6c144d31 | 7832 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7833 | |
6022e75b N |
7834 | int register_md_cluster_operations(struct md_cluster_operations *ops, |
7835 | struct module *module) | |
edb39c9d | 7836 | { |
6022e75b | 7837 | int ret = 0; |
edb39c9d | 7838 | spin_lock(&pers_lock); |
6022e75b N |
7839 | if (md_cluster_ops != NULL) |
7840 | ret = -EALREADY; | |
7841 | else { | |
7842 | md_cluster_ops = ops; | |
7843 | md_cluster_mod = module; | |
7844 | } | |
edb39c9d | 7845 | spin_unlock(&pers_lock); |
6022e75b | 7846 | return ret; |
edb39c9d GR |
7847 | } |
7848 | EXPORT_SYMBOL(register_md_cluster_operations); | |
7849 | ||
7850 | int unregister_md_cluster_operations(void) | |
7851 | { | |
7852 | spin_lock(&pers_lock); | |
7853 | md_cluster_ops = NULL; | |
7854 | spin_unlock(&pers_lock); | |
7855 | return 0; | |
7856 | } | |
7857 | EXPORT_SYMBOL(unregister_md_cluster_operations); | |
7858 | ||
7859 | int md_setup_cluster(struct mddev *mddev, int nodes) | |
7860 | { | |
47a7b0d8 GJ |
7861 | if (!md_cluster_ops) |
7862 | request_module("md-cluster"); | |
edb39c9d | 7863 | spin_lock(&pers_lock); |
47a7b0d8 | 7864 | /* ensure module won't be unloaded */ |
edb39c9d | 7865 | if (!md_cluster_ops || !try_module_get(md_cluster_mod)) { |
9d48739e | 7866 | pr_warn("can't find md-cluster module or get it's reference.\n"); |
edb39c9d GR |
7867 | spin_unlock(&pers_lock); |
7868 | return -ENOENT; | |
7869 | } | |
7870 | spin_unlock(&pers_lock); | |
7871 | ||
cf921cc1 | 7872 | return md_cluster_ops->join(mddev, nodes); |
edb39c9d GR |
7873 | } |
7874 | ||
7875 | void md_cluster_stop(struct mddev *mddev) | |
7876 | { | |
c4ce867f GR |
7877 | if (!md_cluster_ops) |
7878 | return; | |
edb39c9d GR |
7879 | md_cluster_ops->leave(mddev); |
7880 | module_put(md_cluster_mod); | |
7881 | } | |
7882 | ||
fd01b88c | 7883 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7884 | { |
f72ffdd6 | 7885 | struct md_rdev *rdev; |
1da177e4 | 7886 | int idle; |
eea1bf38 | 7887 | int curr_events; |
1da177e4 LT |
7888 | |
7889 | idle = 1; | |
4b80991c N |
7890 | rcu_read_lock(); |
7891 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7892 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7893 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7894 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7895 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7896 | /* sync IO will cause sync_io to increase before the disk_stats |
7897 | * as sync_io is counted when a request starts, and | |
7898 | * disk_stats is counted when it completes. | |
7899 | * So resync activity will cause curr_events to be smaller than | |
7900 | * when there was no such activity. | |
7901 | * non-sync IO will cause disk_stat to increase without | |
7902 | * increasing sync_io so curr_events will (eventually) | |
7903 | * be larger than it was before. Once it becomes | |
7904 | * substantially larger, the test below will cause | |
7905 | * the array to appear non-idle, and resync will slow | |
7906 | * down. | |
7907 | * If there is a lot of outstanding resync activity when | |
7908 | * we set last_event to curr_events, then all that activity | |
7909 | * completing might cause the array to appear non-idle | |
7910 | * and resync will be slowed down even though there might | |
7911 | * not have been non-resync activity. This will only | |
7912 | * happen once though. 'last_events' will soon reflect | |
7913 | * the state where there is little or no outstanding | |
7914 | * resync requests, and further resync activity will | |
7915 | * always make curr_events less than last_events. | |
c0e48521 | 7916 | * |
1da177e4 | 7917 | */ |
eea1bf38 | 7918 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7919 | rdev->last_events = curr_events; |
7920 | idle = 0; | |
7921 | } | |
7922 | } | |
4b80991c | 7923 | rcu_read_unlock(); |
1da177e4 LT |
7924 | return idle; |
7925 | } | |
7926 | ||
fd01b88c | 7927 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7928 | { |
7929 | /* another "blocks" (512byte) blocks have been synced */ | |
7930 | atomic_sub(blocks, &mddev->recovery_active); | |
7931 | wake_up(&mddev->recovery_wait); | |
7932 | if (!ok) { | |
dfc70645 | 7933 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7934 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7935 | md_wakeup_thread(mddev->thread); |
7936 | // stop recovery, signal do_sync .... | |
7937 | } | |
7938 | } | |
6c144d31 | 7939 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 7940 | |
06d91a5f N |
7941 | /* md_write_start(mddev, bi) |
7942 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7943 | * in superblock) before writing, schedule a superblock update |
7944 | * and wait for it to complete. | |
06d91a5f | 7945 | */ |
fd01b88c | 7946 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7947 | { |
0fd62b86 | 7948 | int did_change = 0; |
06d91a5f | 7949 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7950 | return; |
06d91a5f | 7951 | |
f91de92e N |
7952 | BUG_ON(mddev->ro == 1); |
7953 | if (mddev->ro == 2) { | |
7954 | /* need to switch to read/write */ | |
7955 | mddev->ro = 0; | |
7956 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7957 | md_wakeup_thread(mddev->thread); | |
25156198 | 7958 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7959 | did_change = 1; |
f91de92e | 7960 | } |
4ad23a97 N |
7961 | rcu_read_lock(); |
7962 | percpu_ref_get(&mddev->writes_pending); | |
55cc39f3 | 7963 | smp_mb(); /* Match smp_mb in set_in_sync() */ |
31a59e34 N |
7964 | if (mddev->safemode == 1) |
7965 | mddev->safemode = 0; | |
4ad23a97 N |
7966 | /* sync_checkers is always 0 when writes_pending is in per-cpu mode */ |
7967 | if (mddev->in_sync || !mddev->sync_checkers) { | |
85572d7c | 7968 | spin_lock(&mddev->lock); |
3d310eb7 N |
7969 | if (mddev->in_sync) { |
7970 | mddev->in_sync = 0; | |
2953079c SL |
7971 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
7972 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
3d310eb7 | 7973 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7974 | did_change = 1; |
3d310eb7 | 7975 | } |
85572d7c | 7976 | spin_unlock(&mddev->lock); |
06d91a5f | 7977 | } |
4ad23a97 | 7978 | rcu_read_unlock(); |
0fd62b86 | 7979 | if (did_change) |
00bcb4ac | 7980 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7981 | wait_event(mddev->sb_wait, |
2953079c | 7982 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
1da177e4 | 7983 | } |
6c144d31 | 7984 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 7985 | |
49728050 N |
7986 | /* md_write_inc can only be called when md_write_start() has |
7987 | * already been called at least once of the current request. | |
7988 | * It increments the counter and is useful when a single request | |
7989 | * is split into several parts. Each part causes an increment and | |
7990 | * so needs a matching md_write_end(). | |
7991 | * Unlike md_write_start(), it is safe to call md_write_inc() inside | |
7992 | * a spinlocked region. | |
7993 | */ | |
7994 | void md_write_inc(struct mddev *mddev, struct bio *bi) | |
7995 | { | |
7996 | if (bio_data_dir(bi) != WRITE) | |
7997 | return; | |
7998 | WARN_ON_ONCE(mddev->in_sync || mddev->ro); | |
4ad23a97 | 7999 | percpu_ref_get(&mddev->writes_pending); |
49728050 N |
8000 | } |
8001 | EXPORT_SYMBOL(md_write_inc); | |
8002 | ||
fd01b88c | 8003 | void md_write_end(struct mddev *mddev) |
1da177e4 | 8004 | { |
4ad23a97 N |
8005 | percpu_ref_put(&mddev->writes_pending); |
8006 | ||
8007 | if (mddev->safemode == 2) | |
8008 | md_wakeup_thread(mddev->thread); | |
8009 | else if (mddev->safemode_delay) | |
8010 | /* The roundup() ensures this only performs locking once | |
8011 | * every ->safemode_delay jiffies | |
8012 | */ | |
8013 | mod_timer(&mddev->safemode_timer, | |
8014 | roundup(jiffies, mddev->safemode_delay) + | |
8015 | mddev->safemode_delay); | |
1da177e4 | 8016 | } |
4ad23a97 | 8017 | |
6c144d31 | 8018 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 8019 | |
2a2275d6 N |
8020 | /* md_allow_write(mddev) |
8021 | * Calling this ensures that the array is marked 'active' so that writes | |
8022 | * may proceed without blocking. It is important to call this before | |
8023 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
8024 | * Must be called with mddev_lock held. | |
8025 | */ | |
2214c260 | 8026 | void md_allow_write(struct mddev *mddev) |
2a2275d6 N |
8027 | { |
8028 | if (!mddev->pers) | |
2214c260 | 8029 | return; |
2a2275d6 | 8030 | if (mddev->ro) |
2214c260 | 8031 | return; |
1a0fd497 | 8032 | if (!mddev->pers->sync_request) |
2214c260 | 8033 | return; |
2a2275d6 | 8034 | |
85572d7c | 8035 | spin_lock(&mddev->lock); |
2a2275d6 N |
8036 | if (mddev->in_sync) { |
8037 | mddev->in_sync = 0; | |
2953079c SL |
8038 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
8039 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
2a2275d6 N |
8040 | if (mddev->safemode_delay && |
8041 | mddev->safemode == 0) | |
8042 | mddev->safemode = 1; | |
85572d7c | 8043 | spin_unlock(&mddev->lock); |
2a2275d6 | 8044 | md_update_sb(mddev, 0); |
00bcb4ac | 8045 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2214c260 AP |
8046 | /* wait for the dirty state to be recorded in the metadata */ |
8047 | wait_event(mddev->sb_wait, | |
8048 | !test_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags) && | |
8049 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); | |
2a2275d6 | 8050 | } else |
85572d7c | 8051 | spin_unlock(&mddev->lock); |
2a2275d6 N |
8052 | } |
8053 | EXPORT_SYMBOL_GPL(md_allow_write); | |
8054 | ||
1da177e4 LT |
8055 | #define SYNC_MARKS 10 |
8056 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 8057 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 8058 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 8059 | { |
4ed8731d | 8060 | struct mddev *mddev = thread->mddev; |
fd01b88c | 8061 | struct mddev *mddev2; |
1da177e4 LT |
8062 | unsigned int currspeed = 0, |
8063 | window; | |
ac7e50a3 | 8064 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 8065 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 8066 | unsigned long update_time; |
1da177e4 LT |
8067 | sector_t mark_cnt[SYNC_MARKS]; |
8068 | int last_mark,m; | |
8069 | struct list_head *tmp; | |
8070 | sector_t last_check; | |
57afd89f | 8071 | int skipped = 0; |
3cb03002 | 8072 | struct md_rdev *rdev; |
c4a39551 | 8073 | char *desc, *action = NULL; |
7c2c57c9 | 8074 | struct blk_plug plug; |
41a9a0dc | 8075 | int ret; |
1da177e4 LT |
8076 | |
8077 | /* just incase thread restarts... */ | |
8078 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
8079 | return; | |
3991b31e N |
8080 | if (mddev->ro) {/* never try to sync a read-only array */ |
8081 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 8082 | return; |
3991b31e | 8083 | } |
1da177e4 | 8084 | |
41a9a0dc GJ |
8085 | if (mddev_is_clustered(mddev)) { |
8086 | ret = md_cluster_ops->resync_start(mddev); | |
8087 | if (ret) | |
8088 | goto skip; | |
8089 | ||
bb8bf15b | 8090 | set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags); |
41a9a0dc GJ |
8091 | if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
8092 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) || | |
8093 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) | |
8094 | && ((unsigned long long)mddev->curr_resync_completed | |
8095 | < (unsigned long long)mddev->resync_max_sectors)) | |
8096 | goto skip; | |
8097 | } | |
8098 | ||
61df9d91 | 8099 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 8100 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 8101 | desc = "data-check"; |
c4a39551 JB |
8102 | action = "check"; |
8103 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 8104 | desc = "requested-resync"; |
c4a39551 JB |
8105 | action = "repair"; |
8106 | } else | |
61df9d91 N |
8107 | desc = "resync"; |
8108 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
8109 | desc = "reshape"; | |
8110 | else | |
8111 | desc = "recovery"; | |
8112 | ||
c4a39551 JB |
8113 | mddev->last_sync_action = action ?: desc; |
8114 | ||
1da177e4 LT |
8115 | /* we overload curr_resync somewhat here. |
8116 | * 0 == not engaged in resync at all | |
8117 | * 2 == checking that there is no conflict with another sync | |
8118 | * 1 == like 2, but have yielded to allow conflicting resync to | |
8119 | * commense | |
8120 | * other == active in resync - this many blocks | |
8121 | * | |
8122 | * Before starting a resync we must have set curr_resync to | |
8123 | * 2, and then checked that every "conflicting" array has curr_resync | |
8124 | * less than ours. When we find one that is the same or higher | |
8125 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
8126 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
8127 | * This will mean we have to start checking from the beginning again. | |
8128 | * | |
8129 | */ | |
8130 | ||
8131 | do { | |
c622ca54 | 8132 | int mddev2_minor = -1; |
1da177e4 LT |
8133 | mddev->curr_resync = 2; |
8134 | ||
8135 | try_again: | |
404e4b43 | 8136 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 8137 | goto skip; |
29ac4aa3 | 8138 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
8139 | if (mddev2 == mddev) |
8140 | continue; | |
90b08710 BS |
8141 | if (!mddev->parallel_resync |
8142 | && mddev2->curr_resync | |
8143 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
8144 | DEFINE_WAIT(wq); |
8145 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
8146 | /* arbitrarily yield */ | |
8147 | mddev->curr_resync = 1; | |
8148 | wake_up(&resync_wait); | |
8149 | } | |
8150 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
8151 | /* no need to wait here, we can wait the next | |
8152 | * time 'round when curr_resync == 2 | |
8153 | */ | |
8154 | continue; | |
9744197c N |
8155 | /* We need to wait 'interruptible' so as not to |
8156 | * contribute to the load average, and not to | |
8157 | * be caught by 'softlockup' | |
8158 | */ | |
8159 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 8160 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 8161 | mddev2->curr_resync >= mddev->curr_resync) { |
c622ca54 AP |
8162 | if (mddev2_minor != mddev2->md_minor) { |
8163 | mddev2_minor = mddev2->md_minor; | |
9d48739e N |
8164 | pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n", |
8165 | desc, mdname(mddev), | |
8166 | mdname(mddev2)); | |
c622ca54 | 8167 | } |
1da177e4 | 8168 | mddev_put(mddev2); |
9744197c N |
8169 | if (signal_pending(current)) |
8170 | flush_signals(current); | |
1da177e4 LT |
8171 | schedule(); |
8172 | finish_wait(&resync_wait, &wq); | |
8173 | goto try_again; | |
8174 | } | |
8175 | finish_wait(&resync_wait, &wq); | |
8176 | } | |
8177 | } | |
8178 | } while (mddev->curr_resync < 2); | |
8179 | ||
5fd6c1dc | 8180 | j = 0; |
9d88883e | 8181 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 8182 | /* resync follows the size requested by the personality, |
57afd89f | 8183 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
8184 | */ |
8185 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 8186 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 8187 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
8188 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
8189 | j = mddev->resync_min; | |
8190 | else if (!mddev->bitmap) | |
5fd6c1dc | 8191 | j = mddev->recovery_cp; |
5e96ee65 | 8192 | |
ccfcc3c1 | 8193 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 8194 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 8195 | else { |
1da177e4 | 8196 | /* recovery follows the physical size of devices */ |
58c0fed4 | 8197 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 8198 | j = MaxSector; |
4e59ca7d | 8199 | rcu_read_lock(); |
dafb20fa | 8200 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 8201 | if (rdev->raid_disk >= 0 && |
f2076e7d | 8202 | !test_bit(Journal, &rdev->flags) && |
5fd6c1dc N |
8203 | !test_bit(Faulty, &rdev->flags) && |
8204 | !test_bit(In_sync, &rdev->flags) && | |
8205 | rdev->recovery_offset < j) | |
8206 | j = rdev->recovery_offset; | |
4e59ca7d | 8207 | rcu_read_unlock(); |
133d4527 N |
8208 | |
8209 | /* If there is a bitmap, we need to make sure all | |
8210 | * writes that started before we added a spare | |
8211 | * complete before we start doing a recovery. | |
8212 | * Otherwise the write might complete and (via | |
8213 | * bitmap_endwrite) set a bit in the bitmap after the | |
8214 | * recovery has checked that bit and skipped that | |
8215 | * region. | |
8216 | */ | |
8217 | if (mddev->bitmap) { | |
8218 | mddev->pers->quiesce(mddev, 1); | |
8219 | mddev->pers->quiesce(mddev, 0); | |
8220 | } | |
5fd6c1dc | 8221 | } |
1da177e4 | 8222 | |
9d48739e N |
8223 | pr_info("md: %s of RAID array %s\n", desc, mdname(mddev)); |
8224 | pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev)); | |
8225 | pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n", | |
8226 | speed_max(mddev), desc); | |
1da177e4 | 8227 | |
eea1bf38 | 8228 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 8229 | |
57afd89f | 8230 | io_sectors = 0; |
1da177e4 LT |
8231 | for (m = 0; m < SYNC_MARKS; m++) { |
8232 | mark[m] = jiffies; | |
57afd89f | 8233 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
8234 | } |
8235 | last_mark = 0; | |
8236 | mddev->resync_mark = mark[last_mark]; | |
8237 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
8238 | ||
8239 | /* | |
8240 | * Tune reconstruction: | |
8241 | */ | |
8242 | window = 32*(PAGE_SIZE/512); | |
9d48739e N |
8243 | pr_debug("md: using %dk window, over a total of %lluk.\n", |
8244 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
8245 | |
8246 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
8247 | last_check = 0; |
8248 | ||
8249 | if (j>2) { | |
9d48739e N |
8250 | pr_debug("md: resuming %s of %s from checkpoint.\n", |
8251 | desc, mdname(mddev)); | |
1da177e4 | 8252 | mddev->curr_resync = j; |
72f36d59 N |
8253 | } else |
8254 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 8255 | mddev->curr_resync_completed = j; |
72f36d59 N |
8256 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
8257 | md_new_event(mddev); | |
54f89341 | 8258 | update_time = jiffies; |
1da177e4 | 8259 | |
7c2c57c9 | 8260 | blk_start_plug(&plug); |
1da177e4 | 8261 | while (j < max_sectors) { |
57afd89f | 8262 | sector_t sectors; |
1da177e4 | 8263 | |
57afd89f | 8264 | skipped = 0; |
97e4f42d | 8265 | |
7a91ee1f N |
8266 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8267 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
8268 | (mddev->curr_resync - mddev->curr_resync_completed) | |
8269 | > (max_sectors >> 4)) || | |
54f89341 | 8270 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f | 8271 | (j - mddev->curr_resync_completed)*2 |
c5e19d90 N |
8272 | >= mddev->resync_max - mddev->curr_resync_completed || |
8273 | mddev->curr_resync_completed > mddev->resync_max | |
7a91ee1f | 8274 | )) { |
97e4f42d | 8275 | /* time to update curr_resync_completed */ |
97e4f42d N |
8276 | wait_event(mddev->recovery_wait, |
8277 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 8278 | mddev->curr_resync_completed = j; |
35d78c66 | 8279 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
8280 | j > mddev->recovery_cp) | |
8281 | mddev->recovery_cp = j; | |
54f89341 | 8282 | update_time = jiffies; |
2953079c | 8283 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
acb180b0 | 8284 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 8285 | } |
acb180b0 | 8286 | |
c91abf5a N |
8287 | while (j >= mddev->resync_max && |
8288 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
8289 | /* As this condition is controlled by user-space, |
8290 | * we can block indefinitely, so use '_interruptible' | |
8291 | * to avoid triggering warnings. | |
8292 | */ | |
8293 | flush_signals(current); /* just in case */ | |
8294 | wait_event_interruptible(mddev->recovery_wait, | |
8295 | mddev->resync_max > j | |
c91abf5a N |
8296 | || test_bit(MD_RECOVERY_INTR, |
8297 | &mddev->recovery)); | |
e62e58a5 | 8298 | } |
acb180b0 | 8299 | |
c91abf5a N |
8300 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8301 | break; | |
acb180b0 | 8302 | |
09314799 | 8303 | sectors = mddev->pers->sync_request(mddev, j, &skipped); |
57afd89f | 8304 | if (sectors == 0) { |
dfc70645 | 8305 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 8306 | break; |
1da177e4 | 8307 | } |
57afd89f N |
8308 | |
8309 | if (!skipped) { /* actual IO requested */ | |
8310 | io_sectors += sectors; | |
8311 | atomic_add(sectors, &mddev->recovery_active); | |
8312 | } | |
8313 | ||
e875ecea N |
8314 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8315 | break; | |
8316 | ||
1da177e4 | 8317 | j += sectors; |
5ed1df2e N |
8318 | if (j > max_sectors) |
8319 | /* when skipping, extra large numbers can be returned. */ | |
8320 | j = max_sectors; | |
72f36d59 N |
8321 | if (j > 2) |
8322 | mddev->curr_resync = j; | |
ff4e8d9a | 8323 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 8324 | if (last_check == 0) |
e875ecea | 8325 | /* this is the earliest that rebuild will be |
d7603b7e N |
8326 | * visible in /proc/mdstat |
8327 | */ | |
8328 | md_new_event(mddev); | |
57afd89f N |
8329 | |
8330 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
8331 | continue; |
8332 | ||
57afd89f | 8333 | last_check = io_sectors; |
1da177e4 LT |
8334 | repeat: |
8335 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
8336 | /* step marks */ | |
8337 | int next = (last_mark+1) % SYNC_MARKS; | |
8338 | ||
8339 | mddev->resync_mark = mark[next]; | |
8340 | mddev->resync_mark_cnt = mark_cnt[next]; | |
8341 | mark[next] = jiffies; | |
57afd89f | 8342 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
8343 | last_mark = next; |
8344 | } | |
8345 | ||
c91abf5a N |
8346 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8347 | break; | |
1da177e4 LT |
8348 | |
8349 | /* | |
8350 | * this loop exits only if either when we are slower than | |
8351 | * the 'hard' speed limit, or the system was IO-idle for | |
8352 | * a jiffy. | |
8353 | * the system might be non-idle CPU-wise, but we only care | |
8354 | * about not overloading the IO subsystem. (things like an | |
8355 | * e2fsck being done on the RAID array should execute fast) | |
8356 | */ | |
1da177e4 LT |
8357 | cond_resched(); |
8358 | ||
ac7e50a3 XN |
8359 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
8360 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 8361 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 8362 | |
88202a0c | 8363 | if (currspeed > speed_min(mddev)) { |
ac8fa419 | 8364 | if (currspeed > speed_max(mddev)) { |
c0e48521 | 8365 | msleep(500); |
1da177e4 LT |
8366 | goto repeat; |
8367 | } | |
ac8fa419 N |
8368 | if (!is_mddev_idle(mddev, 0)) { |
8369 | /* | |
8370 | * Give other IO more of a chance. | |
8371 | * The faster the devices, the less we wait. | |
8372 | */ | |
8373 | wait_event(mddev->recovery_wait, | |
8374 | !atomic_read(&mddev->recovery_active)); | |
8375 | } | |
1da177e4 LT |
8376 | } |
8377 | } | |
9d48739e N |
8378 | pr_info("md: %s: %s %s.\n",mdname(mddev), desc, |
8379 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
8380 | ? "interrupted" : "done"); | |
1da177e4 LT |
8381 | /* |
8382 | * this also signals 'finished resyncing' to md_stop | |
8383 | */ | |
7c2c57c9 | 8384 | blk_finish_plug(&plug); |
1da177e4 LT |
8385 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
8386 | ||
5ed1df2e N |
8387 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8388 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
1217e1d1 | 8389 | mddev->curr_resync > 3) { |
5ed1df2e N |
8390 | mddev->curr_resync_completed = mddev->curr_resync; |
8391 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
8392 | } | |
09314799 | 8393 | mddev->pers->sync_request(mddev, max_sectors, &skipped); |
1da177e4 | 8394 | |
dfc70645 | 8395 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
1217e1d1 | 8396 | mddev->curr_resync > 3) { |
5fd6c1dc N |
8397 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
8398 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8399 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
9d48739e N |
8400 | pr_debug("md: checkpointing %s of %s.\n", |
8401 | desc, mdname(mddev)); | |
0a19caab | 8402 | if (test_bit(MD_RECOVERY_ERROR, |
8403 | &mddev->recovery)) | |
8404 | mddev->recovery_cp = | |
8405 | mddev->curr_resync_completed; | |
8406 | else | |
8407 | mddev->recovery_cp = | |
8408 | mddev->curr_resync; | |
5fd6c1dc N |
8409 | } |
8410 | } else | |
8411 | mddev->recovery_cp = MaxSector; | |
8412 | } else { | |
8413 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
8414 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 8415 | rcu_read_lock(); |
dafb20fa | 8416 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 8417 | if (rdev->raid_disk >= 0 && |
70fffd0b | 8418 | mddev->delta_disks >= 0 && |
f2076e7d | 8419 | !test_bit(Journal, &rdev->flags) && |
5fd6c1dc N |
8420 | !test_bit(Faulty, &rdev->flags) && |
8421 | !test_bit(In_sync, &rdev->flags) && | |
8422 | rdev->recovery_offset < mddev->curr_resync) | |
8423 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 8424 | rcu_read_unlock(); |
5fd6c1dc | 8425 | } |
1da177e4 | 8426 | } |
db91ff55 | 8427 | skip: |
bb8bf15b GJ |
8428 | /* set CHANGE_PENDING here since maybe another update is needed, |
8429 | * so other nodes are informed. It should be harmless for normal | |
8430 | * raid */ | |
2953079c SL |
8431 | set_mask_bits(&mddev->sb_flags, 0, |
8432 | BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS)); | |
c186b128 | 8433 | |
23da422b | 8434 | spin_lock(&mddev->lock); |
c07b70ad N |
8435 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
8436 | /* We completed so min/max setting can be forgotten if used. */ | |
8437 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8438 | mddev->resync_min = 0; | |
8439 | mddev->resync_max = MaxSector; | |
8440 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8441 | mddev->resync_min = mddev->curr_resync_completed; | |
f7851be7 | 8442 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); |
1da177e4 | 8443 | mddev->curr_resync = 0; |
23da422b N |
8444 | spin_unlock(&mddev->lock); |
8445 | ||
1da177e4 | 8446 | wake_up(&resync_wait); |
1da177e4 | 8447 | md_wakeup_thread(mddev->thread); |
c6207277 | 8448 | return; |
1da177e4 | 8449 | } |
29269553 | 8450 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 8451 | |
746d3207 N |
8452 | static int remove_and_add_spares(struct mddev *mddev, |
8453 | struct md_rdev *this) | |
b4c4c7b8 | 8454 | { |
3cb03002 | 8455 | struct md_rdev *rdev; |
b4c4c7b8 | 8456 | int spares = 0; |
f2a371c5 | 8457 | int removed = 0; |
d787be40 | 8458 | bool remove_some = false; |
b4c4c7b8 | 8459 | |
d787be40 N |
8460 | rdev_for_each(rdev, mddev) { |
8461 | if ((this == NULL || rdev == this) && | |
8462 | rdev->raid_disk >= 0 && | |
8463 | !test_bit(Blocked, &rdev->flags) && | |
8464 | test_bit(Faulty, &rdev->flags) && | |
8465 | atomic_read(&rdev->nr_pending)==0) { | |
8466 | /* Faulty non-Blocked devices with nr_pending == 0 | |
8467 | * never get nr_pending incremented, | |
8468 | * never get Faulty cleared, and never get Blocked set. | |
8469 | * So we can synchronize_rcu now rather than once per device | |
8470 | */ | |
8471 | remove_some = true; | |
8472 | set_bit(RemoveSynchronized, &rdev->flags); | |
8473 | } | |
8474 | } | |
8475 | ||
8476 | if (remove_some) | |
8477 | synchronize_rcu(); | |
8478 | rdev_for_each(rdev, mddev) { | |
746d3207 N |
8479 | if ((this == NULL || rdev == this) && |
8480 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 8481 | !test_bit(Blocked, &rdev->flags) && |
d787be40 | 8482 | ((test_bit(RemoveSynchronized, &rdev->flags) || |
f2076e7d SL |
8483 | (!test_bit(In_sync, &rdev->flags) && |
8484 | !test_bit(Journal, &rdev->flags))) && | |
d787be40 | 8485 | atomic_read(&rdev->nr_pending)==0)) { |
b4c4c7b8 | 8486 | if (mddev->pers->hot_remove_disk( |
b8321b68 | 8487 | mddev, rdev) == 0) { |
36fad858 | 8488 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 8489 | rdev->raid_disk = -1; |
f2a371c5 | 8490 | removed++; |
b4c4c7b8 N |
8491 | } |
8492 | } | |
d787be40 N |
8493 | if (remove_some && test_bit(RemoveSynchronized, &rdev->flags)) |
8494 | clear_bit(RemoveSynchronized, &rdev->flags); | |
8495 | } | |
8496 | ||
90584fc9 JB |
8497 | if (removed && mddev->kobj.sd) |
8498 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 8499 | |
2910ff17 | 8500 | if (this && removed) |
746d3207 N |
8501 | goto no_add; |
8502 | ||
dafb20fa | 8503 | rdev_for_each(rdev, mddev) { |
2910ff17 GR |
8504 | if (this && this != rdev) |
8505 | continue; | |
dbb64f86 GR |
8506 | if (test_bit(Candidate, &rdev->flags)) |
8507 | continue; | |
7bfec5f3 N |
8508 | if (rdev->raid_disk >= 0 && |
8509 | !test_bit(In_sync, &rdev->flags) && | |
f2076e7d | 8510 | !test_bit(Journal, &rdev->flags) && |
7bfec5f3 N |
8511 | !test_bit(Faulty, &rdev->flags)) |
8512 | spares++; | |
7ceb17e8 N |
8513 | if (rdev->raid_disk >= 0) |
8514 | continue; | |
8515 | if (test_bit(Faulty, &rdev->flags)) | |
8516 | continue; | |
f6b6ec5c SL |
8517 | if (!test_bit(Journal, &rdev->flags)) { |
8518 | if (mddev->ro && | |
8519 | ! (rdev->saved_raid_disk >= 0 && | |
8520 | !test_bit(Bitmap_sync, &rdev->flags))) | |
8521 | continue; | |
7ceb17e8 | 8522 | |
f6b6ec5c SL |
8523 | rdev->recovery_offset = 0; |
8524 | } | |
7ceb17e8 N |
8525 | if (mddev->pers-> |
8526 | hot_add_disk(mddev, rdev) == 0) { | |
8527 | if (sysfs_link_rdev(mddev, rdev)) | |
8528 | /* failure here is OK */; | |
f6b6ec5c SL |
8529 | if (!test_bit(Journal, &rdev->flags)) |
8530 | spares++; | |
7ceb17e8 | 8531 | md_new_event(mddev); |
2953079c | 8532 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
dfc70645 | 8533 | } |
b4c4c7b8 | 8534 | } |
746d3207 | 8535 | no_add: |
6dafab6b | 8536 | if (removed) |
2953079c | 8537 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
b4c4c7b8 N |
8538 | return spares; |
8539 | } | |
7ebc0be7 | 8540 | |
ac05f256 N |
8541 | static void md_start_sync(struct work_struct *ws) |
8542 | { | |
8543 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
c186b128 | 8544 | |
ac05f256 N |
8545 | mddev->sync_thread = md_register_thread(md_do_sync, |
8546 | mddev, | |
8547 | "resync"); | |
8548 | if (!mddev->sync_thread) { | |
9d48739e N |
8549 | pr_warn("%s: could not start resync thread...\n", |
8550 | mdname(mddev)); | |
ac05f256 N |
8551 | /* leave the spares where they are, it shouldn't hurt */ |
8552 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8553 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8554 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8555 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
8556 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8557 | wake_up(&resync_wait); |
ac05f256 N |
8558 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8559 | &mddev->recovery)) | |
8560 | if (mddev->sysfs_action) | |
8561 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8562 | } else | |
8563 | md_wakeup_thread(mddev->sync_thread); | |
8564 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8565 | md_new_event(mddev); | |
8566 | } | |
8567 | ||
1da177e4 LT |
8568 | /* |
8569 | * This routine is regularly called by all per-raid-array threads to | |
8570 | * deal with generic issues like resync and super-block update. | |
8571 | * Raid personalities that don't have a thread (linear/raid0) do not | |
8572 | * need this as they never do any recovery or update the superblock. | |
8573 | * | |
8574 | * It does not do any resync itself, but rather "forks" off other threads | |
8575 | * to do that as needed. | |
8576 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
8577 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 8578 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
8579 | * and wakeups up this thread which will reap the thread and finish up. |
8580 | * This thread also removes any faulty devices (with nr_pending == 0). | |
8581 | * | |
8582 | * The overall approach is: | |
8583 | * 1/ if the superblock needs updating, update it. | |
8584 | * 2/ If a recovery thread is running, don't do anything else. | |
8585 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
8586 | * 4/ If there are any faulty devices, remove them. | |
8587 | * 5/ If array is degraded, try to add spares devices | |
8588 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
8589 | */ | |
fd01b88c | 8590 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 8591 | { |
68866e42 JB |
8592 | if (mddev->suspended) |
8593 | return; | |
8594 | ||
5f40402d | 8595 | if (mddev->bitmap) |
aa5cbd10 | 8596 | bitmap_daemon_work(mddev); |
1da177e4 | 8597 | |
fca4d848 | 8598 | if (signal_pending(current)) { |
31a59e34 | 8599 | if (mddev->pers->sync_request && !mddev->external) { |
9d48739e N |
8600 | pr_debug("md: %s in immediate safe mode\n", |
8601 | mdname(mddev)); | |
fca4d848 N |
8602 | mddev->safemode = 2; |
8603 | } | |
8604 | flush_signals(current); | |
8605 | } | |
8606 | ||
c89a8eee N |
8607 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
8608 | return; | |
1da177e4 | 8609 | if ( ! ( |
2953079c | 8610 | (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) || |
1da177e4 | 8611 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 8612 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 8613 | (mddev->external == 0 && mddev->safemode == 1) || |
4ad23a97 | 8614 | (mddev->safemode == 2 |
fca4d848 | 8615 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) |
1da177e4 LT |
8616 | )) |
8617 | return; | |
fca4d848 | 8618 | |
df5b89b3 | 8619 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 8620 | int spares = 0; |
fca4d848 | 8621 | |
c89a8eee | 8622 | if (mddev->ro) { |
ab16bfc7 NB |
8623 | struct md_rdev *rdev; |
8624 | if (!mddev->external && mddev->in_sync) | |
8625 | /* 'Blocked' flag not needed as failed devices | |
8626 | * will be recorded if array switched to read/write. | |
8627 | * Leaving it set will prevent the device | |
8628 | * from being removed. | |
8629 | */ | |
8630 | rdev_for_each(rdev, mddev) | |
8631 | clear_bit(Blocked, &rdev->flags); | |
7ceb17e8 N |
8632 | /* On a read-only array we can: |
8633 | * - remove failed devices | |
8634 | * - add already-in_sync devices if the array itself | |
8635 | * is in-sync. | |
8636 | * As we only add devices that are already in-sync, | |
8637 | * we can activate the spares immediately. | |
c89a8eee | 8638 | */ |
7ceb17e8 | 8639 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
8640 | /* There is no thread, but we need to call |
8641 | * ->spare_active and clear saved_raid_disk | |
8642 | */ | |
2ac295a5 | 8643 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 | 8644 | md_reap_sync_thread(mddev); |
a4a3d26d | 8645 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
8313b8e5 | 8646 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2953079c | 8647 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
c89a8eee N |
8648 | goto unlock; |
8649 | } | |
8650 | ||
659b254f GJ |
8651 | if (mddev_is_clustered(mddev)) { |
8652 | struct md_rdev *rdev; | |
8653 | /* kick the device if another node issued a | |
8654 | * remove disk. | |
8655 | */ | |
8656 | rdev_for_each(rdev, mddev) { | |
8657 | if (test_and_clear_bit(ClusterRemove, &rdev->flags) && | |
8658 | rdev->raid_disk < 0) | |
8659 | md_kick_rdev_from_array(rdev); | |
8660 | } | |
8661 | } | |
8662 | ||
6497709b | 8663 | if (!mddev->external && !mddev->in_sync) { |
85572d7c | 8664 | spin_lock(&mddev->lock); |
6497709b | 8665 | set_in_sync(mddev); |
85572d7c | 8666 | spin_unlock(&mddev->lock); |
fca4d848 | 8667 | } |
fca4d848 | 8668 | |
2953079c | 8669 | if (mddev->sb_flags) |
850b2b42 | 8670 | md_update_sb(mddev, 0); |
06d91a5f | 8671 | |
1da177e4 LT |
8672 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
8673 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
8674 | /* resync/recovery still happening */ | |
8675 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8676 | goto unlock; | |
8677 | } | |
8678 | if (mddev->sync_thread) { | |
a91d5ac0 | 8679 | md_reap_sync_thread(mddev); |
1da177e4 LT |
8680 | goto unlock; |
8681 | } | |
72a23c21 NB |
8682 | /* Set RUNNING before clearing NEEDED to avoid |
8683 | * any transients in the value of "sync_action". | |
8684 | */ | |
72f36d59 | 8685 | mddev->curr_resync_completed = 0; |
23da422b | 8686 | spin_lock(&mddev->lock); |
72a23c21 | 8687 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 8688 | spin_unlock(&mddev->lock); |
24dd469d N |
8689 | /* Clear some bits that don't mean anything, but |
8690 | * might be left set | |
8691 | */ | |
24dd469d N |
8692 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8693 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 8694 | |
ed209584 N |
8695 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
8696 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 8697 | goto not_running; |
1da177e4 LT |
8698 | /* no recovery is running. |
8699 | * remove any failed drives, then | |
8700 | * add spares if possible. | |
72f36d59 | 8701 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
8702 | * the personality to fail the re-add. |
8703 | */ | |
1da177e4 | 8704 | |
b4c4c7b8 | 8705 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
8706 | if (mddev->pers->check_reshape == NULL || |
8707 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 8708 | /* Cannot proceed */ |
ac05f256 | 8709 | goto not_running; |
b4c4c7b8 | 8710 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 8711 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 8712 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
8713 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8714 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 8715 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 8716 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8717 | } else if (mddev->recovery_cp < MaxSector) { |
8718 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 8719 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8720 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
8721 | /* nothing to be done ... */ | |
ac05f256 | 8722 | goto not_running; |
24dd469d | 8723 | |
1da177e4 | 8724 | if (mddev->pers->sync_request) { |
ef99bf48 | 8725 | if (spares) { |
a654b9d8 N |
8726 | /* We are adding a device or devices to an array |
8727 | * which has the bitmap stored on all devices. | |
8728 | * So make sure all bitmap pages get written | |
8729 | */ | |
8730 | bitmap_write_all(mddev->bitmap); | |
8731 | } | |
ac05f256 N |
8732 | INIT_WORK(&mddev->del_work, md_start_sync); |
8733 | queue_work(md_misc_wq, &mddev->del_work); | |
8734 | goto unlock; | |
1da177e4 | 8735 | } |
ac05f256 | 8736 | not_running: |
72a23c21 NB |
8737 | if (!mddev->sync_thread) { |
8738 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8739 | wake_up(&resync_wait); |
72a23c21 NB |
8740 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8741 | &mddev->recovery)) | |
0c3573f1 | 8742 | if (mddev->sysfs_action) |
00bcb4ac | 8743 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 8744 | } |
ac05f256 N |
8745 | unlock: |
8746 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
8747 | mddev_unlock(mddev); |
8748 | } | |
8749 | } | |
6c144d31 | 8750 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8751 | |
a91d5ac0 JB |
8752 | void md_reap_sync_thread(struct mddev *mddev) |
8753 | { | |
8754 | struct md_rdev *rdev; | |
8755 | ||
8756 | /* resync has finished, collect result */ | |
8757 | md_unregister_thread(&mddev->sync_thread); | |
8758 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8759 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8760 | /* success...*/ | |
8761 | /* activate any spares */ | |
8762 | if (mddev->pers->spare_active(mddev)) { | |
8763 | sysfs_notify(&mddev->kobj, NULL, | |
8764 | "degraded"); | |
2953079c | 8765 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a91d5ac0 JB |
8766 | } |
8767 | } | |
8768 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
8769 | mddev->pers->finish_reshape) | |
8770 | mddev->pers->finish_reshape(mddev); | |
8771 | ||
8772 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8773 | * information must be scrapped. |
a91d5ac0 | 8774 | */ |
f466722c N |
8775 | if (!mddev->degraded) |
8776 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8777 | rdev->saved_raid_disk = -1; |
8778 | ||
8779 | md_update_sb(mddev, 1); | |
2953079c | 8780 | /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can |
bb8bf15b GJ |
8781 | * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by |
8782 | * clustered raid */ | |
8783 | if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags)) | |
8784 | md_cluster_ops->resync_finish(mddev); | |
a91d5ac0 | 8785 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
ea358cd0 | 8786 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
a91d5ac0 JB |
8787 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8788 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8789 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8790 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8791 | wake_up(&resync_wait); |
a91d5ac0 JB |
8792 | /* flag recovery needed just to double check */ |
8793 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8794 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8795 | md_new_event(mddev); | |
8796 | if (mddev->event_work.func) | |
8797 | queue_work(md_misc_wq, &mddev->event_work); | |
8798 | } | |
6c144d31 | 8799 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 8800 | |
fd01b88c | 8801 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8802 | { |
00bcb4ac | 8803 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8804 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8805 | !test_bit(Blocked, &rdev->flags) && |
8806 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8807 | msecs_to_jiffies(5000)); |
8808 | rdev_dec_pending(rdev, mddev); | |
8809 | } | |
8810 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8811 | ||
c6563a8c N |
8812 | void md_finish_reshape(struct mddev *mddev) |
8813 | { | |
8814 | /* called be personality module when reshape completes. */ | |
8815 | struct md_rdev *rdev; | |
8816 | ||
8817 | rdev_for_each(rdev, mddev) { | |
8818 | if (rdev->data_offset > rdev->new_data_offset) | |
8819 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8820 | else | |
8821 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8822 | rdev->data_offset = rdev->new_data_offset; | |
8823 | } | |
8824 | } | |
8825 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 | 8826 | |
fc974ee2 | 8827 | /* Bad block management */ |
2230dfe4 | 8828 | |
fc974ee2 | 8829 | /* Returns 1 on success, 0 on failure */ |
3cb03002 | 8830 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8831 | int is_new) |
2230dfe4 | 8832 | { |
85ad1d13 | 8833 | struct mddev *mddev = rdev->mddev; |
c6563a8c N |
8834 | int rv; |
8835 | if (is_new) | |
8836 | s += rdev->new_data_offset; | |
8837 | else | |
8838 | s += rdev->data_offset; | |
fc974ee2 VV |
8839 | rv = badblocks_set(&rdev->badblocks, s, sectors, 0); |
8840 | if (rv == 0) { | |
2230dfe4 | 8841 | /* Make sure they get written out promptly */ |
35b785f7 TM |
8842 | if (test_bit(ExternalBbl, &rdev->flags)) |
8843 | sysfs_notify(&rdev->kobj, NULL, | |
8844 | "unacknowledged_bad_blocks"); | |
8bd2f0a0 | 8845 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2953079c SL |
8846 | set_mask_bits(&mddev->sb_flags, 0, |
8847 | BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING)); | |
2230dfe4 | 8848 | md_wakeup_thread(rdev->mddev->thread); |
fc974ee2 VV |
8849 | return 1; |
8850 | } else | |
8851 | return 0; | |
2230dfe4 N |
8852 | } |
8853 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8854 | ||
c6563a8c N |
8855 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8856 | int is_new) | |
2230dfe4 | 8857 | { |
35b785f7 | 8858 | int rv; |
c6563a8c N |
8859 | if (is_new) |
8860 | s += rdev->new_data_offset; | |
8861 | else | |
8862 | s += rdev->data_offset; | |
35b785f7 TM |
8863 | rv = badblocks_clear(&rdev->badblocks, s, sectors); |
8864 | if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags)) | |
8865 | sysfs_notify(&rdev->kobj, NULL, "bad_blocks"); | |
8866 | return rv; | |
2230dfe4 N |
8867 | } |
8868 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8869 | ||
75c96f85 AB |
8870 | static int md_notify_reboot(struct notifier_block *this, |
8871 | unsigned long code, void *x) | |
1da177e4 LT |
8872 | { |
8873 | struct list_head *tmp; | |
fd01b88c | 8874 | struct mddev *mddev; |
2dba6a91 | 8875 | int need_delay = 0; |
1da177e4 | 8876 | |
c744a65c N |
8877 | for_each_mddev(mddev, tmp) { |
8878 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8879 | if (mddev->pers) |
8880 | __md_stop_writes(mddev); | |
0f62fb22 N |
8881 | if (mddev->persistent) |
8882 | mddev->safemode = 2; | |
c744a65c | 8883 | mddev_unlock(mddev); |
2dba6a91 | 8884 | } |
c744a65c | 8885 | need_delay = 1; |
1da177e4 | 8886 | } |
c744a65c N |
8887 | /* |
8888 | * certain more exotic SCSI devices are known to be | |
8889 | * volatile wrt too early system reboots. While the | |
8890 | * right place to handle this issue is the given | |
8891 | * driver, we do want to have a safe RAID driver ... | |
8892 | */ | |
8893 | if (need_delay) | |
8894 | mdelay(1000*1); | |
8895 | ||
1da177e4 LT |
8896 | return NOTIFY_DONE; |
8897 | } | |
8898 | ||
75c96f85 | 8899 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8900 | .notifier_call = md_notify_reboot, |
8901 | .next = NULL, | |
8902 | .priority = INT_MAX, /* before any real devices */ | |
8903 | }; | |
8904 | ||
8905 | static void md_geninit(void) | |
8906 | { | |
36a4e1fe | 8907 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8908 | |
c7705f34 | 8909 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8910 | } |
8911 | ||
75c96f85 | 8912 | static int __init md_init(void) |
1da177e4 | 8913 | { |
e804ac78 TH |
8914 | int ret = -ENOMEM; |
8915 | ||
ada609ee | 8916 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8917 | if (!md_wq) |
8918 | goto err_wq; | |
8919 | ||
8920 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8921 | if (!md_misc_wq) | |
8922 | goto err_misc_wq; | |
8923 | ||
8924 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8925 | goto err_md; | |
8926 | ||
8927 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8928 | goto err_mdp; | |
8929 | mdp_major = ret; | |
8930 | ||
af5628f0 | 8931 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
8932 | md_probe, NULL, NULL); |
8933 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8934 | md_probe, NULL, NULL); |
8935 | ||
1da177e4 | 8936 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8937 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8938 | |
8939 | md_geninit(); | |
d710e138 | 8940 | return 0; |
1da177e4 | 8941 | |
e804ac78 TH |
8942 | err_mdp: |
8943 | unregister_blkdev(MD_MAJOR, "md"); | |
8944 | err_md: | |
8945 | destroy_workqueue(md_misc_wq); | |
8946 | err_misc_wq: | |
8947 | destroy_workqueue(md_wq); | |
8948 | err_wq: | |
8949 | return ret; | |
8950 | } | |
1da177e4 | 8951 | |
70bcecdb | 8952 | static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev) |
1d7e3e96 | 8953 | { |
70bcecdb GR |
8954 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
8955 | struct md_rdev *rdev2; | |
8956 | int role, ret; | |
8957 | char b[BDEVNAME_SIZE]; | |
1d7e3e96 | 8958 | |
818da59f GJ |
8959 | /* |
8960 | * If size is changed in another node then we need to | |
8961 | * do resize as well. | |
8962 | */ | |
8963 | if (mddev->dev_sectors != le64_to_cpu(sb->size)) { | |
8964 | ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size)); | |
8965 | if (ret) | |
8966 | pr_info("md-cluster: resize failed\n"); | |
8967 | else | |
8968 | bitmap_update_sb(mddev->bitmap); | |
8969 | } | |
8970 | ||
70bcecdb GR |
8971 | /* Check for change of roles in the active devices */ |
8972 | rdev_for_each(rdev2, mddev) { | |
8973 | if (test_bit(Faulty, &rdev2->flags)) | |
8974 | continue; | |
8975 | ||
8976 | /* Check if the roles changed */ | |
8977 | role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]); | |
dbb64f86 GR |
8978 | |
8979 | if (test_bit(Candidate, &rdev2->flags)) { | |
8980 | if (role == 0xfffe) { | |
8981 | pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b)); | |
8982 | md_kick_rdev_from_array(rdev2); | |
8983 | continue; | |
8984 | } | |
8985 | else | |
8986 | clear_bit(Candidate, &rdev2->flags); | |
8987 | } | |
8988 | ||
70bcecdb GR |
8989 | if (role != rdev2->raid_disk) { |
8990 | /* got activated */ | |
8991 | if (rdev2->raid_disk == -1 && role != 0xffff) { | |
8992 | rdev2->saved_raid_disk = role; | |
8993 | ret = remove_and_add_spares(mddev, rdev2); | |
8994 | pr_info("Activated spare: %s\n", | |
9d48739e | 8995 | bdevname(rdev2->bdev,b)); |
a578183e GJ |
8996 | /* wakeup mddev->thread here, so array could |
8997 | * perform resync with the new activated disk */ | |
8998 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8999 | md_wakeup_thread(mddev->thread); | |
9000 | ||
70bcecdb GR |
9001 | } |
9002 | /* device faulty | |
9003 | * We just want to do the minimum to mark the disk | |
9004 | * as faulty. The recovery is performed by the | |
9005 | * one who initiated the error. | |
9006 | */ | |
9007 | if ((role == 0xfffe) || (role == 0xfffd)) { | |
9008 | md_error(mddev, rdev2); | |
9009 | clear_bit(Blocked, &rdev2->flags); | |
9010 | } | |
9011 | } | |
1d7e3e96 | 9012 | } |
70bcecdb | 9013 | |
28c1b9fd GR |
9014 | if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) |
9015 | update_raid_disks(mddev, le32_to_cpu(sb->raid_disks)); | |
70bcecdb GR |
9016 | |
9017 | /* Finally set the event to be up to date */ | |
9018 | mddev->events = le64_to_cpu(sb->events); | |
9019 | } | |
9020 | ||
9021 | static int read_rdev(struct mddev *mddev, struct md_rdev *rdev) | |
9022 | { | |
9023 | int err; | |
9024 | struct page *swapout = rdev->sb_page; | |
9025 | struct mdp_superblock_1 *sb; | |
9026 | ||
9027 | /* Store the sb page of the rdev in the swapout temporary | |
9028 | * variable in case we err in the future | |
9029 | */ | |
9030 | rdev->sb_page = NULL; | |
7f0f0d87 N |
9031 | err = alloc_disk_sb(rdev); |
9032 | if (err == 0) { | |
9033 | ClearPageUptodate(rdev->sb_page); | |
9034 | rdev->sb_loaded = 0; | |
9035 | err = super_types[mddev->major_version]. | |
9036 | load_super(rdev, NULL, mddev->minor_version); | |
9037 | } | |
70bcecdb GR |
9038 | if (err < 0) { |
9039 | pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n", | |
9040 | __func__, __LINE__, rdev->desc_nr, err); | |
7f0f0d87 N |
9041 | if (rdev->sb_page) |
9042 | put_page(rdev->sb_page); | |
70bcecdb GR |
9043 | rdev->sb_page = swapout; |
9044 | rdev->sb_loaded = 1; | |
9045 | return err; | |
1d7e3e96 GR |
9046 | } |
9047 | ||
70bcecdb GR |
9048 | sb = page_address(rdev->sb_page); |
9049 | /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET | |
9050 | * is not set | |
9051 | */ | |
9052 | ||
9053 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)) | |
9054 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
9055 | ||
9056 | /* The other node finished recovery, call spare_active to set | |
9057 | * device In_sync and mddev->degraded | |
9058 | */ | |
9059 | if (rdev->recovery_offset == MaxSector && | |
9060 | !test_bit(In_sync, &rdev->flags) && | |
9061 | mddev->pers->spare_active(mddev)) | |
9062 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
9063 | ||
9064 | put_page(swapout); | |
9065 | return 0; | |
9066 | } | |
9067 | ||
9068 | void md_reload_sb(struct mddev *mddev, int nr) | |
9069 | { | |
9070 | struct md_rdev *rdev; | |
9071 | int err; | |
9072 | ||
9073 | /* Find the rdev */ | |
9074 | rdev_for_each_rcu(rdev, mddev) { | |
9075 | if (rdev->desc_nr == nr) | |
9076 | break; | |
9077 | } | |
9078 | ||
9079 | if (!rdev || rdev->desc_nr != nr) { | |
9080 | pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr); | |
9081 | return; | |
9082 | } | |
9083 | ||
9084 | err = read_rdev(mddev, rdev); | |
9085 | if (err < 0) | |
9086 | return; | |
9087 | ||
9088 | check_sb_changes(mddev, rdev); | |
9089 | ||
9090 | /* Read all rdev's to update recovery_offset */ | |
9091 | rdev_for_each_rcu(rdev, mddev) | |
9092 | read_rdev(mddev, rdev); | |
1d7e3e96 GR |
9093 | } |
9094 | EXPORT_SYMBOL(md_reload_sb); | |
9095 | ||
1da177e4 LT |
9096 | #ifndef MODULE |
9097 | ||
9098 | /* | |
9099 | * Searches all registered partitions for autorun RAID arrays | |
9100 | * at boot time. | |
9101 | */ | |
4d936ec1 | 9102 | |
5b1f5bc3 | 9103 | static DEFINE_MUTEX(detected_devices_mutex); |
4d936ec1 ME |
9104 | static LIST_HEAD(all_detected_devices); |
9105 | struct detected_devices_node { | |
9106 | struct list_head list; | |
9107 | dev_t dev; | |
9108 | }; | |
1da177e4 LT |
9109 | |
9110 | void md_autodetect_dev(dev_t dev) | |
9111 | { | |
4d936ec1 ME |
9112 | struct detected_devices_node *node_detected_dev; |
9113 | ||
9114 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
9115 | if (node_detected_dev) { | |
9116 | node_detected_dev->dev = dev; | |
5b1f5bc3 | 9117 | mutex_lock(&detected_devices_mutex); |
4d936ec1 | 9118 | list_add_tail(&node_detected_dev->list, &all_detected_devices); |
5b1f5bc3 | 9119 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 9120 | } |
1da177e4 LT |
9121 | } |
9122 | ||
1da177e4 LT |
9123 | static void autostart_arrays(int part) |
9124 | { | |
3cb03002 | 9125 | struct md_rdev *rdev; |
4d936ec1 ME |
9126 | struct detected_devices_node *node_detected_dev; |
9127 | dev_t dev; | |
9128 | int i_scanned, i_passed; | |
1da177e4 | 9129 | |
4d936ec1 ME |
9130 | i_scanned = 0; |
9131 | i_passed = 0; | |
1da177e4 | 9132 | |
9d48739e | 9133 | pr_info("md: Autodetecting RAID arrays.\n"); |
1da177e4 | 9134 | |
5b1f5bc3 | 9135 | mutex_lock(&detected_devices_mutex); |
4d936ec1 ME |
9136 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
9137 | i_scanned++; | |
9138 | node_detected_dev = list_entry(all_detected_devices.next, | |
9139 | struct detected_devices_node, list); | |
9140 | list_del(&node_detected_dev->list); | |
9141 | dev = node_detected_dev->dev; | |
9142 | kfree(node_detected_dev); | |
90bcf133 | 9143 | mutex_unlock(&detected_devices_mutex); |
df968c4e | 9144 | rdev = md_import_device(dev,0, 90); |
90bcf133 | 9145 | mutex_lock(&detected_devices_mutex); |
1da177e4 LT |
9146 | if (IS_ERR(rdev)) |
9147 | continue; | |
9148 | ||
403df478 | 9149 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 9150 | continue; |
403df478 | 9151 | |
d0fae18f | 9152 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 9153 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 9154 | i_passed++; |
1da177e4 | 9155 | } |
5b1f5bc3 | 9156 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 9157 | |
9d48739e | 9158 | pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed); |
1da177e4 LT |
9159 | |
9160 | autorun_devices(part); | |
9161 | } | |
9162 | ||
fdee8ae4 | 9163 | #endif /* !MODULE */ |
1da177e4 LT |
9164 | |
9165 | static __exit void md_exit(void) | |
9166 | { | |
fd01b88c | 9167 | struct mddev *mddev; |
1da177e4 | 9168 | struct list_head *tmp; |
e2f23b60 | 9169 | int delay = 1; |
8ab5e4c1 | 9170 | |
af5628f0 | 9171 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 9172 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 9173 | |
3dbd8c2e | 9174 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
9175 | unregister_blkdev(mdp_major, "mdp"); |
9176 | unregister_reboot_notifier(&md_notifier); | |
9177 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
9178 | |
9179 | /* We cannot unload the modules while some process is | |
9180 | * waiting for us in select() or poll() - wake them up | |
9181 | */ | |
9182 | md_unloading = 1; | |
9183 | while (waitqueue_active(&md_event_waiters)) { | |
9184 | /* not safe to leave yet */ | |
9185 | wake_up(&md_event_waiters); | |
9186 | msleep(delay); | |
9187 | delay += delay; | |
9188 | } | |
1da177e4 | 9189 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 9190 | |
29ac4aa3 | 9191 | for_each_mddev(mddev, tmp) { |
1da177e4 | 9192 | export_array(mddev); |
9356863c | 9193 | mddev->ctime = 0; |
d3374825 | 9194 | mddev->hold_active = 0; |
9356863c N |
9195 | /* |
9196 | * for_each_mddev() will call mddev_put() at the end of each | |
9197 | * iteration. As the mddev is now fully clear, this will | |
9198 | * schedule the mddev for destruction by a workqueue, and the | |
9199 | * destroy_workqueue() below will wait for that to complete. | |
9200 | */ | |
1da177e4 | 9201 | } |
e804ac78 TH |
9202 | destroy_workqueue(md_misc_wq); |
9203 | destroy_workqueue(md_wq); | |
1da177e4 LT |
9204 | } |
9205 | ||
685784aa | 9206 | subsys_initcall(md_init); |
1da177e4 LT |
9207 | module_exit(md_exit) |
9208 | ||
f91de92e N |
9209 | static int get_ro(char *buffer, struct kernel_param *kp) |
9210 | { | |
9211 | return sprintf(buffer, "%d", start_readonly); | |
9212 | } | |
9213 | static int set_ro(const char *val, struct kernel_param *kp) | |
9214 | { | |
4c9309c0 | 9215 | return kstrtouint(val, 10, (unsigned int *)&start_readonly); |
f91de92e N |
9216 | } |
9217 | ||
80ca3a44 N |
9218 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
9219 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 9220 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
78b6350d | 9221 | module_param(create_on_open, bool, S_IRUSR|S_IWUSR); |
f91de92e | 9222 | |
1da177e4 | 9223 | MODULE_LICENSE("GPL"); |
0efb9e61 | 9224 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 9225 | MODULE_ALIAS("md"); |
72008652 | 9226 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |