]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | md.c : Multiple Devices driver for Linux | |
3 | Copyright (C) 1998, 1999, 2000 Ingo Molnar | |
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. | |
33 | */ | |
34 | ||
a6fb0934 | 35 | #include <linux/kthread.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
1da177e4 | 37 | #include <linux/sysctl.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
2a48fc0a | 39 | #include <linux/mutex.h> |
1da177e4 | 40 | #include <linux/buffer_head.h> /* for invalidate_bdev */ |
d7603b7e | 41 | #include <linux/poll.h> |
16f17b39 | 42 | #include <linux/ctype.h> |
e7d2860b | 43 | #include <linux/string.h> |
fb4d8c76 N |
44 | #include <linux/hdreg.h> |
45 | #include <linux/proc_fs.h> | |
46 | #include <linux/random.h> | |
47 | #include <linux/reboot.h> | |
32a7627c | 48 | #include <linux/file.h> |
aa98aa31 | 49 | #include <linux/compat.h> |
25570727 | 50 | #include <linux/delay.h> |
bff61975 N |
51 | #include <linux/raid/md_p.h> |
52 | #include <linux/raid/md_u.h> | |
5a0e3ad6 | 53 | #include <linux/slab.h> |
43b2e5d8 | 54 | #include "md.h" |
ef740c37 | 55 | #include "bitmap.h" |
1da177e4 LT |
56 | |
57 | #define DEBUG 0 | |
58 | #define dprintk(x...) ((void)(DEBUG && printk(x))) | |
59 | ||
1da177e4 | 60 | #ifndef MODULE |
d710e138 | 61 | static void autostart_arrays(int part); |
1da177e4 LT |
62 | #endif |
63 | ||
2604b703 | 64 | static LIST_HEAD(pers_list); |
1da177e4 LT |
65 | static DEFINE_SPINLOCK(pers_lock); |
66 | ||
5e56341d AB |
67 | static void md_print_devices(void); |
68 | ||
90b08710 | 69 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
70 | static struct workqueue_struct *md_wq; |
71 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 72 | |
5e56341d AB |
73 | #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } |
74 | ||
1e50915f RB |
75 | /* |
76 | * Default number of read corrections we'll attempt on an rdev | |
77 | * before ejecting it from the array. We divide the read error | |
78 | * count by 2 for every hour elapsed between read errors. | |
79 | */ | |
80 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
81 | /* |
82 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
83 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
84 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 85 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
86 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
87 | * speed limit - in case reconstruction slows down your system despite | |
88 | * idle IO detection. | |
89 | * | |
90 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 91 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
92 | */ |
93 | ||
94 | static int sysctl_speed_limit_min = 1000; | |
95 | static int sysctl_speed_limit_max = 200000; | |
88202a0c N |
96 | static inline int speed_min(mddev_t *mddev) |
97 | { | |
98 | return mddev->sync_speed_min ? | |
99 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
100 | } | |
101 | ||
102 | static inline int speed_max(mddev_t *mddev) | |
103 | { | |
104 | return mddev->sync_speed_max ? | |
105 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
106 | } | |
1da177e4 LT |
107 | |
108 | static struct ctl_table_header *raid_table_header; | |
109 | ||
110 | static ctl_table raid_table[] = { | |
111 | { | |
1da177e4 LT |
112 | .procname = "speed_limit_min", |
113 | .data = &sysctl_speed_limit_min, | |
114 | .maxlen = sizeof(int), | |
80ca3a44 | 115 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 116 | .proc_handler = proc_dointvec, |
1da177e4 LT |
117 | }, |
118 | { | |
1da177e4 LT |
119 | .procname = "speed_limit_max", |
120 | .data = &sysctl_speed_limit_max, | |
121 | .maxlen = sizeof(int), | |
80ca3a44 | 122 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 123 | .proc_handler = proc_dointvec, |
1da177e4 | 124 | }, |
894d2491 | 125 | { } |
1da177e4 LT |
126 | }; |
127 | ||
128 | static ctl_table raid_dir_table[] = { | |
129 | { | |
1da177e4 LT |
130 | .procname = "raid", |
131 | .maxlen = 0, | |
80ca3a44 | 132 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
133 | .child = raid_table, |
134 | }, | |
894d2491 | 135 | { } |
1da177e4 LT |
136 | }; |
137 | ||
138 | static ctl_table raid_root_table[] = { | |
139 | { | |
1da177e4 LT |
140 | .procname = "dev", |
141 | .maxlen = 0, | |
142 | .mode = 0555, | |
143 | .child = raid_dir_table, | |
144 | }, | |
894d2491 | 145 | { } |
1da177e4 LT |
146 | }; |
147 | ||
83d5cde4 | 148 | static const struct block_device_operations md_fops; |
1da177e4 | 149 | |
f91de92e N |
150 | static int start_readonly; |
151 | ||
a167f663 N |
152 | /* bio_clone_mddev |
153 | * like bio_clone, but with a local bio set | |
154 | */ | |
155 | ||
156 | static void mddev_bio_destructor(struct bio *bio) | |
157 | { | |
158 | mddev_t *mddev, **mddevp; | |
159 | ||
160 | mddevp = (void*)bio; | |
161 | mddev = mddevp[-1]; | |
162 | ||
163 | bio_free(bio, mddev->bio_set); | |
164 | } | |
165 | ||
166 | struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, | |
167 | mddev_t *mddev) | |
168 | { | |
169 | struct bio *b; | |
170 | mddev_t **mddevp; | |
171 | ||
172 | if (!mddev || !mddev->bio_set) | |
173 | return bio_alloc(gfp_mask, nr_iovecs); | |
174 | ||
175 | b = bio_alloc_bioset(gfp_mask, nr_iovecs, | |
176 | mddev->bio_set); | |
177 | if (!b) | |
178 | return NULL; | |
179 | mddevp = (void*)b; | |
180 | mddevp[-1] = mddev; | |
181 | b->bi_destructor = mddev_bio_destructor; | |
182 | return b; | |
183 | } | |
184 | EXPORT_SYMBOL_GPL(bio_alloc_mddev); | |
185 | ||
186 | struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, | |
187 | mddev_t *mddev) | |
188 | { | |
189 | struct bio *b; | |
190 | mddev_t **mddevp; | |
191 | ||
192 | if (!mddev || !mddev->bio_set) | |
193 | return bio_clone(bio, gfp_mask); | |
194 | ||
195 | b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, | |
196 | mddev->bio_set); | |
197 | if (!b) | |
198 | return NULL; | |
199 | mddevp = (void*)b; | |
200 | mddevp[-1] = mddev; | |
201 | b->bi_destructor = mddev_bio_destructor; | |
202 | __bio_clone(b, bio); | |
203 | if (bio_integrity(bio)) { | |
204 | int ret; | |
205 | ||
206 | ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set); | |
207 | ||
208 | if (ret < 0) { | |
209 | bio_put(b); | |
210 | return NULL; | |
211 | } | |
212 | } | |
213 | ||
214 | return b; | |
215 | } | |
216 | EXPORT_SYMBOL_GPL(bio_clone_mddev); | |
217 | ||
d7603b7e N |
218 | /* |
219 | * We have a system wide 'event count' that is incremented | |
220 | * on any 'interesting' event, and readers of /proc/mdstat | |
221 | * can use 'poll' or 'select' to find out when the event | |
222 | * count increases. | |
223 | * | |
224 | * Events are: | |
225 | * start array, stop array, error, add device, remove device, | |
226 | * start build, activate spare | |
227 | */ | |
2989ddbd | 228 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 229 | static atomic_t md_event_count; |
29269553 | 230 | void md_new_event(mddev_t *mddev) |
d7603b7e N |
231 | { |
232 | atomic_inc(&md_event_count); | |
233 | wake_up(&md_event_waiters); | |
234 | } | |
29269553 | 235 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 236 | |
c331eb04 N |
237 | /* Alternate version that can be called from interrupts |
238 | * when calling sysfs_notify isn't needed. | |
239 | */ | |
05381954 | 240 | static void md_new_event_inintr(mddev_t *mddev) |
c331eb04 N |
241 | { |
242 | atomic_inc(&md_event_count); | |
243 | wake_up(&md_event_waiters); | |
244 | } | |
245 | ||
1da177e4 LT |
246 | /* |
247 | * Enables to iterate over all existing md arrays | |
248 | * all_mddevs_lock protects this list. | |
249 | */ | |
250 | static LIST_HEAD(all_mddevs); | |
251 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
252 | ||
253 | ||
254 | /* | |
255 | * iterates through all used mddevs in the system. | |
256 | * We take care to grab the all_mddevs_lock whenever navigating | |
257 | * the list, and to always hold a refcount when unlocked. | |
258 | * Any code which breaks out of this loop while own | |
259 | * a reference to the current mddev and must mddev_put it. | |
260 | */ | |
29ac4aa3 | 261 | #define for_each_mddev(mddev,tmp) \ |
1da177e4 LT |
262 | \ |
263 | for (({ spin_lock(&all_mddevs_lock); \ | |
264 | tmp = all_mddevs.next; \ | |
265 | mddev = NULL;}); \ | |
266 | ({ if (tmp != &all_mddevs) \ | |
267 | mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ | |
268 | spin_unlock(&all_mddevs_lock); \ | |
269 | if (mddev) mddev_put(mddev); \ | |
270 | mddev = list_entry(tmp, mddev_t, all_mddevs); \ | |
271 | tmp != &all_mddevs;}); \ | |
272 | ({ spin_lock(&all_mddevs_lock); \ | |
273 | tmp = tmp->next;}) \ | |
274 | ) | |
275 | ||
276 | ||
409c57f3 N |
277 | /* Rather than calling directly into the personality make_request function, |
278 | * IO requests come here first so that we can check if the device is | |
279 | * being suspended pending a reconfiguration. | |
280 | * We hold a refcount over the call to ->make_request. By the time that | |
281 | * call has finished, the bio has been linked into some internal structure | |
282 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
283 | */ | |
284 | static int md_make_request(struct request_queue *q, struct bio *bio) | |
1da177e4 | 285 | { |
49077326 | 286 | const int rw = bio_data_dir(bio); |
409c57f3 N |
287 | mddev_t *mddev = q->queuedata; |
288 | int rv; | |
49077326 | 289 | int cpu; |
e91ece55 | 290 | unsigned int sectors; |
49077326 | 291 | |
0ca69886 N |
292 | if (mddev == NULL || mddev->pers == NULL |
293 | || !mddev->ready) { | |
409c57f3 N |
294 | bio_io_error(bio); |
295 | return 0; | |
296 | } | |
0ca69886 | 297 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 298 | rcu_read_lock(); |
e9c7469b | 299 | if (mddev->suspended) { |
409c57f3 N |
300 | DEFINE_WAIT(__wait); |
301 | for (;;) { | |
302 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
303 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 304 | if (!mddev->suspended) |
409c57f3 N |
305 | break; |
306 | rcu_read_unlock(); | |
307 | schedule(); | |
308 | rcu_read_lock(); | |
309 | } | |
310 | finish_wait(&mddev->sb_wait, &__wait); | |
311 | } | |
312 | atomic_inc(&mddev->active_io); | |
313 | rcu_read_unlock(); | |
49077326 | 314 | |
e91ece55 CM |
315 | /* |
316 | * save the sectors now since our bio can | |
317 | * go away inside make_request | |
318 | */ | |
319 | sectors = bio_sectors(bio); | |
21a52c6d | 320 | rv = mddev->pers->make_request(mddev, bio); |
49077326 N |
321 | |
322 | cpu = part_stat_lock(); | |
323 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
e91ece55 | 324 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); |
49077326 N |
325 | part_stat_unlock(); |
326 | ||
409c57f3 N |
327 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
328 | wake_up(&mddev->sb_wait); | |
329 | ||
330 | return rv; | |
331 | } | |
332 | ||
9e35b99c N |
333 | /* mddev_suspend makes sure no new requests are submitted |
334 | * to the device, and that any requests that have been submitted | |
335 | * are completely handled. | |
336 | * Once ->stop is called and completes, the module will be completely | |
337 | * unused. | |
338 | */ | |
390ee602 | 339 | void mddev_suspend(mddev_t *mddev) |
409c57f3 N |
340 | { |
341 | BUG_ON(mddev->suspended); | |
342 | mddev->suspended = 1; | |
343 | synchronize_rcu(); | |
344 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
345 | mddev->pers->quiesce(mddev, 1); | |
409c57f3 | 346 | } |
390ee602 | 347 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 348 | |
390ee602 | 349 | void mddev_resume(mddev_t *mddev) |
409c57f3 N |
350 | { |
351 | mddev->suspended = 0; | |
352 | wake_up(&mddev->sb_wait); | |
353 | mddev->pers->quiesce(mddev, 0); | |
0fd018af JB |
354 | |
355 | md_wakeup_thread(mddev->thread); | |
356 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 357 | } |
390ee602 | 358 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 359 | |
3fa841d7 N |
360 | int mddev_congested(mddev_t *mddev, int bits) |
361 | { | |
362 | return mddev->suspended; | |
363 | } | |
364 | EXPORT_SYMBOL(mddev_congested); | |
365 | ||
a2826aa9 | 366 | /* |
e9c7469b | 367 | * Generic flush handling for md |
a2826aa9 N |
368 | */ |
369 | ||
e9c7469b | 370 | static void md_end_flush(struct bio *bio, int err) |
a2826aa9 N |
371 | { |
372 | mdk_rdev_t *rdev = bio->bi_private; | |
373 | mddev_t *mddev = rdev->mddev; | |
a2826aa9 N |
374 | |
375 | rdev_dec_pending(rdev, mddev); | |
376 | ||
377 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 378 | /* The pre-request flush has finished */ |
e804ac78 | 379 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
380 | } |
381 | bio_put(bio); | |
382 | } | |
383 | ||
a7a07e69 N |
384 | static void md_submit_flush_data(struct work_struct *ws); |
385 | ||
a035fc3e | 386 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 387 | { |
a035fc3e | 388 | mddev_t *mddev = container_of(ws, mddev_t, flush_work); |
a2826aa9 N |
389 | mdk_rdev_t *rdev; |
390 | ||
a7a07e69 N |
391 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
392 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 N |
393 | rcu_read_lock(); |
394 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
395 | if (rdev->raid_disk >= 0 && | |
396 | !test_bit(Faulty, &rdev->flags)) { | |
397 | /* Take two references, one is dropped | |
398 | * when request finishes, one after | |
399 | * we reclaim rcu_read_lock | |
400 | */ | |
401 | struct bio *bi; | |
402 | atomic_inc(&rdev->nr_pending); | |
403 | atomic_inc(&rdev->nr_pending); | |
404 | rcu_read_unlock(); | |
a167f663 | 405 | bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev); |
e9c7469b | 406 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
407 | bi->bi_private = rdev; |
408 | bi->bi_bdev = rdev->bdev; | |
409 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 410 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
411 | rcu_read_lock(); |
412 | rdev_dec_pending(rdev, mddev); | |
413 | } | |
414 | rcu_read_unlock(); | |
a7a07e69 N |
415 | if (atomic_dec_and_test(&mddev->flush_pending)) |
416 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
417 | } |
418 | ||
e9c7469b | 419 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 420 | { |
e9c7469b TH |
421 | mddev_t *mddev = container_of(ws, mddev_t, flush_work); |
422 | struct bio *bio = mddev->flush_bio; | |
a2826aa9 | 423 | |
e9c7469b | 424 | if (bio->bi_size == 0) |
a2826aa9 N |
425 | /* an empty barrier - all done */ |
426 | bio_endio(bio, 0); | |
427 | else { | |
e9c7469b | 428 | bio->bi_rw &= ~REQ_FLUSH; |
21a52c6d | 429 | if (mddev->pers->make_request(mddev, bio)) |
a2826aa9 | 430 | generic_make_request(bio); |
a2826aa9 | 431 | } |
2b74e12e N |
432 | |
433 | mddev->flush_bio = NULL; | |
434 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
435 | } |
436 | ||
e9c7469b | 437 | void md_flush_request(mddev_t *mddev, struct bio *bio) |
a2826aa9 N |
438 | { |
439 | spin_lock_irq(&mddev->write_lock); | |
440 | wait_event_lock_irq(mddev->sb_wait, | |
e9c7469b | 441 | !mddev->flush_bio, |
a2826aa9 | 442 | mddev->write_lock, /*nothing*/); |
e9c7469b | 443 | mddev->flush_bio = bio; |
a2826aa9 N |
444 | spin_unlock_irq(&mddev->write_lock); |
445 | ||
a035fc3e N |
446 | INIT_WORK(&mddev->flush_work, submit_flushes); |
447 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 448 | } |
e9c7469b | 449 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 450 | |
97658cdd N |
451 | /* Support for plugging. |
452 | * This mirrors the plugging support in request_queue, but does not | |
453 | * require having a whole queue or request structures. | |
454 | * We allocate an md_plug_cb for each md device and each thread it gets | |
455 | * plugged on. This links tot the private plug_handle structure in the | |
456 | * personality data where we keep a count of the number of outstanding | |
457 | * plugs so other code can see if a plug is active. | |
458 | */ | |
459 | struct md_plug_cb { | |
460 | struct blk_plug_cb cb; | |
461 | mddev_t *mddev; | |
462 | }; | |
463 | ||
464 | static void plugger_unplug(struct blk_plug_cb *cb) | |
465 | { | |
466 | struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb); | |
467 | if (atomic_dec_and_test(&mdcb->mddev->plug_cnt)) | |
468 | md_wakeup_thread(mdcb->mddev->thread); | |
469 | kfree(mdcb); | |
470 | } | |
471 | ||
472 | /* Check that an unplug wakeup will come shortly. | |
473 | * If not, wakeup the md thread immediately | |
474 | */ | |
475 | int mddev_check_plugged(mddev_t *mddev) | |
476 | { | |
477 | struct blk_plug *plug = current->plug; | |
478 | struct md_plug_cb *mdcb; | |
479 | ||
480 | if (!plug) | |
481 | return 0; | |
482 | ||
483 | list_for_each_entry(mdcb, &plug->cb_list, cb.list) { | |
484 | if (mdcb->cb.callback == plugger_unplug && | |
485 | mdcb->mddev == mddev) { | |
486 | /* Already on the list, move to top */ | |
487 | if (mdcb != list_first_entry(&plug->cb_list, | |
488 | struct md_plug_cb, | |
489 | cb.list)) | |
490 | list_move(&mdcb->cb.list, &plug->cb_list); | |
491 | return 1; | |
492 | } | |
493 | } | |
494 | /* Not currently on the callback list */ | |
495 | mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC); | |
496 | if (!mdcb) | |
497 | return 0; | |
498 | ||
499 | mdcb->mddev = mddev; | |
500 | mdcb->cb.callback = plugger_unplug; | |
501 | atomic_inc(&mddev->plug_cnt); | |
502 | list_add(&mdcb->cb.list, &plug->cb_list); | |
503 | return 1; | |
504 | } | |
505 | EXPORT_SYMBOL_GPL(mddev_check_plugged); | |
2ac87401 | 506 | |
1da177e4 LT |
507 | static inline mddev_t *mddev_get(mddev_t *mddev) |
508 | { | |
509 | atomic_inc(&mddev->active); | |
510 | return mddev; | |
511 | } | |
512 | ||
5fd3a17e | 513 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 514 | |
1da177e4 LT |
515 | static void mddev_put(mddev_t *mddev) |
516 | { | |
a167f663 N |
517 | struct bio_set *bs = NULL; |
518 | ||
1da177e4 LT |
519 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
520 | return; | |
d3374825 | 521 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
522 | mddev->ctime == 0 && !mddev->hold_active) { |
523 | /* Array is not configured at all, and not held active, | |
524 | * so destroy it */ | |
1da177e4 | 525 | list_del(&mddev->all_mddevs); |
a167f663 N |
526 | bs = mddev->bio_set; |
527 | mddev->bio_set = NULL; | |
d3374825 | 528 | if (mddev->gendisk) { |
e804ac78 TH |
529 | /* We did a probe so need to clean up. Call |
530 | * queue_work inside the spinlock so that | |
531 | * flush_workqueue() after mddev_find will | |
532 | * succeed in waiting for the work to be done. | |
d3374825 N |
533 | */ |
534 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 535 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
536 | } else |
537 | kfree(mddev); | |
538 | } | |
539 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
540 | if (bs) |
541 | bioset_free(bs); | |
1da177e4 LT |
542 | } |
543 | ||
390ee602 | 544 | void mddev_init(mddev_t *mddev) |
fafd7fb0 N |
545 | { |
546 | mutex_init(&mddev->open_mutex); | |
547 | mutex_init(&mddev->reconfig_mutex); | |
548 | mutex_init(&mddev->bitmap_info.mutex); | |
549 | INIT_LIST_HEAD(&mddev->disks); | |
550 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
551 | init_timer(&mddev->safemode_timer); | |
552 | atomic_set(&mddev->active, 1); | |
553 | atomic_set(&mddev->openers, 0); | |
554 | atomic_set(&mddev->active_io, 0); | |
97658cdd | 555 | atomic_set(&mddev->plug_cnt, 0); |
fafd7fb0 N |
556 | spin_lock_init(&mddev->write_lock); |
557 | atomic_set(&mddev->flush_pending, 0); | |
558 | init_waitqueue_head(&mddev->sb_wait); | |
559 | init_waitqueue_head(&mddev->recovery_wait); | |
560 | mddev->reshape_position = MaxSector; | |
561 | mddev->resync_min = 0; | |
562 | mddev->resync_max = MaxSector; | |
563 | mddev->level = LEVEL_NONE; | |
564 | } | |
390ee602 | 565 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 566 | |
1da177e4 LT |
567 | static mddev_t * mddev_find(dev_t unit) |
568 | { | |
569 | mddev_t *mddev, *new = NULL; | |
570 | ||
8f5f02c4 N |
571 | if (unit && MAJOR(unit) != MD_MAJOR) |
572 | unit &= ~((1<<MdpMinorShift)-1); | |
573 | ||
1da177e4 LT |
574 | retry: |
575 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
576 | |
577 | if (unit) { | |
578 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
579 | if (mddev->unit == unit) { | |
580 | mddev_get(mddev); | |
581 | spin_unlock(&all_mddevs_lock); | |
582 | kfree(new); | |
583 | return mddev; | |
584 | } | |
585 | ||
586 | if (new) { | |
587 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 588 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
589 | new->hold_active = UNTIL_IOCTL; |
590 | return new; | |
1da177e4 | 591 | } |
efeb53c0 N |
592 | } else if (new) { |
593 | /* find an unused unit number */ | |
594 | static int next_minor = 512; | |
595 | int start = next_minor; | |
596 | int is_free = 0; | |
597 | int dev = 0; | |
598 | while (!is_free) { | |
599 | dev = MKDEV(MD_MAJOR, next_minor); | |
600 | next_minor++; | |
601 | if (next_minor > MINORMASK) | |
602 | next_minor = 0; | |
603 | if (next_minor == start) { | |
604 | /* Oh dear, all in use. */ | |
605 | spin_unlock(&all_mddevs_lock); | |
606 | kfree(new); | |
607 | return NULL; | |
608 | } | |
609 | ||
610 | is_free = 1; | |
611 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
612 | if (mddev->unit == dev) { | |
613 | is_free = 0; | |
614 | break; | |
615 | } | |
616 | } | |
617 | new->unit = dev; | |
618 | new->md_minor = MINOR(dev); | |
619 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
620 | list_add(&new->all_mddevs, &all_mddevs); |
621 | spin_unlock(&all_mddevs_lock); | |
622 | return new; | |
623 | } | |
624 | spin_unlock(&all_mddevs_lock); | |
625 | ||
9ffae0cf | 626 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
627 | if (!new) |
628 | return NULL; | |
629 | ||
1da177e4 LT |
630 | new->unit = unit; |
631 | if (MAJOR(unit) == MD_MAJOR) | |
632 | new->md_minor = MINOR(unit); | |
633 | else | |
634 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
635 | ||
fafd7fb0 | 636 | mddev_init(new); |
1da177e4 | 637 | |
1da177e4 LT |
638 | goto retry; |
639 | } | |
640 | ||
641 | static inline int mddev_lock(mddev_t * mddev) | |
642 | { | |
df5b89b3 | 643 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
644 | } |
645 | ||
b522adcd DW |
646 | static inline int mddev_is_locked(mddev_t *mddev) |
647 | { | |
648 | return mutex_is_locked(&mddev->reconfig_mutex); | |
649 | } | |
650 | ||
1da177e4 LT |
651 | static inline int mddev_trylock(mddev_t * mddev) |
652 | { | |
df5b89b3 | 653 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
654 | } |
655 | ||
b6eb127d N |
656 | static struct attribute_group md_redundancy_group; |
657 | ||
a64c876f | 658 | static void mddev_unlock(mddev_t * 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 | |
005eca5e | 693 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
694 | } |
695 | ||
2989ddbd | 696 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 697 | { |
159ec1fc | 698 | mdk_rdev_t *rdev; |
1da177e4 | 699 | |
159ec1fc | 700 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
701 | if (rdev->desc_nr == nr) |
702 | return rdev; | |
159ec1fc | 703 | |
1da177e4 LT |
704 | return NULL; |
705 | } | |
706 | ||
707 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
708 | { | |
1da177e4 LT |
709 | mdk_rdev_t *rdev; |
710 | ||
159ec1fc | 711 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
712 | if (rdev->bdev->bd_dev == dev) |
713 | return rdev; | |
159ec1fc | 714 | |
1da177e4 LT |
715 | return NULL; |
716 | } | |
717 | ||
d9d166c2 | 718 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
719 | { |
720 | struct mdk_personality *pers; | |
d9d166c2 N |
721 | list_for_each_entry(pers, &pers_list, list) { |
722 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 723 | return pers; |
d9d166c2 N |
724 | if (strcmp(pers->name, clevel)==0) |
725 | return pers; | |
726 | } | |
2604b703 N |
727 | return NULL; |
728 | } | |
729 | ||
b73df2d3 | 730 | /* return the offset of the super block in 512byte sectors */ |
57b2caa3 | 731 | static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev) |
1da177e4 | 732 | { |
57b2caa3 | 733 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 734 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
735 | } |
736 | ||
1da177e4 LT |
737 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
738 | { | |
739 | if (rdev->sb_page) | |
740 | MD_BUG(); | |
741 | ||
742 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
743 | if (!rdev->sb_page) { | |
744 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 745 | return -ENOMEM; |
1da177e4 LT |
746 | } |
747 | ||
748 | return 0; | |
749 | } | |
750 | ||
751 | static void free_disk_sb(mdk_rdev_t * rdev) | |
752 | { | |
753 | if (rdev->sb_page) { | |
2d1f3b5d | 754 | put_page(rdev->sb_page); |
1da177e4 LT |
755 | rdev->sb_loaded = 0; |
756 | rdev->sb_page = NULL; | |
0f420358 | 757 | rdev->sb_start = 0; |
dd8ac336 | 758 | rdev->sectors = 0; |
1da177e4 LT |
759 | } |
760 | } | |
761 | ||
762 | ||
6712ecf8 | 763 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
764 | { |
765 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 766 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 767 | |
3a0f5bbb N |
768 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
769 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
770 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
771 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 772 | md_error(mddev, rdev); |
3a0f5bbb | 773 | } |
7bfa19f2 | 774 | |
a9701a30 N |
775 | if (atomic_dec_and_test(&mddev->pending_writes)) |
776 | wake_up(&mddev->sb_wait); | |
f8b58edf | 777 | bio_put(bio); |
7bfa19f2 N |
778 | } |
779 | ||
780 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, | |
781 | sector_t sector, int size, struct page *page) | |
782 | { | |
783 | /* write first size bytes of page to sector of rdev | |
784 | * Increment mddev->pending_writes before returning | |
785 | * and decrement it on completion, waking up sb_wait | |
786 | * if zero is reached. | |
787 | * If an error occurred, call md_error | |
788 | */ | |
a167f663 | 789 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 790 | |
a6ff7e08 | 791 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
7bfa19f2 N |
792 | bio->bi_sector = sector; |
793 | bio_add_page(bio, page, size, 0); | |
794 | bio->bi_private = rdev; | |
795 | bio->bi_end_io = super_written; | |
a9701a30 | 796 | |
7bfa19f2 | 797 | atomic_inc(&mddev->pending_writes); |
721a9602 | 798 | submit_bio(REQ_WRITE | REQ_SYNC | REQ_FLUSH | REQ_FUA, bio); |
a9701a30 N |
799 | } |
800 | ||
801 | void md_super_wait(mddev_t *mddev) | |
802 | { | |
e9c7469b | 803 | /* wait for all superblock writes that were scheduled to complete */ |
a9701a30 N |
804 | DEFINE_WAIT(wq); |
805 | for(;;) { | |
806 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
807 | if (atomic_read(&mddev->pending_writes)==0) | |
808 | break; | |
a9701a30 N |
809 | schedule(); |
810 | } | |
811 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
812 | } |
813 | ||
6712ecf8 | 814 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 815 | { |
1da177e4 | 816 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
817 | } |
818 | ||
2b193363 | 819 | int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size, |
ccebd4c4 | 820 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 821 | { |
a167f663 | 822 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
823 | struct completion event; |
824 | int ret; | |
825 | ||
721a9602 | 826 | rw |= REQ_SYNC; |
1da177e4 | 827 | |
a6ff7e08 JB |
828 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
829 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 JB |
830 | if (metadata_op) |
831 | bio->bi_sector = sector + rdev->sb_start; | |
832 | else | |
833 | bio->bi_sector = sector + rdev->data_offset; | |
1da177e4 LT |
834 | bio_add_page(bio, page, size, 0); |
835 | init_completion(&event); | |
836 | bio->bi_private = &event; | |
837 | bio->bi_end_io = bi_complete; | |
838 | submit_bio(rw, bio); | |
839 | wait_for_completion(&event); | |
840 | ||
841 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
842 | bio_put(bio); | |
843 | return ret; | |
844 | } | |
a8745db2 | 845 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 846 | |
0002b271 | 847 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
848 | { |
849 | char b[BDEVNAME_SIZE]; | |
850 | if (!rdev->sb_page) { | |
851 | MD_BUG(); | |
852 | return -EINVAL; | |
853 | } | |
854 | if (rdev->sb_loaded) | |
855 | return 0; | |
856 | ||
857 | ||
ccebd4c4 | 858 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
859 | goto fail; |
860 | rdev->sb_loaded = 1; | |
861 | return 0; | |
862 | ||
863 | fail: | |
864 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
865 | bdevname(rdev->bdev,b)); | |
866 | return -EINVAL; | |
867 | } | |
868 | ||
869 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
870 | { | |
05710466 AN |
871 | return sb1->set_uuid0 == sb2->set_uuid0 && |
872 | sb1->set_uuid1 == sb2->set_uuid1 && | |
873 | sb1->set_uuid2 == sb2->set_uuid2 && | |
874 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
875 | } |
876 | ||
1da177e4 LT |
877 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
878 | { | |
879 | int ret; | |
880 | mdp_super_t *tmp1, *tmp2; | |
881 | ||
882 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
883 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
884 | ||
885 | if (!tmp1 || !tmp2) { | |
886 | ret = 0; | |
35020f1a | 887 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
888 | goto abort; |
889 | } | |
890 | ||
891 | *tmp1 = *sb1; | |
892 | *tmp2 = *sb2; | |
893 | ||
894 | /* | |
895 | * nr_disks is not constant | |
896 | */ | |
897 | tmp1->nr_disks = 0; | |
898 | tmp2->nr_disks = 0; | |
899 | ||
ce0c8e05 | 900 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 901 | abort: |
990a8baf JJ |
902 | kfree(tmp1); |
903 | kfree(tmp2); | |
1da177e4 LT |
904 | return ret; |
905 | } | |
906 | ||
4d167f09 N |
907 | |
908 | static u32 md_csum_fold(u32 csum) | |
909 | { | |
910 | csum = (csum & 0xffff) + (csum >> 16); | |
911 | return (csum & 0xffff) + (csum >> 16); | |
912 | } | |
913 | ||
1da177e4 LT |
914 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
915 | { | |
4d167f09 N |
916 | u64 newcsum = 0; |
917 | u32 *sb32 = (u32*)sb; | |
918 | int i; | |
1da177e4 LT |
919 | unsigned int disk_csum, csum; |
920 | ||
921 | disk_csum = sb->sb_csum; | |
922 | sb->sb_csum = 0; | |
4d167f09 N |
923 | |
924 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
925 | newcsum += sb32[i]; | |
926 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
927 | ||
928 | ||
929 | #ifdef CONFIG_ALPHA | |
930 | /* This used to use csum_partial, which was wrong for several | |
931 | * reasons including that different results are returned on | |
932 | * different architectures. It isn't critical that we get exactly | |
933 | * the same return value as before (we always csum_fold before | |
934 | * testing, and that removes any differences). However as we | |
935 | * know that csum_partial always returned a 16bit value on | |
936 | * alphas, do a fold to maximise conformity to previous behaviour. | |
937 | */ | |
938 | sb->sb_csum = md_csum_fold(disk_csum); | |
939 | #else | |
1da177e4 | 940 | sb->sb_csum = disk_csum; |
4d167f09 | 941 | #endif |
1da177e4 LT |
942 | return csum; |
943 | } | |
944 | ||
945 | ||
946 | /* | |
947 | * Handle superblock details. | |
948 | * We want to be able to handle multiple superblock formats | |
949 | * so we have a common interface to them all, and an array of | |
950 | * different handlers. | |
951 | * We rely on user-space to write the initial superblock, and support | |
952 | * reading and updating of superblocks. | |
953 | * Interface methods are: | |
954 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
955 | * loads and validates a superblock on dev. | |
956 | * if refdev != NULL, compare superblocks on both devices | |
957 | * Return: | |
958 | * 0 - dev has a superblock that is compatible with refdev | |
959 | * 1 - dev has a superblock that is compatible and newer than refdev | |
960 | * so dev should be used as the refdev in future | |
961 | * -EINVAL superblock incompatible or invalid | |
962 | * -othererror e.g. -EIO | |
963 | * | |
964 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
965 | * Verify that dev is acceptable into mddev. | |
966 | * The first time, mddev->raid_disks will be 0, and data from | |
967 | * dev should be merged in. Subsequent calls check that dev | |
968 | * is new enough. Return 0 or -EINVAL | |
969 | * | |
970 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
971 | * Update the superblock for rdev with data in mddev | |
972 | * This does not write to disc. | |
973 | * | |
974 | */ | |
975 | ||
976 | struct super_type { | |
0cd17fec CW |
977 | char *name; |
978 | struct module *owner; | |
979 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
980 | int minor_version); | |
981 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
982 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
983 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 984 | sector_t num_sectors); |
1da177e4 LT |
985 | }; |
986 | ||
0894cc30 AN |
987 | /* |
988 | * Check that the given mddev has no bitmap. | |
989 | * | |
990 | * This function is called from the run method of all personalities that do not | |
991 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
992 | * has a bitmap. Otherwise, it returns 0. | |
993 | * | |
994 | */ | |
995 | int md_check_no_bitmap(mddev_t *mddev) | |
996 | { | |
c3d9714e | 997 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
998 | return 0; |
999 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
1000 | mdname(mddev), mddev->pers->name); | |
1001 | return 1; | |
1002 | } | |
1003 | EXPORT_SYMBOL(md_check_no_bitmap); | |
1004 | ||
1da177e4 LT |
1005 | /* |
1006 | * load_super for 0.90.0 | |
1007 | */ | |
1008 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1009 | { | |
1010 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
1011 | mdp_super_t *sb; | |
1012 | int ret; | |
1da177e4 LT |
1013 | |
1014 | /* | |
0f420358 | 1015 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
1016 | * it's at the end of the disk. |
1017 | * | |
1018 | * It also happens to be a multiple of 4Kb. | |
1019 | */ | |
57b2caa3 | 1020 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 1021 | |
0002b271 | 1022 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
1023 | if (ret) return ret; |
1024 | ||
1025 | ret = -EINVAL; | |
1026 | ||
1027 | bdevname(rdev->bdev, b); | |
1028 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
1029 | ||
1030 | if (sb->md_magic != MD_SB_MAGIC) { | |
1031 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
1032 | b); | |
1033 | goto abort; | |
1034 | } | |
1035 | ||
1036 | if (sb->major_version != 0 || | |
f6705578 N |
1037 | sb->minor_version < 90 || |
1038 | sb->minor_version > 91) { | |
1da177e4 LT |
1039 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
1040 | sb->major_version, sb->minor_version, | |
1041 | b); | |
1042 | goto abort; | |
1043 | } | |
1044 | ||
1045 | if (sb->raid_disks <= 0) | |
1046 | goto abort; | |
1047 | ||
4d167f09 | 1048 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
1049 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
1050 | b); | |
1051 | goto abort; | |
1052 | } | |
1053 | ||
1054 | rdev->preferred_minor = sb->md_minor; | |
1055 | rdev->data_offset = 0; | |
0002b271 | 1056 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
1057 | |
1058 | if (sb->level == LEVEL_MULTIPATH) | |
1059 | rdev->desc_nr = -1; | |
1060 | else | |
1061 | rdev->desc_nr = sb->this_disk.number; | |
1062 | ||
9a7b2b0f | 1063 | if (!refdev) { |
1da177e4 | 1064 | ret = 1; |
9a7b2b0f | 1065 | } else { |
1da177e4 LT |
1066 | __u64 ev1, ev2; |
1067 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
1068 | if (!uuid_equal(refsb, sb)) { | |
1069 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1070 | b, bdevname(refdev->bdev,b2)); | |
1071 | goto abort; | |
1072 | } | |
1073 | if (!sb_equal(refsb, sb)) { | |
1074 | printk(KERN_WARNING "md: %s has same UUID" | |
1075 | " but different superblock to %s\n", | |
1076 | b, bdevname(refdev->bdev, b2)); | |
1077 | goto abort; | |
1078 | } | |
1079 | ev1 = md_event(sb); | |
1080 | ev2 = md_event(refsb); | |
1081 | if (ev1 > ev2) | |
1082 | ret = 1; | |
1083 | else | |
1084 | ret = 0; | |
1085 | } | |
8190e754 | 1086 | rdev->sectors = rdev->sb_start; |
1da177e4 | 1087 | |
dd8ac336 | 1088 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
1089 | /* "this cannot possibly happen" ... */ |
1090 | ret = -EINVAL; | |
1091 | ||
1da177e4 LT |
1092 | abort: |
1093 | return ret; | |
1094 | } | |
1095 | ||
1096 | /* | |
1097 | * validate_super for 0.90.0 | |
1098 | */ | |
1099 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1100 | { | |
1101 | mdp_disk_t *desc; | |
1102 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 1103 | __u64 ev1 = md_event(sb); |
1da177e4 | 1104 | |
41158c7e | 1105 | rdev->raid_disk = -1; |
c5d79adb N |
1106 | clear_bit(Faulty, &rdev->flags); |
1107 | clear_bit(In_sync, &rdev->flags); | |
1108 | clear_bit(WriteMostly, &rdev->flags); | |
c5d79adb | 1109 | |
1da177e4 LT |
1110 | if (mddev->raid_disks == 0) { |
1111 | mddev->major_version = 0; | |
1112 | mddev->minor_version = sb->minor_version; | |
1113 | mddev->patch_version = sb->patch_version; | |
e691063a | 1114 | mddev->external = 0; |
9d8f0363 | 1115 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1116 | mddev->ctime = sb->ctime; |
1117 | mddev->utime = sb->utime; | |
1118 | mddev->level = sb->level; | |
d9d166c2 | 1119 | mddev->clevel[0] = 0; |
1da177e4 LT |
1120 | mddev->layout = sb->layout; |
1121 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 1122 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 1123 | mddev->events = ev1; |
c3d9714e N |
1124 | mddev->bitmap_info.offset = 0; |
1125 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; | |
1da177e4 | 1126 | |
f6705578 N |
1127 | if (mddev->minor_version >= 91) { |
1128 | mddev->reshape_position = sb->reshape_position; | |
1129 | mddev->delta_disks = sb->delta_disks; | |
1130 | mddev->new_level = sb->new_level; | |
1131 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1132 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
1133 | } else { |
1134 | mddev->reshape_position = MaxSector; | |
1135 | mddev->delta_disks = 0; | |
1136 | mddev->new_level = mddev->level; | |
1137 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1138 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1139 | } |
1140 | ||
1da177e4 LT |
1141 | if (sb->state & (1<<MD_SB_CLEAN)) |
1142 | mddev->recovery_cp = MaxSector; | |
1143 | else { | |
1144 | if (sb->events_hi == sb->cp_events_hi && | |
1145 | sb->events_lo == sb->cp_events_lo) { | |
1146 | mddev->recovery_cp = sb->recovery_cp; | |
1147 | } else | |
1148 | mddev->recovery_cp = 0; | |
1149 | } | |
1150 | ||
1151 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1152 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1153 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1154 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1155 | ||
1156 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1157 | |
1158 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
c3d9714e N |
1159 | mddev->bitmap_info.file == NULL) |
1160 | mddev->bitmap_info.offset = | |
1161 | mddev->bitmap_info.default_offset; | |
a654b9d8 | 1162 | |
41158c7e | 1163 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1164 | /* Insist on good event counter while assembling, except |
1165 | * for spares (which don't need an event count) */ | |
1da177e4 | 1166 | ++ev1; |
be6800a7 N |
1167 | if (sb->disks[rdev->desc_nr].state & ( |
1168 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
1169 | if (ev1 < mddev->events) | |
1170 | return -EINVAL; | |
41158c7e N |
1171 | } else if (mddev->bitmap) { |
1172 | /* if adding to array with a bitmap, then we can accept an | |
1173 | * older device ... but not too old. | |
1174 | */ | |
41158c7e N |
1175 | if (ev1 < mddev->bitmap->events_cleared) |
1176 | return 0; | |
07d84d10 N |
1177 | } else { |
1178 | if (ev1 < mddev->events) | |
1179 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1180 | return 0; | |
1181 | } | |
41158c7e | 1182 | |
1da177e4 | 1183 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1184 | desc = sb->disks + rdev->desc_nr; |
1185 | ||
1186 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1187 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1188 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1189 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1190 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1191 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
1192 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1193 | /* active but not in sync implies recovery up to | |
1194 | * reshape position. We don't know exactly where | |
1195 | * that is, so set to zero for now */ | |
1196 | if (mddev->minor_version >= 91) { | |
1197 | rdev->recovery_offset = 0; | |
1198 | rdev->raid_disk = desc->raid_disk; | |
1199 | } | |
1da177e4 | 1200 | } |
8ddf9efe N |
1201 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1202 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1203 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1204 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1205 | return 0; |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * sync_super for 0.90.0 | |
1210 | */ | |
1211 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1212 | { | |
1213 | mdp_super_t *sb; | |
1da177e4 LT |
1214 | mdk_rdev_t *rdev2; |
1215 | int next_spare = mddev->raid_disks; | |
19133a42 | 1216 | |
1da177e4 LT |
1217 | |
1218 | /* make rdev->sb match mddev data.. | |
1219 | * | |
1220 | * 1/ zero out disks | |
1221 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1222 | * 3/ any empty disks < next_spare become removed | |
1223 | * | |
1224 | * disks[0] gets initialised to REMOVED because | |
1225 | * we cannot be sure from other fields if it has | |
1226 | * been initialised or not. | |
1227 | */ | |
1228 | int i; | |
1229 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1230 | ||
61181565 N |
1231 | rdev->sb_size = MD_SB_BYTES; |
1232 | ||
1da177e4 LT |
1233 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
1234 | ||
1235 | memset(sb, 0, sizeof(*sb)); | |
1236 | ||
1237 | sb->md_magic = MD_SB_MAGIC; | |
1238 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1239 | sb->patch_version = mddev->patch_version; |
1240 | sb->gvalid_words = 0; /* ignored */ | |
1241 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1242 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1243 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1244 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1245 | ||
1246 | sb->ctime = mddev->ctime; | |
1247 | sb->level = mddev->level; | |
58c0fed4 | 1248 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1249 | sb->raid_disks = mddev->raid_disks; |
1250 | sb->md_minor = mddev->md_minor; | |
e691063a | 1251 | sb->not_persistent = 0; |
1da177e4 LT |
1252 | sb->utime = mddev->utime; |
1253 | sb->state = 0; | |
1254 | sb->events_hi = (mddev->events>>32); | |
1255 | sb->events_lo = (u32)mddev->events; | |
1256 | ||
f6705578 N |
1257 | if (mddev->reshape_position == MaxSector) |
1258 | sb->minor_version = 90; | |
1259 | else { | |
1260 | sb->minor_version = 91; | |
1261 | sb->reshape_position = mddev->reshape_position; | |
1262 | sb->new_level = mddev->new_level; | |
1263 | sb->delta_disks = mddev->delta_disks; | |
1264 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1265 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1266 | } |
1267 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1268 | if (mddev->in_sync) |
1269 | { | |
1270 | sb->recovery_cp = mddev->recovery_cp; | |
1271 | sb->cp_events_hi = (mddev->events>>32); | |
1272 | sb->cp_events_lo = (u32)mddev->events; | |
1273 | if (mddev->recovery_cp == MaxSector) | |
1274 | sb->state = (1<< MD_SB_CLEAN); | |
1275 | } else | |
1276 | sb->recovery_cp = 0; | |
1277 | ||
1278 | sb->layout = mddev->layout; | |
9d8f0363 | 1279 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1280 | |
c3d9714e | 1281 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1282 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1283 | ||
1da177e4 | 1284 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1285 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1286 | mdp_disk_t *d; |
86e6ffdd | 1287 | int desc_nr; |
0261cd9f N |
1288 | int is_active = test_bit(In_sync, &rdev2->flags); |
1289 | ||
1290 | if (rdev2->raid_disk >= 0 && | |
1291 | sb->minor_version >= 91) | |
1292 | /* we have nowhere to store the recovery_offset, | |
1293 | * but if it is not below the reshape_position, | |
1294 | * we can piggy-back on that. | |
1295 | */ | |
1296 | is_active = 1; | |
1297 | if (rdev2->raid_disk < 0 || | |
1298 | test_bit(Faulty, &rdev2->flags)) | |
1299 | is_active = 0; | |
1300 | if (is_active) | |
86e6ffdd | 1301 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1302 | else |
86e6ffdd | 1303 | desc_nr = next_spare++; |
19133a42 | 1304 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1305 | d = &sb->disks[rdev2->desc_nr]; |
1306 | nr_disks++; | |
1307 | d->number = rdev2->desc_nr; | |
1308 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1309 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1310 | if (is_active) |
1da177e4 LT |
1311 | d->raid_disk = rdev2->raid_disk; |
1312 | else | |
1313 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1314 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1315 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1316 | else if (is_active) { |
1da177e4 | 1317 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1318 | if (test_bit(In_sync, &rdev2->flags)) |
1319 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1320 | active++; |
1321 | working++; | |
1322 | } else { | |
1323 | d->state = 0; | |
1324 | spare++; | |
1325 | working++; | |
1326 | } | |
8ddf9efe N |
1327 | if (test_bit(WriteMostly, &rdev2->flags)) |
1328 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1329 | } |
1da177e4 LT |
1330 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1331 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1332 | mdp_disk_t *d = &sb->disks[i]; | |
1333 | if (d->state == 0 && d->number == 0) { | |
1334 | d->number = i; | |
1335 | d->raid_disk = i; | |
1336 | d->state = (1<<MD_DISK_REMOVED); | |
1337 | d->state |= (1<<MD_DISK_FAULTY); | |
1338 | failed++; | |
1339 | } | |
1340 | } | |
1341 | sb->nr_disks = nr_disks; | |
1342 | sb->active_disks = active; | |
1343 | sb->working_disks = working; | |
1344 | sb->failed_disks = failed; | |
1345 | sb->spare_disks = spare; | |
1346 | ||
1347 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1348 | sb->sb_csum = calc_sb_csum(sb); | |
1349 | } | |
1350 | ||
0cd17fec CW |
1351 | /* |
1352 | * rdev_size_change for 0.90.0 | |
1353 | */ | |
1354 | static unsigned long long | |
15f4a5fd | 1355 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1356 | { |
58c0fed4 | 1357 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1358 | return 0; /* component must fit device */ |
c3d9714e | 1359 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1360 | return 0; /* can't move bitmap */ |
57b2caa3 | 1361 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1362 | if (!num_sectors || num_sectors > rdev->sb_start) |
1363 | num_sectors = rdev->sb_start; | |
0f420358 | 1364 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1365 | rdev->sb_page); |
1366 | md_super_wait(rdev->mddev); | |
c26a44ed | 1367 | return num_sectors; |
0cd17fec CW |
1368 | } |
1369 | ||
1370 | ||
1da177e4 LT |
1371 | /* |
1372 | * version 1 superblock | |
1373 | */ | |
1374 | ||
1c05b4bc | 1375 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1376 | { |
1c05b4bc N |
1377 | __le32 disk_csum; |
1378 | u32 csum; | |
1da177e4 LT |
1379 | unsigned long long newcsum; |
1380 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1381 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1382 | int i; |
1383 | ||
1384 | disk_csum = sb->sb_csum; | |
1385 | sb->sb_csum = 0; | |
1386 | newcsum = 0; | |
1387 | for (i=0; size>=4; size -= 4 ) | |
1388 | newcsum += le32_to_cpu(*isuper++); | |
1389 | ||
1390 | if (size == 2) | |
1c05b4bc | 1391 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1392 | |
1393 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1394 | sb->sb_csum = disk_csum; | |
1395 | return cpu_to_le32(csum); | |
1396 | } | |
1397 | ||
1398 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1399 | { | |
1400 | struct mdp_superblock_1 *sb; | |
1401 | int ret; | |
0f420358 | 1402 | sector_t sb_start; |
1da177e4 | 1403 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1404 | int bmask; |
1da177e4 LT |
1405 | |
1406 | /* | |
0f420358 | 1407 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1408 | * It is always aligned to a 4K boundary and |
1409 | * depeding on minor_version, it can be: | |
1410 | * 0: At least 8K, but less than 12K, from end of device | |
1411 | * 1: At start of device | |
1412 | * 2: 4K from start of device. | |
1413 | */ | |
1414 | switch(minor_version) { | |
1415 | case 0: | |
77304d2a | 1416 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1417 | sb_start -= 8*2; |
1418 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1419 | break; |
1420 | case 1: | |
0f420358 | 1421 | sb_start = 0; |
1da177e4 LT |
1422 | break; |
1423 | case 2: | |
0f420358 | 1424 | sb_start = 8; |
1da177e4 LT |
1425 | break; |
1426 | default: | |
1427 | return -EINVAL; | |
1428 | } | |
0f420358 | 1429 | rdev->sb_start = sb_start; |
1da177e4 | 1430 | |
0002b271 N |
1431 | /* superblock is rarely larger than 1K, but it can be larger, |
1432 | * and it is safe to read 4k, so we do that | |
1433 | */ | |
1434 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1435 | if (ret) return ret; |
1436 | ||
1437 | ||
1438 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1439 | ||
1440 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1441 | sb->major_version != cpu_to_le32(1) || | |
1442 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1443 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1444 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1445 | return -EINVAL; |
1446 | ||
1447 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1448 | printk("md: invalid superblock checksum on %s\n", | |
1449 | bdevname(rdev->bdev,b)); | |
1450 | return -EINVAL; | |
1451 | } | |
1452 | if (le64_to_cpu(sb->data_size) < 10) { | |
1453 | printk("md: data_size too small on %s\n", | |
1454 | bdevname(rdev->bdev,b)); | |
1455 | return -EINVAL; | |
1456 | } | |
e11e93fa | 1457 | |
1da177e4 LT |
1458 | rdev->preferred_minor = 0xffff; |
1459 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1460 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1461 | |
0002b271 | 1462 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1463 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1464 | if (rdev->sb_size & bmask) |
a1801f85 N |
1465 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1466 | ||
1467 | if (minor_version | |
0f420358 | 1468 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1469 | return -EINVAL; |
0002b271 | 1470 | |
31b65a0d N |
1471 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1472 | rdev->desc_nr = -1; | |
1473 | else | |
1474 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1475 | ||
9a7b2b0f | 1476 | if (!refdev) { |
8ed75463 | 1477 | ret = 1; |
9a7b2b0f | 1478 | } else { |
1da177e4 LT |
1479 | __u64 ev1, ev2; |
1480 | struct mdp_superblock_1 *refsb = | |
1481 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1482 | ||
1483 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1484 | sb->level != refsb->level || | |
1485 | sb->layout != refsb->layout || | |
1486 | sb->chunksize != refsb->chunksize) { | |
1487 | printk(KERN_WARNING "md: %s has strangely different" | |
1488 | " superblock to %s\n", | |
1489 | bdevname(rdev->bdev,b), | |
1490 | bdevname(refdev->bdev,b2)); | |
1491 | return -EINVAL; | |
1492 | } | |
1493 | ev1 = le64_to_cpu(sb->events); | |
1494 | ev2 = le64_to_cpu(refsb->events); | |
1495 | ||
1496 | if (ev1 > ev2) | |
8ed75463 N |
1497 | ret = 1; |
1498 | else | |
1499 | ret = 0; | |
1da177e4 | 1500 | } |
a1801f85 | 1501 | if (minor_version) |
77304d2a | 1502 | rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 1503 | le64_to_cpu(sb->data_offset); |
1da177e4 | 1504 | else |
dd8ac336 AN |
1505 | rdev->sectors = rdev->sb_start; |
1506 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1507 | return -EINVAL; |
dd8ac336 | 1508 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1509 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1510 | return -EINVAL; |
8ed75463 | 1511 | return ret; |
1da177e4 LT |
1512 | } |
1513 | ||
1514 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1515 | { | |
1516 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1517 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1518 | |
41158c7e | 1519 | rdev->raid_disk = -1; |
c5d79adb N |
1520 | clear_bit(Faulty, &rdev->flags); |
1521 | clear_bit(In_sync, &rdev->flags); | |
1522 | clear_bit(WriteMostly, &rdev->flags); | |
c5d79adb | 1523 | |
1da177e4 LT |
1524 | if (mddev->raid_disks == 0) { |
1525 | mddev->major_version = 1; | |
1526 | mddev->patch_version = 0; | |
e691063a | 1527 | mddev->external = 0; |
9d8f0363 | 1528 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1529 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1530 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1531 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1532 | mddev->clevel[0] = 0; |
1da177e4 LT |
1533 | mddev->layout = le32_to_cpu(sb->layout); |
1534 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1535 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1536 | mddev->events = ev1; |
c3d9714e N |
1537 | mddev->bitmap_info.offset = 0; |
1538 | mddev->bitmap_info.default_offset = 1024 >> 9; | |
1da177e4 LT |
1539 | |
1540 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1541 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1542 | ||
1543 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1544 | |
71c0805c | 1545 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
c3d9714e N |
1546 | mddev->bitmap_info.file == NULL ) |
1547 | mddev->bitmap_info.offset = | |
1548 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
e11e93fa | 1549 | |
f6705578 N |
1550 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1551 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1552 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1553 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1554 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1555 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1556 | } else { |
1557 | mddev->reshape_position = MaxSector; | |
1558 | mddev->delta_disks = 0; | |
1559 | mddev->new_level = mddev->level; | |
1560 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1561 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1562 | } |
1563 | ||
41158c7e | 1564 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1565 | /* Insist of good event counter while assembling, except for |
1566 | * spares (which don't need an event count) */ | |
1da177e4 | 1567 | ++ev1; |
be6800a7 N |
1568 | if (rdev->desc_nr >= 0 && |
1569 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1570 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1571 | if (ev1 < mddev->events) | |
1572 | return -EINVAL; | |
41158c7e N |
1573 | } else if (mddev->bitmap) { |
1574 | /* If adding to array with a bitmap, then we can accept an | |
1575 | * older device, but not too old. | |
1576 | */ | |
41158c7e N |
1577 | if (ev1 < mddev->bitmap->events_cleared) |
1578 | return 0; | |
07d84d10 N |
1579 | } else { |
1580 | if (ev1 < mddev->events) | |
1581 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1582 | return 0; | |
1583 | } | |
1da177e4 LT |
1584 | if (mddev->level != LEVEL_MULTIPATH) { |
1585 | int role; | |
3673f305 N |
1586 | if (rdev->desc_nr < 0 || |
1587 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1588 | role = 0xffff; | |
1589 | rdev->desc_nr = -1; | |
1590 | } else | |
1591 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1592 | switch(role) { |
1593 | case 0xffff: /* spare */ | |
1da177e4 LT |
1594 | break; |
1595 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1596 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1597 | break; |
1598 | default: | |
5fd6c1dc N |
1599 | if ((le32_to_cpu(sb->feature_map) & |
1600 | MD_FEATURE_RECOVERY_OFFSET)) | |
1601 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1602 | else | |
1603 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1604 | rdev->raid_disk = role; |
1605 | break; | |
1606 | } | |
8ddf9efe N |
1607 | if (sb->devflags & WriteMostly1) |
1608 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1609 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1610 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1611 | |
1da177e4 LT |
1612 | return 0; |
1613 | } | |
1614 | ||
1615 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1616 | { | |
1617 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1618 | mdk_rdev_t *rdev2; |
1619 | int max_dev, i; | |
1620 | /* make rdev->sb match mddev and rdev data. */ | |
1621 | ||
1622 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1623 | ||
1624 | sb->feature_map = 0; | |
1625 | sb->pad0 = 0; | |
5fd6c1dc | 1626 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1627 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1628 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1629 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1630 | ||
1631 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1632 | sb->events = cpu_to_le64(mddev->events); | |
1633 | if (mddev->in_sync) | |
1634 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1635 | else | |
1636 | sb->resync_offset = cpu_to_le64(0); | |
1637 | ||
1c05b4bc | 1638 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1639 | |
f0ca340c | 1640 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1641 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1642 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1643 | sb->level = cpu_to_le32(mddev->level); |
1644 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1645 | |
c3d9714e N |
1646 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1647 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1648 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1649 | } |
5fd6c1dc N |
1650 | |
1651 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1652 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1653 | sb->feature_map |= |
1654 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1655 | sb->recovery_offset = | |
1656 | cpu_to_le64(rdev->recovery_offset); | |
5fd6c1dc N |
1657 | } |
1658 | ||
f6705578 N |
1659 | if (mddev->reshape_position != MaxSector) { |
1660 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1661 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1662 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1663 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1664 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1665 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1666 | } |
a654b9d8 | 1667 | |
1da177e4 | 1668 | max_dev = 0; |
159ec1fc | 1669 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1670 | if (rdev2->desc_nr+1 > max_dev) |
1671 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1672 | |
70471daf N |
1673 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1674 | int bmask; | |
a778b73f | 1675 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1676 | rdev->sb_size = max_dev * 2 + 256; |
1677 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1678 | if (rdev->sb_size & bmask) | |
1679 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1680 | } else |
1681 | max_dev = le32_to_cpu(sb->max_dev); | |
1682 | ||
1da177e4 LT |
1683 | for (i=0; i<max_dev;i++) |
1684 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1685 | ||
159ec1fc | 1686 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1687 | i = rdev2->desc_nr; |
b2d444d7 | 1688 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1689 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1690 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1691 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1692 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1693 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1694 | else |
1695 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1696 | } | |
1697 | ||
1da177e4 LT |
1698 | sb->sb_csum = calc_sb_1_csum(sb); |
1699 | } | |
1700 | ||
0cd17fec | 1701 | static unsigned long long |
15f4a5fd | 1702 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1703 | { |
1704 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1705 | sector_t max_sectors; |
58c0fed4 | 1706 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1707 | return 0; /* component must fit device */ |
0f420358 | 1708 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1709 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1710 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1711 | max_sectors -= rdev->data_offset; |
1712 | if (!num_sectors || num_sectors > max_sectors) | |
1713 | num_sectors = max_sectors; | |
c3d9714e | 1714 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1715 | /* minor version 0 with bitmap we can't move */ |
1716 | return 0; | |
1717 | } else { | |
1718 | /* minor version 0; superblock after data */ | |
0f420358 | 1719 | sector_t sb_start; |
77304d2a | 1720 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1721 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1722 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1723 | if (!num_sectors || num_sectors > max_sectors) |
1724 | num_sectors = max_sectors; | |
0f420358 | 1725 | rdev->sb_start = sb_start; |
0cd17fec CW |
1726 | } |
1727 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1728 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1729 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1730 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1731 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1732 | rdev->sb_page); |
1733 | md_super_wait(rdev->mddev); | |
c26a44ed | 1734 | return num_sectors; |
0cd17fec | 1735 | } |
1da177e4 | 1736 | |
75c96f85 | 1737 | static struct super_type super_types[] = { |
1da177e4 LT |
1738 | [0] = { |
1739 | .name = "0.90.0", | |
1740 | .owner = THIS_MODULE, | |
0cd17fec CW |
1741 | .load_super = super_90_load, |
1742 | .validate_super = super_90_validate, | |
1743 | .sync_super = super_90_sync, | |
1744 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1745 | }, |
1746 | [1] = { | |
1747 | .name = "md-1", | |
1748 | .owner = THIS_MODULE, | |
0cd17fec CW |
1749 | .load_super = super_1_load, |
1750 | .validate_super = super_1_validate, | |
1751 | .sync_super = super_1_sync, | |
1752 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1753 | }, |
1754 | }; | |
1da177e4 | 1755 | |
076f968b JB |
1756 | static void sync_super(mddev_t *mddev, mdk_rdev_t *rdev) |
1757 | { | |
1758 | if (mddev->sync_super) { | |
1759 | mddev->sync_super(mddev, rdev); | |
1760 | return; | |
1761 | } | |
1762 | ||
1763 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1764 | ||
1765 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1766 | } | |
1767 | ||
1da177e4 LT |
1768 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) |
1769 | { | |
7dd5e7c3 | 1770 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1771 | |
4b80991c N |
1772 | rcu_read_lock(); |
1773 | rdev_for_each_rcu(rdev, mddev1) | |
1774 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1775 | if (rdev->bdev->bd_contains == |
4b80991c N |
1776 | rdev2->bdev->bd_contains) { |
1777 | rcu_read_unlock(); | |
7dd5e7c3 | 1778 | return 1; |
4b80991c N |
1779 | } |
1780 | rcu_read_unlock(); | |
1da177e4 LT |
1781 | return 0; |
1782 | } | |
1783 | ||
1784 | static LIST_HEAD(pending_raid_disks); | |
1785 | ||
ac5e7113 AN |
1786 | /* |
1787 | * Try to register data integrity profile for an mddev | |
1788 | * | |
1789 | * This is called when an array is started and after a disk has been kicked | |
1790 | * from the array. It only succeeds if all working and active component devices | |
1791 | * are integrity capable with matching profiles. | |
1792 | */ | |
1793 | int md_integrity_register(mddev_t *mddev) | |
1794 | { | |
1795 | mdk_rdev_t *rdev, *reference = NULL; | |
1796 | ||
1797 | if (list_empty(&mddev->disks)) | |
1798 | return 0; /* nothing to do */ | |
629acb6a JB |
1799 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
1800 | return 0; /* shouldn't register, or already is */ | |
ac5e7113 AN |
1801 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1802 | /* skip spares and non-functional disks */ | |
1803 | if (test_bit(Faulty, &rdev->flags)) | |
1804 | continue; | |
1805 | if (rdev->raid_disk < 0) | |
1806 | continue; | |
ac5e7113 AN |
1807 | if (!reference) { |
1808 | /* Use the first rdev as the reference */ | |
1809 | reference = rdev; | |
1810 | continue; | |
1811 | } | |
1812 | /* does this rdev's profile match the reference profile? */ | |
1813 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1814 | rdev->bdev->bd_disk) < 0) | |
1815 | return -EINVAL; | |
1816 | } | |
89078d57 MP |
1817 | if (!reference || !bdev_get_integrity(reference->bdev)) |
1818 | return 0; | |
ac5e7113 AN |
1819 | /* |
1820 | * All component devices are integrity capable and have matching | |
1821 | * profiles, register the common profile for the md device. | |
1822 | */ | |
1823 | if (blk_integrity_register(mddev->gendisk, | |
1824 | bdev_get_integrity(reference->bdev)) != 0) { | |
1825 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1826 | mdname(mddev)); | |
1827 | return -EINVAL; | |
1828 | } | |
a91a2785 MP |
1829 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
1830 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
1831 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
1832 | mdname(mddev)); | |
1833 | return -EINVAL; | |
1834 | } | |
ac5e7113 AN |
1835 | return 0; |
1836 | } | |
1837 | EXPORT_SYMBOL(md_integrity_register); | |
1838 | ||
1839 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
1840 | void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) | |
3f9d99c1 | 1841 | { |
3f9d99c1 | 1842 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); |
ac5e7113 | 1843 | struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); |
3f9d99c1 | 1844 | |
ac5e7113 | 1845 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1846 | return; |
ac5e7113 | 1847 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1848 | return; |
ac5e7113 AN |
1849 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1850 | rdev->bdev->bd_disk) >= 0) | |
1851 | return; | |
1852 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1853 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1854 | } |
ac5e7113 | 1855 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1856 | |
1da177e4 LT |
1857 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1858 | { | |
7dd5e7c3 | 1859 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1860 | struct kobject *ko; |
1edf80d3 | 1861 | char *s; |
5e55e2f5 | 1862 | int err; |
1da177e4 LT |
1863 | |
1864 | if (rdev->mddev) { | |
1865 | MD_BUG(); | |
1866 | return -EINVAL; | |
1867 | } | |
11e2ede0 DW |
1868 | |
1869 | /* prevent duplicates */ | |
1870 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1871 | return -EEXIST; | |
1872 | ||
dd8ac336 AN |
1873 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1874 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1875 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1876 | if (mddev->pers) { |
1877 | /* Cannot change size, so fail | |
1878 | * If mddev->level <= 0, then we don't care | |
1879 | * about aligning sizes (e.g. linear) | |
1880 | */ | |
1881 | if (mddev->level > 0) | |
1882 | return -ENOSPC; | |
1883 | } else | |
dd8ac336 | 1884 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1885 | } |
1da177e4 LT |
1886 | |
1887 | /* Verify rdev->desc_nr is unique. | |
1888 | * If it is -1, assign a free number, else | |
1889 | * check number is not in use | |
1890 | */ | |
1891 | if (rdev->desc_nr < 0) { | |
1892 | int choice = 0; | |
1893 | if (mddev->pers) choice = mddev->raid_disks; | |
1894 | while (find_rdev_nr(mddev, choice)) | |
1895 | choice++; | |
1896 | rdev->desc_nr = choice; | |
1897 | } else { | |
1898 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1899 | return -EBUSY; | |
1900 | } | |
de01dfad N |
1901 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1902 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1903 | mdname(mddev), mddev->max_disks); | |
1904 | return -EBUSY; | |
1905 | } | |
19133a42 | 1906 | bdevname(rdev->bdev,b); |
649316b2 | 1907 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1908 | *s = '!'; |
649316b2 | 1909 | |
1da177e4 | 1910 | rdev->mddev = mddev; |
19133a42 | 1911 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1912 | |
b2d6db58 | 1913 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1914 | goto fail; |
86e6ffdd | 1915 | |
0762b8bd | 1916 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
1917 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
1918 | /* failure here is OK */; | |
1919 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 1920 | |
4b80991c | 1921 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 1922 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
1923 | |
1924 | /* May as well allow recovery to be retried once */ | |
1925 | mddev->recovery_disabled = 0; | |
3f9d99c1 | 1926 | |
1da177e4 | 1927 | return 0; |
5e55e2f5 N |
1928 | |
1929 | fail: | |
1930 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1931 | b, mdname(mddev)); | |
1932 | return err; | |
1da177e4 LT |
1933 | } |
1934 | ||
177a99b2 | 1935 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1936 | { |
1937 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1938 | kobject_del(&rdev->kobj); | |
177a99b2 | 1939 | kobject_put(&rdev->kobj); |
5792a285 N |
1940 | } |
1941 | ||
1da177e4 LT |
1942 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1943 | { | |
1944 | char b[BDEVNAME_SIZE]; | |
1945 | if (!rdev->mddev) { | |
1946 | MD_BUG(); | |
1947 | return; | |
1948 | } | |
49731baa | 1949 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1950 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1951 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1952 | rdev->mddev = NULL; | |
86e6ffdd | 1953 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1954 | sysfs_put(rdev->sysfs_state); |
1955 | rdev->sysfs_state = NULL; | |
5792a285 | 1956 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1957 | * writing to 'remove' to "dev/state". We also need |
1958 | * to delay it due to rcu usage. | |
5792a285 | 1959 | */ |
4b80991c | 1960 | synchronize_rcu(); |
177a99b2 N |
1961 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1962 | kobject_get(&rdev->kobj); | |
e804ac78 | 1963 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
1964 | } |
1965 | ||
1966 | /* | |
1967 | * prevent the device from being mounted, repartitioned or | |
1968 | * otherwise reused by a RAID array (or any other kernel | |
1969 | * subsystem), by bd_claiming the device. | |
1970 | */ | |
c5d79adb | 1971 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1972 | { |
1973 | int err = 0; | |
1974 | struct block_device *bdev; | |
1975 | char b[BDEVNAME_SIZE]; | |
1976 | ||
d4d77629 TH |
1977 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
1978 | shared ? (mdk_rdev_t *)lock_rdev : rdev); | |
1da177e4 LT |
1979 | if (IS_ERR(bdev)) { |
1980 | printk(KERN_ERR "md: could not open %s.\n", | |
1981 | __bdevname(dev, b)); | |
1982 | return PTR_ERR(bdev); | |
1983 | } | |
1da177e4 LT |
1984 | rdev->bdev = bdev; |
1985 | return err; | |
1986 | } | |
1987 | ||
1988 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1989 | { | |
1990 | struct block_device *bdev = rdev->bdev; | |
1991 | rdev->bdev = NULL; | |
1992 | if (!bdev) | |
1993 | MD_BUG(); | |
e525fd89 | 1994 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
1995 | } |
1996 | ||
1997 | void md_autodetect_dev(dev_t dev); | |
1998 | ||
1999 | static void export_rdev(mdk_rdev_t * rdev) | |
2000 | { | |
2001 | char b[BDEVNAME_SIZE]; | |
2002 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
2003 | bdevname(rdev->bdev,b)); | |
2004 | if (rdev->mddev) | |
2005 | MD_BUG(); | |
2006 | free_disk_sb(rdev); | |
1da177e4 | 2007 | #ifndef MODULE |
d0fae18f N |
2008 | if (test_bit(AutoDetected, &rdev->flags)) |
2009 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2010 | #endif |
2011 | unlock_rdev(rdev); | |
86e6ffdd | 2012 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2013 | } |
2014 | ||
2015 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
2016 | { | |
2017 | unbind_rdev_from_array(rdev); | |
2018 | export_rdev(rdev); | |
2019 | } | |
2020 | ||
2021 | static void export_array(mddev_t *mddev) | |
2022 | { | |
159ec1fc | 2023 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 2024 | |
d089c6af | 2025 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
2026 | if (!rdev->mddev) { |
2027 | MD_BUG(); | |
2028 | continue; | |
2029 | } | |
2030 | kick_rdev_from_array(rdev); | |
2031 | } | |
2032 | if (!list_empty(&mddev->disks)) | |
2033 | MD_BUG(); | |
2034 | mddev->raid_disks = 0; | |
2035 | mddev->major_version = 0; | |
2036 | } | |
2037 | ||
2038 | static void print_desc(mdp_disk_t *desc) | |
2039 | { | |
2040 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
2041 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
2042 | } | |
2043 | ||
cd2ac932 | 2044 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
2045 | { |
2046 | int i; | |
2047 | ||
2048 | printk(KERN_INFO | |
2049 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
2050 | sb->major_version, sb->minor_version, sb->patch_version, | |
2051 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
2052 | sb->ctime); | |
2053 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
2054 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
2055 | sb->md_minor, sb->layout, sb->chunk_size); | |
2056 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
2057 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
2058 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
2059 | sb->failed_disks, sb->spare_disks, | |
2060 | sb->sb_csum, (unsigned long)sb->events_lo); | |
2061 | ||
2062 | printk(KERN_INFO); | |
2063 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2064 | mdp_disk_t *desc; | |
2065 | ||
2066 | desc = sb->disks + i; | |
2067 | if (desc->number || desc->major || desc->minor || | |
2068 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
2069 | printk(" D %2d: ", i); | |
2070 | print_desc(desc); | |
2071 | } | |
2072 | } | |
2073 | printk(KERN_INFO "md: THIS: "); | |
2074 | print_desc(&sb->this_disk); | |
cd2ac932 | 2075 | } |
1da177e4 | 2076 | |
cd2ac932 CR |
2077 | static void print_sb_1(struct mdp_superblock_1 *sb) |
2078 | { | |
2079 | __u8 *uuid; | |
2080 | ||
2081 | uuid = sb->set_uuid; | |
ad361c98 | 2082 | printk(KERN_INFO |
7b75c2f8 | 2083 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 2084 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
2085 | le32_to_cpu(sb->major_version), |
2086 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 2087 | uuid, |
cd2ac932 CR |
2088 | sb->set_name, |
2089 | (unsigned long long)le64_to_cpu(sb->ctime) | |
2090 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
2091 | ||
2092 | uuid = sb->device_uuid; | |
ad361c98 JP |
2093 | printk(KERN_INFO |
2094 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 2095 | " RO:%llu\n" |
7b75c2f8 | 2096 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
2097 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
2098 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
2099 | le32_to_cpu(sb->level), |
2100 | (unsigned long long)le64_to_cpu(sb->size), | |
2101 | le32_to_cpu(sb->raid_disks), | |
2102 | le32_to_cpu(sb->layout), | |
2103 | le32_to_cpu(sb->chunksize), | |
2104 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2105 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2106 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2107 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2108 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2109 | uuid, |
cd2ac932 CR |
2110 | sb->devflags, |
2111 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2112 | (unsigned long long)le64_to_cpu(sb->events), | |
2113 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2114 | le32_to_cpu(sb->sb_csum), | |
2115 | le32_to_cpu(sb->max_dev) | |
2116 | ); | |
1da177e4 LT |
2117 | } |
2118 | ||
cd2ac932 | 2119 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
2120 | { |
2121 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2122 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2123 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2124 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2125 | rdev->desc_nr); | |
1da177e4 | 2126 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2127 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2128 | switch (major_version) { | |
2129 | case 0: | |
2130 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
2131 | break; | |
2132 | case 1: | |
2133 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
2134 | break; | |
2135 | } | |
1da177e4 LT |
2136 | } else |
2137 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2138 | } | |
2139 | ||
5e56341d | 2140 | static void md_print_devices(void) |
1da177e4 | 2141 | { |
159ec1fc | 2142 | struct list_head *tmp; |
1da177e4 LT |
2143 | mdk_rdev_t *rdev; |
2144 | mddev_t *mddev; | |
2145 | char b[BDEVNAME_SIZE]; | |
2146 | ||
2147 | printk("\n"); | |
2148 | printk("md: **********************************\n"); | |
2149 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2150 | printk("md: **********************************\n"); | |
29ac4aa3 | 2151 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2152 | |
32a7627c N |
2153 | if (mddev->bitmap) |
2154 | bitmap_print_sb(mddev->bitmap); | |
2155 | else | |
2156 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 2157 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
2158 | printk("<%s>", bdevname(rdev->bdev,b)); |
2159 | printk("\n"); | |
2160 | ||
159ec1fc | 2161 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 2162 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2163 | } |
2164 | printk("md: **********************************\n"); | |
2165 | printk("\n"); | |
2166 | } | |
2167 | ||
2168 | ||
42543769 | 2169 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 2170 | { |
42543769 N |
2171 | /* Update each superblock (in-memory image), but |
2172 | * if we are allowed to, skip spares which already | |
2173 | * have the right event counter, or have one earlier | |
2174 | * (which would mean they aren't being marked as dirty | |
2175 | * with the rest of the array) | |
2176 | */ | |
1da177e4 | 2177 | mdk_rdev_t *rdev; |
159ec1fc | 2178 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
2179 | if (rdev->sb_events == mddev->events || |
2180 | (nospares && | |
2181 | rdev->raid_disk < 0 && | |
42543769 N |
2182 | rdev->sb_events+1 == mddev->events)) { |
2183 | /* Don't update this superblock */ | |
2184 | rdev->sb_loaded = 2; | |
2185 | } else { | |
076f968b | 2186 | sync_super(mddev, rdev); |
42543769 N |
2187 | rdev->sb_loaded = 1; |
2188 | } | |
1da177e4 LT |
2189 | } |
2190 | } | |
2191 | ||
850b2b42 | 2192 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 2193 | { |
1da177e4 | 2194 | mdk_rdev_t *rdev; |
06d91a5f | 2195 | int sync_req; |
42543769 | 2196 | int nospares = 0; |
1da177e4 | 2197 | |
1da177e4 | 2198 | repeat: |
3a3a5ddb N |
2199 | /* First make sure individual recovery_offsets are correct */ |
2200 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
2201 | if (rdev->raid_disk >= 0 && | |
2202 | mddev->delta_disks >= 0 && | |
2203 | !test_bit(In_sync, &rdev->flags) && | |
2204 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2205 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2206 | ||
2207 | } | |
bd52b746 | 2208 | if (!mddev->persistent) { |
070dc6dd | 2209 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2210 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
d97a41dc N |
2211 | if (!mddev->external) |
2212 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
3a3a5ddb N |
2213 | wake_up(&mddev->sb_wait); |
2214 | return; | |
2215 | } | |
2216 | ||
a9701a30 | 2217 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2218 | |
3a3a5ddb N |
2219 | mddev->utime = get_seconds(); |
2220 | ||
850b2b42 N |
2221 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2222 | force_change = 1; | |
2223 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2224 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2225 | * though if events isn't the right even/odd, we will have to do | |
2226 | * spares after all | |
2227 | */ | |
2228 | nospares = 1; | |
2229 | if (force_change) | |
2230 | nospares = 0; | |
2231 | if (mddev->degraded) | |
84692195 N |
2232 | /* If the array is degraded, then skipping spares is both |
2233 | * dangerous and fairly pointless. | |
2234 | * Dangerous because a device that was removed from the array | |
2235 | * might have a event_count that still looks up-to-date, | |
2236 | * so it can be re-added without a resync. | |
2237 | * Pointless because if there are any spares to skip, | |
2238 | * then a recovery will happen and soon that array won't | |
2239 | * be degraded any more and the spare can go back to sleep then. | |
2240 | */ | |
850b2b42 | 2241 | nospares = 0; |
84692195 | 2242 | |
06d91a5f | 2243 | sync_req = mddev->in_sync; |
42543769 N |
2244 | |
2245 | /* If this is just a dirty<->clean transition, and the array is clean | |
2246 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2247 | if (nospares |
42543769 | 2248 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2249 | && mddev->can_decrease_events |
2250 | && mddev->events != 1) { | |
42543769 | 2251 | mddev->events--; |
a8707c08 N |
2252 | mddev->can_decrease_events = 0; |
2253 | } else { | |
42543769 N |
2254 | /* otherwise we have to go forward and ... */ |
2255 | mddev->events ++; | |
a8707c08 | 2256 | mddev->can_decrease_events = nospares; |
42543769 | 2257 | } |
1da177e4 LT |
2258 | |
2259 | if (!mddev->events) { | |
2260 | /* | |
2261 | * oops, this 64-bit counter should never wrap. | |
2262 | * Either we are in around ~1 trillion A.C., assuming | |
2263 | * 1 reboot per second, or we have a bug: | |
2264 | */ | |
2265 | MD_BUG(); | |
2266 | mddev->events --; | |
2267 | } | |
e691063a | 2268 | sync_sbs(mddev, nospares); |
a9701a30 | 2269 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2270 | |
2271 | dprintk(KERN_INFO | |
2272 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2273 | mdname(mddev),mddev->in_sync); | |
2274 | ||
4ad13663 | 2275 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2276 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2277 | char b[BDEVNAME_SIZE]; |
2278 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2279 | if (rdev->sb_loaded != 1) |
2280 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2281 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2282 | dprintk("(skipping faulty "); |
2283 | ||
2284 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2285 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2286 | md_super_write(mddev,rdev, |
0f420358 | 2287 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2288 | rdev->sb_page); |
2289 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2290 | bdevname(rdev->bdev,b), | |
0f420358 | 2291 | (unsigned long long)rdev->sb_start); |
42543769 | 2292 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2293 | |
1da177e4 LT |
2294 | } else |
2295 | dprintk(")\n"); | |
7bfa19f2 | 2296 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2297 | /* only need to write one superblock... */ |
2298 | break; | |
2299 | } | |
a9701a30 | 2300 | md_super_wait(mddev); |
850b2b42 | 2301 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2302 | |
a9701a30 | 2303 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2304 | if (mddev->in_sync != sync_req || |
2305 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2306 | /* have to write it out again */ |
a9701a30 | 2307 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2308 | goto repeat; |
2309 | } | |
850b2b42 | 2310 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2311 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2312 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2313 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2314 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2315 | |
1da177e4 LT |
2316 | } |
2317 | ||
7f6ce769 | 2318 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2319 | * We want to accept with case. For this we use cmd_match. |
2320 | */ | |
2321 | static int cmd_match(const char *cmd, const char *str) | |
2322 | { | |
2323 | /* See if cmd, written into a sysfs file, matches | |
2324 | * str. They must either be the same, or cmd can | |
2325 | * have a trailing newline | |
2326 | */ | |
2327 | while (*cmd && *str && *cmd == *str) { | |
2328 | cmd++; | |
2329 | str++; | |
2330 | } | |
2331 | if (*cmd == '\n') | |
2332 | cmd++; | |
2333 | if (*str || *cmd) | |
2334 | return 0; | |
2335 | return 1; | |
2336 | } | |
2337 | ||
86e6ffdd N |
2338 | struct rdev_sysfs_entry { |
2339 | struct attribute attr; | |
2340 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2341 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2342 | }; | |
2343 | ||
2344 | static ssize_t | |
96de1e66 | 2345 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2346 | { |
2347 | char *sep = ""; | |
20a49ff6 | 2348 | size_t len = 0; |
86e6ffdd | 2349 | |
b2d444d7 | 2350 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2351 | len+= sprintf(page+len, "%sfaulty",sep); |
2352 | sep = ","; | |
2353 | } | |
b2d444d7 | 2354 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2355 | len += sprintf(page+len, "%sin_sync",sep); |
2356 | sep = ","; | |
2357 | } | |
f655675b N |
2358 | if (test_bit(WriteMostly, &rdev->flags)) { |
2359 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2360 | sep = ","; | |
2361 | } | |
6bfe0b49 DW |
2362 | if (test_bit(Blocked, &rdev->flags)) { |
2363 | len += sprintf(page+len, "%sblocked", sep); | |
2364 | sep = ","; | |
2365 | } | |
b2d444d7 N |
2366 | if (!test_bit(Faulty, &rdev->flags) && |
2367 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2368 | len += sprintf(page+len, "%sspare", sep); |
2369 | sep = ","; | |
2370 | } | |
2371 | return len+sprintf(page+len, "\n"); | |
2372 | } | |
2373 | ||
45dc2de1 N |
2374 | static ssize_t |
2375 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2376 | { | |
2377 | /* can write | |
2378 | * faulty - simulates and error | |
2379 | * remove - disconnects the device | |
f655675b N |
2380 | * writemostly - sets write_mostly |
2381 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2382 | * blocked - sets the Blocked flag |
2383 | * -blocked - clears the Blocked flag | |
6d56e278 | 2384 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2385 | */ |
2386 | int err = -EINVAL; | |
2387 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2388 | md_error(rdev->mddev, rdev); | |
2389 | err = 0; | |
2390 | } else if (cmd_match(buf, "remove")) { | |
2391 | if (rdev->raid_disk >= 0) | |
2392 | err = -EBUSY; | |
2393 | else { | |
2394 | mddev_t *mddev = rdev->mddev; | |
2395 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2396 | if (mddev->pers) |
2397 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2398 | md_new_event(mddev); |
2399 | err = 0; | |
2400 | } | |
f655675b N |
2401 | } else if (cmd_match(buf, "writemostly")) { |
2402 | set_bit(WriteMostly, &rdev->flags); | |
2403 | err = 0; | |
2404 | } else if (cmd_match(buf, "-writemostly")) { | |
2405 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2406 | err = 0; |
2407 | } else if (cmd_match(buf, "blocked")) { | |
2408 | set_bit(Blocked, &rdev->flags); | |
2409 | err = 0; | |
2410 | } else if (cmd_match(buf, "-blocked")) { | |
2411 | clear_bit(Blocked, &rdev->flags); | |
2412 | wake_up(&rdev->blocked_wait); | |
2413 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2414 | md_wakeup_thread(rdev->mddev->thread); | |
2415 | ||
6d56e278 N |
2416 | err = 0; |
2417 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2418 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2419 | err = 0; |
45dc2de1 | 2420 | } |
00bcb4ac N |
2421 | if (!err) |
2422 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2423 | return err ? err : len; |
2424 | } | |
80ca3a44 N |
2425 | static struct rdev_sysfs_entry rdev_state = |
2426 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2427 | |
4dbcdc75 N |
2428 | static ssize_t |
2429 | errors_show(mdk_rdev_t *rdev, char *page) | |
2430 | { | |
2431 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2432 | } | |
2433 | ||
2434 | static ssize_t | |
2435 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2436 | { | |
2437 | char *e; | |
2438 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2439 | if (*buf && (*e == 0 || *e == '\n')) { | |
2440 | atomic_set(&rdev->corrected_errors, n); | |
2441 | return len; | |
2442 | } | |
2443 | return -EINVAL; | |
2444 | } | |
2445 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2446 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2447 | |
014236d2 N |
2448 | static ssize_t |
2449 | slot_show(mdk_rdev_t *rdev, char *page) | |
2450 | { | |
2451 | if (rdev->raid_disk < 0) | |
2452 | return sprintf(page, "none\n"); | |
2453 | else | |
2454 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2455 | } | |
2456 | ||
2457 | static ssize_t | |
2458 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2459 | { | |
2460 | char *e; | |
c303da6d N |
2461 | int err; |
2462 | char nm[20]; | |
014236d2 N |
2463 | int slot = simple_strtoul(buf, &e, 10); |
2464 | if (strncmp(buf, "none", 4)==0) | |
2465 | slot = -1; | |
2466 | else if (e==buf || (*e && *e!= '\n')) | |
2467 | return -EINVAL; | |
6c2fce2e | 2468 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2469 | /* Setting 'slot' on an active array requires also |
2470 | * updating the 'rd%d' link, and communicating | |
2471 | * with the personality with ->hot_*_disk. | |
2472 | * For now we only support removing | |
2473 | * failed/spare devices. This normally happens automatically, | |
2474 | * but not when the metadata is externally managed. | |
2475 | */ | |
c303da6d N |
2476 | if (rdev->raid_disk == -1) |
2477 | return -EEXIST; | |
2478 | /* personality does all needed checks */ | |
01393f3d | 2479 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d N |
2480 | return -EINVAL; |
2481 | err = rdev->mddev->pers-> | |
2482 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2483 | if (err) | |
2484 | return err; | |
2485 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2486 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
b7103107 | 2487 | rdev->raid_disk = -1; |
c303da6d N |
2488 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2489 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2490 | } else if (rdev->mddev->pers) { |
2491 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2492 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2493 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2494 | */ |
2495 | ||
2496 | if (rdev->raid_disk != -1) | |
2497 | return -EBUSY; | |
2498 | ||
c6751b2b N |
2499 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2500 | return -EBUSY; | |
2501 | ||
6c2fce2e NB |
2502 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2503 | return -EINVAL; | |
2504 | ||
159ec1fc | 2505 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2506 | if (rdev2->raid_disk == slot) |
2507 | return -EEXIST; | |
2508 | ||
ba1b41b6 N |
2509 | if (slot >= rdev->mddev->raid_disks && |
2510 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2511 | return -ENOSPC; | |
2512 | ||
6c2fce2e NB |
2513 | rdev->raid_disk = slot; |
2514 | if (test_bit(In_sync, &rdev->flags)) | |
2515 | rdev->saved_raid_disk = slot; | |
2516 | else | |
2517 | rdev->saved_raid_disk = -1; | |
2518 | err = rdev->mddev->pers-> | |
2519 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2520 | if (err) { |
6c2fce2e | 2521 | rdev->raid_disk = -1; |
6c2fce2e | 2522 | return err; |
52664732 | 2523 | } else |
00bcb4ac | 2524 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6c2fce2e NB |
2525 | sprintf(nm, "rd%d", rdev->raid_disk); |
2526 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
00bcb4ac | 2527 | /* failure here is OK */; |
6c2fce2e | 2528 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2529 | } else { |
ba1b41b6 N |
2530 | if (slot >= rdev->mddev->raid_disks && |
2531 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2532 | return -ENOSPC; |
2533 | rdev->raid_disk = slot; | |
2534 | /* assume it is working */ | |
c5d79adb N |
2535 | clear_bit(Faulty, &rdev->flags); |
2536 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2537 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2538 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2539 | } |
014236d2 N |
2540 | return len; |
2541 | } | |
2542 | ||
2543 | ||
2544 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2545 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2546 | |
93c8cad0 N |
2547 | static ssize_t |
2548 | offset_show(mdk_rdev_t *rdev, char *page) | |
2549 | { | |
6961ece4 | 2550 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2551 | } |
2552 | ||
2553 | static ssize_t | |
2554 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2555 | { | |
2556 | char *e; | |
2557 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2558 | if (e==buf || (*e && *e != '\n')) | |
2559 | return -EINVAL; | |
8ed0a521 | 2560 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2561 | return -EBUSY; |
dd8ac336 | 2562 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2563 | /* Must set offset before size, so overlap checks |
2564 | * can be sane */ | |
2565 | return -EBUSY; | |
93c8cad0 N |
2566 | rdev->data_offset = offset; |
2567 | return len; | |
2568 | } | |
2569 | ||
2570 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2571 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2572 | |
83303b61 N |
2573 | static ssize_t |
2574 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2575 | { | |
dd8ac336 | 2576 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2577 | } |
2578 | ||
c5d79adb N |
2579 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2580 | { | |
2581 | /* check if two start/length pairs overlap */ | |
2582 | if (s1+l1 <= s2) | |
2583 | return 0; | |
2584 | if (s2+l2 <= s1) | |
2585 | return 0; | |
2586 | return 1; | |
2587 | } | |
2588 | ||
b522adcd DW |
2589 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2590 | { | |
2591 | unsigned long long blocks; | |
2592 | sector_t new; | |
2593 | ||
2594 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2595 | return -EINVAL; | |
2596 | ||
2597 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2598 | return -EINVAL; /* sector conversion overflow */ | |
2599 | ||
2600 | new = blocks * 2; | |
2601 | if (new != blocks * 2) | |
2602 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2603 | ||
2604 | *sectors = new; | |
2605 | return 0; | |
2606 | } | |
2607 | ||
83303b61 N |
2608 | static ssize_t |
2609 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2610 | { | |
27c529bb | 2611 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2612 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2613 | sector_t sectors; |
27c529bb | 2614 | |
b522adcd | 2615 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2616 | return -EINVAL; |
0cd17fec | 2617 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2618 | if (my_mddev->persistent) { |
dd8ac336 AN |
2619 | sectors = super_types[my_mddev->major_version]. |
2620 | rdev_size_change(rdev, sectors); | |
2621 | if (!sectors) | |
0cd17fec | 2622 | return -EBUSY; |
dd8ac336 | 2623 | } else if (!sectors) |
77304d2a | 2624 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2625 | rdev->data_offset; |
0cd17fec | 2626 | } |
dd8ac336 | 2627 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2628 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2629 | |
dd8ac336 AN |
2630 | rdev->sectors = sectors; |
2631 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2632 | /* need to check that all other rdevs with the same ->bdev |
2633 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2634 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2635 | * we have to change it back, we will have the lock again. |
2636 | */ | |
2637 | mddev_t *mddev; | |
2638 | int overlap = 0; | |
159ec1fc | 2639 | struct list_head *tmp; |
c5d79adb | 2640 | |
27c529bb | 2641 | mddev_unlock(my_mddev); |
29ac4aa3 | 2642 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2643 | mdk_rdev_t *rdev2; |
2644 | ||
2645 | mddev_lock(mddev); | |
159ec1fc | 2646 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
f21e9ff7 N |
2647 | if (rdev->bdev == rdev2->bdev && |
2648 | rdev != rdev2 && | |
2649 | overlaps(rdev->data_offset, rdev->sectors, | |
2650 | rdev2->data_offset, | |
2651 | rdev2->sectors)) { | |
c5d79adb N |
2652 | overlap = 1; |
2653 | break; | |
2654 | } | |
2655 | mddev_unlock(mddev); | |
2656 | if (overlap) { | |
2657 | mddev_put(mddev); | |
2658 | break; | |
2659 | } | |
2660 | } | |
27c529bb | 2661 | mddev_lock(my_mddev); |
c5d79adb N |
2662 | if (overlap) { |
2663 | /* Someone else could have slipped in a size | |
2664 | * change here, but doing so is just silly. | |
dd8ac336 | 2665 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2666 | * safe, and trust userspace not to race with |
2667 | * itself | |
2668 | */ | |
dd8ac336 | 2669 | rdev->sectors = oldsectors; |
c5d79adb N |
2670 | return -EBUSY; |
2671 | } | |
2672 | } | |
83303b61 N |
2673 | return len; |
2674 | } | |
2675 | ||
2676 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2677 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2678 | |
06e3c817 DW |
2679 | |
2680 | static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) | |
2681 | { | |
2682 | unsigned long long recovery_start = rdev->recovery_offset; | |
2683 | ||
2684 | if (test_bit(In_sync, &rdev->flags) || | |
2685 | recovery_start == MaxSector) | |
2686 | return sprintf(page, "none\n"); | |
2687 | ||
2688 | return sprintf(page, "%llu\n", recovery_start); | |
2689 | } | |
2690 | ||
2691 | static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2692 | { | |
2693 | unsigned long long recovery_start; | |
2694 | ||
2695 | if (cmd_match(buf, "none")) | |
2696 | recovery_start = MaxSector; | |
2697 | else if (strict_strtoull(buf, 10, &recovery_start)) | |
2698 | return -EINVAL; | |
2699 | ||
2700 | if (rdev->mddev->pers && | |
2701 | rdev->raid_disk >= 0) | |
2702 | return -EBUSY; | |
2703 | ||
2704 | rdev->recovery_offset = recovery_start; | |
2705 | if (recovery_start == MaxSector) | |
2706 | set_bit(In_sync, &rdev->flags); | |
2707 | else | |
2708 | clear_bit(In_sync, &rdev->flags); | |
2709 | return len; | |
2710 | } | |
2711 | ||
2712 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2713 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2714 | ||
86e6ffdd N |
2715 | static struct attribute *rdev_default_attrs[] = { |
2716 | &rdev_state.attr, | |
4dbcdc75 | 2717 | &rdev_errors.attr, |
014236d2 | 2718 | &rdev_slot.attr, |
93c8cad0 | 2719 | &rdev_offset.attr, |
83303b61 | 2720 | &rdev_size.attr, |
06e3c817 | 2721 | &rdev_recovery_start.attr, |
86e6ffdd N |
2722 | NULL, |
2723 | }; | |
2724 | static ssize_t | |
2725 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2726 | { | |
2727 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2728 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2729 | mddev_t *mddev = rdev->mddev; |
2730 | ssize_t rv; | |
86e6ffdd N |
2731 | |
2732 | if (!entry->show) | |
2733 | return -EIO; | |
27c529bb N |
2734 | |
2735 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2736 | if (!rv) { | |
2737 | if (rdev->mddev == NULL) | |
2738 | rv = -EBUSY; | |
2739 | else | |
2740 | rv = entry->show(rdev, page); | |
2741 | mddev_unlock(mddev); | |
2742 | } | |
2743 | return rv; | |
86e6ffdd N |
2744 | } |
2745 | ||
2746 | static ssize_t | |
2747 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2748 | const char *page, size_t length) | |
2749 | { | |
2750 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2751 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2752 | ssize_t rv; |
2753 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2754 | |
2755 | if (!entry->store) | |
2756 | return -EIO; | |
67463acb N |
2757 | if (!capable(CAP_SYS_ADMIN)) |
2758 | return -EACCES; | |
27c529bb | 2759 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2760 | if (!rv) { |
27c529bb N |
2761 | if (rdev->mddev == NULL) |
2762 | rv = -EBUSY; | |
2763 | else | |
2764 | rv = entry->store(rdev, page, length); | |
6a51830e | 2765 | mddev_unlock(mddev); |
ca388059 N |
2766 | } |
2767 | return rv; | |
86e6ffdd N |
2768 | } |
2769 | ||
2770 | static void rdev_free(struct kobject *ko) | |
2771 | { | |
2772 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2773 | kfree(rdev); | |
2774 | } | |
52cf25d0 | 2775 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
2776 | .show = rdev_attr_show, |
2777 | .store = rdev_attr_store, | |
2778 | }; | |
2779 | static struct kobj_type rdev_ktype = { | |
2780 | .release = rdev_free, | |
2781 | .sysfs_ops = &rdev_sysfs_ops, | |
2782 | .default_attrs = rdev_default_attrs, | |
2783 | }; | |
2784 | ||
e8bb9a83 N |
2785 | void md_rdev_init(mdk_rdev_t *rdev) |
2786 | { | |
2787 | rdev->desc_nr = -1; | |
2788 | rdev->saved_raid_disk = -1; | |
2789 | rdev->raid_disk = -1; | |
2790 | rdev->flags = 0; | |
2791 | rdev->data_offset = 0; | |
2792 | rdev->sb_events = 0; | |
2793 | rdev->last_read_error.tv_sec = 0; | |
2794 | rdev->last_read_error.tv_nsec = 0; | |
2795 | atomic_set(&rdev->nr_pending, 0); | |
2796 | atomic_set(&rdev->read_errors, 0); | |
2797 | atomic_set(&rdev->corrected_errors, 0); | |
2798 | ||
2799 | INIT_LIST_HEAD(&rdev->same_set); | |
2800 | init_waitqueue_head(&rdev->blocked_wait); | |
2801 | } | |
2802 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
2803 | /* |
2804 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2805 | * | |
2806 | * mark the device faulty if: | |
2807 | * | |
2808 | * - the device is nonexistent (zero size) | |
2809 | * - the device has no valid superblock | |
2810 | * | |
2811 | * a faulty rdev _never_ has rdev->sb set. | |
2812 | */ | |
2813 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2814 | { | |
2815 | char b[BDEVNAME_SIZE]; | |
2816 | int err; | |
2817 | mdk_rdev_t *rdev; | |
2818 | sector_t size; | |
2819 | ||
9ffae0cf | 2820 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2821 | if (!rdev) { |
2822 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2823 | return ERR_PTR(-ENOMEM); | |
2824 | } | |
1da177e4 | 2825 | |
e8bb9a83 | 2826 | md_rdev_init(rdev); |
1da177e4 LT |
2827 | if ((err = alloc_disk_sb(rdev))) |
2828 | goto abort_free; | |
2829 | ||
c5d79adb | 2830 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2831 | if (err) |
2832 | goto abort_free; | |
2833 | ||
f9cb074b | 2834 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2835 | |
77304d2a | 2836 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 LT |
2837 | if (!size) { |
2838 | printk(KERN_WARNING | |
2839 | "md: %s has zero or unknown size, marking faulty!\n", | |
2840 | bdevname(rdev->bdev,b)); | |
2841 | err = -EINVAL; | |
2842 | goto abort_free; | |
2843 | } | |
2844 | ||
2845 | if (super_format >= 0) { | |
2846 | err = super_types[super_format]. | |
2847 | load_super(rdev, NULL, super_minor); | |
2848 | if (err == -EINVAL) { | |
df968c4e N |
2849 | printk(KERN_WARNING |
2850 | "md: %s does not have a valid v%d.%d " | |
2851 | "superblock, not importing!\n", | |
2852 | bdevname(rdev->bdev,b), | |
2853 | super_format, super_minor); | |
1da177e4 LT |
2854 | goto abort_free; |
2855 | } | |
2856 | if (err < 0) { | |
2857 | printk(KERN_WARNING | |
2858 | "md: could not read %s's sb, not importing!\n", | |
2859 | bdevname(rdev->bdev,b)); | |
2860 | goto abort_free; | |
2861 | } | |
2862 | } | |
6bfe0b49 | 2863 | |
1da177e4 LT |
2864 | return rdev; |
2865 | ||
2866 | abort_free: | |
2867 | if (rdev->sb_page) { | |
2868 | if (rdev->bdev) | |
2869 | unlock_rdev(rdev); | |
2870 | free_disk_sb(rdev); | |
2871 | } | |
2872 | kfree(rdev); | |
2873 | return ERR_PTR(err); | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * Check a full RAID array for plausibility | |
2878 | */ | |
2879 | ||
2880 | ||
a757e64c | 2881 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2882 | { |
2883 | int i; | |
159ec1fc | 2884 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2885 | char b[BDEVNAME_SIZE]; |
2886 | ||
2887 | freshest = NULL; | |
d089c6af | 2888 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2889 | switch (super_types[mddev->major_version]. |
2890 | load_super(rdev, freshest, mddev->minor_version)) { | |
2891 | case 1: | |
2892 | freshest = rdev; | |
2893 | break; | |
2894 | case 0: | |
2895 | break; | |
2896 | default: | |
2897 | printk( KERN_ERR \ | |
2898 | "md: fatal superblock inconsistency in %s" | |
2899 | " -- removing from array\n", | |
2900 | bdevname(rdev->bdev,b)); | |
2901 | kick_rdev_from_array(rdev); | |
2902 | } | |
2903 | ||
2904 | ||
2905 | super_types[mddev->major_version]. | |
2906 | validate_super(mddev, freshest); | |
2907 | ||
2908 | i = 0; | |
d089c6af | 2909 | rdev_for_each(rdev, tmp, mddev) { |
233fca36 N |
2910 | if (mddev->max_disks && |
2911 | (rdev->desc_nr >= mddev->max_disks || | |
2912 | i > mddev->max_disks)) { | |
de01dfad N |
2913 | printk(KERN_WARNING |
2914 | "md: %s: %s: only %d devices permitted\n", | |
2915 | mdname(mddev), bdevname(rdev->bdev, b), | |
2916 | mddev->max_disks); | |
2917 | kick_rdev_from_array(rdev); | |
2918 | continue; | |
2919 | } | |
1da177e4 LT |
2920 | if (rdev != freshest) |
2921 | if (super_types[mddev->major_version]. | |
2922 | validate_super(mddev, rdev)) { | |
2923 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2924 | " from array!\n", | |
2925 | bdevname(rdev->bdev,b)); | |
2926 | kick_rdev_from_array(rdev); | |
2927 | continue; | |
2928 | } | |
2929 | if (mddev->level == LEVEL_MULTIPATH) { | |
2930 | rdev->desc_nr = i++; | |
2931 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2932 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 2933 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
2934 | rdev->raid_disk = -1; |
2935 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2936 | } |
2937 | } | |
1da177e4 LT |
2938 | } |
2939 | ||
72e02075 N |
2940 | /* Read a fixed-point number. |
2941 | * Numbers in sysfs attributes should be in "standard" units where | |
2942 | * possible, so time should be in seconds. | |
2943 | * However we internally use a a much smaller unit such as | |
2944 | * milliseconds or jiffies. | |
2945 | * This function takes a decimal number with a possible fractional | |
2946 | * component, and produces an integer which is the result of | |
2947 | * multiplying that number by 10^'scale'. | |
2948 | * all without any floating-point arithmetic. | |
2949 | */ | |
2950 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
2951 | { | |
2952 | unsigned long result = 0; | |
2953 | long decimals = -1; | |
2954 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
2955 | if (*cp == '.') | |
2956 | decimals = 0; | |
2957 | else if (decimals < scale) { | |
2958 | unsigned int value; | |
2959 | value = *cp - '0'; | |
2960 | result = result * 10 + value; | |
2961 | if (decimals >= 0) | |
2962 | decimals++; | |
2963 | } | |
2964 | cp++; | |
2965 | } | |
2966 | if (*cp == '\n') | |
2967 | cp++; | |
2968 | if (*cp) | |
2969 | return -EINVAL; | |
2970 | if (decimals < 0) | |
2971 | decimals = 0; | |
2972 | while (decimals < scale) { | |
2973 | result *= 10; | |
2974 | decimals ++; | |
2975 | } | |
2976 | *res = result; | |
2977 | return 0; | |
2978 | } | |
2979 | ||
2980 | ||
19052c0e N |
2981 | static void md_safemode_timeout(unsigned long data); |
2982 | ||
16f17b39 N |
2983 | static ssize_t |
2984 | safe_delay_show(mddev_t *mddev, char *page) | |
2985 | { | |
2986 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2987 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2988 | } | |
2989 | static ssize_t | |
2990 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2991 | { | |
16f17b39 | 2992 | unsigned long msec; |
97ce0a7f | 2993 | |
72e02075 | 2994 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 2995 | return -EINVAL; |
16f17b39 N |
2996 | if (msec == 0) |
2997 | mddev->safemode_delay = 0; | |
2998 | else { | |
19052c0e | 2999 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
3000 | mddev->safemode_delay = (msec*HZ)/1000; |
3001 | if (mddev->safemode_delay == 0) | |
3002 | mddev->safemode_delay = 1; | |
19052c0e N |
3003 | if (mddev->safemode_delay < old_delay) |
3004 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
3005 | } |
3006 | return len; | |
3007 | } | |
3008 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3009 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3010 | |
eae1701f | 3011 | static ssize_t |
96de1e66 | 3012 | level_show(mddev_t *mddev, char *page) |
eae1701f | 3013 | { |
2604b703 | 3014 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 3015 | if (p) |
eae1701f | 3016 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
3017 | else if (mddev->clevel[0]) |
3018 | return sprintf(page, "%s\n", mddev->clevel); | |
3019 | else if (mddev->level != LEVEL_NONE) | |
3020 | return sprintf(page, "%d\n", mddev->level); | |
3021 | else | |
3022 | return 0; | |
eae1701f N |
3023 | } |
3024 | ||
d9d166c2 N |
3025 | static ssize_t |
3026 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
3027 | { | |
f2859af6 | 3028 | char clevel[16]; |
20a49ff6 | 3029 | ssize_t rv = len; |
245f46c2 | 3030 | struct mdk_personality *pers; |
f2859af6 | 3031 | long level; |
245f46c2 | 3032 | void *priv; |
3a981b03 | 3033 | mdk_rdev_t *rdev; |
245f46c2 N |
3034 | |
3035 | if (mddev->pers == NULL) { | |
3036 | if (len == 0) | |
3037 | return 0; | |
3038 | if (len >= sizeof(mddev->clevel)) | |
3039 | return -ENOSPC; | |
3040 | strncpy(mddev->clevel, buf, len); | |
3041 | if (mddev->clevel[len-1] == '\n') | |
3042 | len--; | |
3043 | mddev->clevel[len] = 0; | |
3044 | mddev->level = LEVEL_NONE; | |
3045 | return rv; | |
3046 | } | |
3047 | ||
3048 | /* request to change the personality. Need to ensure: | |
3049 | * - array is not engaged in resync/recovery/reshape | |
3050 | * - old personality can be suspended | |
3051 | * - new personality will access other array. | |
3052 | */ | |
3053 | ||
bb4f1e9d N |
3054 | if (mddev->sync_thread || |
3055 | mddev->reshape_position != MaxSector || | |
3056 | mddev->sysfs_active) | |
d9d166c2 | 3057 | return -EBUSY; |
245f46c2 N |
3058 | |
3059 | if (!mddev->pers->quiesce) { | |
3060 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3061 | mdname(mddev), mddev->pers->name); | |
3062 | return -EINVAL; | |
3063 | } | |
3064 | ||
3065 | /* Now find the new personality */ | |
f2859af6 | 3066 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3067 | return -EINVAL; |
f2859af6 DW |
3068 | strncpy(clevel, buf, len); |
3069 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3070 | len--; |
f2859af6 DW |
3071 | clevel[len] = 0; |
3072 | if (strict_strtol(clevel, 10, &level)) | |
3073 | level = LEVEL_NONE; | |
245f46c2 | 3074 | |
f2859af6 DW |
3075 | if (request_module("md-%s", clevel) != 0) |
3076 | request_module("md-level-%s", clevel); | |
245f46c2 | 3077 | spin_lock(&pers_lock); |
f2859af6 | 3078 | pers = find_pers(level, clevel); |
245f46c2 N |
3079 | if (!pers || !try_module_get(pers->owner)) { |
3080 | spin_unlock(&pers_lock); | |
f2859af6 | 3081 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3082 | return -EINVAL; |
3083 | } | |
3084 | spin_unlock(&pers_lock); | |
3085 | ||
3086 | if (pers == mddev->pers) { | |
3087 | /* Nothing to do! */ | |
3088 | module_put(pers->owner); | |
3089 | return rv; | |
3090 | } | |
3091 | if (!pers->takeover) { | |
3092 | module_put(pers->owner); | |
3093 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3094 | mdname(mddev), clevel); |
245f46c2 N |
3095 | return -EINVAL; |
3096 | } | |
3097 | ||
e93f68a1 N |
3098 | list_for_each_entry(rdev, &mddev->disks, same_set) |
3099 | rdev->new_raid_disk = rdev->raid_disk; | |
3100 | ||
245f46c2 N |
3101 | /* ->takeover must set new_* and/or delta_disks |
3102 | * if it succeeds, and may set them when it fails. | |
3103 | */ | |
3104 | priv = pers->takeover(mddev); | |
3105 | if (IS_ERR(priv)) { | |
3106 | mddev->new_level = mddev->level; | |
3107 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3108 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3109 | mddev->raid_disks -= mddev->delta_disks; |
3110 | mddev->delta_disks = 0; | |
3111 | module_put(pers->owner); | |
3112 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3113 | mdname(mddev), clevel); |
245f46c2 N |
3114 | return PTR_ERR(priv); |
3115 | } | |
3116 | ||
3117 | /* Looks like we have a winner */ | |
3118 | mddev_suspend(mddev); | |
3119 | mddev->pers->stop(mddev); | |
a64c876f N |
3120 | |
3121 | if (mddev->pers->sync_request == NULL && | |
3122 | pers->sync_request != NULL) { | |
3123 | /* need to add the md_redundancy_group */ | |
3124 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3125 | printk(KERN_WARNING | |
3126 | "md: cannot register extra attributes for %s\n", | |
3127 | mdname(mddev)); | |
19fdb9ee | 3128 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); |
a64c876f N |
3129 | } |
3130 | if (mddev->pers->sync_request != NULL && | |
3131 | pers->sync_request == NULL) { | |
3132 | /* need to remove the md_redundancy_group */ | |
3133 | if (mddev->to_remove == NULL) | |
3134 | mddev->to_remove = &md_redundancy_group; | |
3135 | } | |
3136 | ||
54071b38 TM |
3137 | if (mddev->pers->sync_request == NULL && |
3138 | mddev->external) { | |
3139 | /* We are converting from a no-redundancy array | |
3140 | * to a redundancy array and metadata is managed | |
3141 | * externally so we need to be sure that writes | |
3142 | * won't block due to a need to transition | |
3143 | * clean->dirty | |
3144 | * until external management is started. | |
3145 | */ | |
3146 | mddev->in_sync = 0; | |
3147 | mddev->safemode_delay = 0; | |
3148 | mddev->safemode = 0; | |
3149 | } | |
3150 | ||
e93f68a1 N |
3151 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
3152 | char nm[20]; | |
3153 | if (rdev->raid_disk < 0) | |
3154 | continue; | |
bf2cb0da | 3155 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3156 | rdev->new_raid_disk = -1; |
3157 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3158 | continue; | |
3159 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3160 | sysfs_remove_link(&mddev->kobj, nm); | |
3161 | } | |
3162 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
3163 | if (rdev->raid_disk < 0) | |
3164 | continue; | |
3165 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3166 | continue; | |
3167 | rdev->raid_disk = rdev->new_raid_disk; | |
3168 | if (rdev->raid_disk < 0) | |
3a981b03 | 3169 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 N |
3170 | else { |
3171 | char nm[20]; | |
3172 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3173 | if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) | |
3174 | printk("md: cannot register %s for %s after level change\n", | |
3175 | nm, mdname(mddev)); | |
3a981b03 | 3176 | } |
e93f68a1 N |
3177 | } |
3178 | ||
3179 | module_put(mddev->pers->owner); | |
245f46c2 N |
3180 | mddev->pers = pers; |
3181 | mddev->private = priv; | |
3182 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3183 | mddev->level = mddev->new_level; | |
3184 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3185 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3186 | mddev->delta_disks = 0; |
fee68723 | 3187 | mddev->degraded = 0; |
9af204cf TM |
3188 | if (mddev->pers->sync_request == NULL) { |
3189 | /* this is now an array without redundancy, so | |
3190 | * it must always be in_sync | |
3191 | */ | |
3192 | mddev->in_sync = 1; | |
3193 | del_timer_sync(&mddev->safemode_timer); | |
3194 | } | |
245f46c2 N |
3195 | pers->run(mddev); |
3196 | mddev_resume(mddev); | |
3197 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3198 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3199 | md_wakeup_thread(mddev->thread); | |
5cac7861 | 3200 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3201 | md_new_event(mddev); |
d9d166c2 N |
3202 | return rv; |
3203 | } | |
3204 | ||
3205 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3206 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3207 | |
d4dbd025 N |
3208 | |
3209 | static ssize_t | |
3210 | layout_show(mddev_t *mddev, char *page) | |
3211 | { | |
3212 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3213 | if (mddev->reshape_position != MaxSector && |
3214 | mddev->layout != mddev->new_layout) | |
3215 | return sprintf(page, "%d (%d)\n", | |
3216 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3217 | return sprintf(page, "%d\n", mddev->layout); |
3218 | } | |
3219 | ||
3220 | static ssize_t | |
3221 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
3222 | { | |
3223 | char *e; | |
3224 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3225 | |
3226 | if (!*buf || (*e && *e != '\n')) | |
3227 | return -EINVAL; | |
3228 | ||
b3546035 N |
3229 | if (mddev->pers) { |
3230 | int err; | |
50ac168a | 3231 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3232 | return -EBUSY; |
597a711b | 3233 | mddev->new_layout = n; |
50ac168a | 3234 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3235 | if (err) { |
3236 | mddev->new_layout = mddev->layout; | |
b3546035 | 3237 | return err; |
597a711b | 3238 | } |
b3546035 | 3239 | } else { |
08a02ecd | 3240 | mddev->new_layout = n; |
b3546035 N |
3241 | if (mddev->reshape_position == MaxSector) |
3242 | mddev->layout = n; | |
3243 | } | |
d4dbd025 N |
3244 | return len; |
3245 | } | |
3246 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3247 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3248 | |
3249 | ||
eae1701f | 3250 | static ssize_t |
96de1e66 | 3251 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 3252 | { |
bb636547 N |
3253 | if (mddev->raid_disks == 0) |
3254 | return 0; | |
08a02ecd N |
3255 | if (mddev->reshape_position != MaxSector && |
3256 | mddev->delta_disks != 0) | |
3257 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3258 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3259 | return sprintf(page, "%d\n", mddev->raid_disks); |
3260 | } | |
3261 | ||
da943b99 N |
3262 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
3263 | ||
3264 | static ssize_t | |
3265 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
3266 | { | |
da943b99 N |
3267 | char *e; |
3268 | int rv = 0; | |
3269 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3270 | ||
3271 | if (!*buf || (*e && *e != '\n')) | |
3272 | return -EINVAL; | |
3273 | ||
3274 | if (mddev->pers) | |
3275 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
3276 | else if (mddev->reshape_position != MaxSector) { |
3277 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
3278 | mddev->delta_disks = n - olddisks; | |
3279 | mddev->raid_disks = n; | |
3280 | } else | |
da943b99 N |
3281 | mddev->raid_disks = n; |
3282 | return rv ? rv : len; | |
3283 | } | |
3284 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3285 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3286 | |
3b34380a N |
3287 | static ssize_t |
3288 | chunk_size_show(mddev_t *mddev, char *page) | |
3289 | { | |
08a02ecd | 3290 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3291 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3292 | return sprintf(page, "%d (%d)\n", | |
3293 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3294 | mddev->chunk_sectors << 9); |
3295 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3296 | } |
3297 | ||
3298 | static ssize_t | |
3299 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3300 | { | |
3b34380a N |
3301 | char *e; |
3302 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3303 | ||
3b34380a N |
3304 | if (!*buf || (*e && *e != '\n')) |
3305 | return -EINVAL; | |
3306 | ||
b3546035 N |
3307 | if (mddev->pers) { |
3308 | int err; | |
50ac168a | 3309 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3310 | return -EBUSY; |
597a711b | 3311 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3312 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3313 | if (err) { |
3314 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3315 | return err; |
597a711b | 3316 | } |
b3546035 | 3317 | } else { |
664e7c41 | 3318 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3319 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3320 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3321 | } |
3b34380a N |
3322 | return len; |
3323 | } | |
3324 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3325 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3326 | |
a94213b1 N |
3327 | static ssize_t |
3328 | resync_start_show(mddev_t *mddev, char *page) | |
3329 | { | |
d1a7c503 N |
3330 | if (mddev->recovery_cp == MaxSector) |
3331 | return sprintf(page, "none\n"); | |
a94213b1 N |
3332 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3333 | } | |
3334 | ||
3335 | static ssize_t | |
3336 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
3337 | { | |
a94213b1 N |
3338 | char *e; |
3339 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3340 | ||
b098636c | 3341 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
a94213b1 | 3342 | return -EBUSY; |
06e3c817 DW |
3343 | if (cmd_match(buf, "none")) |
3344 | n = MaxSector; | |
3345 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3346 | return -EINVAL; |
3347 | ||
3348 | mddev->recovery_cp = n; | |
3349 | return len; | |
3350 | } | |
3351 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3352 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3353 | |
9e653b63 N |
3354 | /* |
3355 | * The array state can be: | |
3356 | * | |
3357 | * clear | |
3358 | * No devices, no size, no level | |
3359 | * Equivalent to STOP_ARRAY ioctl | |
3360 | * inactive | |
3361 | * May have some settings, but array is not active | |
3362 | * all IO results in error | |
3363 | * When written, doesn't tear down array, but just stops it | |
3364 | * suspended (not supported yet) | |
3365 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3366 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3367 | * readonly |
3368 | * no resync can happen. no superblocks get written. | |
3369 | * write requests fail | |
3370 | * read-auto | |
3371 | * like readonly, but behaves like 'clean' on a write request. | |
3372 | * | |
3373 | * clean - no pending writes, but otherwise active. | |
3374 | * When written to inactive array, starts without resync | |
3375 | * If a write request arrives then | |
3376 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3377 | * if not known, block and switch to write-pending | |
3378 | * If written to an active array that has pending writes, then fails. | |
3379 | * active | |
3380 | * fully active: IO and resync can be happening. | |
3381 | * When written to inactive array, starts with resync | |
3382 | * | |
3383 | * write-pending | |
3384 | * clean, but writes are blocked waiting for 'active' to be written. | |
3385 | * | |
3386 | * active-idle | |
3387 | * like active, but no writes have been seen for a while (100msec). | |
3388 | * | |
3389 | */ | |
3390 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3391 | write_pending, active_idle, bad_word}; | |
05381954 | 3392 | static char *array_states[] = { |
9e653b63 N |
3393 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3394 | "write-pending", "active-idle", NULL }; | |
3395 | ||
3396 | static int match_word(const char *word, char **list) | |
3397 | { | |
3398 | int n; | |
3399 | for (n=0; list[n]; n++) | |
3400 | if (cmd_match(word, list[n])) | |
3401 | break; | |
3402 | return n; | |
3403 | } | |
3404 | ||
3405 | static ssize_t | |
3406 | array_state_show(mddev_t *mddev, char *page) | |
3407 | { | |
3408 | enum array_state st = inactive; | |
3409 | ||
3410 | if (mddev->pers) | |
3411 | switch(mddev->ro) { | |
3412 | case 1: | |
3413 | st = readonly; | |
3414 | break; | |
3415 | case 2: | |
3416 | st = read_auto; | |
3417 | break; | |
3418 | case 0: | |
3419 | if (mddev->in_sync) | |
3420 | st = clean; | |
070dc6dd | 3421 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3422 | st = write_pending; |
9e653b63 N |
3423 | else if (mddev->safemode) |
3424 | st = active_idle; | |
3425 | else | |
3426 | st = active; | |
3427 | } | |
3428 | else { | |
3429 | if (list_empty(&mddev->disks) && | |
3430 | mddev->raid_disks == 0 && | |
58c0fed4 | 3431 | mddev->dev_sectors == 0) |
9e653b63 N |
3432 | st = clear; |
3433 | else | |
3434 | st = inactive; | |
3435 | } | |
3436 | return sprintf(page, "%s\n", array_states[st]); | |
3437 | } | |
3438 | ||
df5b20cf | 3439 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
a4bd82d0 | 3440 | static int md_set_readonly(mddev_t * mddev, int is_open); |
9e653b63 N |
3441 | static int do_md_run(mddev_t * mddev); |
3442 | static int restart_array(mddev_t *mddev); | |
3443 | ||
3444 | static ssize_t | |
3445 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3446 | { | |
3447 | int err = -EINVAL; | |
3448 | enum array_state st = match_word(buf, array_states); | |
3449 | switch(st) { | |
3450 | case bad_word: | |
3451 | break; | |
3452 | case clear: | |
3453 | /* stopping an active array */ | |
f2ea68cf | 3454 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3455 | return -EBUSY; |
df5b20cf | 3456 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3457 | break; |
3458 | case inactive: | |
3459 | /* stopping an active array */ | |
3460 | if (mddev->pers) { | |
f2ea68cf | 3461 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3462 | return -EBUSY; |
df5b20cf | 3463 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3464 | } else |
3465 | err = 0; /* already inactive */ | |
9e653b63 N |
3466 | break; |
3467 | case suspended: | |
3468 | break; /* not supported yet */ | |
3469 | case readonly: | |
3470 | if (mddev->pers) | |
a4bd82d0 | 3471 | err = md_set_readonly(mddev, 0); |
9e653b63 N |
3472 | else { |
3473 | mddev->ro = 1; | |
648b629e | 3474 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3475 | err = do_md_run(mddev); |
3476 | } | |
3477 | break; | |
3478 | case read_auto: | |
9e653b63 | 3479 | if (mddev->pers) { |
80268ee9 | 3480 | if (mddev->ro == 0) |
a4bd82d0 | 3481 | err = md_set_readonly(mddev, 0); |
80268ee9 | 3482 | else if (mddev->ro == 1) |
648b629e N |
3483 | err = restart_array(mddev); |
3484 | if (err == 0) { | |
3485 | mddev->ro = 2; | |
3486 | set_disk_ro(mddev->gendisk, 0); | |
3487 | } | |
9e653b63 N |
3488 | } else { |
3489 | mddev->ro = 2; | |
3490 | err = do_md_run(mddev); | |
3491 | } | |
3492 | break; | |
3493 | case clean: | |
3494 | if (mddev->pers) { | |
3495 | restart_array(mddev); | |
3496 | spin_lock_irq(&mddev->write_lock); | |
3497 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3498 | if (mddev->in_sync == 0) { |
3499 | mddev->in_sync = 1; | |
31a59e34 N |
3500 | if (mddev->safemode == 1) |
3501 | mddev->safemode = 0; | |
070dc6dd | 3502 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3503 | } |
3504 | err = 0; | |
3505 | } else | |
3506 | err = -EBUSY; | |
9e653b63 | 3507 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3508 | } else |
3509 | err = -EINVAL; | |
9e653b63 N |
3510 | break; |
3511 | case active: | |
3512 | if (mddev->pers) { | |
3513 | restart_array(mddev); | |
070dc6dd | 3514 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3515 | wake_up(&mddev->sb_wait); |
3516 | err = 0; | |
3517 | } else { | |
3518 | mddev->ro = 0; | |
648b629e | 3519 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3520 | err = do_md_run(mddev); |
3521 | } | |
3522 | break; | |
3523 | case write_pending: | |
3524 | case active_idle: | |
3525 | /* these cannot be set */ | |
3526 | break; | |
3527 | } | |
3528 | if (err) | |
3529 | return err; | |
0fd62b86 | 3530 | else { |
00bcb4ac | 3531 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3532 | return len; |
0fd62b86 | 3533 | } |
9e653b63 | 3534 | } |
80ca3a44 N |
3535 | static struct md_sysfs_entry md_array_state = |
3536 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3537 | |
1e50915f RB |
3538 | static ssize_t |
3539 | max_corrected_read_errors_show(mddev_t *mddev, char *page) { | |
3540 | return sprintf(page, "%d\n", | |
3541 | atomic_read(&mddev->max_corr_read_errors)); | |
3542 | } | |
3543 | ||
3544 | static ssize_t | |
3545 | max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) | |
3546 | { | |
3547 | char *e; | |
3548 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3549 | ||
3550 | if (*buf && (*e == 0 || *e == '\n')) { | |
3551 | atomic_set(&mddev->max_corr_read_errors, n); | |
3552 | return len; | |
3553 | } | |
3554 | return -EINVAL; | |
3555 | } | |
3556 | ||
3557 | static struct md_sysfs_entry max_corr_read_errors = | |
3558 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3559 | max_corrected_read_errors_store); | |
3560 | ||
6d7ff738 N |
3561 | static ssize_t |
3562 | null_show(mddev_t *mddev, char *page) | |
3563 | { | |
3564 | return -EINVAL; | |
3565 | } | |
3566 | ||
3567 | static ssize_t | |
3568 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3569 | { | |
3570 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3571 | /* The new device is added to the array. | |
3572 | * If the array has a persistent superblock, we read the | |
3573 | * superblock to initialise info and check validity. | |
3574 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3575 | * which mainly checks size. | |
3576 | */ | |
3577 | char *e; | |
3578 | int major = simple_strtoul(buf, &e, 10); | |
3579 | int minor; | |
3580 | dev_t dev; | |
3581 | mdk_rdev_t *rdev; | |
3582 | int err; | |
3583 | ||
3584 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3585 | return -EINVAL; | |
3586 | minor = simple_strtoul(e+1, &e, 10); | |
3587 | if (*e && *e != '\n') | |
3588 | return -EINVAL; | |
3589 | dev = MKDEV(major, minor); | |
3590 | if (major != MAJOR(dev) || | |
3591 | minor != MINOR(dev)) | |
3592 | return -EOVERFLOW; | |
3593 | ||
3594 | ||
3595 | if (mddev->persistent) { | |
3596 | rdev = md_import_device(dev, mddev->major_version, | |
3597 | mddev->minor_version); | |
3598 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3599 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3600 | mdk_rdev_t, same_set); | |
3601 | err = super_types[mddev->major_version] | |
3602 | .load_super(rdev, rdev0, mddev->minor_version); | |
3603 | if (err < 0) | |
3604 | goto out; | |
3605 | } | |
c5d79adb N |
3606 | } else if (mddev->external) |
3607 | rdev = md_import_device(dev, -2, -1); | |
3608 | else | |
6d7ff738 N |
3609 | rdev = md_import_device(dev, -1, -1); |
3610 | ||
3611 | if (IS_ERR(rdev)) | |
3612 | return PTR_ERR(rdev); | |
3613 | err = bind_rdev_to_array(rdev, mddev); | |
3614 | out: | |
3615 | if (err) | |
3616 | export_rdev(rdev); | |
3617 | return err ? err : len; | |
3618 | } | |
3619 | ||
3620 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3621 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3622 | |
9b1d1dac PC |
3623 | static ssize_t |
3624 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3625 | { | |
3626 | char *end; | |
3627 | unsigned long chunk, end_chunk; | |
3628 | ||
3629 | if (!mddev->bitmap) | |
3630 | goto out; | |
3631 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3632 | while (*buf) { | |
3633 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3634 | if (buf == end) break; | |
3635 | if (*end == '-') { /* range */ | |
3636 | buf = end + 1; | |
3637 | end_chunk = simple_strtoul(buf, &end, 0); | |
3638 | if (buf == end) break; | |
3639 | } | |
3640 | if (*end && !isspace(*end)) break; | |
3641 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3642 | buf = skip_spaces(end); |
9b1d1dac PC |
3643 | } |
3644 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3645 | out: | |
3646 | return len; | |
3647 | } | |
3648 | ||
3649 | static struct md_sysfs_entry md_bitmap = | |
3650 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3651 | ||
a35b0d69 N |
3652 | static ssize_t |
3653 | size_show(mddev_t *mddev, char *page) | |
3654 | { | |
58c0fed4 AN |
3655 | return sprintf(page, "%llu\n", |
3656 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3657 | } |
3658 | ||
d71f9f88 | 3659 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3660 | |
3661 | static ssize_t | |
3662 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3663 | { | |
3664 | /* If array is inactive, we can reduce the component size, but | |
3665 | * not increase it (except from 0). | |
3666 | * If array is active, we can try an on-line resize | |
3667 | */ | |
b522adcd DW |
3668 | sector_t sectors; |
3669 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3670 | |
58c0fed4 AN |
3671 | if (err < 0) |
3672 | return err; | |
a35b0d69 | 3673 | if (mddev->pers) { |
58c0fed4 | 3674 | err = update_size(mddev, sectors); |
850b2b42 | 3675 | md_update_sb(mddev, 1); |
a35b0d69 | 3676 | } else { |
58c0fed4 AN |
3677 | if (mddev->dev_sectors == 0 || |
3678 | mddev->dev_sectors > sectors) | |
3679 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3680 | else |
3681 | err = -ENOSPC; | |
3682 | } | |
3683 | return err ? err : len; | |
3684 | } | |
3685 | ||
3686 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3687 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3688 | |
8bb93aac N |
3689 | |
3690 | /* Metdata version. | |
e691063a N |
3691 | * This is one of |
3692 | * 'none' for arrays with no metadata (good luck...) | |
3693 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3694 | * or N.M for internally known formats |
3695 | */ | |
3696 | static ssize_t | |
3697 | metadata_show(mddev_t *mddev, char *page) | |
3698 | { | |
3699 | if (mddev->persistent) | |
3700 | return sprintf(page, "%d.%d\n", | |
3701 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3702 | else if (mddev->external) |
3703 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3704 | else |
3705 | return sprintf(page, "none\n"); | |
3706 | } | |
3707 | ||
3708 | static ssize_t | |
3709 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3710 | { | |
3711 | int major, minor; | |
3712 | char *e; | |
ea43ddd8 N |
3713 | /* Changing the details of 'external' metadata is |
3714 | * always permitted. Otherwise there must be | |
3715 | * no devices attached to the array. | |
3716 | */ | |
3717 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3718 | ; | |
3719 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3720 | return -EBUSY; |
3721 | ||
3722 | if (cmd_match(buf, "none")) { | |
3723 | mddev->persistent = 0; | |
e691063a N |
3724 | mddev->external = 0; |
3725 | mddev->major_version = 0; | |
3726 | mddev->minor_version = 90; | |
3727 | return len; | |
3728 | } | |
3729 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3730 | size_t namelen = len-9; |
e691063a N |
3731 | if (namelen >= sizeof(mddev->metadata_type)) |
3732 | namelen = sizeof(mddev->metadata_type)-1; | |
3733 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3734 | mddev->metadata_type[namelen] = 0; | |
3735 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3736 | mddev->metadata_type[--namelen] = 0; | |
3737 | mddev->persistent = 0; | |
3738 | mddev->external = 1; | |
8bb93aac N |
3739 | mddev->major_version = 0; |
3740 | mddev->minor_version = 90; | |
3741 | return len; | |
3742 | } | |
3743 | major = simple_strtoul(buf, &e, 10); | |
3744 | if (e==buf || *e != '.') | |
3745 | return -EINVAL; | |
3746 | buf = e+1; | |
3747 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3748 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3749 | return -EINVAL; |
50511da3 | 3750 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3751 | return -ENOENT; |
3752 | mddev->major_version = major; | |
3753 | mddev->minor_version = minor; | |
3754 | mddev->persistent = 1; | |
e691063a | 3755 | mddev->external = 0; |
8bb93aac N |
3756 | return len; |
3757 | } | |
3758 | ||
3759 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3760 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3761 | |
24dd469d | 3762 | static ssize_t |
7eec314d | 3763 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3764 | { |
7eec314d | 3765 | char *type = "idle"; |
b6a9ce68 N |
3766 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3767 | type = "frozen"; | |
3768 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3769 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3770 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3771 | type = "reshape"; | |
3772 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3773 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3774 | type = "resync"; | |
3775 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3776 | type = "check"; | |
3777 | else | |
3778 | type = "repair"; | |
72a23c21 | 3779 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3780 | type = "recover"; |
3781 | } | |
3782 | return sprintf(page, "%s\n", type); | |
3783 | } | |
3784 | ||
7ebc0be7 N |
3785 | static void reap_sync_thread(mddev_t *mddev); |
3786 | ||
24dd469d | 3787 | static ssize_t |
7eec314d | 3788 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3789 | { |
7eec314d N |
3790 | if (!mddev->pers || !mddev->pers->sync_request) |
3791 | return -EINVAL; | |
3792 | ||
b6a9ce68 N |
3793 | if (cmd_match(page, "frozen")) |
3794 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3795 | else | |
3796 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3797 | ||
3798 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3799 | if (mddev->sync_thread) { |
3800 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 3801 | reap_sync_thread(mddev); |
7eec314d | 3802 | } |
03c902e1 N |
3803 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3804 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3805 | return -EBUSY; |
72a23c21 NB |
3806 | else if (cmd_match(page, "resync")) |
3807 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3808 | else if (cmd_match(page, "recover")) { | |
3809 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3810 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3811 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3812 | int err; |
3813 | if (mddev->pers->start_reshape == NULL) | |
3814 | return -EINVAL; | |
3815 | err = mddev->pers->start_reshape(mddev); | |
3816 | if (err) | |
3817 | return err; | |
a99ac971 | 3818 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3819 | } else { |
bce74dac | 3820 | if (cmd_match(page, "check")) |
7eec314d | 3821 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3822 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3823 | return -EINVAL; |
3824 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3825 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3826 | } |
03c902e1 | 3827 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3828 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 3829 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
3830 | return len; |
3831 | } | |
3832 | ||
9d88883e | 3833 | static ssize_t |
96de1e66 | 3834 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3835 | { |
3836 | return sprintf(page, "%llu\n", | |
3837 | (unsigned long long) mddev->resync_mismatches); | |
3838 | } | |
3839 | ||
80ca3a44 N |
3840 | static struct md_sysfs_entry md_scan_mode = |
3841 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3842 | |
96de1e66 | 3843 | |
80ca3a44 | 3844 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3845 | |
88202a0c N |
3846 | static ssize_t |
3847 | sync_min_show(mddev_t *mddev, char *page) | |
3848 | { | |
3849 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3850 | mddev->sync_speed_min ? "local": "system"); | |
3851 | } | |
3852 | ||
3853 | static ssize_t | |
3854 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3855 | { | |
3856 | int min; | |
3857 | char *e; | |
3858 | if (strncmp(buf, "system", 6)==0) { | |
3859 | mddev->sync_speed_min = 0; | |
3860 | return len; | |
3861 | } | |
3862 | min = simple_strtoul(buf, &e, 10); | |
3863 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3864 | return -EINVAL; | |
3865 | mddev->sync_speed_min = min; | |
3866 | return len; | |
3867 | } | |
3868 | ||
3869 | static struct md_sysfs_entry md_sync_min = | |
3870 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3871 | ||
3872 | static ssize_t | |
3873 | sync_max_show(mddev_t *mddev, char *page) | |
3874 | { | |
3875 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3876 | mddev->sync_speed_max ? "local": "system"); | |
3877 | } | |
3878 | ||
3879 | static ssize_t | |
3880 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3881 | { | |
3882 | int max; | |
3883 | char *e; | |
3884 | if (strncmp(buf, "system", 6)==0) { | |
3885 | mddev->sync_speed_max = 0; | |
3886 | return len; | |
3887 | } | |
3888 | max = simple_strtoul(buf, &e, 10); | |
3889 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3890 | return -EINVAL; | |
3891 | mddev->sync_speed_max = max; | |
3892 | return len; | |
3893 | } | |
3894 | ||
3895 | static struct md_sysfs_entry md_sync_max = | |
3896 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3897 | ||
d7f3d291 IP |
3898 | static ssize_t |
3899 | degraded_show(mddev_t *mddev, char *page) | |
3900 | { | |
3901 | return sprintf(page, "%d\n", mddev->degraded); | |
3902 | } | |
3903 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3904 | |
90b08710 BS |
3905 | static ssize_t |
3906 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3907 | { | |
3908 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3909 | } | |
3910 | ||
3911 | static ssize_t | |
3912 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3913 | { | |
3914 | long n; | |
3915 | ||
3916 | if (strict_strtol(buf, 10, &n)) | |
3917 | return -EINVAL; | |
3918 | ||
3919 | if (n != 0 && n != 1) | |
3920 | return -EINVAL; | |
3921 | ||
3922 | mddev->parallel_resync = n; | |
3923 | ||
3924 | if (mddev->sync_thread) | |
3925 | wake_up(&resync_wait); | |
3926 | ||
3927 | return len; | |
3928 | } | |
3929 | ||
3930 | /* force parallel resync, even with shared block devices */ | |
3931 | static struct md_sysfs_entry md_sync_force_parallel = | |
3932 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3933 | sync_force_parallel_show, sync_force_parallel_store); | |
3934 | ||
88202a0c N |
3935 | static ssize_t |
3936 | sync_speed_show(mddev_t *mddev, char *page) | |
3937 | { | |
3938 | unsigned long resync, dt, db; | |
d1a7c503 N |
3939 | if (mddev->curr_resync == 0) |
3940 | return sprintf(page, "none\n"); | |
9687a60c AN |
3941 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3942 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3943 | if (!dt) dt++; |
9687a60c AN |
3944 | db = resync - mddev->resync_mark_cnt; |
3945 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3946 | } |
3947 | ||
80ca3a44 | 3948 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3949 | |
3950 | static ssize_t | |
3951 | sync_completed_show(mddev_t *mddev, char *page) | |
3952 | { | |
13ae864b | 3953 | unsigned long long max_sectors, resync; |
88202a0c | 3954 | |
acb180b0 N |
3955 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3956 | return sprintf(page, "none\n"); | |
3957 | ||
88202a0c | 3958 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3959 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3960 | else |
58c0fed4 | 3961 | max_sectors = mddev->dev_sectors; |
88202a0c | 3962 | |
acb180b0 | 3963 | resync = mddev->curr_resync_completed; |
13ae864b | 3964 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
3965 | } |
3966 | ||
80ca3a44 | 3967 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3968 | |
5e96ee65 NB |
3969 | static ssize_t |
3970 | min_sync_show(mddev_t *mddev, char *page) | |
3971 | { | |
3972 | return sprintf(page, "%llu\n", | |
3973 | (unsigned long long)mddev->resync_min); | |
3974 | } | |
3975 | static ssize_t | |
3976 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3977 | { | |
3978 | unsigned long long min; | |
3979 | if (strict_strtoull(buf, 10, &min)) | |
3980 | return -EINVAL; | |
3981 | if (min > mddev->resync_max) | |
3982 | return -EINVAL; | |
3983 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3984 | return -EBUSY; | |
3985 | ||
3986 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3987 | if (mddev->chunk_sectors) { |
2ac06c33 | 3988 | sector_t temp = min; |
9d8f0363 | 3989 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3990 | return -EINVAL; |
3991 | } | |
3992 | mddev->resync_min = min; | |
3993 | ||
3994 | return len; | |
3995 | } | |
3996 | ||
3997 | static struct md_sysfs_entry md_min_sync = | |
3998 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3999 | ||
c6207277 N |
4000 | static ssize_t |
4001 | max_sync_show(mddev_t *mddev, char *page) | |
4002 | { | |
4003 | if (mddev->resync_max == MaxSector) | |
4004 | return sprintf(page, "max\n"); | |
4005 | else | |
4006 | return sprintf(page, "%llu\n", | |
4007 | (unsigned long long)mddev->resync_max); | |
4008 | } | |
4009 | static ssize_t | |
4010 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
4011 | { | |
4012 | if (strncmp(buf, "max", 3) == 0) | |
4013 | mddev->resync_max = MaxSector; | |
4014 | else { | |
5e96ee65 NB |
4015 | unsigned long long max; |
4016 | if (strict_strtoull(buf, 10, &max)) | |
4017 | return -EINVAL; | |
4018 | if (max < mddev->resync_min) | |
c6207277 N |
4019 | return -EINVAL; |
4020 | if (max < mddev->resync_max && | |
4d484a4a | 4021 | mddev->ro == 0 && |
c6207277 N |
4022 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4023 | return -EBUSY; | |
4024 | ||
4025 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4026 | if (mddev->chunk_sectors) { |
2ac06c33 | 4027 | sector_t temp = max; |
9d8f0363 | 4028 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
4029 | return -EINVAL; |
4030 | } | |
4031 | mddev->resync_max = max; | |
4032 | } | |
4033 | wake_up(&mddev->recovery_wait); | |
4034 | return len; | |
4035 | } | |
4036 | ||
4037 | static struct md_sysfs_entry md_max_sync = | |
4038 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4039 | ||
e464eafd N |
4040 | static ssize_t |
4041 | suspend_lo_show(mddev_t *mddev, char *page) | |
4042 | { | |
4043 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4044 | } | |
4045 | ||
4046 | static ssize_t | |
4047 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
4048 | { | |
4049 | char *e; | |
4050 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4051 | unsigned long long old = mddev->suspend_lo; |
e464eafd | 4052 | |
b8d966ef N |
4053 | if (mddev->pers == NULL || |
4054 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4055 | return -EINVAL; |
4056 | if (buf == e || (*e && *e != '\n')) | |
4057 | return -EINVAL; | |
23ddff37 N |
4058 | |
4059 | mddev->suspend_lo = new; | |
4060 | if (new >= old) | |
4061 | /* Shrinking suspended region */ | |
e464eafd | 4062 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4063 | else { |
4064 | /* Expanding suspended region - need to wait */ | |
4065 | mddev->pers->quiesce(mddev, 1); | |
4066 | mddev->pers->quiesce(mddev, 0); | |
4067 | } | |
4068 | return len; | |
e464eafd N |
4069 | } |
4070 | static struct md_sysfs_entry md_suspend_lo = | |
4071 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4072 | ||
4073 | ||
4074 | static ssize_t | |
4075 | suspend_hi_show(mddev_t *mddev, char *page) | |
4076 | { | |
4077 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4078 | } | |
4079 | ||
4080 | static ssize_t | |
4081 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
4082 | { | |
4083 | char *e; | |
4084 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4085 | unsigned long long old = mddev->suspend_hi; |
e464eafd | 4086 | |
b8d966ef N |
4087 | if (mddev->pers == NULL || |
4088 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4089 | return -EINVAL; |
4090 | if (buf == e || (*e && *e != '\n')) | |
4091 | return -EINVAL; | |
23ddff37 N |
4092 | |
4093 | mddev->suspend_hi = new; | |
4094 | if (new <= old) | |
4095 | /* Shrinking suspended region */ | |
4096 | mddev->pers->quiesce(mddev, 2); | |
4097 | else { | |
4098 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4099 | mddev->pers->quiesce(mddev, 1); |
4100 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 N |
4101 | } |
4102 | return len; | |
e464eafd N |
4103 | } |
4104 | static struct md_sysfs_entry md_suspend_hi = | |
4105 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4106 | ||
08a02ecd N |
4107 | static ssize_t |
4108 | reshape_position_show(mddev_t *mddev, char *page) | |
4109 | { | |
4110 | if (mddev->reshape_position != MaxSector) | |
4111 | return sprintf(page, "%llu\n", | |
4112 | (unsigned long long)mddev->reshape_position); | |
4113 | strcpy(page, "none\n"); | |
4114 | return 5; | |
4115 | } | |
4116 | ||
4117 | static ssize_t | |
4118 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
4119 | { | |
4120 | char *e; | |
4121 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4122 | if (mddev->pers) | |
4123 | return -EBUSY; | |
4124 | if (buf == e || (*e && *e != '\n')) | |
4125 | return -EINVAL; | |
4126 | mddev->reshape_position = new; | |
4127 | mddev->delta_disks = 0; | |
4128 | mddev->new_level = mddev->level; | |
4129 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4130 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
4131 | return len; |
4132 | } | |
4133 | ||
4134 | static struct md_sysfs_entry md_reshape_position = | |
4135 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4136 | reshape_position_store); | |
4137 | ||
b522adcd DW |
4138 | static ssize_t |
4139 | array_size_show(mddev_t *mddev, char *page) | |
4140 | { | |
4141 | if (mddev->external_size) | |
4142 | return sprintf(page, "%llu\n", | |
4143 | (unsigned long long)mddev->array_sectors/2); | |
4144 | else | |
4145 | return sprintf(page, "default\n"); | |
4146 | } | |
4147 | ||
4148 | static ssize_t | |
4149 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
4150 | { | |
4151 | sector_t sectors; | |
4152 | ||
4153 | if (strncmp(buf, "default", 7) == 0) { | |
4154 | if (mddev->pers) | |
4155 | sectors = mddev->pers->size(mddev, 0, 0); | |
4156 | else | |
4157 | sectors = mddev->array_sectors; | |
4158 | ||
4159 | mddev->external_size = 0; | |
4160 | } else { | |
4161 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4162 | return -EINVAL; | |
4163 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4164 | return -E2BIG; |
b522adcd DW |
4165 | |
4166 | mddev->external_size = 1; | |
4167 | } | |
4168 | ||
4169 | mddev->array_sectors = sectors; | |
cbe6ef1d N |
4170 | if (mddev->pers) { |
4171 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 4172 | revalidate_disk(mddev->gendisk); |
cbe6ef1d | 4173 | } |
b522adcd DW |
4174 | return len; |
4175 | } | |
4176 | ||
4177 | static struct md_sysfs_entry md_array_size = | |
4178 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4179 | array_size_store); | |
e464eafd | 4180 | |
eae1701f N |
4181 | static struct attribute *md_default_attrs[] = { |
4182 | &md_level.attr, | |
d4dbd025 | 4183 | &md_layout.attr, |
eae1701f | 4184 | &md_raid_disks.attr, |
3b34380a | 4185 | &md_chunk_size.attr, |
a35b0d69 | 4186 | &md_size.attr, |
a94213b1 | 4187 | &md_resync_start.attr, |
8bb93aac | 4188 | &md_metadata.attr, |
6d7ff738 | 4189 | &md_new_device.attr, |
16f17b39 | 4190 | &md_safe_delay.attr, |
9e653b63 | 4191 | &md_array_state.attr, |
08a02ecd | 4192 | &md_reshape_position.attr, |
b522adcd | 4193 | &md_array_size.attr, |
1e50915f | 4194 | &max_corr_read_errors.attr, |
411036fa N |
4195 | NULL, |
4196 | }; | |
4197 | ||
4198 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4199 | &md_scan_mode.attr, |
9d88883e | 4200 | &md_mismatches.attr, |
88202a0c N |
4201 | &md_sync_min.attr, |
4202 | &md_sync_max.attr, | |
4203 | &md_sync_speed.attr, | |
90b08710 | 4204 | &md_sync_force_parallel.attr, |
88202a0c | 4205 | &md_sync_completed.attr, |
5e96ee65 | 4206 | &md_min_sync.attr, |
c6207277 | 4207 | &md_max_sync.attr, |
e464eafd N |
4208 | &md_suspend_lo.attr, |
4209 | &md_suspend_hi.attr, | |
9b1d1dac | 4210 | &md_bitmap.attr, |
d7f3d291 | 4211 | &md_degraded.attr, |
eae1701f N |
4212 | NULL, |
4213 | }; | |
411036fa N |
4214 | static struct attribute_group md_redundancy_group = { |
4215 | .name = NULL, | |
4216 | .attrs = md_redundancy_attrs, | |
4217 | }; | |
4218 | ||
eae1701f N |
4219 | |
4220 | static ssize_t | |
4221 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4222 | { | |
4223 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4224 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4225 | ssize_t rv; |
eae1701f N |
4226 | |
4227 | if (!entry->show) | |
4228 | return -EIO; | |
5dc5cf7d IM |
4229 | rv = mddev_lock(mddev); |
4230 | if (!rv) { | |
4231 | rv = entry->show(mddev, page); | |
4232 | mddev_unlock(mddev); | |
4233 | } | |
96de1e66 | 4234 | return rv; |
eae1701f N |
4235 | } |
4236 | ||
4237 | static ssize_t | |
4238 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4239 | const char *page, size_t length) | |
4240 | { | |
4241 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4242 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4243 | ssize_t rv; |
eae1701f N |
4244 | |
4245 | if (!entry->store) | |
4246 | return -EIO; | |
67463acb N |
4247 | if (!capable(CAP_SYS_ADMIN)) |
4248 | return -EACCES; | |
5dc5cf7d | 4249 | rv = mddev_lock(mddev); |
d3374825 N |
4250 | if (mddev->hold_active == UNTIL_IOCTL) |
4251 | mddev->hold_active = 0; | |
5dc5cf7d IM |
4252 | if (!rv) { |
4253 | rv = entry->store(mddev, page, length); | |
4254 | mddev_unlock(mddev); | |
4255 | } | |
96de1e66 | 4256 | return rv; |
eae1701f N |
4257 | } |
4258 | ||
4259 | static void md_free(struct kobject *ko) | |
4260 | { | |
4261 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
4262 | |
4263 | if (mddev->sysfs_state) | |
4264 | sysfs_put(mddev->sysfs_state); | |
4265 | ||
4266 | if (mddev->gendisk) { | |
4267 | del_gendisk(mddev->gendisk); | |
4268 | put_disk(mddev->gendisk); | |
4269 | } | |
4270 | if (mddev->queue) | |
4271 | blk_cleanup_queue(mddev->queue); | |
4272 | ||
eae1701f N |
4273 | kfree(mddev); |
4274 | } | |
4275 | ||
52cf25d0 | 4276 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4277 | .show = md_attr_show, |
4278 | .store = md_attr_store, | |
4279 | }; | |
4280 | static struct kobj_type md_ktype = { | |
4281 | .release = md_free, | |
4282 | .sysfs_ops = &md_sysfs_ops, | |
4283 | .default_attrs = md_default_attrs, | |
4284 | }; | |
4285 | ||
1da177e4 LT |
4286 | int mdp_major = 0; |
4287 | ||
5fd3a17e DW |
4288 | static void mddev_delayed_delete(struct work_struct *ws) |
4289 | { | |
4290 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
4291 | ||
43a70507 | 4292 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4293 | kobject_del(&mddev->kobj); |
4294 | kobject_put(&mddev->kobj); | |
4295 | } | |
4296 | ||
efeb53c0 | 4297 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4298 | { |
48c9c27b | 4299 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
4300 | mddev_t *mddev = mddev_find(dev); |
4301 | struct gendisk *disk; | |
efeb53c0 N |
4302 | int partitioned; |
4303 | int shift; | |
4304 | int unit; | |
3830c62f | 4305 | int error; |
1da177e4 LT |
4306 | |
4307 | if (!mddev) | |
efeb53c0 N |
4308 | return -ENODEV; |
4309 | ||
4310 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4311 | shift = partitioned ? MdpMinorShift : 0; | |
4312 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4313 | |
e804ac78 TH |
4314 | /* wait for any previous instance of this device to be |
4315 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4316 | */ |
e804ac78 | 4317 | flush_workqueue(md_misc_wq); |
d3374825 | 4318 | |
48c9c27b | 4319 | mutex_lock(&disks_mutex); |
0909dc44 N |
4320 | error = -EEXIST; |
4321 | if (mddev->gendisk) | |
4322 | goto abort; | |
efeb53c0 N |
4323 | |
4324 | if (name) { | |
4325 | /* Need to ensure that 'name' is not a duplicate. | |
4326 | */ | |
4327 | mddev_t *mddev2; | |
4328 | spin_lock(&all_mddevs_lock); | |
4329 | ||
4330 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4331 | if (mddev2->gendisk && | |
4332 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4333 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4334 | goto abort; |
efeb53c0 N |
4335 | } |
4336 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4337 | } |
8b765398 | 4338 | |
0909dc44 | 4339 | error = -ENOMEM; |
8b765398 | 4340 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4341 | if (!mddev->queue) |
4342 | goto abort; | |
409c57f3 N |
4343 | mddev->queue->queuedata = mddev; |
4344 | ||
409c57f3 | 4345 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 4346 | |
1da177e4 LT |
4347 | disk = alloc_disk(1 << shift); |
4348 | if (!disk) { | |
8b765398 N |
4349 | blk_cleanup_queue(mddev->queue); |
4350 | mddev->queue = NULL; | |
0909dc44 | 4351 | goto abort; |
1da177e4 | 4352 | } |
efeb53c0 | 4353 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4354 | disk->first_minor = unit << shift; |
efeb53c0 N |
4355 | if (name) |
4356 | strcpy(disk->disk_name, name); | |
4357 | else if (partitioned) | |
1da177e4 | 4358 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4359 | else |
1da177e4 | 4360 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4361 | disk->fops = &md_fops; |
4362 | disk->private_data = mddev; | |
4363 | disk->queue = mddev->queue; | |
b0140891 | 4364 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4365 | /* Allow extended partitions. This makes the |
d3374825 | 4366 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4367 | * remove it now. |
4368 | */ | |
4369 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4370 | mddev->gendisk = disk; |
b0140891 N |
4371 | /* As soon as we call add_disk(), another thread could get |
4372 | * through to md_open, so make sure it doesn't get too far | |
4373 | */ | |
4374 | mutex_lock(&mddev->open_mutex); | |
4375 | add_disk(disk); | |
4376 | ||
ed9e1982 TH |
4377 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4378 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4379 | if (error) { |
4380 | /* This isn't possible, but as kobject_init_and_add is marked | |
4381 | * __must_check, we must do something with the result | |
4382 | */ | |
5e55e2f5 N |
4383 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4384 | disk->disk_name); | |
0909dc44 N |
4385 | error = 0; |
4386 | } | |
00bcb4ac N |
4387 | if (mddev->kobj.sd && |
4388 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4389 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4390 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4391 | abort: |
4392 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4393 | if (!error && mddev->kobj.sd) { |
3830c62f | 4394 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4395 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4396 | } |
d3374825 | 4397 | mddev_put(mddev); |
0909dc44 | 4398 | return error; |
efeb53c0 N |
4399 | } |
4400 | ||
4401 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4402 | { | |
4403 | md_alloc(dev, NULL); | |
1da177e4 LT |
4404 | return NULL; |
4405 | } | |
4406 | ||
efeb53c0 N |
4407 | static int add_named_array(const char *val, struct kernel_param *kp) |
4408 | { | |
4409 | /* val must be "md_*" where * is not all digits. | |
4410 | * We allocate an array with a large free minor number, and | |
4411 | * set the name to val. val must not already be an active name. | |
4412 | */ | |
4413 | int len = strlen(val); | |
4414 | char buf[DISK_NAME_LEN]; | |
4415 | ||
4416 | while (len && val[len-1] == '\n') | |
4417 | len--; | |
4418 | if (len >= DISK_NAME_LEN) | |
4419 | return -E2BIG; | |
4420 | strlcpy(buf, val, len+1); | |
4421 | if (strncmp(buf, "md_", 3) != 0) | |
4422 | return -EINVAL; | |
4423 | return md_alloc(0, buf); | |
4424 | } | |
4425 | ||
1da177e4 LT |
4426 | static void md_safemode_timeout(unsigned long data) |
4427 | { | |
4428 | mddev_t *mddev = (mddev_t *) data; | |
4429 | ||
0fd62b86 NB |
4430 | if (!atomic_read(&mddev->writes_pending)) { |
4431 | mddev->safemode = 1; | |
4432 | if (mddev->external) | |
00bcb4ac | 4433 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4434 | } |
1da177e4 LT |
4435 | md_wakeup_thread(mddev->thread); |
4436 | } | |
4437 | ||
6ff8d8ec | 4438 | static int start_dirty_degraded; |
1da177e4 | 4439 | |
390ee602 | 4440 | int md_run(mddev_t *mddev) |
1da177e4 | 4441 | { |
2604b703 | 4442 | int err; |
1da177e4 | 4443 | mdk_rdev_t *rdev; |
2604b703 | 4444 | struct mdk_personality *pers; |
1da177e4 | 4445 | |
a757e64c N |
4446 | if (list_empty(&mddev->disks)) |
4447 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4448 | return -EINVAL; |
1da177e4 LT |
4449 | |
4450 | if (mddev->pers) | |
4451 | return -EBUSY; | |
bb4f1e9d N |
4452 | /* Cannot run until previous stop completes properly */ |
4453 | if (mddev->sysfs_active) | |
4454 | return -EBUSY; | |
b6eb127d | 4455 | |
1da177e4 LT |
4456 | /* |
4457 | * Analyze all RAID superblock(s) | |
4458 | */ | |
1ec4a939 N |
4459 | if (!mddev->raid_disks) { |
4460 | if (!mddev->persistent) | |
4461 | return -EINVAL; | |
a757e64c | 4462 | analyze_sbs(mddev); |
1ec4a939 | 4463 | } |
1da177e4 | 4464 | |
d9d166c2 N |
4465 | if (mddev->level != LEVEL_NONE) |
4466 | request_module("md-level-%d", mddev->level); | |
4467 | else if (mddev->clevel[0]) | |
4468 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4469 | |
4470 | /* | |
4471 | * Drop all container device buffers, from now on | |
4472 | * the only valid external interface is through the md | |
4473 | * device. | |
1da177e4 | 4474 | */ |
159ec1fc | 4475 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 4476 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4477 | continue; |
4478 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4479 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4480 | |
4481 | /* perform some consistency tests on the device. | |
4482 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4483 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4484 | */ |
a6ff7e08 JB |
4485 | if (rdev->meta_bdev) { |
4486 | /* Nothing to check */; | |
4487 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
4488 | if (mddev->dev_sectors && |
4489 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4490 | > rdev->sb_start) { |
f0d76d70 N |
4491 | printk("md: %s: data overlaps metadata\n", |
4492 | mdname(mddev)); | |
4493 | return -EINVAL; | |
4494 | } | |
4495 | } else { | |
0f420358 | 4496 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4497 | > rdev->data_offset) { |
4498 | printk("md: %s: metadata overlaps data\n", | |
4499 | mdname(mddev)); | |
4500 | return -EINVAL; | |
4501 | } | |
4502 | } | |
00bcb4ac | 4503 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4504 | } |
4505 | ||
a167f663 N |
4506 | if (mddev->bio_set == NULL) |
4507 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev)); | |
4508 | ||
1da177e4 | 4509 | spin_lock(&pers_lock); |
d9d166c2 | 4510 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4511 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4512 | spin_unlock(&pers_lock); |
d9d166c2 N |
4513 | if (mddev->level != LEVEL_NONE) |
4514 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4515 | mddev->level); | |
4516 | else | |
4517 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4518 | mddev->clevel); | |
1da177e4 LT |
4519 | return -EINVAL; |
4520 | } | |
2604b703 | 4521 | mddev->pers = pers; |
1da177e4 | 4522 | spin_unlock(&pers_lock); |
34817e8c N |
4523 | if (mddev->level != pers->level) { |
4524 | mddev->level = pers->level; | |
4525 | mddev->new_level = pers->level; | |
4526 | } | |
d9d166c2 | 4527 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4528 | |
f6705578 | 4529 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4530 | pers->start_reshape == NULL) { |
f6705578 N |
4531 | /* This personality cannot handle reshaping... */ |
4532 | mddev->pers = NULL; | |
4533 | module_put(pers->owner); | |
4534 | return -EINVAL; | |
4535 | } | |
4536 | ||
7dd5e7c3 N |
4537 | if (pers->sync_request) { |
4538 | /* Warn if this is a potentially silly | |
4539 | * configuration. | |
4540 | */ | |
4541 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4542 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4543 | int warned = 0; |
159ec1fc CR |
4544 | |
4545 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4546 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4547 | if (rdev < rdev2 && |
4548 | rdev->bdev->bd_contains == | |
4549 | rdev2->bdev->bd_contains) { | |
4550 | printk(KERN_WARNING | |
4551 | "%s: WARNING: %s appears to be" | |
4552 | " on the same physical disk as" | |
4553 | " %s.\n", | |
4554 | mdname(mddev), | |
4555 | bdevname(rdev->bdev,b), | |
4556 | bdevname(rdev2->bdev,b2)); | |
4557 | warned = 1; | |
4558 | } | |
4559 | } | |
159ec1fc | 4560 | |
7dd5e7c3 N |
4561 | if (warned) |
4562 | printk(KERN_WARNING | |
4563 | "True protection against single-disk" | |
4564 | " failure might be compromised.\n"); | |
4565 | } | |
4566 | ||
657390d2 | 4567 | mddev->recovery = 0; |
58c0fed4 AN |
4568 | /* may be over-ridden by personality */ |
4569 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4570 | ||
6ff8d8ec | 4571 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4572 | |
0f9552b5 | 4573 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
4574 | mddev->ro = 2; /* read-only, but switch on first write */ |
4575 | ||
b15c2e57 | 4576 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4577 | if (err) |
4578 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4579 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4580 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4581 | " but 'external_size' not in effect?\n", __func__); | |
4582 | printk(KERN_ERR | |
4583 | "md: invalid array_size %llu > default size %llu\n", | |
4584 | (unsigned long long)mddev->array_sectors / 2, | |
4585 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4586 | err = -EINVAL; | |
4587 | mddev->pers->stop(mddev); | |
4588 | } | |
4589 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4590 | err = bitmap_create(mddev); |
4591 | if (err) { | |
4592 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4593 | mdname(mddev), err); | |
4594 | mddev->pers->stop(mddev); | |
4595 | } | |
4596 | } | |
1da177e4 | 4597 | if (err) { |
1da177e4 LT |
4598 | module_put(mddev->pers->owner); |
4599 | mddev->pers = NULL; | |
32a7627c N |
4600 | bitmap_destroy(mddev); |
4601 | return err; | |
1da177e4 | 4602 | } |
5e55e2f5 | 4603 | if (mddev->pers->sync_request) { |
00bcb4ac N |
4604 | if (mddev->kobj.sd && |
4605 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
4606 | printk(KERN_WARNING |
4607 | "md: cannot register extra attributes for %s\n", | |
4608 | mdname(mddev)); | |
00bcb4ac | 4609 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4610 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4611 | mddev->ro = 0; |
4612 | ||
1da177e4 | 4613 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
4614 | atomic_set(&mddev->max_corr_read_errors, |
4615 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
4616 | mddev->safemode = 0; |
4617 | mddev->safemode_timer.function = md_safemode_timeout; | |
4618 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4619 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4620 | mddev->in_sync = 1; |
0ca69886 N |
4621 | smp_wmb(); |
4622 | mddev->ready = 1; | |
159ec1fc | 4623 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4624 | if (rdev->raid_disk >= 0) { |
4625 | char nm[20]; | |
4626 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 | 4627 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
00bcb4ac | 4628 | /* failure here is OK */; |
86e6ffdd | 4629 | } |
1da177e4 LT |
4630 | |
4631 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4632 | ||
850b2b42 N |
4633 | if (mddev->flags) |
4634 | md_update_sb(mddev, 0); | |
1da177e4 | 4635 | |
d7603b7e | 4636 | md_new_event(mddev); |
00bcb4ac N |
4637 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
4638 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 4639 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
4640 | return 0; |
4641 | } | |
390ee602 | 4642 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 4643 | |
fe60b014 N |
4644 | static int do_md_run(mddev_t *mddev) |
4645 | { | |
4646 | int err; | |
4647 | ||
4648 | err = md_run(mddev); | |
4649 | if (err) | |
4650 | goto out; | |
69e51b44 N |
4651 | err = bitmap_load(mddev); |
4652 | if (err) { | |
4653 | bitmap_destroy(mddev); | |
4654 | goto out; | |
4655 | } | |
0fd018af JB |
4656 | |
4657 | md_wakeup_thread(mddev->thread); | |
4658 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
4659 | ||
fe60b014 N |
4660 | set_capacity(mddev->gendisk, mddev->array_sectors); |
4661 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 4662 | mddev->changed = 1; |
fe60b014 N |
4663 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
4664 | out: | |
4665 | return err; | |
4666 | } | |
4667 | ||
1da177e4 LT |
4668 | static int restart_array(mddev_t *mddev) |
4669 | { | |
4670 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4671 | |
80fab1d7 | 4672 | /* Complain if it has no devices */ |
1da177e4 | 4673 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4674 | return -ENXIO; |
4675 | if (!mddev->pers) | |
4676 | return -EINVAL; | |
4677 | if (!mddev->ro) | |
4678 | return -EBUSY; | |
4679 | mddev->safemode = 0; | |
4680 | mddev->ro = 0; | |
4681 | set_disk_ro(disk, 0); | |
4682 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4683 | mdname(mddev)); | |
4684 | /* Kick recovery or resync if necessary */ | |
4685 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4686 | md_wakeup_thread(mddev->thread); | |
4687 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 4688 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 4689 | return 0; |
1da177e4 LT |
4690 | } |
4691 | ||
acc55e22 N |
4692 | /* similar to deny_write_access, but accounts for our holding a reference |
4693 | * to the file ourselves */ | |
4694 | static int deny_bitmap_write_access(struct file * file) | |
4695 | { | |
4696 | struct inode *inode = file->f_mapping->host; | |
4697 | ||
4698 | spin_lock(&inode->i_lock); | |
4699 | if (atomic_read(&inode->i_writecount) > 1) { | |
4700 | spin_unlock(&inode->i_lock); | |
4701 | return -ETXTBSY; | |
4702 | } | |
4703 | atomic_set(&inode->i_writecount, -1); | |
4704 | spin_unlock(&inode->i_lock); | |
4705 | ||
4706 | return 0; | |
4707 | } | |
4708 | ||
43a70507 | 4709 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
4710 | { |
4711 | struct inode *inode = file->f_mapping->host; | |
4712 | ||
4713 | spin_lock(&inode->i_lock); | |
4714 | atomic_set(&inode->i_writecount, 1); | |
4715 | spin_unlock(&inode->i_lock); | |
4716 | } | |
4717 | ||
6177b472 N |
4718 | static void md_clean(mddev_t *mddev) |
4719 | { | |
4720 | mddev->array_sectors = 0; | |
4721 | mddev->external_size = 0; | |
4722 | mddev->dev_sectors = 0; | |
4723 | mddev->raid_disks = 0; | |
4724 | mddev->recovery_cp = 0; | |
4725 | mddev->resync_min = 0; | |
4726 | mddev->resync_max = MaxSector; | |
4727 | mddev->reshape_position = MaxSector; | |
4728 | mddev->external = 0; | |
4729 | mddev->persistent = 0; | |
4730 | mddev->level = LEVEL_NONE; | |
4731 | mddev->clevel[0] = 0; | |
4732 | mddev->flags = 0; | |
4733 | mddev->ro = 0; | |
4734 | mddev->metadata_type[0] = 0; | |
4735 | mddev->chunk_sectors = 0; | |
4736 | mddev->ctime = mddev->utime = 0; | |
4737 | mddev->layout = 0; | |
4738 | mddev->max_disks = 0; | |
4739 | mddev->events = 0; | |
a8707c08 | 4740 | mddev->can_decrease_events = 0; |
6177b472 N |
4741 | mddev->delta_disks = 0; |
4742 | mddev->new_level = LEVEL_NONE; | |
4743 | mddev->new_layout = 0; | |
4744 | mddev->new_chunk_sectors = 0; | |
4745 | mddev->curr_resync = 0; | |
4746 | mddev->resync_mismatches = 0; | |
4747 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4748 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4749 | mddev->recovery = 0; | |
4750 | mddev->in_sync = 0; | |
f0b4f7e2 | 4751 | mddev->changed = 0; |
6177b472 | 4752 | mddev->degraded = 0; |
6177b472 N |
4753 | mddev->safemode = 0; |
4754 | mddev->bitmap_info.offset = 0; | |
4755 | mddev->bitmap_info.default_offset = 0; | |
4756 | mddev->bitmap_info.chunksize = 0; | |
4757 | mddev->bitmap_info.daemon_sleep = 0; | |
4758 | mddev->bitmap_info.max_write_behind = 0; | |
4759 | } | |
4760 | ||
defad61a | 4761 | static void __md_stop_writes(mddev_t *mddev) |
a047e125 N |
4762 | { |
4763 | if (mddev->sync_thread) { | |
4764 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4765 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 4766 | reap_sync_thread(mddev); |
a047e125 N |
4767 | } |
4768 | ||
4769 | del_timer_sync(&mddev->safemode_timer); | |
4770 | ||
4771 | bitmap_flush(mddev); | |
4772 | md_super_wait(mddev); | |
4773 | ||
4774 | if (!mddev->in_sync || mddev->flags) { | |
4775 | /* mark array as shutdown cleanly */ | |
4776 | mddev->in_sync = 1; | |
4777 | md_update_sb(mddev, 1); | |
4778 | } | |
4779 | } | |
defad61a N |
4780 | |
4781 | void md_stop_writes(mddev_t *mddev) | |
4782 | { | |
4783 | mddev_lock(mddev); | |
4784 | __md_stop_writes(mddev); | |
4785 | mddev_unlock(mddev); | |
4786 | } | |
390ee602 | 4787 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 4788 | |
390ee602 | 4789 | void md_stop(mddev_t *mddev) |
6177b472 | 4790 | { |
0ca69886 | 4791 | mddev->ready = 0; |
6177b472 N |
4792 | mddev->pers->stop(mddev); |
4793 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
4794 | mddev->to_remove = &md_redundancy_group; | |
4795 | module_put(mddev->pers->owner); | |
4796 | mddev->pers = NULL; | |
cca9cf90 | 4797 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 4798 | } |
390ee602 | 4799 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 4800 | |
a4bd82d0 N |
4801 | static int md_set_readonly(mddev_t *mddev, int is_open) |
4802 | { | |
4803 | int err = 0; | |
4804 | mutex_lock(&mddev->open_mutex); | |
4805 | if (atomic_read(&mddev->openers) > is_open) { | |
4806 | printk("md: %s still in use.\n",mdname(mddev)); | |
4807 | err = -EBUSY; | |
4808 | goto out; | |
4809 | } | |
4810 | if (mddev->pers) { | |
defad61a | 4811 | __md_stop_writes(mddev); |
a4bd82d0 N |
4812 | |
4813 | err = -ENXIO; | |
4814 | if (mddev->ro==1) | |
4815 | goto out; | |
4816 | mddev->ro = 1; | |
4817 | set_disk_ro(mddev->gendisk, 1); | |
4818 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 4819 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
a4bd82d0 N |
4820 | err = 0; |
4821 | } | |
4822 | out: | |
4823 | mutex_unlock(&mddev->open_mutex); | |
4824 | return err; | |
4825 | } | |
4826 | ||
9e653b63 N |
4827 | /* mode: |
4828 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
4829 | * 2 - stop but do not disassemble array |
4830 | */ | |
df5b20cf | 4831 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 | 4832 | { |
1da177e4 | 4833 | struct gendisk *disk = mddev->gendisk; |
c4647292 | 4834 | mdk_rdev_t *rdev; |
1da177e4 | 4835 | |
c8c00a69 | 4836 | mutex_lock(&mddev->open_mutex); |
bb4f1e9d N |
4837 | if (atomic_read(&mddev->openers) > is_open || |
4838 | mddev->sysfs_active) { | |
df5b20cf | 4839 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 N |
4840 | mutex_unlock(&mddev->open_mutex); |
4841 | return -EBUSY; | |
4842 | } | |
1da177e4 | 4843 | |
6e17b027 | 4844 | if (mddev->pers) { |
a4bd82d0 N |
4845 | if (mddev->ro) |
4846 | set_disk_ro(disk, 0); | |
409c57f3 | 4847 | |
defad61a | 4848 | __md_stop_writes(mddev); |
a4bd82d0 N |
4849 | md_stop(mddev); |
4850 | mddev->queue->merge_bvec_fn = NULL; | |
a4bd82d0 | 4851 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 4852 | |
a4bd82d0 | 4853 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 4854 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 4855 | |
a4bd82d0 N |
4856 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4857 | if (rdev->raid_disk >= 0) { | |
4858 | char nm[20]; | |
4859 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4860 | sysfs_remove_link(&mddev->kobj, nm); | |
4861 | } | |
c4647292 | 4862 | |
a4bd82d0 | 4863 | set_capacity(disk, 0); |
6e17b027 | 4864 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 4865 | mddev->changed = 1; |
a4bd82d0 | 4866 | revalidate_disk(disk); |
0d4ca600 | 4867 | |
a4bd82d0 N |
4868 | if (mddev->ro) |
4869 | mddev->ro = 0; | |
6e17b027 N |
4870 | } else |
4871 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
4872 | /* |
4873 | * Free resources if final stop | |
4874 | */ | |
9e653b63 | 4875 | if (mode == 0) { |
1da177e4 LT |
4876 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4877 | ||
978f946b | 4878 | bitmap_destroy(mddev); |
c3d9714e N |
4879 | if (mddev->bitmap_info.file) { |
4880 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
4881 | fput(mddev->bitmap_info.file); | |
4882 | mddev->bitmap_info.file = NULL; | |
978f946b | 4883 | } |
c3d9714e | 4884 | mddev->bitmap_info.offset = 0; |
978f946b | 4885 | |
1da177e4 LT |
4886 | export_array(mddev); |
4887 | ||
6177b472 | 4888 | md_clean(mddev); |
934d9c23 | 4889 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4890 | if (mddev->hold_active == UNTIL_STOP) |
4891 | mddev->hold_active = 0; | |
a4bd82d0 | 4892 | } |
3f9d99c1 | 4893 | blk_integrity_unregister(disk); |
d7603b7e | 4894 | md_new_event(mddev); |
00bcb4ac | 4895 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 4896 | return 0; |
1da177e4 LT |
4897 | } |
4898 | ||
fdee8ae4 | 4899 | #ifndef MODULE |
1da177e4 LT |
4900 | static void autorun_array(mddev_t *mddev) |
4901 | { | |
4902 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4903 | int err; |
4904 | ||
a757e64c | 4905 | if (list_empty(&mddev->disks)) |
1da177e4 | 4906 | return; |
1da177e4 LT |
4907 | |
4908 | printk(KERN_INFO "md: running: "); | |
4909 | ||
159ec1fc | 4910 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4911 | char b[BDEVNAME_SIZE]; |
4912 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4913 | } | |
4914 | printk("\n"); | |
4915 | ||
d710e138 | 4916 | err = do_md_run(mddev); |
1da177e4 LT |
4917 | if (err) { |
4918 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4919 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4920 | } |
4921 | } | |
4922 | ||
4923 | /* | |
4924 | * lets try to run arrays based on all disks that have arrived | |
4925 | * until now. (those are in pending_raid_disks) | |
4926 | * | |
4927 | * the method: pick the first pending disk, collect all disks with | |
4928 | * the same UUID, remove all from the pending list and put them into | |
4929 | * the 'same_array' list. Then order this list based on superblock | |
4930 | * update time (freshest comes first), kick out 'old' disks and | |
4931 | * compare superblocks. If everything's fine then run it. | |
4932 | * | |
4933 | * If "unit" is allocated, then bump its reference count | |
4934 | */ | |
4935 | static void autorun_devices(int part) | |
4936 | { | |
159ec1fc | 4937 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4938 | mddev_t *mddev; |
4939 | char b[BDEVNAME_SIZE]; | |
4940 | ||
4941 | printk(KERN_INFO "md: autorun ...\n"); | |
4942 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4943 | int unit; |
1da177e4 | 4944 | dev_t dev; |
ad01c9e3 | 4945 | LIST_HEAD(candidates); |
1da177e4 LT |
4946 | rdev0 = list_entry(pending_raid_disks.next, |
4947 | mdk_rdev_t, same_set); | |
4948 | ||
4949 | printk(KERN_INFO "md: considering %s ...\n", | |
4950 | bdevname(rdev0->bdev,b)); | |
4951 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4952 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4953 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4954 | printk(KERN_INFO "md: adding %s ...\n", | |
4955 | bdevname(rdev->bdev,b)); | |
4956 | list_move(&rdev->same_set, &candidates); | |
4957 | } | |
4958 | /* | |
4959 | * now we have a set of devices, with all of them having | |
4960 | * mostly sane superblocks. It's time to allocate the | |
4961 | * mddev. | |
4962 | */ | |
e8703fe1 N |
4963 | if (part) { |
4964 | dev = MKDEV(mdp_major, | |
4965 | rdev0->preferred_minor << MdpMinorShift); | |
4966 | unit = MINOR(dev) >> MdpMinorShift; | |
4967 | } else { | |
4968 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4969 | unit = MINOR(dev); | |
4970 | } | |
4971 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4972 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4973 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4974 | break; | |
4975 | } | |
1da177e4 LT |
4976 | |
4977 | md_probe(dev, NULL, NULL); | |
4978 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4979 | if (!mddev || !mddev->gendisk) { |
4980 | if (mddev) | |
4981 | mddev_put(mddev); | |
4982 | printk(KERN_ERR | |
1da177e4 LT |
4983 | "md: cannot allocate memory for md drive.\n"); |
4984 | break; | |
4985 | } | |
4986 | if (mddev_lock(mddev)) | |
4987 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4988 | mdname(mddev)); | |
4989 | else if (mddev->raid_disks || mddev->major_version | |
4990 | || !list_empty(&mddev->disks)) { | |
4991 | printk(KERN_WARNING | |
4992 | "md: %s already running, cannot run %s\n", | |
4993 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4994 | mddev_unlock(mddev); | |
4995 | } else { | |
4996 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4997 | mddev->persistent = 1; |
159ec1fc | 4998 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4999 | list_del_init(&rdev->same_set); |
5000 | if (bind_rdev_to_array(rdev, mddev)) | |
5001 | export_rdev(rdev); | |
5002 | } | |
5003 | autorun_array(mddev); | |
5004 | mddev_unlock(mddev); | |
5005 | } | |
5006 | /* on success, candidates will be empty, on error | |
5007 | * it won't... | |
5008 | */ | |
159ec1fc | 5009 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5010 | list_del_init(&rdev->same_set); |
1da177e4 | 5011 | export_rdev(rdev); |
4b80991c | 5012 | } |
1da177e4 LT |
5013 | mddev_put(mddev); |
5014 | } | |
5015 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5016 | } | |
fdee8ae4 | 5017 | #endif /* !MODULE */ |
1da177e4 | 5018 | |
1da177e4 LT |
5019 | static int get_version(void __user * arg) |
5020 | { | |
5021 | mdu_version_t ver; | |
5022 | ||
5023 | ver.major = MD_MAJOR_VERSION; | |
5024 | ver.minor = MD_MINOR_VERSION; | |
5025 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5026 | ||
5027 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5028 | return -EFAULT; | |
5029 | ||
5030 | return 0; | |
5031 | } | |
5032 | ||
5033 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
5034 | { | |
5035 | mdu_array_info_t info; | |
a9f326eb | 5036 | int nr,working,insync,failed,spare; |
1da177e4 | 5037 | mdk_rdev_t *rdev; |
1da177e4 | 5038 | |
a9f326eb | 5039 | nr=working=insync=failed=spare=0; |
159ec1fc | 5040 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 5041 | nr++; |
b2d444d7 | 5042 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5043 | failed++; |
5044 | else { | |
5045 | working++; | |
b2d444d7 | 5046 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 5047 | insync++; |
1da177e4 LT |
5048 | else |
5049 | spare++; | |
5050 | } | |
5051 | } | |
5052 | ||
5053 | info.major_version = mddev->major_version; | |
5054 | info.minor_version = mddev->minor_version; | |
5055 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5056 | info.ctime = mddev->ctime; | |
5057 | info.level = mddev->level; | |
58c0fed4 AN |
5058 | info.size = mddev->dev_sectors / 2; |
5059 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5060 | info.size = -1; |
1da177e4 LT |
5061 | info.nr_disks = nr; |
5062 | info.raid_disks = mddev->raid_disks; | |
5063 | info.md_minor = mddev->md_minor; | |
5064 | info.not_persistent= !mddev->persistent; | |
5065 | ||
5066 | info.utime = mddev->utime; | |
5067 | info.state = 0; | |
5068 | if (mddev->in_sync) | |
5069 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5070 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5071 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5072 | info.active_disks = insync; |
1da177e4 LT |
5073 | info.working_disks = working; |
5074 | info.failed_disks = failed; | |
5075 | info.spare_disks = spare; | |
5076 | ||
5077 | info.layout = mddev->layout; | |
9d8f0363 | 5078 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5079 | |
5080 | if (copy_to_user(arg, &info, sizeof(info))) | |
5081 | return -EFAULT; | |
5082 | ||
5083 | return 0; | |
5084 | } | |
5085 | ||
87162a28 | 5086 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
5087 | { |
5088 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5089 | char *ptr, *buf = NULL; | |
5090 | int err = -ENOMEM; | |
5091 | ||
b5470dc5 DW |
5092 | if (md_allow_write(mddev)) |
5093 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
5094 | else | |
5095 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 5096 | |
32a7627c N |
5097 | if (!file) |
5098 | goto out; | |
5099 | ||
5100 | /* bitmap disabled, zero the first byte and copy out */ | |
5101 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
5102 | file->pathname[0] = '\0'; | |
5103 | goto copy_out; | |
5104 | } | |
5105 | ||
5106 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5107 | if (!buf) | |
5108 | goto out; | |
5109 | ||
6bcfd601 CH |
5110 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
5111 | if (IS_ERR(ptr)) | |
32a7627c N |
5112 | goto out; |
5113 | ||
5114 | strcpy(file->pathname, ptr); | |
5115 | ||
5116 | copy_out: | |
5117 | err = 0; | |
5118 | if (copy_to_user(arg, file, sizeof(*file))) | |
5119 | err = -EFAULT; | |
5120 | out: | |
5121 | kfree(buf); | |
5122 | kfree(file); | |
5123 | return err; | |
5124 | } | |
5125 | ||
1da177e4 LT |
5126 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
5127 | { | |
5128 | mdu_disk_info_t info; | |
1da177e4 LT |
5129 | mdk_rdev_t *rdev; |
5130 | ||
5131 | if (copy_from_user(&info, arg, sizeof(info))) | |
5132 | return -EFAULT; | |
5133 | ||
26ef379f | 5134 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
5135 | if (rdev) { |
5136 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5137 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5138 | info.raid_disk = rdev->raid_disk; | |
5139 | info.state = 0; | |
b2d444d7 | 5140 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5141 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5142 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5143 | info.state |= (1<<MD_DISK_ACTIVE); |
5144 | info.state |= (1<<MD_DISK_SYNC); | |
5145 | } | |
8ddf9efe N |
5146 | if (test_bit(WriteMostly, &rdev->flags)) |
5147 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5148 | } else { |
5149 | info.major = info.minor = 0; | |
5150 | info.raid_disk = -1; | |
5151 | info.state = (1<<MD_DISK_REMOVED); | |
5152 | } | |
5153 | ||
5154 | if (copy_to_user(arg, &info, sizeof(info))) | |
5155 | return -EFAULT; | |
5156 | ||
5157 | return 0; | |
5158 | } | |
5159 | ||
5160 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
5161 | { | |
5162 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
5163 | mdk_rdev_t *rdev; | |
5164 | dev_t dev = MKDEV(info->major,info->minor); | |
5165 | ||
5166 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5167 | return -EOVERFLOW; | |
5168 | ||
5169 | if (!mddev->raid_disks) { | |
5170 | int err; | |
5171 | /* expecting a device which has a superblock */ | |
5172 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5173 | if (IS_ERR(rdev)) { | |
5174 | printk(KERN_WARNING | |
5175 | "md: md_import_device returned %ld\n", | |
5176 | PTR_ERR(rdev)); | |
5177 | return PTR_ERR(rdev); | |
5178 | } | |
5179 | if (!list_empty(&mddev->disks)) { | |
5180 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
5181 | mdk_rdev_t, same_set); | |
a9f326eb | 5182 | err = super_types[mddev->major_version] |
1da177e4 LT |
5183 | .load_super(rdev, rdev0, mddev->minor_version); |
5184 | if (err < 0) { | |
5185 | printk(KERN_WARNING | |
5186 | "md: %s has different UUID to %s\n", | |
5187 | bdevname(rdev->bdev,b), | |
5188 | bdevname(rdev0->bdev,b2)); | |
5189 | export_rdev(rdev); | |
5190 | return -EINVAL; | |
5191 | } | |
5192 | } | |
5193 | err = bind_rdev_to_array(rdev, mddev); | |
5194 | if (err) | |
5195 | export_rdev(rdev); | |
5196 | return err; | |
5197 | } | |
5198 | ||
5199 | /* | |
5200 | * add_new_disk can be used once the array is assembled | |
5201 | * to add "hot spares". They must already have a superblock | |
5202 | * written | |
5203 | */ | |
5204 | if (mddev->pers) { | |
5205 | int err; | |
5206 | if (!mddev->pers->hot_add_disk) { | |
5207 | printk(KERN_WARNING | |
5208 | "%s: personality does not support diskops!\n", | |
5209 | mdname(mddev)); | |
5210 | return -EINVAL; | |
5211 | } | |
7b1e35f6 N |
5212 | if (mddev->persistent) |
5213 | rdev = md_import_device(dev, mddev->major_version, | |
5214 | mddev->minor_version); | |
5215 | else | |
5216 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5217 | if (IS_ERR(rdev)) { |
5218 | printk(KERN_WARNING | |
5219 | "md: md_import_device returned %ld\n", | |
5220 | PTR_ERR(rdev)); | |
5221 | return PTR_ERR(rdev); | |
5222 | } | |
1a855a06 | 5223 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5224 | if (!mddev->persistent) { |
5225 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5226 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5227 | rdev->raid_disk = info->raid_disk; |
bf572541 N |
5228 | set_bit(In_sync, &rdev->flags); |
5229 | } else | |
41158c7e N |
5230 | rdev->raid_disk = -1; |
5231 | } else | |
5232 | super_types[mddev->major_version]. | |
5233 | validate_super(mddev, rdev); | |
bedd86b7 N |
5234 | if ((info->state & (1<<MD_DISK_SYNC)) && |
5235 | (!test_bit(In_sync, &rdev->flags) || | |
5236 | rdev->raid_disk != info->raid_disk)) { | |
5237 | /* This was a hot-add request, but events doesn't | |
5238 | * match, so reject it. | |
5239 | */ | |
5240 | export_rdev(rdev); | |
5241 | return -EINVAL; | |
5242 | } | |
5243 | ||
1a855a06 N |
5244 | if (test_bit(In_sync, &rdev->flags)) |
5245 | rdev->saved_raid_disk = rdev->raid_disk; | |
5246 | else | |
5247 | rdev->saved_raid_disk = -1; | |
41158c7e | 5248 | |
b2d444d7 | 5249 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5250 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5251 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5252 | else |
5253 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5254 | |
1da177e4 LT |
5255 | rdev->raid_disk = -1; |
5256 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5257 | if (!err && !mddev->pers->hot_remove_disk) { |
5258 | /* If there is hot_add_disk but no hot_remove_disk | |
5259 | * then added disks for geometry changes, | |
5260 | * and should be added immediately. | |
5261 | */ | |
5262 | super_types[mddev->major_version]. | |
5263 | validate_super(mddev, rdev); | |
5264 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5265 | if (err) | |
5266 | unbind_rdev_from_array(rdev); | |
5267 | } | |
1da177e4 LT |
5268 | if (err) |
5269 | export_rdev(rdev); | |
52664732 | 5270 | else |
00bcb4ac | 5271 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5272 | |
17571284 | 5273 | md_update_sb(mddev, 1); |
72a23c21 NB |
5274 | if (mddev->degraded) |
5275 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5276 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
9864c005 | 5277 | if (!err) |
5278 | md_new_event(mddev); | |
005eca5e | 5279 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5280 | return err; |
5281 | } | |
5282 | ||
5283 | /* otherwise, add_new_disk is only allowed | |
5284 | * for major_version==0 superblocks | |
5285 | */ | |
5286 | if (mddev->major_version != 0) { | |
5287 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5288 | mdname(mddev)); | |
5289 | return -EINVAL; | |
5290 | } | |
5291 | ||
5292 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5293 | int err; | |
d710e138 | 5294 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5295 | if (IS_ERR(rdev)) { |
5296 | printk(KERN_WARNING | |
5297 | "md: error, md_import_device() returned %ld\n", | |
5298 | PTR_ERR(rdev)); | |
5299 | return PTR_ERR(rdev); | |
5300 | } | |
5301 | rdev->desc_nr = info->number; | |
5302 | if (info->raid_disk < mddev->raid_disks) | |
5303 | rdev->raid_disk = info->raid_disk; | |
5304 | else | |
5305 | rdev->raid_disk = -1; | |
5306 | ||
1da177e4 | 5307 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5308 | if (info->state & (1<<MD_DISK_SYNC)) |
5309 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5310 | |
8ddf9efe N |
5311 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5312 | set_bit(WriteMostly, &rdev->flags); | |
5313 | ||
1da177e4 LT |
5314 | if (!mddev->persistent) { |
5315 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5316 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5317 | } else | |
57b2caa3 | 5318 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5319 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5320 | |
2bf071bf N |
5321 | err = bind_rdev_to_array(rdev, mddev); |
5322 | if (err) { | |
5323 | export_rdev(rdev); | |
5324 | return err; | |
5325 | } | |
1da177e4 LT |
5326 | } |
5327 | ||
5328 | return 0; | |
5329 | } | |
5330 | ||
5331 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
5332 | { | |
5333 | char b[BDEVNAME_SIZE]; | |
5334 | mdk_rdev_t *rdev; | |
5335 | ||
1da177e4 LT |
5336 | rdev = find_rdev(mddev, dev); |
5337 | if (!rdev) | |
5338 | return -ENXIO; | |
5339 | ||
5340 | if (rdev->raid_disk >= 0) | |
5341 | goto busy; | |
5342 | ||
5343 | kick_rdev_from_array(rdev); | |
850b2b42 | 5344 | md_update_sb(mddev, 1); |
d7603b7e | 5345 | md_new_event(mddev); |
1da177e4 LT |
5346 | |
5347 | return 0; | |
5348 | busy: | |
fdefa4d8 | 5349 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5350 | bdevname(rdev->bdev,b), mdname(mddev)); |
5351 | return -EBUSY; | |
5352 | } | |
5353 | ||
5354 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
5355 | { | |
5356 | char b[BDEVNAME_SIZE]; | |
5357 | int err; | |
1da177e4 LT |
5358 | mdk_rdev_t *rdev; |
5359 | ||
5360 | if (!mddev->pers) | |
5361 | return -ENODEV; | |
5362 | ||
5363 | if (mddev->major_version != 0) { | |
5364 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5365 | " version-0 superblocks.\n", | |
5366 | mdname(mddev)); | |
5367 | return -EINVAL; | |
5368 | } | |
5369 | if (!mddev->pers->hot_add_disk) { | |
5370 | printk(KERN_WARNING | |
5371 | "%s: personality does not support diskops!\n", | |
5372 | mdname(mddev)); | |
5373 | return -EINVAL; | |
5374 | } | |
5375 | ||
d710e138 | 5376 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5377 | if (IS_ERR(rdev)) { |
5378 | printk(KERN_WARNING | |
5379 | "md: error, md_import_device() returned %ld\n", | |
5380 | PTR_ERR(rdev)); | |
5381 | return -EINVAL; | |
5382 | } | |
5383 | ||
5384 | if (mddev->persistent) | |
57b2caa3 | 5385 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5386 | else |
77304d2a | 5387 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5388 | |
8190e754 | 5389 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5390 | |
b2d444d7 | 5391 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5392 | printk(KERN_WARNING |
5393 | "md: can not hot-add faulty %s disk to %s!\n", | |
5394 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5395 | err = -EINVAL; | |
5396 | goto abort_export; | |
5397 | } | |
b2d444d7 | 5398 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5399 | rdev->desc_nr = -1; |
5842730d | 5400 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5401 | err = bind_rdev_to_array(rdev, mddev); |
5402 | if (err) | |
5403 | goto abort_export; | |
1da177e4 LT |
5404 | |
5405 | /* | |
5406 | * The rest should better be atomic, we can have disk failures | |
5407 | * noticed in interrupt contexts ... | |
5408 | */ | |
5409 | ||
1da177e4 LT |
5410 | rdev->raid_disk = -1; |
5411 | ||
850b2b42 | 5412 | md_update_sb(mddev, 1); |
1da177e4 LT |
5413 | |
5414 | /* | |
5415 | * Kick recovery, maybe this spare has to be added to the | |
5416 | * array immediately. | |
5417 | */ | |
5418 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5419 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5420 | md_new_event(mddev); |
1da177e4 LT |
5421 | return 0; |
5422 | ||
1da177e4 LT |
5423 | abort_export: |
5424 | export_rdev(rdev); | |
5425 | return err; | |
5426 | } | |
5427 | ||
32a7627c N |
5428 | static int set_bitmap_file(mddev_t *mddev, int fd) |
5429 | { | |
5430 | int err; | |
5431 | ||
36fa3063 N |
5432 | if (mddev->pers) { |
5433 | if (!mddev->pers->quiesce) | |
5434 | return -EBUSY; | |
5435 | if (mddev->recovery || mddev->sync_thread) | |
5436 | return -EBUSY; | |
5437 | /* we should be able to change the bitmap.. */ | |
5438 | } | |
32a7627c | 5439 | |
32a7627c | 5440 | |
36fa3063 N |
5441 | if (fd >= 0) { |
5442 | if (mddev->bitmap) | |
5443 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5444 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5445 | |
c3d9714e | 5446 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5447 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5448 | mdname(mddev)); | |
5449 | return -EBADF; | |
5450 | } | |
5451 | ||
c3d9714e | 5452 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
5453 | if (err) { |
5454 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5455 | mdname(mddev)); | |
c3d9714e N |
5456 | fput(mddev->bitmap_info.file); |
5457 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5458 | return err; |
5459 | } | |
c3d9714e | 5460 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5461 | } else if (mddev->bitmap == NULL) |
5462 | return -ENOENT; /* cannot remove what isn't there */ | |
5463 | err = 0; | |
5464 | if (mddev->pers) { | |
5465 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 5466 | if (fd >= 0) { |
36fa3063 | 5467 | err = bitmap_create(mddev); |
69e51b44 N |
5468 | if (!err) |
5469 | err = bitmap_load(mddev); | |
5470 | } | |
d7375ab3 | 5471 | if (fd < 0 || err) { |
36fa3063 | 5472 | bitmap_destroy(mddev); |
d7375ab3 N |
5473 | fd = -1; /* make sure to put the file */ |
5474 | } | |
36fa3063 | 5475 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5476 | } |
5477 | if (fd < 0) { | |
c3d9714e N |
5478 | if (mddev->bitmap_info.file) { |
5479 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5480 | fput(mddev->bitmap_info.file); | |
acc55e22 | 5481 | } |
c3d9714e | 5482 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
5483 | } |
5484 | ||
32a7627c N |
5485 | return err; |
5486 | } | |
5487 | ||
1da177e4 LT |
5488 | /* |
5489 | * set_array_info is used two different ways | |
5490 | * The original usage is when creating a new array. | |
5491 | * In this usage, raid_disks is > 0 and it together with | |
5492 | * level, size, not_persistent,layout,chunksize determine the | |
5493 | * shape of the array. | |
5494 | * This will always create an array with a type-0.90.0 superblock. | |
5495 | * The newer usage is when assembling an array. | |
5496 | * In this case raid_disks will be 0, and the major_version field is | |
5497 | * use to determine which style super-blocks are to be found on the devices. | |
5498 | * The minor and patch _version numbers are also kept incase the | |
5499 | * super_block handler wishes to interpret them. | |
5500 | */ | |
5501 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5502 | { | |
5503 | ||
5504 | if (info->raid_disks == 0) { | |
5505 | /* just setting version number for superblock loading */ | |
5506 | if (info->major_version < 0 || | |
50511da3 | 5507 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5508 | super_types[info->major_version].name == NULL) { |
5509 | /* maybe try to auto-load a module? */ | |
5510 | printk(KERN_INFO | |
5511 | "md: superblock version %d not known\n", | |
5512 | info->major_version); | |
5513 | return -EINVAL; | |
5514 | } | |
5515 | mddev->major_version = info->major_version; | |
5516 | mddev->minor_version = info->minor_version; | |
5517 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5518 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
5519 | /* ensure mddev_put doesn't delete this now that there |
5520 | * is some minimal configuration. | |
5521 | */ | |
5522 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
5523 | return 0; |
5524 | } | |
5525 | mddev->major_version = MD_MAJOR_VERSION; | |
5526 | mddev->minor_version = MD_MINOR_VERSION; | |
5527 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5528 | mddev->ctime = get_seconds(); | |
5529 | ||
5530 | mddev->level = info->level; | |
17115e03 | 5531 | mddev->clevel[0] = 0; |
58c0fed4 | 5532 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5533 | mddev->raid_disks = info->raid_disks; |
5534 | /* don't set md_minor, it is determined by which /dev/md* was | |
5535 | * openned | |
5536 | */ | |
5537 | if (info->state & (1<<MD_SB_CLEAN)) | |
5538 | mddev->recovery_cp = MaxSector; | |
5539 | else | |
5540 | mddev->recovery_cp = 0; | |
5541 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5542 | mddev->external = 0; |
1da177e4 LT |
5543 | |
5544 | mddev->layout = info->layout; | |
9d8f0363 | 5545 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5546 | |
5547 | mddev->max_disks = MD_SB_DISKS; | |
5548 | ||
e691063a N |
5549 | if (mddev->persistent) |
5550 | mddev->flags = 0; | |
850b2b42 | 5551 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5552 | |
c3d9714e N |
5553 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
5554 | mddev->bitmap_info.offset = 0; | |
b2a2703c | 5555 | |
f6705578 N |
5556 | mddev->reshape_position = MaxSector; |
5557 | ||
1da177e4 LT |
5558 | /* |
5559 | * Generate a 128 bit UUID | |
5560 | */ | |
5561 | get_random_bytes(mddev->uuid, 16); | |
5562 | ||
f6705578 | 5563 | mddev->new_level = mddev->level; |
664e7c41 | 5564 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5565 | mddev->new_layout = mddev->layout; |
5566 | mddev->delta_disks = 0; | |
5567 | ||
1da177e4 LT |
5568 | return 0; |
5569 | } | |
5570 | ||
1f403624 DW |
5571 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5572 | { | |
b522adcd DW |
5573 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5574 | ||
5575 | if (mddev->external_size) | |
5576 | return; | |
5577 | ||
1f403624 DW |
5578 | mddev->array_sectors = array_sectors; |
5579 | } | |
5580 | EXPORT_SYMBOL(md_set_array_sectors); | |
5581 | ||
d71f9f88 | 5582 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5583 | { |
159ec1fc | 5584 | mdk_rdev_t *rdev; |
a35b0d69 | 5585 | int rv; |
d71f9f88 | 5586 | int fit = (num_sectors == 0); |
a35b0d69 N |
5587 | |
5588 | if (mddev->pers->resize == NULL) | |
5589 | return -EINVAL; | |
d71f9f88 AN |
5590 | /* The "num_sectors" is the number of sectors of each device that |
5591 | * is used. This can only make sense for arrays with redundancy. | |
5592 | * linear and raid0 always use whatever space is available. We can only | |
5593 | * consider changing this number if no resync or reconstruction is | |
5594 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5595 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5596 | * of each device. If num_sectors is zero, we find the largest size |
5597 | * that fits. | |
a35b0d69 N |
5598 | */ |
5599 | if (mddev->sync_thread) | |
5600 | return -EBUSY; | |
dba034ee N |
5601 | if (mddev->bitmap) |
5602 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5603 | * grow, and re-add. | |
5604 | */ | |
5605 | return -EBUSY; | |
159ec1fc | 5606 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5607 | sector_t avail = rdev->sectors; |
01ab5662 | 5608 | |
d71f9f88 AN |
5609 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5610 | num_sectors = avail; | |
5611 | if (avail < num_sectors) | |
a35b0d69 N |
5612 | return -ENOSPC; |
5613 | } | |
d71f9f88 | 5614 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5615 | if (!rv) |
5616 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5617 | return rv; |
5618 | } | |
5619 | ||
da943b99 N |
5620 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5621 | { | |
5622 | int rv; | |
5623 | /* change the number of raid disks */ | |
63c70c4f | 5624 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5625 | return -EINVAL; |
5626 | if (raid_disks <= 0 || | |
233fca36 | 5627 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 5628 | return -EINVAL; |
63c70c4f | 5629 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5630 | return -EBUSY; |
63c70c4f N |
5631 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5632 | ||
5633 | rv = mddev->pers->check_reshape(mddev); | |
de171cb9 N |
5634 | if (rv < 0) |
5635 | mddev->delta_disks = 0; | |
da943b99 N |
5636 | return rv; |
5637 | } | |
5638 | ||
5639 | ||
1da177e4 LT |
5640 | /* |
5641 | * update_array_info is used to change the configuration of an | |
5642 | * on-line array. | |
5643 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5644 | * fields in the info are checked against the array. | |
5645 | * Any differences that cannot be handled will cause an error. | |
5646 | * Normally, only one change can be managed at a time. | |
5647 | */ | |
5648 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5649 | { | |
5650 | int rv = 0; | |
5651 | int cnt = 0; | |
36fa3063 N |
5652 | int state = 0; |
5653 | ||
5654 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 5655 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5656 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
5657 | |
5658 | if (mddev->major_version != info->major_version || | |
5659 | mddev->minor_version != info->minor_version || | |
5660 | /* mddev->patch_version != info->patch_version || */ | |
5661 | mddev->ctime != info->ctime || | |
5662 | mddev->level != info->level || | |
5663 | /* mddev->layout != info->layout || */ | |
5664 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5665 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5666 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5667 | ((state^info->state) & 0xfffffe00) | |
5668 | ) | |
1da177e4 LT |
5669 | return -EINVAL; |
5670 | /* Check there is only one change */ | |
58c0fed4 AN |
5671 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5672 | cnt++; | |
5673 | if (mddev->raid_disks != info->raid_disks) | |
5674 | cnt++; | |
5675 | if (mddev->layout != info->layout) | |
5676 | cnt++; | |
5677 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5678 | cnt++; | |
5679 | if (cnt == 0) | |
5680 | return 0; | |
5681 | if (cnt > 1) | |
5682 | return -EINVAL; | |
1da177e4 LT |
5683 | |
5684 | if (mddev->layout != info->layout) { | |
5685 | /* Change layout | |
5686 | * we don't need to do anything at the md level, the | |
5687 | * personality will take care of it all. | |
5688 | */ | |
50ac168a | 5689 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5690 | return -EINVAL; |
597a711b N |
5691 | else { |
5692 | mddev->new_layout = info->layout; | |
50ac168a | 5693 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5694 | if (rv) |
5695 | mddev->new_layout = mddev->layout; | |
5696 | return rv; | |
5697 | } | |
1da177e4 | 5698 | } |
58c0fed4 | 5699 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5700 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5701 | |
da943b99 N |
5702 | if (mddev->raid_disks != info->raid_disks) |
5703 | rv = update_raid_disks(mddev, info->raid_disks); | |
5704 | ||
36fa3063 N |
5705 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5706 | if (mddev->pers->quiesce == NULL) | |
5707 | return -EINVAL; | |
5708 | if (mddev->recovery || mddev->sync_thread) | |
5709 | return -EBUSY; | |
5710 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5711 | /* add the bitmap */ | |
5712 | if (mddev->bitmap) | |
5713 | return -EEXIST; | |
c3d9714e | 5714 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 5715 | return -EINVAL; |
c3d9714e N |
5716 | mddev->bitmap_info.offset = |
5717 | mddev->bitmap_info.default_offset; | |
36fa3063 N |
5718 | mddev->pers->quiesce(mddev, 1); |
5719 | rv = bitmap_create(mddev); | |
69e51b44 N |
5720 | if (!rv) |
5721 | rv = bitmap_load(mddev); | |
36fa3063 N |
5722 | if (rv) |
5723 | bitmap_destroy(mddev); | |
5724 | mddev->pers->quiesce(mddev, 0); | |
5725 | } else { | |
5726 | /* remove the bitmap */ | |
5727 | if (!mddev->bitmap) | |
5728 | return -ENOENT; | |
5729 | if (mddev->bitmap->file) | |
5730 | return -EINVAL; | |
5731 | mddev->pers->quiesce(mddev, 1); | |
5732 | bitmap_destroy(mddev); | |
5733 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 5734 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
5735 | } |
5736 | } | |
850b2b42 | 5737 | md_update_sb(mddev, 1); |
1da177e4 LT |
5738 | return rv; |
5739 | } | |
5740 | ||
5741 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5742 | { | |
5743 | mdk_rdev_t *rdev; | |
5744 | ||
5745 | if (mddev->pers == NULL) | |
5746 | return -ENODEV; | |
5747 | ||
5748 | rdev = find_rdev(mddev, dev); | |
5749 | if (!rdev) | |
5750 | return -ENODEV; | |
5751 | ||
5752 | md_error(mddev, rdev); | |
5753 | return 0; | |
5754 | } | |
5755 | ||
2f9618ce AN |
5756 | /* |
5757 | * We have a problem here : there is no easy way to give a CHS | |
5758 | * virtual geometry. We currently pretend that we have a 2 heads | |
5759 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5760 | * dosfs just mad... ;-) | |
5761 | */ | |
a885c8c4 CH |
5762 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5763 | { | |
5764 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5765 | ||
5766 | geo->heads = 2; | |
5767 | geo->sectors = 4; | |
49ce6cea | 5768 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
5769 | return 0; |
5770 | } | |
5771 | ||
a39907fa | 5772 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5773 | unsigned int cmd, unsigned long arg) |
5774 | { | |
5775 | int err = 0; | |
5776 | void __user *argp = (void __user *)arg; | |
1da177e4 | 5777 | mddev_t *mddev = NULL; |
e2218350 | 5778 | int ro; |
1da177e4 LT |
5779 | |
5780 | if (!capable(CAP_SYS_ADMIN)) | |
5781 | return -EACCES; | |
5782 | ||
5783 | /* | |
5784 | * Commands dealing with the RAID driver but not any | |
5785 | * particular array: | |
5786 | */ | |
5787 | switch (cmd) | |
5788 | { | |
5789 | case RAID_VERSION: | |
5790 | err = get_version(argp); | |
5791 | goto done; | |
5792 | ||
5793 | case PRINT_RAID_DEBUG: | |
5794 | err = 0; | |
5795 | md_print_devices(); | |
5796 | goto done; | |
5797 | ||
5798 | #ifndef MODULE | |
5799 | case RAID_AUTORUN: | |
5800 | err = 0; | |
5801 | autostart_arrays(arg); | |
5802 | goto done; | |
5803 | #endif | |
5804 | default:; | |
5805 | } | |
5806 | ||
5807 | /* | |
5808 | * Commands creating/starting a new array: | |
5809 | */ | |
5810 | ||
a39907fa | 5811 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5812 | |
5813 | if (!mddev) { | |
5814 | BUG(); | |
5815 | goto abort; | |
5816 | } | |
5817 | ||
1da177e4 LT |
5818 | err = mddev_lock(mddev); |
5819 | if (err) { | |
5820 | printk(KERN_INFO | |
5821 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5822 | err, cmd); | |
5823 | goto abort; | |
5824 | } | |
5825 | ||
5826 | switch (cmd) | |
5827 | { | |
5828 | case SET_ARRAY_INFO: | |
5829 | { | |
5830 | mdu_array_info_t info; | |
5831 | if (!arg) | |
5832 | memset(&info, 0, sizeof(info)); | |
5833 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5834 | err = -EFAULT; | |
5835 | goto abort_unlock; | |
5836 | } | |
5837 | if (mddev->pers) { | |
5838 | err = update_array_info(mddev, &info); | |
5839 | if (err) { | |
5840 | printk(KERN_WARNING "md: couldn't update" | |
5841 | " array info. %d\n", err); | |
5842 | goto abort_unlock; | |
5843 | } | |
5844 | goto done_unlock; | |
5845 | } | |
5846 | if (!list_empty(&mddev->disks)) { | |
5847 | printk(KERN_WARNING | |
5848 | "md: array %s already has disks!\n", | |
5849 | mdname(mddev)); | |
5850 | err = -EBUSY; | |
5851 | goto abort_unlock; | |
5852 | } | |
5853 | if (mddev->raid_disks) { | |
5854 | printk(KERN_WARNING | |
5855 | "md: array %s already initialised!\n", | |
5856 | mdname(mddev)); | |
5857 | err = -EBUSY; | |
5858 | goto abort_unlock; | |
5859 | } | |
5860 | err = set_array_info(mddev, &info); | |
5861 | if (err) { | |
5862 | printk(KERN_WARNING "md: couldn't set" | |
5863 | " array info. %d\n", err); | |
5864 | goto abort_unlock; | |
5865 | } | |
5866 | } | |
5867 | goto done_unlock; | |
5868 | ||
5869 | default:; | |
5870 | } | |
5871 | ||
5872 | /* | |
5873 | * Commands querying/configuring an existing array: | |
5874 | */ | |
32a7627c | 5875 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5876 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5877 | if ((!mddev->raid_disks && !mddev->external) |
5878 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5879 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5880 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5881 | err = -ENODEV; |
5882 | goto abort_unlock; | |
5883 | } | |
5884 | ||
5885 | /* | |
5886 | * Commands even a read-only array can execute: | |
5887 | */ | |
5888 | switch (cmd) | |
5889 | { | |
5890 | case GET_ARRAY_INFO: | |
5891 | err = get_array_info(mddev, argp); | |
5892 | goto done_unlock; | |
5893 | ||
32a7627c | 5894 | case GET_BITMAP_FILE: |
87162a28 | 5895 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5896 | goto done_unlock; |
5897 | ||
1da177e4 LT |
5898 | case GET_DISK_INFO: |
5899 | err = get_disk_info(mddev, argp); | |
5900 | goto done_unlock; | |
5901 | ||
5902 | case RESTART_ARRAY_RW: | |
5903 | err = restart_array(mddev); | |
5904 | goto done_unlock; | |
5905 | ||
5906 | case STOP_ARRAY: | |
d710e138 | 5907 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5908 | goto done_unlock; |
5909 | ||
5910 | case STOP_ARRAY_RO: | |
a4bd82d0 | 5911 | err = md_set_readonly(mddev, 1); |
1da177e4 LT |
5912 | goto done_unlock; |
5913 | ||
e2218350 DW |
5914 | case BLKROSET: |
5915 | if (get_user(ro, (int __user *)(arg))) { | |
5916 | err = -EFAULT; | |
5917 | goto done_unlock; | |
5918 | } | |
5919 | err = -EINVAL; | |
5920 | ||
5921 | /* if the bdev is going readonly the value of mddev->ro | |
5922 | * does not matter, no writes are coming | |
5923 | */ | |
5924 | if (ro) | |
5925 | goto done_unlock; | |
5926 | ||
5927 | /* are we are already prepared for writes? */ | |
5928 | if (mddev->ro != 1) | |
5929 | goto done_unlock; | |
5930 | ||
5931 | /* transitioning to readauto need only happen for | |
5932 | * arrays that call md_write_start | |
5933 | */ | |
5934 | if (mddev->pers) { | |
5935 | err = restart_array(mddev); | |
5936 | if (err == 0) { | |
5937 | mddev->ro = 2; | |
5938 | set_disk_ro(mddev->gendisk, 0); | |
5939 | } | |
5940 | } | |
5941 | goto done_unlock; | |
1da177e4 LT |
5942 | } |
5943 | ||
5944 | /* | |
5945 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5946 | * superblock, so we do not allow them on read-only arrays. |
5947 | * However non-MD ioctls (e.g. get-size) will still come through | |
5948 | * here and hit the 'default' below, so only disallow | |
5949 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5950 | */ |
bb57fc64 | 5951 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5952 | if (mddev->ro == 2) { |
5953 | mddev->ro = 0; | |
00bcb4ac | 5954 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 NB |
5955 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5956 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5957 | } else { |
5958 | err = -EROFS; | |
5959 | goto abort_unlock; | |
5960 | } | |
1da177e4 LT |
5961 | } |
5962 | ||
5963 | switch (cmd) | |
5964 | { | |
5965 | case ADD_NEW_DISK: | |
5966 | { | |
5967 | mdu_disk_info_t info; | |
5968 | if (copy_from_user(&info, argp, sizeof(info))) | |
5969 | err = -EFAULT; | |
5970 | else | |
5971 | err = add_new_disk(mddev, &info); | |
5972 | goto done_unlock; | |
5973 | } | |
5974 | ||
5975 | case HOT_REMOVE_DISK: | |
5976 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5977 | goto done_unlock; | |
5978 | ||
5979 | case HOT_ADD_DISK: | |
5980 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5981 | goto done_unlock; | |
5982 | ||
5983 | case SET_DISK_FAULTY: | |
5984 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5985 | goto done_unlock; | |
5986 | ||
5987 | case RUN_ARRAY: | |
d710e138 | 5988 | err = do_md_run(mddev); |
1da177e4 LT |
5989 | goto done_unlock; |
5990 | ||
32a7627c N |
5991 | case SET_BITMAP_FILE: |
5992 | err = set_bitmap_file(mddev, (int)arg); | |
5993 | goto done_unlock; | |
5994 | ||
1da177e4 | 5995 | default: |
1da177e4 LT |
5996 | err = -EINVAL; |
5997 | goto abort_unlock; | |
5998 | } | |
5999 | ||
6000 | done_unlock: | |
6001 | abort_unlock: | |
d3374825 N |
6002 | if (mddev->hold_active == UNTIL_IOCTL && |
6003 | err != -EINVAL) | |
6004 | mddev->hold_active = 0; | |
1da177e4 LT |
6005 | mddev_unlock(mddev); |
6006 | ||
6007 | return err; | |
6008 | done: | |
6009 | if (err) | |
6010 | MD_BUG(); | |
6011 | abort: | |
6012 | return err; | |
6013 | } | |
aa98aa31 AB |
6014 | #ifdef CONFIG_COMPAT |
6015 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6016 | unsigned int cmd, unsigned long arg) | |
6017 | { | |
6018 | switch (cmd) { | |
6019 | case HOT_REMOVE_DISK: | |
6020 | case HOT_ADD_DISK: | |
6021 | case SET_DISK_FAULTY: | |
6022 | case SET_BITMAP_FILE: | |
6023 | /* These take in integer arg, do not convert */ | |
6024 | break; | |
6025 | default: | |
6026 | arg = (unsigned long)compat_ptr(arg); | |
6027 | break; | |
6028 | } | |
6029 | ||
6030 | return md_ioctl(bdev, mode, cmd, arg); | |
6031 | } | |
6032 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6033 | |
a39907fa | 6034 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6035 | { |
6036 | /* | |
6037 | * Succeed if we can lock the mddev, which confirms that | |
6038 | * it isn't being stopped right now. | |
6039 | */ | |
d3374825 | 6040 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6041 | int err; |
6042 | ||
d3374825 N |
6043 | if (mddev->gendisk != bdev->bd_disk) { |
6044 | /* we are racing with mddev_put which is discarding this | |
6045 | * bd_disk. | |
6046 | */ | |
6047 | mddev_put(mddev); | |
6048 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6049 | flush_workqueue(md_misc_wq); |
d3374825 N |
6050 | /* Then retry the open from the top */ |
6051 | return -ERESTARTSYS; | |
6052 | } | |
6053 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6054 | ||
c8c00a69 | 6055 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6056 | goto out; |
6057 | ||
6058 | err = 0; | |
f2ea68cf | 6059 | atomic_inc(&mddev->openers); |
c8c00a69 | 6060 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6061 | |
f0b4f7e2 | 6062 | check_disk_change(bdev); |
1da177e4 LT |
6063 | out: |
6064 | return err; | |
6065 | } | |
6066 | ||
a39907fa | 6067 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6068 | { |
a39907fa | 6069 | mddev_t *mddev = disk->private_data; |
1da177e4 | 6070 | |
52e5f9d1 | 6071 | BUG_ON(!mddev); |
f2ea68cf | 6072 | atomic_dec(&mddev->openers); |
1da177e4 LT |
6073 | mddev_put(mddev); |
6074 | ||
6075 | return 0; | |
6076 | } | |
f0b4f7e2 N |
6077 | |
6078 | static int md_media_changed(struct gendisk *disk) | |
6079 | { | |
6080 | mddev_t *mddev = disk->private_data; | |
6081 | ||
6082 | return mddev->changed; | |
6083 | } | |
6084 | ||
6085 | static int md_revalidate(struct gendisk *disk) | |
6086 | { | |
6087 | mddev_t *mddev = disk->private_data; | |
6088 | ||
6089 | mddev->changed = 0; | |
6090 | return 0; | |
6091 | } | |
83d5cde4 | 6092 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6093 | { |
6094 | .owner = THIS_MODULE, | |
a39907fa AV |
6095 | .open = md_open, |
6096 | .release = md_release, | |
b492b852 | 6097 | .ioctl = md_ioctl, |
aa98aa31 AB |
6098 | #ifdef CONFIG_COMPAT |
6099 | .compat_ioctl = md_compat_ioctl, | |
6100 | #endif | |
a885c8c4 | 6101 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6102 | .media_changed = md_media_changed, |
6103 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6104 | }; |
6105 | ||
75c96f85 | 6106 | static int md_thread(void * arg) |
1da177e4 LT |
6107 | { |
6108 | mdk_thread_t *thread = arg; | |
6109 | ||
1da177e4 LT |
6110 | /* |
6111 | * md_thread is a 'system-thread', it's priority should be very | |
6112 | * high. We avoid resource deadlocks individually in each | |
6113 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6114 | * the very same RT priority as kswapd, thus we will never get | |
6115 | * into a priority inversion deadlock. | |
6116 | * | |
6117 | * we definitely have to have equal or higher priority than | |
6118 | * bdflush, otherwise bdflush will deadlock if there are too | |
6119 | * many dirty RAID5 blocks. | |
6120 | */ | |
1da177e4 | 6121 | |
6985c43f | 6122 | allow_signal(SIGKILL); |
a6fb0934 | 6123 | while (!kthread_should_stop()) { |
1da177e4 | 6124 | |
93588e22 N |
6125 | /* We need to wait INTERRUPTIBLE so that |
6126 | * we don't add to the load-average. | |
6127 | * That means we need to be sure no signals are | |
6128 | * pending | |
6129 | */ | |
6130 | if (signal_pending(current)) | |
6131 | flush_signals(current); | |
6132 | ||
6133 | wait_event_interruptible_timeout | |
6134 | (thread->wqueue, | |
6135 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6136 | || kthread_should_stop(), | |
6137 | thread->timeout); | |
1da177e4 | 6138 | |
6c987910 N |
6139 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6140 | if (!kthread_should_stop()) | |
589a594b | 6141 | thread->run(thread->mddev); |
1da177e4 | 6142 | } |
a6fb0934 | 6143 | |
1da177e4 LT |
6144 | return 0; |
6145 | } | |
6146 | ||
6147 | void md_wakeup_thread(mdk_thread_t *thread) | |
6148 | { | |
6149 | if (thread) { | |
6150 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
6151 | set_bit(THREAD_WAKEUP, &thread->flags); | |
6152 | wake_up(&thread->wqueue); | |
6153 | } | |
6154 | } | |
6155 | ||
6156 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
6157 | const char *name) | |
6158 | { | |
6159 | mdk_thread_t *thread; | |
1da177e4 | 6160 | |
9ffae0cf | 6161 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
6162 | if (!thread) |
6163 | return NULL; | |
6164 | ||
1da177e4 LT |
6165 | init_waitqueue_head(&thread->wqueue); |
6166 | ||
1da177e4 LT |
6167 | thread->run = run; |
6168 | thread->mddev = mddev; | |
32a7627c | 6169 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6170 | thread->tsk = kthread_run(md_thread, thread, |
6171 | "%s_%s", | |
6172 | mdname(thread->mddev), | |
6173 | name ?: mddev->pers->name); | |
a6fb0934 | 6174 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6175 | kfree(thread); |
6176 | return NULL; | |
6177 | } | |
1da177e4 LT |
6178 | return thread; |
6179 | } | |
6180 | ||
1da177e4 LT |
6181 | void md_unregister_thread(mdk_thread_t *thread) |
6182 | { | |
e0cf8f04 N |
6183 | if (!thread) |
6184 | return; | |
ba25f9dc | 6185 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
6186 | |
6187 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6188 | kfree(thread); |
6189 | } | |
6190 | ||
6191 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
6192 | { | |
6193 | if (!mddev) { | |
6194 | MD_BUG(); | |
6195 | return; | |
6196 | } | |
6197 | ||
b2d444d7 | 6198 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6199 | return; |
6bfe0b49 DW |
6200 | |
6201 | if (mddev->external) | |
6202 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 6203 | /* |
1da177e4 LT |
6204 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
6205 | mdname(mddev), | |
6206 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
6207 | __builtin_return_address(0),__builtin_return_address(1), | |
6208 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 6209 | */ |
d0a0a5ee AM |
6210 | if (!mddev->pers) |
6211 | return; | |
1da177e4 LT |
6212 | if (!mddev->pers->error_handler) |
6213 | return; | |
6214 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6215 | if (mddev->degraded) |
6216 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6217 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6218 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6219 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6220 | md_wakeup_thread(mddev->thread); | |
768a418d | 6221 | if (mddev->event_work.func) |
e804ac78 | 6222 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6223 | md_new_event_inintr(mddev); |
1da177e4 LT |
6224 | } |
6225 | ||
6226 | /* seq_file implementation /proc/mdstat */ | |
6227 | ||
6228 | static void status_unused(struct seq_file *seq) | |
6229 | { | |
6230 | int i = 0; | |
6231 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6232 | |
6233 | seq_printf(seq, "unused devices: "); | |
6234 | ||
159ec1fc | 6235 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6236 | char b[BDEVNAME_SIZE]; |
6237 | i++; | |
6238 | seq_printf(seq, "%s ", | |
6239 | bdevname(rdev->bdev,b)); | |
6240 | } | |
6241 | if (!i) | |
6242 | seq_printf(seq, "<none>"); | |
6243 | ||
6244 | seq_printf(seq, "\n"); | |
6245 | } | |
6246 | ||
6247 | ||
6248 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
6249 | { | |
dd71cf6b N |
6250 | sector_t max_sectors, resync, res; |
6251 | unsigned long dt, db; | |
6252 | sector_t rt; | |
4588b42e N |
6253 | int scale; |
6254 | unsigned int per_milli; | |
1da177e4 | 6255 | |
dd71cf6b | 6256 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6257 | |
6258 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 6259 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6260 | else |
dd71cf6b | 6261 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6262 | |
6263 | /* | |
6264 | * Should not happen. | |
6265 | */ | |
dd71cf6b | 6266 | if (!max_sectors) { |
1da177e4 LT |
6267 | MD_BUG(); |
6268 | return; | |
6269 | } | |
4588b42e | 6270 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6271 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6272 | * u32, as those are the requirements for sector_div. |
6273 | * Thus 'scale' must be at least 10 | |
6274 | */ | |
6275 | scale = 10; | |
6276 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6277 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6278 | scale++; |
6279 | } | |
6280 | res = (resync>>scale)*1000; | |
dd71cf6b | 6281 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6282 | |
6283 | per_milli = res; | |
1da177e4 | 6284 | { |
4588b42e | 6285 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6286 | seq_printf(seq, "["); |
6287 | for (i = 0; i < x; i++) | |
6288 | seq_printf(seq, "="); | |
6289 | seq_printf(seq, ">"); | |
6290 | for (i = 0; i < y; i++) | |
6291 | seq_printf(seq, "."); | |
6292 | seq_printf(seq, "] "); | |
6293 | } | |
4588b42e | 6294 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6295 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6296 | "reshape" : | |
61df9d91 N |
6297 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6298 | "check" : | |
6299 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6300 | "resync" : "recovery"))), | |
6301 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6302 | (unsigned long long) resync/2, |
6303 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6304 | |
6305 | /* | |
1da177e4 LT |
6306 | * dt: time from mark until now |
6307 | * db: blocks written from mark until now | |
6308 | * rt: remaining time | |
dd71cf6b N |
6309 | * |
6310 | * rt is a sector_t, so could be 32bit or 64bit. | |
6311 | * So we divide before multiply in case it is 32bit and close | |
6312 | * to the limit. | |
25985edc | 6313 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
6314 | * near the end of resync when the number of remaining sectors |
6315 | * is close to 'db'. | |
6316 | * We then divide rt by 32 after multiplying by db to compensate. | |
6317 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6318 | */ |
6319 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6320 | if (!dt) dt++; | |
ff4e8d9a N |
6321 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6322 | - mddev->resync_mark_cnt; | |
1da177e4 | 6323 | |
dd71cf6b N |
6324 | rt = max_sectors - resync; /* number of remaining sectors */ |
6325 | sector_div(rt, db/32+1); | |
6326 | rt *= dt; | |
6327 | rt >>= 5; | |
6328 | ||
6329 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6330 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6331 | |
ff4e8d9a | 6332 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6333 | } |
6334 | ||
6335 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6336 | { | |
6337 | struct list_head *tmp; | |
6338 | loff_t l = *pos; | |
6339 | mddev_t *mddev; | |
6340 | ||
6341 | if (l >= 0x10000) | |
6342 | return NULL; | |
6343 | if (!l--) | |
6344 | /* header */ | |
6345 | return (void*)1; | |
6346 | ||
6347 | spin_lock(&all_mddevs_lock); | |
6348 | list_for_each(tmp,&all_mddevs) | |
6349 | if (!l--) { | |
6350 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
6351 | mddev_get(mddev); | |
6352 | spin_unlock(&all_mddevs_lock); | |
6353 | return mddev; | |
6354 | } | |
6355 | spin_unlock(&all_mddevs_lock); | |
6356 | if (!l--) | |
6357 | return (void*)2;/* tail */ | |
6358 | return NULL; | |
6359 | } | |
6360 | ||
6361 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6362 | { | |
6363 | struct list_head *tmp; | |
6364 | mddev_t *next_mddev, *mddev = v; | |
6365 | ||
6366 | ++*pos; | |
6367 | if (v == (void*)2) | |
6368 | return NULL; | |
6369 | ||
6370 | spin_lock(&all_mddevs_lock); | |
6371 | if (v == (void*)1) | |
6372 | tmp = all_mddevs.next; | |
6373 | else | |
6374 | tmp = mddev->all_mddevs.next; | |
6375 | if (tmp != &all_mddevs) | |
6376 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
6377 | else { | |
6378 | next_mddev = (void*)2; | |
6379 | *pos = 0x10000; | |
6380 | } | |
6381 | spin_unlock(&all_mddevs_lock); | |
6382 | ||
6383 | if (v != (void*)1) | |
6384 | mddev_put(mddev); | |
6385 | return next_mddev; | |
6386 | ||
6387 | } | |
6388 | ||
6389 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6390 | { | |
6391 | mddev_t *mddev = v; | |
6392 | ||
6393 | if (mddev && v != (void*)1 && v != (void*)2) | |
6394 | mddev_put(mddev); | |
6395 | } | |
6396 | ||
d7603b7e N |
6397 | struct mdstat_info { |
6398 | int event; | |
6399 | }; | |
6400 | ||
1da177e4 LT |
6401 | static int md_seq_show(struct seq_file *seq, void *v) |
6402 | { | |
6403 | mddev_t *mddev = v; | |
dd8ac336 | 6404 | sector_t sectors; |
1da177e4 | 6405 | mdk_rdev_t *rdev; |
d7603b7e | 6406 | struct mdstat_info *mi = seq->private; |
32a7627c | 6407 | struct bitmap *bitmap; |
1da177e4 LT |
6408 | |
6409 | if (v == (void*)1) { | |
2604b703 | 6410 | struct mdk_personality *pers; |
1da177e4 LT |
6411 | seq_printf(seq, "Personalities : "); |
6412 | spin_lock(&pers_lock); | |
2604b703 N |
6413 | list_for_each_entry(pers, &pers_list, list) |
6414 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6415 | |
6416 | spin_unlock(&pers_lock); | |
6417 | seq_printf(seq, "\n"); | |
d7603b7e | 6418 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
6419 | return 0; |
6420 | } | |
6421 | if (v == (void*)2) { | |
6422 | status_unused(seq); | |
6423 | return 0; | |
6424 | } | |
6425 | ||
5dc5cf7d | 6426 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6427 | return -EINTR; |
5dc5cf7d | 6428 | |
1da177e4 LT |
6429 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
6430 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
6431 | mddev->pers ? "" : "in"); | |
6432 | if (mddev->pers) { | |
f91de92e | 6433 | if (mddev->ro==1) |
1da177e4 | 6434 | seq_printf(seq, " (read-only)"); |
f91de92e | 6435 | if (mddev->ro==2) |
52720ae7 | 6436 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
6437 | seq_printf(seq, " %s", mddev->pers->name); |
6438 | } | |
6439 | ||
dd8ac336 | 6440 | sectors = 0; |
159ec1fc | 6441 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
6442 | char b[BDEVNAME_SIZE]; |
6443 | seq_printf(seq, " %s[%d]", | |
6444 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
6445 | if (test_bit(WriteMostly, &rdev->flags)) |
6446 | seq_printf(seq, "(W)"); | |
b2d444d7 | 6447 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6448 | seq_printf(seq, "(F)"); |
6449 | continue; | |
b325a32e N |
6450 | } else if (rdev->raid_disk < 0) |
6451 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 6452 | sectors += rdev->sectors; |
1da177e4 LT |
6453 | } |
6454 | ||
6455 | if (!list_empty(&mddev->disks)) { | |
6456 | if (mddev->pers) | |
6457 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
6458 | (unsigned long long) |
6459 | mddev->array_sectors / 2); | |
1da177e4 LT |
6460 | else |
6461 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 6462 | (unsigned long long)sectors / 2); |
1da177e4 | 6463 | } |
1cd6bf19 N |
6464 | if (mddev->persistent) { |
6465 | if (mddev->major_version != 0 || | |
6466 | mddev->minor_version != 90) { | |
6467 | seq_printf(seq," super %d.%d", | |
6468 | mddev->major_version, | |
6469 | mddev->minor_version); | |
6470 | } | |
e691063a N |
6471 | } else if (mddev->external) |
6472 | seq_printf(seq, " super external:%s", | |
6473 | mddev->metadata_type); | |
6474 | else | |
1cd6bf19 | 6475 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
6476 | |
6477 | if (mddev->pers) { | |
d710e138 | 6478 | mddev->pers->status(seq, mddev); |
1da177e4 | 6479 | seq_printf(seq, "\n "); |
8e1b39d6 N |
6480 | if (mddev->pers->sync_request) { |
6481 | if (mddev->curr_resync > 2) { | |
d710e138 | 6482 | status_resync(seq, mddev); |
8e1b39d6 N |
6483 | seq_printf(seq, "\n "); |
6484 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
6485 | seq_printf(seq, "\tresync=DELAYED\n "); | |
6486 | else if (mddev->recovery_cp < MaxSector) | |
6487 | seq_printf(seq, "\tresync=PENDING\n "); | |
6488 | } | |
32a7627c N |
6489 | } else |
6490 | seq_printf(seq, "\n "); | |
6491 | ||
6492 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
6493 | unsigned long chunk_kb; |
6494 | unsigned long flags; | |
32a7627c | 6495 | spin_lock_irqsave(&bitmap->lock, flags); |
42a04b50 | 6496 | chunk_kb = mddev->bitmap_info.chunksize >> 10; |
32a7627c N |
6497 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " |
6498 | "%lu%s chunk", | |
6499 | bitmap->pages - bitmap->missing_pages, | |
6500 | bitmap->pages, | |
6501 | (bitmap->pages - bitmap->missing_pages) | |
6502 | << (PAGE_SHIFT - 10), | |
42a04b50 | 6503 | chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, |
32a7627c | 6504 | chunk_kb ? "KB" : "B"); |
78d742d8 N |
6505 | if (bitmap->file) { |
6506 | seq_printf(seq, ", file: "); | |
c32c2f63 | 6507 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 6508 | } |
78d742d8 | 6509 | |
32a7627c N |
6510 | seq_printf(seq, "\n"); |
6511 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
6512 | } |
6513 | ||
6514 | seq_printf(seq, "\n"); | |
6515 | } | |
6516 | mddev_unlock(mddev); | |
6517 | ||
6518 | return 0; | |
6519 | } | |
6520 | ||
110518bc | 6521 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6522 | .start = md_seq_start, |
6523 | .next = md_seq_next, | |
6524 | .stop = md_seq_stop, | |
6525 | .show = md_seq_show, | |
6526 | }; | |
6527 | ||
6528 | static int md_seq_open(struct inode *inode, struct file *file) | |
6529 | { | |
6530 | int error; | |
d7603b7e N |
6531 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6532 | if (mi == NULL) | |
6533 | return -ENOMEM; | |
1da177e4 LT |
6534 | |
6535 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6536 | if (error) |
6537 | kfree(mi); | |
6538 | else { | |
6539 | struct seq_file *p = file->private_data; | |
6540 | p->private = mi; | |
6541 | mi->event = atomic_read(&md_event_count); | |
6542 | } | |
1da177e4 LT |
6543 | return error; |
6544 | } | |
6545 | ||
d7603b7e N |
6546 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6547 | { | |
6548 | struct seq_file *m = filp->private_data; | |
6549 | struct mdstat_info *mi = m->private; | |
6550 | int mask; | |
6551 | ||
6552 | poll_wait(filp, &md_event_waiters, wait); | |
6553 | ||
6554 | /* always allow read */ | |
6555 | mask = POLLIN | POLLRDNORM; | |
6556 | ||
6557 | if (mi->event != atomic_read(&md_event_count)) | |
6558 | mask |= POLLERR | POLLPRI; | |
6559 | return mask; | |
6560 | } | |
6561 | ||
fa027c2a | 6562 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6563 | .owner = THIS_MODULE, |
1da177e4 LT |
6564 | .open = md_seq_open, |
6565 | .read = seq_read, | |
6566 | .llseek = seq_lseek, | |
c3f94b40 | 6567 | .release = seq_release_private, |
d7603b7e | 6568 | .poll = mdstat_poll, |
1da177e4 LT |
6569 | }; |
6570 | ||
2604b703 | 6571 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6572 | { |
1da177e4 | 6573 | spin_lock(&pers_lock); |
2604b703 N |
6574 | list_add_tail(&p->list, &pers_list); |
6575 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6576 | spin_unlock(&pers_lock); |
6577 | return 0; | |
6578 | } | |
6579 | ||
2604b703 | 6580 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6581 | { |
2604b703 | 6582 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6583 | spin_lock(&pers_lock); |
2604b703 | 6584 | list_del_init(&p->list); |
1da177e4 LT |
6585 | spin_unlock(&pers_lock); |
6586 | return 0; | |
6587 | } | |
6588 | ||
eea1bf38 | 6589 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6590 | { |
6591 | mdk_rdev_t * rdev; | |
1da177e4 | 6592 | int idle; |
eea1bf38 | 6593 | int curr_events; |
1da177e4 LT |
6594 | |
6595 | idle = 1; | |
4b80991c N |
6596 | rcu_read_lock(); |
6597 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6598 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6599 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6600 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6601 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6602 | /* sync IO will cause sync_io to increase before the disk_stats |
6603 | * as sync_io is counted when a request starts, and | |
6604 | * disk_stats is counted when it completes. | |
6605 | * So resync activity will cause curr_events to be smaller than | |
6606 | * when there was no such activity. | |
6607 | * non-sync IO will cause disk_stat to increase without | |
6608 | * increasing sync_io so curr_events will (eventually) | |
6609 | * be larger than it was before. Once it becomes | |
6610 | * substantially larger, the test below will cause | |
6611 | * the array to appear non-idle, and resync will slow | |
6612 | * down. | |
6613 | * If there is a lot of outstanding resync activity when | |
6614 | * we set last_event to curr_events, then all that activity | |
6615 | * completing might cause the array to appear non-idle | |
6616 | * and resync will be slowed down even though there might | |
6617 | * not have been non-resync activity. This will only | |
6618 | * happen once though. 'last_events' will soon reflect | |
6619 | * the state where there is little or no outstanding | |
6620 | * resync requests, and further resync activity will | |
6621 | * always make curr_events less than last_events. | |
c0e48521 | 6622 | * |
1da177e4 | 6623 | */ |
eea1bf38 | 6624 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6625 | rdev->last_events = curr_events; |
6626 | idle = 0; | |
6627 | } | |
6628 | } | |
4b80991c | 6629 | rcu_read_unlock(); |
1da177e4 LT |
6630 | return idle; |
6631 | } | |
6632 | ||
6633 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6634 | { | |
6635 | /* another "blocks" (512byte) blocks have been synced */ | |
6636 | atomic_sub(blocks, &mddev->recovery_active); | |
6637 | wake_up(&mddev->recovery_wait); | |
6638 | if (!ok) { | |
dfc70645 | 6639 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6640 | md_wakeup_thread(mddev->thread); |
6641 | // stop recovery, signal do_sync .... | |
6642 | } | |
6643 | } | |
6644 | ||
6645 | ||
06d91a5f N |
6646 | /* md_write_start(mddev, bi) |
6647 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6648 | * in superblock) before writing, schedule a superblock update |
6649 | * and wait for it to complete. | |
06d91a5f | 6650 | */ |
3d310eb7 | 6651 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6652 | { |
0fd62b86 | 6653 | int did_change = 0; |
06d91a5f | 6654 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6655 | return; |
06d91a5f | 6656 | |
f91de92e N |
6657 | BUG_ON(mddev->ro == 1); |
6658 | if (mddev->ro == 2) { | |
6659 | /* need to switch to read/write */ | |
6660 | mddev->ro = 0; | |
6661 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6662 | md_wakeup_thread(mddev->thread); | |
25156198 | 6663 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6664 | did_change = 1; |
f91de92e | 6665 | } |
06d91a5f | 6666 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6667 | if (mddev->safemode == 1) |
6668 | mddev->safemode = 0; | |
06d91a5f | 6669 | if (mddev->in_sync) { |
a9701a30 | 6670 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6671 | if (mddev->in_sync) { |
6672 | mddev->in_sync = 0; | |
850b2b42 | 6673 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 6674 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 6675 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6676 | did_change = 1; |
3d310eb7 | 6677 | } |
a9701a30 | 6678 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6679 | } |
0fd62b86 | 6680 | if (did_change) |
00bcb4ac | 6681 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 6682 | wait_event(mddev->sb_wait, |
09a44cc1 | 6683 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 LT |
6684 | } |
6685 | ||
6686 | void md_write_end(mddev_t *mddev) | |
6687 | { | |
6688 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6689 | if (mddev->safemode == 2) | |
6690 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6691 | else if (mddev->safemode_delay) |
1da177e4 LT |
6692 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6693 | } | |
6694 | } | |
6695 | ||
2a2275d6 N |
6696 | /* md_allow_write(mddev) |
6697 | * Calling this ensures that the array is marked 'active' so that writes | |
6698 | * may proceed without blocking. It is important to call this before | |
6699 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6700 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6701 | * |
6702 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6703 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6704 | */ |
b5470dc5 | 6705 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6706 | { |
6707 | if (!mddev->pers) | |
b5470dc5 | 6708 | return 0; |
2a2275d6 | 6709 | if (mddev->ro) |
b5470dc5 | 6710 | return 0; |
1a0fd497 | 6711 | if (!mddev->pers->sync_request) |
b5470dc5 | 6712 | return 0; |
2a2275d6 N |
6713 | |
6714 | spin_lock_irq(&mddev->write_lock); | |
6715 | if (mddev->in_sync) { | |
6716 | mddev->in_sync = 0; | |
6717 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 6718 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
6719 | if (mddev->safemode_delay && |
6720 | mddev->safemode == 0) | |
6721 | mddev->safemode = 1; | |
6722 | spin_unlock_irq(&mddev->write_lock); | |
6723 | md_update_sb(mddev, 0); | |
00bcb4ac | 6724 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
6725 | } else |
6726 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 | 6727 | |
070dc6dd | 6728 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
6729 | return -EAGAIN; |
6730 | else | |
6731 | return 0; | |
2a2275d6 N |
6732 | } |
6733 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6734 | ||
1da177e4 LT |
6735 | #define SYNC_MARKS 10 |
6736 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6737 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6738 | { |
6739 | mddev_t *mddev2; | |
6740 | unsigned int currspeed = 0, | |
6741 | window; | |
57afd89f | 6742 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6743 | unsigned long mark[SYNC_MARKS]; |
6744 | sector_t mark_cnt[SYNC_MARKS]; | |
6745 | int last_mark,m; | |
6746 | struct list_head *tmp; | |
6747 | sector_t last_check; | |
57afd89f | 6748 | int skipped = 0; |
5fd6c1dc | 6749 | mdk_rdev_t *rdev; |
61df9d91 | 6750 | char *desc; |
1da177e4 LT |
6751 | |
6752 | /* just incase thread restarts... */ | |
6753 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6754 | return; | |
5fd6c1dc N |
6755 | if (mddev->ro) /* never try to sync a read-only array */ |
6756 | return; | |
1da177e4 | 6757 | |
61df9d91 N |
6758 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6759 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6760 | desc = "data-check"; | |
6761 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6762 | desc = "requested-resync"; | |
6763 | else | |
6764 | desc = "resync"; | |
6765 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6766 | desc = "reshape"; | |
6767 | else | |
6768 | desc = "recovery"; | |
6769 | ||
1da177e4 LT |
6770 | /* we overload curr_resync somewhat here. |
6771 | * 0 == not engaged in resync at all | |
6772 | * 2 == checking that there is no conflict with another sync | |
6773 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6774 | * commense | |
6775 | * other == active in resync - this many blocks | |
6776 | * | |
6777 | * Before starting a resync we must have set curr_resync to | |
6778 | * 2, and then checked that every "conflicting" array has curr_resync | |
6779 | * less than ours. When we find one that is the same or higher | |
6780 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6781 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6782 | * This will mean we have to start checking from the beginning again. | |
6783 | * | |
6784 | */ | |
6785 | ||
6786 | do { | |
6787 | mddev->curr_resync = 2; | |
6788 | ||
6789 | try_again: | |
404e4b43 | 6790 | if (kthread_should_stop()) |
6985c43f | 6791 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
404e4b43 N |
6792 | |
6793 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
1da177e4 | 6794 | goto skip; |
29ac4aa3 | 6795 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6796 | if (mddev2 == mddev) |
6797 | continue; | |
90b08710 BS |
6798 | if (!mddev->parallel_resync |
6799 | && mddev2->curr_resync | |
6800 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6801 | DEFINE_WAIT(wq); |
6802 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6803 | /* arbitrarily yield */ | |
6804 | mddev->curr_resync = 1; | |
6805 | wake_up(&resync_wait); | |
6806 | } | |
6807 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6808 | /* no need to wait here, we can wait the next | |
6809 | * time 'round when curr_resync == 2 | |
6810 | */ | |
6811 | continue; | |
9744197c N |
6812 | /* We need to wait 'interruptible' so as not to |
6813 | * contribute to the load average, and not to | |
6814 | * be caught by 'softlockup' | |
6815 | */ | |
6816 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6817 | if (!kthread_should_stop() && |
8712e553 | 6818 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6819 | printk(KERN_INFO "md: delaying %s of %s" |
6820 | " until %s has finished (they" | |
1da177e4 | 6821 | " share one or more physical units)\n", |
61df9d91 | 6822 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6823 | mddev_put(mddev2); |
9744197c N |
6824 | if (signal_pending(current)) |
6825 | flush_signals(current); | |
1da177e4 LT |
6826 | schedule(); |
6827 | finish_wait(&resync_wait, &wq); | |
6828 | goto try_again; | |
6829 | } | |
6830 | finish_wait(&resync_wait, &wq); | |
6831 | } | |
6832 | } | |
6833 | } while (mddev->curr_resync < 2); | |
6834 | ||
5fd6c1dc | 6835 | j = 0; |
9d88883e | 6836 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6837 | /* resync follows the size requested by the personality, |
57afd89f | 6838 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6839 | */ |
6840 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6841 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6842 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6843 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6844 | j = mddev->resync_min; | |
6845 | else if (!mddev->bitmap) | |
5fd6c1dc | 6846 | j = mddev->recovery_cp; |
5e96ee65 | 6847 | |
ccfcc3c1 | 6848 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6849 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6850 | else { |
1da177e4 | 6851 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6852 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6853 | j = MaxSector; |
4e59ca7d DW |
6854 | rcu_read_lock(); |
6855 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6856 | if (rdev->raid_disk >= 0 && |
6857 | !test_bit(Faulty, &rdev->flags) && | |
6858 | !test_bit(In_sync, &rdev->flags) && | |
6859 | rdev->recovery_offset < j) | |
6860 | j = rdev->recovery_offset; | |
4e59ca7d | 6861 | rcu_read_unlock(); |
5fd6c1dc | 6862 | } |
1da177e4 | 6863 | |
61df9d91 N |
6864 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6865 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6866 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6867 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6868 | "(but not more than %d KB/sec) for %s.\n", |
6869 | speed_max(mddev), desc); | |
1da177e4 | 6870 | |
eea1bf38 | 6871 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6872 | |
57afd89f | 6873 | io_sectors = 0; |
1da177e4 LT |
6874 | for (m = 0; m < SYNC_MARKS; m++) { |
6875 | mark[m] = jiffies; | |
57afd89f | 6876 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6877 | } |
6878 | last_mark = 0; | |
6879 | mddev->resync_mark = mark[last_mark]; | |
6880 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6881 | ||
6882 | /* | |
6883 | * Tune reconstruction: | |
6884 | */ | |
6885 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
6886 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
6887 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
6888 | |
6889 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6890 | last_check = 0; |
6891 | ||
6892 | if (j>2) { | |
6893 | printk(KERN_INFO | |
61df9d91 N |
6894 | "md: resuming %s of %s from checkpoint.\n", |
6895 | desc, mdname(mddev)); | |
1da177e4 LT |
6896 | mddev->curr_resync = j; |
6897 | } | |
75d3da43 | 6898 | mddev->curr_resync_completed = j; |
1da177e4 LT |
6899 | |
6900 | while (j < max_sectors) { | |
57afd89f | 6901 | sector_t sectors; |
1da177e4 | 6902 | |
57afd89f | 6903 | skipped = 0; |
97e4f42d | 6904 | |
7a91ee1f N |
6905 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6906 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6907 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6908 | > (max_sectors >> 4)) || | |
6909 | (j - mddev->curr_resync_completed)*2 | |
6910 | >= mddev->resync_max - mddev->curr_resync_completed | |
6911 | )) { | |
97e4f42d | 6912 | /* time to update curr_resync_completed */ |
97e4f42d N |
6913 | wait_event(mddev->recovery_wait, |
6914 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 6915 | mddev->curr_resync_completed = j; |
070dc6dd | 6916 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 6917 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6918 | } |
acb180b0 | 6919 | |
e62e58a5 N |
6920 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6921 | /* As this condition is controlled by user-space, | |
6922 | * we can block indefinitely, so use '_interruptible' | |
6923 | * to avoid triggering warnings. | |
6924 | */ | |
6925 | flush_signals(current); /* just in case */ | |
6926 | wait_event_interruptible(mddev->recovery_wait, | |
6927 | mddev->resync_max > j | |
6928 | || kthread_should_stop()); | |
6929 | } | |
acb180b0 N |
6930 | |
6931 | if (kthread_should_stop()) | |
6932 | goto interrupted; | |
6933 | ||
57afd89f | 6934 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6935 | currspeed < speed_min(mddev)); |
57afd89f | 6936 | if (sectors == 0) { |
dfc70645 | 6937 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6938 | goto out; |
6939 | } | |
57afd89f N |
6940 | |
6941 | if (!skipped) { /* actual IO requested */ | |
6942 | io_sectors += sectors; | |
6943 | atomic_add(sectors, &mddev->recovery_active); | |
6944 | } | |
6945 | ||
1da177e4 LT |
6946 | j += sectors; |
6947 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6948 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6949 | if (last_check == 0) |
6950 | /* this is the earliers that rebuilt will be | |
6951 | * visible in /proc/mdstat | |
6952 | */ | |
6953 | md_new_event(mddev); | |
57afd89f N |
6954 | |
6955 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6956 | continue; |
6957 | ||
57afd89f | 6958 | last_check = io_sectors; |
1da177e4 | 6959 | |
dfc70645 | 6960 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6961 | break; |
6962 | ||
6963 | repeat: | |
6964 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6965 | /* step marks */ | |
6966 | int next = (last_mark+1) % SYNC_MARKS; | |
6967 | ||
6968 | mddev->resync_mark = mark[next]; | |
6969 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6970 | mark[next] = jiffies; | |
57afd89f | 6971 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6972 | last_mark = next; |
6973 | } | |
6974 | ||
6975 | ||
c6207277 N |
6976 | if (kthread_should_stop()) |
6977 | goto interrupted; | |
6978 | ||
1da177e4 LT |
6979 | |
6980 | /* | |
6981 | * this loop exits only if either when we are slower than | |
6982 | * the 'hard' speed limit, or the system was IO-idle for | |
6983 | * a jiffy. | |
6984 | * the system might be non-idle CPU-wise, but we only care | |
6985 | * about not overloading the IO subsystem. (things like an | |
6986 | * e2fsck being done on the RAID array should execute fast) | |
6987 | */ | |
1da177e4 LT |
6988 | cond_resched(); |
6989 | ||
57afd89f N |
6990 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6991 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6992 | |
88202a0c N |
6993 | if (currspeed > speed_min(mddev)) { |
6994 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6995 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6996 | msleep(500); |
1da177e4 LT |
6997 | goto repeat; |
6998 | } | |
6999 | } | |
7000 | } | |
61df9d91 | 7001 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
7002 | /* |
7003 | * this also signals 'finished resyncing' to md_stop | |
7004 | */ | |
7005 | out: | |
1da177e4 LT |
7006 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7007 | ||
7008 | /* tell personality that we are finished */ | |
57afd89f | 7009 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 7010 | |
dfc70645 | 7011 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7012 | mddev->curr_resync > 2) { |
7013 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7014 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7015 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7016 | printk(KERN_INFO | |
61df9d91 N |
7017 | "md: checkpointing %s of %s.\n", |
7018 | desc, mdname(mddev)); | |
5fd6c1dc N |
7019 | mddev->recovery_cp = mddev->curr_resync; |
7020 | } | |
7021 | } else | |
7022 | mddev->recovery_cp = MaxSector; | |
7023 | } else { | |
7024 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7025 | mddev->curr_resync = MaxSector; | |
4e59ca7d DW |
7026 | rcu_read_lock(); |
7027 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc | 7028 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7029 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7030 | !test_bit(Faulty, &rdev->flags) && |
7031 | !test_bit(In_sync, &rdev->flags) && | |
7032 | rdev->recovery_offset < mddev->curr_resync) | |
7033 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7034 | rcu_read_unlock(); |
5fd6c1dc | 7035 | } |
1da177e4 | 7036 | } |
17571284 | 7037 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7038 | |
1da177e4 | 7039 | skip: |
c07b70ad N |
7040 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7041 | /* We completed so min/max setting can be forgotten if used. */ | |
7042 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7043 | mddev->resync_min = 0; | |
7044 | mddev->resync_max = MaxSector; | |
7045 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7046 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 LT |
7047 | mddev->curr_resync = 0; |
7048 | wake_up(&resync_wait); | |
7049 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
7050 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
7051 | return; |
7052 | ||
7053 | interrupted: | |
7054 | /* | |
7055 | * got a signal, exit. | |
7056 | */ | |
7057 | printk(KERN_INFO | |
7058 | "md: md_do_sync() got signal ... exiting\n"); | |
7059 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7060 | goto out; | |
7061 | ||
1da177e4 | 7062 | } |
29269553 | 7063 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 7064 | |
b4c4c7b8 N |
7065 | static int remove_and_add_spares(mddev_t *mddev) |
7066 | { | |
7067 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
7068 | int spares = 0; |
7069 | ||
97e4f42d N |
7070 | mddev->curr_resync_completed = 0; |
7071 | ||
159ec1fc | 7072 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 7073 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 7074 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7075 | (test_bit(Faulty, &rdev->flags) || |
7076 | ! test_bit(In_sync, &rdev->flags)) && | |
7077 | atomic_read(&rdev->nr_pending)==0) { | |
7078 | if (mddev->pers->hot_remove_disk( | |
7079 | mddev, rdev->raid_disk)==0) { | |
7080 | char nm[20]; | |
7081 | sprintf(nm,"rd%d", rdev->raid_disk); | |
7082 | sysfs_remove_link(&mddev->kobj, nm); | |
7083 | rdev->raid_disk = -1; | |
7084 | } | |
7085 | } | |
7086 | ||
a8c42c7f | 7087 | if (mddev->degraded && !mddev->recovery_disabled) { |
159ec1fc | 7088 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 7089 | if (rdev->raid_disk >= 0 && |
e5427135 | 7090 | !test_bit(In_sync, &rdev->flags) && |
4274215d | 7091 | !test_bit(Faulty, &rdev->flags) && |
e5427135 | 7092 | !test_bit(Blocked, &rdev->flags)) |
dfc70645 | 7093 | spares++; |
b4c4c7b8 N |
7094 | if (rdev->raid_disk < 0 |
7095 | && !test_bit(Faulty, &rdev->flags)) { | |
7096 | rdev->recovery_offset = 0; | |
199050ea NB |
7097 | if (mddev->pers-> |
7098 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
7099 | char nm[20]; |
7100 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
7101 | if (sysfs_create_link(&mddev->kobj, |
7102 | &rdev->kobj, nm)) | |
00bcb4ac | 7103 | /* failure here is OK */; |
b4c4c7b8 N |
7104 | spares++; |
7105 | md_new_event(mddev); | |
93be75ff | 7106 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
b4c4c7b8 N |
7107 | } else |
7108 | break; | |
7109 | } | |
dfc70645 | 7110 | } |
b4c4c7b8 N |
7111 | } |
7112 | return spares; | |
7113 | } | |
7ebc0be7 N |
7114 | |
7115 | static void reap_sync_thread(mddev_t *mddev) | |
7116 | { | |
7117 | mdk_rdev_t *rdev; | |
7118 | ||
7119 | /* resync has finished, collect result */ | |
7120 | md_unregister_thread(mddev->sync_thread); | |
7121 | mddev->sync_thread = NULL; | |
7122 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
7123 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
7124 | /* success...*/ | |
7125 | /* activate any spares */ | |
7126 | if (mddev->pers->spare_active(mddev)) | |
7127 | sysfs_notify(&mddev->kobj, NULL, | |
7128 | "degraded"); | |
7129 | } | |
7130 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
7131 | mddev->pers->finish_reshape) | |
7132 | mddev->pers->finish_reshape(mddev); | |
7133 | md_update_sb(mddev, 1); | |
7134 | ||
7135 | /* if array is no-longer degraded, then any saved_raid_disk | |
7136 | * information must be scrapped | |
7137 | */ | |
7138 | if (!mddev->degraded) | |
7139 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
7140 | rdev->saved_raid_disk = -1; | |
7141 | ||
7142 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7143 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7144 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7145 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7146 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7147 | /* flag recovery needed just to double check */ | |
7148 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7149 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7150 | md_new_event(mddev); | |
7151 | } | |
7152 | ||
1da177e4 LT |
7153 | /* |
7154 | * This routine is regularly called by all per-raid-array threads to | |
7155 | * deal with generic issues like resync and super-block update. | |
7156 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7157 | * need this as they never do any recovery or update the superblock. | |
7158 | * | |
7159 | * It does not do any resync itself, but rather "forks" off other threads | |
7160 | * to do that as needed. | |
7161 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7162 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7163 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7164 | * and wakeups up this thread which will reap the thread and finish up. |
7165 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7166 | * | |
7167 | * The overall approach is: | |
7168 | * 1/ if the superblock needs updating, update it. | |
7169 | * 2/ If a recovery thread is running, don't do anything else. | |
7170 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7171 | * 4/ If there are any faulty devices, remove them. | |
7172 | * 5/ If array is degraded, try to add spares devices | |
7173 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7174 | */ | |
7175 | void md_check_recovery(mddev_t *mddev) | |
7176 | { | |
68866e42 JB |
7177 | if (mddev->suspended) |
7178 | return; | |
7179 | ||
5f40402d | 7180 | if (mddev->bitmap) |
aa5cbd10 | 7181 | bitmap_daemon_work(mddev); |
1da177e4 LT |
7182 | |
7183 | if (mddev->ro) | |
7184 | return; | |
fca4d848 N |
7185 | |
7186 | if (signal_pending(current)) { | |
31a59e34 | 7187 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7188 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7189 | mdname(mddev)); | |
7190 | mddev->safemode = 2; | |
7191 | } | |
7192 | flush_signals(current); | |
7193 | } | |
7194 | ||
c89a8eee N |
7195 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7196 | return; | |
1da177e4 | 7197 | if ( ! ( |
126925c0 | 7198 | (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7199 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7200 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7201 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7202 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7203 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7204 | )) |
7205 | return; | |
fca4d848 | 7206 | |
df5b89b3 | 7207 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7208 | int spares = 0; |
fca4d848 | 7209 | |
c89a8eee N |
7210 | if (mddev->ro) { |
7211 | /* Only thing we do on a ro array is remove | |
7212 | * failed devices. | |
7213 | */ | |
a8c42c7f N |
7214 | mdk_rdev_t *rdev; |
7215 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
7216 | if (rdev->raid_disk >= 0 && | |
7217 | !test_bit(Blocked, &rdev->flags) && | |
7218 | test_bit(Faulty, &rdev->flags) && | |
7219 | atomic_read(&rdev->nr_pending)==0) { | |
7220 | if (mddev->pers->hot_remove_disk( | |
7221 | mddev, rdev->raid_disk)==0) { | |
7222 | char nm[20]; | |
7223 | sprintf(nm,"rd%d", rdev->raid_disk); | |
7224 | sysfs_remove_link(&mddev->kobj, nm); | |
7225 | rdev->raid_disk = -1; | |
7226 | } | |
7227 | } | |
c89a8eee N |
7228 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7229 | goto unlock; | |
7230 | } | |
7231 | ||
31a59e34 | 7232 | if (!mddev->external) { |
0fd62b86 | 7233 | int did_change = 0; |
31a59e34 N |
7234 | spin_lock_irq(&mddev->write_lock); |
7235 | if (mddev->safemode && | |
7236 | !atomic_read(&mddev->writes_pending) && | |
7237 | !mddev->in_sync && | |
7238 | mddev->recovery_cp == MaxSector) { | |
7239 | mddev->in_sync = 1; | |
0fd62b86 | 7240 | did_change = 1; |
070dc6dd | 7241 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7242 | } |
7243 | if (mddev->safemode == 1) | |
7244 | mddev->safemode = 0; | |
7245 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7246 | if (did_change) |
00bcb4ac | 7247 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7248 | } |
fca4d848 | 7249 | |
850b2b42 N |
7250 | if (mddev->flags) |
7251 | md_update_sb(mddev, 0); | |
06d91a5f | 7252 | |
1da177e4 LT |
7253 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7254 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7255 | /* resync/recovery still happening */ | |
7256 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7257 | goto unlock; | |
7258 | } | |
7259 | if (mddev->sync_thread) { | |
7ebc0be7 | 7260 | reap_sync_thread(mddev); |
1da177e4 LT |
7261 | goto unlock; |
7262 | } | |
72a23c21 NB |
7263 | /* Set RUNNING before clearing NEEDED to avoid |
7264 | * any transients in the value of "sync_action". | |
7265 | */ | |
7266 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7267 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
7268 | /* Clear some bits that don't mean anything, but |
7269 | * might be left set | |
7270 | */ | |
24dd469d N |
7271 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7272 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7273 | |
5fd6c1dc N |
7274 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
7275 | goto unlock; | |
1da177e4 LT |
7276 | /* no recovery is running. |
7277 | * remove any failed drives, then | |
7278 | * add spares if possible. | |
7279 | * Spare are also removed and re-added, to allow | |
7280 | * the personality to fail the re-add. | |
7281 | */ | |
1da177e4 | 7282 | |
b4c4c7b8 | 7283 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7284 | if (mddev->pers->check_reshape == NULL || |
7285 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7286 | /* Cannot proceed */ |
7287 | goto unlock; | |
7288 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7289 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 7290 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
7291 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7292 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7293 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7294 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7295 | } else if (mddev->recovery_cp < MaxSector) { |
7296 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7297 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7298 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7299 | /* nothing to be done ... */ | |
1da177e4 | 7300 | goto unlock; |
24dd469d | 7301 | |
1da177e4 | 7302 | if (mddev->pers->sync_request) { |
a654b9d8 N |
7303 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
7304 | /* We are adding a device or devices to an array | |
7305 | * which has the bitmap stored on all devices. | |
7306 | * So make sure all bitmap pages get written | |
7307 | */ | |
7308 | bitmap_write_all(mddev->bitmap); | |
7309 | } | |
1da177e4 LT |
7310 | mddev->sync_thread = md_register_thread(md_do_sync, |
7311 | mddev, | |
0da3c619 | 7312 | "resync"); |
1da177e4 LT |
7313 | if (!mddev->sync_thread) { |
7314 | printk(KERN_ERR "%s: could not start resync" | |
7315 | " thread...\n", | |
7316 | mdname(mddev)); | |
7317 | /* leave the spares where they are, it shouldn't hurt */ | |
7ebc0be7 N |
7318 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
7319 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7320 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7321 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7322 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
d7603b7e | 7323 | } else |
1da177e4 | 7324 | md_wakeup_thread(mddev->sync_thread); |
00bcb4ac | 7325 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7326 | md_new_event(mddev); |
1da177e4 LT |
7327 | } |
7328 | unlock: | |
72a23c21 NB |
7329 | if (!mddev->sync_thread) { |
7330 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7331 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7332 | &mddev->recovery)) | |
0c3573f1 | 7333 | if (mddev->sysfs_action) |
00bcb4ac | 7334 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7335 | } |
1da177e4 LT |
7336 | mddev_unlock(mddev); |
7337 | } | |
7338 | } | |
7339 | ||
6bfe0b49 DW |
7340 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
7341 | { | |
00bcb4ac | 7342 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 DW |
7343 | wait_event_timeout(rdev->blocked_wait, |
7344 | !test_bit(Blocked, &rdev->flags), | |
7345 | msecs_to_jiffies(5000)); | |
7346 | rdev_dec_pending(rdev, mddev); | |
7347 | } | |
7348 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7349 | ||
75c96f85 AB |
7350 | static int md_notify_reboot(struct notifier_block *this, |
7351 | unsigned long code, void *x) | |
1da177e4 LT |
7352 | { |
7353 | struct list_head *tmp; | |
7354 | mddev_t *mddev; | |
7355 | ||
7356 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
7357 | ||
7358 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
7359 | ||
29ac4aa3 | 7360 | for_each_mddev(mddev, tmp) |
c71d4887 | 7361 | if (mddev_trylock(mddev)) { |
2b25000b N |
7362 | /* Force a switch to readonly even array |
7363 | * appears to still be in use. Hence | |
7364 | * the '100'. | |
7365 | */ | |
a4bd82d0 | 7366 | md_set_readonly(mddev, 100); |
c71d4887 NB |
7367 | mddev_unlock(mddev); |
7368 | } | |
1da177e4 LT |
7369 | /* |
7370 | * certain more exotic SCSI devices are known to be | |
7371 | * volatile wrt too early system reboots. While the | |
7372 | * right place to handle this issue is the given | |
7373 | * driver, we do want to have a safe RAID driver ... | |
7374 | */ | |
7375 | mdelay(1000*1); | |
7376 | } | |
7377 | return NOTIFY_DONE; | |
7378 | } | |
7379 | ||
75c96f85 | 7380 | static struct notifier_block md_notifier = { |
1da177e4 LT |
7381 | .notifier_call = md_notify_reboot, |
7382 | .next = NULL, | |
7383 | .priority = INT_MAX, /* before any real devices */ | |
7384 | }; | |
7385 | ||
7386 | static void md_geninit(void) | |
7387 | { | |
1da177e4 LT |
7388 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
7389 | ||
c7705f34 | 7390 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
7391 | } |
7392 | ||
75c96f85 | 7393 | static int __init md_init(void) |
1da177e4 | 7394 | { |
e804ac78 TH |
7395 | int ret = -ENOMEM; |
7396 | ||
ada609ee | 7397 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
7398 | if (!md_wq) |
7399 | goto err_wq; | |
7400 | ||
7401 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
7402 | if (!md_misc_wq) | |
7403 | goto err_misc_wq; | |
7404 | ||
7405 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
7406 | goto err_md; | |
7407 | ||
7408 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
7409 | goto err_mdp; | |
7410 | mdp_major = ret; | |
7411 | ||
3dbd8c2e | 7412 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
7413 | md_probe, NULL, NULL); |
7414 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
7415 | md_probe, NULL, NULL); |
7416 | ||
1da177e4 | 7417 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 7418 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
7419 | |
7420 | md_geninit(); | |
d710e138 | 7421 | return 0; |
1da177e4 | 7422 | |
e804ac78 TH |
7423 | err_mdp: |
7424 | unregister_blkdev(MD_MAJOR, "md"); | |
7425 | err_md: | |
7426 | destroy_workqueue(md_misc_wq); | |
7427 | err_misc_wq: | |
7428 | destroy_workqueue(md_wq); | |
7429 | err_wq: | |
7430 | return ret; | |
7431 | } | |
1da177e4 LT |
7432 | |
7433 | #ifndef MODULE | |
7434 | ||
7435 | /* | |
7436 | * Searches all registered partitions for autorun RAID arrays | |
7437 | * at boot time. | |
7438 | */ | |
4d936ec1 ME |
7439 | |
7440 | static LIST_HEAD(all_detected_devices); | |
7441 | struct detected_devices_node { | |
7442 | struct list_head list; | |
7443 | dev_t dev; | |
7444 | }; | |
1da177e4 LT |
7445 | |
7446 | void md_autodetect_dev(dev_t dev) | |
7447 | { | |
4d936ec1 ME |
7448 | struct detected_devices_node *node_detected_dev; |
7449 | ||
7450 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
7451 | if (node_detected_dev) { | |
7452 | node_detected_dev->dev = dev; | |
7453 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
7454 | } else { | |
7455 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
7456 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
7457 | } | |
1da177e4 LT |
7458 | } |
7459 | ||
7460 | ||
7461 | static void autostart_arrays(int part) | |
7462 | { | |
7463 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
7464 | struct detected_devices_node *node_detected_dev; |
7465 | dev_t dev; | |
7466 | int i_scanned, i_passed; | |
1da177e4 | 7467 | |
4d936ec1 ME |
7468 | i_scanned = 0; |
7469 | i_passed = 0; | |
1da177e4 | 7470 | |
4d936ec1 | 7471 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 7472 | |
4d936ec1 ME |
7473 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
7474 | i_scanned++; | |
7475 | node_detected_dev = list_entry(all_detected_devices.next, | |
7476 | struct detected_devices_node, list); | |
7477 | list_del(&node_detected_dev->list); | |
7478 | dev = node_detected_dev->dev; | |
7479 | kfree(node_detected_dev); | |
df968c4e | 7480 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
7481 | if (IS_ERR(rdev)) |
7482 | continue; | |
7483 | ||
b2d444d7 | 7484 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7485 | MD_BUG(); |
7486 | continue; | |
7487 | } | |
d0fae18f | 7488 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 7489 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 7490 | i_passed++; |
1da177e4 | 7491 | } |
4d936ec1 ME |
7492 | |
7493 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
7494 | i_scanned, i_passed); | |
1da177e4 LT |
7495 | |
7496 | autorun_devices(part); | |
7497 | } | |
7498 | ||
fdee8ae4 | 7499 | #endif /* !MODULE */ |
1da177e4 LT |
7500 | |
7501 | static __exit void md_exit(void) | |
7502 | { | |
7503 | mddev_t *mddev; | |
7504 | struct list_head *tmp; | |
8ab5e4c1 | 7505 | |
3dbd8c2e | 7506 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 7507 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 7508 | |
3dbd8c2e | 7509 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
7510 | unregister_blkdev(mdp_major, "mdp"); |
7511 | unregister_reboot_notifier(&md_notifier); | |
7512 | unregister_sysctl_table(raid_table_header); | |
7513 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 7514 | for_each_mddev(mddev, tmp) { |
1da177e4 | 7515 | export_array(mddev); |
d3374825 | 7516 | mddev->hold_active = 0; |
1da177e4 | 7517 | } |
e804ac78 TH |
7518 | destroy_workqueue(md_misc_wq); |
7519 | destroy_workqueue(md_wq); | |
1da177e4 LT |
7520 | } |
7521 | ||
685784aa | 7522 | subsys_initcall(md_init); |
1da177e4 LT |
7523 | module_exit(md_exit) |
7524 | ||
f91de92e N |
7525 | static int get_ro(char *buffer, struct kernel_param *kp) |
7526 | { | |
7527 | return sprintf(buffer, "%d", start_readonly); | |
7528 | } | |
7529 | static int set_ro(const char *val, struct kernel_param *kp) | |
7530 | { | |
7531 | char *e; | |
7532 | int num = simple_strtoul(val, &e, 10); | |
7533 | if (*val && (*e == '\0' || *e == '\n')) { | |
7534 | start_readonly = num; | |
4dbcdc75 | 7535 | return 0; |
f91de92e N |
7536 | } |
7537 | return -EINVAL; | |
7538 | } | |
7539 | ||
80ca3a44 N |
7540 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
7541 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 7542 | |
efeb53c0 | 7543 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 7544 | |
1da177e4 LT |
7545 | EXPORT_SYMBOL(register_md_personality); |
7546 | EXPORT_SYMBOL(unregister_md_personality); | |
7547 | EXPORT_SYMBOL(md_error); | |
7548 | EXPORT_SYMBOL(md_done_sync); | |
7549 | EXPORT_SYMBOL(md_write_start); | |
7550 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
7551 | EXPORT_SYMBOL(md_register_thread); |
7552 | EXPORT_SYMBOL(md_unregister_thread); | |
7553 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
7554 | EXPORT_SYMBOL(md_check_recovery); |
7555 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7556 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 7557 | MODULE_ALIAS("md"); |
72008652 | 7558 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |