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 | */ | |
b3143b9a N |
269 | static bool is_suspended(struct mddev *mddev, struct bio *bio) |
270 | { | |
271 | if (mddev->suspended) | |
272 | return true; | |
273 | if (bio_data_dir(bio) != WRITE) | |
274 | return false; | |
275 | if (mddev->suspend_lo >= mddev->suspend_hi) | |
276 | return false; | |
277 | if (bio->bi_iter.bi_sector >= mddev->suspend_hi) | |
278 | return false; | |
279 | if (bio_end_sector(bio) < mddev->suspend_lo) | |
280 | return false; | |
281 | return true; | |
282 | } | |
283 | ||
393debc2 SL |
284 | void md_handle_request(struct mddev *mddev, struct bio *bio) |
285 | { | |
286 | check_suspended: | |
287 | rcu_read_lock(); | |
b3143b9a | 288 | if (is_suspended(mddev, bio)) { |
393debc2 SL |
289 | DEFINE_WAIT(__wait); |
290 | for (;;) { | |
291 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
292 | TASK_UNINTERRUPTIBLE); | |
b3143b9a | 293 | if (!is_suspended(mddev, bio)) |
393debc2 SL |
294 | break; |
295 | rcu_read_unlock(); | |
296 | schedule(); | |
297 | rcu_read_lock(); | |
298 | } | |
299 | finish_wait(&mddev->sb_wait, &__wait); | |
300 | } | |
301 | atomic_inc(&mddev->active_io); | |
302 | rcu_read_unlock(); | |
303 | ||
304 | if (!mddev->pers->make_request(mddev, bio)) { | |
305 | atomic_dec(&mddev->active_io); | |
306 | wake_up(&mddev->sb_wait); | |
307 | goto check_suspended; | |
308 | } | |
309 | ||
310 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) | |
311 | wake_up(&mddev->sb_wait); | |
312 | } | |
313 | EXPORT_SYMBOL(md_handle_request); | |
314 | ||
dece1635 | 315 | static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 316 | { |
49077326 | 317 | const int rw = bio_data_dir(bio); |
fd01b88c | 318 | struct mddev *mddev = q->queuedata; |
e91ece55 | 319 | unsigned int sectors; |
74672d06 | 320 | int cpu; |
49077326 | 321 | |
af67c31f | 322 | blk_queue_split(q, &bio); |
54efd50b | 323 | |
274d8cbd | 324 | if (mddev == NULL || mddev->pers == NULL) { |
409c57f3 | 325 | bio_io_error(bio); |
dece1635 | 326 | return BLK_QC_T_NONE; |
409c57f3 | 327 | } |
bbfa57c0 | 328 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
4246a0b6 | 329 | if (bio_sectors(bio) != 0) |
4e4cbee9 | 330 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 331 | bio_endio(bio); |
dece1635 | 332 | return BLK_QC_T_NONE; |
bbfa57c0 | 333 | } |
49077326 | 334 | |
e91ece55 CM |
335 | /* |
336 | * save the sectors now since our bio can | |
337 | * go away inside make_request | |
338 | */ | |
339 | sectors = bio_sectors(bio); | |
9c573de3 | 340 | /* bio could be mergeable after passing to underlayer */ |
1eff9d32 | 341 | bio->bi_opf &= ~REQ_NOMERGE; |
393debc2 SL |
342 | |
343 | md_handle_request(mddev, bio); | |
49077326 | 344 | |
74672d06 GZ |
345 | cpu = part_stat_lock(); |
346 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
347 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); | |
348 | part_stat_unlock(); | |
49077326 | 349 | |
dece1635 | 350 | return BLK_QC_T_NONE; |
409c57f3 N |
351 | } |
352 | ||
9e35b99c N |
353 | /* mddev_suspend makes sure no new requests are submitted |
354 | * to the device, and that any requests that have been submitted | |
355 | * are completely handled. | |
afa0f557 N |
356 | * Once mddev_detach() is called and completes, the module will be |
357 | * completely unused. | |
9e35b99c | 358 | */ |
fd01b88c | 359 | void mddev_suspend(struct mddev *mddev) |
409c57f3 | 360 | { |
092398dc | 361 | WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk); |
4d5324f7 | 362 | lockdep_assert_held(&mddev->reconfig_mutex); |
0dc10e50 MP |
363 | if (mddev->suspended++) |
364 | return; | |
409c57f3 | 365 | synchronize_rcu(); |
cc27b0c7 | 366 | wake_up(&mddev->sb_wait); |
35bfc521 N |
367 | set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags); |
368 | smp_mb__after_atomic(); | |
409c57f3 N |
369 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); |
370 | mddev->pers->quiesce(mddev, 1); | |
35bfc521 N |
371 | clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags); |
372 | wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags)); | |
0d9f4f13 JB |
373 | |
374 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 375 | } |
390ee602 | 376 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 377 | |
fd01b88c | 378 | void mddev_resume(struct mddev *mddev) |
409c57f3 | 379 | { |
4d5324f7 | 380 | lockdep_assert_held(&mddev->reconfig_mutex); |
0dc10e50 MP |
381 | if (--mddev->suspended) |
382 | return; | |
409c57f3 N |
383 | wake_up(&mddev->sb_wait); |
384 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 385 | |
47525e59 | 386 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
387 | md_wakeup_thread(mddev->thread); |
388 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 389 | } |
390ee602 | 390 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 391 | |
fd01b88c | 392 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 | 393 | { |
5c675f83 N |
394 | struct md_personality *pers = mddev->pers; |
395 | int ret = 0; | |
396 | ||
397 | rcu_read_lock(); | |
398 | if (mddev->suspended) | |
399 | ret = 1; | |
400 | else if (pers && pers->congested) | |
401 | ret = pers->congested(mddev, bits); | |
402 | rcu_read_unlock(); | |
403 | return ret; | |
404 | } | |
405 | EXPORT_SYMBOL_GPL(mddev_congested); | |
406 | static int md_congested(void *data, int bits) | |
407 | { | |
408 | struct mddev *mddev = data; | |
409 | return mddev_congested(mddev, bits); | |
3fa841d7 | 410 | } |
3fa841d7 | 411 | |
a2826aa9 | 412 | /* |
e9c7469b | 413 | * Generic flush handling for md |
a2826aa9 N |
414 | */ |
415 | ||
4246a0b6 | 416 | static void md_end_flush(struct bio *bio) |
a2826aa9 | 417 | { |
3cb03002 | 418 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 419 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
420 | |
421 | rdev_dec_pending(rdev, mddev); | |
422 | ||
423 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 424 | /* The pre-request flush has finished */ |
e804ac78 | 425 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
426 | } |
427 | bio_put(bio); | |
428 | } | |
429 | ||
a7a07e69 N |
430 | static void md_submit_flush_data(struct work_struct *ws); |
431 | ||
a035fc3e | 432 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 433 | { |
fd01b88c | 434 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 435 | struct md_rdev *rdev; |
a2826aa9 | 436 | |
a7a07e69 N |
437 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
438 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 439 | rcu_read_lock(); |
dafb20fa | 440 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
441 | if (rdev->raid_disk >= 0 && |
442 | !test_bit(Faulty, &rdev->flags)) { | |
443 | /* Take two references, one is dropped | |
444 | * when request finishes, one after | |
445 | * we reclaim rcu_read_lock | |
446 | */ | |
447 | struct bio *bi; | |
448 | atomic_inc(&rdev->nr_pending); | |
449 | atomic_inc(&rdev->nr_pending); | |
450 | rcu_read_unlock(); | |
b5e1b8ce | 451 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 452 | bi->bi_end_io = md_end_flush; |
a2826aa9 | 453 | bi->bi_private = rdev; |
74d46992 | 454 | bio_set_dev(bi, rdev->bdev); |
70fd7614 | 455 | bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
a2826aa9 | 456 | atomic_inc(&mddev->flush_pending); |
4e49ea4a | 457 | submit_bio(bi); |
a2826aa9 N |
458 | rcu_read_lock(); |
459 | rdev_dec_pending(rdev, mddev); | |
460 | } | |
461 | rcu_read_unlock(); | |
a7a07e69 N |
462 | if (atomic_dec_and_test(&mddev->flush_pending)) |
463 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
464 | } |
465 | ||
e9c7469b | 466 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 467 | { |
fd01b88c | 468 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 469 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 470 | |
79bf31a3 SL |
471 | /* |
472 | * must reset flush_bio before calling into md_handle_request to avoid a | |
473 | * deadlock, because other bios passed md_handle_request suspend check | |
474 | * could wait for this and below md_handle_request could wait for those | |
475 | * bios because of suspend check | |
476 | */ | |
477 | mddev->flush_bio = NULL; | |
478 | wake_up(&mddev->sb_wait); | |
479 | ||
4f024f37 | 480 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 | 481 | /* an empty barrier - all done */ |
4246a0b6 | 482 | bio_endio(bio); |
a2826aa9 | 483 | else { |
1eff9d32 | 484 | bio->bi_opf &= ~REQ_PREFLUSH; |
79bf31a3 | 485 | md_handle_request(mddev, bio); |
a2826aa9 | 486 | } |
a2826aa9 N |
487 | } |
488 | ||
fd01b88c | 489 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 | 490 | { |
85572d7c | 491 | spin_lock_irq(&mddev->lock); |
a2826aa9 | 492 | wait_event_lock_irq(mddev->sb_wait, |
e9c7469b | 493 | !mddev->flush_bio, |
85572d7c | 494 | mddev->lock); |
e9c7469b | 495 | mddev->flush_bio = bio; |
85572d7c | 496 | spin_unlock_irq(&mddev->lock); |
a2826aa9 | 497 | |
a035fc3e N |
498 | INIT_WORK(&mddev->flush_work, submit_flushes); |
499 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 500 | } |
e9c7469b | 501 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 502 | |
fd01b88c | 503 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
504 | { |
505 | atomic_inc(&mddev->active); | |
506 | return mddev; | |
507 | } | |
508 | ||
5fd3a17e | 509 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 510 | |
fd01b88c | 511 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 512 | { |
7184ef8b | 513 | struct bio_set *bs = NULL, *sync_bs = NULL; |
a167f663 | 514 | |
1da177e4 LT |
515 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
516 | return; | |
d3374825 | 517 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
518 | mddev->ctime == 0 && !mddev->hold_active) { |
519 | /* Array is not configured at all, and not held active, | |
520 | * so destroy it */ | |
af8a2434 | 521 | list_del_init(&mddev->all_mddevs); |
a167f663 | 522 | bs = mddev->bio_set; |
7184ef8b | 523 | sync_bs = mddev->sync_set; |
a167f663 | 524 | mddev->bio_set = NULL; |
5a85071c | 525 | mddev->sync_set = NULL; |
d3374825 | 526 | if (mddev->gendisk) { |
e804ac78 TH |
527 | /* We did a probe so need to clean up. Call |
528 | * queue_work inside the spinlock so that | |
529 | * flush_workqueue() after mddev_find will | |
530 | * succeed in waiting for the work to be done. | |
d3374825 N |
531 | */ |
532 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 533 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
534 | } else |
535 | kfree(mddev); | |
536 | } | |
537 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
538 | if (bs) |
539 | bioset_free(bs); | |
7184ef8b SL |
540 | if (sync_bs) |
541 | bioset_free(sync_bs); | |
1da177e4 LT |
542 | } |
543 | ||
8376d3c1 | 544 | static void md_safemode_timeout(struct timer_list *t); |
25b2edfa | 545 | |
fd01b88c | 546 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
547 | { |
548 | mutex_init(&mddev->open_mutex); | |
549 | mutex_init(&mddev->reconfig_mutex); | |
550 | mutex_init(&mddev->bitmap_info.mutex); | |
551 | INIT_LIST_HEAD(&mddev->disks); | |
552 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
8376d3c1 | 553 | timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0); |
fafd7fb0 N |
554 | atomic_set(&mddev->active, 1); |
555 | atomic_set(&mddev->openers, 0); | |
556 | atomic_set(&mddev->active_io, 0); | |
85572d7c | 557 | spin_lock_init(&mddev->lock); |
fafd7fb0 N |
558 | atomic_set(&mddev->flush_pending, 0); |
559 | init_waitqueue_head(&mddev->sb_wait); | |
560 | init_waitqueue_head(&mddev->recovery_wait); | |
561 | mddev->reshape_position = MaxSector; | |
2c810cdd | 562 | mddev->reshape_backwards = 0; |
c4a39551 | 563 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
564 | mddev->resync_min = 0; |
565 | mddev->resync_max = MaxSector; | |
566 | mddev->level = LEVEL_NONE; | |
567 | } | |
390ee602 | 568 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 569 | |
f72ffdd6 | 570 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 571 | { |
fd01b88c | 572 | struct mddev *mddev, *new = NULL; |
1da177e4 | 573 | |
8f5f02c4 N |
574 | if (unit && MAJOR(unit) != MD_MAJOR) |
575 | unit &= ~((1<<MdpMinorShift)-1); | |
576 | ||
1da177e4 LT |
577 | retry: |
578 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
579 | |
580 | if (unit) { | |
581 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
582 | if (mddev->unit == unit) { | |
583 | mddev_get(mddev); | |
584 | spin_unlock(&all_mddevs_lock); | |
585 | kfree(new); | |
586 | return mddev; | |
587 | } | |
588 | ||
589 | if (new) { | |
590 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 591 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
592 | new->hold_active = UNTIL_IOCTL; |
593 | return new; | |
1da177e4 | 594 | } |
efeb53c0 N |
595 | } else if (new) { |
596 | /* find an unused unit number */ | |
597 | static int next_minor = 512; | |
598 | int start = next_minor; | |
599 | int is_free = 0; | |
600 | int dev = 0; | |
601 | while (!is_free) { | |
602 | dev = MKDEV(MD_MAJOR, next_minor); | |
603 | next_minor++; | |
604 | if (next_minor > MINORMASK) | |
605 | next_minor = 0; | |
606 | if (next_minor == start) { | |
607 | /* Oh dear, all in use. */ | |
608 | spin_unlock(&all_mddevs_lock); | |
609 | kfree(new); | |
610 | return NULL; | |
611 | } | |
f72ffdd6 | 612 | |
efeb53c0 N |
613 | is_free = 1; |
614 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
615 | if (mddev->unit == dev) { | |
616 | is_free = 0; | |
617 | break; | |
618 | } | |
619 | } | |
620 | new->unit = dev; | |
621 | new->md_minor = MINOR(dev); | |
622 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
623 | list_add(&new->all_mddevs, &all_mddevs); |
624 | spin_unlock(&all_mddevs_lock); | |
625 | return new; | |
626 | } | |
627 | spin_unlock(&all_mddevs_lock); | |
628 | ||
9ffae0cf | 629 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
630 | if (!new) |
631 | return NULL; | |
632 | ||
1da177e4 LT |
633 | new->unit = unit; |
634 | if (MAJOR(unit) == MD_MAJOR) | |
635 | new->md_minor = MINOR(unit); | |
636 | else | |
637 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
638 | ||
fafd7fb0 | 639 | mddev_init(new); |
1da177e4 | 640 | |
1da177e4 LT |
641 | goto retry; |
642 | } | |
643 | ||
b6eb127d N |
644 | static struct attribute_group md_redundancy_group; |
645 | ||
5c47daf6 | 646 | void mddev_unlock(struct mddev *mddev) |
1da177e4 | 647 | { |
a64c876f | 648 | if (mddev->to_remove) { |
b6eb127d N |
649 | /* These cannot be removed under reconfig_mutex as |
650 | * an access to the files will try to take reconfig_mutex | |
651 | * while holding the file unremovable, which leads to | |
652 | * a deadlock. | |
bb4f1e9d N |
653 | * So hold set sysfs_active while the remove in happeing, |
654 | * and anything else which might set ->to_remove or my | |
655 | * otherwise change the sysfs namespace will fail with | |
656 | * -EBUSY if sysfs_active is still set. | |
657 | * We set sysfs_active under reconfig_mutex and elsewhere | |
658 | * test it under the same mutex to ensure its correct value | |
659 | * is seen. | |
b6eb127d | 660 | */ |
a64c876f N |
661 | struct attribute_group *to_remove = mddev->to_remove; |
662 | mddev->to_remove = NULL; | |
bb4f1e9d | 663 | mddev->sysfs_active = 1; |
b6eb127d N |
664 | mutex_unlock(&mddev->reconfig_mutex); |
665 | ||
00bcb4ac N |
666 | if (mddev->kobj.sd) { |
667 | if (to_remove != &md_redundancy_group) | |
668 | sysfs_remove_group(&mddev->kobj, to_remove); | |
669 | if (mddev->pers == NULL || | |
670 | mddev->pers->sync_request == NULL) { | |
671 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
672 | if (mddev->sysfs_action) | |
673 | sysfs_put(mddev->sysfs_action); | |
674 | mddev->sysfs_action = NULL; | |
675 | } | |
a64c876f | 676 | } |
bb4f1e9d | 677 | mddev->sysfs_active = 0; |
b6eb127d N |
678 | } else |
679 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 680 | |
751e67ca CD |
681 | /* As we've dropped the mutex we need a spinlock to |
682 | * make sure the thread doesn't disappear | |
01f96c0a N |
683 | */ |
684 | spin_lock(&pers_lock); | |
005eca5e | 685 | md_wakeup_thread(mddev->thread); |
4d5324f7 | 686 | wake_up(&mddev->sb_wait); |
01f96c0a | 687 | spin_unlock(&pers_lock); |
1da177e4 | 688 | } |
5c47daf6 | 689 | EXPORT_SYMBOL_GPL(mddev_unlock); |
1da177e4 | 690 | |
57d051dc | 691 | struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr) |
1ca69c4b N |
692 | { |
693 | struct md_rdev *rdev; | |
694 | ||
695 | rdev_for_each_rcu(rdev, mddev) | |
696 | if (rdev->desc_nr == nr) | |
697 | return rdev; | |
698 | ||
699 | return NULL; | |
700 | } | |
57d051dc | 701 | EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu); |
1ca69c4b N |
702 | |
703 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 704 | { |
3cb03002 | 705 | struct md_rdev *rdev; |
1da177e4 | 706 | |
dafb20fa | 707 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
708 | if (rdev->bdev->bd_dev == dev) |
709 | return rdev; | |
159ec1fc | 710 | |
1da177e4 LT |
711 | return NULL; |
712 | } | |
713 | ||
1ca69c4b N |
714 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
715 | { | |
716 | struct md_rdev *rdev; | |
717 | ||
718 | rdev_for_each_rcu(rdev, mddev) | |
719 | if (rdev->bdev->bd_dev == dev) | |
720 | return rdev; | |
721 | ||
722 | return NULL; | |
723 | } | |
724 | ||
84fc4b56 | 725 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 726 | { |
84fc4b56 | 727 | struct md_personality *pers; |
d9d166c2 N |
728 | list_for_each_entry(pers, &pers_list, list) { |
729 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 730 | return pers; |
d9d166c2 N |
731 | if (strcmp(pers->name, clevel)==0) |
732 | return pers; | |
733 | } | |
2604b703 N |
734 | return NULL; |
735 | } | |
736 | ||
b73df2d3 | 737 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 738 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 739 | { |
57b2caa3 | 740 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 741 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
742 | } |
743 | ||
f72ffdd6 | 744 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 745 | { |
1da177e4 | 746 | rdev->sb_page = alloc_page(GFP_KERNEL); |
7f0f0d87 | 747 | if (!rdev->sb_page) |
ebc24337 | 748 | return -ENOMEM; |
1da177e4 LT |
749 | return 0; |
750 | } | |
751 | ||
545c8795 | 752 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
753 | { |
754 | if (rdev->sb_page) { | |
2d1f3b5d | 755 | put_page(rdev->sb_page); |
1da177e4 LT |
756 | rdev->sb_loaded = 0; |
757 | rdev->sb_page = NULL; | |
0f420358 | 758 | rdev->sb_start = 0; |
dd8ac336 | 759 | rdev->sectors = 0; |
1da177e4 | 760 | } |
2699b672 N |
761 | if (rdev->bb_page) { |
762 | put_page(rdev->bb_page); | |
763 | rdev->bb_page = NULL; | |
764 | } | |
d3b407fb | 765 | badblocks_exit(&rdev->badblocks); |
1da177e4 | 766 | } |
545c8795 | 767 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 768 | |
4246a0b6 | 769 | static void super_written(struct bio *bio) |
7bfa19f2 | 770 | { |
3cb03002 | 771 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 772 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 773 | |
4e4cbee9 CH |
774 | if (bio->bi_status) { |
775 | pr_err("md: super_written gets error=%d\n", bio->bi_status); | |
a9701a30 | 776 | md_error(mddev, rdev); |
46533ff7 N |
777 | if (!test_bit(Faulty, &rdev->flags) |
778 | && (bio->bi_opf & MD_FAILFAST)) { | |
2953079c | 779 | set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags); |
46533ff7 N |
780 | set_bit(LastDev, &rdev->flags); |
781 | } | |
782 | } else | |
783 | clear_bit(LastDev, &rdev->flags); | |
7bfa19f2 | 784 | |
a9701a30 N |
785 | if (atomic_dec_and_test(&mddev->pending_writes)) |
786 | wake_up(&mddev->sb_wait); | |
ed3b98c7 | 787 | rdev_dec_pending(rdev, mddev); |
f8b58edf | 788 | bio_put(bio); |
7bfa19f2 N |
789 | } |
790 | ||
fd01b88c | 791 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
792 | sector_t sector, int size, struct page *page) |
793 | { | |
794 | /* write first size bytes of page to sector of rdev | |
795 | * Increment mddev->pending_writes before returning | |
796 | * and decrement it on completion, waking up sb_wait | |
797 | * if zero is reached. | |
798 | * If an error occurred, call md_error | |
799 | */ | |
46533ff7 N |
800 | struct bio *bio; |
801 | int ff = 0; | |
802 | ||
803 | if (test_bit(Faulty, &rdev->flags)) | |
804 | return; | |
805 | ||
5a85071c | 806 | bio = md_bio_alloc_sync(mddev); |
7bfa19f2 | 807 | |
ed3b98c7 SL |
808 | atomic_inc(&rdev->nr_pending); |
809 | ||
74d46992 | 810 | bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev); |
4f024f37 | 811 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
812 | bio_add_page(bio, page, size, 0); |
813 | bio->bi_private = rdev; | |
814 | bio->bi_end_io = super_written; | |
46533ff7 N |
815 | |
816 | if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) && | |
817 | test_bit(FailFast, &rdev->flags) && | |
818 | !test_bit(LastDev, &rdev->flags)) | |
819 | ff = MD_FAILFAST; | |
5a8948f8 | 820 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff; |
a9701a30 | 821 | |
7bfa19f2 | 822 | atomic_inc(&mddev->pending_writes); |
4e49ea4a | 823 | submit_bio(bio); |
a9701a30 N |
824 | } |
825 | ||
46533ff7 | 826 | int md_super_wait(struct mddev *mddev) |
a9701a30 | 827 | { |
e9c7469b | 828 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 829 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
2953079c | 830 | if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags)) |
46533ff7 N |
831 | return -EAGAIN; |
832 | return 0; | |
7bfa19f2 N |
833 | } |
834 | ||
3cb03002 | 835 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
796a5cf0 | 836 | struct page *page, int op, int op_flags, bool metadata_op) |
1da177e4 | 837 | { |
5a85071c | 838 | struct bio *bio = md_bio_alloc_sync(rdev->mddev); |
1da177e4 LT |
839 | int ret; |
840 | ||
74d46992 CH |
841 | if (metadata_op && rdev->meta_bdev) |
842 | bio_set_dev(bio, rdev->meta_bdev); | |
843 | else | |
844 | bio_set_dev(bio, rdev->bdev); | |
796a5cf0 | 845 | bio_set_op_attrs(bio, op, op_flags); |
ccebd4c4 | 846 | if (metadata_op) |
4f024f37 | 847 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
848 | else if (rdev->mddev->reshape_position != MaxSector && |
849 | (rdev->mddev->reshape_backwards == | |
850 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 851 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 852 | else |
4f024f37 | 853 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 854 | bio_add_page(bio, page, size, 0); |
4e49ea4a MC |
855 | |
856 | submit_bio_wait(bio); | |
1da177e4 | 857 | |
4e4cbee9 | 858 | ret = !bio->bi_status; |
1da177e4 LT |
859 | bio_put(bio); |
860 | return ret; | |
861 | } | |
a8745db2 | 862 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 863 | |
f72ffdd6 | 864 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
865 | { |
866 | char b[BDEVNAME_SIZE]; | |
403df478 | 867 | |
1da177e4 LT |
868 | if (rdev->sb_loaded) |
869 | return 0; | |
870 | ||
796a5cf0 | 871 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) |
1da177e4 LT |
872 | goto fail; |
873 | rdev->sb_loaded = 1; | |
874 | return 0; | |
875 | ||
876 | fail: | |
9d48739e N |
877 | pr_err("md: disabled device %s, could not read superblock.\n", |
878 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
879 | return -EINVAL; |
880 | } | |
881 | ||
e6fd2093 | 882 | static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
1da177e4 | 883 | { |
f72ffdd6 | 884 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
885 | sb1->set_uuid1 == sb2->set_uuid1 && |
886 | sb1->set_uuid2 == sb2->set_uuid2 && | |
887 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
888 | } |
889 | ||
e6fd2093 | 890 | static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
1da177e4 LT |
891 | { |
892 | int ret; | |
893 | mdp_super_t *tmp1, *tmp2; | |
894 | ||
895 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
896 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
897 | ||
898 | if (!tmp1 || !tmp2) { | |
899 | ret = 0; | |
1da177e4 LT |
900 | goto abort; |
901 | } | |
902 | ||
903 | *tmp1 = *sb1; | |
904 | *tmp2 = *sb2; | |
905 | ||
906 | /* | |
907 | * nr_disks is not constant | |
908 | */ | |
909 | tmp1->nr_disks = 0; | |
910 | tmp2->nr_disks = 0; | |
911 | ||
ce0c8e05 | 912 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 913 | abort: |
990a8baf JJ |
914 | kfree(tmp1); |
915 | kfree(tmp2); | |
1da177e4 LT |
916 | return ret; |
917 | } | |
918 | ||
4d167f09 N |
919 | static u32 md_csum_fold(u32 csum) |
920 | { | |
921 | csum = (csum & 0xffff) + (csum >> 16); | |
922 | return (csum & 0xffff) + (csum >> 16); | |
923 | } | |
924 | ||
f72ffdd6 | 925 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 926 | { |
4d167f09 N |
927 | u64 newcsum = 0; |
928 | u32 *sb32 = (u32*)sb; | |
929 | int i; | |
1da177e4 LT |
930 | unsigned int disk_csum, csum; |
931 | ||
932 | disk_csum = sb->sb_csum; | |
933 | sb->sb_csum = 0; | |
4d167f09 N |
934 | |
935 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
936 | newcsum += sb32[i]; | |
937 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
938 | ||
4d167f09 N |
939 | #ifdef CONFIG_ALPHA |
940 | /* This used to use csum_partial, which was wrong for several | |
941 | * reasons including that different results are returned on | |
942 | * different architectures. It isn't critical that we get exactly | |
943 | * the same return value as before (we always csum_fold before | |
944 | * testing, and that removes any differences). However as we | |
945 | * know that csum_partial always returned a 16bit value on | |
946 | * alphas, do a fold to maximise conformity to previous behaviour. | |
947 | */ | |
948 | sb->sb_csum = md_csum_fold(disk_csum); | |
949 | #else | |
1da177e4 | 950 | sb->sb_csum = disk_csum; |
4d167f09 | 951 | #endif |
1da177e4 LT |
952 | return csum; |
953 | } | |
954 | ||
1da177e4 LT |
955 | /* |
956 | * Handle superblock details. | |
957 | * We want to be able to handle multiple superblock formats | |
958 | * so we have a common interface to them all, and an array of | |
959 | * different handlers. | |
960 | * We rely on user-space to write the initial superblock, and support | |
961 | * reading and updating of superblocks. | |
962 | * Interface methods are: | |
3cb03002 | 963 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
964 | * loads and validates a superblock on dev. |
965 | * if refdev != NULL, compare superblocks on both devices | |
966 | * Return: | |
967 | * 0 - dev has a superblock that is compatible with refdev | |
968 | * 1 - dev has a superblock that is compatible and newer than refdev | |
969 | * so dev should be used as the refdev in future | |
970 | * -EINVAL superblock incompatible or invalid | |
971 | * -othererror e.g. -EIO | |
972 | * | |
fd01b88c | 973 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
974 | * Verify that dev is acceptable into mddev. |
975 | * The first time, mddev->raid_disks will be 0, and data from | |
976 | * dev should be merged in. Subsequent calls check that dev | |
977 | * is new enough. Return 0 or -EINVAL | |
978 | * | |
fd01b88c | 979 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
980 | * Update the superblock for rdev with data in mddev |
981 | * This does not write to disc. | |
982 | * | |
983 | */ | |
984 | ||
985 | struct super_type { | |
0cd17fec CW |
986 | char *name; |
987 | struct module *owner; | |
c6563a8c N |
988 | int (*load_super)(struct md_rdev *rdev, |
989 | struct md_rdev *refdev, | |
0cd17fec | 990 | int minor_version); |
c6563a8c N |
991 | int (*validate_super)(struct mddev *mddev, |
992 | struct md_rdev *rdev); | |
993 | void (*sync_super)(struct mddev *mddev, | |
994 | struct md_rdev *rdev); | |
3cb03002 | 995 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 996 | sector_t num_sectors); |
c6563a8c N |
997 | int (*allow_new_offset)(struct md_rdev *rdev, |
998 | unsigned long long new_offset); | |
1da177e4 LT |
999 | }; |
1000 | ||
0894cc30 AN |
1001 | /* |
1002 | * Check that the given mddev has no bitmap. | |
1003 | * | |
1004 | * This function is called from the run method of all personalities that do not | |
1005 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
1006 | * has a bitmap. Otherwise, it returns 0. | |
1007 | * | |
1008 | */ | |
fd01b88c | 1009 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 1010 | { |
c3d9714e | 1011 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 | 1012 | return 0; |
9d48739e | 1013 | pr_warn("%s: bitmaps are not supported for %s\n", |
0894cc30 AN |
1014 | mdname(mddev), mddev->pers->name); |
1015 | return 1; | |
1016 | } | |
1017 | EXPORT_SYMBOL(md_check_no_bitmap); | |
1018 | ||
1da177e4 | 1019 | /* |
f72ffdd6 | 1020 | * load_super for 0.90.0 |
1da177e4 | 1021 | */ |
3cb03002 | 1022 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1023 | { |
1024 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
1025 | mdp_super_t *sb; | |
1026 | int ret; | |
1da177e4 LT |
1027 | |
1028 | /* | |
0f420358 | 1029 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
1030 | * it's at the end of the disk. |
1031 | * | |
1032 | * It also happens to be a multiple of 4Kb. | |
1033 | */ | |
57b2caa3 | 1034 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 1035 | |
0002b271 | 1036 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
9d48739e N |
1037 | if (ret) |
1038 | return ret; | |
1da177e4 LT |
1039 | |
1040 | ret = -EINVAL; | |
1041 | ||
1042 | bdevname(rdev->bdev, b); | |
65a06f06 | 1043 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1044 | |
1045 | if (sb->md_magic != MD_SB_MAGIC) { | |
9d48739e | 1046 | pr_warn("md: invalid raid superblock magic on %s\n", b); |
1da177e4 LT |
1047 | goto abort; |
1048 | } | |
1049 | ||
1050 | if (sb->major_version != 0 || | |
f6705578 N |
1051 | sb->minor_version < 90 || |
1052 | sb->minor_version > 91) { | |
9d48739e N |
1053 | pr_warn("Bad version number %d.%d on %s\n", |
1054 | sb->major_version, sb->minor_version, b); | |
1da177e4 LT |
1055 | goto abort; |
1056 | } | |
1057 | ||
1058 | if (sb->raid_disks <= 0) | |
1059 | goto abort; | |
1060 | ||
4d167f09 | 1061 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
9d48739e | 1062 | pr_warn("md: invalid superblock checksum on %s\n", b); |
1da177e4 LT |
1063 | goto abort; |
1064 | } | |
1065 | ||
1066 | rdev->preferred_minor = sb->md_minor; | |
1067 | rdev->data_offset = 0; | |
c6563a8c | 1068 | rdev->new_data_offset = 0; |
0002b271 | 1069 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1070 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1071 | |
1072 | if (sb->level == LEVEL_MULTIPATH) | |
1073 | rdev->desc_nr = -1; | |
1074 | else | |
1075 | rdev->desc_nr = sb->this_disk.number; | |
1076 | ||
9a7b2b0f | 1077 | if (!refdev) { |
1da177e4 | 1078 | ret = 1; |
9a7b2b0f | 1079 | } else { |
1da177e4 | 1080 | __u64 ev1, ev2; |
65a06f06 | 1081 | mdp_super_t *refsb = page_address(refdev->sb_page); |
e6fd2093 | 1082 | if (!md_uuid_equal(refsb, sb)) { |
9d48739e | 1083 | pr_warn("md: %s has different UUID to %s\n", |
1da177e4 LT |
1084 | b, bdevname(refdev->bdev,b2)); |
1085 | goto abort; | |
1086 | } | |
e6fd2093 | 1087 | if (!md_sb_equal(refsb, sb)) { |
9d48739e N |
1088 | pr_warn("md: %s has same UUID but different superblock to %s\n", |
1089 | b, bdevname(refdev->bdev, b2)); | |
1da177e4 LT |
1090 | goto abort; |
1091 | } | |
1092 | ev1 = md_event(sb); | |
1093 | ev2 = md_event(refsb); | |
1094 | if (ev1 > ev2) | |
1095 | ret = 1; | |
f72ffdd6 | 1096 | else |
1da177e4 LT |
1097 | ret = 0; |
1098 | } | |
8190e754 | 1099 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1100 | /* Limit to 4TB as metadata cannot record more than that. |
1101 | * (not needed for Linear and RAID0 as metadata doesn't | |
1102 | * record this size) | |
1103 | */ | |
3312c951 AB |
1104 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) && |
1105 | sb->level >= 1) | |
1106 | rdev->sectors = (sector_t)(2ULL << 32) - 2; | |
1da177e4 | 1107 | |
27a7b260 | 1108 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1109 | /* "this cannot possibly happen" ... */ |
1110 | ret = -EINVAL; | |
1111 | ||
1da177e4 LT |
1112 | abort: |
1113 | return ret; | |
1114 | } | |
1115 | ||
1116 | /* | |
1117 | * validate_super for 0.90.0 | |
1118 | */ | |
fd01b88c | 1119 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1120 | { |
1121 | mdp_disk_t *desc; | |
65a06f06 | 1122 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1123 | __u64 ev1 = md_event(sb); |
1da177e4 | 1124 | |
41158c7e | 1125 | rdev->raid_disk = -1; |
c5d79adb N |
1126 | clear_bit(Faulty, &rdev->flags); |
1127 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1128 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1129 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1130 | |
1da177e4 LT |
1131 | if (mddev->raid_disks == 0) { |
1132 | mddev->major_version = 0; | |
1133 | mddev->minor_version = sb->minor_version; | |
1134 | mddev->patch_version = sb->patch_version; | |
e691063a | 1135 | mddev->external = 0; |
9d8f0363 | 1136 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1137 | mddev->ctime = sb->ctime; |
1138 | mddev->utime = sb->utime; | |
1139 | mddev->level = sb->level; | |
d9d166c2 | 1140 | mddev->clevel[0] = 0; |
1da177e4 LT |
1141 | mddev->layout = sb->layout; |
1142 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1143 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1144 | mddev->events = ev1; |
c3d9714e | 1145 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1146 | mddev->bitmap_info.space = 0; |
1147 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1148 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1149 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1150 | mddev->reshape_backwards = 0; |
1da177e4 | 1151 | |
f6705578 N |
1152 | if (mddev->minor_version >= 91) { |
1153 | mddev->reshape_position = sb->reshape_position; | |
1154 | mddev->delta_disks = sb->delta_disks; | |
1155 | mddev->new_level = sb->new_level; | |
1156 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1157 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1158 | if (mddev->delta_disks < 0) |
1159 | mddev->reshape_backwards = 1; | |
f6705578 N |
1160 | } else { |
1161 | mddev->reshape_position = MaxSector; | |
1162 | mddev->delta_disks = 0; | |
1163 | mddev->new_level = mddev->level; | |
1164 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1165 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1166 | } |
1167 | ||
1da177e4 LT |
1168 | if (sb->state & (1<<MD_SB_CLEAN)) |
1169 | mddev->recovery_cp = MaxSector; | |
1170 | else { | |
f72ffdd6 | 1171 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1172 | sb->events_lo == sb->cp_events_lo) { |
1173 | mddev->recovery_cp = sb->recovery_cp; | |
1174 | } else | |
1175 | mddev->recovery_cp = 0; | |
1176 | } | |
1177 | ||
1178 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1179 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1180 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1181 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1182 | ||
1183 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1184 | |
1185 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1186 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1187 | mddev->bitmap_info.offset = |
1188 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1189 | mddev->bitmap_info.space = |
c9ad020f | 1190 | mddev->bitmap_info.default_space; |
6409bb05 | 1191 | } |
a654b9d8 | 1192 | |
41158c7e | 1193 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1194 | /* Insist on good event counter while assembling, except |
1195 | * for spares (which don't need an event count) */ | |
1da177e4 | 1196 | ++ev1; |
be6800a7 N |
1197 | if (sb->disks[rdev->desc_nr].state & ( |
1198 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1199 | if (ev1 < mddev->events) |
be6800a7 | 1200 | return -EINVAL; |
41158c7e N |
1201 | } else if (mddev->bitmap) { |
1202 | /* if adding to array with a bitmap, then we can accept an | |
1203 | * older device ... but not too old. | |
1204 | */ | |
41158c7e N |
1205 | if (ev1 < mddev->bitmap->events_cleared) |
1206 | return 0; | |
8313b8e5 N |
1207 | if (ev1 < mddev->events) |
1208 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1209 | } else { |
1210 | if (ev1 < mddev->events) | |
1211 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1212 | return 0; | |
1213 | } | |
41158c7e | 1214 | |
1da177e4 | 1215 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1216 | desc = sb->disks + rdev->desc_nr; |
1217 | ||
1218 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1219 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1220 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1221 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1222 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1223 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1224 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1225 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1226 | /* active but not in sync implies recovery up to | |
1227 | * reshape position. We don't know exactly where | |
1228 | * that is, so set to zero for now */ | |
1229 | if (mddev->minor_version >= 91) { | |
1230 | rdev->recovery_offset = 0; | |
1231 | rdev->raid_disk = desc->raid_disk; | |
1232 | } | |
1da177e4 | 1233 | } |
8ddf9efe N |
1234 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1235 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1236 | if (desc->state & (1<<MD_DISK_FAILFAST)) |
1237 | set_bit(FailFast, &rdev->flags); | |
41158c7e | 1238 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1239 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1240 | return 0; |
1241 | } | |
1242 | ||
1243 | /* | |
1244 | * sync_super for 0.90.0 | |
1245 | */ | |
fd01b88c | 1246 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1247 | { |
1248 | mdp_super_t *sb; | |
3cb03002 | 1249 | struct md_rdev *rdev2; |
1da177e4 | 1250 | int next_spare = mddev->raid_disks; |
19133a42 | 1251 | |
1da177e4 LT |
1252 | /* make rdev->sb match mddev data.. |
1253 | * | |
1254 | * 1/ zero out disks | |
1255 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1256 | * 3/ any empty disks < next_spare become removed | |
1257 | * | |
1258 | * disks[0] gets initialised to REMOVED because | |
1259 | * we cannot be sure from other fields if it has | |
1260 | * been initialised or not. | |
1261 | */ | |
1262 | int i; | |
1263 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1264 | ||
61181565 N |
1265 | rdev->sb_size = MD_SB_BYTES; |
1266 | ||
65a06f06 | 1267 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1268 | |
1269 | memset(sb, 0, sizeof(*sb)); | |
1270 | ||
1271 | sb->md_magic = MD_SB_MAGIC; | |
1272 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1273 | sb->patch_version = mddev->patch_version; |
1274 | sb->gvalid_words = 0; /* ignored */ | |
1275 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1276 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1277 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1278 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1279 | ||
9ebc6ef1 | 1280 | sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 1281 | sb->level = mddev->level; |
58c0fed4 | 1282 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1283 | sb->raid_disks = mddev->raid_disks; |
1284 | sb->md_minor = mddev->md_minor; | |
e691063a | 1285 | sb->not_persistent = 0; |
9ebc6ef1 | 1286 | sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
1287 | sb->state = 0; |
1288 | sb->events_hi = (mddev->events>>32); | |
1289 | sb->events_lo = (u32)mddev->events; | |
1290 | ||
f6705578 N |
1291 | if (mddev->reshape_position == MaxSector) |
1292 | sb->minor_version = 90; | |
1293 | else { | |
1294 | sb->minor_version = 91; | |
1295 | sb->reshape_position = mddev->reshape_position; | |
1296 | sb->new_level = mddev->new_level; | |
1297 | sb->delta_disks = mddev->delta_disks; | |
1298 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1299 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1300 | } |
1301 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1302 | if (mddev->in_sync) |
1303 | { | |
1304 | sb->recovery_cp = mddev->recovery_cp; | |
1305 | sb->cp_events_hi = (mddev->events>>32); | |
1306 | sb->cp_events_lo = (u32)mddev->events; | |
1307 | if (mddev->recovery_cp == MaxSector) | |
1308 | sb->state = (1<< MD_SB_CLEAN); | |
1309 | } else | |
1310 | sb->recovery_cp = 0; | |
1311 | ||
1312 | sb->layout = mddev->layout; | |
9d8f0363 | 1313 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1314 | |
c3d9714e | 1315 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1316 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1317 | ||
1da177e4 | 1318 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1319 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1320 | mdp_disk_t *d; |
86e6ffdd | 1321 | int desc_nr; |
0261cd9f N |
1322 | int is_active = test_bit(In_sync, &rdev2->flags); |
1323 | ||
1324 | if (rdev2->raid_disk >= 0 && | |
1325 | sb->minor_version >= 91) | |
1326 | /* we have nowhere to store the recovery_offset, | |
1327 | * but if it is not below the reshape_position, | |
1328 | * we can piggy-back on that. | |
1329 | */ | |
1330 | is_active = 1; | |
1331 | if (rdev2->raid_disk < 0 || | |
1332 | test_bit(Faulty, &rdev2->flags)) | |
1333 | is_active = 0; | |
1334 | if (is_active) | |
86e6ffdd | 1335 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1336 | else |
86e6ffdd | 1337 | desc_nr = next_spare++; |
19133a42 | 1338 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1339 | d = &sb->disks[rdev2->desc_nr]; |
1340 | nr_disks++; | |
1341 | d->number = rdev2->desc_nr; | |
1342 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1343 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1344 | if (is_active) |
1da177e4 LT |
1345 | d->raid_disk = rdev2->raid_disk; |
1346 | else | |
1347 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1348 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1349 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1350 | else if (is_active) { |
1da177e4 | 1351 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1352 | if (test_bit(In_sync, &rdev2->flags)) |
1353 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1354 | active++; |
1355 | working++; | |
1356 | } else { | |
1357 | d->state = 0; | |
1358 | spare++; | |
1359 | working++; | |
1360 | } | |
8ddf9efe N |
1361 | if (test_bit(WriteMostly, &rdev2->flags)) |
1362 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
1363 | if (test_bit(FailFast, &rdev2->flags)) |
1364 | d->state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 | 1365 | } |
1da177e4 LT |
1366 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1367 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1368 | mdp_disk_t *d = &sb->disks[i]; | |
1369 | if (d->state == 0 && d->number == 0) { | |
1370 | d->number = i; | |
1371 | d->raid_disk = i; | |
1372 | d->state = (1<<MD_DISK_REMOVED); | |
1373 | d->state |= (1<<MD_DISK_FAULTY); | |
1374 | failed++; | |
1375 | } | |
1376 | } | |
1377 | sb->nr_disks = nr_disks; | |
1378 | sb->active_disks = active; | |
1379 | sb->working_disks = working; | |
1380 | sb->failed_disks = failed; | |
1381 | sb->spare_disks = spare; | |
1382 | ||
1383 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1384 | sb->sb_csum = calc_sb_csum(sb); | |
1385 | } | |
1386 | ||
0cd17fec CW |
1387 | /* |
1388 | * rdev_size_change for 0.90.0 | |
1389 | */ | |
1390 | static unsigned long long | |
3cb03002 | 1391 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1392 | { |
58c0fed4 | 1393 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1394 | return 0; /* component must fit device */ |
c3d9714e | 1395 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1396 | return 0; /* can't move bitmap */ |
57b2caa3 | 1397 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1398 | if (!num_sectors || num_sectors > rdev->sb_start) |
1399 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1400 | /* Limit to 4TB as metadata cannot record more than that. |
1401 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1402 | */ | |
3312c951 AB |
1403 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) && |
1404 | rdev->mddev->level >= 1) | |
1405 | num_sectors = (sector_t)(2ULL << 32) - 2; | |
46533ff7 N |
1406 | do { |
1407 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
0cd17fec | 1408 | rdev->sb_page); |
46533ff7 | 1409 | } while (md_super_wait(rdev->mddev) < 0); |
c26a44ed | 1410 | return num_sectors; |
0cd17fec CW |
1411 | } |
1412 | ||
c6563a8c N |
1413 | static int |
1414 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1415 | { | |
1416 | /* non-zero offset changes not possible with v0.90 */ | |
1417 | return new_offset == 0; | |
1418 | } | |
0cd17fec | 1419 | |
1da177e4 LT |
1420 | /* |
1421 | * version 1 superblock | |
1422 | */ | |
1423 | ||
f72ffdd6 | 1424 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1425 | { |
1c05b4bc N |
1426 | __le32 disk_csum; |
1427 | u32 csum; | |
1da177e4 LT |
1428 | unsigned long long newcsum; |
1429 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1430 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1431 | |
1432 | disk_csum = sb->sb_csum; | |
1433 | sb->sb_csum = 0; | |
1434 | newcsum = 0; | |
1f3c9907 | 1435 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1436 | newcsum += le32_to_cpu(*isuper++); |
1437 | ||
1438 | if (size == 2) | |
1c05b4bc | 1439 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1440 | |
1441 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1442 | sb->sb_csum = disk_csum; | |
1443 | return cpu_to_le32(csum); | |
1444 | } | |
1445 | ||
3cb03002 | 1446 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1447 | { |
1448 | struct mdp_superblock_1 *sb; | |
1449 | int ret; | |
0f420358 | 1450 | sector_t sb_start; |
c6563a8c | 1451 | sector_t sectors; |
1da177e4 | 1452 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1453 | int bmask; |
1da177e4 LT |
1454 | |
1455 | /* | |
0f420358 | 1456 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1457 | * It is always aligned to a 4K boundary and |
1458 | * depeding on minor_version, it can be: | |
1459 | * 0: At least 8K, but less than 12K, from end of device | |
1460 | * 1: At start of device | |
1461 | * 2: 4K from start of device. | |
1462 | */ | |
1463 | switch(minor_version) { | |
1464 | case 0: | |
77304d2a | 1465 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1466 | sb_start -= 8*2; |
1467 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1468 | break; |
1469 | case 1: | |
0f420358 | 1470 | sb_start = 0; |
1da177e4 LT |
1471 | break; |
1472 | case 2: | |
0f420358 | 1473 | sb_start = 8; |
1da177e4 LT |
1474 | break; |
1475 | default: | |
1476 | return -EINVAL; | |
1477 | } | |
0f420358 | 1478 | rdev->sb_start = sb_start; |
1da177e4 | 1479 | |
0002b271 N |
1480 | /* superblock is rarely larger than 1K, but it can be larger, |
1481 | * and it is safe to read 4k, so we do that | |
1482 | */ | |
1483 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1484 | if (ret) return ret; |
1485 | ||
65a06f06 | 1486 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1487 | |
1488 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1489 | sb->major_version != cpu_to_le32(1) || | |
1490 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1491 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1492 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1493 | return -EINVAL; |
1494 | ||
1495 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
9d48739e | 1496 | pr_warn("md: invalid superblock checksum on %s\n", |
1da177e4 LT |
1497 | bdevname(rdev->bdev,b)); |
1498 | return -EINVAL; | |
1499 | } | |
1500 | if (le64_to_cpu(sb->data_size) < 10) { | |
9d48739e N |
1501 | pr_warn("md: data_size too small on %s\n", |
1502 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1503 | return -EINVAL; |
1504 | } | |
c6563a8c N |
1505 | if (sb->pad0 || |
1506 | sb->pad3[0] || | |
1507 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1508 | /* Some padding is non-zero, might be a new feature */ | |
1509 | return -EINVAL; | |
e11e93fa | 1510 | |
1da177e4 LT |
1511 | rdev->preferred_minor = 0xffff; |
1512 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1513 | rdev->new_data_offset = rdev->data_offset; |
1514 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1515 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1516 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1517 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1518 | |
0002b271 | 1519 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1520 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1521 | if (rdev->sb_size & bmask) |
a1801f85 N |
1522 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1523 | ||
1524 | if (minor_version | |
0f420358 | 1525 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1526 | return -EINVAL; |
c6563a8c N |
1527 | if (minor_version |
1528 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1529 | return -EINVAL; | |
0002b271 | 1530 | |
31b65a0d N |
1531 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1532 | rdev->desc_nr = -1; | |
1533 | else | |
1534 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1535 | ||
2699b672 N |
1536 | if (!rdev->bb_page) { |
1537 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1538 | if (!rdev->bb_page) | |
1539 | return -ENOMEM; | |
1540 | } | |
1541 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1542 | rdev->badblocks.count == 0) { | |
1543 | /* need to load the bad block list. | |
1544 | * Currently we limit it to one page. | |
1545 | */ | |
1546 | s32 offset; | |
1547 | sector_t bb_sector; | |
1548 | u64 *bbp; | |
1549 | int i; | |
1550 | int sectors = le16_to_cpu(sb->bblog_size); | |
1551 | if (sectors > (PAGE_SIZE / 512)) | |
1552 | return -EINVAL; | |
1553 | offset = le32_to_cpu(sb->bblog_offset); | |
1554 | if (offset == 0) | |
1555 | return -EINVAL; | |
1556 | bb_sector = (long long)offset; | |
1557 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
796a5cf0 | 1558 | rdev->bb_page, REQ_OP_READ, 0, true)) |
2699b672 N |
1559 | return -EIO; |
1560 | bbp = (u64 *)page_address(rdev->bb_page); | |
1561 | rdev->badblocks.shift = sb->bblog_shift; | |
1562 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1563 | u64 bb = le64_to_cpu(*bbp); | |
1564 | int count = bb & (0x3ff); | |
1565 | u64 sector = bb >> 10; | |
1566 | sector <<= sb->bblog_shift; | |
1567 | count <<= sb->bblog_shift; | |
1568 | if (bb + 1 == 0) | |
1569 | break; | |
fc974ee2 | 1570 | if (badblocks_set(&rdev->badblocks, sector, count, 1)) |
2699b672 N |
1571 | return -EINVAL; |
1572 | } | |
486adf72 N |
1573 | } else if (sb->bblog_offset != 0) |
1574 | rdev->badblocks.shift = 0; | |
2699b672 | 1575 | |
ddc08823 PB |
1576 | if ((le32_to_cpu(sb->feature_map) & |
1577 | (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) { | |
ea0213e0 AP |
1578 | rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset); |
1579 | rdev->ppl.size = le16_to_cpu(sb->ppl.size); | |
1580 | rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset; | |
1581 | } | |
1582 | ||
9a7b2b0f | 1583 | if (!refdev) { |
8ed75463 | 1584 | ret = 1; |
9a7b2b0f | 1585 | } else { |
1da177e4 | 1586 | __u64 ev1, ev2; |
65a06f06 | 1587 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1588 | |
1589 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1590 | sb->level != refsb->level || | |
1591 | sb->layout != refsb->layout || | |
1592 | sb->chunksize != refsb->chunksize) { | |
9d48739e | 1593 | pr_warn("md: %s has strangely different superblock to %s\n", |
1da177e4 LT |
1594 | bdevname(rdev->bdev,b), |
1595 | bdevname(refdev->bdev,b2)); | |
1596 | return -EINVAL; | |
1597 | } | |
1598 | ev1 = le64_to_cpu(sb->events); | |
1599 | ev2 = le64_to_cpu(refsb->events); | |
1600 | ||
1601 | if (ev1 > ev2) | |
8ed75463 N |
1602 | ret = 1; |
1603 | else | |
1604 | ret = 0; | |
1da177e4 | 1605 | } |
c6563a8c N |
1606 | if (minor_version) { |
1607 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1608 | sectors -= rdev->data_offset; | |
1609 | } else | |
1610 | sectors = rdev->sb_start; | |
1611 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1612 | return -EINVAL; |
dd8ac336 | 1613 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1614 | return ret; |
1da177e4 LT |
1615 | } |
1616 | ||
fd01b88c | 1617 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1618 | { |
65a06f06 | 1619 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1620 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1621 | |
41158c7e | 1622 | rdev->raid_disk = -1; |
c5d79adb N |
1623 | clear_bit(Faulty, &rdev->flags); |
1624 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1625 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1626 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1627 | |
1da177e4 LT |
1628 | if (mddev->raid_disks == 0) { |
1629 | mddev->major_version = 1; | |
1630 | mddev->patch_version = 0; | |
e691063a | 1631 | mddev->external = 0; |
9d8f0363 | 1632 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
9ebc6ef1 DD |
1633 | mddev->ctime = le64_to_cpu(sb->ctime); |
1634 | mddev->utime = le64_to_cpu(sb->utime); | |
1da177e4 | 1635 | mddev->level = le32_to_cpu(sb->level); |
d9d166c2 | 1636 | mddev->clevel[0] = 0; |
1da177e4 LT |
1637 | mddev->layout = le32_to_cpu(sb->layout); |
1638 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1639 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1640 | mddev->events = ev1; |
c3d9714e | 1641 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1642 | mddev->bitmap_info.space = 0; |
1643 | /* Default location for bitmap is 1K after superblock | |
1644 | * using 3K - total of 4K | |
1645 | */ | |
c3d9714e | 1646 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1647 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1648 | mddev->reshape_backwards = 0; |
1649 | ||
1da177e4 LT |
1650 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1651 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1652 | ||
1653 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1654 | |
71c0805c | 1655 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1656 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1657 | mddev->bitmap_info.offset = |
1658 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1659 | /* Metadata doesn't record how much space is available. |
1660 | * For 1.0, we assume we can use up to the superblock | |
1661 | * if before, else to 4K beyond superblock. | |
1662 | * For others, assume no change is possible. | |
1663 | */ | |
1664 | if (mddev->minor_version > 0) | |
1665 | mddev->bitmap_info.space = 0; | |
1666 | else if (mddev->bitmap_info.offset > 0) | |
1667 | mddev->bitmap_info.space = | |
1668 | 8 - mddev->bitmap_info.offset; | |
1669 | else | |
1670 | mddev->bitmap_info.space = | |
1671 | -mddev->bitmap_info.offset; | |
1672 | } | |
e11e93fa | 1673 | |
f6705578 N |
1674 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1675 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1676 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1677 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1678 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1679 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1680 | if (mddev->delta_disks < 0 || |
1681 | (mddev->delta_disks == 0 && | |
1682 | (le32_to_cpu(sb->feature_map) | |
1683 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1684 | mddev->reshape_backwards = 1; | |
f6705578 N |
1685 | } else { |
1686 | mddev->reshape_position = MaxSector; | |
1687 | mddev->delta_disks = 0; | |
1688 | mddev->new_level = mddev->level; | |
1689 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1690 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1691 | } |
1692 | ||
486b0f7b | 1693 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) |
a62ab49e | 1694 | set_bit(MD_HAS_JOURNAL, &mddev->flags); |
ea0213e0 | 1695 | |
ddc08823 PB |
1696 | if (le32_to_cpu(sb->feature_map) & |
1697 | (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) { | |
ea0213e0 AP |
1698 | if (le32_to_cpu(sb->feature_map) & |
1699 | (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL)) | |
1700 | return -EINVAL; | |
ddc08823 PB |
1701 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) && |
1702 | (le32_to_cpu(sb->feature_map) & | |
1703 | MD_FEATURE_MULTIPLE_PPLS)) | |
1704 | return -EINVAL; | |
ea0213e0 AP |
1705 | set_bit(MD_HAS_PPL, &mddev->flags); |
1706 | } | |
41158c7e | 1707 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1708 | /* Insist of good event counter while assembling, except for |
1709 | * spares (which don't need an event count) */ | |
1da177e4 | 1710 | ++ev1; |
be6800a7 N |
1711 | if (rdev->desc_nr >= 0 && |
1712 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
a3dfbdaa SL |
1713 | (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX || |
1714 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)) | |
be6800a7 N |
1715 | if (ev1 < mddev->events) |
1716 | return -EINVAL; | |
41158c7e N |
1717 | } else if (mddev->bitmap) { |
1718 | /* If adding to array with a bitmap, then we can accept an | |
1719 | * older device, but not too old. | |
1720 | */ | |
41158c7e N |
1721 | if (ev1 < mddev->bitmap->events_cleared) |
1722 | return 0; | |
8313b8e5 N |
1723 | if (ev1 < mddev->events) |
1724 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1725 | } else { |
1726 | if (ev1 < mddev->events) | |
1727 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1728 | return 0; | |
1729 | } | |
1da177e4 LT |
1730 | if (mddev->level != LEVEL_MULTIPATH) { |
1731 | int role; | |
3673f305 N |
1732 | if (rdev->desc_nr < 0 || |
1733 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
c4d4c91b | 1734 | role = MD_DISK_ROLE_SPARE; |
3673f305 N |
1735 | rdev->desc_nr = -1; |
1736 | } else | |
1737 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 | 1738 | switch(role) { |
c4d4c91b | 1739 | case MD_DISK_ROLE_SPARE: /* spare */ |
1da177e4 | 1740 | break; |
c4d4c91b | 1741 | case MD_DISK_ROLE_FAULTY: /* faulty */ |
b2d444d7 | 1742 | set_bit(Faulty, &rdev->flags); |
1da177e4 | 1743 | break; |
bac624f3 SL |
1744 | case MD_DISK_ROLE_JOURNAL: /* journal device */ |
1745 | if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) { | |
1746 | /* journal device without journal feature */ | |
9d48739e | 1747 | pr_warn("md: journal device provided without journal feature, ignoring the device\n"); |
bac624f3 SL |
1748 | return -EINVAL; |
1749 | } | |
1750 | set_bit(Journal, &rdev->flags); | |
3069aa8d | 1751 | rdev->journal_tail = le64_to_cpu(sb->journal_tail); |
9b15603d | 1752 | rdev->raid_disk = 0; |
bac624f3 | 1753 | break; |
1da177e4 | 1754 | default: |
f466722c | 1755 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1756 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1757 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1758 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1759 | if (!(le32_to_cpu(sb->feature_map) & |
1760 | MD_FEATURE_RECOVERY_BITMAP)) | |
1761 | rdev->saved_raid_disk = -1; | |
1762 | } else | |
5fd6c1dc | 1763 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1764 | rdev->raid_disk = role; |
1765 | break; | |
1766 | } | |
8ddf9efe N |
1767 | if (sb->devflags & WriteMostly1) |
1768 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1769 | if (sb->devflags & FailFast1) |
1770 | set_bit(FailFast, &rdev->flags); | |
2d78f8c4 N |
1771 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1772 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1773 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1774 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1775 | |
1da177e4 LT |
1776 | return 0; |
1777 | } | |
1778 | ||
fd01b88c | 1779 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1780 | { |
1781 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1782 | struct md_rdev *rdev2; |
1da177e4 LT |
1783 | int max_dev, i; |
1784 | /* make rdev->sb match mddev and rdev data. */ | |
1785 | ||
65a06f06 | 1786 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1787 | |
1788 | sb->feature_map = 0; | |
1789 | sb->pad0 = 0; | |
5fd6c1dc | 1790 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1791 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1792 | ||
1793 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1794 | sb->events = cpu_to_le64(mddev->events); | |
1795 | if (mddev->in_sync) | |
1796 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
bd18f646 SL |
1797 | else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags)) |
1798 | sb->resync_offset = cpu_to_le64(MaxSector); | |
1da177e4 LT |
1799 | else |
1800 | sb->resync_offset = cpu_to_le64(0); | |
1801 | ||
1c05b4bc | 1802 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1803 | |
f0ca340c | 1804 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1805 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1806 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1807 | sb->level = cpu_to_le32(mddev->level); |
1808 | sb->layout = cpu_to_le32(mddev->layout); | |
688834e6 N |
1809 | if (test_bit(FailFast, &rdev->flags)) |
1810 | sb->devflags |= FailFast1; | |
1811 | else | |
1812 | sb->devflags &= ~FailFast1; | |
f0ca340c | 1813 | |
aeb9b211 N |
1814 | if (test_bit(WriteMostly, &rdev->flags)) |
1815 | sb->devflags |= WriteMostly1; | |
1816 | else | |
1817 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1818 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1819 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1820 | |
c3d9714e N |
1821 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1822 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1823 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1824 | } |
5fd6c1dc | 1825 | |
f2076e7d | 1826 | if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) && |
97e4f42d | 1827 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1828 | sb->feature_map |= |
1829 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1830 | sb->recovery_offset = | |
1831 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1832 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1833 | sb->feature_map |= | |
1834 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1835 | } |
3069aa8d SL |
1836 | /* Note: recovery_offset and journal_tail share space */ |
1837 | if (test_bit(Journal, &rdev->flags)) | |
1838 | sb->journal_tail = cpu_to_le64(rdev->journal_tail); | |
2d78f8c4 N |
1839 | if (test_bit(Replacement, &rdev->flags)) |
1840 | sb->feature_map |= | |
1841 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1842 | |
f6705578 N |
1843 | if (mddev->reshape_position != MaxSector) { |
1844 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1845 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1846 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1847 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1848 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1849 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1850 | if (mddev->delta_disks == 0 && |
1851 | mddev->reshape_backwards) | |
1852 | sb->feature_map | |
1853 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1854 | if (rdev->new_data_offset != rdev->data_offset) { |
1855 | sb->feature_map | |
1856 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1857 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1858 | - rdev->data_offset)); | |
1859 | } | |
f6705578 | 1860 | } |
a654b9d8 | 1861 | |
3c462c88 GR |
1862 | if (mddev_is_clustered(mddev)) |
1863 | sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED); | |
1864 | ||
2699b672 N |
1865 | if (rdev->badblocks.count == 0) |
1866 | /* Nothing to do for bad blocks*/ ; | |
1867 | else if (sb->bblog_offset == 0) | |
1868 | /* Cannot record bad blocks on this device */ | |
1869 | md_error(mddev, rdev); | |
1870 | else { | |
1871 | struct badblocks *bb = &rdev->badblocks; | |
1872 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1873 | u64 *p = bb->page; | |
1874 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1875 | if (bb->changed) { | |
1876 | unsigned seq; | |
1877 | ||
1878 | retry: | |
1879 | seq = read_seqbegin(&bb->lock); | |
1880 | ||
1881 | memset(bbp, 0xff, PAGE_SIZE); | |
1882 | ||
1883 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1884 | u64 internal_bb = p[i]; |
2699b672 N |
1885 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1886 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1887 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1888 | } |
d0962936 | 1889 | bb->changed = 0; |
2699b672 N |
1890 | if (read_seqretry(&bb->lock, seq)) |
1891 | goto retry; | |
1892 | ||
1893 | bb->sector = (rdev->sb_start + | |
1894 | (int)le32_to_cpu(sb->bblog_offset)); | |
1895 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1896 | } |
1897 | } | |
1898 | ||
1da177e4 | 1899 | max_dev = 0; |
dafb20fa | 1900 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1901 | if (rdev2->desc_nr+1 > max_dev) |
1902 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1903 | |
70471daf N |
1904 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1905 | int bmask; | |
a778b73f | 1906 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1907 | rdev->sb_size = max_dev * 2 + 256; |
1908 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1909 | if (rdev->sb_size & bmask) | |
1910 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1911 | } else |
1912 | max_dev = le32_to_cpu(sb->max_dev); | |
1913 | ||
1da177e4 | 1914 | for (i=0; i<max_dev;i++) |
8df72024 | 1915 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE); |
f72ffdd6 | 1916 | |
a97b7896 SL |
1917 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) |
1918 | sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL); | |
f72ffdd6 | 1919 | |
ea0213e0 | 1920 | if (test_bit(MD_HAS_PPL, &mddev->flags)) { |
ddc08823 PB |
1921 | if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags)) |
1922 | sb->feature_map |= | |
1923 | cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS); | |
1924 | else | |
1925 | sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL); | |
ea0213e0 AP |
1926 | sb->ppl.offset = cpu_to_le16(rdev->ppl.offset); |
1927 | sb->ppl.size = cpu_to_le16(rdev->ppl.size); | |
1928 | } | |
1929 | ||
dafb20fa | 1930 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1931 | i = rdev2->desc_nr; |
b2d444d7 | 1932 | if (test_bit(Faulty, &rdev2->flags)) |
c4d4c91b | 1933 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY); |
b2d444d7 | 1934 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1935 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
a97b7896 | 1936 | else if (test_bit(Journal, &rdev2->flags)) |
bac624f3 | 1937 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL); |
93be75ff | 1938 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1939 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 | 1940 | else |
c4d4c91b | 1941 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE); |
1da177e4 LT |
1942 | } |
1943 | ||
1da177e4 LT |
1944 | sb->sb_csum = calc_sb_1_csum(sb); |
1945 | } | |
1946 | ||
0cd17fec | 1947 | static unsigned long long |
3cb03002 | 1948 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1949 | { |
1950 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1951 | sector_t max_sectors; |
58c0fed4 | 1952 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1953 | return 0; /* component must fit device */ |
c6563a8c N |
1954 | if (rdev->data_offset != rdev->new_data_offset) |
1955 | return 0; /* too confusing */ | |
0f420358 | 1956 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1957 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1958 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1959 | max_sectors -= rdev->data_offset; |
1960 | if (!num_sectors || num_sectors > max_sectors) | |
1961 | num_sectors = max_sectors; | |
c3d9714e | 1962 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1963 | /* minor version 0 with bitmap we can't move */ |
1964 | return 0; | |
1965 | } else { | |
1966 | /* minor version 0; superblock after data */ | |
0f420358 | 1967 | sector_t sb_start; |
77304d2a | 1968 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1969 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1970 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1971 | if (!num_sectors || num_sectors > max_sectors) |
1972 | num_sectors = max_sectors; | |
0f420358 | 1973 | rdev->sb_start = sb_start; |
0cd17fec | 1974 | } |
65a06f06 | 1975 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1976 | sb->data_size = cpu_to_le64(num_sectors); |
3fb632e4 | 1977 | sb->super_offset = cpu_to_le64(rdev->sb_start); |
0cd17fec | 1978 | sb->sb_csum = calc_sb_1_csum(sb); |
46533ff7 N |
1979 | do { |
1980 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
1981 | rdev->sb_page); | |
1982 | } while (md_super_wait(rdev->mddev) < 0); | |
c26a44ed | 1983 | return num_sectors; |
c6563a8c N |
1984 | |
1985 | } | |
1986 | ||
1987 | static int | |
1988 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1989 | unsigned long long new_offset) | |
1990 | { | |
1991 | /* All necessary checks on new >= old have been done */ | |
1992 | struct bitmap *bitmap; | |
1993 | if (new_offset >= rdev->data_offset) | |
1994 | return 1; | |
1995 | ||
1996 | /* with 1.0 metadata, there is no metadata to tread on | |
1997 | * so we can always move back */ | |
1998 | if (rdev->mddev->minor_version == 0) | |
1999 | return 1; | |
2000 | ||
2001 | /* otherwise we must be sure not to step on | |
2002 | * any metadata, so stay: | |
2003 | * 36K beyond start of superblock | |
2004 | * beyond end of badblocks | |
2005 | * beyond write-intent bitmap | |
2006 | */ | |
2007 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
2008 | return 0; | |
2009 | bitmap = rdev->mddev->bitmap; | |
2010 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
2011 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 2012 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
2013 | return 0; |
2014 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
2015 | return 0; | |
2016 | ||
2017 | return 1; | |
0cd17fec | 2018 | } |
1da177e4 | 2019 | |
75c96f85 | 2020 | static struct super_type super_types[] = { |
1da177e4 LT |
2021 | [0] = { |
2022 | .name = "0.90.0", | |
2023 | .owner = THIS_MODULE, | |
0cd17fec CW |
2024 | .load_super = super_90_load, |
2025 | .validate_super = super_90_validate, | |
2026 | .sync_super = super_90_sync, | |
2027 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 2028 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
2029 | }, |
2030 | [1] = { | |
2031 | .name = "md-1", | |
2032 | .owner = THIS_MODULE, | |
0cd17fec CW |
2033 | .load_super = super_1_load, |
2034 | .validate_super = super_1_validate, | |
2035 | .sync_super = super_1_sync, | |
2036 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 2037 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
2038 | }, |
2039 | }; | |
1da177e4 | 2040 | |
fd01b88c | 2041 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
2042 | { |
2043 | if (mddev->sync_super) { | |
2044 | mddev->sync_super(mddev, rdev); | |
2045 | return; | |
2046 | } | |
2047 | ||
2048 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
2049 | ||
2050 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
2051 | } | |
2052 | ||
fd01b88c | 2053 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 2054 | { |
3cb03002 | 2055 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 2056 | |
4b80991c | 2057 | rcu_read_lock(); |
0b020e85 SL |
2058 | rdev_for_each_rcu(rdev, mddev1) { |
2059 | if (test_bit(Faulty, &rdev->flags) || | |
2060 | test_bit(Journal, &rdev->flags) || | |
2061 | rdev->raid_disk == -1) | |
2062 | continue; | |
2063 | rdev_for_each_rcu(rdev2, mddev2) { | |
2064 | if (test_bit(Faulty, &rdev2->flags) || | |
2065 | test_bit(Journal, &rdev2->flags) || | |
2066 | rdev2->raid_disk == -1) | |
2067 | continue; | |
7dd5e7c3 | 2068 | if (rdev->bdev->bd_contains == |
4b80991c N |
2069 | rdev2->bdev->bd_contains) { |
2070 | rcu_read_unlock(); | |
7dd5e7c3 | 2071 | return 1; |
4b80991c | 2072 | } |
0b020e85 SL |
2073 | } |
2074 | } | |
4b80991c | 2075 | rcu_read_unlock(); |
1da177e4 LT |
2076 | return 0; |
2077 | } | |
2078 | ||
2079 | static LIST_HEAD(pending_raid_disks); | |
2080 | ||
ac5e7113 AN |
2081 | /* |
2082 | * Try to register data integrity profile for an mddev | |
2083 | * | |
2084 | * This is called when an array is started and after a disk has been kicked | |
2085 | * from the array. It only succeeds if all working and active component devices | |
2086 | * are integrity capable with matching profiles. | |
2087 | */ | |
fd01b88c | 2088 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 2089 | { |
3cb03002 | 2090 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
2091 | |
2092 | if (list_empty(&mddev->disks)) | |
2093 | return 0; /* nothing to do */ | |
629acb6a JB |
2094 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
2095 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 2096 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
2097 | /* skip spares and non-functional disks */ |
2098 | if (test_bit(Faulty, &rdev->flags)) | |
2099 | continue; | |
2100 | if (rdev->raid_disk < 0) | |
2101 | continue; | |
ac5e7113 AN |
2102 | if (!reference) { |
2103 | /* Use the first rdev as the reference */ | |
2104 | reference = rdev; | |
2105 | continue; | |
2106 | } | |
2107 | /* does this rdev's profile match the reference profile? */ | |
2108 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
2109 | rdev->bdev->bd_disk) < 0) | |
2110 | return -EINVAL; | |
2111 | } | |
89078d57 MP |
2112 | if (!reference || !bdev_get_integrity(reference->bdev)) |
2113 | return 0; | |
ac5e7113 AN |
2114 | /* |
2115 | * All component devices are integrity capable and have matching | |
2116 | * profiles, register the common profile for the md device. | |
2117 | */ | |
25520d55 MP |
2118 | blk_integrity_register(mddev->gendisk, |
2119 | bdev_get_integrity(reference->bdev)); | |
2120 | ||
9d48739e | 2121 | pr_debug("md: data integrity enabled on %s\n", mdname(mddev)); |
a91a2785 | 2122 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { |
9d48739e | 2123 | pr_err("md: failed to create integrity pool for %s\n", |
a91a2785 MP |
2124 | mdname(mddev)); |
2125 | return -EINVAL; | |
2126 | } | |
ac5e7113 AN |
2127 | return 0; |
2128 | } | |
2129 | EXPORT_SYMBOL(md_integrity_register); | |
2130 | ||
1501efad DW |
2131 | /* |
2132 | * Attempt to add an rdev, but only if it is consistent with the current | |
2133 | * integrity profile | |
2134 | */ | |
2135 | int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) | |
3f9d99c1 | 2136 | { |
2863b9eb JB |
2137 | struct blk_integrity *bi_rdev; |
2138 | struct blk_integrity *bi_mddev; | |
1501efad | 2139 | char name[BDEVNAME_SIZE]; |
2863b9eb JB |
2140 | |
2141 | if (!mddev->gendisk) | |
1501efad | 2142 | return 0; |
2863b9eb JB |
2143 | |
2144 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2145 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2146 | |
ac5e7113 | 2147 | if (!bi_mddev) /* nothing to do */ |
1501efad DW |
2148 | return 0; |
2149 | ||
2150 | if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) { | |
9d48739e N |
2151 | pr_err("%s: incompatible integrity profile for %s\n", |
2152 | mdname(mddev), bdevname(rdev->bdev, name)); | |
1501efad DW |
2153 | return -ENXIO; |
2154 | } | |
2155 | ||
2156 | return 0; | |
3f9d99c1 | 2157 | } |
ac5e7113 | 2158 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2159 | |
f72ffdd6 | 2160 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2161 | { |
7dd5e7c3 | 2162 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2163 | struct kobject *ko; |
5e55e2f5 | 2164 | int err; |
1da177e4 | 2165 | |
11e2ede0 DW |
2166 | /* prevent duplicates */ |
2167 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2168 | return -EEXIST; | |
2169 | ||
97b20ef7 N |
2170 | if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) && |
2171 | mddev->pers) | |
2172 | return -EROFS; | |
2173 | ||
dd8ac336 | 2174 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
f6b6ec5c SL |
2175 | if (!test_bit(Journal, &rdev->flags) && |
2176 | rdev->sectors && | |
2177 | (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2178 | if (mddev->pers) { |
2179 | /* Cannot change size, so fail | |
2180 | * If mddev->level <= 0, then we don't care | |
2181 | * about aligning sizes (e.g. linear) | |
2182 | */ | |
2183 | if (mddev->level > 0) | |
2184 | return -ENOSPC; | |
2185 | } else | |
dd8ac336 | 2186 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2187 | } |
1da177e4 LT |
2188 | |
2189 | /* Verify rdev->desc_nr is unique. | |
2190 | * If it is -1, assign a free number, else | |
2191 | * check number is not in use | |
2192 | */ | |
4878e9eb | 2193 | rcu_read_lock(); |
1da177e4 LT |
2194 | if (rdev->desc_nr < 0) { |
2195 | int choice = 0; | |
4878e9eb N |
2196 | if (mddev->pers) |
2197 | choice = mddev->raid_disks; | |
57d051dc | 2198 | while (md_find_rdev_nr_rcu(mddev, choice)) |
1da177e4 LT |
2199 | choice++; |
2200 | rdev->desc_nr = choice; | |
2201 | } else { | |
57d051dc | 2202 | if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
4878e9eb | 2203 | rcu_read_unlock(); |
1da177e4 | 2204 | return -EBUSY; |
4878e9eb | 2205 | } |
1da177e4 | 2206 | } |
4878e9eb | 2207 | rcu_read_unlock(); |
f6b6ec5c SL |
2208 | if (!test_bit(Journal, &rdev->flags) && |
2209 | mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { | |
9d48739e N |
2210 | pr_warn("md: %s: array is limited to %d devices\n", |
2211 | mdname(mddev), mddev->max_disks); | |
de01dfad N |
2212 | return -EBUSY; |
2213 | } | |
19133a42 | 2214 | bdevname(rdev->bdev,b); |
90a9befb | 2215 | strreplace(b, '/', '!'); |
649316b2 | 2216 | |
1da177e4 | 2217 | rdev->mddev = mddev; |
9d48739e | 2218 | pr_debug("md: bind<%s>\n", b); |
86e6ffdd | 2219 | |
b2d6db58 | 2220 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2221 | goto fail; |
86e6ffdd | 2222 | |
0762b8bd | 2223 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2224 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2225 | /* failure here is OK */; | |
2226 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2227 | |
4b80991c | 2228 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2229 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2230 | |
2231 | /* May as well allow recovery to be retried once */ | |
5389042f | 2232 | mddev->recovery_disabled++; |
3f9d99c1 | 2233 | |
1da177e4 | 2234 | return 0; |
5e55e2f5 N |
2235 | |
2236 | fail: | |
9d48739e N |
2237 | pr_warn("md: failed to register dev-%s for %s\n", |
2238 | b, mdname(mddev)); | |
5e55e2f5 | 2239 | return err; |
1da177e4 LT |
2240 | } |
2241 | ||
177a99b2 | 2242 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2243 | { |
3cb03002 | 2244 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2245 | kobject_del(&rdev->kobj); |
177a99b2 | 2246 | kobject_put(&rdev->kobj); |
5792a285 N |
2247 | } |
2248 | ||
f72ffdd6 | 2249 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2250 | { |
2251 | char b[BDEVNAME_SIZE]; | |
403df478 | 2252 | |
49731baa | 2253 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2254 | list_del_rcu(&rdev->same_set); |
9d48739e | 2255 | pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1da177e4 | 2256 | rdev->mddev = NULL; |
86e6ffdd | 2257 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2258 | sysfs_put(rdev->sysfs_state); |
2259 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2260 | rdev->badblocks.count = 0; |
5792a285 | 2261 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2262 | * writing to 'remove' to "dev/state". We also need |
2263 | * to delay it due to rcu usage. | |
5792a285 | 2264 | */ |
4b80991c | 2265 | synchronize_rcu(); |
177a99b2 N |
2266 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2267 | kobject_get(&rdev->kobj); | |
e804ac78 | 2268 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2269 | } |
2270 | ||
2271 | /* | |
2272 | * prevent the device from being mounted, repartitioned or | |
2273 | * otherwise reused by a RAID array (or any other kernel | |
2274 | * subsystem), by bd_claiming the device. | |
2275 | */ | |
3cb03002 | 2276 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2277 | { |
2278 | int err = 0; | |
2279 | struct block_device *bdev; | |
2280 | char b[BDEVNAME_SIZE]; | |
2281 | ||
d4d77629 | 2282 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2283 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 | 2284 | if (IS_ERR(bdev)) { |
9d48739e | 2285 | pr_warn("md: could not open %s.\n", __bdevname(dev, b)); |
1da177e4 LT |
2286 | return PTR_ERR(bdev); |
2287 | } | |
1da177e4 LT |
2288 | rdev->bdev = bdev; |
2289 | return err; | |
2290 | } | |
2291 | ||
3cb03002 | 2292 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2293 | { |
2294 | struct block_device *bdev = rdev->bdev; | |
2295 | rdev->bdev = NULL; | |
e525fd89 | 2296 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2297 | } |
2298 | ||
2299 | void md_autodetect_dev(dev_t dev); | |
2300 | ||
f72ffdd6 | 2301 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2302 | { |
2303 | char b[BDEVNAME_SIZE]; | |
403df478 | 2304 | |
9d48739e | 2305 | pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b)); |
545c8795 | 2306 | md_rdev_clear(rdev); |
1da177e4 | 2307 | #ifndef MODULE |
d0fae18f N |
2308 | if (test_bit(AutoDetected, &rdev->flags)) |
2309 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2310 | #endif |
2311 | unlock_rdev(rdev); | |
86e6ffdd | 2312 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2313 | } |
2314 | ||
fb56dfef | 2315 | void md_kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2316 | { |
2317 | unbind_rdev_from_array(rdev); | |
2318 | export_rdev(rdev); | |
2319 | } | |
fb56dfef | 2320 | EXPORT_SYMBOL_GPL(md_kick_rdev_from_array); |
1da177e4 | 2321 | |
fd01b88c | 2322 | static void export_array(struct mddev *mddev) |
1da177e4 | 2323 | { |
0638bb0e | 2324 | struct md_rdev *rdev; |
1da177e4 | 2325 | |
0638bb0e N |
2326 | while (!list_empty(&mddev->disks)) { |
2327 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2328 | same_set); | |
fb56dfef | 2329 | md_kick_rdev_from_array(rdev); |
1da177e4 | 2330 | } |
1da177e4 LT |
2331 | mddev->raid_disks = 0; |
2332 | mddev->major_version = 0; | |
2333 | } | |
2334 | ||
6497709b N |
2335 | static bool set_in_sync(struct mddev *mddev) |
2336 | { | |
efa4b77b | 2337 | lockdep_assert_held(&mddev->lock); |
4ad23a97 N |
2338 | if (!mddev->in_sync) { |
2339 | mddev->sync_checkers++; | |
2340 | spin_unlock(&mddev->lock); | |
2341 | percpu_ref_switch_to_atomic_sync(&mddev->writes_pending); | |
2342 | spin_lock(&mddev->lock); | |
2343 | if (!mddev->in_sync && | |
2344 | percpu_ref_is_zero(&mddev->writes_pending)) { | |
6497709b | 2345 | mddev->in_sync = 1; |
4ad23a97 N |
2346 | /* |
2347 | * Ensure ->in_sync is visible before we clear | |
2348 | * ->sync_checkers. | |
2349 | */ | |
55cc39f3 | 2350 | smp_mb(); |
6497709b N |
2351 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2352 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
2353 | } | |
4ad23a97 N |
2354 | if (--mddev->sync_checkers == 0) |
2355 | percpu_ref_switch_to_percpu(&mddev->writes_pending); | |
6497709b N |
2356 | } |
2357 | if (mddev->safemode == 1) | |
2358 | mddev->safemode = 0; | |
2359 | return mddev->in_sync; | |
2360 | } | |
2361 | ||
f72ffdd6 | 2362 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2363 | { |
42543769 N |
2364 | /* Update each superblock (in-memory image), but |
2365 | * if we are allowed to, skip spares which already | |
2366 | * have the right event counter, or have one earlier | |
2367 | * (which would mean they aren't being marked as dirty | |
2368 | * with the rest of the array) | |
2369 | */ | |
3cb03002 | 2370 | struct md_rdev *rdev; |
dafb20fa | 2371 | rdev_for_each(rdev, mddev) { |
42543769 N |
2372 | if (rdev->sb_events == mddev->events || |
2373 | (nospares && | |
2374 | rdev->raid_disk < 0 && | |
42543769 N |
2375 | rdev->sb_events+1 == mddev->events)) { |
2376 | /* Don't update this superblock */ | |
2377 | rdev->sb_loaded = 2; | |
2378 | } else { | |
076f968b | 2379 | sync_super(mddev, rdev); |
42543769 N |
2380 | rdev->sb_loaded = 1; |
2381 | } | |
1da177e4 LT |
2382 | } |
2383 | } | |
2384 | ||
2aa82191 GR |
2385 | static bool does_sb_need_changing(struct mddev *mddev) |
2386 | { | |
2387 | struct md_rdev *rdev; | |
2388 | struct mdp_superblock_1 *sb; | |
2389 | int role; | |
2390 | ||
2391 | /* Find a good rdev */ | |
2392 | rdev_for_each(rdev, mddev) | |
2393 | if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags)) | |
2394 | break; | |
2395 | ||
2396 | /* No good device found. */ | |
2397 | if (!rdev) | |
2398 | return false; | |
2399 | ||
2400 | sb = page_address(rdev->sb_page); | |
2401 | /* Check if a device has become faulty or a spare become active */ | |
2402 | rdev_for_each(rdev, mddev) { | |
2403 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
2404 | /* Device activated? */ | |
2405 | if (role == 0xffff && rdev->raid_disk >=0 && | |
2406 | !test_bit(Faulty, &rdev->flags)) | |
2407 | return true; | |
2408 | /* Device turned faulty? */ | |
2409 | if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd)) | |
2410 | return true; | |
2411 | } | |
2412 | ||
2413 | /* Check if any mddev parameters have changed */ | |
2414 | if ((mddev->dev_sectors != le64_to_cpu(sb->size)) || | |
2415 | (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) || | |
13459213 | 2416 | (mddev->layout != le32_to_cpu(sb->layout)) || |
2aa82191 GR |
2417 | (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) || |
2418 | (mddev->chunk_sectors != le32_to_cpu(sb->chunksize))) | |
2419 | return true; | |
2420 | ||
2421 | return false; | |
2422 | } | |
2423 | ||
1aee41f6 | 2424 | void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2425 | { |
3cb03002 | 2426 | struct md_rdev *rdev; |
06d91a5f | 2427 | int sync_req; |
42543769 | 2428 | int nospares = 0; |
2699b672 | 2429 | int any_badblocks_changed = 0; |
23b63f9f | 2430 | int ret = -1; |
1da177e4 | 2431 | |
d87f064f N |
2432 | if (mddev->ro) { |
2433 | if (force_change) | |
2953079c | 2434 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
d87f064f N |
2435 | return; |
2436 | } | |
2aa82191 | 2437 | |
2c97cf13 | 2438 | repeat: |
2aa82191 | 2439 | if (mddev_is_clustered(mddev)) { |
2953079c | 2440 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
2aa82191 | 2441 | force_change = 1; |
2953079c | 2442 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
85ad1d13 | 2443 | nospares = 1; |
23b63f9f | 2444 | ret = md_cluster_ops->metadata_update_start(mddev); |
2aa82191 GR |
2445 | /* Has someone else has updated the sb */ |
2446 | if (!does_sb_need_changing(mddev)) { | |
23b63f9f GJ |
2447 | if (ret == 0) |
2448 | md_cluster_ops->metadata_update_cancel(mddev); | |
2953079c SL |
2449 | bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2450 | BIT(MD_SB_CHANGE_DEVS) | | |
2451 | BIT(MD_SB_CHANGE_CLEAN)); | |
2aa82191 GR |
2452 | return; |
2453 | } | |
2454 | } | |
2c97cf13 | 2455 | |
db0505d3 N |
2456 | /* |
2457 | * First make sure individual recovery_offsets are correct | |
2458 | * curr_resync_completed can only be used during recovery. | |
2459 | * During reshape/resync it might use array-addresses rather | |
2460 | * that device addresses. | |
2461 | */ | |
dafb20fa | 2462 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2463 | if (rdev->raid_disk >= 0 && |
2464 | mddev->delta_disks >= 0 && | |
db0505d3 N |
2465 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
2466 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) && | |
2467 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
f2076e7d | 2468 | !test_bit(Journal, &rdev->flags) && |
3a3a5ddb N |
2469 | !test_bit(In_sync, &rdev->flags) && |
2470 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2471 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2472 | ||
f72ffdd6 | 2473 | } |
bd52b746 | 2474 | if (!mddev->persistent) { |
2953079c SL |
2475 | clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2476 | clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
de393cde | 2477 | if (!mddev->external) { |
2953079c | 2478 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
dafb20fa | 2479 | rdev_for_each(rdev, mddev) { |
de393cde | 2480 | if (rdev->badblocks.changed) { |
d0962936 | 2481 | rdev->badblocks.changed = 0; |
fc974ee2 | 2482 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2483 | md_error(mddev, rdev); |
2484 | } | |
2485 | clear_bit(Blocked, &rdev->flags); | |
2486 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2487 | wake_up(&rdev->blocked_wait); | |
2488 | } | |
2489 | } | |
3a3a5ddb N |
2490 | wake_up(&mddev->sb_wait); |
2491 | return; | |
2492 | } | |
2493 | ||
85572d7c | 2494 | spin_lock(&mddev->lock); |
84692195 | 2495 | |
9ebc6ef1 | 2496 | mddev->utime = ktime_get_real_seconds(); |
3a3a5ddb | 2497 | |
2953079c | 2498 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
850b2b42 | 2499 | force_change = 1; |
2953079c | 2500 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
850b2b42 N |
2501 | /* just a clean<-> dirty transition, possibly leave spares alone, |
2502 | * though if events isn't the right even/odd, we will have to do | |
2503 | * spares after all | |
2504 | */ | |
2505 | nospares = 1; | |
2506 | if (force_change) | |
2507 | nospares = 0; | |
2508 | if (mddev->degraded) | |
84692195 N |
2509 | /* If the array is degraded, then skipping spares is both |
2510 | * dangerous and fairly pointless. | |
2511 | * Dangerous because a device that was removed from the array | |
2512 | * might have a event_count that still looks up-to-date, | |
2513 | * so it can be re-added without a resync. | |
2514 | * Pointless because if there are any spares to skip, | |
2515 | * then a recovery will happen and soon that array won't | |
2516 | * be degraded any more and the spare can go back to sleep then. | |
2517 | */ | |
850b2b42 | 2518 | nospares = 0; |
84692195 | 2519 | |
06d91a5f | 2520 | sync_req = mddev->in_sync; |
42543769 N |
2521 | |
2522 | /* If this is just a dirty<->clean transition, and the array is clean | |
2523 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2524 | if (nospares |
42543769 | 2525 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2526 | && mddev->can_decrease_events |
2527 | && mddev->events != 1) { | |
42543769 | 2528 | mddev->events--; |
a8707c08 N |
2529 | mddev->can_decrease_events = 0; |
2530 | } else { | |
42543769 N |
2531 | /* otherwise we have to go forward and ... */ |
2532 | mddev->events ++; | |
a8707c08 | 2533 | mddev->can_decrease_events = nospares; |
42543769 | 2534 | } |
1da177e4 | 2535 | |
403df478 N |
2536 | /* |
2537 | * This 64-bit counter should never wrap. | |
2538 | * Either we are in around ~1 trillion A.C., assuming | |
2539 | * 1 reboot per second, or we have a bug... | |
2540 | */ | |
2541 | WARN_ON(mddev->events == 0); | |
2699b672 | 2542 | |
dafb20fa | 2543 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2544 | if (rdev->badblocks.changed) |
2545 | any_badblocks_changed++; | |
de393cde N |
2546 | if (test_bit(Faulty, &rdev->flags)) |
2547 | set_bit(FaultRecorded, &rdev->flags); | |
2548 | } | |
2699b672 | 2549 | |
e691063a | 2550 | sync_sbs(mddev, nospares); |
85572d7c | 2551 | spin_unlock(&mddev->lock); |
1da177e4 | 2552 | |
36a4e1fe N |
2553 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2554 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2555 | |
504634f6 SL |
2556 | if (mddev->queue) |
2557 | blk_add_trace_msg(mddev->queue, "md md_update_sb"); | |
46533ff7 | 2558 | rewrite: |
4ad13663 | 2559 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2560 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2561 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2562 | |
42543769 N |
2563 | if (rdev->sb_loaded != 1) |
2564 | continue; /* no noise on spare devices */ | |
1da177e4 | 2565 | |
f466722c | 2566 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2567 | md_super_write(mddev,rdev, |
0f420358 | 2568 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2569 | rdev->sb_page); |
36a4e1fe N |
2570 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2571 | bdevname(rdev->bdev, b), | |
2572 | (unsigned long long)rdev->sb_start); | |
42543769 | 2573 | rdev->sb_events = mddev->events; |
2699b672 N |
2574 | if (rdev->badblocks.size) { |
2575 | md_super_write(mddev, rdev, | |
2576 | rdev->badblocks.sector, | |
2577 | rdev->badblocks.size << 9, | |
2578 | rdev->bb_page); | |
2579 | rdev->badblocks.size = 0; | |
2580 | } | |
7bfa19f2 | 2581 | |
f466722c | 2582 | } else |
36a4e1fe N |
2583 | pr_debug("md: %s (skipping faulty)\n", |
2584 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2585 | |
7bfa19f2 | 2586 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2587 | /* only need to write one superblock... */ |
2588 | break; | |
2589 | } | |
46533ff7 N |
2590 | if (md_super_wait(mddev) < 0) |
2591 | goto rewrite; | |
2953079c | 2592 | /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2593 | |
2c97cf13 GJ |
2594 | if (mddev_is_clustered(mddev) && ret == 0) |
2595 | md_cluster_ops->metadata_update_finish(mddev); | |
2596 | ||
850b2b42 | 2597 | if (mddev->in_sync != sync_req || |
2953079c SL |
2598 | !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2599 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN))) | |
06d91a5f | 2600 | /* have to write it out again */ |
06d91a5f | 2601 | goto repeat; |
3d310eb7 | 2602 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2603 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2604 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2605 | |
dafb20fa | 2606 | rdev_for_each(rdev, mddev) { |
de393cde N |
2607 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2608 | clear_bit(Blocked, &rdev->flags); | |
2609 | ||
2610 | if (any_badblocks_changed) | |
fc974ee2 | 2611 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2612 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2613 | wake_up(&rdev->blocked_wait); | |
2614 | } | |
1da177e4 | 2615 | } |
1aee41f6 | 2616 | EXPORT_SYMBOL(md_update_sb); |
1da177e4 | 2617 | |
a6da4ef8 GR |
2618 | static int add_bound_rdev(struct md_rdev *rdev) |
2619 | { | |
2620 | struct mddev *mddev = rdev->mddev; | |
2621 | int err = 0; | |
87d4d916 | 2622 | bool add_journal = test_bit(Journal, &rdev->flags); |
a6da4ef8 | 2623 | |
87d4d916 | 2624 | if (!mddev->pers->hot_remove_disk || add_journal) { |
a6da4ef8 GR |
2625 | /* If there is hot_add_disk but no hot_remove_disk |
2626 | * then added disks for geometry changes, | |
2627 | * and should be added immediately. | |
2628 | */ | |
2629 | super_types[mddev->major_version]. | |
2630 | validate_super(mddev, rdev); | |
87d4d916 SL |
2631 | if (add_journal) |
2632 | mddev_suspend(mddev); | |
a6da4ef8 | 2633 | err = mddev->pers->hot_add_disk(mddev, rdev); |
87d4d916 SL |
2634 | if (add_journal) |
2635 | mddev_resume(mddev); | |
a6da4ef8 | 2636 | if (err) { |
db767672 | 2637 | md_kick_rdev_from_array(rdev); |
a6da4ef8 GR |
2638 | return err; |
2639 | } | |
2640 | } | |
2641 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2642 | ||
2953079c | 2643 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a6da4ef8 GR |
2644 | if (mddev->degraded) |
2645 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
2646 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2647 | md_new_event(mddev); | |
2648 | md_wakeup_thread(mddev->thread); | |
2649 | return 0; | |
2650 | } | |
1da177e4 | 2651 | |
7f6ce769 | 2652 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2653 | * We want to accept with case. For this we use cmd_match. |
2654 | */ | |
2655 | static int cmd_match(const char *cmd, const char *str) | |
2656 | { | |
2657 | /* See if cmd, written into a sysfs file, matches | |
2658 | * str. They must either be the same, or cmd can | |
2659 | * have a trailing newline | |
2660 | */ | |
2661 | while (*cmd && *str && *cmd == *str) { | |
2662 | cmd++; | |
2663 | str++; | |
2664 | } | |
2665 | if (*cmd == '\n') | |
2666 | cmd++; | |
2667 | if (*str || *cmd) | |
2668 | return 0; | |
2669 | return 1; | |
2670 | } | |
2671 | ||
86e6ffdd N |
2672 | struct rdev_sysfs_entry { |
2673 | struct attribute attr; | |
3cb03002 N |
2674 | ssize_t (*show)(struct md_rdev *, char *); |
2675 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2676 | }; |
2677 | ||
2678 | static ssize_t | |
3cb03002 | 2679 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd | 2680 | { |
35b785f7 | 2681 | char *sep = ","; |
20a49ff6 | 2682 | size_t len = 0; |
6aa7de05 | 2683 | unsigned long flags = READ_ONCE(rdev->flags); |
86e6ffdd | 2684 | |
758bfc8a | 2685 | if (test_bit(Faulty, &flags) || |
dcbcb486 TM |
2686 | (!test_bit(ExternalBbl, &flags) && |
2687 | rdev->badblocks.unacked_exist)) | |
35b785f7 TM |
2688 | len += sprintf(page+len, "faulty%s", sep); |
2689 | if (test_bit(In_sync, &flags)) | |
2690 | len += sprintf(page+len, "in_sync%s", sep); | |
2691 | if (test_bit(Journal, &flags)) | |
2692 | len += sprintf(page+len, "journal%s", sep); | |
2693 | if (test_bit(WriteMostly, &flags)) | |
2694 | len += sprintf(page+len, "write_mostly%s", sep); | |
758bfc8a | 2695 | if (test_bit(Blocked, &flags) || |
52c64152 | 2696 | (rdev->badblocks.unacked_exist |
35b785f7 TM |
2697 | && !test_bit(Faulty, &flags))) |
2698 | len += sprintf(page+len, "blocked%s", sep); | |
758bfc8a | 2699 | if (!test_bit(Faulty, &flags) && |
f2076e7d | 2700 | !test_bit(Journal, &flags) && |
35b785f7 TM |
2701 | !test_bit(In_sync, &flags)) |
2702 | len += sprintf(page+len, "spare%s", sep); | |
2703 | if (test_bit(WriteErrorSeen, &flags)) | |
2704 | len += sprintf(page+len, "write_error%s", sep); | |
2705 | if (test_bit(WantReplacement, &flags)) | |
2706 | len += sprintf(page+len, "want_replacement%s", sep); | |
2707 | if (test_bit(Replacement, &flags)) | |
2708 | len += sprintf(page+len, "replacement%s", sep); | |
2709 | if (test_bit(ExternalBbl, &flags)) | |
2710 | len += sprintf(page+len, "external_bbl%s", sep); | |
688834e6 N |
2711 | if (test_bit(FailFast, &flags)) |
2712 | len += sprintf(page+len, "failfast%s", sep); | |
35b785f7 TM |
2713 | |
2714 | if (len) | |
2715 | len -= strlen(sep); | |
2d78f8c4 | 2716 | |
86e6ffdd N |
2717 | return len+sprintf(page+len, "\n"); |
2718 | } | |
2719 | ||
45dc2de1 | 2720 | static ssize_t |
3cb03002 | 2721 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2722 | { |
2723 | /* can write | |
de393cde | 2724 | * faulty - simulates an error |
45dc2de1 | 2725 | * remove - disconnects the device |
f655675b N |
2726 | * writemostly - sets write_mostly |
2727 | * -writemostly - clears write_mostly | |
de393cde N |
2728 | * blocked - sets the Blocked flags |
2729 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2730 | * insync - sets Insync providing device isn't active |
f466722c N |
2731 | * -insync - clear Insync for a device with a slot assigned, |
2732 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2733 | * write_error - sets WriteErrorSeen |
2734 | * -write_error - clears WriteErrorSeen | |
688834e6 | 2735 | * {,-}failfast - set/clear FailFast |
45dc2de1 N |
2736 | */ |
2737 | int err = -EINVAL; | |
2738 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2739 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2740 | if (test_bit(Faulty, &rdev->flags)) |
2741 | err = 0; | |
2742 | else | |
2743 | err = -EBUSY; | |
45dc2de1 | 2744 | } else if (cmd_match(buf, "remove")) { |
5d881783 SL |
2745 | if (rdev->mddev->pers) { |
2746 | clear_bit(Blocked, &rdev->flags); | |
2747 | remove_and_add_spares(rdev->mddev, rdev); | |
2748 | } | |
45dc2de1 N |
2749 | if (rdev->raid_disk >= 0) |
2750 | err = -EBUSY; | |
2751 | else { | |
fd01b88c | 2752 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2753 | err = 0; |
a9720903 GJ |
2754 | if (mddev_is_clustered(mddev)) |
2755 | err = md_cluster_ops->remove_disk(mddev, rdev); | |
2756 | ||
2757 | if (err == 0) { | |
2758 | md_kick_rdev_from_array(rdev); | |
060b0689 | 2759 | if (mddev->pers) { |
2953079c | 2760 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
2761 | md_wakeup_thread(mddev->thread); |
2762 | } | |
a9720903 GJ |
2763 | md_new_event(mddev); |
2764 | } | |
45dc2de1 | 2765 | } |
f655675b N |
2766 | } else if (cmd_match(buf, "writemostly")) { |
2767 | set_bit(WriteMostly, &rdev->flags); | |
2768 | err = 0; | |
2769 | } else if (cmd_match(buf, "-writemostly")) { | |
2770 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2771 | err = 0; |
2772 | } else if (cmd_match(buf, "blocked")) { | |
2773 | set_bit(Blocked, &rdev->flags); | |
2774 | err = 0; | |
2775 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2776 | if (!test_bit(Faulty, &rdev->flags) && |
dcbcb486 | 2777 | !test_bit(ExternalBbl, &rdev->flags) && |
7da64a0a | 2778 | rdev->badblocks.unacked_exist) { |
de393cde N |
2779 | /* metadata handler doesn't understand badblocks, |
2780 | * so we need to fail the device | |
2781 | */ | |
2782 | md_error(rdev->mddev, rdev); | |
2783 | } | |
6bfe0b49 | 2784 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2785 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2786 | wake_up(&rdev->blocked_wait); |
2787 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2788 | md_wakeup_thread(rdev->mddev->thread); | |
2789 | ||
6d56e278 N |
2790 | err = 0; |
2791 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2792 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2793 | err = 0; |
688834e6 N |
2794 | } else if (cmd_match(buf, "failfast")) { |
2795 | set_bit(FailFast, &rdev->flags); | |
2796 | err = 0; | |
2797 | } else if (cmd_match(buf, "-failfast")) { | |
2798 | clear_bit(FailFast, &rdev->flags); | |
2799 | err = 0; | |
f2076e7d SL |
2800 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 && |
2801 | !test_bit(Journal, &rdev->flags)) { | |
e1960f8c N |
2802 | if (rdev->mddev->pers == NULL) { |
2803 | clear_bit(In_sync, &rdev->flags); | |
2804 | rdev->saved_raid_disk = rdev->raid_disk; | |
2805 | rdev->raid_disk = -1; | |
2806 | err = 0; | |
2807 | } | |
d7a9d443 N |
2808 | } else if (cmd_match(buf, "write_error")) { |
2809 | set_bit(WriteErrorSeen, &rdev->flags); | |
2810 | err = 0; | |
2811 | } else if (cmd_match(buf, "-write_error")) { | |
2812 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2813 | err = 0; | |
2d78f8c4 N |
2814 | } else if (cmd_match(buf, "want_replacement")) { |
2815 | /* Any non-spare device that is not a replacement can | |
2816 | * become want_replacement at any time, but we then need to | |
2817 | * check if recovery is needed. | |
2818 | */ | |
2819 | if (rdev->raid_disk >= 0 && | |
f2076e7d | 2820 | !test_bit(Journal, &rdev->flags) && |
2d78f8c4 N |
2821 | !test_bit(Replacement, &rdev->flags)) |
2822 | set_bit(WantReplacement, &rdev->flags); | |
2823 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2824 | md_wakeup_thread(rdev->mddev->thread); | |
2825 | err = 0; | |
2826 | } else if (cmd_match(buf, "-want_replacement")) { | |
2827 | /* Clearing 'want_replacement' is always allowed. | |
2828 | * Once replacements starts it is too late though. | |
2829 | */ | |
2830 | err = 0; | |
2831 | clear_bit(WantReplacement, &rdev->flags); | |
2832 | } else if (cmd_match(buf, "replacement")) { | |
2833 | /* Can only set a device as a replacement when array has not | |
2834 | * yet been started. Once running, replacement is automatic | |
2835 | * from spares, or by assigning 'slot'. | |
2836 | */ | |
2837 | if (rdev->mddev->pers) | |
2838 | err = -EBUSY; | |
2839 | else { | |
2840 | set_bit(Replacement, &rdev->flags); | |
2841 | err = 0; | |
2842 | } | |
2843 | } else if (cmd_match(buf, "-replacement")) { | |
2844 | /* Similarly, can only clear Replacement before start */ | |
2845 | if (rdev->mddev->pers) | |
2846 | err = -EBUSY; | |
2847 | else { | |
2848 | clear_bit(Replacement, &rdev->flags); | |
2849 | err = 0; | |
2850 | } | |
a6da4ef8 GR |
2851 | } else if (cmd_match(buf, "re-add")) { |
2852 | if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) { | |
97f6cd39 GR |
2853 | /* clear_bit is performed _after_ all the devices |
2854 | * have their local Faulty bit cleared. If any writes | |
2855 | * happen in the meantime in the local node, they | |
2856 | * will land in the local bitmap, which will be synced | |
2857 | * by this node eventually | |
2858 | */ | |
2859 | if (!mddev_is_clustered(rdev->mddev) || | |
2860 | (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) { | |
2861 | clear_bit(Faulty, &rdev->flags); | |
2862 | err = add_bound_rdev(rdev); | |
2863 | } | |
a6da4ef8 GR |
2864 | } else |
2865 | err = -EBUSY; | |
35b785f7 TM |
2866 | } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) { |
2867 | set_bit(ExternalBbl, &rdev->flags); | |
2868 | rdev->badblocks.shift = 0; | |
2869 | err = 0; | |
2870 | } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) { | |
2871 | clear_bit(ExternalBbl, &rdev->flags); | |
2872 | err = 0; | |
45dc2de1 | 2873 | } |
00bcb4ac N |
2874 | if (!err) |
2875 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2876 | return err ? err : len; |
2877 | } | |
80ca3a44 | 2878 | static struct rdev_sysfs_entry rdev_state = |
750f199e | 2879 | __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store); |
86e6ffdd | 2880 | |
4dbcdc75 | 2881 | static ssize_t |
3cb03002 | 2882 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2883 | { |
2884 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2885 | } | |
2886 | ||
2887 | static ssize_t | |
3cb03002 | 2888 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 | 2889 | { |
4c9309c0 AD |
2890 | unsigned int n; |
2891 | int rv; | |
2892 | ||
2893 | rv = kstrtouint(buf, 10, &n); | |
2894 | if (rv < 0) | |
2895 | return rv; | |
2896 | atomic_set(&rdev->corrected_errors, n); | |
2897 | return len; | |
4dbcdc75 N |
2898 | } |
2899 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2900 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2901 | |
014236d2 | 2902 | static ssize_t |
3cb03002 | 2903 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 | 2904 | { |
f2076e7d SL |
2905 | if (test_bit(Journal, &rdev->flags)) |
2906 | return sprintf(page, "journal\n"); | |
2907 | else if (rdev->raid_disk < 0) | |
014236d2 N |
2908 | return sprintf(page, "none\n"); |
2909 | else | |
2910 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2911 | } | |
2912 | ||
2913 | static ssize_t | |
3cb03002 | 2914 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 | 2915 | { |
4c9309c0 | 2916 | int slot; |
c303da6d | 2917 | int err; |
4c9309c0 | 2918 | |
f2076e7d SL |
2919 | if (test_bit(Journal, &rdev->flags)) |
2920 | return -EBUSY; | |
014236d2 N |
2921 | if (strncmp(buf, "none", 4)==0) |
2922 | slot = -1; | |
4c9309c0 AD |
2923 | else { |
2924 | err = kstrtouint(buf, 10, (unsigned int *)&slot); | |
2925 | if (err < 0) | |
2926 | return err; | |
2927 | } | |
6c2fce2e | 2928 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2929 | /* Setting 'slot' on an active array requires also |
2930 | * updating the 'rd%d' link, and communicating | |
2931 | * with the personality with ->hot_*_disk. | |
2932 | * For now we only support removing | |
2933 | * failed/spare devices. This normally happens automatically, | |
2934 | * but not when the metadata is externally managed. | |
2935 | */ | |
c303da6d N |
2936 | if (rdev->raid_disk == -1) |
2937 | return -EEXIST; | |
2938 | /* personality does all needed checks */ | |
01393f3d | 2939 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2940 | return -EINVAL; |
746d3207 N |
2941 | clear_bit(Blocked, &rdev->flags); |
2942 | remove_and_add_spares(rdev->mddev, rdev); | |
2943 | if (rdev->raid_disk >= 0) | |
2944 | return -EBUSY; | |
c303da6d N |
2945 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2946 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2947 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2948 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2949 | * if we ever get bitmaps working here. |
6c2fce2e | 2950 | */ |
cb01c549 | 2951 | int err; |
6c2fce2e NB |
2952 | |
2953 | if (rdev->raid_disk != -1) | |
2954 | return -EBUSY; | |
2955 | ||
c6751b2b N |
2956 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2957 | return -EBUSY; | |
2958 | ||
6c2fce2e NB |
2959 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2960 | return -EINVAL; | |
2961 | ||
ba1b41b6 N |
2962 | if (slot >= rdev->mddev->raid_disks && |
2963 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2964 | return -ENOSPC; | |
2965 | ||
6c2fce2e NB |
2966 | rdev->raid_disk = slot; |
2967 | if (test_bit(In_sync, &rdev->flags)) | |
2968 | rdev->saved_raid_disk = slot; | |
2969 | else | |
2970 | rdev->saved_raid_disk = -1; | |
d30519fc | 2971 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2972 | clear_bit(Bitmap_sync, &rdev->flags); |
cb01c549 GR |
2973 | err = rdev->mddev->pers-> |
2974 | hot_add_disk(rdev->mddev, rdev); | |
2975 | if (err) { | |
2976 | rdev->raid_disk = -1; | |
2977 | return err; | |
2978 | } else | |
2979 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2980 | if (sysfs_link_rdev(rdev->mddev, rdev)) | |
2981 | /* failure here is OK */; | |
6c2fce2e | 2982 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2983 | } else { |
ba1b41b6 N |
2984 | if (slot >= rdev->mddev->raid_disks && |
2985 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2986 | return -ENOSPC; |
2987 | rdev->raid_disk = slot; | |
2988 | /* assume it is working */ | |
c5d79adb N |
2989 | clear_bit(Faulty, &rdev->flags); |
2990 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2991 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2992 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2993 | } |
014236d2 N |
2994 | return len; |
2995 | } | |
2996 | ||
014236d2 | 2997 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2998 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2999 | |
93c8cad0 | 3000 | static ssize_t |
3cb03002 | 3001 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 3002 | { |
6961ece4 | 3003 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
3004 | } |
3005 | ||
3006 | static ssize_t | |
3cb03002 | 3007 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 3008 | { |
c6563a8c | 3009 | unsigned long long offset; |
b29bebd6 | 3010 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 3011 | return -EINVAL; |
8ed0a521 | 3012 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 3013 | return -EBUSY; |
dd8ac336 | 3014 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
3015 | /* Must set offset before size, so overlap checks |
3016 | * can be sane */ | |
3017 | return -EBUSY; | |
93c8cad0 | 3018 | rdev->data_offset = offset; |
25f7fd47 | 3019 | rdev->new_data_offset = offset; |
93c8cad0 N |
3020 | return len; |
3021 | } | |
3022 | ||
3023 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 3024 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 3025 | |
c6563a8c N |
3026 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
3027 | { | |
3028 | return sprintf(page, "%llu\n", | |
3029 | (unsigned long long)rdev->new_data_offset); | |
3030 | } | |
3031 | ||
3032 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
3033 | const char *buf, size_t len) | |
3034 | { | |
3035 | unsigned long long new_offset; | |
3036 | struct mddev *mddev = rdev->mddev; | |
3037 | ||
b29bebd6 | 3038 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
3039 | return -EINVAL; |
3040 | ||
f851b60d N |
3041 | if (mddev->sync_thread || |
3042 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
3043 | return -EBUSY; |
3044 | if (new_offset == rdev->data_offset) | |
3045 | /* reset is always permitted */ | |
3046 | ; | |
3047 | else if (new_offset > rdev->data_offset) { | |
3048 | /* must not push array size beyond rdev_sectors */ | |
3049 | if (new_offset - rdev->data_offset | |
3050 | + mddev->dev_sectors > rdev->sectors) | |
3051 | return -E2BIG; | |
3052 | } | |
3053 | /* Metadata worries about other space details. */ | |
3054 | ||
3055 | /* decreasing the offset is inconsistent with a backwards | |
3056 | * reshape. | |
3057 | */ | |
3058 | if (new_offset < rdev->data_offset && | |
3059 | mddev->reshape_backwards) | |
3060 | return -EINVAL; | |
3061 | /* Increasing offset is inconsistent with forwards | |
3062 | * reshape. reshape_direction should be set to | |
3063 | * 'backwards' first. | |
3064 | */ | |
3065 | if (new_offset > rdev->data_offset && | |
3066 | !mddev->reshape_backwards) | |
3067 | return -EINVAL; | |
3068 | ||
3069 | if (mddev->pers && mddev->persistent && | |
3070 | !super_types[mddev->major_version] | |
3071 | .allow_new_offset(rdev, new_offset)) | |
3072 | return -E2BIG; | |
3073 | rdev->new_data_offset = new_offset; | |
3074 | if (new_offset > rdev->data_offset) | |
3075 | mddev->reshape_backwards = 1; | |
3076 | else if (new_offset < rdev->data_offset) | |
3077 | mddev->reshape_backwards = 0; | |
3078 | ||
3079 | return len; | |
3080 | } | |
3081 | static struct rdev_sysfs_entry rdev_new_offset = | |
3082 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
3083 | ||
83303b61 | 3084 | static ssize_t |
3cb03002 | 3085 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 3086 | { |
dd8ac336 | 3087 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
3088 | } |
3089 | ||
c5d79adb N |
3090 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
3091 | { | |
3092 | /* check if two start/length pairs overlap */ | |
3093 | if (s1+l1 <= s2) | |
3094 | return 0; | |
3095 | if (s2+l2 <= s1) | |
3096 | return 0; | |
3097 | return 1; | |
3098 | } | |
3099 | ||
b522adcd DW |
3100 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
3101 | { | |
3102 | unsigned long long blocks; | |
3103 | sector_t new; | |
3104 | ||
b29bebd6 | 3105 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
3106 | return -EINVAL; |
3107 | ||
3108 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
3109 | return -EINVAL; /* sector conversion overflow */ | |
3110 | ||
3111 | new = blocks * 2; | |
3112 | if (new != blocks * 2) | |
3113 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
3114 | ||
3115 | *sectors = new; | |
3116 | return 0; | |
3117 | } | |
3118 | ||
83303b61 | 3119 | static ssize_t |
3cb03002 | 3120 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 3121 | { |
fd01b88c | 3122 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 3123 | sector_t oldsectors = rdev->sectors; |
b522adcd | 3124 | sector_t sectors; |
27c529bb | 3125 | |
f2076e7d SL |
3126 | if (test_bit(Journal, &rdev->flags)) |
3127 | return -EBUSY; | |
b522adcd | 3128 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 3129 | return -EINVAL; |
c6563a8c N |
3130 | if (rdev->data_offset != rdev->new_data_offset) |
3131 | return -EINVAL; /* too confusing */ | |
0cd17fec | 3132 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 3133 | if (my_mddev->persistent) { |
dd8ac336 AN |
3134 | sectors = super_types[my_mddev->major_version]. |
3135 | rdev_size_change(rdev, sectors); | |
3136 | if (!sectors) | |
0cd17fec | 3137 | return -EBUSY; |
dd8ac336 | 3138 | } else if (!sectors) |
77304d2a | 3139 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 3140 | rdev->data_offset; |
a6468539 N |
3141 | if (!my_mddev->pers->resize) |
3142 | /* Cannot change size for RAID0 or Linear etc */ | |
3143 | return -EINVAL; | |
0cd17fec | 3144 | } |
dd8ac336 | 3145 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 3146 | return -EINVAL; /* component must fit device */ |
0cd17fec | 3147 | |
dd8ac336 AN |
3148 | rdev->sectors = sectors; |
3149 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
3150 | /* Need to check that all other rdevs with the same |
3151 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
3152 | * the rdev lists safely. | |
3153 | * This check does not provide a hard guarantee, it | |
3154 | * just helps avoid dangerous mistakes. | |
c5d79adb | 3155 | */ |
fd01b88c | 3156 | struct mddev *mddev; |
c5d79adb | 3157 | int overlap = 0; |
159ec1fc | 3158 | struct list_head *tmp; |
c5d79adb | 3159 | |
8b1afc3d | 3160 | rcu_read_lock(); |
29ac4aa3 | 3161 | for_each_mddev(mddev, tmp) { |
3cb03002 | 3162 | struct md_rdev *rdev2; |
c5d79adb | 3163 | |
dafb20fa | 3164 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
3165 | if (rdev->bdev == rdev2->bdev && |
3166 | rdev != rdev2 && | |
3167 | overlaps(rdev->data_offset, rdev->sectors, | |
3168 | rdev2->data_offset, | |
3169 | rdev2->sectors)) { | |
c5d79adb N |
3170 | overlap = 1; |
3171 | break; | |
3172 | } | |
c5d79adb N |
3173 | if (overlap) { |
3174 | mddev_put(mddev); | |
3175 | break; | |
3176 | } | |
3177 | } | |
8b1afc3d | 3178 | rcu_read_unlock(); |
c5d79adb N |
3179 | if (overlap) { |
3180 | /* Someone else could have slipped in a size | |
3181 | * change here, but doing so is just silly. | |
dd8ac336 | 3182 | * We put oldsectors back because we *know* it is |
c5d79adb N |
3183 | * safe, and trust userspace not to race with |
3184 | * itself | |
3185 | */ | |
dd8ac336 | 3186 | rdev->sectors = oldsectors; |
c5d79adb N |
3187 | return -EBUSY; |
3188 | } | |
3189 | } | |
83303b61 N |
3190 | return len; |
3191 | } | |
3192 | ||
3193 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 3194 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 3195 | |
3cb03002 | 3196 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
3197 | { |
3198 | unsigned long long recovery_start = rdev->recovery_offset; | |
3199 | ||
3200 | if (test_bit(In_sync, &rdev->flags) || | |
3201 | recovery_start == MaxSector) | |
3202 | return sprintf(page, "none\n"); | |
3203 | ||
3204 | return sprintf(page, "%llu\n", recovery_start); | |
3205 | } | |
3206 | ||
3cb03002 | 3207 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
3208 | { |
3209 | unsigned long long recovery_start; | |
3210 | ||
3211 | if (cmd_match(buf, "none")) | |
3212 | recovery_start = MaxSector; | |
b29bebd6 | 3213 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
3214 | return -EINVAL; |
3215 | ||
3216 | if (rdev->mddev->pers && | |
3217 | rdev->raid_disk >= 0) | |
3218 | return -EBUSY; | |
3219 | ||
3220 | rdev->recovery_offset = recovery_start; | |
3221 | if (recovery_start == MaxSector) | |
3222 | set_bit(In_sync, &rdev->flags); | |
3223 | else | |
3224 | clear_bit(In_sync, &rdev->flags); | |
3225 | return len; | |
3226 | } | |
3227 | ||
3228 | static struct rdev_sysfs_entry rdev_recovery_start = | |
3229 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
3230 | ||
fc974ee2 VV |
3231 | /* sysfs access to bad-blocks list. |
3232 | * We present two files. | |
3233 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
3234 | * are recorded as bad. The list is truncated to fit within | |
3235 | * the one-page limit of sysfs. | |
3236 | * Writing "sector length" to this file adds an acknowledged | |
3237 | * bad block list. | |
3238 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
3239 | * been acknowledged. Writing to this file adds bad blocks | |
3240 | * without acknowledging them. This is largely for testing. | |
3241 | */ | |
3cb03002 | 3242 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3243 | { |
3244 | return badblocks_show(&rdev->badblocks, page, 0); | |
3245 | } | |
3cb03002 | 3246 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 3247 | { |
de393cde N |
3248 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
3249 | /* Maybe that ack was all we needed */ | |
3250 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
3251 | wake_up(&rdev->blocked_wait); | |
3252 | return rv; | |
16c791a5 N |
3253 | } |
3254 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
3255 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
3256 | ||
3cb03002 | 3257 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3258 | { |
3259 | return badblocks_show(&rdev->badblocks, page, 1); | |
3260 | } | |
3cb03002 | 3261 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
3262 | { |
3263 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
3264 | } | |
3265 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
3266 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
3267 | ||
664aed04 AP |
3268 | static ssize_t |
3269 | ppl_sector_show(struct md_rdev *rdev, char *page) | |
3270 | { | |
3271 | return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector); | |
3272 | } | |
3273 | ||
3274 | static ssize_t | |
3275 | ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len) | |
3276 | { | |
3277 | unsigned long long sector; | |
3278 | ||
3279 | if (kstrtoull(buf, 10, §or) < 0) | |
3280 | return -EINVAL; | |
3281 | if (sector != (sector_t)sector) | |
3282 | return -EINVAL; | |
3283 | ||
3284 | if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) && | |
3285 | rdev->raid_disk >= 0) | |
3286 | return -EBUSY; | |
3287 | ||
3288 | if (rdev->mddev->persistent) { | |
3289 | if (rdev->mddev->major_version == 0) | |
3290 | return -EINVAL; | |
3291 | if ((sector > rdev->sb_start && | |
3292 | sector - rdev->sb_start > S16_MAX) || | |
3293 | (sector < rdev->sb_start && | |
3294 | rdev->sb_start - sector > -S16_MIN)) | |
3295 | return -EINVAL; | |
3296 | rdev->ppl.offset = sector - rdev->sb_start; | |
3297 | } else if (!rdev->mddev->external) { | |
3298 | return -EBUSY; | |
3299 | } | |
3300 | rdev->ppl.sector = sector; | |
3301 | return len; | |
3302 | } | |
3303 | ||
3304 | static struct rdev_sysfs_entry rdev_ppl_sector = | |
3305 | __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store); | |
3306 | ||
3307 | static ssize_t | |
3308 | ppl_size_show(struct md_rdev *rdev, char *page) | |
3309 | { | |
3310 | return sprintf(page, "%u\n", rdev->ppl.size); | |
3311 | } | |
3312 | ||
3313 | static ssize_t | |
3314 | ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len) | |
3315 | { | |
3316 | unsigned int size; | |
3317 | ||
3318 | if (kstrtouint(buf, 10, &size) < 0) | |
3319 | return -EINVAL; | |
3320 | ||
3321 | if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) && | |
3322 | rdev->raid_disk >= 0) | |
3323 | return -EBUSY; | |
3324 | ||
3325 | if (rdev->mddev->persistent) { | |
3326 | if (rdev->mddev->major_version == 0) | |
3327 | return -EINVAL; | |
3328 | if (size > U16_MAX) | |
3329 | return -EINVAL; | |
3330 | } else if (!rdev->mddev->external) { | |
3331 | return -EBUSY; | |
3332 | } | |
3333 | rdev->ppl.size = size; | |
3334 | return len; | |
3335 | } | |
3336 | ||
3337 | static struct rdev_sysfs_entry rdev_ppl_size = | |
3338 | __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store); | |
3339 | ||
86e6ffdd N |
3340 | static struct attribute *rdev_default_attrs[] = { |
3341 | &rdev_state.attr, | |
4dbcdc75 | 3342 | &rdev_errors.attr, |
014236d2 | 3343 | &rdev_slot.attr, |
93c8cad0 | 3344 | &rdev_offset.attr, |
c6563a8c | 3345 | &rdev_new_offset.attr, |
83303b61 | 3346 | &rdev_size.attr, |
06e3c817 | 3347 | &rdev_recovery_start.attr, |
16c791a5 N |
3348 | &rdev_bad_blocks.attr, |
3349 | &rdev_unack_bad_blocks.attr, | |
664aed04 AP |
3350 | &rdev_ppl_sector.attr, |
3351 | &rdev_ppl_size.attr, | |
86e6ffdd N |
3352 | NULL, |
3353 | }; | |
3354 | static ssize_t | |
3355 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3356 | { | |
3357 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3358 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
3359 | |
3360 | if (!entry->show) | |
3361 | return -EIO; | |
758bfc8a N |
3362 | if (!rdev->mddev) |
3363 | return -EBUSY; | |
3364 | return entry->show(rdev, page); | |
86e6ffdd N |
3365 | } |
3366 | ||
3367 | static ssize_t | |
3368 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3369 | const char *page, size_t length) | |
3370 | { | |
3371 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3372 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3373 | ssize_t rv; |
fd01b88c | 3374 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3375 | |
3376 | if (!entry->store) | |
3377 | return -EIO; | |
67463acb N |
3378 | if (!capable(CAP_SYS_ADMIN)) |
3379 | return -EACCES; | |
27c529bb | 3380 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3381 | if (!rv) { |
27c529bb N |
3382 | if (rdev->mddev == NULL) |
3383 | rv = -EBUSY; | |
3384 | else | |
3385 | rv = entry->store(rdev, page, length); | |
6a51830e | 3386 | mddev_unlock(mddev); |
ca388059 N |
3387 | } |
3388 | return rv; | |
86e6ffdd N |
3389 | } |
3390 | ||
3391 | static void rdev_free(struct kobject *ko) | |
3392 | { | |
3cb03002 | 3393 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3394 | kfree(rdev); |
3395 | } | |
52cf25d0 | 3396 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3397 | .show = rdev_attr_show, |
3398 | .store = rdev_attr_store, | |
3399 | }; | |
3400 | static struct kobj_type rdev_ktype = { | |
3401 | .release = rdev_free, | |
3402 | .sysfs_ops = &rdev_sysfs_ops, | |
3403 | .default_attrs = rdev_default_attrs, | |
3404 | }; | |
3405 | ||
3cb03002 | 3406 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3407 | { |
3408 | rdev->desc_nr = -1; | |
3409 | rdev->saved_raid_disk = -1; | |
3410 | rdev->raid_disk = -1; | |
3411 | rdev->flags = 0; | |
3412 | rdev->data_offset = 0; | |
c6563a8c | 3413 | rdev->new_data_offset = 0; |
e8bb9a83 | 3414 | rdev->sb_events = 0; |
0e3ef49e | 3415 | rdev->last_read_error = 0; |
2699b672 N |
3416 | rdev->sb_loaded = 0; |
3417 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3418 | atomic_set(&rdev->nr_pending, 0); |
3419 | atomic_set(&rdev->read_errors, 0); | |
3420 | atomic_set(&rdev->corrected_errors, 0); | |
3421 | ||
3422 | INIT_LIST_HEAD(&rdev->same_set); | |
3423 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3424 | |
3425 | /* Add space to store bad block list. | |
3426 | * This reserves the space even on arrays where it cannot | |
3427 | * be used - I wonder if that matters | |
3428 | */ | |
fc974ee2 | 3429 | return badblocks_init(&rdev->badblocks, 0); |
e8bb9a83 N |
3430 | } |
3431 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3432 | /* |
3433 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3434 | * | |
3435 | * mark the device faulty if: | |
3436 | * | |
3437 | * - the device is nonexistent (zero size) | |
3438 | * - the device has no valid superblock | |
3439 | * | |
3440 | * a faulty rdev _never_ has rdev->sb set. | |
3441 | */ | |
3cb03002 | 3442 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3443 | { |
3444 | char b[BDEVNAME_SIZE]; | |
3445 | int err; | |
3cb03002 | 3446 | struct md_rdev *rdev; |
1da177e4 LT |
3447 | sector_t size; |
3448 | ||
9ffae0cf | 3449 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
9d48739e | 3450 | if (!rdev) |
1da177e4 | 3451 | return ERR_PTR(-ENOMEM); |
1da177e4 | 3452 | |
2230dfe4 N |
3453 | err = md_rdev_init(rdev); |
3454 | if (err) | |
3455 | goto abort_free; | |
3456 | err = alloc_disk_sb(rdev); | |
3457 | if (err) | |
1da177e4 LT |
3458 | goto abort_free; |
3459 | ||
c5d79adb | 3460 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3461 | if (err) |
3462 | goto abort_free; | |
3463 | ||
f9cb074b | 3464 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3465 | |
77304d2a | 3466 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3467 | if (!size) { |
9d48739e | 3468 | pr_warn("md: %s has zero or unknown size, marking faulty!\n", |
1da177e4 LT |
3469 | bdevname(rdev->bdev,b)); |
3470 | err = -EINVAL; | |
3471 | goto abort_free; | |
3472 | } | |
3473 | ||
3474 | if (super_format >= 0) { | |
3475 | err = super_types[super_format]. | |
3476 | load_super(rdev, NULL, super_minor); | |
3477 | if (err == -EINVAL) { | |
9d48739e | 3478 | pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n", |
df968c4e | 3479 | bdevname(rdev->bdev,b), |
9d48739e | 3480 | super_format, super_minor); |
1da177e4 LT |
3481 | goto abort_free; |
3482 | } | |
3483 | if (err < 0) { | |
9d48739e | 3484 | pr_warn("md: could not read %s's sb, not importing!\n", |
1da177e4 LT |
3485 | bdevname(rdev->bdev,b)); |
3486 | goto abort_free; | |
3487 | } | |
3488 | } | |
6bfe0b49 | 3489 | |
1da177e4 LT |
3490 | return rdev; |
3491 | ||
3492 | abort_free: | |
2699b672 N |
3493 | if (rdev->bdev) |
3494 | unlock_rdev(rdev); | |
545c8795 | 3495 | md_rdev_clear(rdev); |
1da177e4 LT |
3496 | kfree(rdev); |
3497 | return ERR_PTR(err); | |
3498 | } | |
3499 | ||
3500 | /* | |
3501 | * Check a full RAID array for plausibility | |
3502 | */ | |
3503 | ||
f72ffdd6 | 3504 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3505 | { |
3506 | int i; | |
3cb03002 | 3507 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3508 | char b[BDEVNAME_SIZE]; |
3509 | ||
3510 | freshest = NULL; | |
dafb20fa | 3511 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3512 | switch (super_types[mddev->major_version]. |
3513 | load_super(rdev, freshest, mddev->minor_version)) { | |
3514 | case 1: | |
3515 | freshest = rdev; | |
3516 | break; | |
3517 | case 0: | |
3518 | break; | |
3519 | default: | |
9d48739e | 3520 | pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n", |
1da177e4 | 3521 | bdevname(rdev->bdev,b)); |
fb56dfef | 3522 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3523 | } |
3524 | ||
1da177e4 LT |
3525 | super_types[mddev->major_version]. |
3526 | validate_super(mddev, freshest); | |
3527 | ||
3528 | i = 0; | |
dafb20fa | 3529 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3530 | if (mddev->max_disks && |
3531 | (rdev->desc_nr >= mddev->max_disks || | |
3532 | i > mddev->max_disks)) { | |
9d48739e N |
3533 | pr_warn("md: %s: %s: only %d devices permitted\n", |
3534 | mdname(mddev), bdevname(rdev->bdev, b), | |
3535 | mddev->max_disks); | |
fb56dfef | 3536 | md_kick_rdev_from_array(rdev); |
de01dfad N |
3537 | continue; |
3538 | } | |
1aee41f6 | 3539 | if (rdev != freshest) { |
1da177e4 LT |
3540 | if (super_types[mddev->major_version]. |
3541 | validate_super(mddev, rdev)) { | |
9d48739e | 3542 | pr_warn("md: kicking non-fresh %s from array!\n", |
1da177e4 | 3543 | bdevname(rdev->bdev,b)); |
fb56dfef | 3544 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3545 | continue; |
3546 | } | |
1aee41f6 | 3547 | } |
1da177e4 LT |
3548 | if (mddev->level == LEVEL_MULTIPATH) { |
3549 | rdev->desc_nr = i++; | |
3550 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3551 | set_bit(In_sync, &rdev->flags); |
f2076e7d SL |
3552 | } else if (rdev->raid_disk >= |
3553 | (mddev->raid_disks - min(0, mddev->delta_disks)) && | |
3554 | !test_bit(Journal, &rdev->flags)) { | |
a778b73f N |
3555 | rdev->raid_disk = -1; |
3556 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3557 | } |
3558 | } | |
1da177e4 LT |
3559 | } |
3560 | ||
72e02075 N |
3561 | /* Read a fixed-point number. |
3562 | * Numbers in sysfs attributes should be in "standard" units where | |
3563 | * possible, so time should be in seconds. | |
f72ffdd6 | 3564 | * However we internally use a a much smaller unit such as |
72e02075 N |
3565 | * milliseconds or jiffies. |
3566 | * This function takes a decimal number with a possible fractional | |
3567 | * component, and produces an integer which is the result of | |
3568 | * multiplying that number by 10^'scale'. | |
3569 | * all without any floating-point arithmetic. | |
3570 | */ | |
3571 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3572 | { | |
3573 | unsigned long result = 0; | |
3574 | long decimals = -1; | |
3575 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3576 | if (*cp == '.') | |
3577 | decimals = 0; | |
3578 | else if (decimals < scale) { | |
3579 | unsigned int value; | |
3580 | value = *cp - '0'; | |
3581 | result = result * 10 + value; | |
3582 | if (decimals >= 0) | |
3583 | decimals++; | |
3584 | } | |
3585 | cp++; | |
3586 | } | |
3587 | if (*cp == '\n') | |
3588 | cp++; | |
3589 | if (*cp) | |
3590 | return -EINVAL; | |
3591 | if (decimals < 0) | |
3592 | decimals = 0; | |
3593 | while (decimals < scale) { | |
3594 | result *= 10; | |
3595 | decimals ++; | |
3596 | } | |
3597 | *res = result; | |
3598 | return 0; | |
3599 | } | |
3600 | ||
16f17b39 | 3601 | static ssize_t |
fd01b88c | 3602 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3603 | { |
3604 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3605 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3606 | } | |
3607 | static ssize_t | |
fd01b88c | 3608 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3609 | { |
16f17b39 | 3610 | unsigned long msec; |
97ce0a7f | 3611 | |
28c1b9fd | 3612 | if (mddev_is_clustered(mddev)) { |
9d48739e | 3613 | pr_warn("md: Safemode is disabled for clustered mode\n"); |
28c1b9fd GR |
3614 | return -EINVAL; |
3615 | } | |
3616 | ||
72e02075 | 3617 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3618 | return -EINVAL; |
16f17b39 N |
3619 | if (msec == 0) |
3620 | mddev->safemode_delay = 0; | |
3621 | else { | |
19052c0e | 3622 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3623 | unsigned long new_delay = (msec*HZ)/1000; |
3624 | ||
3625 | if (new_delay == 0) | |
3626 | new_delay = 1; | |
3627 | mddev->safemode_delay = new_delay; | |
3628 | if (new_delay < old_delay || old_delay == 0) | |
3629 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3630 | } |
3631 | return len; | |
3632 | } | |
3633 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3634 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3635 | |
eae1701f | 3636 | static ssize_t |
fd01b88c | 3637 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3638 | { |
36d091f4 N |
3639 | struct md_personality *p; |
3640 | int ret; | |
3641 | spin_lock(&mddev->lock); | |
3642 | p = mddev->pers; | |
d9d166c2 | 3643 | if (p) |
36d091f4 | 3644 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3645 | else if (mddev->clevel[0]) |
36d091f4 | 3646 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3647 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3648 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3649 | else |
36d091f4 N |
3650 | ret = 0; |
3651 | spin_unlock(&mddev->lock); | |
3652 | return ret; | |
eae1701f N |
3653 | } |
3654 | ||
d9d166c2 | 3655 | static ssize_t |
fd01b88c | 3656 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3657 | { |
f2859af6 | 3658 | char clevel[16]; |
6791875e N |
3659 | ssize_t rv; |
3660 | size_t slen = len; | |
db721d32 | 3661 | struct md_personality *pers, *oldpers; |
f2859af6 | 3662 | long level; |
db721d32 | 3663 | void *priv, *oldpriv; |
3cb03002 | 3664 | struct md_rdev *rdev; |
245f46c2 | 3665 | |
6791875e N |
3666 | if (slen == 0 || slen >= sizeof(clevel)) |
3667 | return -EINVAL; | |
3668 | ||
3669 | rv = mddev_lock(mddev); | |
3670 | if (rv) | |
3671 | return rv; | |
3672 | ||
245f46c2 | 3673 | if (mddev->pers == NULL) { |
6791875e N |
3674 | strncpy(mddev->clevel, buf, slen); |
3675 | if (mddev->clevel[slen-1] == '\n') | |
3676 | slen--; | |
3677 | mddev->clevel[slen] = 0; | |
245f46c2 | 3678 | mddev->level = LEVEL_NONE; |
6791875e N |
3679 | rv = len; |
3680 | goto out_unlock; | |
245f46c2 | 3681 | } |
6791875e | 3682 | rv = -EROFS; |
bd8839e0 | 3683 | if (mddev->ro) |
6791875e | 3684 | goto out_unlock; |
245f46c2 N |
3685 | |
3686 | /* request to change the personality. Need to ensure: | |
3687 | * - array is not engaged in resync/recovery/reshape | |
3688 | * - old personality can be suspended | |
3689 | * - new personality will access other array. | |
3690 | */ | |
3691 | ||
6791875e | 3692 | rv = -EBUSY; |
bb4f1e9d | 3693 | if (mddev->sync_thread || |
f851b60d | 3694 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3695 | mddev->reshape_position != MaxSector || |
3696 | mddev->sysfs_active) | |
6791875e | 3697 | goto out_unlock; |
245f46c2 | 3698 | |
6791875e | 3699 | rv = -EINVAL; |
245f46c2 | 3700 | if (!mddev->pers->quiesce) { |
9d48739e N |
3701 | pr_warn("md: %s: %s does not support online personality change\n", |
3702 | mdname(mddev), mddev->pers->name); | |
6791875e | 3703 | goto out_unlock; |
245f46c2 N |
3704 | } |
3705 | ||
3706 | /* Now find the new personality */ | |
6791875e N |
3707 | strncpy(clevel, buf, slen); |
3708 | if (clevel[slen-1] == '\n') | |
3709 | slen--; | |
3710 | clevel[slen] = 0; | |
b29bebd6 | 3711 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3712 | level = LEVEL_NONE; |
245f46c2 | 3713 | |
f2859af6 DW |
3714 | if (request_module("md-%s", clevel) != 0) |
3715 | request_module("md-level-%s", clevel); | |
245f46c2 | 3716 | spin_lock(&pers_lock); |
f2859af6 | 3717 | pers = find_pers(level, clevel); |
245f46c2 N |
3718 | if (!pers || !try_module_get(pers->owner)) { |
3719 | spin_unlock(&pers_lock); | |
9d48739e | 3720 | pr_warn("md: personality %s not loaded\n", clevel); |
6791875e N |
3721 | rv = -EINVAL; |
3722 | goto out_unlock; | |
245f46c2 N |
3723 | } |
3724 | spin_unlock(&pers_lock); | |
3725 | ||
3726 | if (pers == mddev->pers) { | |
3727 | /* Nothing to do! */ | |
3728 | module_put(pers->owner); | |
6791875e N |
3729 | rv = len; |
3730 | goto out_unlock; | |
245f46c2 N |
3731 | } |
3732 | if (!pers->takeover) { | |
3733 | module_put(pers->owner); | |
9d48739e N |
3734 | pr_warn("md: %s: %s does not support personality takeover\n", |
3735 | mdname(mddev), clevel); | |
6791875e N |
3736 | rv = -EINVAL; |
3737 | goto out_unlock; | |
245f46c2 N |
3738 | } |
3739 | ||
dafb20fa | 3740 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3741 | rdev->new_raid_disk = rdev->raid_disk; |
3742 | ||
245f46c2 N |
3743 | /* ->takeover must set new_* and/or delta_disks |
3744 | * if it succeeds, and may set them when it fails. | |
3745 | */ | |
3746 | priv = pers->takeover(mddev); | |
3747 | if (IS_ERR(priv)) { | |
3748 | mddev->new_level = mddev->level; | |
3749 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3750 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3751 | mddev->raid_disks -= mddev->delta_disks; |
3752 | mddev->delta_disks = 0; | |
2c810cdd | 3753 | mddev->reshape_backwards = 0; |
245f46c2 | 3754 | module_put(pers->owner); |
9d48739e N |
3755 | pr_warn("md: %s: %s would not accept array\n", |
3756 | mdname(mddev), clevel); | |
6791875e N |
3757 | rv = PTR_ERR(priv); |
3758 | goto out_unlock; | |
245f46c2 N |
3759 | } |
3760 | ||
3761 | /* Looks like we have a winner */ | |
3762 | mddev_suspend(mddev); | |
5aa61f42 | 3763 | mddev_detach(mddev); |
36d091f4 N |
3764 | |
3765 | spin_lock(&mddev->lock); | |
db721d32 N |
3766 | oldpers = mddev->pers; |
3767 | oldpriv = mddev->private; | |
3768 | mddev->pers = pers; | |
3769 | mddev->private = priv; | |
3770 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3771 | mddev->level = mddev->new_level; | |
3772 | mddev->layout = mddev->new_layout; | |
3773 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3774 | mddev->delta_disks = 0; | |
3775 | mddev->reshape_backwards = 0; | |
3776 | mddev->degraded = 0; | |
36d091f4 | 3777 | spin_unlock(&mddev->lock); |
db721d32 N |
3778 | |
3779 | if (oldpers->sync_request == NULL && | |
3780 | mddev->external) { | |
3781 | /* We are converting from a no-redundancy array | |
3782 | * to a redundancy array and metadata is managed | |
3783 | * externally so we need to be sure that writes | |
3784 | * won't block due to a need to transition | |
3785 | * clean->dirty | |
3786 | * until external management is started. | |
3787 | */ | |
3788 | mddev->in_sync = 0; | |
3789 | mddev->safemode_delay = 0; | |
3790 | mddev->safemode = 0; | |
3791 | } | |
f72ffdd6 | 3792 | |
db721d32 N |
3793 | oldpers->free(mddev, oldpriv); |
3794 | ||
3795 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3796 | pers->sync_request != NULL) { |
3797 | /* need to add the md_redundancy_group */ | |
3798 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
3799 | pr_warn("md: cannot register extra attributes for %s\n", |
3800 | mdname(mddev)); | |
388975cc | 3801 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3802 | } |
db721d32 | 3803 | if (oldpers->sync_request != NULL && |
a64c876f N |
3804 | pers->sync_request == NULL) { |
3805 | /* need to remove the md_redundancy_group */ | |
3806 | if (mddev->to_remove == NULL) | |
3807 | mddev->to_remove = &md_redundancy_group; | |
3808 | } | |
3809 | ||
4cb9da7d AO |
3810 | module_put(oldpers->owner); |
3811 | ||
dafb20fa | 3812 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3813 | if (rdev->raid_disk < 0) |
3814 | continue; | |
bf2cb0da | 3815 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3816 | rdev->new_raid_disk = -1; |
3817 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3818 | continue; | |
36fad858 | 3819 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3820 | } |
dafb20fa | 3821 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3822 | if (rdev->raid_disk < 0) |
3823 | continue; | |
3824 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3825 | continue; | |
3826 | rdev->raid_disk = rdev->new_raid_disk; | |
3827 | if (rdev->raid_disk < 0) | |
3a981b03 | 3828 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3829 | else { |
36fad858 | 3830 | if (sysfs_link_rdev(mddev, rdev)) |
9d48739e N |
3831 | pr_warn("md: cannot register rd%d for %s after level change\n", |
3832 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3833 | } |
e93f68a1 N |
3834 | } |
3835 | ||
db721d32 | 3836 | if (pers->sync_request == NULL) { |
9af204cf TM |
3837 | /* this is now an array without redundancy, so |
3838 | * it must always be in_sync | |
3839 | */ | |
3840 | mddev->in_sync = 1; | |
3841 | del_timer_sync(&mddev->safemode_timer); | |
3842 | } | |
02e5f5c0 | 3843 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3844 | pers->run(mddev); |
2953079c | 3845 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
47525e59 | 3846 | mddev_resume(mddev); |
830778a1 N |
3847 | if (!mddev->thread) |
3848 | md_update_sb(mddev, 1); | |
5cac7861 | 3849 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3850 | md_new_event(mddev); |
6791875e N |
3851 | rv = len; |
3852 | out_unlock: | |
3853 | mddev_unlock(mddev); | |
d9d166c2 N |
3854 | return rv; |
3855 | } | |
3856 | ||
3857 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3858 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3859 | |
d4dbd025 | 3860 | static ssize_t |
fd01b88c | 3861 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3862 | { |
3863 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3864 | if (mddev->reshape_position != MaxSector && |
3865 | mddev->layout != mddev->new_layout) | |
3866 | return sprintf(page, "%d (%d)\n", | |
3867 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3868 | return sprintf(page, "%d\n", mddev->layout); |
3869 | } | |
3870 | ||
3871 | static ssize_t | |
fd01b88c | 3872 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 | 3873 | { |
4c9309c0 | 3874 | unsigned int n; |
6791875e | 3875 | int err; |
d4dbd025 | 3876 | |
4c9309c0 AD |
3877 | err = kstrtouint(buf, 10, &n); |
3878 | if (err < 0) | |
3879 | return err; | |
6791875e N |
3880 | err = mddev_lock(mddev); |
3881 | if (err) | |
3882 | return err; | |
d4dbd025 | 3883 | |
b3546035 | 3884 | if (mddev->pers) { |
50ac168a | 3885 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3886 | err = -EBUSY; |
3887 | else if (mddev->ro) | |
3888 | err = -EROFS; | |
3889 | else { | |
3890 | mddev->new_layout = n; | |
3891 | err = mddev->pers->check_reshape(mddev); | |
3892 | if (err) | |
3893 | mddev->new_layout = mddev->layout; | |
597a711b | 3894 | } |
b3546035 | 3895 | } else { |
08a02ecd | 3896 | mddev->new_layout = n; |
b3546035 N |
3897 | if (mddev->reshape_position == MaxSector) |
3898 | mddev->layout = n; | |
3899 | } | |
6791875e N |
3900 | mddev_unlock(mddev); |
3901 | return err ?: len; | |
d4dbd025 N |
3902 | } |
3903 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3904 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3905 | |
eae1701f | 3906 | static ssize_t |
fd01b88c | 3907 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3908 | { |
bb636547 N |
3909 | if (mddev->raid_disks == 0) |
3910 | return 0; | |
08a02ecd N |
3911 | if (mddev->reshape_position != MaxSector && |
3912 | mddev->delta_disks != 0) | |
3913 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3914 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3915 | return sprintf(page, "%d\n", mddev->raid_disks); |
3916 | } | |
3917 | ||
fd01b88c | 3918 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3919 | |
3920 | static ssize_t | |
fd01b88c | 3921 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3922 | { |
4c9309c0 | 3923 | unsigned int n; |
6791875e | 3924 | int err; |
da943b99 | 3925 | |
4c9309c0 AD |
3926 | err = kstrtouint(buf, 10, &n); |
3927 | if (err < 0) | |
3928 | return err; | |
da943b99 | 3929 | |
6791875e N |
3930 | err = mddev_lock(mddev); |
3931 | if (err) | |
3932 | return err; | |
da943b99 | 3933 | if (mddev->pers) |
6791875e | 3934 | err = update_raid_disks(mddev, n); |
08a02ecd | 3935 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3936 | struct md_rdev *rdev; |
08a02ecd | 3937 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3938 | |
6791875e | 3939 | err = -EINVAL; |
c6563a8c N |
3940 | rdev_for_each(rdev, mddev) { |
3941 | if (olddisks < n && | |
3942 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3943 | goto out_unlock; |
c6563a8c N |
3944 | if (olddisks > n && |
3945 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3946 | goto out_unlock; |
c6563a8c | 3947 | } |
6791875e | 3948 | err = 0; |
08a02ecd N |
3949 | mddev->delta_disks = n - olddisks; |
3950 | mddev->raid_disks = n; | |
2c810cdd | 3951 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3952 | } else |
da943b99 | 3953 | mddev->raid_disks = n; |
6791875e N |
3954 | out_unlock: |
3955 | mddev_unlock(mddev); | |
3956 | return err ? err : len; | |
da943b99 N |
3957 | } |
3958 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3959 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3960 | |
3b34380a | 3961 | static ssize_t |
fd01b88c | 3962 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3963 | { |
08a02ecd | 3964 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3965 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3966 | return sprintf(page, "%d (%d)\n", | |
3967 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3968 | mddev->chunk_sectors << 9); |
3969 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3970 | } |
3971 | ||
3972 | static ssize_t | |
fd01b88c | 3973 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3974 | { |
4c9309c0 | 3975 | unsigned long n; |
6791875e | 3976 | int err; |
3b34380a | 3977 | |
4c9309c0 AD |
3978 | err = kstrtoul(buf, 10, &n); |
3979 | if (err < 0) | |
3980 | return err; | |
3b34380a | 3981 | |
6791875e N |
3982 | err = mddev_lock(mddev); |
3983 | if (err) | |
3984 | return err; | |
b3546035 | 3985 | if (mddev->pers) { |
50ac168a | 3986 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3987 | err = -EBUSY; |
3988 | else if (mddev->ro) | |
3989 | err = -EROFS; | |
3990 | else { | |
3991 | mddev->new_chunk_sectors = n >> 9; | |
3992 | err = mddev->pers->check_reshape(mddev); | |
3993 | if (err) | |
3994 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 3995 | } |
b3546035 | 3996 | } else { |
664e7c41 | 3997 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3998 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3999 | mddev->chunk_sectors = n >> 9; |
b3546035 | 4000 | } |
6791875e N |
4001 | mddev_unlock(mddev); |
4002 | return err ?: len; | |
3b34380a N |
4003 | } |
4004 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 4005 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 4006 | |
a94213b1 | 4007 | static ssize_t |
fd01b88c | 4008 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 4009 | { |
d1a7c503 N |
4010 | if (mddev->recovery_cp == MaxSector) |
4011 | return sprintf(page, "none\n"); | |
a94213b1 N |
4012 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
4013 | } | |
4014 | ||
4015 | static ssize_t | |
fd01b88c | 4016 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 4017 | { |
4c9309c0 | 4018 | unsigned long long n; |
6791875e | 4019 | int err; |
4c9309c0 AD |
4020 | |
4021 | if (cmd_match(buf, "none")) | |
4022 | n = MaxSector; | |
4023 | else { | |
4024 | err = kstrtoull(buf, 10, &n); | |
4025 | if (err < 0) | |
4026 | return err; | |
4027 | if (n != (sector_t)n) | |
4028 | return -EINVAL; | |
4029 | } | |
a94213b1 | 4030 | |
6791875e N |
4031 | err = mddev_lock(mddev); |
4032 | if (err) | |
4033 | return err; | |
b098636c | 4034 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e | 4035 | err = -EBUSY; |
a94213b1 | 4036 | |
6791875e N |
4037 | if (!err) { |
4038 | mddev->recovery_cp = n; | |
4039 | if (mddev->pers) | |
2953079c | 4040 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
6791875e N |
4041 | } |
4042 | mddev_unlock(mddev); | |
4043 | return err ?: len; | |
a94213b1 N |
4044 | } |
4045 | static struct md_sysfs_entry md_resync_start = | |
750f199e N |
4046 | __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR, |
4047 | resync_start_show, resync_start_store); | |
a94213b1 | 4048 | |
9e653b63 N |
4049 | /* |
4050 | * The array state can be: | |
4051 | * | |
4052 | * clear | |
4053 | * No devices, no size, no level | |
4054 | * Equivalent to STOP_ARRAY ioctl | |
4055 | * inactive | |
4056 | * May have some settings, but array is not active | |
4057 | * all IO results in error | |
4058 | * When written, doesn't tear down array, but just stops it | |
4059 | * suspended (not supported yet) | |
4060 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 4061 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
4062 | * readonly |
4063 | * no resync can happen. no superblocks get written. | |
4064 | * write requests fail | |
4065 | * read-auto | |
4066 | * like readonly, but behaves like 'clean' on a write request. | |
4067 | * | |
4068 | * clean - no pending writes, but otherwise active. | |
4069 | * When written to inactive array, starts without resync | |
4070 | * If a write request arrives then | |
4071 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
4072 | * if not known, block and switch to write-pending | |
4073 | * If written to an active array that has pending writes, then fails. | |
4074 | * active | |
4075 | * fully active: IO and resync can be happening. | |
4076 | * When written to inactive array, starts with resync | |
4077 | * | |
4078 | * write-pending | |
4079 | * clean, but writes are blocked waiting for 'active' to be written. | |
4080 | * | |
4081 | * active-idle | |
4082 | * like active, but no writes have been seen for a while (100msec). | |
4083 | * | |
4084 | */ | |
4085 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
4086 | write_pending, active_idle, bad_word}; | |
05381954 | 4087 | static char *array_states[] = { |
9e653b63 N |
4088 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
4089 | "write-pending", "active-idle", NULL }; | |
4090 | ||
4091 | static int match_word(const char *word, char **list) | |
4092 | { | |
4093 | int n; | |
4094 | for (n=0; list[n]; n++) | |
4095 | if (cmd_match(word, list[n])) | |
4096 | break; | |
4097 | return n; | |
4098 | } | |
4099 | ||
4100 | static ssize_t | |
fd01b88c | 4101 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
4102 | { |
4103 | enum array_state st = inactive; | |
4104 | ||
4105 | if (mddev->pers) | |
4106 | switch(mddev->ro) { | |
4107 | case 1: | |
4108 | st = readonly; | |
4109 | break; | |
4110 | case 2: | |
4111 | st = read_auto; | |
4112 | break; | |
4113 | case 0: | |
55cc39f3 | 4114 | spin_lock(&mddev->lock); |
2953079c | 4115 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
e691063a | 4116 | st = write_pending; |
16f88949 TM |
4117 | else if (mddev->in_sync) |
4118 | st = clean; | |
9e653b63 N |
4119 | else if (mddev->safemode) |
4120 | st = active_idle; | |
4121 | else | |
4122 | st = active; | |
55cc39f3 | 4123 | spin_unlock(&mddev->lock); |
9e653b63 N |
4124 | } |
4125 | else { | |
4126 | if (list_empty(&mddev->disks) && | |
4127 | mddev->raid_disks == 0 && | |
58c0fed4 | 4128 | mddev->dev_sectors == 0) |
9e653b63 N |
4129 | st = clear; |
4130 | else | |
4131 | st = inactive; | |
4132 | } | |
4133 | return sprintf(page, "%s\n", array_states[st]); | |
4134 | } | |
4135 | ||
f72ffdd6 N |
4136 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
4137 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
4138 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 4139 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
4140 | |
4141 | static ssize_t | |
fd01b88c | 4142 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 4143 | { |
6497709b | 4144 | int err = 0; |
9e653b63 | 4145 | enum array_state st = match_word(buf, array_states); |
6791875e N |
4146 | |
4147 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
4148 | /* don't take reconfig_mutex when toggling between | |
4149 | * clean and active | |
4150 | */ | |
4151 | spin_lock(&mddev->lock); | |
4152 | if (st == active) { | |
4153 | restart_array(mddev); | |
2953079c | 4154 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
91a6c4ad | 4155 | md_wakeup_thread(mddev->thread); |
6791875e | 4156 | wake_up(&mddev->sb_wait); |
6791875e N |
4157 | } else /* st == clean */ { |
4158 | restart_array(mddev); | |
6497709b | 4159 | if (!set_in_sync(mddev)) |
6791875e N |
4160 | err = -EBUSY; |
4161 | } | |
573275b5 TM |
4162 | if (!err) |
4163 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
6791875e | 4164 | spin_unlock(&mddev->lock); |
c008f1d3 | 4165 | return err ?: len; |
6791875e N |
4166 | } |
4167 | err = mddev_lock(mddev); | |
4168 | if (err) | |
4169 | return err; | |
4170 | err = -EINVAL; | |
9e653b63 N |
4171 | switch(st) { |
4172 | case bad_word: | |
4173 | break; | |
4174 | case clear: | |
4175 | /* stopping an active array */ | |
a05b7ea0 | 4176 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
4177 | break; |
4178 | case inactive: | |
4179 | /* stopping an active array */ | |
90cf195d | 4180 | if (mddev->pers) |
a05b7ea0 | 4181 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 4182 | else |
e691063a | 4183 | err = 0; /* already inactive */ |
9e653b63 N |
4184 | break; |
4185 | case suspended: | |
4186 | break; /* not supported yet */ | |
4187 | case readonly: | |
4188 | if (mddev->pers) | |
a05b7ea0 | 4189 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
4190 | else { |
4191 | mddev->ro = 1; | |
648b629e | 4192 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
4193 | err = do_md_run(mddev); |
4194 | } | |
4195 | break; | |
4196 | case read_auto: | |
9e653b63 | 4197 | if (mddev->pers) { |
80268ee9 | 4198 | if (mddev->ro == 0) |
a05b7ea0 | 4199 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 4200 | else if (mddev->ro == 1) |
648b629e N |
4201 | err = restart_array(mddev); |
4202 | if (err == 0) { | |
4203 | mddev->ro = 2; | |
4204 | set_disk_ro(mddev->gendisk, 0); | |
4205 | } | |
9e653b63 N |
4206 | } else { |
4207 | mddev->ro = 2; | |
4208 | err = do_md_run(mddev); | |
4209 | } | |
4210 | break; | |
4211 | case clean: | |
4212 | if (mddev->pers) { | |
339421de SL |
4213 | err = restart_array(mddev); |
4214 | if (err) | |
4215 | break; | |
85572d7c | 4216 | spin_lock(&mddev->lock); |
6497709b | 4217 | if (!set_in_sync(mddev)) |
e691063a | 4218 | err = -EBUSY; |
85572d7c | 4219 | spin_unlock(&mddev->lock); |
5bf29597 N |
4220 | } else |
4221 | err = -EINVAL; | |
9e653b63 N |
4222 | break; |
4223 | case active: | |
4224 | if (mddev->pers) { | |
339421de SL |
4225 | err = restart_array(mddev); |
4226 | if (err) | |
4227 | break; | |
2953079c | 4228 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
9e653b63 N |
4229 | wake_up(&mddev->sb_wait); |
4230 | err = 0; | |
4231 | } else { | |
4232 | mddev->ro = 0; | |
648b629e | 4233 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
4234 | err = do_md_run(mddev); |
4235 | } | |
4236 | break; | |
4237 | case write_pending: | |
4238 | case active_idle: | |
4239 | /* these cannot be set */ | |
4240 | break; | |
4241 | } | |
6791875e N |
4242 | |
4243 | if (!err) { | |
1d23f178 N |
4244 | if (mddev->hold_active == UNTIL_IOCTL) |
4245 | mddev->hold_active = 0; | |
00bcb4ac | 4246 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4247 | } |
6791875e N |
4248 | mddev_unlock(mddev); |
4249 | return err ?: len; | |
9e653b63 | 4250 | } |
80ca3a44 | 4251 | static struct md_sysfs_entry md_array_state = |
750f199e | 4252 | __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); |
9e653b63 | 4253 | |
1e50915f | 4254 | static ssize_t |
fd01b88c | 4255 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
4256 | return sprintf(page, "%d\n", |
4257 | atomic_read(&mddev->max_corr_read_errors)); | |
4258 | } | |
4259 | ||
4260 | static ssize_t | |
fd01b88c | 4261 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f | 4262 | { |
4c9309c0 AD |
4263 | unsigned int n; |
4264 | int rv; | |
1e50915f | 4265 | |
4c9309c0 AD |
4266 | rv = kstrtouint(buf, 10, &n); |
4267 | if (rv < 0) | |
4268 | return rv; | |
4269 | atomic_set(&mddev->max_corr_read_errors, n); | |
4270 | return len; | |
1e50915f RB |
4271 | } |
4272 | ||
4273 | static struct md_sysfs_entry max_corr_read_errors = | |
4274 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
4275 | max_corrected_read_errors_store); | |
4276 | ||
6d7ff738 | 4277 | static ssize_t |
fd01b88c | 4278 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
4279 | { |
4280 | return -EINVAL; | |
4281 | } | |
4282 | ||
4283 | static ssize_t | |
fd01b88c | 4284 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
4285 | { |
4286 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
4287 | /* The new device is added to the array. | |
4288 | * If the array has a persistent superblock, we read the | |
4289 | * superblock to initialise info and check validity. | |
4290 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
4291 | * which mainly checks size. | |
4292 | */ | |
4293 | char *e; | |
4294 | int major = simple_strtoul(buf, &e, 10); | |
4295 | int minor; | |
4296 | dev_t dev; | |
3cb03002 | 4297 | struct md_rdev *rdev; |
6d7ff738 N |
4298 | int err; |
4299 | ||
4300 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
4301 | return -EINVAL; | |
4302 | minor = simple_strtoul(e+1, &e, 10); | |
4303 | if (*e && *e != '\n') | |
4304 | return -EINVAL; | |
4305 | dev = MKDEV(major, minor); | |
4306 | if (major != MAJOR(dev) || | |
4307 | minor != MINOR(dev)) | |
4308 | return -EOVERFLOW; | |
4309 | ||
6791875e N |
4310 | flush_workqueue(md_misc_wq); |
4311 | ||
4312 | err = mddev_lock(mddev); | |
4313 | if (err) | |
4314 | return err; | |
6d7ff738 N |
4315 | if (mddev->persistent) { |
4316 | rdev = md_import_device(dev, mddev->major_version, | |
4317 | mddev->minor_version); | |
4318 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4319 | struct md_rdev *rdev0 |
4320 | = list_entry(mddev->disks.next, | |
4321 | struct md_rdev, same_set); | |
6d7ff738 N |
4322 | err = super_types[mddev->major_version] |
4323 | .load_super(rdev, rdev0, mddev->minor_version); | |
4324 | if (err < 0) | |
4325 | goto out; | |
4326 | } | |
c5d79adb N |
4327 | } else if (mddev->external) |
4328 | rdev = md_import_device(dev, -2, -1); | |
4329 | else | |
6d7ff738 N |
4330 | rdev = md_import_device(dev, -1, -1); |
4331 | ||
9a8c0fa8 N |
4332 | if (IS_ERR(rdev)) { |
4333 | mddev_unlock(mddev); | |
6d7ff738 | 4334 | return PTR_ERR(rdev); |
9a8c0fa8 | 4335 | } |
6d7ff738 N |
4336 | err = bind_rdev_to_array(rdev, mddev); |
4337 | out: | |
4338 | if (err) | |
4339 | export_rdev(rdev); | |
6791875e | 4340 | mddev_unlock(mddev); |
5492c46e AO |
4341 | if (!err) |
4342 | md_new_event(mddev); | |
6d7ff738 N |
4343 | return err ? err : len; |
4344 | } | |
4345 | ||
4346 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4347 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4348 | |
9b1d1dac | 4349 | static ssize_t |
fd01b88c | 4350 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4351 | { |
4352 | char *end; | |
4353 | unsigned long chunk, end_chunk; | |
6791875e | 4354 | int err; |
9b1d1dac | 4355 | |
6791875e N |
4356 | err = mddev_lock(mddev); |
4357 | if (err) | |
4358 | return err; | |
9b1d1dac PC |
4359 | if (!mddev->bitmap) |
4360 | goto out; | |
4361 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4362 | while (*buf) { | |
4363 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4364 | if (buf == end) break; | |
4365 | if (*end == '-') { /* range */ | |
4366 | buf = end + 1; | |
4367 | end_chunk = simple_strtoul(buf, &end, 0); | |
4368 | if (buf == end) break; | |
4369 | } | |
4370 | if (*end && !isspace(*end)) break; | |
4371 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4372 | buf = skip_spaces(end); |
9b1d1dac PC |
4373 | } |
4374 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4375 | out: | |
6791875e | 4376 | mddev_unlock(mddev); |
9b1d1dac PC |
4377 | return len; |
4378 | } | |
4379 | ||
4380 | static struct md_sysfs_entry md_bitmap = | |
4381 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4382 | ||
a35b0d69 | 4383 | static ssize_t |
fd01b88c | 4384 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4385 | { |
58c0fed4 AN |
4386 | return sprintf(page, "%llu\n", |
4387 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4388 | } |
4389 | ||
fd01b88c | 4390 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4391 | |
4392 | static ssize_t | |
fd01b88c | 4393 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4394 | { |
4395 | /* If array is inactive, we can reduce the component size, but | |
4396 | * not increase it (except from 0). | |
4397 | * If array is active, we can try an on-line resize | |
4398 | */ | |
b522adcd DW |
4399 | sector_t sectors; |
4400 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4401 | |
58c0fed4 AN |
4402 | if (err < 0) |
4403 | return err; | |
6791875e N |
4404 | err = mddev_lock(mddev); |
4405 | if (err) | |
4406 | return err; | |
a35b0d69 | 4407 | if (mddev->pers) { |
58c0fed4 | 4408 | err = update_size(mddev, sectors); |
4ba1e788 XN |
4409 | if (err == 0) |
4410 | md_update_sb(mddev, 1); | |
a35b0d69 | 4411 | } else { |
58c0fed4 AN |
4412 | if (mddev->dev_sectors == 0 || |
4413 | mddev->dev_sectors > sectors) | |
4414 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4415 | else |
4416 | err = -ENOSPC; | |
4417 | } | |
6791875e | 4418 | mddev_unlock(mddev); |
a35b0d69 N |
4419 | return err ? err : len; |
4420 | } | |
4421 | ||
4422 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4423 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4424 | |
83f0d77a | 4425 | /* Metadata version. |
e691063a N |
4426 | * This is one of |
4427 | * 'none' for arrays with no metadata (good luck...) | |
4428 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4429 | * or N.M for internally known formats |
4430 | */ | |
4431 | static ssize_t | |
fd01b88c | 4432 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4433 | { |
4434 | if (mddev->persistent) | |
4435 | return sprintf(page, "%d.%d\n", | |
4436 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4437 | else if (mddev->external) |
4438 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4439 | else |
4440 | return sprintf(page, "none\n"); | |
4441 | } | |
4442 | ||
4443 | static ssize_t | |
fd01b88c | 4444 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4445 | { |
4446 | int major, minor; | |
4447 | char *e; | |
6791875e | 4448 | int err; |
ea43ddd8 N |
4449 | /* Changing the details of 'external' metadata is |
4450 | * always permitted. Otherwise there must be | |
4451 | * no devices attached to the array. | |
4452 | */ | |
6791875e N |
4453 | |
4454 | err = mddev_lock(mddev); | |
4455 | if (err) | |
4456 | return err; | |
4457 | err = -EBUSY; | |
ea43ddd8 N |
4458 | if (mddev->external && strncmp(buf, "external:", 9) == 0) |
4459 | ; | |
4460 | else if (!list_empty(&mddev->disks)) | |
6791875e | 4461 | goto out_unlock; |
8bb93aac | 4462 | |
6791875e | 4463 | err = 0; |
8bb93aac N |
4464 | if (cmd_match(buf, "none")) { |
4465 | mddev->persistent = 0; | |
e691063a N |
4466 | mddev->external = 0; |
4467 | mddev->major_version = 0; | |
4468 | mddev->minor_version = 90; | |
6791875e | 4469 | goto out_unlock; |
e691063a N |
4470 | } |
4471 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4472 | size_t namelen = len-9; |
e691063a N |
4473 | if (namelen >= sizeof(mddev->metadata_type)) |
4474 | namelen = sizeof(mddev->metadata_type)-1; | |
4475 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4476 | mddev->metadata_type[namelen] = 0; | |
4477 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4478 | mddev->metadata_type[--namelen] = 0; | |
4479 | mddev->persistent = 0; | |
4480 | mddev->external = 1; | |
8bb93aac N |
4481 | mddev->major_version = 0; |
4482 | mddev->minor_version = 90; | |
6791875e | 4483 | goto out_unlock; |
8bb93aac N |
4484 | } |
4485 | major = simple_strtoul(buf, &e, 10); | |
6791875e | 4486 | err = -EINVAL; |
8bb93aac | 4487 | if (e==buf || *e != '.') |
6791875e | 4488 | goto out_unlock; |
8bb93aac N |
4489 | buf = e+1; |
4490 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4491 | if (e==buf || (*e && *e != '\n') ) |
6791875e N |
4492 | goto out_unlock; |
4493 | err = -ENOENT; | |
50511da3 | 4494 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
6791875e | 4495 | goto out_unlock; |
8bb93aac N |
4496 | mddev->major_version = major; |
4497 | mddev->minor_version = minor; | |
4498 | mddev->persistent = 1; | |
e691063a | 4499 | mddev->external = 0; |
6791875e N |
4500 | err = 0; |
4501 | out_unlock: | |
4502 | mddev_unlock(mddev); | |
4503 | return err ?: len; | |
8bb93aac N |
4504 | } |
4505 | ||
4506 | static struct md_sysfs_entry md_metadata = | |
750f199e | 4507 | __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4508 | |
24dd469d | 4509 | static ssize_t |
fd01b88c | 4510 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4511 | { |
7eec314d | 4512 | char *type = "idle"; |
b7b17c9b N |
4513 | unsigned long recovery = mddev->recovery; |
4514 | if (test_bit(MD_RECOVERY_FROZEN, &recovery)) | |
b6a9ce68 | 4515 | type = "frozen"; |
b7b17c9b N |
4516 | else if (test_bit(MD_RECOVERY_RUNNING, &recovery) || |
4517 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) { | |
4518 | if (test_bit(MD_RECOVERY_RESHAPE, &recovery)) | |
ccfcc3c1 | 4519 | type = "reshape"; |
b7b17c9b N |
4520 | else if (test_bit(MD_RECOVERY_SYNC, &recovery)) { |
4521 | if (!test_bit(MD_RECOVERY_REQUESTED, &recovery)) | |
24dd469d | 4522 | type = "resync"; |
b7b17c9b | 4523 | else if (test_bit(MD_RECOVERY_CHECK, &recovery)) |
24dd469d N |
4524 | type = "check"; |
4525 | else | |
4526 | type = "repair"; | |
b7b17c9b | 4527 | } else if (test_bit(MD_RECOVERY_RECOVER, &recovery)) |
24dd469d | 4528 | type = "recover"; |
985ca973 N |
4529 | else if (mddev->reshape_position != MaxSector) |
4530 | type = "reshape"; | |
24dd469d N |
4531 | } |
4532 | return sprintf(page, "%s\n", type); | |
4533 | } | |
4534 | ||
4535 | static ssize_t | |
fd01b88c | 4536 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4537 | { |
7eec314d N |
4538 | if (!mddev->pers || !mddev->pers->sync_request) |
4539 | return -EINVAL; | |
4540 | ||
b6a9ce68 N |
4541 | |
4542 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
56ccc112 N |
4543 | if (cmd_match(page, "frozen")) |
4544 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4545 | else | |
4546 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
8e8e2518 N |
4547 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
4548 | mddev_lock(mddev) == 0) { | |
4549 | flush_workqueue(md_misc_wq); | |
4550 | if (mddev->sync_thread) { | |
4551 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e | 4552 | md_reap_sync_thread(mddev); |
6791875e | 4553 | } |
8e8e2518 | 4554 | mddev_unlock(mddev); |
7eec314d | 4555 | } |
312045ee | 4556 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
24dd469d | 4557 | return -EBUSY; |
72a23c21 | 4558 | else if (cmd_match(page, "resync")) |
56ccc112 | 4559 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4560 | else if (cmd_match(page, "recover")) { |
56ccc112 | 4561 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4562 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
72a23c21 | 4563 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4564 | int err; |
4565 | if (mddev->pers->start_reshape == NULL) | |
4566 | return -EINVAL; | |
6791875e N |
4567 | err = mddev_lock(mddev); |
4568 | if (!err) { | |
312045ee N |
4569 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4570 | err = -EBUSY; | |
4571 | else { | |
4572 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4573 | err = mddev->pers->start_reshape(mddev); | |
4574 | } | |
6791875e N |
4575 | mddev_unlock(mddev); |
4576 | } | |
16484bf5 N |
4577 | if (err) |
4578 | return err; | |
a99ac971 | 4579 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4580 | } else { |
bce74dac | 4581 | if (cmd_match(page, "check")) |
7eec314d | 4582 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4583 | else if (!cmd_match(page, "repair")) |
7eec314d | 4584 | return -EINVAL; |
56ccc112 | 4585 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
7eec314d N |
4586 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
4587 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4588 | } |
48c26ddc N |
4589 | if (mddev->ro == 2) { |
4590 | /* A write to sync_action is enough to justify | |
4591 | * canceling read-auto mode | |
4592 | */ | |
4593 | mddev->ro = 0; | |
4594 | md_wakeup_thread(mddev->sync_thread); | |
4595 | } | |
03c902e1 | 4596 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4597 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4598 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4599 | return len; |
4600 | } | |
4601 | ||
c4a39551 | 4602 | static struct md_sysfs_entry md_scan_mode = |
750f199e | 4603 | __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
c4a39551 JB |
4604 | |
4605 | static ssize_t | |
4606 | last_sync_action_show(struct mddev *mddev, char *page) | |
4607 | { | |
4608 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4609 | } | |
4610 | ||
4611 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4612 | ||
9d88883e | 4613 | static ssize_t |
fd01b88c | 4614 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4615 | { |
4616 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4617 | (unsigned long long) |
4618 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4619 | } |
4620 | ||
80ca3a44 | 4621 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4622 | |
88202a0c | 4623 | static ssize_t |
fd01b88c | 4624 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4625 | { |
4626 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4627 | mddev->sync_speed_min ? "local": "system"); | |
4628 | } | |
4629 | ||
4630 | static ssize_t | |
fd01b88c | 4631 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4632 | { |
4c9309c0 AD |
4633 | unsigned int min; |
4634 | int rv; | |
4635 | ||
88202a0c | 4636 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4637 | min = 0; |
4638 | } else { | |
4639 | rv = kstrtouint(buf, 10, &min); | |
4640 | if (rv < 0) | |
4641 | return rv; | |
4642 | if (min == 0) | |
4643 | return -EINVAL; | |
88202a0c | 4644 | } |
88202a0c N |
4645 | mddev->sync_speed_min = min; |
4646 | return len; | |
4647 | } | |
4648 | ||
4649 | static struct md_sysfs_entry md_sync_min = | |
4650 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4651 | ||
4652 | static ssize_t | |
fd01b88c | 4653 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4654 | { |
4655 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4656 | mddev->sync_speed_max ? "local": "system"); | |
4657 | } | |
4658 | ||
4659 | static ssize_t | |
fd01b88c | 4660 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4661 | { |
4c9309c0 AD |
4662 | unsigned int max; |
4663 | int rv; | |
4664 | ||
88202a0c | 4665 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4666 | max = 0; |
4667 | } else { | |
4668 | rv = kstrtouint(buf, 10, &max); | |
4669 | if (rv < 0) | |
4670 | return rv; | |
4671 | if (max == 0) | |
4672 | return -EINVAL; | |
88202a0c | 4673 | } |
88202a0c N |
4674 | mddev->sync_speed_max = max; |
4675 | return len; | |
4676 | } | |
4677 | ||
4678 | static struct md_sysfs_entry md_sync_max = | |
4679 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4680 | ||
d7f3d291 | 4681 | static ssize_t |
fd01b88c | 4682 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4683 | { |
4684 | return sprintf(page, "%d\n", mddev->degraded); | |
4685 | } | |
4686 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4687 | |
90b08710 | 4688 | static ssize_t |
fd01b88c | 4689 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4690 | { |
4691 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4692 | } | |
4693 | ||
4694 | static ssize_t | |
fd01b88c | 4695 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4696 | { |
4697 | long n; | |
4698 | ||
b29bebd6 | 4699 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4700 | return -EINVAL; |
4701 | ||
4702 | if (n != 0 && n != 1) | |
4703 | return -EINVAL; | |
4704 | ||
4705 | mddev->parallel_resync = n; | |
4706 | ||
4707 | if (mddev->sync_thread) | |
4708 | wake_up(&resync_wait); | |
4709 | ||
4710 | return len; | |
4711 | } | |
4712 | ||
4713 | /* force parallel resync, even with shared block devices */ | |
4714 | static struct md_sysfs_entry md_sync_force_parallel = | |
4715 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4716 | sync_force_parallel_show, sync_force_parallel_store); | |
4717 | ||
88202a0c | 4718 | static ssize_t |
fd01b88c | 4719 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4720 | { |
4721 | unsigned long resync, dt, db; | |
d1a7c503 N |
4722 | if (mddev->curr_resync == 0) |
4723 | return sprintf(page, "none\n"); | |
9687a60c AN |
4724 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4725 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4726 | if (!dt) dt++; |
9687a60c AN |
4727 | db = resync - mddev->resync_mark_cnt; |
4728 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4729 | } |
4730 | ||
80ca3a44 | 4731 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4732 | |
4733 | static ssize_t | |
fd01b88c | 4734 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4735 | { |
13ae864b | 4736 | unsigned long long max_sectors, resync; |
88202a0c | 4737 | |
acb180b0 N |
4738 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4739 | return sprintf(page, "none\n"); | |
4740 | ||
72f36d59 N |
4741 | if (mddev->curr_resync == 1 || |
4742 | mddev->curr_resync == 2) | |
4743 | return sprintf(page, "delayed\n"); | |
4744 | ||
c804cdec N |
4745 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4746 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4747 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4748 | else |
58c0fed4 | 4749 | max_sectors = mddev->dev_sectors; |
88202a0c | 4750 | |
acb180b0 | 4751 | resync = mddev->curr_resync_completed; |
13ae864b | 4752 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4753 | } |
4754 | ||
750f199e N |
4755 | static struct md_sysfs_entry md_sync_completed = |
4756 | __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL); | |
88202a0c | 4757 | |
5e96ee65 | 4758 | static ssize_t |
fd01b88c | 4759 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4760 | { |
4761 | return sprintf(page, "%llu\n", | |
4762 | (unsigned long long)mddev->resync_min); | |
4763 | } | |
4764 | static ssize_t | |
fd01b88c | 4765 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4766 | { |
4767 | unsigned long long min; | |
23da422b | 4768 | int err; |
23da422b | 4769 | |
b29bebd6 | 4770 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4771 | return -EINVAL; |
23da422b N |
4772 | |
4773 | spin_lock(&mddev->lock); | |
4774 | err = -EINVAL; | |
5e96ee65 | 4775 | if (min > mddev->resync_max) |
23da422b N |
4776 | goto out_unlock; |
4777 | ||
4778 | err = -EBUSY; | |
5e96ee65 | 4779 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4780 | goto out_unlock; |
5e96ee65 | 4781 | |
50c37b13 N |
4782 | /* Round down to multiple of 4K for safety */ |
4783 | mddev->resync_min = round_down(min, 8); | |
23da422b | 4784 | err = 0; |
5e96ee65 | 4785 | |
23da422b N |
4786 | out_unlock: |
4787 | spin_unlock(&mddev->lock); | |
4788 | return err ?: len; | |
5e96ee65 NB |
4789 | } |
4790 | ||
4791 | static struct md_sysfs_entry md_min_sync = | |
4792 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4793 | ||
c6207277 | 4794 | static ssize_t |
fd01b88c | 4795 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4796 | { |
4797 | if (mddev->resync_max == MaxSector) | |
4798 | return sprintf(page, "max\n"); | |
4799 | else | |
4800 | return sprintf(page, "%llu\n", | |
4801 | (unsigned long long)mddev->resync_max); | |
4802 | } | |
4803 | static ssize_t | |
fd01b88c | 4804 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4805 | { |
23da422b N |
4806 | int err; |
4807 | spin_lock(&mddev->lock); | |
c6207277 N |
4808 | if (strncmp(buf, "max", 3) == 0) |
4809 | mddev->resync_max = MaxSector; | |
4810 | else { | |
5e96ee65 | 4811 | unsigned long long max; |
23da422b N |
4812 | int chunk; |
4813 | ||
4814 | err = -EINVAL; | |
b29bebd6 | 4815 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4816 | goto out_unlock; |
5e96ee65 | 4817 | if (max < mddev->resync_min) |
23da422b N |
4818 | goto out_unlock; |
4819 | ||
4820 | err = -EBUSY; | |
c6207277 | 4821 | if (max < mddev->resync_max && |
4d484a4a | 4822 | mddev->ro == 0 && |
c6207277 | 4823 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4824 | goto out_unlock; |
c6207277 N |
4825 | |
4826 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4827 | chunk = mddev->chunk_sectors; |
4828 | if (chunk) { | |
2ac06c33 | 4829 | sector_t temp = max; |
23da422b N |
4830 | |
4831 | err = -EINVAL; | |
4832 | if (sector_div(temp, chunk)) | |
4833 | goto out_unlock; | |
c6207277 N |
4834 | } |
4835 | mddev->resync_max = max; | |
4836 | } | |
4837 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4838 | err = 0; |
4839 | out_unlock: | |
4840 | spin_unlock(&mddev->lock); | |
4841 | return err ?: len; | |
c6207277 N |
4842 | } |
4843 | ||
4844 | static struct md_sysfs_entry md_max_sync = | |
4845 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4846 | ||
e464eafd | 4847 | static ssize_t |
fd01b88c | 4848 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4849 | { |
4850 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4851 | } | |
4852 | ||
4853 | static ssize_t | |
fd01b88c | 4854 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4855 | { |
b03e0ccb | 4856 | unsigned long long new; |
6791875e | 4857 | int err; |
e464eafd | 4858 | |
4c9309c0 AD |
4859 | err = kstrtoull(buf, 10, &new); |
4860 | if (err < 0) | |
4861 | return err; | |
4862 | if (new != (sector_t)new) | |
e464eafd | 4863 | return -EINVAL; |
23ddff37 | 4864 | |
6791875e N |
4865 | err = mddev_lock(mddev); |
4866 | if (err) | |
4867 | return err; | |
4868 | err = -EINVAL; | |
4869 | if (mddev->pers == NULL || | |
4870 | mddev->pers->quiesce == NULL) | |
4871 | goto unlock; | |
b03e0ccb | 4872 | mddev_suspend(mddev); |
23ddff37 | 4873 | mddev->suspend_lo = new; |
b03e0ccb N |
4874 | mddev_resume(mddev); |
4875 | ||
6791875e N |
4876 | err = 0; |
4877 | unlock: | |
4878 | mddev_unlock(mddev); | |
4879 | return err ?: len; | |
e464eafd N |
4880 | } |
4881 | static struct md_sysfs_entry md_suspend_lo = | |
4882 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4883 | ||
e464eafd | 4884 | static ssize_t |
fd01b88c | 4885 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4886 | { |
4887 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4888 | } | |
4889 | ||
4890 | static ssize_t | |
fd01b88c | 4891 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4892 | { |
b03e0ccb | 4893 | unsigned long long new; |
6791875e | 4894 | int err; |
e464eafd | 4895 | |
4c9309c0 AD |
4896 | err = kstrtoull(buf, 10, &new); |
4897 | if (err < 0) | |
4898 | return err; | |
4899 | if (new != (sector_t)new) | |
e464eafd | 4900 | return -EINVAL; |
23ddff37 | 4901 | |
6791875e N |
4902 | err = mddev_lock(mddev); |
4903 | if (err) | |
4904 | return err; | |
4905 | err = -EINVAL; | |
b03e0ccb | 4906 | if (mddev->pers == NULL) |
6791875e | 4907 | goto unlock; |
b03e0ccb N |
4908 | |
4909 | mddev_suspend(mddev); | |
23ddff37 | 4910 | mddev->suspend_hi = new; |
b03e0ccb N |
4911 | mddev_resume(mddev); |
4912 | ||
6791875e N |
4913 | err = 0; |
4914 | unlock: | |
4915 | mddev_unlock(mddev); | |
4916 | return err ?: len; | |
e464eafd N |
4917 | } |
4918 | static struct md_sysfs_entry md_suspend_hi = | |
4919 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4920 | ||
08a02ecd | 4921 | static ssize_t |
fd01b88c | 4922 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4923 | { |
4924 | if (mddev->reshape_position != MaxSector) | |
4925 | return sprintf(page, "%llu\n", | |
4926 | (unsigned long long)mddev->reshape_position); | |
4927 | strcpy(page, "none\n"); | |
4928 | return 5; | |
4929 | } | |
4930 | ||
4931 | static ssize_t | |
fd01b88c | 4932 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4933 | { |
c6563a8c | 4934 | struct md_rdev *rdev; |
4c9309c0 | 4935 | unsigned long long new; |
6791875e | 4936 | int err; |
6791875e | 4937 | |
4c9309c0 AD |
4938 | err = kstrtoull(buf, 10, &new); |
4939 | if (err < 0) | |
4940 | return err; | |
4941 | if (new != (sector_t)new) | |
08a02ecd | 4942 | return -EINVAL; |
6791875e N |
4943 | err = mddev_lock(mddev); |
4944 | if (err) | |
4945 | return err; | |
4946 | err = -EBUSY; | |
4947 | if (mddev->pers) | |
4948 | goto unlock; | |
08a02ecd N |
4949 | mddev->reshape_position = new; |
4950 | mddev->delta_disks = 0; | |
2c810cdd | 4951 | mddev->reshape_backwards = 0; |
08a02ecd N |
4952 | mddev->new_level = mddev->level; |
4953 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4954 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4955 | rdev_for_each(rdev, mddev) |
4956 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4957 | err = 0; |
4958 | unlock: | |
4959 | mddev_unlock(mddev); | |
4960 | return err ?: len; | |
08a02ecd N |
4961 | } |
4962 | ||
4963 | static struct md_sysfs_entry md_reshape_position = | |
4964 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4965 | reshape_position_store); | |
4966 | ||
2c810cdd N |
4967 | static ssize_t |
4968 | reshape_direction_show(struct mddev *mddev, char *page) | |
4969 | { | |
4970 | return sprintf(page, "%s\n", | |
4971 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4972 | } | |
4973 | ||
4974 | static ssize_t | |
4975 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4976 | { | |
4977 | int backwards = 0; | |
6791875e N |
4978 | int err; |
4979 | ||
2c810cdd N |
4980 | if (cmd_match(buf, "forwards")) |
4981 | backwards = 0; | |
4982 | else if (cmd_match(buf, "backwards")) | |
4983 | backwards = 1; | |
4984 | else | |
4985 | return -EINVAL; | |
4986 | if (mddev->reshape_backwards == backwards) | |
4987 | return len; | |
4988 | ||
6791875e N |
4989 | err = mddev_lock(mddev); |
4990 | if (err) | |
4991 | return err; | |
2c810cdd N |
4992 | /* check if we are allowed to change */ |
4993 | if (mddev->delta_disks) | |
6791875e N |
4994 | err = -EBUSY; |
4995 | else if (mddev->persistent && | |
2c810cdd | 4996 | mddev->major_version == 0) |
6791875e N |
4997 | err = -EINVAL; |
4998 | else | |
4999 | mddev->reshape_backwards = backwards; | |
5000 | mddev_unlock(mddev); | |
5001 | return err ?: len; | |
2c810cdd N |
5002 | } |
5003 | ||
5004 | static struct md_sysfs_entry md_reshape_direction = | |
5005 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
5006 | reshape_direction_store); | |
5007 | ||
b522adcd | 5008 | static ssize_t |
fd01b88c | 5009 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
5010 | { |
5011 | if (mddev->external_size) | |
5012 | return sprintf(page, "%llu\n", | |
5013 | (unsigned long long)mddev->array_sectors/2); | |
5014 | else | |
5015 | return sprintf(page, "default\n"); | |
5016 | } | |
5017 | ||
5018 | static ssize_t | |
fd01b88c | 5019 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
5020 | { |
5021 | sector_t sectors; | |
6791875e N |
5022 | int err; |
5023 | ||
5024 | err = mddev_lock(mddev); | |
5025 | if (err) | |
5026 | return err; | |
b522adcd | 5027 | |
ab5a98b1 | 5028 | /* cluster raid doesn't support change array_sectors */ |
b670883b ZL |
5029 | if (mddev_is_clustered(mddev)) { |
5030 | mddev_unlock(mddev); | |
ab5a98b1 | 5031 | return -EINVAL; |
b670883b | 5032 | } |
ab5a98b1 | 5033 | |
b522adcd DW |
5034 | if (strncmp(buf, "default", 7) == 0) { |
5035 | if (mddev->pers) | |
5036 | sectors = mddev->pers->size(mddev, 0, 0); | |
5037 | else | |
5038 | sectors = mddev->array_sectors; | |
5039 | ||
5040 | mddev->external_size = 0; | |
5041 | } else { | |
5042 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
5043 | err = -EINVAL; |
5044 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
5045 | err = -E2BIG; | |
5046 | else | |
5047 | mddev->external_size = 1; | |
b522adcd DW |
5048 | } |
5049 | ||
6791875e N |
5050 | if (!err) { |
5051 | mddev->array_sectors = sectors; | |
5052 | if (mddev->pers) { | |
5053 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
5054 | revalidate_disk(mddev->gendisk); | |
5055 | } | |
cbe6ef1d | 5056 | } |
6791875e N |
5057 | mddev_unlock(mddev); |
5058 | return err ?: len; | |
b522adcd DW |
5059 | } |
5060 | ||
5061 | static struct md_sysfs_entry md_array_size = | |
5062 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
5063 | array_size_store); | |
e464eafd | 5064 | |
664aed04 AP |
5065 | static ssize_t |
5066 | consistency_policy_show(struct mddev *mddev, char *page) | |
5067 | { | |
5068 | int ret; | |
5069 | ||
5070 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { | |
5071 | ret = sprintf(page, "journal\n"); | |
5072 | } else if (test_bit(MD_HAS_PPL, &mddev->flags)) { | |
5073 | ret = sprintf(page, "ppl\n"); | |
5074 | } else if (mddev->bitmap) { | |
5075 | ret = sprintf(page, "bitmap\n"); | |
5076 | } else if (mddev->pers) { | |
5077 | if (mddev->pers->sync_request) | |
5078 | ret = sprintf(page, "resync\n"); | |
5079 | else | |
5080 | ret = sprintf(page, "none\n"); | |
5081 | } else { | |
5082 | ret = sprintf(page, "unknown\n"); | |
5083 | } | |
5084 | ||
5085 | return ret; | |
5086 | } | |
5087 | ||
5088 | static ssize_t | |
5089 | consistency_policy_store(struct mddev *mddev, const char *buf, size_t len) | |
5090 | { | |
ba903a3e AP |
5091 | int err = 0; |
5092 | ||
664aed04 | 5093 | if (mddev->pers) { |
ba903a3e AP |
5094 | if (mddev->pers->change_consistency_policy) |
5095 | err = mddev->pers->change_consistency_policy(mddev, buf); | |
5096 | else | |
5097 | err = -EBUSY; | |
664aed04 AP |
5098 | } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) { |
5099 | set_bit(MD_HAS_PPL, &mddev->flags); | |
664aed04 | 5100 | } else { |
ba903a3e | 5101 | err = -EINVAL; |
664aed04 | 5102 | } |
ba903a3e AP |
5103 | |
5104 | return err ? err : len; | |
664aed04 AP |
5105 | } |
5106 | ||
5107 | static struct md_sysfs_entry md_consistency_policy = | |
5108 | __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show, | |
5109 | consistency_policy_store); | |
5110 | ||
eae1701f N |
5111 | static struct attribute *md_default_attrs[] = { |
5112 | &md_level.attr, | |
d4dbd025 | 5113 | &md_layout.attr, |
eae1701f | 5114 | &md_raid_disks.attr, |
3b34380a | 5115 | &md_chunk_size.attr, |
a35b0d69 | 5116 | &md_size.attr, |
a94213b1 | 5117 | &md_resync_start.attr, |
8bb93aac | 5118 | &md_metadata.attr, |
6d7ff738 | 5119 | &md_new_device.attr, |
16f17b39 | 5120 | &md_safe_delay.attr, |
9e653b63 | 5121 | &md_array_state.attr, |
08a02ecd | 5122 | &md_reshape_position.attr, |
2c810cdd | 5123 | &md_reshape_direction.attr, |
b522adcd | 5124 | &md_array_size.attr, |
1e50915f | 5125 | &max_corr_read_errors.attr, |
664aed04 | 5126 | &md_consistency_policy.attr, |
411036fa N |
5127 | NULL, |
5128 | }; | |
5129 | ||
5130 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 5131 | &md_scan_mode.attr, |
c4a39551 | 5132 | &md_last_scan_mode.attr, |
9d88883e | 5133 | &md_mismatches.attr, |
88202a0c N |
5134 | &md_sync_min.attr, |
5135 | &md_sync_max.attr, | |
5136 | &md_sync_speed.attr, | |
90b08710 | 5137 | &md_sync_force_parallel.attr, |
88202a0c | 5138 | &md_sync_completed.attr, |
5e96ee65 | 5139 | &md_min_sync.attr, |
c6207277 | 5140 | &md_max_sync.attr, |
e464eafd N |
5141 | &md_suspend_lo.attr, |
5142 | &md_suspend_hi.attr, | |
9b1d1dac | 5143 | &md_bitmap.attr, |
d7f3d291 | 5144 | &md_degraded.attr, |
eae1701f N |
5145 | NULL, |
5146 | }; | |
411036fa N |
5147 | static struct attribute_group md_redundancy_group = { |
5148 | .name = NULL, | |
5149 | .attrs = md_redundancy_attrs, | |
5150 | }; | |
5151 | ||
eae1701f N |
5152 | static ssize_t |
5153 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
5154 | { | |
5155 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 5156 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 5157 | ssize_t rv; |
eae1701f N |
5158 | |
5159 | if (!entry->show) | |
5160 | return -EIO; | |
af8a2434 N |
5161 | spin_lock(&all_mddevs_lock); |
5162 | if (list_empty(&mddev->all_mddevs)) { | |
5163 | spin_unlock(&all_mddevs_lock); | |
5164 | return -EBUSY; | |
5165 | } | |
5166 | mddev_get(mddev); | |
5167 | spin_unlock(&all_mddevs_lock); | |
5168 | ||
b7b17c9b | 5169 | rv = entry->show(mddev, page); |
af8a2434 | 5170 | mddev_put(mddev); |
96de1e66 | 5171 | return rv; |
eae1701f N |
5172 | } |
5173 | ||
5174 | static ssize_t | |
5175 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
5176 | const char *page, size_t length) | |
5177 | { | |
5178 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 5179 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 5180 | ssize_t rv; |
eae1701f N |
5181 | |
5182 | if (!entry->store) | |
5183 | return -EIO; | |
67463acb N |
5184 | if (!capable(CAP_SYS_ADMIN)) |
5185 | return -EACCES; | |
af8a2434 N |
5186 | spin_lock(&all_mddevs_lock); |
5187 | if (list_empty(&mddev->all_mddevs)) { | |
5188 | spin_unlock(&all_mddevs_lock); | |
5189 | return -EBUSY; | |
5190 | } | |
5191 | mddev_get(mddev); | |
5192 | spin_unlock(&all_mddevs_lock); | |
6791875e | 5193 | rv = entry->store(mddev, page, length); |
af8a2434 | 5194 | mddev_put(mddev); |
96de1e66 | 5195 | return rv; |
eae1701f N |
5196 | } |
5197 | ||
5198 | static void md_free(struct kobject *ko) | |
5199 | { | |
fd01b88c | 5200 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
5201 | |
5202 | if (mddev->sysfs_state) | |
5203 | sysfs_put(mddev->sysfs_state); | |
5204 | ||
6cd18e71 N |
5205 | if (mddev->queue) |
5206 | blk_cleanup_queue(mddev->queue); | |
a21d1504 N |
5207 | if (mddev->gendisk) { |
5208 | del_gendisk(mddev->gendisk); | |
5209 | put_disk(mddev->gendisk); | |
5210 | } | |
4ad23a97 | 5211 | percpu_ref_exit(&mddev->writes_pending); |
a21d1504 | 5212 | |
eae1701f N |
5213 | kfree(mddev); |
5214 | } | |
5215 | ||
52cf25d0 | 5216 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
5217 | .show = md_attr_show, |
5218 | .store = md_attr_store, | |
5219 | }; | |
5220 | static struct kobj_type md_ktype = { | |
5221 | .release = md_free, | |
5222 | .sysfs_ops = &md_sysfs_ops, | |
5223 | .default_attrs = md_default_attrs, | |
5224 | }; | |
5225 | ||
1da177e4 LT |
5226 | int mdp_major = 0; |
5227 | ||
5fd3a17e DW |
5228 | static void mddev_delayed_delete(struct work_struct *ws) |
5229 | { | |
fd01b88c | 5230 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 5231 | |
43a70507 | 5232 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
5233 | kobject_del(&mddev->kobj); |
5234 | kobject_put(&mddev->kobj); | |
5235 | } | |
5236 | ||
4ad23a97 N |
5237 | static void no_op(struct percpu_ref *r) {} |
5238 | ||
a415c0f1 N |
5239 | int mddev_init_writes_pending(struct mddev *mddev) |
5240 | { | |
5241 | if (mddev->writes_pending.percpu_count_ptr) | |
5242 | return 0; | |
5243 | if (percpu_ref_init(&mddev->writes_pending, no_op, 0, GFP_KERNEL) < 0) | |
5244 | return -ENOMEM; | |
5245 | /* We want to start with the refcount at zero */ | |
5246 | percpu_ref_put(&mddev->writes_pending); | |
5247 | return 0; | |
5248 | } | |
5249 | EXPORT_SYMBOL_GPL(mddev_init_writes_pending); | |
5250 | ||
efeb53c0 | 5251 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 5252 | { |
039b7225 N |
5253 | /* |
5254 | * If dev is zero, name is the name of a device to allocate with | |
5255 | * an arbitrary minor number. It will be "md_???" | |
5256 | * If dev is non-zero it must be a device number with a MAJOR of | |
5257 | * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then | |
5258 | * the device is being created by opening a node in /dev. | |
5259 | * If "name" is not NULL, the device is being created by | |
5260 | * writing to /sys/module/md_mod/parameters/new_array. | |
5261 | */ | |
48c9c27b | 5262 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 5263 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 5264 | struct gendisk *disk; |
efeb53c0 N |
5265 | int partitioned; |
5266 | int shift; | |
5267 | int unit; | |
3830c62f | 5268 | int error; |
1da177e4 LT |
5269 | |
5270 | if (!mddev) | |
efeb53c0 N |
5271 | return -ENODEV; |
5272 | ||
5273 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
5274 | shift = partitioned ? MdpMinorShift : 0; | |
5275 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 5276 | |
e804ac78 TH |
5277 | /* wait for any previous instance of this device to be |
5278 | * completely removed (mddev_delayed_delete). | |
d3374825 | 5279 | */ |
e804ac78 | 5280 | flush_workqueue(md_misc_wq); |
d3374825 | 5281 | |
48c9c27b | 5282 | mutex_lock(&disks_mutex); |
0909dc44 N |
5283 | error = -EEXIST; |
5284 | if (mddev->gendisk) | |
5285 | goto abort; | |
efeb53c0 | 5286 | |
039b7225 | 5287 | if (name && !dev) { |
efeb53c0 N |
5288 | /* Need to ensure that 'name' is not a duplicate. |
5289 | */ | |
fd01b88c | 5290 | struct mddev *mddev2; |
efeb53c0 N |
5291 | spin_lock(&all_mddevs_lock); |
5292 | ||
5293 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
5294 | if (mddev2->gendisk && | |
5295 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
5296 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 5297 | goto abort; |
efeb53c0 N |
5298 | } |
5299 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 5300 | } |
039b7225 N |
5301 | if (name && dev) |
5302 | /* | |
5303 | * Creating /dev/mdNNN via "newarray", so adjust hold_active. | |
5304 | */ | |
5305 | mddev->hold_active = UNTIL_STOP; | |
8b765398 | 5306 | |
0909dc44 | 5307 | error = -ENOMEM; |
8b765398 | 5308 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
5309 | if (!mddev->queue) |
5310 | goto abort; | |
409c57f3 N |
5311 | mddev->queue->queuedata = mddev; |
5312 | ||
409c57f3 | 5313 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 5314 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 5315 | |
1da177e4 LT |
5316 | disk = alloc_disk(1 << shift); |
5317 | if (!disk) { | |
8b765398 N |
5318 | blk_cleanup_queue(mddev->queue); |
5319 | mddev->queue = NULL; | |
0909dc44 | 5320 | goto abort; |
1da177e4 | 5321 | } |
efeb53c0 | 5322 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 5323 | disk->first_minor = unit << shift; |
efeb53c0 N |
5324 | if (name) |
5325 | strcpy(disk->disk_name, name); | |
5326 | else if (partitioned) | |
1da177e4 | 5327 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 5328 | else |
1da177e4 | 5329 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
5330 | disk->fops = &md_fops; |
5331 | disk->private_data = mddev; | |
5332 | disk->queue = mddev->queue; | |
56883a7e | 5333 | blk_queue_write_cache(mddev->queue, true, true); |
92850bbd | 5334 | /* Allow extended partitions. This makes the |
d3374825 | 5335 | * 'mdp' device redundant, but we can't really |
92850bbd N |
5336 | * remove it now. |
5337 | */ | |
5338 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 5339 | mddev->gendisk = disk; |
b0140891 N |
5340 | /* As soon as we call add_disk(), another thread could get |
5341 | * through to md_open, so make sure it doesn't get too far | |
5342 | */ | |
5343 | mutex_lock(&mddev->open_mutex); | |
5344 | add_disk(disk); | |
5345 | ||
ed9e1982 TH |
5346 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
5347 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
5348 | if (error) { |
5349 | /* This isn't possible, but as kobject_init_and_add is marked | |
5350 | * __must_check, we must do something with the result | |
5351 | */ | |
9d48739e N |
5352 | pr_debug("md: cannot register %s/md - name in use\n", |
5353 | disk->disk_name); | |
0909dc44 N |
5354 | error = 0; |
5355 | } | |
00bcb4ac N |
5356 | if (mddev->kobj.sd && |
5357 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
9d48739e | 5358 | pr_debug("pointless warning\n"); |
b0140891 | 5359 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
5360 | abort: |
5361 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 5362 | if (!error && mddev->kobj.sd) { |
3830c62f | 5363 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 5364 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 5365 | } |
d3374825 | 5366 | mddev_put(mddev); |
0909dc44 | 5367 | return error; |
efeb53c0 N |
5368 | } |
5369 | ||
5370 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
5371 | { | |
78b6350d N |
5372 | if (create_on_open) |
5373 | md_alloc(dev, NULL); | |
1da177e4 LT |
5374 | return NULL; |
5375 | } | |
5376 | ||
e4dca7b7 | 5377 | static int add_named_array(const char *val, const struct kernel_param *kp) |
efeb53c0 | 5378 | { |
039b7225 N |
5379 | /* |
5380 | * val must be "md_*" or "mdNNN". | |
5381 | * For "md_*" we allocate an array with a large free minor number, and | |
efeb53c0 | 5382 | * set the name to val. val must not already be an active name. |
039b7225 N |
5383 | * For "mdNNN" we allocate an array with the minor number NNN |
5384 | * which must not already be in use. | |
efeb53c0 N |
5385 | */ |
5386 | int len = strlen(val); | |
5387 | char buf[DISK_NAME_LEN]; | |
039b7225 | 5388 | unsigned long devnum; |
efeb53c0 N |
5389 | |
5390 | while (len && val[len-1] == '\n') | |
5391 | len--; | |
5392 | if (len >= DISK_NAME_LEN) | |
5393 | return -E2BIG; | |
5394 | strlcpy(buf, val, len+1); | |
039b7225 N |
5395 | if (strncmp(buf, "md_", 3) == 0) |
5396 | return md_alloc(0, buf); | |
5397 | if (strncmp(buf, "md", 2) == 0 && | |
5398 | isdigit(buf[2]) && | |
5399 | kstrtoul(buf+2, 10, &devnum) == 0 && | |
5400 | devnum <= MINORMASK) | |
5401 | return md_alloc(MKDEV(MD_MAJOR, devnum), NULL); | |
5402 | ||
5403 | return -EINVAL; | |
efeb53c0 N |
5404 | } |
5405 | ||
8376d3c1 | 5406 | static void md_safemode_timeout(struct timer_list *t) |
1da177e4 | 5407 | { |
8376d3c1 | 5408 | struct mddev *mddev = from_timer(mddev, t, safemode_timer); |
1da177e4 | 5409 | |
4ad23a97 N |
5410 | mddev->safemode = 1; |
5411 | if (mddev->external) | |
5412 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
5413 | ||
1da177e4 LT |
5414 | md_wakeup_thread(mddev->thread); |
5415 | } | |
5416 | ||
6ff8d8ec | 5417 | static int start_dirty_degraded; |
1da177e4 | 5418 | |
fd01b88c | 5419 | int md_run(struct mddev *mddev) |
1da177e4 | 5420 | { |
2604b703 | 5421 | int err; |
3cb03002 | 5422 | struct md_rdev *rdev; |
84fc4b56 | 5423 | struct md_personality *pers; |
1da177e4 | 5424 | |
a757e64c N |
5425 | if (list_empty(&mddev->disks)) |
5426 | /* cannot run an array with no devices.. */ | |
1da177e4 | 5427 | return -EINVAL; |
1da177e4 LT |
5428 | |
5429 | if (mddev->pers) | |
5430 | return -EBUSY; | |
bb4f1e9d N |
5431 | /* Cannot run until previous stop completes properly */ |
5432 | if (mddev->sysfs_active) | |
5433 | return -EBUSY; | |
b6eb127d | 5434 | |
1da177e4 LT |
5435 | /* |
5436 | * Analyze all RAID superblock(s) | |
5437 | */ | |
1ec4a939 N |
5438 | if (!mddev->raid_disks) { |
5439 | if (!mddev->persistent) | |
5440 | return -EINVAL; | |
a757e64c | 5441 | analyze_sbs(mddev); |
1ec4a939 | 5442 | } |
1da177e4 | 5443 | |
d9d166c2 N |
5444 | if (mddev->level != LEVEL_NONE) |
5445 | request_module("md-level-%d", mddev->level); | |
5446 | else if (mddev->clevel[0]) | |
5447 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
5448 | |
5449 | /* | |
5450 | * Drop all container device buffers, from now on | |
5451 | * the only valid external interface is through the md | |
5452 | * device. | |
1da177e4 | 5453 | */ |
dafb20fa | 5454 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 5455 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5456 | continue; |
5457 | sync_blockdev(rdev->bdev); | |
f98393a6 | 5458 | invalidate_bdev(rdev->bdev); |
97b20ef7 N |
5459 | if (mddev->ro != 1 && |
5460 | (bdev_read_only(rdev->bdev) || | |
5461 | bdev_read_only(rdev->meta_bdev))) { | |
5462 | mddev->ro = 1; | |
5463 | if (mddev->gendisk) | |
5464 | set_disk_ro(mddev->gendisk, 1); | |
5465 | } | |
f0d76d70 N |
5466 | |
5467 | /* perform some consistency tests on the device. | |
5468 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 5469 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 5470 | */ |
a6ff7e08 JB |
5471 | if (rdev->meta_bdev) { |
5472 | /* Nothing to check */; | |
5473 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5474 | if (mddev->dev_sectors && |
5475 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5476 | > rdev->sb_start) { |
9d48739e N |
5477 | pr_warn("md: %s: data overlaps metadata\n", |
5478 | mdname(mddev)); | |
f0d76d70 N |
5479 | return -EINVAL; |
5480 | } | |
5481 | } else { | |
0f420358 | 5482 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 | 5483 | > rdev->data_offset) { |
9d48739e N |
5484 | pr_warn("md: %s: metadata overlaps data\n", |
5485 | mdname(mddev)); | |
f0d76d70 N |
5486 | return -EINVAL; |
5487 | } | |
5488 | } | |
00bcb4ac | 5489 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5490 | } |
5491 | ||
10273170 | 5492 | if (mddev->bio_set == NULL) { |
011067b0 | 5493 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
10273170 ML |
5494 | if (!mddev->bio_set) |
5495 | return -ENOMEM; | |
5496 | } | |
5a85071c | 5497 | if (mddev->sync_set == NULL) { |
026d15f6 | 5498 | mddev->sync_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
5a85071c N |
5499 | if (!mddev->sync_set) |
5500 | return -ENOMEM; | |
5501 | } | |
a167f663 | 5502 | |
1da177e4 | 5503 | spin_lock(&pers_lock); |
d9d166c2 | 5504 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5505 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5506 | spin_unlock(&pers_lock); |
d9d166c2 | 5507 | if (mddev->level != LEVEL_NONE) |
9d48739e N |
5508 | pr_warn("md: personality for level %d is not loaded!\n", |
5509 | mddev->level); | |
d9d166c2 | 5510 | else |
9d48739e N |
5511 | pr_warn("md: personality for level %s is not loaded!\n", |
5512 | mddev->clevel); | |
1da177e4 LT |
5513 | return -EINVAL; |
5514 | } | |
1da177e4 | 5515 | spin_unlock(&pers_lock); |
34817e8c N |
5516 | if (mddev->level != pers->level) { |
5517 | mddev->level = pers->level; | |
5518 | mddev->new_level = pers->level; | |
5519 | } | |
d9d166c2 | 5520 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5521 | |
f6705578 | 5522 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5523 | pers->start_reshape == NULL) { |
f6705578 | 5524 | /* This personality cannot handle reshaping... */ |
f6705578 N |
5525 | module_put(pers->owner); |
5526 | return -EINVAL; | |
5527 | } | |
5528 | ||
7dd5e7c3 N |
5529 | if (pers->sync_request) { |
5530 | /* Warn if this is a potentially silly | |
5531 | * configuration. | |
5532 | */ | |
5533 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5534 | struct md_rdev *rdev2; |
7dd5e7c3 | 5535 | int warned = 0; |
159ec1fc | 5536 | |
dafb20fa N |
5537 | rdev_for_each(rdev, mddev) |
5538 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5539 | if (rdev < rdev2 && |
5540 | rdev->bdev->bd_contains == | |
5541 | rdev2->bdev->bd_contains) { | |
9d48739e N |
5542 | pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n", |
5543 | mdname(mddev), | |
5544 | bdevname(rdev->bdev,b), | |
5545 | bdevname(rdev2->bdev,b2)); | |
7dd5e7c3 N |
5546 | warned = 1; |
5547 | } | |
5548 | } | |
159ec1fc | 5549 | |
7dd5e7c3 | 5550 | if (warned) |
9d48739e | 5551 | pr_warn("True protection against single-disk failure might be compromised.\n"); |
7dd5e7c3 N |
5552 | } |
5553 | ||
657390d2 | 5554 | mddev->recovery = 0; |
58c0fed4 AN |
5555 | /* may be over-ridden by personality */ |
5556 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5557 | ||
6ff8d8ec | 5558 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5559 | |
0f9552b5 | 5560 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5561 | mddev->ro = 2; /* read-only, but switch on first write */ |
5562 | ||
a85dd7b8 SL |
5563 | /* |
5564 | * NOTE: some pers->run(), for example r5l_recovery_log(), wakes | |
5565 | * up mddev->thread. It is important to initialize critical | |
5566 | * resources for mddev->thread BEFORE calling pers->run(). | |
5567 | */ | |
36d091f4 | 5568 | err = pers->run(mddev); |
13e53df3 | 5569 | if (err) |
9d48739e | 5570 | pr_warn("md: pers->run() failed ...\n"); |
36d091f4 | 5571 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
9d48739e N |
5572 | WARN_ONCE(!mddev->external_size, |
5573 | "%s: default size too small, but 'external_size' not in effect?\n", | |
5574 | __func__); | |
5575 | pr_warn("md: invalid array_size %llu > default size %llu\n", | |
5576 | (unsigned long long)mddev->array_sectors / 2, | |
5577 | (unsigned long long)pers->size(mddev, 0, 0) / 2); | |
b522adcd | 5578 | err = -EINVAL; |
b522adcd | 5579 | } |
36d091f4 | 5580 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5581 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
f9209a32 GR |
5582 | struct bitmap *bitmap; |
5583 | ||
5584 | bitmap = bitmap_create(mddev, -1); | |
5585 | if (IS_ERR(bitmap)) { | |
5586 | err = PTR_ERR(bitmap); | |
9d48739e N |
5587 | pr_warn("%s: failed to create bitmap (%d)\n", |
5588 | mdname(mddev), err); | |
f9209a32 GR |
5589 | } else |
5590 | mddev->bitmap = bitmap; | |
5591 | ||
b15c2e57 | 5592 | } |
1da177e4 | 5593 | if (err) { |
5aa61f42 | 5594 | mddev_detach(mddev); |
0c35bd47 N |
5595 | if (mddev->private) |
5596 | pers->free(mddev, mddev->private); | |
bd691922 | 5597 | mddev->private = NULL; |
36d091f4 | 5598 | module_put(pers->owner); |
32a7627c N |
5599 | bitmap_destroy(mddev); |
5600 | return err; | |
1da177e4 | 5601 | } |
5c675f83 | 5602 | if (mddev->queue) { |
bb086a89 SL |
5603 | bool nonrot = true; |
5604 | ||
5605 | rdev_for_each(rdev, mddev) { | |
5606 | if (rdev->raid_disk >= 0 && | |
5607 | !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
5608 | nonrot = false; | |
5609 | break; | |
5610 | } | |
5611 | } | |
5612 | if (mddev->degraded) | |
5613 | nonrot = false; | |
5614 | if (nonrot) | |
5615 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue); | |
5616 | else | |
5617 | queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue); | |
dc3b17cc JK |
5618 | mddev->queue->backing_dev_info->congested_data = mddev; |
5619 | mddev->queue->backing_dev_info->congested_fn = md_congested; | |
5c675f83 | 5620 | } |
36d091f4 | 5621 | if (pers->sync_request) { |
00bcb4ac N |
5622 | if (mddev->kobj.sd && |
5623 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
5624 | pr_warn("md: cannot register extra attributes for %s\n", |
5625 | mdname(mddev)); | |
00bcb4ac | 5626 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5627 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5628 | mddev->ro = 0; |
5629 | ||
1e50915f RB |
5630 | atomic_set(&mddev->max_corr_read_errors, |
5631 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 | 5632 | mddev->safemode = 0; |
28c1b9fd GR |
5633 | if (mddev_is_clustered(mddev)) |
5634 | mddev->safemode_delay = 0; | |
5635 | else | |
5636 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ | |
1da177e4 | 5637 | mddev->in_sync = 1; |
0ca69886 | 5638 | smp_wmb(); |
36d091f4 N |
5639 | spin_lock(&mddev->lock); |
5640 | mddev->pers = pers; | |
36d091f4 | 5641 | spin_unlock(&mddev->lock); |
dafb20fa | 5642 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5643 | if (rdev->raid_disk >= 0) |
5644 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5645 | /* failure here is OK */; |
f72ffdd6 | 5646 | |
a4a3d26d N |
5647 | if (mddev->degraded && !mddev->ro) |
5648 | /* This ensures that recovering status is reported immediately | |
5649 | * via sysfs - until a lack of spares is confirmed. | |
5650 | */ | |
5651 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
1da177e4 | 5652 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5653 | |
2953079c | 5654 | if (mddev->sb_flags) |
850b2b42 | 5655 | md_update_sb(mddev, 0); |
1da177e4 | 5656 | |
d7603b7e | 5657 | md_new_event(mddev); |
00bcb4ac N |
5658 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5659 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5660 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5661 | return 0; |
5662 | } | |
390ee602 | 5663 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5664 | |
fd01b88c | 5665 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5666 | { |
5667 | int err; | |
5668 | ||
5669 | err = md_run(mddev); | |
5670 | if (err) | |
5671 | goto out; | |
69e51b44 N |
5672 | err = bitmap_load(mddev); |
5673 | if (err) { | |
5674 | bitmap_destroy(mddev); | |
5675 | goto out; | |
5676 | } | |
0fd018af | 5677 | |
28c1b9fd GR |
5678 | if (mddev_is_clustered(mddev)) |
5679 | md_allow_write(mddev); | |
5680 | ||
0fd018af JB |
5681 | md_wakeup_thread(mddev->thread); |
5682 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5683 | ||
fe60b014 N |
5684 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5685 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5686 | mddev->changed = 1; |
fe60b014 N |
5687 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5688 | out: | |
5689 | return err; | |
5690 | } | |
5691 | ||
fd01b88c | 5692 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5693 | { |
5694 | struct gendisk *disk = mddev->gendisk; | |
97b20ef7 N |
5695 | struct md_rdev *rdev; |
5696 | bool has_journal = false; | |
5697 | bool has_readonly = false; | |
1da177e4 | 5698 | |
80fab1d7 | 5699 | /* Complain if it has no devices */ |
1da177e4 | 5700 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5701 | return -ENXIO; |
5702 | if (!mddev->pers) | |
5703 | return -EINVAL; | |
5704 | if (!mddev->ro) | |
5705 | return -EBUSY; | |
339421de | 5706 | |
97b20ef7 N |
5707 | rcu_read_lock(); |
5708 | rdev_for_each_rcu(rdev, mddev) { | |
5709 | if (test_bit(Journal, &rdev->flags) && | |
5710 | !test_bit(Faulty, &rdev->flags)) | |
5711 | has_journal = true; | |
5712 | if (bdev_read_only(rdev->bdev)) | |
5713 | has_readonly = true; | |
5714 | } | |
5715 | rcu_read_unlock(); | |
5716 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal) | |
339421de | 5717 | /* Don't restart rw with journal missing/faulty */ |
339421de | 5718 | return -EINVAL; |
97b20ef7 N |
5719 | if (has_readonly) |
5720 | return -EROFS; | |
339421de | 5721 | |
80fab1d7 AN |
5722 | mddev->safemode = 0; |
5723 | mddev->ro = 0; | |
5724 | set_disk_ro(disk, 0); | |
9d48739e | 5725 | pr_debug("md: %s switched to read-write mode.\n", mdname(mddev)); |
80fab1d7 AN |
5726 | /* Kick recovery or resync if necessary */ |
5727 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5728 | md_wakeup_thread(mddev->thread); | |
5729 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5730 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5731 | return 0; |
1da177e4 LT |
5732 | } |
5733 | ||
fd01b88c | 5734 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5735 | { |
5736 | mddev->array_sectors = 0; | |
5737 | mddev->external_size = 0; | |
5738 | mddev->dev_sectors = 0; | |
5739 | mddev->raid_disks = 0; | |
5740 | mddev->recovery_cp = 0; | |
5741 | mddev->resync_min = 0; | |
5742 | mddev->resync_max = MaxSector; | |
5743 | mddev->reshape_position = MaxSector; | |
5744 | mddev->external = 0; | |
5745 | mddev->persistent = 0; | |
5746 | mddev->level = LEVEL_NONE; | |
5747 | mddev->clevel[0] = 0; | |
5748 | mddev->flags = 0; | |
2953079c | 5749 | mddev->sb_flags = 0; |
6177b472 N |
5750 | mddev->ro = 0; |
5751 | mddev->metadata_type[0] = 0; | |
5752 | mddev->chunk_sectors = 0; | |
5753 | mddev->ctime = mddev->utime = 0; | |
5754 | mddev->layout = 0; | |
5755 | mddev->max_disks = 0; | |
5756 | mddev->events = 0; | |
a8707c08 | 5757 | mddev->can_decrease_events = 0; |
6177b472 | 5758 | mddev->delta_disks = 0; |
2c810cdd | 5759 | mddev->reshape_backwards = 0; |
6177b472 N |
5760 | mddev->new_level = LEVEL_NONE; |
5761 | mddev->new_layout = 0; | |
5762 | mddev->new_chunk_sectors = 0; | |
5763 | mddev->curr_resync = 0; | |
7f7583d4 | 5764 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5765 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5766 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5767 | mddev->recovery = 0; | |
5768 | mddev->in_sync = 0; | |
f0b4f7e2 | 5769 | mddev->changed = 0; |
6177b472 | 5770 | mddev->degraded = 0; |
6177b472 | 5771 | mddev->safemode = 0; |
bd691922 | 5772 | mddev->private = NULL; |
c20c33f0 | 5773 | mddev->cluster_info = NULL; |
6177b472 N |
5774 | mddev->bitmap_info.offset = 0; |
5775 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5776 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5777 | mddev->bitmap_info.chunksize = 0; |
5778 | mddev->bitmap_info.daemon_sleep = 0; | |
5779 | mddev->bitmap_info.max_write_behind = 0; | |
c20c33f0 | 5780 | mddev->bitmap_info.nodes = 0; |
6177b472 N |
5781 | } |
5782 | ||
fd01b88c | 5783 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5784 | { |
6b6204ee | 5785 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5786 | flush_workqueue(md_misc_wq); |
a047e125 | 5787 | if (mddev->sync_thread) { |
a047e125 | 5788 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5789 | md_reap_sync_thread(mddev); |
a047e125 N |
5790 | } |
5791 | ||
5792 | del_timer_sync(&mddev->safemode_timer); | |
5793 | ||
034e33f5 SL |
5794 | if (mddev->pers && mddev->pers->quiesce) { |
5795 | mddev->pers->quiesce(mddev, 1); | |
5796 | mddev->pers->quiesce(mddev, 0); | |
5797 | } | |
a047e125 | 5798 | bitmap_flush(mddev); |
a047e125 | 5799 | |
b6d428c6 | 5800 | if (mddev->ro == 0 && |
28c1b9fd | 5801 | ((!mddev->in_sync && !mddev_is_clustered(mddev)) || |
2953079c | 5802 | mddev->sb_flags)) { |
a047e125 | 5803 | /* mark array as shutdown cleanly */ |
28c1b9fd GR |
5804 | if (!mddev_is_clustered(mddev)) |
5805 | mddev->in_sync = 1; | |
a047e125 N |
5806 | md_update_sb(mddev, 1); |
5807 | } | |
5808 | } | |
defad61a | 5809 | |
fd01b88c | 5810 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5811 | { |
29f097c4 | 5812 | mddev_lock_nointr(mddev); |
defad61a N |
5813 | __md_stop_writes(mddev); |
5814 | mddev_unlock(mddev); | |
5815 | } | |
390ee602 | 5816 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5817 | |
5aa61f42 N |
5818 | static void mddev_detach(struct mddev *mddev) |
5819 | { | |
48df498d | 5820 | bitmap_wait_behind_writes(mddev); |
36d091f4 | 5821 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5822 | mddev->pers->quiesce(mddev, 1); |
5823 | mddev->pers->quiesce(mddev, 0); | |
5824 | } | |
5825 | md_unregister_thread(&mddev->thread); | |
5826 | if (mddev->queue) | |
5827 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5828 | } | |
5829 | ||
5eff3c43 | 5830 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5831 | { |
36d091f4 | 5832 | struct md_personality *pers = mddev->pers; |
48df498d | 5833 | bitmap_destroy(mddev); |
5aa61f42 | 5834 | mddev_detach(mddev); |
ee5d004f N |
5835 | /* Ensure ->event_work is done */ |
5836 | flush_workqueue(md_misc_wq); | |
36d091f4 | 5837 | spin_lock(&mddev->lock); |
6177b472 | 5838 | mddev->pers = NULL; |
36d091f4 N |
5839 | spin_unlock(&mddev->lock); |
5840 | pers->free(mddev, mddev->private); | |
bd691922 | 5841 | mddev->private = NULL; |
36d091f4 N |
5842 | if (pers->sync_request && mddev->to_remove == NULL) |
5843 | mddev->to_remove = &md_redundancy_group; | |
5844 | module_put(pers->owner); | |
cca9cf90 | 5845 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5846 | } |
5eff3c43 N |
5847 | |
5848 | void md_stop(struct mddev *mddev) | |
5849 | { | |
5850 | /* stop the array and free an attached data structures. | |
5851 | * This is called from dm-raid | |
5852 | */ | |
5853 | __md_stop(mddev); | |
0202ce8a | 5854 | if (mddev->bio_set) { |
5eff3c43 | 5855 | bioset_free(mddev->bio_set); |
0202ce8a ZK |
5856 | mddev->bio_set = NULL; |
5857 | } | |
5858 | if (mddev->sync_set) { | |
5859 | bioset_free(mddev->sync_set); | |
5860 | mddev->sync_set = NULL; | |
5861 | } | |
5eff3c43 N |
5862 | } |
5863 | ||
390ee602 | 5864 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5865 | |
a05b7ea0 | 5866 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5867 | { |
5868 | int err = 0; | |
30b8feb7 N |
5869 | int did_freeze = 0; |
5870 | ||
5871 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5872 | did_freeze = 1; | |
5873 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5874 | md_wakeup_thread(mddev->thread); | |
5875 | } | |
f851b60d | 5876 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5877 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5878 | if (mddev->sync_thread) |
30b8feb7 N |
5879 | /* Thread might be blocked waiting for metadata update |
5880 | * which will now never happen */ | |
5881 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5882 | |
2953079c | 5883 | if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
88724bfa | 5884 | return -EBUSY; |
30b8feb7 | 5885 | mddev_unlock(mddev); |
f851b60d N |
5886 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5887 | &mddev->recovery)); | |
88724bfa | 5888 | wait_event(mddev->sb_wait, |
2953079c | 5889 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
30b8feb7 N |
5890 | mddev_lock_nointr(mddev); |
5891 | ||
a4bd82d0 | 5892 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5893 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5894 | mddev->sync_thread || |
af8d8e6f | 5895 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5896 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5897 | if (did_freeze) { |
5898 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5899 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5900 | md_wakeup_thread(mddev->thread); |
5901 | } | |
a4bd82d0 N |
5902 | err = -EBUSY; |
5903 | goto out; | |
5904 | } | |
5905 | if (mddev->pers) { | |
defad61a | 5906 | __md_stop_writes(mddev); |
a4bd82d0 N |
5907 | |
5908 | err = -ENXIO; | |
5909 | if (mddev->ro==1) | |
5910 | goto out; | |
5911 | mddev->ro = 1; | |
5912 | set_disk_ro(mddev->gendisk, 1); | |
5913 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d N |
5914 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5915 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5916 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5917 | err = 0; |
a4bd82d0 N |
5918 | } |
5919 | out: | |
5920 | mutex_unlock(&mddev->open_mutex); | |
5921 | return err; | |
5922 | } | |
5923 | ||
9e653b63 N |
5924 | /* mode: |
5925 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5926 | * 2 - stop but do not disassemble array |
5927 | */ | |
f72ffdd6 | 5928 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5929 | struct block_device *bdev) |
1da177e4 | 5930 | { |
1da177e4 | 5931 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5932 | struct md_rdev *rdev; |
30b8feb7 N |
5933 | int did_freeze = 0; |
5934 | ||
5935 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5936 | did_freeze = 1; | |
5937 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5938 | md_wakeup_thread(mddev->thread); | |
5939 | } | |
f851b60d | 5940 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5941 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5942 | if (mddev->sync_thread) |
30b8feb7 N |
5943 | /* Thread might be blocked waiting for metadata update |
5944 | * which will now never happen */ | |
5945 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5946 | |
30b8feb7 | 5947 | mddev_unlock(mddev); |
f851b60d N |
5948 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
5949 | !test_bit(MD_RECOVERY_RUNNING, | |
5950 | &mddev->recovery))); | |
30b8feb7 | 5951 | mddev_lock_nointr(mddev); |
1da177e4 | 5952 | |
c8c00a69 | 5953 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5954 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5955 | mddev->sysfs_active || |
5956 | mddev->sync_thread || | |
af8d8e6f | 5957 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5958 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5959 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5960 | if (did_freeze) { |
5961 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5962 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5963 | md_wakeup_thread(mddev->thread); |
5964 | } | |
260fa034 N |
5965 | return -EBUSY; |
5966 | } | |
6e17b027 | 5967 | if (mddev->pers) { |
a4bd82d0 N |
5968 | if (mddev->ro) |
5969 | set_disk_ro(disk, 0); | |
409c57f3 | 5970 | |
defad61a | 5971 | __md_stop_writes(mddev); |
5eff3c43 | 5972 | __md_stop(mddev); |
dc3b17cc | 5973 | mddev->queue->backing_dev_info->congested_fn = NULL; |
6177b472 | 5974 | |
a4bd82d0 | 5975 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5976 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5977 | |
dafb20fa | 5978 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5979 | if (rdev->raid_disk >= 0) |
5980 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5981 | |
a4bd82d0 | 5982 | set_capacity(disk, 0); |
6e17b027 | 5983 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5984 | mddev->changed = 1; |
a4bd82d0 | 5985 | revalidate_disk(disk); |
0d4ca600 | 5986 | |
a4bd82d0 N |
5987 | if (mddev->ro) |
5988 | mddev->ro = 0; | |
6e17b027 N |
5989 | } else |
5990 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5991 | /* |
5992 | * Free resources if final stop | |
5993 | */ | |
9e653b63 | 5994 | if (mode == 0) { |
9d48739e | 5995 | pr_info("md: %s stopped.\n", mdname(mddev)); |
1da177e4 | 5996 | |
c3d9714e | 5997 | if (mddev->bitmap_info.file) { |
4af1a041 N |
5998 | struct file *f = mddev->bitmap_info.file; |
5999 | spin_lock(&mddev->lock); | |
c3d9714e | 6000 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
6001 | spin_unlock(&mddev->lock); |
6002 | fput(f); | |
978f946b | 6003 | } |
c3d9714e | 6004 | mddev->bitmap_info.offset = 0; |
978f946b | 6005 | |
1da177e4 LT |
6006 | export_array(mddev); |
6007 | ||
6177b472 | 6008 | md_clean(mddev); |
efeb53c0 N |
6009 | if (mddev->hold_active == UNTIL_STOP) |
6010 | mddev->hold_active = 0; | |
a4bd82d0 | 6011 | } |
d7603b7e | 6012 | md_new_event(mddev); |
00bcb4ac | 6013 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 6014 | return 0; |
1da177e4 LT |
6015 | } |
6016 | ||
fdee8ae4 | 6017 | #ifndef MODULE |
fd01b88c | 6018 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 6019 | { |
3cb03002 | 6020 | struct md_rdev *rdev; |
1da177e4 LT |
6021 | int err; |
6022 | ||
a757e64c | 6023 | if (list_empty(&mddev->disks)) |
1da177e4 | 6024 | return; |
1da177e4 | 6025 | |
9d48739e | 6026 | pr_info("md: running: "); |
1da177e4 | 6027 | |
dafb20fa | 6028 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6029 | char b[BDEVNAME_SIZE]; |
9d48739e | 6030 | pr_cont("<%s>", bdevname(rdev->bdev,b)); |
1da177e4 | 6031 | } |
9d48739e | 6032 | pr_cont("\n"); |
1da177e4 | 6033 | |
d710e138 | 6034 | err = do_md_run(mddev); |
1da177e4 | 6035 | if (err) { |
9d48739e | 6036 | pr_warn("md: do_md_run() returned %d\n", err); |
a05b7ea0 | 6037 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
6038 | } |
6039 | } | |
6040 | ||
6041 | /* | |
6042 | * lets try to run arrays based on all disks that have arrived | |
6043 | * until now. (those are in pending_raid_disks) | |
6044 | * | |
6045 | * the method: pick the first pending disk, collect all disks with | |
6046 | * the same UUID, remove all from the pending list and put them into | |
6047 | * the 'same_array' list. Then order this list based on superblock | |
6048 | * update time (freshest comes first), kick out 'old' disks and | |
6049 | * compare superblocks. If everything's fine then run it. | |
6050 | * | |
6051 | * If "unit" is allocated, then bump its reference count | |
6052 | */ | |
6053 | static void autorun_devices(int part) | |
6054 | { | |
3cb03002 | 6055 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 6056 | struct mddev *mddev; |
1da177e4 LT |
6057 | char b[BDEVNAME_SIZE]; |
6058 | ||
9d48739e | 6059 | pr_info("md: autorun ...\n"); |
1da177e4 | 6060 | while (!list_empty(&pending_raid_disks)) { |
e8703fe1 | 6061 | int unit; |
1da177e4 | 6062 | dev_t dev; |
ad01c9e3 | 6063 | LIST_HEAD(candidates); |
1da177e4 | 6064 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 6065 | struct md_rdev, same_set); |
1da177e4 | 6066 | |
9d48739e | 6067 | pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b)); |
1da177e4 | 6068 | INIT_LIST_HEAD(&candidates); |
159ec1fc | 6069 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 | 6070 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
9d48739e N |
6071 | pr_debug("md: adding %s ...\n", |
6072 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
6073 | list_move(&rdev->same_set, &candidates); |
6074 | } | |
6075 | /* | |
6076 | * now we have a set of devices, with all of them having | |
6077 | * mostly sane superblocks. It's time to allocate the | |
6078 | * mddev. | |
6079 | */ | |
e8703fe1 N |
6080 | if (part) { |
6081 | dev = MKDEV(mdp_major, | |
6082 | rdev0->preferred_minor << MdpMinorShift); | |
6083 | unit = MINOR(dev) >> MdpMinorShift; | |
6084 | } else { | |
6085 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
6086 | unit = MINOR(dev); | |
6087 | } | |
6088 | if (rdev0->preferred_minor != unit) { | |
9d48739e N |
6089 | pr_warn("md: unit number in %s is bad: %d\n", |
6090 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
1da177e4 LT |
6091 | break; |
6092 | } | |
1da177e4 LT |
6093 | |
6094 | md_probe(dev, NULL, NULL); | |
6095 | mddev = mddev_find(dev); | |
9bbbca3a NB |
6096 | if (!mddev || !mddev->gendisk) { |
6097 | if (mddev) | |
6098 | mddev_put(mddev); | |
1da177e4 LT |
6099 | break; |
6100 | } | |
f72ffdd6 | 6101 | if (mddev_lock(mddev)) |
9d48739e | 6102 | pr_warn("md: %s locked, cannot run\n", mdname(mddev)); |
1da177e4 LT |
6103 | else if (mddev->raid_disks || mddev->major_version |
6104 | || !list_empty(&mddev->disks)) { | |
9d48739e | 6105 | pr_warn("md: %s already running, cannot run %s\n", |
1da177e4 LT |
6106 | mdname(mddev), bdevname(rdev0->bdev,b)); |
6107 | mddev_unlock(mddev); | |
6108 | } else { | |
9d48739e | 6109 | pr_debug("md: created %s\n", mdname(mddev)); |
1ec4a939 | 6110 | mddev->persistent = 1; |
159ec1fc | 6111 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
6112 | list_del_init(&rdev->same_set); |
6113 | if (bind_rdev_to_array(rdev, mddev)) | |
6114 | export_rdev(rdev); | |
6115 | } | |
6116 | autorun_array(mddev); | |
6117 | mddev_unlock(mddev); | |
6118 | } | |
6119 | /* on success, candidates will be empty, on error | |
6120 | * it won't... | |
6121 | */ | |
159ec1fc | 6122 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 6123 | list_del_init(&rdev->same_set); |
1da177e4 | 6124 | export_rdev(rdev); |
4b80991c | 6125 | } |
1da177e4 LT |
6126 | mddev_put(mddev); |
6127 | } | |
9d48739e | 6128 | pr_info("md: ... autorun DONE.\n"); |
1da177e4 | 6129 | } |
fdee8ae4 | 6130 | #endif /* !MODULE */ |
1da177e4 | 6131 | |
f72ffdd6 | 6132 | static int get_version(void __user *arg) |
1da177e4 LT |
6133 | { |
6134 | mdu_version_t ver; | |
6135 | ||
6136 | ver.major = MD_MAJOR_VERSION; | |
6137 | ver.minor = MD_MINOR_VERSION; | |
6138 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
6139 | ||
6140 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
6141 | return -EFAULT; | |
6142 | ||
6143 | return 0; | |
6144 | } | |
6145 | ||
f72ffdd6 | 6146 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
6147 | { |
6148 | mdu_array_info_t info; | |
a9f326eb | 6149 | int nr,working,insync,failed,spare; |
3cb03002 | 6150 | struct md_rdev *rdev; |
1da177e4 | 6151 | |
1ca69c4b N |
6152 | nr = working = insync = failed = spare = 0; |
6153 | rcu_read_lock(); | |
6154 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6155 | nr++; |
b2d444d7 | 6156 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
6157 | failed++; |
6158 | else { | |
6159 | working++; | |
b2d444d7 | 6160 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 6161 | insync++; |
b347af81 SL |
6162 | else if (test_bit(Journal, &rdev->flags)) |
6163 | /* TODO: add journal count to md_u.h */ | |
6164 | ; | |
1da177e4 LT |
6165 | else |
6166 | spare++; | |
6167 | } | |
6168 | } | |
1ca69c4b | 6169 | rcu_read_unlock(); |
1da177e4 LT |
6170 | |
6171 | info.major_version = mddev->major_version; | |
6172 | info.minor_version = mddev->minor_version; | |
6173 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6174 | info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 6175 | info.level = mddev->level; |
58c0fed4 AN |
6176 | info.size = mddev->dev_sectors / 2; |
6177 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 6178 | info.size = -1; |
1da177e4 LT |
6179 | info.nr_disks = nr; |
6180 | info.raid_disks = mddev->raid_disks; | |
6181 | info.md_minor = mddev->md_minor; | |
6182 | info.not_persistent= !mddev->persistent; | |
6183 | ||
9ebc6ef1 | 6184 | info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
6185 | info.state = 0; |
6186 | if (mddev->in_sync) | |
6187 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 6188 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 6189 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
ca8895d9 GR |
6190 | if (mddev_is_clustered(mddev)) |
6191 | info.state |= (1<<MD_SB_CLUSTERED); | |
a9f326eb | 6192 | info.active_disks = insync; |
1da177e4 LT |
6193 | info.working_disks = working; |
6194 | info.failed_disks = failed; | |
6195 | info.spare_disks = spare; | |
6196 | ||
6197 | info.layout = mddev->layout; | |
9d8f0363 | 6198 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
6199 | |
6200 | if (copy_to_user(arg, &info, sizeof(info))) | |
6201 | return -EFAULT; | |
6202 | ||
6203 | return 0; | |
6204 | } | |
6205 | ||
f72ffdd6 | 6206 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
6207 | { |
6208 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 6209 | char *ptr; |
4af1a041 | 6210 | int err; |
32a7627c | 6211 | |
b6878d9e | 6212 | file = kzalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 6213 | if (!file) |
4af1a041 | 6214 | return -ENOMEM; |
32a7627c | 6215 | |
4af1a041 N |
6216 | err = 0; |
6217 | spin_lock(&mddev->lock); | |
25eafe1a BR |
6218 | /* bitmap enabled */ |
6219 | if (mddev->bitmap_info.file) { | |
6220 | ptr = file_path(mddev->bitmap_info.file, file->pathname, | |
6221 | sizeof(file->pathname)); | |
6222 | if (IS_ERR(ptr)) | |
6223 | err = PTR_ERR(ptr); | |
6224 | else | |
6225 | memmove(file->pathname, ptr, | |
6226 | sizeof(file->pathname)-(ptr-file->pathname)); | |
6227 | } | |
4af1a041 | 6228 | spin_unlock(&mddev->lock); |
32a7627c | 6229 | |
4af1a041 N |
6230 | if (err == 0 && |
6231 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 6232 | err = -EFAULT; |
4af1a041 | 6233 | |
32a7627c N |
6234 | kfree(file); |
6235 | return err; | |
6236 | } | |
6237 | ||
f72ffdd6 | 6238 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
6239 | { |
6240 | mdu_disk_info_t info; | |
3cb03002 | 6241 | struct md_rdev *rdev; |
1da177e4 LT |
6242 | |
6243 | if (copy_from_user(&info, arg, sizeof(info))) | |
6244 | return -EFAULT; | |
6245 | ||
1ca69c4b | 6246 | rcu_read_lock(); |
57d051dc | 6247 | rdev = md_find_rdev_nr_rcu(mddev, info.number); |
1da177e4 LT |
6248 | if (rdev) { |
6249 | info.major = MAJOR(rdev->bdev->bd_dev); | |
6250 | info.minor = MINOR(rdev->bdev->bd_dev); | |
6251 | info.raid_disk = rdev->raid_disk; | |
6252 | info.state = 0; | |
b2d444d7 | 6253 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6254 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 6255 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
6256 | info.state |= (1<<MD_DISK_ACTIVE); |
6257 | info.state |= (1<<MD_DISK_SYNC); | |
6258 | } | |
9efdca16 | 6259 | if (test_bit(Journal, &rdev->flags)) |
bac624f3 | 6260 | info.state |= (1<<MD_DISK_JOURNAL); |
8ddf9efe N |
6261 | if (test_bit(WriteMostly, &rdev->flags)) |
6262 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
6263 | if (test_bit(FailFast, &rdev->flags)) |
6264 | info.state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 LT |
6265 | } else { |
6266 | info.major = info.minor = 0; | |
6267 | info.raid_disk = -1; | |
6268 | info.state = (1<<MD_DISK_REMOVED); | |
6269 | } | |
1ca69c4b | 6270 | rcu_read_unlock(); |
1da177e4 LT |
6271 | |
6272 | if (copy_to_user(arg, &info, sizeof(info))) | |
6273 | return -EFAULT; | |
6274 | ||
6275 | return 0; | |
6276 | } | |
6277 | ||
f72ffdd6 | 6278 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
6279 | { |
6280 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 6281 | struct md_rdev *rdev; |
1da177e4 LT |
6282 | dev_t dev = MKDEV(info->major,info->minor); |
6283 | ||
1aee41f6 GR |
6284 | if (mddev_is_clustered(mddev) && |
6285 | !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) { | |
9d48739e N |
6286 | pr_warn("%s: Cannot add to clustered mddev.\n", |
6287 | mdname(mddev)); | |
1aee41f6 GR |
6288 | return -EINVAL; |
6289 | } | |
6290 | ||
1da177e4 LT |
6291 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) |
6292 | return -EOVERFLOW; | |
6293 | ||
6294 | if (!mddev->raid_disks) { | |
6295 | int err; | |
6296 | /* expecting a device which has a superblock */ | |
6297 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
6298 | if (IS_ERR(rdev)) { | |
9d48739e | 6299 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6300 | PTR_ERR(rdev)); |
6301 | return PTR_ERR(rdev); | |
6302 | } | |
6303 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
6304 | struct md_rdev *rdev0 |
6305 | = list_entry(mddev->disks.next, | |
6306 | struct md_rdev, same_set); | |
a9f326eb | 6307 | err = super_types[mddev->major_version] |
1da177e4 LT |
6308 | .load_super(rdev, rdev0, mddev->minor_version); |
6309 | if (err < 0) { | |
9d48739e | 6310 | pr_warn("md: %s has different UUID to %s\n", |
f72ffdd6 | 6311 | bdevname(rdev->bdev,b), |
1da177e4 LT |
6312 | bdevname(rdev0->bdev,b2)); |
6313 | export_rdev(rdev); | |
6314 | return -EINVAL; | |
6315 | } | |
6316 | } | |
6317 | err = bind_rdev_to_array(rdev, mddev); | |
6318 | if (err) | |
6319 | export_rdev(rdev); | |
6320 | return err; | |
6321 | } | |
6322 | ||
6323 | /* | |
6324 | * add_new_disk can be used once the array is assembled | |
6325 | * to add "hot spares". They must already have a superblock | |
6326 | * written | |
6327 | */ | |
6328 | if (mddev->pers) { | |
6329 | int err; | |
6330 | if (!mddev->pers->hot_add_disk) { | |
9d48739e N |
6331 | pr_warn("%s: personality does not support diskops!\n", |
6332 | mdname(mddev)); | |
1da177e4 LT |
6333 | return -EINVAL; |
6334 | } | |
7b1e35f6 N |
6335 | if (mddev->persistent) |
6336 | rdev = md_import_device(dev, mddev->major_version, | |
6337 | mddev->minor_version); | |
6338 | else | |
6339 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 6340 | if (IS_ERR(rdev)) { |
9d48739e | 6341 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6342 | PTR_ERR(rdev)); |
6343 | return PTR_ERR(rdev); | |
6344 | } | |
1a855a06 | 6345 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
6346 | if (!mddev->persistent) { |
6347 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 6348 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 6349 | rdev->raid_disk = info->raid_disk; |
bf572541 | 6350 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 6351 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 6352 | } else |
41158c7e | 6353 | rdev->raid_disk = -1; |
f466722c | 6354 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
6355 | } else |
6356 | super_types[mddev->major_version]. | |
6357 | validate_super(mddev, rdev); | |
bedd86b7 | 6358 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 6359 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
6360 | /* This was a hot-add request, but events doesn't |
6361 | * match, so reject it. | |
6362 | */ | |
6363 | export_rdev(rdev); | |
6364 | return -EINVAL; | |
6365 | } | |
6366 | ||
b2d444d7 | 6367 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
6368 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6369 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
6370 | else |
6371 | clear_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6372 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6373 | set_bit(FailFast, &rdev->flags); | |
6374 | else | |
6375 | clear_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6376 | |
f6b6ec5c SL |
6377 | if (info->state & (1<<MD_DISK_JOURNAL)) { |
6378 | struct md_rdev *rdev2; | |
6379 | bool has_journal = false; | |
6380 | ||
6381 | /* make sure no existing journal disk */ | |
6382 | rdev_for_each(rdev2, mddev) { | |
6383 | if (test_bit(Journal, &rdev2->flags)) { | |
6384 | has_journal = true; | |
6385 | break; | |
6386 | } | |
6387 | } | |
230b55fa | 6388 | if (has_journal || mddev->bitmap) { |
f6b6ec5c SL |
6389 | export_rdev(rdev); |
6390 | return -EBUSY; | |
6391 | } | |
bac624f3 | 6392 | set_bit(Journal, &rdev->flags); |
f6b6ec5c | 6393 | } |
1aee41f6 GR |
6394 | /* |
6395 | * check whether the device shows up in other nodes | |
6396 | */ | |
6397 | if (mddev_is_clustered(mddev)) { | |
dbb64f86 | 6398 | if (info->state & (1 << MD_DISK_CANDIDATE)) |
1aee41f6 | 6399 | set_bit(Candidate, &rdev->flags); |
dbb64f86 | 6400 | else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) { |
1aee41f6 | 6401 | /* --add initiated by this node */ |
dbb64f86 | 6402 | err = md_cluster_ops->add_new_disk(mddev, rdev); |
1aee41f6 | 6403 | if (err) { |
1aee41f6 GR |
6404 | export_rdev(rdev); |
6405 | return err; | |
6406 | } | |
6407 | } | |
6408 | } | |
6409 | ||
1da177e4 LT |
6410 | rdev->raid_disk = -1; |
6411 | err = bind_rdev_to_array(rdev, mddev); | |
dbb64f86 | 6412 | |
1da177e4 LT |
6413 | if (err) |
6414 | export_rdev(rdev); | |
dbb64f86 GR |
6415 | |
6416 | if (mddev_is_clustered(mddev)) { | |
e566aef1 GJ |
6417 | if (info->state & (1 << MD_DISK_CANDIDATE)) { |
6418 | if (!err) { | |
6419 | err = md_cluster_ops->new_disk_ack(mddev, | |
6420 | err == 0); | |
6421 | if (err) | |
6422 | md_kick_rdev_from_array(rdev); | |
6423 | } | |
6424 | } else { | |
dbb64f86 GR |
6425 | if (err) |
6426 | md_cluster_ops->add_new_disk_cancel(mddev); | |
6427 | else | |
6428 | err = add_bound_rdev(rdev); | |
6429 | } | |
6430 | ||
6431 | } else if (!err) | |
a6da4ef8 | 6432 | err = add_bound_rdev(rdev); |
dbb64f86 | 6433 | |
1da177e4 LT |
6434 | return err; |
6435 | } | |
6436 | ||
6437 | /* otherwise, add_new_disk is only allowed | |
6438 | * for major_version==0 superblocks | |
6439 | */ | |
6440 | if (mddev->major_version != 0) { | |
9d48739e | 6441 | pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev)); |
1da177e4 LT |
6442 | return -EINVAL; |
6443 | } | |
6444 | ||
6445 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
6446 | int err; | |
d710e138 | 6447 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6448 | if (IS_ERR(rdev)) { |
9d48739e | 6449 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6450 | PTR_ERR(rdev)); |
6451 | return PTR_ERR(rdev); | |
6452 | } | |
6453 | rdev->desc_nr = info->number; | |
6454 | if (info->raid_disk < mddev->raid_disks) | |
6455 | rdev->raid_disk = info->raid_disk; | |
6456 | else | |
6457 | rdev->raid_disk = -1; | |
6458 | ||
1da177e4 | 6459 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
6460 | if (info->state & (1<<MD_DISK_SYNC)) |
6461 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 6462 | |
8ddf9efe N |
6463 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6464 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6465 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6466 | set_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6467 | |
1da177e4 | 6468 | if (!mddev->persistent) { |
9d48739e | 6469 | pr_debug("md: nonpersistent superblock ...\n"); |
77304d2a MS |
6470 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
6471 | } else | |
57b2caa3 | 6472 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 6473 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6474 | |
2bf071bf N |
6475 | err = bind_rdev_to_array(rdev, mddev); |
6476 | if (err) { | |
6477 | export_rdev(rdev); | |
6478 | return err; | |
6479 | } | |
1da177e4 LT |
6480 | } |
6481 | ||
6482 | return 0; | |
6483 | } | |
6484 | ||
f72ffdd6 | 6485 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6486 | { |
6487 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 6488 | struct md_rdev *rdev; |
1da177e4 | 6489 | |
1da177e4 LT |
6490 | rdev = find_rdev(mddev, dev); |
6491 | if (!rdev) | |
6492 | return -ENXIO; | |
6493 | ||
2910ff17 GR |
6494 | if (rdev->raid_disk < 0) |
6495 | goto kick_rdev; | |
293467aa | 6496 | |
3ea8929d N |
6497 | clear_bit(Blocked, &rdev->flags); |
6498 | remove_and_add_spares(mddev, rdev); | |
6499 | ||
1da177e4 LT |
6500 | if (rdev->raid_disk >= 0) |
6501 | goto busy; | |
6502 | ||
2910ff17 | 6503 | kick_rdev: |
54a88392 | 6504 | if (mddev_is_clustered(mddev)) |
88bcfef7 GR |
6505 | md_cluster_ops->remove_disk(mddev, rdev); |
6506 | ||
fb56dfef | 6507 | md_kick_rdev_from_array(rdev); |
2953079c | 6508 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6509 | if (mddev->thread) |
6510 | md_wakeup_thread(mddev->thread); | |
6511 | else | |
6512 | md_update_sb(mddev, 1); | |
d7603b7e | 6513 | md_new_event(mddev); |
1da177e4 LT |
6514 | |
6515 | return 0; | |
6516 | busy: | |
9d48739e N |
6517 | pr_debug("md: cannot remove active disk %s from %s ...\n", |
6518 | bdevname(rdev->bdev,b), mdname(mddev)); | |
1da177e4 LT |
6519 | return -EBUSY; |
6520 | } | |
6521 | ||
f72ffdd6 | 6522 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6523 | { |
6524 | char b[BDEVNAME_SIZE]; | |
6525 | int err; | |
3cb03002 | 6526 | struct md_rdev *rdev; |
1da177e4 LT |
6527 | |
6528 | if (!mddev->pers) | |
6529 | return -ENODEV; | |
6530 | ||
6531 | if (mddev->major_version != 0) { | |
9d48739e | 6532 | pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n", |
1da177e4 LT |
6533 | mdname(mddev)); |
6534 | return -EINVAL; | |
6535 | } | |
6536 | if (!mddev->pers->hot_add_disk) { | |
9d48739e | 6537 | pr_warn("%s: personality does not support diskops!\n", |
1da177e4 LT |
6538 | mdname(mddev)); |
6539 | return -EINVAL; | |
6540 | } | |
6541 | ||
d710e138 | 6542 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6543 | if (IS_ERR(rdev)) { |
9d48739e | 6544 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6545 | PTR_ERR(rdev)); |
6546 | return -EINVAL; | |
6547 | } | |
6548 | ||
6549 | if (mddev->persistent) | |
57b2caa3 | 6550 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 6551 | else |
77304d2a | 6552 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 6553 | |
8190e754 | 6554 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6555 | |
b2d444d7 | 6556 | if (test_bit(Faulty, &rdev->flags)) { |
9d48739e | 6557 | pr_warn("md: can not hot-add faulty %s disk to %s!\n", |
1da177e4 LT |
6558 | bdevname(rdev->bdev,b), mdname(mddev)); |
6559 | err = -EINVAL; | |
6560 | goto abort_export; | |
6561 | } | |
293467aa | 6562 | |
b2d444d7 | 6563 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6564 | rdev->desc_nr = -1; |
5842730d | 6565 | rdev->saved_raid_disk = -1; |
2bf071bf N |
6566 | err = bind_rdev_to_array(rdev, mddev); |
6567 | if (err) | |
2aa82191 | 6568 | goto abort_export; |
1da177e4 LT |
6569 | |
6570 | /* | |
6571 | * The rest should better be atomic, we can have disk failures | |
6572 | * noticed in interrupt contexts ... | |
6573 | */ | |
6574 | ||
1da177e4 LT |
6575 | rdev->raid_disk = -1; |
6576 | ||
2953079c | 6577 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6578 | if (!mddev->thread) |
6579 | md_update_sb(mddev, 1); | |
1da177e4 LT |
6580 | /* |
6581 | * Kick recovery, maybe this spare has to be added to the | |
6582 | * array immediately. | |
6583 | */ | |
6584 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6585 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 6586 | md_new_event(mddev); |
1da177e4 LT |
6587 | return 0; |
6588 | ||
1da177e4 LT |
6589 | abort_export: |
6590 | export_rdev(rdev); | |
6591 | return err; | |
6592 | } | |
6593 | ||
fd01b88c | 6594 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 6595 | { |
035328c2 | 6596 | int err = 0; |
32a7627c | 6597 | |
36fa3063 | 6598 | if (mddev->pers) { |
d66b1b39 | 6599 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
6600 | return -EBUSY; |
6601 | if (mddev->recovery || mddev->sync_thread) | |
6602 | return -EBUSY; | |
6603 | /* we should be able to change the bitmap.. */ | |
6604 | } | |
32a7627c | 6605 | |
36fa3063 | 6606 | if (fd >= 0) { |
035328c2 | 6607 | struct inode *inode; |
1e594bb2 N |
6608 | struct file *f; |
6609 | ||
6610 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 6611 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6612 | f = fget(fd); |
32a7627c | 6613 | |
1e594bb2 | 6614 | if (f == NULL) { |
9d48739e N |
6615 | pr_warn("%s: error: failed to get bitmap file\n", |
6616 | mdname(mddev)); | |
36fa3063 N |
6617 | return -EBADF; |
6618 | } | |
6619 | ||
1e594bb2 | 6620 | inode = f->f_mapping->host; |
035328c2 | 6621 | if (!S_ISREG(inode->i_mode)) { |
9d48739e N |
6622 | pr_warn("%s: error: bitmap file must be a regular file\n", |
6623 | mdname(mddev)); | |
035328c2 | 6624 | err = -EBADF; |
1e594bb2 | 6625 | } else if (!(f->f_mode & FMODE_WRITE)) { |
9d48739e N |
6626 | pr_warn("%s: error: bitmap file must open for write\n", |
6627 | mdname(mddev)); | |
035328c2 N |
6628 | err = -EBADF; |
6629 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
9d48739e N |
6630 | pr_warn("%s: error: bitmap file is already in use\n", |
6631 | mdname(mddev)); | |
035328c2 N |
6632 | err = -EBUSY; |
6633 | } | |
6634 | if (err) { | |
1e594bb2 | 6635 | fput(f); |
36fa3063 N |
6636 | return err; |
6637 | } | |
1e594bb2 | 6638 | mddev->bitmap_info.file = f; |
c3d9714e | 6639 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6640 | } else if (mddev->bitmap == NULL) |
6641 | return -ENOENT; /* cannot remove what isn't there */ | |
6642 | err = 0; | |
6643 | if (mddev->pers) { | |
69e51b44 | 6644 | if (fd >= 0) { |
f9209a32 GR |
6645 | struct bitmap *bitmap; |
6646 | ||
6647 | bitmap = bitmap_create(mddev, -1); | |
9e1cc0a5 | 6648 | mddev_suspend(mddev); |
f9209a32 GR |
6649 | if (!IS_ERR(bitmap)) { |
6650 | mddev->bitmap = bitmap; | |
69e51b44 | 6651 | err = bitmap_load(mddev); |
ba599aca N |
6652 | } else |
6653 | err = PTR_ERR(bitmap); | |
52a0d49d N |
6654 | if (err) { |
6655 | bitmap_destroy(mddev); | |
6656 | fd = -1; | |
6657 | } | |
9e1cc0a5 | 6658 | mddev_resume(mddev); |
52a0d49d | 6659 | } else if (fd < 0) { |
9e1cc0a5 | 6660 | mddev_suspend(mddev); |
36fa3063 | 6661 | bitmap_destroy(mddev); |
9e1cc0a5 | 6662 | mddev_resume(mddev); |
d7375ab3 | 6663 | } |
d7375ab3 N |
6664 | } |
6665 | if (fd < 0) { | |
4af1a041 N |
6666 | struct file *f = mddev->bitmap_info.file; |
6667 | if (f) { | |
6668 | spin_lock(&mddev->lock); | |
6669 | mddev->bitmap_info.file = NULL; | |
6670 | spin_unlock(&mddev->lock); | |
6671 | fput(f); | |
6672 | } | |
36fa3063 N |
6673 | } |
6674 | ||
32a7627c N |
6675 | return err; |
6676 | } | |
6677 | ||
1da177e4 LT |
6678 | /* |
6679 | * set_array_info is used two different ways | |
6680 | * The original usage is when creating a new array. | |
6681 | * In this usage, raid_disks is > 0 and it together with | |
6682 | * level, size, not_persistent,layout,chunksize determine the | |
6683 | * shape of the array. | |
6684 | * This will always create an array with a type-0.90.0 superblock. | |
6685 | * The newer usage is when assembling an array. | |
6686 | * In this case raid_disks will be 0, and the major_version field is | |
6687 | * use to determine which style super-blocks are to be found on the devices. | |
6688 | * The minor and patch _version numbers are also kept incase the | |
6689 | * super_block handler wishes to interpret them. | |
6690 | */ | |
f72ffdd6 | 6691 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6692 | { |
6693 | ||
6694 | if (info->raid_disks == 0) { | |
6695 | /* just setting version number for superblock loading */ | |
6696 | if (info->major_version < 0 || | |
50511da3 | 6697 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6698 | super_types[info->major_version].name == NULL) { |
6699 | /* maybe try to auto-load a module? */ | |
9d48739e | 6700 | pr_warn("md: superblock version %d not known\n", |
1da177e4 LT |
6701 | info->major_version); |
6702 | return -EINVAL; | |
6703 | } | |
6704 | mddev->major_version = info->major_version; | |
6705 | mddev->minor_version = info->minor_version; | |
6706 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6707 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6708 | /* ensure mddev_put doesn't delete this now that there |
6709 | * is some minimal configuration. | |
6710 | */ | |
9ebc6ef1 | 6711 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6712 | return 0; |
6713 | } | |
6714 | mddev->major_version = MD_MAJOR_VERSION; | |
6715 | mddev->minor_version = MD_MINOR_VERSION; | |
6716 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6717 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6718 | |
6719 | mddev->level = info->level; | |
17115e03 | 6720 | mddev->clevel[0] = 0; |
58c0fed4 | 6721 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6722 | mddev->raid_disks = info->raid_disks; |
6723 | /* don't set md_minor, it is determined by which /dev/md* was | |
6724 | * openned | |
6725 | */ | |
6726 | if (info->state & (1<<MD_SB_CLEAN)) | |
6727 | mddev->recovery_cp = MaxSector; | |
6728 | else | |
6729 | mddev->recovery_cp = 0; | |
6730 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6731 | mddev->external = 0; |
1da177e4 LT |
6732 | |
6733 | mddev->layout = info->layout; | |
9d8f0363 | 6734 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 | 6735 | |
2953079c | 6736 | if (mddev->persistent) { |
1b3bae49 N |
6737 | mddev->max_disks = MD_SB_DISKS; |
6738 | mddev->flags = 0; | |
6739 | mddev->sb_flags = 0; | |
2953079c SL |
6740 | } |
6741 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
1da177e4 | 6742 | |
c3d9714e | 6743 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6744 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6745 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6746 | |
f6705578 N |
6747 | mddev->reshape_position = MaxSector; |
6748 | ||
1da177e4 LT |
6749 | /* |
6750 | * Generate a 128 bit UUID | |
6751 | */ | |
6752 | get_random_bytes(mddev->uuid, 16); | |
6753 | ||
f6705578 | 6754 | mddev->new_level = mddev->level; |
664e7c41 | 6755 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6756 | mddev->new_layout = mddev->layout; |
6757 | mddev->delta_disks = 0; | |
2c810cdd | 6758 | mddev->reshape_backwards = 0; |
f6705578 | 6759 | |
1da177e4 LT |
6760 | return 0; |
6761 | } | |
6762 | ||
fd01b88c | 6763 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6764 | { |
efa4b77b | 6765 | lockdep_assert_held(&mddev->reconfig_mutex); |
b522adcd DW |
6766 | |
6767 | if (mddev->external_size) | |
6768 | return; | |
6769 | ||
1f403624 DW |
6770 | mddev->array_sectors = array_sectors; |
6771 | } | |
6772 | EXPORT_SYMBOL(md_set_array_sectors); | |
6773 | ||
fd01b88c | 6774 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6775 | { |
3cb03002 | 6776 | struct md_rdev *rdev; |
a35b0d69 | 6777 | int rv; |
d71f9f88 | 6778 | int fit = (num_sectors == 0); |
818da59f | 6779 | sector_t old_dev_sectors = mddev->dev_sectors; |
ab5a98b1 | 6780 | |
a35b0d69 N |
6781 | if (mddev->pers->resize == NULL) |
6782 | return -EINVAL; | |
d71f9f88 AN |
6783 | /* The "num_sectors" is the number of sectors of each device that |
6784 | * is used. This can only make sense for arrays with redundancy. | |
6785 | * linear and raid0 always use whatever space is available. We can only | |
6786 | * consider changing this number if no resync or reconstruction is | |
6787 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6788 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6789 | * of each device. If num_sectors is zero, we find the largest size |
6790 | * that fits. | |
a35b0d69 | 6791 | */ |
f851b60d N |
6792 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6793 | mddev->sync_thread) | |
a35b0d69 | 6794 | return -EBUSY; |
bd8839e0 N |
6795 | if (mddev->ro) |
6796 | return -EROFS; | |
a4a6125a | 6797 | |
dafb20fa | 6798 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6799 | sector_t avail = rdev->sectors; |
01ab5662 | 6800 | |
d71f9f88 AN |
6801 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6802 | num_sectors = avail; | |
6803 | if (avail < num_sectors) | |
a35b0d69 N |
6804 | return -ENOSPC; |
6805 | } | |
d71f9f88 | 6806 | rv = mddev->pers->resize(mddev, num_sectors); |
c9483634 | 6807 | if (!rv) { |
818da59f GJ |
6808 | if (mddev_is_clustered(mddev)) |
6809 | md_cluster_ops->update_size(mddev, old_dev_sectors); | |
6810 | else if (mddev->queue) { | |
c9483634 GJ |
6811 | set_capacity(mddev->gendisk, mddev->array_sectors); |
6812 | revalidate_disk(mddev->gendisk); | |
6813 | } | |
6814 | } | |
a35b0d69 N |
6815 | return rv; |
6816 | } | |
6817 | ||
fd01b88c | 6818 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6819 | { |
6820 | int rv; | |
c6563a8c | 6821 | struct md_rdev *rdev; |
da943b99 | 6822 | /* change the number of raid disks */ |
63c70c4f | 6823 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6824 | return -EINVAL; |
bd8839e0 N |
6825 | if (mddev->ro) |
6826 | return -EROFS; | |
da943b99 | 6827 | if (raid_disks <= 0 || |
233fca36 | 6828 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6829 | return -EINVAL; |
f851b60d N |
6830 | if (mddev->sync_thread || |
6831 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6832 | mddev->reshape_position != MaxSector) | |
da943b99 | 6833 | return -EBUSY; |
c6563a8c N |
6834 | |
6835 | rdev_for_each(rdev, mddev) { | |
6836 | if (mddev->raid_disks < raid_disks && | |
6837 | rdev->data_offset < rdev->new_data_offset) | |
6838 | return -EINVAL; | |
6839 | if (mddev->raid_disks > raid_disks && | |
6840 | rdev->data_offset > rdev->new_data_offset) | |
6841 | return -EINVAL; | |
6842 | } | |
6843 | ||
63c70c4f | 6844 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6845 | if (mddev->delta_disks < 0) |
6846 | mddev->reshape_backwards = 1; | |
6847 | else if (mddev->delta_disks > 0) | |
6848 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6849 | |
6850 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6851 | if (rv < 0) { |
de171cb9 | 6852 | mddev->delta_disks = 0; |
2c810cdd N |
6853 | mddev->reshape_backwards = 0; |
6854 | } | |
da943b99 N |
6855 | return rv; |
6856 | } | |
6857 | ||
1da177e4 LT |
6858 | /* |
6859 | * update_array_info is used to change the configuration of an | |
6860 | * on-line array. | |
6861 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6862 | * fields in the info are checked against the array. | |
6863 | * Any differences that cannot be handled will cause an error. | |
6864 | * Normally, only one change can be managed at a time. | |
6865 | */ | |
fd01b88c | 6866 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6867 | { |
6868 | int rv = 0; | |
6869 | int cnt = 0; | |
36fa3063 N |
6870 | int state = 0; |
6871 | ||
6872 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6873 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6874 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6875 | |
6876 | if (mddev->major_version != info->major_version || | |
6877 | mddev->minor_version != info->minor_version || | |
6878 | /* mddev->patch_version != info->patch_version || */ | |
6879 | mddev->ctime != info->ctime || | |
6880 | mddev->level != info->level || | |
6881 | /* mddev->layout != info->layout || */ | |
4e023612 | 6882 | mddev->persistent != !info->not_persistent || |
9d8f0363 | 6883 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6884 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6885 | ((state^info->state) & 0xfffffe00) | |
6886 | ) | |
1da177e4 LT |
6887 | return -EINVAL; |
6888 | /* Check there is only one change */ | |
58c0fed4 AN |
6889 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6890 | cnt++; | |
6891 | if (mddev->raid_disks != info->raid_disks) | |
6892 | cnt++; | |
6893 | if (mddev->layout != info->layout) | |
6894 | cnt++; | |
6895 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6896 | cnt++; | |
6897 | if (cnt == 0) | |
6898 | return 0; | |
6899 | if (cnt > 1) | |
6900 | return -EINVAL; | |
1da177e4 LT |
6901 | |
6902 | if (mddev->layout != info->layout) { | |
6903 | /* Change layout | |
6904 | * we don't need to do anything at the md level, the | |
6905 | * personality will take care of it all. | |
6906 | */ | |
50ac168a | 6907 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6908 | return -EINVAL; |
597a711b N |
6909 | else { |
6910 | mddev->new_layout = info->layout; | |
50ac168a | 6911 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6912 | if (rv) |
6913 | mddev->new_layout = mddev->layout; | |
6914 | return rv; | |
6915 | } | |
1da177e4 | 6916 | } |
58c0fed4 | 6917 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6918 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6919 | |
da943b99 N |
6920 | if (mddev->raid_disks != info->raid_disks) |
6921 | rv = update_raid_disks(mddev, info->raid_disks); | |
6922 | ||
36fa3063 | 6923 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
293467aa GR |
6924 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) { |
6925 | rv = -EINVAL; | |
6926 | goto err; | |
6927 | } | |
6928 | if (mddev->recovery || mddev->sync_thread) { | |
6929 | rv = -EBUSY; | |
6930 | goto err; | |
6931 | } | |
36fa3063 | 6932 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { |
f9209a32 | 6933 | struct bitmap *bitmap; |
36fa3063 | 6934 | /* add the bitmap */ |
293467aa GR |
6935 | if (mddev->bitmap) { |
6936 | rv = -EEXIST; | |
6937 | goto err; | |
6938 | } | |
6939 | if (mddev->bitmap_info.default_offset == 0) { | |
6940 | rv = -EINVAL; | |
6941 | goto err; | |
6942 | } | |
c3d9714e N |
6943 | mddev->bitmap_info.offset = |
6944 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6945 | mddev->bitmap_info.space = |
6946 | mddev->bitmap_info.default_space; | |
f9209a32 | 6947 | bitmap = bitmap_create(mddev, -1); |
9e1cc0a5 | 6948 | mddev_suspend(mddev); |
f9209a32 GR |
6949 | if (!IS_ERR(bitmap)) { |
6950 | mddev->bitmap = bitmap; | |
69e51b44 | 6951 | rv = bitmap_load(mddev); |
ba599aca N |
6952 | } else |
6953 | rv = PTR_ERR(bitmap); | |
36fa3063 N |
6954 | if (rv) |
6955 | bitmap_destroy(mddev); | |
9e1cc0a5 | 6956 | mddev_resume(mddev); |
36fa3063 N |
6957 | } else { |
6958 | /* remove the bitmap */ | |
293467aa GR |
6959 | if (!mddev->bitmap) { |
6960 | rv = -ENOENT; | |
6961 | goto err; | |
6962 | } | |
6963 | if (mddev->bitmap->storage.file) { | |
6964 | rv = -EINVAL; | |
6965 | goto err; | |
6966 | } | |
f6a2dc64 GJ |
6967 | if (mddev->bitmap_info.nodes) { |
6968 | /* hold PW on all the bitmap lock */ | |
6969 | if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) { | |
9d48739e | 6970 | pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n"); |
f6a2dc64 GJ |
6971 | rv = -EPERM; |
6972 | md_cluster_ops->unlock_all_bitmaps(mddev); | |
6973 | goto err; | |
6974 | } | |
6975 | ||
6976 | mddev->bitmap_info.nodes = 0; | |
6977 | md_cluster_ops->leave(mddev); | |
6978 | } | |
9e1cc0a5 | 6979 | mddev_suspend(mddev); |
36fa3063 | 6980 | bitmap_destroy(mddev); |
9e1cc0a5 | 6981 | mddev_resume(mddev); |
c3d9714e | 6982 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6983 | } |
6984 | } | |
850b2b42 | 6985 | md_update_sb(mddev, 1); |
293467aa GR |
6986 | return rv; |
6987 | err: | |
1da177e4 LT |
6988 | return rv; |
6989 | } | |
6990 | ||
fd01b88c | 6991 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6992 | { |
3cb03002 | 6993 | struct md_rdev *rdev; |
1ca69c4b | 6994 | int err = 0; |
1da177e4 LT |
6995 | |
6996 | if (mddev->pers == NULL) | |
6997 | return -ENODEV; | |
6998 | ||
1ca69c4b N |
6999 | rcu_read_lock(); |
7000 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 7001 | if (!rdev) |
1ca69c4b N |
7002 | err = -ENODEV; |
7003 | else { | |
7004 | md_error(mddev, rdev); | |
7005 | if (!test_bit(Faulty, &rdev->flags)) | |
7006 | err = -EBUSY; | |
7007 | } | |
7008 | rcu_read_unlock(); | |
7009 | return err; | |
1da177e4 LT |
7010 | } |
7011 | ||
2f9618ce AN |
7012 | /* |
7013 | * We have a problem here : there is no easy way to give a CHS | |
7014 | * virtual geometry. We currently pretend that we have a 2 heads | |
7015 | * 4 sectors (with a BIG number of cylinders...). This drives | |
7016 | * dosfs just mad... ;-) | |
7017 | */ | |
a885c8c4 CH |
7018 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
7019 | { | |
fd01b88c | 7020 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
7021 | |
7022 | geo->heads = 2; | |
7023 | geo->sectors = 4; | |
49ce6cea | 7024 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
7025 | return 0; |
7026 | } | |
7027 | ||
cb335f88 NS |
7028 | static inline bool md_ioctl_valid(unsigned int cmd) |
7029 | { | |
7030 | switch (cmd) { | |
7031 | case ADD_NEW_DISK: | |
7032 | case BLKROSET: | |
7033 | case GET_ARRAY_INFO: | |
7034 | case GET_BITMAP_FILE: | |
7035 | case GET_DISK_INFO: | |
7036 | case HOT_ADD_DISK: | |
7037 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
7038 | case RAID_AUTORUN: |
7039 | case RAID_VERSION: | |
7040 | case RESTART_ARRAY_RW: | |
7041 | case RUN_ARRAY: | |
7042 | case SET_ARRAY_INFO: | |
7043 | case SET_BITMAP_FILE: | |
7044 | case SET_DISK_FAULTY: | |
7045 | case STOP_ARRAY: | |
7046 | case STOP_ARRAY_RO: | |
1aee41f6 | 7047 | case CLUSTERED_DISK_NACK: |
cb335f88 NS |
7048 | return true; |
7049 | default: | |
7050 | return false; | |
7051 | } | |
7052 | } | |
7053 | ||
a39907fa | 7054 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
7055 | unsigned int cmd, unsigned long arg) |
7056 | { | |
7057 | int err = 0; | |
7058 | void __user *argp = (void __user *)arg; | |
fd01b88c | 7059 | struct mddev *mddev = NULL; |
e2218350 | 7060 | int ro; |
065e519e | 7061 | bool did_set_md_closing = false; |
1da177e4 | 7062 | |
cb335f88 NS |
7063 | if (!md_ioctl_valid(cmd)) |
7064 | return -ENOTTY; | |
7065 | ||
506c9e44 N |
7066 | switch (cmd) { |
7067 | case RAID_VERSION: | |
7068 | case GET_ARRAY_INFO: | |
7069 | case GET_DISK_INFO: | |
7070 | break; | |
7071 | default: | |
7072 | if (!capable(CAP_SYS_ADMIN)) | |
7073 | return -EACCES; | |
7074 | } | |
1da177e4 LT |
7075 | |
7076 | /* | |
7077 | * Commands dealing with the RAID driver but not any | |
7078 | * particular array: | |
7079 | */ | |
c02c0aeb N |
7080 | switch (cmd) { |
7081 | case RAID_VERSION: | |
7082 | err = get_version(argp); | |
3adc28d8 | 7083 | goto out; |
1da177e4 | 7084 | |
1da177e4 | 7085 | #ifndef MODULE |
c02c0aeb N |
7086 | case RAID_AUTORUN: |
7087 | err = 0; | |
7088 | autostart_arrays(arg); | |
3adc28d8 | 7089 | goto out; |
1da177e4 | 7090 | #endif |
c02c0aeb | 7091 | default:; |
1da177e4 LT |
7092 | } |
7093 | ||
7094 | /* | |
7095 | * Commands creating/starting a new array: | |
7096 | */ | |
7097 | ||
a39907fa | 7098 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
7099 | |
7100 | if (!mddev) { | |
7101 | BUG(); | |
3adc28d8 | 7102 | goto out; |
1da177e4 LT |
7103 | } |
7104 | ||
1ca69c4b N |
7105 | /* Some actions do not requires the mutex */ |
7106 | switch (cmd) { | |
7107 | case GET_ARRAY_INFO: | |
7108 | if (!mddev->raid_disks && !mddev->external) | |
7109 | err = -ENODEV; | |
7110 | else | |
7111 | err = get_array_info(mddev, argp); | |
3adc28d8 | 7112 | goto out; |
1ca69c4b N |
7113 | |
7114 | case GET_DISK_INFO: | |
7115 | if (!mddev->raid_disks && !mddev->external) | |
7116 | err = -ENODEV; | |
7117 | else | |
7118 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 7119 | goto out; |
1ca69c4b N |
7120 | |
7121 | case SET_DISK_FAULTY: | |
7122 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7123 | goto out; |
4af1a041 N |
7124 | |
7125 | case GET_BITMAP_FILE: | |
7126 | err = get_bitmap_file(mddev, argp); | |
7127 | goto out; | |
7128 | ||
1ca69c4b N |
7129 | } |
7130 | ||
a7a3f08d N |
7131 | if (cmd == ADD_NEW_DISK) |
7132 | /* need to ensure md_delayed_delete() has completed */ | |
7133 | flush_workqueue(md_misc_wq); | |
7134 | ||
90f5f7ad HR |
7135 | if (cmd == HOT_REMOVE_DISK) |
7136 | /* need to ensure recovery thread has run */ | |
7137 | wait_event_interruptible_timeout(mddev->sb_wait, | |
7138 | !test_bit(MD_RECOVERY_NEEDED, | |
82a301cb | 7139 | &mddev->recovery), |
90f5f7ad | 7140 | msecs_to_jiffies(5000)); |
260fa034 N |
7141 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
7142 | /* Need to flush page cache, and ensure no-one else opens | |
7143 | * and writes | |
7144 | */ | |
7145 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 7146 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
7147 | mutex_unlock(&mddev->open_mutex); |
7148 | err = -EBUSY; | |
3adc28d8 | 7149 | goto out; |
260fa034 | 7150 | } |
065e519e | 7151 | WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags)); |
af8d8e6f | 7152 | set_bit(MD_CLOSING, &mddev->flags); |
065e519e | 7153 | did_set_md_closing = true; |
260fa034 N |
7154 | mutex_unlock(&mddev->open_mutex); |
7155 | sync_blockdev(bdev); | |
7156 | } | |
1da177e4 LT |
7157 | err = mddev_lock(mddev); |
7158 | if (err) { | |
9d48739e N |
7159 | pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n", |
7160 | err, cmd); | |
3adc28d8 | 7161 | goto out; |
1da177e4 LT |
7162 | } |
7163 | ||
c02c0aeb N |
7164 | if (cmd == SET_ARRAY_INFO) { |
7165 | mdu_array_info_t info; | |
7166 | if (!arg) | |
7167 | memset(&info, 0, sizeof(info)); | |
7168 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
7169 | err = -EFAULT; | |
3adc28d8 | 7170 | goto unlock; |
c02c0aeb N |
7171 | } |
7172 | if (mddev->pers) { | |
7173 | err = update_array_info(mddev, &info); | |
7174 | if (err) { | |
9d48739e | 7175 | pr_warn("md: couldn't update array info. %d\n", err); |
3adc28d8 | 7176 | goto unlock; |
1da177e4 | 7177 | } |
3adc28d8 | 7178 | goto unlock; |
c02c0aeb N |
7179 | } |
7180 | if (!list_empty(&mddev->disks)) { | |
9d48739e | 7181 | pr_warn("md: array %s already has disks!\n", mdname(mddev)); |
c02c0aeb | 7182 | err = -EBUSY; |
3adc28d8 | 7183 | goto unlock; |
c02c0aeb N |
7184 | } |
7185 | if (mddev->raid_disks) { | |
9d48739e | 7186 | pr_warn("md: array %s already initialised!\n", mdname(mddev)); |
c02c0aeb | 7187 | err = -EBUSY; |
3adc28d8 | 7188 | goto unlock; |
c02c0aeb N |
7189 | } |
7190 | err = set_array_info(mddev, &info); | |
7191 | if (err) { | |
9d48739e | 7192 | pr_warn("md: couldn't set array info. %d\n", err); |
3adc28d8 | 7193 | goto unlock; |
c02c0aeb | 7194 | } |
3adc28d8 | 7195 | goto unlock; |
1da177e4 LT |
7196 | } |
7197 | ||
7198 | /* | |
7199 | * Commands querying/configuring an existing array: | |
7200 | */ | |
32a7627c | 7201 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 7202 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
7203 | if ((!mddev->raid_disks && !mddev->external) |
7204 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
7205 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
7206 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 7207 | err = -ENODEV; |
3adc28d8 | 7208 | goto unlock; |
1da177e4 LT |
7209 | } |
7210 | ||
7211 | /* | |
7212 | * Commands even a read-only array can execute: | |
7213 | */ | |
c02c0aeb | 7214 | switch (cmd) { |
c02c0aeb N |
7215 | case RESTART_ARRAY_RW: |
7216 | err = restart_array(mddev); | |
3adc28d8 | 7217 | goto unlock; |
1da177e4 | 7218 | |
c02c0aeb N |
7219 | case STOP_ARRAY: |
7220 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 7221 | goto unlock; |
1da177e4 | 7222 | |
c02c0aeb N |
7223 | case STOP_ARRAY_RO: |
7224 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 7225 | goto unlock; |
1da177e4 | 7226 | |
3ea8929d N |
7227 | case HOT_REMOVE_DISK: |
7228 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7229 | goto unlock; |
3ea8929d | 7230 | |
7ceb17e8 N |
7231 | case ADD_NEW_DISK: |
7232 | /* We can support ADD_NEW_DISK on read-only arrays | |
466ad292 | 7233 | * only if we are re-adding a preexisting device. |
7ceb17e8 N |
7234 | * So require mddev->pers and MD_DISK_SYNC. |
7235 | */ | |
7236 | if (mddev->pers) { | |
7237 | mdu_disk_info_t info; | |
7238 | if (copy_from_user(&info, argp, sizeof(info))) | |
7239 | err = -EFAULT; | |
7240 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
7241 | /* Need to clear read-only for this */ | |
7242 | break; | |
7243 | else | |
7244 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7245 | goto unlock; |
7ceb17e8 N |
7246 | } |
7247 | break; | |
7248 | ||
c02c0aeb N |
7249 | case BLKROSET: |
7250 | if (get_user(ro, (int __user *)(arg))) { | |
7251 | err = -EFAULT; | |
3adc28d8 | 7252 | goto unlock; |
c02c0aeb N |
7253 | } |
7254 | err = -EINVAL; | |
e2218350 | 7255 | |
c02c0aeb N |
7256 | /* if the bdev is going readonly the value of mddev->ro |
7257 | * does not matter, no writes are coming | |
7258 | */ | |
7259 | if (ro) | |
3adc28d8 | 7260 | goto unlock; |
e2218350 | 7261 | |
c02c0aeb N |
7262 | /* are we are already prepared for writes? */ |
7263 | if (mddev->ro != 1) | |
3adc28d8 | 7264 | goto unlock; |
e2218350 | 7265 | |
c02c0aeb N |
7266 | /* transitioning to readauto need only happen for |
7267 | * arrays that call md_write_start | |
7268 | */ | |
7269 | if (mddev->pers) { | |
7270 | err = restart_array(mddev); | |
7271 | if (err == 0) { | |
7272 | mddev->ro = 2; | |
7273 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 7274 | } |
c02c0aeb | 7275 | } |
3adc28d8 | 7276 | goto unlock; |
1da177e4 LT |
7277 | } |
7278 | ||
7279 | /* | |
7280 | * The remaining ioctls are changing the state of the | |
f91de92e | 7281 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 7282 | */ |
326eb17d | 7283 | if (mddev->ro && mddev->pers) { |
f91de92e N |
7284 | if (mddev->ro == 2) { |
7285 | mddev->ro = 0; | |
00bcb4ac | 7286 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 7287 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
7288 | /* mddev_unlock will wake thread */ |
7289 | /* If a device failed while we were read-only, we | |
7290 | * need to make sure the metadata is updated now. | |
7291 | */ | |
2953079c | 7292 | if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) { |
f3378b48 N |
7293 | mddev_unlock(mddev); |
7294 | wait_event(mddev->sb_wait, | |
2953079c SL |
7295 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) && |
7296 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); | |
29f097c4 | 7297 | mddev_lock_nointr(mddev); |
f3378b48 | 7298 | } |
f91de92e N |
7299 | } else { |
7300 | err = -EROFS; | |
3adc28d8 | 7301 | goto unlock; |
f91de92e | 7302 | } |
1da177e4 LT |
7303 | } |
7304 | ||
c02c0aeb N |
7305 | switch (cmd) { |
7306 | case ADD_NEW_DISK: | |
1da177e4 | 7307 | { |
c02c0aeb N |
7308 | mdu_disk_info_t info; |
7309 | if (copy_from_user(&info, argp, sizeof(info))) | |
7310 | err = -EFAULT; | |
7311 | else | |
7312 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7313 | goto unlock; |
c02c0aeb | 7314 | } |
1da177e4 | 7315 | |
1aee41f6 GR |
7316 | case CLUSTERED_DISK_NACK: |
7317 | if (mddev_is_clustered(mddev)) | |
7318 | md_cluster_ops->new_disk_ack(mddev, false); | |
7319 | else | |
7320 | err = -EINVAL; | |
7321 | goto unlock; | |
7322 | ||
c02c0aeb N |
7323 | case HOT_ADD_DISK: |
7324 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7325 | goto unlock; |
1da177e4 | 7326 | |
c02c0aeb N |
7327 | case RUN_ARRAY: |
7328 | err = do_md_run(mddev); | |
3adc28d8 | 7329 | goto unlock; |
1da177e4 | 7330 | |
c02c0aeb N |
7331 | case SET_BITMAP_FILE: |
7332 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 7333 | goto unlock; |
32a7627c | 7334 | |
c02c0aeb N |
7335 | default: |
7336 | err = -EINVAL; | |
3adc28d8 | 7337 | goto unlock; |
1da177e4 LT |
7338 | } |
7339 | ||
3adc28d8 | 7340 | unlock: |
d3374825 N |
7341 | if (mddev->hold_active == UNTIL_IOCTL && |
7342 | err != -EINVAL) | |
7343 | mddev->hold_active = 0; | |
1da177e4 | 7344 | mddev_unlock(mddev); |
3adc28d8 | 7345 | out: |
065e519e N |
7346 | if(did_set_md_closing) |
7347 | clear_bit(MD_CLOSING, &mddev->flags); | |
1da177e4 LT |
7348 | return err; |
7349 | } | |
aa98aa31 AB |
7350 | #ifdef CONFIG_COMPAT |
7351 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
7352 | unsigned int cmd, unsigned long arg) | |
7353 | { | |
7354 | switch (cmd) { | |
7355 | case HOT_REMOVE_DISK: | |
7356 | case HOT_ADD_DISK: | |
7357 | case SET_DISK_FAULTY: | |
7358 | case SET_BITMAP_FILE: | |
7359 | /* These take in integer arg, do not convert */ | |
7360 | break; | |
7361 | default: | |
7362 | arg = (unsigned long)compat_ptr(arg); | |
7363 | break; | |
7364 | } | |
7365 | ||
7366 | return md_ioctl(bdev, mode, cmd, arg); | |
7367 | } | |
7368 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 7369 | |
a39907fa | 7370 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
7371 | { |
7372 | /* | |
7373 | * Succeed if we can lock the mddev, which confirms that | |
7374 | * it isn't being stopped right now. | |
7375 | */ | |
fd01b88c | 7376 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
7377 | int err; |
7378 | ||
0c098220 YL |
7379 | if (!mddev) |
7380 | return -ENODEV; | |
7381 | ||
d3374825 N |
7382 | if (mddev->gendisk != bdev->bd_disk) { |
7383 | /* we are racing with mddev_put which is discarding this | |
7384 | * bd_disk. | |
7385 | */ | |
7386 | mddev_put(mddev); | |
7387 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 7388 | flush_workqueue(md_misc_wq); |
d3374825 N |
7389 | /* Then retry the open from the top */ |
7390 | return -ERESTARTSYS; | |
7391 | } | |
7392 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
7393 | ||
c8c00a69 | 7394 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
7395 | goto out; |
7396 | ||
af8d8e6f GJ |
7397 | if (test_bit(MD_CLOSING, &mddev->flags)) { |
7398 | mutex_unlock(&mddev->open_mutex); | |
e2342ca8 N |
7399 | err = -ENODEV; |
7400 | goto out; | |
af8d8e6f GJ |
7401 | } |
7402 | ||
1da177e4 | 7403 | err = 0; |
f2ea68cf | 7404 | atomic_inc(&mddev->openers); |
c8c00a69 | 7405 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 7406 | |
f0b4f7e2 | 7407 | check_disk_change(bdev); |
1da177e4 | 7408 | out: |
e2342ca8 N |
7409 | if (err) |
7410 | mddev_put(mddev); | |
1da177e4 LT |
7411 | return err; |
7412 | } | |
7413 | ||
db2a144b | 7414 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 7415 | { |
f72ffdd6 | 7416 | struct mddev *mddev = disk->private_data; |
1da177e4 | 7417 | |
52e5f9d1 | 7418 | BUG_ON(!mddev); |
f2ea68cf | 7419 | atomic_dec(&mddev->openers); |
1da177e4 | 7420 | mddev_put(mddev); |
1da177e4 | 7421 | } |
f0b4f7e2 N |
7422 | |
7423 | static int md_media_changed(struct gendisk *disk) | |
7424 | { | |
fd01b88c | 7425 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7426 | |
7427 | return mddev->changed; | |
7428 | } | |
7429 | ||
7430 | static int md_revalidate(struct gendisk *disk) | |
7431 | { | |
fd01b88c | 7432 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7433 | |
7434 | mddev->changed = 0; | |
7435 | return 0; | |
7436 | } | |
83d5cde4 | 7437 | static const struct block_device_operations md_fops = |
1da177e4 LT |
7438 | { |
7439 | .owner = THIS_MODULE, | |
a39907fa AV |
7440 | .open = md_open, |
7441 | .release = md_release, | |
b492b852 | 7442 | .ioctl = md_ioctl, |
aa98aa31 AB |
7443 | #ifdef CONFIG_COMPAT |
7444 | .compat_ioctl = md_compat_ioctl, | |
7445 | #endif | |
a885c8c4 | 7446 | .getgeo = md_getgeo, |
f0b4f7e2 N |
7447 | .media_changed = md_media_changed, |
7448 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
7449 | }; |
7450 | ||
f72ffdd6 | 7451 | static int md_thread(void *arg) |
1da177e4 | 7452 | { |
2b8bf345 | 7453 | struct md_thread *thread = arg; |
1da177e4 | 7454 | |
1da177e4 LT |
7455 | /* |
7456 | * md_thread is a 'system-thread', it's priority should be very | |
7457 | * high. We avoid resource deadlocks individually in each | |
7458 | * raid personality. (RAID5 does preallocation) We also use RR and | |
7459 | * the very same RT priority as kswapd, thus we will never get | |
7460 | * into a priority inversion deadlock. | |
7461 | * | |
7462 | * we definitely have to have equal or higher priority than | |
7463 | * bdflush, otherwise bdflush will deadlock if there are too | |
7464 | * many dirty RAID5 blocks. | |
7465 | */ | |
1da177e4 | 7466 | |
6985c43f | 7467 | allow_signal(SIGKILL); |
a6fb0934 | 7468 | while (!kthread_should_stop()) { |
1da177e4 | 7469 | |
93588e22 N |
7470 | /* We need to wait INTERRUPTIBLE so that |
7471 | * we don't add to the load-average. | |
7472 | * That means we need to be sure no signals are | |
7473 | * pending | |
7474 | */ | |
7475 | if (signal_pending(current)) | |
7476 | flush_signals(current); | |
7477 | ||
7478 | wait_event_interruptible_timeout | |
7479 | (thread->wqueue, | |
7480 | test_bit(THREAD_WAKEUP, &thread->flags) | |
ce1ccd07 | 7481 | || kthread_should_stop() || kthread_should_park(), |
93588e22 | 7482 | thread->timeout); |
1da177e4 | 7483 | |
6c987910 | 7484 | clear_bit(THREAD_WAKEUP, &thread->flags); |
ce1ccd07 SL |
7485 | if (kthread_should_park()) |
7486 | kthread_parkme(); | |
6c987910 | 7487 | if (!kthread_should_stop()) |
4ed8731d | 7488 | thread->run(thread); |
1da177e4 | 7489 | } |
a6fb0934 | 7490 | |
1da177e4 LT |
7491 | return 0; |
7492 | } | |
7493 | ||
2b8bf345 | 7494 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
7495 | { |
7496 | if (thread) { | |
36a4e1fe | 7497 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
d1d90147 GJ |
7498 | set_bit(THREAD_WAKEUP, &thread->flags); |
7499 | wake_up(&thread->wqueue); | |
1da177e4 LT |
7500 | } |
7501 | } | |
6c144d31 | 7502 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 7503 | |
4ed8731d SL |
7504 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
7505 | struct mddev *mddev, const char *name) | |
1da177e4 | 7506 | { |
2b8bf345 | 7507 | struct md_thread *thread; |
1da177e4 | 7508 | |
2b8bf345 | 7509 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
7510 | if (!thread) |
7511 | return NULL; | |
7512 | ||
1da177e4 LT |
7513 | init_waitqueue_head(&thread->wqueue); |
7514 | ||
1da177e4 LT |
7515 | thread->run = run; |
7516 | thread->mddev = mddev; | |
32a7627c | 7517 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
7518 | thread->tsk = kthread_run(md_thread, thread, |
7519 | "%s_%s", | |
7520 | mdname(thread->mddev), | |
0232605d | 7521 | name); |
a6fb0934 | 7522 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
7523 | kfree(thread); |
7524 | return NULL; | |
7525 | } | |
1da177e4 LT |
7526 | return thread; |
7527 | } | |
6c144d31 | 7528 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 7529 | |
2b8bf345 | 7530 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 7531 | { |
2b8bf345 | 7532 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
7533 | if (!thread) |
7534 | return; | |
36a4e1fe | 7535 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
7536 | /* Locking ensures that mddev_unlock does not wake_up a |
7537 | * non-existent thread | |
7538 | */ | |
7539 | spin_lock(&pers_lock); | |
7540 | *threadp = NULL; | |
7541 | spin_unlock(&pers_lock); | |
a6fb0934 N |
7542 | |
7543 | kthread_stop(thread->tsk); | |
1da177e4 LT |
7544 | kfree(thread); |
7545 | } | |
6c144d31 | 7546 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 7547 | |
fd01b88c | 7548 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7549 | { |
b2d444d7 | 7550 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 7551 | return; |
6bfe0b49 | 7552 | |
de393cde | 7553 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
7554 | return; |
7555 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
7556 | if (mddev->degraded) |
7557 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 7558 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
7559 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7560 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7561 | md_wakeup_thread(mddev->thread); | |
768a418d | 7562 | if (mddev->event_work.func) |
e804ac78 | 7563 | queue_work(md_misc_wq, &mddev->event_work); |
bb9ef716 | 7564 | md_new_event(mddev); |
1da177e4 | 7565 | } |
6c144d31 | 7566 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
7567 | |
7568 | /* seq_file implementation /proc/mdstat */ | |
7569 | ||
7570 | static void status_unused(struct seq_file *seq) | |
7571 | { | |
7572 | int i = 0; | |
3cb03002 | 7573 | struct md_rdev *rdev; |
1da177e4 LT |
7574 | |
7575 | seq_printf(seq, "unused devices: "); | |
7576 | ||
159ec1fc | 7577 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
7578 | char b[BDEVNAME_SIZE]; |
7579 | i++; | |
7580 | seq_printf(seq, "%s ", | |
7581 | bdevname(rdev->bdev,b)); | |
7582 | } | |
7583 | if (!i) | |
7584 | seq_printf(seq, "<none>"); | |
7585 | ||
7586 | seq_printf(seq, "\n"); | |
7587 | } | |
7588 | ||
f7851be7 | 7589 | static int status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7590 | { |
dd71cf6b N |
7591 | sector_t max_sectors, resync, res; |
7592 | unsigned long dt, db; | |
7593 | sector_t rt; | |
4588b42e N |
7594 | int scale; |
7595 | unsigned int per_milli; | |
1da177e4 | 7596 | |
c804cdec N |
7597 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7598 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 7599 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 7600 | else |
dd71cf6b | 7601 | max_sectors = mddev->dev_sectors; |
1da177e4 | 7602 | |
f7851be7 N |
7603 | resync = mddev->curr_resync; |
7604 | if (resync <= 3) { | |
7605 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7606 | /* Still cleaning up */ | |
7607 | resync = max_sectors; | |
d2e2ec82 ND |
7608 | } else if (resync > max_sectors) |
7609 | resync = max_sectors; | |
7610 | else | |
f7851be7 N |
7611 | resync -= atomic_read(&mddev->recovery_active); |
7612 | ||
7613 | if (resync == 0) { | |
7614 | if (mddev->recovery_cp < MaxSector) { | |
7615 | seq_printf(seq, "\tresync=PENDING"); | |
7616 | return 1; | |
7617 | } | |
7618 | return 0; | |
7619 | } | |
7620 | if (resync < 3) { | |
7621 | seq_printf(seq, "\tresync=DELAYED"); | |
7622 | return 1; | |
7623 | } | |
7624 | ||
403df478 | 7625 | WARN_ON(max_sectors == 0); |
4588b42e | 7626 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 7627 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
7628 | * u32, as those are the requirements for sector_div. |
7629 | * Thus 'scale' must be at least 10 | |
7630 | */ | |
7631 | scale = 10; | |
7632 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 7633 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
7634 | scale++; |
7635 | } | |
7636 | res = (resync>>scale)*1000; | |
dd71cf6b | 7637 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
7638 | |
7639 | per_milli = res; | |
1da177e4 | 7640 | { |
4588b42e | 7641 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
7642 | seq_printf(seq, "["); |
7643 | for (i = 0; i < x; i++) | |
7644 | seq_printf(seq, "="); | |
7645 | seq_printf(seq, ">"); | |
7646 | for (i = 0; i < y; i++) | |
7647 | seq_printf(seq, "."); | |
7648 | seq_printf(seq, "] "); | |
7649 | } | |
4588b42e | 7650 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
7651 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
7652 | "reshape" : | |
61df9d91 N |
7653 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
7654 | "check" : | |
7655 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
7656 | "resync" : "recovery"))), | |
7657 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
7658 | (unsigned long long) resync/2, |
7659 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
7660 | |
7661 | /* | |
1da177e4 LT |
7662 | * dt: time from mark until now |
7663 | * db: blocks written from mark until now | |
7664 | * rt: remaining time | |
dd71cf6b N |
7665 | * |
7666 | * rt is a sector_t, so could be 32bit or 64bit. | |
7667 | * So we divide before multiply in case it is 32bit and close | |
7668 | * to the limit. | |
25985edc | 7669 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
7670 | * near the end of resync when the number of remaining sectors |
7671 | * is close to 'db'. | |
7672 | * We then divide rt by 32 after multiplying by db to compensate. | |
7673 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
7674 | */ |
7675 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
7676 | if (!dt) dt++; | |
ff4e8d9a N |
7677 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
7678 | - mddev->resync_mark_cnt; | |
1da177e4 | 7679 | |
dd71cf6b N |
7680 | rt = max_sectors - resync; /* number of remaining sectors */ |
7681 | sector_div(rt, db/32+1); | |
7682 | rt *= dt; | |
7683 | rt >>= 5; | |
7684 | ||
7685 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
7686 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 7687 | |
ff4e8d9a | 7688 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
f7851be7 | 7689 | return 1; |
1da177e4 LT |
7690 | } |
7691 | ||
7692 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7693 | { | |
7694 | struct list_head *tmp; | |
7695 | loff_t l = *pos; | |
fd01b88c | 7696 | struct mddev *mddev; |
1da177e4 LT |
7697 | |
7698 | if (l >= 0x10000) | |
7699 | return NULL; | |
7700 | if (!l--) | |
7701 | /* header */ | |
7702 | return (void*)1; | |
7703 | ||
7704 | spin_lock(&all_mddevs_lock); | |
7705 | list_for_each(tmp,&all_mddevs) | |
7706 | if (!l--) { | |
fd01b88c | 7707 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7708 | mddev_get(mddev); |
7709 | spin_unlock(&all_mddevs_lock); | |
7710 | return mddev; | |
7711 | } | |
7712 | spin_unlock(&all_mddevs_lock); | |
7713 | if (!l--) | |
7714 | return (void*)2;/* tail */ | |
7715 | return NULL; | |
7716 | } | |
7717 | ||
7718 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7719 | { | |
7720 | struct list_head *tmp; | |
fd01b88c | 7721 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7722 | |
1da177e4 LT |
7723 | ++*pos; |
7724 | if (v == (void*)2) | |
7725 | return NULL; | |
7726 | ||
7727 | spin_lock(&all_mddevs_lock); | |
7728 | if (v == (void*)1) | |
7729 | tmp = all_mddevs.next; | |
7730 | else | |
7731 | tmp = mddev->all_mddevs.next; | |
7732 | if (tmp != &all_mddevs) | |
fd01b88c | 7733 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7734 | else { |
7735 | next_mddev = (void*)2; | |
7736 | *pos = 0x10000; | |
f72ffdd6 | 7737 | } |
1da177e4 LT |
7738 | spin_unlock(&all_mddevs_lock); |
7739 | ||
7740 | if (v != (void*)1) | |
7741 | mddev_put(mddev); | |
7742 | return next_mddev; | |
7743 | ||
7744 | } | |
7745 | ||
7746 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7747 | { | |
fd01b88c | 7748 | struct mddev *mddev = v; |
1da177e4 LT |
7749 | |
7750 | if (mddev && v != (void*)1 && v != (void*)2) | |
7751 | mddev_put(mddev); | |
7752 | } | |
7753 | ||
7754 | static int md_seq_show(struct seq_file *seq, void *v) | |
7755 | { | |
fd01b88c | 7756 | struct mddev *mddev = v; |
dd8ac336 | 7757 | sector_t sectors; |
3cb03002 | 7758 | struct md_rdev *rdev; |
1da177e4 LT |
7759 | |
7760 | if (v == (void*)1) { | |
84fc4b56 | 7761 | struct md_personality *pers; |
1da177e4 LT |
7762 | seq_printf(seq, "Personalities : "); |
7763 | spin_lock(&pers_lock); | |
2604b703 N |
7764 | list_for_each_entry(pers, &pers_list, list) |
7765 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7766 | |
7767 | spin_unlock(&pers_lock); | |
7768 | seq_printf(seq, "\n"); | |
f1514638 | 7769 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7770 | return 0; |
7771 | } | |
7772 | if (v == (void*)2) { | |
7773 | status_unused(seq); | |
7774 | return 0; | |
7775 | } | |
7776 | ||
36d091f4 | 7777 | spin_lock(&mddev->lock); |
1da177e4 LT |
7778 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7779 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7780 | mddev->pers ? "" : "in"); | |
7781 | if (mddev->pers) { | |
f91de92e | 7782 | if (mddev->ro==1) |
1da177e4 | 7783 | seq_printf(seq, " (read-only)"); |
f91de92e | 7784 | if (mddev->ro==2) |
52720ae7 | 7785 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7786 | seq_printf(seq, " %s", mddev->pers->name); |
7787 | } | |
7788 | ||
dd8ac336 | 7789 | sectors = 0; |
f97fcad3 N |
7790 | rcu_read_lock(); |
7791 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7792 | char b[BDEVNAME_SIZE]; |
7793 | seq_printf(seq, " %s[%d]", | |
7794 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7795 | if (test_bit(WriteMostly, &rdev->flags)) |
7796 | seq_printf(seq, "(W)"); | |
9efdca16 SL |
7797 | if (test_bit(Journal, &rdev->flags)) |
7798 | seq_printf(seq, "(J)"); | |
b2d444d7 | 7799 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7800 | seq_printf(seq, "(F)"); |
7801 | continue; | |
2d78f8c4 N |
7802 | } |
7803 | if (rdev->raid_disk < 0) | |
b325a32e | 7804 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7805 | if (test_bit(Replacement, &rdev->flags)) |
7806 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7807 | sectors += rdev->sectors; |
1da177e4 | 7808 | } |
f97fcad3 | 7809 | rcu_read_unlock(); |
1da177e4 LT |
7810 | |
7811 | if (!list_empty(&mddev->disks)) { | |
7812 | if (mddev->pers) | |
7813 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7814 | (unsigned long long) |
7815 | mddev->array_sectors / 2); | |
1da177e4 LT |
7816 | else |
7817 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7818 | (unsigned long long)sectors / 2); |
1da177e4 | 7819 | } |
1cd6bf19 N |
7820 | if (mddev->persistent) { |
7821 | if (mddev->major_version != 0 || | |
7822 | mddev->minor_version != 90) { | |
7823 | seq_printf(seq," super %d.%d", | |
7824 | mddev->major_version, | |
7825 | mddev->minor_version); | |
7826 | } | |
e691063a N |
7827 | } else if (mddev->external) |
7828 | seq_printf(seq, " super external:%s", | |
7829 | mddev->metadata_type); | |
7830 | else | |
1cd6bf19 | 7831 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7832 | |
7833 | if (mddev->pers) { | |
d710e138 | 7834 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7835 | seq_printf(seq, "\n "); |
8e1b39d6 | 7836 | if (mddev->pers->sync_request) { |
f7851be7 | 7837 | if (status_resync(seq, mddev)) |
8e1b39d6 | 7838 | seq_printf(seq, "\n "); |
8e1b39d6 | 7839 | } |
32a7627c N |
7840 | } else |
7841 | seq_printf(seq, "\n "); | |
7842 | ||
57148964 | 7843 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7844 | |
7845 | seq_printf(seq, "\n"); | |
7846 | } | |
36d091f4 | 7847 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7848 | |
1da177e4 LT |
7849 | return 0; |
7850 | } | |
7851 | ||
110518bc | 7852 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7853 | .start = md_seq_start, |
7854 | .next = md_seq_next, | |
7855 | .stop = md_seq_stop, | |
7856 | .show = md_seq_show, | |
7857 | }; | |
7858 | ||
7859 | static int md_seq_open(struct inode *inode, struct file *file) | |
7860 | { | |
f1514638 | 7861 | struct seq_file *seq; |
1da177e4 LT |
7862 | int error; |
7863 | ||
7864 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7865 | if (error) |
f1514638 KS |
7866 | return error; |
7867 | ||
7868 | seq = file->private_data; | |
7869 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7870 | return error; |
7871 | } | |
7872 | ||
e2f23b60 | 7873 | static int md_unloading; |
d7603b7e N |
7874 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7875 | { | |
f1514638 | 7876 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7877 | int mask; |
7878 | ||
e2f23b60 | 7879 | if (md_unloading) |
7d7e64f2 | 7880 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI; |
d7603b7e N |
7881 | poll_wait(filp, &md_event_waiters, wait); |
7882 | ||
7883 | /* always allow read */ | |
7884 | mask = POLLIN | POLLRDNORM; | |
7885 | ||
f1514638 | 7886 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7887 | mask |= POLLERR | POLLPRI; |
7888 | return mask; | |
7889 | } | |
7890 | ||
fa027c2a | 7891 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7892 | .owner = THIS_MODULE, |
1da177e4 LT |
7893 | .open = md_seq_open, |
7894 | .read = seq_read, | |
7895 | .llseek = seq_lseek, | |
26e13043 | 7896 | .release = seq_release, |
d7603b7e | 7897 | .poll = mdstat_poll, |
1da177e4 LT |
7898 | }; |
7899 | ||
84fc4b56 | 7900 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7901 | { |
9d48739e N |
7902 | pr_debug("md: %s personality registered for level %d\n", |
7903 | p->name, p->level); | |
1da177e4 | 7904 | spin_lock(&pers_lock); |
2604b703 | 7905 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7906 | spin_unlock(&pers_lock); |
7907 | return 0; | |
7908 | } | |
6c144d31 | 7909 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7910 | |
84fc4b56 | 7911 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7912 | { |
9d48739e | 7913 | pr_debug("md: %s personality unregistered\n", p->name); |
1da177e4 | 7914 | spin_lock(&pers_lock); |
2604b703 | 7915 | list_del_init(&p->list); |
1da177e4 LT |
7916 | spin_unlock(&pers_lock); |
7917 | return 0; | |
7918 | } | |
6c144d31 | 7919 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7920 | |
6022e75b N |
7921 | int register_md_cluster_operations(struct md_cluster_operations *ops, |
7922 | struct module *module) | |
edb39c9d | 7923 | { |
6022e75b | 7924 | int ret = 0; |
edb39c9d | 7925 | spin_lock(&pers_lock); |
6022e75b N |
7926 | if (md_cluster_ops != NULL) |
7927 | ret = -EALREADY; | |
7928 | else { | |
7929 | md_cluster_ops = ops; | |
7930 | md_cluster_mod = module; | |
7931 | } | |
edb39c9d | 7932 | spin_unlock(&pers_lock); |
6022e75b | 7933 | return ret; |
edb39c9d GR |
7934 | } |
7935 | EXPORT_SYMBOL(register_md_cluster_operations); | |
7936 | ||
7937 | int unregister_md_cluster_operations(void) | |
7938 | { | |
7939 | spin_lock(&pers_lock); | |
7940 | md_cluster_ops = NULL; | |
7941 | spin_unlock(&pers_lock); | |
7942 | return 0; | |
7943 | } | |
7944 | EXPORT_SYMBOL(unregister_md_cluster_operations); | |
7945 | ||
7946 | int md_setup_cluster(struct mddev *mddev, int nodes) | |
7947 | { | |
47a7b0d8 GJ |
7948 | if (!md_cluster_ops) |
7949 | request_module("md-cluster"); | |
edb39c9d | 7950 | spin_lock(&pers_lock); |
47a7b0d8 | 7951 | /* ensure module won't be unloaded */ |
edb39c9d | 7952 | if (!md_cluster_ops || !try_module_get(md_cluster_mod)) { |
9d48739e | 7953 | pr_warn("can't find md-cluster module or get it's reference.\n"); |
edb39c9d GR |
7954 | spin_unlock(&pers_lock); |
7955 | return -ENOENT; | |
7956 | } | |
7957 | spin_unlock(&pers_lock); | |
7958 | ||
cf921cc1 | 7959 | return md_cluster_ops->join(mddev, nodes); |
edb39c9d GR |
7960 | } |
7961 | ||
7962 | void md_cluster_stop(struct mddev *mddev) | |
7963 | { | |
c4ce867f GR |
7964 | if (!md_cluster_ops) |
7965 | return; | |
edb39c9d GR |
7966 | md_cluster_ops->leave(mddev); |
7967 | module_put(md_cluster_mod); | |
7968 | } | |
7969 | ||
fd01b88c | 7970 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7971 | { |
f72ffdd6 | 7972 | struct md_rdev *rdev; |
1da177e4 | 7973 | int idle; |
eea1bf38 | 7974 | int curr_events; |
1da177e4 LT |
7975 | |
7976 | idle = 1; | |
4b80991c N |
7977 | rcu_read_lock(); |
7978 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7979 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7980 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7981 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7982 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7983 | /* sync IO will cause sync_io to increase before the disk_stats |
7984 | * as sync_io is counted when a request starts, and | |
7985 | * disk_stats is counted when it completes. | |
7986 | * So resync activity will cause curr_events to be smaller than | |
7987 | * when there was no such activity. | |
7988 | * non-sync IO will cause disk_stat to increase without | |
7989 | * increasing sync_io so curr_events will (eventually) | |
7990 | * be larger than it was before. Once it becomes | |
7991 | * substantially larger, the test below will cause | |
7992 | * the array to appear non-idle, and resync will slow | |
7993 | * down. | |
7994 | * If there is a lot of outstanding resync activity when | |
7995 | * we set last_event to curr_events, then all that activity | |
7996 | * completing might cause the array to appear non-idle | |
7997 | * and resync will be slowed down even though there might | |
7998 | * not have been non-resync activity. This will only | |
7999 | * happen once though. 'last_events' will soon reflect | |
8000 | * the state where there is little or no outstanding | |
8001 | * resync requests, and further resync activity will | |
8002 | * always make curr_events less than last_events. | |
c0e48521 | 8003 | * |
1da177e4 | 8004 | */ |
eea1bf38 | 8005 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
8006 | rdev->last_events = curr_events; |
8007 | idle = 0; | |
8008 | } | |
8009 | } | |
4b80991c | 8010 | rcu_read_unlock(); |
1da177e4 LT |
8011 | return idle; |
8012 | } | |
8013 | ||
fd01b88c | 8014 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
8015 | { |
8016 | /* another "blocks" (512byte) blocks have been synced */ | |
8017 | atomic_sub(blocks, &mddev->recovery_active); | |
8018 | wake_up(&mddev->recovery_wait); | |
8019 | if (!ok) { | |
dfc70645 | 8020 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 8021 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
8022 | md_wakeup_thread(mddev->thread); |
8023 | // stop recovery, signal do_sync .... | |
8024 | } | |
8025 | } | |
6c144d31 | 8026 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 8027 | |
06d91a5f N |
8028 | /* md_write_start(mddev, bi) |
8029 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
8030 | * in superblock) before writing, schedule a superblock update |
8031 | * and wait for it to complete. | |
cc27b0c7 N |
8032 | * A return value of 'false' means that the write wasn't recorded |
8033 | * and cannot proceed as the array is being suspend. | |
06d91a5f | 8034 | */ |
cc27b0c7 | 8035 | bool md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 8036 | { |
0fd62b86 | 8037 | int did_change = 0; |
06d91a5f | 8038 | if (bio_data_dir(bi) != WRITE) |
cc27b0c7 | 8039 | return true; |
06d91a5f | 8040 | |
f91de92e N |
8041 | BUG_ON(mddev->ro == 1); |
8042 | if (mddev->ro == 2) { | |
8043 | /* need to switch to read/write */ | |
8044 | mddev->ro = 0; | |
8045 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8046 | md_wakeup_thread(mddev->thread); | |
25156198 | 8047 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 8048 | did_change = 1; |
f91de92e | 8049 | } |
4ad23a97 N |
8050 | rcu_read_lock(); |
8051 | percpu_ref_get(&mddev->writes_pending); | |
55cc39f3 | 8052 | smp_mb(); /* Match smp_mb in set_in_sync() */ |
31a59e34 N |
8053 | if (mddev->safemode == 1) |
8054 | mddev->safemode = 0; | |
4ad23a97 | 8055 | /* sync_checkers is always 0 when writes_pending is in per-cpu mode */ |
81fe48e9 | 8056 | if (mddev->in_sync || mddev->sync_checkers) { |
85572d7c | 8057 | spin_lock(&mddev->lock); |
3d310eb7 N |
8058 | if (mddev->in_sync) { |
8059 | mddev->in_sync = 0; | |
2953079c SL |
8060 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
8061 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
3d310eb7 | 8062 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 8063 | did_change = 1; |
3d310eb7 | 8064 | } |
85572d7c | 8065 | spin_unlock(&mddev->lock); |
06d91a5f | 8066 | } |
4ad23a97 | 8067 | rcu_read_unlock(); |
0fd62b86 | 8068 | if (did_change) |
00bcb4ac | 8069 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 8070 | wait_event(mddev->sb_wait, |
d47c8ad2 N |
8071 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) || |
8072 | mddev->suspended); | |
cc27b0c7 N |
8073 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
8074 | percpu_ref_put(&mddev->writes_pending); | |
8075 | return false; | |
8076 | } | |
8077 | return true; | |
1da177e4 | 8078 | } |
6c144d31 | 8079 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 8080 | |
49728050 N |
8081 | /* md_write_inc can only be called when md_write_start() has |
8082 | * already been called at least once of the current request. | |
8083 | * It increments the counter and is useful when a single request | |
8084 | * is split into several parts. Each part causes an increment and | |
8085 | * so needs a matching md_write_end(). | |
8086 | * Unlike md_write_start(), it is safe to call md_write_inc() inside | |
8087 | * a spinlocked region. | |
8088 | */ | |
8089 | void md_write_inc(struct mddev *mddev, struct bio *bi) | |
8090 | { | |
8091 | if (bio_data_dir(bi) != WRITE) | |
8092 | return; | |
8093 | WARN_ON_ONCE(mddev->in_sync || mddev->ro); | |
4ad23a97 | 8094 | percpu_ref_get(&mddev->writes_pending); |
49728050 N |
8095 | } |
8096 | EXPORT_SYMBOL(md_write_inc); | |
8097 | ||
fd01b88c | 8098 | void md_write_end(struct mddev *mddev) |
1da177e4 | 8099 | { |
4ad23a97 N |
8100 | percpu_ref_put(&mddev->writes_pending); |
8101 | ||
8102 | if (mddev->safemode == 2) | |
8103 | md_wakeup_thread(mddev->thread); | |
8104 | else if (mddev->safemode_delay) | |
8105 | /* The roundup() ensures this only performs locking once | |
8106 | * every ->safemode_delay jiffies | |
8107 | */ | |
8108 | mod_timer(&mddev->safemode_timer, | |
8109 | roundup(jiffies, mddev->safemode_delay) + | |
8110 | mddev->safemode_delay); | |
1da177e4 | 8111 | } |
4ad23a97 | 8112 | |
6c144d31 | 8113 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 8114 | |
2a2275d6 N |
8115 | /* md_allow_write(mddev) |
8116 | * Calling this ensures that the array is marked 'active' so that writes | |
8117 | * may proceed without blocking. It is important to call this before | |
8118 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
8119 | * Must be called with mddev_lock held. | |
8120 | */ | |
2214c260 | 8121 | void md_allow_write(struct mddev *mddev) |
2a2275d6 N |
8122 | { |
8123 | if (!mddev->pers) | |
2214c260 | 8124 | return; |
2a2275d6 | 8125 | if (mddev->ro) |
2214c260 | 8126 | return; |
1a0fd497 | 8127 | if (!mddev->pers->sync_request) |
2214c260 | 8128 | return; |
2a2275d6 | 8129 | |
85572d7c | 8130 | spin_lock(&mddev->lock); |
2a2275d6 N |
8131 | if (mddev->in_sync) { |
8132 | mddev->in_sync = 0; | |
2953079c SL |
8133 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
8134 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
2a2275d6 N |
8135 | if (mddev->safemode_delay && |
8136 | mddev->safemode == 0) | |
8137 | mddev->safemode = 1; | |
85572d7c | 8138 | spin_unlock(&mddev->lock); |
2a2275d6 | 8139 | md_update_sb(mddev, 0); |
00bcb4ac | 8140 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2214c260 AP |
8141 | /* wait for the dirty state to be recorded in the metadata */ |
8142 | wait_event(mddev->sb_wait, | |
2214c260 | 8143 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
2a2275d6 | 8144 | } else |
85572d7c | 8145 | spin_unlock(&mddev->lock); |
2a2275d6 N |
8146 | } |
8147 | EXPORT_SYMBOL_GPL(md_allow_write); | |
8148 | ||
1da177e4 LT |
8149 | #define SYNC_MARKS 10 |
8150 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 8151 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 8152 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 8153 | { |
4ed8731d | 8154 | struct mddev *mddev = thread->mddev; |
fd01b88c | 8155 | struct mddev *mddev2; |
1da177e4 LT |
8156 | unsigned int currspeed = 0, |
8157 | window; | |
ac7e50a3 | 8158 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 8159 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 8160 | unsigned long update_time; |
1da177e4 LT |
8161 | sector_t mark_cnt[SYNC_MARKS]; |
8162 | int last_mark,m; | |
8163 | struct list_head *tmp; | |
8164 | sector_t last_check; | |
57afd89f | 8165 | int skipped = 0; |
3cb03002 | 8166 | struct md_rdev *rdev; |
c4a39551 | 8167 | char *desc, *action = NULL; |
7c2c57c9 | 8168 | struct blk_plug plug; |
41a9a0dc | 8169 | int ret; |
1da177e4 LT |
8170 | |
8171 | /* just incase thread restarts... */ | |
8172 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
8173 | return; | |
3991b31e N |
8174 | if (mddev->ro) {/* never try to sync a read-only array */ |
8175 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 8176 | return; |
3991b31e | 8177 | } |
1da177e4 | 8178 | |
41a9a0dc GJ |
8179 | if (mddev_is_clustered(mddev)) { |
8180 | ret = md_cluster_ops->resync_start(mddev); | |
8181 | if (ret) | |
8182 | goto skip; | |
8183 | ||
bb8bf15b | 8184 | set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags); |
41a9a0dc GJ |
8185 | if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
8186 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) || | |
8187 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) | |
8188 | && ((unsigned long long)mddev->curr_resync_completed | |
8189 | < (unsigned long long)mddev->resync_max_sectors)) | |
8190 | goto skip; | |
8191 | } | |
8192 | ||
61df9d91 | 8193 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 8194 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 8195 | desc = "data-check"; |
c4a39551 JB |
8196 | action = "check"; |
8197 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 8198 | desc = "requested-resync"; |
c4a39551 JB |
8199 | action = "repair"; |
8200 | } else | |
61df9d91 N |
8201 | desc = "resync"; |
8202 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
8203 | desc = "reshape"; | |
8204 | else | |
8205 | desc = "recovery"; | |
8206 | ||
c4a39551 JB |
8207 | mddev->last_sync_action = action ?: desc; |
8208 | ||
1da177e4 LT |
8209 | /* we overload curr_resync somewhat here. |
8210 | * 0 == not engaged in resync at all | |
8211 | * 2 == checking that there is no conflict with another sync | |
8212 | * 1 == like 2, but have yielded to allow conflicting resync to | |
8213 | * commense | |
8214 | * other == active in resync - this many blocks | |
8215 | * | |
8216 | * Before starting a resync we must have set curr_resync to | |
8217 | * 2, and then checked that every "conflicting" array has curr_resync | |
8218 | * less than ours. When we find one that is the same or higher | |
8219 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
8220 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
8221 | * This will mean we have to start checking from the beginning again. | |
8222 | * | |
8223 | */ | |
8224 | ||
8225 | do { | |
c622ca54 | 8226 | int mddev2_minor = -1; |
1da177e4 LT |
8227 | mddev->curr_resync = 2; |
8228 | ||
8229 | try_again: | |
404e4b43 | 8230 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 8231 | goto skip; |
29ac4aa3 | 8232 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
8233 | if (mddev2 == mddev) |
8234 | continue; | |
90b08710 BS |
8235 | if (!mddev->parallel_resync |
8236 | && mddev2->curr_resync | |
8237 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
8238 | DEFINE_WAIT(wq); |
8239 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
8240 | /* arbitrarily yield */ | |
8241 | mddev->curr_resync = 1; | |
8242 | wake_up(&resync_wait); | |
8243 | } | |
8244 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
8245 | /* no need to wait here, we can wait the next | |
8246 | * time 'round when curr_resync == 2 | |
8247 | */ | |
8248 | continue; | |
9744197c N |
8249 | /* We need to wait 'interruptible' so as not to |
8250 | * contribute to the load average, and not to | |
8251 | * be caught by 'softlockup' | |
8252 | */ | |
8253 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 8254 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 8255 | mddev2->curr_resync >= mddev->curr_resync) { |
c622ca54 AP |
8256 | if (mddev2_minor != mddev2->md_minor) { |
8257 | mddev2_minor = mddev2->md_minor; | |
9d48739e N |
8258 | pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n", |
8259 | desc, mdname(mddev), | |
8260 | mdname(mddev2)); | |
c622ca54 | 8261 | } |
1da177e4 | 8262 | mddev_put(mddev2); |
9744197c N |
8263 | if (signal_pending(current)) |
8264 | flush_signals(current); | |
1da177e4 LT |
8265 | schedule(); |
8266 | finish_wait(&resync_wait, &wq); | |
8267 | goto try_again; | |
8268 | } | |
8269 | finish_wait(&resync_wait, &wq); | |
8270 | } | |
8271 | } | |
8272 | } while (mddev->curr_resync < 2); | |
8273 | ||
5fd6c1dc | 8274 | j = 0; |
9d88883e | 8275 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 8276 | /* resync follows the size requested by the personality, |
57afd89f | 8277 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
8278 | */ |
8279 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 8280 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 8281 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
8282 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
8283 | j = mddev->resync_min; | |
8284 | else if (!mddev->bitmap) | |
5fd6c1dc | 8285 | j = mddev->recovery_cp; |
5e96ee65 | 8286 | |
ccfcc3c1 | 8287 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 8288 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 8289 | else { |
1da177e4 | 8290 | /* recovery follows the physical size of devices */ |
58c0fed4 | 8291 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 8292 | j = MaxSector; |
4e59ca7d | 8293 | rcu_read_lock(); |
dafb20fa | 8294 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 8295 | if (rdev->raid_disk >= 0 && |
f2076e7d | 8296 | !test_bit(Journal, &rdev->flags) && |
5fd6c1dc N |
8297 | !test_bit(Faulty, &rdev->flags) && |
8298 | !test_bit(In_sync, &rdev->flags) && | |
8299 | rdev->recovery_offset < j) | |
8300 | j = rdev->recovery_offset; | |
4e59ca7d | 8301 | rcu_read_unlock(); |
133d4527 N |
8302 | |
8303 | /* If there is a bitmap, we need to make sure all | |
8304 | * writes that started before we added a spare | |
8305 | * complete before we start doing a recovery. | |
8306 | * Otherwise the write might complete and (via | |
8307 | * bitmap_endwrite) set a bit in the bitmap after the | |
8308 | * recovery has checked that bit and skipped that | |
8309 | * region. | |
8310 | */ | |
8311 | if (mddev->bitmap) { | |
8312 | mddev->pers->quiesce(mddev, 1); | |
8313 | mddev->pers->quiesce(mddev, 0); | |
8314 | } | |
5fd6c1dc | 8315 | } |
1da177e4 | 8316 | |
9d48739e N |
8317 | pr_info("md: %s of RAID array %s\n", desc, mdname(mddev)); |
8318 | pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev)); | |
8319 | pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n", | |
8320 | speed_max(mddev), desc); | |
1da177e4 | 8321 | |
eea1bf38 | 8322 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 8323 | |
57afd89f | 8324 | io_sectors = 0; |
1da177e4 LT |
8325 | for (m = 0; m < SYNC_MARKS; m++) { |
8326 | mark[m] = jiffies; | |
57afd89f | 8327 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
8328 | } |
8329 | last_mark = 0; | |
8330 | mddev->resync_mark = mark[last_mark]; | |
8331 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
8332 | ||
8333 | /* | |
8334 | * Tune reconstruction: | |
8335 | */ | |
8336 | window = 32*(PAGE_SIZE/512); | |
9d48739e N |
8337 | pr_debug("md: using %dk window, over a total of %lluk.\n", |
8338 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
8339 | |
8340 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
8341 | last_check = 0; |
8342 | ||
8343 | if (j>2) { | |
9d48739e N |
8344 | pr_debug("md: resuming %s of %s from checkpoint.\n", |
8345 | desc, mdname(mddev)); | |
1da177e4 | 8346 | mddev->curr_resync = j; |
72f36d59 N |
8347 | } else |
8348 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 8349 | mddev->curr_resync_completed = j; |
72f36d59 N |
8350 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
8351 | md_new_event(mddev); | |
54f89341 | 8352 | update_time = jiffies; |
1da177e4 | 8353 | |
7c2c57c9 | 8354 | blk_start_plug(&plug); |
1da177e4 | 8355 | while (j < max_sectors) { |
57afd89f | 8356 | sector_t sectors; |
1da177e4 | 8357 | |
57afd89f | 8358 | skipped = 0; |
97e4f42d | 8359 | |
7a91ee1f N |
8360 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8361 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
8362 | (mddev->curr_resync - mddev->curr_resync_completed) | |
8363 | > (max_sectors >> 4)) || | |
54f89341 | 8364 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f | 8365 | (j - mddev->curr_resync_completed)*2 |
c5e19d90 N |
8366 | >= mddev->resync_max - mddev->curr_resync_completed || |
8367 | mddev->curr_resync_completed > mddev->resync_max | |
7a91ee1f | 8368 | )) { |
97e4f42d | 8369 | /* time to update curr_resync_completed */ |
97e4f42d N |
8370 | wait_event(mddev->recovery_wait, |
8371 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 8372 | mddev->curr_resync_completed = j; |
35d78c66 | 8373 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
8374 | j > mddev->recovery_cp) | |
8375 | mddev->recovery_cp = j; | |
54f89341 | 8376 | update_time = jiffies; |
2953079c | 8377 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
acb180b0 | 8378 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 8379 | } |
acb180b0 | 8380 | |
c91abf5a N |
8381 | while (j >= mddev->resync_max && |
8382 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
8383 | /* As this condition is controlled by user-space, |
8384 | * we can block indefinitely, so use '_interruptible' | |
8385 | * to avoid triggering warnings. | |
8386 | */ | |
8387 | flush_signals(current); /* just in case */ | |
8388 | wait_event_interruptible(mddev->recovery_wait, | |
8389 | mddev->resync_max > j | |
c91abf5a N |
8390 | || test_bit(MD_RECOVERY_INTR, |
8391 | &mddev->recovery)); | |
e62e58a5 | 8392 | } |
acb180b0 | 8393 | |
c91abf5a N |
8394 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8395 | break; | |
acb180b0 | 8396 | |
09314799 | 8397 | sectors = mddev->pers->sync_request(mddev, j, &skipped); |
57afd89f | 8398 | if (sectors == 0) { |
dfc70645 | 8399 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 8400 | break; |
1da177e4 | 8401 | } |
57afd89f N |
8402 | |
8403 | if (!skipped) { /* actual IO requested */ | |
8404 | io_sectors += sectors; | |
8405 | atomic_add(sectors, &mddev->recovery_active); | |
8406 | } | |
8407 | ||
e875ecea N |
8408 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8409 | break; | |
8410 | ||
1da177e4 | 8411 | j += sectors; |
5ed1df2e N |
8412 | if (j > max_sectors) |
8413 | /* when skipping, extra large numbers can be returned. */ | |
8414 | j = max_sectors; | |
72f36d59 N |
8415 | if (j > 2) |
8416 | mddev->curr_resync = j; | |
ff4e8d9a | 8417 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 8418 | if (last_check == 0) |
e875ecea | 8419 | /* this is the earliest that rebuild will be |
d7603b7e N |
8420 | * visible in /proc/mdstat |
8421 | */ | |
8422 | md_new_event(mddev); | |
57afd89f N |
8423 | |
8424 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
8425 | continue; |
8426 | ||
57afd89f | 8427 | last_check = io_sectors; |
1da177e4 LT |
8428 | repeat: |
8429 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
8430 | /* step marks */ | |
8431 | int next = (last_mark+1) % SYNC_MARKS; | |
8432 | ||
8433 | mddev->resync_mark = mark[next]; | |
8434 | mddev->resync_mark_cnt = mark_cnt[next]; | |
8435 | mark[next] = jiffies; | |
57afd89f | 8436 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
8437 | last_mark = next; |
8438 | } | |
8439 | ||
c91abf5a N |
8440 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8441 | break; | |
1da177e4 LT |
8442 | |
8443 | /* | |
8444 | * this loop exits only if either when we are slower than | |
8445 | * the 'hard' speed limit, or the system was IO-idle for | |
8446 | * a jiffy. | |
8447 | * the system might be non-idle CPU-wise, but we only care | |
8448 | * about not overloading the IO subsystem. (things like an | |
8449 | * e2fsck being done on the RAID array should execute fast) | |
8450 | */ | |
1da177e4 LT |
8451 | cond_resched(); |
8452 | ||
ac7e50a3 XN |
8453 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
8454 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 8455 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 8456 | |
88202a0c | 8457 | if (currspeed > speed_min(mddev)) { |
ac8fa419 | 8458 | if (currspeed > speed_max(mddev)) { |
c0e48521 | 8459 | msleep(500); |
1da177e4 LT |
8460 | goto repeat; |
8461 | } | |
ac8fa419 N |
8462 | if (!is_mddev_idle(mddev, 0)) { |
8463 | /* | |
8464 | * Give other IO more of a chance. | |
8465 | * The faster the devices, the less we wait. | |
8466 | */ | |
8467 | wait_event(mddev->recovery_wait, | |
8468 | !atomic_read(&mddev->recovery_active)); | |
8469 | } | |
1da177e4 LT |
8470 | } |
8471 | } | |
9d48739e N |
8472 | pr_info("md: %s: %s %s.\n",mdname(mddev), desc, |
8473 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
8474 | ? "interrupted" : "done"); | |
1da177e4 LT |
8475 | /* |
8476 | * this also signals 'finished resyncing' to md_stop | |
8477 | */ | |
7c2c57c9 | 8478 | blk_finish_plug(&plug); |
1da177e4 LT |
8479 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
8480 | ||
5ed1df2e N |
8481 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8482 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
1217e1d1 | 8483 | mddev->curr_resync > 3) { |
5ed1df2e N |
8484 | mddev->curr_resync_completed = mddev->curr_resync; |
8485 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
8486 | } | |
09314799 | 8487 | mddev->pers->sync_request(mddev, max_sectors, &skipped); |
1da177e4 | 8488 | |
dfc70645 | 8489 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
1217e1d1 | 8490 | mddev->curr_resync > 3) { |
5fd6c1dc N |
8491 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
8492 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8493 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
9d48739e N |
8494 | pr_debug("md: checkpointing %s of %s.\n", |
8495 | desc, mdname(mddev)); | |
0a19caab | 8496 | if (test_bit(MD_RECOVERY_ERROR, |
8497 | &mddev->recovery)) | |
8498 | mddev->recovery_cp = | |
8499 | mddev->curr_resync_completed; | |
8500 | else | |
8501 | mddev->recovery_cp = | |
8502 | mddev->curr_resync; | |
5fd6c1dc N |
8503 | } |
8504 | } else | |
8505 | mddev->recovery_cp = MaxSector; | |
8506 | } else { | |
8507 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
8508 | mddev->curr_resync = MaxSector; | |
db0505d3 N |
8509 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8510 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) { | |
8511 | rcu_read_lock(); | |
8512 | rdev_for_each_rcu(rdev, mddev) | |
8513 | if (rdev->raid_disk >= 0 && | |
8514 | mddev->delta_disks >= 0 && | |
8515 | !test_bit(Journal, &rdev->flags) && | |
8516 | !test_bit(Faulty, &rdev->flags) && | |
8517 | !test_bit(In_sync, &rdev->flags) && | |
8518 | rdev->recovery_offset < mddev->curr_resync) | |
8519 | rdev->recovery_offset = mddev->curr_resync; | |
8520 | rcu_read_unlock(); | |
8521 | } | |
5fd6c1dc | 8522 | } |
1da177e4 | 8523 | } |
db91ff55 | 8524 | skip: |
bb8bf15b GJ |
8525 | /* set CHANGE_PENDING here since maybe another update is needed, |
8526 | * so other nodes are informed. It should be harmless for normal | |
8527 | * raid */ | |
2953079c SL |
8528 | set_mask_bits(&mddev->sb_flags, 0, |
8529 | BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS)); | |
c186b128 | 8530 | |
23da422b | 8531 | spin_lock(&mddev->lock); |
c07b70ad N |
8532 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
8533 | /* We completed so min/max setting can be forgotten if used. */ | |
8534 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8535 | mddev->resync_min = 0; | |
8536 | mddev->resync_max = MaxSector; | |
8537 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8538 | mddev->resync_min = mddev->curr_resync_completed; | |
f7851be7 | 8539 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); |
1da177e4 | 8540 | mddev->curr_resync = 0; |
23da422b N |
8541 | spin_unlock(&mddev->lock); |
8542 | ||
1da177e4 | 8543 | wake_up(&resync_wait); |
1da177e4 | 8544 | md_wakeup_thread(mddev->thread); |
c6207277 | 8545 | return; |
1da177e4 | 8546 | } |
29269553 | 8547 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 8548 | |
746d3207 N |
8549 | static int remove_and_add_spares(struct mddev *mddev, |
8550 | struct md_rdev *this) | |
b4c4c7b8 | 8551 | { |
3cb03002 | 8552 | struct md_rdev *rdev; |
b4c4c7b8 | 8553 | int spares = 0; |
f2a371c5 | 8554 | int removed = 0; |
d787be40 | 8555 | bool remove_some = false; |
b4c4c7b8 | 8556 | |
d787be40 N |
8557 | rdev_for_each(rdev, mddev) { |
8558 | if ((this == NULL || rdev == this) && | |
8559 | rdev->raid_disk >= 0 && | |
8560 | !test_bit(Blocked, &rdev->flags) && | |
8561 | test_bit(Faulty, &rdev->flags) && | |
8562 | atomic_read(&rdev->nr_pending)==0) { | |
8563 | /* Faulty non-Blocked devices with nr_pending == 0 | |
8564 | * never get nr_pending incremented, | |
8565 | * never get Faulty cleared, and never get Blocked set. | |
8566 | * So we can synchronize_rcu now rather than once per device | |
8567 | */ | |
8568 | remove_some = true; | |
8569 | set_bit(RemoveSynchronized, &rdev->flags); | |
8570 | } | |
8571 | } | |
8572 | ||
8573 | if (remove_some) | |
8574 | synchronize_rcu(); | |
8575 | rdev_for_each(rdev, mddev) { | |
746d3207 N |
8576 | if ((this == NULL || rdev == this) && |
8577 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 8578 | !test_bit(Blocked, &rdev->flags) && |
d787be40 | 8579 | ((test_bit(RemoveSynchronized, &rdev->flags) || |
f2076e7d SL |
8580 | (!test_bit(In_sync, &rdev->flags) && |
8581 | !test_bit(Journal, &rdev->flags))) && | |
d787be40 | 8582 | atomic_read(&rdev->nr_pending)==0)) { |
b4c4c7b8 | 8583 | if (mddev->pers->hot_remove_disk( |
b8321b68 | 8584 | mddev, rdev) == 0) { |
36fad858 | 8585 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 8586 | rdev->raid_disk = -1; |
f2a371c5 | 8587 | removed++; |
b4c4c7b8 N |
8588 | } |
8589 | } | |
d787be40 N |
8590 | if (remove_some && test_bit(RemoveSynchronized, &rdev->flags)) |
8591 | clear_bit(RemoveSynchronized, &rdev->flags); | |
8592 | } | |
8593 | ||
90584fc9 JB |
8594 | if (removed && mddev->kobj.sd) |
8595 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 8596 | |
2910ff17 | 8597 | if (this && removed) |
746d3207 N |
8598 | goto no_add; |
8599 | ||
dafb20fa | 8600 | rdev_for_each(rdev, mddev) { |
2910ff17 GR |
8601 | if (this && this != rdev) |
8602 | continue; | |
dbb64f86 GR |
8603 | if (test_bit(Candidate, &rdev->flags)) |
8604 | continue; | |
7bfec5f3 N |
8605 | if (rdev->raid_disk >= 0 && |
8606 | !test_bit(In_sync, &rdev->flags) && | |
f2076e7d | 8607 | !test_bit(Journal, &rdev->flags) && |
7bfec5f3 N |
8608 | !test_bit(Faulty, &rdev->flags)) |
8609 | spares++; | |
7ceb17e8 N |
8610 | if (rdev->raid_disk >= 0) |
8611 | continue; | |
8612 | if (test_bit(Faulty, &rdev->flags)) | |
8613 | continue; | |
f6b6ec5c SL |
8614 | if (!test_bit(Journal, &rdev->flags)) { |
8615 | if (mddev->ro && | |
8616 | ! (rdev->saved_raid_disk >= 0 && | |
8617 | !test_bit(Bitmap_sync, &rdev->flags))) | |
8618 | continue; | |
7ceb17e8 | 8619 | |
f6b6ec5c SL |
8620 | rdev->recovery_offset = 0; |
8621 | } | |
7ceb17e8 N |
8622 | if (mddev->pers-> |
8623 | hot_add_disk(mddev, rdev) == 0) { | |
8624 | if (sysfs_link_rdev(mddev, rdev)) | |
8625 | /* failure here is OK */; | |
f6b6ec5c SL |
8626 | if (!test_bit(Journal, &rdev->flags)) |
8627 | spares++; | |
7ceb17e8 | 8628 | md_new_event(mddev); |
2953079c | 8629 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
dfc70645 | 8630 | } |
b4c4c7b8 | 8631 | } |
746d3207 | 8632 | no_add: |
6dafab6b | 8633 | if (removed) |
2953079c | 8634 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
b4c4c7b8 N |
8635 | return spares; |
8636 | } | |
7ebc0be7 | 8637 | |
ac05f256 N |
8638 | static void md_start_sync(struct work_struct *ws) |
8639 | { | |
8640 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
c186b128 | 8641 | |
ac05f256 N |
8642 | mddev->sync_thread = md_register_thread(md_do_sync, |
8643 | mddev, | |
8644 | "resync"); | |
8645 | if (!mddev->sync_thread) { | |
9d48739e N |
8646 | pr_warn("%s: could not start resync thread...\n", |
8647 | mdname(mddev)); | |
ac05f256 N |
8648 | /* leave the spares where they are, it shouldn't hurt */ |
8649 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8650 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8651 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8652 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
8653 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8654 | wake_up(&resync_wait); |
ac05f256 N |
8655 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8656 | &mddev->recovery)) | |
8657 | if (mddev->sysfs_action) | |
8658 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8659 | } else | |
8660 | md_wakeup_thread(mddev->sync_thread); | |
8661 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8662 | md_new_event(mddev); | |
8663 | } | |
8664 | ||
1da177e4 LT |
8665 | /* |
8666 | * This routine is regularly called by all per-raid-array threads to | |
8667 | * deal with generic issues like resync and super-block update. | |
8668 | * Raid personalities that don't have a thread (linear/raid0) do not | |
8669 | * need this as they never do any recovery or update the superblock. | |
8670 | * | |
8671 | * It does not do any resync itself, but rather "forks" off other threads | |
8672 | * to do that as needed. | |
8673 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
8674 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 8675 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
8676 | * and wakeups up this thread which will reap the thread and finish up. |
8677 | * This thread also removes any faulty devices (with nr_pending == 0). | |
8678 | * | |
8679 | * The overall approach is: | |
8680 | * 1/ if the superblock needs updating, update it. | |
8681 | * 2/ If a recovery thread is running, don't do anything else. | |
8682 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
8683 | * 4/ If there are any faulty devices, remove them. | |
8684 | * 5/ If array is degraded, try to add spares devices | |
8685 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
8686 | */ | |
fd01b88c | 8687 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 8688 | { |
68866e42 JB |
8689 | if (mddev->suspended) |
8690 | return; | |
8691 | ||
5f40402d | 8692 | if (mddev->bitmap) |
aa5cbd10 | 8693 | bitmap_daemon_work(mddev); |
1da177e4 | 8694 | |
fca4d848 | 8695 | if (signal_pending(current)) { |
31a59e34 | 8696 | if (mddev->pers->sync_request && !mddev->external) { |
9d48739e N |
8697 | pr_debug("md: %s in immediate safe mode\n", |
8698 | mdname(mddev)); | |
fca4d848 N |
8699 | mddev->safemode = 2; |
8700 | } | |
8701 | flush_signals(current); | |
8702 | } | |
8703 | ||
c89a8eee N |
8704 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
8705 | return; | |
1da177e4 | 8706 | if ( ! ( |
2953079c | 8707 | (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) || |
1da177e4 | 8708 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 8709 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 8710 | (mddev->external == 0 && mddev->safemode == 1) || |
4ad23a97 | 8711 | (mddev->safemode == 2 |
fca4d848 | 8712 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) |
1da177e4 LT |
8713 | )) |
8714 | return; | |
fca4d848 | 8715 | |
df5b89b3 | 8716 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 8717 | int spares = 0; |
fca4d848 | 8718 | |
afc1f55c | 8719 | if (!mddev->external && mddev->safemode == 1) |
33182d15 N |
8720 | mddev->safemode = 0; |
8721 | ||
c89a8eee | 8722 | if (mddev->ro) { |
ab16bfc7 NB |
8723 | struct md_rdev *rdev; |
8724 | if (!mddev->external && mddev->in_sync) | |
8725 | /* 'Blocked' flag not needed as failed devices | |
8726 | * will be recorded if array switched to read/write. | |
8727 | * Leaving it set will prevent the device | |
8728 | * from being removed. | |
8729 | */ | |
8730 | rdev_for_each(rdev, mddev) | |
8731 | clear_bit(Blocked, &rdev->flags); | |
7ceb17e8 N |
8732 | /* On a read-only array we can: |
8733 | * - remove failed devices | |
8734 | * - add already-in_sync devices if the array itself | |
8735 | * is in-sync. | |
8736 | * As we only add devices that are already in-sync, | |
8737 | * we can activate the spares immediately. | |
c89a8eee | 8738 | */ |
7ceb17e8 | 8739 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
8740 | /* There is no thread, but we need to call |
8741 | * ->spare_active and clear saved_raid_disk | |
8742 | */ | |
2ac295a5 | 8743 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 | 8744 | md_reap_sync_thread(mddev); |
a4a3d26d | 8745 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
8313b8e5 | 8746 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2953079c | 8747 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
c89a8eee N |
8748 | goto unlock; |
8749 | } | |
8750 | ||
659b254f GJ |
8751 | if (mddev_is_clustered(mddev)) { |
8752 | struct md_rdev *rdev; | |
8753 | /* kick the device if another node issued a | |
8754 | * remove disk. | |
8755 | */ | |
8756 | rdev_for_each(rdev, mddev) { | |
8757 | if (test_and_clear_bit(ClusterRemove, &rdev->flags) && | |
8758 | rdev->raid_disk < 0) | |
8759 | md_kick_rdev_from_array(rdev); | |
8760 | } | |
8761 | } | |
8762 | ||
6497709b | 8763 | if (!mddev->external && !mddev->in_sync) { |
85572d7c | 8764 | spin_lock(&mddev->lock); |
6497709b | 8765 | set_in_sync(mddev); |
85572d7c | 8766 | spin_unlock(&mddev->lock); |
fca4d848 | 8767 | } |
fca4d848 | 8768 | |
2953079c | 8769 | if (mddev->sb_flags) |
850b2b42 | 8770 | md_update_sb(mddev, 0); |
06d91a5f | 8771 | |
1da177e4 LT |
8772 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
8773 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
8774 | /* resync/recovery still happening */ | |
8775 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8776 | goto unlock; | |
8777 | } | |
8778 | if (mddev->sync_thread) { | |
a91d5ac0 | 8779 | md_reap_sync_thread(mddev); |
1da177e4 LT |
8780 | goto unlock; |
8781 | } | |
72a23c21 NB |
8782 | /* Set RUNNING before clearing NEEDED to avoid |
8783 | * any transients in the value of "sync_action". | |
8784 | */ | |
72f36d59 | 8785 | mddev->curr_resync_completed = 0; |
23da422b | 8786 | spin_lock(&mddev->lock); |
72a23c21 | 8787 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 8788 | spin_unlock(&mddev->lock); |
24dd469d N |
8789 | /* Clear some bits that don't mean anything, but |
8790 | * might be left set | |
8791 | */ | |
24dd469d N |
8792 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8793 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 8794 | |
ed209584 N |
8795 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
8796 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 8797 | goto not_running; |
1da177e4 LT |
8798 | /* no recovery is running. |
8799 | * remove any failed drives, then | |
8800 | * add spares if possible. | |
72f36d59 | 8801 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
8802 | * the personality to fail the re-add. |
8803 | */ | |
1da177e4 | 8804 | |
b4c4c7b8 | 8805 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
8806 | if (mddev->pers->check_reshape == NULL || |
8807 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 8808 | /* Cannot proceed */ |
ac05f256 | 8809 | goto not_running; |
b4c4c7b8 | 8810 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 8811 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 8812 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
8813 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8814 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 8815 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 8816 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8817 | } else if (mddev->recovery_cp < MaxSector) { |
8818 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 8819 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8820 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
8821 | /* nothing to be done ... */ | |
ac05f256 | 8822 | goto not_running; |
24dd469d | 8823 | |
1da177e4 | 8824 | if (mddev->pers->sync_request) { |
ef99bf48 | 8825 | if (spares) { |
a654b9d8 N |
8826 | /* We are adding a device or devices to an array |
8827 | * which has the bitmap stored on all devices. | |
8828 | * So make sure all bitmap pages get written | |
8829 | */ | |
8830 | bitmap_write_all(mddev->bitmap); | |
8831 | } | |
ac05f256 N |
8832 | INIT_WORK(&mddev->del_work, md_start_sync); |
8833 | queue_work(md_misc_wq, &mddev->del_work); | |
8834 | goto unlock; | |
1da177e4 | 8835 | } |
ac05f256 | 8836 | not_running: |
72a23c21 NB |
8837 | if (!mddev->sync_thread) { |
8838 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8839 | wake_up(&resync_wait); |
72a23c21 NB |
8840 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8841 | &mddev->recovery)) | |
0c3573f1 | 8842 | if (mddev->sysfs_action) |
00bcb4ac | 8843 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 8844 | } |
ac05f256 N |
8845 | unlock: |
8846 | wake_up(&mddev->sb_wait); | |
1da177e4 | 8847 | mddev_unlock(mddev); |
35bfc521 N |
8848 | } else if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) { |
8849 | /* Write superblock - thread that called mddev_suspend() | |
8850 | * holds reconfig_mutex for us. | |
8851 | */ | |
8852 | set_bit(MD_UPDATING_SB, &mddev->flags); | |
8853 | smp_mb__after_atomic(); | |
8854 | if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags)) | |
8855 | md_update_sb(mddev, 0); | |
8856 | clear_bit_unlock(MD_UPDATING_SB, &mddev->flags); | |
8857 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
8858 | } |
8859 | } | |
6c144d31 | 8860 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8861 | |
a91d5ac0 JB |
8862 | void md_reap_sync_thread(struct mddev *mddev) |
8863 | { | |
8864 | struct md_rdev *rdev; | |
8865 | ||
8866 | /* resync has finished, collect result */ | |
8867 | md_unregister_thread(&mddev->sync_thread); | |
8868 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8869 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8870 | /* success...*/ | |
8871 | /* activate any spares */ | |
8872 | if (mddev->pers->spare_active(mddev)) { | |
8873 | sysfs_notify(&mddev->kobj, NULL, | |
8874 | "degraded"); | |
2953079c | 8875 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a91d5ac0 JB |
8876 | } |
8877 | } | |
8878 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
8879 | mddev->pers->finish_reshape) | |
8880 | mddev->pers->finish_reshape(mddev); | |
8881 | ||
8882 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8883 | * information must be scrapped. |
a91d5ac0 | 8884 | */ |
f466722c N |
8885 | if (!mddev->degraded) |
8886 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8887 | rdev->saved_raid_disk = -1; |
8888 | ||
8889 | md_update_sb(mddev, 1); | |
2953079c | 8890 | /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can |
bb8bf15b GJ |
8891 | * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by |
8892 | * clustered raid */ | |
8893 | if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags)) | |
8894 | md_cluster_ops->resync_finish(mddev); | |
a91d5ac0 | 8895 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
ea358cd0 | 8896 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
a91d5ac0 JB |
8897 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8898 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8899 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8900 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8901 | wake_up(&resync_wait); |
a91d5ac0 JB |
8902 | /* flag recovery needed just to double check */ |
8903 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8904 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8905 | md_new_event(mddev); | |
8906 | if (mddev->event_work.func) | |
8907 | queue_work(md_misc_wq, &mddev->event_work); | |
8908 | } | |
6c144d31 | 8909 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 8910 | |
fd01b88c | 8911 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8912 | { |
00bcb4ac | 8913 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8914 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8915 | !test_bit(Blocked, &rdev->flags) && |
8916 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8917 | msecs_to_jiffies(5000)); |
8918 | rdev_dec_pending(rdev, mddev); | |
8919 | } | |
8920 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8921 | ||
c6563a8c N |
8922 | void md_finish_reshape(struct mddev *mddev) |
8923 | { | |
8924 | /* called be personality module when reshape completes. */ | |
8925 | struct md_rdev *rdev; | |
8926 | ||
8927 | rdev_for_each(rdev, mddev) { | |
8928 | if (rdev->data_offset > rdev->new_data_offset) | |
8929 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8930 | else | |
8931 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8932 | rdev->data_offset = rdev->new_data_offset; | |
8933 | } | |
8934 | } | |
8935 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 | 8936 | |
fc974ee2 | 8937 | /* Bad block management */ |
2230dfe4 | 8938 | |
fc974ee2 | 8939 | /* Returns 1 on success, 0 on failure */ |
3cb03002 | 8940 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8941 | int is_new) |
2230dfe4 | 8942 | { |
85ad1d13 | 8943 | struct mddev *mddev = rdev->mddev; |
c6563a8c N |
8944 | int rv; |
8945 | if (is_new) | |
8946 | s += rdev->new_data_offset; | |
8947 | else | |
8948 | s += rdev->data_offset; | |
fc974ee2 VV |
8949 | rv = badblocks_set(&rdev->badblocks, s, sectors, 0); |
8950 | if (rv == 0) { | |
2230dfe4 | 8951 | /* Make sure they get written out promptly */ |
35b785f7 TM |
8952 | if (test_bit(ExternalBbl, &rdev->flags)) |
8953 | sysfs_notify(&rdev->kobj, NULL, | |
8954 | "unacknowledged_bad_blocks"); | |
8bd2f0a0 | 8955 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2953079c SL |
8956 | set_mask_bits(&mddev->sb_flags, 0, |
8957 | BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING)); | |
2230dfe4 | 8958 | md_wakeup_thread(rdev->mddev->thread); |
fc974ee2 VV |
8959 | return 1; |
8960 | } else | |
8961 | return 0; | |
2230dfe4 N |
8962 | } |
8963 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8964 | ||
c6563a8c N |
8965 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8966 | int is_new) | |
2230dfe4 | 8967 | { |
35b785f7 | 8968 | int rv; |
c6563a8c N |
8969 | if (is_new) |
8970 | s += rdev->new_data_offset; | |
8971 | else | |
8972 | s += rdev->data_offset; | |
35b785f7 TM |
8973 | rv = badblocks_clear(&rdev->badblocks, s, sectors); |
8974 | if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags)) | |
8975 | sysfs_notify(&rdev->kobj, NULL, "bad_blocks"); | |
8976 | return rv; | |
2230dfe4 N |
8977 | } |
8978 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8979 | ||
75c96f85 AB |
8980 | static int md_notify_reboot(struct notifier_block *this, |
8981 | unsigned long code, void *x) | |
1da177e4 LT |
8982 | { |
8983 | struct list_head *tmp; | |
fd01b88c | 8984 | struct mddev *mddev; |
2dba6a91 | 8985 | int need_delay = 0; |
1da177e4 | 8986 | |
c744a65c N |
8987 | for_each_mddev(mddev, tmp) { |
8988 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8989 | if (mddev->pers) |
8990 | __md_stop_writes(mddev); | |
0f62fb22 N |
8991 | if (mddev->persistent) |
8992 | mddev->safemode = 2; | |
c744a65c | 8993 | mddev_unlock(mddev); |
2dba6a91 | 8994 | } |
c744a65c | 8995 | need_delay = 1; |
1da177e4 | 8996 | } |
c744a65c N |
8997 | /* |
8998 | * certain more exotic SCSI devices are known to be | |
8999 | * volatile wrt too early system reboots. While the | |
9000 | * right place to handle this issue is the given | |
9001 | * driver, we do want to have a safe RAID driver ... | |
9002 | */ | |
9003 | if (need_delay) | |
9004 | mdelay(1000*1); | |
9005 | ||
1da177e4 LT |
9006 | return NOTIFY_DONE; |
9007 | } | |
9008 | ||
75c96f85 | 9009 | static struct notifier_block md_notifier = { |
1da177e4 LT |
9010 | .notifier_call = md_notify_reboot, |
9011 | .next = NULL, | |
9012 | .priority = INT_MAX, /* before any real devices */ | |
9013 | }; | |
9014 | ||
9015 | static void md_geninit(void) | |
9016 | { | |
36a4e1fe | 9017 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 9018 | |
c7705f34 | 9019 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
9020 | } |
9021 | ||
75c96f85 | 9022 | static int __init md_init(void) |
1da177e4 | 9023 | { |
e804ac78 TH |
9024 | int ret = -ENOMEM; |
9025 | ||
ada609ee | 9026 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
9027 | if (!md_wq) |
9028 | goto err_wq; | |
9029 | ||
9030 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
9031 | if (!md_misc_wq) | |
9032 | goto err_misc_wq; | |
9033 | ||
9034 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
9035 | goto err_md; | |
9036 | ||
9037 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
9038 | goto err_mdp; | |
9039 | mdp_major = ret; | |
9040 | ||
af5628f0 | 9041 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
9042 | md_probe, NULL, NULL); |
9043 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
9044 | md_probe, NULL, NULL); |
9045 | ||
1da177e4 | 9046 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 9047 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
9048 | |
9049 | md_geninit(); | |
d710e138 | 9050 | return 0; |
1da177e4 | 9051 | |
e804ac78 TH |
9052 | err_mdp: |
9053 | unregister_blkdev(MD_MAJOR, "md"); | |
9054 | err_md: | |
9055 | destroy_workqueue(md_misc_wq); | |
9056 | err_misc_wq: | |
9057 | destroy_workqueue(md_wq); | |
9058 | err_wq: | |
9059 | return ret; | |
9060 | } | |
1da177e4 | 9061 | |
70bcecdb | 9062 | static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev) |
1d7e3e96 | 9063 | { |
70bcecdb GR |
9064 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
9065 | struct md_rdev *rdev2; | |
9066 | int role, ret; | |
9067 | char b[BDEVNAME_SIZE]; | |
1d7e3e96 | 9068 | |
818da59f GJ |
9069 | /* |
9070 | * If size is changed in another node then we need to | |
9071 | * do resize as well. | |
9072 | */ | |
9073 | if (mddev->dev_sectors != le64_to_cpu(sb->size)) { | |
9074 | ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size)); | |
9075 | if (ret) | |
9076 | pr_info("md-cluster: resize failed\n"); | |
9077 | else | |
9078 | bitmap_update_sb(mddev->bitmap); | |
9079 | } | |
9080 | ||
70bcecdb GR |
9081 | /* Check for change of roles in the active devices */ |
9082 | rdev_for_each(rdev2, mddev) { | |
9083 | if (test_bit(Faulty, &rdev2->flags)) | |
9084 | continue; | |
9085 | ||
9086 | /* Check if the roles changed */ | |
9087 | role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]); | |
dbb64f86 GR |
9088 | |
9089 | if (test_bit(Candidate, &rdev2->flags)) { | |
9090 | if (role == 0xfffe) { | |
9091 | pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b)); | |
9092 | md_kick_rdev_from_array(rdev2); | |
9093 | continue; | |
9094 | } | |
9095 | else | |
9096 | clear_bit(Candidate, &rdev2->flags); | |
9097 | } | |
9098 | ||
70bcecdb GR |
9099 | if (role != rdev2->raid_disk) { |
9100 | /* got activated */ | |
9101 | if (rdev2->raid_disk == -1 && role != 0xffff) { | |
9102 | rdev2->saved_raid_disk = role; | |
9103 | ret = remove_and_add_spares(mddev, rdev2); | |
9104 | pr_info("Activated spare: %s\n", | |
9d48739e | 9105 | bdevname(rdev2->bdev,b)); |
a578183e GJ |
9106 | /* wakeup mddev->thread here, so array could |
9107 | * perform resync with the new activated disk */ | |
9108 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
9109 | md_wakeup_thread(mddev->thread); | |
9110 | ||
70bcecdb GR |
9111 | } |
9112 | /* device faulty | |
9113 | * We just want to do the minimum to mark the disk | |
9114 | * as faulty. The recovery is performed by the | |
9115 | * one who initiated the error. | |
9116 | */ | |
9117 | if ((role == 0xfffe) || (role == 0xfffd)) { | |
9118 | md_error(mddev, rdev2); | |
9119 | clear_bit(Blocked, &rdev2->flags); | |
9120 | } | |
9121 | } | |
1d7e3e96 | 9122 | } |
70bcecdb | 9123 | |
28c1b9fd GR |
9124 | if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) |
9125 | update_raid_disks(mddev, le32_to_cpu(sb->raid_disks)); | |
70bcecdb GR |
9126 | |
9127 | /* Finally set the event to be up to date */ | |
9128 | mddev->events = le64_to_cpu(sb->events); | |
9129 | } | |
9130 | ||
9131 | static int read_rdev(struct mddev *mddev, struct md_rdev *rdev) | |
9132 | { | |
9133 | int err; | |
9134 | struct page *swapout = rdev->sb_page; | |
9135 | struct mdp_superblock_1 *sb; | |
9136 | ||
9137 | /* Store the sb page of the rdev in the swapout temporary | |
9138 | * variable in case we err in the future | |
9139 | */ | |
9140 | rdev->sb_page = NULL; | |
7f0f0d87 N |
9141 | err = alloc_disk_sb(rdev); |
9142 | if (err == 0) { | |
9143 | ClearPageUptodate(rdev->sb_page); | |
9144 | rdev->sb_loaded = 0; | |
9145 | err = super_types[mddev->major_version]. | |
9146 | load_super(rdev, NULL, mddev->minor_version); | |
9147 | } | |
70bcecdb GR |
9148 | if (err < 0) { |
9149 | pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n", | |
9150 | __func__, __LINE__, rdev->desc_nr, err); | |
7f0f0d87 N |
9151 | if (rdev->sb_page) |
9152 | put_page(rdev->sb_page); | |
70bcecdb GR |
9153 | rdev->sb_page = swapout; |
9154 | rdev->sb_loaded = 1; | |
9155 | return err; | |
1d7e3e96 GR |
9156 | } |
9157 | ||
70bcecdb GR |
9158 | sb = page_address(rdev->sb_page); |
9159 | /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET | |
9160 | * is not set | |
9161 | */ | |
9162 | ||
9163 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)) | |
9164 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
9165 | ||
9166 | /* The other node finished recovery, call spare_active to set | |
9167 | * device In_sync and mddev->degraded | |
9168 | */ | |
9169 | if (rdev->recovery_offset == MaxSector && | |
9170 | !test_bit(In_sync, &rdev->flags) && | |
9171 | mddev->pers->spare_active(mddev)) | |
9172 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
9173 | ||
9174 | put_page(swapout); | |
9175 | return 0; | |
9176 | } | |
9177 | ||
9178 | void md_reload_sb(struct mddev *mddev, int nr) | |
9179 | { | |
9180 | struct md_rdev *rdev; | |
9181 | int err; | |
9182 | ||
9183 | /* Find the rdev */ | |
9184 | rdev_for_each_rcu(rdev, mddev) { | |
9185 | if (rdev->desc_nr == nr) | |
9186 | break; | |
9187 | } | |
9188 | ||
9189 | if (!rdev || rdev->desc_nr != nr) { | |
9190 | pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr); | |
9191 | return; | |
9192 | } | |
9193 | ||
9194 | err = read_rdev(mddev, rdev); | |
9195 | if (err < 0) | |
9196 | return; | |
9197 | ||
9198 | check_sb_changes(mddev, rdev); | |
9199 | ||
9200 | /* Read all rdev's to update recovery_offset */ | |
9201 | rdev_for_each_rcu(rdev, mddev) | |
9202 | read_rdev(mddev, rdev); | |
1d7e3e96 GR |
9203 | } |
9204 | EXPORT_SYMBOL(md_reload_sb); | |
9205 | ||
1da177e4 LT |
9206 | #ifndef MODULE |
9207 | ||
9208 | /* | |
9209 | * Searches all registered partitions for autorun RAID arrays | |
9210 | * at boot time. | |
9211 | */ | |
4d936ec1 | 9212 | |
5b1f5bc3 | 9213 | static DEFINE_MUTEX(detected_devices_mutex); |
4d936ec1 ME |
9214 | static LIST_HEAD(all_detected_devices); |
9215 | struct detected_devices_node { | |
9216 | struct list_head list; | |
9217 | dev_t dev; | |
9218 | }; | |
1da177e4 LT |
9219 | |
9220 | void md_autodetect_dev(dev_t dev) | |
9221 | { | |
4d936ec1 ME |
9222 | struct detected_devices_node *node_detected_dev; |
9223 | ||
9224 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
9225 | if (node_detected_dev) { | |
9226 | node_detected_dev->dev = dev; | |
5b1f5bc3 | 9227 | mutex_lock(&detected_devices_mutex); |
4d936ec1 | 9228 | list_add_tail(&node_detected_dev->list, &all_detected_devices); |
5b1f5bc3 | 9229 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 9230 | } |
1da177e4 LT |
9231 | } |
9232 | ||
1da177e4 LT |
9233 | static void autostart_arrays(int part) |
9234 | { | |
3cb03002 | 9235 | struct md_rdev *rdev; |
4d936ec1 ME |
9236 | struct detected_devices_node *node_detected_dev; |
9237 | dev_t dev; | |
9238 | int i_scanned, i_passed; | |
1da177e4 | 9239 | |
4d936ec1 ME |
9240 | i_scanned = 0; |
9241 | i_passed = 0; | |
1da177e4 | 9242 | |
9d48739e | 9243 | pr_info("md: Autodetecting RAID arrays.\n"); |
1da177e4 | 9244 | |
5b1f5bc3 | 9245 | mutex_lock(&detected_devices_mutex); |
4d936ec1 ME |
9246 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
9247 | i_scanned++; | |
9248 | node_detected_dev = list_entry(all_detected_devices.next, | |
9249 | struct detected_devices_node, list); | |
9250 | list_del(&node_detected_dev->list); | |
9251 | dev = node_detected_dev->dev; | |
9252 | kfree(node_detected_dev); | |
90bcf133 | 9253 | mutex_unlock(&detected_devices_mutex); |
df968c4e | 9254 | rdev = md_import_device(dev,0, 90); |
90bcf133 | 9255 | mutex_lock(&detected_devices_mutex); |
1da177e4 LT |
9256 | if (IS_ERR(rdev)) |
9257 | continue; | |
9258 | ||
403df478 | 9259 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 9260 | continue; |
403df478 | 9261 | |
d0fae18f | 9262 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 9263 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 9264 | i_passed++; |
1da177e4 | 9265 | } |
5b1f5bc3 | 9266 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 9267 | |
9d48739e | 9268 | pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed); |
1da177e4 LT |
9269 | |
9270 | autorun_devices(part); | |
9271 | } | |
9272 | ||
fdee8ae4 | 9273 | #endif /* !MODULE */ |
1da177e4 LT |
9274 | |
9275 | static __exit void md_exit(void) | |
9276 | { | |
fd01b88c | 9277 | struct mddev *mddev; |
1da177e4 | 9278 | struct list_head *tmp; |
e2f23b60 | 9279 | int delay = 1; |
8ab5e4c1 | 9280 | |
af5628f0 | 9281 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 9282 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 9283 | |
3dbd8c2e | 9284 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
9285 | unregister_blkdev(mdp_major, "mdp"); |
9286 | unregister_reboot_notifier(&md_notifier); | |
9287 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
9288 | |
9289 | /* We cannot unload the modules while some process is | |
9290 | * waiting for us in select() or poll() - wake them up | |
9291 | */ | |
9292 | md_unloading = 1; | |
9293 | while (waitqueue_active(&md_event_waiters)) { | |
9294 | /* not safe to leave yet */ | |
9295 | wake_up(&md_event_waiters); | |
9296 | msleep(delay); | |
9297 | delay += delay; | |
9298 | } | |
1da177e4 | 9299 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 9300 | |
29ac4aa3 | 9301 | for_each_mddev(mddev, tmp) { |
1da177e4 | 9302 | export_array(mddev); |
9356863c | 9303 | mddev->ctime = 0; |
d3374825 | 9304 | mddev->hold_active = 0; |
9356863c N |
9305 | /* |
9306 | * for_each_mddev() will call mddev_put() at the end of each | |
9307 | * iteration. As the mddev is now fully clear, this will | |
9308 | * schedule the mddev for destruction by a workqueue, and the | |
9309 | * destroy_workqueue() below will wait for that to complete. | |
9310 | */ | |
1da177e4 | 9311 | } |
e804ac78 TH |
9312 | destroy_workqueue(md_misc_wq); |
9313 | destroy_workqueue(md_wq); | |
1da177e4 LT |
9314 | } |
9315 | ||
685784aa | 9316 | subsys_initcall(md_init); |
1da177e4 LT |
9317 | module_exit(md_exit) |
9318 | ||
e4dca7b7 | 9319 | static int get_ro(char *buffer, const struct kernel_param *kp) |
f91de92e N |
9320 | { |
9321 | return sprintf(buffer, "%d", start_readonly); | |
9322 | } | |
e4dca7b7 | 9323 | static int set_ro(const char *val, const struct kernel_param *kp) |
f91de92e | 9324 | { |
4c9309c0 | 9325 | return kstrtouint(val, 10, (unsigned int *)&start_readonly); |
f91de92e N |
9326 | } |
9327 | ||
80ca3a44 N |
9328 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
9329 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 9330 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
78b6350d | 9331 | module_param(create_on_open, bool, S_IRUSR|S_IWUSR); |
f91de92e | 9332 | |
1da177e4 | 9333 | MODULE_LICENSE("GPL"); |
0efb9e61 | 9334 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 9335 | MODULE_ALIAS("md"); |
72008652 | 9336 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |