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 | ||
28dec870 KO |
87 | static struct kobj_type md_ktype; |
88 | ||
edb39c9d | 89 | struct md_cluster_operations *md_cluster_ops; |
589a1c49 | 90 | EXPORT_SYMBOL(md_cluster_ops); |
edb39c9d GR |
91 | struct module *md_cluster_mod; |
92 | EXPORT_SYMBOL(md_cluster_mod); | |
93 | ||
90b08710 | 94 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
95 | static struct workqueue_struct *md_wq; |
96 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 97 | |
746d3207 N |
98 | static int remove_and_add_spares(struct mddev *mddev, |
99 | struct md_rdev *this); | |
5aa61f42 | 100 | static void mddev_detach(struct mddev *mddev); |
746d3207 | 101 | |
1e50915f RB |
102 | /* |
103 | * Default number of read corrections we'll attempt on an rdev | |
104 | * before ejecting it from the array. We divide the read error | |
105 | * count by 2 for every hour elapsed between read errors. | |
106 | */ | |
107 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
108 | /* |
109 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
110 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
111 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 112 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
113 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
114 | * speed limit - in case reconstruction slows down your system despite | |
115 | * idle IO detection. | |
116 | * | |
117 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 118 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
119 | */ |
120 | ||
121 | static int sysctl_speed_limit_min = 1000; | |
122 | static int sysctl_speed_limit_max = 200000; | |
fd01b88c | 123 | static inline int speed_min(struct mddev *mddev) |
88202a0c N |
124 | { |
125 | return mddev->sync_speed_min ? | |
126 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
127 | } | |
128 | ||
fd01b88c | 129 | static inline int speed_max(struct mddev *mddev) |
88202a0c N |
130 | { |
131 | return mddev->sync_speed_max ? | |
132 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
133 | } | |
1da177e4 | 134 | |
5a409b4f XN |
135 | static void * flush_info_alloc(gfp_t gfp_flags, void *data) |
136 | { | |
137 | return kzalloc(sizeof(struct flush_info), gfp_flags); | |
138 | } | |
139 | static void flush_info_free(void *flush_info, void *data) | |
140 | { | |
141 | kfree(flush_info); | |
142 | } | |
143 | ||
144 | static void * flush_bio_alloc(gfp_t gfp_flags, void *data) | |
145 | { | |
146 | return kzalloc(sizeof(struct flush_bio), gfp_flags); | |
147 | } | |
148 | static void flush_bio_free(void *flush_bio, void *data) | |
149 | { | |
150 | kfree(flush_bio); | |
151 | } | |
152 | ||
1da177e4 LT |
153 | static struct ctl_table_header *raid_table_header; |
154 | ||
82592c38 | 155 | static struct ctl_table raid_table[] = { |
1da177e4 | 156 | { |
1da177e4 LT |
157 | .procname = "speed_limit_min", |
158 | .data = &sysctl_speed_limit_min, | |
159 | .maxlen = sizeof(int), | |
80ca3a44 | 160 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 161 | .proc_handler = proc_dointvec, |
1da177e4 LT |
162 | }, |
163 | { | |
1da177e4 LT |
164 | .procname = "speed_limit_max", |
165 | .data = &sysctl_speed_limit_max, | |
166 | .maxlen = sizeof(int), | |
80ca3a44 | 167 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 168 | .proc_handler = proc_dointvec, |
1da177e4 | 169 | }, |
894d2491 | 170 | { } |
1da177e4 LT |
171 | }; |
172 | ||
82592c38 | 173 | static struct ctl_table raid_dir_table[] = { |
1da177e4 | 174 | { |
1da177e4 LT |
175 | .procname = "raid", |
176 | .maxlen = 0, | |
80ca3a44 | 177 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
178 | .child = raid_table, |
179 | }, | |
894d2491 | 180 | { } |
1da177e4 LT |
181 | }; |
182 | ||
82592c38 | 183 | static struct ctl_table raid_root_table[] = { |
1da177e4 | 184 | { |
1da177e4 LT |
185 | .procname = "dev", |
186 | .maxlen = 0, | |
187 | .mode = 0555, | |
188 | .child = raid_dir_table, | |
189 | }, | |
894d2491 | 190 | { } |
1da177e4 LT |
191 | }; |
192 | ||
83d5cde4 | 193 | static const struct block_device_operations md_fops; |
1da177e4 | 194 | |
f91de92e N |
195 | static int start_readonly; |
196 | ||
78b6350d N |
197 | /* |
198 | * The original mechanism for creating an md device is to create | |
199 | * a device node in /dev and to open it. This causes races with device-close. | |
200 | * The preferred method is to write to the "new_array" module parameter. | |
201 | * This can avoid races. | |
202 | * Setting create_on_open to false disables the original mechanism | |
203 | * so all the races disappear. | |
204 | */ | |
205 | static bool create_on_open = true; | |
206 | ||
a167f663 | 207 | /* bio_clone_mddev |
9b10f6a9 | 208 | * like bio_clone_bioset, but with a local bio set |
a167f663 N |
209 | */ |
210 | ||
a167f663 | 211 | struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, |
fd01b88c | 212 | struct mddev *mddev) |
a167f663 N |
213 | { |
214 | struct bio *b; | |
a167f663 | 215 | |
afeee514 | 216 | if (!mddev || !bioset_initialized(&mddev->bio_set)) |
a167f663 N |
217 | return bio_alloc(gfp_mask, nr_iovecs); |
218 | ||
afeee514 | 219 | b = bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set); |
a167f663 N |
220 | if (!b) |
221 | return NULL; | |
a167f663 N |
222 | return b; |
223 | } | |
224 | EXPORT_SYMBOL_GPL(bio_alloc_mddev); | |
225 | ||
5a85071c N |
226 | static struct bio *md_bio_alloc_sync(struct mddev *mddev) |
227 | { | |
afeee514 | 228 | if (!mddev || !bioset_initialized(&mddev->sync_set)) |
5a85071c N |
229 | return bio_alloc(GFP_NOIO, 1); |
230 | ||
afeee514 | 231 | return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set); |
5a85071c N |
232 | } |
233 | ||
d7603b7e N |
234 | /* |
235 | * We have a system wide 'event count' that is incremented | |
236 | * on any 'interesting' event, and readers of /proc/mdstat | |
237 | * can use 'poll' or 'select' to find out when the event | |
238 | * count increases. | |
239 | * | |
240 | * Events are: | |
241 | * start array, stop array, error, add device, remove device, | |
242 | * start build, activate spare | |
243 | */ | |
2989ddbd | 244 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 245 | static atomic_t md_event_count; |
fd01b88c | 246 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
247 | { |
248 | atomic_inc(&md_event_count); | |
249 | wake_up(&md_event_waiters); | |
250 | } | |
29269553 | 251 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 252 | |
1da177e4 LT |
253 | /* |
254 | * Enables to iterate over all existing md arrays | |
255 | * all_mddevs_lock protects this list. | |
256 | */ | |
257 | static LIST_HEAD(all_mddevs); | |
258 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
259 | ||
1da177e4 LT |
260 | /* |
261 | * iterates through all used mddevs in the system. | |
262 | * We take care to grab the all_mddevs_lock whenever navigating | |
263 | * the list, and to always hold a refcount when unlocked. | |
264 | * Any code which breaks out of this loop while own | |
265 | * a reference to the current mddev and must mddev_put it. | |
266 | */ | |
fd01b88c | 267 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 | 268 | \ |
f72ffdd6 | 269 | for (({ spin_lock(&all_mddevs_lock); \ |
fd01b88c N |
270 | _tmp = all_mddevs.next; \ |
271 | _mddev = NULL;}); \ | |
272 | ({ if (_tmp != &all_mddevs) \ | |
273 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 274 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
275 | if (_mddev) mddev_put(_mddev); \ |
276 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
277 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 278 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 279 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
280 | ) |
281 | ||
409c57f3 N |
282 | /* Rather than calling directly into the personality make_request function, |
283 | * IO requests come here first so that we can check if the device is | |
284 | * being suspended pending a reconfiguration. | |
285 | * We hold a refcount over the call to ->make_request. By the time that | |
286 | * call has finished, the bio has been linked into some internal structure | |
287 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
288 | */ | |
b3143b9a N |
289 | static bool is_suspended(struct mddev *mddev, struct bio *bio) |
290 | { | |
291 | if (mddev->suspended) | |
292 | return true; | |
293 | if (bio_data_dir(bio) != WRITE) | |
294 | return false; | |
295 | if (mddev->suspend_lo >= mddev->suspend_hi) | |
296 | return false; | |
297 | if (bio->bi_iter.bi_sector >= mddev->suspend_hi) | |
298 | return false; | |
299 | if (bio_end_sector(bio) < mddev->suspend_lo) | |
300 | return false; | |
301 | return true; | |
302 | } | |
303 | ||
393debc2 SL |
304 | void md_handle_request(struct mddev *mddev, struct bio *bio) |
305 | { | |
306 | check_suspended: | |
307 | rcu_read_lock(); | |
b3143b9a | 308 | if (is_suspended(mddev, bio)) { |
393debc2 SL |
309 | DEFINE_WAIT(__wait); |
310 | for (;;) { | |
311 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
312 | TASK_UNINTERRUPTIBLE); | |
b3143b9a | 313 | if (!is_suspended(mddev, bio)) |
393debc2 SL |
314 | break; |
315 | rcu_read_unlock(); | |
316 | schedule(); | |
317 | rcu_read_lock(); | |
318 | } | |
319 | finish_wait(&mddev->sb_wait, &__wait); | |
320 | } | |
321 | atomic_inc(&mddev->active_io); | |
322 | rcu_read_unlock(); | |
323 | ||
324 | if (!mddev->pers->make_request(mddev, bio)) { | |
325 | atomic_dec(&mddev->active_io); | |
326 | wake_up(&mddev->sb_wait); | |
327 | goto check_suspended; | |
328 | } | |
329 | ||
330 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) | |
331 | wake_up(&mddev->sb_wait); | |
332 | } | |
333 | EXPORT_SYMBOL(md_handle_request); | |
334 | ||
dece1635 | 335 | static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 336 | { |
49077326 | 337 | const int rw = bio_data_dir(bio); |
ddcf35d3 | 338 | const int sgrp = op_stat_group(bio_op(bio)); |
fd01b88c | 339 | struct mddev *mddev = q->queuedata; |
e91ece55 | 340 | unsigned int sectors; |
74672d06 | 341 | int cpu; |
49077326 | 342 | |
af67c31f | 343 | blk_queue_split(q, &bio); |
54efd50b | 344 | |
274d8cbd | 345 | if (mddev == NULL || mddev->pers == NULL) { |
409c57f3 | 346 | bio_io_error(bio); |
dece1635 | 347 | return BLK_QC_T_NONE; |
409c57f3 | 348 | } |
bbfa57c0 | 349 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
4246a0b6 | 350 | if (bio_sectors(bio) != 0) |
4e4cbee9 | 351 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 352 | bio_endio(bio); |
dece1635 | 353 | return BLK_QC_T_NONE; |
bbfa57c0 | 354 | } |
49077326 | 355 | |
e91ece55 CM |
356 | /* |
357 | * save the sectors now since our bio can | |
358 | * go away inside make_request | |
359 | */ | |
360 | sectors = bio_sectors(bio); | |
9c573de3 | 361 | /* bio could be mergeable after passing to underlayer */ |
1eff9d32 | 362 | bio->bi_opf &= ~REQ_NOMERGE; |
393debc2 SL |
363 | |
364 | md_handle_request(mddev, bio); | |
49077326 | 365 | |
74672d06 | 366 | cpu = part_stat_lock(); |
ddcf35d3 MC |
367 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[sgrp]); |
368 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[sgrp], sectors); | |
74672d06 | 369 | part_stat_unlock(); |
49077326 | 370 | |
dece1635 | 371 | return BLK_QC_T_NONE; |
409c57f3 N |
372 | } |
373 | ||
9e35b99c N |
374 | /* mddev_suspend makes sure no new requests are submitted |
375 | * to the device, and that any requests that have been submitted | |
376 | * are completely handled. | |
afa0f557 N |
377 | * Once mddev_detach() is called and completes, the module will be |
378 | * completely unused. | |
9e35b99c | 379 | */ |
fd01b88c | 380 | void mddev_suspend(struct mddev *mddev) |
409c57f3 | 381 | { |
092398dc | 382 | WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk); |
4d5324f7 | 383 | lockdep_assert_held(&mddev->reconfig_mutex); |
0dc10e50 MP |
384 | if (mddev->suspended++) |
385 | return; | |
409c57f3 | 386 | synchronize_rcu(); |
cc27b0c7 | 387 | wake_up(&mddev->sb_wait); |
35bfc521 N |
388 | set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags); |
389 | smp_mb__after_atomic(); | |
409c57f3 N |
390 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); |
391 | mddev->pers->quiesce(mddev, 1); | |
35bfc521 N |
392 | clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags); |
393 | wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags)); | |
0d9f4f13 JB |
394 | |
395 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 396 | } |
390ee602 | 397 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 398 | |
fd01b88c | 399 | void mddev_resume(struct mddev *mddev) |
409c57f3 | 400 | { |
4d5324f7 | 401 | lockdep_assert_held(&mddev->reconfig_mutex); |
0dc10e50 MP |
402 | if (--mddev->suspended) |
403 | return; | |
409c57f3 N |
404 | wake_up(&mddev->sb_wait); |
405 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 406 | |
47525e59 | 407 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
408 | md_wakeup_thread(mddev->thread); |
409 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 410 | } |
390ee602 | 411 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 412 | |
fd01b88c | 413 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 | 414 | { |
5c675f83 N |
415 | struct md_personality *pers = mddev->pers; |
416 | int ret = 0; | |
417 | ||
418 | rcu_read_lock(); | |
419 | if (mddev->suspended) | |
420 | ret = 1; | |
421 | else if (pers && pers->congested) | |
422 | ret = pers->congested(mddev, bits); | |
423 | rcu_read_unlock(); | |
424 | return ret; | |
425 | } | |
426 | EXPORT_SYMBOL_GPL(mddev_congested); | |
427 | static int md_congested(void *data, int bits) | |
428 | { | |
429 | struct mddev *mddev = data; | |
430 | return mddev_congested(mddev, bits); | |
3fa841d7 | 431 | } |
3fa841d7 | 432 | |
a2826aa9 | 433 | /* |
e9c7469b | 434 | * Generic flush handling for md |
a2826aa9 | 435 | */ |
5a409b4f XN |
436 | static void submit_flushes(struct work_struct *ws) |
437 | { | |
438 | struct flush_info *fi = container_of(ws, struct flush_info, flush_work); | |
439 | struct mddev *mddev = fi->mddev; | |
440 | struct bio *bio = fi->bio; | |
441 | ||
442 | bio->bi_opf &= ~REQ_PREFLUSH; | |
443 | md_handle_request(mddev, bio); | |
444 | ||
445 | mempool_free(fi, mddev->flush_pool); | |
446 | } | |
a2826aa9 | 447 | |
5a409b4f | 448 | static void md_end_flush(struct bio *fbio) |
a2826aa9 | 449 | { |
5a409b4f XN |
450 | struct flush_bio *fb = fbio->bi_private; |
451 | struct md_rdev *rdev = fb->rdev; | |
452 | struct flush_info *fi = fb->fi; | |
453 | struct bio *bio = fi->bio; | |
454 | struct mddev *mddev = fi->mddev; | |
a2826aa9 N |
455 | |
456 | rdev_dec_pending(rdev, mddev); | |
457 | ||
5a409b4f XN |
458 | if (atomic_dec_and_test(&fi->flush_pending)) { |
459 | if (bio->bi_iter.bi_size == 0) | |
460 | /* an empty barrier - all done */ | |
461 | bio_endio(bio); | |
462 | else { | |
463 | INIT_WORK(&fi->flush_work, submit_flushes); | |
464 | queue_work(md_wq, &fi->flush_work); | |
465 | } | |
a2826aa9 | 466 | } |
a2826aa9 | 467 | |
5a409b4f XN |
468 | mempool_free(fb, mddev->flush_bio_pool); |
469 | bio_put(fbio); | |
470 | } | |
a7a07e69 | 471 | |
5a409b4f | 472 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 | 473 | { |
3cb03002 | 474 | struct md_rdev *rdev; |
5a409b4f XN |
475 | struct flush_info *fi; |
476 | ||
477 | fi = mempool_alloc(mddev->flush_pool, GFP_NOIO); | |
478 | ||
479 | fi->bio = bio; | |
480 | fi->mddev = mddev; | |
481 | atomic_set(&fi->flush_pending, 1); | |
a2826aa9 N |
482 | |
483 | rcu_read_lock(); | |
dafb20fa | 484 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
485 | if (rdev->raid_disk >= 0 && |
486 | !test_bit(Faulty, &rdev->flags)) { | |
487 | /* Take two references, one is dropped | |
488 | * when request finishes, one after | |
489 | * we reclaim rcu_read_lock | |
490 | */ | |
491 | struct bio *bi; | |
5a409b4f | 492 | struct flush_bio *fb; |
a2826aa9 N |
493 | atomic_inc(&rdev->nr_pending); |
494 | atomic_inc(&rdev->nr_pending); | |
495 | rcu_read_unlock(); | |
5a409b4f XN |
496 | |
497 | fb = mempool_alloc(mddev->flush_bio_pool, GFP_NOIO); | |
498 | fb->fi = fi; | |
499 | fb->rdev = rdev; | |
500 | ||
b5e1b8ce | 501 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
74d46992 | 502 | bio_set_dev(bi, rdev->bdev); |
5a409b4f XN |
503 | bi->bi_end_io = md_end_flush; |
504 | bi->bi_private = fb; | |
70fd7614 | 505 | bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
5a409b4f XN |
506 | |
507 | atomic_inc(&fi->flush_pending); | |
4e49ea4a | 508 | submit_bio(bi); |
5a409b4f | 509 | |
a2826aa9 N |
510 | rcu_read_lock(); |
511 | rdev_dec_pending(rdev, mddev); | |
512 | } | |
513 | rcu_read_unlock(); | |
a2826aa9 | 514 | |
5a409b4f XN |
515 | if (atomic_dec_and_test(&fi->flush_pending)) { |
516 | if (bio->bi_iter.bi_size == 0) | |
517 | /* an empty barrier - all done */ | |
518 | bio_endio(bio); | |
519 | else { | |
520 | INIT_WORK(&fi->flush_work, submit_flushes); | |
521 | queue_work(md_wq, &fi->flush_work); | |
522 | } | |
a2826aa9 | 523 | } |
a2826aa9 | 524 | } |
e9c7469b | 525 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 526 | |
fd01b88c | 527 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
528 | { |
529 | atomic_inc(&mddev->active); | |
530 | return mddev; | |
531 | } | |
532 | ||
5fd3a17e | 533 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 534 | |
fd01b88c | 535 | static void mddev_put(struct mddev *mddev) |
1da177e4 LT |
536 | { |
537 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) | |
538 | return; | |
d3374825 | 539 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
540 | mddev->ctime == 0 && !mddev->hold_active) { |
541 | /* Array is not configured at all, and not held active, | |
542 | * so destroy it */ | |
af8a2434 | 543 | list_del_init(&mddev->all_mddevs); |
28dec870 KO |
544 | |
545 | /* | |
546 | * Call queue_work inside the spinlock so that | |
547 | * flush_workqueue() after mddev_find will succeed in waiting | |
548 | * for the work to be done. | |
549 | */ | |
550 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
551 | queue_work(md_misc_wq, &mddev->del_work); | |
d3374825 N |
552 | } |
553 | spin_unlock(&all_mddevs_lock); | |
1da177e4 LT |
554 | } |
555 | ||
8376d3c1 | 556 | static void md_safemode_timeout(struct timer_list *t); |
25b2edfa | 557 | |
fd01b88c | 558 | void mddev_init(struct mddev *mddev) |
fafd7fb0 | 559 | { |
28dec870 | 560 | kobject_init(&mddev->kobj, &md_ktype); |
fafd7fb0 N |
561 | mutex_init(&mddev->open_mutex); |
562 | mutex_init(&mddev->reconfig_mutex); | |
563 | mutex_init(&mddev->bitmap_info.mutex); | |
564 | INIT_LIST_HEAD(&mddev->disks); | |
565 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
8376d3c1 | 566 | timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0); |
fafd7fb0 N |
567 | atomic_set(&mddev->active, 1); |
568 | atomic_set(&mddev->openers, 0); | |
569 | atomic_set(&mddev->active_io, 0); | |
85572d7c | 570 | spin_lock_init(&mddev->lock); |
fafd7fb0 N |
571 | init_waitqueue_head(&mddev->sb_wait); |
572 | init_waitqueue_head(&mddev->recovery_wait); | |
573 | mddev->reshape_position = MaxSector; | |
2c810cdd | 574 | mddev->reshape_backwards = 0; |
c4a39551 | 575 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
576 | mddev->resync_min = 0; |
577 | mddev->resync_max = MaxSector; | |
578 | mddev->level = LEVEL_NONE; | |
579 | } | |
390ee602 | 580 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 581 | |
f72ffdd6 | 582 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 583 | { |
fd01b88c | 584 | struct mddev *mddev, *new = NULL; |
1da177e4 | 585 | |
8f5f02c4 N |
586 | if (unit && MAJOR(unit) != MD_MAJOR) |
587 | unit &= ~((1<<MdpMinorShift)-1); | |
588 | ||
1da177e4 LT |
589 | retry: |
590 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
591 | |
592 | if (unit) { | |
593 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
594 | if (mddev->unit == unit) { | |
595 | mddev_get(mddev); | |
596 | spin_unlock(&all_mddevs_lock); | |
597 | kfree(new); | |
598 | return mddev; | |
599 | } | |
600 | ||
601 | if (new) { | |
602 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 603 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
604 | new->hold_active = UNTIL_IOCTL; |
605 | return new; | |
1da177e4 | 606 | } |
efeb53c0 N |
607 | } else if (new) { |
608 | /* find an unused unit number */ | |
609 | static int next_minor = 512; | |
610 | int start = next_minor; | |
611 | int is_free = 0; | |
612 | int dev = 0; | |
613 | while (!is_free) { | |
614 | dev = MKDEV(MD_MAJOR, next_minor); | |
615 | next_minor++; | |
616 | if (next_minor > MINORMASK) | |
617 | next_minor = 0; | |
618 | if (next_minor == start) { | |
619 | /* Oh dear, all in use. */ | |
620 | spin_unlock(&all_mddevs_lock); | |
621 | kfree(new); | |
622 | return NULL; | |
623 | } | |
f72ffdd6 | 624 | |
efeb53c0 N |
625 | is_free = 1; |
626 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
627 | if (mddev->unit == dev) { | |
628 | is_free = 0; | |
629 | break; | |
630 | } | |
631 | } | |
632 | new->unit = dev; | |
633 | new->md_minor = MINOR(dev); | |
634 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
635 | list_add(&new->all_mddevs, &all_mddevs); |
636 | spin_unlock(&all_mddevs_lock); | |
637 | return new; | |
638 | } | |
639 | spin_unlock(&all_mddevs_lock); | |
640 | ||
9ffae0cf | 641 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
642 | if (!new) |
643 | return NULL; | |
644 | ||
1da177e4 LT |
645 | new->unit = unit; |
646 | if (MAJOR(unit) == MD_MAJOR) | |
647 | new->md_minor = MINOR(unit); | |
648 | else | |
649 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
650 | ||
fafd7fb0 | 651 | mddev_init(new); |
1da177e4 | 652 | |
1da177e4 LT |
653 | goto retry; |
654 | } | |
655 | ||
b6eb127d N |
656 | static struct attribute_group md_redundancy_group; |
657 | ||
5c47daf6 | 658 | void mddev_unlock(struct mddev *mddev) |
1da177e4 | 659 | { |
a64c876f | 660 | if (mddev->to_remove) { |
b6eb127d N |
661 | /* These cannot be removed under reconfig_mutex as |
662 | * an access to the files will try to take reconfig_mutex | |
663 | * while holding the file unremovable, which leads to | |
664 | * a deadlock. | |
bb4f1e9d N |
665 | * So hold set sysfs_active while the remove in happeing, |
666 | * and anything else which might set ->to_remove or my | |
667 | * otherwise change the sysfs namespace will fail with | |
668 | * -EBUSY if sysfs_active is still set. | |
669 | * We set sysfs_active under reconfig_mutex and elsewhere | |
670 | * test it under the same mutex to ensure its correct value | |
671 | * is seen. | |
b6eb127d | 672 | */ |
a64c876f N |
673 | struct attribute_group *to_remove = mddev->to_remove; |
674 | mddev->to_remove = NULL; | |
bb4f1e9d | 675 | mddev->sysfs_active = 1; |
b6eb127d N |
676 | mutex_unlock(&mddev->reconfig_mutex); |
677 | ||
00bcb4ac N |
678 | if (mddev->kobj.sd) { |
679 | if (to_remove != &md_redundancy_group) | |
680 | sysfs_remove_group(&mddev->kobj, to_remove); | |
681 | if (mddev->pers == NULL || | |
682 | mddev->pers->sync_request == NULL) { | |
683 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
684 | if (mddev->sysfs_action) | |
685 | sysfs_put(mddev->sysfs_action); | |
686 | mddev->sysfs_action = NULL; | |
687 | } | |
a64c876f | 688 | } |
bb4f1e9d | 689 | mddev->sysfs_active = 0; |
b6eb127d N |
690 | } else |
691 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 692 | |
751e67ca CD |
693 | /* As we've dropped the mutex we need a spinlock to |
694 | * make sure the thread doesn't disappear | |
01f96c0a N |
695 | */ |
696 | spin_lock(&pers_lock); | |
005eca5e | 697 | md_wakeup_thread(mddev->thread); |
4d5324f7 | 698 | wake_up(&mddev->sb_wait); |
01f96c0a | 699 | spin_unlock(&pers_lock); |
1da177e4 | 700 | } |
5c47daf6 | 701 | EXPORT_SYMBOL_GPL(mddev_unlock); |
1da177e4 | 702 | |
57d051dc | 703 | struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr) |
1ca69c4b N |
704 | { |
705 | struct md_rdev *rdev; | |
706 | ||
707 | rdev_for_each_rcu(rdev, mddev) | |
708 | if (rdev->desc_nr == nr) | |
709 | return rdev; | |
710 | ||
711 | return NULL; | |
712 | } | |
57d051dc | 713 | EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu); |
1ca69c4b N |
714 | |
715 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 716 | { |
3cb03002 | 717 | struct md_rdev *rdev; |
1da177e4 | 718 | |
dafb20fa | 719 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
720 | if (rdev->bdev->bd_dev == dev) |
721 | return rdev; | |
159ec1fc | 722 | |
1da177e4 LT |
723 | return NULL; |
724 | } | |
725 | ||
1532d9e8 | 726 | struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev) |
1ca69c4b N |
727 | { |
728 | struct md_rdev *rdev; | |
729 | ||
730 | rdev_for_each_rcu(rdev, mddev) | |
731 | if (rdev->bdev->bd_dev == dev) | |
732 | return rdev; | |
733 | ||
734 | return NULL; | |
735 | } | |
1532d9e8 | 736 | EXPORT_SYMBOL_GPL(md_find_rdev_rcu); |
1ca69c4b | 737 | |
84fc4b56 | 738 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 739 | { |
84fc4b56 | 740 | struct md_personality *pers; |
d9d166c2 N |
741 | list_for_each_entry(pers, &pers_list, list) { |
742 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 743 | return pers; |
d9d166c2 N |
744 | if (strcmp(pers->name, clevel)==0) |
745 | return pers; | |
746 | } | |
2604b703 N |
747 | return NULL; |
748 | } | |
749 | ||
b73df2d3 | 750 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 751 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 752 | { |
57b2caa3 | 753 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 754 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
755 | } |
756 | ||
f72ffdd6 | 757 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 758 | { |
1da177e4 | 759 | rdev->sb_page = alloc_page(GFP_KERNEL); |
7f0f0d87 | 760 | if (!rdev->sb_page) |
ebc24337 | 761 | return -ENOMEM; |
1da177e4 LT |
762 | return 0; |
763 | } | |
764 | ||
545c8795 | 765 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
766 | { |
767 | if (rdev->sb_page) { | |
2d1f3b5d | 768 | put_page(rdev->sb_page); |
1da177e4 LT |
769 | rdev->sb_loaded = 0; |
770 | rdev->sb_page = NULL; | |
0f420358 | 771 | rdev->sb_start = 0; |
dd8ac336 | 772 | rdev->sectors = 0; |
1da177e4 | 773 | } |
2699b672 N |
774 | if (rdev->bb_page) { |
775 | put_page(rdev->bb_page); | |
776 | rdev->bb_page = NULL; | |
777 | } | |
d3b407fb | 778 | badblocks_exit(&rdev->badblocks); |
1da177e4 | 779 | } |
545c8795 | 780 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 781 | |
4246a0b6 | 782 | static void super_written(struct bio *bio) |
7bfa19f2 | 783 | { |
3cb03002 | 784 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 785 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 786 | |
4e4cbee9 CH |
787 | if (bio->bi_status) { |
788 | pr_err("md: super_written gets error=%d\n", bio->bi_status); | |
a9701a30 | 789 | md_error(mddev, rdev); |
46533ff7 N |
790 | if (!test_bit(Faulty, &rdev->flags) |
791 | && (bio->bi_opf & MD_FAILFAST)) { | |
2953079c | 792 | set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags); |
46533ff7 N |
793 | set_bit(LastDev, &rdev->flags); |
794 | } | |
795 | } else | |
796 | clear_bit(LastDev, &rdev->flags); | |
7bfa19f2 | 797 | |
a9701a30 N |
798 | if (atomic_dec_and_test(&mddev->pending_writes)) |
799 | wake_up(&mddev->sb_wait); | |
ed3b98c7 | 800 | rdev_dec_pending(rdev, mddev); |
f8b58edf | 801 | bio_put(bio); |
7bfa19f2 N |
802 | } |
803 | ||
fd01b88c | 804 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
805 | sector_t sector, int size, struct page *page) |
806 | { | |
807 | /* write first size bytes of page to sector of rdev | |
808 | * Increment mddev->pending_writes before returning | |
809 | * and decrement it on completion, waking up sb_wait | |
810 | * if zero is reached. | |
811 | * If an error occurred, call md_error | |
812 | */ | |
46533ff7 N |
813 | struct bio *bio; |
814 | int ff = 0; | |
815 | ||
4b6c1060 HM |
816 | if (!page) |
817 | return; | |
818 | ||
46533ff7 N |
819 | if (test_bit(Faulty, &rdev->flags)) |
820 | return; | |
821 | ||
5a85071c | 822 | bio = md_bio_alloc_sync(mddev); |
7bfa19f2 | 823 | |
ed3b98c7 SL |
824 | atomic_inc(&rdev->nr_pending); |
825 | ||
74d46992 | 826 | bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev); |
4f024f37 | 827 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
828 | bio_add_page(bio, page, size, 0); |
829 | bio->bi_private = rdev; | |
830 | bio->bi_end_io = super_written; | |
46533ff7 N |
831 | |
832 | if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) && | |
833 | test_bit(FailFast, &rdev->flags) && | |
834 | !test_bit(LastDev, &rdev->flags)) | |
835 | ff = MD_FAILFAST; | |
5a8948f8 | 836 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff; |
a9701a30 | 837 | |
7bfa19f2 | 838 | atomic_inc(&mddev->pending_writes); |
4e49ea4a | 839 | submit_bio(bio); |
a9701a30 N |
840 | } |
841 | ||
46533ff7 | 842 | int md_super_wait(struct mddev *mddev) |
a9701a30 | 843 | { |
e9c7469b | 844 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 845 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
2953079c | 846 | if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags)) |
46533ff7 N |
847 | return -EAGAIN; |
848 | return 0; | |
7bfa19f2 N |
849 | } |
850 | ||
3cb03002 | 851 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
796a5cf0 | 852 | struct page *page, int op, int op_flags, bool metadata_op) |
1da177e4 | 853 | { |
5a85071c | 854 | struct bio *bio = md_bio_alloc_sync(rdev->mddev); |
1da177e4 LT |
855 | int ret; |
856 | ||
74d46992 CH |
857 | if (metadata_op && rdev->meta_bdev) |
858 | bio_set_dev(bio, rdev->meta_bdev); | |
859 | else | |
860 | bio_set_dev(bio, rdev->bdev); | |
796a5cf0 | 861 | bio_set_op_attrs(bio, op, op_flags); |
ccebd4c4 | 862 | if (metadata_op) |
4f024f37 | 863 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
864 | else if (rdev->mddev->reshape_position != MaxSector && |
865 | (rdev->mddev->reshape_backwards == | |
866 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 867 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 868 | else |
4f024f37 | 869 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 870 | bio_add_page(bio, page, size, 0); |
4e49ea4a MC |
871 | |
872 | submit_bio_wait(bio); | |
1da177e4 | 873 | |
4e4cbee9 | 874 | ret = !bio->bi_status; |
1da177e4 LT |
875 | bio_put(bio); |
876 | return ret; | |
877 | } | |
a8745db2 | 878 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 879 | |
f72ffdd6 | 880 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
881 | { |
882 | char b[BDEVNAME_SIZE]; | |
403df478 | 883 | |
1da177e4 LT |
884 | if (rdev->sb_loaded) |
885 | return 0; | |
886 | ||
796a5cf0 | 887 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) |
1da177e4 LT |
888 | goto fail; |
889 | rdev->sb_loaded = 1; | |
890 | return 0; | |
891 | ||
892 | fail: | |
9d48739e N |
893 | pr_err("md: disabled device %s, could not read superblock.\n", |
894 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
895 | return -EINVAL; |
896 | } | |
897 | ||
e6fd2093 | 898 | static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
1da177e4 | 899 | { |
f72ffdd6 | 900 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
901 | sb1->set_uuid1 == sb2->set_uuid1 && |
902 | sb1->set_uuid2 == sb2->set_uuid2 && | |
903 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
904 | } |
905 | ||
e6fd2093 | 906 | static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
1da177e4 LT |
907 | { |
908 | int ret; | |
909 | mdp_super_t *tmp1, *tmp2; | |
910 | ||
911 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
912 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
913 | ||
914 | if (!tmp1 || !tmp2) { | |
915 | ret = 0; | |
1da177e4 LT |
916 | goto abort; |
917 | } | |
918 | ||
919 | *tmp1 = *sb1; | |
920 | *tmp2 = *sb2; | |
921 | ||
922 | /* | |
923 | * nr_disks is not constant | |
924 | */ | |
925 | tmp1->nr_disks = 0; | |
926 | tmp2->nr_disks = 0; | |
927 | ||
ce0c8e05 | 928 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 929 | abort: |
990a8baf JJ |
930 | kfree(tmp1); |
931 | kfree(tmp2); | |
1da177e4 LT |
932 | return ret; |
933 | } | |
934 | ||
4d167f09 N |
935 | static u32 md_csum_fold(u32 csum) |
936 | { | |
937 | csum = (csum & 0xffff) + (csum >> 16); | |
938 | return (csum & 0xffff) + (csum >> 16); | |
939 | } | |
940 | ||
f72ffdd6 | 941 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 942 | { |
4d167f09 N |
943 | u64 newcsum = 0; |
944 | u32 *sb32 = (u32*)sb; | |
945 | int i; | |
1da177e4 LT |
946 | unsigned int disk_csum, csum; |
947 | ||
948 | disk_csum = sb->sb_csum; | |
949 | sb->sb_csum = 0; | |
4d167f09 N |
950 | |
951 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
952 | newcsum += sb32[i]; | |
953 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
954 | ||
4d167f09 N |
955 | #ifdef CONFIG_ALPHA |
956 | /* This used to use csum_partial, which was wrong for several | |
957 | * reasons including that different results are returned on | |
958 | * different architectures. It isn't critical that we get exactly | |
959 | * the same return value as before (we always csum_fold before | |
960 | * testing, and that removes any differences). However as we | |
961 | * know that csum_partial always returned a 16bit value on | |
962 | * alphas, do a fold to maximise conformity to previous behaviour. | |
963 | */ | |
964 | sb->sb_csum = md_csum_fold(disk_csum); | |
965 | #else | |
1da177e4 | 966 | sb->sb_csum = disk_csum; |
4d167f09 | 967 | #endif |
1da177e4 LT |
968 | return csum; |
969 | } | |
970 | ||
1da177e4 LT |
971 | /* |
972 | * Handle superblock details. | |
973 | * We want to be able to handle multiple superblock formats | |
974 | * so we have a common interface to them all, and an array of | |
975 | * different handlers. | |
976 | * We rely on user-space to write the initial superblock, and support | |
977 | * reading and updating of superblocks. | |
978 | * Interface methods are: | |
3cb03002 | 979 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
980 | * loads and validates a superblock on dev. |
981 | * if refdev != NULL, compare superblocks on both devices | |
982 | * Return: | |
983 | * 0 - dev has a superblock that is compatible with refdev | |
984 | * 1 - dev has a superblock that is compatible and newer than refdev | |
985 | * so dev should be used as the refdev in future | |
986 | * -EINVAL superblock incompatible or invalid | |
987 | * -othererror e.g. -EIO | |
988 | * | |
fd01b88c | 989 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
990 | * Verify that dev is acceptable into mddev. |
991 | * The first time, mddev->raid_disks will be 0, and data from | |
992 | * dev should be merged in. Subsequent calls check that dev | |
993 | * is new enough. Return 0 or -EINVAL | |
994 | * | |
fd01b88c | 995 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
996 | * Update the superblock for rdev with data in mddev |
997 | * This does not write to disc. | |
998 | * | |
999 | */ | |
1000 | ||
1001 | struct super_type { | |
0cd17fec CW |
1002 | char *name; |
1003 | struct module *owner; | |
c6563a8c N |
1004 | int (*load_super)(struct md_rdev *rdev, |
1005 | struct md_rdev *refdev, | |
0cd17fec | 1006 | int minor_version); |
c6563a8c N |
1007 | int (*validate_super)(struct mddev *mddev, |
1008 | struct md_rdev *rdev); | |
1009 | void (*sync_super)(struct mddev *mddev, | |
1010 | struct md_rdev *rdev); | |
3cb03002 | 1011 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 1012 | sector_t num_sectors); |
c6563a8c N |
1013 | int (*allow_new_offset)(struct md_rdev *rdev, |
1014 | unsigned long long new_offset); | |
1da177e4 LT |
1015 | }; |
1016 | ||
0894cc30 AN |
1017 | /* |
1018 | * Check that the given mddev has no bitmap. | |
1019 | * | |
1020 | * This function is called from the run method of all personalities that do not | |
1021 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
1022 | * has a bitmap. Otherwise, it returns 0. | |
1023 | * | |
1024 | */ | |
fd01b88c | 1025 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 1026 | { |
c3d9714e | 1027 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 | 1028 | return 0; |
9d48739e | 1029 | pr_warn("%s: bitmaps are not supported for %s\n", |
0894cc30 AN |
1030 | mdname(mddev), mddev->pers->name); |
1031 | return 1; | |
1032 | } | |
1033 | EXPORT_SYMBOL(md_check_no_bitmap); | |
1034 | ||
1da177e4 | 1035 | /* |
f72ffdd6 | 1036 | * load_super for 0.90.0 |
1da177e4 | 1037 | */ |
3cb03002 | 1038 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1039 | { |
1040 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
1041 | mdp_super_t *sb; | |
1042 | int ret; | |
1da177e4 LT |
1043 | |
1044 | /* | |
0f420358 | 1045 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
1046 | * it's at the end of the disk. |
1047 | * | |
1048 | * It also happens to be a multiple of 4Kb. | |
1049 | */ | |
57b2caa3 | 1050 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 1051 | |
0002b271 | 1052 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
9d48739e N |
1053 | if (ret) |
1054 | return ret; | |
1da177e4 LT |
1055 | |
1056 | ret = -EINVAL; | |
1057 | ||
1058 | bdevname(rdev->bdev, b); | |
65a06f06 | 1059 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1060 | |
1061 | if (sb->md_magic != MD_SB_MAGIC) { | |
9d48739e | 1062 | pr_warn("md: invalid raid superblock magic on %s\n", b); |
1da177e4 LT |
1063 | goto abort; |
1064 | } | |
1065 | ||
1066 | if (sb->major_version != 0 || | |
f6705578 N |
1067 | sb->minor_version < 90 || |
1068 | sb->minor_version > 91) { | |
9d48739e N |
1069 | pr_warn("Bad version number %d.%d on %s\n", |
1070 | sb->major_version, sb->minor_version, b); | |
1da177e4 LT |
1071 | goto abort; |
1072 | } | |
1073 | ||
1074 | if (sb->raid_disks <= 0) | |
1075 | goto abort; | |
1076 | ||
4d167f09 | 1077 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
9d48739e | 1078 | pr_warn("md: invalid superblock checksum on %s\n", b); |
1da177e4 LT |
1079 | goto abort; |
1080 | } | |
1081 | ||
1082 | rdev->preferred_minor = sb->md_minor; | |
1083 | rdev->data_offset = 0; | |
c6563a8c | 1084 | rdev->new_data_offset = 0; |
0002b271 | 1085 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1086 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1087 | |
1088 | if (sb->level == LEVEL_MULTIPATH) | |
1089 | rdev->desc_nr = -1; | |
1090 | else | |
1091 | rdev->desc_nr = sb->this_disk.number; | |
1092 | ||
9a7b2b0f | 1093 | if (!refdev) { |
1da177e4 | 1094 | ret = 1; |
9a7b2b0f | 1095 | } else { |
1da177e4 | 1096 | __u64 ev1, ev2; |
65a06f06 | 1097 | mdp_super_t *refsb = page_address(refdev->sb_page); |
e6fd2093 | 1098 | if (!md_uuid_equal(refsb, sb)) { |
9d48739e | 1099 | pr_warn("md: %s has different UUID to %s\n", |
1da177e4 LT |
1100 | b, bdevname(refdev->bdev,b2)); |
1101 | goto abort; | |
1102 | } | |
e6fd2093 | 1103 | if (!md_sb_equal(refsb, sb)) { |
9d48739e N |
1104 | pr_warn("md: %s has same UUID but different superblock to %s\n", |
1105 | b, bdevname(refdev->bdev, b2)); | |
1da177e4 LT |
1106 | goto abort; |
1107 | } | |
1108 | ev1 = md_event(sb); | |
1109 | ev2 = md_event(refsb); | |
1110 | if (ev1 > ev2) | |
1111 | ret = 1; | |
f72ffdd6 | 1112 | else |
1da177e4 LT |
1113 | ret = 0; |
1114 | } | |
8190e754 | 1115 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1116 | /* Limit to 4TB as metadata cannot record more than that. |
1117 | * (not needed for Linear and RAID0 as metadata doesn't | |
1118 | * record this size) | |
1119 | */ | |
3312c951 AB |
1120 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) && |
1121 | sb->level >= 1) | |
1122 | rdev->sectors = (sector_t)(2ULL << 32) - 2; | |
1da177e4 | 1123 | |
27a7b260 | 1124 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1125 | /* "this cannot possibly happen" ... */ |
1126 | ret = -EINVAL; | |
1127 | ||
1da177e4 LT |
1128 | abort: |
1129 | return ret; | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * validate_super for 0.90.0 | |
1134 | */ | |
fd01b88c | 1135 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1136 | { |
1137 | mdp_disk_t *desc; | |
65a06f06 | 1138 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1139 | __u64 ev1 = md_event(sb); |
1da177e4 | 1140 | |
41158c7e | 1141 | rdev->raid_disk = -1; |
c5d79adb N |
1142 | clear_bit(Faulty, &rdev->flags); |
1143 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1144 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1145 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1146 | |
1da177e4 LT |
1147 | if (mddev->raid_disks == 0) { |
1148 | mddev->major_version = 0; | |
1149 | mddev->minor_version = sb->minor_version; | |
1150 | mddev->patch_version = sb->patch_version; | |
e691063a | 1151 | mddev->external = 0; |
9d8f0363 | 1152 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1153 | mddev->ctime = sb->ctime; |
1154 | mddev->utime = sb->utime; | |
1155 | mddev->level = sb->level; | |
d9d166c2 | 1156 | mddev->clevel[0] = 0; |
1da177e4 LT |
1157 | mddev->layout = sb->layout; |
1158 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1159 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1160 | mddev->events = ev1; |
c3d9714e | 1161 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1162 | mddev->bitmap_info.space = 0; |
1163 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1164 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1165 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1166 | mddev->reshape_backwards = 0; |
1da177e4 | 1167 | |
f6705578 N |
1168 | if (mddev->minor_version >= 91) { |
1169 | mddev->reshape_position = sb->reshape_position; | |
1170 | mddev->delta_disks = sb->delta_disks; | |
1171 | mddev->new_level = sb->new_level; | |
1172 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1173 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1174 | if (mddev->delta_disks < 0) |
1175 | mddev->reshape_backwards = 1; | |
f6705578 N |
1176 | } else { |
1177 | mddev->reshape_position = MaxSector; | |
1178 | mddev->delta_disks = 0; | |
1179 | mddev->new_level = mddev->level; | |
1180 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1181 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1182 | } |
1183 | ||
1da177e4 LT |
1184 | if (sb->state & (1<<MD_SB_CLEAN)) |
1185 | mddev->recovery_cp = MaxSector; | |
1186 | else { | |
f72ffdd6 | 1187 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1188 | sb->events_lo == sb->cp_events_lo) { |
1189 | mddev->recovery_cp = sb->recovery_cp; | |
1190 | } else | |
1191 | mddev->recovery_cp = 0; | |
1192 | } | |
1193 | ||
1194 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1195 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1196 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1197 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1198 | ||
1199 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1200 | |
1201 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1202 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1203 | mddev->bitmap_info.offset = |
1204 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1205 | mddev->bitmap_info.space = |
c9ad020f | 1206 | mddev->bitmap_info.default_space; |
6409bb05 | 1207 | } |
a654b9d8 | 1208 | |
41158c7e | 1209 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1210 | /* Insist on good event counter while assembling, except |
1211 | * for spares (which don't need an event count) */ | |
1da177e4 | 1212 | ++ev1; |
be6800a7 N |
1213 | if (sb->disks[rdev->desc_nr].state & ( |
1214 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1215 | if (ev1 < mddev->events) |
be6800a7 | 1216 | return -EINVAL; |
41158c7e N |
1217 | } else if (mddev->bitmap) { |
1218 | /* if adding to array with a bitmap, then we can accept an | |
1219 | * older device ... but not too old. | |
1220 | */ | |
41158c7e N |
1221 | if (ev1 < mddev->bitmap->events_cleared) |
1222 | return 0; | |
8313b8e5 N |
1223 | if (ev1 < mddev->events) |
1224 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1225 | } else { |
1226 | if (ev1 < mddev->events) | |
1227 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1228 | return 0; | |
1229 | } | |
41158c7e | 1230 | |
1da177e4 | 1231 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1232 | desc = sb->disks + rdev->desc_nr; |
1233 | ||
1234 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1235 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1236 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1237 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1238 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1239 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1240 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1241 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1242 | /* active but not in sync implies recovery up to | |
1243 | * reshape position. We don't know exactly where | |
1244 | * that is, so set to zero for now */ | |
1245 | if (mddev->minor_version >= 91) { | |
1246 | rdev->recovery_offset = 0; | |
1247 | rdev->raid_disk = desc->raid_disk; | |
1248 | } | |
1da177e4 | 1249 | } |
8ddf9efe N |
1250 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1251 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1252 | if (desc->state & (1<<MD_DISK_FAILFAST)) |
1253 | set_bit(FailFast, &rdev->flags); | |
41158c7e | 1254 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1255 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1256 | return 0; |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * sync_super for 0.90.0 | |
1261 | */ | |
fd01b88c | 1262 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1263 | { |
1264 | mdp_super_t *sb; | |
3cb03002 | 1265 | struct md_rdev *rdev2; |
1da177e4 | 1266 | int next_spare = mddev->raid_disks; |
19133a42 | 1267 | |
1da177e4 LT |
1268 | /* make rdev->sb match mddev data.. |
1269 | * | |
1270 | * 1/ zero out disks | |
1271 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1272 | * 3/ any empty disks < next_spare become removed | |
1273 | * | |
1274 | * disks[0] gets initialised to REMOVED because | |
1275 | * we cannot be sure from other fields if it has | |
1276 | * been initialised or not. | |
1277 | */ | |
1278 | int i; | |
1279 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1280 | ||
61181565 N |
1281 | rdev->sb_size = MD_SB_BYTES; |
1282 | ||
65a06f06 | 1283 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1284 | |
1285 | memset(sb, 0, sizeof(*sb)); | |
1286 | ||
1287 | sb->md_magic = MD_SB_MAGIC; | |
1288 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1289 | sb->patch_version = mddev->patch_version; |
1290 | sb->gvalid_words = 0; /* ignored */ | |
1291 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1292 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1293 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1294 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1295 | ||
9ebc6ef1 | 1296 | sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 1297 | sb->level = mddev->level; |
58c0fed4 | 1298 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1299 | sb->raid_disks = mddev->raid_disks; |
1300 | sb->md_minor = mddev->md_minor; | |
e691063a | 1301 | sb->not_persistent = 0; |
9ebc6ef1 | 1302 | sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
1303 | sb->state = 0; |
1304 | sb->events_hi = (mddev->events>>32); | |
1305 | sb->events_lo = (u32)mddev->events; | |
1306 | ||
f6705578 N |
1307 | if (mddev->reshape_position == MaxSector) |
1308 | sb->minor_version = 90; | |
1309 | else { | |
1310 | sb->minor_version = 91; | |
1311 | sb->reshape_position = mddev->reshape_position; | |
1312 | sb->new_level = mddev->new_level; | |
1313 | sb->delta_disks = mddev->delta_disks; | |
1314 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1315 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1316 | } |
1317 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1318 | if (mddev->in_sync) |
1319 | { | |
1320 | sb->recovery_cp = mddev->recovery_cp; | |
1321 | sb->cp_events_hi = (mddev->events>>32); | |
1322 | sb->cp_events_lo = (u32)mddev->events; | |
1323 | if (mddev->recovery_cp == MaxSector) | |
1324 | sb->state = (1<< MD_SB_CLEAN); | |
1325 | } else | |
1326 | sb->recovery_cp = 0; | |
1327 | ||
1328 | sb->layout = mddev->layout; | |
9d8f0363 | 1329 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1330 | |
c3d9714e | 1331 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1332 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1333 | ||
1da177e4 | 1334 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1335 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1336 | mdp_disk_t *d; |
86e6ffdd | 1337 | int desc_nr; |
0261cd9f N |
1338 | int is_active = test_bit(In_sync, &rdev2->flags); |
1339 | ||
1340 | if (rdev2->raid_disk >= 0 && | |
1341 | sb->minor_version >= 91) | |
1342 | /* we have nowhere to store the recovery_offset, | |
1343 | * but if it is not below the reshape_position, | |
1344 | * we can piggy-back on that. | |
1345 | */ | |
1346 | is_active = 1; | |
1347 | if (rdev2->raid_disk < 0 || | |
1348 | test_bit(Faulty, &rdev2->flags)) | |
1349 | is_active = 0; | |
1350 | if (is_active) | |
86e6ffdd | 1351 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1352 | else |
86e6ffdd | 1353 | desc_nr = next_spare++; |
19133a42 | 1354 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1355 | d = &sb->disks[rdev2->desc_nr]; |
1356 | nr_disks++; | |
1357 | d->number = rdev2->desc_nr; | |
1358 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1359 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1360 | if (is_active) |
1da177e4 LT |
1361 | d->raid_disk = rdev2->raid_disk; |
1362 | else | |
1363 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1364 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1365 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1366 | else if (is_active) { |
1da177e4 | 1367 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1368 | if (test_bit(In_sync, &rdev2->flags)) |
1369 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1370 | active++; |
1371 | working++; | |
1372 | } else { | |
1373 | d->state = 0; | |
1374 | spare++; | |
1375 | working++; | |
1376 | } | |
8ddf9efe N |
1377 | if (test_bit(WriteMostly, &rdev2->flags)) |
1378 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
1379 | if (test_bit(FailFast, &rdev2->flags)) |
1380 | d->state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 | 1381 | } |
1da177e4 LT |
1382 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1383 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1384 | mdp_disk_t *d = &sb->disks[i]; | |
1385 | if (d->state == 0 && d->number == 0) { | |
1386 | d->number = i; | |
1387 | d->raid_disk = i; | |
1388 | d->state = (1<<MD_DISK_REMOVED); | |
1389 | d->state |= (1<<MD_DISK_FAULTY); | |
1390 | failed++; | |
1391 | } | |
1392 | } | |
1393 | sb->nr_disks = nr_disks; | |
1394 | sb->active_disks = active; | |
1395 | sb->working_disks = working; | |
1396 | sb->failed_disks = failed; | |
1397 | sb->spare_disks = spare; | |
1398 | ||
1399 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1400 | sb->sb_csum = calc_sb_csum(sb); | |
1401 | } | |
1402 | ||
0cd17fec CW |
1403 | /* |
1404 | * rdev_size_change for 0.90.0 | |
1405 | */ | |
1406 | static unsigned long long | |
3cb03002 | 1407 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1408 | { |
58c0fed4 | 1409 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1410 | return 0; /* component must fit device */ |
c3d9714e | 1411 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1412 | return 0; /* can't move bitmap */ |
57b2caa3 | 1413 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1414 | if (!num_sectors || num_sectors > rdev->sb_start) |
1415 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1416 | /* Limit to 4TB as metadata cannot record more than that. |
1417 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1418 | */ | |
3312c951 AB |
1419 | if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) && |
1420 | rdev->mddev->level >= 1) | |
1421 | num_sectors = (sector_t)(2ULL << 32) - 2; | |
46533ff7 N |
1422 | do { |
1423 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
0cd17fec | 1424 | rdev->sb_page); |
46533ff7 | 1425 | } while (md_super_wait(rdev->mddev) < 0); |
c26a44ed | 1426 | return num_sectors; |
0cd17fec CW |
1427 | } |
1428 | ||
c6563a8c N |
1429 | static int |
1430 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1431 | { | |
1432 | /* non-zero offset changes not possible with v0.90 */ | |
1433 | return new_offset == 0; | |
1434 | } | |
0cd17fec | 1435 | |
1da177e4 LT |
1436 | /* |
1437 | * version 1 superblock | |
1438 | */ | |
1439 | ||
f72ffdd6 | 1440 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1441 | { |
1c05b4bc N |
1442 | __le32 disk_csum; |
1443 | u32 csum; | |
1da177e4 LT |
1444 | unsigned long long newcsum; |
1445 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1446 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1447 | |
1448 | disk_csum = sb->sb_csum; | |
1449 | sb->sb_csum = 0; | |
1450 | newcsum = 0; | |
1f3c9907 | 1451 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1452 | newcsum += le32_to_cpu(*isuper++); |
1453 | ||
1454 | if (size == 2) | |
1c05b4bc | 1455 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1456 | |
1457 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1458 | sb->sb_csum = disk_csum; | |
1459 | return cpu_to_le32(csum); | |
1460 | } | |
1461 | ||
3cb03002 | 1462 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1463 | { |
1464 | struct mdp_superblock_1 *sb; | |
1465 | int ret; | |
0f420358 | 1466 | sector_t sb_start; |
c6563a8c | 1467 | sector_t sectors; |
1da177e4 | 1468 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1469 | int bmask; |
1da177e4 LT |
1470 | |
1471 | /* | |
0f420358 | 1472 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1473 | * It is always aligned to a 4K boundary and |
1474 | * depeding on minor_version, it can be: | |
1475 | * 0: At least 8K, but less than 12K, from end of device | |
1476 | * 1: At start of device | |
1477 | * 2: 4K from start of device. | |
1478 | */ | |
1479 | switch(minor_version) { | |
1480 | case 0: | |
77304d2a | 1481 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1482 | sb_start -= 8*2; |
1483 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1484 | break; |
1485 | case 1: | |
0f420358 | 1486 | sb_start = 0; |
1da177e4 LT |
1487 | break; |
1488 | case 2: | |
0f420358 | 1489 | sb_start = 8; |
1da177e4 LT |
1490 | break; |
1491 | default: | |
1492 | return -EINVAL; | |
1493 | } | |
0f420358 | 1494 | rdev->sb_start = sb_start; |
1da177e4 | 1495 | |
0002b271 N |
1496 | /* superblock is rarely larger than 1K, but it can be larger, |
1497 | * and it is safe to read 4k, so we do that | |
1498 | */ | |
1499 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1500 | if (ret) return ret; |
1501 | ||
65a06f06 | 1502 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1503 | |
1504 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1505 | sb->major_version != cpu_to_le32(1) || | |
1506 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1507 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1508 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1509 | return -EINVAL; |
1510 | ||
1511 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
9d48739e | 1512 | pr_warn("md: invalid superblock checksum on %s\n", |
1da177e4 LT |
1513 | bdevname(rdev->bdev,b)); |
1514 | return -EINVAL; | |
1515 | } | |
1516 | if (le64_to_cpu(sb->data_size) < 10) { | |
9d48739e N |
1517 | pr_warn("md: data_size too small on %s\n", |
1518 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1519 | return -EINVAL; |
1520 | } | |
c6563a8c N |
1521 | if (sb->pad0 || |
1522 | sb->pad3[0] || | |
1523 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1524 | /* Some padding is non-zero, might be a new feature */ | |
1525 | return -EINVAL; | |
e11e93fa | 1526 | |
1da177e4 LT |
1527 | rdev->preferred_minor = 0xffff; |
1528 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1529 | rdev->new_data_offset = rdev->data_offset; |
1530 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1531 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1532 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1533 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1534 | |
0002b271 | 1535 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1536 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1537 | if (rdev->sb_size & bmask) |
a1801f85 N |
1538 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1539 | ||
1540 | if (minor_version | |
0f420358 | 1541 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1542 | return -EINVAL; |
c6563a8c N |
1543 | if (minor_version |
1544 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1545 | return -EINVAL; | |
0002b271 | 1546 | |
31b65a0d N |
1547 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1548 | rdev->desc_nr = -1; | |
1549 | else | |
1550 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1551 | ||
2699b672 N |
1552 | if (!rdev->bb_page) { |
1553 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1554 | if (!rdev->bb_page) | |
1555 | return -ENOMEM; | |
1556 | } | |
1557 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1558 | rdev->badblocks.count == 0) { | |
1559 | /* need to load the bad block list. | |
1560 | * Currently we limit it to one page. | |
1561 | */ | |
1562 | s32 offset; | |
1563 | sector_t bb_sector; | |
1564 | u64 *bbp; | |
1565 | int i; | |
1566 | int sectors = le16_to_cpu(sb->bblog_size); | |
1567 | if (sectors > (PAGE_SIZE / 512)) | |
1568 | return -EINVAL; | |
1569 | offset = le32_to_cpu(sb->bblog_offset); | |
1570 | if (offset == 0) | |
1571 | return -EINVAL; | |
1572 | bb_sector = (long long)offset; | |
1573 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
796a5cf0 | 1574 | rdev->bb_page, REQ_OP_READ, 0, true)) |
2699b672 N |
1575 | return -EIO; |
1576 | bbp = (u64 *)page_address(rdev->bb_page); | |
1577 | rdev->badblocks.shift = sb->bblog_shift; | |
1578 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1579 | u64 bb = le64_to_cpu(*bbp); | |
1580 | int count = bb & (0x3ff); | |
1581 | u64 sector = bb >> 10; | |
1582 | sector <<= sb->bblog_shift; | |
1583 | count <<= sb->bblog_shift; | |
1584 | if (bb + 1 == 0) | |
1585 | break; | |
fc974ee2 | 1586 | if (badblocks_set(&rdev->badblocks, sector, count, 1)) |
2699b672 N |
1587 | return -EINVAL; |
1588 | } | |
486adf72 N |
1589 | } else if (sb->bblog_offset != 0) |
1590 | rdev->badblocks.shift = 0; | |
2699b672 | 1591 | |
ddc08823 PB |
1592 | if ((le32_to_cpu(sb->feature_map) & |
1593 | (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) { | |
ea0213e0 AP |
1594 | rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset); |
1595 | rdev->ppl.size = le16_to_cpu(sb->ppl.size); | |
1596 | rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset; | |
1597 | } | |
1598 | ||
9a7b2b0f | 1599 | if (!refdev) { |
8ed75463 | 1600 | ret = 1; |
9a7b2b0f | 1601 | } else { |
1da177e4 | 1602 | __u64 ev1, ev2; |
65a06f06 | 1603 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1604 | |
1605 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1606 | sb->level != refsb->level || | |
1607 | sb->layout != refsb->layout || | |
1608 | sb->chunksize != refsb->chunksize) { | |
9d48739e | 1609 | pr_warn("md: %s has strangely different superblock to %s\n", |
1da177e4 LT |
1610 | bdevname(rdev->bdev,b), |
1611 | bdevname(refdev->bdev,b2)); | |
1612 | return -EINVAL; | |
1613 | } | |
1614 | ev1 = le64_to_cpu(sb->events); | |
1615 | ev2 = le64_to_cpu(refsb->events); | |
1616 | ||
1617 | if (ev1 > ev2) | |
8ed75463 N |
1618 | ret = 1; |
1619 | else | |
1620 | ret = 0; | |
1da177e4 | 1621 | } |
c6563a8c N |
1622 | if (minor_version) { |
1623 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1624 | sectors -= rdev->data_offset; | |
1625 | } else | |
1626 | sectors = rdev->sb_start; | |
1627 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1628 | return -EINVAL; |
dd8ac336 | 1629 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1630 | return ret; |
1da177e4 LT |
1631 | } |
1632 | ||
fd01b88c | 1633 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1634 | { |
65a06f06 | 1635 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1636 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1637 | |
41158c7e | 1638 | rdev->raid_disk = -1; |
c5d79adb N |
1639 | clear_bit(Faulty, &rdev->flags); |
1640 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1641 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1642 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1643 | |
1da177e4 LT |
1644 | if (mddev->raid_disks == 0) { |
1645 | mddev->major_version = 1; | |
1646 | mddev->patch_version = 0; | |
e691063a | 1647 | mddev->external = 0; |
9d8f0363 | 1648 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
9ebc6ef1 DD |
1649 | mddev->ctime = le64_to_cpu(sb->ctime); |
1650 | mddev->utime = le64_to_cpu(sb->utime); | |
1da177e4 | 1651 | mddev->level = le32_to_cpu(sb->level); |
d9d166c2 | 1652 | mddev->clevel[0] = 0; |
1da177e4 LT |
1653 | mddev->layout = le32_to_cpu(sb->layout); |
1654 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1655 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1656 | mddev->events = ev1; |
c3d9714e | 1657 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1658 | mddev->bitmap_info.space = 0; |
1659 | /* Default location for bitmap is 1K after superblock | |
1660 | * using 3K - total of 4K | |
1661 | */ | |
c3d9714e | 1662 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1663 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1664 | mddev->reshape_backwards = 0; |
1665 | ||
1da177e4 LT |
1666 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1667 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1668 | ||
1669 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1670 | |
71c0805c | 1671 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1672 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1673 | mddev->bitmap_info.offset = |
1674 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1675 | /* Metadata doesn't record how much space is available. |
1676 | * For 1.0, we assume we can use up to the superblock | |
1677 | * if before, else to 4K beyond superblock. | |
1678 | * For others, assume no change is possible. | |
1679 | */ | |
1680 | if (mddev->minor_version > 0) | |
1681 | mddev->bitmap_info.space = 0; | |
1682 | else if (mddev->bitmap_info.offset > 0) | |
1683 | mddev->bitmap_info.space = | |
1684 | 8 - mddev->bitmap_info.offset; | |
1685 | else | |
1686 | mddev->bitmap_info.space = | |
1687 | -mddev->bitmap_info.offset; | |
1688 | } | |
e11e93fa | 1689 | |
f6705578 N |
1690 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1691 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1692 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1693 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1694 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1695 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1696 | if (mddev->delta_disks < 0 || |
1697 | (mddev->delta_disks == 0 && | |
1698 | (le32_to_cpu(sb->feature_map) | |
1699 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1700 | mddev->reshape_backwards = 1; | |
f6705578 N |
1701 | } else { |
1702 | mddev->reshape_position = MaxSector; | |
1703 | mddev->delta_disks = 0; | |
1704 | mddev->new_level = mddev->level; | |
1705 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1706 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1707 | } |
1708 | ||
486b0f7b | 1709 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) |
a62ab49e | 1710 | set_bit(MD_HAS_JOURNAL, &mddev->flags); |
ea0213e0 | 1711 | |
ddc08823 PB |
1712 | if (le32_to_cpu(sb->feature_map) & |
1713 | (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) { | |
ea0213e0 AP |
1714 | if (le32_to_cpu(sb->feature_map) & |
1715 | (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL)) | |
1716 | return -EINVAL; | |
ddc08823 PB |
1717 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) && |
1718 | (le32_to_cpu(sb->feature_map) & | |
1719 | MD_FEATURE_MULTIPLE_PPLS)) | |
1720 | return -EINVAL; | |
ea0213e0 AP |
1721 | set_bit(MD_HAS_PPL, &mddev->flags); |
1722 | } | |
41158c7e | 1723 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1724 | /* Insist of good event counter while assembling, except for |
1725 | * spares (which don't need an event count) */ | |
1da177e4 | 1726 | ++ev1; |
be6800a7 N |
1727 | if (rdev->desc_nr >= 0 && |
1728 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
a3dfbdaa SL |
1729 | (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX || |
1730 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)) | |
be6800a7 N |
1731 | if (ev1 < mddev->events) |
1732 | return -EINVAL; | |
41158c7e N |
1733 | } else if (mddev->bitmap) { |
1734 | /* If adding to array with a bitmap, then we can accept an | |
1735 | * older device, but not too old. | |
1736 | */ | |
41158c7e N |
1737 | if (ev1 < mddev->bitmap->events_cleared) |
1738 | return 0; | |
8313b8e5 N |
1739 | if (ev1 < mddev->events) |
1740 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1741 | } else { |
1742 | if (ev1 < mddev->events) | |
1743 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1744 | return 0; | |
1745 | } | |
1da177e4 LT |
1746 | if (mddev->level != LEVEL_MULTIPATH) { |
1747 | int role; | |
3673f305 N |
1748 | if (rdev->desc_nr < 0 || |
1749 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
c4d4c91b | 1750 | role = MD_DISK_ROLE_SPARE; |
3673f305 N |
1751 | rdev->desc_nr = -1; |
1752 | } else | |
1753 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 | 1754 | switch(role) { |
c4d4c91b | 1755 | case MD_DISK_ROLE_SPARE: /* spare */ |
1da177e4 | 1756 | break; |
c4d4c91b | 1757 | case MD_DISK_ROLE_FAULTY: /* faulty */ |
b2d444d7 | 1758 | set_bit(Faulty, &rdev->flags); |
1da177e4 | 1759 | break; |
bac624f3 SL |
1760 | case MD_DISK_ROLE_JOURNAL: /* journal device */ |
1761 | if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) { | |
1762 | /* journal device without journal feature */ | |
9d48739e | 1763 | pr_warn("md: journal device provided without journal feature, ignoring the device\n"); |
bac624f3 SL |
1764 | return -EINVAL; |
1765 | } | |
1766 | set_bit(Journal, &rdev->flags); | |
3069aa8d | 1767 | rdev->journal_tail = le64_to_cpu(sb->journal_tail); |
9b15603d | 1768 | rdev->raid_disk = 0; |
bac624f3 | 1769 | break; |
1da177e4 | 1770 | default: |
f466722c | 1771 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1772 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1773 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1774 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1775 | if (!(le32_to_cpu(sb->feature_map) & |
1776 | MD_FEATURE_RECOVERY_BITMAP)) | |
1777 | rdev->saved_raid_disk = -1; | |
1778 | } else | |
5fd6c1dc | 1779 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1780 | rdev->raid_disk = role; |
1781 | break; | |
1782 | } | |
8ddf9efe N |
1783 | if (sb->devflags & WriteMostly1) |
1784 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
1785 | if (sb->devflags & FailFast1) |
1786 | set_bit(FailFast, &rdev->flags); | |
2d78f8c4 N |
1787 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1788 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1789 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1790 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1791 | |
1da177e4 LT |
1792 | return 0; |
1793 | } | |
1794 | ||
fd01b88c | 1795 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1796 | { |
1797 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1798 | struct md_rdev *rdev2; |
1da177e4 LT |
1799 | int max_dev, i; |
1800 | /* make rdev->sb match mddev and rdev data. */ | |
1801 | ||
65a06f06 | 1802 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1803 | |
1804 | sb->feature_map = 0; | |
1805 | sb->pad0 = 0; | |
5fd6c1dc | 1806 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1807 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1808 | ||
1809 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1810 | sb->events = cpu_to_le64(mddev->events); | |
1811 | if (mddev->in_sync) | |
1812 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
bd18f646 SL |
1813 | else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags)) |
1814 | sb->resync_offset = cpu_to_le64(MaxSector); | |
1da177e4 LT |
1815 | else |
1816 | sb->resync_offset = cpu_to_le64(0); | |
1817 | ||
1c05b4bc | 1818 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1819 | |
f0ca340c | 1820 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1821 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1822 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1823 | sb->level = cpu_to_le32(mddev->level); |
1824 | sb->layout = cpu_to_le32(mddev->layout); | |
688834e6 N |
1825 | if (test_bit(FailFast, &rdev->flags)) |
1826 | sb->devflags |= FailFast1; | |
1827 | else | |
1828 | sb->devflags &= ~FailFast1; | |
f0ca340c | 1829 | |
aeb9b211 N |
1830 | if (test_bit(WriteMostly, &rdev->flags)) |
1831 | sb->devflags |= WriteMostly1; | |
1832 | else | |
1833 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1834 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1835 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1836 | |
c3d9714e N |
1837 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1838 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1839 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1840 | } |
5fd6c1dc | 1841 | |
f2076e7d | 1842 | if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) && |
97e4f42d | 1843 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1844 | sb->feature_map |= |
1845 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1846 | sb->recovery_offset = | |
1847 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1848 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1849 | sb->feature_map |= | |
1850 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1851 | } |
3069aa8d SL |
1852 | /* Note: recovery_offset and journal_tail share space */ |
1853 | if (test_bit(Journal, &rdev->flags)) | |
1854 | sb->journal_tail = cpu_to_le64(rdev->journal_tail); | |
2d78f8c4 N |
1855 | if (test_bit(Replacement, &rdev->flags)) |
1856 | sb->feature_map |= | |
1857 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1858 | |
f6705578 N |
1859 | if (mddev->reshape_position != MaxSector) { |
1860 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1861 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1862 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1863 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1864 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1865 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1866 | if (mddev->delta_disks == 0 && |
1867 | mddev->reshape_backwards) | |
1868 | sb->feature_map | |
1869 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1870 | if (rdev->new_data_offset != rdev->data_offset) { |
1871 | sb->feature_map | |
1872 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1873 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1874 | - rdev->data_offset)); | |
1875 | } | |
f6705578 | 1876 | } |
a654b9d8 | 1877 | |
3c462c88 GR |
1878 | if (mddev_is_clustered(mddev)) |
1879 | sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED); | |
1880 | ||
2699b672 N |
1881 | if (rdev->badblocks.count == 0) |
1882 | /* Nothing to do for bad blocks*/ ; | |
1883 | else if (sb->bblog_offset == 0) | |
1884 | /* Cannot record bad blocks on this device */ | |
1885 | md_error(mddev, rdev); | |
1886 | else { | |
1887 | struct badblocks *bb = &rdev->badblocks; | |
1888 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1889 | u64 *p = bb->page; | |
1890 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1891 | if (bb->changed) { | |
1892 | unsigned seq; | |
1893 | ||
1894 | retry: | |
1895 | seq = read_seqbegin(&bb->lock); | |
1896 | ||
1897 | memset(bbp, 0xff, PAGE_SIZE); | |
1898 | ||
1899 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1900 | u64 internal_bb = p[i]; |
2699b672 N |
1901 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1902 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1903 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1904 | } |
d0962936 | 1905 | bb->changed = 0; |
2699b672 N |
1906 | if (read_seqretry(&bb->lock, seq)) |
1907 | goto retry; | |
1908 | ||
1909 | bb->sector = (rdev->sb_start + | |
1910 | (int)le32_to_cpu(sb->bblog_offset)); | |
1911 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1912 | } |
1913 | } | |
1914 | ||
1da177e4 | 1915 | max_dev = 0; |
dafb20fa | 1916 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1917 | if (rdev2->desc_nr+1 > max_dev) |
1918 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1919 | |
70471daf N |
1920 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1921 | int bmask; | |
a778b73f | 1922 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1923 | rdev->sb_size = max_dev * 2 + 256; |
1924 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1925 | if (rdev->sb_size & bmask) | |
1926 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1927 | } else |
1928 | max_dev = le32_to_cpu(sb->max_dev); | |
1929 | ||
1da177e4 | 1930 | for (i=0; i<max_dev;i++) |
8df72024 | 1931 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE); |
f72ffdd6 | 1932 | |
a97b7896 SL |
1933 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) |
1934 | sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL); | |
f72ffdd6 | 1935 | |
ea0213e0 | 1936 | if (test_bit(MD_HAS_PPL, &mddev->flags)) { |
ddc08823 PB |
1937 | if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags)) |
1938 | sb->feature_map |= | |
1939 | cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS); | |
1940 | else | |
1941 | sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL); | |
ea0213e0 AP |
1942 | sb->ppl.offset = cpu_to_le16(rdev->ppl.offset); |
1943 | sb->ppl.size = cpu_to_le16(rdev->ppl.size); | |
1944 | } | |
1945 | ||
dafb20fa | 1946 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1947 | i = rdev2->desc_nr; |
b2d444d7 | 1948 | if (test_bit(Faulty, &rdev2->flags)) |
c4d4c91b | 1949 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY); |
b2d444d7 | 1950 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1951 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
a97b7896 | 1952 | else if (test_bit(Journal, &rdev2->flags)) |
bac624f3 | 1953 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL); |
93be75ff | 1954 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1955 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 | 1956 | else |
c4d4c91b | 1957 | sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE); |
1da177e4 LT |
1958 | } |
1959 | ||
1da177e4 LT |
1960 | sb->sb_csum = calc_sb_1_csum(sb); |
1961 | } | |
1962 | ||
0cd17fec | 1963 | static unsigned long long |
3cb03002 | 1964 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1965 | { |
1966 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1967 | sector_t max_sectors; |
58c0fed4 | 1968 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1969 | return 0; /* component must fit device */ |
c6563a8c N |
1970 | if (rdev->data_offset != rdev->new_data_offset) |
1971 | return 0; /* too confusing */ | |
0f420358 | 1972 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1973 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1974 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1975 | max_sectors -= rdev->data_offset; |
1976 | if (!num_sectors || num_sectors > max_sectors) | |
1977 | num_sectors = max_sectors; | |
c3d9714e | 1978 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1979 | /* minor version 0 with bitmap we can't move */ |
1980 | return 0; | |
1981 | } else { | |
1982 | /* minor version 0; superblock after data */ | |
0f420358 | 1983 | sector_t sb_start; |
77304d2a | 1984 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1985 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1986 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1987 | if (!num_sectors || num_sectors > max_sectors) |
1988 | num_sectors = max_sectors; | |
0f420358 | 1989 | rdev->sb_start = sb_start; |
0cd17fec | 1990 | } |
65a06f06 | 1991 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1992 | sb->data_size = cpu_to_le64(num_sectors); |
3fb632e4 | 1993 | sb->super_offset = cpu_to_le64(rdev->sb_start); |
0cd17fec | 1994 | sb->sb_csum = calc_sb_1_csum(sb); |
46533ff7 N |
1995 | do { |
1996 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, | |
1997 | rdev->sb_page); | |
1998 | } while (md_super_wait(rdev->mddev) < 0); | |
c26a44ed | 1999 | return num_sectors; |
c6563a8c N |
2000 | |
2001 | } | |
2002 | ||
2003 | static int | |
2004 | super_1_allow_new_offset(struct md_rdev *rdev, | |
2005 | unsigned long long new_offset) | |
2006 | { | |
2007 | /* All necessary checks on new >= old have been done */ | |
2008 | struct bitmap *bitmap; | |
2009 | if (new_offset >= rdev->data_offset) | |
2010 | return 1; | |
2011 | ||
2012 | /* with 1.0 metadata, there is no metadata to tread on | |
2013 | * so we can always move back */ | |
2014 | if (rdev->mddev->minor_version == 0) | |
2015 | return 1; | |
2016 | ||
2017 | /* otherwise we must be sure not to step on | |
2018 | * any metadata, so stay: | |
2019 | * 36K beyond start of superblock | |
2020 | * beyond end of badblocks | |
2021 | * beyond write-intent bitmap | |
2022 | */ | |
2023 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
2024 | return 0; | |
2025 | bitmap = rdev->mddev->bitmap; | |
2026 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
2027 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 2028 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
2029 | return 0; |
2030 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
2031 | return 0; | |
2032 | ||
2033 | return 1; | |
0cd17fec | 2034 | } |
1da177e4 | 2035 | |
75c96f85 | 2036 | static struct super_type super_types[] = { |
1da177e4 LT |
2037 | [0] = { |
2038 | .name = "0.90.0", | |
2039 | .owner = THIS_MODULE, | |
0cd17fec CW |
2040 | .load_super = super_90_load, |
2041 | .validate_super = super_90_validate, | |
2042 | .sync_super = super_90_sync, | |
2043 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 2044 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
2045 | }, |
2046 | [1] = { | |
2047 | .name = "md-1", | |
2048 | .owner = THIS_MODULE, | |
0cd17fec CW |
2049 | .load_super = super_1_load, |
2050 | .validate_super = super_1_validate, | |
2051 | .sync_super = super_1_sync, | |
2052 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 2053 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
2054 | }, |
2055 | }; | |
1da177e4 | 2056 | |
fd01b88c | 2057 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
2058 | { |
2059 | if (mddev->sync_super) { | |
2060 | mddev->sync_super(mddev, rdev); | |
2061 | return; | |
2062 | } | |
2063 | ||
2064 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
2065 | ||
2066 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
2067 | } | |
2068 | ||
fd01b88c | 2069 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 2070 | { |
3cb03002 | 2071 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 2072 | |
4b80991c | 2073 | rcu_read_lock(); |
0b020e85 SL |
2074 | rdev_for_each_rcu(rdev, mddev1) { |
2075 | if (test_bit(Faulty, &rdev->flags) || | |
2076 | test_bit(Journal, &rdev->flags) || | |
2077 | rdev->raid_disk == -1) | |
2078 | continue; | |
2079 | rdev_for_each_rcu(rdev2, mddev2) { | |
2080 | if (test_bit(Faulty, &rdev2->flags) || | |
2081 | test_bit(Journal, &rdev2->flags) || | |
2082 | rdev2->raid_disk == -1) | |
2083 | continue; | |
7dd5e7c3 | 2084 | if (rdev->bdev->bd_contains == |
4b80991c N |
2085 | rdev2->bdev->bd_contains) { |
2086 | rcu_read_unlock(); | |
7dd5e7c3 | 2087 | return 1; |
4b80991c | 2088 | } |
0b020e85 SL |
2089 | } |
2090 | } | |
4b80991c | 2091 | rcu_read_unlock(); |
1da177e4 LT |
2092 | return 0; |
2093 | } | |
2094 | ||
2095 | static LIST_HEAD(pending_raid_disks); | |
2096 | ||
ac5e7113 AN |
2097 | /* |
2098 | * Try to register data integrity profile for an mddev | |
2099 | * | |
2100 | * This is called when an array is started and after a disk has been kicked | |
2101 | * from the array. It only succeeds if all working and active component devices | |
2102 | * are integrity capable with matching profiles. | |
2103 | */ | |
fd01b88c | 2104 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 2105 | { |
3cb03002 | 2106 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
2107 | |
2108 | if (list_empty(&mddev->disks)) | |
2109 | return 0; /* nothing to do */ | |
629acb6a JB |
2110 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
2111 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 2112 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
2113 | /* skip spares and non-functional disks */ |
2114 | if (test_bit(Faulty, &rdev->flags)) | |
2115 | continue; | |
2116 | if (rdev->raid_disk < 0) | |
2117 | continue; | |
ac5e7113 AN |
2118 | if (!reference) { |
2119 | /* Use the first rdev as the reference */ | |
2120 | reference = rdev; | |
2121 | continue; | |
2122 | } | |
2123 | /* does this rdev's profile match the reference profile? */ | |
2124 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
2125 | rdev->bdev->bd_disk) < 0) | |
2126 | return -EINVAL; | |
2127 | } | |
89078d57 MP |
2128 | if (!reference || !bdev_get_integrity(reference->bdev)) |
2129 | return 0; | |
ac5e7113 AN |
2130 | /* |
2131 | * All component devices are integrity capable and have matching | |
2132 | * profiles, register the common profile for the md device. | |
2133 | */ | |
25520d55 MP |
2134 | blk_integrity_register(mddev->gendisk, |
2135 | bdev_get_integrity(reference->bdev)); | |
2136 | ||
9d48739e | 2137 | pr_debug("md: data integrity enabled on %s\n", mdname(mddev)); |
afeee514 | 2138 | if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) { |
9d48739e | 2139 | pr_err("md: failed to create integrity pool for %s\n", |
a91a2785 MP |
2140 | mdname(mddev)); |
2141 | return -EINVAL; | |
2142 | } | |
ac5e7113 AN |
2143 | return 0; |
2144 | } | |
2145 | EXPORT_SYMBOL(md_integrity_register); | |
2146 | ||
1501efad DW |
2147 | /* |
2148 | * Attempt to add an rdev, but only if it is consistent with the current | |
2149 | * integrity profile | |
2150 | */ | |
2151 | int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) | |
3f9d99c1 | 2152 | { |
2863b9eb JB |
2153 | struct blk_integrity *bi_rdev; |
2154 | struct blk_integrity *bi_mddev; | |
1501efad | 2155 | char name[BDEVNAME_SIZE]; |
2863b9eb JB |
2156 | |
2157 | if (!mddev->gendisk) | |
1501efad | 2158 | return 0; |
2863b9eb JB |
2159 | |
2160 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2161 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2162 | |
ac5e7113 | 2163 | if (!bi_mddev) /* nothing to do */ |
1501efad DW |
2164 | return 0; |
2165 | ||
2166 | if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) { | |
9d48739e N |
2167 | pr_err("%s: incompatible integrity profile for %s\n", |
2168 | mdname(mddev), bdevname(rdev->bdev, name)); | |
1501efad DW |
2169 | return -ENXIO; |
2170 | } | |
2171 | ||
2172 | return 0; | |
3f9d99c1 | 2173 | } |
ac5e7113 | 2174 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2175 | |
f72ffdd6 | 2176 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2177 | { |
7dd5e7c3 | 2178 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2179 | struct kobject *ko; |
5e55e2f5 | 2180 | int err; |
1da177e4 | 2181 | |
11e2ede0 DW |
2182 | /* prevent duplicates */ |
2183 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2184 | return -EEXIST; | |
2185 | ||
97b20ef7 N |
2186 | if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) && |
2187 | mddev->pers) | |
2188 | return -EROFS; | |
2189 | ||
dd8ac336 | 2190 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
f6b6ec5c SL |
2191 | if (!test_bit(Journal, &rdev->flags) && |
2192 | rdev->sectors && | |
2193 | (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2194 | if (mddev->pers) { |
2195 | /* Cannot change size, so fail | |
2196 | * If mddev->level <= 0, then we don't care | |
2197 | * about aligning sizes (e.g. linear) | |
2198 | */ | |
2199 | if (mddev->level > 0) | |
2200 | return -ENOSPC; | |
2201 | } else | |
dd8ac336 | 2202 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2203 | } |
1da177e4 LT |
2204 | |
2205 | /* Verify rdev->desc_nr is unique. | |
2206 | * If it is -1, assign a free number, else | |
2207 | * check number is not in use | |
2208 | */ | |
4878e9eb | 2209 | rcu_read_lock(); |
1da177e4 LT |
2210 | if (rdev->desc_nr < 0) { |
2211 | int choice = 0; | |
4878e9eb N |
2212 | if (mddev->pers) |
2213 | choice = mddev->raid_disks; | |
57d051dc | 2214 | while (md_find_rdev_nr_rcu(mddev, choice)) |
1da177e4 LT |
2215 | choice++; |
2216 | rdev->desc_nr = choice; | |
2217 | } else { | |
57d051dc | 2218 | if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
4878e9eb | 2219 | rcu_read_unlock(); |
1da177e4 | 2220 | return -EBUSY; |
4878e9eb | 2221 | } |
1da177e4 | 2222 | } |
4878e9eb | 2223 | rcu_read_unlock(); |
f6b6ec5c SL |
2224 | if (!test_bit(Journal, &rdev->flags) && |
2225 | mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { | |
9d48739e N |
2226 | pr_warn("md: %s: array is limited to %d devices\n", |
2227 | mdname(mddev), mddev->max_disks); | |
de01dfad N |
2228 | return -EBUSY; |
2229 | } | |
19133a42 | 2230 | bdevname(rdev->bdev,b); |
90a9befb | 2231 | strreplace(b, '/', '!'); |
649316b2 | 2232 | |
1da177e4 | 2233 | rdev->mddev = mddev; |
9d48739e | 2234 | pr_debug("md: bind<%s>\n", b); |
86e6ffdd | 2235 | |
b2d6db58 | 2236 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2237 | goto fail; |
86e6ffdd | 2238 | |
0762b8bd | 2239 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2240 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2241 | /* failure here is OK */; | |
2242 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2243 | |
4b80991c | 2244 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2245 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2246 | |
2247 | /* May as well allow recovery to be retried once */ | |
5389042f | 2248 | mddev->recovery_disabled++; |
3f9d99c1 | 2249 | |
1da177e4 | 2250 | return 0; |
5e55e2f5 N |
2251 | |
2252 | fail: | |
9d48739e N |
2253 | pr_warn("md: failed to register dev-%s for %s\n", |
2254 | b, mdname(mddev)); | |
5e55e2f5 | 2255 | return err; |
1da177e4 LT |
2256 | } |
2257 | ||
177a99b2 | 2258 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2259 | { |
3cb03002 | 2260 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2261 | kobject_del(&rdev->kobj); |
177a99b2 | 2262 | kobject_put(&rdev->kobj); |
5792a285 N |
2263 | } |
2264 | ||
f72ffdd6 | 2265 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2266 | { |
2267 | char b[BDEVNAME_SIZE]; | |
403df478 | 2268 | |
49731baa | 2269 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2270 | list_del_rcu(&rdev->same_set); |
9d48739e | 2271 | pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1da177e4 | 2272 | rdev->mddev = NULL; |
86e6ffdd | 2273 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2274 | sysfs_put(rdev->sysfs_state); |
2275 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2276 | rdev->badblocks.count = 0; |
5792a285 | 2277 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2278 | * writing to 'remove' to "dev/state". We also need |
2279 | * to delay it due to rcu usage. | |
5792a285 | 2280 | */ |
4b80991c | 2281 | synchronize_rcu(); |
177a99b2 N |
2282 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2283 | kobject_get(&rdev->kobj); | |
e804ac78 | 2284 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2285 | } |
2286 | ||
2287 | /* | |
2288 | * prevent the device from being mounted, repartitioned or | |
2289 | * otherwise reused by a RAID array (or any other kernel | |
2290 | * subsystem), by bd_claiming the device. | |
2291 | */ | |
3cb03002 | 2292 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2293 | { |
2294 | int err = 0; | |
2295 | struct block_device *bdev; | |
2296 | char b[BDEVNAME_SIZE]; | |
2297 | ||
d4d77629 | 2298 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2299 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 | 2300 | if (IS_ERR(bdev)) { |
9d48739e | 2301 | pr_warn("md: could not open %s.\n", __bdevname(dev, b)); |
1da177e4 LT |
2302 | return PTR_ERR(bdev); |
2303 | } | |
1da177e4 LT |
2304 | rdev->bdev = bdev; |
2305 | return err; | |
2306 | } | |
2307 | ||
3cb03002 | 2308 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2309 | { |
2310 | struct block_device *bdev = rdev->bdev; | |
2311 | rdev->bdev = NULL; | |
e525fd89 | 2312 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2313 | } |
2314 | ||
2315 | void md_autodetect_dev(dev_t dev); | |
2316 | ||
f72ffdd6 | 2317 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2318 | { |
2319 | char b[BDEVNAME_SIZE]; | |
403df478 | 2320 | |
9d48739e | 2321 | pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b)); |
545c8795 | 2322 | md_rdev_clear(rdev); |
1da177e4 | 2323 | #ifndef MODULE |
d0fae18f N |
2324 | if (test_bit(AutoDetected, &rdev->flags)) |
2325 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2326 | #endif |
2327 | unlock_rdev(rdev); | |
86e6ffdd | 2328 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2329 | } |
2330 | ||
fb56dfef | 2331 | void md_kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2332 | { |
2333 | unbind_rdev_from_array(rdev); | |
2334 | export_rdev(rdev); | |
2335 | } | |
fb56dfef | 2336 | EXPORT_SYMBOL_GPL(md_kick_rdev_from_array); |
1da177e4 | 2337 | |
fd01b88c | 2338 | static void export_array(struct mddev *mddev) |
1da177e4 | 2339 | { |
0638bb0e | 2340 | struct md_rdev *rdev; |
1da177e4 | 2341 | |
0638bb0e N |
2342 | while (!list_empty(&mddev->disks)) { |
2343 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2344 | same_set); | |
fb56dfef | 2345 | md_kick_rdev_from_array(rdev); |
1da177e4 | 2346 | } |
1da177e4 LT |
2347 | mddev->raid_disks = 0; |
2348 | mddev->major_version = 0; | |
2349 | } | |
2350 | ||
6497709b N |
2351 | static bool set_in_sync(struct mddev *mddev) |
2352 | { | |
efa4b77b | 2353 | lockdep_assert_held(&mddev->lock); |
4ad23a97 N |
2354 | if (!mddev->in_sync) { |
2355 | mddev->sync_checkers++; | |
2356 | spin_unlock(&mddev->lock); | |
2357 | percpu_ref_switch_to_atomic_sync(&mddev->writes_pending); | |
2358 | spin_lock(&mddev->lock); | |
2359 | if (!mddev->in_sync && | |
2360 | percpu_ref_is_zero(&mddev->writes_pending)) { | |
6497709b | 2361 | mddev->in_sync = 1; |
4ad23a97 N |
2362 | /* |
2363 | * Ensure ->in_sync is visible before we clear | |
2364 | * ->sync_checkers. | |
2365 | */ | |
55cc39f3 | 2366 | smp_mb(); |
6497709b N |
2367 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2368 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
2369 | } | |
4ad23a97 N |
2370 | if (--mddev->sync_checkers == 0) |
2371 | percpu_ref_switch_to_percpu(&mddev->writes_pending); | |
6497709b N |
2372 | } |
2373 | if (mddev->safemode == 1) | |
2374 | mddev->safemode = 0; | |
2375 | return mddev->in_sync; | |
2376 | } | |
2377 | ||
f72ffdd6 | 2378 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2379 | { |
42543769 N |
2380 | /* Update each superblock (in-memory image), but |
2381 | * if we are allowed to, skip spares which already | |
2382 | * have the right event counter, or have one earlier | |
2383 | * (which would mean they aren't being marked as dirty | |
2384 | * with the rest of the array) | |
2385 | */ | |
3cb03002 | 2386 | struct md_rdev *rdev; |
dafb20fa | 2387 | rdev_for_each(rdev, mddev) { |
42543769 N |
2388 | if (rdev->sb_events == mddev->events || |
2389 | (nospares && | |
2390 | rdev->raid_disk < 0 && | |
42543769 N |
2391 | rdev->sb_events+1 == mddev->events)) { |
2392 | /* Don't update this superblock */ | |
2393 | rdev->sb_loaded = 2; | |
2394 | } else { | |
076f968b | 2395 | sync_super(mddev, rdev); |
42543769 N |
2396 | rdev->sb_loaded = 1; |
2397 | } | |
1da177e4 LT |
2398 | } |
2399 | } | |
2400 | ||
2aa82191 GR |
2401 | static bool does_sb_need_changing(struct mddev *mddev) |
2402 | { | |
2403 | struct md_rdev *rdev; | |
2404 | struct mdp_superblock_1 *sb; | |
2405 | int role; | |
2406 | ||
2407 | /* Find a good rdev */ | |
2408 | rdev_for_each(rdev, mddev) | |
2409 | if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags)) | |
2410 | break; | |
2411 | ||
2412 | /* No good device found. */ | |
2413 | if (!rdev) | |
2414 | return false; | |
2415 | ||
2416 | sb = page_address(rdev->sb_page); | |
2417 | /* Check if a device has become faulty or a spare become active */ | |
2418 | rdev_for_each(rdev, mddev) { | |
2419 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
2420 | /* Device activated? */ | |
2421 | if (role == 0xffff && rdev->raid_disk >=0 && | |
2422 | !test_bit(Faulty, &rdev->flags)) | |
2423 | return true; | |
2424 | /* Device turned faulty? */ | |
2425 | if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd)) | |
2426 | return true; | |
2427 | } | |
2428 | ||
2429 | /* Check if any mddev parameters have changed */ | |
2430 | if ((mddev->dev_sectors != le64_to_cpu(sb->size)) || | |
2431 | (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) || | |
13459213 | 2432 | (mddev->layout != le32_to_cpu(sb->layout)) || |
2aa82191 GR |
2433 | (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) || |
2434 | (mddev->chunk_sectors != le32_to_cpu(sb->chunksize))) | |
2435 | return true; | |
2436 | ||
2437 | return false; | |
2438 | } | |
2439 | ||
1aee41f6 | 2440 | void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2441 | { |
3cb03002 | 2442 | struct md_rdev *rdev; |
06d91a5f | 2443 | int sync_req; |
42543769 | 2444 | int nospares = 0; |
2699b672 | 2445 | int any_badblocks_changed = 0; |
23b63f9f | 2446 | int ret = -1; |
1da177e4 | 2447 | |
d87f064f N |
2448 | if (mddev->ro) { |
2449 | if (force_change) | |
2953079c | 2450 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
d87f064f N |
2451 | return; |
2452 | } | |
2aa82191 | 2453 | |
2c97cf13 | 2454 | repeat: |
2aa82191 | 2455 | if (mddev_is_clustered(mddev)) { |
2953079c | 2456 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
2aa82191 | 2457 | force_change = 1; |
2953079c | 2458 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
85ad1d13 | 2459 | nospares = 1; |
23b63f9f | 2460 | ret = md_cluster_ops->metadata_update_start(mddev); |
2aa82191 GR |
2461 | /* Has someone else has updated the sb */ |
2462 | if (!does_sb_need_changing(mddev)) { | |
23b63f9f GJ |
2463 | if (ret == 0) |
2464 | md_cluster_ops->metadata_update_cancel(mddev); | |
2953079c SL |
2465 | bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2466 | BIT(MD_SB_CHANGE_DEVS) | | |
2467 | BIT(MD_SB_CHANGE_CLEAN)); | |
2aa82191 GR |
2468 | return; |
2469 | } | |
2470 | } | |
2c97cf13 | 2471 | |
db0505d3 N |
2472 | /* |
2473 | * First make sure individual recovery_offsets are correct | |
2474 | * curr_resync_completed can only be used during recovery. | |
2475 | * During reshape/resync it might use array-addresses rather | |
2476 | * that device addresses. | |
2477 | */ | |
dafb20fa | 2478 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2479 | if (rdev->raid_disk >= 0 && |
2480 | mddev->delta_disks >= 0 && | |
db0505d3 N |
2481 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
2482 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) && | |
2483 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
f2076e7d | 2484 | !test_bit(Journal, &rdev->flags) && |
3a3a5ddb N |
2485 | !test_bit(In_sync, &rdev->flags) && |
2486 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2487 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2488 | ||
f72ffdd6 | 2489 | } |
bd52b746 | 2490 | if (!mddev->persistent) { |
2953079c SL |
2491 | clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
2492 | clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
de393cde | 2493 | if (!mddev->external) { |
2953079c | 2494 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
dafb20fa | 2495 | rdev_for_each(rdev, mddev) { |
de393cde | 2496 | if (rdev->badblocks.changed) { |
d0962936 | 2497 | rdev->badblocks.changed = 0; |
fc974ee2 | 2498 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2499 | md_error(mddev, rdev); |
2500 | } | |
2501 | clear_bit(Blocked, &rdev->flags); | |
2502 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2503 | wake_up(&rdev->blocked_wait); | |
2504 | } | |
2505 | } | |
3a3a5ddb N |
2506 | wake_up(&mddev->sb_wait); |
2507 | return; | |
2508 | } | |
2509 | ||
85572d7c | 2510 | spin_lock(&mddev->lock); |
84692195 | 2511 | |
9ebc6ef1 | 2512 | mddev->utime = ktime_get_real_seconds(); |
3a3a5ddb | 2513 | |
2953079c | 2514 | if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) |
850b2b42 | 2515 | force_change = 1; |
2953079c | 2516 | if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags)) |
850b2b42 N |
2517 | /* just a clean<-> dirty transition, possibly leave spares alone, |
2518 | * though if events isn't the right even/odd, we will have to do | |
2519 | * spares after all | |
2520 | */ | |
2521 | nospares = 1; | |
2522 | if (force_change) | |
2523 | nospares = 0; | |
2524 | if (mddev->degraded) | |
84692195 N |
2525 | /* If the array is degraded, then skipping spares is both |
2526 | * dangerous and fairly pointless. | |
2527 | * Dangerous because a device that was removed from the array | |
2528 | * might have a event_count that still looks up-to-date, | |
2529 | * so it can be re-added without a resync. | |
2530 | * Pointless because if there are any spares to skip, | |
2531 | * then a recovery will happen and soon that array won't | |
2532 | * be degraded any more and the spare can go back to sleep then. | |
2533 | */ | |
850b2b42 | 2534 | nospares = 0; |
84692195 | 2535 | |
06d91a5f | 2536 | sync_req = mddev->in_sync; |
42543769 N |
2537 | |
2538 | /* If this is just a dirty<->clean transition, and the array is clean | |
2539 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2540 | if (nospares |
42543769 | 2541 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2542 | && mddev->can_decrease_events |
2543 | && mddev->events != 1) { | |
42543769 | 2544 | mddev->events--; |
a8707c08 N |
2545 | mddev->can_decrease_events = 0; |
2546 | } else { | |
42543769 N |
2547 | /* otherwise we have to go forward and ... */ |
2548 | mddev->events ++; | |
a8707c08 | 2549 | mddev->can_decrease_events = nospares; |
42543769 | 2550 | } |
1da177e4 | 2551 | |
403df478 N |
2552 | /* |
2553 | * This 64-bit counter should never wrap. | |
2554 | * Either we are in around ~1 trillion A.C., assuming | |
2555 | * 1 reboot per second, or we have a bug... | |
2556 | */ | |
2557 | WARN_ON(mddev->events == 0); | |
2699b672 | 2558 | |
dafb20fa | 2559 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2560 | if (rdev->badblocks.changed) |
2561 | any_badblocks_changed++; | |
de393cde N |
2562 | if (test_bit(Faulty, &rdev->flags)) |
2563 | set_bit(FaultRecorded, &rdev->flags); | |
2564 | } | |
2699b672 | 2565 | |
e691063a | 2566 | sync_sbs(mddev, nospares); |
85572d7c | 2567 | spin_unlock(&mddev->lock); |
1da177e4 | 2568 | |
36a4e1fe N |
2569 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2570 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2571 | |
504634f6 SL |
2572 | if (mddev->queue) |
2573 | blk_add_trace_msg(mddev->queue, "md md_update_sb"); | |
46533ff7 | 2574 | rewrite: |
4ad13663 | 2575 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2576 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2577 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2578 | |
42543769 N |
2579 | if (rdev->sb_loaded != 1) |
2580 | continue; /* no noise on spare devices */ | |
1da177e4 | 2581 | |
f466722c | 2582 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2583 | md_super_write(mddev,rdev, |
0f420358 | 2584 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2585 | rdev->sb_page); |
36a4e1fe N |
2586 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2587 | bdevname(rdev->bdev, b), | |
2588 | (unsigned long long)rdev->sb_start); | |
42543769 | 2589 | rdev->sb_events = mddev->events; |
2699b672 N |
2590 | if (rdev->badblocks.size) { |
2591 | md_super_write(mddev, rdev, | |
2592 | rdev->badblocks.sector, | |
2593 | rdev->badblocks.size << 9, | |
2594 | rdev->bb_page); | |
2595 | rdev->badblocks.size = 0; | |
2596 | } | |
7bfa19f2 | 2597 | |
f466722c | 2598 | } else |
36a4e1fe N |
2599 | pr_debug("md: %s (skipping faulty)\n", |
2600 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2601 | |
7bfa19f2 | 2602 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2603 | /* only need to write one superblock... */ |
2604 | break; | |
2605 | } | |
46533ff7 N |
2606 | if (md_super_wait(mddev) < 0) |
2607 | goto rewrite; | |
2953079c | 2608 | /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2609 | |
2c97cf13 GJ |
2610 | if (mddev_is_clustered(mddev) && ret == 0) |
2611 | md_cluster_ops->metadata_update_finish(mddev); | |
2612 | ||
850b2b42 | 2613 | if (mddev->in_sync != sync_req || |
2953079c SL |
2614 | !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING), |
2615 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN))) | |
06d91a5f | 2616 | /* have to write it out again */ |
06d91a5f | 2617 | goto repeat; |
3d310eb7 | 2618 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2619 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2620 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2621 | |
dafb20fa | 2622 | rdev_for_each(rdev, mddev) { |
de393cde N |
2623 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2624 | clear_bit(Blocked, &rdev->flags); | |
2625 | ||
2626 | if (any_badblocks_changed) | |
fc974ee2 | 2627 | ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2628 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2629 | wake_up(&rdev->blocked_wait); | |
2630 | } | |
1da177e4 | 2631 | } |
1aee41f6 | 2632 | EXPORT_SYMBOL(md_update_sb); |
1da177e4 | 2633 | |
a6da4ef8 GR |
2634 | static int add_bound_rdev(struct md_rdev *rdev) |
2635 | { | |
2636 | struct mddev *mddev = rdev->mddev; | |
2637 | int err = 0; | |
87d4d916 | 2638 | bool add_journal = test_bit(Journal, &rdev->flags); |
a6da4ef8 | 2639 | |
87d4d916 | 2640 | if (!mddev->pers->hot_remove_disk || add_journal) { |
a6da4ef8 GR |
2641 | /* If there is hot_add_disk but no hot_remove_disk |
2642 | * then added disks for geometry changes, | |
2643 | * and should be added immediately. | |
2644 | */ | |
2645 | super_types[mddev->major_version]. | |
2646 | validate_super(mddev, rdev); | |
87d4d916 SL |
2647 | if (add_journal) |
2648 | mddev_suspend(mddev); | |
a6da4ef8 | 2649 | err = mddev->pers->hot_add_disk(mddev, rdev); |
87d4d916 SL |
2650 | if (add_journal) |
2651 | mddev_resume(mddev); | |
a6da4ef8 | 2652 | if (err) { |
db767672 | 2653 | md_kick_rdev_from_array(rdev); |
a6da4ef8 GR |
2654 | return err; |
2655 | } | |
2656 | } | |
2657 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2658 | ||
2953079c | 2659 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a6da4ef8 GR |
2660 | if (mddev->degraded) |
2661 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
2662 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2663 | md_new_event(mddev); | |
2664 | md_wakeup_thread(mddev->thread); | |
2665 | return 0; | |
2666 | } | |
1da177e4 | 2667 | |
7f6ce769 | 2668 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2669 | * We want to accept with case. For this we use cmd_match. |
2670 | */ | |
2671 | static int cmd_match(const char *cmd, const char *str) | |
2672 | { | |
2673 | /* See if cmd, written into a sysfs file, matches | |
2674 | * str. They must either be the same, or cmd can | |
2675 | * have a trailing newline | |
2676 | */ | |
2677 | while (*cmd && *str && *cmd == *str) { | |
2678 | cmd++; | |
2679 | str++; | |
2680 | } | |
2681 | if (*cmd == '\n') | |
2682 | cmd++; | |
2683 | if (*str || *cmd) | |
2684 | return 0; | |
2685 | return 1; | |
2686 | } | |
2687 | ||
86e6ffdd N |
2688 | struct rdev_sysfs_entry { |
2689 | struct attribute attr; | |
3cb03002 N |
2690 | ssize_t (*show)(struct md_rdev *, char *); |
2691 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2692 | }; |
2693 | ||
2694 | static ssize_t | |
3cb03002 | 2695 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd | 2696 | { |
35b785f7 | 2697 | char *sep = ","; |
20a49ff6 | 2698 | size_t len = 0; |
6aa7de05 | 2699 | unsigned long flags = READ_ONCE(rdev->flags); |
86e6ffdd | 2700 | |
758bfc8a | 2701 | if (test_bit(Faulty, &flags) || |
dcbcb486 TM |
2702 | (!test_bit(ExternalBbl, &flags) && |
2703 | rdev->badblocks.unacked_exist)) | |
35b785f7 TM |
2704 | len += sprintf(page+len, "faulty%s", sep); |
2705 | if (test_bit(In_sync, &flags)) | |
2706 | len += sprintf(page+len, "in_sync%s", sep); | |
2707 | if (test_bit(Journal, &flags)) | |
2708 | len += sprintf(page+len, "journal%s", sep); | |
2709 | if (test_bit(WriteMostly, &flags)) | |
2710 | len += sprintf(page+len, "write_mostly%s", sep); | |
758bfc8a | 2711 | if (test_bit(Blocked, &flags) || |
52c64152 | 2712 | (rdev->badblocks.unacked_exist |
35b785f7 TM |
2713 | && !test_bit(Faulty, &flags))) |
2714 | len += sprintf(page+len, "blocked%s", sep); | |
758bfc8a | 2715 | if (!test_bit(Faulty, &flags) && |
f2076e7d | 2716 | !test_bit(Journal, &flags) && |
35b785f7 TM |
2717 | !test_bit(In_sync, &flags)) |
2718 | len += sprintf(page+len, "spare%s", sep); | |
2719 | if (test_bit(WriteErrorSeen, &flags)) | |
2720 | len += sprintf(page+len, "write_error%s", sep); | |
2721 | if (test_bit(WantReplacement, &flags)) | |
2722 | len += sprintf(page+len, "want_replacement%s", sep); | |
2723 | if (test_bit(Replacement, &flags)) | |
2724 | len += sprintf(page+len, "replacement%s", sep); | |
2725 | if (test_bit(ExternalBbl, &flags)) | |
2726 | len += sprintf(page+len, "external_bbl%s", sep); | |
688834e6 N |
2727 | if (test_bit(FailFast, &flags)) |
2728 | len += sprintf(page+len, "failfast%s", sep); | |
35b785f7 TM |
2729 | |
2730 | if (len) | |
2731 | len -= strlen(sep); | |
2d78f8c4 | 2732 | |
86e6ffdd N |
2733 | return len+sprintf(page+len, "\n"); |
2734 | } | |
2735 | ||
45dc2de1 | 2736 | static ssize_t |
3cb03002 | 2737 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2738 | { |
2739 | /* can write | |
de393cde | 2740 | * faulty - simulates an error |
45dc2de1 | 2741 | * remove - disconnects the device |
f655675b N |
2742 | * writemostly - sets write_mostly |
2743 | * -writemostly - clears write_mostly | |
de393cde N |
2744 | * blocked - sets the Blocked flags |
2745 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2746 | * insync - sets Insync providing device isn't active |
f466722c N |
2747 | * -insync - clear Insync for a device with a slot assigned, |
2748 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2749 | * write_error - sets WriteErrorSeen |
2750 | * -write_error - clears WriteErrorSeen | |
688834e6 | 2751 | * {,-}failfast - set/clear FailFast |
45dc2de1 N |
2752 | */ |
2753 | int err = -EINVAL; | |
2754 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2755 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2756 | if (test_bit(Faulty, &rdev->flags)) |
2757 | err = 0; | |
2758 | else | |
2759 | err = -EBUSY; | |
45dc2de1 | 2760 | } else if (cmd_match(buf, "remove")) { |
5d881783 SL |
2761 | if (rdev->mddev->pers) { |
2762 | clear_bit(Blocked, &rdev->flags); | |
2763 | remove_and_add_spares(rdev->mddev, rdev); | |
2764 | } | |
45dc2de1 N |
2765 | if (rdev->raid_disk >= 0) |
2766 | err = -EBUSY; | |
2767 | else { | |
fd01b88c | 2768 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2769 | err = 0; |
a9720903 GJ |
2770 | if (mddev_is_clustered(mddev)) |
2771 | err = md_cluster_ops->remove_disk(mddev, rdev); | |
2772 | ||
2773 | if (err == 0) { | |
2774 | md_kick_rdev_from_array(rdev); | |
060b0689 | 2775 | if (mddev->pers) { |
2953079c | 2776 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
2777 | md_wakeup_thread(mddev->thread); |
2778 | } | |
a9720903 GJ |
2779 | md_new_event(mddev); |
2780 | } | |
45dc2de1 | 2781 | } |
f655675b N |
2782 | } else if (cmd_match(buf, "writemostly")) { |
2783 | set_bit(WriteMostly, &rdev->flags); | |
2784 | err = 0; | |
2785 | } else if (cmd_match(buf, "-writemostly")) { | |
2786 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2787 | err = 0; |
2788 | } else if (cmd_match(buf, "blocked")) { | |
2789 | set_bit(Blocked, &rdev->flags); | |
2790 | err = 0; | |
2791 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2792 | if (!test_bit(Faulty, &rdev->flags) && |
dcbcb486 | 2793 | !test_bit(ExternalBbl, &rdev->flags) && |
7da64a0a | 2794 | rdev->badblocks.unacked_exist) { |
de393cde N |
2795 | /* metadata handler doesn't understand badblocks, |
2796 | * so we need to fail the device | |
2797 | */ | |
2798 | md_error(rdev->mddev, rdev); | |
2799 | } | |
6bfe0b49 | 2800 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2801 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2802 | wake_up(&rdev->blocked_wait); |
2803 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2804 | md_wakeup_thread(rdev->mddev->thread); | |
2805 | ||
6d56e278 N |
2806 | err = 0; |
2807 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2808 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2809 | err = 0; |
688834e6 N |
2810 | } else if (cmd_match(buf, "failfast")) { |
2811 | set_bit(FailFast, &rdev->flags); | |
2812 | err = 0; | |
2813 | } else if (cmd_match(buf, "-failfast")) { | |
2814 | clear_bit(FailFast, &rdev->flags); | |
2815 | err = 0; | |
f2076e7d SL |
2816 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 && |
2817 | !test_bit(Journal, &rdev->flags)) { | |
e1960f8c N |
2818 | if (rdev->mddev->pers == NULL) { |
2819 | clear_bit(In_sync, &rdev->flags); | |
2820 | rdev->saved_raid_disk = rdev->raid_disk; | |
2821 | rdev->raid_disk = -1; | |
2822 | err = 0; | |
2823 | } | |
d7a9d443 N |
2824 | } else if (cmd_match(buf, "write_error")) { |
2825 | set_bit(WriteErrorSeen, &rdev->flags); | |
2826 | err = 0; | |
2827 | } else if (cmd_match(buf, "-write_error")) { | |
2828 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2829 | err = 0; | |
2d78f8c4 N |
2830 | } else if (cmd_match(buf, "want_replacement")) { |
2831 | /* Any non-spare device that is not a replacement can | |
2832 | * become want_replacement at any time, but we then need to | |
2833 | * check if recovery is needed. | |
2834 | */ | |
2835 | if (rdev->raid_disk >= 0 && | |
f2076e7d | 2836 | !test_bit(Journal, &rdev->flags) && |
2d78f8c4 N |
2837 | !test_bit(Replacement, &rdev->flags)) |
2838 | set_bit(WantReplacement, &rdev->flags); | |
2839 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2840 | md_wakeup_thread(rdev->mddev->thread); | |
2841 | err = 0; | |
2842 | } else if (cmd_match(buf, "-want_replacement")) { | |
2843 | /* Clearing 'want_replacement' is always allowed. | |
2844 | * Once replacements starts it is too late though. | |
2845 | */ | |
2846 | err = 0; | |
2847 | clear_bit(WantReplacement, &rdev->flags); | |
2848 | } else if (cmd_match(buf, "replacement")) { | |
2849 | /* Can only set a device as a replacement when array has not | |
2850 | * yet been started. Once running, replacement is automatic | |
2851 | * from spares, or by assigning 'slot'. | |
2852 | */ | |
2853 | if (rdev->mddev->pers) | |
2854 | err = -EBUSY; | |
2855 | else { | |
2856 | set_bit(Replacement, &rdev->flags); | |
2857 | err = 0; | |
2858 | } | |
2859 | } else if (cmd_match(buf, "-replacement")) { | |
2860 | /* Similarly, can only clear Replacement before start */ | |
2861 | if (rdev->mddev->pers) | |
2862 | err = -EBUSY; | |
2863 | else { | |
2864 | clear_bit(Replacement, &rdev->flags); | |
2865 | err = 0; | |
2866 | } | |
a6da4ef8 | 2867 | } else if (cmd_match(buf, "re-add")) { |
011abdc9 N |
2868 | if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) && |
2869 | rdev->saved_raid_disk >= 0) { | |
97f6cd39 GR |
2870 | /* clear_bit is performed _after_ all the devices |
2871 | * have their local Faulty bit cleared. If any writes | |
2872 | * happen in the meantime in the local node, they | |
2873 | * will land in the local bitmap, which will be synced | |
2874 | * by this node eventually | |
2875 | */ | |
2876 | if (!mddev_is_clustered(rdev->mddev) || | |
2877 | (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) { | |
2878 | clear_bit(Faulty, &rdev->flags); | |
2879 | err = add_bound_rdev(rdev); | |
2880 | } | |
a6da4ef8 GR |
2881 | } else |
2882 | err = -EBUSY; | |
35b785f7 TM |
2883 | } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) { |
2884 | set_bit(ExternalBbl, &rdev->flags); | |
2885 | rdev->badblocks.shift = 0; | |
2886 | err = 0; | |
2887 | } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) { | |
2888 | clear_bit(ExternalBbl, &rdev->flags); | |
2889 | err = 0; | |
45dc2de1 | 2890 | } |
00bcb4ac N |
2891 | if (!err) |
2892 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2893 | return err ? err : len; |
2894 | } | |
80ca3a44 | 2895 | static struct rdev_sysfs_entry rdev_state = |
750f199e | 2896 | __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store); |
86e6ffdd | 2897 | |
4dbcdc75 | 2898 | static ssize_t |
3cb03002 | 2899 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2900 | { |
2901 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2902 | } | |
2903 | ||
2904 | static ssize_t | |
3cb03002 | 2905 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 | 2906 | { |
4c9309c0 AD |
2907 | unsigned int n; |
2908 | int rv; | |
2909 | ||
2910 | rv = kstrtouint(buf, 10, &n); | |
2911 | if (rv < 0) | |
2912 | return rv; | |
2913 | atomic_set(&rdev->corrected_errors, n); | |
2914 | return len; | |
4dbcdc75 N |
2915 | } |
2916 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2917 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2918 | |
014236d2 | 2919 | static ssize_t |
3cb03002 | 2920 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 | 2921 | { |
f2076e7d SL |
2922 | if (test_bit(Journal, &rdev->flags)) |
2923 | return sprintf(page, "journal\n"); | |
2924 | else if (rdev->raid_disk < 0) | |
014236d2 N |
2925 | return sprintf(page, "none\n"); |
2926 | else | |
2927 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2928 | } | |
2929 | ||
2930 | static ssize_t | |
3cb03002 | 2931 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 | 2932 | { |
4c9309c0 | 2933 | int slot; |
c303da6d | 2934 | int err; |
4c9309c0 | 2935 | |
f2076e7d SL |
2936 | if (test_bit(Journal, &rdev->flags)) |
2937 | return -EBUSY; | |
014236d2 N |
2938 | if (strncmp(buf, "none", 4)==0) |
2939 | slot = -1; | |
4c9309c0 AD |
2940 | else { |
2941 | err = kstrtouint(buf, 10, (unsigned int *)&slot); | |
2942 | if (err < 0) | |
2943 | return err; | |
2944 | } | |
6c2fce2e | 2945 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2946 | /* Setting 'slot' on an active array requires also |
2947 | * updating the 'rd%d' link, and communicating | |
2948 | * with the personality with ->hot_*_disk. | |
2949 | * For now we only support removing | |
2950 | * failed/spare devices. This normally happens automatically, | |
2951 | * but not when the metadata is externally managed. | |
2952 | */ | |
c303da6d N |
2953 | if (rdev->raid_disk == -1) |
2954 | return -EEXIST; | |
2955 | /* personality does all needed checks */ | |
01393f3d | 2956 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2957 | return -EINVAL; |
746d3207 N |
2958 | clear_bit(Blocked, &rdev->flags); |
2959 | remove_and_add_spares(rdev->mddev, rdev); | |
2960 | if (rdev->raid_disk >= 0) | |
2961 | return -EBUSY; | |
c303da6d N |
2962 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2963 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2964 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2965 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2966 | * if we ever get bitmaps working here. |
6c2fce2e | 2967 | */ |
cb01c549 | 2968 | int err; |
6c2fce2e NB |
2969 | |
2970 | if (rdev->raid_disk != -1) | |
2971 | return -EBUSY; | |
2972 | ||
c6751b2b N |
2973 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2974 | return -EBUSY; | |
2975 | ||
6c2fce2e NB |
2976 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2977 | return -EINVAL; | |
2978 | ||
ba1b41b6 N |
2979 | if (slot >= rdev->mddev->raid_disks && |
2980 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2981 | return -ENOSPC; | |
2982 | ||
6c2fce2e NB |
2983 | rdev->raid_disk = slot; |
2984 | if (test_bit(In_sync, &rdev->flags)) | |
2985 | rdev->saved_raid_disk = slot; | |
2986 | else | |
2987 | rdev->saved_raid_disk = -1; | |
d30519fc | 2988 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2989 | clear_bit(Bitmap_sync, &rdev->flags); |
cb01c549 GR |
2990 | err = rdev->mddev->pers-> |
2991 | hot_add_disk(rdev->mddev, rdev); | |
2992 | if (err) { | |
2993 | rdev->raid_disk = -1; | |
2994 | return err; | |
2995 | } else | |
2996 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2997 | if (sysfs_link_rdev(rdev->mddev, rdev)) | |
2998 | /* failure here is OK */; | |
6c2fce2e | 2999 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 3000 | } else { |
ba1b41b6 N |
3001 | if (slot >= rdev->mddev->raid_disks && |
3002 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
3003 | return -ENOSPC; |
3004 | rdev->raid_disk = slot; | |
3005 | /* assume it is working */ | |
c5d79adb N |
3006 | clear_bit(Faulty, &rdev->flags); |
3007 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 3008 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 3009 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 3010 | } |
014236d2 N |
3011 | return len; |
3012 | } | |
3013 | ||
014236d2 | 3014 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 3015 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 3016 | |
93c8cad0 | 3017 | static ssize_t |
3cb03002 | 3018 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 3019 | { |
6961ece4 | 3020 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
3021 | } |
3022 | ||
3023 | static ssize_t | |
3cb03002 | 3024 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 3025 | { |
c6563a8c | 3026 | unsigned long long offset; |
b29bebd6 | 3027 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 3028 | return -EINVAL; |
8ed0a521 | 3029 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 3030 | return -EBUSY; |
dd8ac336 | 3031 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
3032 | /* Must set offset before size, so overlap checks |
3033 | * can be sane */ | |
3034 | return -EBUSY; | |
93c8cad0 | 3035 | rdev->data_offset = offset; |
25f7fd47 | 3036 | rdev->new_data_offset = offset; |
93c8cad0 N |
3037 | return len; |
3038 | } | |
3039 | ||
3040 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 3041 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 3042 | |
c6563a8c N |
3043 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
3044 | { | |
3045 | return sprintf(page, "%llu\n", | |
3046 | (unsigned long long)rdev->new_data_offset); | |
3047 | } | |
3048 | ||
3049 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
3050 | const char *buf, size_t len) | |
3051 | { | |
3052 | unsigned long long new_offset; | |
3053 | struct mddev *mddev = rdev->mddev; | |
3054 | ||
b29bebd6 | 3055 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
3056 | return -EINVAL; |
3057 | ||
f851b60d N |
3058 | if (mddev->sync_thread || |
3059 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
3060 | return -EBUSY; |
3061 | if (new_offset == rdev->data_offset) | |
3062 | /* reset is always permitted */ | |
3063 | ; | |
3064 | else if (new_offset > rdev->data_offset) { | |
3065 | /* must not push array size beyond rdev_sectors */ | |
3066 | if (new_offset - rdev->data_offset | |
3067 | + mddev->dev_sectors > rdev->sectors) | |
3068 | return -E2BIG; | |
3069 | } | |
3070 | /* Metadata worries about other space details. */ | |
3071 | ||
3072 | /* decreasing the offset is inconsistent with a backwards | |
3073 | * reshape. | |
3074 | */ | |
3075 | if (new_offset < rdev->data_offset && | |
3076 | mddev->reshape_backwards) | |
3077 | return -EINVAL; | |
3078 | /* Increasing offset is inconsistent with forwards | |
3079 | * reshape. reshape_direction should be set to | |
3080 | * 'backwards' first. | |
3081 | */ | |
3082 | if (new_offset > rdev->data_offset && | |
3083 | !mddev->reshape_backwards) | |
3084 | return -EINVAL; | |
3085 | ||
3086 | if (mddev->pers && mddev->persistent && | |
3087 | !super_types[mddev->major_version] | |
3088 | .allow_new_offset(rdev, new_offset)) | |
3089 | return -E2BIG; | |
3090 | rdev->new_data_offset = new_offset; | |
3091 | if (new_offset > rdev->data_offset) | |
3092 | mddev->reshape_backwards = 1; | |
3093 | else if (new_offset < rdev->data_offset) | |
3094 | mddev->reshape_backwards = 0; | |
3095 | ||
3096 | return len; | |
3097 | } | |
3098 | static struct rdev_sysfs_entry rdev_new_offset = | |
3099 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
3100 | ||
83303b61 | 3101 | static ssize_t |
3cb03002 | 3102 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 3103 | { |
dd8ac336 | 3104 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
3105 | } |
3106 | ||
c5d79adb N |
3107 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
3108 | { | |
3109 | /* check if two start/length pairs overlap */ | |
3110 | if (s1+l1 <= s2) | |
3111 | return 0; | |
3112 | if (s2+l2 <= s1) | |
3113 | return 0; | |
3114 | return 1; | |
3115 | } | |
3116 | ||
b522adcd DW |
3117 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
3118 | { | |
3119 | unsigned long long blocks; | |
3120 | sector_t new; | |
3121 | ||
b29bebd6 | 3122 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
3123 | return -EINVAL; |
3124 | ||
3125 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
3126 | return -EINVAL; /* sector conversion overflow */ | |
3127 | ||
3128 | new = blocks * 2; | |
3129 | if (new != blocks * 2) | |
3130 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
3131 | ||
3132 | *sectors = new; | |
3133 | return 0; | |
3134 | } | |
3135 | ||
83303b61 | 3136 | static ssize_t |
3cb03002 | 3137 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 3138 | { |
fd01b88c | 3139 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 3140 | sector_t oldsectors = rdev->sectors; |
b522adcd | 3141 | sector_t sectors; |
27c529bb | 3142 | |
f2076e7d SL |
3143 | if (test_bit(Journal, &rdev->flags)) |
3144 | return -EBUSY; | |
b522adcd | 3145 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 3146 | return -EINVAL; |
c6563a8c N |
3147 | if (rdev->data_offset != rdev->new_data_offset) |
3148 | return -EINVAL; /* too confusing */ | |
0cd17fec | 3149 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 3150 | if (my_mddev->persistent) { |
dd8ac336 AN |
3151 | sectors = super_types[my_mddev->major_version]. |
3152 | rdev_size_change(rdev, sectors); | |
3153 | if (!sectors) | |
0cd17fec | 3154 | return -EBUSY; |
dd8ac336 | 3155 | } else if (!sectors) |
77304d2a | 3156 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 3157 | rdev->data_offset; |
a6468539 N |
3158 | if (!my_mddev->pers->resize) |
3159 | /* Cannot change size for RAID0 or Linear etc */ | |
3160 | return -EINVAL; | |
0cd17fec | 3161 | } |
dd8ac336 | 3162 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 3163 | return -EINVAL; /* component must fit device */ |
0cd17fec | 3164 | |
dd8ac336 AN |
3165 | rdev->sectors = sectors; |
3166 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
3167 | /* Need to check that all other rdevs with the same |
3168 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
3169 | * the rdev lists safely. | |
3170 | * This check does not provide a hard guarantee, it | |
3171 | * just helps avoid dangerous mistakes. | |
c5d79adb | 3172 | */ |
fd01b88c | 3173 | struct mddev *mddev; |
c5d79adb | 3174 | int overlap = 0; |
159ec1fc | 3175 | struct list_head *tmp; |
c5d79adb | 3176 | |
8b1afc3d | 3177 | rcu_read_lock(); |
29ac4aa3 | 3178 | for_each_mddev(mddev, tmp) { |
3cb03002 | 3179 | struct md_rdev *rdev2; |
c5d79adb | 3180 | |
dafb20fa | 3181 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
3182 | if (rdev->bdev == rdev2->bdev && |
3183 | rdev != rdev2 && | |
3184 | overlaps(rdev->data_offset, rdev->sectors, | |
3185 | rdev2->data_offset, | |
3186 | rdev2->sectors)) { | |
c5d79adb N |
3187 | overlap = 1; |
3188 | break; | |
3189 | } | |
c5d79adb N |
3190 | if (overlap) { |
3191 | mddev_put(mddev); | |
3192 | break; | |
3193 | } | |
3194 | } | |
8b1afc3d | 3195 | rcu_read_unlock(); |
c5d79adb N |
3196 | if (overlap) { |
3197 | /* Someone else could have slipped in a size | |
3198 | * change here, but doing so is just silly. | |
dd8ac336 | 3199 | * We put oldsectors back because we *know* it is |
c5d79adb N |
3200 | * safe, and trust userspace not to race with |
3201 | * itself | |
3202 | */ | |
dd8ac336 | 3203 | rdev->sectors = oldsectors; |
c5d79adb N |
3204 | return -EBUSY; |
3205 | } | |
3206 | } | |
83303b61 N |
3207 | return len; |
3208 | } | |
3209 | ||
3210 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 3211 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 3212 | |
3cb03002 | 3213 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
3214 | { |
3215 | unsigned long long recovery_start = rdev->recovery_offset; | |
3216 | ||
3217 | if (test_bit(In_sync, &rdev->flags) || | |
3218 | recovery_start == MaxSector) | |
3219 | return sprintf(page, "none\n"); | |
3220 | ||
3221 | return sprintf(page, "%llu\n", recovery_start); | |
3222 | } | |
3223 | ||
3cb03002 | 3224 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
3225 | { |
3226 | unsigned long long recovery_start; | |
3227 | ||
3228 | if (cmd_match(buf, "none")) | |
3229 | recovery_start = MaxSector; | |
b29bebd6 | 3230 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
3231 | return -EINVAL; |
3232 | ||
3233 | if (rdev->mddev->pers && | |
3234 | rdev->raid_disk >= 0) | |
3235 | return -EBUSY; | |
3236 | ||
3237 | rdev->recovery_offset = recovery_start; | |
3238 | if (recovery_start == MaxSector) | |
3239 | set_bit(In_sync, &rdev->flags); | |
3240 | else | |
3241 | clear_bit(In_sync, &rdev->flags); | |
3242 | return len; | |
3243 | } | |
3244 | ||
3245 | static struct rdev_sysfs_entry rdev_recovery_start = | |
3246 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
3247 | ||
fc974ee2 VV |
3248 | /* sysfs access to bad-blocks list. |
3249 | * We present two files. | |
3250 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
3251 | * are recorded as bad. The list is truncated to fit within | |
3252 | * the one-page limit of sysfs. | |
3253 | * Writing "sector length" to this file adds an acknowledged | |
3254 | * bad block list. | |
3255 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
3256 | * been acknowledged. Writing to this file adds bad blocks | |
3257 | * without acknowledging them. This is largely for testing. | |
3258 | */ | |
3cb03002 | 3259 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3260 | { |
3261 | return badblocks_show(&rdev->badblocks, page, 0); | |
3262 | } | |
3cb03002 | 3263 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 3264 | { |
de393cde N |
3265 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
3266 | /* Maybe that ack was all we needed */ | |
3267 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
3268 | wake_up(&rdev->blocked_wait); | |
3269 | return rv; | |
16c791a5 N |
3270 | } |
3271 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
3272 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
3273 | ||
3cb03002 | 3274 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
3275 | { |
3276 | return badblocks_show(&rdev->badblocks, page, 1); | |
3277 | } | |
3cb03002 | 3278 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
3279 | { |
3280 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
3281 | } | |
3282 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
3283 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
3284 | ||
664aed04 AP |
3285 | static ssize_t |
3286 | ppl_sector_show(struct md_rdev *rdev, char *page) | |
3287 | { | |
3288 | return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector); | |
3289 | } | |
3290 | ||
3291 | static ssize_t | |
3292 | ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len) | |
3293 | { | |
3294 | unsigned long long sector; | |
3295 | ||
3296 | if (kstrtoull(buf, 10, §or) < 0) | |
3297 | return -EINVAL; | |
3298 | if (sector != (sector_t)sector) | |
3299 | return -EINVAL; | |
3300 | ||
3301 | if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) && | |
3302 | rdev->raid_disk >= 0) | |
3303 | return -EBUSY; | |
3304 | ||
3305 | if (rdev->mddev->persistent) { | |
3306 | if (rdev->mddev->major_version == 0) | |
3307 | return -EINVAL; | |
3308 | if ((sector > rdev->sb_start && | |
3309 | sector - rdev->sb_start > S16_MAX) || | |
3310 | (sector < rdev->sb_start && | |
3311 | rdev->sb_start - sector > -S16_MIN)) | |
3312 | return -EINVAL; | |
3313 | rdev->ppl.offset = sector - rdev->sb_start; | |
3314 | } else if (!rdev->mddev->external) { | |
3315 | return -EBUSY; | |
3316 | } | |
3317 | rdev->ppl.sector = sector; | |
3318 | return len; | |
3319 | } | |
3320 | ||
3321 | static struct rdev_sysfs_entry rdev_ppl_sector = | |
3322 | __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store); | |
3323 | ||
3324 | static ssize_t | |
3325 | ppl_size_show(struct md_rdev *rdev, char *page) | |
3326 | { | |
3327 | return sprintf(page, "%u\n", rdev->ppl.size); | |
3328 | } | |
3329 | ||
3330 | static ssize_t | |
3331 | ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len) | |
3332 | { | |
3333 | unsigned int size; | |
3334 | ||
3335 | if (kstrtouint(buf, 10, &size) < 0) | |
3336 | return -EINVAL; | |
3337 | ||
3338 | if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) && | |
3339 | rdev->raid_disk >= 0) | |
3340 | return -EBUSY; | |
3341 | ||
3342 | if (rdev->mddev->persistent) { | |
3343 | if (rdev->mddev->major_version == 0) | |
3344 | return -EINVAL; | |
3345 | if (size > U16_MAX) | |
3346 | return -EINVAL; | |
3347 | } else if (!rdev->mddev->external) { | |
3348 | return -EBUSY; | |
3349 | } | |
3350 | rdev->ppl.size = size; | |
3351 | return len; | |
3352 | } | |
3353 | ||
3354 | static struct rdev_sysfs_entry rdev_ppl_size = | |
3355 | __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store); | |
3356 | ||
86e6ffdd N |
3357 | static struct attribute *rdev_default_attrs[] = { |
3358 | &rdev_state.attr, | |
4dbcdc75 | 3359 | &rdev_errors.attr, |
014236d2 | 3360 | &rdev_slot.attr, |
93c8cad0 | 3361 | &rdev_offset.attr, |
c6563a8c | 3362 | &rdev_new_offset.attr, |
83303b61 | 3363 | &rdev_size.attr, |
06e3c817 | 3364 | &rdev_recovery_start.attr, |
16c791a5 N |
3365 | &rdev_bad_blocks.attr, |
3366 | &rdev_unack_bad_blocks.attr, | |
664aed04 AP |
3367 | &rdev_ppl_sector.attr, |
3368 | &rdev_ppl_size.attr, | |
86e6ffdd N |
3369 | NULL, |
3370 | }; | |
3371 | static ssize_t | |
3372 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3373 | { | |
3374 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3375 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
3376 | |
3377 | if (!entry->show) | |
3378 | return -EIO; | |
758bfc8a N |
3379 | if (!rdev->mddev) |
3380 | return -EBUSY; | |
3381 | return entry->show(rdev, page); | |
86e6ffdd N |
3382 | } |
3383 | ||
3384 | static ssize_t | |
3385 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3386 | const char *page, size_t length) | |
3387 | { | |
3388 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3389 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3390 | ssize_t rv; |
fd01b88c | 3391 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3392 | |
3393 | if (!entry->store) | |
3394 | return -EIO; | |
67463acb N |
3395 | if (!capable(CAP_SYS_ADMIN)) |
3396 | return -EACCES; | |
27c529bb | 3397 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3398 | if (!rv) { |
27c529bb N |
3399 | if (rdev->mddev == NULL) |
3400 | rv = -EBUSY; | |
3401 | else | |
3402 | rv = entry->store(rdev, page, length); | |
6a51830e | 3403 | mddev_unlock(mddev); |
ca388059 N |
3404 | } |
3405 | return rv; | |
86e6ffdd N |
3406 | } |
3407 | ||
3408 | static void rdev_free(struct kobject *ko) | |
3409 | { | |
3cb03002 | 3410 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3411 | kfree(rdev); |
3412 | } | |
52cf25d0 | 3413 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3414 | .show = rdev_attr_show, |
3415 | .store = rdev_attr_store, | |
3416 | }; | |
3417 | static struct kobj_type rdev_ktype = { | |
3418 | .release = rdev_free, | |
3419 | .sysfs_ops = &rdev_sysfs_ops, | |
3420 | .default_attrs = rdev_default_attrs, | |
3421 | }; | |
3422 | ||
3cb03002 | 3423 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3424 | { |
3425 | rdev->desc_nr = -1; | |
3426 | rdev->saved_raid_disk = -1; | |
3427 | rdev->raid_disk = -1; | |
3428 | rdev->flags = 0; | |
3429 | rdev->data_offset = 0; | |
c6563a8c | 3430 | rdev->new_data_offset = 0; |
e8bb9a83 | 3431 | rdev->sb_events = 0; |
0e3ef49e | 3432 | rdev->last_read_error = 0; |
2699b672 N |
3433 | rdev->sb_loaded = 0; |
3434 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3435 | atomic_set(&rdev->nr_pending, 0); |
3436 | atomic_set(&rdev->read_errors, 0); | |
3437 | atomic_set(&rdev->corrected_errors, 0); | |
3438 | ||
3439 | INIT_LIST_HEAD(&rdev->same_set); | |
3440 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3441 | |
3442 | /* Add space to store bad block list. | |
3443 | * This reserves the space even on arrays where it cannot | |
3444 | * be used - I wonder if that matters | |
3445 | */ | |
fc974ee2 | 3446 | return badblocks_init(&rdev->badblocks, 0); |
e8bb9a83 N |
3447 | } |
3448 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3449 | /* |
3450 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3451 | * | |
3452 | * mark the device faulty if: | |
3453 | * | |
3454 | * - the device is nonexistent (zero size) | |
3455 | * - the device has no valid superblock | |
3456 | * | |
3457 | * a faulty rdev _never_ has rdev->sb set. | |
3458 | */ | |
3cb03002 | 3459 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3460 | { |
3461 | char b[BDEVNAME_SIZE]; | |
3462 | int err; | |
3cb03002 | 3463 | struct md_rdev *rdev; |
1da177e4 LT |
3464 | sector_t size; |
3465 | ||
9ffae0cf | 3466 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
9d48739e | 3467 | if (!rdev) |
1da177e4 | 3468 | return ERR_PTR(-ENOMEM); |
1da177e4 | 3469 | |
2230dfe4 N |
3470 | err = md_rdev_init(rdev); |
3471 | if (err) | |
3472 | goto abort_free; | |
3473 | err = alloc_disk_sb(rdev); | |
3474 | if (err) | |
1da177e4 LT |
3475 | goto abort_free; |
3476 | ||
c5d79adb | 3477 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3478 | if (err) |
3479 | goto abort_free; | |
3480 | ||
f9cb074b | 3481 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3482 | |
77304d2a | 3483 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3484 | if (!size) { |
9d48739e | 3485 | pr_warn("md: %s has zero or unknown size, marking faulty!\n", |
1da177e4 LT |
3486 | bdevname(rdev->bdev,b)); |
3487 | err = -EINVAL; | |
3488 | goto abort_free; | |
3489 | } | |
3490 | ||
3491 | if (super_format >= 0) { | |
3492 | err = super_types[super_format]. | |
3493 | load_super(rdev, NULL, super_minor); | |
3494 | if (err == -EINVAL) { | |
9d48739e | 3495 | pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n", |
df968c4e | 3496 | bdevname(rdev->bdev,b), |
9d48739e | 3497 | super_format, super_minor); |
1da177e4 LT |
3498 | goto abort_free; |
3499 | } | |
3500 | if (err < 0) { | |
9d48739e | 3501 | pr_warn("md: could not read %s's sb, not importing!\n", |
1da177e4 LT |
3502 | bdevname(rdev->bdev,b)); |
3503 | goto abort_free; | |
3504 | } | |
3505 | } | |
6bfe0b49 | 3506 | |
1da177e4 LT |
3507 | return rdev; |
3508 | ||
3509 | abort_free: | |
2699b672 N |
3510 | if (rdev->bdev) |
3511 | unlock_rdev(rdev); | |
545c8795 | 3512 | md_rdev_clear(rdev); |
1da177e4 LT |
3513 | kfree(rdev); |
3514 | return ERR_PTR(err); | |
3515 | } | |
3516 | ||
3517 | /* | |
3518 | * Check a full RAID array for plausibility | |
3519 | */ | |
3520 | ||
f72ffdd6 | 3521 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3522 | { |
3523 | int i; | |
3cb03002 | 3524 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3525 | char b[BDEVNAME_SIZE]; |
3526 | ||
3527 | freshest = NULL; | |
dafb20fa | 3528 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3529 | switch (super_types[mddev->major_version]. |
3530 | load_super(rdev, freshest, mddev->minor_version)) { | |
3531 | case 1: | |
3532 | freshest = rdev; | |
3533 | break; | |
3534 | case 0: | |
3535 | break; | |
3536 | default: | |
9d48739e | 3537 | pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n", |
1da177e4 | 3538 | bdevname(rdev->bdev,b)); |
fb56dfef | 3539 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3540 | } |
3541 | ||
1da177e4 LT |
3542 | super_types[mddev->major_version]. |
3543 | validate_super(mddev, freshest); | |
3544 | ||
3545 | i = 0; | |
dafb20fa | 3546 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3547 | if (mddev->max_disks && |
3548 | (rdev->desc_nr >= mddev->max_disks || | |
3549 | i > mddev->max_disks)) { | |
9d48739e N |
3550 | pr_warn("md: %s: %s: only %d devices permitted\n", |
3551 | mdname(mddev), bdevname(rdev->bdev, b), | |
3552 | mddev->max_disks); | |
fb56dfef | 3553 | md_kick_rdev_from_array(rdev); |
de01dfad N |
3554 | continue; |
3555 | } | |
1aee41f6 | 3556 | if (rdev != freshest) { |
1da177e4 LT |
3557 | if (super_types[mddev->major_version]. |
3558 | validate_super(mddev, rdev)) { | |
9d48739e | 3559 | pr_warn("md: kicking non-fresh %s from array!\n", |
1da177e4 | 3560 | bdevname(rdev->bdev,b)); |
fb56dfef | 3561 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3562 | continue; |
3563 | } | |
1aee41f6 | 3564 | } |
1da177e4 LT |
3565 | if (mddev->level == LEVEL_MULTIPATH) { |
3566 | rdev->desc_nr = i++; | |
3567 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3568 | set_bit(In_sync, &rdev->flags); |
f2076e7d SL |
3569 | } else if (rdev->raid_disk >= |
3570 | (mddev->raid_disks - min(0, mddev->delta_disks)) && | |
3571 | !test_bit(Journal, &rdev->flags)) { | |
a778b73f N |
3572 | rdev->raid_disk = -1; |
3573 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3574 | } |
3575 | } | |
1da177e4 LT |
3576 | } |
3577 | ||
72e02075 N |
3578 | /* Read a fixed-point number. |
3579 | * Numbers in sysfs attributes should be in "standard" units where | |
3580 | * possible, so time should be in seconds. | |
f72ffdd6 | 3581 | * However we internally use a a much smaller unit such as |
72e02075 N |
3582 | * milliseconds or jiffies. |
3583 | * This function takes a decimal number with a possible fractional | |
3584 | * component, and produces an integer which is the result of | |
3585 | * multiplying that number by 10^'scale'. | |
3586 | * all without any floating-point arithmetic. | |
3587 | */ | |
3588 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3589 | { | |
3590 | unsigned long result = 0; | |
3591 | long decimals = -1; | |
3592 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3593 | if (*cp == '.') | |
3594 | decimals = 0; | |
3595 | else if (decimals < scale) { | |
3596 | unsigned int value; | |
3597 | value = *cp - '0'; | |
3598 | result = result * 10 + value; | |
3599 | if (decimals >= 0) | |
3600 | decimals++; | |
3601 | } | |
3602 | cp++; | |
3603 | } | |
3604 | if (*cp == '\n') | |
3605 | cp++; | |
3606 | if (*cp) | |
3607 | return -EINVAL; | |
3608 | if (decimals < 0) | |
3609 | decimals = 0; | |
3610 | while (decimals < scale) { | |
3611 | result *= 10; | |
3612 | decimals ++; | |
3613 | } | |
3614 | *res = result; | |
3615 | return 0; | |
3616 | } | |
3617 | ||
16f17b39 | 3618 | static ssize_t |
fd01b88c | 3619 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3620 | { |
3621 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3622 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3623 | } | |
3624 | static ssize_t | |
fd01b88c | 3625 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3626 | { |
16f17b39 | 3627 | unsigned long msec; |
97ce0a7f | 3628 | |
28c1b9fd | 3629 | if (mddev_is_clustered(mddev)) { |
9d48739e | 3630 | pr_warn("md: Safemode is disabled for clustered mode\n"); |
28c1b9fd GR |
3631 | return -EINVAL; |
3632 | } | |
3633 | ||
72e02075 | 3634 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3635 | return -EINVAL; |
16f17b39 N |
3636 | if (msec == 0) |
3637 | mddev->safemode_delay = 0; | |
3638 | else { | |
19052c0e | 3639 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3640 | unsigned long new_delay = (msec*HZ)/1000; |
3641 | ||
3642 | if (new_delay == 0) | |
3643 | new_delay = 1; | |
3644 | mddev->safemode_delay = new_delay; | |
3645 | if (new_delay < old_delay || old_delay == 0) | |
3646 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3647 | } |
3648 | return len; | |
3649 | } | |
3650 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3651 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3652 | |
eae1701f | 3653 | static ssize_t |
fd01b88c | 3654 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3655 | { |
36d091f4 N |
3656 | struct md_personality *p; |
3657 | int ret; | |
3658 | spin_lock(&mddev->lock); | |
3659 | p = mddev->pers; | |
d9d166c2 | 3660 | if (p) |
36d091f4 | 3661 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3662 | else if (mddev->clevel[0]) |
36d091f4 | 3663 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3664 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3665 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3666 | else |
36d091f4 N |
3667 | ret = 0; |
3668 | spin_unlock(&mddev->lock); | |
3669 | return ret; | |
eae1701f N |
3670 | } |
3671 | ||
d9d166c2 | 3672 | static ssize_t |
fd01b88c | 3673 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3674 | { |
f2859af6 | 3675 | char clevel[16]; |
6791875e N |
3676 | ssize_t rv; |
3677 | size_t slen = len; | |
db721d32 | 3678 | struct md_personality *pers, *oldpers; |
f2859af6 | 3679 | long level; |
db721d32 | 3680 | void *priv, *oldpriv; |
3cb03002 | 3681 | struct md_rdev *rdev; |
245f46c2 | 3682 | |
6791875e N |
3683 | if (slen == 0 || slen >= sizeof(clevel)) |
3684 | return -EINVAL; | |
3685 | ||
3686 | rv = mddev_lock(mddev); | |
3687 | if (rv) | |
3688 | return rv; | |
3689 | ||
245f46c2 | 3690 | if (mddev->pers == NULL) { |
6791875e N |
3691 | strncpy(mddev->clevel, buf, slen); |
3692 | if (mddev->clevel[slen-1] == '\n') | |
3693 | slen--; | |
3694 | mddev->clevel[slen] = 0; | |
245f46c2 | 3695 | mddev->level = LEVEL_NONE; |
6791875e N |
3696 | rv = len; |
3697 | goto out_unlock; | |
245f46c2 | 3698 | } |
6791875e | 3699 | rv = -EROFS; |
bd8839e0 | 3700 | if (mddev->ro) |
6791875e | 3701 | goto out_unlock; |
245f46c2 N |
3702 | |
3703 | /* request to change the personality. Need to ensure: | |
3704 | * - array is not engaged in resync/recovery/reshape | |
3705 | * - old personality can be suspended | |
3706 | * - new personality will access other array. | |
3707 | */ | |
3708 | ||
6791875e | 3709 | rv = -EBUSY; |
bb4f1e9d | 3710 | if (mddev->sync_thread || |
f851b60d | 3711 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3712 | mddev->reshape_position != MaxSector || |
3713 | mddev->sysfs_active) | |
6791875e | 3714 | goto out_unlock; |
245f46c2 | 3715 | |
6791875e | 3716 | rv = -EINVAL; |
245f46c2 | 3717 | if (!mddev->pers->quiesce) { |
9d48739e N |
3718 | pr_warn("md: %s: %s does not support online personality change\n", |
3719 | mdname(mddev), mddev->pers->name); | |
6791875e | 3720 | goto out_unlock; |
245f46c2 N |
3721 | } |
3722 | ||
3723 | /* Now find the new personality */ | |
6791875e N |
3724 | strncpy(clevel, buf, slen); |
3725 | if (clevel[slen-1] == '\n') | |
3726 | slen--; | |
3727 | clevel[slen] = 0; | |
b29bebd6 | 3728 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3729 | level = LEVEL_NONE; |
245f46c2 | 3730 | |
f2859af6 DW |
3731 | if (request_module("md-%s", clevel) != 0) |
3732 | request_module("md-level-%s", clevel); | |
245f46c2 | 3733 | spin_lock(&pers_lock); |
f2859af6 | 3734 | pers = find_pers(level, clevel); |
245f46c2 N |
3735 | if (!pers || !try_module_get(pers->owner)) { |
3736 | spin_unlock(&pers_lock); | |
9d48739e | 3737 | pr_warn("md: personality %s not loaded\n", clevel); |
6791875e N |
3738 | rv = -EINVAL; |
3739 | goto out_unlock; | |
245f46c2 N |
3740 | } |
3741 | spin_unlock(&pers_lock); | |
3742 | ||
3743 | if (pers == mddev->pers) { | |
3744 | /* Nothing to do! */ | |
3745 | module_put(pers->owner); | |
6791875e N |
3746 | rv = len; |
3747 | goto out_unlock; | |
245f46c2 N |
3748 | } |
3749 | if (!pers->takeover) { | |
3750 | module_put(pers->owner); | |
9d48739e N |
3751 | pr_warn("md: %s: %s does not support personality takeover\n", |
3752 | mdname(mddev), clevel); | |
6791875e N |
3753 | rv = -EINVAL; |
3754 | goto out_unlock; | |
245f46c2 N |
3755 | } |
3756 | ||
dafb20fa | 3757 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3758 | rdev->new_raid_disk = rdev->raid_disk; |
3759 | ||
245f46c2 N |
3760 | /* ->takeover must set new_* and/or delta_disks |
3761 | * if it succeeds, and may set them when it fails. | |
3762 | */ | |
3763 | priv = pers->takeover(mddev); | |
3764 | if (IS_ERR(priv)) { | |
3765 | mddev->new_level = mddev->level; | |
3766 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3767 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3768 | mddev->raid_disks -= mddev->delta_disks; |
3769 | mddev->delta_disks = 0; | |
2c810cdd | 3770 | mddev->reshape_backwards = 0; |
245f46c2 | 3771 | module_put(pers->owner); |
9d48739e N |
3772 | pr_warn("md: %s: %s would not accept array\n", |
3773 | mdname(mddev), clevel); | |
6791875e N |
3774 | rv = PTR_ERR(priv); |
3775 | goto out_unlock; | |
245f46c2 N |
3776 | } |
3777 | ||
3778 | /* Looks like we have a winner */ | |
3779 | mddev_suspend(mddev); | |
5aa61f42 | 3780 | mddev_detach(mddev); |
36d091f4 N |
3781 | |
3782 | spin_lock(&mddev->lock); | |
db721d32 N |
3783 | oldpers = mddev->pers; |
3784 | oldpriv = mddev->private; | |
3785 | mddev->pers = pers; | |
3786 | mddev->private = priv; | |
3787 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3788 | mddev->level = mddev->new_level; | |
3789 | mddev->layout = mddev->new_layout; | |
3790 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3791 | mddev->delta_disks = 0; | |
3792 | mddev->reshape_backwards = 0; | |
3793 | mddev->degraded = 0; | |
36d091f4 | 3794 | spin_unlock(&mddev->lock); |
db721d32 N |
3795 | |
3796 | if (oldpers->sync_request == NULL && | |
3797 | mddev->external) { | |
3798 | /* We are converting from a no-redundancy array | |
3799 | * to a redundancy array and metadata is managed | |
3800 | * externally so we need to be sure that writes | |
3801 | * won't block due to a need to transition | |
3802 | * clean->dirty | |
3803 | * until external management is started. | |
3804 | */ | |
3805 | mddev->in_sync = 0; | |
3806 | mddev->safemode_delay = 0; | |
3807 | mddev->safemode = 0; | |
3808 | } | |
f72ffdd6 | 3809 | |
db721d32 N |
3810 | oldpers->free(mddev, oldpriv); |
3811 | ||
3812 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3813 | pers->sync_request != NULL) { |
3814 | /* need to add the md_redundancy_group */ | |
3815 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
3816 | pr_warn("md: cannot register extra attributes for %s\n", |
3817 | mdname(mddev)); | |
388975cc | 3818 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3819 | } |
db721d32 | 3820 | if (oldpers->sync_request != NULL && |
a64c876f N |
3821 | pers->sync_request == NULL) { |
3822 | /* need to remove the md_redundancy_group */ | |
3823 | if (mddev->to_remove == NULL) | |
3824 | mddev->to_remove = &md_redundancy_group; | |
3825 | } | |
3826 | ||
4cb9da7d AO |
3827 | module_put(oldpers->owner); |
3828 | ||
dafb20fa | 3829 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3830 | if (rdev->raid_disk < 0) |
3831 | continue; | |
bf2cb0da | 3832 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3833 | rdev->new_raid_disk = -1; |
3834 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3835 | continue; | |
36fad858 | 3836 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3837 | } |
dafb20fa | 3838 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3839 | if (rdev->raid_disk < 0) |
3840 | continue; | |
3841 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3842 | continue; | |
3843 | rdev->raid_disk = rdev->new_raid_disk; | |
3844 | if (rdev->raid_disk < 0) | |
3a981b03 | 3845 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3846 | else { |
36fad858 | 3847 | if (sysfs_link_rdev(mddev, rdev)) |
9d48739e N |
3848 | pr_warn("md: cannot register rd%d for %s after level change\n", |
3849 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3850 | } |
e93f68a1 N |
3851 | } |
3852 | ||
db721d32 | 3853 | if (pers->sync_request == NULL) { |
9af204cf TM |
3854 | /* this is now an array without redundancy, so |
3855 | * it must always be in_sync | |
3856 | */ | |
3857 | mddev->in_sync = 1; | |
3858 | del_timer_sync(&mddev->safemode_timer); | |
3859 | } | |
02e5f5c0 | 3860 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3861 | pers->run(mddev); |
2953079c | 3862 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
47525e59 | 3863 | mddev_resume(mddev); |
830778a1 N |
3864 | if (!mddev->thread) |
3865 | md_update_sb(mddev, 1); | |
5cac7861 | 3866 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3867 | md_new_event(mddev); |
6791875e N |
3868 | rv = len; |
3869 | out_unlock: | |
3870 | mddev_unlock(mddev); | |
d9d166c2 N |
3871 | return rv; |
3872 | } | |
3873 | ||
3874 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3875 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3876 | |
d4dbd025 | 3877 | static ssize_t |
fd01b88c | 3878 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3879 | { |
3880 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3881 | if (mddev->reshape_position != MaxSector && |
3882 | mddev->layout != mddev->new_layout) | |
3883 | return sprintf(page, "%d (%d)\n", | |
3884 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3885 | return sprintf(page, "%d\n", mddev->layout); |
3886 | } | |
3887 | ||
3888 | static ssize_t | |
fd01b88c | 3889 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 | 3890 | { |
4c9309c0 | 3891 | unsigned int n; |
6791875e | 3892 | int err; |
d4dbd025 | 3893 | |
4c9309c0 AD |
3894 | err = kstrtouint(buf, 10, &n); |
3895 | if (err < 0) | |
3896 | return err; | |
6791875e N |
3897 | err = mddev_lock(mddev); |
3898 | if (err) | |
3899 | return err; | |
d4dbd025 | 3900 | |
b3546035 | 3901 | if (mddev->pers) { |
50ac168a | 3902 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3903 | err = -EBUSY; |
3904 | else if (mddev->ro) | |
3905 | err = -EROFS; | |
3906 | else { | |
3907 | mddev->new_layout = n; | |
3908 | err = mddev->pers->check_reshape(mddev); | |
3909 | if (err) | |
3910 | mddev->new_layout = mddev->layout; | |
597a711b | 3911 | } |
b3546035 | 3912 | } else { |
08a02ecd | 3913 | mddev->new_layout = n; |
b3546035 N |
3914 | if (mddev->reshape_position == MaxSector) |
3915 | mddev->layout = n; | |
3916 | } | |
6791875e N |
3917 | mddev_unlock(mddev); |
3918 | return err ?: len; | |
d4dbd025 N |
3919 | } |
3920 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3921 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3922 | |
eae1701f | 3923 | static ssize_t |
fd01b88c | 3924 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3925 | { |
bb636547 N |
3926 | if (mddev->raid_disks == 0) |
3927 | return 0; | |
08a02ecd N |
3928 | if (mddev->reshape_position != MaxSector && |
3929 | mddev->delta_disks != 0) | |
3930 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3931 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3932 | return sprintf(page, "%d\n", mddev->raid_disks); |
3933 | } | |
3934 | ||
fd01b88c | 3935 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3936 | |
3937 | static ssize_t | |
fd01b88c | 3938 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3939 | { |
4c9309c0 | 3940 | unsigned int n; |
6791875e | 3941 | int err; |
da943b99 | 3942 | |
4c9309c0 AD |
3943 | err = kstrtouint(buf, 10, &n); |
3944 | if (err < 0) | |
3945 | return err; | |
da943b99 | 3946 | |
6791875e N |
3947 | err = mddev_lock(mddev); |
3948 | if (err) | |
3949 | return err; | |
da943b99 | 3950 | if (mddev->pers) |
6791875e | 3951 | err = update_raid_disks(mddev, n); |
08a02ecd | 3952 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3953 | struct md_rdev *rdev; |
08a02ecd | 3954 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3955 | |
6791875e | 3956 | err = -EINVAL; |
c6563a8c N |
3957 | rdev_for_each(rdev, mddev) { |
3958 | if (olddisks < n && | |
3959 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3960 | goto out_unlock; |
c6563a8c N |
3961 | if (olddisks > n && |
3962 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3963 | goto out_unlock; |
c6563a8c | 3964 | } |
6791875e | 3965 | err = 0; |
08a02ecd N |
3966 | mddev->delta_disks = n - olddisks; |
3967 | mddev->raid_disks = n; | |
2c810cdd | 3968 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3969 | } else |
da943b99 | 3970 | mddev->raid_disks = n; |
6791875e N |
3971 | out_unlock: |
3972 | mddev_unlock(mddev); | |
3973 | return err ? err : len; | |
da943b99 N |
3974 | } |
3975 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3976 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3977 | |
3b34380a | 3978 | static ssize_t |
fd01b88c | 3979 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3980 | { |
08a02ecd | 3981 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3982 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3983 | return sprintf(page, "%d (%d)\n", | |
3984 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3985 | mddev->chunk_sectors << 9); |
3986 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3987 | } |
3988 | ||
3989 | static ssize_t | |
fd01b88c | 3990 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3991 | { |
4c9309c0 | 3992 | unsigned long n; |
6791875e | 3993 | int err; |
3b34380a | 3994 | |
4c9309c0 AD |
3995 | err = kstrtoul(buf, 10, &n); |
3996 | if (err < 0) | |
3997 | return err; | |
3b34380a | 3998 | |
6791875e N |
3999 | err = mddev_lock(mddev); |
4000 | if (err) | |
4001 | return err; | |
b3546035 | 4002 | if (mddev->pers) { |
50ac168a | 4003 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
4004 | err = -EBUSY; |
4005 | else if (mddev->ro) | |
4006 | err = -EROFS; | |
4007 | else { | |
4008 | mddev->new_chunk_sectors = n >> 9; | |
4009 | err = mddev->pers->check_reshape(mddev); | |
4010 | if (err) | |
4011 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 4012 | } |
b3546035 | 4013 | } else { |
664e7c41 | 4014 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 4015 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 4016 | mddev->chunk_sectors = n >> 9; |
b3546035 | 4017 | } |
6791875e N |
4018 | mddev_unlock(mddev); |
4019 | return err ?: len; | |
3b34380a N |
4020 | } |
4021 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 4022 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 4023 | |
a94213b1 | 4024 | static ssize_t |
fd01b88c | 4025 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 4026 | { |
d1a7c503 N |
4027 | if (mddev->recovery_cp == MaxSector) |
4028 | return sprintf(page, "none\n"); | |
a94213b1 N |
4029 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
4030 | } | |
4031 | ||
4032 | static ssize_t | |
fd01b88c | 4033 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 4034 | { |
4c9309c0 | 4035 | unsigned long long n; |
6791875e | 4036 | int err; |
4c9309c0 AD |
4037 | |
4038 | if (cmd_match(buf, "none")) | |
4039 | n = MaxSector; | |
4040 | else { | |
4041 | err = kstrtoull(buf, 10, &n); | |
4042 | if (err < 0) | |
4043 | return err; | |
4044 | if (n != (sector_t)n) | |
4045 | return -EINVAL; | |
4046 | } | |
a94213b1 | 4047 | |
6791875e N |
4048 | err = mddev_lock(mddev); |
4049 | if (err) | |
4050 | return err; | |
b098636c | 4051 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e | 4052 | err = -EBUSY; |
a94213b1 | 4053 | |
6791875e N |
4054 | if (!err) { |
4055 | mddev->recovery_cp = n; | |
4056 | if (mddev->pers) | |
2953079c | 4057 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
6791875e N |
4058 | } |
4059 | mddev_unlock(mddev); | |
4060 | return err ?: len; | |
a94213b1 N |
4061 | } |
4062 | static struct md_sysfs_entry md_resync_start = | |
750f199e N |
4063 | __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR, |
4064 | resync_start_show, resync_start_store); | |
a94213b1 | 4065 | |
9e653b63 N |
4066 | /* |
4067 | * The array state can be: | |
4068 | * | |
4069 | * clear | |
4070 | * No devices, no size, no level | |
4071 | * Equivalent to STOP_ARRAY ioctl | |
4072 | * inactive | |
4073 | * May have some settings, but array is not active | |
4074 | * all IO results in error | |
4075 | * When written, doesn't tear down array, but just stops it | |
4076 | * suspended (not supported yet) | |
4077 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 4078 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
4079 | * readonly |
4080 | * no resync can happen. no superblocks get written. | |
4081 | * write requests fail | |
4082 | * read-auto | |
4083 | * like readonly, but behaves like 'clean' on a write request. | |
4084 | * | |
4085 | * clean - no pending writes, but otherwise active. | |
4086 | * When written to inactive array, starts without resync | |
4087 | * If a write request arrives then | |
4088 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
4089 | * if not known, block and switch to write-pending | |
4090 | * If written to an active array that has pending writes, then fails. | |
4091 | * active | |
4092 | * fully active: IO and resync can be happening. | |
4093 | * When written to inactive array, starts with resync | |
4094 | * | |
4095 | * write-pending | |
4096 | * clean, but writes are blocked waiting for 'active' to be written. | |
4097 | * | |
4098 | * active-idle | |
4099 | * like active, but no writes have been seen for a while (100msec). | |
4100 | * | |
4101 | */ | |
4102 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
4103 | write_pending, active_idle, bad_word}; | |
05381954 | 4104 | static char *array_states[] = { |
9e653b63 N |
4105 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
4106 | "write-pending", "active-idle", NULL }; | |
4107 | ||
4108 | static int match_word(const char *word, char **list) | |
4109 | { | |
4110 | int n; | |
4111 | for (n=0; list[n]; n++) | |
4112 | if (cmd_match(word, list[n])) | |
4113 | break; | |
4114 | return n; | |
4115 | } | |
4116 | ||
4117 | static ssize_t | |
fd01b88c | 4118 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
4119 | { |
4120 | enum array_state st = inactive; | |
4121 | ||
4122 | if (mddev->pers) | |
4123 | switch(mddev->ro) { | |
4124 | case 1: | |
4125 | st = readonly; | |
4126 | break; | |
4127 | case 2: | |
4128 | st = read_auto; | |
4129 | break; | |
4130 | case 0: | |
55cc39f3 | 4131 | spin_lock(&mddev->lock); |
2953079c | 4132 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
e691063a | 4133 | st = write_pending; |
16f88949 TM |
4134 | else if (mddev->in_sync) |
4135 | st = clean; | |
9e653b63 N |
4136 | else if (mddev->safemode) |
4137 | st = active_idle; | |
4138 | else | |
4139 | st = active; | |
55cc39f3 | 4140 | spin_unlock(&mddev->lock); |
9e653b63 N |
4141 | } |
4142 | else { | |
4143 | if (list_empty(&mddev->disks) && | |
4144 | mddev->raid_disks == 0 && | |
58c0fed4 | 4145 | mddev->dev_sectors == 0) |
9e653b63 N |
4146 | st = clear; |
4147 | else | |
4148 | st = inactive; | |
4149 | } | |
4150 | return sprintf(page, "%s\n", array_states[st]); | |
4151 | } | |
4152 | ||
f72ffdd6 N |
4153 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
4154 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
4155 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 4156 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
4157 | |
4158 | static ssize_t | |
fd01b88c | 4159 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 4160 | { |
6497709b | 4161 | int err = 0; |
9e653b63 | 4162 | enum array_state st = match_word(buf, array_states); |
6791875e N |
4163 | |
4164 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
4165 | /* don't take reconfig_mutex when toggling between | |
4166 | * clean and active | |
4167 | */ | |
4168 | spin_lock(&mddev->lock); | |
4169 | if (st == active) { | |
4170 | restart_array(mddev); | |
2953079c | 4171 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
91a6c4ad | 4172 | md_wakeup_thread(mddev->thread); |
6791875e | 4173 | wake_up(&mddev->sb_wait); |
6791875e N |
4174 | } else /* st == clean */ { |
4175 | restart_array(mddev); | |
6497709b | 4176 | if (!set_in_sync(mddev)) |
6791875e N |
4177 | err = -EBUSY; |
4178 | } | |
573275b5 TM |
4179 | if (!err) |
4180 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
6791875e | 4181 | spin_unlock(&mddev->lock); |
c008f1d3 | 4182 | return err ?: len; |
6791875e N |
4183 | } |
4184 | err = mddev_lock(mddev); | |
4185 | if (err) | |
4186 | return err; | |
4187 | err = -EINVAL; | |
9e653b63 N |
4188 | switch(st) { |
4189 | case bad_word: | |
4190 | break; | |
4191 | case clear: | |
4192 | /* stopping an active array */ | |
a05b7ea0 | 4193 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
4194 | break; |
4195 | case inactive: | |
4196 | /* stopping an active array */ | |
90cf195d | 4197 | if (mddev->pers) |
a05b7ea0 | 4198 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 4199 | else |
e691063a | 4200 | err = 0; /* already inactive */ |
9e653b63 N |
4201 | break; |
4202 | case suspended: | |
4203 | break; /* not supported yet */ | |
4204 | case readonly: | |
4205 | if (mddev->pers) | |
a05b7ea0 | 4206 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
4207 | else { |
4208 | mddev->ro = 1; | |
648b629e | 4209 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
4210 | err = do_md_run(mddev); |
4211 | } | |
4212 | break; | |
4213 | case read_auto: | |
9e653b63 | 4214 | if (mddev->pers) { |
80268ee9 | 4215 | if (mddev->ro == 0) |
a05b7ea0 | 4216 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 4217 | else if (mddev->ro == 1) |
648b629e N |
4218 | err = restart_array(mddev); |
4219 | if (err == 0) { | |
4220 | mddev->ro = 2; | |
4221 | set_disk_ro(mddev->gendisk, 0); | |
4222 | } | |
9e653b63 N |
4223 | } else { |
4224 | mddev->ro = 2; | |
4225 | err = do_md_run(mddev); | |
4226 | } | |
4227 | break; | |
4228 | case clean: | |
4229 | if (mddev->pers) { | |
339421de SL |
4230 | err = restart_array(mddev); |
4231 | if (err) | |
4232 | break; | |
85572d7c | 4233 | spin_lock(&mddev->lock); |
6497709b | 4234 | if (!set_in_sync(mddev)) |
e691063a | 4235 | err = -EBUSY; |
85572d7c | 4236 | spin_unlock(&mddev->lock); |
5bf29597 N |
4237 | } else |
4238 | err = -EINVAL; | |
9e653b63 N |
4239 | break; |
4240 | case active: | |
4241 | if (mddev->pers) { | |
339421de SL |
4242 | err = restart_array(mddev); |
4243 | if (err) | |
4244 | break; | |
2953079c | 4245 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
9e653b63 N |
4246 | wake_up(&mddev->sb_wait); |
4247 | err = 0; | |
4248 | } else { | |
4249 | mddev->ro = 0; | |
648b629e | 4250 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
4251 | err = do_md_run(mddev); |
4252 | } | |
4253 | break; | |
4254 | case write_pending: | |
4255 | case active_idle: | |
4256 | /* these cannot be set */ | |
4257 | break; | |
4258 | } | |
6791875e N |
4259 | |
4260 | if (!err) { | |
1d23f178 N |
4261 | if (mddev->hold_active == UNTIL_IOCTL) |
4262 | mddev->hold_active = 0; | |
00bcb4ac | 4263 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4264 | } |
6791875e N |
4265 | mddev_unlock(mddev); |
4266 | return err ?: len; | |
9e653b63 | 4267 | } |
80ca3a44 | 4268 | static struct md_sysfs_entry md_array_state = |
750f199e | 4269 | __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); |
9e653b63 | 4270 | |
1e50915f | 4271 | static ssize_t |
fd01b88c | 4272 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
4273 | return sprintf(page, "%d\n", |
4274 | atomic_read(&mddev->max_corr_read_errors)); | |
4275 | } | |
4276 | ||
4277 | static ssize_t | |
fd01b88c | 4278 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f | 4279 | { |
4c9309c0 AD |
4280 | unsigned int n; |
4281 | int rv; | |
1e50915f | 4282 | |
4c9309c0 AD |
4283 | rv = kstrtouint(buf, 10, &n); |
4284 | if (rv < 0) | |
4285 | return rv; | |
4286 | atomic_set(&mddev->max_corr_read_errors, n); | |
4287 | return len; | |
1e50915f RB |
4288 | } |
4289 | ||
4290 | static struct md_sysfs_entry max_corr_read_errors = | |
4291 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
4292 | max_corrected_read_errors_store); | |
4293 | ||
6d7ff738 | 4294 | static ssize_t |
fd01b88c | 4295 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
4296 | { |
4297 | return -EINVAL; | |
4298 | } | |
4299 | ||
4300 | static ssize_t | |
fd01b88c | 4301 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
4302 | { |
4303 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
4304 | /* The new device is added to the array. | |
4305 | * If the array has a persistent superblock, we read the | |
4306 | * superblock to initialise info and check validity. | |
4307 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
4308 | * which mainly checks size. | |
4309 | */ | |
4310 | char *e; | |
4311 | int major = simple_strtoul(buf, &e, 10); | |
4312 | int minor; | |
4313 | dev_t dev; | |
3cb03002 | 4314 | struct md_rdev *rdev; |
6d7ff738 N |
4315 | int err; |
4316 | ||
4317 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
4318 | return -EINVAL; | |
4319 | minor = simple_strtoul(e+1, &e, 10); | |
4320 | if (*e && *e != '\n') | |
4321 | return -EINVAL; | |
4322 | dev = MKDEV(major, minor); | |
4323 | if (major != MAJOR(dev) || | |
4324 | minor != MINOR(dev)) | |
4325 | return -EOVERFLOW; | |
4326 | ||
6791875e N |
4327 | flush_workqueue(md_misc_wq); |
4328 | ||
4329 | err = mddev_lock(mddev); | |
4330 | if (err) | |
4331 | return err; | |
6d7ff738 N |
4332 | if (mddev->persistent) { |
4333 | rdev = md_import_device(dev, mddev->major_version, | |
4334 | mddev->minor_version); | |
4335 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4336 | struct md_rdev *rdev0 |
4337 | = list_entry(mddev->disks.next, | |
4338 | struct md_rdev, same_set); | |
6d7ff738 N |
4339 | err = super_types[mddev->major_version] |
4340 | .load_super(rdev, rdev0, mddev->minor_version); | |
4341 | if (err < 0) | |
4342 | goto out; | |
4343 | } | |
c5d79adb N |
4344 | } else if (mddev->external) |
4345 | rdev = md_import_device(dev, -2, -1); | |
4346 | else | |
6d7ff738 N |
4347 | rdev = md_import_device(dev, -1, -1); |
4348 | ||
9a8c0fa8 N |
4349 | if (IS_ERR(rdev)) { |
4350 | mddev_unlock(mddev); | |
6d7ff738 | 4351 | return PTR_ERR(rdev); |
9a8c0fa8 | 4352 | } |
6d7ff738 N |
4353 | err = bind_rdev_to_array(rdev, mddev); |
4354 | out: | |
4355 | if (err) | |
4356 | export_rdev(rdev); | |
6791875e | 4357 | mddev_unlock(mddev); |
5492c46e AO |
4358 | if (!err) |
4359 | md_new_event(mddev); | |
6d7ff738 N |
4360 | return err ? err : len; |
4361 | } | |
4362 | ||
4363 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4364 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4365 | |
9b1d1dac | 4366 | static ssize_t |
fd01b88c | 4367 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4368 | { |
4369 | char *end; | |
4370 | unsigned long chunk, end_chunk; | |
6791875e | 4371 | int err; |
9b1d1dac | 4372 | |
6791875e N |
4373 | err = mddev_lock(mddev); |
4374 | if (err) | |
4375 | return err; | |
9b1d1dac PC |
4376 | if (!mddev->bitmap) |
4377 | goto out; | |
4378 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4379 | while (*buf) { | |
4380 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4381 | if (buf == end) break; | |
4382 | if (*end == '-') { /* range */ | |
4383 | buf = end + 1; | |
4384 | end_chunk = simple_strtoul(buf, &end, 0); | |
4385 | if (buf == end) break; | |
4386 | } | |
4387 | if (*end && !isspace(*end)) break; | |
4388 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4389 | buf = skip_spaces(end); |
9b1d1dac PC |
4390 | } |
4391 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4392 | out: | |
6791875e | 4393 | mddev_unlock(mddev); |
9b1d1dac PC |
4394 | return len; |
4395 | } | |
4396 | ||
4397 | static struct md_sysfs_entry md_bitmap = | |
4398 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4399 | ||
a35b0d69 | 4400 | static ssize_t |
fd01b88c | 4401 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4402 | { |
58c0fed4 AN |
4403 | return sprintf(page, "%llu\n", |
4404 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4405 | } |
4406 | ||
fd01b88c | 4407 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4408 | |
4409 | static ssize_t | |
fd01b88c | 4410 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4411 | { |
4412 | /* If array is inactive, we can reduce the component size, but | |
4413 | * not increase it (except from 0). | |
4414 | * If array is active, we can try an on-line resize | |
4415 | */ | |
b522adcd DW |
4416 | sector_t sectors; |
4417 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4418 | |
58c0fed4 AN |
4419 | if (err < 0) |
4420 | return err; | |
6791875e N |
4421 | err = mddev_lock(mddev); |
4422 | if (err) | |
4423 | return err; | |
a35b0d69 | 4424 | if (mddev->pers) { |
58c0fed4 | 4425 | err = update_size(mddev, sectors); |
4ba1e788 XN |
4426 | if (err == 0) |
4427 | md_update_sb(mddev, 1); | |
a35b0d69 | 4428 | } else { |
58c0fed4 AN |
4429 | if (mddev->dev_sectors == 0 || |
4430 | mddev->dev_sectors > sectors) | |
4431 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4432 | else |
4433 | err = -ENOSPC; | |
4434 | } | |
6791875e | 4435 | mddev_unlock(mddev); |
a35b0d69 N |
4436 | return err ? err : len; |
4437 | } | |
4438 | ||
4439 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4440 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4441 | |
83f0d77a | 4442 | /* Metadata version. |
e691063a N |
4443 | * This is one of |
4444 | * 'none' for arrays with no metadata (good luck...) | |
4445 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4446 | * or N.M for internally known formats |
4447 | */ | |
4448 | static ssize_t | |
fd01b88c | 4449 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4450 | { |
4451 | if (mddev->persistent) | |
4452 | return sprintf(page, "%d.%d\n", | |
4453 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4454 | else if (mddev->external) |
4455 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4456 | else |
4457 | return sprintf(page, "none\n"); | |
4458 | } | |
4459 | ||
4460 | static ssize_t | |
fd01b88c | 4461 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4462 | { |
4463 | int major, minor; | |
4464 | char *e; | |
6791875e | 4465 | int err; |
ea43ddd8 N |
4466 | /* Changing the details of 'external' metadata is |
4467 | * always permitted. Otherwise there must be | |
4468 | * no devices attached to the array. | |
4469 | */ | |
6791875e N |
4470 | |
4471 | err = mddev_lock(mddev); | |
4472 | if (err) | |
4473 | return err; | |
4474 | err = -EBUSY; | |
ea43ddd8 N |
4475 | if (mddev->external && strncmp(buf, "external:", 9) == 0) |
4476 | ; | |
4477 | else if (!list_empty(&mddev->disks)) | |
6791875e | 4478 | goto out_unlock; |
8bb93aac | 4479 | |
6791875e | 4480 | err = 0; |
8bb93aac N |
4481 | if (cmd_match(buf, "none")) { |
4482 | mddev->persistent = 0; | |
e691063a N |
4483 | mddev->external = 0; |
4484 | mddev->major_version = 0; | |
4485 | mddev->minor_version = 90; | |
6791875e | 4486 | goto out_unlock; |
e691063a N |
4487 | } |
4488 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4489 | size_t namelen = len-9; |
e691063a N |
4490 | if (namelen >= sizeof(mddev->metadata_type)) |
4491 | namelen = sizeof(mddev->metadata_type)-1; | |
4492 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4493 | mddev->metadata_type[namelen] = 0; | |
4494 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4495 | mddev->metadata_type[--namelen] = 0; | |
4496 | mddev->persistent = 0; | |
4497 | mddev->external = 1; | |
8bb93aac N |
4498 | mddev->major_version = 0; |
4499 | mddev->minor_version = 90; | |
6791875e | 4500 | goto out_unlock; |
8bb93aac N |
4501 | } |
4502 | major = simple_strtoul(buf, &e, 10); | |
6791875e | 4503 | err = -EINVAL; |
8bb93aac | 4504 | if (e==buf || *e != '.') |
6791875e | 4505 | goto out_unlock; |
8bb93aac N |
4506 | buf = e+1; |
4507 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4508 | if (e==buf || (*e && *e != '\n') ) |
6791875e N |
4509 | goto out_unlock; |
4510 | err = -ENOENT; | |
50511da3 | 4511 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
6791875e | 4512 | goto out_unlock; |
8bb93aac N |
4513 | mddev->major_version = major; |
4514 | mddev->minor_version = minor; | |
4515 | mddev->persistent = 1; | |
e691063a | 4516 | mddev->external = 0; |
6791875e N |
4517 | err = 0; |
4518 | out_unlock: | |
4519 | mddev_unlock(mddev); | |
4520 | return err ?: len; | |
8bb93aac N |
4521 | } |
4522 | ||
4523 | static struct md_sysfs_entry md_metadata = | |
750f199e | 4524 | __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4525 | |
24dd469d | 4526 | static ssize_t |
fd01b88c | 4527 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4528 | { |
7eec314d | 4529 | char *type = "idle"; |
b7b17c9b N |
4530 | unsigned long recovery = mddev->recovery; |
4531 | if (test_bit(MD_RECOVERY_FROZEN, &recovery)) | |
b6a9ce68 | 4532 | type = "frozen"; |
b7b17c9b N |
4533 | else if (test_bit(MD_RECOVERY_RUNNING, &recovery) || |
4534 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) { | |
4535 | if (test_bit(MD_RECOVERY_RESHAPE, &recovery)) | |
ccfcc3c1 | 4536 | type = "reshape"; |
b7b17c9b N |
4537 | else if (test_bit(MD_RECOVERY_SYNC, &recovery)) { |
4538 | if (!test_bit(MD_RECOVERY_REQUESTED, &recovery)) | |
24dd469d | 4539 | type = "resync"; |
b7b17c9b | 4540 | else if (test_bit(MD_RECOVERY_CHECK, &recovery)) |
24dd469d N |
4541 | type = "check"; |
4542 | else | |
4543 | type = "repair"; | |
b7b17c9b | 4544 | } else if (test_bit(MD_RECOVERY_RECOVER, &recovery)) |
24dd469d | 4545 | type = "recover"; |
985ca973 N |
4546 | else if (mddev->reshape_position != MaxSector) |
4547 | type = "reshape"; | |
24dd469d N |
4548 | } |
4549 | return sprintf(page, "%s\n", type); | |
4550 | } | |
4551 | ||
4552 | static ssize_t | |
fd01b88c | 4553 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4554 | { |
7eec314d N |
4555 | if (!mddev->pers || !mddev->pers->sync_request) |
4556 | return -EINVAL; | |
4557 | ||
b6a9ce68 N |
4558 | |
4559 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
56ccc112 N |
4560 | if (cmd_match(page, "frozen")) |
4561 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4562 | else | |
4563 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
8e8e2518 N |
4564 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
4565 | mddev_lock(mddev) == 0) { | |
4566 | flush_workqueue(md_misc_wq); | |
4567 | if (mddev->sync_thread) { | |
4568 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e | 4569 | md_reap_sync_thread(mddev); |
6791875e | 4570 | } |
8e8e2518 | 4571 | mddev_unlock(mddev); |
7eec314d | 4572 | } |
312045ee | 4573 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
24dd469d | 4574 | return -EBUSY; |
72a23c21 | 4575 | else if (cmd_match(page, "resync")) |
56ccc112 | 4576 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4577 | else if (cmd_match(page, "recover")) { |
56ccc112 | 4578 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4579 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
72a23c21 | 4580 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4581 | int err; |
4582 | if (mddev->pers->start_reshape == NULL) | |
4583 | return -EINVAL; | |
6791875e N |
4584 | err = mddev_lock(mddev); |
4585 | if (!err) { | |
312045ee N |
4586 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4587 | err = -EBUSY; | |
4588 | else { | |
4589 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4590 | err = mddev->pers->start_reshape(mddev); | |
4591 | } | |
6791875e N |
4592 | mddev_unlock(mddev); |
4593 | } | |
16484bf5 N |
4594 | if (err) |
4595 | return err; | |
a99ac971 | 4596 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4597 | } else { |
bce74dac | 4598 | if (cmd_match(page, "check")) |
7eec314d | 4599 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4600 | else if (!cmd_match(page, "repair")) |
7eec314d | 4601 | return -EINVAL; |
56ccc112 | 4602 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
7eec314d N |
4603 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
4604 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4605 | } |
48c26ddc N |
4606 | if (mddev->ro == 2) { |
4607 | /* A write to sync_action is enough to justify | |
4608 | * canceling read-auto mode | |
4609 | */ | |
4610 | mddev->ro = 0; | |
4611 | md_wakeup_thread(mddev->sync_thread); | |
4612 | } | |
03c902e1 | 4613 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4614 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4615 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4616 | return len; |
4617 | } | |
4618 | ||
c4a39551 | 4619 | static struct md_sysfs_entry md_scan_mode = |
750f199e | 4620 | __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
c4a39551 JB |
4621 | |
4622 | static ssize_t | |
4623 | last_sync_action_show(struct mddev *mddev, char *page) | |
4624 | { | |
4625 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4626 | } | |
4627 | ||
4628 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4629 | ||
9d88883e | 4630 | static ssize_t |
fd01b88c | 4631 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4632 | { |
4633 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4634 | (unsigned long long) |
4635 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4636 | } |
4637 | ||
80ca3a44 | 4638 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4639 | |
88202a0c | 4640 | static ssize_t |
fd01b88c | 4641 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4642 | { |
4643 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4644 | mddev->sync_speed_min ? "local": "system"); | |
4645 | } | |
4646 | ||
4647 | static ssize_t | |
fd01b88c | 4648 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4649 | { |
4c9309c0 AD |
4650 | unsigned int min; |
4651 | int rv; | |
4652 | ||
88202a0c | 4653 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4654 | min = 0; |
4655 | } else { | |
4656 | rv = kstrtouint(buf, 10, &min); | |
4657 | if (rv < 0) | |
4658 | return rv; | |
4659 | if (min == 0) | |
4660 | return -EINVAL; | |
88202a0c | 4661 | } |
88202a0c N |
4662 | mddev->sync_speed_min = min; |
4663 | return len; | |
4664 | } | |
4665 | ||
4666 | static struct md_sysfs_entry md_sync_min = | |
4667 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4668 | ||
4669 | static ssize_t | |
fd01b88c | 4670 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4671 | { |
4672 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4673 | mddev->sync_speed_max ? "local": "system"); | |
4674 | } | |
4675 | ||
4676 | static ssize_t | |
fd01b88c | 4677 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4678 | { |
4c9309c0 AD |
4679 | unsigned int max; |
4680 | int rv; | |
4681 | ||
88202a0c | 4682 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4683 | max = 0; |
4684 | } else { | |
4685 | rv = kstrtouint(buf, 10, &max); | |
4686 | if (rv < 0) | |
4687 | return rv; | |
4688 | if (max == 0) | |
4689 | return -EINVAL; | |
88202a0c | 4690 | } |
88202a0c N |
4691 | mddev->sync_speed_max = max; |
4692 | return len; | |
4693 | } | |
4694 | ||
4695 | static struct md_sysfs_entry md_sync_max = | |
4696 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4697 | ||
d7f3d291 | 4698 | static ssize_t |
fd01b88c | 4699 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4700 | { |
4701 | return sprintf(page, "%d\n", mddev->degraded); | |
4702 | } | |
4703 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4704 | |
90b08710 | 4705 | static ssize_t |
fd01b88c | 4706 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4707 | { |
4708 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4709 | } | |
4710 | ||
4711 | static ssize_t | |
fd01b88c | 4712 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4713 | { |
4714 | long n; | |
4715 | ||
b29bebd6 | 4716 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4717 | return -EINVAL; |
4718 | ||
4719 | if (n != 0 && n != 1) | |
4720 | return -EINVAL; | |
4721 | ||
4722 | mddev->parallel_resync = n; | |
4723 | ||
4724 | if (mddev->sync_thread) | |
4725 | wake_up(&resync_wait); | |
4726 | ||
4727 | return len; | |
4728 | } | |
4729 | ||
4730 | /* force parallel resync, even with shared block devices */ | |
4731 | static struct md_sysfs_entry md_sync_force_parallel = | |
4732 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4733 | sync_force_parallel_show, sync_force_parallel_store); | |
4734 | ||
88202a0c | 4735 | static ssize_t |
fd01b88c | 4736 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4737 | { |
4738 | unsigned long resync, dt, db; | |
d1a7c503 N |
4739 | if (mddev->curr_resync == 0) |
4740 | return sprintf(page, "none\n"); | |
9687a60c AN |
4741 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4742 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4743 | if (!dt) dt++; |
9687a60c AN |
4744 | db = resync - mddev->resync_mark_cnt; |
4745 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4746 | } |
4747 | ||
80ca3a44 | 4748 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4749 | |
4750 | static ssize_t | |
fd01b88c | 4751 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4752 | { |
13ae864b | 4753 | unsigned long long max_sectors, resync; |
88202a0c | 4754 | |
acb180b0 N |
4755 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4756 | return sprintf(page, "none\n"); | |
4757 | ||
72f36d59 N |
4758 | if (mddev->curr_resync == 1 || |
4759 | mddev->curr_resync == 2) | |
4760 | return sprintf(page, "delayed\n"); | |
4761 | ||
c804cdec N |
4762 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4763 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4764 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4765 | else |
58c0fed4 | 4766 | max_sectors = mddev->dev_sectors; |
88202a0c | 4767 | |
acb180b0 | 4768 | resync = mddev->curr_resync_completed; |
13ae864b | 4769 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4770 | } |
4771 | ||
750f199e N |
4772 | static struct md_sysfs_entry md_sync_completed = |
4773 | __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL); | |
88202a0c | 4774 | |
5e96ee65 | 4775 | static ssize_t |
fd01b88c | 4776 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4777 | { |
4778 | return sprintf(page, "%llu\n", | |
4779 | (unsigned long long)mddev->resync_min); | |
4780 | } | |
4781 | static ssize_t | |
fd01b88c | 4782 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4783 | { |
4784 | unsigned long long min; | |
23da422b | 4785 | int err; |
23da422b | 4786 | |
b29bebd6 | 4787 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4788 | return -EINVAL; |
23da422b N |
4789 | |
4790 | spin_lock(&mddev->lock); | |
4791 | err = -EINVAL; | |
5e96ee65 | 4792 | if (min > mddev->resync_max) |
23da422b N |
4793 | goto out_unlock; |
4794 | ||
4795 | err = -EBUSY; | |
5e96ee65 | 4796 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4797 | goto out_unlock; |
5e96ee65 | 4798 | |
50c37b13 N |
4799 | /* Round down to multiple of 4K for safety */ |
4800 | mddev->resync_min = round_down(min, 8); | |
23da422b | 4801 | err = 0; |
5e96ee65 | 4802 | |
23da422b N |
4803 | out_unlock: |
4804 | spin_unlock(&mddev->lock); | |
4805 | return err ?: len; | |
5e96ee65 NB |
4806 | } |
4807 | ||
4808 | static struct md_sysfs_entry md_min_sync = | |
4809 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4810 | ||
c6207277 | 4811 | static ssize_t |
fd01b88c | 4812 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4813 | { |
4814 | if (mddev->resync_max == MaxSector) | |
4815 | return sprintf(page, "max\n"); | |
4816 | else | |
4817 | return sprintf(page, "%llu\n", | |
4818 | (unsigned long long)mddev->resync_max); | |
4819 | } | |
4820 | static ssize_t | |
fd01b88c | 4821 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4822 | { |
23da422b N |
4823 | int err; |
4824 | spin_lock(&mddev->lock); | |
c6207277 N |
4825 | if (strncmp(buf, "max", 3) == 0) |
4826 | mddev->resync_max = MaxSector; | |
4827 | else { | |
5e96ee65 | 4828 | unsigned long long max; |
23da422b N |
4829 | int chunk; |
4830 | ||
4831 | err = -EINVAL; | |
b29bebd6 | 4832 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4833 | goto out_unlock; |
5e96ee65 | 4834 | if (max < mddev->resync_min) |
23da422b N |
4835 | goto out_unlock; |
4836 | ||
4837 | err = -EBUSY; | |
c6207277 | 4838 | if (max < mddev->resync_max && |
4d484a4a | 4839 | mddev->ro == 0 && |
c6207277 | 4840 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4841 | goto out_unlock; |
c6207277 N |
4842 | |
4843 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4844 | chunk = mddev->chunk_sectors; |
4845 | if (chunk) { | |
2ac06c33 | 4846 | sector_t temp = max; |
23da422b N |
4847 | |
4848 | err = -EINVAL; | |
4849 | if (sector_div(temp, chunk)) | |
4850 | goto out_unlock; | |
c6207277 N |
4851 | } |
4852 | mddev->resync_max = max; | |
4853 | } | |
4854 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4855 | err = 0; |
4856 | out_unlock: | |
4857 | spin_unlock(&mddev->lock); | |
4858 | return err ?: len; | |
c6207277 N |
4859 | } |
4860 | ||
4861 | static struct md_sysfs_entry md_max_sync = | |
4862 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4863 | ||
e464eafd | 4864 | static ssize_t |
fd01b88c | 4865 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4866 | { |
4867 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4868 | } | |
4869 | ||
4870 | static ssize_t | |
fd01b88c | 4871 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4872 | { |
b03e0ccb | 4873 | unsigned long long new; |
6791875e | 4874 | int err; |
e464eafd | 4875 | |
4c9309c0 AD |
4876 | err = kstrtoull(buf, 10, &new); |
4877 | if (err < 0) | |
4878 | return err; | |
4879 | if (new != (sector_t)new) | |
e464eafd | 4880 | return -EINVAL; |
23ddff37 | 4881 | |
6791875e N |
4882 | err = mddev_lock(mddev); |
4883 | if (err) | |
4884 | return err; | |
4885 | err = -EINVAL; | |
4886 | if (mddev->pers == NULL || | |
4887 | mddev->pers->quiesce == NULL) | |
4888 | goto unlock; | |
b03e0ccb | 4889 | mddev_suspend(mddev); |
23ddff37 | 4890 | mddev->suspend_lo = new; |
b03e0ccb N |
4891 | mddev_resume(mddev); |
4892 | ||
6791875e N |
4893 | err = 0; |
4894 | unlock: | |
4895 | mddev_unlock(mddev); | |
4896 | return err ?: len; | |
e464eafd N |
4897 | } |
4898 | static struct md_sysfs_entry md_suspend_lo = | |
4899 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4900 | ||
e464eafd | 4901 | static ssize_t |
fd01b88c | 4902 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4903 | { |
4904 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4905 | } | |
4906 | ||
4907 | static ssize_t | |
fd01b88c | 4908 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4909 | { |
b03e0ccb | 4910 | unsigned long long new; |
6791875e | 4911 | int err; |
e464eafd | 4912 | |
4c9309c0 AD |
4913 | err = kstrtoull(buf, 10, &new); |
4914 | if (err < 0) | |
4915 | return err; | |
4916 | if (new != (sector_t)new) | |
e464eafd | 4917 | return -EINVAL; |
23ddff37 | 4918 | |
6791875e N |
4919 | err = mddev_lock(mddev); |
4920 | if (err) | |
4921 | return err; | |
4922 | err = -EINVAL; | |
b03e0ccb | 4923 | if (mddev->pers == NULL) |
6791875e | 4924 | goto unlock; |
b03e0ccb N |
4925 | |
4926 | mddev_suspend(mddev); | |
23ddff37 | 4927 | mddev->suspend_hi = new; |
b03e0ccb N |
4928 | mddev_resume(mddev); |
4929 | ||
6791875e N |
4930 | err = 0; |
4931 | unlock: | |
4932 | mddev_unlock(mddev); | |
4933 | return err ?: len; | |
e464eafd N |
4934 | } |
4935 | static struct md_sysfs_entry md_suspend_hi = | |
4936 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4937 | ||
08a02ecd | 4938 | static ssize_t |
fd01b88c | 4939 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4940 | { |
4941 | if (mddev->reshape_position != MaxSector) | |
4942 | return sprintf(page, "%llu\n", | |
4943 | (unsigned long long)mddev->reshape_position); | |
4944 | strcpy(page, "none\n"); | |
4945 | return 5; | |
4946 | } | |
4947 | ||
4948 | static ssize_t | |
fd01b88c | 4949 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4950 | { |
c6563a8c | 4951 | struct md_rdev *rdev; |
4c9309c0 | 4952 | unsigned long long new; |
6791875e | 4953 | int err; |
6791875e | 4954 | |
4c9309c0 AD |
4955 | err = kstrtoull(buf, 10, &new); |
4956 | if (err < 0) | |
4957 | return err; | |
4958 | if (new != (sector_t)new) | |
08a02ecd | 4959 | return -EINVAL; |
6791875e N |
4960 | err = mddev_lock(mddev); |
4961 | if (err) | |
4962 | return err; | |
4963 | err = -EBUSY; | |
4964 | if (mddev->pers) | |
4965 | goto unlock; | |
08a02ecd N |
4966 | mddev->reshape_position = new; |
4967 | mddev->delta_disks = 0; | |
2c810cdd | 4968 | mddev->reshape_backwards = 0; |
08a02ecd N |
4969 | mddev->new_level = mddev->level; |
4970 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4971 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4972 | rdev_for_each(rdev, mddev) |
4973 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4974 | err = 0; |
4975 | unlock: | |
4976 | mddev_unlock(mddev); | |
4977 | return err ?: len; | |
08a02ecd N |
4978 | } |
4979 | ||
4980 | static struct md_sysfs_entry md_reshape_position = | |
4981 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4982 | reshape_position_store); | |
4983 | ||
2c810cdd N |
4984 | static ssize_t |
4985 | reshape_direction_show(struct mddev *mddev, char *page) | |
4986 | { | |
4987 | return sprintf(page, "%s\n", | |
4988 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4989 | } | |
4990 | ||
4991 | static ssize_t | |
4992 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4993 | { | |
4994 | int backwards = 0; | |
6791875e N |
4995 | int err; |
4996 | ||
2c810cdd N |
4997 | if (cmd_match(buf, "forwards")) |
4998 | backwards = 0; | |
4999 | else if (cmd_match(buf, "backwards")) | |
5000 | backwards = 1; | |
5001 | else | |
5002 | return -EINVAL; | |
5003 | if (mddev->reshape_backwards == backwards) | |
5004 | return len; | |
5005 | ||
6791875e N |
5006 | err = mddev_lock(mddev); |
5007 | if (err) | |
5008 | return err; | |
2c810cdd N |
5009 | /* check if we are allowed to change */ |
5010 | if (mddev->delta_disks) | |
6791875e N |
5011 | err = -EBUSY; |
5012 | else if (mddev->persistent && | |
2c810cdd | 5013 | mddev->major_version == 0) |
6791875e N |
5014 | err = -EINVAL; |
5015 | else | |
5016 | mddev->reshape_backwards = backwards; | |
5017 | mddev_unlock(mddev); | |
5018 | return err ?: len; | |
2c810cdd N |
5019 | } |
5020 | ||
5021 | static struct md_sysfs_entry md_reshape_direction = | |
5022 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
5023 | reshape_direction_store); | |
5024 | ||
b522adcd | 5025 | static ssize_t |
fd01b88c | 5026 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
5027 | { |
5028 | if (mddev->external_size) | |
5029 | return sprintf(page, "%llu\n", | |
5030 | (unsigned long long)mddev->array_sectors/2); | |
5031 | else | |
5032 | return sprintf(page, "default\n"); | |
5033 | } | |
5034 | ||
5035 | static ssize_t | |
fd01b88c | 5036 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
5037 | { |
5038 | sector_t sectors; | |
6791875e N |
5039 | int err; |
5040 | ||
5041 | err = mddev_lock(mddev); | |
5042 | if (err) | |
5043 | return err; | |
b522adcd | 5044 | |
ab5a98b1 | 5045 | /* cluster raid doesn't support change array_sectors */ |
b670883b ZL |
5046 | if (mddev_is_clustered(mddev)) { |
5047 | mddev_unlock(mddev); | |
ab5a98b1 | 5048 | return -EINVAL; |
b670883b | 5049 | } |
ab5a98b1 | 5050 | |
b522adcd DW |
5051 | if (strncmp(buf, "default", 7) == 0) { |
5052 | if (mddev->pers) | |
5053 | sectors = mddev->pers->size(mddev, 0, 0); | |
5054 | else | |
5055 | sectors = mddev->array_sectors; | |
5056 | ||
5057 | mddev->external_size = 0; | |
5058 | } else { | |
5059 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
5060 | err = -EINVAL; |
5061 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
5062 | err = -E2BIG; | |
5063 | else | |
5064 | mddev->external_size = 1; | |
b522adcd DW |
5065 | } |
5066 | ||
6791875e N |
5067 | if (!err) { |
5068 | mddev->array_sectors = sectors; | |
5069 | if (mddev->pers) { | |
5070 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
5071 | revalidate_disk(mddev->gendisk); | |
5072 | } | |
cbe6ef1d | 5073 | } |
6791875e N |
5074 | mddev_unlock(mddev); |
5075 | return err ?: len; | |
b522adcd DW |
5076 | } |
5077 | ||
5078 | static struct md_sysfs_entry md_array_size = | |
5079 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
5080 | array_size_store); | |
e464eafd | 5081 | |
664aed04 AP |
5082 | static ssize_t |
5083 | consistency_policy_show(struct mddev *mddev, char *page) | |
5084 | { | |
5085 | int ret; | |
5086 | ||
5087 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { | |
5088 | ret = sprintf(page, "journal\n"); | |
5089 | } else if (test_bit(MD_HAS_PPL, &mddev->flags)) { | |
5090 | ret = sprintf(page, "ppl\n"); | |
5091 | } else if (mddev->bitmap) { | |
5092 | ret = sprintf(page, "bitmap\n"); | |
5093 | } else if (mddev->pers) { | |
5094 | if (mddev->pers->sync_request) | |
5095 | ret = sprintf(page, "resync\n"); | |
5096 | else | |
5097 | ret = sprintf(page, "none\n"); | |
5098 | } else { | |
5099 | ret = sprintf(page, "unknown\n"); | |
5100 | } | |
5101 | ||
5102 | return ret; | |
5103 | } | |
5104 | ||
5105 | static ssize_t | |
5106 | consistency_policy_store(struct mddev *mddev, const char *buf, size_t len) | |
5107 | { | |
ba903a3e AP |
5108 | int err = 0; |
5109 | ||
664aed04 | 5110 | if (mddev->pers) { |
ba903a3e AP |
5111 | if (mddev->pers->change_consistency_policy) |
5112 | err = mddev->pers->change_consistency_policy(mddev, buf); | |
5113 | else | |
5114 | err = -EBUSY; | |
664aed04 AP |
5115 | } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) { |
5116 | set_bit(MD_HAS_PPL, &mddev->flags); | |
664aed04 | 5117 | } else { |
ba903a3e | 5118 | err = -EINVAL; |
664aed04 | 5119 | } |
ba903a3e AP |
5120 | |
5121 | return err ? err : len; | |
664aed04 AP |
5122 | } |
5123 | ||
5124 | static struct md_sysfs_entry md_consistency_policy = | |
5125 | __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show, | |
5126 | consistency_policy_store); | |
5127 | ||
eae1701f N |
5128 | static struct attribute *md_default_attrs[] = { |
5129 | &md_level.attr, | |
d4dbd025 | 5130 | &md_layout.attr, |
eae1701f | 5131 | &md_raid_disks.attr, |
3b34380a | 5132 | &md_chunk_size.attr, |
a35b0d69 | 5133 | &md_size.attr, |
a94213b1 | 5134 | &md_resync_start.attr, |
8bb93aac | 5135 | &md_metadata.attr, |
6d7ff738 | 5136 | &md_new_device.attr, |
16f17b39 | 5137 | &md_safe_delay.attr, |
9e653b63 | 5138 | &md_array_state.attr, |
08a02ecd | 5139 | &md_reshape_position.attr, |
2c810cdd | 5140 | &md_reshape_direction.attr, |
b522adcd | 5141 | &md_array_size.attr, |
1e50915f | 5142 | &max_corr_read_errors.attr, |
664aed04 | 5143 | &md_consistency_policy.attr, |
411036fa N |
5144 | NULL, |
5145 | }; | |
5146 | ||
5147 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 5148 | &md_scan_mode.attr, |
c4a39551 | 5149 | &md_last_scan_mode.attr, |
9d88883e | 5150 | &md_mismatches.attr, |
88202a0c N |
5151 | &md_sync_min.attr, |
5152 | &md_sync_max.attr, | |
5153 | &md_sync_speed.attr, | |
90b08710 | 5154 | &md_sync_force_parallel.attr, |
88202a0c | 5155 | &md_sync_completed.attr, |
5e96ee65 | 5156 | &md_min_sync.attr, |
c6207277 | 5157 | &md_max_sync.attr, |
e464eafd N |
5158 | &md_suspend_lo.attr, |
5159 | &md_suspend_hi.attr, | |
9b1d1dac | 5160 | &md_bitmap.attr, |
d7f3d291 | 5161 | &md_degraded.attr, |
eae1701f N |
5162 | NULL, |
5163 | }; | |
411036fa N |
5164 | static struct attribute_group md_redundancy_group = { |
5165 | .name = NULL, | |
5166 | .attrs = md_redundancy_attrs, | |
5167 | }; | |
5168 | ||
eae1701f N |
5169 | static ssize_t |
5170 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
5171 | { | |
5172 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 5173 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 5174 | ssize_t rv; |
eae1701f N |
5175 | |
5176 | if (!entry->show) | |
5177 | return -EIO; | |
af8a2434 N |
5178 | spin_lock(&all_mddevs_lock); |
5179 | if (list_empty(&mddev->all_mddevs)) { | |
5180 | spin_unlock(&all_mddevs_lock); | |
5181 | return -EBUSY; | |
5182 | } | |
5183 | mddev_get(mddev); | |
5184 | spin_unlock(&all_mddevs_lock); | |
5185 | ||
b7b17c9b | 5186 | rv = entry->show(mddev, page); |
af8a2434 | 5187 | mddev_put(mddev); |
96de1e66 | 5188 | return rv; |
eae1701f N |
5189 | } |
5190 | ||
5191 | static ssize_t | |
5192 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
5193 | const char *page, size_t length) | |
5194 | { | |
5195 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 5196 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 5197 | ssize_t rv; |
eae1701f N |
5198 | |
5199 | if (!entry->store) | |
5200 | return -EIO; | |
67463acb N |
5201 | if (!capable(CAP_SYS_ADMIN)) |
5202 | return -EACCES; | |
af8a2434 N |
5203 | spin_lock(&all_mddevs_lock); |
5204 | if (list_empty(&mddev->all_mddevs)) { | |
5205 | spin_unlock(&all_mddevs_lock); | |
5206 | return -EBUSY; | |
5207 | } | |
5208 | mddev_get(mddev); | |
5209 | spin_unlock(&all_mddevs_lock); | |
6791875e | 5210 | rv = entry->store(mddev, page, length); |
af8a2434 | 5211 | mddev_put(mddev); |
96de1e66 | 5212 | return rv; |
eae1701f N |
5213 | } |
5214 | ||
5215 | static void md_free(struct kobject *ko) | |
5216 | { | |
fd01b88c | 5217 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
5218 | |
5219 | if (mddev->sysfs_state) | |
5220 | sysfs_put(mddev->sysfs_state); | |
5221 | ||
d8115c35 BVA |
5222 | if (mddev->gendisk) |
5223 | del_gendisk(mddev->gendisk); | |
6cd18e71 N |
5224 | if (mddev->queue) |
5225 | blk_cleanup_queue(mddev->queue); | |
d8115c35 | 5226 | if (mddev->gendisk) |
a21d1504 | 5227 | put_disk(mddev->gendisk); |
4ad23a97 | 5228 | percpu_ref_exit(&mddev->writes_pending); |
a21d1504 | 5229 | |
28dec870 KO |
5230 | bioset_exit(&mddev->bio_set); |
5231 | bioset_exit(&mddev->sync_set); | |
eae1701f N |
5232 | kfree(mddev); |
5233 | } | |
5234 | ||
52cf25d0 | 5235 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
5236 | .show = md_attr_show, |
5237 | .store = md_attr_store, | |
5238 | }; | |
5239 | static struct kobj_type md_ktype = { | |
5240 | .release = md_free, | |
5241 | .sysfs_ops = &md_sysfs_ops, | |
5242 | .default_attrs = md_default_attrs, | |
5243 | }; | |
5244 | ||
1da177e4 LT |
5245 | int mdp_major = 0; |
5246 | ||
5fd3a17e DW |
5247 | static void mddev_delayed_delete(struct work_struct *ws) |
5248 | { | |
fd01b88c | 5249 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 5250 | |
43a70507 | 5251 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
5252 | kobject_del(&mddev->kobj); |
5253 | kobject_put(&mddev->kobj); | |
5254 | } | |
5255 | ||
4ad23a97 N |
5256 | static void no_op(struct percpu_ref *r) {} |
5257 | ||
a415c0f1 N |
5258 | int mddev_init_writes_pending(struct mddev *mddev) |
5259 | { | |
5260 | if (mddev->writes_pending.percpu_count_ptr) | |
5261 | return 0; | |
5262 | if (percpu_ref_init(&mddev->writes_pending, no_op, 0, GFP_KERNEL) < 0) | |
5263 | return -ENOMEM; | |
5264 | /* We want to start with the refcount at zero */ | |
5265 | percpu_ref_put(&mddev->writes_pending); | |
5266 | return 0; | |
5267 | } | |
5268 | EXPORT_SYMBOL_GPL(mddev_init_writes_pending); | |
5269 | ||
efeb53c0 | 5270 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 5271 | { |
039b7225 N |
5272 | /* |
5273 | * If dev is zero, name is the name of a device to allocate with | |
5274 | * an arbitrary minor number. It will be "md_???" | |
5275 | * If dev is non-zero it must be a device number with a MAJOR of | |
5276 | * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then | |
5277 | * the device is being created by opening a node in /dev. | |
5278 | * If "name" is not NULL, the device is being created by | |
5279 | * writing to /sys/module/md_mod/parameters/new_array. | |
5280 | */ | |
48c9c27b | 5281 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 5282 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 5283 | struct gendisk *disk; |
efeb53c0 N |
5284 | int partitioned; |
5285 | int shift; | |
5286 | int unit; | |
3830c62f | 5287 | int error; |
1da177e4 LT |
5288 | |
5289 | if (!mddev) | |
efeb53c0 N |
5290 | return -ENODEV; |
5291 | ||
5292 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
5293 | shift = partitioned ? MdpMinorShift : 0; | |
5294 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 5295 | |
e804ac78 TH |
5296 | /* wait for any previous instance of this device to be |
5297 | * completely removed (mddev_delayed_delete). | |
d3374825 | 5298 | */ |
e804ac78 | 5299 | flush_workqueue(md_misc_wq); |
d3374825 | 5300 | |
48c9c27b | 5301 | mutex_lock(&disks_mutex); |
0909dc44 N |
5302 | error = -EEXIST; |
5303 | if (mddev->gendisk) | |
5304 | goto abort; | |
efeb53c0 | 5305 | |
039b7225 | 5306 | if (name && !dev) { |
efeb53c0 N |
5307 | /* Need to ensure that 'name' is not a duplicate. |
5308 | */ | |
fd01b88c | 5309 | struct mddev *mddev2; |
efeb53c0 N |
5310 | spin_lock(&all_mddevs_lock); |
5311 | ||
5312 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
5313 | if (mddev2->gendisk && | |
5314 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
5315 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 5316 | goto abort; |
efeb53c0 N |
5317 | } |
5318 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 5319 | } |
039b7225 N |
5320 | if (name && dev) |
5321 | /* | |
5322 | * Creating /dev/mdNNN via "newarray", so adjust hold_active. | |
5323 | */ | |
5324 | mddev->hold_active = UNTIL_STOP; | |
8b765398 | 5325 | |
0909dc44 | 5326 | error = -ENOMEM; |
8b765398 | 5327 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
5328 | if (!mddev->queue) |
5329 | goto abort; | |
409c57f3 N |
5330 | mddev->queue->queuedata = mddev; |
5331 | ||
409c57f3 | 5332 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 5333 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 5334 | |
1da177e4 LT |
5335 | disk = alloc_disk(1 << shift); |
5336 | if (!disk) { | |
8b765398 N |
5337 | blk_cleanup_queue(mddev->queue); |
5338 | mddev->queue = NULL; | |
0909dc44 | 5339 | goto abort; |
1da177e4 | 5340 | } |
efeb53c0 | 5341 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 5342 | disk->first_minor = unit << shift; |
efeb53c0 N |
5343 | if (name) |
5344 | strcpy(disk->disk_name, name); | |
5345 | else if (partitioned) | |
1da177e4 | 5346 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 5347 | else |
1da177e4 | 5348 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
5349 | disk->fops = &md_fops; |
5350 | disk->private_data = mddev; | |
5351 | disk->queue = mddev->queue; | |
56883a7e | 5352 | blk_queue_write_cache(mddev->queue, true, true); |
92850bbd | 5353 | /* Allow extended partitions. This makes the |
d3374825 | 5354 | * 'mdp' device redundant, but we can't really |
92850bbd N |
5355 | * remove it now. |
5356 | */ | |
5357 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 5358 | mddev->gendisk = disk; |
b0140891 N |
5359 | /* As soon as we call add_disk(), another thread could get |
5360 | * through to md_open, so make sure it doesn't get too far | |
5361 | */ | |
5362 | mutex_lock(&mddev->open_mutex); | |
5363 | add_disk(disk); | |
5364 | ||
28dec870 | 5365 | error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md"); |
0909dc44 N |
5366 | if (error) { |
5367 | /* This isn't possible, but as kobject_init_and_add is marked | |
5368 | * __must_check, we must do something with the result | |
5369 | */ | |
9d48739e N |
5370 | pr_debug("md: cannot register %s/md - name in use\n", |
5371 | disk->disk_name); | |
0909dc44 N |
5372 | error = 0; |
5373 | } | |
00bcb4ac N |
5374 | if (mddev->kobj.sd && |
5375 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
9d48739e | 5376 | pr_debug("pointless warning\n"); |
b0140891 | 5377 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
5378 | abort: |
5379 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 5380 | if (!error && mddev->kobj.sd) { |
3830c62f | 5381 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 5382 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 5383 | } |
d3374825 | 5384 | mddev_put(mddev); |
0909dc44 | 5385 | return error; |
efeb53c0 N |
5386 | } |
5387 | ||
5388 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
5389 | { | |
78b6350d N |
5390 | if (create_on_open) |
5391 | md_alloc(dev, NULL); | |
1da177e4 LT |
5392 | return NULL; |
5393 | } | |
5394 | ||
e4dca7b7 | 5395 | static int add_named_array(const char *val, const struct kernel_param *kp) |
efeb53c0 | 5396 | { |
039b7225 N |
5397 | /* |
5398 | * val must be "md_*" or "mdNNN". | |
5399 | * For "md_*" we allocate an array with a large free minor number, and | |
efeb53c0 | 5400 | * set the name to val. val must not already be an active name. |
039b7225 N |
5401 | * For "mdNNN" we allocate an array with the minor number NNN |
5402 | * which must not already be in use. | |
efeb53c0 N |
5403 | */ |
5404 | int len = strlen(val); | |
5405 | char buf[DISK_NAME_LEN]; | |
039b7225 | 5406 | unsigned long devnum; |
efeb53c0 N |
5407 | |
5408 | while (len && val[len-1] == '\n') | |
5409 | len--; | |
5410 | if (len >= DISK_NAME_LEN) | |
5411 | return -E2BIG; | |
5412 | strlcpy(buf, val, len+1); | |
039b7225 N |
5413 | if (strncmp(buf, "md_", 3) == 0) |
5414 | return md_alloc(0, buf); | |
5415 | if (strncmp(buf, "md", 2) == 0 && | |
5416 | isdigit(buf[2]) && | |
5417 | kstrtoul(buf+2, 10, &devnum) == 0 && | |
5418 | devnum <= MINORMASK) | |
5419 | return md_alloc(MKDEV(MD_MAJOR, devnum), NULL); | |
5420 | ||
5421 | return -EINVAL; | |
efeb53c0 N |
5422 | } |
5423 | ||
8376d3c1 | 5424 | static void md_safemode_timeout(struct timer_list *t) |
1da177e4 | 5425 | { |
8376d3c1 | 5426 | struct mddev *mddev = from_timer(mddev, t, safemode_timer); |
1da177e4 | 5427 | |
4ad23a97 N |
5428 | mddev->safemode = 1; |
5429 | if (mddev->external) | |
5430 | sysfs_notify_dirent_safe(mddev->sysfs_state); | |
5431 | ||
1da177e4 LT |
5432 | md_wakeup_thread(mddev->thread); |
5433 | } | |
5434 | ||
6ff8d8ec | 5435 | static int start_dirty_degraded; |
1da177e4 | 5436 | |
fd01b88c | 5437 | int md_run(struct mddev *mddev) |
1da177e4 | 5438 | { |
2604b703 | 5439 | int err; |
3cb03002 | 5440 | struct md_rdev *rdev; |
84fc4b56 | 5441 | struct md_personality *pers; |
1da177e4 | 5442 | |
a757e64c N |
5443 | if (list_empty(&mddev->disks)) |
5444 | /* cannot run an array with no devices.. */ | |
1da177e4 | 5445 | return -EINVAL; |
1da177e4 LT |
5446 | |
5447 | if (mddev->pers) | |
5448 | return -EBUSY; | |
bb4f1e9d N |
5449 | /* Cannot run until previous stop completes properly */ |
5450 | if (mddev->sysfs_active) | |
5451 | return -EBUSY; | |
b6eb127d | 5452 | |
1da177e4 LT |
5453 | /* |
5454 | * Analyze all RAID superblock(s) | |
5455 | */ | |
1ec4a939 N |
5456 | if (!mddev->raid_disks) { |
5457 | if (!mddev->persistent) | |
5458 | return -EINVAL; | |
a757e64c | 5459 | analyze_sbs(mddev); |
1ec4a939 | 5460 | } |
1da177e4 | 5461 | |
d9d166c2 N |
5462 | if (mddev->level != LEVEL_NONE) |
5463 | request_module("md-level-%d", mddev->level); | |
5464 | else if (mddev->clevel[0]) | |
5465 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
5466 | |
5467 | /* | |
5468 | * Drop all container device buffers, from now on | |
5469 | * the only valid external interface is through the md | |
5470 | * device. | |
1da177e4 | 5471 | */ |
4b6c1060 | 5472 | mddev->has_superblocks = false; |
dafb20fa | 5473 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 5474 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5475 | continue; |
5476 | sync_blockdev(rdev->bdev); | |
f98393a6 | 5477 | invalidate_bdev(rdev->bdev); |
97b20ef7 N |
5478 | if (mddev->ro != 1 && |
5479 | (bdev_read_only(rdev->bdev) || | |
5480 | bdev_read_only(rdev->meta_bdev))) { | |
5481 | mddev->ro = 1; | |
5482 | if (mddev->gendisk) | |
5483 | set_disk_ro(mddev->gendisk, 1); | |
5484 | } | |
f0d76d70 | 5485 | |
4b6c1060 HM |
5486 | if (rdev->sb_page) |
5487 | mddev->has_superblocks = true; | |
5488 | ||
f0d76d70 N |
5489 | /* perform some consistency tests on the device. |
5490 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 5491 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 5492 | */ |
a6ff7e08 JB |
5493 | if (rdev->meta_bdev) { |
5494 | /* Nothing to check */; | |
5495 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5496 | if (mddev->dev_sectors && |
5497 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5498 | > rdev->sb_start) { |
9d48739e N |
5499 | pr_warn("md: %s: data overlaps metadata\n", |
5500 | mdname(mddev)); | |
f0d76d70 N |
5501 | return -EINVAL; |
5502 | } | |
5503 | } else { | |
0f420358 | 5504 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 | 5505 | > rdev->data_offset) { |
9d48739e N |
5506 | pr_warn("md: %s: metadata overlaps data\n", |
5507 | mdname(mddev)); | |
f0d76d70 N |
5508 | return -EINVAL; |
5509 | } | |
5510 | } | |
00bcb4ac | 5511 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5512 | } |
5513 | ||
afeee514 KO |
5514 | if (!bioset_initialized(&mddev->bio_set)) { |
5515 | err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); | |
5516 | if (err) | |
5517 | return err; | |
10273170 | 5518 | } |
afeee514 KO |
5519 | if (!bioset_initialized(&mddev->sync_set)) { |
5520 | err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); | |
5521 | if (err) | |
28dec870 | 5522 | return err; |
5a85071c | 5523 | } |
5a409b4f XN |
5524 | if (mddev->flush_pool == NULL) { |
5525 | mddev->flush_pool = mempool_create(NR_FLUSH_INFOS, flush_info_alloc, | |
5526 | flush_info_free, mddev); | |
5527 | if (!mddev->flush_pool) { | |
5528 | err = -ENOMEM; | |
5529 | goto abort; | |
5530 | } | |
5531 | } | |
5532 | if (mddev->flush_bio_pool == NULL) { | |
5533 | mddev->flush_bio_pool = mempool_create(NR_FLUSH_BIOS, flush_bio_alloc, | |
5534 | flush_bio_free, mddev); | |
5535 | if (!mddev->flush_bio_pool) { | |
5536 | err = -ENOMEM; | |
5537 | goto abort; | |
5538 | } | |
5539 | } | |
a167f663 | 5540 | |
1da177e4 | 5541 | spin_lock(&pers_lock); |
d9d166c2 | 5542 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5543 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5544 | spin_unlock(&pers_lock); |
d9d166c2 | 5545 | if (mddev->level != LEVEL_NONE) |
9d48739e N |
5546 | pr_warn("md: personality for level %d is not loaded!\n", |
5547 | mddev->level); | |
d9d166c2 | 5548 | else |
9d48739e N |
5549 | pr_warn("md: personality for level %s is not loaded!\n", |
5550 | mddev->clevel); | |
bfc9dfdc SL |
5551 | err = -EINVAL; |
5552 | goto abort; | |
1da177e4 | 5553 | } |
1da177e4 | 5554 | spin_unlock(&pers_lock); |
34817e8c N |
5555 | if (mddev->level != pers->level) { |
5556 | mddev->level = pers->level; | |
5557 | mddev->new_level = pers->level; | |
5558 | } | |
d9d166c2 | 5559 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5560 | |
f6705578 | 5561 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5562 | pers->start_reshape == NULL) { |
f6705578 | 5563 | /* This personality cannot handle reshaping... */ |
f6705578 | 5564 | module_put(pers->owner); |
bfc9dfdc SL |
5565 | err = -EINVAL; |
5566 | goto abort; | |
f6705578 N |
5567 | } |
5568 | ||
7dd5e7c3 N |
5569 | if (pers->sync_request) { |
5570 | /* Warn if this is a potentially silly | |
5571 | * configuration. | |
5572 | */ | |
5573 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5574 | struct md_rdev *rdev2; |
7dd5e7c3 | 5575 | int warned = 0; |
159ec1fc | 5576 | |
dafb20fa N |
5577 | rdev_for_each(rdev, mddev) |
5578 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5579 | if (rdev < rdev2 && |
5580 | rdev->bdev->bd_contains == | |
5581 | rdev2->bdev->bd_contains) { | |
9d48739e N |
5582 | pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n", |
5583 | mdname(mddev), | |
5584 | bdevname(rdev->bdev,b), | |
5585 | bdevname(rdev2->bdev,b2)); | |
7dd5e7c3 N |
5586 | warned = 1; |
5587 | } | |
5588 | } | |
159ec1fc | 5589 | |
7dd5e7c3 | 5590 | if (warned) |
9d48739e | 5591 | pr_warn("True protection against single-disk failure might be compromised.\n"); |
7dd5e7c3 N |
5592 | } |
5593 | ||
657390d2 | 5594 | mddev->recovery = 0; |
58c0fed4 AN |
5595 | /* may be over-ridden by personality */ |
5596 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5597 | ||
6ff8d8ec | 5598 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5599 | |
0f9552b5 | 5600 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5601 | mddev->ro = 2; /* read-only, but switch on first write */ |
5602 | ||
36d091f4 | 5603 | err = pers->run(mddev); |
13e53df3 | 5604 | if (err) |
9d48739e | 5605 | pr_warn("md: pers->run() failed ...\n"); |
36d091f4 | 5606 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
9d48739e N |
5607 | WARN_ONCE(!mddev->external_size, |
5608 | "%s: default size too small, but 'external_size' not in effect?\n", | |
5609 | __func__); | |
5610 | pr_warn("md: invalid array_size %llu > default size %llu\n", | |
5611 | (unsigned long long)mddev->array_sectors / 2, | |
5612 | (unsigned long long)pers->size(mddev, 0, 0) / 2); | |
b522adcd | 5613 | err = -EINVAL; |
b522adcd | 5614 | } |
36d091f4 | 5615 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5616 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
f9209a32 GR |
5617 | struct bitmap *bitmap; |
5618 | ||
5619 | bitmap = bitmap_create(mddev, -1); | |
5620 | if (IS_ERR(bitmap)) { | |
5621 | err = PTR_ERR(bitmap); | |
9d48739e N |
5622 | pr_warn("%s: failed to create bitmap (%d)\n", |
5623 | mdname(mddev), err); | |
f9209a32 GR |
5624 | } else |
5625 | mddev->bitmap = bitmap; | |
5626 | ||
b15c2e57 | 5627 | } |
1da177e4 | 5628 | if (err) { |
5aa61f42 | 5629 | mddev_detach(mddev); |
0c35bd47 N |
5630 | if (mddev->private) |
5631 | pers->free(mddev, mddev->private); | |
bd691922 | 5632 | mddev->private = NULL; |
36d091f4 | 5633 | module_put(pers->owner); |
32a7627c | 5634 | bitmap_destroy(mddev); |
bfc9dfdc | 5635 | goto abort; |
1da177e4 | 5636 | } |
5c675f83 | 5637 | if (mddev->queue) { |
bb086a89 SL |
5638 | bool nonrot = true; |
5639 | ||
5640 | rdev_for_each(rdev, mddev) { | |
5641 | if (rdev->raid_disk >= 0 && | |
5642 | !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
5643 | nonrot = false; | |
5644 | break; | |
5645 | } | |
5646 | } | |
5647 | if (mddev->degraded) | |
5648 | nonrot = false; | |
5649 | if (nonrot) | |
8b904b5b | 5650 | blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue); |
bb086a89 | 5651 | else |
8b904b5b | 5652 | blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue); |
dc3b17cc JK |
5653 | mddev->queue->backing_dev_info->congested_data = mddev; |
5654 | mddev->queue->backing_dev_info->congested_fn = md_congested; | |
5c675f83 | 5655 | } |
36d091f4 | 5656 | if (pers->sync_request) { |
00bcb4ac N |
5657 | if (mddev->kobj.sd && |
5658 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
9d48739e N |
5659 | pr_warn("md: cannot register extra attributes for %s\n", |
5660 | mdname(mddev)); | |
00bcb4ac | 5661 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5662 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5663 | mddev->ro = 0; |
5664 | ||
1e50915f RB |
5665 | atomic_set(&mddev->max_corr_read_errors, |
5666 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 | 5667 | mddev->safemode = 0; |
28c1b9fd GR |
5668 | if (mddev_is_clustered(mddev)) |
5669 | mddev->safemode_delay = 0; | |
5670 | else | |
5671 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ | |
1da177e4 | 5672 | mddev->in_sync = 1; |
0ca69886 | 5673 | smp_wmb(); |
36d091f4 N |
5674 | spin_lock(&mddev->lock); |
5675 | mddev->pers = pers; | |
36d091f4 | 5676 | spin_unlock(&mddev->lock); |
dafb20fa | 5677 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5678 | if (rdev->raid_disk >= 0) |
5679 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5680 | /* failure here is OK */; |
f72ffdd6 | 5681 | |
a4a3d26d N |
5682 | if (mddev->degraded && !mddev->ro) |
5683 | /* This ensures that recovering status is reported immediately | |
5684 | * via sysfs - until a lack of spares is confirmed. | |
5685 | */ | |
5686 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
1da177e4 | 5687 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5688 | |
2953079c | 5689 | if (mddev->sb_flags) |
850b2b42 | 5690 | md_update_sb(mddev, 0); |
1da177e4 | 5691 | |
d7603b7e | 5692 | md_new_event(mddev); |
00bcb4ac N |
5693 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5694 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5695 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 | 5696 | return 0; |
b126194c XN |
5697 | |
5698 | abort: | |
5a409b4f XN |
5699 | if (mddev->flush_bio_pool) { |
5700 | mempool_destroy(mddev->flush_bio_pool); | |
5701 | mddev->flush_bio_pool = NULL; | |
5702 | } | |
5703 | if (mddev->flush_pool){ | |
5704 | mempool_destroy(mddev->flush_pool); | |
5705 | mddev->flush_pool = NULL; | |
5706 | } | |
b126194c XN |
5707 | |
5708 | return err; | |
1da177e4 | 5709 | } |
390ee602 | 5710 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5711 | |
fd01b88c | 5712 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5713 | { |
5714 | int err; | |
5715 | ||
5716 | err = md_run(mddev); | |
5717 | if (err) | |
5718 | goto out; | |
69e51b44 N |
5719 | err = bitmap_load(mddev); |
5720 | if (err) { | |
5721 | bitmap_destroy(mddev); | |
5722 | goto out; | |
5723 | } | |
0fd018af | 5724 | |
28c1b9fd GR |
5725 | if (mddev_is_clustered(mddev)) |
5726 | md_allow_write(mddev); | |
5727 | ||
d5d885fd SL |
5728 | /* run start up tasks that require md_thread */ |
5729 | md_start(mddev); | |
5730 | ||
0fd018af JB |
5731 | md_wakeup_thread(mddev->thread); |
5732 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5733 | ||
fe60b014 N |
5734 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5735 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5736 | mddev->changed = 1; |
fe60b014 N |
5737 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5738 | out: | |
5739 | return err; | |
5740 | } | |
5741 | ||
d5d885fd SL |
5742 | int md_start(struct mddev *mddev) |
5743 | { | |
5744 | int ret = 0; | |
5745 | ||
5746 | if (mddev->pers->start) { | |
5747 | set_bit(MD_RECOVERY_WAIT, &mddev->recovery); | |
5748 | md_wakeup_thread(mddev->thread); | |
5749 | ret = mddev->pers->start(mddev); | |
5750 | clear_bit(MD_RECOVERY_WAIT, &mddev->recovery); | |
5751 | md_wakeup_thread(mddev->sync_thread); | |
5752 | } | |
5753 | return ret; | |
5754 | } | |
5755 | EXPORT_SYMBOL_GPL(md_start); | |
5756 | ||
fd01b88c | 5757 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5758 | { |
5759 | struct gendisk *disk = mddev->gendisk; | |
97b20ef7 N |
5760 | struct md_rdev *rdev; |
5761 | bool has_journal = false; | |
5762 | bool has_readonly = false; | |
1da177e4 | 5763 | |
80fab1d7 | 5764 | /* Complain if it has no devices */ |
1da177e4 | 5765 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5766 | return -ENXIO; |
5767 | if (!mddev->pers) | |
5768 | return -EINVAL; | |
5769 | if (!mddev->ro) | |
5770 | return -EBUSY; | |
339421de | 5771 | |
97b20ef7 N |
5772 | rcu_read_lock(); |
5773 | rdev_for_each_rcu(rdev, mddev) { | |
5774 | if (test_bit(Journal, &rdev->flags) && | |
5775 | !test_bit(Faulty, &rdev->flags)) | |
5776 | has_journal = true; | |
5777 | if (bdev_read_only(rdev->bdev)) | |
5778 | has_readonly = true; | |
5779 | } | |
5780 | rcu_read_unlock(); | |
5781 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal) | |
339421de | 5782 | /* Don't restart rw with journal missing/faulty */ |
339421de | 5783 | return -EINVAL; |
97b20ef7 N |
5784 | if (has_readonly) |
5785 | return -EROFS; | |
339421de | 5786 | |
80fab1d7 AN |
5787 | mddev->safemode = 0; |
5788 | mddev->ro = 0; | |
5789 | set_disk_ro(disk, 0); | |
9d48739e | 5790 | pr_debug("md: %s switched to read-write mode.\n", mdname(mddev)); |
80fab1d7 AN |
5791 | /* Kick recovery or resync if necessary */ |
5792 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5793 | md_wakeup_thread(mddev->thread); | |
5794 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5795 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5796 | return 0; |
1da177e4 LT |
5797 | } |
5798 | ||
fd01b88c | 5799 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5800 | { |
5801 | mddev->array_sectors = 0; | |
5802 | mddev->external_size = 0; | |
5803 | mddev->dev_sectors = 0; | |
5804 | mddev->raid_disks = 0; | |
5805 | mddev->recovery_cp = 0; | |
5806 | mddev->resync_min = 0; | |
5807 | mddev->resync_max = MaxSector; | |
5808 | mddev->reshape_position = MaxSector; | |
5809 | mddev->external = 0; | |
5810 | mddev->persistent = 0; | |
5811 | mddev->level = LEVEL_NONE; | |
5812 | mddev->clevel[0] = 0; | |
5813 | mddev->flags = 0; | |
2953079c | 5814 | mddev->sb_flags = 0; |
6177b472 N |
5815 | mddev->ro = 0; |
5816 | mddev->metadata_type[0] = 0; | |
5817 | mddev->chunk_sectors = 0; | |
5818 | mddev->ctime = mddev->utime = 0; | |
5819 | mddev->layout = 0; | |
5820 | mddev->max_disks = 0; | |
5821 | mddev->events = 0; | |
a8707c08 | 5822 | mddev->can_decrease_events = 0; |
6177b472 | 5823 | mddev->delta_disks = 0; |
2c810cdd | 5824 | mddev->reshape_backwards = 0; |
6177b472 N |
5825 | mddev->new_level = LEVEL_NONE; |
5826 | mddev->new_layout = 0; | |
5827 | mddev->new_chunk_sectors = 0; | |
5828 | mddev->curr_resync = 0; | |
7f7583d4 | 5829 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5830 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5831 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5832 | mddev->recovery = 0; | |
5833 | mddev->in_sync = 0; | |
f0b4f7e2 | 5834 | mddev->changed = 0; |
6177b472 | 5835 | mddev->degraded = 0; |
6177b472 | 5836 | mddev->safemode = 0; |
bd691922 | 5837 | mddev->private = NULL; |
c20c33f0 | 5838 | mddev->cluster_info = NULL; |
6177b472 N |
5839 | mddev->bitmap_info.offset = 0; |
5840 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5841 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5842 | mddev->bitmap_info.chunksize = 0; |
5843 | mddev->bitmap_info.daemon_sleep = 0; | |
5844 | mddev->bitmap_info.max_write_behind = 0; | |
c20c33f0 | 5845 | mddev->bitmap_info.nodes = 0; |
6177b472 N |
5846 | } |
5847 | ||
fd01b88c | 5848 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5849 | { |
6b6204ee | 5850 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5851 | flush_workqueue(md_misc_wq); |
a047e125 | 5852 | if (mddev->sync_thread) { |
a047e125 | 5853 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5854 | md_reap_sync_thread(mddev); |
a047e125 N |
5855 | } |
5856 | ||
5857 | del_timer_sync(&mddev->safemode_timer); | |
5858 | ||
034e33f5 SL |
5859 | if (mddev->pers && mddev->pers->quiesce) { |
5860 | mddev->pers->quiesce(mddev, 1); | |
5861 | mddev->pers->quiesce(mddev, 0); | |
5862 | } | |
a047e125 | 5863 | bitmap_flush(mddev); |
a047e125 | 5864 | |
b6d428c6 | 5865 | if (mddev->ro == 0 && |
28c1b9fd | 5866 | ((!mddev->in_sync && !mddev_is_clustered(mddev)) || |
2953079c | 5867 | mddev->sb_flags)) { |
a047e125 | 5868 | /* mark array as shutdown cleanly */ |
28c1b9fd GR |
5869 | if (!mddev_is_clustered(mddev)) |
5870 | mddev->in_sync = 1; | |
a047e125 N |
5871 | md_update_sb(mddev, 1); |
5872 | } | |
5873 | } | |
defad61a | 5874 | |
fd01b88c | 5875 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5876 | { |
29f097c4 | 5877 | mddev_lock_nointr(mddev); |
defad61a N |
5878 | __md_stop_writes(mddev); |
5879 | mddev_unlock(mddev); | |
5880 | } | |
390ee602 | 5881 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5882 | |
5aa61f42 N |
5883 | static void mddev_detach(struct mddev *mddev) |
5884 | { | |
48df498d | 5885 | bitmap_wait_behind_writes(mddev); |
36d091f4 | 5886 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5887 | mddev->pers->quiesce(mddev, 1); |
5888 | mddev->pers->quiesce(mddev, 0); | |
5889 | } | |
5890 | md_unregister_thread(&mddev->thread); | |
5891 | if (mddev->queue) | |
5892 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5893 | } | |
5894 | ||
5eff3c43 | 5895 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5896 | { |
36d091f4 | 5897 | struct md_personality *pers = mddev->pers; |
48df498d | 5898 | bitmap_destroy(mddev); |
5aa61f42 | 5899 | mddev_detach(mddev); |
ee5d004f N |
5900 | /* Ensure ->event_work is done */ |
5901 | flush_workqueue(md_misc_wq); | |
36d091f4 | 5902 | spin_lock(&mddev->lock); |
6177b472 | 5903 | mddev->pers = NULL; |
36d091f4 N |
5904 | spin_unlock(&mddev->lock); |
5905 | pers->free(mddev, mddev->private); | |
bd691922 | 5906 | mddev->private = NULL; |
36d091f4 N |
5907 | if (pers->sync_request && mddev->to_remove == NULL) |
5908 | mddev->to_remove = &md_redundancy_group; | |
5909 | module_put(pers->owner); | |
cca9cf90 | 5910 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5911 | } |
5eff3c43 N |
5912 | |
5913 | void md_stop(struct mddev *mddev) | |
5914 | { | |
5915 | /* stop the array and free an attached data structures. | |
5916 | * This is called from dm-raid | |
5917 | */ | |
5918 | __md_stop(mddev); | |
5a409b4f XN |
5919 | if (mddev->flush_bio_pool) { |
5920 | mempool_destroy(mddev->flush_bio_pool); | |
5921 | mddev->flush_bio_pool = NULL; | |
5922 | } | |
5923 | if (mddev->flush_pool) { | |
5924 | mempool_destroy(mddev->flush_pool); | |
5925 | mddev->flush_pool = NULL; | |
5926 | } | |
afeee514 KO |
5927 | bioset_exit(&mddev->bio_set); |
5928 | bioset_exit(&mddev->sync_set); | |
5eff3c43 N |
5929 | } |
5930 | ||
390ee602 | 5931 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5932 | |
a05b7ea0 | 5933 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5934 | { |
5935 | int err = 0; | |
30b8feb7 N |
5936 | int did_freeze = 0; |
5937 | ||
5938 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5939 | did_freeze = 1; | |
5940 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5941 | md_wakeup_thread(mddev->thread); | |
5942 | } | |
f851b60d | 5943 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5944 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5945 | if (mddev->sync_thread) |
30b8feb7 N |
5946 | /* Thread might be blocked waiting for metadata update |
5947 | * which will now never happen */ | |
5948 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5949 | |
2953079c | 5950 | if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
88724bfa | 5951 | return -EBUSY; |
30b8feb7 | 5952 | mddev_unlock(mddev); |
f851b60d N |
5953 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5954 | &mddev->recovery)); | |
88724bfa | 5955 | wait_event(mddev->sb_wait, |
2953079c | 5956 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
30b8feb7 N |
5957 | mddev_lock_nointr(mddev); |
5958 | ||
a4bd82d0 | 5959 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5960 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5961 | mddev->sync_thread || |
af8d8e6f | 5962 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 5963 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5964 | if (did_freeze) { |
5965 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5966 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5967 | md_wakeup_thread(mddev->thread); |
5968 | } | |
a4bd82d0 N |
5969 | err = -EBUSY; |
5970 | goto out; | |
5971 | } | |
5972 | if (mddev->pers) { | |
defad61a | 5973 | __md_stop_writes(mddev); |
a4bd82d0 N |
5974 | |
5975 | err = -ENXIO; | |
5976 | if (mddev->ro==1) | |
5977 | goto out; | |
5978 | mddev->ro = 1; | |
5979 | set_disk_ro(mddev->gendisk, 1); | |
5980 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d N |
5981 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5982 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5983 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5984 | err = 0; |
a4bd82d0 N |
5985 | } |
5986 | out: | |
5987 | mutex_unlock(&mddev->open_mutex); | |
5988 | return err; | |
5989 | } | |
5990 | ||
9e653b63 N |
5991 | /* mode: |
5992 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5993 | * 2 - stop but do not disassemble array |
5994 | */ | |
f72ffdd6 | 5995 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5996 | struct block_device *bdev) |
1da177e4 | 5997 | { |
1da177e4 | 5998 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5999 | struct md_rdev *rdev; |
30b8feb7 N |
6000 | int did_freeze = 0; |
6001 | ||
6002 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
6003 | did_freeze = 1; | |
6004 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
6005 | md_wakeup_thread(mddev->thread); | |
6006 | } | |
f851b60d | 6007 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 6008 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 6009 | if (mddev->sync_thread) |
30b8feb7 N |
6010 | /* Thread might be blocked waiting for metadata update |
6011 | * which will now never happen */ | |
6012 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 6013 | |
30b8feb7 | 6014 | mddev_unlock(mddev); |
f851b60d N |
6015 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
6016 | !test_bit(MD_RECOVERY_RUNNING, | |
6017 | &mddev->recovery))); | |
30b8feb7 | 6018 | mddev_lock_nointr(mddev); |
1da177e4 | 6019 | |
c8c00a69 | 6020 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 6021 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
6022 | mddev->sysfs_active || |
6023 | mddev->sync_thread || | |
af8d8e6f | 6024 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) { |
9d48739e | 6025 | pr_warn("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 6026 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
6027 | if (did_freeze) { |
6028 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 6029 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
6030 | md_wakeup_thread(mddev->thread); |
6031 | } | |
260fa034 N |
6032 | return -EBUSY; |
6033 | } | |
6e17b027 | 6034 | if (mddev->pers) { |
a4bd82d0 N |
6035 | if (mddev->ro) |
6036 | set_disk_ro(disk, 0); | |
409c57f3 | 6037 | |
defad61a | 6038 | __md_stop_writes(mddev); |
5eff3c43 | 6039 | __md_stop(mddev); |
dc3b17cc | 6040 | mddev->queue->backing_dev_info->congested_fn = NULL; |
6177b472 | 6041 | |
a4bd82d0 | 6042 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 6043 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 6044 | |
dafb20fa | 6045 | rdev_for_each(rdev, mddev) |
36fad858 NK |
6046 | if (rdev->raid_disk >= 0) |
6047 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 6048 | |
a4bd82d0 | 6049 | set_capacity(disk, 0); |
6e17b027 | 6050 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 6051 | mddev->changed = 1; |
a4bd82d0 | 6052 | revalidate_disk(disk); |
0d4ca600 | 6053 | |
a4bd82d0 N |
6054 | if (mddev->ro) |
6055 | mddev->ro = 0; | |
6e17b027 N |
6056 | } else |
6057 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
6058 | /* |
6059 | * Free resources if final stop | |
6060 | */ | |
9e653b63 | 6061 | if (mode == 0) { |
9d48739e | 6062 | pr_info("md: %s stopped.\n", mdname(mddev)); |
1da177e4 | 6063 | |
c3d9714e | 6064 | if (mddev->bitmap_info.file) { |
4af1a041 N |
6065 | struct file *f = mddev->bitmap_info.file; |
6066 | spin_lock(&mddev->lock); | |
c3d9714e | 6067 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
6068 | spin_unlock(&mddev->lock); |
6069 | fput(f); | |
978f946b | 6070 | } |
c3d9714e | 6071 | mddev->bitmap_info.offset = 0; |
978f946b | 6072 | |
1da177e4 LT |
6073 | export_array(mddev); |
6074 | ||
6177b472 | 6075 | md_clean(mddev); |
efeb53c0 N |
6076 | if (mddev->hold_active == UNTIL_STOP) |
6077 | mddev->hold_active = 0; | |
a4bd82d0 | 6078 | } |
d7603b7e | 6079 | md_new_event(mddev); |
00bcb4ac | 6080 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 6081 | return 0; |
1da177e4 LT |
6082 | } |
6083 | ||
fdee8ae4 | 6084 | #ifndef MODULE |
fd01b88c | 6085 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 6086 | { |
3cb03002 | 6087 | struct md_rdev *rdev; |
1da177e4 LT |
6088 | int err; |
6089 | ||
a757e64c | 6090 | if (list_empty(&mddev->disks)) |
1da177e4 | 6091 | return; |
1da177e4 | 6092 | |
9d48739e | 6093 | pr_info("md: running: "); |
1da177e4 | 6094 | |
dafb20fa | 6095 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6096 | char b[BDEVNAME_SIZE]; |
9d48739e | 6097 | pr_cont("<%s>", bdevname(rdev->bdev,b)); |
1da177e4 | 6098 | } |
9d48739e | 6099 | pr_cont("\n"); |
1da177e4 | 6100 | |
d710e138 | 6101 | err = do_md_run(mddev); |
1da177e4 | 6102 | if (err) { |
9d48739e | 6103 | pr_warn("md: do_md_run() returned %d\n", err); |
a05b7ea0 | 6104 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
6105 | } |
6106 | } | |
6107 | ||
6108 | /* | |
6109 | * lets try to run arrays based on all disks that have arrived | |
6110 | * until now. (those are in pending_raid_disks) | |
6111 | * | |
6112 | * the method: pick the first pending disk, collect all disks with | |
6113 | * the same UUID, remove all from the pending list and put them into | |
6114 | * the 'same_array' list. Then order this list based on superblock | |
6115 | * update time (freshest comes first), kick out 'old' disks and | |
6116 | * compare superblocks. If everything's fine then run it. | |
6117 | * | |
6118 | * If "unit" is allocated, then bump its reference count | |
6119 | */ | |
6120 | static void autorun_devices(int part) | |
6121 | { | |
3cb03002 | 6122 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 6123 | struct mddev *mddev; |
1da177e4 LT |
6124 | char b[BDEVNAME_SIZE]; |
6125 | ||
9d48739e | 6126 | pr_info("md: autorun ...\n"); |
1da177e4 | 6127 | while (!list_empty(&pending_raid_disks)) { |
e8703fe1 | 6128 | int unit; |
1da177e4 | 6129 | dev_t dev; |
ad01c9e3 | 6130 | LIST_HEAD(candidates); |
1da177e4 | 6131 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 6132 | struct md_rdev, same_set); |
1da177e4 | 6133 | |
9d48739e | 6134 | pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b)); |
1da177e4 | 6135 | INIT_LIST_HEAD(&candidates); |
159ec1fc | 6136 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 | 6137 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
9d48739e N |
6138 | pr_debug("md: adding %s ...\n", |
6139 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
6140 | list_move(&rdev->same_set, &candidates); |
6141 | } | |
6142 | /* | |
6143 | * now we have a set of devices, with all of them having | |
6144 | * mostly sane superblocks. It's time to allocate the | |
6145 | * mddev. | |
6146 | */ | |
e8703fe1 N |
6147 | if (part) { |
6148 | dev = MKDEV(mdp_major, | |
6149 | rdev0->preferred_minor << MdpMinorShift); | |
6150 | unit = MINOR(dev) >> MdpMinorShift; | |
6151 | } else { | |
6152 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
6153 | unit = MINOR(dev); | |
6154 | } | |
6155 | if (rdev0->preferred_minor != unit) { | |
9d48739e N |
6156 | pr_warn("md: unit number in %s is bad: %d\n", |
6157 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
1da177e4 LT |
6158 | break; |
6159 | } | |
1da177e4 LT |
6160 | |
6161 | md_probe(dev, NULL, NULL); | |
6162 | mddev = mddev_find(dev); | |
9bbbca3a NB |
6163 | if (!mddev || !mddev->gendisk) { |
6164 | if (mddev) | |
6165 | mddev_put(mddev); | |
1da177e4 LT |
6166 | break; |
6167 | } | |
f72ffdd6 | 6168 | if (mddev_lock(mddev)) |
9d48739e | 6169 | pr_warn("md: %s locked, cannot run\n", mdname(mddev)); |
1da177e4 LT |
6170 | else if (mddev->raid_disks || mddev->major_version |
6171 | || !list_empty(&mddev->disks)) { | |
9d48739e | 6172 | pr_warn("md: %s already running, cannot run %s\n", |
1da177e4 LT |
6173 | mdname(mddev), bdevname(rdev0->bdev,b)); |
6174 | mddev_unlock(mddev); | |
6175 | } else { | |
9d48739e | 6176 | pr_debug("md: created %s\n", mdname(mddev)); |
1ec4a939 | 6177 | mddev->persistent = 1; |
159ec1fc | 6178 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
6179 | list_del_init(&rdev->same_set); |
6180 | if (bind_rdev_to_array(rdev, mddev)) | |
6181 | export_rdev(rdev); | |
6182 | } | |
6183 | autorun_array(mddev); | |
6184 | mddev_unlock(mddev); | |
6185 | } | |
6186 | /* on success, candidates will be empty, on error | |
6187 | * it won't... | |
6188 | */ | |
159ec1fc | 6189 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 6190 | list_del_init(&rdev->same_set); |
1da177e4 | 6191 | export_rdev(rdev); |
4b80991c | 6192 | } |
1da177e4 LT |
6193 | mddev_put(mddev); |
6194 | } | |
9d48739e | 6195 | pr_info("md: ... autorun DONE.\n"); |
1da177e4 | 6196 | } |
fdee8ae4 | 6197 | #endif /* !MODULE */ |
1da177e4 | 6198 | |
f72ffdd6 | 6199 | static int get_version(void __user *arg) |
1da177e4 LT |
6200 | { |
6201 | mdu_version_t ver; | |
6202 | ||
6203 | ver.major = MD_MAJOR_VERSION; | |
6204 | ver.minor = MD_MINOR_VERSION; | |
6205 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
6206 | ||
6207 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
6208 | return -EFAULT; | |
6209 | ||
6210 | return 0; | |
6211 | } | |
6212 | ||
f72ffdd6 | 6213 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
6214 | { |
6215 | mdu_array_info_t info; | |
a9f326eb | 6216 | int nr,working,insync,failed,spare; |
3cb03002 | 6217 | struct md_rdev *rdev; |
1da177e4 | 6218 | |
1ca69c4b N |
6219 | nr = working = insync = failed = spare = 0; |
6220 | rcu_read_lock(); | |
6221 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6222 | nr++; |
b2d444d7 | 6223 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
6224 | failed++; |
6225 | else { | |
6226 | working++; | |
b2d444d7 | 6227 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 6228 | insync++; |
b347af81 SL |
6229 | else if (test_bit(Journal, &rdev->flags)) |
6230 | /* TODO: add journal count to md_u.h */ | |
6231 | ; | |
1da177e4 LT |
6232 | else |
6233 | spare++; | |
6234 | } | |
6235 | } | |
1ca69c4b | 6236 | rcu_read_unlock(); |
1da177e4 LT |
6237 | |
6238 | info.major_version = mddev->major_version; | |
6239 | info.minor_version = mddev->minor_version; | |
6240 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6241 | info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX); |
1da177e4 | 6242 | info.level = mddev->level; |
58c0fed4 AN |
6243 | info.size = mddev->dev_sectors / 2; |
6244 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 6245 | info.size = -1; |
1da177e4 LT |
6246 | info.nr_disks = nr; |
6247 | info.raid_disks = mddev->raid_disks; | |
6248 | info.md_minor = mddev->md_minor; | |
6249 | info.not_persistent= !mddev->persistent; | |
6250 | ||
9ebc6ef1 | 6251 | info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX); |
1da177e4 LT |
6252 | info.state = 0; |
6253 | if (mddev->in_sync) | |
6254 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 6255 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 6256 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
ca8895d9 GR |
6257 | if (mddev_is_clustered(mddev)) |
6258 | info.state |= (1<<MD_SB_CLUSTERED); | |
a9f326eb | 6259 | info.active_disks = insync; |
1da177e4 LT |
6260 | info.working_disks = working; |
6261 | info.failed_disks = failed; | |
6262 | info.spare_disks = spare; | |
6263 | ||
6264 | info.layout = mddev->layout; | |
9d8f0363 | 6265 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
6266 | |
6267 | if (copy_to_user(arg, &info, sizeof(info))) | |
6268 | return -EFAULT; | |
6269 | ||
6270 | return 0; | |
6271 | } | |
6272 | ||
f72ffdd6 | 6273 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
6274 | { |
6275 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 6276 | char *ptr; |
4af1a041 | 6277 | int err; |
32a7627c | 6278 | |
b6878d9e | 6279 | file = kzalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 6280 | if (!file) |
4af1a041 | 6281 | return -ENOMEM; |
32a7627c | 6282 | |
4af1a041 N |
6283 | err = 0; |
6284 | spin_lock(&mddev->lock); | |
25eafe1a BR |
6285 | /* bitmap enabled */ |
6286 | if (mddev->bitmap_info.file) { | |
6287 | ptr = file_path(mddev->bitmap_info.file, file->pathname, | |
6288 | sizeof(file->pathname)); | |
6289 | if (IS_ERR(ptr)) | |
6290 | err = PTR_ERR(ptr); | |
6291 | else | |
6292 | memmove(file->pathname, ptr, | |
6293 | sizeof(file->pathname)-(ptr-file->pathname)); | |
6294 | } | |
4af1a041 | 6295 | spin_unlock(&mddev->lock); |
32a7627c | 6296 | |
4af1a041 N |
6297 | if (err == 0 && |
6298 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 6299 | err = -EFAULT; |
4af1a041 | 6300 | |
32a7627c N |
6301 | kfree(file); |
6302 | return err; | |
6303 | } | |
6304 | ||
f72ffdd6 | 6305 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
6306 | { |
6307 | mdu_disk_info_t info; | |
3cb03002 | 6308 | struct md_rdev *rdev; |
1da177e4 LT |
6309 | |
6310 | if (copy_from_user(&info, arg, sizeof(info))) | |
6311 | return -EFAULT; | |
6312 | ||
1ca69c4b | 6313 | rcu_read_lock(); |
57d051dc | 6314 | rdev = md_find_rdev_nr_rcu(mddev, info.number); |
1da177e4 LT |
6315 | if (rdev) { |
6316 | info.major = MAJOR(rdev->bdev->bd_dev); | |
6317 | info.minor = MINOR(rdev->bdev->bd_dev); | |
6318 | info.raid_disk = rdev->raid_disk; | |
6319 | info.state = 0; | |
b2d444d7 | 6320 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6321 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 6322 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
6323 | info.state |= (1<<MD_DISK_ACTIVE); |
6324 | info.state |= (1<<MD_DISK_SYNC); | |
6325 | } | |
9efdca16 | 6326 | if (test_bit(Journal, &rdev->flags)) |
bac624f3 | 6327 | info.state |= (1<<MD_DISK_JOURNAL); |
8ddf9efe N |
6328 | if (test_bit(WriteMostly, &rdev->flags)) |
6329 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
688834e6 N |
6330 | if (test_bit(FailFast, &rdev->flags)) |
6331 | info.state |= (1<<MD_DISK_FAILFAST); | |
1da177e4 LT |
6332 | } else { |
6333 | info.major = info.minor = 0; | |
6334 | info.raid_disk = -1; | |
6335 | info.state = (1<<MD_DISK_REMOVED); | |
6336 | } | |
1ca69c4b | 6337 | rcu_read_unlock(); |
1da177e4 LT |
6338 | |
6339 | if (copy_to_user(arg, &info, sizeof(info))) | |
6340 | return -EFAULT; | |
6341 | ||
6342 | return 0; | |
6343 | } | |
6344 | ||
f72ffdd6 | 6345 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
6346 | { |
6347 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 6348 | struct md_rdev *rdev; |
1da177e4 LT |
6349 | dev_t dev = MKDEV(info->major,info->minor); |
6350 | ||
1aee41f6 GR |
6351 | if (mddev_is_clustered(mddev) && |
6352 | !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) { | |
9d48739e N |
6353 | pr_warn("%s: Cannot add to clustered mddev.\n", |
6354 | mdname(mddev)); | |
1aee41f6 GR |
6355 | return -EINVAL; |
6356 | } | |
6357 | ||
1da177e4 LT |
6358 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) |
6359 | return -EOVERFLOW; | |
6360 | ||
6361 | if (!mddev->raid_disks) { | |
6362 | int err; | |
6363 | /* expecting a device which has a superblock */ | |
6364 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
6365 | if (IS_ERR(rdev)) { | |
9d48739e | 6366 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6367 | PTR_ERR(rdev)); |
6368 | return PTR_ERR(rdev); | |
6369 | } | |
6370 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
6371 | struct md_rdev *rdev0 |
6372 | = list_entry(mddev->disks.next, | |
6373 | struct md_rdev, same_set); | |
a9f326eb | 6374 | err = super_types[mddev->major_version] |
1da177e4 LT |
6375 | .load_super(rdev, rdev0, mddev->minor_version); |
6376 | if (err < 0) { | |
9d48739e | 6377 | pr_warn("md: %s has different UUID to %s\n", |
f72ffdd6 | 6378 | bdevname(rdev->bdev,b), |
1da177e4 LT |
6379 | bdevname(rdev0->bdev,b2)); |
6380 | export_rdev(rdev); | |
6381 | return -EINVAL; | |
6382 | } | |
6383 | } | |
6384 | err = bind_rdev_to_array(rdev, mddev); | |
6385 | if (err) | |
6386 | export_rdev(rdev); | |
6387 | return err; | |
6388 | } | |
6389 | ||
6390 | /* | |
6391 | * add_new_disk can be used once the array is assembled | |
6392 | * to add "hot spares". They must already have a superblock | |
6393 | * written | |
6394 | */ | |
6395 | if (mddev->pers) { | |
6396 | int err; | |
6397 | if (!mddev->pers->hot_add_disk) { | |
9d48739e N |
6398 | pr_warn("%s: personality does not support diskops!\n", |
6399 | mdname(mddev)); | |
1da177e4 LT |
6400 | return -EINVAL; |
6401 | } | |
7b1e35f6 N |
6402 | if (mddev->persistent) |
6403 | rdev = md_import_device(dev, mddev->major_version, | |
6404 | mddev->minor_version); | |
6405 | else | |
6406 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 6407 | if (IS_ERR(rdev)) { |
9d48739e | 6408 | pr_warn("md: md_import_device returned %ld\n", |
1da177e4 LT |
6409 | PTR_ERR(rdev)); |
6410 | return PTR_ERR(rdev); | |
6411 | } | |
1a855a06 | 6412 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
6413 | if (!mddev->persistent) { |
6414 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 6415 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 6416 | rdev->raid_disk = info->raid_disk; |
bf572541 | 6417 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 6418 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 6419 | } else |
41158c7e | 6420 | rdev->raid_disk = -1; |
f466722c | 6421 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
6422 | } else |
6423 | super_types[mddev->major_version]. | |
6424 | validate_super(mddev, rdev); | |
bedd86b7 | 6425 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 6426 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
6427 | /* This was a hot-add request, but events doesn't |
6428 | * match, so reject it. | |
6429 | */ | |
6430 | export_rdev(rdev); | |
6431 | return -EINVAL; | |
6432 | } | |
6433 | ||
b2d444d7 | 6434 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
6435 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6436 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
6437 | else |
6438 | clear_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6439 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6440 | set_bit(FailFast, &rdev->flags); | |
6441 | else | |
6442 | clear_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6443 | |
f6b6ec5c SL |
6444 | if (info->state & (1<<MD_DISK_JOURNAL)) { |
6445 | struct md_rdev *rdev2; | |
6446 | bool has_journal = false; | |
6447 | ||
6448 | /* make sure no existing journal disk */ | |
6449 | rdev_for_each(rdev2, mddev) { | |
6450 | if (test_bit(Journal, &rdev2->flags)) { | |
6451 | has_journal = true; | |
6452 | break; | |
6453 | } | |
6454 | } | |
230b55fa | 6455 | if (has_journal || mddev->bitmap) { |
f6b6ec5c SL |
6456 | export_rdev(rdev); |
6457 | return -EBUSY; | |
6458 | } | |
bac624f3 | 6459 | set_bit(Journal, &rdev->flags); |
f6b6ec5c | 6460 | } |
1aee41f6 GR |
6461 | /* |
6462 | * check whether the device shows up in other nodes | |
6463 | */ | |
6464 | if (mddev_is_clustered(mddev)) { | |
dbb64f86 | 6465 | if (info->state & (1 << MD_DISK_CANDIDATE)) |
1aee41f6 | 6466 | set_bit(Candidate, &rdev->flags); |
dbb64f86 | 6467 | else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) { |
1aee41f6 | 6468 | /* --add initiated by this node */ |
dbb64f86 | 6469 | err = md_cluster_ops->add_new_disk(mddev, rdev); |
1aee41f6 | 6470 | if (err) { |
1aee41f6 GR |
6471 | export_rdev(rdev); |
6472 | return err; | |
6473 | } | |
6474 | } | |
6475 | } | |
6476 | ||
1da177e4 LT |
6477 | rdev->raid_disk = -1; |
6478 | err = bind_rdev_to_array(rdev, mddev); | |
dbb64f86 | 6479 | |
1da177e4 LT |
6480 | if (err) |
6481 | export_rdev(rdev); | |
dbb64f86 GR |
6482 | |
6483 | if (mddev_is_clustered(mddev)) { | |
e566aef1 GJ |
6484 | if (info->state & (1 << MD_DISK_CANDIDATE)) { |
6485 | if (!err) { | |
6486 | err = md_cluster_ops->new_disk_ack(mddev, | |
6487 | err == 0); | |
6488 | if (err) | |
6489 | md_kick_rdev_from_array(rdev); | |
6490 | } | |
6491 | } else { | |
dbb64f86 GR |
6492 | if (err) |
6493 | md_cluster_ops->add_new_disk_cancel(mddev); | |
6494 | else | |
6495 | err = add_bound_rdev(rdev); | |
6496 | } | |
6497 | ||
6498 | } else if (!err) | |
a6da4ef8 | 6499 | err = add_bound_rdev(rdev); |
dbb64f86 | 6500 | |
1da177e4 LT |
6501 | return err; |
6502 | } | |
6503 | ||
6504 | /* otherwise, add_new_disk is only allowed | |
6505 | * for major_version==0 superblocks | |
6506 | */ | |
6507 | if (mddev->major_version != 0) { | |
9d48739e | 6508 | pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev)); |
1da177e4 LT |
6509 | return -EINVAL; |
6510 | } | |
6511 | ||
6512 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
6513 | int err; | |
d710e138 | 6514 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6515 | if (IS_ERR(rdev)) { |
9d48739e | 6516 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6517 | PTR_ERR(rdev)); |
6518 | return PTR_ERR(rdev); | |
6519 | } | |
6520 | rdev->desc_nr = info->number; | |
6521 | if (info->raid_disk < mddev->raid_disks) | |
6522 | rdev->raid_disk = info->raid_disk; | |
6523 | else | |
6524 | rdev->raid_disk = -1; | |
6525 | ||
1da177e4 | 6526 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
6527 | if (info->state & (1<<MD_DISK_SYNC)) |
6528 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 6529 | |
8ddf9efe N |
6530 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
6531 | set_bit(WriteMostly, &rdev->flags); | |
688834e6 N |
6532 | if (info->state & (1<<MD_DISK_FAILFAST)) |
6533 | set_bit(FailFast, &rdev->flags); | |
8ddf9efe | 6534 | |
1da177e4 | 6535 | if (!mddev->persistent) { |
9d48739e | 6536 | pr_debug("md: nonpersistent superblock ...\n"); |
77304d2a MS |
6537 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
6538 | } else | |
57b2caa3 | 6539 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 6540 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6541 | |
2bf071bf N |
6542 | err = bind_rdev_to_array(rdev, mddev); |
6543 | if (err) { | |
6544 | export_rdev(rdev); | |
6545 | return err; | |
6546 | } | |
1da177e4 LT |
6547 | } |
6548 | ||
6549 | return 0; | |
6550 | } | |
6551 | ||
f72ffdd6 | 6552 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6553 | { |
6554 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 6555 | struct md_rdev *rdev; |
1da177e4 | 6556 | |
c42a0e26 YY |
6557 | if (!mddev->pers) |
6558 | return -ENODEV; | |
6559 | ||
1da177e4 LT |
6560 | rdev = find_rdev(mddev, dev); |
6561 | if (!rdev) | |
6562 | return -ENXIO; | |
6563 | ||
2910ff17 GR |
6564 | if (rdev->raid_disk < 0) |
6565 | goto kick_rdev; | |
293467aa | 6566 | |
3ea8929d N |
6567 | clear_bit(Blocked, &rdev->flags); |
6568 | remove_and_add_spares(mddev, rdev); | |
6569 | ||
1da177e4 LT |
6570 | if (rdev->raid_disk >= 0) |
6571 | goto busy; | |
6572 | ||
2910ff17 | 6573 | kick_rdev: |
54a88392 | 6574 | if (mddev_is_clustered(mddev)) |
88bcfef7 GR |
6575 | md_cluster_ops->remove_disk(mddev, rdev); |
6576 | ||
fb56dfef | 6577 | md_kick_rdev_from_array(rdev); |
2953079c | 6578 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6579 | if (mddev->thread) |
6580 | md_wakeup_thread(mddev->thread); | |
6581 | else | |
6582 | md_update_sb(mddev, 1); | |
d7603b7e | 6583 | md_new_event(mddev); |
1da177e4 LT |
6584 | |
6585 | return 0; | |
6586 | busy: | |
9d48739e N |
6587 | pr_debug("md: cannot remove active disk %s from %s ...\n", |
6588 | bdevname(rdev->bdev,b), mdname(mddev)); | |
1da177e4 LT |
6589 | return -EBUSY; |
6590 | } | |
6591 | ||
f72ffdd6 | 6592 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6593 | { |
6594 | char b[BDEVNAME_SIZE]; | |
6595 | int err; | |
3cb03002 | 6596 | struct md_rdev *rdev; |
1da177e4 LT |
6597 | |
6598 | if (!mddev->pers) | |
6599 | return -ENODEV; | |
6600 | ||
6601 | if (mddev->major_version != 0) { | |
9d48739e | 6602 | pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n", |
1da177e4 LT |
6603 | mdname(mddev)); |
6604 | return -EINVAL; | |
6605 | } | |
6606 | if (!mddev->pers->hot_add_disk) { | |
9d48739e | 6607 | pr_warn("%s: personality does not support diskops!\n", |
1da177e4 LT |
6608 | mdname(mddev)); |
6609 | return -EINVAL; | |
6610 | } | |
6611 | ||
d710e138 | 6612 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6613 | if (IS_ERR(rdev)) { |
9d48739e | 6614 | pr_warn("md: error, md_import_device() returned %ld\n", |
1da177e4 LT |
6615 | PTR_ERR(rdev)); |
6616 | return -EINVAL; | |
6617 | } | |
6618 | ||
6619 | if (mddev->persistent) | |
57b2caa3 | 6620 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 6621 | else |
77304d2a | 6622 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 6623 | |
8190e754 | 6624 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6625 | |
b2d444d7 | 6626 | if (test_bit(Faulty, &rdev->flags)) { |
9d48739e | 6627 | pr_warn("md: can not hot-add faulty %s disk to %s!\n", |
1da177e4 LT |
6628 | bdevname(rdev->bdev,b), mdname(mddev)); |
6629 | err = -EINVAL; | |
6630 | goto abort_export; | |
6631 | } | |
293467aa | 6632 | |
b2d444d7 | 6633 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6634 | rdev->desc_nr = -1; |
5842730d | 6635 | rdev->saved_raid_disk = -1; |
2bf071bf N |
6636 | err = bind_rdev_to_array(rdev, mddev); |
6637 | if (err) | |
2aa82191 | 6638 | goto abort_export; |
1da177e4 LT |
6639 | |
6640 | /* | |
6641 | * The rest should better be atomic, we can have disk failures | |
6642 | * noticed in interrupt contexts ... | |
6643 | */ | |
6644 | ||
1da177e4 LT |
6645 | rdev->raid_disk = -1; |
6646 | ||
2953079c | 6647 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
060b0689 N |
6648 | if (!mddev->thread) |
6649 | md_update_sb(mddev, 1); | |
1da177e4 LT |
6650 | /* |
6651 | * Kick recovery, maybe this spare has to be added to the | |
6652 | * array immediately. | |
6653 | */ | |
6654 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6655 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 6656 | md_new_event(mddev); |
1da177e4 LT |
6657 | return 0; |
6658 | ||
1da177e4 LT |
6659 | abort_export: |
6660 | export_rdev(rdev); | |
6661 | return err; | |
6662 | } | |
6663 | ||
fd01b88c | 6664 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 6665 | { |
035328c2 | 6666 | int err = 0; |
32a7627c | 6667 | |
36fa3063 | 6668 | if (mddev->pers) { |
d66b1b39 | 6669 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
6670 | return -EBUSY; |
6671 | if (mddev->recovery || mddev->sync_thread) | |
6672 | return -EBUSY; | |
6673 | /* we should be able to change the bitmap.. */ | |
6674 | } | |
32a7627c | 6675 | |
36fa3063 | 6676 | if (fd >= 0) { |
035328c2 | 6677 | struct inode *inode; |
1e594bb2 N |
6678 | struct file *f; |
6679 | ||
6680 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 6681 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6682 | f = fget(fd); |
32a7627c | 6683 | |
1e594bb2 | 6684 | if (f == NULL) { |
9d48739e N |
6685 | pr_warn("%s: error: failed to get bitmap file\n", |
6686 | mdname(mddev)); | |
36fa3063 N |
6687 | return -EBADF; |
6688 | } | |
6689 | ||
1e594bb2 | 6690 | inode = f->f_mapping->host; |
035328c2 | 6691 | if (!S_ISREG(inode->i_mode)) { |
9d48739e N |
6692 | pr_warn("%s: error: bitmap file must be a regular file\n", |
6693 | mdname(mddev)); | |
035328c2 | 6694 | err = -EBADF; |
1e594bb2 | 6695 | } else if (!(f->f_mode & FMODE_WRITE)) { |
9d48739e N |
6696 | pr_warn("%s: error: bitmap file must open for write\n", |
6697 | mdname(mddev)); | |
035328c2 N |
6698 | err = -EBADF; |
6699 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
9d48739e N |
6700 | pr_warn("%s: error: bitmap file is already in use\n", |
6701 | mdname(mddev)); | |
035328c2 N |
6702 | err = -EBUSY; |
6703 | } | |
6704 | if (err) { | |
1e594bb2 | 6705 | fput(f); |
36fa3063 N |
6706 | return err; |
6707 | } | |
1e594bb2 | 6708 | mddev->bitmap_info.file = f; |
c3d9714e | 6709 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6710 | } else if (mddev->bitmap == NULL) |
6711 | return -ENOENT; /* cannot remove what isn't there */ | |
6712 | err = 0; | |
6713 | if (mddev->pers) { | |
69e51b44 | 6714 | if (fd >= 0) { |
f9209a32 GR |
6715 | struct bitmap *bitmap; |
6716 | ||
6717 | bitmap = bitmap_create(mddev, -1); | |
9e1cc0a5 | 6718 | mddev_suspend(mddev); |
f9209a32 GR |
6719 | if (!IS_ERR(bitmap)) { |
6720 | mddev->bitmap = bitmap; | |
69e51b44 | 6721 | err = bitmap_load(mddev); |
ba599aca N |
6722 | } else |
6723 | err = PTR_ERR(bitmap); | |
52a0d49d N |
6724 | if (err) { |
6725 | bitmap_destroy(mddev); | |
6726 | fd = -1; | |
6727 | } | |
9e1cc0a5 | 6728 | mddev_resume(mddev); |
52a0d49d | 6729 | } else if (fd < 0) { |
9e1cc0a5 | 6730 | mddev_suspend(mddev); |
36fa3063 | 6731 | bitmap_destroy(mddev); |
9e1cc0a5 | 6732 | mddev_resume(mddev); |
d7375ab3 | 6733 | } |
d7375ab3 N |
6734 | } |
6735 | if (fd < 0) { | |
4af1a041 N |
6736 | struct file *f = mddev->bitmap_info.file; |
6737 | if (f) { | |
6738 | spin_lock(&mddev->lock); | |
6739 | mddev->bitmap_info.file = NULL; | |
6740 | spin_unlock(&mddev->lock); | |
6741 | fput(f); | |
6742 | } | |
36fa3063 N |
6743 | } |
6744 | ||
32a7627c N |
6745 | return err; |
6746 | } | |
6747 | ||
1da177e4 LT |
6748 | /* |
6749 | * set_array_info is used two different ways | |
6750 | * The original usage is when creating a new array. | |
6751 | * In this usage, raid_disks is > 0 and it together with | |
6752 | * level, size, not_persistent,layout,chunksize determine the | |
6753 | * shape of the array. | |
6754 | * This will always create an array with a type-0.90.0 superblock. | |
6755 | * The newer usage is when assembling an array. | |
6756 | * In this case raid_disks will be 0, and the major_version field is | |
6757 | * use to determine which style super-blocks are to be found on the devices. | |
6758 | * The minor and patch _version numbers are also kept incase the | |
6759 | * super_block handler wishes to interpret them. | |
6760 | */ | |
f72ffdd6 | 6761 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6762 | { |
6763 | ||
6764 | if (info->raid_disks == 0) { | |
6765 | /* just setting version number for superblock loading */ | |
6766 | if (info->major_version < 0 || | |
50511da3 | 6767 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6768 | super_types[info->major_version].name == NULL) { |
6769 | /* maybe try to auto-load a module? */ | |
9d48739e | 6770 | pr_warn("md: superblock version %d not known\n", |
1da177e4 LT |
6771 | info->major_version); |
6772 | return -EINVAL; | |
6773 | } | |
6774 | mddev->major_version = info->major_version; | |
6775 | mddev->minor_version = info->minor_version; | |
6776 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6777 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6778 | /* ensure mddev_put doesn't delete this now that there |
6779 | * is some minimal configuration. | |
6780 | */ | |
9ebc6ef1 | 6781 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6782 | return 0; |
6783 | } | |
6784 | mddev->major_version = MD_MAJOR_VERSION; | |
6785 | mddev->minor_version = MD_MINOR_VERSION; | |
6786 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
9ebc6ef1 | 6787 | mddev->ctime = ktime_get_real_seconds(); |
1da177e4 LT |
6788 | |
6789 | mddev->level = info->level; | |
17115e03 | 6790 | mddev->clevel[0] = 0; |
58c0fed4 | 6791 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6792 | mddev->raid_disks = info->raid_disks; |
6793 | /* don't set md_minor, it is determined by which /dev/md* was | |
6794 | * openned | |
6795 | */ | |
6796 | if (info->state & (1<<MD_SB_CLEAN)) | |
6797 | mddev->recovery_cp = MaxSector; | |
6798 | else | |
6799 | mddev->recovery_cp = 0; | |
6800 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6801 | mddev->external = 0; |
1da177e4 LT |
6802 | |
6803 | mddev->layout = info->layout; | |
9d8f0363 | 6804 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 | 6805 | |
2953079c | 6806 | if (mddev->persistent) { |
1b3bae49 N |
6807 | mddev->max_disks = MD_SB_DISKS; |
6808 | mddev->flags = 0; | |
6809 | mddev->sb_flags = 0; | |
2953079c SL |
6810 | } |
6811 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); | |
1da177e4 | 6812 | |
c3d9714e | 6813 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6814 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6815 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6816 | |
f6705578 N |
6817 | mddev->reshape_position = MaxSector; |
6818 | ||
1da177e4 LT |
6819 | /* |
6820 | * Generate a 128 bit UUID | |
6821 | */ | |
6822 | get_random_bytes(mddev->uuid, 16); | |
6823 | ||
f6705578 | 6824 | mddev->new_level = mddev->level; |
664e7c41 | 6825 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6826 | mddev->new_layout = mddev->layout; |
6827 | mddev->delta_disks = 0; | |
2c810cdd | 6828 | mddev->reshape_backwards = 0; |
f6705578 | 6829 | |
1da177e4 LT |
6830 | return 0; |
6831 | } | |
6832 | ||
fd01b88c | 6833 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6834 | { |
efa4b77b | 6835 | lockdep_assert_held(&mddev->reconfig_mutex); |
b522adcd DW |
6836 | |
6837 | if (mddev->external_size) | |
6838 | return; | |
6839 | ||
1f403624 DW |
6840 | mddev->array_sectors = array_sectors; |
6841 | } | |
6842 | EXPORT_SYMBOL(md_set_array_sectors); | |
6843 | ||
fd01b88c | 6844 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6845 | { |
3cb03002 | 6846 | struct md_rdev *rdev; |
a35b0d69 | 6847 | int rv; |
d71f9f88 | 6848 | int fit = (num_sectors == 0); |
818da59f | 6849 | sector_t old_dev_sectors = mddev->dev_sectors; |
ab5a98b1 | 6850 | |
a35b0d69 N |
6851 | if (mddev->pers->resize == NULL) |
6852 | return -EINVAL; | |
d71f9f88 AN |
6853 | /* The "num_sectors" is the number of sectors of each device that |
6854 | * is used. This can only make sense for arrays with redundancy. | |
6855 | * linear and raid0 always use whatever space is available. We can only | |
6856 | * consider changing this number if no resync or reconstruction is | |
6857 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6858 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6859 | * of each device. If num_sectors is zero, we find the largest size |
6860 | * that fits. | |
a35b0d69 | 6861 | */ |
f851b60d N |
6862 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6863 | mddev->sync_thread) | |
a35b0d69 | 6864 | return -EBUSY; |
bd8839e0 N |
6865 | if (mddev->ro) |
6866 | return -EROFS; | |
a4a6125a | 6867 | |
dafb20fa | 6868 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6869 | sector_t avail = rdev->sectors; |
01ab5662 | 6870 | |
d71f9f88 AN |
6871 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6872 | num_sectors = avail; | |
6873 | if (avail < num_sectors) | |
a35b0d69 N |
6874 | return -ENOSPC; |
6875 | } | |
d71f9f88 | 6876 | rv = mddev->pers->resize(mddev, num_sectors); |
c9483634 | 6877 | if (!rv) { |
818da59f GJ |
6878 | if (mddev_is_clustered(mddev)) |
6879 | md_cluster_ops->update_size(mddev, old_dev_sectors); | |
6880 | else if (mddev->queue) { | |
c9483634 GJ |
6881 | set_capacity(mddev->gendisk, mddev->array_sectors); |
6882 | revalidate_disk(mddev->gendisk); | |
6883 | } | |
6884 | } | |
a35b0d69 N |
6885 | return rv; |
6886 | } | |
6887 | ||
fd01b88c | 6888 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6889 | { |
6890 | int rv; | |
c6563a8c | 6891 | struct md_rdev *rdev; |
da943b99 | 6892 | /* change the number of raid disks */ |
63c70c4f | 6893 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6894 | return -EINVAL; |
bd8839e0 N |
6895 | if (mddev->ro) |
6896 | return -EROFS; | |
da943b99 | 6897 | if (raid_disks <= 0 || |
233fca36 | 6898 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6899 | return -EINVAL; |
f851b60d N |
6900 | if (mddev->sync_thread || |
6901 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6902 | mddev->reshape_position != MaxSector) | |
da943b99 | 6903 | return -EBUSY; |
c6563a8c N |
6904 | |
6905 | rdev_for_each(rdev, mddev) { | |
6906 | if (mddev->raid_disks < raid_disks && | |
6907 | rdev->data_offset < rdev->new_data_offset) | |
6908 | return -EINVAL; | |
6909 | if (mddev->raid_disks > raid_disks && | |
6910 | rdev->data_offset > rdev->new_data_offset) | |
6911 | return -EINVAL; | |
6912 | } | |
6913 | ||
63c70c4f | 6914 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6915 | if (mddev->delta_disks < 0) |
6916 | mddev->reshape_backwards = 1; | |
6917 | else if (mddev->delta_disks > 0) | |
6918 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6919 | |
6920 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6921 | if (rv < 0) { |
de171cb9 | 6922 | mddev->delta_disks = 0; |
2c810cdd N |
6923 | mddev->reshape_backwards = 0; |
6924 | } | |
da943b99 N |
6925 | return rv; |
6926 | } | |
6927 | ||
1da177e4 LT |
6928 | /* |
6929 | * update_array_info is used to change the configuration of an | |
6930 | * on-line array. | |
6931 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6932 | * fields in the info are checked against the array. | |
6933 | * Any differences that cannot be handled will cause an error. | |
6934 | * Normally, only one change can be managed at a time. | |
6935 | */ | |
fd01b88c | 6936 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6937 | { |
6938 | int rv = 0; | |
6939 | int cnt = 0; | |
36fa3063 N |
6940 | int state = 0; |
6941 | ||
6942 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6943 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6944 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6945 | |
6946 | if (mddev->major_version != info->major_version || | |
6947 | mddev->minor_version != info->minor_version || | |
6948 | /* mddev->patch_version != info->patch_version || */ | |
6949 | mddev->ctime != info->ctime || | |
6950 | mddev->level != info->level || | |
6951 | /* mddev->layout != info->layout || */ | |
4e023612 | 6952 | mddev->persistent != !info->not_persistent || |
9d8f0363 | 6953 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6954 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6955 | ((state^info->state) & 0xfffffe00) | |
6956 | ) | |
1da177e4 LT |
6957 | return -EINVAL; |
6958 | /* Check there is only one change */ | |
58c0fed4 AN |
6959 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6960 | cnt++; | |
6961 | if (mddev->raid_disks != info->raid_disks) | |
6962 | cnt++; | |
6963 | if (mddev->layout != info->layout) | |
6964 | cnt++; | |
6965 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6966 | cnt++; | |
6967 | if (cnt == 0) | |
6968 | return 0; | |
6969 | if (cnt > 1) | |
6970 | return -EINVAL; | |
1da177e4 LT |
6971 | |
6972 | if (mddev->layout != info->layout) { | |
6973 | /* Change layout | |
6974 | * we don't need to do anything at the md level, the | |
6975 | * personality will take care of it all. | |
6976 | */ | |
50ac168a | 6977 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6978 | return -EINVAL; |
597a711b N |
6979 | else { |
6980 | mddev->new_layout = info->layout; | |
50ac168a | 6981 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6982 | if (rv) |
6983 | mddev->new_layout = mddev->layout; | |
6984 | return rv; | |
6985 | } | |
1da177e4 | 6986 | } |
58c0fed4 | 6987 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6988 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6989 | |
da943b99 N |
6990 | if (mddev->raid_disks != info->raid_disks) |
6991 | rv = update_raid_disks(mddev, info->raid_disks); | |
6992 | ||
36fa3063 | 6993 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
293467aa GR |
6994 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) { |
6995 | rv = -EINVAL; | |
6996 | goto err; | |
6997 | } | |
6998 | if (mddev->recovery || mddev->sync_thread) { | |
6999 | rv = -EBUSY; | |
7000 | goto err; | |
7001 | } | |
36fa3063 | 7002 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { |
f9209a32 | 7003 | struct bitmap *bitmap; |
36fa3063 | 7004 | /* add the bitmap */ |
293467aa GR |
7005 | if (mddev->bitmap) { |
7006 | rv = -EEXIST; | |
7007 | goto err; | |
7008 | } | |
7009 | if (mddev->bitmap_info.default_offset == 0) { | |
7010 | rv = -EINVAL; | |
7011 | goto err; | |
7012 | } | |
c3d9714e N |
7013 | mddev->bitmap_info.offset = |
7014 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
7015 | mddev->bitmap_info.space = |
7016 | mddev->bitmap_info.default_space; | |
f9209a32 | 7017 | bitmap = bitmap_create(mddev, -1); |
9e1cc0a5 | 7018 | mddev_suspend(mddev); |
f9209a32 GR |
7019 | if (!IS_ERR(bitmap)) { |
7020 | mddev->bitmap = bitmap; | |
69e51b44 | 7021 | rv = bitmap_load(mddev); |
ba599aca N |
7022 | } else |
7023 | rv = PTR_ERR(bitmap); | |
36fa3063 N |
7024 | if (rv) |
7025 | bitmap_destroy(mddev); | |
9e1cc0a5 | 7026 | mddev_resume(mddev); |
36fa3063 N |
7027 | } else { |
7028 | /* remove the bitmap */ | |
293467aa GR |
7029 | if (!mddev->bitmap) { |
7030 | rv = -ENOENT; | |
7031 | goto err; | |
7032 | } | |
7033 | if (mddev->bitmap->storage.file) { | |
7034 | rv = -EINVAL; | |
7035 | goto err; | |
7036 | } | |
f6a2dc64 GJ |
7037 | if (mddev->bitmap_info.nodes) { |
7038 | /* hold PW on all the bitmap lock */ | |
7039 | if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) { | |
9d48739e | 7040 | pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n"); |
f6a2dc64 GJ |
7041 | rv = -EPERM; |
7042 | md_cluster_ops->unlock_all_bitmaps(mddev); | |
7043 | goto err; | |
7044 | } | |
7045 | ||
7046 | mddev->bitmap_info.nodes = 0; | |
7047 | md_cluster_ops->leave(mddev); | |
7048 | } | |
9e1cc0a5 | 7049 | mddev_suspend(mddev); |
36fa3063 | 7050 | bitmap_destroy(mddev); |
9e1cc0a5 | 7051 | mddev_resume(mddev); |
c3d9714e | 7052 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
7053 | } |
7054 | } | |
850b2b42 | 7055 | md_update_sb(mddev, 1); |
293467aa GR |
7056 | return rv; |
7057 | err: | |
1da177e4 LT |
7058 | return rv; |
7059 | } | |
7060 | ||
fd01b88c | 7061 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 7062 | { |
3cb03002 | 7063 | struct md_rdev *rdev; |
1ca69c4b | 7064 | int err = 0; |
1da177e4 LT |
7065 | |
7066 | if (mddev->pers == NULL) | |
7067 | return -ENODEV; | |
7068 | ||
1ca69c4b | 7069 | rcu_read_lock(); |
1532d9e8 | 7070 | rdev = md_find_rdev_rcu(mddev, dev); |
1da177e4 | 7071 | if (!rdev) |
1ca69c4b N |
7072 | err = -ENODEV; |
7073 | else { | |
7074 | md_error(mddev, rdev); | |
7075 | if (!test_bit(Faulty, &rdev->flags)) | |
7076 | err = -EBUSY; | |
7077 | } | |
7078 | rcu_read_unlock(); | |
7079 | return err; | |
1da177e4 LT |
7080 | } |
7081 | ||
2f9618ce AN |
7082 | /* |
7083 | * We have a problem here : there is no easy way to give a CHS | |
7084 | * virtual geometry. We currently pretend that we have a 2 heads | |
7085 | * 4 sectors (with a BIG number of cylinders...). This drives | |
7086 | * dosfs just mad... ;-) | |
7087 | */ | |
a885c8c4 CH |
7088 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
7089 | { | |
fd01b88c | 7090 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
7091 | |
7092 | geo->heads = 2; | |
7093 | geo->sectors = 4; | |
49ce6cea | 7094 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
7095 | return 0; |
7096 | } | |
7097 | ||
cb335f88 NS |
7098 | static inline bool md_ioctl_valid(unsigned int cmd) |
7099 | { | |
7100 | switch (cmd) { | |
7101 | case ADD_NEW_DISK: | |
7102 | case BLKROSET: | |
7103 | case GET_ARRAY_INFO: | |
7104 | case GET_BITMAP_FILE: | |
7105 | case GET_DISK_INFO: | |
7106 | case HOT_ADD_DISK: | |
7107 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
7108 | case RAID_AUTORUN: |
7109 | case RAID_VERSION: | |
7110 | case RESTART_ARRAY_RW: | |
7111 | case RUN_ARRAY: | |
7112 | case SET_ARRAY_INFO: | |
7113 | case SET_BITMAP_FILE: | |
7114 | case SET_DISK_FAULTY: | |
7115 | case STOP_ARRAY: | |
7116 | case STOP_ARRAY_RO: | |
1aee41f6 | 7117 | case CLUSTERED_DISK_NACK: |
cb335f88 NS |
7118 | return true; |
7119 | default: | |
7120 | return false; | |
7121 | } | |
7122 | } | |
7123 | ||
a39907fa | 7124 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
7125 | unsigned int cmd, unsigned long arg) |
7126 | { | |
7127 | int err = 0; | |
7128 | void __user *argp = (void __user *)arg; | |
fd01b88c | 7129 | struct mddev *mddev = NULL; |
e2218350 | 7130 | int ro; |
065e519e | 7131 | bool did_set_md_closing = false; |
1da177e4 | 7132 | |
cb335f88 NS |
7133 | if (!md_ioctl_valid(cmd)) |
7134 | return -ENOTTY; | |
7135 | ||
506c9e44 N |
7136 | switch (cmd) { |
7137 | case RAID_VERSION: | |
7138 | case GET_ARRAY_INFO: | |
7139 | case GET_DISK_INFO: | |
7140 | break; | |
7141 | default: | |
7142 | if (!capable(CAP_SYS_ADMIN)) | |
7143 | return -EACCES; | |
7144 | } | |
1da177e4 LT |
7145 | |
7146 | /* | |
7147 | * Commands dealing with the RAID driver but not any | |
7148 | * particular array: | |
7149 | */ | |
c02c0aeb N |
7150 | switch (cmd) { |
7151 | case RAID_VERSION: | |
7152 | err = get_version(argp); | |
3adc28d8 | 7153 | goto out; |
1da177e4 | 7154 | |
1da177e4 | 7155 | #ifndef MODULE |
c02c0aeb N |
7156 | case RAID_AUTORUN: |
7157 | err = 0; | |
7158 | autostart_arrays(arg); | |
3adc28d8 | 7159 | goto out; |
1da177e4 | 7160 | #endif |
c02c0aeb | 7161 | default:; |
1da177e4 LT |
7162 | } |
7163 | ||
7164 | /* | |
7165 | * Commands creating/starting a new array: | |
7166 | */ | |
7167 | ||
a39907fa | 7168 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
7169 | |
7170 | if (!mddev) { | |
7171 | BUG(); | |
3adc28d8 | 7172 | goto out; |
1da177e4 LT |
7173 | } |
7174 | ||
1ca69c4b N |
7175 | /* Some actions do not requires the mutex */ |
7176 | switch (cmd) { | |
7177 | case GET_ARRAY_INFO: | |
7178 | if (!mddev->raid_disks && !mddev->external) | |
7179 | err = -ENODEV; | |
7180 | else | |
7181 | err = get_array_info(mddev, argp); | |
3adc28d8 | 7182 | goto out; |
1ca69c4b N |
7183 | |
7184 | case GET_DISK_INFO: | |
7185 | if (!mddev->raid_disks && !mddev->external) | |
7186 | err = -ENODEV; | |
7187 | else | |
7188 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 7189 | goto out; |
1ca69c4b N |
7190 | |
7191 | case SET_DISK_FAULTY: | |
7192 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7193 | goto out; |
4af1a041 N |
7194 | |
7195 | case GET_BITMAP_FILE: | |
7196 | err = get_bitmap_file(mddev, argp); | |
7197 | goto out; | |
7198 | ||
1ca69c4b N |
7199 | } |
7200 | ||
a7a3f08d N |
7201 | if (cmd == ADD_NEW_DISK) |
7202 | /* need to ensure md_delayed_delete() has completed */ | |
7203 | flush_workqueue(md_misc_wq); | |
7204 | ||
90f5f7ad HR |
7205 | if (cmd == HOT_REMOVE_DISK) |
7206 | /* need to ensure recovery thread has run */ | |
7207 | wait_event_interruptible_timeout(mddev->sb_wait, | |
7208 | !test_bit(MD_RECOVERY_NEEDED, | |
82a301cb | 7209 | &mddev->recovery), |
90f5f7ad | 7210 | msecs_to_jiffies(5000)); |
260fa034 N |
7211 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
7212 | /* Need to flush page cache, and ensure no-one else opens | |
7213 | * and writes | |
7214 | */ | |
7215 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 7216 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
7217 | mutex_unlock(&mddev->open_mutex); |
7218 | err = -EBUSY; | |
3adc28d8 | 7219 | goto out; |
260fa034 | 7220 | } |
065e519e | 7221 | WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags)); |
af8d8e6f | 7222 | set_bit(MD_CLOSING, &mddev->flags); |
065e519e | 7223 | did_set_md_closing = true; |
260fa034 N |
7224 | mutex_unlock(&mddev->open_mutex); |
7225 | sync_blockdev(bdev); | |
7226 | } | |
1da177e4 LT |
7227 | err = mddev_lock(mddev); |
7228 | if (err) { | |
9d48739e N |
7229 | pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n", |
7230 | err, cmd); | |
3adc28d8 | 7231 | goto out; |
1da177e4 LT |
7232 | } |
7233 | ||
c02c0aeb N |
7234 | if (cmd == SET_ARRAY_INFO) { |
7235 | mdu_array_info_t info; | |
7236 | if (!arg) | |
7237 | memset(&info, 0, sizeof(info)); | |
7238 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
7239 | err = -EFAULT; | |
3adc28d8 | 7240 | goto unlock; |
c02c0aeb N |
7241 | } |
7242 | if (mddev->pers) { | |
7243 | err = update_array_info(mddev, &info); | |
7244 | if (err) { | |
9d48739e | 7245 | pr_warn("md: couldn't update array info. %d\n", err); |
3adc28d8 | 7246 | goto unlock; |
1da177e4 | 7247 | } |
3adc28d8 | 7248 | goto unlock; |
c02c0aeb N |
7249 | } |
7250 | if (!list_empty(&mddev->disks)) { | |
9d48739e | 7251 | pr_warn("md: array %s already has disks!\n", mdname(mddev)); |
c02c0aeb | 7252 | err = -EBUSY; |
3adc28d8 | 7253 | goto unlock; |
c02c0aeb N |
7254 | } |
7255 | if (mddev->raid_disks) { | |
9d48739e | 7256 | pr_warn("md: array %s already initialised!\n", mdname(mddev)); |
c02c0aeb | 7257 | err = -EBUSY; |
3adc28d8 | 7258 | goto unlock; |
c02c0aeb N |
7259 | } |
7260 | err = set_array_info(mddev, &info); | |
7261 | if (err) { | |
9d48739e | 7262 | pr_warn("md: couldn't set array info. %d\n", err); |
3adc28d8 | 7263 | goto unlock; |
c02c0aeb | 7264 | } |
3adc28d8 | 7265 | goto unlock; |
1da177e4 LT |
7266 | } |
7267 | ||
7268 | /* | |
7269 | * Commands querying/configuring an existing array: | |
7270 | */ | |
32a7627c | 7271 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 7272 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
7273 | if ((!mddev->raid_disks && !mddev->external) |
7274 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
7275 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
7276 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 7277 | err = -ENODEV; |
3adc28d8 | 7278 | goto unlock; |
1da177e4 LT |
7279 | } |
7280 | ||
7281 | /* | |
7282 | * Commands even a read-only array can execute: | |
7283 | */ | |
c02c0aeb | 7284 | switch (cmd) { |
c02c0aeb N |
7285 | case RESTART_ARRAY_RW: |
7286 | err = restart_array(mddev); | |
3adc28d8 | 7287 | goto unlock; |
1da177e4 | 7288 | |
c02c0aeb N |
7289 | case STOP_ARRAY: |
7290 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 7291 | goto unlock; |
1da177e4 | 7292 | |
c02c0aeb N |
7293 | case STOP_ARRAY_RO: |
7294 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 7295 | goto unlock; |
1da177e4 | 7296 | |
3ea8929d N |
7297 | case HOT_REMOVE_DISK: |
7298 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7299 | goto unlock; |
3ea8929d | 7300 | |
7ceb17e8 N |
7301 | case ADD_NEW_DISK: |
7302 | /* We can support ADD_NEW_DISK on read-only arrays | |
466ad292 | 7303 | * only if we are re-adding a preexisting device. |
7ceb17e8 N |
7304 | * So require mddev->pers and MD_DISK_SYNC. |
7305 | */ | |
7306 | if (mddev->pers) { | |
7307 | mdu_disk_info_t info; | |
7308 | if (copy_from_user(&info, argp, sizeof(info))) | |
7309 | err = -EFAULT; | |
7310 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
7311 | /* Need to clear read-only for this */ | |
7312 | break; | |
7313 | else | |
7314 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7315 | goto unlock; |
7ceb17e8 N |
7316 | } |
7317 | break; | |
7318 | ||
c02c0aeb N |
7319 | case BLKROSET: |
7320 | if (get_user(ro, (int __user *)(arg))) { | |
7321 | err = -EFAULT; | |
3adc28d8 | 7322 | goto unlock; |
c02c0aeb N |
7323 | } |
7324 | err = -EINVAL; | |
e2218350 | 7325 | |
c02c0aeb N |
7326 | /* if the bdev is going readonly the value of mddev->ro |
7327 | * does not matter, no writes are coming | |
7328 | */ | |
7329 | if (ro) | |
3adc28d8 | 7330 | goto unlock; |
e2218350 | 7331 | |
c02c0aeb N |
7332 | /* are we are already prepared for writes? */ |
7333 | if (mddev->ro != 1) | |
3adc28d8 | 7334 | goto unlock; |
e2218350 | 7335 | |
c02c0aeb N |
7336 | /* transitioning to readauto need only happen for |
7337 | * arrays that call md_write_start | |
7338 | */ | |
7339 | if (mddev->pers) { | |
7340 | err = restart_array(mddev); | |
7341 | if (err == 0) { | |
7342 | mddev->ro = 2; | |
7343 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 7344 | } |
c02c0aeb | 7345 | } |
3adc28d8 | 7346 | goto unlock; |
1da177e4 LT |
7347 | } |
7348 | ||
7349 | /* | |
7350 | * The remaining ioctls are changing the state of the | |
f91de92e | 7351 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 7352 | */ |
326eb17d | 7353 | if (mddev->ro && mddev->pers) { |
f91de92e N |
7354 | if (mddev->ro == 2) { |
7355 | mddev->ro = 0; | |
00bcb4ac | 7356 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 7357 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
7358 | /* mddev_unlock will wake thread */ |
7359 | /* If a device failed while we were read-only, we | |
7360 | * need to make sure the metadata is updated now. | |
7361 | */ | |
2953079c | 7362 | if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) { |
f3378b48 N |
7363 | mddev_unlock(mddev); |
7364 | wait_event(mddev->sb_wait, | |
2953079c SL |
7365 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) && |
7366 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); | |
29f097c4 | 7367 | mddev_lock_nointr(mddev); |
f3378b48 | 7368 | } |
f91de92e N |
7369 | } else { |
7370 | err = -EROFS; | |
3adc28d8 | 7371 | goto unlock; |
f91de92e | 7372 | } |
1da177e4 LT |
7373 | } |
7374 | ||
c02c0aeb N |
7375 | switch (cmd) { |
7376 | case ADD_NEW_DISK: | |
1da177e4 | 7377 | { |
c02c0aeb N |
7378 | mdu_disk_info_t info; |
7379 | if (copy_from_user(&info, argp, sizeof(info))) | |
7380 | err = -EFAULT; | |
7381 | else | |
7382 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 7383 | goto unlock; |
c02c0aeb | 7384 | } |
1da177e4 | 7385 | |
1aee41f6 GR |
7386 | case CLUSTERED_DISK_NACK: |
7387 | if (mddev_is_clustered(mddev)) | |
7388 | md_cluster_ops->new_disk_ack(mddev, false); | |
7389 | else | |
7390 | err = -EINVAL; | |
7391 | goto unlock; | |
7392 | ||
c02c0aeb N |
7393 | case HOT_ADD_DISK: |
7394 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 7395 | goto unlock; |
1da177e4 | 7396 | |
c02c0aeb N |
7397 | case RUN_ARRAY: |
7398 | err = do_md_run(mddev); | |
3adc28d8 | 7399 | goto unlock; |
1da177e4 | 7400 | |
c02c0aeb N |
7401 | case SET_BITMAP_FILE: |
7402 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 7403 | goto unlock; |
32a7627c | 7404 | |
c02c0aeb N |
7405 | default: |
7406 | err = -EINVAL; | |
3adc28d8 | 7407 | goto unlock; |
1da177e4 LT |
7408 | } |
7409 | ||
3adc28d8 | 7410 | unlock: |
d3374825 N |
7411 | if (mddev->hold_active == UNTIL_IOCTL && |
7412 | err != -EINVAL) | |
7413 | mddev->hold_active = 0; | |
1da177e4 | 7414 | mddev_unlock(mddev); |
3adc28d8 | 7415 | out: |
065e519e N |
7416 | if(did_set_md_closing) |
7417 | clear_bit(MD_CLOSING, &mddev->flags); | |
1da177e4 LT |
7418 | return err; |
7419 | } | |
aa98aa31 AB |
7420 | #ifdef CONFIG_COMPAT |
7421 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
7422 | unsigned int cmd, unsigned long arg) | |
7423 | { | |
7424 | switch (cmd) { | |
7425 | case HOT_REMOVE_DISK: | |
7426 | case HOT_ADD_DISK: | |
7427 | case SET_DISK_FAULTY: | |
7428 | case SET_BITMAP_FILE: | |
7429 | /* These take in integer arg, do not convert */ | |
7430 | break; | |
7431 | default: | |
7432 | arg = (unsigned long)compat_ptr(arg); | |
7433 | break; | |
7434 | } | |
7435 | ||
7436 | return md_ioctl(bdev, mode, cmd, arg); | |
7437 | } | |
7438 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 7439 | |
a39907fa | 7440 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
7441 | { |
7442 | /* | |
7443 | * Succeed if we can lock the mddev, which confirms that | |
7444 | * it isn't being stopped right now. | |
7445 | */ | |
fd01b88c | 7446 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
7447 | int err; |
7448 | ||
0c098220 YL |
7449 | if (!mddev) |
7450 | return -ENODEV; | |
7451 | ||
d3374825 N |
7452 | if (mddev->gendisk != bdev->bd_disk) { |
7453 | /* we are racing with mddev_put which is discarding this | |
7454 | * bd_disk. | |
7455 | */ | |
7456 | mddev_put(mddev); | |
7457 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 7458 | flush_workqueue(md_misc_wq); |
d3374825 N |
7459 | /* Then retry the open from the top */ |
7460 | return -ERESTARTSYS; | |
7461 | } | |
7462 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
7463 | ||
c8c00a69 | 7464 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
7465 | goto out; |
7466 | ||
af8d8e6f GJ |
7467 | if (test_bit(MD_CLOSING, &mddev->flags)) { |
7468 | mutex_unlock(&mddev->open_mutex); | |
e2342ca8 N |
7469 | err = -ENODEV; |
7470 | goto out; | |
af8d8e6f GJ |
7471 | } |
7472 | ||
1da177e4 | 7473 | err = 0; |
f2ea68cf | 7474 | atomic_inc(&mddev->openers); |
c8c00a69 | 7475 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 7476 | |
f0b4f7e2 | 7477 | check_disk_change(bdev); |
1da177e4 | 7478 | out: |
e2342ca8 N |
7479 | if (err) |
7480 | mddev_put(mddev); | |
1da177e4 LT |
7481 | return err; |
7482 | } | |
7483 | ||
db2a144b | 7484 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 7485 | { |
f72ffdd6 | 7486 | struct mddev *mddev = disk->private_data; |
1da177e4 | 7487 | |
52e5f9d1 | 7488 | BUG_ON(!mddev); |
f2ea68cf | 7489 | atomic_dec(&mddev->openers); |
1da177e4 | 7490 | mddev_put(mddev); |
1da177e4 | 7491 | } |
f0b4f7e2 N |
7492 | |
7493 | static int md_media_changed(struct gendisk *disk) | |
7494 | { | |
fd01b88c | 7495 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7496 | |
7497 | return mddev->changed; | |
7498 | } | |
7499 | ||
7500 | static int md_revalidate(struct gendisk *disk) | |
7501 | { | |
fd01b88c | 7502 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
7503 | |
7504 | mddev->changed = 0; | |
7505 | return 0; | |
7506 | } | |
83d5cde4 | 7507 | static const struct block_device_operations md_fops = |
1da177e4 LT |
7508 | { |
7509 | .owner = THIS_MODULE, | |
a39907fa AV |
7510 | .open = md_open, |
7511 | .release = md_release, | |
b492b852 | 7512 | .ioctl = md_ioctl, |
aa98aa31 AB |
7513 | #ifdef CONFIG_COMPAT |
7514 | .compat_ioctl = md_compat_ioctl, | |
7515 | #endif | |
a885c8c4 | 7516 | .getgeo = md_getgeo, |
f0b4f7e2 N |
7517 | .media_changed = md_media_changed, |
7518 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
7519 | }; |
7520 | ||
f72ffdd6 | 7521 | static int md_thread(void *arg) |
1da177e4 | 7522 | { |
2b8bf345 | 7523 | struct md_thread *thread = arg; |
1da177e4 | 7524 | |
1da177e4 LT |
7525 | /* |
7526 | * md_thread is a 'system-thread', it's priority should be very | |
7527 | * high. We avoid resource deadlocks individually in each | |
7528 | * raid personality. (RAID5 does preallocation) We also use RR and | |
7529 | * the very same RT priority as kswapd, thus we will never get | |
7530 | * into a priority inversion deadlock. | |
7531 | * | |
7532 | * we definitely have to have equal or higher priority than | |
7533 | * bdflush, otherwise bdflush will deadlock if there are too | |
7534 | * many dirty RAID5 blocks. | |
7535 | */ | |
1da177e4 | 7536 | |
6985c43f | 7537 | allow_signal(SIGKILL); |
a6fb0934 | 7538 | while (!kthread_should_stop()) { |
1da177e4 | 7539 | |
93588e22 N |
7540 | /* We need to wait INTERRUPTIBLE so that |
7541 | * we don't add to the load-average. | |
7542 | * That means we need to be sure no signals are | |
7543 | * pending | |
7544 | */ | |
7545 | if (signal_pending(current)) | |
7546 | flush_signals(current); | |
7547 | ||
7548 | wait_event_interruptible_timeout | |
7549 | (thread->wqueue, | |
7550 | test_bit(THREAD_WAKEUP, &thread->flags) | |
ce1ccd07 | 7551 | || kthread_should_stop() || kthread_should_park(), |
93588e22 | 7552 | thread->timeout); |
1da177e4 | 7553 | |
6c987910 | 7554 | clear_bit(THREAD_WAKEUP, &thread->flags); |
ce1ccd07 SL |
7555 | if (kthread_should_park()) |
7556 | kthread_parkme(); | |
6c987910 | 7557 | if (!kthread_should_stop()) |
4ed8731d | 7558 | thread->run(thread); |
1da177e4 | 7559 | } |
a6fb0934 | 7560 | |
1da177e4 LT |
7561 | return 0; |
7562 | } | |
7563 | ||
2b8bf345 | 7564 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
7565 | { |
7566 | if (thread) { | |
36a4e1fe | 7567 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
d1d90147 GJ |
7568 | set_bit(THREAD_WAKEUP, &thread->flags); |
7569 | wake_up(&thread->wqueue); | |
1da177e4 LT |
7570 | } |
7571 | } | |
6c144d31 | 7572 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 7573 | |
4ed8731d SL |
7574 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
7575 | struct mddev *mddev, const char *name) | |
1da177e4 | 7576 | { |
2b8bf345 | 7577 | struct md_thread *thread; |
1da177e4 | 7578 | |
2b8bf345 | 7579 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
7580 | if (!thread) |
7581 | return NULL; | |
7582 | ||
1da177e4 LT |
7583 | init_waitqueue_head(&thread->wqueue); |
7584 | ||
1da177e4 LT |
7585 | thread->run = run; |
7586 | thread->mddev = mddev; | |
32a7627c | 7587 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
7588 | thread->tsk = kthread_run(md_thread, thread, |
7589 | "%s_%s", | |
7590 | mdname(thread->mddev), | |
0232605d | 7591 | name); |
a6fb0934 | 7592 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
7593 | kfree(thread); |
7594 | return NULL; | |
7595 | } | |
1da177e4 LT |
7596 | return thread; |
7597 | } | |
6c144d31 | 7598 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 7599 | |
2b8bf345 | 7600 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 7601 | { |
2b8bf345 | 7602 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
7603 | if (!thread) |
7604 | return; | |
36a4e1fe | 7605 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
7606 | /* Locking ensures that mddev_unlock does not wake_up a |
7607 | * non-existent thread | |
7608 | */ | |
7609 | spin_lock(&pers_lock); | |
7610 | *threadp = NULL; | |
7611 | spin_unlock(&pers_lock); | |
a6fb0934 N |
7612 | |
7613 | kthread_stop(thread->tsk); | |
1da177e4 LT |
7614 | kfree(thread); |
7615 | } | |
6c144d31 | 7616 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 7617 | |
fd01b88c | 7618 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7619 | { |
b2d444d7 | 7620 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 7621 | return; |
6bfe0b49 | 7622 | |
de393cde | 7623 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
7624 | return; |
7625 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
7626 | if (mddev->degraded) |
7627 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 7628 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
7629 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7630 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7631 | md_wakeup_thread(mddev->thread); | |
768a418d | 7632 | if (mddev->event_work.func) |
e804ac78 | 7633 | queue_work(md_misc_wq, &mddev->event_work); |
bb9ef716 | 7634 | md_new_event(mddev); |
1da177e4 | 7635 | } |
6c144d31 | 7636 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
7637 | |
7638 | /* seq_file implementation /proc/mdstat */ | |
7639 | ||
7640 | static void status_unused(struct seq_file *seq) | |
7641 | { | |
7642 | int i = 0; | |
3cb03002 | 7643 | struct md_rdev *rdev; |
1da177e4 LT |
7644 | |
7645 | seq_printf(seq, "unused devices: "); | |
7646 | ||
159ec1fc | 7647 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
7648 | char b[BDEVNAME_SIZE]; |
7649 | i++; | |
7650 | seq_printf(seq, "%s ", | |
7651 | bdevname(rdev->bdev,b)); | |
7652 | } | |
7653 | if (!i) | |
7654 | seq_printf(seq, "<none>"); | |
7655 | ||
7656 | seq_printf(seq, "\n"); | |
7657 | } | |
7658 | ||
f7851be7 | 7659 | static int status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7660 | { |
dd71cf6b N |
7661 | sector_t max_sectors, resync, res; |
7662 | unsigned long dt, db; | |
7663 | sector_t rt; | |
4588b42e N |
7664 | int scale; |
7665 | unsigned int per_milli; | |
1da177e4 | 7666 | |
c804cdec N |
7667 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7668 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 7669 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 7670 | else |
dd71cf6b | 7671 | max_sectors = mddev->dev_sectors; |
1da177e4 | 7672 | |
f7851be7 N |
7673 | resync = mddev->curr_resync; |
7674 | if (resync <= 3) { | |
7675 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7676 | /* Still cleaning up */ | |
7677 | resync = max_sectors; | |
d2e2ec82 ND |
7678 | } else if (resync > max_sectors) |
7679 | resync = max_sectors; | |
7680 | else | |
f7851be7 N |
7681 | resync -= atomic_read(&mddev->recovery_active); |
7682 | ||
7683 | if (resync == 0) { | |
7684 | if (mddev->recovery_cp < MaxSector) { | |
7685 | seq_printf(seq, "\tresync=PENDING"); | |
7686 | return 1; | |
7687 | } | |
7688 | return 0; | |
7689 | } | |
7690 | if (resync < 3) { | |
7691 | seq_printf(seq, "\tresync=DELAYED"); | |
7692 | return 1; | |
7693 | } | |
7694 | ||
403df478 | 7695 | WARN_ON(max_sectors == 0); |
4588b42e | 7696 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 7697 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
7698 | * u32, as those are the requirements for sector_div. |
7699 | * Thus 'scale' must be at least 10 | |
7700 | */ | |
7701 | scale = 10; | |
7702 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 7703 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
7704 | scale++; |
7705 | } | |
7706 | res = (resync>>scale)*1000; | |
dd71cf6b | 7707 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
7708 | |
7709 | per_milli = res; | |
1da177e4 | 7710 | { |
4588b42e | 7711 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
7712 | seq_printf(seq, "["); |
7713 | for (i = 0; i < x; i++) | |
7714 | seq_printf(seq, "="); | |
7715 | seq_printf(seq, ">"); | |
7716 | for (i = 0; i < y; i++) | |
7717 | seq_printf(seq, "."); | |
7718 | seq_printf(seq, "] "); | |
7719 | } | |
4588b42e | 7720 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
7721 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
7722 | "reshape" : | |
61df9d91 N |
7723 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
7724 | "check" : | |
7725 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
7726 | "resync" : "recovery"))), | |
7727 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
7728 | (unsigned long long) resync/2, |
7729 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
7730 | |
7731 | /* | |
1da177e4 LT |
7732 | * dt: time from mark until now |
7733 | * db: blocks written from mark until now | |
7734 | * rt: remaining time | |
dd71cf6b N |
7735 | * |
7736 | * rt is a sector_t, so could be 32bit or 64bit. | |
7737 | * So we divide before multiply in case it is 32bit and close | |
7738 | * to the limit. | |
25985edc | 7739 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
7740 | * near the end of resync when the number of remaining sectors |
7741 | * is close to 'db'. | |
7742 | * We then divide rt by 32 after multiplying by db to compensate. | |
7743 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
7744 | */ |
7745 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
7746 | if (!dt) dt++; | |
ff4e8d9a N |
7747 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
7748 | - mddev->resync_mark_cnt; | |
1da177e4 | 7749 | |
dd71cf6b N |
7750 | rt = max_sectors - resync; /* number of remaining sectors */ |
7751 | sector_div(rt, db/32+1); | |
7752 | rt *= dt; | |
7753 | rt >>= 5; | |
7754 | ||
7755 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
7756 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 7757 | |
ff4e8d9a | 7758 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
f7851be7 | 7759 | return 1; |
1da177e4 LT |
7760 | } |
7761 | ||
7762 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7763 | { | |
7764 | struct list_head *tmp; | |
7765 | loff_t l = *pos; | |
fd01b88c | 7766 | struct mddev *mddev; |
1da177e4 LT |
7767 | |
7768 | if (l >= 0x10000) | |
7769 | return NULL; | |
7770 | if (!l--) | |
7771 | /* header */ | |
7772 | return (void*)1; | |
7773 | ||
7774 | spin_lock(&all_mddevs_lock); | |
7775 | list_for_each(tmp,&all_mddevs) | |
7776 | if (!l--) { | |
fd01b88c | 7777 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7778 | mddev_get(mddev); |
7779 | spin_unlock(&all_mddevs_lock); | |
7780 | return mddev; | |
7781 | } | |
7782 | spin_unlock(&all_mddevs_lock); | |
7783 | if (!l--) | |
7784 | return (void*)2;/* tail */ | |
7785 | return NULL; | |
7786 | } | |
7787 | ||
7788 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7789 | { | |
7790 | struct list_head *tmp; | |
fd01b88c | 7791 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7792 | |
1da177e4 LT |
7793 | ++*pos; |
7794 | if (v == (void*)2) | |
7795 | return NULL; | |
7796 | ||
7797 | spin_lock(&all_mddevs_lock); | |
7798 | if (v == (void*)1) | |
7799 | tmp = all_mddevs.next; | |
7800 | else | |
7801 | tmp = mddev->all_mddevs.next; | |
7802 | if (tmp != &all_mddevs) | |
fd01b88c | 7803 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7804 | else { |
7805 | next_mddev = (void*)2; | |
7806 | *pos = 0x10000; | |
f72ffdd6 | 7807 | } |
1da177e4 LT |
7808 | spin_unlock(&all_mddevs_lock); |
7809 | ||
7810 | if (v != (void*)1) | |
7811 | mddev_put(mddev); | |
7812 | return next_mddev; | |
7813 | ||
7814 | } | |
7815 | ||
7816 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7817 | { | |
fd01b88c | 7818 | struct mddev *mddev = v; |
1da177e4 LT |
7819 | |
7820 | if (mddev && v != (void*)1 && v != (void*)2) | |
7821 | mddev_put(mddev); | |
7822 | } | |
7823 | ||
7824 | static int md_seq_show(struct seq_file *seq, void *v) | |
7825 | { | |
fd01b88c | 7826 | struct mddev *mddev = v; |
dd8ac336 | 7827 | sector_t sectors; |
3cb03002 | 7828 | struct md_rdev *rdev; |
1da177e4 LT |
7829 | |
7830 | if (v == (void*)1) { | |
84fc4b56 | 7831 | struct md_personality *pers; |
1da177e4 LT |
7832 | seq_printf(seq, "Personalities : "); |
7833 | spin_lock(&pers_lock); | |
2604b703 N |
7834 | list_for_each_entry(pers, &pers_list, list) |
7835 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7836 | |
7837 | spin_unlock(&pers_lock); | |
7838 | seq_printf(seq, "\n"); | |
f1514638 | 7839 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7840 | return 0; |
7841 | } | |
7842 | if (v == (void*)2) { | |
7843 | status_unused(seq); | |
7844 | return 0; | |
7845 | } | |
7846 | ||
36d091f4 | 7847 | spin_lock(&mddev->lock); |
1da177e4 LT |
7848 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7849 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7850 | mddev->pers ? "" : "in"); | |
7851 | if (mddev->pers) { | |
f91de92e | 7852 | if (mddev->ro==1) |
1da177e4 | 7853 | seq_printf(seq, " (read-only)"); |
f91de92e | 7854 | if (mddev->ro==2) |
52720ae7 | 7855 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7856 | seq_printf(seq, " %s", mddev->pers->name); |
7857 | } | |
7858 | ||
dd8ac336 | 7859 | sectors = 0; |
f97fcad3 N |
7860 | rcu_read_lock(); |
7861 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7862 | char b[BDEVNAME_SIZE]; |
7863 | seq_printf(seq, " %s[%d]", | |
7864 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7865 | if (test_bit(WriteMostly, &rdev->flags)) |
7866 | seq_printf(seq, "(W)"); | |
9efdca16 SL |
7867 | if (test_bit(Journal, &rdev->flags)) |
7868 | seq_printf(seq, "(J)"); | |
b2d444d7 | 7869 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7870 | seq_printf(seq, "(F)"); |
7871 | continue; | |
2d78f8c4 N |
7872 | } |
7873 | if (rdev->raid_disk < 0) | |
b325a32e | 7874 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7875 | if (test_bit(Replacement, &rdev->flags)) |
7876 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7877 | sectors += rdev->sectors; |
1da177e4 | 7878 | } |
f97fcad3 | 7879 | rcu_read_unlock(); |
1da177e4 LT |
7880 | |
7881 | if (!list_empty(&mddev->disks)) { | |
7882 | if (mddev->pers) | |
7883 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7884 | (unsigned long long) |
7885 | mddev->array_sectors / 2); | |
1da177e4 LT |
7886 | else |
7887 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7888 | (unsigned long long)sectors / 2); |
1da177e4 | 7889 | } |
1cd6bf19 N |
7890 | if (mddev->persistent) { |
7891 | if (mddev->major_version != 0 || | |
7892 | mddev->minor_version != 90) { | |
7893 | seq_printf(seq," super %d.%d", | |
7894 | mddev->major_version, | |
7895 | mddev->minor_version); | |
7896 | } | |
e691063a N |
7897 | } else if (mddev->external) |
7898 | seq_printf(seq, " super external:%s", | |
7899 | mddev->metadata_type); | |
7900 | else | |
1cd6bf19 | 7901 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7902 | |
7903 | if (mddev->pers) { | |
d710e138 | 7904 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7905 | seq_printf(seq, "\n "); |
8e1b39d6 | 7906 | if (mddev->pers->sync_request) { |
f7851be7 | 7907 | if (status_resync(seq, mddev)) |
8e1b39d6 | 7908 | seq_printf(seq, "\n "); |
8e1b39d6 | 7909 | } |
32a7627c N |
7910 | } else |
7911 | seq_printf(seq, "\n "); | |
7912 | ||
57148964 | 7913 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7914 | |
7915 | seq_printf(seq, "\n"); | |
7916 | } | |
36d091f4 | 7917 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7918 | |
1da177e4 LT |
7919 | return 0; |
7920 | } | |
7921 | ||
110518bc | 7922 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7923 | .start = md_seq_start, |
7924 | .next = md_seq_next, | |
7925 | .stop = md_seq_stop, | |
7926 | .show = md_seq_show, | |
7927 | }; | |
7928 | ||
7929 | static int md_seq_open(struct inode *inode, struct file *file) | |
7930 | { | |
f1514638 | 7931 | struct seq_file *seq; |
1da177e4 LT |
7932 | int error; |
7933 | ||
7934 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7935 | if (error) |
f1514638 KS |
7936 | return error; |
7937 | ||
7938 | seq = file->private_data; | |
7939 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7940 | return error; |
7941 | } | |
7942 | ||
e2f23b60 | 7943 | static int md_unloading; |
afc9a42b | 7944 | static __poll_t mdstat_poll(struct file *filp, poll_table *wait) |
d7603b7e | 7945 | { |
f1514638 | 7946 | struct seq_file *seq = filp->private_data; |
afc9a42b | 7947 | __poll_t mask; |
d7603b7e | 7948 | |
e2f23b60 | 7949 | if (md_unloading) |
a9a08845 | 7950 | return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI; |
d7603b7e N |
7951 | poll_wait(filp, &md_event_waiters, wait); |
7952 | ||
7953 | /* always allow read */ | |
a9a08845 | 7954 | mask = EPOLLIN | EPOLLRDNORM; |
d7603b7e | 7955 | |
f1514638 | 7956 | if (seq->poll_event != atomic_read(&md_event_count)) |
a9a08845 | 7957 | mask |= EPOLLERR | EPOLLPRI; |
d7603b7e N |
7958 | return mask; |
7959 | } | |
7960 | ||
fa027c2a | 7961 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7962 | .owner = THIS_MODULE, |
1da177e4 LT |
7963 | .open = md_seq_open, |
7964 | .read = seq_read, | |
7965 | .llseek = seq_lseek, | |
26e13043 | 7966 | .release = seq_release, |
d7603b7e | 7967 | .poll = mdstat_poll, |
1da177e4 LT |
7968 | }; |
7969 | ||
84fc4b56 | 7970 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7971 | { |
9d48739e N |
7972 | pr_debug("md: %s personality registered for level %d\n", |
7973 | p->name, p->level); | |
1da177e4 | 7974 | spin_lock(&pers_lock); |
2604b703 | 7975 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7976 | spin_unlock(&pers_lock); |
7977 | return 0; | |
7978 | } | |
6c144d31 | 7979 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7980 | |
84fc4b56 | 7981 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7982 | { |
9d48739e | 7983 | pr_debug("md: %s personality unregistered\n", p->name); |
1da177e4 | 7984 | spin_lock(&pers_lock); |
2604b703 | 7985 | list_del_init(&p->list); |
1da177e4 LT |
7986 | spin_unlock(&pers_lock); |
7987 | return 0; | |
7988 | } | |
6c144d31 | 7989 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7990 | |
6022e75b N |
7991 | int register_md_cluster_operations(struct md_cluster_operations *ops, |
7992 | struct module *module) | |
edb39c9d | 7993 | { |
6022e75b | 7994 | int ret = 0; |
edb39c9d | 7995 | spin_lock(&pers_lock); |
6022e75b N |
7996 | if (md_cluster_ops != NULL) |
7997 | ret = -EALREADY; | |
7998 | else { | |
7999 | md_cluster_ops = ops; | |
8000 | md_cluster_mod = module; | |
8001 | } | |
edb39c9d | 8002 | spin_unlock(&pers_lock); |
6022e75b | 8003 | return ret; |
edb39c9d GR |
8004 | } |
8005 | EXPORT_SYMBOL(register_md_cluster_operations); | |
8006 | ||
8007 | int unregister_md_cluster_operations(void) | |
8008 | { | |
8009 | spin_lock(&pers_lock); | |
8010 | md_cluster_ops = NULL; | |
8011 | spin_unlock(&pers_lock); | |
8012 | return 0; | |
8013 | } | |
8014 | EXPORT_SYMBOL(unregister_md_cluster_operations); | |
8015 | ||
8016 | int md_setup_cluster(struct mddev *mddev, int nodes) | |
8017 | { | |
47a7b0d8 GJ |
8018 | if (!md_cluster_ops) |
8019 | request_module("md-cluster"); | |
edb39c9d | 8020 | spin_lock(&pers_lock); |
47a7b0d8 | 8021 | /* ensure module won't be unloaded */ |
edb39c9d | 8022 | if (!md_cluster_ops || !try_module_get(md_cluster_mod)) { |
9d48739e | 8023 | pr_warn("can't find md-cluster module or get it's reference.\n"); |
edb39c9d GR |
8024 | spin_unlock(&pers_lock); |
8025 | return -ENOENT; | |
8026 | } | |
8027 | spin_unlock(&pers_lock); | |
8028 | ||
cf921cc1 | 8029 | return md_cluster_ops->join(mddev, nodes); |
edb39c9d GR |
8030 | } |
8031 | ||
8032 | void md_cluster_stop(struct mddev *mddev) | |
8033 | { | |
c4ce867f GR |
8034 | if (!md_cluster_ops) |
8035 | return; | |
edb39c9d GR |
8036 | md_cluster_ops->leave(mddev); |
8037 | module_put(md_cluster_mod); | |
8038 | } | |
8039 | ||
fd01b88c | 8040 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 8041 | { |
f72ffdd6 | 8042 | struct md_rdev *rdev; |
1da177e4 | 8043 | int idle; |
eea1bf38 | 8044 | int curr_events; |
1da177e4 LT |
8045 | |
8046 | idle = 1; | |
4b80991c N |
8047 | rcu_read_lock(); |
8048 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 8049 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
59767fbd | 8050 | curr_events = (int)part_stat_read_accum(&disk->part0, sectors) - |
eea1bf38 | 8051 | atomic_read(&disk->sync_io); |
713f6ab1 N |
8052 | /* sync IO will cause sync_io to increase before the disk_stats |
8053 | * as sync_io is counted when a request starts, and | |
8054 | * disk_stats is counted when it completes. | |
8055 | * So resync activity will cause curr_events to be smaller than | |
8056 | * when there was no such activity. | |
8057 | * non-sync IO will cause disk_stat to increase without | |
8058 | * increasing sync_io so curr_events will (eventually) | |
8059 | * be larger than it was before. Once it becomes | |
8060 | * substantially larger, the test below will cause | |
8061 | * the array to appear non-idle, and resync will slow | |
8062 | * down. | |
8063 | * If there is a lot of outstanding resync activity when | |
8064 | * we set last_event to curr_events, then all that activity | |
8065 | * completing might cause the array to appear non-idle | |
8066 | * and resync will be slowed down even though there might | |
8067 | * not have been non-resync activity. This will only | |
8068 | * happen once though. 'last_events' will soon reflect | |
8069 | * the state where there is little or no outstanding | |
8070 | * resync requests, and further resync activity will | |
8071 | * always make curr_events less than last_events. | |
c0e48521 | 8072 | * |
1da177e4 | 8073 | */ |
eea1bf38 | 8074 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
8075 | rdev->last_events = curr_events; |
8076 | idle = 0; | |
8077 | } | |
8078 | } | |
4b80991c | 8079 | rcu_read_unlock(); |
1da177e4 LT |
8080 | return idle; |
8081 | } | |
8082 | ||
fd01b88c | 8083 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
8084 | { |
8085 | /* another "blocks" (512byte) blocks have been synced */ | |
8086 | atomic_sub(blocks, &mddev->recovery_active); | |
8087 | wake_up(&mddev->recovery_wait); | |
8088 | if (!ok) { | |
dfc70645 | 8089 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 8090 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
8091 | md_wakeup_thread(mddev->thread); |
8092 | // stop recovery, signal do_sync .... | |
8093 | } | |
8094 | } | |
6c144d31 | 8095 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 8096 | |
06d91a5f N |
8097 | /* md_write_start(mddev, bi) |
8098 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
8099 | * in superblock) before writing, schedule a superblock update |
8100 | * and wait for it to complete. | |
cc27b0c7 N |
8101 | * A return value of 'false' means that the write wasn't recorded |
8102 | * and cannot proceed as the array is being suspend. | |
06d91a5f | 8103 | */ |
cc27b0c7 | 8104 | bool md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 8105 | { |
0fd62b86 | 8106 | int did_change = 0; |
4b6c1060 | 8107 | |
06d91a5f | 8108 | if (bio_data_dir(bi) != WRITE) |
cc27b0c7 | 8109 | return true; |
06d91a5f | 8110 | |
f91de92e N |
8111 | BUG_ON(mddev->ro == 1); |
8112 | if (mddev->ro == 2) { | |
8113 | /* need to switch to read/write */ | |
8114 | mddev->ro = 0; | |
8115 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8116 | md_wakeup_thread(mddev->thread); | |
25156198 | 8117 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 8118 | did_change = 1; |
f91de92e | 8119 | } |
4ad23a97 N |
8120 | rcu_read_lock(); |
8121 | percpu_ref_get(&mddev->writes_pending); | |
55cc39f3 | 8122 | smp_mb(); /* Match smp_mb in set_in_sync() */ |
31a59e34 N |
8123 | if (mddev->safemode == 1) |
8124 | mddev->safemode = 0; | |
4ad23a97 | 8125 | /* sync_checkers is always 0 when writes_pending is in per-cpu mode */ |
81fe48e9 | 8126 | if (mddev->in_sync || mddev->sync_checkers) { |
85572d7c | 8127 | spin_lock(&mddev->lock); |
3d310eb7 N |
8128 | if (mddev->in_sync) { |
8129 | mddev->in_sync = 0; | |
2953079c SL |
8130 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
8131 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
3d310eb7 | 8132 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 8133 | did_change = 1; |
3d310eb7 | 8134 | } |
85572d7c | 8135 | spin_unlock(&mddev->lock); |
06d91a5f | 8136 | } |
4ad23a97 | 8137 | rcu_read_unlock(); |
0fd62b86 | 8138 | if (did_change) |
00bcb4ac | 8139 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
4b6c1060 HM |
8140 | if (!mddev->has_superblocks) |
8141 | return true; | |
09a44cc1 | 8142 | wait_event(mddev->sb_wait, |
d47c8ad2 N |
8143 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) || |
8144 | mddev->suspended); | |
cc27b0c7 N |
8145 | if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
8146 | percpu_ref_put(&mddev->writes_pending); | |
8147 | return false; | |
8148 | } | |
8149 | return true; | |
1da177e4 | 8150 | } |
6c144d31 | 8151 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 8152 | |
49728050 N |
8153 | /* md_write_inc can only be called when md_write_start() has |
8154 | * already been called at least once of the current request. | |
8155 | * It increments the counter and is useful when a single request | |
8156 | * is split into several parts. Each part causes an increment and | |
8157 | * so needs a matching md_write_end(). | |
8158 | * Unlike md_write_start(), it is safe to call md_write_inc() inside | |
8159 | * a spinlocked region. | |
8160 | */ | |
8161 | void md_write_inc(struct mddev *mddev, struct bio *bi) | |
8162 | { | |
8163 | if (bio_data_dir(bi) != WRITE) | |
8164 | return; | |
8165 | WARN_ON_ONCE(mddev->in_sync || mddev->ro); | |
4ad23a97 | 8166 | percpu_ref_get(&mddev->writes_pending); |
49728050 N |
8167 | } |
8168 | EXPORT_SYMBOL(md_write_inc); | |
8169 | ||
fd01b88c | 8170 | void md_write_end(struct mddev *mddev) |
1da177e4 | 8171 | { |
4ad23a97 N |
8172 | percpu_ref_put(&mddev->writes_pending); |
8173 | ||
8174 | if (mddev->safemode == 2) | |
8175 | md_wakeup_thread(mddev->thread); | |
8176 | else if (mddev->safemode_delay) | |
8177 | /* The roundup() ensures this only performs locking once | |
8178 | * every ->safemode_delay jiffies | |
8179 | */ | |
8180 | mod_timer(&mddev->safemode_timer, | |
8181 | roundup(jiffies, mddev->safemode_delay) + | |
8182 | mddev->safemode_delay); | |
1da177e4 | 8183 | } |
4ad23a97 | 8184 | |
6c144d31 | 8185 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 8186 | |
2a2275d6 N |
8187 | /* md_allow_write(mddev) |
8188 | * Calling this ensures that the array is marked 'active' so that writes | |
8189 | * may proceed without blocking. It is important to call this before | |
8190 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
8191 | * Must be called with mddev_lock held. | |
8192 | */ | |
2214c260 | 8193 | void md_allow_write(struct mddev *mddev) |
2a2275d6 N |
8194 | { |
8195 | if (!mddev->pers) | |
2214c260 | 8196 | return; |
2a2275d6 | 8197 | if (mddev->ro) |
2214c260 | 8198 | return; |
1a0fd497 | 8199 | if (!mddev->pers->sync_request) |
2214c260 | 8200 | return; |
2a2275d6 | 8201 | |
85572d7c | 8202 | spin_lock(&mddev->lock); |
2a2275d6 N |
8203 | if (mddev->in_sync) { |
8204 | mddev->in_sync = 0; | |
2953079c SL |
8205 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
8206 | set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); | |
2a2275d6 N |
8207 | if (mddev->safemode_delay && |
8208 | mddev->safemode == 0) | |
8209 | mddev->safemode = 1; | |
85572d7c | 8210 | spin_unlock(&mddev->lock); |
2a2275d6 | 8211 | md_update_sb(mddev, 0); |
00bcb4ac | 8212 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2214c260 AP |
8213 | /* wait for the dirty state to be recorded in the metadata */ |
8214 | wait_event(mddev->sb_wait, | |
2214c260 | 8215 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
2a2275d6 | 8216 | } else |
85572d7c | 8217 | spin_unlock(&mddev->lock); |
2a2275d6 N |
8218 | } |
8219 | EXPORT_SYMBOL_GPL(md_allow_write); | |
8220 | ||
1da177e4 LT |
8221 | #define SYNC_MARKS 10 |
8222 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 8223 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 8224 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 8225 | { |
4ed8731d | 8226 | struct mddev *mddev = thread->mddev; |
fd01b88c | 8227 | struct mddev *mddev2; |
1da177e4 LT |
8228 | unsigned int currspeed = 0, |
8229 | window; | |
ac7e50a3 | 8230 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 8231 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 8232 | unsigned long update_time; |
1da177e4 LT |
8233 | sector_t mark_cnt[SYNC_MARKS]; |
8234 | int last_mark,m; | |
8235 | struct list_head *tmp; | |
8236 | sector_t last_check; | |
57afd89f | 8237 | int skipped = 0; |
3cb03002 | 8238 | struct md_rdev *rdev; |
c4a39551 | 8239 | char *desc, *action = NULL; |
7c2c57c9 | 8240 | struct blk_plug plug; |
41a9a0dc | 8241 | int ret; |
1da177e4 LT |
8242 | |
8243 | /* just incase thread restarts... */ | |
d5d885fd SL |
8244 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
8245 | test_bit(MD_RECOVERY_WAIT, &mddev->recovery)) | |
1da177e4 | 8246 | return; |
3991b31e N |
8247 | if (mddev->ro) {/* never try to sync a read-only array */ |
8248 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 8249 | return; |
3991b31e | 8250 | } |
1da177e4 | 8251 | |
41a9a0dc GJ |
8252 | if (mddev_is_clustered(mddev)) { |
8253 | ret = md_cluster_ops->resync_start(mddev); | |
8254 | if (ret) | |
8255 | goto skip; | |
8256 | ||
bb8bf15b | 8257 | set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags); |
41a9a0dc GJ |
8258 | if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
8259 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) || | |
8260 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) | |
8261 | && ((unsigned long long)mddev->curr_resync_completed | |
8262 | < (unsigned long long)mddev->resync_max_sectors)) | |
8263 | goto skip; | |
8264 | } | |
8265 | ||
61df9d91 | 8266 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 8267 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 8268 | desc = "data-check"; |
c4a39551 JB |
8269 | action = "check"; |
8270 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 8271 | desc = "requested-resync"; |
c4a39551 JB |
8272 | action = "repair"; |
8273 | } else | |
61df9d91 N |
8274 | desc = "resync"; |
8275 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
8276 | desc = "reshape"; | |
8277 | else | |
8278 | desc = "recovery"; | |
8279 | ||
c4a39551 JB |
8280 | mddev->last_sync_action = action ?: desc; |
8281 | ||
1da177e4 LT |
8282 | /* we overload curr_resync somewhat here. |
8283 | * 0 == not engaged in resync at all | |
8284 | * 2 == checking that there is no conflict with another sync | |
8285 | * 1 == like 2, but have yielded to allow conflicting resync to | |
8286 | * commense | |
8287 | * other == active in resync - this many blocks | |
8288 | * | |
8289 | * Before starting a resync we must have set curr_resync to | |
8290 | * 2, and then checked that every "conflicting" array has curr_resync | |
8291 | * less than ours. When we find one that is the same or higher | |
8292 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
8293 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
8294 | * This will mean we have to start checking from the beginning again. | |
8295 | * | |
8296 | */ | |
8297 | ||
8298 | do { | |
c622ca54 | 8299 | int mddev2_minor = -1; |
1da177e4 LT |
8300 | mddev->curr_resync = 2; |
8301 | ||
8302 | try_again: | |
404e4b43 | 8303 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 8304 | goto skip; |
29ac4aa3 | 8305 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
8306 | if (mddev2 == mddev) |
8307 | continue; | |
90b08710 BS |
8308 | if (!mddev->parallel_resync |
8309 | && mddev2->curr_resync | |
8310 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
8311 | DEFINE_WAIT(wq); |
8312 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
8313 | /* arbitrarily yield */ | |
8314 | mddev->curr_resync = 1; | |
8315 | wake_up(&resync_wait); | |
8316 | } | |
8317 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
8318 | /* no need to wait here, we can wait the next | |
8319 | * time 'round when curr_resync == 2 | |
8320 | */ | |
8321 | continue; | |
9744197c N |
8322 | /* We need to wait 'interruptible' so as not to |
8323 | * contribute to the load average, and not to | |
8324 | * be caught by 'softlockup' | |
8325 | */ | |
8326 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 8327 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 8328 | mddev2->curr_resync >= mddev->curr_resync) { |
c622ca54 AP |
8329 | if (mddev2_minor != mddev2->md_minor) { |
8330 | mddev2_minor = mddev2->md_minor; | |
9d48739e N |
8331 | pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n", |
8332 | desc, mdname(mddev), | |
8333 | mdname(mddev2)); | |
c622ca54 | 8334 | } |
1da177e4 | 8335 | mddev_put(mddev2); |
9744197c N |
8336 | if (signal_pending(current)) |
8337 | flush_signals(current); | |
1da177e4 LT |
8338 | schedule(); |
8339 | finish_wait(&resync_wait, &wq); | |
8340 | goto try_again; | |
8341 | } | |
8342 | finish_wait(&resync_wait, &wq); | |
8343 | } | |
8344 | } | |
8345 | } while (mddev->curr_resync < 2); | |
8346 | ||
5fd6c1dc | 8347 | j = 0; |
9d88883e | 8348 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 8349 | /* resync follows the size requested by the personality, |
57afd89f | 8350 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
8351 | */ |
8352 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 8353 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 8354 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
8355 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
8356 | j = mddev->resync_min; | |
8357 | else if (!mddev->bitmap) | |
5fd6c1dc | 8358 | j = mddev->recovery_cp; |
5e96ee65 | 8359 | |
ccfcc3c1 | 8360 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 8361 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 8362 | else { |
1da177e4 | 8363 | /* recovery follows the physical size of devices */ |
58c0fed4 | 8364 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 8365 | j = MaxSector; |
4e59ca7d | 8366 | rcu_read_lock(); |
dafb20fa | 8367 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 8368 | if (rdev->raid_disk >= 0 && |
f2076e7d | 8369 | !test_bit(Journal, &rdev->flags) && |
5fd6c1dc N |
8370 | !test_bit(Faulty, &rdev->flags) && |
8371 | !test_bit(In_sync, &rdev->flags) && | |
8372 | rdev->recovery_offset < j) | |
8373 | j = rdev->recovery_offset; | |
4e59ca7d | 8374 | rcu_read_unlock(); |
133d4527 N |
8375 | |
8376 | /* If there is a bitmap, we need to make sure all | |
8377 | * writes that started before we added a spare | |
8378 | * complete before we start doing a recovery. | |
8379 | * Otherwise the write might complete and (via | |
8380 | * bitmap_endwrite) set a bit in the bitmap after the | |
8381 | * recovery has checked that bit and skipped that | |
8382 | * region. | |
8383 | */ | |
8384 | if (mddev->bitmap) { | |
8385 | mddev->pers->quiesce(mddev, 1); | |
8386 | mddev->pers->quiesce(mddev, 0); | |
8387 | } | |
5fd6c1dc | 8388 | } |
1da177e4 | 8389 | |
9d48739e N |
8390 | pr_info("md: %s of RAID array %s\n", desc, mdname(mddev)); |
8391 | pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev)); | |
8392 | pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n", | |
8393 | speed_max(mddev), desc); | |
1da177e4 | 8394 | |
eea1bf38 | 8395 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 8396 | |
57afd89f | 8397 | io_sectors = 0; |
1da177e4 LT |
8398 | for (m = 0; m < SYNC_MARKS; m++) { |
8399 | mark[m] = jiffies; | |
57afd89f | 8400 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
8401 | } |
8402 | last_mark = 0; | |
8403 | mddev->resync_mark = mark[last_mark]; | |
8404 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
8405 | ||
8406 | /* | |
8407 | * Tune reconstruction: | |
8408 | */ | |
8409 | window = 32*(PAGE_SIZE/512); | |
9d48739e N |
8410 | pr_debug("md: using %dk window, over a total of %lluk.\n", |
8411 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
8412 | |
8413 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
8414 | last_check = 0; |
8415 | ||
8416 | if (j>2) { | |
9d48739e N |
8417 | pr_debug("md: resuming %s of %s from checkpoint.\n", |
8418 | desc, mdname(mddev)); | |
1da177e4 | 8419 | mddev->curr_resync = j; |
72f36d59 N |
8420 | } else |
8421 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 8422 | mddev->curr_resync_completed = j; |
72f36d59 N |
8423 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
8424 | md_new_event(mddev); | |
54f89341 | 8425 | update_time = jiffies; |
1da177e4 | 8426 | |
7c2c57c9 | 8427 | blk_start_plug(&plug); |
1da177e4 | 8428 | while (j < max_sectors) { |
57afd89f | 8429 | sector_t sectors; |
1da177e4 | 8430 | |
57afd89f | 8431 | skipped = 0; |
97e4f42d | 8432 | |
7a91ee1f N |
8433 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8434 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
8435 | (mddev->curr_resync - mddev->curr_resync_completed) | |
8436 | > (max_sectors >> 4)) || | |
54f89341 | 8437 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f | 8438 | (j - mddev->curr_resync_completed)*2 |
c5e19d90 N |
8439 | >= mddev->resync_max - mddev->curr_resync_completed || |
8440 | mddev->curr_resync_completed > mddev->resync_max | |
7a91ee1f | 8441 | )) { |
97e4f42d | 8442 | /* time to update curr_resync_completed */ |
97e4f42d N |
8443 | wait_event(mddev->recovery_wait, |
8444 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 8445 | mddev->curr_resync_completed = j; |
35d78c66 | 8446 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
8447 | j > mddev->recovery_cp) | |
8448 | mddev->recovery_cp = j; | |
54f89341 | 8449 | update_time = jiffies; |
2953079c | 8450 | set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); |
acb180b0 | 8451 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 8452 | } |
acb180b0 | 8453 | |
c91abf5a N |
8454 | while (j >= mddev->resync_max && |
8455 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
8456 | /* As this condition is controlled by user-space, |
8457 | * we can block indefinitely, so use '_interruptible' | |
8458 | * to avoid triggering warnings. | |
8459 | */ | |
8460 | flush_signals(current); /* just in case */ | |
8461 | wait_event_interruptible(mddev->recovery_wait, | |
8462 | mddev->resync_max > j | |
c91abf5a N |
8463 | || test_bit(MD_RECOVERY_INTR, |
8464 | &mddev->recovery)); | |
e62e58a5 | 8465 | } |
acb180b0 | 8466 | |
c91abf5a N |
8467 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8468 | break; | |
acb180b0 | 8469 | |
09314799 | 8470 | sectors = mddev->pers->sync_request(mddev, j, &skipped); |
57afd89f | 8471 | if (sectors == 0) { |
dfc70645 | 8472 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 8473 | break; |
1da177e4 | 8474 | } |
57afd89f N |
8475 | |
8476 | if (!skipped) { /* actual IO requested */ | |
8477 | io_sectors += sectors; | |
8478 | atomic_add(sectors, &mddev->recovery_active); | |
8479 | } | |
8480 | ||
e875ecea N |
8481 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8482 | break; | |
8483 | ||
1da177e4 | 8484 | j += sectors; |
5ed1df2e N |
8485 | if (j > max_sectors) |
8486 | /* when skipping, extra large numbers can be returned. */ | |
8487 | j = max_sectors; | |
72f36d59 N |
8488 | if (j > 2) |
8489 | mddev->curr_resync = j; | |
ff4e8d9a | 8490 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 8491 | if (last_check == 0) |
e875ecea | 8492 | /* this is the earliest that rebuild will be |
d7603b7e N |
8493 | * visible in /proc/mdstat |
8494 | */ | |
8495 | md_new_event(mddev); | |
57afd89f N |
8496 | |
8497 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
8498 | continue; |
8499 | ||
57afd89f | 8500 | last_check = io_sectors; |
1da177e4 LT |
8501 | repeat: |
8502 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
8503 | /* step marks */ | |
8504 | int next = (last_mark+1) % SYNC_MARKS; | |
8505 | ||
8506 | mddev->resync_mark = mark[next]; | |
8507 | mddev->resync_mark_cnt = mark_cnt[next]; | |
8508 | mark[next] = jiffies; | |
57afd89f | 8509 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
8510 | last_mark = next; |
8511 | } | |
8512 | ||
c91abf5a N |
8513 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
8514 | break; | |
1da177e4 LT |
8515 | |
8516 | /* | |
8517 | * this loop exits only if either when we are slower than | |
8518 | * the 'hard' speed limit, or the system was IO-idle for | |
8519 | * a jiffy. | |
8520 | * the system might be non-idle CPU-wise, but we only care | |
8521 | * about not overloading the IO subsystem. (things like an | |
8522 | * e2fsck being done on the RAID array should execute fast) | |
8523 | */ | |
1da177e4 LT |
8524 | cond_resched(); |
8525 | ||
ac7e50a3 XN |
8526 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
8527 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 8528 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 8529 | |
88202a0c | 8530 | if (currspeed > speed_min(mddev)) { |
ac8fa419 | 8531 | if (currspeed > speed_max(mddev)) { |
c0e48521 | 8532 | msleep(500); |
1da177e4 LT |
8533 | goto repeat; |
8534 | } | |
ac8fa419 N |
8535 | if (!is_mddev_idle(mddev, 0)) { |
8536 | /* | |
8537 | * Give other IO more of a chance. | |
8538 | * The faster the devices, the less we wait. | |
8539 | */ | |
8540 | wait_event(mddev->recovery_wait, | |
8541 | !atomic_read(&mddev->recovery_active)); | |
8542 | } | |
1da177e4 LT |
8543 | } |
8544 | } | |
9d48739e N |
8545 | pr_info("md: %s: %s %s.\n",mdname(mddev), desc, |
8546 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
8547 | ? "interrupted" : "done"); | |
1da177e4 LT |
8548 | /* |
8549 | * this also signals 'finished resyncing' to md_stop | |
8550 | */ | |
7c2c57c9 | 8551 | blk_finish_plug(&plug); |
1da177e4 LT |
8552 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
8553 | ||
5ed1df2e N |
8554 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8555 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
1217e1d1 | 8556 | mddev->curr_resync > 3) { |
5ed1df2e N |
8557 | mddev->curr_resync_completed = mddev->curr_resync; |
8558 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
8559 | } | |
09314799 | 8560 | mddev->pers->sync_request(mddev, max_sectors, &skipped); |
1da177e4 | 8561 | |
dfc70645 | 8562 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
1217e1d1 | 8563 | mddev->curr_resync > 3) { |
5fd6c1dc N |
8564 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
8565 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8566 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
9d48739e N |
8567 | pr_debug("md: checkpointing %s of %s.\n", |
8568 | desc, mdname(mddev)); | |
0a19caab | 8569 | if (test_bit(MD_RECOVERY_ERROR, |
8570 | &mddev->recovery)) | |
8571 | mddev->recovery_cp = | |
8572 | mddev->curr_resync_completed; | |
8573 | else | |
8574 | mddev->recovery_cp = | |
8575 | mddev->curr_resync; | |
5fd6c1dc N |
8576 | } |
8577 | } else | |
8578 | mddev->recovery_cp = MaxSector; | |
8579 | } else { | |
8580 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
8581 | mddev->curr_resync = MaxSector; | |
db0505d3 N |
8582 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8583 | test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) { | |
8584 | rcu_read_lock(); | |
8585 | rdev_for_each_rcu(rdev, mddev) | |
8586 | if (rdev->raid_disk >= 0 && | |
8587 | mddev->delta_disks >= 0 && | |
8588 | !test_bit(Journal, &rdev->flags) && | |
8589 | !test_bit(Faulty, &rdev->flags) && | |
8590 | !test_bit(In_sync, &rdev->flags) && | |
8591 | rdev->recovery_offset < mddev->curr_resync) | |
8592 | rdev->recovery_offset = mddev->curr_resync; | |
8593 | rcu_read_unlock(); | |
8594 | } | |
5fd6c1dc | 8595 | } |
1da177e4 | 8596 | } |
db91ff55 | 8597 | skip: |
bb8bf15b GJ |
8598 | /* set CHANGE_PENDING here since maybe another update is needed, |
8599 | * so other nodes are informed. It should be harmless for normal | |
8600 | * raid */ | |
2953079c SL |
8601 | set_mask_bits(&mddev->sb_flags, 0, |
8602 | BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS)); | |
c186b128 | 8603 | |
8876391e BC |
8604 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8605 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8606 | mddev->delta_disks > 0 && | |
8607 | mddev->pers->finish_reshape && | |
8608 | mddev->pers->size && | |
8609 | mddev->queue) { | |
8610 | mddev_lock_nointr(mddev); | |
8611 | md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0)); | |
8612 | mddev_unlock(mddev); | |
8613 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
8614 | revalidate_disk(mddev->gendisk); | |
8615 | } | |
8616 | ||
23da422b | 8617 | spin_lock(&mddev->lock); |
c07b70ad N |
8618 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
8619 | /* We completed so min/max setting can be forgotten if used. */ | |
8620 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8621 | mddev->resync_min = 0; | |
8622 | mddev->resync_max = MaxSector; | |
8623 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
8624 | mddev->resync_min = mddev->curr_resync_completed; | |
f7851be7 | 8625 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); |
1da177e4 | 8626 | mddev->curr_resync = 0; |
23da422b N |
8627 | spin_unlock(&mddev->lock); |
8628 | ||
1da177e4 | 8629 | wake_up(&resync_wait); |
1da177e4 | 8630 | md_wakeup_thread(mddev->thread); |
c6207277 | 8631 | return; |
1da177e4 | 8632 | } |
29269553 | 8633 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 8634 | |
746d3207 N |
8635 | static int remove_and_add_spares(struct mddev *mddev, |
8636 | struct md_rdev *this) | |
b4c4c7b8 | 8637 | { |
3cb03002 | 8638 | struct md_rdev *rdev; |
b4c4c7b8 | 8639 | int spares = 0; |
f2a371c5 | 8640 | int removed = 0; |
d787be40 | 8641 | bool remove_some = false; |
b4c4c7b8 | 8642 | |
39772f0a N |
8643 | if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
8644 | /* Mustn't remove devices when resync thread is running */ | |
8645 | return 0; | |
8646 | ||
d787be40 N |
8647 | rdev_for_each(rdev, mddev) { |
8648 | if ((this == NULL || rdev == this) && | |
8649 | rdev->raid_disk >= 0 && | |
8650 | !test_bit(Blocked, &rdev->flags) && | |
8651 | test_bit(Faulty, &rdev->flags) && | |
8652 | atomic_read(&rdev->nr_pending)==0) { | |
8653 | /* Faulty non-Blocked devices with nr_pending == 0 | |
8654 | * never get nr_pending incremented, | |
8655 | * never get Faulty cleared, and never get Blocked set. | |
8656 | * So we can synchronize_rcu now rather than once per device | |
8657 | */ | |
8658 | remove_some = true; | |
8659 | set_bit(RemoveSynchronized, &rdev->flags); | |
8660 | } | |
8661 | } | |
8662 | ||
8663 | if (remove_some) | |
8664 | synchronize_rcu(); | |
8665 | rdev_for_each(rdev, mddev) { | |
746d3207 N |
8666 | if ((this == NULL || rdev == this) && |
8667 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 8668 | !test_bit(Blocked, &rdev->flags) && |
d787be40 | 8669 | ((test_bit(RemoveSynchronized, &rdev->flags) || |
f2076e7d SL |
8670 | (!test_bit(In_sync, &rdev->flags) && |
8671 | !test_bit(Journal, &rdev->flags))) && | |
d787be40 | 8672 | atomic_read(&rdev->nr_pending)==0)) { |
b4c4c7b8 | 8673 | if (mddev->pers->hot_remove_disk( |
b8321b68 | 8674 | mddev, rdev) == 0) { |
36fad858 | 8675 | sysfs_unlink_rdev(mddev, rdev); |
011abdc9 | 8676 | rdev->saved_raid_disk = rdev->raid_disk; |
b4c4c7b8 | 8677 | rdev->raid_disk = -1; |
f2a371c5 | 8678 | removed++; |
b4c4c7b8 N |
8679 | } |
8680 | } | |
d787be40 N |
8681 | if (remove_some && test_bit(RemoveSynchronized, &rdev->flags)) |
8682 | clear_bit(RemoveSynchronized, &rdev->flags); | |
8683 | } | |
8684 | ||
90584fc9 JB |
8685 | if (removed && mddev->kobj.sd) |
8686 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 8687 | |
2910ff17 | 8688 | if (this && removed) |
746d3207 N |
8689 | goto no_add; |
8690 | ||
dafb20fa | 8691 | rdev_for_each(rdev, mddev) { |
2910ff17 GR |
8692 | if (this && this != rdev) |
8693 | continue; | |
dbb64f86 GR |
8694 | if (test_bit(Candidate, &rdev->flags)) |
8695 | continue; | |
7bfec5f3 N |
8696 | if (rdev->raid_disk >= 0 && |
8697 | !test_bit(In_sync, &rdev->flags) && | |
f2076e7d | 8698 | !test_bit(Journal, &rdev->flags) && |
7bfec5f3 N |
8699 | !test_bit(Faulty, &rdev->flags)) |
8700 | spares++; | |
7ceb17e8 N |
8701 | if (rdev->raid_disk >= 0) |
8702 | continue; | |
8703 | if (test_bit(Faulty, &rdev->flags)) | |
8704 | continue; | |
f6b6ec5c SL |
8705 | if (!test_bit(Journal, &rdev->flags)) { |
8706 | if (mddev->ro && | |
8707 | ! (rdev->saved_raid_disk >= 0 && | |
8708 | !test_bit(Bitmap_sync, &rdev->flags))) | |
8709 | continue; | |
7ceb17e8 | 8710 | |
f6b6ec5c SL |
8711 | rdev->recovery_offset = 0; |
8712 | } | |
7ceb17e8 N |
8713 | if (mddev->pers-> |
8714 | hot_add_disk(mddev, rdev) == 0) { | |
8715 | if (sysfs_link_rdev(mddev, rdev)) | |
8716 | /* failure here is OK */; | |
f6b6ec5c SL |
8717 | if (!test_bit(Journal, &rdev->flags)) |
8718 | spares++; | |
7ceb17e8 | 8719 | md_new_event(mddev); |
2953079c | 8720 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
dfc70645 | 8721 | } |
b4c4c7b8 | 8722 | } |
746d3207 | 8723 | no_add: |
6dafab6b | 8724 | if (removed) |
2953079c | 8725 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
b4c4c7b8 N |
8726 | return spares; |
8727 | } | |
7ebc0be7 | 8728 | |
ac05f256 N |
8729 | static void md_start_sync(struct work_struct *ws) |
8730 | { | |
8731 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
c186b128 | 8732 | |
ac05f256 N |
8733 | mddev->sync_thread = md_register_thread(md_do_sync, |
8734 | mddev, | |
8735 | "resync"); | |
8736 | if (!mddev->sync_thread) { | |
9d48739e N |
8737 | pr_warn("%s: could not start resync thread...\n", |
8738 | mdname(mddev)); | |
ac05f256 N |
8739 | /* leave the spares where they are, it shouldn't hurt */ |
8740 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8741 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8742 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8743 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
8744 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8745 | wake_up(&resync_wait); |
ac05f256 N |
8746 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8747 | &mddev->recovery)) | |
8748 | if (mddev->sysfs_action) | |
8749 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8750 | } else | |
8751 | md_wakeup_thread(mddev->sync_thread); | |
8752 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8753 | md_new_event(mddev); | |
8754 | } | |
8755 | ||
1da177e4 LT |
8756 | /* |
8757 | * This routine is regularly called by all per-raid-array threads to | |
8758 | * deal with generic issues like resync and super-block update. | |
8759 | * Raid personalities that don't have a thread (linear/raid0) do not | |
8760 | * need this as they never do any recovery or update the superblock. | |
8761 | * | |
8762 | * It does not do any resync itself, but rather "forks" off other threads | |
8763 | * to do that as needed. | |
8764 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
8765 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 8766 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
8767 | * and wakeups up this thread which will reap the thread and finish up. |
8768 | * This thread also removes any faulty devices (with nr_pending == 0). | |
8769 | * | |
8770 | * The overall approach is: | |
8771 | * 1/ if the superblock needs updating, update it. | |
8772 | * 2/ If a recovery thread is running, don't do anything else. | |
8773 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
8774 | * 4/ If there are any faulty devices, remove them. | |
8775 | * 5/ If array is degraded, try to add spares devices | |
8776 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
8777 | */ | |
fd01b88c | 8778 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 8779 | { |
68866e42 JB |
8780 | if (mddev->suspended) |
8781 | return; | |
8782 | ||
5f40402d | 8783 | if (mddev->bitmap) |
aa5cbd10 | 8784 | bitmap_daemon_work(mddev); |
1da177e4 | 8785 | |
fca4d848 | 8786 | if (signal_pending(current)) { |
31a59e34 | 8787 | if (mddev->pers->sync_request && !mddev->external) { |
9d48739e N |
8788 | pr_debug("md: %s in immediate safe mode\n", |
8789 | mdname(mddev)); | |
fca4d848 N |
8790 | mddev->safemode = 2; |
8791 | } | |
8792 | flush_signals(current); | |
8793 | } | |
8794 | ||
c89a8eee N |
8795 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
8796 | return; | |
1da177e4 | 8797 | if ( ! ( |
2953079c | 8798 | (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) || |
1da177e4 | 8799 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 8800 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 8801 | (mddev->external == 0 && mddev->safemode == 1) || |
4ad23a97 | 8802 | (mddev->safemode == 2 |
fca4d848 | 8803 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) |
1da177e4 LT |
8804 | )) |
8805 | return; | |
fca4d848 | 8806 | |
df5b89b3 | 8807 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 8808 | int spares = 0; |
fca4d848 | 8809 | |
afc1f55c | 8810 | if (!mddev->external && mddev->safemode == 1) |
33182d15 N |
8811 | mddev->safemode = 0; |
8812 | ||
c89a8eee | 8813 | if (mddev->ro) { |
ab16bfc7 NB |
8814 | struct md_rdev *rdev; |
8815 | if (!mddev->external && mddev->in_sync) | |
8816 | /* 'Blocked' flag not needed as failed devices | |
8817 | * will be recorded if array switched to read/write. | |
8818 | * Leaving it set will prevent the device | |
8819 | * from being removed. | |
8820 | */ | |
8821 | rdev_for_each(rdev, mddev) | |
8822 | clear_bit(Blocked, &rdev->flags); | |
7ceb17e8 N |
8823 | /* On a read-only array we can: |
8824 | * - remove failed devices | |
8825 | * - add already-in_sync devices if the array itself | |
8826 | * is in-sync. | |
8827 | * As we only add devices that are already in-sync, | |
8828 | * we can activate the spares immediately. | |
c89a8eee | 8829 | */ |
7ceb17e8 | 8830 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
8831 | /* There is no thread, but we need to call |
8832 | * ->spare_active and clear saved_raid_disk | |
8833 | */ | |
2ac295a5 | 8834 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 | 8835 | md_reap_sync_thread(mddev); |
a4a3d26d | 8836 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
8313b8e5 | 8837 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2953079c | 8838 | clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags); |
c89a8eee N |
8839 | goto unlock; |
8840 | } | |
8841 | ||
659b254f GJ |
8842 | if (mddev_is_clustered(mddev)) { |
8843 | struct md_rdev *rdev; | |
8844 | /* kick the device if another node issued a | |
8845 | * remove disk. | |
8846 | */ | |
8847 | rdev_for_each(rdev, mddev) { | |
8848 | if (test_and_clear_bit(ClusterRemove, &rdev->flags) && | |
8849 | rdev->raid_disk < 0) | |
8850 | md_kick_rdev_from_array(rdev); | |
8851 | } | |
8852 | } | |
8853 | ||
6497709b | 8854 | if (!mddev->external && !mddev->in_sync) { |
85572d7c | 8855 | spin_lock(&mddev->lock); |
6497709b | 8856 | set_in_sync(mddev); |
85572d7c | 8857 | spin_unlock(&mddev->lock); |
fca4d848 | 8858 | } |
fca4d848 | 8859 | |
2953079c | 8860 | if (mddev->sb_flags) |
850b2b42 | 8861 | md_update_sb(mddev, 0); |
06d91a5f | 8862 | |
1da177e4 LT |
8863 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
8864 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
8865 | /* resync/recovery still happening */ | |
8866 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8867 | goto unlock; | |
8868 | } | |
8869 | if (mddev->sync_thread) { | |
a91d5ac0 | 8870 | md_reap_sync_thread(mddev); |
1da177e4 LT |
8871 | goto unlock; |
8872 | } | |
72a23c21 NB |
8873 | /* Set RUNNING before clearing NEEDED to avoid |
8874 | * any transients in the value of "sync_action". | |
8875 | */ | |
72f36d59 | 8876 | mddev->curr_resync_completed = 0; |
23da422b | 8877 | spin_lock(&mddev->lock); |
72a23c21 | 8878 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 8879 | spin_unlock(&mddev->lock); |
24dd469d N |
8880 | /* Clear some bits that don't mean anything, but |
8881 | * might be left set | |
8882 | */ | |
24dd469d N |
8883 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8884 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 8885 | |
ed209584 N |
8886 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
8887 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 8888 | goto not_running; |
1da177e4 LT |
8889 | /* no recovery is running. |
8890 | * remove any failed drives, then | |
8891 | * add spares if possible. | |
72f36d59 | 8892 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
8893 | * the personality to fail the re-add. |
8894 | */ | |
1da177e4 | 8895 | |
b4c4c7b8 | 8896 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
8897 | if (mddev->pers->check_reshape == NULL || |
8898 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 8899 | /* Cannot proceed */ |
ac05f256 | 8900 | goto not_running; |
b4c4c7b8 | 8901 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 8902 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 8903 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
8904 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8905 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 8906 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 8907 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8908 | } else if (mddev->recovery_cp < MaxSector) { |
8909 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 8910 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8911 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
8912 | /* nothing to be done ... */ | |
ac05f256 | 8913 | goto not_running; |
24dd469d | 8914 | |
1da177e4 | 8915 | if (mddev->pers->sync_request) { |
ef99bf48 | 8916 | if (spares) { |
a654b9d8 N |
8917 | /* We are adding a device or devices to an array |
8918 | * which has the bitmap stored on all devices. | |
8919 | * So make sure all bitmap pages get written | |
8920 | */ | |
8921 | bitmap_write_all(mddev->bitmap); | |
8922 | } | |
ac05f256 N |
8923 | INIT_WORK(&mddev->del_work, md_start_sync); |
8924 | queue_work(md_misc_wq, &mddev->del_work); | |
8925 | goto unlock; | |
1da177e4 | 8926 | } |
ac05f256 | 8927 | not_running: |
72a23c21 NB |
8928 | if (!mddev->sync_thread) { |
8929 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8930 | wake_up(&resync_wait); |
72a23c21 NB |
8931 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8932 | &mddev->recovery)) | |
0c3573f1 | 8933 | if (mddev->sysfs_action) |
00bcb4ac | 8934 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 8935 | } |
ac05f256 N |
8936 | unlock: |
8937 | wake_up(&mddev->sb_wait); | |
1da177e4 | 8938 | mddev_unlock(mddev); |
35bfc521 N |
8939 | } else if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) { |
8940 | /* Write superblock - thread that called mddev_suspend() | |
8941 | * holds reconfig_mutex for us. | |
8942 | */ | |
8943 | set_bit(MD_UPDATING_SB, &mddev->flags); | |
8944 | smp_mb__after_atomic(); | |
8945 | if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags)) | |
8946 | md_update_sb(mddev, 0); | |
8947 | clear_bit_unlock(MD_UPDATING_SB, &mddev->flags); | |
8948 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
8949 | } |
8950 | } | |
6c144d31 | 8951 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8952 | |
a91d5ac0 JB |
8953 | void md_reap_sync_thread(struct mddev *mddev) |
8954 | { | |
8955 | struct md_rdev *rdev; | |
8956 | ||
8957 | /* resync has finished, collect result */ | |
8958 | md_unregister_thread(&mddev->sync_thread); | |
8959 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8960 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8961 | /* success...*/ | |
8962 | /* activate any spares */ | |
8963 | if (mddev->pers->spare_active(mddev)) { | |
8964 | sysfs_notify(&mddev->kobj, NULL, | |
8965 | "degraded"); | |
2953079c | 8966 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
a91d5ac0 JB |
8967 | } |
8968 | } | |
8969 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
8970 | mddev->pers->finish_reshape) | |
8971 | mddev->pers->finish_reshape(mddev); | |
8972 | ||
8973 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8974 | * information must be scrapped. |
a91d5ac0 | 8975 | */ |
f466722c N |
8976 | if (!mddev->degraded) |
8977 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8978 | rdev->saved_raid_disk = -1; |
8979 | ||
8980 | md_update_sb(mddev, 1); | |
2953079c | 8981 | /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can |
bb8bf15b GJ |
8982 | * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by |
8983 | * clustered raid */ | |
8984 | if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags)) | |
8985 | md_cluster_ops->resync_finish(mddev); | |
a91d5ac0 | 8986 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
ea358cd0 | 8987 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
a91d5ac0 JB |
8988 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8989 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8990 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8991 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8992 | wake_up(&resync_wait); |
a91d5ac0 JB |
8993 | /* flag recovery needed just to double check */ |
8994 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8995 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8996 | md_new_event(mddev); | |
8997 | if (mddev->event_work.func) | |
8998 | queue_work(md_misc_wq, &mddev->event_work); | |
8999 | } | |
6c144d31 | 9000 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 9001 | |
fd01b88c | 9002 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 9003 | { |
00bcb4ac | 9004 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 9005 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
9006 | !test_bit(Blocked, &rdev->flags) && |
9007 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
9008 | msecs_to_jiffies(5000)); |
9009 | rdev_dec_pending(rdev, mddev); | |
9010 | } | |
9011 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
9012 | ||
c6563a8c N |
9013 | void md_finish_reshape(struct mddev *mddev) |
9014 | { | |
9015 | /* called be personality module when reshape completes. */ | |
9016 | struct md_rdev *rdev; | |
9017 | ||
9018 | rdev_for_each(rdev, mddev) { | |
9019 | if (rdev->data_offset > rdev->new_data_offset) | |
9020 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
9021 | else | |
9022 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
9023 | rdev->data_offset = rdev->new_data_offset; | |
9024 | } | |
9025 | } | |
9026 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 | 9027 | |
fc974ee2 | 9028 | /* Bad block management */ |
2230dfe4 | 9029 | |
fc974ee2 | 9030 | /* Returns 1 on success, 0 on failure */ |
3cb03002 | 9031 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 9032 | int is_new) |
2230dfe4 | 9033 | { |
85ad1d13 | 9034 | struct mddev *mddev = rdev->mddev; |
c6563a8c N |
9035 | int rv; |
9036 | if (is_new) | |
9037 | s += rdev->new_data_offset; | |
9038 | else | |
9039 | s += rdev->data_offset; | |
fc974ee2 VV |
9040 | rv = badblocks_set(&rdev->badblocks, s, sectors, 0); |
9041 | if (rv == 0) { | |
2230dfe4 | 9042 | /* Make sure they get written out promptly */ |
35b785f7 TM |
9043 | if (test_bit(ExternalBbl, &rdev->flags)) |
9044 | sysfs_notify(&rdev->kobj, NULL, | |
9045 | "unacknowledged_bad_blocks"); | |
8bd2f0a0 | 9046 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2953079c SL |
9047 | set_mask_bits(&mddev->sb_flags, 0, |
9048 | BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING)); | |
2230dfe4 | 9049 | md_wakeup_thread(rdev->mddev->thread); |
fc974ee2 VV |
9050 | return 1; |
9051 | } else | |
9052 | return 0; | |
2230dfe4 N |
9053 | } |
9054 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
9055 | ||
c6563a8c N |
9056 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
9057 | int is_new) | |
2230dfe4 | 9058 | { |
35b785f7 | 9059 | int rv; |
c6563a8c N |
9060 | if (is_new) |
9061 | s += rdev->new_data_offset; | |
9062 | else | |
9063 | s += rdev->data_offset; | |
35b785f7 TM |
9064 | rv = badblocks_clear(&rdev->badblocks, s, sectors); |
9065 | if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags)) | |
9066 | sysfs_notify(&rdev->kobj, NULL, "bad_blocks"); | |
9067 | return rv; | |
2230dfe4 N |
9068 | } |
9069 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
9070 | ||
75c96f85 AB |
9071 | static int md_notify_reboot(struct notifier_block *this, |
9072 | unsigned long code, void *x) | |
1da177e4 LT |
9073 | { |
9074 | struct list_head *tmp; | |
fd01b88c | 9075 | struct mddev *mddev; |
2dba6a91 | 9076 | int need_delay = 0; |
1da177e4 | 9077 | |
c744a65c N |
9078 | for_each_mddev(mddev, tmp) { |
9079 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
9080 | if (mddev->pers) |
9081 | __md_stop_writes(mddev); | |
0f62fb22 N |
9082 | if (mddev->persistent) |
9083 | mddev->safemode = 2; | |
c744a65c | 9084 | mddev_unlock(mddev); |
2dba6a91 | 9085 | } |
c744a65c | 9086 | need_delay = 1; |
1da177e4 | 9087 | } |
c744a65c N |
9088 | /* |
9089 | * certain more exotic SCSI devices are known to be | |
9090 | * volatile wrt too early system reboots. While the | |
9091 | * right place to handle this issue is the given | |
9092 | * driver, we do want to have a safe RAID driver ... | |
9093 | */ | |
9094 | if (need_delay) | |
9095 | mdelay(1000*1); | |
9096 | ||
1da177e4 LT |
9097 | return NOTIFY_DONE; |
9098 | } | |
9099 | ||
75c96f85 | 9100 | static struct notifier_block md_notifier = { |
1da177e4 LT |
9101 | .notifier_call = md_notify_reboot, |
9102 | .next = NULL, | |
9103 | .priority = INT_MAX, /* before any real devices */ | |
9104 | }; | |
9105 | ||
9106 | static void md_geninit(void) | |
9107 | { | |
36a4e1fe | 9108 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 9109 | |
c7705f34 | 9110 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
9111 | } |
9112 | ||
75c96f85 | 9113 | static int __init md_init(void) |
1da177e4 | 9114 | { |
e804ac78 TH |
9115 | int ret = -ENOMEM; |
9116 | ||
ada609ee | 9117 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
9118 | if (!md_wq) |
9119 | goto err_wq; | |
9120 | ||
9121 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
9122 | if (!md_misc_wq) | |
9123 | goto err_misc_wq; | |
9124 | ||
9125 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
9126 | goto err_md; | |
9127 | ||
9128 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
9129 | goto err_mdp; | |
9130 | mdp_major = ret; | |
9131 | ||
af5628f0 | 9132 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
9133 | md_probe, NULL, NULL); |
9134 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
9135 | md_probe, NULL, NULL); |
9136 | ||
1da177e4 | 9137 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 9138 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
9139 | |
9140 | md_geninit(); | |
d710e138 | 9141 | return 0; |
1da177e4 | 9142 | |
e804ac78 TH |
9143 | err_mdp: |
9144 | unregister_blkdev(MD_MAJOR, "md"); | |
9145 | err_md: | |
9146 | destroy_workqueue(md_misc_wq); | |
9147 | err_misc_wq: | |
9148 | destroy_workqueue(md_wq); | |
9149 | err_wq: | |
9150 | return ret; | |
9151 | } | |
1da177e4 | 9152 | |
70bcecdb | 9153 | static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev) |
1d7e3e96 | 9154 | { |
70bcecdb GR |
9155 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
9156 | struct md_rdev *rdev2; | |
9157 | int role, ret; | |
9158 | char b[BDEVNAME_SIZE]; | |
1d7e3e96 | 9159 | |
818da59f GJ |
9160 | /* |
9161 | * If size is changed in another node then we need to | |
9162 | * do resize as well. | |
9163 | */ | |
9164 | if (mddev->dev_sectors != le64_to_cpu(sb->size)) { | |
9165 | ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size)); | |
9166 | if (ret) | |
9167 | pr_info("md-cluster: resize failed\n"); | |
9168 | else | |
9169 | bitmap_update_sb(mddev->bitmap); | |
9170 | } | |
9171 | ||
70bcecdb GR |
9172 | /* Check for change of roles in the active devices */ |
9173 | rdev_for_each(rdev2, mddev) { | |
9174 | if (test_bit(Faulty, &rdev2->flags)) | |
9175 | continue; | |
9176 | ||
9177 | /* Check if the roles changed */ | |
9178 | role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]); | |
dbb64f86 GR |
9179 | |
9180 | if (test_bit(Candidate, &rdev2->flags)) { | |
9181 | if (role == 0xfffe) { | |
9182 | pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b)); | |
9183 | md_kick_rdev_from_array(rdev2); | |
9184 | continue; | |
9185 | } | |
9186 | else | |
9187 | clear_bit(Candidate, &rdev2->flags); | |
9188 | } | |
9189 | ||
70bcecdb GR |
9190 | if (role != rdev2->raid_disk) { |
9191 | /* got activated */ | |
9192 | if (rdev2->raid_disk == -1 && role != 0xffff) { | |
9193 | rdev2->saved_raid_disk = role; | |
9194 | ret = remove_and_add_spares(mddev, rdev2); | |
9195 | pr_info("Activated spare: %s\n", | |
9d48739e | 9196 | bdevname(rdev2->bdev,b)); |
a578183e GJ |
9197 | /* wakeup mddev->thread here, so array could |
9198 | * perform resync with the new activated disk */ | |
9199 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
9200 | md_wakeup_thread(mddev->thread); | |
9201 | ||
70bcecdb GR |
9202 | } |
9203 | /* device faulty | |
9204 | * We just want to do the minimum to mark the disk | |
9205 | * as faulty. The recovery is performed by the | |
9206 | * one who initiated the error. | |
9207 | */ | |
9208 | if ((role == 0xfffe) || (role == 0xfffd)) { | |
9209 | md_error(mddev, rdev2); | |
9210 | clear_bit(Blocked, &rdev2->flags); | |
9211 | } | |
9212 | } | |
1d7e3e96 | 9213 | } |
70bcecdb | 9214 | |
28c1b9fd GR |
9215 | if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) |
9216 | update_raid_disks(mddev, le32_to_cpu(sb->raid_disks)); | |
70bcecdb GR |
9217 | |
9218 | /* Finally set the event to be up to date */ | |
9219 | mddev->events = le64_to_cpu(sb->events); | |
9220 | } | |
9221 | ||
9222 | static int read_rdev(struct mddev *mddev, struct md_rdev *rdev) | |
9223 | { | |
9224 | int err; | |
9225 | struct page *swapout = rdev->sb_page; | |
9226 | struct mdp_superblock_1 *sb; | |
9227 | ||
9228 | /* Store the sb page of the rdev in the swapout temporary | |
9229 | * variable in case we err in the future | |
9230 | */ | |
9231 | rdev->sb_page = NULL; | |
7f0f0d87 N |
9232 | err = alloc_disk_sb(rdev); |
9233 | if (err == 0) { | |
9234 | ClearPageUptodate(rdev->sb_page); | |
9235 | rdev->sb_loaded = 0; | |
9236 | err = super_types[mddev->major_version]. | |
9237 | load_super(rdev, NULL, mddev->minor_version); | |
9238 | } | |
70bcecdb GR |
9239 | if (err < 0) { |
9240 | pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n", | |
9241 | __func__, __LINE__, rdev->desc_nr, err); | |
7f0f0d87 N |
9242 | if (rdev->sb_page) |
9243 | put_page(rdev->sb_page); | |
70bcecdb GR |
9244 | rdev->sb_page = swapout; |
9245 | rdev->sb_loaded = 1; | |
9246 | return err; | |
1d7e3e96 GR |
9247 | } |
9248 | ||
70bcecdb GR |
9249 | sb = page_address(rdev->sb_page); |
9250 | /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET | |
9251 | * is not set | |
9252 | */ | |
9253 | ||
9254 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)) | |
9255 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
9256 | ||
9257 | /* The other node finished recovery, call spare_active to set | |
9258 | * device In_sync and mddev->degraded | |
9259 | */ | |
9260 | if (rdev->recovery_offset == MaxSector && | |
9261 | !test_bit(In_sync, &rdev->flags) && | |
9262 | mddev->pers->spare_active(mddev)) | |
9263 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
9264 | ||
9265 | put_page(swapout); | |
9266 | return 0; | |
9267 | } | |
9268 | ||
9269 | void md_reload_sb(struct mddev *mddev, int nr) | |
9270 | { | |
9271 | struct md_rdev *rdev; | |
9272 | int err; | |
9273 | ||
9274 | /* Find the rdev */ | |
9275 | rdev_for_each_rcu(rdev, mddev) { | |
9276 | if (rdev->desc_nr == nr) | |
9277 | break; | |
9278 | } | |
9279 | ||
9280 | if (!rdev || rdev->desc_nr != nr) { | |
9281 | pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr); | |
9282 | return; | |
9283 | } | |
9284 | ||
9285 | err = read_rdev(mddev, rdev); | |
9286 | if (err < 0) | |
9287 | return; | |
9288 | ||
9289 | check_sb_changes(mddev, rdev); | |
9290 | ||
9291 | /* Read all rdev's to update recovery_offset */ | |
0ea9924a GJ |
9292 | rdev_for_each_rcu(rdev, mddev) { |
9293 | if (!test_bit(Faulty, &rdev->flags)) | |
9294 | read_rdev(mddev, rdev); | |
9295 | } | |
1d7e3e96 GR |
9296 | } |
9297 | EXPORT_SYMBOL(md_reload_sb); | |
9298 | ||
1da177e4 LT |
9299 | #ifndef MODULE |
9300 | ||
9301 | /* | |
9302 | * Searches all registered partitions for autorun RAID arrays | |
9303 | * at boot time. | |
9304 | */ | |
4d936ec1 | 9305 | |
5b1f5bc3 | 9306 | static DEFINE_MUTEX(detected_devices_mutex); |
4d936ec1 ME |
9307 | static LIST_HEAD(all_detected_devices); |
9308 | struct detected_devices_node { | |
9309 | struct list_head list; | |
9310 | dev_t dev; | |
9311 | }; | |
1da177e4 LT |
9312 | |
9313 | void md_autodetect_dev(dev_t dev) | |
9314 | { | |
4d936ec1 ME |
9315 | struct detected_devices_node *node_detected_dev; |
9316 | ||
9317 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
9318 | if (node_detected_dev) { | |
9319 | node_detected_dev->dev = dev; | |
5b1f5bc3 | 9320 | mutex_lock(&detected_devices_mutex); |
4d936ec1 | 9321 | list_add_tail(&node_detected_dev->list, &all_detected_devices); |
5b1f5bc3 | 9322 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 9323 | } |
1da177e4 LT |
9324 | } |
9325 | ||
1da177e4 LT |
9326 | static void autostart_arrays(int part) |
9327 | { | |
3cb03002 | 9328 | struct md_rdev *rdev; |
4d936ec1 ME |
9329 | struct detected_devices_node *node_detected_dev; |
9330 | dev_t dev; | |
9331 | int i_scanned, i_passed; | |
1da177e4 | 9332 | |
4d936ec1 ME |
9333 | i_scanned = 0; |
9334 | i_passed = 0; | |
1da177e4 | 9335 | |
9d48739e | 9336 | pr_info("md: Autodetecting RAID arrays.\n"); |
1da177e4 | 9337 | |
5b1f5bc3 | 9338 | mutex_lock(&detected_devices_mutex); |
4d936ec1 ME |
9339 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
9340 | i_scanned++; | |
9341 | node_detected_dev = list_entry(all_detected_devices.next, | |
9342 | struct detected_devices_node, list); | |
9343 | list_del(&node_detected_dev->list); | |
9344 | dev = node_detected_dev->dev; | |
9345 | kfree(node_detected_dev); | |
90bcf133 | 9346 | mutex_unlock(&detected_devices_mutex); |
df968c4e | 9347 | rdev = md_import_device(dev,0, 90); |
90bcf133 | 9348 | mutex_lock(&detected_devices_mutex); |
1da177e4 LT |
9349 | if (IS_ERR(rdev)) |
9350 | continue; | |
9351 | ||
403df478 | 9352 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 9353 | continue; |
403df478 | 9354 | |
d0fae18f | 9355 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 9356 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 9357 | i_passed++; |
1da177e4 | 9358 | } |
5b1f5bc3 | 9359 | mutex_unlock(&detected_devices_mutex); |
4d936ec1 | 9360 | |
9d48739e | 9361 | pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed); |
1da177e4 LT |
9362 | |
9363 | autorun_devices(part); | |
9364 | } | |
9365 | ||
fdee8ae4 | 9366 | #endif /* !MODULE */ |
1da177e4 LT |
9367 | |
9368 | static __exit void md_exit(void) | |
9369 | { | |
fd01b88c | 9370 | struct mddev *mddev; |
1da177e4 | 9371 | struct list_head *tmp; |
e2f23b60 | 9372 | int delay = 1; |
8ab5e4c1 | 9373 | |
af5628f0 | 9374 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 9375 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 9376 | |
3dbd8c2e | 9377 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
9378 | unregister_blkdev(mdp_major, "mdp"); |
9379 | unregister_reboot_notifier(&md_notifier); | |
9380 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
9381 | |
9382 | /* We cannot unload the modules while some process is | |
9383 | * waiting for us in select() or poll() - wake them up | |
9384 | */ | |
9385 | md_unloading = 1; | |
9386 | while (waitqueue_active(&md_event_waiters)) { | |
9387 | /* not safe to leave yet */ | |
9388 | wake_up(&md_event_waiters); | |
9389 | msleep(delay); | |
9390 | delay += delay; | |
9391 | } | |
1da177e4 | 9392 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 9393 | |
29ac4aa3 | 9394 | for_each_mddev(mddev, tmp) { |
1da177e4 | 9395 | export_array(mddev); |
9356863c | 9396 | mddev->ctime = 0; |
d3374825 | 9397 | mddev->hold_active = 0; |
9356863c N |
9398 | /* |
9399 | * for_each_mddev() will call mddev_put() at the end of each | |
9400 | * iteration. As the mddev is now fully clear, this will | |
9401 | * schedule the mddev for destruction by a workqueue, and the | |
9402 | * destroy_workqueue() below will wait for that to complete. | |
9403 | */ | |
1da177e4 | 9404 | } |
e804ac78 TH |
9405 | destroy_workqueue(md_misc_wq); |
9406 | destroy_workqueue(md_wq); | |
1da177e4 LT |
9407 | } |
9408 | ||
685784aa | 9409 | subsys_initcall(md_init); |
1da177e4 LT |
9410 | module_exit(md_exit) |
9411 | ||
e4dca7b7 | 9412 | static int get_ro(char *buffer, const struct kernel_param *kp) |
f91de92e N |
9413 | { |
9414 | return sprintf(buffer, "%d", start_readonly); | |
9415 | } | |
e4dca7b7 | 9416 | static int set_ro(const char *val, const struct kernel_param *kp) |
f91de92e | 9417 | { |
4c9309c0 | 9418 | return kstrtouint(val, 10, (unsigned int *)&start_readonly); |
f91de92e N |
9419 | } |
9420 | ||
80ca3a44 N |
9421 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
9422 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 9423 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
78b6350d | 9424 | module_param(create_on_open, bool, S_IRUSR|S_IWUSR); |
f91de92e | 9425 | |
1da177e4 | 9426 | MODULE_LICENSE("GPL"); |
0efb9e61 | 9427 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 9428 | MODULE_ALIAS("md"); |
72008652 | 9429 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |