]>
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" |
935fe098 | 72 | #include "md-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 | 187 | /* bio_clone_mddev |
9b10f6a9 | 188 | * like bio_clone_bioset, but with a local bio set |
a167f663 N |
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 | ||
5a85071c N |
206 | static struct bio *md_bio_alloc_sync(struct mddev *mddev) |
207 | { | |
7f053a6a | 208 | if (!mddev || !mddev->sync_set) |
5a85071c N |
209 | return bio_alloc(GFP_NOIO, 1); |
210 | ||
211 | return bio_alloc_bioset(GFP_NOIO, 1, mddev->sync_set); | |
212 | } | |
213 | ||
d7603b7e N |
214 | /* |
215 | * We have a system wide 'event count' that is incremented | |
216 | * on any 'interesting' event, and readers of /proc/mdstat | |
217 | * can use 'poll' or 'select' to find out when the event | |
218 | * count increases. | |
219 | * | |
220 | * Events are: | |
221 | * start array, stop array, error, add device, remove device, | |
222 | * start build, activate spare | |
223 | */ | |
2989ddbd | 224 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 225 | static atomic_t md_event_count; |
fd01b88c | 226 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
227 | { |
228 | atomic_inc(&md_event_count); | |
229 | wake_up(&md_event_waiters); | |
230 | } | |
29269553 | 231 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 232 | |
1da177e4 LT |
233 | /* |
234 | * Enables to iterate over all existing md arrays | |
235 | * all_mddevs_lock protects this list. | |
236 | */ | |
237 | static LIST_HEAD(all_mddevs); | |
238 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
239 | ||
1da177e4 LT |
240 | /* |
241 | * iterates through all used mddevs in the system. | |
242 | * We take care to grab the all_mddevs_lock whenever navigating | |
243 | * the list, and to always hold a refcount when unlocked. | |
244 | * Any code which breaks out of this loop while own | |
245 | * a reference to the current mddev and must mddev_put it. | |
246 | */ | |
fd01b88c | 247 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 | 248 | \ |
f72ffdd6 | 249 | for (({ spin_lock(&all_mddevs_lock); \ |
fd01b88c N |
250 | _tmp = all_mddevs.next; \ |
251 | _mddev = NULL;}); \ | |
252 | ({ if (_tmp != &all_mddevs) \ | |
253 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 254 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
255 | if (_mddev) mddev_put(_mddev); \ |
256 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
257 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 258 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 259 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
260 | ) |
261 | ||
409c57f3 N |
262 | /* Rather than calling directly into the personality make_request function, |
263 | * IO requests come here first so that we can check if the device is | |
264 | * being suspended pending a reconfiguration. | |
265 | * We hold a refcount over the call to ->make_request. By the time that | |
266 | * call has finished, the bio has been linked into some internal structure | |
267 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
268 | */ | |
393debc2 SL |
269 | void md_handle_request(struct mddev *mddev, struct bio *bio) |
270 | { | |
271 | check_suspended: | |
272 | rcu_read_lock(); | |
273 | if (mddev->suspended) { | |
274 | DEFINE_WAIT(__wait); | |
275 | for (;;) { | |
276 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
277 | TASK_UNINTERRUPTIBLE); | |
278 | if (!mddev->suspended) | |
279 | break; | |
280 | rcu_read_unlock(); | |
281 | schedule(); | |
282 | rcu_read_lock(); | |
283 | } | |
284 | finish_wait(&mddev->sb_wait, &__wait); | |
285 | } | |
286 | atomic_inc(&mddev->active_io); | |
287 | rcu_read_unlock(); | |
288 | ||
289 | if (!mddev->pers->make_request(mddev, bio)) { | |
290 | atomic_dec(&mddev->active_io); | |
291 | wake_up(&mddev->sb_wait); | |
292 | goto check_suspended; | |
293 | } | |
294 | ||
295 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) | |
296 | wake_up(&mddev->sb_wait); | |
297 | } | |
298 | EXPORT_SYMBOL(md_handle_request); | |
299 | ||
dece1635 | 300 | static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 301 | { |
49077326 | 302 | const int rw = bio_data_dir(bio); |
fd01b88c | 303 | struct mddev *mddev = q->queuedata; |
e91ece55 | 304 | unsigned int sectors; |
74672d06 | 305 | int cpu; |
49077326 | 306 | |
af67c31f | 307 | blk_queue_split(q, &bio); |
54efd50b | 308 | |
274d8cbd | 309 | if (mddev == NULL || mddev->pers == NULL) { |
409c57f3 | 310 | bio_io_error(bio); |
dece1635 | 311 | return BLK_QC_T_NONE; |
409c57f3 | 312 | } |
bbfa57c0 | 313 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
4246a0b6 | 314 | if (bio_sectors(bio) != 0) |
4e4cbee9 | 315 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 316 | bio_endio(bio); |
dece1635 | 317 | return BLK_QC_T_NONE; |
bbfa57c0 | 318 | } |
49077326 | 319 | |
e91ece55 CM |
320 | /* |
321 | * save the sectors now since our bio can | |
322 | * go away inside make_request | |
323 | */ | |
324 | sectors = bio_sectors(bio); | |
9c573de3 | 325 | /* bio could be mergeable after passing to underlayer */ |
1eff9d32 | 326 | bio->bi_opf &= ~REQ_NOMERGE; |
393debc2 SL |
327 | |
328 | md_handle_request(mddev, bio); | |
49077326 | 329 | |
74672d06 GZ |
330 | cpu = part_stat_lock(); |
331 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
332 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); | |
333 | part_stat_unlock(); | |
49077326 | 334 | |
dece1635 | 335 | return BLK_QC_T_NONE; |
409c57f3 N |
336 | } |
337 | ||
9e35b99c N |
338 | /* mddev_suspend makes sure no new requests are submitted |
339 | * to the device, and that any requests that have been submitted | |
340 | * are completely handled. | |
afa0f557 N |
341 | * Once mddev_detach() is called and completes, the module will be |
342 | * completely unused. | |
9e35b99c | 343 | */ |
fd01b88c | 344 | void mddev_suspend(struct mddev *mddev) |
409c57f3 | 345 | { |
092398dc | 346 | WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk); |
0dc10e50 MP |
347 | if (mddev->suspended++) |
348 | return; | |
409c57f3 | 349 | synchronize_rcu(); |
cc27b0c7 | 350 | wake_up(&mddev->sb_wait); |
409c57f3 N |
351 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); |
352 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
353 | |
354 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 355 | } |
390ee602 | 356 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 357 | |
fd01b88c | 358 | void mddev_resume(struct mddev *mddev) |
409c57f3 | 359 | { |
0dc10e50 MP |
360 | if (--mddev->suspended) |
361 | return; | |
409c57f3 N |
362 | wake_up(&mddev->sb_wait); |
363 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 364 | |
47525e59 | 365 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
366 | md_wakeup_thread(mddev->thread); |
367 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 368 | } |
390ee602 | 369 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 370 | |
fd01b88c | 371 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 | 372 | { |
5c675f83 N |
373 | struct md_personality *pers = mddev->pers; |
374 | int ret = 0; | |
375 | ||
376 | rcu_read_lock(); | |
377 | if (mddev->suspended) | |
378 | ret = 1; | |
379 | else if (pers && pers->congested) | |
380 | ret = pers->congested(mddev, bits); | |
381 | rcu_read_unlock(); | |
382 | return ret; | |
383 | } | |
384 | EXPORT_SYMBOL_GPL(mddev_congested); | |
385 | static int md_congested(void *data, int bits) | |
386 | { | |
387 | struct mddev *mddev = data; | |
388 | return mddev_congested(mddev, bits); | |
3fa841d7 | 389 | } |
3fa841d7 | 390 | |
a2826aa9 | 391 | /* |
e9c7469b | 392 | * Generic flush handling for md |
a2826aa9 N |
393 | */ |
394 | ||
4246a0b6 | 395 | static void md_end_flush(struct bio *bio) |
a2826aa9 | 396 | { |
3cb03002 | 397 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 398 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
399 | |
400 | rdev_dec_pending(rdev, mddev); | |
401 | ||
402 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 403 | /* The pre-request flush has finished */ |
e804ac78 | 404 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
405 | } |
406 | bio_put(bio); | |
407 | } | |
408 | ||
a7a07e69 N |
409 | static void md_submit_flush_data(struct work_struct *ws); |
410 | ||
a035fc3e | 411 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 412 | { |
fd01b88c | 413 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 414 | struct md_rdev *rdev; |
a2826aa9 | 415 | |
a7a07e69 N |
416 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
417 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 418 | rcu_read_lock(); |
dafb20fa | 419 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
420 | if (rdev->raid_disk >= 0 && |
421 | !test_bit(Faulty, &rdev->flags)) { | |
422 | /* Take two references, one is dropped | |
423 | * when request finishes, one after | |
424 | * we reclaim rcu_read_lock | |
425 | */ | |
426 | struct bio *bi; | |
427 | atomic_inc(&rdev->nr_pending); | |
428 | atomic_inc(&rdev->nr_pending); | |
429 | rcu_read_unlock(); | |
b5e1b8ce | 430 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 431 | bi->bi_end_io = md_end_flush; |
a2826aa9 | 432 | bi->bi_private = rdev; |
74d46992 | 433 | bio_set_dev(bi, rdev->bdev); |
70fd7614 | 434 | bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
a2826aa9 | 435 | atomic_inc(&mddev->flush_pending); |
4e49ea4a | 436 | submit_bio(bi); |
a2826aa9 N |
437 | rcu_read_lock(); |
438 | rdev_dec_pending(rdev, mddev); | |
439 | } | |
440 | rcu_read_unlock(); | |
a7a07e69 N |
441 | if (atomic_dec_and_test(&mddev->flush_pending)) |
442 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
443 | } |
444 | ||
e9c7469b | 445 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 446 | { |
fd01b88c | 447 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 448 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 449 | |
79bf31a3 SL |
450 | /* |
451 | * must reset flush_bio before calling into md_handle_request to avoid a | |
452 | * deadlock, because other bios passed md_handle_request suspend check | |
453 | * could wait for this and below md_handle_request could wait for those | |
454 | * bios because of suspend check | |
455 | */ | |
456 | mddev->flush_bio = NULL; | |
457 | wake_up(&mddev->sb_wait); | |
458 | ||
4f024f37 | 459 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 | 460 | /* an empty barrier - all done */ |
4246a0b6 | 461 | bio_endio(bio); |
a2826aa9 | 462 | else { |
1eff9d32 | 463 | bio->bi_opf &= ~REQ_PREFLUSH; |
79bf31a3 | 464 | md_handle_request(mddev, bio); |
a2826aa9 | 465 | } |
a2826aa9 N |
466 | } |
467 | ||
fd01b88c | 468 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 | 469 | { |
85572d7c | 470 | spin_lock_irq(&mddev->lock); |
a2826aa9 | 471 | wait_event_lock_irq(mddev->sb_wait, |
e9c7469b | 472 | !mddev->flush_bio, |
85572d7c | 473 | mddev->lock); |
e9c7469b | 474 | mddev->flush_bio = bio; |
85572d7c | 475 | spin_unlock_irq(&mddev->lock); |
a2826aa9 | 476 | |
a035fc3e N |
477 | INIT_WORK(&mddev->flush_work, submit_flushes); |
478 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 479 | } |
e9c7469b | 480 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 481 | |
fd01b88c | 482 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
483 | { |
484 | atomic_inc(&mddev->active); | |
485 | return mddev; | |
486 | } | |
487 | ||
5fd3a17e | 488 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 489 | |
fd01b88c | 490 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 491 | { |
7184ef8b | 492 | struct bio_set *bs = NULL, *sync_bs = NULL; |
a167f663 | 493 | |
1da177e4 LT |
494 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
495 | return; | |
d3374825 | 496 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
497 | mddev->ctime == 0 && !mddev->hold_active) { |
498 | /* Array is not configured at all, and not held active, | |
499 | * so destroy it */ | |
af8a2434 | 500 | list_del_init(&mddev->all_mddevs); |
a167f663 | 501 | bs = mddev->bio_set; |
7184ef8b | 502 | sync_bs = mddev->sync_set; |
a167f663 | 503 | mddev->bio_set = NULL; |
5a85071c | 504 | mddev->sync_set = NULL; |
d3374825 | 505 | if (mddev->gendisk) { |
e804ac78 TH |
506 | /* We did a probe so need to clean up. Call |
507 | * queue_work inside the spinlock so that | |
508 | * flush_workqueue() after mddev_find will | |
509 | * succeed in waiting for the work to be done. | |
d3374825 N |
510 | */ |
511 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 512 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
513 | } else |
514 | kfree(mddev); | |
515 | } | |
516 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
517 | if (bs) |
518 | bioset_free(bs); | |
7184ef8b SL |
519 | if (sync_bs) |
520 | bioset_free(sync_bs); | |
1da177e4 LT |
521 | } |
522 | ||
25b2edfa SL |
523 | static void md_safemode_timeout(unsigned long data); |
524 | ||
fd01b88c | 525 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
526 | { |
527 | mutex_init(&mddev->open_mutex); | |
528 | mutex_init(&mddev->reconfig_mutex); | |
529 | mutex_init(&mddev->bitmap_info.mutex); | |
530 | INIT_LIST_HEAD(&mddev->disks); | |
531 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
25b2edfa SL |
532 | setup_timer(&mddev->safemode_timer, md_safemode_timeout, |
533 | (unsigned long) mddev); | |
fafd7fb0 N |
534 | atomic_set(&mddev->active, 1); |
535 | atomic_set(&mddev->openers, 0); | |
536 | atomic_set(&mddev->active_io, 0); | |
85572d7c | 537 | spin_lock_init(&mddev->lock); |
fafd7fb0 N |
538 | atomic_set(&mddev->flush_pending, 0); |
539 | init_waitqueue_head(&mddev->sb_wait); | |
540 | init_waitqueue_head(&mddev->recovery_wait); | |
541 | mddev->reshape_position = MaxSector; | |
2c810cdd | 542 | mddev->reshape_backwards = 0; |
c4a39551 | 543 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
544 | mddev->resync_min = 0; |
545 | mddev->resync_max = MaxSector; | |
546 | mddev->level = LEVEL_NONE; | |
547 | } | |
390ee602 | 548 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 549 | |
f72ffdd6 | 550 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 551 | { |
fd01b88c | 552 | struct mddev *mddev, *new = NULL; |
1da177e4 | 553 | |
8f5f02c4 N |
554 | if (unit && MAJOR(unit) != MD_MAJOR) |
555 | unit &= ~((1<<MdpMinorShift)-1); | |
556 | ||
1da177e4 LT |
557 | retry: |
558 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
559 | |
560 | if (unit) { | |
561 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
562 | if (mddev->unit == unit) { | |
563 | mddev_get(mddev); | |
564 | spin_unlock(&all_mddevs_lock); | |
565 | kfree(new); | |
566 | return mddev; | |
567 | } | |
568 | ||
569 | if (new) { | |
570 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 571 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
572 | new->hold_active = UNTIL_IOCTL; |
573 | return new; | |
1da177e4 | 574 | } |
efeb53c0 N |
575 | } else if (new) { |
576 | /* find an unused unit number */ | |
577 | static int next_minor = 512; | |
578 | int start = next_minor; | |
579 | int is_free = 0; | |
580 | int dev = 0; | |
581 | while (!is_free) { | |
582 | dev = MKDEV(MD_MAJOR, next_minor); | |
583 | next_minor++; | |
584 | if (next_minor > MINORMASK) | |
585 | next_minor = 0; | |
586 | if (next_minor == start) { | |
587 | /* Oh dear, all in use. */ | |
588 | spin_unlock(&all_mddevs_lock); | |
589 | kfree(new); | |
590 | return NULL; | |
591 | } | |
f72ffdd6 | 592 | |
efeb53c0 N |
593 | is_free = 1; |
594 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
595 | if (mddev->unit == dev) { | |
596 | is_free = 0; | |
597 | break; | |
598 | } | |
599 | } | |
600 | new->unit = dev; | |
601 | new->md_minor = MINOR(dev); | |
602 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
603 | list_add(&new->all_mddevs, &all_mddevs); |
604 | spin_unlock(&all_mddevs_lock); | |
605 | return new; | |
606 | } | |
607 | spin_unlock(&all_mddevs_lock); | |
608 | ||
9ffae0cf | 609 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
610 | if (!new) |
611 | return NULL; | |
612 | ||
1da177e4 LT |
613 | new->unit = unit; |
614 | if (MAJOR(unit) == MD_MAJOR) | |
615 | new->md_minor = MINOR(unit); | |
616 | else | |
617 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
618 | ||
fafd7fb0 | 619 | mddev_init(new); |
1da177e4 | 620 | |
1da177e4 LT |
621 | goto retry; |
622 | } | |
623 | ||
b6eb127d N |
624 | static struct attribute_group md_redundancy_group; |
625 | ||
5c47daf6 | 626 | void mddev_unlock(struct mddev *mddev) |
1da177e4 | 627 | { |
a64c876f | 628 | if (mddev->to_remove) { |
b6eb127d N |
629 | /* These cannot be removed under reconfig_mutex as |
630 | * an access to the files will try to take reconfig_mutex | |
631 | * while holding the file unremovable, which leads to | |
632 | * a deadlock. | |
bb4f1e9d N |
633 | * So hold set sysfs_active while the remove in happeing, |
634 | * and anything else which might set ->to_remove or my | |
635 | * otherwise change the sysfs namespace will fail with | |
636 | * -EBUSY if sysfs_active is still set. | |
637 | * We set sysfs_active under reconfig_mutex and elsewhere | |
638 | * test it under the same mutex to ensure its correct value | |
639 | * is seen. | |
b6eb127d | 640 | */ |
a64c876f N |
641 | struct attribute_group *to_remove = mddev->to_remove; |
642 | mddev->to_remove = NULL; | |
bb4f1e9d | 643 | mddev->sysfs_active = 1; |
b6eb127d N |
644 | mutex_unlock(&mddev->reconfig_mutex); |
645 | ||
00bcb4ac N |
646 | if (mddev->kobj.sd) { |
647 | if (to_remove != &md_redundancy_group) | |
648 | sysfs_remove_group(&mddev->kobj, to_remove); | |
649 | if (mddev->pers == NULL || | |
650 | mddev->pers->sync_request == NULL) { | |
651 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
652 | if (mddev->sysfs_action) | |
653 | sysfs_put(mddev->sysfs_action); | |
654 | mddev->sysfs_action = NULL; | |
655 | } | |
a64c876f | 656 | } |
bb4f1e9d | 657 | mddev->sysfs_active = 0; |
b6eb127d N |
658 | } else |
659 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 660 | |
751e67ca CD |
661 | /* As we've dropped the mutex we need a spinlock to |
662 | * make sure the thread doesn't disappear | |
01f96c0a N |
663 | */ |
664 | spin_lock(&pers_lock); | |
005eca5e | 665 | md_wakeup_thread(mddev->thread); |
01f96c0a | 666 | spin_unlock(&pers_lock); |
1da177e4 | 667 | } |
5c47daf6 | 668 | EXPORT_SYMBOL_GPL(mddev_unlock); |
1da177e4 | 669 | |
57d051dc | 670 | struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr) |
1ca69c4b N |
671 | { |
672 | struct md_rdev *rdev; | |
673 | ||
674 | rdev_for_each_rcu(rdev, mddev) | |
675 | if (rdev->desc_nr == nr) | |
676 | return rdev; | |
677 | ||
678 | return NULL; | |
679 | } | |
57d051dc | 680 | EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu); |
1ca69c4b N |
681 | |
682 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 683 | { |
3cb03002 | 684 | struct md_rdev *rdev; |
1da177e4 | 685 | |
dafb20fa | 686 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
687 | if (rdev->bdev->bd_dev == dev) |
688 | return rdev; | |
159ec1fc | 689 | |
1da177e4 LT |
690 | return NULL; |
691 | } | |
692 | ||
1ca69c4b N |
693 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
694 | { | |
695 | struct md_rdev *rdev; | |
696 | ||
697 | rdev_for_each_rcu(rdev, mddev) | |
698 | if (rdev->bdev->bd_dev == dev) | |
699 | return rdev; | |
700 | ||
701 | return NULL; | |
702 | } | |
703 | ||
84fc4b56 | 704 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 705 | { |
84fc4b56 | 706 | struct md_personality *pers; |
d9d166c2 N |
707 | list_for_each_entry(pers, &pers_list, list) { |
708 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 709 | return pers; |
d9d166c2 N |
710 | if (strcmp(pers->name, clevel)==0) |
711 | return pers; | |
712 | } | |
2604b703 N |
713 | return NULL; |
714 | } | |
715 | ||
b73df2d3 | 716 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 717 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 718 | { |
57b2caa3 | 719 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 720 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
721 | } |
722 | ||
f72ffdd6 | 723 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 724 | { |
1da177e4 | 725 | rdev->sb_page = alloc_page(GFP_KERNEL); |
7f0f0d87 | 726 | if (!rdev->sb_page) |
ebc24337 | 727 | return -ENOMEM; |
1da177e4 LT |
728 | return 0; |
729 | } | |
730 | ||
545c8795 | 731 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
732 | { |
733 | if (rdev->sb_page) { | |
2d1f3b5d | 734 | put_page(rdev->sb_page); |
1da177e4 LT |
735 | rdev->sb_loaded = 0; |
736 | rdev->sb_page = NULL; | |
0f420358 | 737 | rdev->sb_start = 0; |
dd8ac336 | 738 | rdev->sectors = 0; |
1da177e4 | 739 | } |
2699b672 N |
740 | if (rdev->bb_page) { |
741 | put_page(rdev->bb_page); | |
742 | rdev->bb_page = NULL; | |
743 | } | |
d3b407fb | 744 | badblocks_exit(&rdev->badblocks); |
1da177e4 | 745 | } |
545c8795 | 746 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 747 | |
4246a0b6 | 748 | static void super_written(struct bio *bio) |
7bfa19f2 | 749 | { |
3cb03002 | 750 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 751 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 752 | |
4e4cbee9 CH |
753 | if (bio->bi_status) { |
754 | pr_err("md: super_written gets error=%d\n", bio->bi_status); | |
a9701a30 | 755 | md_error(mddev, rdev); |
46533ff7 N |
756 | if (!test_bit(Faulty, &rdev->flags) |
757 | && (bio->bi_opf & MD_FAILFAST)) { | |
2953079c | 758 | set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags); |
46533ff7 N |
759 | set_bit(LastDev, &rdev->flags); |
760 | } | |
761 | } else | |
762 | clear_bit(LastDev, &rdev->flags); | |
7bfa19f2 | 763 | |
a9701a30 N |
764 | if (atomic_dec_and_test(&mddev->pending_writes)) |
765 | wake_up(&mddev->sb_wait); | |
ed3b98c7 | 766 | rdev_dec_pending(rdev, mddev); |
f8b58edf | 767 | bio_put(bio); |
7bfa19f2 N |
768 | } |
769 | ||
fd01b88c | 770 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
771 | sector_t sector, int size, struct page *page) |
772 | { | |
773 | /* write first size bytes of page to sector of rdev | |
774 | * Increment mddev->pending_writes before returning | |
775 | * and decrement it on completion, waking up sb_wait | |
776 | * if zero is reached. | |
777 | * If an error occurred, call md_error | |
778 | */ | |
46533ff7 N |
779 | struct bio *bio; |
780 | int ff = 0; | |
781 | ||
782 | if (test_bit(Faulty, &rdev->flags)) | |
783 | return; | |
784 | ||
5a85071c | 785 | bio = md_bio_alloc_sync(mddev); |
7bfa19f2 | 786 | |
ed3b98c7 SL |
787 | atomic_inc(&rdev->nr_pending); |
788 | ||
74d46992 | 789 | bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev); |
4f024f37 | 790 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
791 | bio_add_page(bio, page, size, 0); |
792 | bio->bi_private = rdev; | |
793 | bio->bi_end_io = super_written; | |
46533ff7 N |
794 | |
795 | if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) && | |
796 | test_bit(FailFast, &rdev->flags) && | |
797 | !test_bit(LastDev, &rdev->flags)) | |
798 | ff = MD_FAILFAST; | |
5a8948f8 | 799 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff; |
a9701a30 | 800 | |
7bfa19f2 | 801 | atomic_inc(&mddev->pending_writes); |
4e49ea4a | 802 | submit_bio(bio); |
a9701a30 N |
803 | } |
804 | ||
46533ff7 | 805 | int md_super_wait(struct mddev *mddev) |
a9701a30 | 806 | { |
e9c7469b | 807 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 808 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
2953079c | 809 | if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags)) |
46533ff7 N |
810 | return -EAGAIN; |
811 | return 0; | |
7bfa19f2 N |
812 | } |
813 | ||
3cb03002 | 814 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
796a5cf0 | 815 | struct page *page, int op, int op_flags, bool metadata_op) |
1da177e4 | 816 | { |
5a85071c | 817 | struct bio *bio = md_bio_alloc_sync(rdev->mddev); |
1da177e4 LT |
818 | int ret; |
819 | ||
74d46992 CH |
820 | if (metadata_op && rdev->meta_bdev) |
821 | bio_set_dev(bio, rdev->meta_bdev); | |
822 | else | |
823 | bio_set_dev(bio, rdev->bdev); | |
796a5cf0 | 824 | bio_set_op_attrs(bio, op, op_flags); |
ccebd4c4 | 825 | if (metadata_op) |
4f024f37 | 826 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
827 | else if (rdev->mddev->reshape_position != MaxSector && |
828 | (rdev->mddev->reshape_backwards == | |
829 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 830 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 831 | else |
4f024f37 | 832 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 833 | bio_add_page(bio, page, size, 0); |
4e49ea4a MC |
834 | |
835 | submit_bio_wait(bio); | |
1da177e4 | 836 | |
4e4cbee9 | 837 | ret = !bio->bi_status; |
1da177e4 LT |
838 | bio_put(bio); |
839 | return ret; | |
840 | } | |
a8745db2 | 841 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 842 | |
f72ffdd6 | 843 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
844 | { |
845 | char b[BDEVNAME_SIZE]; | |
403df478 | 846 | |
1da177e4 LT |
847 | if (rdev->sb_loaded) |
848 | return 0; | |
849 | ||
796a5cf0 | 850 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) |
1da177e4 LT |
851 | goto fail; |
852 | rdev->sb_loaded = 1; | |
853 | return 0; | |
854 | ||
855 | fail: | |
9d48739e N |
856 | pr_err("md: disabled device %s, could not read superblock.\n", |
857 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
858 | return -EINVAL; |
859 | } | |
860 | ||
e6fd2093 | 861 | static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
1da177e4 | 862 | { |
f72ffdd6 | 863 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
864 | sb1->set_uuid1 == sb2->set_uuid1 && |
865 | sb1->set_uuid2 == sb2->set_uuid2 && | |
866 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
867 | } |
868 | ||
e6fd2093 | 869 | static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
1da177e4 LT |
870 | { |
871 | int ret; | |
872 | mdp_super_t *tmp1, *tmp2; | |
873 | ||
874 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
875 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
876 | ||
877 | if (!tmp1 || !tmp2) { | |
878 | ret = 0; | |
1da177e4 LT |
879 | goto abort; |
880 | } | |
881 | ||
882 | *tmp1 = *sb1; | |
883 | *tmp2 = *sb2; | |
884 | ||
885 | /* | |
886 | * nr_disks is not constant | |
887 | */ | |
888 | tmp1->nr_disks = 0; | |
889 | tmp2->nr_disks = 0; | |
890 | ||
ce0c8e05 | 891 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 892 | abort: |
990a8baf JJ |
893 | kfree(tmp1); |
894 | kfree(tmp2); | |
1da177e4 LT |
895 | return ret; |
896 | } | |
897 | ||
4d167f09 N |
898 | static u32 md_csum_fold(u32 csum) |
899 | { | |
900 | csum = (csum & 0xffff) + (csum >> 16); | |
901 | return (csum & 0xffff) + (csum >> 16); | |
902 | } | |
903 | ||
f72ffdd6 | 904 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 905 | { |
4d167f09 N |
906 | u64 newcsum = 0; |
907 | u32 *sb32 = (u32*)sb; | |
908 | int i; | |
1da177e4 LT |
909 | unsigned int disk_csum, csum; |
910 | ||
911 | disk_csum = sb->sb_csum; | |
912 | sb->sb_csum = 0; | |
4d167f09 N |
913 | |
914 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
915 | newcsum += sb32[i]; | |
916 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
917 | ||
4d167f09 N |
918 | #ifdef CONFIG_ALPHA |
919 | /* This used to use csum_partial, which was wrong for several | |
920 | * reasons including that different results are returned on | |
921 | * different architectures. It isn't critical that we get exactly | |
922 | * the same return value as before (we always csum_fold before | |
923 | * testing, and that removes any differences). However as we | |
924 | * know that csum_partial always returned a 16bit value on | |
925 | * alphas, do a fold to maximise conformity to previous behaviour. | |
926 | */ | |
927 | sb->sb_csum = md_csum_fold(disk_csum); | |
928 | #else | |
1da177e4 | 929 | sb->sb_csum = disk_csum; |
4d167f09 | 930 | #endif |
1da177e4 LT |
931 | return csum; |
932 | } | |
933 | ||
1da177e4 LT |
934 | /* |
935 | * Handle superblock details. | |
936 | * We want to be able to handle multiple superblock formats | |
937 | * so we have a common interface to them all, and an array of | |
938 | * different handlers. | |
939 | * We rely on user-space to write the initial superblock, and support | |
940 | * reading and updating of superblocks. | |
941 | * Interface methods are: | |
3cb03002 | 942 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
943 | * loads and validates a superblock on dev. |
944 | * if refdev != NULL, compare superblocks on both devices | |
945 | * Return: | |
946 | * 0 - dev has a superblock that is compatible with refdev | |
947 | * 1 - dev has a superblock that is compatible and newer than refdev | |
948 | * so dev should be used as the refdev in future | |
949 | * -EINVAL superblock incompatible or invalid | |
950 | * -othererror e.g. -EIO | |
951 | * | |
fd01b88c | 952 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
953 | * Verify that dev is acceptable into mddev. |
954 | * The first time, mddev->raid_disks will be 0, and data from | |
955 | * dev should be merged in. Subsequent calls check that dev | |
956 | * is new enough. Return 0 or -EINVAL | |
957 | * | |
fd01b88c | 958 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
959 | * Update the superblock for rdev with data in mddev |
960 | * This does not write to disc. | |
961 | * | |
962 | */ | |
963 | ||
964 | struct super_type { | |
0cd17fec CW |
965 | char *name; |
966 | struct module *owner; | |
c6563a8c N |
967 | int (*load_super)(struct md_rdev *rdev, |
968 | struct md_rdev *refdev, | |
0cd17fec | 969 | int minor_version); |
c6563a8c N |
970 | int (*validate_super)(struct mddev *mddev, |
971 | struct md_rdev *rdev); | |
972 | void (*sync_super)(struct mddev *mddev, | |
973 | struct md_rdev *rdev); | |
3cb03002 | 974 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 975 | sector_t num_sectors); |
c6563a8c N |
976 | int (*allow_new_offset)(struct md_rdev *rdev, |
977 | unsigned long long new_offset); | |
1da177e4 LT |
978 | }; |
979 | ||
0894cc30 AN |
980 | /* |
981 | * Check that the given mddev has no bitmap. | |
982 | * | |
983 | * This function is called from the run method of all personalities that do not | |
984 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
985 | * has a bitmap. Otherwise, it returns 0. | |
986 | * | |
987 | */ | |
fd01b88c | 988 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 989 | { |
c3d9714e | 990 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 | 991 | return 0; |
9d48739e | 992 | pr_warn("%s: bitmaps are not supported for %s\n", |
0894cc30 AN |
993 | mdname(mddev), mddev->pers->name); |
994 | return 1; | |
995 | } | |
996 | EXPORT_SYMBOL(md_check_no_bitmap); | |
997 | ||
1da177e4 | 998 | /* |
f72ffdd6 | 999 | * load_super for 0.90.0 |
1da177e4 | 1000 | */ |
3cb03002 | 1001 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1002 | { |
1003 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
1004 | mdp_super_t *sb; | |
1005 | int ret; | |
1da177e4 LT |
1006 | |
1007 | /* | |
0f420358 | 1008 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
1009 | * it's at the end of the disk. |
1010 | * | |
1011 | * It also happens to be a multiple of 4Kb. | |
1012 | */ | |
57b2caa3 | 1013 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 1014 | |
0002b271 | 1015 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
9d48739e N |
1016 | if (ret) |
1017 | return ret; | |
1da177e4 LT |
1018 | |
1019 | ret = -EINVAL; | |
1020 | ||
1021 | bdevname(rdev->bdev, b); | |
65a06f06 | 1022 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1023 | |
1024 | if (sb->md_magic != MD_SB_MAGIC) { | |
9d48739e | 1025 | pr_warn("md: invalid raid superblock magic on %s\n", b); |
1da177e4 LT |
1026 | goto abort; |
1027 | } | |
1028 | ||
1029 | if (sb->major_version != 0 || | |
f6705578 N |
1030 | sb->minor_version < 90 || |
1031 | sb->minor_version > 91) { | |
9d48739e N |
1032 | pr_warn("Bad version number %d.%d on %s\n", |
1033 | sb->major_version, sb->minor_version, b); | |
1da177e4 LT |
1034 | goto abort; |
1035 | } | |
1036 | ||
1037 | if (sb->raid_disks <= 0) | |
1038 | goto abort; | |
1039 | ||
4d167f09 | 1040 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
9d48739e | 1041 | pr_warn("md: invalid superblock checksum on %s\n", b); |
1da177e4 LT |
1042 | goto abort; |
1043 | } | |
1044 | ||
1045 | rdev->preferred_minor = sb->md_minor; | |
1046 | rdev->data_offset = 0; | |
c6563a8c | 1047 | rdev->new_data_offset = 0; |
0002b271 | 1048 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1049 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1050 | |
1051 | if (sb->level == LEVEL_MULTIPATH) | |
1052 | rdev->desc_nr = -1; | |
1053 | else | |
1054 | rdev->desc_nr = sb->this_disk.number; | |
1055 | ||
9a7b2b0f | 1056 | if (!refdev) { |
1da177e4 | 1057 | ret = 1; |
9a7b2b0f | 1058 | } else { |
1da177e4 | 1059 | __u64 ev1, ev2; |
65a06f06 | 1060 | mdp_super_t *refsb = page_address(refdev->sb_page); |
e6fd2093 | 1061 | if (!md_uuid_equal(refsb, sb)) { |
9d48739e | 1062 | pr_warn("md: %s has different UUID to %s\n", |
1da177e4 LT |
1063 | b, bdevname(refdev->bdev,b2)); |
1064 | goto abort; | |
1065 | } | |
e6fd2093 | 1066 | if (!md_sb_equal(refsb, sb)) { |
9d48739e N |
1067 | pr_warn("md: %s has same UUID but different superblock to %s\n", |
1068 | b, bdevname(refdev->bdev, b2)); | |
1da177e4 LT |
1069 | goto abort; |
1070 | } | |
1071 | ev1 = md_event(sb); | |
1072 | ev2 = md_event(refsb); | |
1073 | if (ev1 > ev2) | |
1074 | ret = 1; | |
f72ffdd6 | 1075 | else |
1da177e4 LT |
1076 | ret = 0; |
1077 | } | |
8190e754 | 1078 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1079 | /* Limit to 4TB as metadata cannot record more than that. |
1080 | * (not needed for Linear and RAID0 as metadata doesn't | |
1081 | * record this size) | |
1082 | */ | |
3312c951 AB |
1083 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) && |
1084 | sb->level >= 1) | |
1085 | rdev->sectors = (sector_t)(2ULL << 32) - 2; | |
1da177e4 | 1086 | |
27a7b260 | 1087 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1088 | /* "this cannot possibly happen" ... */ |
1089 | ret = -EINVAL; | |
1090 | ||
1da177e4 LT |
1091 | abort: |
1092 | return ret; | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * validate_super for 0.90.0 | |
1097 | */ | |
fd01b88c | 1098 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1099 | { |
1100 | mdp_disk_t *desc; | |
65a06f06 | 1101 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1102 | __u64 ev1 = md_event(sb); |
1da177e4 | 1103 | |
41158c7e | 1104 | rdev->raid_disk = -1; |
c5d79adb N |
1105 | clear_bit(Faulty, &rdev->flags); |
1106 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1107 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1108 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1109 | |
1da177e4 LT |
1110 | if (mddev->raid_disks == 0) { |
1111 | mddev->major_version = 0; | |
1112 | mddev->minor_version = sb->minor_version; | |
1113 | mddev->patch_version = sb->patch_version; | |
e691063a | 1114 | mddev->external = 0; |
9d8f0363 | 1115 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1116 | mddev->ctime = sb->ctime; |
1117 | mddev->utime = sb->utime; | |
1118 | mddev->level = sb->level; | |
d9d166c2 | 1119 | mddev->clevel[0] = 0; |
1da177e4 LT |
1120 | mddev->layout = sb->layout; |
1121 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1122 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1123 | mddev->events = ev1; |
c3d9714e | 1124 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1125 | mddev->bitmap_info.space = 0; |
1126 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1127 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1128 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1129 | mddev->reshape_backwards = 0; |
1da177e4 | 1130 | |
f6705578 N |
1131 | if (mddev->minor_version >= 91) { |
1132 | mddev->reshape_position = sb->reshape_position; | |
1133 | mddev->delta_disks = sb->delta_disks; | |
1134 | mddev->new_level = sb->new_level; | |
1135 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1136 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1137 | if (mddev->delta_disks < 0) |
1138 | mddev->reshape_backwards = 1; | |
f6705578 N |
1139 | } else { |
1140 | mddev->reshape_position = MaxSector; | |
1141 | mddev->delta_disks = 0; | |
1142 | mddev->new_level = mddev->level; | |
1143 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1144 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1145 | } |
1146 | ||
1da177e4 LT |
1147 | if (sb->state & (1<<MD_SB_CLEAN)) |
1148 | mddev->recovery_cp = MaxSector; | |
1149 | else { | |
f72ffdd6 | 1150 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1151 | sb->events_lo == sb->cp_events_lo) { |
1152 | mddev->recovery_cp = sb->recovery_cp; | |
1153 | } else | |
1154 | mddev->recovery_cp = 0; | |
1155 | } | |
1156 | ||
1157 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1158 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1159 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1160 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1161 | ||
1162 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1163 | |
1164 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1165 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1166 | mddev->bitmap_info.offset = |
1167 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1168 | mddev->bitmap_info.space = |
c9ad020f | 1169 | mddev->bitmap_info.default_space; |
6409bb05 | 1170 | } |
a654b9d8 | 1171 | |
41158c7e | 1172 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1173 | /* Insist on good event counter while assembling, except |
1174 | * for spares (which don't need an event count) */ | |
1da177e4 | 1175 | ++ev1; |
be6800a7 N |
1176 | if (sb->disks[rdev->desc_nr].state & ( |
1177 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1178 | if (ev1 < mddev->events) |
be6800a7 | 1179 | return -EINVAL; |
41158c7e N |
1180 | } else if (mddev->bitmap) { |
1181 | /* if adding to array with a bitmap, then we can accept an | |
1182 | * older device ... but not too old. | |
1183 | */ | |
41158c7e N |
1184 | if (ev1 < mddev->bitmap->events_cleared) |
1185 | return 0; | |
8313b8e5 N |
1186 | if (ev1 < mddev->events) |
1187 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1188 | } else { |
1189 | if (ev1 < mddev->events) | |
1190 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1191 | return 0; | |
1192 | } | |
41158c7e | 1193 | |
1da177e4 | 1194 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1195 | desc = sb->disks + rdev->desc_nr; |
1196 | ||
1197 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1198 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1199 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1200 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1201 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1202 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1203 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1204 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1205 | /* active but not in sync implies recovery up to | |
1206 | * reshape position. We don't know exactly where | |
1207 | * that is, so set to zero for now */ | |
1208 | if (mddev->minor_version >= 91) { | |
1209 | rdev->recovery_offset = 0; | |
1210 | rdev->raid_disk = desc->raid_disk; | |
1211 | } | |
1da177e4 | 1212 | } |
8ddf9efe N |
1213 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1214 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1215 | if (desc->state & (1<<MD_DISK_FAILFAST)) |
1216 | set_bit(FailFast, &rdev->flags); | |
41158c7e | 1217 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1218 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1219 | return 0; |
1220 | } | |
1221 | ||
1222 | /* | |
1223 | * sync_super for 0.90.0 | |
1224 | */ | |
fd01b88c | 1225 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1226 | { |
1227 | mdp_super_t *sb; | |
3cb03002 | 1228 | struct md_rdev *rdev2; |
1da177e4 | 1229 | int next_spare = mddev->raid_disks; |
19133a42 | 1230 | |
1da177e4 LT |
1231 | /* make rdev->sb match mddev data.. |
1232 | * | |
1233 | * 1/ zero out disks | |
1234 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1235 | * 3/ any empty disks < next_spare become removed | |
1236 | * | |
1237 | * disks[0] gets initialised to REMOVED because | |
1238 | * we cannot be sure from other fields if it has | |
1239 | * been initialised or not. | |
1240 | */ | |
1241 | int i; | |
1242 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1243 | ||
61181565 N |
1244 | rdev->sb_size = MD_SB_BYTES; |
1245 | ||
65a06f06 | 1246 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1247 | |
1248 | memset(sb, 0, sizeof(*sb)); | |
1249 | ||
1250 | sb->md_magic = MD_SB_MAGIC; | |
1251 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1252 | sb->patch_version = mddev->patch_version; |
1253 | sb->gvalid_words = 0; /* ignored */ | |
1254 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1255 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1256 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1257 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1258 | ||
9ebc6ef1 | 1259 | sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 1260 | sb->level = mddev->level; |
58c0fed4 | 1261 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1262 | sb->raid_disks = mddev->raid_disks; |
1263 | sb->md_minor = mddev->md_minor; | |
e691063a | 1264 | sb->not_persistent = 0; |
9ebc6ef1 | 1265 | sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
1266 | sb->state = 0; |
1267 | sb->events_hi = (mddev->events>>32); | |
1268 | sb->events_lo = (u32)mddev->events; | |
1269 | ||
f6705578 N |
1270 | if (mddev->reshape_position == MaxSector) |
1271 | sb->minor_version = 90; | |
1272 | else { | |
1273 | sb->minor_version = 91; | |
1274 | sb->reshape_position = mddev->reshape_position; | |
1275 | sb->new_level = mddev->new_level; | |
1276 | sb->delta_disks = mddev->delta_disks; | |
1277 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1278 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1279 | } |
1280 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1281 | if (mddev->in_sync) |
1282 | { | |
1283 | sb->recovery_cp = mddev->recovery_cp; | |
1284 | sb->cp_events_hi = (mddev->events>>32); | |
1285 | sb->cp_events_lo = (u32)mddev->events; | |
1286 | if (mddev->recovery_cp == MaxSector) | |
1287 | sb->state = (1<< MD_SB_CLEAN); | |
1288 | } else | |
1289 | sb->recovery_cp = 0; | |
1290 | ||
1291 | sb->layout = mddev->layout; | |
9d8f0363 | 1292 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1293 | |
c3d9714e | 1294 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1295 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1296 | ||
1da177e4 | 1297 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1298 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1299 | mdp_disk_t *d; |
86e6ffdd | 1300 | int desc_nr; |
0261cd9f N |
1301 | int is_active = test_bit(In_sync, &rdev2->flags); |
1302 | ||
1303 | if (rdev2->raid_disk >= 0 && | |
1304 | sb->minor_version >= 91) | |
1305 | /* we have nowhere to store the recovery_offset, | |
1306 | * but if it is not below the reshape_position, | |
1307 | * we can piggy-back on that. | |
1308 | */ | |
1309 | is_active = 1; | |
1310 | if (rdev2->raid_disk < 0 || | |
1311 | test_bit(Faulty, &rdev2->flags)) | |
1312 | is_active = 0; | |
1313 | if (is_active) | |
86e6ffdd | 1314 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1315 | else |
86e6ffdd | 1316 | desc_nr = next_spare++; |
19133a42 | 1317 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1318 | d = &sb->disks[rdev2->desc_nr]; |
1319 | nr_disks++; | |
1320 | d->number = rdev2->desc_nr; | |
1321 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1322 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1323 | if (is_active) |
1da177e4 LT |
1324 | d->raid_disk = rdev2->raid_disk; |
1325 | else | |
1326 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1327 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1328 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1329 | else if (is_active) { |
1da177e4 | 1330 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1331 | if (test_bit(In_sync, &rdev2->flags)) |
1332 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1333 | active++; |
1334 | working++; | |
1335 | } else { | |
1336 | d->state = 0; | |
1337 | spare++; | |
1338 | working++; | |
1339 | } | |
8ddf9efe N |
1340 | if (test_bit(WriteMostly, &rdev2->flags)) |
1341 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
1342 | if (test_bit(FailFast, &rdev2->flags)) |
1343 | d->state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 | 1344 | } |
1da177e4 LT |
1345 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1346 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1347 | mdp_disk_t *d = &sb->disks[i]; | |
1348 | if (d->state == 0 && d->number == 0) { | |
1349 | d->number = i; | |
1350 | d->raid_disk = i; | |
1351 | d->state = (1<<MD_DISK_REMOVED); | |
1352 | d->state |= (1<<MD_DISK_FAULTY); | |
1353 | failed++; | |
1354 | } | |
1355 | } | |
1356 | sb->nr_disks = nr_disks; | |
1357 | sb->active_disks = active; | |
1358 | sb->working_disks = working; | |
1359 | sb->failed_disks = failed; | |
1360 | sb->spare_disks = spare; | |
1361 | ||
1362 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1363 | sb->sb_csum = calc_sb_csum(sb); | |
1364 | } | |
1365 | ||
0cd17fec CW |
1366 | /* |
1367 | * rdev_size_change for 0.90.0 | |
1368 | */ | |
1369 | static unsigned long long | |
3cb03002 | 1370 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1371 | { |
58c0fed4 | 1372 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1373 | return 0; /* component must fit device */ |
c3d9714e | 1374 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1375 | return 0; /* can't move bitmap */ |
57b2caa3 | 1376 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1377 | if (!num_sectors || num_sectors > rdev->sb_start) |
1378 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1379 | /* Limit to 4TB as metadata cannot record more than that. |
1380 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1381 | */ | |
3312c951 AB |
1382 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) && |
1383 | rdev->mddev->level >= 1) | |
1384 | num_sectors = (sector_t)(2ULL << 32) - 2; | |
46533ff7 N |
1385 | do { |
1386 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
0cd17fec | 1387 | rdev->sb_page); |
46533ff7 | 1388 | } while (md_super_wait(rdev->mddev) < 0); |
c26a44ed | 1389 | return num_sectors; |
0cd17fec CW |
1390 | } |
1391 | ||
c6563a8c N |
1392 | static int |
1393 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1394 | { | |
1395 | /* non-zero offset changes not possible with v0.90 */ | |
1396 | return new_offset == 0; | |
1397 | } | |
0cd17fec | 1398 | |
1da177e4 LT |
1399 | /* |
1400 | * version 1 superblock | |
1401 | */ | |
1402 | ||
f72ffdd6 | 1403 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1404 | { |
1c05b4bc N |
1405 | __le32 disk_csum; |
1406 | u32 csum; | |
1da177e4 LT |
1407 | unsigned long long newcsum; |
1408 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1409 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1410 | |
1411 | disk_csum = sb->sb_csum; | |
1412 | sb->sb_csum = 0; | |
1413 | newcsum = 0; | |
1f3c9907 | 1414 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1415 | newcsum += le32_to_cpu(*isuper++); |
1416 | ||
1417 | if (size == 2) | |
1c05b4bc | 1418 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1419 | |
1420 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1421 | sb->sb_csum = disk_csum; | |
1422 | return cpu_to_le32(csum); | |
1423 | } | |
1424 | ||
3cb03002 | 1425 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1426 | { |
1427 | struct mdp_superblock_1 *sb; | |
1428 | int ret; | |
0f420358 | 1429 | sector_t sb_start; |
c6563a8c | 1430 | sector_t sectors; |
1da177e4 | 1431 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1432 | int bmask; |
1da177e4 LT |
1433 | |
1434 | /* | |
0f420358 | 1435 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1436 | * It is always aligned to a 4K boundary and |
1437 | * depeding on minor_version, it can be: | |
1438 | * 0: At least 8K, but less than 12K, from end of device | |
1439 | * 1: At start of device | |
1440 | * 2: 4K from start of device. | |
1441 | */ | |
1442 | switch(minor_version) { | |
1443 | case 0: | |
77304d2a | 1444 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1445 | sb_start -= 8*2; |
1446 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1447 | break; |
1448 | case 1: | |
0f420358 | 1449 | sb_start = 0; |
1da177e4 LT |
1450 | break; |
1451 | case 2: | |
0f420358 | 1452 | sb_start = 8; |
1da177e4 LT |
1453 | break; |
1454 | default: | |
1455 | return -EINVAL; | |
1456 | } | |
0f420358 | 1457 | rdev->sb_start = sb_start; |
1da177e4 | 1458 | |
0002b271 N |
1459 | /* superblock is rarely larger than 1K, but it can be larger, |
1460 | * and it is safe to read 4k, so we do that | |
1461 | */ | |
1462 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1463 | if (ret) return ret; |
1464 | ||
65a06f06 | 1465 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1466 | |
1467 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1468 | sb->major_version != cpu_to_le32(1) || | |
1469 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1470 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1471 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1472 | return -EINVAL; |
1473 | ||
1474 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
9d48739e | 1475 | pr_warn("md: invalid superblock checksum on %s\n", |
1da177e4 LT |
1476 | bdevname(rdev->bdev,b)); |
1477 | return -EINVAL; | |
1478 | } | |
1479 | if (le64_to_cpu(sb->data_size) < 10) { | |
9d48739e N |
1480 | pr_warn("md: data_size too small on %s\n", |
1481 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1482 | return -EINVAL; |
1483 | } | |
c6563a8c N |
1484 | if (sb->pad0 || |
1485 | sb->pad3[0] || | |
1486 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1487 | /* Some padding is non-zero, might be a new feature */ | |
1488 | return -EINVAL; | |
e11e93fa | 1489 | |
1da177e4 LT |
1490 | rdev->preferred_minor = 0xffff; |
1491 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1492 | rdev->new_data_offset = rdev->data_offset; |
1493 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1494 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1495 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1496 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1497 | |
0002b271 | 1498 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1499 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1500 | if (rdev->sb_size & bmask) |
a1801f85 N |
1501 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1502 | ||
1503 | if (minor_version | |
0f420358 | 1504 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1505 | return -EINVAL; |
c6563a8c N |
1506 | if (minor_version |
1507 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1508 | return -EINVAL; | |
0002b271 | 1509 | |
31b65a0d N |
1510 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1511 | rdev->desc_nr = -1; | |
1512 | else | |
1513 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1514 | ||
2699b672 N |
1515 | if (!rdev->bb_page) { |
1516 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1517 | if (!rdev->bb_page) | |
1518 | return -ENOMEM; | |
1519 | } | |
1520 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1521 | rdev->badblocks.count == 0) { | |
1522 | /* need to load the bad block list. | |
1523 | * Currently we limit it to one page. | |
1524 | */ | |
1525 | s32 offset; | |
1526 | sector_t bb_sector; | |
1527 | u64 *bbp; | |
1528 | int i; | |
1529 | int sectors = le16_to_cpu(sb->bblog_size); | |
1530 | if (sectors > (PAGE_SIZE / 512)) | |
1531 | return -EINVAL; | |
1532 | offset = le32_to_cpu(sb->bblog_offset); | |
1533 | if (offset == 0) | |
1534 | return -EINVAL; | |
1535 | bb_sector = (long long)offset; | |
1536 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
796a5cf0 | 1537 | rdev->bb_page, REQ_OP_READ, 0, true)) |
2699b672 N |
1538 | return -EIO; |
1539 | bbp = (u64 *)page_address(rdev->bb_page); | |
1540 | rdev->badblocks.shift = sb->bblog_shift; | |
1541 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1542 | u64 bb = le64_to_cpu(*bbp); | |
1543 | int count = bb & (0x3ff); | |
1544 | u64 sector = bb >> 10; | |
1545 | sector <<= sb->bblog_shift; | |
1546 | count <<= sb->bblog_shift; | |
1547 | if (bb + 1 == 0) | |
1548 | break; | |
fc974ee2 | 1549 | if (badblocks_set(&rdev->badblocks, sector, count, 1)) |
2699b672 N |
1550 | return -EINVAL; |
1551 | } | |
486adf72 N |
1552 | } else if (sb->bblog_offset != 0) |
1553 | rdev->badblocks.shift = 0; | |
2699b672 | 1554 | |
ddc08823 PB |
1555 | if ((le32_to_cpu(sb->feature_map) & |
1556 | (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) { | |
ea0213e0 AP |
1557 | rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset); |
1558 | rdev->ppl.size = le16_to_cpu(sb->ppl.size); | |
1559 | rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset; | |
1560 | } | |
1561 | ||
9a7b2b0f | 1562 | if (!refdev) { |
8ed75463 | 1563 | ret = 1; |
9a7b2b0f | 1564 | } else { |
1da177e4 | 1565 | __u64 ev1, ev2; |
65a06f06 | 1566 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1567 | |
1568 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1569 | sb->level != refsb->level || | |
1570 | sb->layout != refsb->layout || | |
1571 | sb->chunksize != refsb->chunksize) { | |
9d48739e | 1572 | pr_warn("md: %s has strangely different superblock to %s\n", |
1da177e4 LT |
1573 | bdevname(rdev->bdev,b), |
1574 | bdevname(refdev->bdev,b2)); | |
1575 | return -EINVAL; | |
1576 | } | |
1577 | ev1 = le64_to_cpu(sb->events); | |
1578 | ev2 = le64_to_cpu(refsb->events); | |
1579 | ||
1580 | if (ev1 > ev2) | |
8ed75463 N |
1581 | ret = 1; |
1582 | else | |
1583 | ret = 0; | |
1da177e4 | 1584 | } |
c6563a8c N |
1585 | if (minor_version) { |
1586 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1587 | sectors -= rdev->data_offset; | |
1588 | } else | |
1589 | sectors = rdev->sb_start; | |
1590 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1591 | return -EINVAL; |
dd8ac336 | 1592 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1593 | return ret; |
1da177e4 LT |
1594 | } |
1595 | ||
fd01b88c | 1596 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1597 | { |
65a06f06 | 1598 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1599 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1600 | |
41158c7e | 1601 | rdev->raid_disk = -1; |
c5d79adb N |
1602 | clear_bit(Faulty, &rdev->flags); |
1603 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1604 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1605 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1606 | |
1da177e4 LT |
1607 | if (mddev->raid_disks == 0) { |
1608 | mddev->major_version = 1; | |
1609 | mddev->patch_version = 0; | |
e691063a | 1610 | mddev->external = 0; |
9d8f0363 | 1611 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
9ebc6ef1 DD |
1612 | mddev->ctime = le64_to_cpu(sb->ctime); |
1613 | mddev->utime = le64_to_cpu(sb->utime); | |
1da177e4 | 1614 | mddev->level = le32_to_cpu(sb->level); |
d9d166c2 | 1615 | mddev->clevel[0] = 0; |
1da177e4 LT |
1616 | mddev->layout = le32_to_cpu(sb->layout); |
1617 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1618 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1619 | mddev->events = ev1; |
c3d9714e | 1620 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1621 | mddev->bitmap_info.space = 0; |
1622 | /* Default location for bitmap is 1K after superblock | |
1623 | * using 3K - total of 4K | |
1624 | */ | |
c3d9714e | 1625 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1626 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1627 | mddev->reshape_backwards = 0; |
1628 | ||
1da177e4 LT |
1629 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1630 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1631 | ||
1632 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1633 | |
71c0805c | 1634 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1635 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1636 | mddev->bitmap_info.offset = |
1637 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1638 | /* Metadata doesn't record how much space is available. |
1639 | * For 1.0, we assume we can use up to the superblock | |
1640 | * if before, else to 4K beyond superblock. | |
1641 | * For others, assume no change is possible. | |
1642 | */ | |
1643 | if (mddev->minor_version > 0) | |
1644 | mddev->bitmap_info.space = 0; | |
1645 | else if (mddev->bitmap_info.offset > 0) | |
1646 | mddev->bitmap_info.space = | |
1647 | 8 - mddev->bitmap_info.offset; | |
1648 | else | |
1649 | mddev->bitmap_info.space = | |
1650 | -mddev->bitmap_info.offset; | |
1651 | } | |
e11e93fa | 1652 | |
f6705578 N |
1653 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1654 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1655 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1656 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1657 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1658 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1659 | if (mddev->delta_disks < 0 || |
1660 | (mddev->delta_disks == 0 && | |
1661 | (le32_to_cpu(sb->feature_map) | |
1662 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1663 | mddev->reshape_backwards = 1; | |
f6705578 N |
1664 | } else { |
1665 | mddev->reshape_position = MaxSector; | |
1666 | mddev->delta_disks = 0; | |
1667 | mddev->new_level = mddev->level; | |
1668 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1669 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1670 | } |
1671 | ||
486b0f7b | 1672 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) |
a62ab49e | 1673 | set_bit(MD_HAS_JOURNAL, &mddev->flags); |
ea0213e0 | 1674 | |
ddc08823 PB |
1675 | if (le32_to_cpu(sb->feature_map) & |
1676 | (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) { | |
ea0213e0 AP |
1677 | if (le32_to_cpu(sb->feature_map) & |
1678 | (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL)) | |
1679 | return -EINVAL; | |
ddc08823 PB |
1680 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) && |
1681 | (le32_to_cpu(sb->feature_map) & | |
1682 | MD_FEATURE_MULTIPLE_PPLS)) | |
1683 | return -EINVAL; | |
ea0213e0 AP |
1684 | set_bit(MD_HAS_PPL, &mddev->flags); |
1685 | } | |
41158c7e | 1686 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1687 | /* Insist of good event counter while assembling, except for |
1688 | * spares (which don't need an event count) */ | |
1da177e4 | 1689 | ++ev1; |
be6800a7 N |
1690 | if (rdev->desc_nr >= 0 && |
1691 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
a3dfbdaa SL |
1692 | (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX || |
1693 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)) | |
be6800a7 N |
1694 | if (ev1 < mddev->events) |
1695 | return -EINVAL; | |
41158c7e N |
1696 | } else if (mddev->bitmap) { |
1697 | /* If adding to array with a bitmap, then we can accept an | |
1698 | * older device, but not too old. | |
1699 | */ | |
41158c7e N |
1700 | if (ev1 < mddev->bitmap->events_cleared) |
1701 | return 0; | |
8313b8e5 N |
1702 | if (ev1 < mddev->events) |
1703 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1704 | } else { |
1705 | if (ev1 < mddev->events) | |
1706 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1707 | return 0; | |
1708 | } | |
1da177e4 LT |
1709 | if (mddev->level != LEVEL_MULTIPATH) { |
1710 | int role; | |
3673f305 N |
1711 | if (rdev->desc_nr < 0 || |
1712 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
c4d4c91b | 1713 | role = MD_DISK_ROLE_SPARE; |
3673f305 N |
1714 | rdev->desc_nr = -1; |
1715 | } else | |
1716 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 | 1717 | switch(role) { |
c4d4c91b | 1718 | case MD_DISK_ROLE_SPARE: /* spare */ |
1da177e4 | 1719 | break; |
c4d4c91b | 1720 | case MD_DISK_ROLE_FAULTY: /* faulty */ |
b2d444d7 | 1721 | set_bit(Faulty, &rdev->flags); |
1da177e4 | 1722 | break; |
bac624f3 SL |
1723 | case MD_DISK_ROLE_JOURNAL: /* journal device */ |
1724 | if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) { | |
1725 | /* journal device without journal feature */ | |
9d48739e | 1726 | pr_warn("md: journal device provided without journal feature, ignoring the device\n"); |
bac624f3 SL |
1727 | return -EINVAL; |
1728 | } | |
1729 | set_bit(Journal, &rdev->flags); | |
3069aa8d | 1730 | rdev->journal_tail = le64_to_cpu(sb->journal_tail); |
9b15603d | 1731 | rdev->raid_disk = 0; |
bac624f3 | 1732 | break; |
1da177e4 | 1733 | default: |
f466722c | 1734 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1735 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1736 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1737 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1738 | if (!(le32_to_cpu(sb->feature_map) & |
1739 | MD_FEATURE_RECOVERY_BITMAP)) | |
1740 | rdev->saved_raid_disk = -1; | |
1741 | } else | |
5fd6c1dc | 1742 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1743 | rdev->raid_disk = role; |
1744 | break; | |
1745 | } | |
8ddf9efe N |
1746 | if (sb->devflags & WriteMostly1) |
1747 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1748 | if (sb->devflags & FailFast1) |
1749 | set_bit(FailFast, &rdev->flags); | |
2d78f8c4 N |
1750 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1751 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1752 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1753 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1754 | |
1da177e4 LT |
1755 | return 0; |
1756 | } | |
1757 | ||
fd01b88c | 1758 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1759 | { |
1760 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1761 | struct md_rdev *rdev2; |
1da177e4 LT |
1762 | int max_dev, i; |
1763 | /* make rdev->sb match mddev and rdev data. */ | |
1764 | ||
65a06f06 | 1765 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1766 | |
1767 | sb->feature_map = 0; | |
1768 | sb->pad0 = 0; | |
5fd6c1dc | 1769 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1770 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1771 | ||
1772 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1773 | sb->events = cpu_to_le64(mddev->events); | |
1774 | if (mddev->in_sync) | |
1775 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
bd18f646 SL |
1776 | else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags)) |
1777 | sb->resync_offset = cpu_to_le64(MaxSector); | |
1da177e4 LT |
1778 | else |
1779 | sb->resync_offset = cpu_to_le64(0); | |
1780 | ||
1c05b4bc | 1781 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1782 | |
f0ca340c | 1783 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1784 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1785 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1786 | sb->level = cpu_to_le32(mddev->level); |
1787 | sb->layout = cpu_to_le32(mddev->layout); | |
688834e6 N |
1788 | if (test_bit(FailFast, &rdev->flags)) |
1789 | sb->devflags |= FailFast1; | |
1790 | else | |
1791 | sb->devflags &= ~FailFast1; | |
f0ca340c | 1792 | |
aeb9b211 N |
1793 | if (test_bit(WriteMostly, &rdev->flags)) |
1794 | sb->devflags |= WriteMostly1; | |
1795 | else | |
1796 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1797 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1798 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1799 | |
c3d9714e N |
1800 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1801 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1802 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1803 | } |
5fd6c1dc | 1804 | |
f2076e7d | 1805 | if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) && |
97e4f42d | 1806 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1807 | sb->feature_map |= |
1808 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1809 | sb->recovery_offset = | |
1810 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1811 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1812 | sb->feature_map |= | |
1813 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1814 | } |
3069aa8d SL |
1815 | /* Note: recovery_offset and journal_tail share space */ |
1816 | if (test_bit(Journal, &rdev->flags)) | |
1817 | sb->journal_tail = cpu_to_le64(rdev->journal_tail); | |
2d78f8c4 N |
1818 | if (test_bit(Replacement, &rdev->flags)) |
1819 | sb->feature_map |= | |
1820 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1821 | |
f6705578 N |
1822 | if (mddev->reshape_position != MaxSector) { |
1823 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1824 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1825 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1826 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1827 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1828 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1829 | if (mddev->delta_disks == 0 && |
1830 | mddev->reshape_backwards) | |
1831 | sb->feature_map | |
1832 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1833 | if (rdev->new_data_offset != rdev->data_offset) { |
1834 | sb->feature_map | |
1835 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1836 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1837 | - rdev->data_offset)); | |
1838 | } | |
f6705578 | 1839 | } |
a654b9d8 | 1840 | |
3c462c88 GR |
1841 | if (mddev_is_clustered(mddev)) |
1842 | sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED); | |
1843 | ||
2699b672 N |
1844 | if (rdev->badblocks.count == 0) |
1845 | /* Nothing to do for bad blocks*/ ; | |
1846 | else if (sb->bblog_offset == 0) | |
1847 | /* Cannot record bad blocks on this device */ | |
1848 | md_error(mddev, rdev); | |
1849 | else { | |
1850 | struct badblocks *bb = &rdev->badblocks; | |
1851 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1852 | u64 *p = bb->page; | |
1853 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1854 | if (bb->changed) { | |
1855 | unsigned seq; | |
1856 | ||
1857 | retry: | |
1858 | seq = read_seqbegin(&bb->lock); | |
1859 | ||
1860 | memset(bbp, 0xff, PAGE_SIZE); | |
1861 | ||
1862 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1863 | u64 internal_bb = p[i]; |
2699b672 N |
1864 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1865 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1866 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1867 | } |
d0962936 | 1868 | bb->changed = 0; |
2699b672 N |
1869 | if (read_seqretry(&bb->lock, seq)) |
1870 | goto retry; | |
1871 | ||
1872 | bb->sector = (rdev->sb_start + | |
1873 | (int)le32_to_cpu(sb->bblog_offset)); | |
1874 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1875 | } |
1876 | } | |
1877 | ||
1da177e4 | 1878 | max_dev = 0; |
dafb20fa | 1879 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1880 | if (rdev2->desc_nr+1 > max_dev) |
1881 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1882 | |
70471daf N |
1883 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1884 | int bmask; | |
a778b73f | 1885 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1886 | rdev->sb_size = max_dev * 2 + 256; |
1887 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1888 | if (rdev->sb_size & bmask) | |
1889 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1890 | } else |
1891 | max_dev = le32_to_cpu(sb->max_dev); | |
1892 | ||
1da177e4 | 1893 | for (i=0; i<max_dev;i++) |
8df72024 | 1894 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE); |
f72ffdd6 | 1895 | |
a97b7896 SL |
1896 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) |
1897 | sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL); | |
f72ffdd6 | 1898 | |
ea0213e0 | 1899 | if (test_bit(MD_HAS_PPL, &mddev->flags)) { |
ddc08823 PB |
1900 | if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags)) |
1901 | sb->feature_map |= | |
1902 | cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS); | |
1903 | else | |
1904 | sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL); | |
ea0213e0 AP |
1905 | sb->ppl.offset = cpu_to_le16(rdev->ppl.offset); |
1906 | sb->ppl.size = cpu_to_le16(rdev->ppl.size); | |
1907 | } | |
1908 | ||
dafb20fa | 1909 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1910 | i = rdev2->desc_nr; |
b2d444d7 | 1911 | if (test_bit(Faulty, &rdev2->flags)) |
c4d4c91b | 1912 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY); |
b2d444d7 | 1913 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1914 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
a97b7896 | 1915 | else if (test_bit(Journal, &rdev2->flags)) |
bac624f3 | 1916 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL); |
93be75ff | 1917 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1918 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 | 1919 | else |
c4d4c91b | 1920 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE); |
1da177e4 LT |
1921 | } |
1922 | ||
1da177e4 LT |
1923 | sb->sb_csum = calc_sb_1_csum(sb); |
1924 | } | |
1925 | ||
0cd17fec | 1926 | static unsigned long long |
3cb03002 | 1927 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1928 | { |
1929 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1930 | sector_t max_sectors; |
58c0fed4 | 1931 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1932 | return 0; /* component must fit device */ |
c6563a8c N |
1933 | if (rdev->data_offset != rdev->new_data_offset) |
1934 | return 0; /* too confusing */ | |
0f420358 | 1935 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1936 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1937 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1938 | max_sectors -= rdev->data_offset; |
1939 | if (!num_sectors || num_sectors > max_sectors) | |
1940 | num_sectors = max_sectors; | |
c3d9714e | 1941 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1942 | /* minor version 0 with bitmap we can't move */ |
1943 | return 0; | |
1944 | } else { | |
1945 | /* minor version 0; superblock after data */ | |
0f420358 | 1946 | sector_t sb_start; |
77304d2a | 1947 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1948 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1949 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1950 | if (!num_sectors || num_sectors > max_sectors) |
1951 | num_sectors = max_sectors; | |
0f420358 | 1952 | rdev->sb_start = sb_start; |
0cd17fec | 1953 | } |
65a06f06 | 1954 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1955 | sb->data_size = cpu_to_le64(num_sectors); |
3fb632e4 | 1956 | sb->super_offset = cpu_to_le64(rdev->sb_start); |
0cd17fec | 1957 | sb->sb_csum = calc_sb_1_csum(sb); |
46533ff7 N |
1958 | do { |
1959 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
1960 | rdev->sb_page); | |
1961 | } while (md_super_wait(rdev->mddev) < 0); | |
c26a44ed | 1962 | return num_sectors; |
c6563a8c N |
1963 | |
1964 | } | |
1965 | ||
1966 | static int | |
1967 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1968 | unsigned long long new_offset) | |
1969 | { | |
1970 | /* All necessary checks on new >= old have been done */ | |
1971 | struct bitmap *bitmap; | |
1972 | if (new_offset >= rdev->data_offset) | |
1973 | return 1; | |
1974 | ||
1975 | /* with 1.0 metadata, there is no metadata to tread on | |
1976 | * so we can always move back */ | |
1977 | if (rdev->mddev->minor_version == 0) | |
1978 | return 1; | |
1979 | ||
1980 | /* otherwise we must be sure not to step on | |
1981 | * any metadata, so stay: | |
1982 | * 36K beyond start of superblock | |
1983 | * beyond end of badblocks | |
1984 | * beyond write-intent bitmap | |
1985 | */ | |
1986 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1987 | return 0; | |
1988 | bitmap = rdev->mddev->bitmap; | |
1989 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1990 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1991 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1992 | return 0; |
1993 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1994 | return 0; | |
1995 | ||
1996 | return 1; | |
0cd17fec | 1997 | } |
1da177e4 | 1998 | |
75c96f85 | 1999 | static struct super_type super_types[] = { |
1da177e4 LT |
2000 | [0] = { |
2001 | .name = "0.90.0", | |
2002 | .owner = THIS_MODULE, | |
0cd17fec CW |
2003 | .load_super = super_90_load, |
2004 | .validate_super = super_90_validate, | |
2005 | .sync_super = super_90_sync, | |
2006 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 2007 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
2008 | }, |
2009 | [1] = { | |
2010 | .name = "md-1", | |
2011 | .owner = THIS_MODULE, | |
0cd17fec CW |
2012 | .load_super = super_1_load, |
2013 | .validate_super = super_1_validate, | |
2014 | .sync_super = super_1_sync, | |
2015 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 2016 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
2017 | }, |
2018 | }; | |
1da177e4 | 2019 | |
fd01b88c | 2020 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
2021 | { |
2022 | if (mddev->sync_super) { | |
2023 | mddev->sync_super(mddev, rdev); | |
2024 | return; | |
2025 | } | |
2026 | ||
2027 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
2028 | ||
2029 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
2030 | } | |
2031 | ||
fd01b88c | 2032 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 2033 | { |
3cb03002 | 2034 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 2035 | |
4b80991c | 2036 | rcu_read_lock(); |
0b020e85 SL |
2037 | rdev_for_each_rcu(rdev, mddev1) { |
2038 | if (test_bit(Faulty, &rdev->flags) || | |
2039 | test_bit(Journal, &rdev->flags) || | |
2040 | rdev->raid_disk == -1) | |
2041 | continue; | |
2042 | rdev_for_each_rcu(rdev2, mddev2) { | |
2043 | if (test_bit(Faulty, &rdev2->flags) || | |
2044 | test_bit(Journal, &rdev2->flags) || | |
2045 | rdev2->raid_disk == -1) | |
2046 | continue; | |
7dd5e7c3 | 2047 | if (rdev->bdev->bd_contains == |
4b80991c N |
2048 | rdev2->bdev->bd_contains) { |
2049 | rcu_read_unlock(); | |
7dd5e7c3 | 2050 | return 1; |
4b80991c | 2051 | } |
0b020e85 SL |
2052 | } |
2053 | } | |
4b80991c | 2054 | rcu_read_unlock(); |
1da177e4 LT |
2055 | return 0; |
2056 | } | |
2057 | ||
2058 | static LIST_HEAD(pending_raid_disks); | |
2059 | ||
ac5e7113 AN |
2060 | /* |
2061 | * Try to register data integrity profile for an mddev | |
2062 | * | |
2063 | * This is called when an array is started and after a disk has been kicked | |
2064 | * from the array. It only succeeds if all working and active component devices | |
2065 | * are integrity capable with matching profiles. | |
2066 | */ | |
fd01b88c | 2067 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 2068 | { |
3cb03002 | 2069 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
2070 | |
2071 | if (list_empty(&mddev->disks)) | |
2072 | return 0; /* nothing to do */ | |
629acb6a JB |
2073 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
2074 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 2075 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
2076 | /* skip spares and non-functional disks */ |
2077 | if (test_bit(Faulty, &rdev->flags)) | |
2078 | continue; | |
2079 | if (rdev->raid_disk < 0) | |
2080 | continue; | |
ac5e7113 AN |
2081 | if (!reference) { |
2082 | /* Use the first rdev as the reference */ | |
2083 | reference = rdev; | |
2084 | continue; | |
2085 | } | |
2086 | /* does this rdev's profile match the reference profile? */ | |
2087 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
2088 | rdev->bdev->bd_disk) < 0) | |
2089 | return -EINVAL; | |
2090 | } | |
89078d57 MP |
2091 | if (!reference || !bdev_get_integrity(reference->bdev)) |
2092 | return 0; | |
ac5e7113 AN |
2093 | /* |
2094 | * All component devices are integrity capable and have matching | |
2095 | * profiles, register the common profile for the md device. | |
2096 | */ | |
25520d55 MP |
2097 | blk_integrity_register(mddev->gendisk, |
2098 | bdev_get_integrity(reference->bdev)); | |
2099 | ||
9d48739e | 2100 | pr_debug("md: data integrity enabled on %s\n", mdname(mddev)); |
a91a2785 | 2101 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { |
9d48739e | 2102 | pr_err("md: failed to create integrity pool for %s\n", |
a91a2785 MP |
2103 | mdname(mddev)); |
2104 | return -EINVAL; | |
2105 | } | |
ac5e7113 AN |
2106 | return 0; |
2107 | } | |
2108 | EXPORT_SYMBOL(md_integrity_register); | |
2109 | ||
1501efad DW |
2110 | /* |
2111 | * Attempt to add an rdev, but only if it is consistent with the current | |
2112 | * integrity profile | |
2113 | */ | |
2114 | int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) | |
3f9d99c1 | 2115 | { |
2863b9eb JB |
2116 | struct blk_integrity *bi_rdev; |
2117 | struct blk_integrity *bi_mddev; | |
1501efad | 2118 | char name[BDEVNAME_SIZE]; |
2863b9eb JB |
2119 | |
2120 | if (!mddev->gendisk) | |
1501efad | 2121 | return 0; |
2863b9eb JB |
2122 | |
2123 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2124 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2125 | |
ac5e7113 | 2126 | if (!bi_mddev) /* nothing to do */ |
1501efad DW |
2127 | return 0; |
2128 | ||
2129 | if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) { | |
9d48739e N |
2130 | pr_err("%s: incompatible integrity profile for %s\n", |
2131 | mdname(mddev), bdevname(rdev->bdev, name)); | |
1501efad DW |
2132 | return -ENXIO; |
2133 | } | |
2134 | ||
2135 | return 0; | |
3f9d99c1 | 2136 | } |
ac5e7113 | 2137 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2138 | |
f72ffdd6 | 2139 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2140 | { |
7dd5e7c3 | 2141 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2142 | struct kobject *ko; |
5e55e2f5 | 2143 | int err; |
1da177e4 | 2144 | |
11e2ede0 DW |
2145 | /* prevent duplicates */ |
2146 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2147 | return -EEXIST; | |
2148 | ||
97b20ef7 N |
2149 | if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) && |
2150 | mddev->pers) | |
2151 | return -EROFS; | |
2152 | ||
dd8ac336 | 2153 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
f6b6ec5c SL |
2154 | if (!test_bit(Journal, &rdev->flags) && |
2155 | rdev->sectors && | |
2156 | (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2157 | if (mddev->pers) { |
2158 | /* Cannot change size, so fail | |
2159 | * If mddev->level <= 0, then we don't care | |
2160 | * about aligning sizes (e.g. linear) | |
2161 | */ | |
2162 | if (mddev->level > 0) | |
2163 | return -ENOSPC; | |
2164 | } else | |
dd8ac336 | 2165 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2166 | } |
1da177e4 LT |
2167 | |
2168 | /* Verify rdev->desc_nr is unique. | |
2169 | * If it is -1, assign a free number, else | |
2170 | * check number is not in use | |
2171 | */ | |
4878e9eb | 2172 | rcu_read_lock(); |
1da177e4 LT |
2173 | if (rdev->desc_nr < 0) { |
2174 | int choice = 0; | |
4878e9eb N |
2175 | if (mddev->pers) |
2176 | choice = mddev->raid_disks; | |
57d051dc | 2177 | while (md_find_rdev_nr_rcu(mddev, choice)) |
1da177e4 LT |
2178 | choice++; |
2179 | rdev->desc_nr = choice; | |
2180 | } else { | |
57d051dc | 2181 | if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
4878e9eb | 2182 | rcu_read_unlock(); |
1da177e4 | 2183 | return -EBUSY; |
4878e9eb | 2184 | } |
1da177e4 | 2185 | } |
4878e9eb | 2186 | rcu_read_unlock(); |
f6b6ec5c SL |
2187 | if (!test_bit(Journal, &rdev->flags) && |
2188 | mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { | |
9d48739e N |
2189 | pr_warn("md: %s: array is limited to %d devices\n", |
2190 | mdname(mddev), mddev->max_disks); | |
de01dfad N |
2191 | return -EBUSY; |
2192 | } | |
19133a42 | 2193 | bdevname(rdev->bdev,b); |
90a9befb | 2194 | strreplace(b, '/', '!'); |
649316b2 | 2195 | |
1da177e4 | 2196 | rdev->mddev = mddev; |
9d48739e | 2197 | pr_debug("md: bind<%s>\n", b); |
86e6ffdd | 2198 | |
b2d6db58 | 2199 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2200 | goto fail; |
86e6ffdd | 2201 | |
0762b8bd | 2202 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2203 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2204 | /* failure here is OK */; | |
2205 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2206 | |
4b80991c | 2207 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2208 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2209 | |
2210 | /* May as well allow recovery to be retried once */ | |
5389042f | 2211 | mddev->recovery_disabled++; |
3f9d99c1 | 2212 | |
1da177e4 | 2213 | return 0; |
5e55e2f5 N |
2214 | |
2215 | fail: | |
9d48739e N |
2216 | pr_warn("md: failed to register dev-%s for %s\n", |
2217 | b, mdname(mddev)); | |
5e55e2f5 | 2218 | return err; |
1da177e4 LT |
2219 | } |
2220 | ||
177a99b2 | 2221 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2222 | { |
3cb03002 | 2223 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2224 | kobject_del(&rdev->kobj); |
177a99b2 | 2225 | kobject_put(&rdev->kobj); |
5792a285 N |
2226 | } |
2227 | ||
f72ffdd6 | 2228 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2229 | { |
2230 | char b[BDEVNAME_SIZE]; | |
403df478 | 2231 | |
49731baa | 2232 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2233 | list_del_rcu(&rdev->same_set); |
9d48739e | 2234 | pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1da177e4 | 2235 | rdev->mddev = NULL; |
86e6ffdd | 2236 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2237 | sysfs_put(rdev->sysfs_state); |
2238 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2239 | rdev->badblocks.count = 0; |
5792a285 | 2240 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2241 | * writing to 'remove' to "dev/state". We also need |
2242 | * to delay it due to rcu usage. | |
5792a285 | 2243 | */ |
4b80991c | 2244 | synchronize_rcu(); |
177a99b2 N |
2245 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2246 | kobject_get(&rdev->kobj); | |
e804ac78 | 2247 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2248 | } |
2249 | ||
2250 | /* | |
2251 | * prevent the device from being mounted, repartitioned or | |
2252 | * otherwise reused by a RAID array (or any other kernel | |
2253 | * subsystem), by bd_claiming the device. | |
2254 | */ | |
3cb03002 | 2255 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2256 | { |
2257 | int err = 0; | |
2258 | struct block_device *bdev; | |
2259 | char b[BDEVNAME_SIZE]; | |
2260 | ||
d4d77629 | 2261 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2262 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 | 2263 | if (IS_ERR(bdev)) { |
9d48739e | 2264 | pr_warn("md: could not open %s.\n", __bdevname(dev, b)); |
1da177e4 LT |
2265 | return PTR_ERR(bdev); |
2266 | } | |
1da177e4 LT |
2267 | rdev->bdev = bdev; |
2268 | return err; | |
2269 | } | |
2270 | ||
3cb03002 | 2271 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2272 | { |
2273 | struct block_device *bdev = rdev->bdev; | |
2274 | rdev->bdev = NULL; | |
e525fd89 | 2275 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2276 | } |
2277 | ||
2278 | void md_autodetect_dev(dev_t dev); | |
2279 | ||
f72ffdd6 | 2280 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2281 | { |
2282 | char b[BDEVNAME_SIZE]; | |
403df478 | 2283 | |
9d48739e | 2284 | pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b)); |
545c8795 | 2285 | md_rdev_clear(rdev); |
1da177e4 | 2286 | #ifndef MODULE |
d0fae18f N |
2287 | if (test_bit(AutoDetected, &rdev->flags)) |
2288 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2289 | #endif |
2290 | unlock_rdev(rdev); | |
86e6ffdd | 2291 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2292 | } |
2293 | ||
fb56dfef | 2294 | void md_kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2295 | { |
2296 | unbind_rdev_from_array(rdev); | |
2297 | export_rdev(rdev); | |
2298 | } | |
fb56dfef | 2299 | EXPORT_SYMBOL_GPL(md_kick_rdev_from_array); |
1da177e4 | 2300 | |
fd01b88c | 2301 | static void export_array(struct mddev *mddev) |
1da177e4 | 2302 | { |
0638bb0e | 2303 | struct md_rdev *rdev; |
1da177e4 | 2304 | |
0638bb0e N |
2305 | while (!list_empty(&mddev->disks)) { |
2306 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2307 | same_set); | |
fb56dfef | 2308 | md_kick_rdev_from_array(rdev); |
1da177e4 | 2309 | } |
1da177e4 LT |
2310 | mddev->raid_disks = 0; |
2311 | mddev->major_version = 0; | |
2312 | } | |
2313 | ||
6497709b N |
2314 | static bool set_in_sync(struct mddev *mddev) |
2315 | { | |
ed9b66d2 | 2316 | WARN_ON_ONCE(NR_CPUS != 1 && !spin_is_locked(&mddev->lock)); |
4ad23a97 N |
2317 | if (!mddev->in_sync) { |
2318 | mddev->sync_checkers++; | |
2319 | spin_unlock(&mddev->lock); | |
2320 | percpu_ref_switch_to_atomic_sync(&mddev->writes_pending); | |
2321 | spin_lock(&mddev->lock); | |
2322 | if (!mddev->in_sync && | |
2323 | percpu_ref_is_zero(&mddev->writes_pending)) { | |
6497709b | 2324 | mddev->in_sync = 1; |
4ad23a97 N |
2325 | /* |
2326 | * Ensure ->in_sync is visible before we clear | |
2327 | * ->sync_checkers. | |
2328 | */ | |
55cc39f3 | 2329 | smp_mb(); |
6497709b N |
2330 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2331 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
2332 | } | |
4ad23a97 N |
2333 | if (--mddev->sync_checkers == 0) |
2334 | percpu_ref_switch_to_percpu(&mddev->writes_pending); | |
6497709b N |
2335 | } |
2336 | if (mddev->safemode == 1) | |
2337 | mddev->safemode = 0; | |
2338 | return mddev->in_sync; | |
2339 | } | |
2340 | ||
f72ffdd6 | 2341 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2342 | { |
42543769 N |
2343 | /* Update each superblock (in-memory image), but |
2344 | * if we are allowed to, skip spares which already | |
2345 | * have the right event counter, or have one earlier | |
2346 | * (which would mean they aren't being marked as dirty | |
2347 | * with the rest of the array) | |
2348 | */ | |
3cb03002 | 2349 | struct md_rdev *rdev; |
dafb20fa | 2350 | rdev_for_each(rdev, mddev) { |
42543769 N |
2351 | if (rdev->sb_events == mddev->events || |
2352 | (nospares && | |
2353 | rdev->raid_disk < 0 && | |
42543769 N |
2354 | rdev->sb_events+1 == mddev->events)) { |
2355 | /* Don't update this superblock */ | |
2356 | rdev->sb_loaded = 2; | |
2357 | } else { | |
076f968b | 2358 | sync_super(mddev, rdev); |
42543769 N |
2359 | rdev->sb_loaded = 1; |
2360 | } | |
1da177e4 LT |
2361 | } |
2362 | } | |
2363 | ||
2aa82191 GR |
2364 | static bool does_sb_need_changing(struct mddev *mddev) |
2365 | { | |
2366 | struct md_rdev *rdev; | |
2367 | struct mdp_superblock_1 *sb; | |
2368 | int role; | |
2369 | ||
2370 | /* Find a good rdev */ | |
2371 | rdev_for_each(rdev, mddev) | |
2372 | if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags)) | |
2373 | break; | |
2374 | ||
2375 | /* No good device found. */ | |
2376 | if (!rdev) | |
2377 | return false; | |
2378 | ||
2379 | sb = page_address(rdev->sb_page); | |
2380 | /* Check if a device has become faulty or a spare become active */ | |
2381 | rdev_for_each(rdev, mddev) { | |
2382 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
2383 | /* Device activated? */ | |
2384 | if (role == 0xffff && rdev->raid_disk >=0 && | |
2385 | !test_bit(Faulty, &rdev->flags)) | |
2386 | return true; | |
2387 | /* Device turned faulty? */ | |
2388 | if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd)) | |
2389 | return true; | |
2390 | } | |
2391 | ||
2392 | /* Check if any mddev parameters have changed */ | |
2393 | if ((mddev->dev_sectors != le64_to_cpu(sb->size)) || | |
2394 | (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) || | |
13459213 | 2395 | (mddev->layout != le32_to_cpu(sb->layout)) || |
2aa82191 GR |
2396 | (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) || |
2397 | (mddev->chunk_sectors != le32_to_cpu(sb->chunksize))) | |
2398 | return true; | |
2399 | ||
2400 | return false; | |
2401 | } | |
2402 | ||
1aee41f6 | 2403 | void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2404 | { |
3cb03002 | 2405 | struct md_rdev *rdev; |
06d91a5f | 2406 | int sync_req; |
42543769 | 2407 | int nospares = 0; |
2699b672 | 2408 | int any_badblocks_changed = 0; |
23b63f9f | 2409 | int ret = -1; |
1da177e4 | 2410 | |
d87f064f N |
2411 | if (mddev->ro) { |
2412 | if (force_change) | |
2953079c | 2413 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
d87f064f N |
2414 | return; |
2415 | } | |
2aa82191 | 2416 | |
2c97cf13 | 2417 | repeat: |
2aa82191 | 2418 | if (mddev_is_clustered(mddev)) { |
2953079c | 2419 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
2aa82191 | 2420 | force_change = 1; |
2953079c | 2421 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
85ad1d13 | 2422 | nospares = 1; |
23b63f9f | 2423 | ret = md_cluster_ops->metadata_update_start(mddev); |
2aa82191 GR |
2424 | /* Has someone else has updated the sb */ |
2425 | if (!does_sb_need_changing(mddev)) { | |
23b63f9f GJ |
2426 | if (ret == 0) |
2427 | md_cluster_ops->metadata_update_cancel(mddev); | |
2953079c SL |
2428 | bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2429 | BIT(MD_SB_CHANGE_DEVS) | | |
2430 | BIT(MD_SB_CHANGE_CLEAN)); | |
2aa82191 GR |
2431 | return; |
2432 | } | |
2433 | } | |
2c97cf13 | 2434 | |
3a3a5ddb | 2435 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2436 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2437 | if (rdev->raid_disk >= 0 && |
2438 | mddev->delta_disks >= 0 && | |
f2076e7d | 2439 | !test_bit(Journal, &rdev->flags) && |
3a3a5ddb N |
2440 | !test_bit(In_sync, &rdev->flags) && |
2441 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2442 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2443 | ||
f72ffdd6 | 2444 | } |
bd52b746 | 2445 | if (!mddev->persistent) { |
2953079c SL |
2446 | clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2447 | clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
de393cde | 2448 | if (!mddev->external) { |
2953079c | 2449 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
dafb20fa | 2450 | rdev_for_each(rdev, mddev) { |
de393cde | 2451 | if (rdev->badblocks.changed) { |
d0962936 | 2452 | rdev->badblocks.changed = 0; |
fc974ee2 | 2453 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2454 | md_error(mddev, rdev); |
2455 | } | |
2456 | clear_bit(Blocked, &rdev->flags); | |
2457 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2458 | wake_up(&rdev->blocked_wait); | |
2459 | } | |
2460 | } | |
3a3a5ddb N |
2461 | wake_up(&mddev->sb_wait); |
2462 | return; | |
2463 | } | |
2464 | ||
85572d7c | 2465 | spin_lock(&mddev->lock); |
84692195 | 2466 | |
9ebc6ef1 | 2467 | mddev->utime = ktime_get_real_seconds(); |
3a3a5ddb | 2468 | |
2953079c | 2469 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
850b2b42 | 2470 | force_change = 1; |
2953079c | 2471 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
850b2b42 N |
2472 | /* just a clean<-> dirty transition, possibly leave spares alone, |
2473 | * though if events isn't the right even/odd, we will have to do | |
2474 | * spares after all | |
2475 | */ | |
2476 | nospares = 1; | |
2477 | if (force_change) | |
2478 | nospares = 0; | |
2479 | if (mddev->degraded) | |
84692195 N |
2480 | /* If the array is degraded, then skipping spares is both |
2481 | * dangerous and fairly pointless. | |
2482 | * Dangerous because a device that was removed from the array | |
2483 | * might have a event_count that still looks up-to-date, | |
2484 | * so it can be re-added without a resync. | |
2485 | * Pointless because if there are any spares to skip, | |
2486 | * then a recovery will happen and soon that array won't | |
2487 | * be degraded any more and the spare can go back to sleep then. | |
2488 | */ | |
850b2b42 | 2489 | nospares = 0; |
84692195 | 2490 | |
06d91a5f | 2491 | sync_req = mddev->in_sync; |
42543769 N |
2492 | |
2493 | /* If this is just a dirty<->clean transition, and the array is clean | |
2494 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2495 | if (nospares |
42543769 | 2496 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2497 | && mddev->can_decrease_events |
2498 | && mddev->events != 1) { | |
42543769 | 2499 | mddev->events--; |
a8707c08 N |
2500 | mddev->can_decrease_events = 0; |
2501 | } else { | |
42543769 N |
2502 | /* otherwise we have to go forward and ... */ |
2503 | mddev->events ++; | |
a8707c08 | 2504 | mddev->can_decrease_events = nospares; |
42543769 | 2505 | } |
1da177e4 | 2506 | |
403df478 N |
2507 | /* |
2508 | * This 64-bit counter should never wrap. | |
2509 | * Either we are in around ~1 trillion A.C., assuming | |
2510 | * 1 reboot per second, or we have a bug... | |
2511 | */ | |
2512 | WARN_ON(mddev->events == 0); | |
2699b672 | 2513 | |
dafb20fa | 2514 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2515 | if (rdev->badblocks.changed) |
2516 | any_badblocks_changed++; | |
de393cde N |
2517 | if (test_bit(Faulty, &rdev->flags)) |
2518 | set_bit(FaultRecorded, &rdev->flags); | |
2519 | } | |
2699b672 | 2520 | |
e691063a | 2521 | sync_sbs(mddev, nospares); |
85572d7c | 2522 | spin_unlock(&mddev->lock); |
1da177e4 | 2523 | |
36a4e1fe N |
2524 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2525 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2526 | |
504634f6 SL |
2527 | if (mddev->queue) |
2528 | blk_add_trace_msg(mddev->queue, "md md_update_sb"); | |
46533ff7 | 2529 | rewrite: |
4ad13663 | 2530 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2531 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2532 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2533 | |
42543769 N |
2534 | if (rdev->sb_loaded != 1) |
2535 | continue; /* no noise on spare devices */ | |
1da177e4 | 2536 | |
f466722c | 2537 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2538 | md_super_write(mddev,rdev, |
0f420358 | 2539 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2540 | rdev->sb_page); |
36a4e1fe N |
2541 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2542 | bdevname(rdev->bdev, b), | |
2543 | (unsigned long long)rdev->sb_start); | |
42543769 | 2544 | rdev->sb_events = mddev->events; |
2699b672 N |
2545 | if (rdev->badblocks.size) { |
2546 | md_super_write(mddev, rdev, | |
2547 | rdev->badblocks.sector, | |
2548 | rdev->badblocks.size << 9, | |
2549 | rdev->bb_page); | |
2550 | rdev->badblocks.size = 0; | |
2551 | } | |
7bfa19f2 | 2552 | |
f466722c | 2553 | } else |
36a4e1fe N |
2554 | pr_debug("md: %s (skipping faulty)\n", |
2555 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2556 | |
7bfa19f2 | 2557 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2558 | /* only need to write one superblock... */ |
2559 | break; | |
2560 | } | |
46533ff7 N |
2561 | if (md_super_wait(mddev) < 0) |
2562 | goto rewrite; | |
2953079c | 2563 | /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2564 | |
2c97cf13 GJ |
2565 | if (mddev_is_clustered(mddev) && ret == 0) |
2566 | md_cluster_ops->metadata_update_finish(mddev); | |
2567 | ||
850b2b42 | 2568 | if (mddev->in_sync != sync_req || |
2953079c SL |
2569 | !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2570 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN))) | |
06d91a5f | 2571 | /* have to write it out again */ |
06d91a5f | 2572 | goto repeat; |
3d310eb7 | 2573 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2574 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2575 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2576 | |
dafb20fa | 2577 | rdev_for_each(rdev, mddev) { |
de393cde N |
2578 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2579 | clear_bit(Blocked, &rdev->flags); | |
2580 | ||
2581 | if (any_badblocks_changed) | |
fc974ee2 | 2582 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2583 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2584 | wake_up(&rdev->blocked_wait); | |
2585 | } | |
1da177e4 | 2586 | } |
1aee41f6 | 2587 | EXPORT_SYMBOL(md_update_sb); |
1da177e4 | 2588 | |
a6da4ef8 GR |
2589 | static int add_bound_rdev(struct md_rdev *rdev) |
2590 | { | |
2591 | struct mddev *mddev = rdev->mddev; | |
2592 | int err = 0; | |
87d4d916 | 2593 | bool add_journal = test_bit(Journal, &rdev->flags); |
a6da4ef8 | 2594 | |
87d4d916 | 2595 | if (!mddev->pers->hot_remove_disk || add_journal) { |
a6da4ef8 GR |
2596 | /* If there is hot_add_disk but no hot_remove_disk |
2597 | * then added disks for geometry changes, | |
2598 | * and should be added immediately. | |
2599 | */ | |
2600 | super_types[mddev->major_version]. | |
2601 | validate_super(mddev, rdev); | |
87d4d916 SL |
2602 | if (add_journal) |
2603 | mddev_suspend(mddev); | |
a6da4ef8 | 2604 | err = mddev->pers->hot_add_disk(mddev, rdev); |
87d4d916 SL |
2605 | if (add_journal) |
2606 | mddev_resume(mddev); | |
a6da4ef8 | 2607 | if (err) { |
db767672 | 2608 | md_kick_rdev_from_array(rdev); |
a6da4ef8 GR |
2609 | return err; |
2610 | } | |
2611 | } | |
2612 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2613 | ||
2953079c | 2614 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a6da4ef8 GR |
2615 | if (mddev->degraded) |
2616 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
2617 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2618 | md_new_event(mddev); | |
2619 | md_wakeup_thread(mddev->thread); | |
2620 | return 0; | |
2621 | } | |
1da177e4 | 2622 | |
7f6ce769 | 2623 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2624 | * We want to accept with case. For this we use cmd_match. |
2625 | */ | |
2626 | static int cmd_match(const char *cmd, const char *str) | |
2627 | { | |
2628 | /* See if cmd, written into a sysfs file, matches | |
2629 | * str. They must either be the same, or cmd can | |
2630 | * have a trailing newline | |
2631 | */ | |
2632 | while (*cmd && *str && *cmd == *str) { | |
2633 | cmd++; | |
2634 | str++; | |
2635 | } | |
2636 | if (*cmd == '\n') | |
2637 | cmd++; | |
2638 | if (*str || *cmd) | |
2639 | return 0; | |
2640 | return 1; | |
2641 | } | |
2642 | ||
86e6ffdd N |
2643 | struct rdev_sysfs_entry { |
2644 | struct attribute attr; | |
3cb03002 N |
2645 | ssize_t (*show)(struct md_rdev *, char *); |
2646 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2647 | }; |
2648 | ||
2649 | static ssize_t | |
3cb03002 | 2650 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd | 2651 | { |
35b785f7 | 2652 | char *sep = ","; |
20a49ff6 | 2653 | size_t len = 0; |
758bfc8a | 2654 | unsigned long flags = ACCESS_ONCE(rdev->flags); |
86e6ffdd | 2655 | |
758bfc8a | 2656 | if (test_bit(Faulty, &flags) || |
dcbcb486 TM |
2657 | (!test_bit(ExternalBbl, &flags) && |
2658 | rdev->badblocks.unacked_exist)) | |
35b785f7 TM |
2659 | len += sprintf(page+len, "faulty%s", sep); |
2660 | if (test_bit(In_sync, &flags)) | |
2661 | len += sprintf(page+len, "in_sync%s", sep); | |
2662 | if (test_bit(Journal, &flags)) | |
2663 | len += sprintf(page+len, "journal%s", sep); | |
2664 | if (test_bit(WriteMostly, &flags)) | |
2665 | len += sprintf(page+len, "write_mostly%s", sep); | |
758bfc8a | 2666 | if (test_bit(Blocked, &flags) || |
52c64152 | 2667 | (rdev->badblocks.unacked_exist |
35b785f7 TM |
2668 | && !test_bit(Faulty, &flags))) |
2669 | len += sprintf(page+len, "blocked%s", sep); | |
758bfc8a | 2670 | if (!test_bit(Faulty, &flags) && |
f2076e7d | 2671 | !test_bit(Journal, &flags) && |
35b785f7 TM |
2672 | !test_bit(In_sync, &flags)) |
2673 | len += sprintf(page+len, "spare%s", sep); | |
2674 | if (test_bit(WriteErrorSeen, &flags)) | |
2675 | len += sprintf(page+len, "write_error%s", sep); | |
2676 | if (test_bit(WantReplacement, &flags)) | |
2677 | len += sprintf(page+len, "want_replacement%s", sep); | |
2678 | if (test_bit(Replacement, &flags)) | |
2679 | len += sprintf(page+len, "replacement%s", sep); | |
2680 | if (test_bit(ExternalBbl, &flags)) | |
2681 | len += sprintf(page+len, "external_bbl%s", sep); | |
688834e6 N |
2682 | if (test_bit(FailFast, &flags)) |
2683 | len += sprintf(page+len, "failfast%s", sep); | |
35b785f7 TM |
2684 | |
2685 | if (len) | |
2686 | len -= strlen(sep); | |
2d78f8c4 | 2687 | |
86e6ffdd N |
2688 | return len+sprintf(page+len, "\n"); |
2689 | } | |
2690 | ||
45dc2de1 | 2691 | static ssize_t |
3cb03002 | 2692 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2693 | { |
2694 | /* can write | |
de393cde | 2695 | * faulty - simulates an error |
45dc2de1 | 2696 | * remove - disconnects the device |
f655675b N |
2697 | * writemostly - sets write_mostly |
2698 | * -writemostly - clears write_mostly | |
de393cde N |
2699 | * blocked - sets the Blocked flags |
2700 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2701 | * insync - sets Insync providing device isn't active |
f466722c N |
2702 | * -insync - clear Insync for a device with a slot assigned, |
2703 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2704 | * write_error - sets WriteErrorSeen |
2705 | * -write_error - clears WriteErrorSeen | |
688834e6 | 2706 | * {,-}failfast - set/clear FailFast |
45dc2de1 N |
2707 | */ |
2708 | int err = -EINVAL; | |
2709 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2710 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2711 | if (test_bit(Faulty, &rdev->flags)) |
2712 | err = 0; | |
2713 | else | |
2714 | err = -EBUSY; | |
45dc2de1 | 2715 | } else if (cmd_match(buf, "remove")) { |
5d881783 SL |
2716 | if (rdev->mddev->pers) { |
2717 | clear_bit(Blocked, &rdev->flags); | |
2718 | remove_and_add_spares(rdev->mddev, rdev); | |
2719 | } | |
45dc2de1 N |
2720 | if (rdev->raid_disk >= 0) |
2721 | err = -EBUSY; | |
2722 | else { | |
fd01b88c | 2723 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2724 | err = 0; |
a9720903 GJ |
2725 | if (mddev_is_clustered(mddev)) |
2726 | err = md_cluster_ops->remove_disk(mddev, rdev); | |
2727 | ||
2728 | if (err == 0) { | |
2729 | md_kick_rdev_from_array(rdev); | |
060b0689 | 2730 | if (mddev->pers) { |
2953079c | 2731 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
2732 | md_wakeup_thread(mddev->thread); |
2733 | } | |
a9720903 GJ |
2734 | md_new_event(mddev); |
2735 | } | |
45dc2de1 | 2736 | } |
f655675b N |
2737 | } else if (cmd_match(buf, "writemostly")) { |
2738 | set_bit(WriteMostly, &rdev->flags); | |
2739 | err = 0; | |
2740 | } else if (cmd_match(buf, "-writemostly")) { | |
2741 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2742 | err = 0; |
2743 | } else if (cmd_match(buf, "blocked")) { | |
2744 | set_bit(Blocked, &rdev->flags); | |
2745 | err = 0; | |
2746 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2747 | if (!test_bit(Faulty, &rdev->flags) && |
dcbcb486 | 2748 | !test_bit(ExternalBbl, &rdev->flags) && |
7da64a0a | 2749 | rdev->badblocks.unacked_exist) { |
de393cde N |
2750 | /* metadata handler doesn't understand badblocks, |
2751 | * so we need to fail the device | |
2752 | */ | |
2753 | md_error(rdev->mddev, rdev); | |
2754 | } | |
6bfe0b49 | 2755 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2756 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2757 | wake_up(&rdev->blocked_wait); |
2758 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2759 | md_wakeup_thread(rdev->mddev->thread); | |
2760 | ||
6d56e278 N |
2761 | err = 0; |
2762 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2763 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2764 | err = 0; |
688834e6 N |
2765 | } else if (cmd_match(buf, "failfast")) { |
2766 | set_bit(FailFast, &rdev->flags); | |
2767 | err = 0; | |
2768 | } else if (cmd_match(buf, "-failfast")) { | |
2769 | clear_bit(FailFast, &rdev->flags); | |
2770 | err = 0; | |
f2076e7d SL |
2771 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 && |
2772 | !test_bit(Journal, &rdev->flags)) { | |
e1960f8c N |
2773 | if (rdev->mddev->pers == NULL) { |
2774 | clear_bit(In_sync, &rdev->flags); | |
2775 | rdev->saved_raid_disk = rdev->raid_disk; | |
2776 | rdev->raid_disk = -1; | |
2777 | err = 0; | |
2778 | } | |
d7a9d443 N |
2779 | } else if (cmd_match(buf, "write_error")) { |
2780 | set_bit(WriteErrorSeen, &rdev->flags); | |
2781 | err = 0; | |
2782 | } else if (cmd_match(buf, "-write_error")) { | |
2783 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2784 | err = 0; | |
2d78f8c4 N |
2785 | } else if (cmd_match(buf, "want_replacement")) { |
2786 | /* Any non-spare device that is not a replacement can | |
2787 | * become want_replacement at any time, but we then need to | |
2788 | * check if recovery is needed. | |
2789 | */ | |
2790 | if (rdev->raid_disk >= 0 && | |
f2076e7d | 2791 | !test_bit(Journal, &rdev->flags) && |
2d78f8c4 N |
2792 | !test_bit(Replacement, &rdev->flags)) |
2793 | set_bit(WantReplacement, &rdev->flags); | |
2794 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2795 | md_wakeup_thread(rdev->mddev->thread); | |
2796 | err = 0; | |
2797 | } else if (cmd_match(buf, "-want_replacement")) { | |
2798 | /* Clearing 'want_replacement' is always allowed. | |
2799 | * Once replacements starts it is too late though. | |
2800 | */ | |
2801 | err = 0; | |
2802 | clear_bit(WantReplacement, &rdev->flags); | |
2803 | } else if (cmd_match(buf, "replacement")) { | |
2804 | /* Can only set a device as a replacement when array has not | |
2805 | * yet been started. Once running, replacement is automatic | |
2806 | * from spares, or by assigning 'slot'. | |
2807 | */ | |
2808 | if (rdev->mddev->pers) | |
2809 | err = -EBUSY; | |
2810 | else { | |
2811 | set_bit(Replacement, &rdev->flags); | |
2812 | err = 0; | |
2813 | } | |
2814 | } else if (cmd_match(buf, "-replacement")) { | |
2815 | /* Similarly, can only clear Replacement before start */ | |
2816 | if (rdev->mddev->pers) | |
2817 | err = -EBUSY; | |
2818 | else { | |
2819 | clear_bit(Replacement, &rdev->flags); | |
2820 | err = 0; | |
2821 | } | |
a6da4ef8 GR |
2822 | } else if (cmd_match(buf, "re-add")) { |
2823 | if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) { | |
97f6cd39 GR |
2824 | /* clear_bit is performed _after_ all the devices |
2825 | * have their local Faulty bit cleared. If any writes | |
2826 | * happen in the meantime in the local node, they | |
2827 | * will land in the local bitmap, which will be synced | |
2828 | * by this node eventually | |
2829 | */ | |
2830 | if (!mddev_is_clustered(rdev->mddev) || | |
2831 | (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) { | |
2832 | clear_bit(Faulty, &rdev->flags); | |
2833 | err = add_bound_rdev(rdev); | |
2834 | } | |
a6da4ef8 GR |
2835 | } else |
2836 | err = -EBUSY; | |
35b785f7 TM |
2837 | } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) { |
2838 | set_bit(ExternalBbl, &rdev->flags); | |
2839 | rdev->badblocks.shift = 0; | |
2840 | err = 0; | |
2841 | } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) { | |
2842 | clear_bit(ExternalBbl, &rdev->flags); | |
2843 | err = 0; | |
45dc2de1 | 2844 | } |
00bcb4ac N |
2845 | if (!err) |
2846 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2847 | return err ? err : len; |
2848 | } | |
80ca3a44 | 2849 | static struct rdev_sysfs_entry rdev_state = |
750f199e | 2850 | __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store); |
86e6ffdd | 2851 | |
4dbcdc75 | 2852 | static ssize_t |
3cb03002 | 2853 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2854 | { |
2855 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2856 | } | |
2857 | ||
2858 | static ssize_t | |
3cb03002 | 2859 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 | 2860 | { |
4c9309c0 AD |
2861 | unsigned int n; |
2862 | int rv; | |
2863 | ||
2864 | rv = kstrtouint(buf, 10, &n); | |
2865 | if (rv < 0) | |
2866 | return rv; | |
2867 | atomic_set(&rdev->corrected_errors, n); | |
2868 | return len; | |
4dbcdc75 N |
2869 | } |
2870 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2871 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2872 | |
014236d2 | 2873 | static ssize_t |
3cb03002 | 2874 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 | 2875 | { |
f2076e7d SL |
2876 | if (test_bit(Journal, &rdev->flags)) |
2877 | return sprintf(page, "journal\n"); | |
2878 | else if (rdev->raid_disk < 0) | |
014236d2 N |
2879 | return sprintf(page, "none\n"); |
2880 | else | |
2881 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2882 | } | |
2883 | ||
2884 | static ssize_t | |
3cb03002 | 2885 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 | 2886 | { |
4c9309c0 | 2887 | int slot; |
c303da6d | 2888 | int err; |
4c9309c0 | 2889 | |
f2076e7d SL |
2890 | if (test_bit(Journal, &rdev->flags)) |
2891 | return -EBUSY; | |
014236d2 N |
2892 | if (strncmp(buf, "none", 4)==0) |
2893 | slot = -1; | |
4c9309c0 AD |
2894 | else { |
2895 | err = kstrtouint(buf, 10, (unsigned int *)&slot); | |
2896 | if (err < 0) | |
2897 | return err; | |
2898 | } | |
6c2fce2e | 2899 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2900 | /* Setting 'slot' on an active array requires also |
2901 | * updating the 'rd%d' link, and communicating | |
2902 | * with the personality with ->hot_*_disk. | |
2903 | * For now we only support removing | |
2904 | * failed/spare devices. This normally happens automatically, | |
2905 | * but not when the metadata is externally managed. | |
2906 | */ | |
c303da6d N |
2907 | if (rdev->raid_disk == -1) |
2908 | return -EEXIST; | |
2909 | /* personality does all needed checks */ | |
01393f3d | 2910 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2911 | return -EINVAL; |
746d3207 N |
2912 | clear_bit(Blocked, &rdev->flags); |
2913 | remove_and_add_spares(rdev->mddev, rdev); | |
2914 | if (rdev->raid_disk >= 0) | |
2915 | return -EBUSY; | |
c303da6d N |
2916 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2917 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2918 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2919 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2920 | * if we ever get bitmaps working here. |
6c2fce2e | 2921 | */ |
cb01c549 | 2922 | int err; |
6c2fce2e NB |
2923 | |
2924 | if (rdev->raid_disk != -1) | |
2925 | return -EBUSY; | |
2926 | ||
c6751b2b N |
2927 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2928 | return -EBUSY; | |
2929 | ||
6c2fce2e NB |
2930 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2931 | return -EINVAL; | |
2932 | ||
ba1b41b6 N |
2933 | if (slot >= rdev->mddev->raid_disks && |
2934 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2935 | return -ENOSPC; | |
2936 | ||
6c2fce2e NB |
2937 | rdev->raid_disk = slot; |
2938 | if (test_bit(In_sync, &rdev->flags)) | |
2939 | rdev->saved_raid_disk = slot; | |
2940 | else | |
2941 | rdev->saved_raid_disk = -1; | |
d30519fc | 2942 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2943 | clear_bit(Bitmap_sync, &rdev->flags); |
cb01c549 GR |
2944 | err = rdev->mddev->pers-> |
2945 | hot_add_disk(rdev->mddev, rdev); | |
2946 | if (err) { | |
2947 | rdev->raid_disk = -1; | |
2948 | return err; | |
2949 | } else | |
2950 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2951 | if (sysfs_link_rdev(rdev->mddev, rdev)) | |
2952 | /* failure here is OK */; | |
6c2fce2e | 2953 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2954 | } else { |
ba1b41b6 N |
2955 | if (slot >= rdev->mddev->raid_disks && |
2956 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2957 | return -ENOSPC; |
2958 | rdev->raid_disk = slot; | |
2959 | /* assume it is working */ | |
c5d79adb N |
2960 | clear_bit(Faulty, &rdev->flags); |
2961 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2962 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2963 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2964 | } |
014236d2 N |
2965 | return len; |
2966 | } | |
2967 | ||
014236d2 | 2968 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2969 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2970 | |
93c8cad0 | 2971 | static ssize_t |
3cb03002 | 2972 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2973 | { |
6961ece4 | 2974 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2975 | } |
2976 | ||
2977 | static ssize_t | |
3cb03002 | 2978 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2979 | { |
c6563a8c | 2980 | unsigned long long offset; |
b29bebd6 | 2981 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2982 | return -EINVAL; |
8ed0a521 | 2983 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2984 | return -EBUSY; |
dd8ac336 | 2985 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2986 | /* Must set offset before size, so overlap checks |
2987 | * can be sane */ | |
2988 | return -EBUSY; | |
93c8cad0 | 2989 | rdev->data_offset = offset; |
25f7fd47 | 2990 | rdev->new_data_offset = offset; |
93c8cad0 N |
2991 | return len; |
2992 | } | |
2993 | ||
2994 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2995 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2996 | |
c6563a8c N |
2997 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2998 | { | |
2999 | return sprintf(page, "%llu\n", | |
3000 | (unsigned long long)rdev->new_data_offset); | |
3001 | } | |
3002 | ||
3003 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
3004 | const char *buf, size_t len) | |
3005 | { | |
3006 | unsigned long long new_offset; | |
3007 | struct mddev *mddev = rdev->mddev; | |
3008 | ||
b29bebd6 | 3009 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
3010 | return -EINVAL; |
3011 | ||
f851b60d N |
3012 | if (mddev->sync_thread || |
3013 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
3014 | return -EBUSY; |
3015 | if (new_offset == rdev->data_offset) | |
3016 | /* reset is always permitted */ | |
3017 | ; | |
3018 | else if (new_offset > rdev->data_offset) { | |
3019 | /* must not push array size beyond rdev_sectors */ | |
3020 | if (new_offset - rdev->data_offset | |
3021 | + mddev->dev_sectors > rdev->sectors) | |
3022 | return -E2BIG; | |
3023 | } | |
3024 | /* Metadata worries about other space details. */ | |
3025 | ||
3026 | /* decreasing the offset is inconsistent with a backwards | |
3027 | * reshape. | |
3028 | */ | |
3029 | if (new_offset < rdev->data_offset && | |
3030 | mddev->reshape_backwards) | |
3031 | return -EINVAL; | |
3032 | /* Increasing offset is inconsistent with forwards | |
3033 | * reshape. reshape_direction should be set to | |
3034 | * 'backwards' first. | |
3035 | */ | |
3036 | if (new_offset > rdev->data_offset && | |
3037 | !mddev->reshape_backwards) | |
3038 | return -EINVAL; | |
3039 | ||
3040 | if (mddev->pers && mddev->persistent && | |
3041 | !super_types[mddev->major_version] | |
3042 | .allow_new_offset(rdev, new_offset)) | |
3043 | return -E2BIG; | |
3044 | rdev->new_data_offset = new_offset; | |
3045 | if (new_offset > rdev->data_offset) | |
3046 | mddev->reshape_backwards = 1; | |
3047 | else if (new_offset < rdev->data_offset) | |
3048 | mddev->reshape_backwards = 0; | |
3049 | ||
3050 | return len; | |
3051 | } | |
3052 | static struct rdev_sysfs_entry rdev_new_offset = | |
3053 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
3054 | ||
83303b61 | 3055 | static ssize_t |
3cb03002 | 3056 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 3057 | { |
dd8ac336 | 3058 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
3059 | } |
3060 | ||
c5d79adb N |
3061 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
3062 | { | |
3063 | /* check if two start/length pairs overlap */ | |
3064 | if (s1+l1 <= s2) | |
3065 | return 0; | |
3066 | if (s2+l2 <= s1) | |
3067 | return 0; | |
3068 | return 1; | |
3069 | } | |
3070 | ||
b522adcd DW |
3071 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
3072 | { | |
3073 | unsigned long long blocks; | |
3074 | sector_t new; | |
3075 | ||
b29bebd6 | 3076 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
3077 | return -EINVAL; |
3078 | ||
3079 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
3080 | return -EINVAL; /* sector conversion overflow */ | |
3081 | ||
3082 | new = blocks * 2; | |
3083 | if (new != blocks * 2) | |
3084 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
3085 | ||
3086 | *sectors = new; | |
3087 | return 0; | |
3088 | } | |
3089 | ||
83303b61 | 3090 | static ssize_t |
3cb03002 | 3091 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 3092 | { |
fd01b88c | 3093 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 3094 | sector_t oldsectors = rdev->sectors; |
b522adcd | 3095 | sector_t sectors; |
27c529bb | 3096 | |
f2076e7d SL |
3097 | if (test_bit(Journal, &rdev->flags)) |
3098 | return -EBUSY; | |
b522adcd | 3099 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 3100 | return -EINVAL; |
c6563a8c N |
3101 | if (rdev->data_offset != rdev->new_data_offset) |
3102 | return -EINVAL; /* too confusing */ | |
0cd17fec | 3103 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 3104 | if (my_mddev->persistent) { |
dd8ac336 AN |
3105 | sectors = super_types[my_mddev->major_version]. |
3106 | rdev_size_change(rdev, sectors); | |
3107 | if (!sectors) | |
0cd17fec | 3108 | return -EBUSY; |
dd8ac336 | 3109 | } else if (!sectors) |
77304d2a | 3110 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 3111 | rdev->data_offset; |
a6468539 N |
3112 | if (!my_mddev->pers->resize) |
3113 | /* Cannot change size for RAID0 or Linear etc */ | |
3114 | return -EINVAL; | |
0cd17fec | 3115 | } |
dd8ac336 | 3116 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 3117 | return -EINVAL; /* component must fit device */ |
0cd17fec | 3118 | |
dd8ac336 AN |
3119 | rdev->sectors = sectors; |
3120 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
3121 | /* Need to check that all other rdevs with the same |
3122 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
3123 | * the rdev lists safely. | |
3124 | * This check does not provide a hard guarantee, it | |
3125 | * just helps avoid dangerous mistakes. | |
c5d79adb | 3126 | */ |
fd01b88c | 3127 | struct mddev *mddev; |
c5d79adb | 3128 | int overlap = 0; |
159ec1fc | 3129 | struct list_head *tmp; |
c5d79adb | 3130 | |
8b1afc3d | 3131 | rcu_read_lock(); |
29ac4aa3 | 3132 | for_each_mddev(mddev, tmp) { |
3cb03002 | 3133 | struct md_rdev *rdev2; |
c5d79adb | 3134 | |
dafb20fa | 3135 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
3136 | if (rdev->bdev == rdev2->bdev && |
3137 | rdev != rdev2 && | |
3138 | overlaps(rdev->data_offset, rdev->sectors, | |
3139 | rdev2->data_offset, | |
3140 | rdev2->sectors)) { | |
c5d79adb N |
3141 | overlap = 1; |
3142 | break; | |
3143 | } | |
c5d79adb N |
3144 | if (overlap) { |
3145 | mddev_put(mddev); | |
3146 | break; | |
3147 | } | |
3148 | } | |
8b1afc3d | 3149 | rcu_read_unlock(); |
c5d79adb N |
3150 | if (overlap) { |
3151 | /* Someone else could have slipped in a size | |
3152 | * change here, but doing so is just silly. | |
dd8ac336 | 3153 | * We put oldsectors back because we *know* it is |
c5d79adb N |
3154 | * safe, and trust userspace not to race with |
3155 | * itself | |
3156 | */ | |
dd8ac336 | 3157 | rdev->sectors = oldsectors; |
c5d79adb N |
3158 | return -EBUSY; |
3159 | } | |
3160 | } | |
83303b61 N |
3161 | return len; |
3162 | } | |
3163 | ||
3164 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 3165 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 3166 | |
3cb03002 | 3167 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
3168 | { |
3169 | unsigned long long recovery_start = rdev->recovery_offset; | |
3170 | ||
3171 | if (test_bit(In_sync, &rdev->flags) || | |
3172 | recovery_start == MaxSector) | |
3173 | return sprintf(page, "none\n"); | |
3174 | ||
3175 | return sprintf(page, "%llu\n", recovery_start); | |
3176 | } | |
3177 | ||
3cb03002 | 3178 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
3179 | { |
3180 | unsigned long long recovery_start; | |
3181 | ||
3182 | if (cmd_match(buf, "none")) | |
3183 | recovery_start = MaxSector; | |
b29bebd6 | 3184 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
3185 | return -EINVAL; |
3186 | ||
3187 | if (rdev->mddev->pers && | |
3188 | rdev->raid_disk >= 0) | |
3189 | return -EBUSY; | |
3190 | ||
3191 | rdev->recovery_offset = recovery_start; | |
3192 | if (recovery_start == MaxSector) | |
3193 | set_bit(In_sync, &rdev->flags); | |
3194 | else | |
3195 | clear_bit(In_sync, &rdev->flags); | |
3196 | return len; | |
3197 | } | |
3198 | ||
3199 | static struct rdev_sysfs_entry rdev_recovery_start = | |
3200 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
3201 | ||
fc974ee2 VV |
3202 | /* sysfs access to bad-blocks list. |
3203 | * We present two files. | |
3204 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
3205 | * are recorded as bad. The list is truncated to fit within | |
3206 | * the one-page limit of sysfs. | |
3207 | * Writing "sector length" to this file adds an acknowledged | |
3208 | * bad block list. | |
3209 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
3210 | * been acknowledged. Writing to this file adds bad blocks | |
3211 | * without acknowledging them. This is largely for testing. | |
3212 | */ | |
3cb03002 | 3213 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3214 | { |
3215 | return badblocks_show(&rdev->badblocks, page, 0); | |
3216 | } | |
3cb03002 | 3217 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 3218 | { |
de393cde N |
3219 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
3220 | /* Maybe that ack was all we needed */ | |
3221 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
3222 | wake_up(&rdev->blocked_wait); | |
3223 | return rv; | |
16c791a5 N |
3224 | } |
3225 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
3226 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
3227 | ||
3cb03002 | 3228 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3229 | { |
3230 | return badblocks_show(&rdev->badblocks, page, 1); | |
3231 | } | |
3cb03002 | 3232 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
3233 | { |
3234 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
3235 | } | |
3236 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
3237 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
3238 | ||
664aed04 AP |
3239 | static ssize_t |
3240 | ppl_sector_show(struct md_rdev *rdev, char *page) | |
3241 | { | |
3242 | return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector); | |
3243 | } | |
3244 | ||
3245 | static ssize_t | |
3246 | ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len) | |
3247 | { | |
3248 | unsigned long long sector; | |
3249 | ||
3250 | if (kstrtoull(buf, 10, §or) < 0) | |
3251 | return -EINVAL; | |
3252 | if (sector != (sector_t)sector) | |
3253 | return -EINVAL; | |
3254 | ||
3255 | if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) && | |
3256 | rdev->raid_disk >= 0) | |
3257 | return -EBUSY; | |
3258 | ||
3259 | if (rdev->mddev->persistent) { | |
3260 | if (rdev->mddev->major_version == 0) | |
3261 | return -EINVAL; | |
3262 | if ((sector > rdev->sb_start && | |
3263 | sector - rdev->sb_start > S16_MAX) || | |
3264 | (sector < rdev->sb_start && | |
3265 | rdev->sb_start - sector > -S16_MIN)) | |
3266 | return -EINVAL; | |
3267 | rdev->ppl.offset = sector - rdev->sb_start; | |
3268 | } else if (!rdev->mddev->external) { | |
3269 | return -EBUSY; | |
3270 | } | |
3271 | rdev->ppl.sector = sector; | |
3272 | return len; | |
3273 | } | |
3274 | ||
3275 | static struct rdev_sysfs_entry rdev_ppl_sector = | |
3276 | __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store); | |
3277 | ||
3278 | static ssize_t | |
3279 | ppl_size_show(struct md_rdev *rdev, char *page) | |
3280 | { | |
3281 | return sprintf(page, "%u\n", rdev->ppl.size); | |
3282 | } | |
3283 | ||
3284 | static ssize_t | |
3285 | ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len) | |
3286 | { | |
3287 | unsigned int size; | |
3288 | ||
3289 | if (kstrtouint(buf, 10, &size) < 0) | |
3290 | return -EINVAL; | |
3291 | ||
3292 | if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) && | |
3293 | rdev->raid_disk >= 0) | |
3294 | return -EBUSY; | |
3295 | ||
3296 | if (rdev->mddev->persistent) { | |
3297 | if (rdev->mddev->major_version == 0) | |
3298 | return -EINVAL; | |
3299 | if (size > U16_MAX) | |
3300 | return -EINVAL; | |
3301 | } else if (!rdev->mddev->external) { | |
3302 | return -EBUSY; | |
3303 | } | |
3304 | rdev->ppl.size = size; | |
3305 | return len; | |
3306 | } | |
3307 | ||
3308 | static struct rdev_sysfs_entry rdev_ppl_size = | |
3309 | __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store); | |
3310 | ||
86e6ffdd N |
3311 | static struct attribute *rdev_default_attrs[] = { |
3312 | &rdev_state.attr, | |
4dbcdc75 | 3313 | &rdev_errors.attr, |
014236d2 | 3314 | &rdev_slot.attr, |
93c8cad0 | 3315 | &rdev_offset.attr, |
c6563a8c | 3316 | &rdev_new_offset.attr, |
83303b61 | 3317 | &rdev_size.attr, |
06e3c817 | 3318 | &rdev_recovery_start.attr, |
16c791a5 N |
3319 | &rdev_bad_blocks.attr, |
3320 | &rdev_unack_bad_blocks.attr, | |
664aed04 AP |
3321 | &rdev_ppl_sector.attr, |
3322 | &rdev_ppl_size.attr, | |
86e6ffdd N |
3323 | NULL, |
3324 | }; | |
3325 | static ssize_t | |
3326 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3327 | { | |
3328 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3329 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
3330 | |
3331 | if (!entry->show) | |
3332 | return -EIO; | |
758bfc8a N |
3333 | if (!rdev->mddev) |
3334 | return -EBUSY; | |
3335 | return entry->show(rdev, page); | |
86e6ffdd N |
3336 | } |
3337 | ||
3338 | static ssize_t | |
3339 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3340 | const char *page, size_t length) | |
3341 | { | |
3342 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3343 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3344 | ssize_t rv; |
fd01b88c | 3345 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3346 | |
3347 | if (!entry->store) | |
3348 | return -EIO; | |
67463acb N |
3349 | if (!capable(CAP_SYS_ADMIN)) |
3350 | return -EACCES; | |
27c529bb | 3351 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3352 | if (!rv) { |
27c529bb N |
3353 | if (rdev->mddev == NULL) |
3354 | rv = -EBUSY; | |
3355 | else | |
3356 | rv = entry->store(rdev, page, length); | |
6a51830e | 3357 | mddev_unlock(mddev); |
ca388059 N |
3358 | } |
3359 | return rv; | |
86e6ffdd N |
3360 | } |
3361 | ||
3362 | static void rdev_free(struct kobject *ko) | |
3363 | { | |
3cb03002 | 3364 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3365 | kfree(rdev); |
3366 | } | |
52cf25d0 | 3367 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3368 | .show = rdev_attr_show, |
3369 | .store = rdev_attr_store, | |
3370 | }; | |
3371 | static struct kobj_type rdev_ktype = { | |
3372 | .release = rdev_free, | |
3373 | .sysfs_ops = &rdev_sysfs_ops, | |
3374 | .default_attrs = rdev_default_attrs, | |
3375 | }; | |
3376 | ||
3cb03002 | 3377 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3378 | { |
3379 | rdev->desc_nr = -1; | |
3380 | rdev->saved_raid_disk = -1; | |
3381 | rdev->raid_disk = -1; | |
3382 | rdev->flags = 0; | |
3383 | rdev->data_offset = 0; | |
c6563a8c | 3384 | rdev->new_data_offset = 0; |
e8bb9a83 | 3385 | rdev->sb_events = 0; |
0e3ef49e | 3386 | rdev->last_read_error = 0; |
2699b672 N |
3387 | rdev->sb_loaded = 0; |
3388 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3389 | atomic_set(&rdev->nr_pending, 0); |
3390 | atomic_set(&rdev->read_errors, 0); | |
3391 | atomic_set(&rdev->corrected_errors, 0); | |
3392 | ||
3393 | INIT_LIST_HEAD(&rdev->same_set); | |
3394 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3395 | |
3396 | /* Add space to store bad block list. | |
3397 | * This reserves the space even on arrays where it cannot | |
3398 | * be used - I wonder if that matters | |
3399 | */ | |
fc974ee2 | 3400 | return badblocks_init(&rdev->badblocks, 0); |
e8bb9a83 N |
3401 | } |
3402 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3403 | /* |
3404 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3405 | * | |
3406 | * mark the device faulty if: | |
3407 | * | |
3408 | * - the device is nonexistent (zero size) | |
3409 | * - the device has no valid superblock | |
3410 | * | |
3411 | * a faulty rdev _never_ has rdev->sb set. | |
3412 | */ | |
3cb03002 | 3413 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3414 | { |
3415 | char b[BDEVNAME_SIZE]; | |
3416 | int err; | |
3cb03002 | 3417 | struct md_rdev *rdev; |
1da177e4 LT |
3418 | sector_t size; |
3419 | ||
9ffae0cf | 3420 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
9d48739e | 3421 | if (!rdev) |
1da177e4 | 3422 | return ERR_PTR(-ENOMEM); |
1da177e4 | 3423 | |
2230dfe4 N |
3424 | err = md_rdev_init(rdev); |
3425 | if (err) | |
3426 | goto abort_free; | |
3427 | err = alloc_disk_sb(rdev); | |
3428 | if (err) | |
1da177e4 LT |
3429 | goto abort_free; |
3430 | ||
c5d79adb | 3431 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3432 | if (err) |
3433 | goto abort_free; | |
3434 | ||
f9cb074b | 3435 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3436 | |
77304d2a | 3437 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3438 | if (!size) { |
9d48739e | 3439 | pr_warn("md: %s has zero or unknown size, marking faulty!\n", |
1da177e4 LT |
3440 | bdevname(rdev->bdev,b)); |
3441 | err = -EINVAL; | |
3442 | goto abort_free; | |
3443 | } | |
3444 | ||
3445 | if (super_format >= 0) { | |
3446 | err = super_types[super_format]. | |
3447 | load_super(rdev, NULL, super_minor); | |
3448 | if (err == -EINVAL) { | |
9d48739e | 3449 | pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n", |
df968c4e | 3450 | bdevname(rdev->bdev,b), |
9d48739e | 3451 | super_format, super_minor); |
1da177e4 LT |
3452 | goto abort_free; |
3453 | } | |
3454 | if (err < 0) { | |
9d48739e | 3455 | pr_warn("md: could not read %s's sb, not importing!\n", |
1da177e4 LT |
3456 | bdevname(rdev->bdev,b)); |
3457 | goto abort_free; | |
3458 | } | |
3459 | } | |
6bfe0b49 | 3460 | |
1da177e4 LT |
3461 | return rdev; |
3462 | ||
3463 | abort_free: | |
2699b672 N |
3464 | if (rdev->bdev) |
3465 | unlock_rdev(rdev); | |
545c8795 | 3466 | md_rdev_clear(rdev); |
1da177e4 LT |
3467 | kfree(rdev); |
3468 | return ERR_PTR(err); | |
3469 | } | |
3470 | ||
3471 | /* | |
3472 | * Check a full RAID array for plausibility | |
3473 | */ | |
3474 | ||
f72ffdd6 | 3475 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3476 | { |
3477 | int i; | |
3cb03002 | 3478 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3479 | char b[BDEVNAME_SIZE]; |
3480 | ||
3481 | freshest = NULL; | |
dafb20fa | 3482 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3483 | switch (super_types[mddev->major_version]. |
3484 | load_super(rdev, freshest, mddev->minor_version)) { | |
3485 | case 1: | |
3486 | freshest = rdev; | |
3487 | break; | |
3488 | case 0: | |
3489 | break; | |
3490 | default: | |
9d48739e | 3491 | pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n", |
1da177e4 | 3492 | bdevname(rdev->bdev,b)); |
fb56dfef | 3493 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3494 | } |
3495 | ||
1da177e4 LT |
3496 | super_types[mddev->major_version]. |
3497 | validate_super(mddev, freshest); | |
3498 | ||
3499 | i = 0; | |
dafb20fa | 3500 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3501 | if (mddev->max_disks && |
3502 | (rdev->desc_nr >= mddev->max_disks || | |
3503 | i > mddev->max_disks)) { | |
9d48739e N |
3504 | pr_warn("md: %s: %s: only %d devices permitted\n", |
3505 | mdname(mddev), bdevname(rdev->bdev, b), | |
3506 | mddev->max_disks); | |
fb56dfef | 3507 | md_kick_rdev_from_array(rdev); |
de01dfad N |
3508 | continue; |
3509 | } | |
1aee41f6 | 3510 | if (rdev != freshest) { |
1da177e4 LT |
3511 | if (super_types[mddev->major_version]. |
3512 | validate_super(mddev, rdev)) { | |
9d48739e | 3513 | pr_warn("md: kicking non-fresh %s from array!\n", |
1da177e4 | 3514 | bdevname(rdev->bdev,b)); |
fb56dfef | 3515 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3516 | continue; |
3517 | } | |
1aee41f6 | 3518 | } |
1da177e4 LT |
3519 | if (mddev->level == LEVEL_MULTIPATH) { |
3520 | rdev->desc_nr = i++; | |
3521 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3522 | set_bit(In_sync, &rdev->flags); |
f2076e7d SL |
3523 | } else if (rdev->raid_disk >= |
3524 | (mddev->raid_disks - min(0, mddev->delta_disks)) && | |
3525 | !test_bit(Journal, &rdev->flags)) { | |
a778b73f N |
3526 | rdev->raid_disk = -1; |
3527 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3528 | } |
3529 | } | |
1da177e4 LT |
3530 | } |
3531 | ||
72e02075 N |
3532 | /* Read a fixed-point number. |
3533 | * Numbers in sysfs attributes should be in "standard" units where | |
3534 | * possible, so time should be in seconds. | |
f72ffdd6 | 3535 | * However we internally use a a much smaller unit such as |
72e02075 N |
3536 | * milliseconds or jiffies. |
3537 | * This function takes a decimal number with a possible fractional | |
3538 | * component, and produces an integer which is the result of | |
3539 | * multiplying that number by 10^'scale'. | |
3540 | * all without any floating-point arithmetic. | |
3541 | */ | |
3542 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3543 | { | |
3544 | unsigned long result = 0; | |
3545 | long decimals = -1; | |
3546 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3547 | if (*cp == '.') | |
3548 | decimals = 0; | |
3549 | else if (decimals < scale) { | |
3550 | unsigned int value; | |
3551 | value = *cp - '0'; | |
3552 | result = result * 10 + value; | |
3553 | if (decimals >= 0) | |
3554 | decimals++; | |
3555 | } | |
3556 | cp++; | |
3557 | } | |
3558 | if (*cp == '\n') | |
3559 | cp++; | |
3560 | if (*cp) | |
3561 | return -EINVAL; | |
3562 | if (decimals < 0) | |
3563 | decimals = 0; | |
3564 | while (decimals < scale) { | |
3565 | result *= 10; | |
3566 | decimals ++; | |
3567 | } | |
3568 | *res = result; | |
3569 | return 0; | |
3570 | } | |
3571 | ||
16f17b39 | 3572 | static ssize_t |
fd01b88c | 3573 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3574 | { |
3575 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3576 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3577 | } | |
3578 | static ssize_t | |
fd01b88c | 3579 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3580 | { |
16f17b39 | 3581 | unsigned long msec; |
97ce0a7f | 3582 | |
28c1b9fd | 3583 | if (mddev_is_clustered(mddev)) { |
9d48739e | 3584 | pr_warn("md: Safemode is disabled for clustered mode\n"); |
28c1b9fd GR |
3585 | return -EINVAL; |
3586 | } | |
3587 | ||
72e02075 | 3588 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3589 | return -EINVAL; |
16f17b39 N |
3590 | if (msec == 0) |
3591 | mddev->safemode_delay = 0; | |
3592 | else { | |
19052c0e | 3593 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3594 | unsigned long new_delay = (msec*HZ)/1000; |
3595 | ||
3596 | if (new_delay == 0) | |
3597 | new_delay = 1; | |
3598 | mddev->safemode_delay = new_delay; | |
3599 | if (new_delay < old_delay || old_delay == 0) | |
3600 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3601 | } |
3602 | return len; | |
3603 | } | |
3604 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3605 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3606 | |
eae1701f | 3607 | static ssize_t |
fd01b88c | 3608 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3609 | { |
36d091f4 N |
3610 | struct md_personality *p; |
3611 | int ret; | |
3612 | spin_lock(&mddev->lock); | |
3613 | p = mddev->pers; | |
d9d166c2 | 3614 | if (p) |
36d091f4 | 3615 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3616 | else if (mddev->clevel[0]) |
36d091f4 | 3617 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3618 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3619 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3620 | else |
36d091f4 N |
3621 | ret = 0; |
3622 | spin_unlock(&mddev->lock); | |
3623 | return ret; | |
eae1701f N |
3624 | } |
3625 | ||
d9d166c2 | 3626 | static ssize_t |
fd01b88c | 3627 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3628 | { |
f2859af6 | 3629 | char clevel[16]; |
6791875e N |
3630 | ssize_t rv; |
3631 | size_t slen = len; | |
db721d32 | 3632 | struct md_personality *pers, *oldpers; |
f2859af6 | 3633 | long level; |
db721d32 | 3634 | void *priv, *oldpriv; |
3cb03002 | 3635 | struct md_rdev *rdev; |
245f46c2 | 3636 | |
6791875e N |
3637 | if (slen == 0 || slen >= sizeof(clevel)) |
3638 | return -EINVAL; | |
3639 | ||
3640 | rv = mddev_lock(mddev); | |
3641 | if (rv) | |
3642 | return rv; | |
3643 | ||
245f46c2 | 3644 | if (mddev->pers == NULL) { |
6791875e N |
3645 | strncpy(mddev->clevel, buf, slen); |
3646 | if (mddev->clevel[slen-1] == '\n') | |
3647 | slen--; | |
3648 | mddev->clevel[slen] = 0; | |
245f46c2 | 3649 | mddev->level = LEVEL_NONE; |
6791875e N |
3650 | rv = len; |
3651 | goto out_unlock; | |
245f46c2 | 3652 | } |
6791875e | 3653 | rv = -EROFS; |
bd8839e0 | 3654 | if (mddev->ro) |
6791875e | 3655 | goto out_unlock; |
245f46c2 N |
3656 | |
3657 | /* request to change the personality. Need to ensure: | |
3658 | * - array is not engaged in resync/recovery/reshape | |
3659 | * - old personality can be suspended | |
3660 | * - new personality will access other array. | |
3661 | */ | |
3662 | ||
6791875e | 3663 | rv = -EBUSY; |
bb4f1e9d | 3664 | if (mddev->sync_thread || |
f851b60d | 3665 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3666 | mddev->reshape_position != MaxSector || |
3667 | mddev->sysfs_active) | |
6791875e | 3668 | goto out_unlock; |
245f46c2 | 3669 | |
6791875e | 3670 | rv = -EINVAL; |
245f46c2 | 3671 | if (!mddev->pers->quiesce) { |
9d48739e N |
3672 | pr_warn("md: %s: %s does not support online personality change\n", |
3673 | mdname(mddev), mddev->pers->name); | |
6791875e | 3674 | goto out_unlock; |
245f46c2 N |
3675 | } |
3676 | ||
3677 | /* Now find the new personality */ | |
6791875e N |
3678 | strncpy(clevel, buf, slen); |
3679 | if (clevel[slen-1] == '\n') | |
3680 | slen--; | |
3681 | clevel[slen] = 0; | |
b29bebd6 | 3682 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3683 | level = LEVEL_NONE; |
245f46c2 | 3684 | |
f2859af6 DW |
3685 | if (request_module("md-%s", clevel) != 0) |
3686 | request_module("md-level-%s", clevel); | |
245f46c2 | 3687 | spin_lock(&pers_lock); |
f2859af6 | 3688 | pers = find_pers(level, clevel); |
245f46c2 N |
3689 | if (!pers || !try_module_get(pers->owner)) { |
3690 | spin_unlock(&pers_lock); | |
9d48739e | 3691 | pr_warn("md: personality %s not loaded\n", clevel); |
6791875e N |
3692 | rv = -EINVAL; |
3693 | goto out_unlock; | |
245f46c2 N |
3694 | } |
3695 | spin_unlock(&pers_lock); | |
3696 | ||
3697 | if (pers == mddev->pers) { | |
3698 | /* Nothing to do! */ | |
3699 | module_put(pers->owner); | |
6791875e N |
3700 | rv = len; |
3701 | goto out_unlock; | |
245f46c2 N |
3702 | } |
3703 | if (!pers->takeover) { | |
3704 | module_put(pers->owner); | |
9d48739e N |
3705 | pr_warn("md: %s: %s does not support personality takeover\n", |
3706 | mdname(mddev), clevel); | |
6791875e N |
3707 | rv = -EINVAL; |
3708 | goto out_unlock; | |
245f46c2 N |
3709 | } |
3710 | ||
dafb20fa | 3711 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3712 | rdev->new_raid_disk = rdev->raid_disk; |
3713 | ||
245f46c2 N |
3714 | /* ->takeover must set new_* and/or delta_disks |
3715 | * if it succeeds, and may set them when it fails. | |
3716 | */ | |
3717 | priv = pers->takeover(mddev); | |
3718 | if (IS_ERR(priv)) { | |
3719 | mddev->new_level = mddev->level; | |
3720 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3721 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3722 | mddev->raid_disks -= mddev->delta_disks; |
3723 | mddev->delta_disks = 0; | |
2c810cdd | 3724 | mddev->reshape_backwards = 0; |
245f46c2 | 3725 | module_put(pers->owner); |
9d48739e N |
3726 | pr_warn("md: %s: %s would not accept array\n", |
3727 | mdname(mddev), clevel); | |
6791875e N |
3728 | rv = PTR_ERR(priv); |
3729 | goto out_unlock; | |
245f46c2 N |
3730 | } |
3731 | ||
3732 | /* Looks like we have a winner */ | |
3733 | mddev_suspend(mddev); | |
5aa61f42 | 3734 | mddev_detach(mddev); |
36d091f4 N |
3735 | |
3736 | spin_lock(&mddev->lock); | |
db721d32 N |
3737 | oldpers = mddev->pers; |
3738 | oldpriv = mddev->private; | |
3739 | mddev->pers = pers; | |
3740 | mddev->private = priv; | |
3741 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3742 | mddev->level = mddev->new_level; | |
3743 | mddev->layout = mddev->new_layout; | |
3744 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3745 | mddev->delta_disks = 0; | |
3746 | mddev->reshape_backwards = 0; | |
3747 | mddev->degraded = 0; | |
36d091f4 | 3748 | spin_unlock(&mddev->lock); |
db721d32 N |
3749 | |
3750 | if (oldpers->sync_request == NULL && | |
3751 | mddev->external) { | |
3752 | /* We are converting from a no-redundancy array | |
3753 | * to a redundancy array and metadata is managed | |
3754 | * externally so we need to be sure that writes | |
3755 | * won't block due to a need to transition | |
3756 | * clean->dirty | |
3757 | * until external management is started. | |
3758 | */ | |
3759 | mddev->in_sync = 0; | |
3760 | mddev->safemode_delay = 0; | |
3761 | mddev->safemode = 0; | |
3762 | } | |
f72ffdd6 | 3763 | |
db721d32 N |
3764 | oldpers->free(mddev, oldpriv); |
3765 | ||
3766 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3767 | pers->sync_request != NULL) { |
3768 | /* need to add the md_redundancy_group */ | |
3769 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
3770 | pr_warn("md: cannot register extra attributes for %s\n", |
3771 | mdname(mddev)); | |
388975cc | 3772 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3773 | } |
db721d32 | 3774 | if (oldpers->sync_request != NULL && |
a64c876f N |
3775 | pers->sync_request == NULL) { |
3776 | /* need to remove the md_redundancy_group */ | |
3777 | if (mddev->to_remove == NULL) | |
3778 | mddev->to_remove = &md_redundancy_group; | |
3779 | } | |
3780 | ||
4cb9da7d AO |
3781 | module_put(oldpers->owner); |
3782 | ||
dafb20fa | 3783 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3784 | if (rdev->raid_disk < 0) |
3785 | continue; | |
bf2cb0da | 3786 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3787 | rdev->new_raid_disk = -1; |
3788 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3789 | continue; | |
36fad858 | 3790 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3791 | } |
dafb20fa | 3792 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3793 | if (rdev->raid_disk < 0) |
3794 | continue; | |
3795 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3796 | continue; | |
3797 | rdev->raid_disk = rdev->new_raid_disk; | |
3798 | if (rdev->raid_disk < 0) | |
3a981b03 | 3799 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3800 | else { |
36fad858 | 3801 | if (sysfs_link_rdev(mddev, rdev)) |
9d48739e N |
3802 | pr_warn("md: cannot register rd%d for %s after level change\n", |
3803 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3804 | } |
e93f68a1 N |
3805 | } |
3806 | ||
db721d32 | 3807 | if (pers->sync_request == NULL) { |
9af204cf TM |
3808 | /* this is now an array without redundancy, so |
3809 | * it must always be in_sync | |
3810 | */ | |
3811 | mddev->in_sync = 1; | |
3812 | del_timer_sync(&mddev->safemode_timer); | |
3813 | } | |
02e5f5c0 | 3814 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3815 | pers->run(mddev); |
2953079c | 3816 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
47525e59 | 3817 | mddev_resume(mddev); |
830778a1 N |
3818 | if (!mddev->thread) |
3819 | md_update_sb(mddev, 1); | |
5cac7861 | 3820 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3821 | md_new_event(mddev); |
6791875e N |
3822 | rv = len; |
3823 | out_unlock: | |
3824 | mddev_unlock(mddev); | |
d9d166c2 N |
3825 | return rv; |
3826 | } | |
3827 | ||
3828 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3829 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3830 | |
d4dbd025 | 3831 | static ssize_t |
fd01b88c | 3832 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3833 | { |
3834 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3835 | if (mddev->reshape_position != MaxSector && |
3836 | mddev->layout != mddev->new_layout) | |
3837 | return sprintf(page, "%d (%d)\n", | |
3838 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3839 | return sprintf(page, "%d\n", mddev->layout); |
3840 | } | |
3841 | ||
3842 | static ssize_t | |
fd01b88c | 3843 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 | 3844 | { |
4c9309c0 | 3845 | unsigned int n; |
6791875e | 3846 | int err; |
d4dbd025 | 3847 | |
4c9309c0 AD |
3848 | err = kstrtouint(buf, 10, &n); |
3849 | if (err < 0) | |
3850 | return err; | |
6791875e N |
3851 | err = mddev_lock(mddev); |
3852 | if (err) | |
3853 | return err; | |
d4dbd025 | 3854 | |
b3546035 | 3855 | if (mddev->pers) { |
50ac168a | 3856 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3857 | err = -EBUSY; |
3858 | else if (mddev->ro) | |
3859 | err = -EROFS; | |
3860 | else { | |
3861 | mddev->new_layout = n; | |
3862 | err = mddev->pers->check_reshape(mddev); | |
3863 | if (err) | |
3864 | mddev->new_layout = mddev->layout; | |
597a711b | 3865 | } |
b3546035 | 3866 | } else { |
08a02ecd | 3867 | mddev->new_layout = n; |
b3546035 N |
3868 | if (mddev->reshape_position == MaxSector) |
3869 | mddev->layout = n; | |
3870 | } | |
6791875e N |
3871 | mddev_unlock(mddev); |
3872 | return err ?: len; | |
d4dbd025 N |
3873 | } |
3874 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3875 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3876 | |
eae1701f | 3877 | static ssize_t |
fd01b88c | 3878 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3879 | { |
bb636547 N |
3880 | if (mddev->raid_disks == 0) |
3881 | return 0; | |
08a02ecd N |
3882 | if (mddev->reshape_position != MaxSector && |
3883 | mddev->delta_disks != 0) | |
3884 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3885 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3886 | return sprintf(page, "%d\n", mddev->raid_disks); |
3887 | } | |
3888 | ||
fd01b88c | 3889 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3890 | |
3891 | static ssize_t | |
fd01b88c | 3892 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3893 | { |
4c9309c0 | 3894 | unsigned int n; |
6791875e | 3895 | int err; |
da943b99 | 3896 | |
4c9309c0 AD |
3897 | err = kstrtouint(buf, 10, &n); |
3898 | if (err < 0) | |
3899 | return err; | |
da943b99 | 3900 | |
6791875e N |
3901 | err = mddev_lock(mddev); |
3902 | if (err) | |
3903 | return err; | |
da943b99 | 3904 | if (mddev->pers) |
6791875e | 3905 | err = update_raid_disks(mddev, n); |
08a02ecd | 3906 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3907 | struct md_rdev *rdev; |
08a02ecd | 3908 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3909 | |
6791875e | 3910 | err = -EINVAL; |
c6563a8c N |
3911 | rdev_for_each(rdev, mddev) { |
3912 | if (olddisks < n && | |
3913 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3914 | goto out_unlock; |
c6563a8c N |
3915 | if (olddisks > n && |
3916 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3917 | goto out_unlock; |
c6563a8c | 3918 | } |
6791875e | 3919 | err = 0; |
08a02ecd N |
3920 | mddev->delta_disks = n - olddisks; |
3921 | mddev->raid_disks = n; | |
2c810cdd | 3922 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3923 | } else |
da943b99 | 3924 | mddev->raid_disks = n; |
6791875e N |
3925 | out_unlock: |
3926 | mddev_unlock(mddev); | |
3927 | return err ? err : len; | |
da943b99 N |
3928 | } |
3929 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3930 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3931 | |
3b34380a | 3932 | static ssize_t |
fd01b88c | 3933 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3934 | { |
08a02ecd | 3935 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3936 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3937 | return sprintf(page, "%d (%d)\n", | |
3938 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3939 | mddev->chunk_sectors << 9); |
3940 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3941 | } |
3942 | ||
3943 | static ssize_t | |
fd01b88c | 3944 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3945 | { |
4c9309c0 | 3946 | unsigned long n; |
6791875e | 3947 | int err; |
3b34380a | 3948 | |
4c9309c0 AD |
3949 | err = kstrtoul(buf, 10, &n); |
3950 | if (err < 0) | |
3951 | return err; | |
3b34380a | 3952 | |
6791875e N |
3953 | err = mddev_lock(mddev); |
3954 | if (err) | |
3955 | return err; | |
b3546035 | 3956 | if (mddev->pers) { |
50ac168a | 3957 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3958 | err = -EBUSY; |
3959 | else if (mddev->ro) | |
3960 | err = -EROFS; | |
3961 | else { | |
3962 | mddev->new_chunk_sectors = n >> 9; | |
3963 | err = mddev->pers->check_reshape(mddev); | |
3964 | if (err) | |
3965 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 3966 | } |
b3546035 | 3967 | } else { |
664e7c41 | 3968 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3969 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3970 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3971 | } |
6791875e N |
3972 | mddev_unlock(mddev); |
3973 | return err ?: len; | |
3b34380a N |
3974 | } |
3975 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3976 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3977 | |
a94213b1 | 3978 | static ssize_t |
fd01b88c | 3979 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3980 | { |
d1a7c503 N |
3981 | if (mddev->recovery_cp == MaxSector) |
3982 | return sprintf(page, "none\n"); | |
a94213b1 N |
3983 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3984 | } | |
3985 | ||
3986 | static ssize_t | |
fd01b88c | 3987 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3988 | { |
4c9309c0 | 3989 | unsigned long long n; |
6791875e | 3990 | int err; |
4c9309c0 AD |
3991 | |
3992 | if (cmd_match(buf, "none")) | |
3993 | n = MaxSector; | |
3994 | else { | |
3995 | err = kstrtoull(buf, 10, &n); | |
3996 | if (err < 0) | |
3997 | return err; | |
3998 | if (n != (sector_t)n) | |
3999 | return -EINVAL; | |
4000 | } | |
a94213b1 | 4001 | |
6791875e N |
4002 | err = mddev_lock(mddev); |
4003 | if (err) | |
4004 | return err; | |
b098636c | 4005 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e | 4006 | err = -EBUSY; |
a94213b1 | 4007 | |
6791875e N |
4008 | if (!err) { |
4009 | mddev->recovery_cp = n; | |
4010 | if (mddev->pers) | |
2953079c | 4011 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
6791875e N |
4012 | } |
4013 | mddev_unlock(mddev); | |
4014 | return err ?: len; | |
a94213b1 N |
4015 | } |
4016 | static struct md_sysfs_entry md_resync_start = | |
750f199e N |
4017 | __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR, |
4018 | resync_start_show, resync_start_store); | |
a94213b1 | 4019 | |
9e653b63 N |
4020 | /* |
4021 | * The array state can be: | |
4022 | * | |
4023 | * clear | |
4024 | * No devices, no size, no level | |
4025 | * Equivalent to STOP_ARRAY ioctl | |
4026 | * inactive | |
4027 | * May have some settings, but array is not active | |
4028 | * all IO results in error | |
4029 | * When written, doesn't tear down array, but just stops it | |
4030 | * suspended (not supported yet) | |
4031 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 4032 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
4033 | * readonly |
4034 | * no resync can happen. no superblocks get written. | |
4035 | * write requests fail | |
4036 | * read-auto | |
4037 | * like readonly, but behaves like 'clean' on a write request. | |
4038 | * | |
4039 | * clean - no pending writes, but otherwise active. | |
4040 | * When written to inactive array, starts without resync | |
4041 | * If a write request arrives then | |
4042 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
4043 | * if not known, block and switch to write-pending | |
4044 | * If written to an active array that has pending writes, then fails. | |
4045 | * active | |
4046 | * fully active: IO and resync can be happening. | |
4047 | * When written to inactive array, starts with resync | |
4048 | * | |
4049 | * write-pending | |
4050 | * clean, but writes are blocked waiting for 'active' to be written. | |
4051 | * | |
4052 | * active-idle | |
4053 | * like active, but no writes have been seen for a while (100msec). | |
4054 | * | |
4055 | */ | |
4056 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
4057 | write_pending, active_idle, bad_word}; | |
05381954 | 4058 | static char *array_states[] = { |
9e653b63 N |
4059 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
4060 | "write-pending", "active-idle", NULL }; | |
4061 | ||
4062 | static int match_word(const char *word, char **list) | |
4063 | { | |
4064 | int n; | |
4065 | for (n=0; list[n]; n++) | |
4066 | if (cmd_match(word, list[n])) | |
4067 | break; | |
4068 | return n; | |
4069 | } | |
4070 | ||
4071 | static ssize_t | |
fd01b88c | 4072 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
4073 | { |
4074 | enum array_state st = inactive; | |
4075 | ||
4076 | if (mddev->pers) | |
4077 | switch(mddev->ro) { | |
4078 | case 1: | |
4079 | st = readonly; | |
4080 | break; | |
4081 | case 2: | |
4082 | st = read_auto; | |
4083 | break; | |
4084 | case 0: | |
55cc39f3 | 4085 | spin_lock(&mddev->lock); |
2953079c | 4086 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
e691063a | 4087 | st = write_pending; |
16f88949 TM |
4088 | else if (mddev->in_sync) |
4089 | st = clean; | |
9e653b63 N |
4090 | else if (mddev->safemode) |
4091 | st = active_idle; | |
4092 | else | |
4093 | st = active; | |
55cc39f3 | 4094 | spin_unlock(&mddev->lock); |
9e653b63 N |
4095 | } |
4096 | else { | |
4097 | if (list_empty(&mddev->disks) && | |
4098 | mddev->raid_disks == 0 && | |
58c0fed4 | 4099 | mddev->dev_sectors == 0) |
9e653b63 N |
4100 | st = clear; |
4101 | else | |
4102 | st = inactive; | |
4103 | } | |
4104 | return sprintf(page, "%s\n", array_states[st]); | |
4105 | } | |
4106 | ||
f72ffdd6 N |
4107 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
4108 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
4109 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 4110 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
4111 | |
4112 | static ssize_t | |
fd01b88c | 4113 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 4114 | { |
6497709b | 4115 | int err = 0; |
9e653b63 | 4116 | enum array_state st = match_word(buf, array_states); |
6791875e N |
4117 | |
4118 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
4119 | /* don't take reconfig_mutex when toggling between | |
4120 | * clean and active | |
4121 | */ | |
4122 | spin_lock(&mddev->lock); | |
4123 | if (st == active) { | |
4124 | restart_array(mddev); | |
2953079c | 4125 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
91a6c4ad | 4126 | md_wakeup_thread(mddev->thread); |
6791875e | 4127 | wake_up(&mddev->sb_wait); |
6791875e N |
4128 | } else /* st == clean */ { |
4129 | restart_array(mddev); | |
6497709b | 4130 | if (!set_in_sync(mddev)) |
6791875e N |
4131 | err = -EBUSY; |
4132 | } | |
573275b5 TM |
4133 | if (!err) |
4134 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
6791875e | 4135 | spin_unlock(&mddev->lock); |
c008f1d3 | 4136 | return err ?: len; |
6791875e N |
4137 | } |
4138 | err = mddev_lock(mddev); | |
4139 | if (err) | |
4140 | return err; | |
4141 | err = -EINVAL; | |
9e653b63 N |
4142 | switch(st) { |
4143 | case bad_word: | |
4144 | break; | |
4145 | case clear: | |
4146 | /* stopping an active array */ | |
a05b7ea0 | 4147 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
4148 | break; |
4149 | case inactive: | |
4150 | /* stopping an active array */ | |
90cf195d | 4151 | if (mddev->pers) |
a05b7ea0 | 4152 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 4153 | else |
e691063a | 4154 | err = 0; /* already inactive */ |
9e653b63 N |
4155 | break; |
4156 | case suspended: | |
4157 | break; /* not supported yet */ | |
4158 | case readonly: | |
4159 | if (mddev->pers) | |
a05b7ea0 | 4160 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
4161 | else { |
4162 | mddev->ro = 1; | |
648b629e | 4163 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
4164 | err = do_md_run(mddev); |
4165 | } | |
4166 | break; | |
4167 | case read_auto: | |
9e653b63 | 4168 | if (mddev->pers) { |
80268ee9 | 4169 | if (mddev->ro == 0) |
a05b7ea0 | 4170 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 4171 | else if (mddev->ro == 1) |
648b629e N |
4172 | err = restart_array(mddev); |
4173 | if (err == 0) { | |
4174 | mddev->ro = 2; | |
4175 | set_disk_ro(mddev->gendisk, 0); | |
4176 | } | |
9e653b63 N |
4177 | } else { |
4178 | mddev->ro = 2; | |
4179 | err = do_md_run(mddev); | |
4180 | } | |
4181 | break; | |
4182 | case clean: | |
4183 | if (mddev->pers) { | |
339421de SL |
4184 | err = restart_array(mddev); |
4185 | if (err) | |
4186 | break; | |
85572d7c | 4187 | spin_lock(&mddev->lock); |
6497709b | 4188 | if (!set_in_sync(mddev)) |
e691063a | 4189 | err = -EBUSY; |
85572d7c | 4190 | spin_unlock(&mddev->lock); |
5bf29597 N |
4191 | } else |
4192 | err = -EINVAL; | |
9e653b63 N |
4193 | break; |
4194 | case active: | |
4195 | if (mddev->pers) { | |
339421de SL |
4196 | err = restart_array(mddev); |
4197 | if (err) | |
4198 | break; | |
2953079c | 4199 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
9e653b63 N |
4200 | wake_up(&mddev->sb_wait); |
4201 | err = 0; | |
4202 | } else { | |
4203 | mddev->ro = 0; | |
648b629e | 4204 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
4205 | err = do_md_run(mddev); |
4206 | } | |
4207 | break; | |
4208 | case write_pending: | |
4209 | case active_idle: | |
4210 | /* these cannot be set */ | |
4211 | break; | |
4212 | } | |
6791875e N |
4213 | |
4214 | if (!err) { | |
1d23f178 N |
4215 | if (mddev->hold_active == UNTIL_IOCTL) |
4216 | mddev->hold_active = 0; | |
00bcb4ac | 4217 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4218 | } |
6791875e N |
4219 | mddev_unlock(mddev); |
4220 | return err ?: len; | |
9e653b63 | 4221 | } |
80ca3a44 | 4222 | static struct md_sysfs_entry md_array_state = |
750f199e | 4223 | __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); |
9e653b63 | 4224 | |
1e50915f | 4225 | static ssize_t |
fd01b88c | 4226 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
4227 | return sprintf(page, "%d\n", |
4228 | atomic_read(&mddev->max_corr_read_errors)); | |
4229 | } | |
4230 | ||
4231 | static ssize_t | |
fd01b88c | 4232 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f | 4233 | { |
4c9309c0 AD |
4234 | unsigned int n; |
4235 | int rv; | |
1e50915f | 4236 | |
4c9309c0 AD |
4237 | rv = kstrtouint(buf, 10, &n); |
4238 | if (rv < 0) | |
4239 | return rv; | |
4240 | atomic_set(&mddev->max_corr_read_errors, n); | |
4241 | return len; | |
1e50915f RB |
4242 | } |
4243 | ||
4244 | static struct md_sysfs_entry max_corr_read_errors = | |
4245 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
4246 | max_corrected_read_errors_store); | |
4247 | ||
6d7ff738 | 4248 | static ssize_t |
fd01b88c | 4249 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
4250 | { |
4251 | return -EINVAL; | |
4252 | } | |
4253 | ||
4254 | static ssize_t | |
fd01b88c | 4255 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
4256 | { |
4257 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
4258 | /* The new device is added to the array. | |
4259 | * If the array has a persistent superblock, we read the | |
4260 | * superblock to initialise info and check validity. | |
4261 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
4262 | * which mainly checks size. | |
4263 | */ | |
4264 | char *e; | |
4265 | int major = simple_strtoul(buf, &e, 10); | |
4266 | int minor; | |
4267 | dev_t dev; | |
3cb03002 | 4268 | struct md_rdev *rdev; |
6d7ff738 N |
4269 | int err; |
4270 | ||
4271 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
4272 | return -EINVAL; | |
4273 | minor = simple_strtoul(e+1, &e, 10); | |
4274 | if (*e && *e != '\n') | |
4275 | return -EINVAL; | |
4276 | dev = MKDEV(major, minor); | |
4277 | if (major != MAJOR(dev) || | |
4278 | minor != MINOR(dev)) | |
4279 | return -EOVERFLOW; | |
4280 | ||
6791875e N |
4281 | flush_workqueue(md_misc_wq); |
4282 | ||
4283 | err = mddev_lock(mddev); | |
4284 | if (err) | |
4285 | return err; | |
6d7ff738 N |
4286 | if (mddev->persistent) { |
4287 | rdev = md_import_device(dev, mddev->major_version, | |
4288 | mddev->minor_version); | |
4289 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4290 | struct md_rdev *rdev0 |
4291 | = list_entry(mddev->disks.next, | |
4292 | struct md_rdev, same_set); | |
6d7ff738 N |
4293 | err = super_types[mddev->major_version] |
4294 | .load_super(rdev, rdev0, mddev->minor_version); | |
4295 | if (err < 0) | |
4296 | goto out; | |
4297 | } | |
c5d79adb N |
4298 | } else if (mddev->external) |
4299 | rdev = md_import_device(dev, -2, -1); | |
4300 | else | |
6d7ff738 N |
4301 | rdev = md_import_device(dev, -1, -1); |
4302 | ||
9a8c0fa8 N |
4303 | if (IS_ERR(rdev)) { |
4304 | mddev_unlock(mddev); | |
6d7ff738 | 4305 | return PTR_ERR(rdev); |
9a8c0fa8 | 4306 | } |
6d7ff738 N |
4307 | err = bind_rdev_to_array(rdev, mddev); |
4308 | out: | |
4309 | if (err) | |
4310 | export_rdev(rdev); | |
6791875e | 4311 | mddev_unlock(mddev); |
5492c46e AO |
4312 | if (!err) |
4313 | md_new_event(mddev); | |
6d7ff738 N |
4314 | return err ? err : len; |
4315 | } | |
4316 | ||
4317 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4318 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4319 | |
9b1d1dac | 4320 | static ssize_t |
fd01b88c | 4321 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4322 | { |
4323 | char *end; | |
4324 | unsigned long chunk, end_chunk; | |
6791875e | 4325 | int err; |
9b1d1dac | 4326 | |
6791875e N |
4327 | err = mddev_lock(mddev); |
4328 | if (err) | |
4329 | return err; | |
9b1d1dac PC |
4330 | if (!mddev->bitmap) |
4331 | goto out; | |
4332 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4333 | while (*buf) { | |
4334 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4335 | if (buf == end) break; | |
4336 | if (*end == '-') { /* range */ | |
4337 | buf = end + 1; | |
4338 | end_chunk = simple_strtoul(buf, &end, 0); | |
4339 | if (buf == end) break; | |
4340 | } | |
4341 | if (*end && !isspace(*end)) break; | |
4342 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4343 | buf = skip_spaces(end); |
9b1d1dac PC |
4344 | } |
4345 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4346 | out: | |
6791875e | 4347 | mddev_unlock(mddev); |
9b1d1dac PC |
4348 | return len; |
4349 | } | |
4350 | ||
4351 | static struct md_sysfs_entry md_bitmap = | |
4352 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4353 | ||
a35b0d69 | 4354 | static ssize_t |
fd01b88c | 4355 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4356 | { |
58c0fed4 AN |
4357 | return sprintf(page, "%llu\n", |
4358 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4359 | } |
4360 | ||
fd01b88c | 4361 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4362 | |
4363 | static ssize_t | |
fd01b88c | 4364 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4365 | { |
4366 | /* If array is inactive, we can reduce the component size, but | |
4367 | * not increase it (except from 0). | |
4368 | * If array is active, we can try an on-line resize | |
4369 | */ | |
b522adcd DW |
4370 | sector_t sectors; |
4371 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4372 | |
58c0fed4 AN |
4373 | if (err < 0) |
4374 | return err; | |
6791875e N |
4375 | err = mddev_lock(mddev); |
4376 | if (err) | |
4377 | return err; | |
a35b0d69 | 4378 | if (mddev->pers) { |
58c0fed4 | 4379 | err = update_size(mddev, sectors); |
4ba1e788 XN |
4380 | if (err == 0) |
4381 | md_update_sb(mddev, 1); | |
a35b0d69 | 4382 | } else { |
58c0fed4 AN |
4383 | if (mddev->dev_sectors == 0 || |
4384 | mddev->dev_sectors > sectors) | |
4385 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4386 | else |
4387 | err = -ENOSPC; | |
4388 | } | |
6791875e | 4389 | mddev_unlock(mddev); |
a35b0d69 N |
4390 | return err ? err : len; |
4391 | } | |
4392 | ||
4393 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4394 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4395 | |
83f0d77a | 4396 | /* Metadata version. |
e691063a N |
4397 | * This is one of |
4398 | * 'none' for arrays with no metadata (good luck...) | |
4399 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4400 | * or N.M for internally known formats |
4401 | */ | |
4402 | static ssize_t | |
fd01b88c | 4403 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4404 | { |
4405 | if (mddev->persistent) | |
4406 | return sprintf(page, "%d.%d\n", | |
4407 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4408 | else if (mddev->external) |
4409 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4410 | else |
4411 | return sprintf(page, "none\n"); | |
4412 | } | |
4413 | ||
4414 | static ssize_t | |
fd01b88c | 4415 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4416 | { |
4417 | int major, minor; | |
4418 | char *e; | |
6791875e | 4419 | int err; |
ea43ddd8 N |
4420 | /* Changing the details of 'external' metadata is |
4421 | * always permitted. Otherwise there must be | |
4422 | * no devices attached to the array. | |
4423 | */ | |
6791875e N |
4424 | |
4425 | err = mddev_lock(mddev); | |
4426 | if (err) | |
4427 | return err; | |
4428 | err = -EBUSY; | |
ea43ddd8 N |
4429 | if (mddev->external && strncmp(buf, "external:", 9) == 0) |
4430 | ; | |
4431 | else if (!list_empty(&mddev->disks)) | |
6791875e | 4432 | goto out_unlock; |
8bb93aac | 4433 | |
6791875e | 4434 | err = 0; |
8bb93aac N |
4435 | if (cmd_match(buf, "none")) { |
4436 | mddev->persistent = 0; | |
e691063a N |
4437 | mddev->external = 0; |
4438 | mddev->major_version = 0; | |
4439 | mddev->minor_version = 90; | |
6791875e | 4440 | goto out_unlock; |
e691063a N |
4441 | } |
4442 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4443 | size_t namelen = len-9; |
e691063a N |
4444 | if (namelen >= sizeof(mddev->metadata_type)) |
4445 | namelen = sizeof(mddev->metadata_type)-1; | |
4446 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4447 | mddev->metadata_type[namelen] = 0; | |
4448 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4449 | mddev->metadata_type[--namelen] = 0; | |
4450 | mddev->persistent = 0; | |
4451 | mddev->external = 1; | |
8bb93aac N |
4452 | mddev->major_version = 0; |
4453 | mddev->minor_version = 90; | |
6791875e | 4454 | goto out_unlock; |
8bb93aac N |
4455 | } |
4456 | major = simple_strtoul(buf, &e, 10); | |
6791875e | 4457 | err = -EINVAL; |
8bb93aac | 4458 | if (e==buf || *e != '.') |
6791875e | 4459 | goto out_unlock; |
8bb93aac N |
4460 | buf = e+1; |
4461 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4462 | if (e==buf || (*e && *e != '\n') ) |
6791875e N |
4463 | goto out_unlock; |
4464 | err = -ENOENT; | |
50511da3 | 4465 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
6791875e | 4466 | goto out_unlock; |
8bb93aac N |
4467 | mddev->major_version = major; |
4468 | mddev->minor_version = minor; | |
4469 | mddev->persistent = 1; | |
e691063a | 4470 | mddev->external = 0; |
6791875e N |
4471 | err = 0; |
4472 | out_unlock: | |
4473 | mddev_unlock(mddev); | |
4474 | return err ?: len; | |
8bb93aac N |
4475 | } |
4476 | ||
4477 | static struct md_sysfs_entry md_metadata = | |
750f199e | 4478 | __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4479 | |
24dd469d | 4480 | static ssize_t |
fd01b88c | 4481 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4482 | { |
7eec314d | 4483 | char *type = "idle"; |
b7b17c9b N |
4484 | unsigned long recovery = mddev->recovery; |
4485 | if (test_bit(MD_RECOVERY_FROZEN, &recovery)) | |
b6a9ce68 | 4486 | type = "frozen"; |
b7b17c9b N |
4487 | else if (test_bit(MD_RECOVERY_RUNNING, &recovery) || |
4488 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) { | |
4489 | if (test_bit(MD_RECOVERY_RESHAPE, &recovery)) | |
ccfcc3c1 | 4490 | type = "reshape"; |
b7b17c9b N |
4491 | else if (test_bit(MD_RECOVERY_SYNC, &recovery)) { |
4492 | if (!test_bit(MD_RECOVERY_REQUESTED, &recovery)) | |
24dd469d | 4493 | type = "resync"; |
b7b17c9b | 4494 | else if (test_bit(MD_RECOVERY_CHECK, &recovery)) |
24dd469d N |
4495 | type = "check"; |
4496 | else | |
4497 | type = "repair"; | |
b7b17c9b | 4498 | } else if (test_bit(MD_RECOVERY_RECOVER, &recovery)) |
24dd469d | 4499 | type = "recover"; |
985ca973 N |
4500 | else if (mddev->reshape_position != MaxSector) |
4501 | type = "reshape"; | |
24dd469d N |
4502 | } |
4503 | return sprintf(page, "%s\n", type); | |
4504 | } | |
4505 | ||
4506 | static ssize_t | |
fd01b88c | 4507 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4508 | { |
7eec314d N |
4509 | if (!mddev->pers || !mddev->pers->sync_request) |
4510 | return -EINVAL; | |
4511 | ||
b6a9ce68 N |
4512 | |
4513 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
56ccc112 N |
4514 | if (cmd_match(page, "frozen")) |
4515 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4516 | else | |
4517 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
8e8e2518 N |
4518 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
4519 | mddev_lock(mddev) == 0) { | |
4520 | flush_workqueue(md_misc_wq); | |
4521 | if (mddev->sync_thread) { | |
4522 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e | 4523 | md_reap_sync_thread(mddev); |
6791875e | 4524 | } |
8e8e2518 | 4525 | mddev_unlock(mddev); |
7eec314d | 4526 | } |
312045ee | 4527 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
24dd469d | 4528 | return -EBUSY; |
72a23c21 | 4529 | else if (cmd_match(page, "resync")) |
56ccc112 | 4530 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4531 | else if (cmd_match(page, "recover")) { |
56ccc112 | 4532 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4533 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
72a23c21 | 4534 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4535 | int err; |
4536 | if (mddev->pers->start_reshape == NULL) | |
4537 | return -EINVAL; | |
6791875e N |
4538 | err = mddev_lock(mddev); |
4539 | if (!err) { | |
312045ee N |
4540 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4541 | err = -EBUSY; | |
4542 | else { | |
4543 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4544 | err = mddev->pers->start_reshape(mddev); | |
4545 | } | |
6791875e N |
4546 | mddev_unlock(mddev); |
4547 | } | |
16484bf5 N |
4548 | if (err) |
4549 | return err; | |
a99ac971 | 4550 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4551 | } else { |
bce74dac | 4552 | if (cmd_match(page, "check")) |
7eec314d | 4553 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4554 | else if (!cmd_match(page, "repair")) |
7eec314d | 4555 | return -EINVAL; |
56ccc112 | 4556 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
7eec314d N |
4557 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
4558 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4559 | } |
48c26ddc N |
4560 | if (mddev->ro == 2) { |
4561 | /* A write to sync_action is enough to justify | |
4562 | * canceling read-auto mode | |
4563 | */ | |
4564 | mddev->ro = 0; | |
4565 | md_wakeup_thread(mddev->sync_thread); | |
4566 | } | |
03c902e1 | 4567 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4568 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4569 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4570 | return len; |
4571 | } | |
4572 | ||
c4a39551 | 4573 | static struct md_sysfs_entry md_scan_mode = |
750f199e | 4574 | __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
c4a39551 JB |
4575 | |
4576 | static ssize_t | |
4577 | last_sync_action_show(struct mddev *mddev, char *page) | |
4578 | { | |
4579 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4580 | } | |
4581 | ||
4582 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4583 | ||
9d88883e | 4584 | static ssize_t |
fd01b88c | 4585 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4586 | { |
4587 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4588 | (unsigned long long) |
4589 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4590 | } |
4591 | ||
80ca3a44 | 4592 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4593 | |
88202a0c | 4594 | static ssize_t |
fd01b88c | 4595 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4596 | { |
4597 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4598 | mddev->sync_speed_min ? "local": "system"); | |
4599 | } | |
4600 | ||
4601 | static ssize_t | |
fd01b88c | 4602 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4603 | { |
4c9309c0 AD |
4604 | unsigned int min; |
4605 | int rv; | |
4606 | ||
88202a0c | 4607 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4608 | min = 0; |
4609 | } else { | |
4610 | rv = kstrtouint(buf, 10, &min); | |
4611 | if (rv < 0) | |
4612 | return rv; | |
4613 | if (min == 0) | |
4614 | return -EINVAL; | |
88202a0c | 4615 | } |
88202a0c N |
4616 | mddev->sync_speed_min = min; |
4617 | return len; | |
4618 | } | |
4619 | ||
4620 | static struct md_sysfs_entry md_sync_min = | |
4621 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4622 | ||
4623 | static ssize_t | |
fd01b88c | 4624 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4625 | { |
4626 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4627 | mddev->sync_speed_max ? "local": "system"); | |
4628 | } | |
4629 | ||
4630 | static ssize_t | |
fd01b88c | 4631 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4632 | { |
4c9309c0 AD |
4633 | unsigned int max; |
4634 | int rv; | |
4635 | ||
88202a0c | 4636 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4637 | max = 0; |
4638 | } else { | |
4639 | rv = kstrtouint(buf, 10, &max); | |
4640 | if (rv < 0) | |
4641 | return rv; | |
4642 | if (max == 0) | |
4643 | return -EINVAL; | |
88202a0c | 4644 | } |
88202a0c N |
4645 | mddev->sync_speed_max = max; |
4646 | return len; | |
4647 | } | |
4648 | ||
4649 | static struct md_sysfs_entry md_sync_max = | |
4650 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4651 | ||
d7f3d291 | 4652 | static ssize_t |
fd01b88c | 4653 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4654 | { |
4655 | return sprintf(page, "%d\n", mddev->degraded); | |
4656 | } | |
4657 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4658 | |
90b08710 | 4659 | static ssize_t |
fd01b88c | 4660 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4661 | { |
4662 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4663 | } | |
4664 | ||
4665 | static ssize_t | |
fd01b88c | 4666 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4667 | { |
4668 | long n; | |
4669 | ||
b29bebd6 | 4670 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4671 | return -EINVAL; |
4672 | ||
4673 | if (n != 0 && n != 1) | |
4674 | return -EINVAL; | |
4675 | ||
4676 | mddev->parallel_resync = n; | |
4677 | ||
4678 | if (mddev->sync_thread) | |
4679 | wake_up(&resync_wait); | |
4680 | ||
4681 | return len; | |
4682 | } | |
4683 | ||
4684 | /* force parallel resync, even with shared block devices */ | |
4685 | static struct md_sysfs_entry md_sync_force_parallel = | |
4686 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4687 | sync_force_parallel_show, sync_force_parallel_store); | |
4688 | ||
88202a0c | 4689 | static ssize_t |
fd01b88c | 4690 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4691 | { |
4692 | unsigned long resync, dt, db; | |
d1a7c503 N |
4693 | if (mddev->curr_resync == 0) |
4694 | return sprintf(page, "none\n"); | |
9687a60c AN |
4695 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4696 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4697 | if (!dt) dt++; |
9687a60c AN |
4698 | db = resync - mddev->resync_mark_cnt; |
4699 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4700 | } |
4701 | ||
80ca3a44 | 4702 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4703 | |
4704 | static ssize_t | |
fd01b88c | 4705 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4706 | { |
13ae864b | 4707 | unsigned long long max_sectors, resync; |
88202a0c | 4708 | |
acb180b0 N |
4709 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4710 | return sprintf(page, "none\n"); | |
4711 | ||
72f36d59 N |
4712 | if (mddev->curr_resync == 1 || |
4713 | mddev->curr_resync == 2) | |
4714 | return sprintf(page, "delayed\n"); | |
4715 | ||
c804cdec N |
4716 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4717 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4718 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4719 | else |
58c0fed4 | 4720 | max_sectors = mddev->dev_sectors; |
88202a0c | 4721 | |
acb180b0 | 4722 | resync = mddev->curr_resync_completed; |
13ae864b | 4723 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4724 | } |
4725 | ||
750f199e N |
4726 | static struct md_sysfs_entry md_sync_completed = |
4727 | __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL); | |
88202a0c | 4728 | |
5e96ee65 | 4729 | static ssize_t |
fd01b88c | 4730 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4731 | { |
4732 | return sprintf(page, "%llu\n", | |
4733 | (unsigned long long)mddev->resync_min); | |
4734 | } | |
4735 | static ssize_t | |
fd01b88c | 4736 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4737 | { |
4738 | unsigned long long min; | |
23da422b | 4739 | int err; |
23da422b | 4740 | |
b29bebd6 | 4741 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4742 | return -EINVAL; |
23da422b N |
4743 | |
4744 | spin_lock(&mddev->lock); | |
4745 | err = -EINVAL; | |
5e96ee65 | 4746 | if (min > mddev->resync_max) |
23da422b N |
4747 | goto out_unlock; |
4748 | ||
4749 | err = -EBUSY; | |
5e96ee65 | 4750 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4751 | goto out_unlock; |
5e96ee65 | 4752 | |
50c37b13 N |
4753 | /* Round down to multiple of 4K for safety */ |
4754 | mddev->resync_min = round_down(min, 8); | |
23da422b | 4755 | err = 0; |
5e96ee65 | 4756 | |
23da422b N |
4757 | out_unlock: |
4758 | spin_unlock(&mddev->lock); | |
4759 | return err ?: len; | |
5e96ee65 NB |
4760 | } |
4761 | ||
4762 | static struct md_sysfs_entry md_min_sync = | |
4763 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4764 | ||
c6207277 | 4765 | static ssize_t |
fd01b88c | 4766 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4767 | { |
4768 | if (mddev->resync_max == MaxSector) | |
4769 | return sprintf(page, "max\n"); | |
4770 | else | |
4771 | return sprintf(page, "%llu\n", | |
4772 | (unsigned long long)mddev->resync_max); | |
4773 | } | |
4774 | static ssize_t | |
fd01b88c | 4775 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4776 | { |
23da422b N |
4777 | int err; |
4778 | spin_lock(&mddev->lock); | |
c6207277 N |
4779 | if (strncmp(buf, "max", 3) == 0) |
4780 | mddev->resync_max = MaxSector; | |
4781 | else { | |
5e96ee65 | 4782 | unsigned long long max; |
23da422b N |
4783 | int chunk; |
4784 | ||
4785 | err = -EINVAL; | |
b29bebd6 | 4786 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4787 | goto out_unlock; |
5e96ee65 | 4788 | if (max < mddev->resync_min) |
23da422b N |
4789 | goto out_unlock; |
4790 | ||
4791 | err = -EBUSY; | |
c6207277 | 4792 | if (max < mddev->resync_max && |
4d484a4a | 4793 | mddev->ro == 0 && |
c6207277 | 4794 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4795 | goto out_unlock; |
c6207277 N |
4796 | |
4797 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4798 | chunk = mddev->chunk_sectors; |
4799 | if (chunk) { | |
2ac06c33 | 4800 | sector_t temp = max; |
23da422b N |
4801 | |
4802 | err = -EINVAL; | |
4803 | if (sector_div(temp, chunk)) | |
4804 | goto out_unlock; | |
c6207277 N |
4805 | } |
4806 | mddev->resync_max = max; | |
4807 | } | |
4808 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4809 | err = 0; |
4810 | out_unlock: | |
4811 | spin_unlock(&mddev->lock); | |
4812 | return err ?: len; | |
c6207277 N |
4813 | } |
4814 | ||
4815 | static struct md_sysfs_entry md_max_sync = | |
4816 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4817 | ||
e464eafd | 4818 | static ssize_t |
fd01b88c | 4819 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4820 | { |
4821 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4822 | } | |
4823 | ||
4824 | static ssize_t | |
fd01b88c | 4825 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4826 | { |
4c9309c0 | 4827 | unsigned long long old, new; |
6791875e | 4828 | int err; |
e464eafd | 4829 | |
4c9309c0 AD |
4830 | err = kstrtoull(buf, 10, &new); |
4831 | if (err < 0) | |
4832 | return err; | |
4833 | if (new != (sector_t)new) | |
e464eafd | 4834 | return -EINVAL; |
23ddff37 | 4835 | |
6791875e N |
4836 | err = mddev_lock(mddev); |
4837 | if (err) | |
4838 | return err; | |
4839 | err = -EINVAL; | |
4840 | if (mddev->pers == NULL || | |
4841 | mddev->pers->quiesce == NULL) | |
4842 | goto unlock; | |
4843 | old = mddev->suspend_lo; | |
23ddff37 N |
4844 | mddev->suspend_lo = new; |
4845 | if (new >= old) | |
4846 | /* Shrinking suspended region */ | |
e464eafd | 4847 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4848 | else { |
4849 | /* Expanding suspended region - need to wait */ | |
4850 | mddev->pers->quiesce(mddev, 1); | |
4851 | mddev->pers->quiesce(mddev, 0); | |
4852 | } | |
6791875e N |
4853 | err = 0; |
4854 | unlock: | |
4855 | mddev_unlock(mddev); | |
4856 | return err ?: len; | |
e464eafd N |
4857 | } |
4858 | static struct md_sysfs_entry md_suspend_lo = | |
4859 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4860 | ||
e464eafd | 4861 | static ssize_t |
fd01b88c | 4862 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4863 | { |
4864 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4865 | } | |
4866 | ||
4867 | static ssize_t | |
fd01b88c | 4868 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4869 | { |
4c9309c0 | 4870 | unsigned long long old, new; |
6791875e | 4871 | int err; |
e464eafd | 4872 | |
4c9309c0 AD |
4873 | err = kstrtoull(buf, 10, &new); |
4874 | if (err < 0) | |
4875 | return err; | |
4876 | if (new != (sector_t)new) | |
e464eafd | 4877 | return -EINVAL; |
23ddff37 | 4878 | |
6791875e N |
4879 | err = mddev_lock(mddev); |
4880 | if (err) | |
4881 | return err; | |
4882 | err = -EINVAL; | |
4883 | if (mddev->pers == NULL || | |
4884 | mddev->pers->quiesce == NULL) | |
4885 | goto unlock; | |
4886 | old = mddev->suspend_hi; | |
23ddff37 N |
4887 | mddev->suspend_hi = new; |
4888 | if (new <= old) | |
4889 | /* Shrinking suspended region */ | |
4890 | mddev->pers->quiesce(mddev, 2); | |
4891 | else { | |
4892 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4893 | mddev->pers->quiesce(mddev, 1); |
4894 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 | 4895 | } |
6791875e N |
4896 | err = 0; |
4897 | unlock: | |
4898 | mddev_unlock(mddev); | |
4899 | return err ?: len; | |
e464eafd N |
4900 | } |
4901 | static struct md_sysfs_entry md_suspend_hi = | |
4902 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4903 | ||
08a02ecd | 4904 | static ssize_t |
fd01b88c | 4905 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4906 | { |
4907 | if (mddev->reshape_position != MaxSector) | |
4908 | return sprintf(page, "%llu\n", | |
4909 | (unsigned long long)mddev->reshape_position); | |
4910 | strcpy(page, "none\n"); | |
4911 | return 5; | |
4912 | } | |
4913 | ||
4914 | static ssize_t | |
fd01b88c | 4915 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4916 | { |
c6563a8c | 4917 | struct md_rdev *rdev; |
4c9309c0 | 4918 | unsigned long long new; |
6791875e | 4919 | int err; |
6791875e | 4920 | |
4c9309c0 AD |
4921 | err = kstrtoull(buf, 10, &new); |
4922 | if (err < 0) | |
4923 | return err; | |
4924 | if (new != (sector_t)new) | |
08a02ecd | 4925 | return -EINVAL; |
6791875e N |
4926 | err = mddev_lock(mddev); |
4927 | if (err) | |
4928 | return err; | |
4929 | err = -EBUSY; | |
4930 | if (mddev->pers) | |
4931 | goto unlock; | |
08a02ecd N |
4932 | mddev->reshape_position = new; |
4933 | mddev->delta_disks = 0; | |
2c810cdd | 4934 | mddev->reshape_backwards = 0; |
08a02ecd N |
4935 | mddev->new_level = mddev->level; |
4936 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4937 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4938 | rdev_for_each(rdev, mddev) |
4939 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4940 | err = 0; |
4941 | unlock: | |
4942 | mddev_unlock(mddev); | |
4943 | return err ?: len; | |
08a02ecd N |
4944 | } |
4945 | ||
4946 | static struct md_sysfs_entry md_reshape_position = | |
4947 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4948 | reshape_position_store); | |
4949 | ||
2c810cdd N |
4950 | static ssize_t |
4951 | reshape_direction_show(struct mddev *mddev, char *page) | |
4952 | { | |
4953 | return sprintf(page, "%s\n", | |
4954 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4955 | } | |
4956 | ||
4957 | static ssize_t | |
4958 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4959 | { | |
4960 | int backwards = 0; | |
6791875e N |
4961 | int err; |
4962 | ||
2c810cdd N |
4963 | if (cmd_match(buf, "forwards")) |
4964 | backwards = 0; | |
4965 | else if (cmd_match(buf, "backwards")) | |
4966 | backwards = 1; | |
4967 | else | |
4968 | return -EINVAL; | |
4969 | if (mddev->reshape_backwards == backwards) | |
4970 | return len; | |
4971 | ||
6791875e N |
4972 | err = mddev_lock(mddev); |
4973 | if (err) | |
4974 | return err; | |
2c810cdd N |
4975 | /* check if we are allowed to change */ |
4976 | if (mddev->delta_disks) | |
6791875e N |
4977 | err = -EBUSY; |
4978 | else if (mddev->persistent && | |
2c810cdd | 4979 | mddev->major_version == 0) |
6791875e N |
4980 | err = -EINVAL; |
4981 | else | |
4982 | mddev->reshape_backwards = backwards; | |
4983 | mddev_unlock(mddev); | |
4984 | return err ?: len; | |
2c810cdd N |
4985 | } |
4986 | ||
4987 | static struct md_sysfs_entry md_reshape_direction = | |
4988 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4989 | reshape_direction_store); | |
4990 | ||
b522adcd | 4991 | static ssize_t |
fd01b88c | 4992 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4993 | { |
4994 | if (mddev->external_size) | |
4995 | return sprintf(page, "%llu\n", | |
4996 | (unsigned long long)mddev->array_sectors/2); | |
4997 | else | |
4998 | return sprintf(page, "default\n"); | |
4999 | } | |
5000 | ||
5001 | static ssize_t | |
fd01b88c | 5002 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
5003 | { |
5004 | sector_t sectors; | |
6791875e N |
5005 | int err; |
5006 | ||
5007 | err = mddev_lock(mddev); | |
5008 | if (err) | |
5009 | return err; | |
b522adcd | 5010 | |
ab5a98b1 | 5011 | /* cluster raid doesn't support change array_sectors */ |
b670883b ZL |
5012 | if (mddev_is_clustered(mddev)) { |
5013 | mddev_unlock(mddev); | |
ab5a98b1 | 5014 | return -EINVAL; |
b670883b | 5015 | } |
ab5a98b1 | 5016 | |
b522adcd DW |
5017 | if (strncmp(buf, "default", 7) == 0) { |
5018 | if (mddev->pers) | |
5019 | sectors = mddev->pers->size(mddev, 0, 0); | |
5020 | else | |
5021 | sectors = mddev->array_sectors; | |
5022 | ||
5023 | mddev->external_size = 0; | |
5024 | } else { | |
5025 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
5026 | err = -EINVAL; |
5027 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
5028 | err = -E2BIG; | |
5029 | else | |
5030 | mddev->external_size = 1; | |
b522adcd DW |
5031 | } |
5032 | ||
6791875e N |
5033 | if (!err) { |
5034 | mddev->array_sectors = sectors; | |
5035 | if (mddev->pers) { | |
5036 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
5037 | revalidate_disk(mddev->gendisk); | |
5038 | } | |
cbe6ef1d | 5039 | } |
6791875e N |
5040 | mddev_unlock(mddev); |
5041 | return err ?: len; | |
b522adcd DW |
5042 | } |
5043 | ||
5044 | static struct md_sysfs_entry md_array_size = | |
5045 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
5046 | array_size_store); | |
e464eafd | 5047 | |
664aed04 AP |
5048 | static ssize_t |
5049 | consistency_policy_show(struct mddev *mddev, char *page) | |
5050 | { | |
5051 | int ret; | |
5052 | ||
5053 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { | |
5054 | ret = sprintf(page, "journal\n"); | |
5055 | } else if (test_bit(MD_HAS_PPL, &mddev->flags)) { | |
5056 | ret = sprintf(page, "ppl\n"); | |
5057 | } else if (mddev->bitmap) { | |
5058 | ret = sprintf(page, "bitmap\n"); | |
5059 | } else if (mddev->pers) { | |
5060 | if (mddev->pers->sync_request) | |
5061 | ret = sprintf(page, "resync\n"); | |
5062 | else | |
5063 | ret = sprintf(page, "none\n"); | |
5064 | } else { | |
5065 | ret = sprintf(page, "unknown\n"); | |
5066 | } | |
5067 | ||
5068 | return ret; | |
5069 | } | |
5070 | ||
5071 | static ssize_t | |
5072 | consistency_policy_store(struct mddev *mddev, const char *buf, size_t len) | |
5073 | { | |
ba903a3e AP |
5074 | int err = 0; |
5075 | ||
664aed04 | 5076 | if (mddev->pers) { |
ba903a3e AP |
5077 | if (mddev->pers->change_consistency_policy) |
5078 | err = mddev->pers->change_consistency_policy(mddev, buf); | |
5079 | else | |
5080 | err = -EBUSY; | |
664aed04 AP |
5081 | } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) { |
5082 | set_bit(MD_HAS_PPL, &mddev->flags); | |
664aed04 | 5083 | } else { |
ba903a3e | 5084 | err = -EINVAL; |
664aed04 | 5085 | } |
ba903a3e AP |
5086 | |
5087 | return err ? err : len; | |
664aed04 AP |
5088 | } |
5089 | ||
5090 | static struct md_sysfs_entry md_consistency_policy = | |
5091 | __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show, | |
5092 | consistency_policy_store); | |
5093 | ||
eae1701f N |
5094 | static struct attribute *md_default_attrs[] = { |
5095 | &md_level.attr, | |
d4dbd025 | 5096 | &md_layout.attr, |
eae1701f | 5097 | &md_raid_disks.attr, |
3b34380a | 5098 | &md_chunk_size.attr, |
a35b0d69 | 5099 | &md_size.attr, |
a94213b1 | 5100 | &md_resync_start.attr, |
8bb93aac | 5101 | &md_metadata.attr, |
6d7ff738 | 5102 | &md_new_device.attr, |
16f17b39 | 5103 | &md_safe_delay.attr, |
9e653b63 | 5104 | &md_array_state.attr, |
08a02ecd | 5105 | &md_reshape_position.attr, |
2c810cdd | 5106 | &md_reshape_direction.attr, |
b522adcd | 5107 | &md_array_size.attr, |
1e50915f | 5108 | &max_corr_read_errors.attr, |
664aed04 | 5109 | &md_consistency_policy.attr, |
411036fa N |
5110 | NULL, |
5111 | }; | |
5112 | ||
5113 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 5114 | &md_scan_mode.attr, |
c4a39551 | 5115 | &md_last_scan_mode.attr, |
9d88883e | 5116 | &md_mismatches.attr, |
88202a0c N |
5117 | &md_sync_min.attr, |
5118 | &md_sync_max.attr, | |
5119 | &md_sync_speed.attr, | |
90b08710 | 5120 | &md_sync_force_parallel.attr, |
88202a0c | 5121 | &md_sync_completed.attr, |
5e96ee65 | 5122 | &md_min_sync.attr, |
c6207277 | 5123 | &md_max_sync.attr, |
e464eafd N |
5124 | &md_suspend_lo.attr, |
5125 | &md_suspend_hi.attr, | |
9b1d1dac | 5126 | &md_bitmap.attr, |
d7f3d291 | 5127 | &md_degraded.attr, |
eae1701f N |
5128 | NULL, |
5129 | }; | |
411036fa N |
5130 | static struct attribute_group md_redundancy_group = { |
5131 | .name = NULL, | |
5132 | .attrs = md_redundancy_attrs, | |
5133 | }; | |
5134 | ||
eae1701f N |
5135 | static ssize_t |
5136 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
5137 | { | |
5138 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 5139 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 5140 | ssize_t rv; |
eae1701f N |
5141 | |
5142 | if (!entry->show) | |
5143 | return -EIO; | |
af8a2434 N |
5144 | spin_lock(&all_mddevs_lock); |
5145 | if (list_empty(&mddev->all_mddevs)) { | |
5146 | spin_unlock(&all_mddevs_lock); | |
5147 | return -EBUSY; | |
5148 | } | |
5149 | mddev_get(mddev); | |
5150 | spin_unlock(&all_mddevs_lock); | |
5151 | ||
b7b17c9b | 5152 | rv = entry->show(mddev, page); |
af8a2434 | 5153 | mddev_put(mddev); |
96de1e66 | 5154 | return rv; |
eae1701f N |
5155 | } |
5156 | ||
5157 | static ssize_t | |
5158 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
5159 | const char *page, size_t length) | |
5160 | { | |
5161 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 5162 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 5163 | ssize_t rv; |
eae1701f N |
5164 | |
5165 | if (!entry->store) | |
5166 | return -EIO; | |
67463acb N |
5167 | if (!capable(CAP_SYS_ADMIN)) |
5168 | return -EACCES; | |
af8a2434 N |
5169 | spin_lock(&all_mddevs_lock); |
5170 | if (list_empty(&mddev->all_mddevs)) { | |
5171 | spin_unlock(&all_mddevs_lock); | |
5172 | return -EBUSY; | |
5173 | } | |
5174 | mddev_get(mddev); | |
5175 | spin_unlock(&all_mddevs_lock); | |
6791875e | 5176 | rv = entry->store(mddev, page, length); |
af8a2434 | 5177 | mddev_put(mddev); |
96de1e66 | 5178 | return rv; |
eae1701f N |
5179 | } |
5180 | ||
5181 | static void md_free(struct kobject *ko) | |
5182 | { | |
fd01b88c | 5183 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
5184 | |
5185 | if (mddev->sysfs_state) | |
5186 | sysfs_put(mddev->sysfs_state); | |
5187 | ||
6cd18e71 N |
5188 | if (mddev->queue) |
5189 | blk_cleanup_queue(mddev->queue); | |
a21d1504 N |
5190 | if (mddev->gendisk) { |
5191 | del_gendisk(mddev->gendisk); | |
5192 | put_disk(mddev->gendisk); | |
5193 | } | |
4ad23a97 | 5194 | percpu_ref_exit(&mddev->writes_pending); |
a21d1504 | 5195 | |
eae1701f N |
5196 | kfree(mddev); |
5197 | } | |
5198 | ||
52cf25d0 | 5199 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
5200 | .show = md_attr_show, |
5201 | .store = md_attr_store, | |
5202 | }; | |
5203 | static struct kobj_type md_ktype = { | |
5204 | .release = md_free, | |
5205 | .sysfs_ops = &md_sysfs_ops, | |
5206 | .default_attrs = md_default_attrs, | |
5207 | }; | |
5208 | ||
1da177e4 LT |
5209 | int mdp_major = 0; |
5210 | ||
5fd3a17e DW |
5211 | static void mddev_delayed_delete(struct work_struct *ws) |
5212 | { | |
fd01b88c | 5213 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 5214 | |
43a70507 | 5215 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
5216 | kobject_del(&mddev->kobj); |
5217 | kobject_put(&mddev->kobj); | |
5218 | } | |
5219 | ||
4ad23a97 N |
5220 | static void no_op(struct percpu_ref *r) {} |
5221 | ||
a415c0f1 N |
5222 | int mddev_init_writes_pending(struct mddev *mddev) |
5223 | { | |
5224 | if (mddev->writes_pending.percpu_count_ptr) | |
5225 | return 0; | |
5226 | if (percpu_ref_init(&mddev->writes_pending, no_op, 0, GFP_KERNEL) < 0) | |
5227 | return -ENOMEM; | |
5228 | /* We want to start with the refcount at zero */ | |
5229 | percpu_ref_put(&mddev->writes_pending); | |
5230 | return 0; | |
5231 | } | |
5232 | EXPORT_SYMBOL_GPL(mddev_init_writes_pending); | |
5233 | ||
efeb53c0 | 5234 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 5235 | { |
039b7225 N |
5236 | /* |
5237 | * If dev is zero, name is the name of a device to allocate with | |
5238 | * an arbitrary minor number. It will be "md_???" | |
5239 | * If dev is non-zero it must be a device number with a MAJOR of | |
5240 | * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then | |
5241 | * the device is being created by opening a node in /dev. | |
5242 | * If "name" is not NULL, the device is being created by | |
5243 | * writing to /sys/module/md_mod/parameters/new_array. | |
5244 | */ | |
48c9c27b | 5245 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 5246 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 5247 | struct gendisk *disk; |
efeb53c0 N |
5248 | int partitioned; |
5249 | int shift; | |
5250 | int unit; | |
3830c62f | 5251 | int error; |
1da177e4 LT |
5252 | |
5253 | if (!mddev) | |
efeb53c0 N |
5254 | return -ENODEV; |
5255 | ||
5256 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
5257 | shift = partitioned ? MdpMinorShift : 0; | |
5258 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 5259 | |
e804ac78 TH |
5260 | /* wait for any previous instance of this device to be |
5261 | * completely removed (mddev_delayed_delete). | |
d3374825 | 5262 | */ |
e804ac78 | 5263 | flush_workqueue(md_misc_wq); |
d3374825 | 5264 | |
48c9c27b | 5265 | mutex_lock(&disks_mutex); |
0909dc44 N |
5266 | error = -EEXIST; |
5267 | if (mddev->gendisk) | |
5268 | goto abort; | |
efeb53c0 | 5269 | |
039b7225 | 5270 | if (name && !dev) { |
efeb53c0 N |
5271 | /* Need to ensure that 'name' is not a duplicate. |
5272 | */ | |
fd01b88c | 5273 | struct mddev *mddev2; |
efeb53c0 N |
5274 | spin_lock(&all_mddevs_lock); |
5275 | ||
5276 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
5277 | if (mddev2->gendisk && | |
5278 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
5279 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 5280 | goto abort; |
efeb53c0 N |
5281 | } |
5282 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 5283 | } |
039b7225 N |
5284 | if (name && dev) |
5285 | /* | |
5286 | * Creating /dev/mdNNN via "newarray", so adjust hold_active. | |
5287 | */ | |
5288 | mddev->hold_active = UNTIL_STOP; | |
8b765398 | 5289 | |
0909dc44 | 5290 | error = -ENOMEM; |
8b765398 | 5291 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
5292 | if (!mddev->queue) |
5293 | goto abort; | |
409c57f3 N |
5294 | mddev->queue->queuedata = mddev; |
5295 | ||
409c57f3 | 5296 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 5297 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 5298 | |
1da177e4 LT |
5299 | disk = alloc_disk(1 << shift); |
5300 | if (!disk) { | |
8b765398 N |
5301 | blk_cleanup_queue(mddev->queue); |
5302 | mddev->queue = NULL; | |
0909dc44 | 5303 | goto abort; |
1da177e4 | 5304 | } |
efeb53c0 | 5305 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 5306 | disk->first_minor = unit << shift; |
efeb53c0 N |
5307 | if (name) |
5308 | strcpy(disk->disk_name, name); | |
5309 | else if (partitioned) | |
1da177e4 | 5310 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 5311 | else |
1da177e4 | 5312 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
5313 | disk->fops = &md_fops; |
5314 | disk->private_data = mddev; | |
5315 | disk->queue = mddev->queue; | |
56883a7e | 5316 | blk_queue_write_cache(mddev->queue, true, true); |
92850bbd | 5317 | /* Allow extended partitions. This makes the |
d3374825 | 5318 | * 'mdp' device redundant, but we can't really |
92850bbd N |
5319 | * remove it now. |
5320 | */ | |
5321 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 5322 | mddev->gendisk = disk; |
b0140891 N |
5323 | /* As soon as we call add_disk(), another thread could get |
5324 | * through to md_open, so make sure it doesn't get too far | |
5325 | */ | |
5326 | mutex_lock(&mddev->open_mutex); | |
5327 | add_disk(disk); | |
5328 | ||
ed9e1982 TH |
5329 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
5330 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
5331 | if (error) { |
5332 | /* This isn't possible, but as kobject_init_and_add is marked | |
5333 | * __must_check, we must do something with the result | |
5334 | */ | |
9d48739e N |
5335 | pr_debug("md: cannot register %s/md - name in use\n", |
5336 | disk->disk_name); | |
0909dc44 N |
5337 | error = 0; |
5338 | } | |
00bcb4ac N |
5339 | if (mddev->kobj.sd && |
5340 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
9d48739e | 5341 | pr_debug("pointless warning\n"); |
b0140891 | 5342 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
5343 | abort: |
5344 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 5345 | if (!error && mddev->kobj.sd) { |
3830c62f | 5346 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 5347 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 5348 | } |
d3374825 | 5349 | mddev_put(mddev); |
0909dc44 | 5350 | return error; |
efeb53c0 N |
5351 | } |
5352 | ||
5353 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
5354 | { | |
78b6350d N |
5355 | if (create_on_open) |
5356 | md_alloc(dev, NULL); | |
1da177e4 LT |
5357 | return NULL; |
5358 | } | |
5359 | ||
efeb53c0 N |
5360 | static int add_named_array(const char *val, struct kernel_param *kp) |
5361 | { | |
039b7225 N |
5362 | /* |
5363 | * val must be "md_*" or "mdNNN". | |
5364 | * For "md_*" we allocate an array with a large free minor number, and | |
efeb53c0 | 5365 | * set the name to val. val must not already be an active name. |
039b7225 N |
5366 | * For "mdNNN" we allocate an array with the minor number NNN |
5367 | * which must not already be in use. | |
efeb53c0 N |
5368 | */ |
5369 | int len = strlen(val); | |
5370 | char buf[DISK_NAME_LEN]; | |
039b7225 | 5371 | unsigned long devnum; |
efeb53c0 N |
5372 | |
5373 | while (len && val[len-1] == '\n') | |
5374 | len--; | |
5375 | if (len >= DISK_NAME_LEN) | |
5376 | return -E2BIG; | |
5377 | strlcpy(buf, val, len+1); | |
039b7225 N |
5378 | if (strncmp(buf, "md_", 3) == 0) |
5379 | return md_alloc(0, buf); | |
5380 | if (strncmp(buf, "md", 2) == 0 && | |
5381 | isdigit(buf[2]) && | |
5382 | kstrtoul(buf+2, 10, &devnum) == 0 && | |
5383 | devnum <= MINORMASK) | |
5384 | return md_alloc(MKDEV(MD_MAJOR, devnum), NULL); | |
5385 | ||
5386 | return -EINVAL; | |
efeb53c0 N |
5387 | } |
5388 | ||
1da177e4 LT |
5389 | static void md_safemode_timeout(unsigned long data) |
5390 | { | |
fd01b88c | 5391 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 5392 | |
4ad23a97 N |
5393 | mddev->safemode = 1; |
5394 | if (mddev->external) | |
5395 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
5396 | ||
1da177e4 LT |
5397 | md_wakeup_thread(mddev->thread); |
5398 | } | |
5399 | ||
6ff8d8ec | 5400 | static int start_dirty_degraded; |
1da177e4 | 5401 | |
fd01b88c | 5402 | int md_run(struct mddev *mddev) |
1da177e4 | 5403 | { |
2604b703 | 5404 | int err; |
3cb03002 | 5405 | struct md_rdev *rdev; |
84fc4b56 | 5406 | struct md_personality *pers; |
1da177e4 | 5407 | |
a757e64c N |
5408 | if (list_empty(&mddev->disks)) |
5409 | /* cannot run an array with no devices.. */ | |
1da177e4 | 5410 | return -EINVAL; |
1da177e4 LT |
5411 | |
5412 | if (mddev->pers) | |
5413 | return -EBUSY; | |
bb4f1e9d N |
5414 | /* Cannot run until previous stop completes properly */ |
5415 | if (mddev->sysfs_active) | |
5416 | return -EBUSY; | |
b6eb127d | 5417 | |
1da177e4 LT |
5418 | /* |
5419 | * Analyze all RAID superblock(s) | |
5420 | */ | |
1ec4a939 N |
5421 | if (!mddev->raid_disks) { |
5422 | if (!mddev->persistent) | |
5423 | return -EINVAL; | |
a757e64c | 5424 | analyze_sbs(mddev); |
1ec4a939 | 5425 | } |
1da177e4 | 5426 | |
d9d166c2 N |
5427 | if (mddev->level != LEVEL_NONE) |
5428 | request_module("md-level-%d", mddev->level); | |
5429 | else if (mddev->clevel[0]) | |
5430 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
5431 | |
5432 | /* | |
5433 | * Drop all container device buffers, from now on | |
5434 | * the only valid external interface is through the md | |
5435 | * device. | |
1da177e4 | 5436 | */ |
dafb20fa | 5437 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 5438 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5439 | continue; |
5440 | sync_blockdev(rdev->bdev); | |
f98393a6 | 5441 | invalidate_bdev(rdev->bdev); |
97b20ef7 N |
5442 | if (mddev->ro != 1 && |
5443 | (bdev_read_only(rdev->bdev) || | |
5444 | bdev_read_only(rdev->meta_bdev))) { | |
5445 | mddev->ro = 1; | |
5446 | if (mddev->gendisk) | |
5447 | set_disk_ro(mddev->gendisk, 1); | |
5448 | } | |
f0d76d70 N |
5449 | |
5450 | /* perform some consistency tests on the device. | |
5451 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 5452 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 5453 | */ |
a6ff7e08 JB |
5454 | if (rdev->meta_bdev) { |
5455 | /* Nothing to check */; | |
5456 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5457 | if (mddev->dev_sectors && |
5458 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5459 | > rdev->sb_start) { |
9d48739e N |
5460 | pr_warn("md: %s: data overlaps metadata\n", |
5461 | mdname(mddev)); | |
f0d76d70 N |
5462 | return -EINVAL; |
5463 | } | |
5464 | } else { | |
0f420358 | 5465 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 | 5466 | > rdev->data_offset) { |
9d48739e N |
5467 | pr_warn("md: %s: metadata overlaps data\n", |
5468 | mdname(mddev)); | |
f0d76d70 N |
5469 | return -EINVAL; |
5470 | } | |
5471 | } | |
00bcb4ac | 5472 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5473 | } |
5474 | ||
10273170 | 5475 | if (mddev->bio_set == NULL) { |
011067b0 | 5476 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
10273170 ML |
5477 | if (!mddev->bio_set) |
5478 | return -ENOMEM; | |
5479 | } | |
5a85071c | 5480 | if (mddev->sync_set == NULL) { |
026d15f6 | 5481 | mddev->sync_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
5a85071c N |
5482 | if (!mddev->sync_set) |
5483 | return -ENOMEM; | |
5484 | } | |
a167f663 | 5485 | |
1da177e4 | 5486 | spin_lock(&pers_lock); |
d9d166c2 | 5487 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5488 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5489 | spin_unlock(&pers_lock); |
d9d166c2 | 5490 | if (mddev->level != LEVEL_NONE) |
9d48739e N |
5491 | pr_warn("md: personality for level %d is not loaded!\n", |
5492 | mddev->level); | |
d9d166c2 | 5493 | else |
9d48739e N |
5494 | pr_warn("md: personality for level %s is not loaded!\n", |
5495 | mddev->clevel); | |
1da177e4 LT |
5496 | return -EINVAL; |
5497 | } | |
1da177e4 | 5498 | spin_unlock(&pers_lock); |
34817e8c N |
5499 | if (mddev->level != pers->level) { |
5500 | mddev->level = pers->level; | |
5501 | mddev->new_level = pers->level; | |
5502 | } | |
d9d166c2 | 5503 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5504 | |
f6705578 | 5505 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5506 | pers->start_reshape == NULL) { |
f6705578 | 5507 | /* This personality cannot handle reshaping... */ |
f6705578 N |
5508 | module_put(pers->owner); |
5509 | return -EINVAL; | |
5510 | } | |
5511 | ||
7dd5e7c3 N |
5512 | if (pers->sync_request) { |
5513 | /* Warn if this is a potentially silly | |
5514 | * configuration. | |
5515 | */ | |
5516 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5517 | struct md_rdev *rdev2; |
7dd5e7c3 | 5518 | int warned = 0; |
159ec1fc | 5519 | |
dafb20fa N |
5520 | rdev_for_each(rdev, mddev) |
5521 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5522 | if (rdev < rdev2 && |
5523 | rdev->bdev->bd_contains == | |
5524 | rdev2->bdev->bd_contains) { | |
9d48739e N |
5525 | pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n", |
5526 | mdname(mddev), | |
5527 | bdevname(rdev->bdev,b), | |
5528 | bdevname(rdev2->bdev,b2)); | |
7dd5e7c3 N |
5529 | warned = 1; |
5530 | } | |
5531 | } | |
159ec1fc | 5532 | |
7dd5e7c3 | 5533 | if (warned) |
9d48739e | 5534 | pr_warn("True protection against single-disk failure might be compromised.\n"); |
7dd5e7c3 N |
5535 | } |
5536 | ||
657390d2 | 5537 | mddev->recovery = 0; |
58c0fed4 AN |
5538 | /* may be over-ridden by personality */ |
5539 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5540 | ||
6ff8d8ec | 5541 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5542 | |
0f9552b5 | 5543 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5544 | mddev->ro = 2; /* read-only, but switch on first write */ |
5545 | ||
a85dd7b8 SL |
5546 | /* |
5547 | * NOTE: some pers->run(), for example r5l_recovery_log(), wakes | |
5548 | * up mddev->thread. It is important to initialize critical | |
5549 | * resources for mddev->thread BEFORE calling pers->run(). | |
5550 | */ | |
36d091f4 | 5551 | err = pers->run(mddev); |
13e53df3 | 5552 | if (err) |
9d48739e | 5553 | pr_warn("md: pers->run() failed ...\n"); |
36d091f4 | 5554 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
9d48739e N |
5555 | WARN_ONCE(!mddev->external_size, |
5556 | "%s: default size too small, but 'external_size' not in effect?\n", | |
5557 | __func__); | |
5558 | pr_warn("md: invalid array_size %llu > default size %llu\n", | |
5559 | (unsigned long long)mddev->array_sectors / 2, | |
5560 | (unsigned long long)pers->size(mddev, 0, 0) / 2); | |
b522adcd | 5561 | err = -EINVAL; |
b522adcd | 5562 | } |
36d091f4 | 5563 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5564 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
f9209a32 GR |
5565 | struct bitmap *bitmap; |
5566 | ||
5567 | bitmap = bitmap_create(mddev, -1); | |
5568 | if (IS_ERR(bitmap)) { | |
5569 | err = PTR_ERR(bitmap); | |
9d48739e N |
5570 | pr_warn("%s: failed to create bitmap (%d)\n", |
5571 | mdname(mddev), err); | |
f9209a32 GR |
5572 | } else |
5573 | mddev->bitmap = bitmap; | |
5574 | ||
b15c2e57 | 5575 | } |
1da177e4 | 5576 | if (err) { |
5aa61f42 | 5577 | mddev_detach(mddev); |
0c35bd47 N |
5578 | if (mddev->private) |
5579 | pers->free(mddev, mddev->private); | |
bd691922 | 5580 | mddev->private = NULL; |
36d091f4 | 5581 | module_put(pers->owner); |
32a7627c N |
5582 | bitmap_destroy(mddev); |
5583 | return err; | |
1da177e4 | 5584 | } |
5c675f83 | 5585 | if (mddev->queue) { |
bb086a89 SL |
5586 | bool nonrot = true; |
5587 | ||
5588 | rdev_for_each(rdev, mddev) { | |
5589 | if (rdev->raid_disk >= 0 && | |
5590 | !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
5591 | nonrot = false; | |
5592 | break; | |
5593 | } | |
5594 | } | |
5595 | if (mddev->degraded) | |
5596 | nonrot = false; | |
5597 | if (nonrot) | |
5598 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue); | |
5599 | else | |
5600 | queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue); | |
dc3b17cc JK |
5601 | mddev->queue->backing_dev_info->congested_data = mddev; |
5602 | mddev->queue->backing_dev_info->congested_fn = md_congested; | |
5c675f83 | 5603 | } |
36d091f4 | 5604 | if (pers->sync_request) { |
00bcb4ac N |
5605 | if (mddev->kobj.sd && |
5606 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
5607 | pr_warn("md: cannot register extra attributes for %s\n", |
5608 | mdname(mddev)); | |
00bcb4ac | 5609 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5610 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5611 | mddev->ro = 0; |
5612 | ||
1e50915f RB |
5613 | atomic_set(&mddev->max_corr_read_errors, |
5614 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 | 5615 | mddev->safemode = 0; |
28c1b9fd GR |
5616 | if (mddev_is_clustered(mddev)) |
5617 | mddev->safemode_delay = 0; | |
5618 | else | |
5619 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ | |
1da177e4 | 5620 | mddev->in_sync = 1; |
0ca69886 | 5621 | smp_wmb(); |
36d091f4 N |
5622 | spin_lock(&mddev->lock); |
5623 | mddev->pers = pers; | |
36d091f4 | 5624 | spin_unlock(&mddev->lock); |
dafb20fa | 5625 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5626 | if (rdev->raid_disk >= 0) |
5627 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5628 | /* failure here is OK */; |
f72ffdd6 | 5629 | |
a4a3d26d N |
5630 | if (mddev->degraded && !mddev->ro) |
5631 | /* This ensures that recovering status is reported immediately | |
5632 | * via sysfs - until a lack of spares is confirmed. | |
5633 | */ | |
5634 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
1da177e4 | 5635 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5636 | |
2953079c | 5637 | if (mddev->sb_flags) |
850b2b42 | 5638 | md_update_sb(mddev, 0); |
1da177e4 | 5639 | |
d7603b7e | 5640 | md_new_event(mddev); |
00bcb4ac N |
5641 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5642 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5643 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5644 | return 0; |
5645 | } | |
390ee602 | 5646 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5647 | |
fd01b88c | 5648 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5649 | { |
5650 | int err; | |
5651 | ||
5652 | err = md_run(mddev); | |
5653 | if (err) | |
5654 | goto out; | |
69e51b44 N |
5655 | err = bitmap_load(mddev); |
5656 | if (err) { | |
5657 | bitmap_destroy(mddev); | |
5658 | goto out; | |
5659 | } | |
0fd018af | 5660 | |
28c1b9fd GR |
5661 | if (mddev_is_clustered(mddev)) |
5662 | md_allow_write(mddev); | |
5663 | ||
0fd018af JB |
5664 | md_wakeup_thread(mddev->thread); |
5665 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5666 | ||
fe60b014 N |
5667 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5668 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5669 | mddev->changed = 1; |
fe60b014 N |
5670 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5671 | out: | |
5672 | return err; | |
5673 | } | |
5674 | ||
fd01b88c | 5675 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5676 | { |
5677 | struct gendisk *disk = mddev->gendisk; | |
97b20ef7 N |
5678 | struct md_rdev *rdev; |
5679 | bool has_journal = false; | |
5680 | bool has_readonly = false; | |
1da177e4 | 5681 | |
80fab1d7 | 5682 | /* Complain if it has no devices */ |
1da177e4 | 5683 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5684 | return -ENXIO; |
5685 | if (!mddev->pers) | |
5686 | return -EINVAL; | |
5687 | if (!mddev->ro) | |
5688 | return -EBUSY; | |
339421de | 5689 | |
97b20ef7 N |
5690 | rcu_read_lock(); |
5691 | rdev_for_each_rcu(rdev, mddev) { | |
5692 | if (test_bit(Journal, &rdev->flags) && | |
5693 | !test_bit(Faulty, &rdev->flags)) | |
5694 | has_journal = true; | |
5695 | if (bdev_read_only(rdev->bdev)) | |
5696 | has_readonly = true; | |
5697 | } | |
5698 | rcu_read_unlock(); | |
5699 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal) | |
339421de | 5700 | /* Don't restart rw with journal missing/faulty */ |
339421de | 5701 | return -EINVAL; |
97b20ef7 N |
5702 | if (has_readonly) |
5703 | return -EROFS; | |
339421de | 5704 | |
80fab1d7 AN |
5705 | mddev->safemode = 0; |
5706 | mddev->ro = 0; | |
5707 | set_disk_ro(disk, 0); | |
9d48739e | 5708 | pr_debug("md: %s switched to read-write mode.\n", mdname(mddev)); |
80fab1d7 AN |
5709 | /* Kick recovery or resync if necessary */ |
5710 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5711 | md_wakeup_thread(mddev->thread); | |
5712 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5713 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5714 | return 0; |
1da177e4 LT |
5715 | } |
5716 | ||
fd01b88c | 5717 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5718 | { |
5719 | mddev->array_sectors = 0; | |
5720 | mddev->external_size = 0; | |
5721 | mddev->dev_sectors = 0; | |
5722 | mddev->raid_disks = 0; | |
5723 | mddev->recovery_cp = 0; | |
5724 | mddev->resync_min = 0; | |
5725 | mddev->resync_max = MaxSector; | |
5726 | mddev->reshape_position = MaxSector; | |
5727 | mddev->external = 0; | |
5728 | mddev->persistent = 0; | |
5729 | mddev->level = LEVEL_NONE; | |
5730 | mddev->clevel[0] = 0; | |
5731 | mddev->flags = 0; | |
2953079c | 5732 | mddev->sb_flags = 0; |
6177b472 N |
5733 | mddev->ro = 0; |
5734 | mddev->metadata_type[0] = 0; | |
5735 | mddev->chunk_sectors = 0; | |
5736 | mddev->ctime = mddev->utime = 0; | |
5737 | mddev->layout = 0; | |
5738 | mddev->max_disks = 0; | |
5739 | mddev->events = 0; | |
a8707c08 | 5740 | mddev->can_decrease_events = 0; |
6177b472 | 5741 | mddev->delta_disks = 0; |
2c810cdd | 5742 | mddev->reshape_backwards = 0; |
6177b472 N |
5743 | mddev->new_level = LEVEL_NONE; |
5744 | mddev->new_layout = 0; | |
5745 | mddev->new_chunk_sectors = 0; | |
5746 | mddev->curr_resync = 0; | |
7f7583d4 | 5747 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5748 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5749 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5750 | mddev->recovery = 0; | |
5751 | mddev->in_sync = 0; | |
f0b4f7e2 | 5752 | mddev->changed = 0; |
6177b472 | 5753 | mddev->degraded = 0; |
6177b472 | 5754 | mddev->safemode = 0; |
bd691922 | 5755 | mddev->private = NULL; |
c20c33f0 | 5756 | mddev->cluster_info = NULL; |
6177b472 N |
5757 | mddev->bitmap_info.offset = 0; |
5758 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5759 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5760 | mddev->bitmap_info.chunksize = 0; |
5761 | mddev->bitmap_info.daemon_sleep = 0; | |
5762 | mddev->bitmap_info.max_write_behind = 0; | |
c20c33f0 | 5763 | mddev->bitmap_info.nodes = 0; |
6177b472 N |
5764 | } |
5765 | ||
fd01b88c | 5766 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5767 | { |
6b6204ee | 5768 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5769 | flush_workqueue(md_misc_wq); |
a047e125 | 5770 | if (mddev->sync_thread) { |
a047e125 | 5771 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5772 | md_reap_sync_thread(mddev); |
a047e125 N |
5773 | } |
5774 | ||
5775 | del_timer_sync(&mddev->safemode_timer); | |
5776 | ||
034e33f5 SL |
5777 | if (mddev->pers && mddev->pers->quiesce) { |
5778 | mddev->pers->quiesce(mddev, 1); | |
5779 | mddev->pers->quiesce(mddev, 0); | |
5780 | } | |
a047e125 | 5781 | bitmap_flush(mddev); |
a047e125 | 5782 | |
b6d428c6 | 5783 | if (mddev->ro == 0 && |
28c1b9fd | 5784 | ((!mddev->in_sync && !mddev_is_clustered(mddev)) || |
2953079c | 5785 | mddev->sb_flags)) { |
a047e125 | 5786 | /* mark array as shutdown cleanly */ |
28c1b9fd GR |
5787 | if (!mddev_is_clustered(mddev)) |
5788 | mddev->in_sync = 1; | |
a047e125 N |
5789 | md_update_sb(mddev, 1); |
5790 | } | |
5791 | } | |
defad61a | 5792 | |
fd01b88c | 5793 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5794 | { |
29f097c4 | 5795 | mddev_lock_nointr(mddev); |
defad61a N |
5796 | __md_stop_writes(mddev); |
5797 | mddev_unlock(mddev); | |
5798 | } | |
390ee602 | 5799 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5800 | |
5aa61f42 N |
5801 | static void mddev_detach(struct mddev *mddev) |
5802 | { | |
48df498d | 5803 | bitmap_wait_behind_writes(mddev); |
36d091f4 | 5804 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5805 | mddev->pers->quiesce(mddev, 1); |
5806 | mddev->pers->quiesce(mddev, 0); | |
5807 | } | |
5808 | md_unregister_thread(&mddev->thread); | |
5809 | if (mddev->queue) | |
5810 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5811 | } | |
5812 | ||
5eff3c43 | 5813 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5814 | { |
36d091f4 | 5815 | struct md_personality *pers = mddev->pers; |
48df498d | 5816 | bitmap_destroy(mddev); |
5aa61f42 | 5817 | mddev_detach(mddev); |
ee5d004f N |
5818 | /* Ensure ->event_work is done */ |
5819 | flush_workqueue(md_misc_wq); | |
36d091f4 | 5820 | spin_lock(&mddev->lock); |
6177b472 | 5821 | mddev->pers = NULL; |
36d091f4 N |
5822 | spin_unlock(&mddev->lock); |
5823 | pers->free(mddev, mddev->private); | |
bd691922 | 5824 | mddev->private = NULL; |
36d091f4 N |
5825 | if (pers->sync_request && mddev->to_remove == NULL) |
5826 | mddev->to_remove = &md_redundancy_group; | |
5827 | module_put(pers->owner); | |
cca9cf90 | 5828 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5829 | } |
5eff3c43 N |
5830 | |
5831 | void md_stop(struct mddev *mddev) | |
5832 | { | |
5833 | /* stop the array and free an attached data structures. | |
5834 | * This is called from dm-raid | |
5835 | */ | |
5836 | __md_stop(mddev); | |
5eff3c43 N |
5837 | if (mddev->bio_set) |
5838 | bioset_free(mddev->bio_set); | |
5839 | } | |
5840 | ||
390ee602 | 5841 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5842 | |
a05b7ea0 | 5843 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5844 | { |
5845 | int err = 0; | |
30b8feb7 N |
5846 | int did_freeze = 0; |
5847 | ||
5848 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5849 | did_freeze = 1; | |
5850 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5851 | md_wakeup_thread(mddev->thread); | |
5852 | } | |
f851b60d | 5853 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5854 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5855 | if (mddev->sync_thread) |
30b8feb7 N |
5856 | /* Thread might be blocked waiting for metadata update |
5857 | * which will now never happen */ | |
5858 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5859 | |
2953079c | 5860 | if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
88724bfa | 5861 | return -EBUSY; |
30b8feb7 | 5862 | mddev_unlock(mddev); |
f851b60d N |
5863 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5864 | &mddev->recovery)); | |
88724bfa | 5865 | wait_event(mddev->sb_wait, |
2953079c | 5866 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
30b8feb7 N |
5867 | mddev_lock_nointr(mddev); |
5868 | ||
a4bd82d0 | 5869 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5870 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5871 | mddev->sync_thread || |
af8d8e6f | 5872 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5873 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5874 | if (did_freeze) { |
5875 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5876 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5877 | md_wakeup_thread(mddev->thread); |
5878 | } | |
a4bd82d0 N |
5879 | err = -EBUSY; |
5880 | goto out; | |
5881 | } | |
5882 | if (mddev->pers) { | |
defad61a | 5883 | __md_stop_writes(mddev); |
a4bd82d0 N |
5884 | |
5885 | err = -ENXIO; | |
5886 | if (mddev->ro==1) | |
5887 | goto out; | |
5888 | mddev->ro = 1; | |
5889 | set_disk_ro(mddev->gendisk, 1); | |
5890 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d N |
5891 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5892 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5893 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5894 | err = 0; |
a4bd82d0 N |
5895 | } |
5896 | out: | |
5897 | mutex_unlock(&mddev->open_mutex); | |
5898 | return err; | |
5899 | } | |
5900 | ||
9e653b63 N |
5901 | /* mode: |
5902 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5903 | * 2 - stop but do not disassemble array |
5904 | */ | |
f72ffdd6 | 5905 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5906 | struct block_device *bdev) |
1da177e4 | 5907 | { |
1da177e4 | 5908 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5909 | struct md_rdev *rdev; |
30b8feb7 N |
5910 | int did_freeze = 0; |
5911 | ||
5912 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5913 | did_freeze = 1; | |
5914 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5915 | md_wakeup_thread(mddev->thread); | |
5916 | } | |
f851b60d | 5917 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5918 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5919 | if (mddev->sync_thread) |
30b8feb7 N |
5920 | /* Thread might be blocked waiting for metadata update |
5921 | * which will now never happen */ | |
5922 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5923 | |
30b8feb7 | 5924 | mddev_unlock(mddev); |
f851b60d N |
5925 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
5926 | !test_bit(MD_RECOVERY_RUNNING, | |
5927 | &mddev->recovery))); | |
30b8feb7 | 5928 | mddev_lock_nointr(mddev); |
1da177e4 | 5929 | |
c8c00a69 | 5930 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5931 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5932 | mddev->sysfs_active || |
5933 | mddev->sync_thread || | |
af8d8e6f | 5934 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5935 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5936 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5937 | if (did_freeze) { |
5938 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5939 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5940 | md_wakeup_thread(mddev->thread); |
5941 | } | |
260fa034 N |
5942 | return -EBUSY; |
5943 | } | |
6e17b027 | 5944 | if (mddev->pers) { |
a4bd82d0 N |
5945 | if (mddev->ro) |
5946 | set_disk_ro(disk, 0); | |
409c57f3 | 5947 | |
defad61a | 5948 | __md_stop_writes(mddev); |
5eff3c43 | 5949 | __md_stop(mddev); |
dc3b17cc | 5950 | mddev->queue->backing_dev_info->congested_fn = NULL; |
6177b472 | 5951 | |
a4bd82d0 | 5952 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5953 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5954 | |
dafb20fa | 5955 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5956 | if (rdev->raid_disk >= 0) |
5957 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5958 | |
a4bd82d0 | 5959 | set_capacity(disk, 0); |
6e17b027 | 5960 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5961 | mddev->changed = 1; |
a4bd82d0 | 5962 | revalidate_disk(disk); |
0d4ca600 | 5963 | |
a4bd82d0 N |
5964 | if (mddev->ro) |
5965 | mddev->ro = 0; | |
6e17b027 N |
5966 | } else |
5967 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5968 | /* |
5969 | * Free resources if final stop | |
5970 | */ | |
9e653b63 | 5971 | if (mode == 0) { |
9d48739e | 5972 | pr_info("md: %s stopped.\n", mdname(mddev)); |
1da177e4 | 5973 | |
c3d9714e | 5974 | if (mddev->bitmap_info.file) { |
4af1a041 N |
5975 | struct file *f = mddev->bitmap_info.file; |
5976 | spin_lock(&mddev->lock); | |
c3d9714e | 5977 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
5978 | spin_unlock(&mddev->lock); |
5979 | fput(f); | |
978f946b | 5980 | } |
c3d9714e | 5981 | mddev->bitmap_info.offset = 0; |
978f946b | 5982 | |
1da177e4 LT |
5983 | export_array(mddev); |
5984 | ||
6177b472 | 5985 | md_clean(mddev); |
efeb53c0 N |
5986 | if (mddev->hold_active == UNTIL_STOP) |
5987 | mddev->hold_active = 0; | |
a4bd82d0 | 5988 | } |
d7603b7e | 5989 | md_new_event(mddev); |
00bcb4ac | 5990 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5991 | return 0; |
1da177e4 LT |
5992 | } |
5993 | ||
fdee8ae4 | 5994 | #ifndef MODULE |
fd01b88c | 5995 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5996 | { |
3cb03002 | 5997 | struct md_rdev *rdev; |
1da177e4 LT |
5998 | int err; |
5999 | ||
a757e64c | 6000 | if (list_empty(&mddev->disks)) |
1da177e4 | 6001 | return; |
1da177e4 | 6002 | |
9d48739e | 6003 | pr_info("md: running: "); |
1da177e4 | 6004 | |
dafb20fa | 6005 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6006 | char b[BDEVNAME_SIZE]; |
9d48739e | 6007 | pr_cont("<%s>", bdevname(rdev->bdev,b)); |
1da177e4 | 6008 | } |
9d48739e | 6009 | pr_cont("\n"); |
1da177e4 | 6010 | |
d710e138 | 6011 | err = do_md_run(mddev); |
1da177e4 | 6012 | if (err) { |
9d48739e | 6013 | pr_warn("md: do_md_run() returned %d\n", err); |
a05b7ea0 | 6014 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
6015 | } |
6016 | } | |
6017 | ||
6018 | /* | |
6019 | * lets try to run arrays based on all disks that have arrived | |
6020 | * until now. (those are in pending_raid_disks) | |
6021 | * | |
6022 | * the method: pick the first pending disk, collect all disks with | |
6023 | * the same UUID, remove all from the pending list and put them into | |
6024 | * the 'same_array' list. Then order this list based on superblock | |
6025 | * update time (freshest comes first), kick out 'old' disks and | |
6026 | * compare superblocks. If everything's fine then run it. | |
6027 | * | |
6028 | * If "unit" is allocated, then bump its reference count | |
6029 | */ | |
6030 | static void autorun_devices(int part) | |
6031 | { | |
3cb03002 | 6032 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 6033 | struct mddev *mddev; |
1da177e4 LT |
6034 | char b[BDEVNAME_SIZE]; |
6035 | ||
9d48739e | 6036 | pr_info("md: autorun ...\n"); |
1da177e4 | 6037 | while (!list_empty(&pending_raid_disks)) { |
e8703fe1 | 6038 | int unit; |
1da177e4 | 6039 | dev_t dev; |
ad01c9e3 | 6040 | LIST_HEAD(candidates); |
1da177e4 | 6041 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 6042 | struct md_rdev, same_set); |
1da177e4 | 6043 | |
9d48739e | 6044 | pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b)); |
1da177e4 | 6045 | INIT_LIST_HEAD(&candidates); |
159ec1fc | 6046 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 | 6047 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
9d48739e N |
6048 | pr_debug("md: adding %s ...\n", |
6049 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
6050 | list_move(&rdev->same_set, &candidates); |
6051 | } | |
6052 | /* | |
6053 | * now we have a set of devices, with all of them having | |
6054 | * mostly sane superblocks. It's time to allocate the | |
6055 | * mddev. | |
6056 | */ | |
e8703fe1 N |
6057 | if (part) { |
6058 | dev = MKDEV(mdp_major, | |
6059 | rdev0->preferred_minor << MdpMinorShift); | |
6060 | unit = MINOR(dev) >> MdpMinorShift; | |
6061 | } else { | |
6062 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
6063 | unit = MINOR(dev); | |
6064 | } | |
6065 | if (rdev0->preferred_minor != unit) { | |
9d48739e N |
6066 | pr_warn("md: unit number in %s is bad: %d\n", |
6067 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
1da177e4 LT |
6068 | break; |
6069 | } | |
1da177e4 LT |
6070 | |
6071 | md_probe(dev, NULL, NULL); | |
6072 | mddev = mddev_find(dev); | |
9bbbca3a NB |
6073 | if (!mddev || !mddev->gendisk) { |
6074 | if (mddev) | |
6075 | mddev_put(mddev); | |
1da177e4 LT |
6076 | break; |
6077 | } | |
f72ffdd6 | 6078 | if (mddev_lock(mddev)) |
9d48739e | 6079 | pr_warn("md: %s locked, cannot run\n", mdname(mddev)); |
1da177e4 LT |
6080 | else if (mddev->raid_disks || mddev->major_version |
6081 | || !list_empty(&mddev->disks)) { | |
9d48739e | 6082 | pr_warn("md: %s already running, cannot run %s\n", |
1da177e4 LT |
6083 | mdname(mddev), bdevname(rdev0->bdev,b)); |
6084 | mddev_unlock(mddev); | |
6085 | } else { | |
9d48739e | 6086 | pr_debug("md: created %s\n", mdname(mddev)); |
1ec4a939 | 6087 | mddev->persistent = 1; |
159ec1fc | 6088 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
6089 | list_del_init(&rdev->same_set); |
6090 | if (bind_rdev_to_array(rdev, mddev)) | |
6091 | export_rdev(rdev); | |
6092 | } | |
6093 | autorun_array(mddev); | |
6094 | mddev_unlock(mddev); | |
6095 | } | |
6096 | /* on success, candidates will be empty, on error | |
6097 | * it won't... | |
6098 | */ | |
159ec1fc | 6099 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 6100 | list_del_init(&rdev->same_set); |
1da177e4 | 6101 | export_rdev(rdev); |
4b80991c | 6102 | } |
1da177e4 LT |
6103 | mddev_put(mddev); |
6104 | } | |
9d48739e | 6105 | pr_info("md: ... autorun DONE.\n"); |
1da177e4 | 6106 | } |
fdee8ae4 | 6107 | #endif /* !MODULE */ |
1da177e4 | 6108 | |
f72ffdd6 | 6109 | static int get_version(void __user *arg) |
1da177e4 LT |
6110 | { |
6111 | mdu_version_t ver; | |
6112 | ||
6113 | ver.major = MD_MAJOR_VERSION; | |
6114 | ver.minor = MD_MINOR_VERSION; | |
6115 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
6116 | ||
6117 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
6118 | return -EFAULT; | |
6119 | ||
6120 | return 0; | |
6121 | } | |
6122 | ||
f72ffdd6 | 6123 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
6124 | { |
6125 | mdu_array_info_t info; | |
a9f326eb | 6126 | int nr,working,insync,failed,spare; |
3cb03002 | 6127 | struct md_rdev *rdev; |
1da177e4 | 6128 | |
1ca69c4b N |
6129 | nr = working = insync = failed = spare = 0; |
6130 | rcu_read_lock(); | |
6131 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6132 | nr++; |
b2d444d7 | 6133 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
6134 | failed++; |
6135 | else { | |
6136 | working++; | |
b2d444d7 | 6137 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 6138 | insync++; |
b347af81 SL |
6139 | else if (test_bit(Journal, &rdev->flags)) |
6140 | /* TODO: add journal count to md_u.h */ | |
6141 | ; | |
1da177e4 LT |
6142 | else |
6143 | spare++; | |
6144 | } | |
6145 | } | |
1ca69c4b | 6146 | rcu_read_unlock(); |
1da177e4 LT |
6147 | |
6148 | info.major_version = mddev->major_version; | |
6149 | info.minor_version = mddev->minor_version; | |
6150 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6151 | info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 6152 | info.level = mddev->level; |
58c0fed4 AN |
6153 | info.size = mddev->dev_sectors / 2; |
6154 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 6155 | info.size = -1; |
1da177e4 LT |
6156 | info.nr_disks = nr; |
6157 | info.raid_disks = mddev->raid_disks; | |
6158 | info.md_minor = mddev->md_minor; | |
6159 | info.not_persistent= !mddev->persistent; | |
6160 | ||
9ebc6ef1 | 6161 | info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
6162 | info.state = 0; |
6163 | if (mddev->in_sync) | |
6164 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 6165 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 6166 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
ca8895d9 GR |
6167 | if (mddev_is_clustered(mddev)) |
6168 | info.state |= (1<<MD_SB_CLUSTERED); | |
a9f326eb | 6169 | info.active_disks = insync; |
1da177e4 LT |
6170 | info.working_disks = working; |
6171 | info.failed_disks = failed; | |
6172 | info.spare_disks = spare; | |
6173 | ||
6174 | info.layout = mddev->layout; | |
9d8f0363 | 6175 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
6176 | |
6177 | if (copy_to_user(arg, &info, sizeof(info))) | |
6178 | return -EFAULT; | |
6179 | ||
6180 | return 0; | |
6181 | } | |
6182 | ||
f72ffdd6 | 6183 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
6184 | { |
6185 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 6186 | char *ptr; |
4af1a041 | 6187 | int err; |
32a7627c | 6188 | |
b6878d9e | 6189 | file = kzalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 6190 | if (!file) |
4af1a041 | 6191 | return -ENOMEM; |
32a7627c | 6192 | |
4af1a041 N |
6193 | err = 0; |
6194 | spin_lock(&mddev->lock); | |
25eafe1a BR |
6195 | /* bitmap enabled */ |
6196 | if (mddev->bitmap_info.file) { | |
6197 | ptr = file_path(mddev->bitmap_info.file, file->pathname, | |
6198 | sizeof(file->pathname)); | |
6199 | if (IS_ERR(ptr)) | |
6200 | err = PTR_ERR(ptr); | |
6201 | else | |
6202 | memmove(file->pathname, ptr, | |
6203 | sizeof(file->pathname)-(ptr-file->pathname)); | |
6204 | } | |
4af1a041 | 6205 | spin_unlock(&mddev->lock); |
32a7627c | 6206 | |
4af1a041 N |
6207 | if (err == 0 && |
6208 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 6209 | err = -EFAULT; |
4af1a041 | 6210 | |
32a7627c N |
6211 | kfree(file); |
6212 | return err; | |
6213 | } | |
6214 | ||
f72ffdd6 | 6215 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
6216 | { |
6217 | mdu_disk_info_t info; | |
3cb03002 | 6218 | struct md_rdev *rdev; |
1da177e4 LT |
6219 | |
6220 | if (copy_from_user(&info, arg, sizeof(info))) | |
6221 | return -EFAULT; | |
6222 | ||
1ca69c4b | 6223 | rcu_read_lock(); |
57d051dc | 6224 | rdev = md_find_rdev_nr_rcu(mddev, info.number); |
1da177e4 LT |
6225 | if (rdev) { |
6226 | info.major = MAJOR(rdev->bdev->bd_dev); | |
6227 | info.minor = MINOR(rdev->bdev->bd_dev); | |
6228 | info.raid_disk = rdev->raid_disk; | |
6229 | info.state = 0; | |
b2d444d7 | 6230 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6231 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 6232 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
6233 | info.state |= (1<<MD_DISK_ACTIVE); |
6234 | info.state |= (1<<MD_DISK_SYNC); | |
6235 | } | |
9efdca16 | 6236 | if (test_bit(Journal, &rdev->flags)) |
bac624f3 | 6237 | info.state |= (1<<MD_DISK_JOURNAL); |
8ddf9efe N |
6238 | if (test_bit(WriteMostly, &rdev->flags)) |
6239 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
6240 | if (test_bit(FailFast, &rdev->flags)) |
6241 | info.state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 LT |
6242 | } else { |
6243 | info.major = info.minor = 0; | |
6244 | info.raid_disk = -1; | |
6245 | info.state = (1<<MD_DISK_REMOVED); | |
6246 | } | |
1ca69c4b | 6247 | rcu_read_unlock(); |
1da177e4 LT |
6248 | |
6249 | if (copy_to_user(arg, &info, sizeof(info))) | |
6250 | return -EFAULT; | |
6251 | ||
6252 | return 0; | |
6253 | } | |
6254 | ||
f72ffdd6 | 6255 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
6256 | { |
6257 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 6258 | struct md_rdev *rdev; |
1da177e4 LT |
6259 | dev_t dev = MKDEV(info->major,info->minor); |
6260 | ||
1aee41f6 GR |
6261 | if (mddev_is_clustered(mddev) && |
6262 | !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) { | |
9d48739e N |
6263 | pr_warn("%s: Cannot add to clustered mddev.\n", |
6264 | mdname(mddev)); | |
1aee41f6 GR |
6265 | return -EINVAL; |
6266 | } | |
6267 | ||
1da177e4 LT |
6268 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) |
6269 | return -EOVERFLOW; | |
6270 | ||
6271 | if (!mddev->raid_disks) { | |
6272 | int err; | |
6273 | /* expecting a device which has a superblock */ | |
6274 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
6275 | if (IS_ERR(rdev)) { | |
9d48739e | 6276 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6277 | PTR_ERR(rdev)); |
6278 | return PTR_ERR(rdev); | |
6279 | } | |
6280 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
6281 | struct md_rdev *rdev0 |
6282 | = list_entry(mddev->disks.next, | |
6283 | struct md_rdev, same_set); | |
a9f326eb | 6284 | err = super_types[mddev->major_version] |
1da177e4 LT |
6285 | .load_super(rdev, rdev0, mddev->minor_version); |
6286 | if (err < 0) { | |
9d48739e | 6287 | pr_warn("md: %s has different UUID to %s\n", |
f72ffdd6 | 6288 | bdevname(rdev->bdev,b), |
1da177e4 LT |
6289 | bdevname(rdev0->bdev,b2)); |
6290 | export_rdev(rdev); | |
6291 | return -EINVAL; | |
6292 | } | |
6293 | } | |
6294 | err = bind_rdev_to_array(rdev, mddev); | |
6295 | if (err) | |
6296 | export_rdev(rdev); | |
6297 | return err; | |
6298 | } | |
6299 | ||
6300 | /* | |
6301 | * add_new_disk can be used once the array is assembled | |
6302 | * to add "hot spares". They must already have a superblock | |
6303 | * written | |
6304 | */ | |
6305 | if (mddev->pers) { | |
6306 | int err; | |
6307 | if (!mddev->pers->hot_add_disk) { | |
9d48739e N |
6308 | pr_warn("%s: personality does not support diskops!\n", |
6309 | mdname(mddev)); | |
1da177e4 LT |
6310 | return -EINVAL; |
6311 | } | |
7b1e35f6 N |
6312 | if (mddev->persistent) |
6313 | rdev = md_import_device(dev, mddev->major_version, | |
6314 | mddev->minor_version); | |
6315 | else | |
6316 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 6317 | if (IS_ERR(rdev)) { |
9d48739e | 6318 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6319 | PTR_ERR(rdev)); |
6320 | return PTR_ERR(rdev); | |
6321 | } | |
1a855a06 | 6322 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
6323 | if (!mddev->persistent) { |
6324 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 6325 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 6326 | rdev->raid_disk = info->raid_disk; |
bf572541 | 6327 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 6328 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 6329 | } else |
41158c7e | 6330 | rdev->raid_disk = -1; |
f466722c | 6331 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
6332 | } else |
6333 | super_types[mddev->major_version]. | |
6334 | validate_super(mddev, rdev); | |
bedd86b7 | 6335 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 6336 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
6337 | /* This was a hot-add request, but events doesn't |
6338 | * match, so reject it. | |
6339 | */ | |
6340 | export_rdev(rdev); | |
6341 | return -EINVAL; | |
6342 | } | |
6343 | ||
b2d444d7 | 6344 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
6345 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6346 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
6347 | else |
6348 | clear_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6349 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6350 | set_bit(FailFast, &rdev->flags); | |
6351 | else | |
6352 | clear_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6353 | |
f6b6ec5c SL |
6354 | if (info->state & (1<<MD_DISK_JOURNAL)) { |
6355 | struct md_rdev *rdev2; | |
6356 | bool has_journal = false; | |
6357 | ||
6358 | /* make sure no existing journal disk */ | |
6359 | rdev_for_each(rdev2, mddev) { | |
6360 | if (test_bit(Journal, &rdev2->flags)) { | |
6361 | has_journal = true; | |
6362 | break; | |
6363 | } | |
6364 | } | |
6365 | if (has_journal) { | |
6366 | export_rdev(rdev); | |
6367 | return -EBUSY; | |
6368 | } | |
bac624f3 | 6369 | set_bit(Journal, &rdev->flags); |
f6b6ec5c | 6370 | } |
1aee41f6 GR |
6371 | /* |
6372 | * check whether the device shows up in other nodes | |
6373 | */ | |
6374 | if (mddev_is_clustered(mddev)) { | |
dbb64f86 | 6375 | if (info->state & (1 << MD_DISK_CANDIDATE)) |
1aee41f6 | 6376 | set_bit(Candidate, &rdev->flags); |
dbb64f86 | 6377 | else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) { |
1aee41f6 | 6378 | /* --add initiated by this node */ |
dbb64f86 | 6379 | err = md_cluster_ops->add_new_disk(mddev, rdev); |
1aee41f6 | 6380 | if (err) { |
1aee41f6 GR |
6381 | export_rdev(rdev); |
6382 | return err; | |
6383 | } | |
6384 | } | |
6385 | } | |
6386 | ||
1da177e4 LT |
6387 | rdev->raid_disk = -1; |
6388 | err = bind_rdev_to_array(rdev, mddev); | |
dbb64f86 | 6389 | |
1da177e4 LT |
6390 | if (err) |
6391 | export_rdev(rdev); | |
dbb64f86 GR |
6392 | |
6393 | if (mddev_is_clustered(mddev)) { | |
e566aef1 GJ |
6394 | if (info->state & (1 << MD_DISK_CANDIDATE)) { |
6395 | if (!err) { | |
6396 | err = md_cluster_ops->new_disk_ack(mddev, | |
6397 | err == 0); | |
6398 | if (err) | |
6399 | md_kick_rdev_from_array(rdev); | |
6400 | } | |
6401 | } else { | |
dbb64f86 GR |
6402 | if (err) |
6403 | md_cluster_ops->add_new_disk_cancel(mddev); | |
6404 | else | |
6405 | err = add_bound_rdev(rdev); | |
6406 | } | |
6407 | ||
6408 | } else if (!err) | |
a6da4ef8 | 6409 | err = add_bound_rdev(rdev); |
dbb64f86 | 6410 | |
1da177e4 LT |
6411 | return err; |
6412 | } | |
6413 | ||
6414 | /* otherwise, add_new_disk is only allowed | |
6415 | * for major_version==0 superblocks | |
6416 | */ | |
6417 | if (mddev->major_version != 0) { | |
9d48739e | 6418 | pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev)); |
1da177e4 LT |
6419 | return -EINVAL; |
6420 | } | |
6421 | ||
6422 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
6423 | int err; | |
d710e138 | 6424 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6425 | if (IS_ERR(rdev)) { |
9d48739e | 6426 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6427 | PTR_ERR(rdev)); |
6428 | return PTR_ERR(rdev); | |
6429 | } | |
6430 | rdev->desc_nr = info->number; | |
6431 | if (info->raid_disk < mddev->raid_disks) | |
6432 | rdev->raid_disk = info->raid_disk; | |
6433 | else | |
6434 | rdev->raid_disk = -1; | |
6435 | ||
1da177e4 | 6436 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
6437 | if (info->state & (1<<MD_DISK_SYNC)) |
6438 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 6439 | |
8ddf9efe N |
6440 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6441 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6442 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6443 | set_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6444 | |
1da177e4 | 6445 | if (!mddev->persistent) { |
9d48739e | 6446 | pr_debug("md: nonpersistent superblock ...\n"); |
77304d2a MS |
6447 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
6448 | } else | |
57b2caa3 | 6449 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 6450 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6451 | |
2bf071bf N |
6452 | err = bind_rdev_to_array(rdev, mddev); |
6453 | if (err) { | |
6454 | export_rdev(rdev); | |
6455 | return err; | |
6456 | } | |
1da177e4 LT |
6457 | } |
6458 | ||
6459 | return 0; | |
6460 | } | |
6461 | ||
f72ffdd6 | 6462 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6463 | { |
6464 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 6465 | struct md_rdev *rdev; |
1da177e4 | 6466 | |
1da177e4 LT |
6467 | rdev = find_rdev(mddev, dev); |
6468 | if (!rdev) | |
6469 | return -ENXIO; | |
6470 | ||
2910ff17 GR |
6471 | if (rdev->raid_disk < 0) |
6472 | goto kick_rdev; | |
293467aa | 6473 | |
3ea8929d N |
6474 | clear_bit(Blocked, &rdev->flags); |
6475 | remove_and_add_spares(mddev, rdev); | |
6476 | ||
1da177e4 LT |
6477 | if (rdev->raid_disk >= 0) |
6478 | goto busy; | |
6479 | ||
2910ff17 | 6480 | kick_rdev: |
54a88392 | 6481 | if (mddev_is_clustered(mddev)) |
88bcfef7 GR |
6482 | md_cluster_ops->remove_disk(mddev, rdev); |
6483 | ||
fb56dfef | 6484 | md_kick_rdev_from_array(rdev); |
2953079c | 6485 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6486 | if (mddev->thread) |
6487 | md_wakeup_thread(mddev->thread); | |
6488 | else | |
6489 | md_update_sb(mddev, 1); | |
d7603b7e | 6490 | md_new_event(mddev); |
1da177e4 LT |
6491 | |
6492 | return 0; | |
6493 | busy: | |
9d48739e N |
6494 | pr_debug("md: cannot remove active disk %s from %s ...\n", |
6495 | bdevname(rdev->bdev,b), mdname(mddev)); | |
1da177e4 LT |
6496 | return -EBUSY; |
6497 | } | |
6498 | ||
f72ffdd6 | 6499 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6500 | { |
6501 | char b[BDEVNAME_SIZE]; | |
6502 | int err; | |
3cb03002 | 6503 | struct md_rdev *rdev; |
1da177e4 LT |
6504 | |
6505 | if (!mddev->pers) | |
6506 | return -ENODEV; | |
6507 | ||
6508 | if (mddev->major_version != 0) { | |
9d48739e | 6509 | pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n", |
1da177e4 LT |
6510 | mdname(mddev)); |
6511 | return -EINVAL; | |
6512 | } | |
6513 | if (!mddev->pers->hot_add_disk) { | |
9d48739e | 6514 | pr_warn("%s: personality does not support diskops!\n", |
1da177e4 LT |
6515 | mdname(mddev)); |
6516 | return -EINVAL; | |
6517 | } | |
6518 | ||
d710e138 | 6519 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6520 | if (IS_ERR(rdev)) { |
9d48739e | 6521 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6522 | PTR_ERR(rdev)); |
6523 | return -EINVAL; | |
6524 | } | |
6525 | ||
6526 | if (mddev->persistent) | |
57b2caa3 | 6527 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 6528 | else |
77304d2a | 6529 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 6530 | |
8190e754 | 6531 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6532 | |
b2d444d7 | 6533 | if (test_bit(Faulty, &rdev->flags)) { |
9d48739e | 6534 | pr_warn("md: can not hot-add faulty %s disk to %s!\n", |
1da177e4 LT |
6535 | bdevname(rdev->bdev,b), mdname(mddev)); |
6536 | err = -EINVAL; | |
6537 | goto abort_export; | |
6538 | } | |
293467aa | 6539 | |
b2d444d7 | 6540 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6541 | rdev->desc_nr = -1; |
5842730d | 6542 | rdev->saved_raid_disk = -1; |
2bf071bf N |
6543 | err = bind_rdev_to_array(rdev, mddev); |
6544 | if (err) | |
2aa82191 | 6545 | goto abort_export; |
1da177e4 LT |
6546 | |
6547 | /* | |
6548 | * The rest should better be atomic, we can have disk failures | |
6549 | * noticed in interrupt contexts ... | |
6550 | */ | |
6551 | ||
1da177e4 LT |
6552 | rdev->raid_disk = -1; |
6553 | ||
2953079c | 6554 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6555 | if (!mddev->thread) |
6556 | md_update_sb(mddev, 1); | |
1da177e4 LT |
6557 | /* |
6558 | * Kick recovery, maybe this spare has to be added to the | |
6559 | * array immediately. | |
6560 | */ | |
6561 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6562 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 6563 | md_new_event(mddev); |
1da177e4 LT |
6564 | return 0; |
6565 | ||
1da177e4 LT |
6566 | abort_export: |
6567 | export_rdev(rdev); | |
6568 | return err; | |
6569 | } | |
6570 | ||
fd01b88c | 6571 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 6572 | { |
035328c2 | 6573 | int err = 0; |
32a7627c | 6574 | |
36fa3063 | 6575 | if (mddev->pers) { |
d66b1b39 | 6576 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
6577 | return -EBUSY; |
6578 | if (mddev->recovery || mddev->sync_thread) | |
6579 | return -EBUSY; | |
6580 | /* we should be able to change the bitmap.. */ | |
6581 | } | |
32a7627c | 6582 | |
36fa3063 | 6583 | if (fd >= 0) { |
035328c2 | 6584 | struct inode *inode; |
1e594bb2 N |
6585 | struct file *f; |
6586 | ||
6587 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 6588 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6589 | f = fget(fd); |
32a7627c | 6590 | |
1e594bb2 | 6591 | if (f == NULL) { |
9d48739e N |
6592 | pr_warn("%s: error: failed to get bitmap file\n", |
6593 | mdname(mddev)); | |
36fa3063 N |
6594 | return -EBADF; |
6595 | } | |
6596 | ||
1e594bb2 | 6597 | inode = f->f_mapping->host; |
035328c2 | 6598 | if (!S_ISREG(inode->i_mode)) { |
9d48739e N |
6599 | pr_warn("%s: error: bitmap file must be a regular file\n", |
6600 | mdname(mddev)); | |
035328c2 | 6601 | err = -EBADF; |
1e594bb2 | 6602 | } else if (!(f->f_mode & FMODE_WRITE)) { |
9d48739e N |
6603 | pr_warn("%s: error: bitmap file must open for write\n", |
6604 | mdname(mddev)); | |
035328c2 N |
6605 | err = -EBADF; |
6606 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
9d48739e N |
6607 | pr_warn("%s: error: bitmap file is already in use\n", |
6608 | mdname(mddev)); | |
035328c2 N |
6609 | err = -EBUSY; |
6610 | } | |
6611 | if (err) { | |
1e594bb2 | 6612 | fput(f); |
36fa3063 N |
6613 | return err; |
6614 | } | |
1e594bb2 | 6615 | mddev->bitmap_info.file = f; |
c3d9714e | 6616 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6617 | } else if (mddev->bitmap == NULL) |
6618 | return -ENOENT; /* cannot remove what isn't there */ | |
6619 | err = 0; | |
6620 | if (mddev->pers) { | |
6621 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6622 | if (fd >= 0) { |
f9209a32 GR |
6623 | struct bitmap *bitmap; |
6624 | ||
6625 | bitmap = bitmap_create(mddev, -1); | |
6626 | if (!IS_ERR(bitmap)) { | |
6627 | mddev->bitmap = bitmap; | |
69e51b44 | 6628 | err = bitmap_load(mddev); |
ba599aca N |
6629 | } else |
6630 | err = PTR_ERR(bitmap); | |
69e51b44 | 6631 | } |
d7375ab3 | 6632 | if (fd < 0 || err) { |
36fa3063 | 6633 | bitmap_destroy(mddev); |
d7375ab3 N |
6634 | fd = -1; /* make sure to put the file */ |
6635 | } | |
36fa3063 | 6636 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6637 | } |
6638 | if (fd < 0) { | |
4af1a041 N |
6639 | struct file *f = mddev->bitmap_info.file; |
6640 | if (f) { | |
6641 | spin_lock(&mddev->lock); | |
6642 | mddev->bitmap_info.file = NULL; | |
6643 | spin_unlock(&mddev->lock); | |
6644 | fput(f); | |
6645 | } | |
36fa3063 N |
6646 | } |
6647 | ||
32a7627c N |
6648 | return err; |
6649 | } | |
6650 | ||
1da177e4 LT |
6651 | /* |
6652 | * set_array_info is used two different ways | |
6653 | * The original usage is when creating a new array. | |
6654 | * In this usage, raid_disks is > 0 and it together with | |
6655 | * level, size, not_persistent,layout,chunksize determine the | |
6656 | * shape of the array. | |
6657 | * This will always create an array with a type-0.90.0 superblock. | |
6658 | * The newer usage is when assembling an array. | |
6659 | * In this case raid_disks will be 0, and the major_version field is | |
6660 | * use to determine which style super-blocks are to be found on the devices. | |
6661 | * The minor and patch _version numbers are also kept incase the | |
6662 | * super_block handler wishes to interpret them. | |
6663 | */ | |
f72ffdd6 | 6664 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6665 | { |
6666 | ||
6667 | if (info->raid_disks == 0) { | |
6668 | /* just setting version number for superblock loading */ | |
6669 | if (info->major_version < 0 || | |
50511da3 | 6670 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6671 | super_types[info->major_version].name == NULL) { |
6672 | /* maybe try to auto-load a module? */ | |
9d48739e | 6673 | pr_warn("md: superblock version %d not known\n", |
1da177e4 LT |
6674 | info->major_version); |
6675 | return -EINVAL; | |
6676 | } | |
6677 | mddev->major_version = info->major_version; | |
6678 | mddev->minor_version = info->minor_version; | |
6679 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6680 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6681 | /* ensure mddev_put doesn't delete this now that there |
6682 | * is some minimal configuration. | |
6683 | */ | |
9ebc6ef1 | 6684 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6685 | return 0; |
6686 | } | |
6687 | mddev->major_version = MD_MAJOR_VERSION; | |
6688 | mddev->minor_version = MD_MINOR_VERSION; | |
6689 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6690 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6691 | |
6692 | mddev->level = info->level; | |
17115e03 | 6693 | mddev->clevel[0] = 0; |
58c0fed4 | 6694 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6695 | mddev->raid_disks = info->raid_disks; |
6696 | /* don't set md_minor, it is determined by which /dev/md* was | |
6697 | * openned | |
6698 | */ | |
6699 | if (info->state & (1<<MD_SB_CLEAN)) | |
6700 | mddev->recovery_cp = MaxSector; | |
6701 | else | |
6702 | mddev->recovery_cp = 0; | |
6703 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6704 | mddev->external = 0; |
1da177e4 LT |
6705 | |
6706 | mddev->layout = info->layout; | |
9d8f0363 | 6707 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 | 6708 | |
2953079c | 6709 | if (mddev->persistent) { |
1b3bae49 N |
6710 | mddev->max_disks = MD_SB_DISKS; |
6711 | mddev->flags = 0; | |
6712 | mddev->sb_flags = 0; | |
2953079c SL |
6713 | } |
6714 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
1da177e4 | 6715 | |
c3d9714e | 6716 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6717 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6718 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6719 | |
f6705578 N |
6720 | mddev->reshape_position = MaxSector; |
6721 | ||
1da177e4 LT |
6722 | /* |
6723 | * Generate a 128 bit UUID | |
6724 | */ | |
6725 | get_random_bytes(mddev->uuid, 16); | |
6726 | ||
f6705578 | 6727 | mddev->new_level = mddev->level; |
664e7c41 | 6728 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6729 | mddev->new_layout = mddev->layout; |
6730 | mddev->delta_disks = 0; | |
2c810cdd | 6731 | mddev->reshape_backwards = 0; |
f6705578 | 6732 | |
1da177e4 LT |
6733 | return 0; |
6734 | } | |
6735 | ||
fd01b88c | 6736 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6737 | { |
b522adcd DW |
6738 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6739 | ||
6740 | if (mddev->external_size) | |
6741 | return; | |
6742 | ||
1f403624 DW |
6743 | mddev->array_sectors = array_sectors; |
6744 | } | |
6745 | EXPORT_SYMBOL(md_set_array_sectors); | |
6746 | ||
fd01b88c | 6747 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6748 | { |
3cb03002 | 6749 | struct md_rdev *rdev; |
a35b0d69 | 6750 | int rv; |
d71f9f88 | 6751 | int fit = (num_sectors == 0); |
818da59f | 6752 | sector_t old_dev_sectors = mddev->dev_sectors; |
ab5a98b1 | 6753 | |
a35b0d69 N |
6754 | if (mddev->pers->resize == NULL) |
6755 | return -EINVAL; | |
d71f9f88 AN |
6756 | /* The "num_sectors" is the number of sectors of each device that |
6757 | * is used. This can only make sense for arrays with redundancy. | |
6758 | * linear and raid0 always use whatever space is available. We can only | |
6759 | * consider changing this number if no resync or reconstruction is | |
6760 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6761 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6762 | * of each device. If num_sectors is zero, we find the largest size |
6763 | * that fits. | |
a35b0d69 | 6764 | */ |
f851b60d N |
6765 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6766 | mddev->sync_thread) | |
a35b0d69 | 6767 | return -EBUSY; |
bd8839e0 N |
6768 | if (mddev->ro) |
6769 | return -EROFS; | |
a4a6125a | 6770 | |
dafb20fa | 6771 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6772 | sector_t avail = rdev->sectors; |
01ab5662 | 6773 | |
d71f9f88 AN |
6774 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6775 | num_sectors = avail; | |
6776 | if (avail < num_sectors) | |
a35b0d69 N |
6777 | return -ENOSPC; |
6778 | } | |
d71f9f88 | 6779 | rv = mddev->pers->resize(mddev, num_sectors); |
c9483634 | 6780 | if (!rv) { |
818da59f GJ |
6781 | if (mddev_is_clustered(mddev)) |
6782 | md_cluster_ops->update_size(mddev, old_dev_sectors); | |
6783 | else if (mddev->queue) { | |
c9483634 GJ |
6784 | set_capacity(mddev->gendisk, mddev->array_sectors); |
6785 | revalidate_disk(mddev->gendisk); | |
6786 | } | |
6787 | } | |
a35b0d69 N |
6788 | return rv; |
6789 | } | |
6790 | ||
fd01b88c | 6791 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6792 | { |
6793 | int rv; | |
c6563a8c | 6794 | struct md_rdev *rdev; |
da943b99 | 6795 | /* change the number of raid disks */ |
63c70c4f | 6796 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6797 | return -EINVAL; |
bd8839e0 N |
6798 | if (mddev->ro) |
6799 | return -EROFS; | |
da943b99 | 6800 | if (raid_disks <= 0 || |
233fca36 | 6801 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6802 | return -EINVAL; |
f851b60d N |
6803 | if (mddev->sync_thread || |
6804 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6805 | mddev->reshape_position != MaxSector) | |
da943b99 | 6806 | return -EBUSY; |
c6563a8c N |
6807 | |
6808 | rdev_for_each(rdev, mddev) { | |
6809 | if (mddev->raid_disks < raid_disks && | |
6810 | rdev->data_offset < rdev->new_data_offset) | |
6811 | return -EINVAL; | |
6812 | if (mddev->raid_disks > raid_disks && | |
6813 | rdev->data_offset > rdev->new_data_offset) | |
6814 | return -EINVAL; | |
6815 | } | |
6816 | ||
63c70c4f | 6817 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6818 | if (mddev->delta_disks < 0) |
6819 | mddev->reshape_backwards = 1; | |
6820 | else if (mddev->delta_disks > 0) | |
6821 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6822 | |
6823 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6824 | if (rv < 0) { |
de171cb9 | 6825 | mddev->delta_disks = 0; |
2c810cdd N |
6826 | mddev->reshape_backwards = 0; |
6827 | } | |
da943b99 N |
6828 | return rv; |
6829 | } | |
6830 | ||
1da177e4 LT |
6831 | /* |
6832 | * update_array_info is used to change the configuration of an | |
6833 | * on-line array. | |
6834 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6835 | * fields in the info are checked against the array. | |
6836 | * Any differences that cannot be handled will cause an error. | |
6837 | * Normally, only one change can be managed at a time. | |
6838 | */ | |
fd01b88c | 6839 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6840 | { |
6841 | int rv = 0; | |
6842 | int cnt = 0; | |
36fa3063 N |
6843 | int state = 0; |
6844 | ||
6845 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6846 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6847 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6848 | |
6849 | if (mddev->major_version != info->major_version || | |
6850 | mddev->minor_version != info->minor_version || | |
6851 | /* mddev->patch_version != info->patch_version || */ | |
6852 | mddev->ctime != info->ctime || | |
6853 | mddev->level != info->level || | |
6854 | /* mddev->layout != info->layout || */ | |
4e023612 | 6855 | mddev->persistent != !info->not_persistent || |
9d8f0363 | 6856 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6857 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6858 | ((state^info->state) & 0xfffffe00) | |
6859 | ) | |
1da177e4 LT |
6860 | return -EINVAL; |
6861 | /* Check there is only one change */ | |
58c0fed4 AN |
6862 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6863 | cnt++; | |
6864 | if (mddev->raid_disks != info->raid_disks) | |
6865 | cnt++; | |
6866 | if (mddev->layout != info->layout) | |
6867 | cnt++; | |
6868 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6869 | cnt++; | |
6870 | if (cnt == 0) | |
6871 | return 0; | |
6872 | if (cnt > 1) | |
6873 | return -EINVAL; | |
1da177e4 LT |
6874 | |
6875 | if (mddev->layout != info->layout) { | |
6876 | /* Change layout | |
6877 | * we don't need to do anything at the md level, the | |
6878 | * personality will take care of it all. | |
6879 | */ | |
50ac168a | 6880 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6881 | return -EINVAL; |
597a711b N |
6882 | else { |
6883 | mddev->new_layout = info->layout; | |
50ac168a | 6884 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6885 | if (rv) |
6886 | mddev->new_layout = mddev->layout; | |
6887 | return rv; | |
6888 | } | |
1da177e4 | 6889 | } |
58c0fed4 | 6890 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6891 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6892 | |
da943b99 N |
6893 | if (mddev->raid_disks != info->raid_disks) |
6894 | rv = update_raid_disks(mddev, info->raid_disks); | |
6895 | ||
36fa3063 | 6896 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
293467aa GR |
6897 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) { |
6898 | rv = -EINVAL; | |
6899 | goto err; | |
6900 | } | |
6901 | if (mddev->recovery || mddev->sync_thread) { | |
6902 | rv = -EBUSY; | |
6903 | goto err; | |
6904 | } | |
36fa3063 | 6905 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { |
f9209a32 | 6906 | struct bitmap *bitmap; |
36fa3063 | 6907 | /* add the bitmap */ |
293467aa GR |
6908 | if (mddev->bitmap) { |
6909 | rv = -EEXIST; | |
6910 | goto err; | |
6911 | } | |
6912 | if (mddev->bitmap_info.default_offset == 0) { | |
6913 | rv = -EINVAL; | |
6914 | goto err; | |
6915 | } | |
c3d9714e N |
6916 | mddev->bitmap_info.offset = |
6917 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6918 | mddev->bitmap_info.space = |
6919 | mddev->bitmap_info.default_space; | |
36fa3063 | 6920 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
6921 | bitmap = bitmap_create(mddev, -1); |
6922 | if (!IS_ERR(bitmap)) { | |
6923 | mddev->bitmap = bitmap; | |
69e51b44 | 6924 | rv = bitmap_load(mddev); |
ba599aca N |
6925 | } else |
6926 | rv = PTR_ERR(bitmap); | |
36fa3063 N |
6927 | if (rv) |
6928 | bitmap_destroy(mddev); | |
6929 | mddev->pers->quiesce(mddev, 0); | |
6930 | } else { | |
6931 | /* remove the bitmap */ | |
293467aa GR |
6932 | if (!mddev->bitmap) { |
6933 | rv = -ENOENT; | |
6934 | goto err; | |
6935 | } | |
6936 | if (mddev->bitmap->storage.file) { | |
6937 | rv = -EINVAL; | |
6938 | goto err; | |
6939 | } | |
f6a2dc64 GJ |
6940 | if (mddev->bitmap_info.nodes) { |
6941 | /* hold PW on all the bitmap lock */ | |
6942 | if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) { | |
9d48739e | 6943 | pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n"); |
f6a2dc64 GJ |
6944 | rv = -EPERM; |
6945 | md_cluster_ops->unlock_all_bitmaps(mddev); | |
6946 | goto err; | |
6947 | } | |
6948 | ||
6949 | mddev->bitmap_info.nodes = 0; | |
6950 | md_cluster_ops->leave(mddev); | |
6951 | } | |
36fa3063 N |
6952 | mddev->pers->quiesce(mddev, 1); |
6953 | bitmap_destroy(mddev); | |
6954 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6955 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6956 | } |
6957 | } | |
850b2b42 | 6958 | md_update_sb(mddev, 1); |
293467aa GR |
6959 | return rv; |
6960 | err: | |
1da177e4 LT |
6961 | return rv; |
6962 | } | |
6963 | ||
fd01b88c | 6964 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6965 | { |
3cb03002 | 6966 | struct md_rdev *rdev; |
1ca69c4b | 6967 | int err = 0; |
1da177e4 LT |
6968 | |
6969 | if (mddev->pers == NULL) | |
6970 | return -ENODEV; | |
6971 | ||
1ca69c4b N |
6972 | rcu_read_lock(); |
6973 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6974 | if (!rdev) |
1ca69c4b N |
6975 | err = -ENODEV; |
6976 | else { | |
6977 | md_error(mddev, rdev); | |
6978 | if (!test_bit(Faulty, &rdev->flags)) | |
6979 | err = -EBUSY; | |
6980 | } | |
6981 | rcu_read_unlock(); | |
6982 | return err; | |
1da177e4 LT |
6983 | } |
6984 | ||
2f9618ce AN |
6985 | /* |
6986 | * We have a problem here : there is no easy way to give a CHS | |
6987 | * virtual geometry. We currently pretend that we have a 2 heads | |
6988 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6989 | * dosfs just mad... ;-) | |
6990 | */ | |
a885c8c4 CH |
6991 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6992 | { | |
fd01b88c | 6993 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6994 | |
6995 | geo->heads = 2; | |
6996 | geo->sectors = 4; | |
49ce6cea | 6997 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6998 | return 0; |
6999 | } | |
7000 | ||
cb335f88 NS |
7001 | static inline bool md_ioctl_valid(unsigned int cmd) |
7002 | { | |
7003 | switch (cmd) { | |
7004 | case ADD_NEW_DISK: | |
7005 | case BLKROSET: | |
7006 | case GET_ARRAY_INFO: | |
7007 | case GET_BITMAP_FILE: | |
7008 | case GET_DISK_INFO: | |
7009 | case HOT_ADD_DISK: | |
7010 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
7011 | case RAID_AUTORUN: |
7012 | case RAID_VERSION: | |
7013 | case RESTART_ARRAY_RW: | |
7014 | case RUN_ARRAY: | |
7015 | case SET_ARRAY_INFO: | |
7016 | case SET_BITMAP_FILE: | |
7017 | case SET_DISK_FAULTY: | |
7018 | case STOP_ARRAY: | |
7019 | case STOP_ARRAY_RO: | |
1aee41f6 | 7020 | case CLUSTERED_DISK_NACK: |
cb335f88 NS |
7021 | return true; |
7022 | default: | |
7023 | return false; | |
7024 | } | |
7025 | } | |
7026 | ||
a39907fa | 7027 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
7028 | unsigned int cmd, unsigned long arg) |
7029 | { | |
7030 | int err = 0; | |
7031 | void __user *argp = (void __user *)arg; | |
fd01b88c | 7032 | struct mddev *mddev = NULL; |
e2218350 | 7033 | int ro; |
065e519e | 7034 | bool did_set_md_closing = false; |
1da177e4 | 7035 | |
cb335f88 NS |
7036 | if (!md_ioctl_valid(cmd)) |
7037 | return -ENOTTY; | |
7038 | ||
506c9e44 N |
7039 | switch (cmd) { |
7040 | case RAID_VERSION: | |
7041 | case GET_ARRAY_INFO: | |
7042 | case GET_DISK_INFO: | |
7043 | break; | |
7044 | default: | |
7045 | if (!capable(CAP_SYS_ADMIN)) | |
7046 | return -EACCES; | |
7047 | } | |
1da177e4 LT |
7048 | |
7049 | /* | |
7050 | * Commands dealing with the RAID driver but not any | |
7051 | * particular array: | |
7052 | */ | |
c02c0aeb N |
7053 | switch (cmd) { |
7054 | case RAID_VERSION: | |
7055 | err = get_version(argp); | |
3adc28d8 | 7056 | goto out; |
1da177e4 | 7057 | |
1da177e4 | 7058 | #ifndef MODULE |
c02c0aeb N |
7059 | case RAID_AUTORUN: |
7060 | err = 0; | |
7061 | autostart_arrays(arg); | |
3adc28d8 | 7062 | goto out; |
1da177e4 | 7063 | #endif |
c02c0aeb | 7064 | default:; |
1da177e4 LT |
7065 | } |
7066 | ||
7067 | /* | |
7068 | * Commands creating/starting a new array: | |
7069 | */ | |
7070 | ||
a39907fa | 7071 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
7072 | |
7073 | if (!mddev) { | |
7074 | BUG(); | |
3adc28d8 | 7075 | goto out; |
1da177e4 LT |
7076 | } |
7077 | ||
1ca69c4b N |
7078 | /* Some actions do not requires the mutex */ |
7079 | switch (cmd) { | |
7080 | case GET_ARRAY_INFO: | |
7081 | if (!mddev->raid_disks && !mddev->external) | |
7082 | err = -ENODEV; | |
7083 | else | |
7084 | err = get_array_info(mddev, argp); | |
3adc28d8 | 7085 | goto out; |
1ca69c4b N |
7086 | |
7087 | case GET_DISK_INFO: | |
7088 | if (!mddev->raid_disks && !mddev->external) | |
7089 | err = -ENODEV; | |
7090 | else | |
7091 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 7092 | goto out; |
1ca69c4b N |
7093 | |
7094 | case SET_DISK_FAULTY: | |
7095 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7096 | goto out; |
4af1a041 N |
7097 | |
7098 | case GET_BITMAP_FILE: | |
7099 | err = get_bitmap_file(mddev, argp); | |
7100 | goto out; | |
7101 | ||
1ca69c4b N |
7102 | } |
7103 | ||
a7a3f08d N |
7104 | if (cmd == ADD_NEW_DISK) |
7105 | /* need to ensure md_delayed_delete() has completed */ | |
7106 | flush_workqueue(md_misc_wq); | |
7107 | ||
90f5f7ad HR |
7108 | if (cmd == HOT_REMOVE_DISK) |
7109 | /* need to ensure recovery thread has run */ | |
7110 | wait_event_interruptible_timeout(mddev->sb_wait, | |
7111 | !test_bit(MD_RECOVERY_NEEDED, | |
82a301cb | 7112 | &mddev->recovery), |
90f5f7ad | 7113 | msecs_to_jiffies(5000)); |
260fa034 N |
7114 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
7115 | /* Need to flush page cache, and ensure no-one else opens | |
7116 | * and writes | |
7117 | */ | |
7118 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 7119 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
7120 | mutex_unlock(&mddev->open_mutex); |
7121 | err = -EBUSY; | |
3adc28d8 | 7122 | goto out; |
260fa034 | 7123 | } |
065e519e | 7124 | WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags)); |
af8d8e6f | 7125 | set_bit(MD_CLOSING, &mddev->flags); |
065e519e | 7126 | did_set_md_closing = true; |
260fa034 N |
7127 | mutex_unlock(&mddev->open_mutex); |
7128 | sync_blockdev(bdev); | |
7129 | } | |
1da177e4 LT |
7130 | err = mddev_lock(mddev); |
7131 | if (err) { | |
9d48739e N |
7132 | pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n", |
7133 | err, cmd); | |
3adc28d8 | 7134 | goto out; |
1da177e4 LT |
7135 | } |
7136 | ||
c02c0aeb N |
7137 | if (cmd == SET_ARRAY_INFO) { |
7138 | mdu_array_info_t info; | |
7139 | if (!arg) | |
7140 | memset(&info, 0, sizeof(info)); | |
7141 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
7142 | err = -EFAULT; | |
3adc28d8 | 7143 | goto unlock; |
c02c0aeb N |
7144 | } |
7145 | if (mddev->pers) { | |
7146 | err = update_array_info(mddev, &info); | |
7147 | if (err) { | |
9d48739e | 7148 | pr_warn("md: couldn't update array info. %d\n", err); |
3adc28d8 | 7149 | goto unlock; |
1da177e4 | 7150 | } |
3adc28d8 | 7151 | goto unlock; |
c02c0aeb N |
7152 | } |
7153 | if (!list_empty(&mddev->disks)) { | |
9d48739e | 7154 | pr_warn("md: array %s already has disks!\n", mdname(mddev)); |
c02c0aeb | 7155 | err = -EBUSY; |
3adc28d8 | 7156 | goto unlock; |
c02c0aeb N |
7157 | } |
7158 | if (mddev->raid_disks) { | |
9d48739e | 7159 | pr_warn("md: array %s already initialised!\n", mdname(mddev)); |
c02c0aeb | 7160 | err = -EBUSY; |
3adc28d8 | 7161 | goto unlock; |
c02c0aeb N |
7162 | } |
7163 | err = set_array_info(mddev, &info); | |
7164 | if (err) { | |
9d48739e | 7165 | pr_warn("md: couldn't set array info. %d\n", err); |
3adc28d8 | 7166 | goto unlock; |
c02c0aeb | 7167 | } |
3adc28d8 | 7168 | goto unlock; |
1da177e4 LT |
7169 | } |
7170 | ||
7171 | /* | |
7172 | * Commands querying/configuring an existing array: | |
7173 | */ | |
32a7627c | 7174 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 7175 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
7176 | if ((!mddev->raid_disks && !mddev->external) |
7177 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
7178 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
7179 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 7180 | err = -ENODEV; |
3adc28d8 | 7181 | goto unlock; |
1da177e4 LT |
7182 | } |
7183 | ||
7184 | /* | |
7185 | * Commands even a read-only array can execute: | |
7186 | */ | |
c02c0aeb | 7187 | switch (cmd) { |
c02c0aeb N |
7188 | case RESTART_ARRAY_RW: |
7189 | err = restart_array(mddev); | |
3adc28d8 | 7190 | goto unlock; |
1da177e4 | 7191 | |
c02c0aeb N |
7192 | case STOP_ARRAY: |
7193 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 7194 | goto unlock; |
1da177e4 | 7195 | |
c02c0aeb N |
7196 | case STOP_ARRAY_RO: |
7197 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 7198 | goto unlock; |
1da177e4 | 7199 | |
3ea8929d N |
7200 | case HOT_REMOVE_DISK: |
7201 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7202 | goto unlock; |
3ea8929d | 7203 | |
7ceb17e8 N |
7204 | case ADD_NEW_DISK: |
7205 | /* We can support ADD_NEW_DISK on read-only arrays | |
466ad292 | 7206 | * only if we are re-adding a preexisting device. |
7ceb17e8 N |
7207 | * So require mddev->pers and MD_DISK_SYNC. |
7208 | */ | |
7209 | if (mddev->pers) { | |
7210 | mdu_disk_info_t info; | |
7211 | if (copy_from_user(&info, argp, sizeof(info))) | |
7212 | err = -EFAULT; | |
7213 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
7214 | /* Need to clear read-only for this */ | |
7215 | break; | |
7216 | else | |
7217 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7218 | goto unlock; |
7ceb17e8 N |
7219 | } |
7220 | break; | |
7221 | ||
c02c0aeb N |
7222 | case BLKROSET: |
7223 | if (get_user(ro, (int __user *)(arg))) { | |
7224 | err = -EFAULT; | |
3adc28d8 | 7225 | goto unlock; |
c02c0aeb N |
7226 | } |
7227 | err = -EINVAL; | |
e2218350 | 7228 | |
c02c0aeb N |
7229 | /* if the bdev is going readonly the value of mddev->ro |
7230 | * does not matter, no writes are coming | |
7231 | */ | |
7232 | if (ro) | |
3adc28d8 | 7233 | goto unlock; |
e2218350 | 7234 | |
c02c0aeb N |
7235 | /* are we are already prepared for writes? */ |
7236 | if (mddev->ro != 1) | |
3adc28d8 | 7237 | goto unlock; |
e2218350 | 7238 | |
c02c0aeb N |
7239 | /* transitioning to readauto need only happen for |
7240 | * arrays that call md_write_start | |
7241 | */ | |
7242 | if (mddev->pers) { | |
7243 | err = restart_array(mddev); | |
7244 | if (err == 0) { | |
7245 | mddev->ro = 2; | |
7246 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 7247 | } |
c02c0aeb | 7248 | } |
3adc28d8 | 7249 | goto unlock; |
1da177e4 LT |
7250 | } |
7251 | ||
7252 | /* | |
7253 | * The remaining ioctls are changing the state of the | |
f91de92e | 7254 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 7255 | */ |
326eb17d | 7256 | if (mddev->ro && mddev->pers) { |
f91de92e N |
7257 | if (mddev->ro == 2) { |
7258 | mddev->ro = 0; | |
00bcb4ac | 7259 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 7260 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
7261 | /* mddev_unlock will wake thread */ |
7262 | /* If a device failed while we were read-only, we | |
7263 | * need to make sure the metadata is updated now. | |
7264 | */ | |
2953079c | 7265 | if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) { |
f3378b48 N |
7266 | mddev_unlock(mddev); |
7267 | wait_event(mddev->sb_wait, | |
2953079c SL |
7268 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) && |
7269 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); | |
29f097c4 | 7270 | mddev_lock_nointr(mddev); |
f3378b48 | 7271 | } |
f91de92e N |
7272 | } else { |
7273 | err = -EROFS; | |
3adc28d8 | 7274 | goto unlock; |
f91de92e | 7275 | } |
1da177e4 LT |
7276 | } |
7277 | ||
c02c0aeb N |
7278 | switch (cmd) { |
7279 | case ADD_NEW_DISK: | |
1da177e4 | 7280 | { |
c02c0aeb N |
7281 | mdu_disk_info_t info; |
7282 | if (copy_from_user(&info, argp, sizeof(info))) | |
7283 | err = -EFAULT; | |
7284 | else | |
7285 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7286 | goto unlock; |
c02c0aeb | 7287 | } |
1da177e4 | 7288 | |
1aee41f6 GR |
7289 | case CLUSTERED_DISK_NACK: |
7290 | if (mddev_is_clustered(mddev)) | |
7291 | md_cluster_ops->new_disk_ack(mddev, false); | |
7292 | else | |
7293 | err = -EINVAL; | |
7294 | goto unlock; | |
7295 | ||
c02c0aeb N |
7296 | case HOT_ADD_DISK: |
7297 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7298 | goto unlock; |
1da177e4 | 7299 | |
c02c0aeb N |
7300 | case RUN_ARRAY: |
7301 | err = do_md_run(mddev); | |
3adc28d8 | 7302 | goto unlock; |
1da177e4 | 7303 | |
c02c0aeb N |
7304 | case SET_BITMAP_FILE: |
7305 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 7306 | goto unlock; |
32a7627c | 7307 | |
c02c0aeb N |
7308 | default: |
7309 | err = -EINVAL; | |
3adc28d8 | 7310 | goto unlock; |
1da177e4 LT |
7311 | } |
7312 | ||
3adc28d8 | 7313 | unlock: |
d3374825 N |
7314 | if (mddev->hold_active == UNTIL_IOCTL && |
7315 | err != -EINVAL) | |
7316 | mddev->hold_active = 0; | |
1da177e4 | 7317 | mddev_unlock(mddev); |
3adc28d8 | 7318 | out: |
065e519e N |
7319 | if(did_set_md_closing) |
7320 | clear_bit(MD_CLOSING, &mddev->flags); | |
1da177e4 LT |
7321 | return err; |
7322 | } | |
aa98aa31 AB |
7323 | #ifdef CONFIG_COMPAT |
7324 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
7325 | unsigned int cmd, unsigned long arg) | |
7326 | { | |
7327 | switch (cmd) { | |
7328 | case HOT_REMOVE_DISK: | |
7329 | case HOT_ADD_DISK: | |
7330 | case SET_DISK_FAULTY: | |
7331 | case SET_BITMAP_FILE: | |
7332 | /* These take in integer arg, do not convert */ | |
7333 | break; | |
7334 | default: | |
7335 | arg = (unsigned long)compat_ptr(arg); | |
7336 | break; | |
7337 | } | |
7338 | ||
7339 | return md_ioctl(bdev, mode, cmd, arg); | |
7340 | } | |
7341 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 7342 | |
a39907fa | 7343 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
7344 | { |
7345 | /* | |
7346 | * Succeed if we can lock the mddev, which confirms that | |
7347 | * it isn't being stopped right now. | |
7348 | */ | |
fd01b88c | 7349 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
7350 | int err; |
7351 | ||
0c098220 YL |
7352 | if (!mddev) |
7353 | return -ENODEV; | |
7354 | ||
d3374825 N |
7355 | if (mddev->gendisk != bdev->bd_disk) { |
7356 | /* we are racing with mddev_put which is discarding this | |
7357 | * bd_disk. | |
7358 | */ | |
7359 | mddev_put(mddev); | |
7360 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 7361 | flush_workqueue(md_misc_wq); |
d3374825 N |
7362 | /* Then retry the open from the top */ |
7363 | return -ERESTARTSYS; | |
7364 | } | |
7365 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
7366 | ||
c8c00a69 | 7367 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
7368 | goto out; |
7369 | ||
af8d8e6f GJ |
7370 | if (test_bit(MD_CLOSING, &mddev->flags)) { |
7371 | mutex_unlock(&mddev->open_mutex); | |
e2342ca8 N |
7372 | err = -ENODEV; |
7373 | goto out; | |
af8d8e6f GJ |
7374 | } |
7375 | ||
1da177e4 | 7376 | err = 0; |
f2ea68cf | 7377 | atomic_inc(&mddev->openers); |
c8c00a69 | 7378 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 7379 | |
f0b4f7e2 | 7380 | check_disk_change(bdev); |
1da177e4 | 7381 | out: |
e2342ca8 N |
7382 | if (err) |
7383 | mddev_put(mddev); | |
1da177e4 LT |
7384 | return err; |
7385 | } | |
7386 | ||
db2a144b | 7387 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 7388 | { |
f72ffdd6 | 7389 | struct mddev *mddev = disk->private_data; |
1da177e4 | 7390 | |
52e5f9d1 | 7391 | BUG_ON(!mddev); |
f2ea68cf | 7392 | atomic_dec(&mddev->openers); |
1da177e4 | 7393 | mddev_put(mddev); |
1da177e4 | 7394 | } |
f0b4f7e2 N |
7395 | |
7396 | static int md_media_changed(struct gendisk *disk) | |
7397 | { | |
fd01b88c | 7398 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7399 | |
7400 | return mddev->changed; | |
7401 | } | |
7402 | ||
7403 | static int md_revalidate(struct gendisk *disk) | |
7404 | { | |
fd01b88c | 7405 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7406 | |
7407 | mddev->changed = 0; | |
7408 | return 0; | |
7409 | } | |
83d5cde4 | 7410 | static const struct block_device_operations md_fops = |
1da177e4 LT |
7411 | { |
7412 | .owner = THIS_MODULE, | |
a39907fa AV |
7413 | .open = md_open, |
7414 | .release = md_release, | |
b492b852 | 7415 | .ioctl = md_ioctl, |
aa98aa31 AB |
7416 | #ifdef CONFIG_COMPAT |
7417 | .compat_ioctl = md_compat_ioctl, | |
7418 | #endif | |
a885c8c4 | 7419 | .getgeo = md_getgeo, |
f0b4f7e2 N |
7420 | .media_changed = md_media_changed, |
7421 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
7422 | }; |
7423 | ||
f72ffdd6 | 7424 | static int md_thread(void *arg) |
1da177e4 | 7425 | { |
2b8bf345 | 7426 | struct md_thread *thread = arg; |
1da177e4 | 7427 | |
1da177e4 LT |
7428 | /* |
7429 | * md_thread is a 'system-thread', it's priority should be very | |
7430 | * high. We avoid resource deadlocks individually in each | |
7431 | * raid personality. (RAID5 does preallocation) We also use RR and | |
7432 | * the very same RT priority as kswapd, thus we will never get | |
7433 | * into a priority inversion deadlock. | |
7434 | * | |
7435 | * we definitely have to have equal or higher priority than | |
7436 | * bdflush, otherwise bdflush will deadlock if there are too | |
7437 | * many dirty RAID5 blocks. | |
7438 | */ | |
1da177e4 | 7439 | |
6985c43f | 7440 | allow_signal(SIGKILL); |
a6fb0934 | 7441 | while (!kthread_should_stop()) { |
1da177e4 | 7442 | |
93588e22 N |
7443 | /* We need to wait INTERRUPTIBLE so that |
7444 | * we don't add to the load-average. | |
7445 | * That means we need to be sure no signals are | |
7446 | * pending | |
7447 | */ | |
7448 | if (signal_pending(current)) | |
7449 | flush_signals(current); | |
7450 | ||
7451 | wait_event_interruptible_timeout | |
7452 | (thread->wqueue, | |
7453 | test_bit(THREAD_WAKEUP, &thread->flags) | |
ce1ccd07 | 7454 | || kthread_should_stop() || kthread_should_park(), |
93588e22 | 7455 | thread->timeout); |
1da177e4 | 7456 | |
6c987910 | 7457 | clear_bit(THREAD_WAKEUP, &thread->flags); |
ce1ccd07 SL |
7458 | if (kthread_should_park()) |
7459 | kthread_parkme(); | |
6c987910 | 7460 | if (!kthread_should_stop()) |
4ed8731d | 7461 | thread->run(thread); |
1da177e4 | 7462 | } |
a6fb0934 | 7463 | |
1da177e4 LT |
7464 | return 0; |
7465 | } | |
7466 | ||
2b8bf345 | 7467 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
7468 | { |
7469 | if (thread) { | |
36a4e1fe | 7470 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
d1d90147 GJ |
7471 | set_bit(THREAD_WAKEUP, &thread->flags); |
7472 | wake_up(&thread->wqueue); | |
1da177e4 LT |
7473 | } |
7474 | } | |
6c144d31 | 7475 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 7476 | |
4ed8731d SL |
7477 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
7478 | struct mddev *mddev, const char *name) | |
1da177e4 | 7479 | { |
2b8bf345 | 7480 | struct md_thread *thread; |
1da177e4 | 7481 | |
2b8bf345 | 7482 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
7483 | if (!thread) |
7484 | return NULL; | |
7485 | ||
1da177e4 LT |
7486 | init_waitqueue_head(&thread->wqueue); |
7487 | ||
1da177e4 LT |
7488 | thread->run = run; |
7489 | thread->mddev = mddev; | |
32a7627c | 7490 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
7491 | thread->tsk = kthread_run(md_thread, thread, |
7492 | "%s_%s", | |
7493 | mdname(thread->mddev), | |
0232605d | 7494 | name); |
a6fb0934 | 7495 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
7496 | kfree(thread); |
7497 | return NULL; | |
7498 | } | |
1da177e4 LT |
7499 | return thread; |
7500 | } | |
6c144d31 | 7501 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 7502 | |
2b8bf345 | 7503 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 7504 | { |
2b8bf345 | 7505 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
7506 | if (!thread) |
7507 | return; | |
36a4e1fe | 7508 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
7509 | /* Locking ensures that mddev_unlock does not wake_up a |
7510 | * non-existent thread | |
7511 | */ | |
7512 | spin_lock(&pers_lock); | |
7513 | *threadp = NULL; | |
7514 | spin_unlock(&pers_lock); | |
a6fb0934 N |
7515 | |
7516 | kthread_stop(thread->tsk); | |
1da177e4 LT |
7517 | kfree(thread); |
7518 | } | |
6c144d31 | 7519 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 7520 | |
fd01b88c | 7521 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7522 | { |
b2d444d7 | 7523 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 7524 | return; |
6bfe0b49 | 7525 | |
de393cde | 7526 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
7527 | return; |
7528 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
7529 | if (mddev->degraded) |
7530 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 7531 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
7532 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7533 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7534 | md_wakeup_thread(mddev->thread); | |
768a418d | 7535 | if (mddev->event_work.func) |
e804ac78 | 7536 | queue_work(md_misc_wq, &mddev->event_work); |
bb9ef716 | 7537 | md_new_event(mddev); |
1da177e4 | 7538 | } |
6c144d31 | 7539 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
7540 | |
7541 | /* seq_file implementation /proc/mdstat */ | |
7542 | ||
7543 | static void status_unused(struct seq_file *seq) | |
7544 | { | |
7545 | int i = 0; | |
3cb03002 | 7546 | struct md_rdev *rdev; |
1da177e4 LT |
7547 | |
7548 | seq_printf(seq, "unused devices: "); | |
7549 | ||
159ec1fc | 7550 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
7551 | char b[BDEVNAME_SIZE]; |
7552 | i++; | |
7553 | seq_printf(seq, "%s ", | |
7554 | bdevname(rdev->bdev,b)); | |
7555 | } | |
7556 | if (!i) | |
7557 | seq_printf(seq, "<none>"); | |
7558 | ||
7559 | seq_printf(seq, "\n"); | |
7560 | } | |
7561 | ||
f7851be7 | 7562 | static int status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7563 | { |
dd71cf6b N |
7564 | sector_t max_sectors, resync, res; |
7565 | unsigned long dt, db; | |
7566 | sector_t rt; | |
4588b42e N |
7567 | int scale; |
7568 | unsigned int per_milli; | |
1da177e4 | 7569 | |
c804cdec N |
7570 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7571 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 7572 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 7573 | else |
dd71cf6b | 7574 | max_sectors = mddev->dev_sectors; |
1da177e4 | 7575 | |
f7851be7 N |
7576 | resync = mddev->curr_resync; |
7577 | if (resync <= 3) { | |
7578 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7579 | /* Still cleaning up */ | |
7580 | resync = max_sectors; | |
7581 | } else | |
7582 | resync -= atomic_read(&mddev->recovery_active); | |
7583 | ||
7584 | if (resync == 0) { | |
7585 | if (mddev->recovery_cp < MaxSector) { | |
7586 | seq_printf(seq, "\tresync=PENDING"); | |
7587 | return 1; | |
7588 | } | |
7589 | return 0; | |
7590 | } | |
7591 | if (resync < 3) { | |
7592 | seq_printf(seq, "\tresync=DELAYED"); | |
7593 | return 1; | |
7594 | } | |
7595 | ||
403df478 | 7596 | WARN_ON(max_sectors == 0); |
4588b42e | 7597 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 7598 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
7599 | * u32, as those are the requirements for sector_div. |
7600 | * Thus 'scale' must be at least 10 | |
7601 | */ | |
7602 | scale = 10; | |
7603 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 7604 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
7605 | scale++; |
7606 | } | |
7607 | res = (resync>>scale)*1000; | |
dd71cf6b | 7608 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
7609 | |
7610 | per_milli = res; | |
1da177e4 | 7611 | { |
4588b42e | 7612 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
7613 | seq_printf(seq, "["); |
7614 | for (i = 0; i < x; i++) | |
7615 | seq_printf(seq, "="); | |
7616 | seq_printf(seq, ">"); | |
7617 | for (i = 0; i < y; i++) | |
7618 | seq_printf(seq, "."); | |
7619 | seq_printf(seq, "] "); | |
7620 | } | |
4588b42e | 7621 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
7622 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
7623 | "reshape" : | |
61df9d91 N |
7624 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
7625 | "check" : | |
7626 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
7627 | "resync" : "recovery"))), | |
7628 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
7629 | (unsigned long long) resync/2, |
7630 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
7631 | |
7632 | /* | |
1da177e4 LT |
7633 | * dt: time from mark until now |
7634 | * db: blocks written from mark until now | |
7635 | * rt: remaining time | |
dd71cf6b N |
7636 | * |
7637 | * rt is a sector_t, so could be 32bit or 64bit. | |
7638 | * So we divide before multiply in case it is 32bit and close | |
7639 | * to the limit. | |
25985edc | 7640 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
7641 | * near the end of resync when the number of remaining sectors |
7642 | * is close to 'db'. | |
7643 | * We then divide rt by 32 after multiplying by db to compensate. | |
7644 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
7645 | */ |
7646 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
7647 | if (!dt) dt++; | |
ff4e8d9a N |
7648 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
7649 | - mddev->resync_mark_cnt; | |
1da177e4 | 7650 | |
dd71cf6b N |
7651 | rt = max_sectors - resync; /* number of remaining sectors */ |
7652 | sector_div(rt, db/32+1); | |
7653 | rt *= dt; | |
7654 | rt >>= 5; | |
7655 | ||
7656 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
7657 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 7658 | |
ff4e8d9a | 7659 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
f7851be7 | 7660 | return 1; |
1da177e4 LT |
7661 | } |
7662 | ||
7663 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7664 | { | |
7665 | struct list_head *tmp; | |
7666 | loff_t l = *pos; | |
fd01b88c | 7667 | struct mddev *mddev; |
1da177e4 LT |
7668 | |
7669 | if (l >= 0x10000) | |
7670 | return NULL; | |
7671 | if (!l--) | |
7672 | /* header */ | |
7673 | return (void*)1; | |
7674 | ||
7675 | spin_lock(&all_mddevs_lock); | |
7676 | list_for_each(tmp,&all_mddevs) | |
7677 | if (!l--) { | |
fd01b88c | 7678 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7679 | mddev_get(mddev); |
7680 | spin_unlock(&all_mddevs_lock); | |
7681 | return mddev; | |
7682 | } | |
7683 | spin_unlock(&all_mddevs_lock); | |
7684 | if (!l--) | |
7685 | return (void*)2;/* tail */ | |
7686 | return NULL; | |
7687 | } | |
7688 | ||
7689 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7690 | { | |
7691 | struct list_head *tmp; | |
fd01b88c | 7692 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7693 | |
1da177e4 LT |
7694 | ++*pos; |
7695 | if (v == (void*)2) | |
7696 | return NULL; | |
7697 | ||
7698 | spin_lock(&all_mddevs_lock); | |
7699 | if (v == (void*)1) | |
7700 | tmp = all_mddevs.next; | |
7701 | else | |
7702 | tmp = mddev->all_mddevs.next; | |
7703 | if (tmp != &all_mddevs) | |
fd01b88c | 7704 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7705 | else { |
7706 | next_mddev = (void*)2; | |
7707 | *pos = 0x10000; | |
f72ffdd6 | 7708 | } |
1da177e4 LT |
7709 | spin_unlock(&all_mddevs_lock); |
7710 | ||
7711 | if (v != (void*)1) | |
7712 | mddev_put(mddev); | |
7713 | return next_mddev; | |
7714 | ||
7715 | } | |
7716 | ||
7717 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7718 | { | |
fd01b88c | 7719 | struct mddev *mddev = v; |
1da177e4 LT |
7720 | |
7721 | if (mddev && v != (void*)1 && v != (void*)2) | |
7722 | mddev_put(mddev); | |
7723 | } | |
7724 | ||
7725 | static int md_seq_show(struct seq_file *seq, void *v) | |
7726 | { | |
fd01b88c | 7727 | struct mddev *mddev = v; |
dd8ac336 | 7728 | sector_t sectors; |
3cb03002 | 7729 | struct md_rdev *rdev; |
1da177e4 LT |
7730 | |
7731 | if (v == (void*)1) { | |
84fc4b56 | 7732 | struct md_personality *pers; |
1da177e4 LT |
7733 | seq_printf(seq, "Personalities : "); |
7734 | spin_lock(&pers_lock); | |
2604b703 N |
7735 | list_for_each_entry(pers, &pers_list, list) |
7736 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7737 | |
7738 | spin_unlock(&pers_lock); | |
7739 | seq_printf(seq, "\n"); | |
f1514638 | 7740 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7741 | return 0; |
7742 | } | |
7743 | if (v == (void*)2) { | |
7744 | status_unused(seq); | |
7745 | return 0; | |
7746 | } | |
7747 | ||
36d091f4 | 7748 | spin_lock(&mddev->lock); |
1da177e4 LT |
7749 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7750 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7751 | mddev->pers ? "" : "in"); | |
7752 | if (mddev->pers) { | |
f91de92e | 7753 | if (mddev->ro==1) |
1da177e4 | 7754 | seq_printf(seq, " (read-only)"); |
f91de92e | 7755 | if (mddev->ro==2) |
52720ae7 | 7756 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7757 | seq_printf(seq, " %s", mddev->pers->name); |
7758 | } | |
7759 | ||
dd8ac336 | 7760 | sectors = 0; |
f97fcad3 N |
7761 | rcu_read_lock(); |
7762 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7763 | char b[BDEVNAME_SIZE]; |
7764 | seq_printf(seq, " %s[%d]", | |
7765 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7766 | if (test_bit(WriteMostly, &rdev->flags)) |
7767 | seq_printf(seq, "(W)"); | |
9efdca16 SL |
7768 | if (test_bit(Journal, &rdev->flags)) |
7769 | seq_printf(seq, "(J)"); | |
b2d444d7 | 7770 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7771 | seq_printf(seq, "(F)"); |
7772 | continue; | |
2d78f8c4 N |
7773 | } |
7774 | if (rdev->raid_disk < 0) | |
b325a32e | 7775 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7776 | if (test_bit(Replacement, &rdev->flags)) |
7777 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7778 | sectors += rdev->sectors; |
1da177e4 | 7779 | } |
f97fcad3 | 7780 | rcu_read_unlock(); |
1da177e4 LT |
7781 | |
7782 | if (!list_empty(&mddev->disks)) { | |
7783 | if (mddev->pers) | |
7784 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7785 | (unsigned long long) |
7786 | mddev->array_sectors / 2); | |
1da177e4 LT |
7787 | else |
7788 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7789 | (unsigned long long)sectors / 2); |
1da177e4 | 7790 | } |
1cd6bf19 N |
7791 | if (mddev->persistent) { |
7792 | if (mddev->major_version != 0 || | |
7793 | mddev->minor_version != 90) { | |
7794 | seq_printf(seq," super %d.%d", | |
7795 | mddev->major_version, | |
7796 | mddev->minor_version); | |
7797 | } | |
e691063a N |
7798 | } else if (mddev->external) |
7799 | seq_printf(seq, " super external:%s", | |
7800 | mddev->metadata_type); | |
7801 | else | |
1cd6bf19 | 7802 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7803 | |
7804 | if (mddev->pers) { | |
d710e138 | 7805 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7806 | seq_printf(seq, "\n "); |
8e1b39d6 | 7807 | if (mddev->pers->sync_request) { |
f7851be7 | 7808 | if (status_resync(seq, mddev)) |
8e1b39d6 | 7809 | seq_printf(seq, "\n "); |
8e1b39d6 | 7810 | } |
32a7627c N |
7811 | } else |
7812 | seq_printf(seq, "\n "); | |
7813 | ||
57148964 | 7814 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7815 | |
7816 | seq_printf(seq, "\n"); | |
7817 | } | |
36d091f4 | 7818 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7819 | |
1da177e4 LT |
7820 | return 0; |
7821 | } | |
7822 | ||
110518bc | 7823 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7824 | .start = md_seq_start, |
7825 | .next = md_seq_next, | |
7826 | .stop = md_seq_stop, | |
7827 | .show = md_seq_show, | |
7828 | }; | |
7829 | ||
7830 | static int md_seq_open(struct inode *inode, struct file *file) | |
7831 | { | |
f1514638 | 7832 | struct seq_file *seq; |
1da177e4 LT |
7833 | int error; |
7834 | ||
7835 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7836 | if (error) |
f1514638 KS |
7837 | return error; |
7838 | ||
7839 | seq = file->private_data; | |
7840 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7841 | return error; |
7842 | } | |
7843 | ||
e2f23b60 | 7844 | static int md_unloading; |
d7603b7e N |
7845 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7846 | { | |
f1514638 | 7847 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7848 | int mask; |
7849 | ||
e2f23b60 | 7850 | if (md_unloading) |
7d7e64f2 | 7851 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI; |
d7603b7e N |
7852 | poll_wait(filp, &md_event_waiters, wait); |
7853 | ||
7854 | /* always allow read */ | |
7855 | mask = POLLIN | POLLRDNORM; | |
7856 | ||
f1514638 | 7857 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7858 | mask |= POLLERR | POLLPRI; |
7859 | return mask; | |
7860 | } | |
7861 | ||
fa027c2a | 7862 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7863 | .owner = THIS_MODULE, |
1da177e4 LT |
7864 | .open = md_seq_open, |
7865 | .read = seq_read, | |
7866 | .llseek = seq_lseek, | |
26e13043 | 7867 | .release = seq_release, |
d7603b7e | 7868 | .poll = mdstat_poll, |
1da177e4 LT |
7869 | }; |
7870 | ||
84fc4b56 | 7871 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7872 | { |
9d48739e N |
7873 | pr_debug("md: %s personality registered for level %d\n", |
7874 | p->name, p->level); | |
1da177e4 | 7875 | spin_lock(&pers_lock); |
2604b703 | 7876 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7877 | spin_unlock(&pers_lock); |
7878 | return 0; | |
7879 | } | |
6c144d31 | 7880 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7881 | |
84fc4b56 | 7882 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7883 | { |
9d48739e | 7884 | pr_debug("md: %s personality unregistered\n", p->name); |
1da177e4 | 7885 | spin_lock(&pers_lock); |
2604b703 | 7886 | list_del_init(&p->list); |
1da177e4 LT |
7887 | spin_unlock(&pers_lock); |
7888 | return 0; | |
7889 | } | |
6c144d31 | 7890 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7891 | |
6022e75b N |
7892 | int register_md_cluster_operations(struct md_cluster_operations *ops, |
7893 | struct module *module) | |
edb39c9d | 7894 | { |
6022e75b | 7895 | int ret = 0; |
edb39c9d | 7896 | spin_lock(&pers_lock); |
6022e75b N |
7897 | if (md_cluster_ops != NULL) |
7898 | ret = -EALREADY; | |
7899 | else { | |
7900 | md_cluster_ops = ops; | |
7901 | md_cluster_mod = module; | |
7902 | } | |
edb39c9d | 7903 | spin_unlock(&pers_lock); |
6022e75b | 7904 | return ret; |
edb39c9d GR |
7905 | } |
7906 | EXPORT_SYMBOL(register_md_cluster_operations); | |
7907 | ||
7908 | int unregister_md_cluster_operations(void) | |
7909 | { | |
7910 | spin_lock(&pers_lock); | |
7911 | md_cluster_ops = NULL; | |
7912 | spin_unlock(&pers_lock); | |
7913 | return 0; | |
7914 | } | |
7915 | EXPORT_SYMBOL(unregister_md_cluster_operations); | |
7916 | ||
7917 | int md_setup_cluster(struct mddev *mddev, int nodes) | |
7918 | { | |
47a7b0d8 GJ |
7919 | if (!md_cluster_ops) |
7920 | request_module("md-cluster"); | |
edb39c9d | 7921 | spin_lock(&pers_lock); |
47a7b0d8 | 7922 | /* ensure module won't be unloaded */ |
edb39c9d | 7923 | if (!md_cluster_ops || !try_module_get(md_cluster_mod)) { |
9d48739e | 7924 | pr_warn("can't find md-cluster module or get it's reference.\n"); |
edb39c9d GR |
7925 | spin_unlock(&pers_lock); |
7926 | return -ENOENT; | |
7927 | } | |
7928 | spin_unlock(&pers_lock); | |
7929 | ||
cf921cc1 | 7930 | return md_cluster_ops->join(mddev, nodes); |
edb39c9d GR |
7931 | } |
7932 | ||
7933 | void md_cluster_stop(struct mddev *mddev) | |
7934 | { | |
c4ce867f GR |
7935 | if (!md_cluster_ops) |
7936 | return; | |
edb39c9d GR |
7937 | md_cluster_ops->leave(mddev); |
7938 | module_put(md_cluster_mod); | |
7939 | } | |
7940 | ||
fd01b88c | 7941 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7942 | { |
f72ffdd6 | 7943 | struct md_rdev *rdev; |
1da177e4 | 7944 | int idle; |
eea1bf38 | 7945 | int curr_events; |
1da177e4 LT |
7946 | |
7947 | idle = 1; | |
4b80991c N |
7948 | rcu_read_lock(); |
7949 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7950 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7951 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7952 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7953 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7954 | /* sync IO will cause sync_io to increase before the disk_stats |
7955 | * as sync_io is counted when a request starts, and | |
7956 | * disk_stats is counted when it completes. | |
7957 | * So resync activity will cause curr_events to be smaller than | |
7958 | * when there was no such activity. | |
7959 | * non-sync IO will cause disk_stat to increase without | |
7960 | * increasing sync_io so curr_events will (eventually) | |
7961 | * be larger than it was before. Once it becomes | |
7962 | * substantially larger, the test below will cause | |
7963 | * the array to appear non-idle, and resync will slow | |
7964 | * down. | |
7965 | * If there is a lot of outstanding resync activity when | |
7966 | * we set last_event to curr_events, then all that activity | |
7967 | * completing might cause the array to appear non-idle | |
7968 | * and resync will be slowed down even though there might | |
7969 | * not have been non-resync activity. This will only | |
7970 | * happen once though. 'last_events' will soon reflect | |
7971 | * the state where there is little or no outstanding | |
7972 | * resync requests, and further resync activity will | |
7973 | * always make curr_events less than last_events. | |
c0e48521 | 7974 | * |
1da177e4 | 7975 | */ |
eea1bf38 | 7976 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7977 | rdev->last_events = curr_events; |
7978 | idle = 0; | |
7979 | } | |
7980 | } | |
4b80991c | 7981 | rcu_read_unlock(); |
1da177e4 LT |
7982 | return idle; |
7983 | } | |
7984 | ||
fd01b88c | 7985 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7986 | { |
7987 | /* another "blocks" (512byte) blocks have been synced */ | |
7988 | atomic_sub(blocks, &mddev->recovery_active); | |
7989 | wake_up(&mddev->recovery_wait); | |
7990 | if (!ok) { | |
dfc70645 | 7991 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7992 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7993 | md_wakeup_thread(mddev->thread); |
7994 | // stop recovery, signal do_sync .... | |
7995 | } | |
7996 | } | |
6c144d31 | 7997 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 7998 | |
06d91a5f N |
7999 | /* md_write_start(mddev, bi) |
8000 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
8001 | * in superblock) before writing, schedule a superblock update |
8002 | * and wait for it to complete. | |
cc27b0c7 N |
8003 | * A return value of 'false' means that the write wasn't recorded |
8004 | * and cannot proceed as the array is being suspend. | |
06d91a5f | 8005 | */ |
cc27b0c7 | 8006 | bool md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 8007 | { |
0fd62b86 | 8008 | int did_change = 0; |
06d91a5f | 8009 | if (bio_data_dir(bi) != WRITE) |
cc27b0c7 | 8010 | return true; |
06d91a5f | 8011 | |
f91de92e N |
8012 | BUG_ON(mddev->ro == 1); |
8013 | if (mddev->ro == 2) { | |
8014 | /* need to switch to read/write */ | |
8015 | mddev->ro = 0; | |
8016 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8017 | md_wakeup_thread(mddev->thread); | |
25156198 | 8018 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 8019 | did_change = 1; |
f91de92e | 8020 | } |
4ad23a97 N |
8021 | rcu_read_lock(); |
8022 | percpu_ref_get(&mddev->writes_pending); | |
55cc39f3 | 8023 | smp_mb(); /* Match smp_mb in set_in_sync() */ |
31a59e34 N |
8024 | if (mddev->safemode == 1) |
8025 | mddev->safemode = 0; | |
4ad23a97 | 8026 | /* sync_checkers is always 0 when writes_pending is in per-cpu mode */ |
81fe48e9 | 8027 | if (mddev->in_sync || mddev->sync_checkers) { |
85572d7c | 8028 | spin_lock(&mddev->lock); |
3d310eb7 N |
8029 | if (mddev->in_sync) { |
8030 | mddev->in_sync = 0; | |
2953079c SL |
8031 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
8032 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
3d310eb7 | 8033 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 8034 | did_change = 1; |
3d310eb7 | 8035 | } |
85572d7c | 8036 | spin_unlock(&mddev->lock); |
06d91a5f | 8037 | } |
4ad23a97 | 8038 | rcu_read_unlock(); |
0fd62b86 | 8039 | if (did_change) |
00bcb4ac | 8040 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 8041 | wait_event(mddev->sb_wait, |
d47c8ad2 N |
8042 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) || |
8043 | mddev->suspended); | |
cc27b0c7 N |
8044 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
8045 | percpu_ref_put(&mddev->writes_pending); | |
8046 | return false; | |
8047 | } | |
8048 | return true; | |
1da177e4 | 8049 | } |
6c144d31 | 8050 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 8051 | |
49728050 N |
8052 | /* md_write_inc can only be called when md_write_start() has |
8053 | * already been called at least once of the current request. | |
8054 | * It increments the counter and is useful when a single request | |
8055 | * is split into several parts. Each part causes an increment and | |
8056 | * so needs a matching md_write_end(). | |
8057 | * Unlike md_write_start(), it is safe to call md_write_inc() inside | |
8058 | * a spinlocked region. | |
8059 | */ | |
8060 | void md_write_inc(struct mddev *mddev, struct bio *bi) | |
8061 | { | |
8062 | if (bio_data_dir(bi) != WRITE) | |
8063 | return; | |
8064 | WARN_ON_ONCE(mddev->in_sync || mddev->ro); | |
4ad23a97 | 8065 | percpu_ref_get(&mddev->writes_pending); |
49728050 N |
8066 | } |
8067 | EXPORT_SYMBOL(md_write_inc); | |
8068 | ||
fd01b88c | 8069 | void md_write_end(struct mddev *mddev) |
1da177e4 | 8070 | { |
4ad23a97 N |
8071 | percpu_ref_put(&mddev->writes_pending); |
8072 | ||
8073 | if (mddev->safemode == 2) | |
8074 | md_wakeup_thread(mddev->thread); | |
8075 | else if (mddev->safemode_delay) | |
8076 | /* The roundup() ensures this only performs locking once | |
8077 | * every ->safemode_delay jiffies | |
8078 | */ | |
8079 | mod_timer(&mddev->safemode_timer, | |
8080 | roundup(jiffies, mddev->safemode_delay) + | |
8081 | mddev->safemode_delay); | |
1da177e4 | 8082 | } |
4ad23a97 | 8083 | |
6c144d31 | 8084 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 8085 | |
2a2275d6 N |
8086 | /* md_allow_write(mddev) |
8087 | * Calling this ensures that the array is marked 'active' so that writes | |
8088 | * may proceed without blocking. It is important to call this before | |
8089 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
8090 | * Must be called with mddev_lock held. | |
8091 | */ | |
2214c260 | 8092 | void md_allow_write(struct mddev *mddev) |
2a2275d6 N |
8093 | { |
8094 | if (!mddev->pers) | |
2214c260 | 8095 | return; |
2a2275d6 | 8096 | if (mddev->ro) |
2214c260 | 8097 | return; |
1a0fd497 | 8098 | if (!mddev->pers->sync_request) |
2214c260 | 8099 | return; |
2a2275d6 | 8100 | |
85572d7c | 8101 | spin_lock(&mddev->lock); |
2a2275d6 N |
8102 | if (mddev->in_sync) { |
8103 | mddev->in_sync = 0; | |
2953079c SL |
8104 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
8105 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
2a2275d6 N |
8106 | if (mddev->safemode_delay && |
8107 | mddev->safemode == 0) | |
8108 | mddev->safemode = 1; | |
85572d7c | 8109 | spin_unlock(&mddev->lock); |
2a2275d6 | 8110 | md_update_sb(mddev, 0); |
00bcb4ac | 8111 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2214c260 AP |
8112 | /* wait for the dirty state to be recorded in the metadata */ |
8113 | wait_event(mddev->sb_wait, | |
8114 | !test_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags) && | |
8115 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); | |
2a2275d6 | 8116 | } else |
85572d7c | 8117 | spin_unlock(&mddev->lock); |
2a2275d6 N |
8118 | } |
8119 | EXPORT_SYMBOL_GPL(md_allow_write); | |
8120 | ||
1da177e4 LT |
8121 | #define SYNC_MARKS 10 |
8122 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 8123 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 8124 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 8125 | { |
4ed8731d | 8126 | struct mddev *mddev = thread->mddev; |
fd01b88c | 8127 | struct mddev *mddev2; |
1da177e4 LT |
8128 | unsigned int currspeed = 0, |
8129 | window; | |
ac7e50a3 | 8130 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 8131 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 8132 | unsigned long update_time; |
1da177e4 LT |
8133 | sector_t mark_cnt[SYNC_MARKS]; |
8134 | int last_mark,m; | |
8135 | struct list_head *tmp; | |
8136 | sector_t last_check; | |
57afd89f | 8137 | int skipped = 0; |
3cb03002 | 8138 | struct md_rdev *rdev; |
c4a39551 | 8139 | char *desc, *action = NULL; |
7c2c57c9 | 8140 | struct blk_plug plug; |
41a9a0dc | 8141 | int ret; |
1da177e4 LT |
8142 | |
8143 | /* just incase thread restarts... */ | |
8144 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
8145 | return; | |
3991b31e N |
8146 | if (mddev->ro) {/* never try to sync a read-only array */ |
8147 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 8148 | return; |
3991b31e | 8149 | } |
1da177e4 | 8150 | |
41a9a0dc GJ |
8151 | if (mddev_is_clustered(mddev)) { |
8152 | ret = md_cluster_ops->resync_start(mddev); | |
8153 | if (ret) | |
8154 | goto skip; | |
8155 | ||
bb8bf15b | 8156 | set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags); |
41a9a0dc GJ |
8157 | if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
8158 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) || | |
8159 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) | |
8160 | && ((unsigned long long)mddev->curr_resync_completed | |
8161 | < (unsigned long long)mddev->resync_max_sectors)) | |
8162 | goto skip; | |
8163 | } | |
8164 | ||
61df9d91 | 8165 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 8166 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 8167 | desc = "data-check"; |
c4a39551 JB |
8168 | action = "check"; |
8169 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 8170 | desc = "requested-resync"; |
c4a39551 JB |
8171 | action = "repair"; |
8172 | } else | |
61df9d91 N |
8173 | desc = "resync"; |
8174 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
8175 | desc = "reshape"; | |
8176 | else | |
8177 | desc = "recovery"; | |
8178 | ||
c4a39551 JB |
8179 | mddev->last_sync_action = action ?: desc; |
8180 | ||
1da177e4 LT |
8181 | /* we overload curr_resync somewhat here. |
8182 | * 0 == not engaged in resync at all | |
8183 | * 2 == checking that there is no conflict with another sync | |
8184 | * 1 == like 2, but have yielded to allow conflicting resync to | |
8185 | * commense | |
8186 | * other == active in resync - this many blocks | |
8187 | * | |
8188 | * Before starting a resync we must have set curr_resync to | |
8189 | * 2, and then checked that every "conflicting" array has curr_resync | |
8190 | * less than ours. When we find one that is the same or higher | |
8191 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
8192 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
8193 | * This will mean we have to start checking from the beginning again. | |
8194 | * | |
8195 | */ | |
8196 | ||
8197 | do { | |
c622ca54 | 8198 | int mddev2_minor = -1; |
1da177e4 LT |
8199 | mddev->curr_resync = 2; |
8200 | ||
8201 | try_again: | |
404e4b43 | 8202 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 8203 | goto skip; |
29ac4aa3 | 8204 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
8205 | if (mddev2 == mddev) |
8206 | continue; | |
90b08710 BS |
8207 | if (!mddev->parallel_resync |
8208 | && mddev2->curr_resync | |
8209 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
8210 | DEFINE_WAIT(wq); |
8211 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
8212 | /* arbitrarily yield */ | |
8213 | mddev->curr_resync = 1; | |
8214 | wake_up(&resync_wait); | |
8215 | } | |
8216 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
8217 | /* no need to wait here, we can wait the next | |
8218 | * time 'round when curr_resync == 2 | |
8219 | */ | |
8220 | continue; | |
9744197c N |
8221 | /* We need to wait 'interruptible' so as not to |
8222 | * contribute to the load average, and not to | |
8223 | * be caught by 'softlockup' | |
8224 | */ | |
8225 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 8226 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 8227 | mddev2->curr_resync >= mddev->curr_resync) { |
c622ca54 AP |
8228 | if (mddev2_minor != mddev2->md_minor) { |
8229 | mddev2_minor = mddev2->md_minor; | |
9d48739e N |
8230 | pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n", |
8231 | desc, mdname(mddev), | |
8232 | mdname(mddev2)); | |
c622ca54 | 8233 | } |
1da177e4 | 8234 | mddev_put(mddev2); |
9744197c N |
8235 | if (signal_pending(current)) |
8236 | flush_signals(current); | |
1da177e4 LT |
8237 | schedule(); |
8238 | finish_wait(&resync_wait, &wq); | |
8239 | goto try_again; | |
8240 | } | |
8241 | finish_wait(&resync_wait, &wq); | |
8242 | } | |
8243 | } | |
8244 | } while (mddev->curr_resync < 2); | |
8245 | ||
5fd6c1dc | 8246 | j = 0; |
9d88883e | 8247 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 8248 | /* resync follows the size requested by the personality, |
57afd89f | 8249 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
8250 | */ |
8251 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 8252 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 8253 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
8254 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
8255 | j = mddev->resync_min; | |
8256 | else if (!mddev->bitmap) | |
5fd6c1dc | 8257 | j = mddev->recovery_cp; |
5e96ee65 | 8258 | |
ccfcc3c1 | 8259 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 8260 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 8261 | else { |
1da177e4 | 8262 | /* recovery follows the physical size of devices */ |
58c0fed4 | 8263 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 8264 | j = MaxSector; |
4e59ca7d | 8265 | rcu_read_lock(); |
dafb20fa | 8266 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 8267 | if (rdev->raid_disk >= 0 && |
f2076e7d | 8268 | !test_bit(Journal, &rdev->flags) && |
5fd6c1dc N |
8269 | !test_bit(Faulty, &rdev->flags) && |
8270 | !test_bit(In_sync, &rdev->flags) && | |
8271 | rdev->recovery_offset < j) | |
8272 | j = rdev->recovery_offset; | |
4e59ca7d | 8273 | rcu_read_unlock(); |
133d4527 N |
8274 | |
8275 | /* If there is a bitmap, we need to make sure all | |
8276 | * writes that started before we added a spare | |
8277 | * complete before we start doing a recovery. | |
8278 | * Otherwise the write might complete and (via | |
8279 | * bitmap_endwrite) set a bit in the bitmap after the | |
8280 | * recovery has checked that bit and skipped that | |
8281 | * region. | |
8282 | */ | |
8283 | if (mddev->bitmap) { | |
8284 | mddev->pers->quiesce(mddev, 1); | |
8285 | mddev->pers->quiesce(mddev, 0); | |
8286 | } | |
5fd6c1dc | 8287 | } |
1da177e4 | 8288 | |
9d48739e N |
8289 | pr_info("md: %s of RAID array %s\n", desc, mdname(mddev)); |
8290 | pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev)); | |
8291 | pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n", | |
8292 | speed_max(mddev), desc); | |
1da177e4 | 8293 | |
eea1bf38 | 8294 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 8295 | |
57afd89f | 8296 | io_sectors = 0; |
1da177e4 LT |
8297 | for (m = 0; m < SYNC_MARKS; m++) { |
8298 | mark[m] = jiffies; | |
57afd89f | 8299 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
8300 | } |
8301 | last_mark = 0; | |
8302 | mddev->resync_mark = mark[last_mark]; | |
8303 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
8304 | ||
8305 | /* | |
8306 | * Tune reconstruction: | |
8307 | */ | |
8308 | window = 32*(PAGE_SIZE/512); | |
9d48739e N |
8309 | pr_debug("md: using %dk window, over a total of %lluk.\n", |
8310 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
8311 | |
8312 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
8313 | last_check = 0; |
8314 | ||
8315 | if (j>2) { | |
9d48739e N |
8316 | pr_debug("md: resuming %s of %s from checkpoint.\n", |
8317 | desc, mdname(mddev)); | |
1da177e4 | 8318 | mddev->curr_resync = j; |
72f36d59 N |
8319 | } else |
8320 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 8321 | mddev->curr_resync_completed = j; |
72f36d59 N |
8322 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
8323 | md_new_event(mddev); | |
54f89341 | 8324 | update_time = jiffies; |
1da177e4 | 8325 | |
7c2c57c9 | 8326 | blk_start_plug(&plug); |
1da177e4 | 8327 | while (j < max_sectors) { |
57afd89f | 8328 | sector_t sectors; |
1da177e4 | 8329 | |
57afd89f | 8330 | skipped = 0; |
97e4f42d | 8331 | |
7a91ee1f N |
8332 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8333 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
8334 | (mddev->curr_resync - mddev->curr_resync_completed) | |
8335 | > (max_sectors >> 4)) || | |
54f89341 | 8336 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f | 8337 | (j - mddev->curr_resync_completed)*2 |
c5e19d90 N |
8338 | >= mddev->resync_max - mddev->curr_resync_completed || |
8339 | mddev->curr_resync_completed > mddev->resync_max | |
7a91ee1f | 8340 | )) { |
97e4f42d | 8341 | /* time to update curr_resync_completed */ |
97e4f42d N |
8342 | wait_event(mddev->recovery_wait, |
8343 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 8344 | mddev->curr_resync_completed = j; |
35d78c66 | 8345 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
8346 | j > mddev->recovery_cp) | |
8347 | mddev->recovery_cp = j; | |
54f89341 | 8348 | update_time = jiffies; |
2953079c | 8349 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
acb180b0 | 8350 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 8351 | } |
acb180b0 | 8352 | |
c91abf5a N |
8353 | while (j >= mddev->resync_max && |
8354 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
8355 | /* As this condition is controlled by user-space, |
8356 | * we can block indefinitely, so use '_interruptible' | |
8357 | * to avoid triggering warnings. | |
8358 | */ | |
8359 | flush_signals(current); /* just in case */ | |
8360 | wait_event_interruptible(mddev->recovery_wait, | |
8361 | mddev->resync_max > j | |
c91abf5a N |
8362 | || test_bit(MD_RECOVERY_INTR, |
8363 | &mddev->recovery)); | |
e62e58a5 | 8364 | } |
acb180b0 | 8365 | |
c91abf5a N |
8366 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8367 | break; | |
acb180b0 | 8368 | |
09314799 | 8369 | sectors = mddev->pers->sync_request(mddev, j, &skipped); |
57afd89f | 8370 | if (sectors == 0) { |
dfc70645 | 8371 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 8372 | break; |
1da177e4 | 8373 | } |
57afd89f N |
8374 | |
8375 | if (!skipped) { /* actual IO requested */ | |
8376 | io_sectors += sectors; | |
8377 | atomic_add(sectors, &mddev->recovery_active); | |
8378 | } | |
8379 | ||
e875ecea N |
8380 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8381 | break; | |
8382 | ||
1da177e4 | 8383 | j += sectors; |
5ed1df2e N |
8384 | if (j > max_sectors) |
8385 | /* when skipping, extra large numbers can be returned. */ | |
8386 | j = max_sectors; | |
72f36d59 N |
8387 | if (j > 2) |
8388 | mddev->curr_resync = j; | |
ff4e8d9a | 8389 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 8390 | if (last_check == 0) |
e875ecea | 8391 | /* this is the earliest that rebuild will be |
d7603b7e N |
8392 | * visible in /proc/mdstat |
8393 | */ | |
8394 | md_new_event(mddev); | |
57afd89f N |
8395 | |
8396 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
8397 | continue; |
8398 | ||
57afd89f | 8399 | last_check = io_sectors; |
1da177e4 LT |
8400 | repeat: |
8401 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
8402 | /* step marks */ | |
8403 | int next = (last_mark+1) % SYNC_MARKS; | |
8404 | ||
8405 | mddev->resync_mark = mark[next]; | |
8406 | mddev->resync_mark_cnt = mark_cnt[next]; | |
8407 | mark[next] = jiffies; | |
57afd89f | 8408 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
8409 | last_mark = next; |
8410 | } | |
8411 | ||
c91abf5a N |
8412 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8413 | break; | |
1da177e4 LT |
8414 | |
8415 | /* | |
8416 | * this loop exits only if either when we are slower than | |
8417 | * the 'hard' speed limit, or the system was IO-idle for | |
8418 | * a jiffy. | |
8419 | * the system might be non-idle CPU-wise, but we only care | |
8420 | * about not overloading the IO subsystem. (things like an | |
8421 | * e2fsck being done on the RAID array should execute fast) | |
8422 | */ | |
1da177e4 LT |
8423 | cond_resched(); |
8424 | ||
ac7e50a3 XN |
8425 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
8426 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 8427 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 8428 | |
88202a0c | 8429 | if (currspeed > speed_min(mddev)) { |
ac8fa419 | 8430 | if (currspeed > speed_max(mddev)) { |
c0e48521 | 8431 | msleep(500); |
1da177e4 LT |
8432 | goto repeat; |
8433 | } | |
ac8fa419 N |
8434 | if (!is_mddev_idle(mddev, 0)) { |
8435 | /* | |
8436 | * Give other IO more of a chance. | |
8437 | * The faster the devices, the less we wait. | |
8438 | */ | |
8439 | wait_event(mddev->recovery_wait, | |
8440 | !atomic_read(&mddev->recovery_active)); | |
8441 | } | |
1da177e4 LT |
8442 | } |
8443 | } | |
9d48739e N |
8444 | pr_info("md: %s: %s %s.\n",mdname(mddev), desc, |
8445 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
8446 | ? "interrupted" : "done"); | |
1da177e4 LT |
8447 | /* |
8448 | * this also signals 'finished resyncing' to md_stop | |
8449 | */ | |
7c2c57c9 | 8450 | blk_finish_plug(&plug); |
1da177e4 LT |
8451 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
8452 | ||
5ed1df2e N |
8453 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8454 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
1217e1d1 | 8455 | mddev->curr_resync > 3) { |
5ed1df2e N |
8456 | mddev->curr_resync_completed = mddev->curr_resync; |
8457 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
8458 | } | |
09314799 | 8459 | mddev->pers->sync_request(mddev, max_sectors, &skipped); |
1da177e4 | 8460 | |
dfc70645 | 8461 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
1217e1d1 | 8462 | mddev->curr_resync > 3) { |
5fd6c1dc N |
8463 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
8464 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8465 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
9d48739e N |
8466 | pr_debug("md: checkpointing %s of %s.\n", |
8467 | desc, mdname(mddev)); | |
0a19caab | 8468 | if (test_bit(MD_RECOVERY_ERROR, |
8469 | &mddev->recovery)) | |
8470 | mddev->recovery_cp = | |
8471 | mddev->curr_resync_completed; | |
8472 | else | |
8473 | mddev->recovery_cp = | |
8474 | mddev->curr_resync; | |
5fd6c1dc N |
8475 | } |
8476 | } else | |
8477 | mddev->recovery_cp = MaxSector; | |
8478 | } else { | |
8479 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
8480 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 8481 | rcu_read_lock(); |
dafb20fa | 8482 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 8483 | if (rdev->raid_disk >= 0 && |
70fffd0b | 8484 | mddev->delta_disks >= 0 && |
f2076e7d | 8485 | !test_bit(Journal, &rdev->flags) && |
5fd6c1dc N |
8486 | !test_bit(Faulty, &rdev->flags) && |
8487 | !test_bit(In_sync, &rdev->flags) && | |
8488 | rdev->recovery_offset < mddev->curr_resync) | |
8489 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 8490 | rcu_read_unlock(); |
5fd6c1dc | 8491 | } |
1da177e4 | 8492 | } |
db91ff55 | 8493 | skip: |
bb8bf15b GJ |
8494 | /* set CHANGE_PENDING here since maybe another update is needed, |
8495 | * so other nodes are informed. It should be harmless for normal | |
8496 | * raid */ | |
2953079c SL |
8497 | set_mask_bits(&mddev->sb_flags, 0, |
8498 | BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS)); | |
c186b128 | 8499 | |
23da422b | 8500 | spin_lock(&mddev->lock); |
c07b70ad N |
8501 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
8502 | /* We completed so min/max setting can be forgotten if used. */ | |
8503 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8504 | mddev->resync_min = 0; | |
8505 | mddev->resync_max = MaxSector; | |
8506 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8507 | mddev->resync_min = mddev->curr_resync_completed; | |
f7851be7 | 8508 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); |
1da177e4 | 8509 | mddev->curr_resync = 0; |
23da422b N |
8510 | spin_unlock(&mddev->lock); |
8511 | ||
1da177e4 | 8512 | wake_up(&resync_wait); |
1da177e4 | 8513 | md_wakeup_thread(mddev->thread); |
c6207277 | 8514 | return; |
1da177e4 | 8515 | } |
29269553 | 8516 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 8517 | |
746d3207 N |
8518 | static int remove_and_add_spares(struct mddev *mddev, |
8519 | struct md_rdev *this) | |
b4c4c7b8 | 8520 | { |
3cb03002 | 8521 | struct md_rdev *rdev; |
b4c4c7b8 | 8522 | int spares = 0; |
f2a371c5 | 8523 | int removed = 0; |
d787be40 | 8524 | bool remove_some = false; |
b4c4c7b8 | 8525 | |
d787be40 N |
8526 | rdev_for_each(rdev, mddev) { |
8527 | if ((this == NULL || rdev == this) && | |
8528 | rdev->raid_disk >= 0 && | |
8529 | !test_bit(Blocked, &rdev->flags) && | |
8530 | test_bit(Faulty, &rdev->flags) && | |
8531 | atomic_read(&rdev->nr_pending)==0) { | |
8532 | /* Faulty non-Blocked devices with nr_pending == 0 | |
8533 | * never get nr_pending incremented, | |
8534 | * never get Faulty cleared, and never get Blocked set. | |
8535 | * So we can synchronize_rcu now rather than once per device | |
8536 | */ | |
8537 | remove_some = true; | |
8538 | set_bit(RemoveSynchronized, &rdev->flags); | |
8539 | } | |
8540 | } | |
8541 | ||
8542 | if (remove_some) | |
8543 | synchronize_rcu(); | |
8544 | rdev_for_each(rdev, mddev) { | |
746d3207 N |
8545 | if ((this == NULL || rdev == this) && |
8546 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 8547 | !test_bit(Blocked, &rdev->flags) && |
d787be40 | 8548 | ((test_bit(RemoveSynchronized, &rdev->flags) || |
f2076e7d SL |
8549 | (!test_bit(In_sync, &rdev->flags) && |
8550 | !test_bit(Journal, &rdev->flags))) && | |
d787be40 | 8551 | atomic_read(&rdev->nr_pending)==0)) { |
b4c4c7b8 | 8552 | if (mddev->pers->hot_remove_disk( |
b8321b68 | 8553 | mddev, rdev) == 0) { |
36fad858 | 8554 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 8555 | rdev->raid_disk = -1; |
f2a371c5 | 8556 | removed++; |
b4c4c7b8 N |
8557 | } |
8558 | } | |
d787be40 N |
8559 | if (remove_some && test_bit(RemoveSynchronized, &rdev->flags)) |
8560 | clear_bit(RemoveSynchronized, &rdev->flags); | |
8561 | } | |
8562 | ||
90584fc9 JB |
8563 | if (removed && mddev->kobj.sd) |
8564 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 8565 | |
2910ff17 | 8566 | if (this && removed) |
746d3207 N |
8567 | goto no_add; |
8568 | ||
dafb20fa | 8569 | rdev_for_each(rdev, mddev) { |
2910ff17 GR |
8570 | if (this && this != rdev) |
8571 | continue; | |
dbb64f86 GR |
8572 | if (test_bit(Candidate, &rdev->flags)) |
8573 | continue; | |
7bfec5f3 N |
8574 | if (rdev->raid_disk >= 0 && |
8575 | !test_bit(In_sync, &rdev->flags) && | |
f2076e7d | 8576 | !test_bit(Journal, &rdev->flags) && |
7bfec5f3 N |
8577 | !test_bit(Faulty, &rdev->flags)) |
8578 | spares++; | |
7ceb17e8 N |
8579 | if (rdev->raid_disk >= 0) |
8580 | continue; | |
8581 | if (test_bit(Faulty, &rdev->flags)) | |
8582 | continue; | |
f6b6ec5c SL |
8583 | if (!test_bit(Journal, &rdev->flags)) { |
8584 | if (mddev->ro && | |
8585 | ! (rdev->saved_raid_disk >= 0 && | |
8586 | !test_bit(Bitmap_sync, &rdev->flags))) | |
8587 | continue; | |
7ceb17e8 | 8588 | |
f6b6ec5c SL |
8589 | rdev->recovery_offset = 0; |
8590 | } | |
7ceb17e8 N |
8591 | if (mddev->pers-> |
8592 | hot_add_disk(mddev, rdev) == 0) { | |
8593 | if (sysfs_link_rdev(mddev, rdev)) | |
8594 | /* failure here is OK */; | |
f6b6ec5c SL |
8595 | if (!test_bit(Journal, &rdev->flags)) |
8596 | spares++; | |
7ceb17e8 | 8597 | md_new_event(mddev); |
2953079c | 8598 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
dfc70645 | 8599 | } |
b4c4c7b8 | 8600 | } |
746d3207 | 8601 | no_add: |
6dafab6b | 8602 | if (removed) |
2953079c | 8603 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
b4c4c7b8 N |
8604 | return spares; |
8605 | } | |
7ebc0be7 | 8606 | |
ac05f256 N |
8607 | static void md_start_sync(struct work_struct *ws) |
8608 | { | |
8609 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
c186b128 | 8610 | |
ac05f256 N |
8611 | mddev->sync_thread = md_register_thread(md_do_sync, |
8612 | mddev, | |
8613 | "resync"); | |
8614 | if (!mddev->sync_thread) { | |
9d48739e N |
8615 | pr_warn("%s: could not start resync thread...\n", |
8616 | mdname(mddev)); | |
ac05f256 N |
8617 | /* leave the spares where they are, it shouldn't hurt */ |
8618 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8619 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8620 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8621 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
8622 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8623 | wake_up(&resync_wait); |
ac05f256 N |
8624 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8625 | &mddev->recovery)) | |
8626 | if (mddev->sysfs_action) | |
8627 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8628 | } else | |
8629 | md_wakeup_thread(mddev->sync_thread); | |
8630 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8631 | md_new_event(mddev); | |
8632 | } | |
8633 | ||
1da177e4 LT |
8634 | /* |
8635 | * This routine is regularly called by all per-raid-array threads to | |
8636 | * deal with generic issues like resync and super-block update. | |
8637 | * Raid personalities that don't have a thread (linear/raid0) do not | |
8638 | * need this as they never do any recovery or update the superblock. | |
8639 | * | |
8640 | * It does not do any resync itself, but rather "forks" off other threads | |
8641 | * to do that as needed. | |
8642 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
8643 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 8644 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
8645 | * and wakeups up this thread which will reap the thread and finish up. |
8646 | * This thread also removes any faulty devices (with nr_pending == 0). | |
8647 | * | |
8648 | * The overall approach is: | |
8649 | * 1/ if the superblock needs updating, update it. | |
8650 | * 2/ If a recovery thread is running, don't do anything else. | |
8651 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
8652 | * 4/ If there are any faulty devices, remove them. | |
8653 | * 5/ If array is degraded, try to add spares devices | |
8654 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
8655 | */ | |
fd01b88c | 8656 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 8657 | { |
68866e42 JB |
8658 | if (mddev->suspended) |
8659 | return; | |
8660 | ||
5f40402d | 8661 | if (mddev->bitmap) |
aa5cbd10 | 8662 | bitmap_daemon_work(mddev); |
1da177e4 | 8663 | |
fca4d848 | 8664 | if (signal_pending(current)) { |
31a59e34 | 8665 | if (mddev->pers->sync_request && !mddev->external) { |
9d48739e N |
8666 | pr_debug("md: %s in immediate safe mode\n", |
8667 | mdname(mddev)); | |
fca4d848 N |
8668 | mddev->safemode = 2; |
8669 | } | |
8670 | flush_signals(current); | |
8671 | } | |
8672 | ||
c89a8eee N |
8673 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
8674 | return; | |
1da177e4 | 8675 | if ( ! ( |
2953079c | 8676 | (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) || |
1da177e4 | 8677 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 8678 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 8679 | (mddev->external == 0 && mddev->safemode == 1) || |
4ad23a97 | 8680 | (mddev->safemode == 2 |
fca4d848 | 8681 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) |
1da177e4 LT |
8682 | )) |
8683 | return; | |
fca4d848 | 8684 | |
df5b89b3 | 8685 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 8686 | int spares = 0; |
fca4d848 | 8687 | |
afc1f55c | 8688 | if (!mddev->external && mddev->safemode == 1) |
33182d15 N |
8689 | mddev->safemode = 0; |
8690 | ||
c89a8eee | 8691 | if (mddev->ro) { |
ab16bfc7 NB |
8692 | struct md_rdev *rdev; |
8693 | if (!mddev->external && mddev->in_sync) | |
8694 | /* 'Blocked' flag not needed as failed devices | |
8695 | * will be recorded if array switched to read/write. | |
8696 | * Leaving it set will prevent the device | |
8697 | * from being removed. | |
8698 | */ | |
8699 | rdev_for_each(rdev, mddev) | |
8700 | clear_bit(Blocked, &rdev->flags); | |
7ceb17e8 N |
8701 | /* On a read-only array we can: |
8702 | * - remove failed devices | |
8703 | * - add already-in_sync devices if the array itself | |
8704 | * is in-sync. | |
8705 | * As we only add devices that are already in-sync, | |
8706 | * we can activate the spares immediately. | |
c89a8eee | 8707 | */ |
7ceb17e8 | 8708 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
8709 | /* There is no thread, but we need to call |
8710 | * ->spare_active and clear saved_raid_disk | |
8711 | */ | |
2ac295a5 | 8712 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 | 8713 | md_reap_sync_thread(mddev); |
a4a3d26d | 8714 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
8313b8e5 | 8715 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2953079c | 8716 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
c89a8eee N |
8717 | goto unlock; |
8718 | } | |
8719 | ||
659b254f GJ |
8720 | if (mddev_is_clustered(mddev)) { |
8721 | struct md_rdev *rdev; | |
8722 | /* kick the device if another node issued a | |
8723 | * remove disk. | |
8724 | */ | |
8725 | rdev_for_each(rdev, mddev) { | |
8726 | if (test_and_clear_bit(ClusterRemove, &rdev->flags) && | |
8727 | rdev->raid_disk < 0) | |
8728 | md_kick_rdev_from_array(rdev); | |
8729 | } | |
8730 | } | |
8731 | ||
6497709b | 8732 | if (!mddev->external && !mddev->in_sync) { |
85572d7c | 8733 | spin_lock(&mddev->lock); |
6497709b | 8734 | set_in_sync(mddev); |
85572d7c | 8735 | spin_unlock(&mddev->lock); |
fca4d848 | 8736 | } |
fca4d848 | 8737 | |
2953079c | 8738 | if (mddev->sb_flags) |
850b2b42 | 8739 | md_update_sb(mddev, 0); |
06d91a5f | 8740 | |
1da177e4 LT |
8741 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
8742 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
8743 | /* resync/recovery still happening */ | |
8744 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8745 | goto unlock; | |
8746 | } | |
8747 | if (mddev->sync_thread) { | |
a91d5ac0 | 8748 | md_reap_sync_thread(mddev); |
1da177e4 LT |
8749 | goto unlock; |
8750 | } | |
72a23c21 NB |
8751 | /* Set RUNNING before clearing NEEDED to avoid |
8752 | * any transients in the value of "sync_action". | |
8753 | */ | |
72f36d59 | 8754 | mddev->curr_resync_completed = 0; |
23da422b | 8755 | spin_lock(&mddev->lock); |
72a23c21 | 8756 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 8757 | spin_unlock(&mddev->lock); |
24dd469d N |
8758 | /* Clear some bits that don't mean anything, but |
8759 | * might be left set | |
8760 | */ | |
24dd469d N |
8761 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8762 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 8763 | |
ed209584 N |
8764 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
8765 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 8766 | goto not_running; |
1da177e4 LT |
8767 | /* no recovery is running. |
8768 | * remove any failed drives, then | |
8769 | * add spares if possible. | |
72f36d59 | 8770 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
8771 | * the personality to fail the re-add. |
8772 | */ | |
1da177e4 | 8773 | |
b4c4c7b8 | 8774 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
8775 | if (mddev->pers->check_reshape == NULL || |
8776 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 8777 | /* Cannot proceed */ |
ac05f256 | 8778 | goto not_running; |
b4c4c7b8 | 8779 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 8780 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 8781 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
8782 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8783 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 8784 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 8785 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8786 | } else if (mddev->recovery_cp < MaxSector) { |
8787 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 8788 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8789 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
8790 | /* nothing to be done ... */ | |
ac05f256 | 8791 | goto not_running; |
24dd469d | 8792 | |
1da177e4 | 8793 | if (mddev->pers->sync_request) { |
ef99bf48 | 8794 | if (spares) { |
a654b9d8 N |
8795 | /* We are adding a device or devices to an array |
8796 | * which has the bitmap stored on all devices. | |
8797 | * So make sure all bitmap pages get written | |
8798 | */ | |
8799 | bitmap_write_all(mddev->bitmap); | |
8800 | } | |
ac05f256 N |
8801 | INIT_WORK(&mddev->del_work, md_start_sync); |
8802 | queue_work(md_misc_wq, &mddev->del_work); | |
8803 | goto unlock; | |
1da177e4 | 8804 | } |
ac05f256 | 8805 | not_running: |
72a23c21 NB |
8806 | if (!mddev->sync_thread) { |
8807 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8808 | wake_up(&resync_wait); |
72a23c21 NB |
8809 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8810 | &mddev->recovery)) | |
0c3573f1 | 8811 | if (mddev->sysfs_action) |
00bcb4ac | 8812 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 8813 | } |
ac05f256 N |
8814 | unlock: |
8815 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
8816 | mddev_unlock(mddev); |
8817 | } | |
8818 | } | |
6c144d31 | 8819 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8820 | |
a91d5ac0 JB |
8821 | void md_reap_sync_thread(struct mddev *mddev) |
8822 | { | |
8823 | struct md_rdev *rdev; | |
8824 | ||
8825 | /* resync has finished, collect result */ | |
8826 | md_unregister_thread(&mddev->sync_thread); | |
8827 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8828 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8829 | /* success...*/ | |
8830 | /* activate any spares */ | |
8831 | if (mddev->pers->spare_active(mddev)) { | |
8832 | sysfs_notify(&mddev->kobj, NULL, | |
8833 | "degraded"); | |
2953079c | 8834 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a91d5ac0 JB |
8835 | } |
8836 | } | |
8837 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
8838 | mddev->pers->finish_reshape) | |
8839 | mddev->pers->finish_reshape(mddev); | |
8840 | ||
8841 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8842 | * information must be scrapped. |
a91d5ac0 | 8843 | */ |
f466722c N |
8844 | if (!mddev->degraded) |
8845 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8846 | rdev->saved_raid_disk = -1; |
8847 | ||
8848 | md_update_sb(mddev, 1); | |
2953079c | 8849 | /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can |
bb8bf15b GJ |
8850 | * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by |
8851 | * clustered raid */ | |
8852 | if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags)) | |
8853 | md_cluster_ops->resync_finish(mddev); | |
a91d5ac0 | 8854 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
ea358cd0 | 8855 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
a91d5ac0 JB |
8856 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8857 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8858 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8859 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8860 | wake_up(&resync_wait); |
a91d5ac0 JB |
8861 | /* flag recovery needed just to double check */ |
8862 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8863 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8864 | md_new_event(mddev); | |
8865 | if (mddev->event_work.func) | |
8866 | queue_work(md_misc_wq, &mddev->event_work); | |
8867 | } | |
6c144d31 | 8868 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 8869 | |
fd01b88c | 8870 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8871 | { |
00bcb4ac | 8872 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8873 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8874 | !test_bit(Blocked, &rdev->flags) && |
8875 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8876 | msecs_to_jiffies(5000)); |
8877 | rdev_dec_pending(rdev, mddev); | |
8878 | } | |
8879 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8880 | ||
c6563a8c N |
8881 | void md_finish_reshape(struct mddev *mddev) |
8882 | { | |
8883 | /* called be personality module when reshape completes. */ | |
8884 | struct md_rdev *rdev; | |
8885 | ||
8886 | rdev_for_each(rdev, mddev) { | |
8887 | if (rdev->data_offset > rdev->new_data_offset) | |
8888 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8889 | else | |
8890 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8891 | rdev->data_offset = rdev->new_data_offset; | |
8892 | } | |
8893 | } | |
8894 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 | 8895 | |
fc974ee2 | 8896 | /* Bad block management */ |
2230dfe4 | 8897 | |
fc974ee2 | 8898 | /* Returns 1 on success, 0 on failure */ |
3cb03002 | 8899 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8900 | int is_new) |
2230dfe4 | 8901 | { |
85ad1d13 | 8902 | struct mddev *mddev = rdev->mddev; |
c6563a8c N |
8903 | int rv; |
8904 | if (is_new) | |
8905 | s += rdev->new_data_offset; | |
8906 | else | |
8907 | s += rdev->data_offset; | |
fc974ee2 VV |
8908 | rv = badblocks_set(&rdev->badblocks, s, sectors, 0); |
8909 | if (rv == 0) { | |
2230dfe4 | 8910 | /* Make sure they get written out promptly */ |
35b785f7 TM |
8911 | if (test_bit(ExternalBbl, &rdev->flags)) |
8912 | sysfs_notify(&rdev->kobj, NULL, | |
8913 | "unacknowledged_bad_blocks"); | |
8bd2f0a0 | 8914 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2953079c SL |
8915 | set_mask_bits(&mddev->sb_flags, 0, |
8916 | BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING)); | |
2230dfe4 | 8917 | md_wakeup_thread(rdev->mddev->thread); |
fc974ee2 VV |
8918 | return 1; |
8919 | } else | |
8920 | return 0; | |
2230dfe4 N |
8921 | } |
8922 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8923 | ||
c6563a8c N |
8924 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8925 | int is_new) | |
2230dfe4 | 8926 | { |
35b785f7 | 8927 | int rv; |
c6563a8c N |
8928 | if (is_new) |
8929 | s += rdev->new_data_offset; | |
8930 | else | |
8931 | s += rdev->data_offset; | |
35b785f7 TM |
8932 | rv = badblocks_clear(&rdev->badblocks, s, sectors); |
8933 | if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags)) | |
8934 | sysfs_notify(&rdev->kobj, NULL, "bad_blocks"); | |
8935 | return rv; | |
2230dfe4 N |
8936 | } |
8937 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8938 | ||
75c96f85 AB |
8939 | static int md_notify_reboot(struct notifier_block *this, |
8940 | unsigned long code, void *x) | |
1da177e4 LT |
8941 | { |
8942 | struct list_head *tmp; | |
fd01b88c | 8943 | struct mddev *mddev; |
2dba6a91 | 8944 | int need_delay = 0; |
1da177e4 | 8945 | |
c744a65c N |
8946 | for_each_mddev(mddev, tmp) { |
8947 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8948 | if (mddev->pers) |
8949 | __md_stop_writes(mddev); | |
0f62fb22 N |
8950 | if (mddev->persistent) |
8951 | mddev->safemode = 2; | |
c744a65c | 8952 | mddev_unlock(mddev); |
2dba6a91 | 8953 | } |
c744a65c | 8954 | need_delay = 1; |
1da177e4 | 8955 | } |
c744a65c N |
8956 | /* |
8957 | * certain more exotic SCSI devices are known to be | |
8958 | * volatile wrt too early system reboots. While the | |
8959 | * right place to handle this issue is the given | |
8960 | * driver, we do want to have a safe RAID driver ... | |
8961 | */ | |
8962 | if (need_delay) | |
8963 | mdelay(1000*1); | |
8964 | ||
1da177e4 LT |
8965 | return NOTIFY_DONE; |
8966 | } | |
8967 | ||
75c96f85 | 8968 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8969 | .notifier_call = md_notify_reboot, |
8970 | .next = NULL, | |
8971 | .priority = INT_MAX, /* before any real devices */ | |
8972 | }; | |
8973 | ||
8974 | static void md_geninit(void) | |
8975 | { | |
36a4e1fe | 8976 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8977 | |
c7705f34 | 8978 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8979 | } |
8980 | ||
75c96f85 | 8981 | static int __init md_init(void) |
1da177e4 | 8982 | { |
e804ac78 TH |
8983 | int ret = -ENOMEM; |
8984 | ||
ada609ee | 8985 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8986 | if (!md_wq) |
8987 | goto err_wq; | |
8988 | ||
8989 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8990 | if (!md_misc_wq) | |
8991 | goto err_misc_wq; | |
8992 | ||
8993 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8994 | goto err_md; | |
8995 | ||
8996 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8997 | goto err_mdp; | |
8998 | mdp_major = ret; | |
8999 | ||
af5628f0 | 9000 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
9001 | md_probe, NULL, NULL); |
9002 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
9003 | md_probe, NULL, NULL); |
9004 | ||
1da177e4 | 9005 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 9006 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
9007 | |
9008 | md_geninit(); | |
d710e138 | 9009 | return 0; |
1da177e4 | 9010 | |
e804ac78 TH |
9011 | err_mdp: |
9012 | unregister_blkdev(MD_MAJOR, "md"); | |
9013 | err_md: | |
9014 | destroy_workqueue(md_misc_wq); | |
9015 | err_misc_wq: | |
9016 | destroy_workqueue(md_wq); | |
9017 | err_wq: | |
9018 | return ret; | |
9019 | } | |
1da177e4 | 9020 | |
70bcecdb | 9021 | static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev) |
1d7e3e96 | 9022 | { |
70bcecdb GR |
9023 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
9024 | struct md_rdev *rdev2; | |
9025 | int role, ret; | |
9026 | char b[BDEVNAME_SIZE]; | |
1d7e3e96 | 9027 | |
818da59f GJ |
9028 | /* |
9029 | * If size is changed in another node then we need to | |
9030 | * do resize as well. | |
9031 | */ | |
9032 | if (mddev->dev_sectors != le64_to_cpu(sb->size)) { | |
9033 | ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size)); | |
9034 | if (ret) | |
9035 | pr_info("md-cluster: resize failed\n"); | |
9036 | else | |
9037 | bitmap_update_sb(mddev->bitmap); | |
9038 | } | |
9039 | ||
70bcecdb GR |
9040 | /* Check for change of roles in the active devices */ |
9041 | rdev_for_each(rdev2, mddev) { | |
9042 | if (test_bit(Faulty, &rdev2->flags)) | |
9043 | continue; | |
9044 | ||
9045 | /* Check if the roles changed */ | |
9046 | role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]); | |
dbb64f86 GR |
9047 | |
9048 | if (test_bit(Candidate, &rdev2->flags)) { | |
9049 | if (role == 0xfffe) { | |
9050 | pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b)); | |
9051 | md_kick_rdev_from_array(rdev2); | |
9052 | continue; | |
9053 | } | |
9054 | else | |
9055 | clear_bit(Candidate, &rdev2->flags); | |
9056 | } | |
9057 | ||
70bcecdb GR |
9058 | if (role != rdev2->raid_disk) { |
9059 | /* got activated */ | |
9060 | if (rdev2->raid_disk == -1 && role != 0xffff) { | |
9061 | rdev2->saved_raid_disk = role; | |
9062 | ret = remove_and_add_spares(mddev, rdev2); | |
9063 | pr_info("Activated spare: %s\n", | |
9d48739e | 9064 | bdevname(rdev2->bdev,b)); |
a578183e GJ |
9065 | /* wakeup mddev->thread here, so array could |
9066 | * perform resync with the new activated disk */ | |
9067 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
9068 | md_wakeup_thread(mddev->thread); | |
9069 | ||
70bcecdb GR |
9070 | } |
9071 | /* device faulty | |
9072 | * We just want to do the minimum to mark the disk | |
9073 | * as faulty. The recovery is performed by the | |
9074 | * one who initiated the error. | |
9075 | */ | |
9076 | if ((role == 0xfffe) || (role == 0xfffd)) { | |
9077 | md_error(mddev, rdev2); | |
9078 | clear_bit(Blocked, &rdev2->flags); | |
9079 | } | |
9080 | } | |
1d7e3e96 | 9081 | } |
70bcecdb | 9082 | |
28c1b9fd GR |
9083 | if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) |
9084 | update_raid_disks(mddev, le32_to_cpu(sb->raid_disks)); | |
70bcecdb GR |
9085 | |
9086 | /* Finally set the event to be up to date */ | |
9087 | mddev->events = le64_to_cpu(sb->events); | |
9088 | } | |
9089 | ||
9090 | static int read_rdev(struct mddev *mddev, struct md_rdev *rdev) | |
9091 | { | |
9092 | int err; | |
9093 | struct page *swapout = rdev->sb_page; | |
9094 | struct mdp_superblock_1 *sb; | |
9095 | ||
9096 | /* Store the sb page of the rdev in the swapout temporary | |
9097 | * variable in case we err in the future | |
9098 | */ | |
9099 | rdev->sb_page = NULL; | |
7f0f0d87 N |
9100 | err = alloc_disk_sb(rdev); |
9101 | if (err == 0) { | |
9102 | ClearPageUptodate(rdev->sb_page); | |
9103 | rdev->sb_loaded = 0; | |
9104 | err = super_types[mddev->major_version]. | |
9105 | load_super(rdev, NULL, mddev->minor_version); | |
9106 | } | |
70bcecdb GR |
9107 | if (err < 0) { |
9108 | pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n", | |
9109 | __func__, __LINE__, rdev->desc_nr, err); | |
7f0f0d87 N |
9110 | if (rdev->sb_page) |
9111 | put_page(rdev->sb_page); | |
70bcecdb GR |
9112 | rdev->sb_page = swapout; |
9113 | rdev->sb_loaded = 1; | |
9114 | return err; | |
1d7e3e96 GR |
9115 | } |
9116 | ||
70bcecdb GR |
9117 | sb = page_address(rdev->sb_page); |
9118 | /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET | |
9119 | * is not set | |
9120 | */ | |
9121 | ||
9122 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)) | |
9123 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
9124 | ||
9125 | /* The other node finished recovery, call spare_active to set | |
9126 | * device In_sync and mddev->degraded | |
9127 | */ | |
9128 | if (rdev->recovery_offset == MaxSector && | |
9129 | !test_bit(In_sync, &rdev->flags) && | |
9130 | mddev->pers->spare_active(mddev)) | |
9131 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
9132 | ||
9133 | put_page(swapout); | |
9134 | return 0; | |
9135 | } | |
9136 | ||
9137 | void md_reload_sb(struct mddev *mddev, int nr) | |
9138 | { | |
9139 | struct md_rdev *rdev; | |
9140 | int err; | |
9141 | ||
9142 | /* Find the rdev */ | |
9143 | rdev_for_each_rcu(rdev, mddev) { | |
9144 | if (rdev->desc_nr == nr) | |
9145 | break; | |
9146 | } | |
9147 | ||
9148 | if (!rdev || rdev->desc_nr != nr) { | |
9149 | pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr); | |
9150 | return; | |
9151 | } | |
9152 | ||
9153 | err = read_rdev(mddev, rdev); | |
9154 | if (err < 0) | |
9155 | return; | |
9156 | ||
9157 | check_sb_changes(mddev, rdev); | |
9158 | ||
9159 | /* Read all rdev's to update recovery_offset */ | |
9160 | rdev_for_each_rcu(rdev, mddev) | |
9161 | read_rdev(mddev, rdev); | |
1d7e3e96 GR |
9162 | } |
9163 | EXPORT_SYMBOL(md_reload_sb); | |
9164 | ||
1da177e4 LT |
9165 | #ifndef MODULE |
9166 | ||
9167 | /* | |
9168 | * Searches all registered partitions for autorun RAID arrays | |
9169 | * at boot time. | |
9170 | */ | |
4d936ec1 | 9171 | |
5b1f5bc3 | 9172 | static DEFINE_MUTEX(detected_devices_mutex); |
4d936ec1 ME |
9173 | static LIST_HEAD(all_detected_devices); |
9174 | struct detected_devices_node { | |
9175 | struct list_head list; | |
9176 | dev_t dev; | |
9177 | }; | |
1da177e4 LT |
9178 | |
9179 | void md_autodetect_dev(dev_t dev) | |
9180 | { | |
4d936ec1 ME |
9181 | struct detected_devices_node *node_detected_dev; |
9182 | ||
9183 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
9184 | if (node_detected_dev) { | |
9185 | node_detected_dev->dev = dev; | |
5b1f5bc3 | 9186 | mutex_lock(&detected_devices_mutex); |
4d936ec1 | 9187 | list_add_tail(&node_detected_dev->list, &all_detected_devices); |
5b1f5bc3 | 9188 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 9189 | } |
1da177e4 LT |
9190 | } |
9191 | ||
1da177e4 LT |
9192 | static void autostart_arrays(int part) |
9193 | { | |
3cb03002 | 9194 | struct md_rdev *rdev; |
4d936ec1 ME |
9195 | struct detected_devices_node *node_detected_dev; |
9196 | dev_t dev; | |
9197 | int i_scanned, i_passed; | |
1da177e4 | 9198 | |
4d936ec1 ME |
9199 | i_scanned = 0; |
9200 | i_passed = 0; | |
1da177e4 | 9201 | |
9d48739e | 9202 | pr_info("md: Autodetecting RAID arrays.\n"); |
1da177e4 | 9203 | |
5b1f5bc3 | 9204 | mutex_lock(&detected_devices_mutex); |
4d936ec1 ME |
9205 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
9206 | i_scanned++; | |
9207 | node_detected_dev = list_entry(all_detected_devices.next, | |
9208 | struct detected_devices_node, list); | |
9209 | list_del(&node_detected_dev->list); | |
9210 | dev = node_detected_dev->dev; | |
9211 | kfree(node_detected_dev); | |
90bcf133 | 9212 | mutex_unlock(&detected_devices_mutex); |
df968c4e | 9213 | rdev = md_import_device(dev,0, 90); |
90bcf133 | 9214 | mutex_lock(&detected_devices_mutex); |
1da177e4 LT |
9215 | if (IS_ERR(rdev)) |
9216 | continue; | |
9217 | ||
403df478 | 9218 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 9219 | continue; |
403df478 | 9220 | |
d0fae18f | 9221 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 9222 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 9223 | i_passed++; |
1da177e4 | 9224 | } |
5b1f5bc3 | 9225 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 9226 | |
9d48739e | 9227 | pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed); |
1da177e4 LT |
9228 | |
9229 | autorun_devices(part); | |
9230 | } | |
9231 | ||
fdee8ae4 | 9232 | #endif /* !MODULE */ |
1da177e4 LT |
9233 | |
9234 | static __exit void md_exit(void) | |
9235 | { | |
fd01b88c | 9236 | struct mddev *mddev; |
1da177e4 | 9237 | struct list_head *tmp; |
e2f23b60 | 9238 | int delay = 1; |
8ab5e4c1 | 9239 | |
af5628f0 | 9240 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 9241 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 9242 | |
3dbd8c2e | 9243 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
9244 | unregister_blkdev(mdp_major, "mdp"); |
9245 | unregister_reboot_notifier(&md_notifier); | |
9246 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
9247 | |
9248 | /* We cannot unload the modules while some process is | |
9249 | * waiting for us in select() or poll() - wake them up | |
9250 | */ | |
9251 | md_unloading = 1; | |
9252 | while (waitqueue_active(&md_event_waiters)) { | |
9253 | /* not safe to leave yet */ | |
9254 | wake_up(&md_event_waiters); | |
9255 | msleep(delay); | |
9256 | delay += delay; | |
9257 | } | |
1da177e4 | 9258 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 9259 | |
29ac4aa3 | 9260 | for_each_mddev(mddev, tmp) { |
1da177e4 | 9261 | export_array(mddev); |
9356863c | 9262 | mddev->ctime = 0; |
d3374825 | 9263 | mddev->hold_active = 0; |
9356863c N |
9264 | /* |
9265 | * for_each_mddev() will call mddev_put() at the end of each | |
9266 | * iteration. As the mddev is now fully clear, this will | |
9267 | * schedule the mddev for destruction by a workqueue, and the | |
9268 | * destroy_workqueue() below will wait for that to complete. | |
9269 | */ | |
1da177e4 | 9270 | } |
e804ac78 TH |
9271 | destroy_workqueue(md_misc_wq); |
9272 | destroy_workqueue(md_wq); | |
1da177e4 LT |
9273 | } |
9274 | ||
685784aa | 9275 | subsys_initcall(md_init); |
1da177e4 LT |
9276 | module_exit(md_exit) |
9277 | ||
f91de92e N |
9278 | static int get_ro(char *buffer, struct kernel_param *kp) |
9279 | { | |
9280 | return sprintf(buffer, "%d", start_readonly); | |
9281 | } | |
9282 | static int set_ro(const char *val, struct kernel_param *kp) | |
9283 | { | |
4c9309c0 | 9284 | return kstrtouint(val, 10, (unsigned int *)&start_readonly); |
f91de92e N |
9285 | } |
9286 | ||
80ca3a44 N |
9287 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
9288 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 9289 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
78b6350d | 9290 | module_param(create_on_open, bool, S_IRUSR|S_IWUSR); |
f91de92e | 9291 | |
1da177e4 | 9292 | MODULE_LICENSE("GPL"); |
0efb9e61 | 9293 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 9294 | MODULE_ALIAS("md"); |
72008652 | 9295 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |