]>
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); | |
1da177e4 | 354 | } |
390ee602 | 355 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 356 | |
3fa841d7 N |
357 | int mddev_congested(mddev_t *mddev, int bits) |
358 | { | |
359 | return mddev->suspended; | |
360 | } | |
361 | EXPORT_SYMBOL(mddev_congested); | |
362 | ||
a2826aa9 | 363 | /* |
e9c7469b | 364 | * Generic flush handling for md |
a2826aa9 N |
365 | */ |
366 | ||
e9c7469b | 367 | static void md_end_flush(struct bio *bio, int err) |
a2826aa9 N |
368 | { |
369 | mdk_rdev_t *rdev = bio->bi_private; | |
370 | mddev_t *mddev = rdev->mddev; | |
a2826aa9 N |
371 | |
372 | rdev_dec_pending(rdev, mddev); | |
373 | ||
374 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 375 | /* The pre-request flush has finished */ |
e804ac78 | 376 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
377 | } |
378 | bio_put(bio); | |
379 | } | |
380 | ||
a7a07e69 N |
381 | static void md_submit_flush_data(struct work_struct *ws); |
382 | ||
a035fc3e | 383 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 384 | { |
a035fc3e | 385 | mddev_t *mddev = container_of(ws, mddev_t, flush_work); |
a2826aa9 N |
386 | mdk_rdev_t *rdev; |
387 | ||
a7a07e69 N |
388 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
389 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 N |
390 | rcu_read_lock(); |
391 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
392 | if (rdev->raid_disk >= 0 && | |
393 | !test_bit(Faulty, &rdev->flags)) { | |
394 | /* Take two references, one is dropped | |
395 | * when request finishes, one after | |
396 | * we reclaim rcu_read_lock | |
397 | */ | |
398 | struct bio *bi; | |
399 | atomic_inc(&rdev->nr_pending); | |
400 | atomic_inc(&rdev->nr_pending); | |
401 | rcu_read_unlock(); | |
a167f663 | 402 | bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev); |
e9c7469b | 403 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
404 | bi->bi_private = rdev; |
405 | bi->bi_bdev = rdev->bdev; | |
406 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 407 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
408 | rcu_read_lock(); |
409 | rdev_dec_pending(rdev, mddev); | |
410 | } | |
411 | rcu_read_unlock(); | |
a7a07e69 N |
412 | if (atomic_dec_and_test(&mddev->flush_pending)) |
413 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
414 | } |
415 | ||
e9c7469b | 416 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 417 | { |
e9c7469b TH |
418 | mddev_t *mddev = container_of(ws, mddev_t, flush_work); |
419 | struct bio *bio = mddev->flush_bio; | |
a2826aa9 | 420 | |
e9c7469b | 421 | if (bio->bi_size == 0) |
a2826aa9 N |
422 | /* an empty barrier - all done */ |
423 | bio_endio(bio, 0); | |
424 | else { | |
e9c7469b | 425 | bio->bi_rw &= ~REQ_FLUSH; |
21a52c6d | 426 | if (mddev->pers->make_request(mddev, bio)) |
a2826aa9 | 427 | generic_make_request(bio); |
a2826aa9 | 428 | } |
2b74e12e N |
429 | |
430 | mddev->flush_bio = NULL; | |
431 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
432 | } |
433 | ||
e9c7469b | 434 | void md_flush_request(mddev_t *mddev, struct bio *bio) |
a2826aa9 N |
435 | { |
436 | spin_lock_irq(&mddev->write_lock); | |
437 | wait_event_lock_irq(mddev->sb_wait, | |
e9c7469b | 438 | !mddev->flush_bio, |
a2826aa9 | 439 | mddev->write_lock, /*nothing*/); |
e9c7469b | 440 | mddev->flush_bio = bio; |
a2826aa9 N |
441 | spin_unlock_irq(&mddev->write_lock); |
442 | ||
a035fc3e N |
443 | INIT_WORK(&mddev->flush_work, submit_flushes); |
444 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 445 | } |
e9c7469b | 446 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 447 | |
97658cdd N |
448 | /* Support for plugging. |
449 | * This mirrors the plugging support in request_queue, but does not | |
450 | * require having a whole queue or request structures. | |
451 | * We allocate an md_plug_cb for each md device and each thread it gets | |
452 | * plugged on. This links tot the private plug_handle structure in the | |
453 | * personality data where we keep a count of the number of outstanding | |
454 | * plugs so other code can see if a plug is active. | |
455 | */ | |
456 | struct md_plug_cb { | |
457 | struct blk_plug_cb cb; | |
458 | mddev_t *mddev; | |
459 | }; | |
460 | ||
461 | static void plugger_unplug(struct blk_plug_cb *cb) | |
462 | { | |
463 | struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb); | |
464 | if (atomic_dec_and_test(&mdcb->mddev->plug_cnt)) | |
465 | md_wakeup_thread(mdcb->mddev->thread); | |
466 | kfree(mdcb); | |
467 | } | |
468 | ||
469 | /* Check that an unplug wakeup will come shortly. | |
470 | * If not, wakeup the md thread immediately | |
471 | */ | |
472 | int mddev_check_plugged(mddev_t *mddev) | |
473 | { | |
474 | struct blk_plug *plug = current->plug; | |
475 | struct md_plug_cb *mdcb; | |
476 | ||
477 | if (!plug) | |
478 | return 0; | |
479 | ||
480 | list_for_each_entry(mdcb, &plug->cb_list, cb.list) { | |
481 | if (mdcb->cb.callback == plugger_unplug && | |
482 | mdcb->mddev == mddev) { | |
483 | /* Already on the list, move to top */ | |
484 | if (mdcb != list_first_entry(&plug->cb_list, | |
485 | struct md_plug_cb, | |
486 | cb.list)) | |
487 | list_move(&mdcb->cb.list, &plug->cb_list); | |
488 | return 1; | |
489 | } | |
490 | } | |
491 | /* Not currently on the callback list */ | |
492 | mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC); | |
493 | if (!mdcb) | |
494 | return 0; | |
495 | ||
496 | mdcb->mddev = mddev; | |
497 | mdcb->cb.callback = plugger_unplug; | |
498 | atomic_inc(&mddev->plug_cnt); | |
499 | list_add(&mdcb->cb.list, &plug->cb_list); | |
500 | return 1; | |
501 | } | |
502 | EXPORT_SYMBOL_GPL(mddev_check_plugged); | |
2ac87401 | 503 | |
1da177e4 LT |
504 | static inline mddev_t *mddev_get(mddev_t *mddev) |
505 | { | |
506 | atomic_inc(&mddev->active); | |
507 | return mddev; | |
508 | } | |
509 | ||
5fd3a17e | 510 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 511 | |
1da177e4 LT |
512 | static void mddev_put(mddev_t *mddev) |
513 | { | |
a167f663 N |
514 | struct bio_set *bs = NULL; |
515 | ||
1da177e4 LT |
516 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
517 | return; | |
d3374825 | 518 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
519 | mddev->ctime == 0 && !mddev->hold_active) { |
520 | /* Array is not configured at all, and not held active, | |
521 | * so destroy it */ | |
1da177e4 | 522 | list_del(&mddev->all_mddevs); |
a167f663 N |
523 | bs = mddev->bio_set; |
524 | mddev->bio_set = NULL; | |
d3374825 | 525 | if (mddev->gendisk) { |
e804ac78 TH |
526 | /* We did a probe so need to clean up. Call |
527 | * queue_work inside the spinlock so that | |
528 | * flush_workqueue() after mddev_find will | |
529 | * succeed in waiting for the work to be done. | |
d3374825 N |
530 | */ |
531 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 532 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
533 | } else |
534 | kfree(mddev); | |
535 | } | |
536 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
537 | if (bs) |
538 | bioset_free(bs); | |
1da177e4 LT |
539 | } |
540 | ||
390ee602 | 541 | void mddev_init(mddev_t *mddev) |
fafd7fb0 N |
542 | { |
543 | mutex_init(&mddev->open_mutex); | |
544 | mutex_init(&mddev->reconfig_mutex); | |
545 | mutex_init(&mddev->bitmap_info.mutex); | |
546 | INIT_LIST_HEAD(&mddev->disks); | |
547 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
548 | init_timer(&mddev->safemode_timer); | |
549 | atomic_set(&mddev->active, 1); | |
550 | atomic_set(&mddev->openers, 0); | |
551 | atomic_set(&mddev->active_io, 0); | |
97658cdd | 552 | atomic_set(&mddev->plug_cnt, 0); |
fafd7fb0 N |
553 | spin_lock_init(&mddev->write_lock); |
554 | atomic_set(&mddev->flush_pending, 0); | |
555 | init_waitqueue_head(&mddev->sb_wait); | |
556 | init_waitqueue_head(&mddev->recovery_wait); | |
557 | mddev->reshape_position = MaxSector; | |
558 | mddev->resync_min = 0; | |
559 | mddev->resync_max = MaxSector; | |
560 | mddev->level = LEVEL_NONE; | |
561 | } | |
390ee602 | 562 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 563 | |
1da177e4 LT |
564 | static mddev_t * mddev_find(dev_t unit) |
565 | { | |
566 | mddev_t *mddev, *new = NULL; | |
567 | ||
8f5f02c4 N |
568 | if (unit && MAJOR(unit) != MD_MAJOR) |
569 | unit &= ~((1<<MdpMinorShift)-1); | |
570 | ||
1da177e4 LT |
571 | retry: |
572 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
573 | |
574 | if (unit) { | |
575 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
576 | if (mddev->unit == unit) { | |
577 | mddev_get(mddev); | |
578 | spin_unlock(&all_mddevs_lock); | |
579 | kfree(new); | |
580 | return mddev; | |
581 | } | |
582 | ||
583 | if (new) { | |
584 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 585 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
586 | new->hold_active = UNTIL_IOCTL; |
587 | return new; | |
1da177e4 | 588 | } |
efeb53c0 N |
589 | } else if (new) { |
590 | /* find an unused unit number */ | |
591 | static int next_minor = 512; | |
592 | int start = next_minor; | |
593 | int is_free = 0; | |
594 | int dev = 0; | |
595 | while (!is_free) { | |
596 | dev = MKDEV(MD_MAJOR, next_minor); | |
597 | next_minor++; | |
598 | if (next_minor > MINORMASK) | |
599 | next_minor = 0; | |
600 | if (next_minor == start) { | |
601 | /* Oh dear, all in use. */ | |
602 | spin_unlock(&all_mddevs_lock); | |
603 | kfree(new); | |
604 | return NULL; | |
605 | } | |
606 | ||
607 | is_free = 1; | |
608 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
609 | if (mddev->unit == dev) { | |
610 | is_free = 0; | |
611 | break; | |
612 | } | |
613 | } | |
614 | new->unit = dev; | |
615 | new->md_minor = MINOR(dev); | |
616 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
617 | list_add(&new->all_mddevs, &all_mddevs); |
618 | spin_unlock(&all_mddevs_lock); | |
619 | return new; | |
620 | } | |
621 | spin_unlock(&all_mddevs_lock); | |
622 | ||
9ffae0cf | 623 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
624 | if (!new) |
625 | return NULL; | |
626 | ||
1da177e4 LT |
627 | new->unit = unit; |
628 | if (MAJOR(unit) == MD_MAJOR) | |
629 | new->md_minor = MINOR(unit); | |
630 | else | |
631 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
632 | ||
fafd7fb0 | 633 | mddev_init(new); |
1da177e4 | 634 | |
1da177e4 LT |
635 | goto retry; |
636 | } | |
637 | ||
638 | static inline int mddev_lock(mddev_t * mddev) | |
639 | { | |
df5b89b3 | 640 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
641 | } |
642 | ||
b522adcd DW |
643 | static inline int mddev_is_locked(mddev_t *mddev) |
644 | { | |
645 | return mutex_is_locked(&mddev->reconfig_mutex); | |
646 | } | |
647 | ||
1da177e4 LT |
648 | static inline int mddev_trylock(mddev_t * mddev) |
649 | { | |
df5b89b3 | 650 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
651 | } |
652 | ||
b6eb127d N |
653 | static struct attribute_group md_redundancy_group; |
654 | ||
a64c876f | 655 | static void mddev_unlock(mddev_t * mddev) |
1da177e4 | 656 | { |
a64c876f | 657 | if (mddev->to_remove) { |
b6eb127d N |
658 | /* These cannot be removed under reconfig_mutex as |
659 | * an access to the files will try to take reconfig_mutex | |
660 | * while holding the file unremovable, which leads to | |
661 | * a deadlock. | |
bb4f1e9d N |
662 | * So hold set sysfs_active while the remove in happeing, |
663 | * and anything else which might set ->to_remove or my | |
664 | * otherwise change the sysfs namespace will fail with | |
665 | * -EBUSY if sysfs_active is still set. | |
666 | * We set sysfs_active under reconfig_mutex and elsewhere | |
667 | * test it under the same mutex to ensure its correct value | |
668 | * is seen. | |
b6eb127d | 669 | */ |
a64c876f N |
670 | struct attribute_group *to_remove = mddev->to_remove; |
671 | mddev->to_remove = NULL; | |
bb4f1e9d | 672 | mddev->sysfs_active = 1; |
b6eb127d N |
673 | mutex_unlock(&mddev->reconfig_mutex); |
674 | ||
00bcb4ac N |
675 | if (mddev->kobj.sd) { |
676 | if (to_remove != &md_redundancy_group) | |
677 | sysfs_remove_group(&mddev->kobj, to_remove); | |
678 | if (mddev->pers == NULL || | |
679 | mddev->pers->sync_request == NULL) { | |
680 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
681 | if (mddev->sysfs_action) | |
682 | sysfs_put(mddev->sysfs_action); | |
683 | mddev->sysfs_action = NULL; | |
684 | } | |
a64c876f | 685 | } |
bb4f1e9d | 686 | mddev->sysfs_active = 0; |
b6eb127d N |
687 | } else |
688 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 689 | |
005eca5e | 690 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
691 | } |
692 | ||
2989ddbd | 693 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 694 | { |
159ec1fc | 695 | mdk_rdev_t *rdev; |
1da177e4 | 696 | |
159ec1fc | 697 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
698 | if (rdev->desc_nr == nr) |
699 | return rdev; | |
159ec1fc | 700 | |
1da177e4 LT |
701 | return NULL; |
702 | } | |
703 | ||
704 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
705 | { | |
1da177e4 LT |
706 | mdk_rdev_t *rdev; |
707 | ||
159ec1fc | 708 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
709 | if (rdev->bdev->bd_dev == dev) |
710 | return rdev; | |
159ec1fc | 711 | |
1da177e4 LT |
712 | return NULL; |
713 | } | |
714 | ||
d9d166c2 | 715 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
716 | { |
717 | struct mdk_personality *pers; | |
d9d166c2 N |
718 | list_for_each_entry(pers, &pers_list, list) { |
719 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 720 | return pers; |
d9d166c2 N |
721 | if (strcmp(pers->name, clevel)==0) |
722 | return pers; | |
723 | } | |
2604b703 N |
724 | return NULL; |
725 | } | |
726 | ||
b73df2d3 | 727 | /* return the offset of the super block in 512byte sectors */ |
57b2caa3 | 728 | static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev) |
1da177e4 | 729 | { |
57b2caa3 | 730 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 731 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
732 | } |
733 | ||
1da177e4 LT |
734 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
735 | { | |
736 | if (rdev->sb_page) | |
737 | MD_BUG(); | |
738 | ||
739 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
740 | if (!rdev->sb_page) { | |
741 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 742 | return -ENOMEM; |
1da177e4 LT |
743 | } |
744 | ||
745 | return 0; | |
746 | } | |
747 | ||
748 | static void free_disk_sb(mdk_rdev_t * rdev) | |
749 | { | |
750 | if (rdev->sb_page) { | |
2d1f3b5d | 751 | put_page(rdev->sb_page); |
1da177e4 LT |
752 | rdev->sb_loaded = 0; |
753 | rdev->sb_page = NULL; | |
0f420358 | 754 | rdev->sb_start = 0; |
dd8ac336 | 755 | rdev->sectors = 0; |
1da177e4 LT |
756 | } |
757 | } | |
758 | ||
759 | ||
6712ecf8 | 760 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
761 | { |
762 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 763 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 764 | |
3a0f5bbb N |
765 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
766 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
767 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
768 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 769 | md_error(mddev, rdev); |
3a0f5bbb | 770 | } |
7bfa19f2 | 771 | |
a9701a30 N |
772 | if (atomic_dec_and_test(&mddev->pending_writes)) |
773 | wake_up(&mddev->sb_wait); | |
f8b58edf | 774 | bio_put(bio); |
7bfa19f2 N |
775 | } |
776 | ||
777 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, | |
778 | sector_t sector, int size, struct page *page) | |
779 | { | |
780 | /* write first size bytes of page to sector of rdev | |
781 | * Increment mddev->pending_writes before returning | |
782 | * and decrement it on completion, waking up sb_wait | |
783 | * if zero is reached. | |
784 | * If an error occurred, call md_error | |
785 | */ | |
a167f663 | 786 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 787 | |
a6ff7e08 | 788 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
7bfa19f2 N |
789 | bio->bi_sector = sector; |
790 | bio_add_page(bio, page, size, 0); | |
791 | bio->bi_private = rdev; | |
792 | bio->bi_end_io = super_written; | |
a9701a30 | 793 | |
7bfa19f2 | 794 | atomic_inc(&mddev->pending_writes); |
721a9602 | 795 | submit_bio(REQ_WRITE | REQ_SYNC | REQ_FLUSH | REQ_FUA, bio); |
a9701a30 N |
796 | } |
797 | ||
798 | void md_super_wait(mddev_t *mddev) | |
799 | { | |
e9c7469b | 800 | /* wait for all superblock writes that were scheduled to complete */ |
a9701a30 N |
801 | DEFINE_WAIT(wq); |
802 | for(;;) { | |
803 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
804 | if (atomic_read(&mddev->pending_writes)==0) | |
805 | break; | |
a9701a30 N |
806 | schedule(); |
807 | } | |
808 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
809 | } |
810 | ||
6712ecf8 | 811 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 812 | { |
1da177e4 | 813 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
814 | } |
815 | ||
2b193363 | 816 | int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size, |
ccebd4c4 | 817 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 818 | { |
a167f663 | 819 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
820 | struct completion event; |
821 | int ret; | |
822 | ||
721a9602 | 823 | rw |= REQ_SYNC; |
1da177e4 | 824 | |
a6ff7e08 JB |
825 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
826 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 JB |
827 | if (metadata_op) |
828 | bio->bi_sector = sector + rdev->sb_start; | |
829 | else | |
830 | bio->bi_sector = sector + rdev->data_offset; | |
1da177e4 LT |
831 | bio_add_page(bio, page, size, 0); |
832 | init_completion(&event); | |
833 | bio->bi_private = &event; | |
834 | bio->bi_end_io = bi_complete; | |
835 | submit_bio(rw, bio); | |
836 | wait_for_completion(&event); | |
837 | ||
838 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
839 | bio_put(bio); | |
840 | return ret; | |
841 | } | |
a8745db2 | 842 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 843 | |
0002b271 | 844 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
845 | { |
846 | char b[BDEVNAME_SIZE]; | |
847 | if (!rdev->sb_page) { | |
848 | MD_BUG(); | |
849 | return -EINVAL; | |
850 | } | |
851 | if (rdev->sb_loaded) | |
852 | return 0; | |
853 | ||
854 | ||
ccebd4c4 | 855 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
856 | goto fail; |
857 | rdev->sb_loaded = 1; | |
858 | return 0; | |
859 | ||
860 | fail: | |
861 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
862 | bdevname(rdev->bdev,b)); | |
863 | return -EINVAL; | |
864 | } | |
865 | ||
866 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
867 | { | |
05710466 AN |
868 | return sb1->set_uuid0 == sb2->set_uuid0 && |
869 | sb1->set_uuid1 == sb2->set_uuid1 && | |
870 | sb1->set_uuid2 == sb2->set_uuid2 && | |
871 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
872 | } |
873 | ||
1da177e4 LT |
874 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
875 | { | |
876 | int ret; | |
877 | mdp_super_t *tmp1, *tmp2; | |
878 | ||
879 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
880 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
881 | ||
882 | if (!tmp1 || !tmp2) { | |
883 | ret = 0; | |
35020f1a | 884 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
885 | goto abort; |
886 | } | |
887 | ||
888 | *tmp1 = *sb1; | |
889 | *tmp2 = *sb2; | |
890 | ||
891 | /* | |
892 | * nr_disks is not constant | |
893 | */ | |
894 | tmp1->nr_disks = 0; | |
895 | tmp2->nr_disks = 0; | |
896 | ||
ce0c8e05 | 897 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 898 | abort: |
990a8baf JJ |
899 | kfree(tmp1); |
900 | kfree(tmp2); | |
1da177e4 LT |
901 | return ret; |
902 | } | |
903 | ||
4d167f09 N |
904 | |
905 | static u32 md_csum_fold(u32 csum) | |
906 | { | |
907 | csum = (csum & 0xffff) + (csum >> 16); | |
908 | return (csum & 0xffff) + (csum >> 16); | |
909 | } | |
910 | ||
1da177e4 LT |
911 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
912 | { | |
4d167f09 N |
913 | u64 newcsum = 0; |
914 | u32 *sb32 = (u32*)sb; | |
915 | int i; | |
1da177e4 LT |
916 | unsigned int disk_csum, csum; |
917 | ||
918 | disk_csum = sb->sb_csum; | |
919 | sb->sb_csum = 0; | |
4d167f09 N |
920 | |
921 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
922 | newcsum += sb32[i]; | |
923 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
924 | ||
925 | ||
926 | #ifdef CONFIG_ALPHA | |
927 | /* This used to use csum_partial, which was wrong for several | |
928 | * reasons including that different results are returned on | |
929 | * different architectures. It isn't critical that we get exactly | |
930 | * the same return value as before (we always csum_fold before | |
931 | * testing, and that removes any differences). However as we | |
932 | * know that csum_partial always returned a 16bit value on | |
933 | * alphas, do a fold to maximise conformity to previous behaviour. | |
934 | */ | |
935 | sb->sb_csum = md_csum_fold(disk_csum); | |
936 | #else | |
1da177e4 | 937 | sb->sb_csum = disk_csum; |
4d167f09 | 938 | #endif |
1da177e4 LT |
939 | return csum; |
940 | } | |
941 | ||
942 | ||
943 | /* | |
944 | * Handle superblock details. | |
945 | * We want to be able to handle multiple superblock formats | |
946 | * so we have a common interface to them all, and an array of | |
947 | * different handlers. | |
948 | * We rely on user-space to write the initial superblock, and support | |
949 | * reading and updating of superblocks. | |
950 | * Interface methods are: | |
951 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
952 | * loads and validates a superblock on dev. | |
953 | * if refdev != NULL, compare superblocks on both devices | |
954 | * Return: | |
955 | * 0 - dev has a superblock that is compatible with refdev | |
956 | * 1 - dev has a superblock that is compatible and newer than refdev | |
957 | * so dev should be used as the refdev in future | |
958 | * -EINVAL superblock incompatible or invalid | |
959 | * -othererror e.g. -EIO | |
960 | * | |
961 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
962 | * Verify that dev is acceptable into mddev. | |
963 | * The first time, mddev->raid_disks will be 0, and data from | |
964 | * dev should be merged in. Subsequent calls check that dev | |
965 | * is new enough. Return 0 or -EINVAL | |
966 | * | |
967 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
968 | * Update the superblock for rdev with data in mddev | |
969 | * This does not write to disc. | |
970 | * | |
971 | */ | |
972 | ||
973 | struct super_type { | |
0cd17fec CW |
974 | char *name; |
975 | struct module *owner; | |
976 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
977 | int minor_version); | |
978 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
979 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
980 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 981 | sector_t num_sectors); |
1da177e4 LT |
982 | }; |
983 | ||
0894cc30 AN |
984 | /* |
985 | * Check that the given mddev has no bitmap. | |
986 | * | |
987 | * This function is called from the run method of all personalities that do not | |
988 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
989 | * has a bitmap. Otherwise, it returns 0. | |
990 | * | |
991 | */ | |
992 | int md_check_no_bitmap(mddev_t *mddev) | |
993 | { | |
c3d9714e | 994 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
995 | return 0; |
996 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
997 | mdname(mddev), mddev->pers->name); | |
998 | return 1; | |
999 | } | |
1000 | EXPORT_SYMBOL(md_check_no_bitmap); | |
1001 | ||
1da177e4 LT |
1002 | /* |
1003 | * load_super for 0.90.0 | |
1004 | */ | |
1005 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1006 | { | |
1007 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
1008 | mdp_super_t *sb; | |
1009 | int ret; | |
1da177e4 LT |
1010 | |
1011 | /* | |
0f420358 | 1012 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
1013 | * it's at the end of the disk. |
1014 | * | |
1015 | * It also happens to be a multiple of 4Kb. | |
1016 | */ | |
57b2caa3 | 1017 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 1018 | |
0002b271 | 1019 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
1020 | if (ret) return ret; |
1021 | ||
1022 | ret = -EINVAL; | |
1023 | ||
1024 | bdevname(rdev->bdev, b); | |
1025 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
1026 | ||
1027 | if (sb->md_magic != MD_SB_MAGIC) { | |
1028 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
1029 | b); | |
1030 | goto abort; | |
1031 | } | |
1032 | ||
1033 | if (sb->major_version != 0 || | |
f6705578 N |
1034 | sb->minor_version < 90 || |
1035 | sb->minor_version > 91) { | |
1da177e4 LT |
1036 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
1037 | sb->major_version, sb->minor_version, | |
1038 | b); | |
1039 | goto abort; | |
1040 | } | |
1041 | ||
1042 | if (sb->raid_disks <= 0) | |
1043 | goto abort; | |
1044 | ||
4d167f09 | 1045 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
1046 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
1047 | b); | |
1048 | goto abort; | |
1049 | } | |
1050 | ||
1051 | rdev->preferred_minor = sb->md_minor; | |
1052 | rdev->data_offset = 0; | |
0002b271 | 1053 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
1054 | |
1055 | if (sb->level == LEVEL_MULTIPATH) | |
1056 | rdev->desc_nr = -1; | |
1057 | else | |
1058 | rdev->desc_nr = sb->this_disk.number; | |
1059 | ||
9a7b2b0f | 1060 | if (!refdev) { |
1da177e4 | 1061 | ret = 1; |
9a7b2b0f | 1062 | } else { |
1da177e4 LT |
1063 | __u64 ev1, ev2; |
1064 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
1065 | if (!uuid_equal(refsb, sb)) { | |
1066 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1067 | b, bdevname(refdev->bdev,b2)); | |
1068 | goto abort; | |
1069 | } | |
1070 | if (!sb_equal(refsb, sb)) { | |
1071 | printk(KERN_WARNING "md: %s has same UUID" | |
1072 | " but different superblock to %s\n", | |
1073 | b, bdevname(refdev->bdev, b2)); | |
1074 | goto abort; | |
1075 | } | |
1076 | ev1 = md_event(sb); | |
1077 | ev2 = md_event(refsb); | |
1078 | if (ev1 > ev2) | |
1079 | ret = 1; | |
1080 | else | |
1081 | ret = 0; | |
1082 | } | |
8190e754 | 1083 | rdev->sectors = rdev->sb_start; |
1da177e4 | 1084 | |
dd8ac336 | 1085 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
1086 | /* "this cannot possibly happen" ... */ |
1087 | ret = -EINVAL; | |
1088 | ||
1da177e4 LT |
1089 | abort: |
1090 | return ret; | |
1091 | } | |
1092 | ||
1093 | /* | |
1094 | * validate_super for 0.90.0 | |
1095 | */ | |
1096 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1097 | { | |
1098 | mdp_disk_t *desc; | |
1099 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 1100 | __u64 ev1 = md_event(sb); |
1da177e4 | 1101 | |
41158c7e | 1102 | rdev->raid_disk = -1; |
c5d79adb N |
1103 | clear_bit(Faulty, &rdev->flags); |
1104 | clear_bit(In_sync, &rdev->flags); | |
1105 | clear_bit(WriteMostly, &rdev->flags); | |
c5d79adb | 1106 | |
1da177e4 LT |
1107 | if (mddev->raid_disks == 0) { |
1108 | mddev->major_version = 0; | |
1109 | mddev->minor_version = sb->minor_version; | |
1110 | mddev->patch_version = sb->patch_version; | |
e691063a | 1111 | mddev->external = 0; |
9d8f0363 | 1112 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1113 | mddev->ctime = sb->ctime; |
1114 | mddev->utime = sb->utime; | |
1115 | mddev->level = sb->level; | |
d9d166c2 | 1116 | mddev->clevel[0] = 0; |
1da177e4 LT |
1117 | mddev->layout = sb->layout; |
1118 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 1119 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 1120 | mddev->events = ev1; |
c3d9714e N |
1121 | mddev->bitmap_info.offset = 0; |
1122 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; | |
1da177e4 | 1123 | |
f6705578 N |
1124 | if (mddev->minor_version >= 91) { |
1125 | mddev->reshape_position = sb->reshape_position; | |
1126 | mddev->delta_disks = sb->delta_disks; | |
1127 | mddev->new_level = sb->new_level; | |
1128 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1129 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
1130 | } else { |
1131 | mddev->reshape_position = MaxSector; | |
1132 | mddev->delta_disks = 0; | |
1133 | mddev->new_level = mddev->level; | |
1134 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1135 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1136 | } |
1137 | ||
1da177e4 LT |
1138 | if (sb->state & (1<<MD_SB_CLEAN)) |
1139 | mddev->recovery_cp = MaxSector; | |
1140 | else { | |
1141 | if (sb->events_hi == sb->cp_events_hi && | |
1142 | sb->events_lo == sb->cp_events_lo) { | |
1143 | mddev->recovery_cp = sb->recovery_cp; | |
1144 | } else | |
1145 | mddev->recovery_cp = 0; | |
1146 | } | |
1147 | ||
1148 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1149 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1150 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1151 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1152 | ||
1153 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1154 | |
1155 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
c3d9714e N |
1156 | mddev->bitmap_info.file == NULL) |
1157 | mddev->bitmap_info.offset = | |
1158 | mddev->bitmap_info.default_offset; | |
a654b9d8 | 1159 | |
41158c7e | 1160 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1161 | /* Insist on good event counter while assembling, except |
1162 | * for spares (which don't need an event count) */ | |
1da177e4 | 1163 | ++ev1; |
be6800a7 N |
1164 | if (sb->disks[rdev->desc_nr].state & ( |
1165 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
1166 | if (ev1 < mddev->events) | |
1167 | return -EINVAL; | |
41158c7e N |
1168 | } else if (mddev->bitmap) { |
1169 | /* if adding to array with a bitmap, then we can accept an | |
1170 | * older device ... but not too old. | |
1171 | */ | |
41158c7e N |
1172 | if (ev1 < mddev->bitmap->events_cleared) |
1173 | return 0; | |
07d84d10 N |
1174 | } else { |
1175 | if (ev1 < mddev->events) | |
1176 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1177 | return 0; | |
1178 | } | |
41158c7e | 1179 | |
1da177e4 | 1180 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1181 | desc = sb->disks + rdev->desc_nr; |
1182 | ||
1183 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1184 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1185 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1186 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1187 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1188 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
1189 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1190 | /* active but not in sync implies recovery up to | |
1191 | * reshape position. We don't know exactly where | |
1192 | * that is, so set to zero for now */ | |
1193 | if (mddev->minor_version >= 91) { | |
1194 | rdev->recovery_offset = 0; | |
1195 | rdev->raid_disk = desc->raid_disk; | |
1196 | } | |
1da177e4 | 1197 | } |
8ddf9efe N |
1198 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1199 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1200 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1201 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1202 | return 0; |
1203 | } | |
1204 | ||
1205 | /* | |
1206 | * sync_super for 0.90.0 | |
1207 | */ | |
1208 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1209 | { | |
1210 | mdp_super_t *sb; | |
1da177e4 LT |
1211 | mdk_rdev_t *rdev2; |
1212 | int next_spare = mddev->raid_disks; | |
19133a42 | 1213 | |
1da177e4 LT |
1214 | |
1215 | /* make rdev->sb match mddev data.. | |
1216 | * | |
1217 | * 1/ zero out disks | |
1218 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1219 | * 3/ any empty disks < next_spare become removed | |
1220 | * | |
1221 | * disks[0] gets initialised to REMOVED because | |
1222 | * we cannot be sure from other fields if it has | |
1223 | * been initialised or not. | |
1224 | */ | |
1225 | int i; | |
1226 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1227 | ||
61181565 N |
1228 | rdev->sb_size = MD_SB_BYTES; |
1229 | ||
1da177e4 LT |
1230 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
1231 | ||
1232 | memset(sb, 0, sizeof(*sb)); | |
1233 | ||
1234 | sb->md_magic = MD_SB_MAGIC; | |
1235 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1236 | sb->patch_version = mddev->patch_version; |
1237 | sb->gvalid_words = 0; /* ignored */ | |
1238 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1239 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1240 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1241 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1242 | ||
1243 | sb->ctime = mddev->ctime; | |
1244 | sb->level = mddev->level; | |
58c0fed4 | 1245 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1246 | sb->raid_disks = mddev->raid_disks; |
1247 | sb->md_minor = mddev->md_minor; | |
e691063a | 1248 | sb->not_persistent = 0; |
1da177e4 LT |
1249 | sb->utime = mddev->utime; |
1250 | sb->state = 0; | |
1251 | sb->events_hi = (mddev->events>>32); | |
1252 | sb->events_lo = (u32)mddev->events; | |
1253 | ||
f6705578 N |
1254 | if (mddev->reshape_position == MaxSector) |
1255 | sb->minor_version = 90; | |
1256 | else { | |
1257 | sb->minor_version = 91; | |
1258 | sb->reshape_position = mddev->reshape_position; | |
1259 | sb->new_level = mddev->new_level; | |
1260 | sb->delta_disks = mddev->delta_disks; | |
1261 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1262 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1263 | } |
1264 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1265 | if (mddev->in_sync) |
1266 | { | |
1267 | sb->recovery_cp = mddev->recovery_cp; | |
1268 | sb->cp_events_hi = (mddev->events>>32); | |
1269 | sb->cp_events_lo = (u32)mddev->events; | |
1270 | if (mddev->recovery_cp == MaxSector) | |
1271 | sb->state = (1<< MD_SB_CLEAN); | |
1272 | } else | |
1273 | sb->recovery_cp = 0; | |
1274 | ||
1275 | sb->layout = mddev->layout; | |
9d8f0363 | 1276 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1277 | |
c3d9714e | 1278 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1279 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1280 | ||
1da177e4 | 1281 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1282 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1283 | mdp_disk_t *d; |
86e6ffdd | 1284 | int desc_nr; |
0261cd9f N |
1285 | int is_active = test_bit(In_sync, &rdev2->flags); |
1286 | ||
1287 | if (rdev2->raid_disk >= 0 && | |
1288 | sb->minor_version >= 91) | |
1289 | /* we have nowhere to store the recovery_offset, | |
1290 | * but if it is not below the reshape_position, | |
1291 | * we can piggy-back on that. | |
1292 | */ | |
1293 | is_active = 1; | |
1294 | if (rdev2->raid_disk < 0 || | |
1295 | test_bit(Faulty, &rdev2->flags)) | |
1296 | is_active = 0; | |
1297 | if (is_active) | |
86e6ffdd | 1298 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1299 | else |
86e6ffdd | 1300 | desc_nr = next_spare++; |
19133a42 | 1301 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1302 | d = &sb->disks[rdev2->desc_nr]; |
1303 | nr_disks++; | |
1304 | d->number = rdev2->desc_nr; | |
1305 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1306 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1307 | if (is_active) |
1da177e4 LT |
1308 | d->raid_disk = rdev2->raid_disk; |
1309 | else | |
1310 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1311 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1312 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1313 | else if (is_active) { |
1da177e4 | 1314 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1315 | if (test_bit(In_sync, &rdev2->flags)) |
1316 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1317 | active++; |
1318 | working++; | |
1319 | } else { | |
1320 | d->state = 0; | |
1321 | spare++; | |
1322 | working++; | |
1323 | } | |
8ddf9efe N |
1324 | if (test_bit(WriteMostly, &rdev2->flags)) |
1325 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1326 | } |
1da177e4 LT |
1327 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1328 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1329 | mdp_disk_t *d = &sb->disks[i]; | |
1330 | if (d->state == 0 && d->number == 0) { | |
1331 | d->number = i; | |
1332 | d->raid_disk = i; | |
1333 | d->state = (1<<MD_DISK_REMOVED); | |
1334 | d->state |= (1<<MD_DISK_FAULTY); | |
1335 | failed++; | |
1336 | } | |
1337 | } | |
1338 | sb->nr_disks = nr_disks; | |
1339 | sb->active_disks = active; | |
1340 | sb->working_disks = working; | |
1341 | sb->failed_disks = failed; | |
1342 | sb->spare_disks = spare; | |
1343 | ||
1344 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1345 | sb->sb_csum = calc_sb_csum(sb); | |
1346 | } | |
1347 | ||
0cd17fec CW |
1348 | /* |
1349 | * rdev_size_change for 0.90.0 | |
1350 | */ | |
1351 | static unsigned long long | |
15f4a5fd | 1352 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1353 | { |
58c0fed4 | 1354 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1355 | return 0; /* component must fit device */ |
c3d9714e | 1356 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1357 | return 0; /* can't move bitmap */ |
57b2caa3 | 1358 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1359 | if (!num_sectors || num_sectors > rdev->sb_start) |
1360 | num_sectors = rdev->sb_start; | |
0f420358 | 1361 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1362 | rdev->sb_page); |
1363 | md_super_wait(rdev->mddev); | |
c26a44ed | 1364 | return num_sectors; |
0cd17fec CW |
1365 | } |
1366 | ||
1367 | ||
1da177e4 LT |
1368 | /* |
1369 | * version 1 superblock | |
1370 | */ | |
1371 | ||
1c05b4bc | 1372 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1373 | { |
1c05b4bc N |
1374 | __le32 disk_csum; |
1375 | u32 csum; | |
1da177e4 LT |
1376 | unsigned long long newcsum; |
1377 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1378 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1379 | int i; |
1380 | ||
1381 | disk_csum = sb->sb_csum; | |
1382 | sb->sb_csum = 0; | |
1383 | newcsum = 0; | |
1384 | for (i=0; size>=4; size -= 4 ) | |
1385 | newcsum += le32_to_cpu(*isuper++); | |
1386 | ||
1387 | if (size == 2) | |
1c05b4bc | 1388 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1389 | |
1390 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1391 | sb->sb_csum = disk_csum; | |
1392 | return cpu_to_le32(csum); | |
1393 | } | |
1394 | ||
1395 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1396 | { | |
1397 | struct mdp_superblock_1 *sb; | |
1398 | int ret; | |
0f420358 | 1399 | sector_t sb_start; |
1da177e4 | 1400 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1401 | int bmask; |
1da177e4 LT |
1402 | |
1403 | /* | |
0f420358 | 1404 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1405 | * It is always aligned to a 4K boundary and |
1406 | * depeding on minor_version, it can be: | |
1407 | * 0: At least 8K, but less than 12K, from end of device | |
1408 | * 1: At start of device | |
1409 | * 2: 4K from start of device. | |
1410 | */ | |
1411 | switch(minor_version) { | |
1412 | case 0: | |
77304d2a | 1413 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1414 | sb_start -= 8*2; |
1415 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1416 | break; |
1417 | case 1: | |
0f420358 | 1418 | sb_start = 0; |
1da177e4 LT |
1419 | break; |
1420 | case 2: | |
0f420358 | 1421 | sb_start = 8; |
1da177e4 LT |
1422 | break; |
1423 | default: | |
1424 | return -EINVAL; | |
1425 | } | |
0f420358 | 1426 | rdev->sb_start = sb_start; |
1da177e4 | 1427 | |
0002b271 N |
1428 | /* superblock is rarely larger than 1K, but it can be larger, |
1429 | * and it is safe to read 4k, so we do that | |
1430 | */ | |
1431 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1432 | if (ret) return ret; |
1433 | ||
1434 | ||
1435 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1436 | ||
1437 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1438 | sb->major_version != cpu_to_le32(1) || | |
1439 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1440 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1441 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1442 | return -EINVAL; |
1443 | ||
1444 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1445 | printk("md: invalid superblock checksum on %s\n", | |
1446 | bdevname(rdev->bdev,b)); | |
1447 | return -EINVAL; | |
1448 | } | |
1449 | if (le64_to_cpu(sb->data_size) < 10) { | |
1450 | printk("md: data_size too small on %s\n", | |
1451 | bdevname(rdev->bdev,b)); | |
1452 | return -EINVAL; | |
1453 | } | |
e11e93fa | 1454 | |
1da177e4 LT |
1455 | rdev->preferred_minor = 0xffff; |
1456 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1457 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1458 | |
0002b271 | 1459 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1460 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1461 | if (rdev->sb_size & bmask) |
a1801f85 N |
1462 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1463 | ||
1464 | if (minor_version | |
0f420358 | 1465 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1466 | return -EINVAL; |
0002b271 | 1467 | |
31b65a0d N |
1468 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1469 | rdev->desc_nr = -1; | |
1470 | else | |
1471 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1472 | ||
9a7b2b0f | 1473 | if (!refdev) { |
8ed75463 | 1474 | ret = 1; |
9a7b2b0f | 1475 | } else { |
1da177e4 LT |
1476 | __u64 ev1, ev2; |
1477 | struct mdp_superblock_1 *refsb = | |
1478 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1479 | ||
1480 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1481 | sb->level != refsb->level || | |
1482 | sb->layout != refsb->layout || | |
1483 | sb->chunksize != refsb->chunksize) { | |
1484 | printk(KERN_WARNING "md: %s has strangely different" | |
1485 | " superblock to %s\n", | |
1486 | bdevname(rdev->bdev,b), | |
1487 | bdevname(refdev->bdev,b2)); | |
1488 | return -EINVAL; | |
1489 | } | |
1490 | ev1 = le64_to_cpu(sb->events); | |
1491 | ev2 = le64_to_cpu(refsb->events); | |
1492 | ||
1493 | if (ev1 > ev2) | |
8ed75463 N |
1494 | ret = 1; |
1495 | else | |
1496 | ret = 0; | |
1da177e4 | 1497 | } |
a1801f85 | 1498 | if (minor_version) |
77304d2a | 1499 | rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 1500 | le64_to_cpu(sb->data_offset); |
1da177e4 | 1501 | else |
dd8ac336 AN |
1502 | rdev->sectors = rdev->sb_start; |
1503 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1504 | return -EINVAL; |
dd8ac336 | 1505 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1506 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1507 | return -EINVAL; |
8ed75463 | 1508 | return ret; |
1da177e4 LT |
1509 | } |
1510 | ||
1511 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1512 | { | |
1513 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1514 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1515 | |
41158c7e | 1516 | rdev->raid_disk = -1; |
c5d79adb N |
1517 | clear_bit(Faulty, &rdev->flags); |
1518 | clear_bit(In_sync, &rdev->flags); | |
1519 | clear_bit(WriteMostly, &rdev->flags); | |
c5d79adb | 1520 | |
1da177e4 LT |
1521 | if (mddev->raid_disks == 0) { |
1522 | mddev->major_version = 1; | |
1523 | mddev->patch_version = 0; | |
e691063a | 1524 | mddev->external = 0; |
9d8f0363 | 1525 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1526 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1527 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1528 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1529 | mddev->clevel[0] = 0; |
1da177e4 LT |
1530 | mddev->layout = le32_to_cpu(sb->layout); |
1531 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1532 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1533 | mddev->events = ev1; |
c3d9714e N |
1534 | mddev->bitmap_info.offset = 0; |
1535 | mddev->bitmap_info.default_offset = 1024 >> 9; | |
1da177e4 LT |
1536 | |
1537 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1538 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1539 | ||
1540 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1541 | |
71c0805c | 1542 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
c3d9714e N |
1543 | mddev->bitmap_info.file == NULL ) |
1544 | mddev->bitmap_info.offset = | |
1545 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
e11e93fa | 1546 | |
f6705578 N |
1547 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1548 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1549 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1550 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1551 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1552 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1553 | } else { |
1554 | mddev->reshape_position = MaxSector; | |
1555 | mddev->delta_disks = 0; | |
1556 | mddev->new_level = mddev->level; | |
1557 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1558 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1559 | } |
1560 | ||
41158c7e | 1561 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1562 | /* Insist of good event counter while assembling, except for |
1563 | * spares (which don't need an event count) */ | |
1da177e4 | 1564 | ++ev1; |
be6800a7 N |
1565 | if (rdev->desc_nr >= 0 && |
1566 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1567 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1568 | if (ev1 < mddev->events) | |
1569 | return -EINVAL; | |
41158c7e N |
1570 | } else if (mddev->bitmap) { |
1571 | /* If adding to array with a bitmap, then we can accept an | |
1572 | * older device, but not too old. | |
1573 | */ | |
41158c7e N |
1574 | if (ev1 < mddev->bitmap->events_cleared) |
1575 | return 0; | |
07d84d10 N |
1576 | } else { |
1577 | if (ev1 < mddev->events) | |
1578 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1579 | return 0; | |
1580 | } | |
1da177e4 LT |
1581 | if (mddev->level != LEVEL_MULTIPATH) { |
1582 | int role; | |
3673f305 N |
1583 | if (rdev->desc_nr < 0 || |
1584 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1585 | role = 0xffff; | |
1586 | rdev->desc_nr = -1; | |
1587 | } else | |
1588 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1589 | switch(role) { |
1590 | case 0xffff: /* spare */ | |
1da177e4 LT |
1591 | break; |
1592 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1593 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1594 | break; |
1595 | default: | |
5fd6c1dc N |
1596 | if ((le32_to_cpu(sb->feature_map) & |
1597 | MD_FEATURE_RECOVERY_OFFSET)) | |
1598 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1599 | else | |
1600 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1601 | rdev->raid_disk = role; |
1602 | break; | |
1603 | } | |
8ddf9efe N |
1604 | if (sb->devflags & WriteMostly1) |
1605 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1606 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1607 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1608 | |
1da177e4 LT |
1609 | return 0; |
1610 | } | |
1611 | ||
1612 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1613 | { | |
1614 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1615 | mdk_rdev_t *rdev2; |
1616 | int max_dev, i; | |
1617 | /* make rdev->sb match mddev and rdev data. */ | |
1618 | ||
1619 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1620 | ||
1621 | sb->feature_map = 0; | |
1622 | sb->pad0 = 0; | |
5fd6c1dc | 1623 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1624 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1625 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1626 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1627 | ||
1628 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1629 | sb->events = cpu_to_le64(mddev->events); | |
1630 | if (mddev->in_sync) | |
1631 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1632 | else | |
1633 | sb->resync_offset = cpu_to_le64(0); | |
1634 | ||
1c05b4bc | 1635 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1636 | |
f0ca340c | 1637 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1638 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1639 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1640 | sb->level = cpu_to_le32(mddev->level); |
1641 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1642 | |
c3d9714e N |
1643 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1644 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1645 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1646 | } |
5fd6c1dc N |
1647 | |
1648 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1649 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1650 | sb->feature_map |= |
1651 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1652 | sb->recovery_offset = | |
1653 | cpu_to_le64(rdev->recovery_offset); | |
5fd6c1dc N |
1654 | } |
1655 | ||
f6705578 N |
1656 | if (mddev->reshape_position != MaxSector) { |
1657 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1658 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1659 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1660 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1661 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1662 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1663 | } |
a654b9d8 | 1664 | |
1da177e4 | 1665 | max_dev = 0; |
159ec1fc | 1666 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1667 | if (rdev2->desc_nr+1 > max_dev) |
1668 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1669 | |
70471daf N |
1670 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1671 | int bmask; | |
a778b73f | 1672 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1673 | rdev->sb_size = max_dev * 2 + 256; |
1674 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1675 | if (rdev->sb_size & bmask) | |
1676 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1677 | } else |
1678 | max_dev = le32_to_cpu(sb->max_dev); | |
1679 | ||
1da177e4 LT |
1680 | for (i=0; i<max_dev;i++) |
1681 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1682 | ||
159ec1fc | 1683 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1684 | i = rdev2->desc_nr; |
b2d444d7 | 1685 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1686 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1687 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1688 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1689 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1690 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1691 | else |
1692 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1693 | } | |
1694 | ||
1da177e4 LT |
1695 | sb->sb_csum = calc_sb_1_csum(sb); |
1696 | } | |
1697 | ||
0cd17fec | 1698 | static unsigned long long |
15f4a5fd | 1699 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1700 | { |
1701 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1702 | sector_t max_sectors; |
58c0fed4 | 1703 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1704 | return 0; /* component must fit device */ |
0f420358 | 1705 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1706 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1707 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1708 | max_sectors -= rdev->data_offset; |
1709 | if (!num_sectors || num_sectors > max_sectors) | |
1710 | num_sectors = max_sectors; | |
c3d9714e | 1711 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1712 | /* minor version 0 with bitmap we can't move */ |
1713 | return 0; | |
1714 | } else { | |
1715 | /* minor version 0; superblock after data */ | |
0f420358 | 1716 | sector_t sb_start; |
77304d2a | 1717 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1718 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1719 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1720 | if (!num_sectors || num_sectors > max_sectors) |
1721 | num_sectors = max_sectors; | |
0f420358 | 1722 | rdev->sb_start = sb_start; |
0cd17fec CW |
1723 | } |
1724 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1725 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1726 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1727 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1728 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1729 | rdev->sb_page); |
1730 | md_super_wait(rdev->mddev); | |
c26a44ed | 1731 | return num_sectors; |
0cd17fec | 1732 | } |
1da177e4 | 1733 | |
75c96f85 | 1734 | static struct super_type super_types[] = { |
1da177e4 LT |
1735 | [0] = { |
1736 | .name = "0.90.0", | |
1737 | .owner = THIS_MODULE, | |
0cd17fec CW |
1738 | .load_super = super_90_load, |
1739 | .validate_super = super_90_validate, | |
1740 | .sync_super = super_90_sync, | |
1741 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1742 | }, |
1743 | [1] = { | |
1744 | .name = "md-1", | |
1745 | .owner = THIS_MODULE, | |
0cd17fec CW |
1746 | .load_super = super_1_load, |
1747 | .validate_super = super_1_validate, | |
1748 | .sync_super = super_1_sync, | |
1749 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1750 | }, |
1751 | }; | |
1da177e4 LT |
1752 | |
1753 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1754 | { | |
7dd5e7c3 | 1755 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1756 | |
4b80991c N |
1757 | rcu_read_lock(); |
1758 | rdev_for_each_rcu(rdev, mddev1) | |
1759 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1760 | if (rdev->bdev->bd_contains == |
4b80991c N |
1761 | rdev2->bdev->bd_contains) { |
1762 | rcu_read_unlock(); | |
7dd5e7c3 | 1763 | return 1; |
4b80991c N |
1764 | } |
1765 | rcu_read_unlock(); | |
1da177e4 LT |
1766 | return 0; |
1767 | } | |
1768 | ||
1769 | static LIST_HEAD(pending_raid_disks); | |
1770 | ||
ac5e7113 AN |
1771 | /* |
1772 | * Try to register data integrity profile for an mddev | |
1773 | * | |
1774 | * This is called when an array is started and after a disk has been kicked | |
1775 | * from the array. It only succeeds if all working and active component devices | |
1776 | * are integrity capable with matching profiles. | |
1777 | */ | |
1778 | int md_integrity_register(mddev_t *mddev) | |
1779 | { | |
1780 | mdk_rdev_t *rdev, *reference = NULL; | |
1781 | ||
1782 | if (list_empty(&mddev->disks)) | |
1783 | return 0; /* nothing to do */ | |
1784 | if (blk_get_integrity(mddev->gendisk)) | |
1785 | return 0; /* already registered */ | |
1786 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
1787 | /* skip spares and non-functional disks */ | |
1788 | if (test_bit(Faulty, &rdev->flags)) | |
1789 | continue; | |
1790 | if (rdev->raid_disk < 0) | |
1791 | continue; | |
ac5e7113 AN |
1792 | if (!reference) { |
1793 | /* Use the first rdev as the reference */ | |
1794 | reference = rdev; | |
1795 | continue; | |
1796 | } | |
1797 | /* does this rdev's profile match the reference profile? */ | |
1798 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1799 | rdev->bdev->bd_disk) < 0) | |
1800 | return -EINVAL; | |
1801 | } | |
89078d57 MP |
1802 | if (!reference || !bdev_get_integrity(reference->bdev)) |
1803 | return 0; | |
ac5e7113 AN |
1804 | /* |
1805 | * All component devices are integrity capable and have matching | |
1806 | * profiles, register the common profile for the md device. | |
1807 | */ | |
1808 | if (blk_integrity_register(mddev->gendisk, | |
1809 | bdev_get_integrity(reference->bdev)) != 0) { | |
1810 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1811 | mdname(mddev)); | |
1812 | return -EINVAL; | |
1813 | } | |
a91a2785 MP |
1814 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
1815 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
1816 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
1817 | mdname(mddev)); | |
1818 | return -EINVAL; | |
1819 | } | |
ac5e7113 AN |
1820 | return 0; |
1821 | } | |
1822 | EXPORT_SYMBOL(md_integrity_register); | |
1823 | ||
1824 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
1825 | void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) | |
3f9d99c1 | 1826 | { |
3f9d99c1 | 1827 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); |
ac5e7113 | 1828 | struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); |
3f9d99c1 | 1829 | |
ac5e7113 | 1830 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1831 | return; |
ac5e7113 | 1832 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1833 | return; |
ac5e7113 AN |
1834 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1835 | rdev->bdev->bd_disk) >= 0) | |
1836 | return; | |
1837 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1838 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1839 | } |
ac5e7113 | 1840 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1841 | |
1da177e4 LT |
1842 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1843 | { | |
7dd5e7c3 | 1844 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1845 | struct kobject *ko; |
1edf80d3 | 1846 | char *s; |
5e55e2f5 | 1847 | int err; |
1da177e4 LT |
1848 | |
1849 | if (rdev->mddev) { | |
1850 | MD_BUG(); | |
1851 | return -EINVAL; | |
1852 | } | |
11e2ede0 DW |
1853 | |
1854 | /* prevent duplicates */ | |
1855 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1856 | return -EEXIST; | |
1857 | ||
dd8ac336 AN |
1858 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1859 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1860 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1861 | if (mddev->pers) { |
1862 | /* Cannot change size, so fail | |
1863 | * If mddev->level <= 0, then we don't care | |
1864 | * about aligning sizes (e.g. linear) | |
1865 | */ | |
1866 | if (mddev->level > 0) | |
1867 | return -ENOSPC; | |
1868 | } else | |
dd8ac336 | 1869 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1870 | } |
1da177e4 LT |
1871 | |
1872 | /* Verify rdev->desc_nr is unique. | |
1873 | * If it is -1, assign a free number, else | |
1874 | * check number is not in use | |
1875 | */ | |
1876 | if (rdev->desc_nr < 0) { | |
1877 | int choice = 0; | |
1878 | if (mddev->pers) choice = mddev->raid_disks; | |
1879 | while (find_rdev_nr(mddev, choice)) | |
1880 | choice++; | |
1881 | rdev->desc_nr = choice; | |
1882 | } else { | |
1883 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1884 | return -EBUSY; | |
1885 | } | |
de01dfad N |
1886 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1887 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1888 | mdname(mddev), mddev->max_disks); | |
1889 | return -EBUSY; | |
1890 | } | |
19133a42 | 1891 | bdevname(rdev->bdev,b); |
649316b2 | 1892 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1893 | *s = '!'; |
649316b2 | 1894 | |
1da177e4 | 1895 | rdev->mddev = mddev; |
19133a42 | 1896 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1897 | |
b2d6db58 | 1898 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1899 | goto fail; |
86e6ffdd | 1900 | |
0762b8bd | 1901 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
1902 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
1903 | /* failure here is OK */; | |
1904 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 1905 | |
4b80991c | 1906 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 1907 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
1908 | |
1909 | /* May as well allow recovery to be retried once */ | |
1910 | mddev->recovery_disabled = 0; | |
3f9d99c1 | 1911 | |
1da177e4 | 1912 | return 0; |
5e55e2f5 N |
1913 | |
1914 | fail: | |
1915 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1916 | b, mdname(mddev)); | |
1917 | return err; | |
1da177e4 LT |
1918 | } |
1919 | ||
177a99b2 | 1920 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1921 | { |
1922 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1923 | kobject_del(&rdev->kobj); | |
177a99b2 | 1924 | kobject_put(&rdev->kobj); |
5792a285 N |
1925 | } |
1926 | ||
1da177e4 LT |
1927 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1928 | { | |
1929 | char b[BDEVNAME_SIZE]; | |
1930 | if (!rdev->mddev) { | |
1931 | MD_BUG(); | |
1932 | return; | |
1933 | } | |
49731baa | 1934 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1935 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1936 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1937 | rdev->mddev = NULL; | |
86e6ffdd | 1938 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1939 | sysfs_put(rdev->sysfs_state); |
1940 | rdev->sysfs_state = NULL; | |
5792a285 | 1941 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1942 | * writing to 'remove' to "dev/state". We also need |
1943 | * to delay it due to rcu usage. | |
5792a285 | 1944 | */ |
4b80991c | 1945 | synchronize_rcu(); |
177a99b2 N |
1946 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1947 | kobject_get(&rdev->kobj); | |
e804ac78 | 1948 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
1949 | } |
1950 | ||
1951 | /* | |
1952 | * prevent the device from being mounted, repartitioned or | |
1953 | * otherwise reused by a RAID array (or any other kernel | |
1954 | * subsystem), by bd_claiming the device. | |
1955 | */ | |
c5d79adb | 1956 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1957 | { |
1958 | int err = 0; | |
1959 | struct block_device *bdev; | |
1960 | char b[BDEVNAME_SIZE]; | |
1961 | ||
d4d77629 TH |
1962 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
1963 | shared ? (mdk_rdev_t *)lock_rdev : rdev); | |
1da177e4 LT |
1964 | if (IS_ERR(bdev)) { |
1965 | printk(KERN_ERR "md: could not open %s.\n", | |
1966 | __bdevname(dev, b)); | |
1967 | return PTR_ERR(bdev); | |
1968 | } | |
1da177e4 LT |
1969 | rdev->bdev = bdev; |
1970 | return err; | |
1971 | } | |
1972 | ||
1973 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1974 | { | |
1975 | struct block_device *bdev = rdev->bdev; | |
1976 | rdev->bdev = NULL; | |
1977 | if (!bdev) | |
1978 | MD_BUG(); | |
e525fd89 | 1979 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
1980 | } |
1981 | ||
1982 | void md_autodetect_dev(dev_t dev); | |
1983 | ||
1984 | static void export_rdev(mdk_rdev_t * rdev) | |
1985 | { | |
1986 | char b[BDEVNAME_SIZE]; | |
1987 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1988 | bdevname(rdev->bdev,b)); | |
1989 | if (rdev->mddev) | |
1990 | MD_BUG(); | |
1991 | free_disk_sb(rdev); | |
1da177e4 | 1992 | #ifndef MODULE |
d0fae18f N |
1993 | if (test_bit(AutoDetected, &rdev->flags)) |
1994 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1995 | #endif |
1996 | unlock_rdev(rdev); | |
86e6ffdd | 1997 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1998 | } |
1999 | ||
2000 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
2001 | { | |
2002 | unbind_rdev_from_array(rdev); | |
2003 | export_rdev(rdev); | |
2004 | } | |
2005 | ||
2006 | static void export_array(mddev_t *mddev) | |
2007 | { | |
159ec1fc | 2008 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 2009 | |
d089c6af | 2010 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
2011 | if (!rdev->mddev) { |
2012 | MD_BUG(); | |
2013 | continue; | |
2014 | } | |
2015 | kick_rdev_from_array(rdev); | |
2016 | } | |
2017 | if (!list_empty(&mddev->disks)) | |
2018 | MD_BUG(); | |
2019 | mddev->raid_disks = 0; | |
2020 | mddev->major_version = 0; | |
2021 | } | |
2022 | ||
2023 | static void print_desc(mdp_disk_t *desc) | |
2024 | { | |
2025 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
2026 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
2027 | } | |
2028 | ||
cd2ac932 | 2029 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
2030 | { |
2031 | int i; | |
2032 | ||
2033 | printk(KERN_INFO | |
2034 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
2035 | sb->major_version, sb->minor_version, sb->patch_version, | |
2036 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
2037 | sb->ctime); | |
2038 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
2039 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
2040 | sb->md_minor, sb->layout, sb->chunk_size); | |
2041 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
2042 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
2043 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
2044 | sb->failed_disks, sb->spare_disks, | |
2045 | sb->sb_csum, (unsigned long)sb->events_lo); | |
2046 | ||
2047 | printk(KERN_INFO); | |
2048 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2049 | mdp_disk_t *desc; | |
2050 | ||
2051 | desc = sb->disks + i; | |
2052 | if (desc->number || desc->major || desc->minor || | |
2053 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
2054 | printk(" D %2d: ", i); | |
2055 | print_desc(desc); | |
2056 | } | |
2057 | } | |
2058 | printk(KERN_INFO "md: THIS: "); | |
2059 | print_desc(&sb->this_disk); | |
cd2ac932 | 2060 | } |
1da177e4 | 2061 | |
cd2ac932 CR |
2062 | static void print_sb_1(struct mdp_superblock_1 *sb) |
2063 | { | |
2064 | __u8 *uuid; | |
2065 | ||
2066 | uuid = sb->set_uuid; | |
ad361c98 | 2067 | printk(KERN_INFO |
7b75c2f8 | 2068 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 2069 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
2070 | le32_to_cpu(sb->major_version), |
2071 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 2072 | uuid, |
cd2ac932 CR |
2073 | sb->set_name, |
2074 | (unsigned long long)le64_to_cpu(sb->ctime) | |
2075 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
2076 | ||
2077 | uuid = sb->device_uuid; | |
ad361c98 JP |
2078 | printk(KERN_INFO |
2079 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 2080 | " RO:%llu\n" |
7b75c2f8 | 2081 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
2082 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
2083 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
2084 | le32_to_cpu(sb->level), |
2085 | (unsigned long long)le64_to_cpu(sb->size), | |
2086 | le32_to_cpu(sb->raid_disks), | |
2087 | le32_to_cpu(sb->layout), | |
2088 | le32_to_cpu(sb->chunksize), | |
2089 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2090 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2091 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2092 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2093 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2094 | uuid, |
cd2ac932 CR |
2095 | sb->devflags, |
2096 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2097 | (unsigned long long)le64_to_cpu(sb->events), | |
2098 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2099 | le32_to_cpu(sb->sb_csum), | |
2100 | le32_to_cpu(sb->max_dev) | |
2101 | ); | |
1da177e4 LT |
2102 | } |
2103 | ||
cd2ac932 | 2104 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
2105 | { |
2106 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2107 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2108 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2109 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2110 | rdev->desc_nr); | |
1da177e4 | 2111 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2112 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2113 | switch (major_version) { | |
2114 | case 0: | |
2115 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
2116 | break; | |
2117 | case 1: | |
2118 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
2119 | break; | |
2120 | } | |
1da177e4 LT |
2121 | } else |
2122 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2123 | } | |
2124 | ||
5e56341d | 2125 | static void md_print_devices(void) |
1da177e4 | 2126 | { |
159ec1fc | 2127 | struct list_head *tmp; |
1da177e4 LT |
2128 | mdk_rdev_t *rdev; |
2129 | mddev_t *mddev; | |
2130 | char b[BDEVNAME_SIZE]; | |
2131 | ||
2132 | printk("\n"); | |
2133 | printk("md: **********************************\n"); | |
2134 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2135 | printk("md: **********************************\n"); | |
29ac4aa3 | 2136 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2137 | |
32a7627c N |
2138 | if (mddev->bitmap) |
2139 | bitmap_print_sb(mddev->bitmap); | |
2140 | else | |
2141 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 2142 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
2143 | printk("<%s>", bdevname(rdev->bdev,b)); |
2144 | printk("\n"); | |
2145 | ||
159ec1fc | 2146 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 2147 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2148 | } |
2149 | printk("md: **********************************\n"); | |
2150 | printk("\n"); | |
2151 | } | |
2152 | ||
2153 | ||
42543769 | 2154 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 2155 | { |
42543769 N |
2156 | /* Update each superblock (in-memory image), but |
2157 | * if we are allowed to, skip spares which already | |
2158 | * have the right event counter, or have one earlier | |
2159 | * (which would mean they aren't being marked as dirty | |
2160 | * with the rest of the array) | |
2161 | */ | |
1da177e4 | 2162 | mdk_rdev_t *rdev; |
159ec1fc | 2163 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
2164 | if (rdev->sb_events == mddev->events || |
2165 | (nospares && | |
2166 | rdev->raid_disk < 0 && | |
42543769 N |
2167 | rdev->sb_events+1 == mddev->events)) { |
2168 | /* Don't update this superblock */ | |
2169 | rdev->sb_loaded = 2; | |
2170 | } else { | |
2171 | super_types[mddev->major_version]. | |
2172 | sync_super(mddev, rdev); | |
2173 | rdev->sb_loaded = 1; | |
2174 | } | |
1da177e4 LT |
2175 | } |
2176 | } | |
2177 | ||
850b2b42 | 2178 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 2179 | { |
1da177e4 | 2180 | mdk_rdev_t *rdev; |
06d91a5f | 2181 | int sync_req; |
42543769 | 2182 | int nospares = 0; |
1da177e4 | 2183 | |
1da177e4 | 2184 | repeat: |
3a3a5ddb N |
2185 | /* First make sure individual recovery_offsets are correct */ |
2186 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
2187 | if (rdev->raid_disk >= 0 && | |
2188 | mddev->delta_disks >= 0 && | |
2189 | !test_bit(In_sync, &rdev->flags) && | |
2190 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2191 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2192 | ||
2193 | } | |
bd52b746 | 2194 | if (!mddev->persistent) { |
070dc6dd | 2195 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2196 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
d97a41dc N |
2197 | if (!mddev->external) |
2198 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
3a3a5ddb N |
2199 | wake_up(&mddev->sb_wait); |
2200 | return; | |
2201 | } | |
2202 | ||
a9701a30 | 2203 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2204 | |
3a3a5ddb N |
2205 | mddev->utime = get_seconds(); |
2206 | ||
850b2b42 N |
2207 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2208 | force_change = 1; | |
2209 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2210 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2211 | * though if events isn't the right even/odd, we will have to do | |
2212 | * spares after all | |
2213 | */ | |
2214 | nospares = 1; | |
2215 | if (force_change) | |
2216 | nospares = 0; | |
2217 | if (mddev->degraded) | |
84692195 N |
2218 | /* If the array is degraded, then skipping spares is both |
2219 | * dangerous and fairly pointless. | |
2220 | * Dangerous because a device that was removed from the array | |
2221 | * might have a event_count that still looks up-to-date, | |
2222 | * so it can be re-added without a resync. | |
2223 | * Pointless because if there are any spares to skip, | |
2224 | * then a recovery will happen and soon that array won't | |
2225 | * be degraded any more and the spare can go back to sleep then. | |
2226 | */ | |
850b2b42 | 2227 | nospares = 0; |
84692195 | 2228 | |
06d91a5f | 2229 | sync_req = mddev->in_sync; |
42543769 N |
2230 | |
2231 | /* If this is just a dirty<->clean transition, and the array is clean | |
2232 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2233 | if (nospares |
42543769 | 2234 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2235 | && mddev->can_decrease_events |
2236 | && mddev->events != 1) { | |
42543769 | 2237 | mddev->events--; |
a8707c08 N |
2238 | mddev->can_decrease_events = 0; |
2239 | } else { | |
42543769 N |
2240 | /* otherwise we have to go forward and ... */ |
2241 | mddev->events ++; | |
a8707c08 | 2242 | mddev->can_decrease_events = nospares; |
42543769 | 2243 | } |
1da177e4 LT |
2244 | |
2245 | if (!mddev->events) { | |
2246 | /* | |
2247 | * oops, this 64-bit counter should never wrap. | |
2248 | * Either we are in around ~1 trillion A.C., assuming | |
2249 | * 1 reboot per second, or we have a bug: | |
2250 | */ | |
2251 | MD_BUG(); | |
2252 | mddev->events --; | |
2253 | } | |
e691063a | 2254 | sync_sbs(mddev, nospares); |
a9701a30 | 2255 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2256 | |
2257 | dprintk(KERN_INFO | |
2258 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2259 | mdname(mddev),mddev->in_sync); | |
2260 | ||
4ad13663 | 2261 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2262 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2263 | char b[BDEVNAME_SIZE]; |
2264 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2265 | if (rdev->sb_loaded != 1) |
2266 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2267 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2268 | dprintk("(skipping faulty "); |
2269 | ||
2270 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2271 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2272 | md_super_write(mddev,rdev, |
0f420358 | 2273 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2274 | rdev->sb_page); |
2275 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2276 | bdevname(rdev->bdev,b), | |
0f420358 | 2277 | (unsigned long long)rdev->sb_start); |
42543769 | 2278 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2279 | |
1da177e4 LT |
2280 | } else |
2281 | dprintk(")\n"); | |
7bfa19f2 | 2282 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2283 | /* only need to write one superblock... */ |
2284 | break; | |
2285 | } | |
a9701a30 | 2286 | md_super_wait(mddev); |
850b2b42 | 2287 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2288 | |
a9701a30 | 2289 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2290 | if (mddev->in_sync != sync_req || |
2291 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2292 | /* have to write it out again */ |
a9701a30 | 2293 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2294 | goto repeat; |
2295 | } | |
850b2b42 | 2296 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2297 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2298 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2299 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2300 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2301 | |
1da177e4 LT |
2302 | } |
2303 | ||
7f6ce769 | 2304 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2305 | * We want to accept with case. For this we use cmd_match. |
2306 | */ | |
2307 | static int cmd_match(const char *cmd, const char *str) | |
2308 | { | |
2309 | /* See if cmd, written into a sysfs file, matches | |
2310 | * str. They must either be the same, or cmd can | |
2311 | * have a trailing newline | |
2312 | */ | |
2313 | while (*cmd && *str && *cmd == *str) { | |
2314 | cmd++; | |
2315 | str++; | |
2316 | } | |
2317 | if (*cmd == '\n') | |
2318 | cmd++; | |
2319 | if (*str || *cmd) | |
2320 | return 0; | |
2321 | return 1; | |
2322 | } | |
2323 | ||
86e6ffdd N |
2324 | struct rdev_sysfs_entry { |
2325 | struct attribute attr; | |
2326 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2327 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2328 | }; | |
2329 | ||
2330 | static ssize_t | |
96de1e66 | 2331 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2332 | { |
2333 | char *sep = ""; | |
20a49ff6 | 2334 | size_t len = 0; |
86e6ffdd | 2335 | |
b2d444d7 | 2336 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2337 | len+= sprintf(page+len, "%sfaulty",sep); |
2338 | sep = ","; | |
2339 | } | |
b2d444d7 | 2340 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2341 | len += sprintf(page+len, "%sin_sync",sep); |
2342 | sep = ","; | |
2343 | } | |
f655675b N |
2344 | if (test_bit(WriteMostly, &rdev->flags)) { |
2345 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2346 | sep = ","; | |
2347 | } | |
6bfe0b49 DW |
2348 | if (test_bit(Blocked, &rdev->flags)) { |
2349 | len += sprintf(page+len, "%sblocked", sep); | |
2350 | sep = ","; | |
2351 | } | |
b2d444d7 N |
2352 | if (!test_bit(Faulty, &rdev->flags) && |
2353 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2354 | len += sprintf(page+len, "%sspare", sep); |
2355 | sep = ","; | |
2356 | } | |
2357 | return len+sprintf(page+len, "\n"); | |
2358 | } | |
2359 | ||
45dc2de1 N |
2360 | static ssize_t |
2361 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2362 | { | |
2363 | /* can write | |
2364 | * faulty - simulates and error | |
2365 | * remove - disconnects the device | |
f655675b N |
2366 | * writemostly - sets write_mostly |
2367 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2368 | * blocked - sets the Blocked flag |
2369 | * -blocked - clears the Blocked flag | |
6d56e278 | 2370 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2371 | */ |
2372 | int err = -EINVAL; | |
2373 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2374 | md_error(rdev->mddev, rdev); | |
2375 | err = 0; | |
2376 | } else if (cmd_match(buf, "remove")) { | |
2377 | if (rdev->raid_disk >= 0) | |
2378 | err = -EBUSY; | |
2379 | else { | |
2380 | mddev_t *mddev = rdev->mddev; | |
2381 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2382 | if (mddev->pers) |
2383 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2384 | md_new_event(mddev); |
2385 | err = 0; | |
2386 | } | |
f655675b N |
2387 | } else if (cmd_match(buf, "writemostly")) { |
2388 | set_bit(WriteMostly, &rdev->flags); | |
2389 | err = 0; | |
2390 | } else if (cmd_match(buf, "-writemostly")) { | |
2391 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2392 | err = 0; |
2393 | } else if (cmd_match(buf, "blocked")) { | |
2394 | set_bit(Blocked, &rdev->flags); | |
2395 | err = 0; | |
2396 | } else if (cmd_match(buf, "-blocked")) { | |
2397 | clear_bit(Blocked, &rdev->flags); | |
2398 | wake_up(&rdev->blocked_wait); | |
2399 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2400 | md_wakeup_thread(rdev->mddev->thread); | |
2401 | ||
6d56e278 N |
2402 | err = 0; |
2403 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2404 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2405 | err = 0; |
45dc2de1 | 2406 | } |
00bcb4ac N |
2407 | if (!err) |
2408 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2409 | return err ? err : len; |
2410 | } | |
80ca3a44 N |
2411 | static struct rdev_sysfs_entry rdev_state = |
2412 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2413 | |
4dbcdc75 N |
2414 | static ssize_t |
2415 | errors_show(mdk_rdev_t *rdev, char *page) | |
2416 | { | |
2417 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2418 | } | |
2419 | ||
2420 | static ssize_t | |
2421 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2422 | { | |
2423 | char *e; | |
2424 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2425 | if (*buf && (*e == 0 || *e == '\n')) { | |
2426 | atomic_set(&rdev->corrected_errors, n); | |
2427 | return len; | |
2428 | } | |
2429 | return -EINVAL; | |
2430 | } | |
2431 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2432 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2433 | |
014236d2 N |
2434 | static ssize_t |
2435 | slot_show(mdk_rdev_t *rdev, char *page) | |
2436 | { | |
2437 | if (rdev->raid_disk < 0) | |
2438 | return sprintf(page, "none\n"); | |
2439 | else | |
2440 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2441 | } | |
2442 | ||
2443 | static ssize_t | |
2444 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2445 | { | |
2446 | char *e; | |
c303da6d N |
2447 | int err; |
2448 | char nm[20]; | |
014236d2 N |
2449 | int slot = simple_strtoul(buf, &e, 10); |
2450 | if (strncmp(buf, "none", 4)==0) | |
2451 | slot = -1; | |
2452 | else if (e==buf || (*e && *e!= '\n')) | |
2453 | return -EINVAL; | |
6c2fce2e | 2454 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2455 | /* Setting 'slot' on an active array requires also |
2456 | * updating the 'rd%d' link, and communicating | |
2457 | * with the personality with ->hot_*_disk. | |
2458 | * For now we only support removing | |
2459 | * failed/spare devices. This normally happens automatically, | |
2460 | * but not when the metadata is externally managed. | |
2461 | */ | |
c303da6d N |
2462 | if (rdev->raid_disk == -1) |
2463 | return -EEXIST; | |
2464 | /* personality does all needed checks */ | |
2465 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2466 | return -EINVAL; | |
2467 | err = rdev->mddev->pers-> | |
2468 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2469 | if (err) | |
2470 | return err; | |
2471 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2472 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
b7103107 | 2473 | rdev->raid_disk = -1; |
c303da6d N |
2474 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2475 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2476 | } else if (rdev->mddev->pers) { |
2477 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2478 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2479 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2480 | */ |
2481 | ||
2482 | if (rdev->raid_disk != -1) | |
2483 | return -EBUSY; | |
2484 | ||
c6751b2b N |
2485 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2486 | return -EBUSY; | |
2487 | ||
6c2fce2e NB |
2488 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2489 | return -EINVAL; | |
2490 | ||
159ec1fc | 2491 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2492 | if (rdev2->raid_disk == slot) |
2493 | return -EEXIST; | |
2494 | ||
ba1b41b6 N |
2495 | if (slot >= rdev->mddev->raid_disks && |
2496 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2497 | return -ENOSPC; | |
2498 | ||
6c2fce2e NB |
2499 | rdev->raid_disk = slot; |
2500 | if (test_bit(In_sync, &rdev->flags)) | |
2501 | rdev->saved_raid_disk = slot; | |
2502 | else | |
2503 | rdev->saved_raid_disk = -1; | |
2504 | err = rdev->mddev->pers-> | |
2505 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2506 | if (err) { |
6c2fce2e | 2507 | rdev->raid_disk = -1; |
6c2fce2e | 2508 | return err; |
52664732 | 2509 | } else |
00bcb4ac | 2510 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6c2fce2e NB |
2511 | sprintf(nm, "rd%d", rdev->raid_disk); |
2512 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
00bcb4ac | 2513 | /* failure here is OK */; |
6c2fce2e | 2514 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2515 | } else { |
ba1b41b6 N |
2516 | if (slot >= rdev->mddev->raid_disks && |
2517 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2518 | return -ENOSPC; |
2519 | rdev->raid_disk = slot; | |
2520 | /* assume it is working */ | |
c5d79adb N |
2521 | clear_bit(Faulty, &rdev->flags); |
2522 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2523 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2524 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2525 | } |
014236d2 N |
2526 | return len; |
2527 | } | |
2528 | ||
2529 | ||
2530 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2531 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2532 | |
93c8cad0 N |
2533 | static ssize_t |
2534 | offset_show(mdk_rdev_t *rdev, char *page) | |
2535 | { | |
6961ece4 | 2536 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2537 | } |
2538 | ||
2539 | static ssize_t | |
2540 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2541 | { | |
2542 | char *e; | |
2543 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2544 | if (e==buf || (*e && *e != '\n')) | |
2545 | return -EINVAL; | |
8ed0a521 | 2546 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2547 | return -EBUSY; |
dd8ac336 | 2548 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2549 | /* Must set offset before size, so overlap checks |
2550 | * can be sane */ | |
2551 | return -EBUSY; | |
93c8cad0 N |
2552 | rdev->data_offset = offset; |
2553 | return len; | |
2554 | } | |
2555 | ||
2556 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2557 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2558 | |
83303b61 N |
2559 | static ssize_t |
2560 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2561 | { | |
dd8ac336 | 2562 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2563 | } |
2564 | ||
c5d79adb N |
2565 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2566 | { | |
2567 | /* check if two start/length pairs overlap */ | |
2568 | if (s1+l1 <= s2) | |
2569 | return 0; | |
2570 | if (s2+l2 <= s1) | |
2571 | return 0; | |
2572 | return 1; | |
2573 | } | |
2574 | ||
b522adcd DW |
2575 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2576 | { | |
2577 | unsigned long long blocks; | |
2578 | sector_t new; | |
2579 | ||
2580 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2581 | return -EINVAL; | |
2582 | ||
2583 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2584 | return -EINVAL; /* sector conversion overflow */ | |
2585 | ||
2586 | new = blocks * 2; | |
2587 | if (new != blocks * 2) | |
2588 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2589 | ||
2590 | *sectors = new; | |
2591 | return 0; | |
2592 | } | |
2593 | ||
83303b61 N |
2594 | static ssize_t |
2595 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2596 | { | |
27c529bb | 2597 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2598 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2599 | sector_t sectors; |
27c529bb | 2600 | |
b522adcd | 2601 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2602 | return -EINVAL; |
0cd17fec | 2603 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2604 | if (my_mddev->persistent) { |
dd8ac336 AN |
2605 | sectors = super_types[my_mddev->major_version]. |
2606 | rdev_size_change(rdev, sectors); | |
2607 | if (!sectors) | |
0cd17fec | 2608 | return -EBUSY; |
dd8ac336 | 2609 | } else if (!sectors) |
77304d2a | 2610 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2611 | rdev->data_offset; |
0cd17fec | 2612 | } |
dd8ac336 | 2613 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2614 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2615 | |
dd8ac336 AN |
2616 | rdev->sectors = sectors; |
2617 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2618 | /* need to check that all other rdevs with the same ->bdev |
2619 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2620 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2621 | * we have to change it back, we will have the lock again. |
2622 | */ | |
2623 | mddev_t *mddev; | |
2624 | int overlap = 0; | |
159ec1fc | 2625 | struct list_head *tmp; |
c5d79adb | 2626 | |
27c529bb | 2627 | mddev_unlock(my_mddev); |
29ac4aa3 | 2628 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2629 | mdk_rdev_t *rdev2; |
2630 | ||
2631 | mddev_lock(mddev); | |
159ec1fc | 2632 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
f21e9ff7 N |
2633 | if (rdev->bdev == rdev2->bdev && |
2634 | rdev != rdev2 && | |
2635 | overlaps(rdev->data_offset, rdev->sectors, | |
2636 | rdev2->data_offset, | |
2637 | rdev2->sectors)) { | |
c5d79adb N |
2638 | overlap = 1; |
2639 | break; | |
2640 | } | |
2641 | mddev_unlock(mddev); | |
2642 | if (overlap) { | |
2643 | mddev_put(mddev); | |
2644 | break; | |
2645 | } | |
2646 | } | |
27c529bb | 2647 | mddev_lock(my_mddev); |
c5d79adb N |
2648 | if (overlap) { |
2649 | /* Someone else could have slipped in a size | |
2650 | * change here, but doing so is just silly. | |
dd8ac336 | 2651 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2652 | * safe, and trust userspace not to race with |
2653 | * itself | |
2654 | */ | |
dd8ac336 | 2655 | rdev->sectors = oldsectors; |
c5d79adb N |
2656 | return -EBUSY; |
2657 | } | |
2658 | } | |
83303b61 N |
2659 | return len; |
2660 | } | |
2661 | ||
2662 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2663 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2664 | |
06e3c817 DW |
2665 | |
2666 | static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) | |
2667 | { | |
2668 | unsigned long long recovery_start = rdev->recovery_offset; | |
2669 | ||
2670 | if (test_bit(In_sync, &rdev->flags) || | |
2671 | recovery_start == MaxSector) | |
2672 | return sprintf(page, "none\n"); | |
2673 | ||
2674 | return sprintf(page, "%llu\n", recovery_start); | |
2675 | } | |
2676 | ||
2677 | static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2678 | { | |
2679 | unsigned long long recovery_start; | |
2680 | ||
2681 | if (cmd_match(buf, "none")) | |
2682 | recovery_start = MaxSector; | |
2683 | else if (strict_strtoull(buf, 10, &recovery_start)) | |
2684 | return -EINVAL; | |
2685 | ||
2686 | if (rdev->mddev->pers && | |
2687 | rdev->raid_disk >= 0) | |
2688 | return -EBUSY; | |
2689 | ||
2690 | rdev->recovery_offset = recovery_start; | |
2691 | if (recovery_start == MaxSector) | |
2692 | set_bit(In_sync, &rdev->flags); | |
2693 | else | |
2694 | clear_bit(In_sync, &rdev->flags); | |
2695 | return len; | |
2696 | } | |
2697 | ||
2698 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2699 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2700 | ||
86e6ffdd N |
2701 | static struct attribute *rdev_default_attrs[] = { |
2702 | &rdev_state.attr, | |
4dbcdc75 | 2703 | &rdev_errors.attr, |
014236d2 | 2704 | &rdev_slot.attr, |
93c8cad0 | 2705 | &rdev_offset.attr, |
83303b61 | 2706 | &rdev_size.attr, |
06e3c817 | 2707 | &rdev_recovery_start.attr, |
86e6ffdd N |
2708 | NULL, |
2709 | }; | |
2710 | static ssize_t | |
2711 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2712 | { | |
2713 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2714 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2715 | mddev_t *mddev = rdev->mddev; |
2716 | ssize_t rv; | |
86e6ffdd N |
2717 | |
2718 | if (!entry->show) | |
2719 | return -EIO; | |
27c529bb N |
2720 | |
2721 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2722 | if (!rv) { | |
2723 | if (rdev->mddev == NULL) | |
2724 | rv = -EBUSY; | |
2725 | else | |
2726 | rv = entry->show(rdev, page); | |
2727 | mddev_unlock(mddev); | |
2728 | } | |
2729 | return rv; | |
86e6ffdd N |
2730 | } |
2731 | ||
2732 | static ssize_t | |
2733 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2734 | const char *page, size_t length) | |
2735 | { | |
2736 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2737 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2738 | ssize_t rv; |
2739 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2740 | |
2741 | if (!entry->store) | |
2742 | return -EIO; | |
67463acb N |
2743 | if (!capable(CAP_SYS_ADMIN)) |
2744 | return -EACCES; | |
27c529bb | 2745 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2746 | if (!rv) { |
27c529bb N |
2747 | if (rdev->mddev == NULL) |
2748 | rv = -EBUSY; | |
2749 | else | |
2750 | rv = entry->store(rdev, page, length); | |
6a51830e | 2751 | mddev_unlock(mddev); |
ca388059 N |
2752 | } |
2753 | return rv; | |
86e6ffdd N |
2754 | } |
2755 | ||
2756 | static void rdev_free(struct kobject *ko) | |
2757 | { | |
2758 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2759 | kfree(rdev); | |
2760 | } | |
52cf25d0 | 2761 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
2762 | .show = rdev_attr_show, |
2763 | .store = rdev_attr_store, | |
2764 | }; | |
2765 | static struct kobj_type rdev_ktype = { | |
2766 | .release = rdev_free, | |
2767 | .sysfs_ops = &rdev_sysfs_ops, | |
2768 | .default_attrs = rdev_default_attrs, | |
2769 | }; | |
2770 | ||
e8bb9a83 N |
2771 | void md_rdev_init(mdk_rdev_t *rdev) |
2772 | { | |
2773 | rdev->desc_nr = -1; | |
2774 | rdev->saved_raid_disk = -1; | |
2775 | rdev->raid_disk = -1; | |
2776 | rdev->flags = 0; | |
2777 | rdev->data_offset = 0; | |
2778 | rdev->sb_events = 0; | |
2779 | rdev->last_read_error.tv_sec = 0; | |
2780 | rdev->last_read_error.tv_nsec = 0; | |
2781 | atomic_set(&rdev->nr_pending, 0); | |
2782 | atomic_set(&rdev->read_errors, 0); | |
2783 | atomic_set(&rdev->corrected_errors, 0); | |
2784 | ||
2785 | INIT_LIST_HEAD(&rdev->same_set); | |
2786 | init_waitqueue_head(&rdev->blocked_wait); | |
2787 | } | |
2788 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
2789 | /* |
2790 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2791 | * | |
2792 | * mark the device faulty if: | |
2793 | * | |
2794 | * - the device is nonexistent (zero size) | |
2795 | * - the device has no valid superblock | |
2796 | * | |
2797 | * a faulty rdev _never_ has rdev->sb set. | |
2798 | */ | |
2799 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2800 | { | |
2801 | char b[BDEVNAME_SIZE]; | |
2802 | int err; | |
2803 | mdk_rdev_t *rdev; | |
2804 | sector_t size; | |
2805 | ||
9ffae0cf | 2806 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2807 | if (!rdev) { |
2808 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2809 | return ERR_PTR(-ENOMEM); | |
2810 | } | |
1da177e4 | 2811 | |
e8bb9a83 | 2812 | md_rdev_init(rdev); |
1da177e4 LT |
2813 | if ((err = alloc_disk_sb(rdev))) |
2814 | goto abort_free; | |
2815 | ||
c5d79adb | 2816 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2817 | if (err) |
2818 | goto abort_free; | |
2819 | ||
f9cb074b | 2820 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2821 | |
77304d2a | 2822 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 LT |
2823 | if (!size) { |
2824 | printk(KERN_WARNING | |
2825 | "md: %s has zero or unknown size, marking faulty!\n", | |
2826 | bdevname(rdev->bdev,b)); | |
2827 | err = -EINVAL; | |
2828 | goto abort_free; | |
2829 | } | |
2830 | ||
2831 | if (super_format >= 0) { | |
2832 | err = super_types[super_format]. | |
2833 | load_super(rdev, NULL, super_minor); | |
2834 | if (err == -EINVAL) { | |
df968c4e N |
2835 | printk(KERN_WARNING |
2836 | "md: %s does not have a valid v%d.%d " | |
2837 | "superblock, not importing!\n", | |
2838 | bdevname(rdev->bdev,b), | |
2839 | super_format, super_minor); | |
1da177e4 LT |
2840 | goto abort_free; |
2841 | } | |
2842 | if (err < 0) { | |
2843 | printk(KERN_WARNING | |
2844 | "md: could not read %s's sb, not importing!\n", | |
2845 | bdevname(rdev->bdev,b)); | |
2846 | goto abort_free; | |
2847 | } | |
2848 | } | |
6bfe0b49 | 2849 | |
1da177e4 LT |
2850 | return rdev; |
2851 | ||
2852 | abort_free: | |
2853 | if (rdev->sb_page) { | |
2854 | if (rdev->bdev) | |
2855 | unlock_rdev(rdev); | |
2856 | free_disk_sb(rdev); | |
2857 | } | |
2858 | kfree(rdev); | |
2859 | return ERR_PTR(err); | |
2860 | } | |
2861 | ||
2862 | /* | |
2863 | * Check a full RAID array for plausibility | |
2864 | */ | |
2865 | ||
2866 | ||
a757e64c | 2867 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2868 | { |
2869 | int i; | |
159ec1fc | 2870 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2871 | char b[BDEVNAME_SIZE]; |
2872 | ||
2873 | freshest = NULL; | |
d089c6af | 2874 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2875 | switch (super_types[mddev->major_version]. |
2876 | load_super(rdev, freshest, mddev->minor_version)) { | |
2877 | case 1: | |
2878 | freshest = rdev; | |
2879 | break; | |
2880 | case 0: | |
2881 | break; | |
2882 | default: | |
2883 | printk( KERN_ERR \ | |
2884 | "md: fatal superblock inconsistency in %s" | |
2885 | " -- removing from array\n", | |
2886 | bdevname(rdev->bdev,b)); | |
2887 | kick_rdev_from_array(rdev); | |
2888 | } | |
2889 | ||
2890 | ||
2891 | super_types[mddev->major_version]. | |
2892 | validate_super(mddev, freshest); | |
2893 | ||
2894 | i = 0; | |
d089c6af | 2895 | rdev_for_each(rdev, tmp, mddev) { |
233fca36 N |
2896 | if (mddev->max_disks && |
2897 | (rdev->desc_nr >= mddev->max_disks || | |
2898 | i > mddev->max_disks)) { | |
de01dfad N |
2899 | printk(KERN_WARNING |
2900 | "md: %s: %s: only %d devices permitted\n", | |
2901 | mdname(mddev), bdevname(rdev->bdev, b), | |
2902 | mddev->max_disks); | |
2903 | kick_rdev_from_array(rdev); | |
2904 | continue; | |
2905 | } | |
1da177e4 LT |
2906 | if (rdev != freshest) |
2907 | if (super_types[mddev->major_version]. | |
2908 | validate_super(mddev, rdev)) { | |
2909 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2910 | " from array!\n", | |
2911 | bdevname(rdev->bdev,b)); | |
2912 | kick_rdev_from_array(rdev); | |
2913 | continue; | |
2914 | } | |
2915 | if (mddev->level == LEVEL_MULTIPATH) { | |
2916 | rdev->desc_nr = i++; | |
2917 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2918 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 2919 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
2920 | rdev->raid_disk = -1; |
2921 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2922 | } |
2923 | } | |
1da177e4 LT |
2924 | } |
2925 | ||
72e02075 N |
2926 | /* Read a fixed-point number. |
2927 | * Numbers in sysfs attributes should be in "standard" units where | |
2928 | * possible, so time should be in seconds. | |
2929 | * However we internally use a a much smaller unit such as | |
2930 | * milliseconds or jiffies. | |
2931 | * This function takes a decimal number with a possible fractional | |
2932 | * component, and produces an integer which is the result of | |
2933 | * multiplying that number by 10^'scale'. | |
2934 | * all without any floating-point arithmetic. | |
2935 | */ | |
2936 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
2937 | { | |
2938 | unsigned long result = 0; | |
2939 | long decimals = -1; | |
2940 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
2941 | if (*cp == '.') | |
2942 | decimals = 0; | |
2943 | else if (decimals < scale) { | |
2944 | unsigned int value; | |
2945 | value = *cp - '0'; | |
2946 | result = result * 10 + value; | |
2947 | if (decimals >= 0) | |
2948 | decimals++; | |
2949 | } | |
2950 | cp++; | |
2951 | } | |
2952 | if (*cp == '\n') | |
2953 | cp++; | |
2954 | if (*cp) | |
2955 | return -EINVAL; | |
2956 | if (decimals < 0) | |
2957 | decimals = 0; | |
2958 | while (decimals < scale) { | |
2959 | result *= 10; | |
2960 | decimals ++; | |
2961 | } | |
2962 | *res = result; | |
2963 | return 0; | |
2964 | } | |
2965 | ||
2966 | ||
19052c0e N |
2967 | static void md_safemode_timeout(unsigned long data); |
2968 | ||
16f17b39 N |
2969 | static ssize_t |
2970 | safe_delay_show(mddev_t *mddev, char *page) | |
2971 | { | |
2972 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2973 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2974 | } | |
2975 | static ssize_t | |
2976 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2977 | { | |
16f17b39 | 2978 | unsigned long msec; |
97ce0a7f | 2979 | |
72e02075 | 2980 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 2981 | return -EINVAL; |
16f17b39 N |
2982 | if (msec == 0) |
2983 | mddev->safemode_delay = 0; | |
2984 | else { | |
19052c0e | 2985 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2986 | mddev->safemode_delay = (msec*HZ)/1000; |
2987 | if (mddev->safemode_delay == 0) | |
2988 | mddev->safemode_delay = 1; | |
19052c0e N |
2989 | if (mddev->safemode_delay < old_delay) |
2990 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2991 | } |
2992 | return len; | |
2993 | } | |
2994 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2995 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2996 | |
eae1701f | 2997 | static ssize_t |
96de1e66 | 2998 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2999 | { |
2604b703 | 3000 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 3001 | if (p) |
eae1701f | 3002 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
3003 | else if (mddev->clevel[0]) |
3004 | return sprintf(page, "%s\n", mddev->clevel); | |
3005 | else if (mddev->level != LEVEL_NONE) | |
3006 | return sprintf(page, "%d\n", mddev->level); | |
3007 | else | |
3008 | return 0; | |
eae1701f N |
3009 | } |
3010 | ||
d9d166c2 N |
3011 | static ssize_t |
3012 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
3013 | { | |
f2859af6 | 3014 | char clevel[16]; |
20a49ff6 | 3015 | ssize_t rv = len; |
245f46c2 | 3016 | struct mdk_personality *pers; |
f2859af6 | 3017 | long level; |
245f46c2 | 3018 | void *priv; |
3a981b03 | 3019 | mdk_rdev_t *rdev; |
245f46c2 N |
3020 | |
3021 | if (mddev->pers == NULL) { | |
3022 | if (len == 0) | |
3023 | return 0; | |
3024 | if (len >= sizeof(mddev->clevel)) | |
3025 | return -ENOSPC; | |
3026 | strncpy(mddev->clevel, buf, len); | |
3027 | if (mddev->clevel[len-1] == '\n') | |
3028 | len--; | |
3029 | mddev->clevel[len] = 0; | |
3030 | mddev->level = LEVEL_NONE; | |
3031 | return rv; | |
3032 | } | |
3033 | ||
3034 | /* request to change the personality. Need to ensure: | |
3035 | * - array is not engaged in resync/recovery/reshape | |
3036 | * - old personality can be suspended | |
3037 | * - new personality will access other array. | |
3038 | */ | |
3039 | ||
bb4f1e9d N |
3040 | if (mddev->sync_thread || |
3041 | mddev->reshape_position != MaxSector || | |
3042 | mddev->sysfs_active) | |
d9d166c2 | 3043 | return -EBUSY; |
245f46c2 N |
3044 | |
3045 | if (!mddev->pers->quiesce) { | |
3046 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3047 | mdname(mddev), mddev->pers->name); | |
3048 | return -EINVAL; | |
3049 | } | |
3050 | ||
3051 | /* Now find the new personality */ | |
f2859af6 | 3052 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3053 | return -EINVAL; |
f2859af6 DW |
3054 | strncpy(clevel, buf, len); |
3055 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3056 | len--; |
f2859af6 DW |
3057 | clevel[len] = 0; |
3058 | if (strict_strtol(clevel, 10, &level)) | |
3059 | level = LEVEL_NONE; | |
245f46c2 | 3060 | |
f2859af6 DW |
3061 | if (request_module("md-%s", clevel) != 0) |
3062 | request_module("md-level-%s", clevel); | |
245f46c2 | 3063 | spin_lock(&pers_lock); |
f2859af6 | 3064 | pers = find_pers(level, clevel); |
245f46c2 N |
3065 | if (!pers || !try_module_get(pers->owner)) { |
3066 | spin_unlock(&pers_lock); | |
f2859af6 | 3067 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3068 | return -EINVAL; |
3069 | } | |
3070 | spin_unlock(&pers_lock); | |
3071 | ||
3072 | if (pers == mddev->pers) { | |
3073 | /* Nothing to do! */ | |
3074 | module_put(pers->owner); | |
3075 | return rv; | |
3076 | } | |
3077 | if (!pers->takeover) { | |
3078 | module_put(pers->owner); | |
3079 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3080 | mdname(mddev), clevel); |
245f46c2 N |
3081 | return -EINVAL; |
3082 | } | |
3083 | ||
e93f68a1 N |
3084 | list_for_each_entry(rdev, &mddev->disks, same_set) |
3085 | rdev->new_raid_disk = rdev->raid_disk; | |
3086 | ||
245f46c2 N |
3087 | /* ->takeover must set new_* and/or delta_disks |
3088 | * if it succeeds, and may set them when it fails. | |
3089 | */ | |
3090 | priv = pers->takeover(mddev); | |
3091 | if (IS_ERR(priv)) { | |
3092 | mddev->new_level = mddev->level; | |
3093 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3094 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3095 | mddev->raid_disks -= mddev->delta_disks; |
3096 | mddev->delta_disks = 0; | |
3097 | module_put(pers->owner); | |
3098 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3099 | mdname(mddev), clevel); |
245f46c2 N |
3100 | return PTR_ERR(priv); |
3101 | } | |
3102 | ||
3103 | /* Looks like we have a winner */ | |
3104 | mddev_suspend(mddev); | |
3105 | mddev->pers->stop(mddev); | |
a64c876f N |
3106 | |
3107 | if (mddev->pers->sync_request == NULL && | |
3108 | pers->sync_request != NULL) { | |
3109 | /* need to add the md_redundancy_group */ | |
3110 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3111 | printk(KERN_WARNING | |
3112 | "md: cannot register extra attributes for %s\n", | |
3113 | mdname(mddev)); | |
19fdb9ee | 3114 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); |
a64c876f N |
3115 | } |
3116 | if (mddev->pers->sync_request != NULL && | |
3117 | pers->sync_request == NULL) { | |
3118 | /* need to remove the md_redundancy_group */ | |
3119 | if (mddev->to_remove == NULL) | |
3120 | mddev->to_remove = &md_redundancy_group; | |
3121 | } | |
3122 | ||
54071b38 TM |
3123 | if (mddev->pers->sync_request == NULL && |
3124 | mddev->external) { | |
3125 | /* We are converting from a no-redundancy array | |
3126 | * to a redundancy array and metadata is managed | |
3127 | * externally so we need to be sure that writes | |
3128 | * won't block due to a need to transition | |
3129 | * clean->dirty | |
3130 | * until external management is started. | |
3131 | */ | |
3132 | mddev->in_sync = 0; | |
3133 | mddev->safemode_delay = 0; | |
3134 | mddev->safemode = 0; | |
3135 | } | |
3136 | ||
e93f68a1 N |
3137 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
3138 | char nm[20]; | |
3139 | if (rdev->raid_disk < 0) | |
3140 | continue; | |
bf2cb0da | 3141 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3142 | rdev->new_raid_disk = -1; |
3143 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3144 | continue; | |
3145 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3146 | sysfs_remove_link(&mddev->kobj, nm); | |
3147 | } | |
3148 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
3149 | if (rdev->raid_disk < 0) | |
3150 | continue; | |
3151 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3152 | continue; | |
3153 | rdev->raid_disk = rdev->new_raid_disk; | |
3154 | if (rdev->raid_disk < 0) | |
3a981b03 | 3155 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 N |
3156 | else { |
3157 | char nm[20]; | |
3158 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3159 | if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) | |
3160 | printk("md: cannot register %s for %s after level change\n", | |
3161 | nm, mdname(mddev)); | |
3a981b03 | 3162 | } |
e93f68a1 N |
3163 | } |
3164 | ||
3165 | module_put(mddev->pers->owner); | |
245f46c2 N |
3166 | mddev->pers = pers; |
3167 | mddev->private = priv; | |
3168 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3169 | mddev->level = mddev->new_level; | |
3170 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3171 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3172 | mddev->delta_disks = 0; |
fee68723 | 3173 | mddev->degraded = 0; |
9af204cf TM |
3174 | if (mddev->pers->sync_request == NULL) { |
3175 | /* this is now an array without redundancy, so | |
3176 | * it must always be in_sync | |
3177 | */ | |
3178 | mddev->in_sync = 1; | |
3179 | del_timer_sync(&mddev->safemode_timer); | |
3180 | } | |
245f46c2 N |
3181 | pers->run(mddev); |
3182 | mddev_resume(mddev); | |
3183 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3184 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3185 | md_wakeup_thread(mddev->thread); | |
5cac7861 | 3186 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3187 | md_new_event(mddev); |
d9d166c2 N |
3188 | return rv; |
3189 | } | |
3190 | ||
3191 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3192 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3193 | |
d4dbd025 N |
3194 | |
3195 | static ssize_t | |
3196 | layout_show(mddev_t *mddev, char *page) | |
3197 | { | |
3198 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3199 | if (mddev->reshape_position != MaxSector && |
3200 | mddev->layout != mddev->new_layout) | |
3201 | return sprintf(page, "%d (%d)\n", | |
3202 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3203 | return sprintf(page, "%d\n", mddev->layout); |
3204 | } | |
3205 | ||
3206 | static ssize_t | |
3207 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
3208 | { | |
3209 | char *e; | |
3210 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3211 | |
3212 | if (!*buf || (*e && *e != '\n')) | |
3213 | return -EINVAL; | |
3214 | ||
b3546035 N |
3215 | if (mddev->pers) { |
3216 | int err; | |
50ac168a | 3217 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3218 | return -EBUSY; |
597a711b | 3219 | mddev->new_layout = n; |
50ac168a | 3220 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3221 | if (err) { |
3222 | mddev->new_layout = mddev->layout; | |
b3546035 | 3223 | return err; |
597a711b | 3224 | } |
b3546035 | 3225 | } else { |
08a02ecd | 3226 | mddev->new_layout = n; |
b3546035 N |
3227 | if (mddev->reshape_position == MaxSector) |
3228 | mddev->layout = n; | |
3229 | } | |
d4dbd025 N |
3230 | return len; |
3231 | } | |
3232 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3233 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3234 | |
3235 | ||
eae1701f | 3236 | static ssize_t |
96de1e66 | 3237 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 3238 | { |
bb636547 N |
3239 | if (mddev->raid_disks == 0) |
3240 | return 0; | |
08a02ecd N |
3241 | if (mddev->reshape_position != MaxSector && |
3242 | mddev->delta_disks != 0) | |
3243 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3244 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3245 | return sprintf(page, "%d\n", mddev->raid_disks); |
3246 | } | |
3247 | ||
da943b99 N |
3248 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
3249 | ||
3250 | static ssize_t | |
3251 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
3252 | { | |
da943b99 N |
3253 | char *e; |
3254 | int rv = 0; | |
3255 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3256 | ||
3257 | if (!*buf || (*e && *e != '\n')) | |
3258 | return -EINVAL; | |
3259 | ||
3260 | if (mddev->pers) | |
3261 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
3262 | else if (mddev->reshape_position != MaxSector) { |
3263 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
3264 | mddev->delta_disks = n - olddisks; | |
3265 | mddev->raid_disks = n; | |
3266 | } else | |
da943b99 N |
3267 | mddev->raid_disks = n; |
3268 | return rv ? rv : len; | |
3269 | } | |
3270 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3271 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3272 | |
3b34380a N |
3273 | static ssize_t |
3274 | chunk_size_show(mddev_t *mddev, char *page) | |
3275 | { | |
08a02ecd | 3276 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3277 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3278 | return sprintf(page, "%d (%d)\n", | |
3279 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3280 | mddev->chunk_sectors << 9); |
3281 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3282 | } |
3283 | ||
3284 | static ssize_t | |
3285 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3286 | { | |
3b34380a N |
3287 | char *e; |
3288 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3289 | ||
3b34380a N |
3290 | if (!*buf || (*e && *e != '\n')) |
3291 | return -EINVAL; | |
3292 | ||
b3546035 N |
3293 | if (mddev->pers) { |
3294 | int err; | |
50ac168a | 3295 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3296 | return -EBUSY; |
597a711b | 3297 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3298 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3299 | if (err) { |
3300 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3301 | return err; |
597a711b | 3302 | } |
b3546035 | 3303 | } else { |
664e7c41 | 3304 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3305 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3306 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3307 | } |
3b34380a N |
3308 | return len; |
3309 | } | |
3310 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3311 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3312 | |
a94213b1 N |
3313 | static ssize_t |
3314 | resync_start_show(mddev_t *mddev, char *page) | |
3315 | { | |
d1a7c503 N |
3316 | if (mddev->recovery_cp == MaxSector) |
3317 | return sprintf(page, "none\n"); | |
a94213b1 N |
3318 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3319 | } | |
3320 | ||
3321 | static ssize_t | |
3322 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
3323 | { | |
a94213b1 N |
3324 | char *e; |
3325 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3326 | ||
3327 | if (mddev->pers) | |
3328 | return -EBUSY; | |
06e3c817 DW |
3329 | if (cmd_match(buf, "none")) |
3330 | n = MaxSector; | |
3331 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3332 | return -EINVAL; |
3333 | ||
3334 | mddev->recovery_cp = n; | |
3335 | return len; | |
3336 | } | |
3337 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3338 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3339 | |
9e653b63 N |
3340 | /* |
3341 | * The array state can be: | |
3342 | * | |
3343 | * clear | |
3344 | * No devices, no size, no level | |
3345 | * Equivalent to STOP_ARRAY ioctl | |
3346 | * inactive | |
3347 | * May have some settings, but array is not active | |
3348 | * all IO results in error | |
3349 | * When written, doesn't tear down array, but just stops it | |
3350 | * suspended (not supported yet) | |
3351 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3352 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3353 | * readonly |
3354 | * no resync can happen. no superblocks get written. | |
3355 | * write requests fail | |
3356 | * read-auto | |
3357 | * like readonly, but behaves like 'clean' on a write request. | |
3358 | * | |
3359 | * clean - no pending writes, but otherwise active. | |
3360 | * When written to inactive array, starts without resync | |
3361 | * If a write request arrives then | |
3362 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3363 | * if not known, block and switch to write-pending | |
3364 | * If written to an active array that has pending writes, then fails. | |
3365 | * active | |
3366 | * fully active: IO and resync can be happening. | |
3367 | * When written to inactive array, starts with resync | |
3368 | * | |
3369 | * write-pending | |
3370 | * clean, but writes are blocked waiting for 'active' to be written. | |
3371 | * | |
3372 | * active-idle | |
3373 | * like active, but no writes have been seen for a while (100msec). | |
3374 | * | |
3375 | */ | |
3376 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3377 | write_pending, active_idle, bad_word}; | |
05381954 | 3378 | static char *array_states[] = { |
9e653b63 N |
3379 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3380 | "write-pending", "active-idle", NULL }; | |
3381 | ||
3382 | static int match_word(const char *word, char **list) | |
3383 | { | |
3384 | int n; | |
3385 | for (n=0; list[n]; n++) | |
3386 | if (cmd_match(word, list[n])) | |
3387 | break; | |
3388 | return n; | |
3389 | } | |
3390 | ||
3391 | static ssize_t | |
3392 | array_state_show(mddev_t *mddev, char *page) | |
3393 | { | |
3394 | enum array_state st = inactive; | |
3395 | ||
3396 | if (mddev->pers) | |
3397 | switch(mddev->ro) { | |
3398 | case 1: | |
3399 | st = readonly; | |
3400 | break; | |
3401 | case 2: | |
3402 | st = read_auto; | |
3403 | break; | |
3404 | case 0: | |
3405 | if (mddev->in_sync) | |
3406 | st = clean; | |
070dc6dd | 3407 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3408 | st = write_pending; |
9e653b63 N |
3409 | else if (mddev->safemode) |
3410 | st = active_idle; | |
3411 | else | |
3412 | st = active; | |
3413 | } | |
3414 | else { | |
3415 | if (list_empty(&mddev->disks) && | |
3416 | mddev->raid_disks == 0 && | |
58c0fed4 | 3417 | mddev->dev_sectors == 0) |
9e653b63 N |
3418 | st = clear; |
3419 | else | |
3420 | st = inactive; | |
3421 | } | |
3422 | return sprintf(page, "%s\n", array_states[st]); | |
3423 | } | |
3424 | ||
df5b20cf | 3425 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
a4bd82d0 | 3426 | static int md_set_readonly(mddev_t * mddev, int is_open); |
9e653b63 N |
3427 | static int do_md_run(mddev_t * mddev); |
3428 | static int restart_array(mddev_t *mddev); | |
3429 | ||
3430 | static ssize_t | |
3431 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3432 | { | |
3433 | int err = -EINVAL; | |
3434 | enum array_state st = match_word(buf, array_states); | |
3435 | switch(st) { | |
3436 | case bad_word: | |
3437 | break; | |
3438 | case clear: | |
3439 | /* stopping an active array */ | |
f2ea68cf | 3440 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3441 | return -EBUSY; |
df5b20cf | 3442 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3443 | break; |
3444 | case inactive: | |
3445 | /* stopping an active array */ | |
3446 | if (mddev->pers) { | |
f2ea68cf | 3447 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3448 | return -EBUSY; |
df5b20cf | 3449 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3450 | } else |
3451 | err = 0; /* already inactive */ | |
9e653b63 N |
3452 | break; |
3453 | case suspended: | |
3454 | break; /* not supported yet */ | |
3455 | case readonly: | |
3456 | if (mddev->pers) | |
a4bd82d0 | 3457 | err = md_set_readonly(mddev, 0); |
9e653b63 N |
3458 | else { |
3459 | mddev->ro = 1; | |
648b629e | 3460 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3461 | err = do_md_run(mddev); |
3462 | } | |
3463 | break; | |
3464 | case read_auto: | |
9e653b63 | 3465 | if (mddev->pers) { |
80268ee9 | 3466 | if (mddev->ro == 0) |
a4bd82d0 | 3467 | err = md_set_readonly(mddev, 0); |
80268ee9 | 3468 | else if (mddev->ro == 1) |
648b629e N |
3469 | err = restart_array(mddev); |
3470 | if (err == 0) { | |
3471 | mddev->ro = 2; | |
3472 | set_disk_ro(mddev->gendisk, 0); | |
3473 | } | |
9e653b63 N |
3474 | } else { |
3475 | mddev->ro = 2; | |
3476 | err = do_md_run(mddev); | |
3477 | } | |
3478 | break; | |
3479 | case clean: | |
3480 | if (mddev->pers) { | |
3481 | restart_array(mddev); | |
3482 | spin_lock_irq(&mddev->write_lock); | |
3483 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3484 | if (mddev->in_sync == 0) { |
3485 | mddev->in_sync = 1; | |
31a59e34 N |
3486 | if (mddev->safemode == 1) |
3487 | mddev->safemode = 0; | |
070dc6dd | 3488 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3489 | } |
3490 | err = 0; | |
3491 | } else | |
3492 | err = -EBUSY; | |
9e653b63 | 3493 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3494 | } else |
3495 | err = -EINVAL; | |
9e653b63 N |
3496 | break; |
3497 | case active: | |
3498 | if (mddev->pers) { | |
3499 | restart_array(mddev); | |
070dc6dd | 3500 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3501 | wake_up(&mddev->sb_wait); |
3502 | err = 0; | |
3503 | } else { | |
3504 | mddev->ro = 0; | |
648b629e | 3505 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3506 | err = do_md_run(mddev); |
3507 | } | |
3508 | break; | |
3509 | case write_pending: | |
3510 | case active_idle: | |
3511 | /* these cannot be set */ | |
3512 | break; | |
3513 | } | |
3514 | if (err) | |
3515 | return err; | |
0fd62b86 | 3516 | else { |
00bcb4ac | 3517 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3518 | return len; |
0fd62b86 | 3519 | } |
9e653b63 | 3520 | } |
80ca3a44 N |
3521 | static struct md_sysfs_entry md_array_state = |
3522 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3523 | |
1e50915f RB |
3524 | static ssize_t |
3525 | max_corrected_read_errors_show(mddev_t *mddev, char *page) { | |
3526 | return sprintf(page, "%d\n", | |
3527 | atomic_read(&mddev->max_corr_read_errors)); | |
3528 | } | |
3529 | ||
3530 | static ssize_t | |
3531 | max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) | |
3532 | { | |
3533 | char *e; | |
3534 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3535 | ||
3536 | if (*buf && (*e == 0 || *e == '\n')) { | |
3537 | atomic_set(&mddev->max_corr_read_errors, n); | |
3538 | return len; | |
3539 | } | |
3540 | return -EINVAL; | |
3541 | } | |
3542 | ||
3543 | static struct md_sysfs_entry max_corr_read_errors = | |
3544 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3545 | max_corrected_read_errors_store); | |
3546 | ||
6d7ff738 N |
3547 | static ssize_t |
3548 | null_show(mddev_t *mddev, char *page) | |
3549 | { | |
3550 | return -EINVAL; | |
3551 | } | |
3552 | ||
3553 | static ssize_t | |
3554 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3555 | { | |
3556 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3557 | /* The new device is added to the array. | |
3558 | * If the array has a persistent superblock, we read the | |
3559 | * superblock to initialise info and check validity. | |
3560 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3561 | * which mainly checks size. | |
3562 | */ | |
3563 | char *e; | |
3564 | int major = simple_strtoul(buf, &e, 10); | |
3565 | int minor; | |
3566 | dev_t dev; | |
3567 | mdk_rdev_t *rdev; | |
3568 | int err; | |
3569 | ||
3570 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3571 | return -EINVAL; | |
3572 | minor = simple_strtoul(e+1, &e, 10); | |
3573 | if (*e && *e != '\n') | |
3574 | return -EINVAL; | |
3575 | dev = MKDEV(major, minor); | |
3576 | if (major != MAJOR(dev) || | |
3577 | minor != MINOR(dev)) | |
3578 | return -EOVERFLOW; | |
3579 | ||
3580 | ||
3581 | if (mddev->persistent) { | |
3582 | rdev = md_import_device(dev, mddev->major_version, | |
3583 | mddev->minor_version); | |
3584 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3585 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3586 | mdk_rdev_t, same_set); | |
3587 | err = super_types[mddev->major_version] | |
3588 | .load_super(rdev, rdev0, mddev->minor_version); | |
3589 | if (err < 0) | |
3590 | goto out; | |
3591 | } | |
c5d79adb N |
3592 | } else if (mddev->external) |
3593 | rdev = md_import_device(dev, -2, -1); | |
3594 | else | |
6d7ff738 N |
3595 | rdev = md_import_device(dev, -1, -1); |
3596 | ||
3597 | if (IS_ERR(rdev)) | |
3598 | return PTR_ERR(rdev); | |
3599 | err = bind_rdev_to_array(rdev, mddev); | |
3600 | out: | |
3601 | if (err) | |
3602 | export_rdev(rdev); | |
3603 | return err ? err : len; | |
3604 | } | |
3605 | ||
3606 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3607 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3608 | |
9b1d1dac PC |
3609 | static ssize_t |
3610 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3611 | { | |
3612 | char *end; | |
3613 | unsigned long chunk, end_chunk; | |
3614 | ||
3615 | if (!mddev->bitmap) | |
3616 | goto out; | |
3617 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3618 | while (*buf) { | |
3619 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3620 | if (buf == end) break; | |
3621 | if (*end == '-') { /* range */ | |
3622 | buf = end + 1; | |
3623 | end_chunk = simple_strtoul(buf, &end, 0); | |
3624 | if (buf == end) break; | |
3625 | } | |
3626 | if (*end && !isspace(*end)) break; | |
3627 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3628 | buf = skip_spaces(end); |
9b1d1dac PC |
3629 | } |
3630 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3631 | out: | |
3632 | return len; | |
3633 | } | |
3634 | ||
3635 | static struct md_sysfs_entry md_bitmap = | |
3636 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3637 | ||
a35b0d69 N |
3638 | static ssize_t |
3639 | size_show(mddev_t *mddev, char *page) | |
3640 | { | |
58c0fed4 AN |
3641 | return sprintf(page, "%llu\n", |
3642 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3643 | } |
3644 | ||
d71f9f88 | 3645 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3646 | |
3647 | static ssize_t | |
3648 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3649 | { | |
3650 | /* If array is inactive, we can reduce the component size, but | |
3651 | * not increase it (except from 0). | |
3652 | * If array is active, we can try an on-line resize | |
3653 | */ | |
b522adcd DW |
3654 | sector_t sectors; |
3655 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3656 | |
58c0fed4 AN |
3657 | if (err < 0) |
3658 | return err; | |
a35b0d69 | 3659 | if (mddev->pers) { |
58c0fed4 | 3660 | err = update_size(mddev, sectors); |
850b2b42 | 3661 | md_update_sb(mddev, 1); |
a35b0d69 | 3662 | } else { |
58c0fed4 AN |
3663 | if (mddev->dev_sectors == 0 || |
3664 | mddev->dev_sectors > sectors) | |
3665 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3666 | else |
3667 | err = -ENOSPC; | |
3668 | } | |
3669 | return err ? err : len; | |
3670 | } | |
3671 | ||
3672 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3673 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3674 | |
8bb93aac N |
3675 | |
3676 | /* Metdata version. | |
e691063a N |
3677 | * This is one of |
3678 | * 'none' for arrays with no metadata (good luck...) | |
3679 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3680 | * or N.M for internally known formats |
3681 | */ | |
3682 | static ssize_t | |
3683 | metadata_show(mddev_t *mddev, char *page) | |
3684 | { | |
3685 | if (mddev->persistent) | |
3686 | return sprintf(page, "%d.%d\n", | |
3687 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3688 | else if (mddev->external) |
3689 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3690 | else |
3691 | return sprintf(page, "none\n"); | |
3692 | } | |
3693 | ||
3694 | static ssize_t | |
3695 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3696 | { | |
3697 | int major, minor; | |
3698 | char *e; | |
ea43ddd8 N |
3699 | /* Changing the details of 'external' metadata is |
3700 | * always permitted. Otherwise there must be | |
3701 | * no devices attached to the array. | |
3702 | */ | |
3703 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3704 | ; | |
3705 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3706 | return -EBUSY; |
3707 | ||
3708 | if (cmd_match(buf, "none")) { | |
3709 | mddev->persistent = 0; | |
e691063a N |
3710 | mddev->external = 0; |
3711 | mddev->major_version = 0; | |
3712 | mddev->minor_version = 90; | |
3713 | return len; | |
3714 | } | |
3715 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3716 | size_t namelen = len-9; |
e691063a N |
3717 | if (namelen >= sizeof(mddev->metadata_type)) |
3718 | namelen = sizeof(mddev->metadata_type)-1; | |
3719 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3720 | mddev->metadata_type[namelen] = 0; | |
3721 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3722 | mddev->metadata_type[--namelen] = 0; | |
3723 | mddev->persistent = 0; | |
3724 | mddev->external = 1; | |
8bb93aac N |
3725 | mddev->major_version = 0; |
3726 | mddev->minor_version = 90; | |
3727 | return len; | |
3728 | } | |
3729 | major = simple_strtoul(buf, &e, 10); | |
3730 | if (e==buf || *e != '.') | |
3731 | return -EINVAL; | |
3732 | buf = e+1; | |
3733 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3734 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3735 | return -EINVAL; |
50511da3 | 3736 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3737 | return -ENOENT; |
3738 | mddev->major_version = major; | |
3739 | mddev->minor_version = minor; | |
3740 | mddev->persistent = 1; | |
e691063a | 3741 | mddev->external = 0; |
8bb93aac N |
3742 | return len; |
3743 | } | |
3744 | ||
3745 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3746 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3747 | |
24dd469d | 3748 | static ssize_t |
7eec314d | 3749 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3750 | { |
7eec314d | 3751 | char *type = "idle"; |
b6a9ce68 N |
3752 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3753 | type = "frozen"; | |
3754 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3755 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3756 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3757 | type = "reshape"; | |
3758 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3759 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3760 | type = "resync"; | |
3761 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3762 | type = "check"; | |
3763 | else | |
3764 | type = "repair"; | |
72a23c21 | 3765 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3766 | type = "recover"; |
3767 | } | |
3768 | return sprintf(page, "%s\n", type); | |
3769 | } | |
3770 | ||
7ebc0be7 N |
3771 | static void reap_sync_thread(mddev_t *mddev); |
3772 | ||
24dd469d | 3773 | static ssize_t |
7eec314d | 3774 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3775 | { |
7eec314d N |
3776 | if (!mddev->pers || !mddev->pers->sync_request) |
3777 | return -EINVAL; | |
3778 | ||
b6a9ce68 N |
3779 | if (cmd_match(page, "frozen")) |
3780 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3781 | else | |
3782 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3783 | ||
3784 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3785 | if (mddev->sync_thread) { |
3786 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 3787 | reap_sync_thread(mddev); |
7eec314d | 3788 | } |
03c902e1 N |
3789 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3790 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3791 | return -EBUSY; |
72a23c21 NB |
3792 | else if (cmd_match(page, "resync")) |
3793 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3794 | else if (cmd_match(page, "recover")) { | |
3795 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3796 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3797 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3798 | int err; |
3799 | if (mddev->pers->start_reshape == NULL) | |
3800 | return -EINVAL; | |
3801 | err = mddev->pers->start_reshape(mddev); | |
3802 | if (err) | |
3803 | return err; | |
a99ac971 | 3804 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3805 | } else { |
bce74dac | 3806 | if (cmd_match(page, "check")) |
7eec314d | 3807 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3808 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3809 | return -EINVAL; |
3810 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3811 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3812 | } |
03c902e1 | 3813 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3814 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 3815 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
3816 | return len; |
3817 | } | |
3818 | ||
9d88883e | 3819 | static ssize_t |
96de1e66 | 3820 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3821 | { |
3822 | return sprintf(page, "%llu\n", | |
3823 | (unsigned long long) mddev->resync_mismatches); | |
3824 | } | |
3825 | ||
80ca3a44 N |
3826 | static struct md_sysfs_entry md_scan_mode = |
3827 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3828 | |
96de1e66 | 3829 | |
80ca3a44 | 3830 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3831 | |
88202a0c N |
3832 | static ssize_t |
3833 | sync_min_show(mddev_t *mddev, char *page) | |
3834 | { | |
3835 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3836 | mddev->sync_speed_min ? "local": "system"); | |
3837 | } | |
3838 | ||
3839 | static ssize_t | |
3840 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3841 | { | |
3842 | int min; | |
3843 | char *e; | |
3844 | if (strncmp(buf, "system", 6)==0) { | |
3845 | mddev->sync_speed_min = 0; | |
3846 | return len; | |
3847 | } | |
3848 | min = simple_strtoul(buf, &e, 10); | |
3849 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3850 | return -EINVAL; | |
3851 | mddev->sync_speed_min = min; | |
3852 | return len; | |
3853 | } | |
3854 | ||
3855 | static struct md_sysfs_entry md_sync_min = | |
3856 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3857 | ||
3858 | static ssize_t | |
3859 | sync_max_show(mddev_t *mddev, char *page) | |
3860 | { | |
3861 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3862 | mddev->sync_speed_max ? "local": "system"); | |
3863 | } | |
3864 | ||
3865 | static ssize_t | |
3866 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3867 | { | |
3868 | int max; | |
3869 | char *e; | |
3870 | if (strncmp(buf, "system", 6)==0) { | |
3871 | mddev->sync_speed_max = 0; | |
3872 | return len; | |
3873 | } | |
3874 | max = simple_strtoul(buf, &e, 10); | |
3875 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3876 | return -EINVAL; | |
3877 | mddev->sync_speed_max = max; | |
3878 | return len; | |
3879 | } | |
3880 | ||
3881 | static struct md_sysfs_entry md_sync_max = | |
3882 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3883 | ||
d7f3d291 IP |
3884 | static ssize_t |
3885 | degraded_show(mddev_t *mddev, char *page) | |
3886 | { | |
3887 | return sprintf(page, "%d\n", mddev->degraded); | |
3888 | } | |
3889 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3890 | |
90b08710 BS |
3891 | static ssize_t |
3892 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3893 | { | |
3894 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3895 | } | |
3896 | ||
3897 | static ssize_t | |
3898 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3899 | { | |
3900 | long n; | |
3901 | ||
3902 | if (strict_strtol(buf, 10, &n)) | |
3903 | return -EINVAL; | |
3904 | ||
3905 | if (n != 0 && n != 1) | |
3906 | return -EINVAL; | |
3907 | ||
3908 | mddev->parallel_resync = n; | |
3909 | ||
3910 | if (mddev->sync_thread) | |
3911 | wake_up(&resync_wait); | |
3912 | ||
3913 | return len; | |
3914 | } | |
3915 | ||
3916 | /* force parallel resync, even with shared block devices */ | |
3917 | static struct md_sysfs_entry md_sync_force_parallel = | |
3918 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3919 | sync_force_parallel_show, sync_force_parallel_store); | |
3920 | ||
88202a0c N |
3921 | static ssize_t |
3922 | sync_speed_show(mddev_t *mddev, char *page) | |
3923 | { | |
3924 | unsigned long resync, dt, db; | |
d1a7c503 N |
3925 | if (mddev->curr_resync == 0) |
3926 | return sprintf(page, "none\n"); | |
9687a60c AN |
3927 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3928 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3929 | if (!dt) dt++; |
9687a60c AN |
3930 | db = resync - mddev->resync_mark_cnt; |
3931 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3932 | } |
3933 | ||
80ca3a44 | 3934 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3935 | |
3936 | static ssize_t | |
3937 | sync_completed_show(mddev_t *mddev, char *page) | |
3938 | { | |
13ae864b | 3939 | unsigned long long max_sectors, resync; |
88202a0c | 3940 | |
acb180b0 N |
3941 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3942 | return sprintf(page, "none\n"); | |
3943 | ||
88202a0c | 3944 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3945 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3946 | else |
58c0fed4 | 3947 | max_sectors = mddev->dev_sectors; |
88202a0c | 3948 | |
acb180b0 | 3949 | resync = mddev->curr_resync_completed; |
13ae864b | 3950 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
3951 | } |
3952 | ||
80ca3a44 | 3953 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3954 | |
5e96ee65 NB |
3955 | static ssize_t |
3956 | min_sync_show(mddev_t *mddev, char *page) | |
3957 | { | |
3958 | return sprintf(page, "%llu\n", | |
3959 | (unsigned long long)mddev->resync_min); | |
3960 | } | |
3961 | static ssize_t | |
3962 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3963 | { | |
3964 | unsigned long long min; | |
3965 | if (strict_strtoull(buf, 10, &min)) | |
3966 | return -EINVAL; | |
3967 | if (min > mddev->resync_max) | |
3968 | return -EINVAL; | |
3969 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3970 | return -EBUSY; | |
3971 | ||
3972 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3973 | if (mddev->chunk_sectors) { |
2ac06c33 | 3974 | sector_t temp = min; |
9d8f0363 | 3975 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3976 | return -EINVAL; |
3977 | } | |
3978 | mddev->resync_min = min; | |
3979 | ||
3980 | return len; | |
3981 | } | |
3982 | ||
3983 | static struct md_sysfs_entry md_min_sync = | |
3984 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3985 | ||
c6207277 N |
3986 | static ssize_t |
3987 | max_sync_show(mddev_t *mddev, char *page) | |
3988 | { | |
3989 | if (mddev->resync_max == MaxSector) | |
3990 | return sprintf(page, "max\n"); | |
3991 | else | |
3992 | return sprintf(page, "%llu\n", | |
3993 | (unsigned long long)mddev->resync_max); | |
3994 | } | |
3995 | static ssize_t | |
3996 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3997 | { | |
3998 | if (strncmp(buf, "max", 3) == 0) | |
3999 | mddev->resync_max = MaxSector; | |
4000 | else { | |
5e96ee65 NB |
4001 | unsigned long long max; |
4002 | if (strict_strtoull(buf, 10, &max)) | |
4003 | return -EINVAL; | |
4004 | if (max < mddev->resync_min) | |
c6207277 N |
4005 | return -EINVAL; |
4006 | if (max < mddev->resync_max && | |
4d484a4a | 4007 | mddev->ro == 0 && |
c6207277 N |
4008 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4009 | return -EBUSY; | |
4010 | ||
4011 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4012 | if (mddev->chunk_sectors) { |
2ac06c33 | 4013 | sector_t temp = max; |
9d8f0363 | 4014 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
4015 | return -EINVAL; |
4016 | } | |
4017 | mddev->resync_max = max; | |
4018 | } | |
4019 | wake_up(&mddev->recovery_wait); | |
4020 | return len; | |
4021 | } | |
4022 | ||
4023 | static struct md_sysfs_entry md_max_sync = | |
4024 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4025 | ||
e464eafd N |
4026 | static ssize_t |
4027 | suspend_lo_show(mddev_t *mddev, char *page) | |
4028 | { | |
4029 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4030 | } | |
4031 | ||
4032 | static ssize_t | |
4033 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
4034 | { | |
4035 | char *e; | |
4036 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4037 | unsigned long long old = mddev->suspend_lo; |
e464eafd | 4038 | |
b8d966ef N |
4039 | if (mddev->pers == NULL || |
4040 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4041 | return -EINVAL; |
4042 | if (buf == e || (*e && *e != '\n')) | |
4043 | return -EINVAL; | |
23ddff37 N |
4044 | |
4045 | mddev->suspend_lo = new; | |
4046 | if (new >= old) | |
4047 | /* Shrinking suspended region */ | |
e464eafd | 4048 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4049 | else { |
4050 | /* Expanding suspended region - need to wait */ | |
4051 | mddev->pers->quiesce(mddev, 1); | |
4052 | mddev->pers->quiesce(mddev, 0); | |
4053 | } | |
4054 | return len; | |
e464eafd N |
4055 | } |
4056 | static struct md_sysfs_entry md_suspend_lo = | |
4057 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4058 | ||
4059 | ||
4060 | static ssize_t | |
4061 | suspend_hi_show(mddev_t *mddev, char *page) | |
4062 | { | |
4063 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4064 | } | |
4065 | ||
4066 | static ssize_t | |
4067 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
4068 | { | |
4069 | char *e; | |
4070 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4071 | unsigned long long old = mddev->suspend_hi; |
e464eafd | 4072 | |
b8d966ef N |
4073 | if (mddev->pers == NULL || |
4074 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4075 | return -EINVAL; |
4076 | if (buf == e || (*e && *e != '\n')) | |
4077 | return -EINVAL; | |
23ddff37 N |
4078 | |
4079 | mddev->suspend_hi = new; | |
4080 | if (new <= old) | |
4081 | /* Shrinking suspended region */ | |
4082 | mddev->pers->quiesce(mddev, 2); | |
4083 | else { | |
4084 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4085 | mddev->pers->quiesce(mddev, 1); |
4086 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 N |
4087 | } |
4088 | return len; | |
e464eafd N |
4089 | } |
4090 | static struct md_sysfs_entry md_suspend_hi = | |
4091 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4092 | ||
08a02ecd N |
4093 | static ssize_t |
4094 | reshape_position_show(mddev_t *mddev, char *page) | |
4095 | { | |
4096 | if (mddev->reshape_position != MaxSector) | |
4097 | return sprintf(page, "%llu\n", | |
4098 | (unsigned long long)mddev->reshape_position); | |
4099 | strcpy(page, "none\n"); | |
4100 | return 5; | |
4101 | } | |
4102 | ||
4103 | static ssize_t | |
4104 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
4105 | { | |
4106 | char *e; | |
4107 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4108 | if (mddev->pers) | |
4109 | return -EBUSY; | |
4110 | if (buf == e || (*e && *e != '\n')) | |
4111 | return -EINVAL; | |
4112 | mddev->reshape_position = new; | |
4113 | mddev->delta_disks = 0; | |
4114 | mddev->new_level = mddev->level; | |
4115 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4116 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
4117 | return len; |
4118 | } | |
4119 | ||
4120 | static struct md_sysfs_entry md_reshape_position = | |
4121 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4122 | reshape_position_store); | |
4123 | ||
b522adcd DW |
4124 | static ssize_t |
4125 | array_size_show(mddev_t *mddev, char *page) | |
4126 | { | |
4127 | if (mddev->external_size) | |
4128 | return sprintf(page, "%llu\n", | |
4129 | (unsigned long long)mddev->array_sectors/2); | |
4130 | else | |
4131 | return sprintf(page, "default\n"); | |
4132 | } | |
4133 | ||
4134 | static ssize_t | |
4135 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
4136 | { | |
4137 | sector_t sectors; | |
4138 | ||
4139 | if (strncmp(buf, "default", 7) == 0) { | |
4140 | if (mddev->pers) | |
4141 | sectors = mddev->pers->size(mddev, 0, 0); | |
4142 | else | |
4143 | sectors = mddev->array_sectors; | |
4144 | ||
4145 | mddev->external_size = 0; | |
4146 | } else { | |
4147 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4148 | return -EINVAL; | |
4149 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4150 | return -E2BIG; |
b522adcd DW |
4151 | |
4152 | mddev->external_size = 1; | |
4153 | } | |
4154 | ||
4155 | mddev->array_sectors = sectors; | |
cbe6ef1d N |
4156 | if (mddev->pers) { |
4157 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 4158 | revalidate_disk(mddev->gendisk); |
cbe6ef1d | 4159 | } |
b522adcd DW |
4160 | return len; |
4161 | } | |
4162 | ||
4163 | static struct md_sysfs_entry md_array_size = | |
4164 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4165 | array_size_store); | |
e464eafd | 4166 | |
eae1701f N |
4167 | static struct attribute *md_default_attrs[] = { |
4168 | &md_level.attr, | |
d4dbd025 | 4169 | &md_layout.attr, |
eae1701f | 4170 | &md_raid_disks.attr, |
3b34380a | 4171 | &md_chunk_size.attr, |
a35b0d69 | 4172 | &md_size.attr, |
a94213b1 | 4173 | &md_resync_start.attr, |
8bb93aac | 4174 | &md_metadata.attr, |
6d7ff738 | 4175 | &md_new_device.attr, |
16f17b39 | 4176 | &md_safe_delay.attr, |
9e653b63 | 4177 | &md_array_state.attr, |
08a02ecd | 4178 | &md_reshape_position.attr, |
b522adcd | 4179 | &md_array_size.attr, |
1e50915f | 4180 | &max_corr_read_errors.attr, |
411036fa N |
4181 | NULL, |
4182 | }; | |
4183 | ||
4184 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4185 | &md_scan_mode.attr, |
9d88883e | 4186 | &md_mismatches.attr, |
88202a0c N |
4187 | &md_sync_min.attr, |
4188 | &md_sync_max.attr, | |
4189 | &md_sync_speed.attr, | |
90b08710 | 4190 | &md_sync_force_parallel.attr, |
88202a0c | 4191 | &md_sync_completed.attr, |
5e96ee65 | 4192 | &md_min_sync.attr, |
c6207277 | 4193 | &md_max_sync.attr, |
e464eafd N |
4194 | &md_suspend_lo.attr, |
4195 | &md_suspend_hi.attr, | |
9b1d1dac | 4196 | &md_bitmap.attr, |
d7f3d291 | 4197 | &md_degraded.attr, |
eae1701f N |
4198 | NULL, |
4199 | }; | |
411036fa N |
4200 | static struct attribute_group md_redundancy_group = { |
4201 | .name = NULL, | |
4202 | .attrs = md_redundancy_attrs, | |
4203 | }; | |
4204 | ||
eae1701f N |
4205 | |
4206 | static ssize_t | |
4207 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4208 | { | |
4209 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4210 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4211 | ssize_t rv; |
eae1701f N |
4212 | |
4213 | if (!entry->show) | |
4214 | return -EIO; | |
5dc5cf7d IM |
4215 | rv = mddev_lock(mddev); |
4216 | if (!rv) { | |
4217 | rv = entry->show(mddev, page); | |
4218 | mddev_unlock(mddev); | |
4219 | } | |
96de1e66 | 4220 | return rv; |
eae1701f N |
4221 | } |
4222 | ||
4223 | static ssize_t | |
4224 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4225 | const char *page, size_t length) | |
4226 | { | |
4227 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4228 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4229 | ssize_t rv; |
eae1701f N |
4230 | |
4231 | if (!entry->store) | |
4232 | return -EIO; | |
67463acb N |
4233 | if (!capable(CAP_SYS_ADMIN)) |
4234 | return -EACCES; | |
5dc5cf7d | 4235 | rv = mddev_lock(mddev); |
d3374825 N |
4236 | if (mddev->hold_active == UNTIL_IOCTL) |
4237 | mddev->hold_active = 0; | |
5dc5cf7d IM |
4238 | if (!rv) { |
4239 | rv = entry->store(mddev, page, length); | |
4240 | mddev_unlock(mddev); | |
4241 | } | |
96de1e66 | 4242 | return rv; |
eae1701f N |
4243 | } |
4244 | ||
4245 | static void md_free(struct kobject *ko) | |
4246 | { | |
4247 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
4248 | |
4249 | if (mddev->sysfs_state) | |
4250 | sysfs_put(mddev->sysfs_state); | |
4251 | ||
4252 | if (mddev->gendisk) { | |
4253 | del_gendisk(mddev->gendisk); | |
4254 | put_disk(mddev->gendisk); | |
4255 | } | |
4256 | if (mddev->queue) | |
4257 | blk_cleanup_queue(mddev->queue); | |
4258 | ||
eae1701f N |
4259 | kfree(mddev); |
4260 | } | |
4261 | ||
52cf25d0 | 4262 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4263 | .show = md_attr_show, |
4264 | .store = md_attr_store, | |
4265 | }; | |
4266 | static struct kobj_type md_ktype = { | |
4267 | .release = md_free, | |
4268 | .sysfs_ops = &md_sysfs_ops, | |
4269 | .default_attrs = md_default_attrs, | |
4270 | }; | |
4271 | ||
1da177e4 LT |
4272 | int mdp_major = 0; |
4273 | ||
5fd3a17e DW |
4274 | static void mddev_delayed_delete(struct work_struct *ws) |
4275 | { | |
4276 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
4277 | ||
43a70507 | 4278 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4279 | kobject_del(&mddev->kobj); |
4280 | kobject_put(&mddev->kobj); | |
4281 | } | |
4282 | ||
efeb53c0 | 4283 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4284 | { |
48c9c27b | 4285 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
4286 | mddev_t *mddev = mddev_find(dev); |
4287 | struct gendisk *disk; | |
efeb53c0 N |
4288 | int partitioned; |
4289 | int shift; | |
4290 | int unit; | |
3830c62f | 4291 | int error; |
1da177e4 LT |
4292 | |
4293 | if (!mddev) | |
efeb53c0 N |
4294 | return -ENODEV; |
4295 | ||
4296 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4297 | shift = partitioned ? MdpMinorShift : 0; | |
4298 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4299 | |
e804ac78 TH |
4300 | /* wait for any previous instance of this device to be |
4301 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4302 | */ |
e804ac78 | 4303 | flush_workqueue(md_misc_wq); |
d3374825 | 4304 | |
48c9c27b | 4305 | mutex_lock(&disks_mutex); |
0909dc44 N |
4306 | error = -EEXIST; |
4307 | if (mddev->gendisk) | |
4308 | goto abort; | |
efeb53c0 N |
4309 | |
4310 | if (name) { | |
4311 | /* Need to ensure that 'name' is not a duplicate. | |
4312 | */ | |
4313 | mddev_t *mddev2; | |
4314 | spin_lock(&all_mddevs_lock); | |
4315 | ||
4316 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4317 | if (mddev2->gendisk && | |
4318 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4319 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4320 | goto abort; |
efeb53c0 N |
4321 | } |
4322 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4323 | } |
8b765398 | 4324 | |
0909dc44 | 4325 | error = -ENOMEM; |
8b765398 | 4326 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4327 | if (!mddev->queue) |
4328 | goto abort; | |
409c57f3 N |
4329 | mddev->queue->queuedata = mddev; |
4330 | ||
409c57f3 | 4331 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 4332 | |
1da177e4 LT |
4333 | disk = alloc_disk(1 << shift); |
4334 | if (!disk) { | |
8b765398 N |
4335 | blk_cleanup_queue(mddev->queue); |
4336 | mddev->queue = NULL; | |
0909dc44 | 4337 | goto abort; |
1da177e4 | 4338 | } |
efeb53c0 | 4339 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4340 | disk->first_minor = unit << shift; |
efeb53c0 N |
4341 | if (name) |
4342 | strcpy(disk->disk_name, name); | |
4343 | else if (partitioned) | |
1da177e4 | 4344 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4345 | else |
1da177e4 | 4346 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4347 | disk->fops = &md_fops; |
4348 | disk->private_data = mddev; | |
4349 | disk->queue = mddev->queue; | |
b0140891 | 4350 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4351 | /* Allow extended partitions. This makes the |
d3374825 | 4352 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4353 | * remove it now. |
4354 | */ | |
4355 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4356 | mddev->gendisk = disk; |
b0140891 N |
4357 | /* As soon as we call add_disk(), another thread could get |
4358 | * through to md_open, so make sure it doesn't get too far | |
4359 | */ | |
4360 | mutex_lock(&mddev->open_mutex); | |
4361 | add_disk(disk); | |
4362 | ||
ed9e1982 TH |
4363 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4364 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4365 | if (error) { |
4366 | /* This isn't possible, but as kobject_init_and_add is marked | |
4367 | * __must_check, we must do something with the result | |
4368 | */ | |
5e55e2f5 N |
4369 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4370 | disk->disk_name); | |
0909dc44 N |
4371 | error = 0; |
4372 | } | |
00bcb4ac N |
4373 | if (mddev->kobj.sd && |
4374 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4375 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4376 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4377 | abort: |
4378 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4379 | if (!error && mddev->kobj.sd) { |
3830c62f | 4380 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4381 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4382 | } |
d3374825 | 4383 | mddev_put(mddev); |
0909dc44 | 4384 | return error; |
efeb53c0 N |
4385 | } |
4386 | ||
4387 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4388 | { | |
4389 | md_alloc(dev, NULL); | |
1da177e4 LT |
4390 | return NULL; |
4391 | } | |
4392 | ||
efeb53c0 N |
4393 | static int add_named_array(const char *val, struct kernel_param *kp) |
4394 | { | |
4395 | /* val must be "md_*" where * is not all digits. | |
4396 | * We allocate an array with a large free minor number, and | |
4397 | * set the name to val. val must not already be an active name. | |
4398 | */ | |
4399 | int len = strlen(val); | |
4400 | char buf[DISK_NAME_LEN]; | |
4401 | ||
4402 | while (len && val[len-1] == '\n') | |
4403 | len--; | |
4404 | if (len >= DISK_NAME_LEN) | |
4405 | return -E2BIG; | |
4406 | strlcpy(buf, val, len+1); | |
4407 | if (strncmp(buf, "md_", 3) != 0) | |
4408 | return -EINVAL; | |
4409 | return md_alloc(0, buf); | |
4410 | } | |
4411 | ||
1da177e4 LT |
4412 | static void md_safemode_timeout(unsigned long data) |
4413 | { | |
4414 | mddev_t *mddev = (mddev_t *) data; | |
4415 | ||
0fd62b86 NB |
4416 | if (!atomic_read(&mddev->writes_pending)) { |
4417 | mddev->safemode = 1; | |
4418 | if (mddev->external) | |
00bcb4ac | 4419 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4420 | } |
1da177e4 LT |
4421 | md_wakeup_thread(mddev->thread); |
4422 | } | |
4423 | ||
6ff8d8ec | 4424 | static int start_dirty_degraded; |
1da177e4 | 4425 | |
390ee602 | 4426 | int md_run(mddev_t *mddev) |
1da177e4 | 4427 | { |
2604b703 | 4428 | int err; |
1da177e4 | 4429 | mdk_rdev_t *rdev; |
2604b703 | 4430 | struct mdk_personality *pers; |
1da177e4 | 4431 | |
a757e64c N |
4432 | if (list_empty(&mddev->disks)) |
4433 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4434 | return -EINVAL; |
1da177e4 LT |
4435 | |
4436 | if (mddev->pers) | |
4437 | return -EBUSY; | |
bb4f1e9d N |
4438 | /* Cannot run until previous stop completes properly */ |
4439 | if (mddev->sysfs_active) | |
4440 | return -EBUSY; | |
b6eb127d | 4441 | |
1da177e4 LT |
4442 | /* |
4443 | * Analyze all RAID superblock(s) | |
4444 | */ | |
1ec4a939 N |
4445 | if (!mddev->raid_disks) { |
4446 | if (!mddev->persistent) | |
4447 | return -EINVAL; | |
a757e64c | 4448 | analyze_sbs(mddev); |
1ec4a939 | 4449 | } |
1da177e4 | 4450 | |
d9d166c2 N |
4451 | if (mddev->level != LEVEL_NONE) |
4452 | request_module("md-level-%d", mddev->level); | |
4453 | else if (mddev->clevel[0]) | |
4454 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4455 | |
4456 | /* | |
4457 | * Drop all container device buffers, from now on | |
4458 | * the only valid external interface is through the md | |
4459 | * device. | |
1da177e4 | 4460 | */ |
159ec1fc | 4461 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 4462 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4463 | continue; |
4464 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4465 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4466 | |
4467 | /* perform some consistency tests on the device. | |
4468 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4469 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4470 | */ |
a6ff7e08 JB |
4471 | if (rdev->meta_bdev) { |
4472 | /* Nothing to check */; | |
4473 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
4474 | if (mddev->dev_sectors && |
4475 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4476 | > rdev->sb_start) { |
f0d76d70 N |
4477 | printk("md: %s: data overlaps metadata\n", |
4478 | mdname(mddev)); | |
4479 | return -EINVAL; | |
4480 | } | |
4481 | } else { | |
0f420358 | 4482 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4483 | > rdev->data_offset) { |
4484 | printk("md: %s: metadata overlaps data\n", | |
4485 | mdname(mddev)); | |
4486 | return -EINVAL; | |
4487 | } | |
4488 | } | |
00bcb4ac | 4489 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4490 | } |
4491 | ||
a167f663 N |
4492 | if (mddev->bio_set == NULL) |
4493 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev)); | |
4494 | ||
1da177e4 | 4495 | spin_lock(&pers_lock); |
d9d166c2 | 4496 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4497 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4498 | spin_unlock(&pers_lock); |
d9d166c2 N |
4499 | if (mddev->level != LEVEL_NONE) |
4500 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4501 | mddev->level); | |
4502 | else | |
4503 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4504 | mddev->clevel); | |
1da177e4 LT |
4505 | return -EINVAL; |
4506 | } | |
2604b703 | 4507 | mddev->pers = pers; |
1da177e4 | 4508 | spin_unlock(&pers_lock); |
34817e8c N |
4509 | if (mddev->level != pers->level) { |
4510 | mddev->level = pers->level; | |
4511 | mddev->new_level = pers->level; | |
4512 | } | |
d9d166c2 | 4513 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4514 | |
f6705578 | 4515 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4516 | pers->start_reshape == NULL) { |
f6705578 N |
4517 | /* This personality cannot handle reshaping... */ |
4518 | mddev->pers = NULL; | |
4519 | module_put(pers->owner); | |
4520 | return -EINVAL; | |
4521 | } | |
4522 | ||
7dd5e7c3 N |
4523 | if (pers->sync_request) { |
4524 | /* Warn if this is a potentially silly | |
4525 | * configuration. | |
4526 | */ | |
4527 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4528 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4529 | int warned = 0; |
159ec1fc CR |
4530 | |
4531 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4532 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4533 | if (rdev < rdev2 && |
4534 | rdev->bdev->bd_contains == | |
4535 | rdev2->bdev->bd_contains) { | |
4536 | printk(KERN_WARNING | |
4537 | "%s: WARNING: %s appears to be" | |
4538 | " on the same physical disk as" | |
4539 | " %s.\n", | |
4540 | mdname(mddev), | |
4541 | bdevname(rdev->bdev,b), | |
4542 | bdevname(rdev2->bdev,b2)); | |
4543 | warned = 1; | |
4544 | } | |
4545 | } | |
159ec1fc | 4546 | |
7dd5e7c3 N |
4547 | if (warned) |
4548 | printk(KERN_WARNING | |
4549 | "True protection against single-disk" | |
4550 | " failure might be compromised.\n"); | |
4551 | } | |
4552 | ||
657390d2 | 4553 | mddev->recovery = 0; |
58c0fed4 AN |
4554 | /* may be over-ridden by personality */ |
4555 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4556 | ||
6ff8d8ec | 4557 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4558 | |
0f9552b5 | 4559 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
4560 | mddev->ro = 2; /* read-only, but switch on first write */ |
4561 | ||
b15c2e57 | 4562 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4563 | if (err) |
4564 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4565 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4566 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4567 | " but 'external_size' not in effect?\n", __func__); | |
4568 | printk(KERN_ERR | |
4569 | "md: invalid array_size %llu > default size %llu\n", | |
4570 | (unsigned long long)mddev->array_sectors / 2, | |
4571 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4572 | err = -EINVAL; | |
4573 | mddev->pers->stop(mddev); | |
4574 | } | |
4575 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4576 | err = bitmap_create(mddev); |
4577 | if (err) { | |
4578 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4579 | mdname(mddev), err); | |
4580 | mddev->pers->stop(mddev); | |
4581 | } | |
4582 | } | |
1da177e4 | 4583 | if (err) { |
1da177e4 LT |
4584 | module_put(mddev->pers->owner); |
4585 | mddev->pers = NULL; | |
32a7627c N |
4586 | bitmap_destroy(mddev); |
4587 | return err; | |
1da177e4 | 4588 | } |
5e55e2f5 | 4589 | if (mddev->pers->sync_request) { |
00bcb4ac N |
4590 | if (mddev->kobj.sd && |
4591 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
4592 | printk(KERN_WARNING |
4593 | "md: cannot register extra attributes for %s\n", | |
4594 | mdname(mddev)); | |
00bcb4ac | 4595 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4596 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4597 | mddev->ro = 0; |
4598 | ||
1da177e4 | 4599 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
4600 | atomic_set(&mddev->max_corr_read_errors, |
4601 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
4602 | mddev->safemode = 0; |
4603 | mddev->safemode_timer.function = md_safemode_timeout; | |
4604 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4605 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4606 | mddev->in_sync = 1; |
0ca69886 N |
4607 | smp_wmb(); |
4608 | mddev->ready = 1; | |
159ec1fc | 4609 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4610 | if (rdev->raid_disk >= 0) { |
4611 | char nm[20]; | |
4612 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 | 4613 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
00bcb4ac | 4614 | /* failure here is OK */; |
86e6ffdd | 4615 | } |
1da177e4 LT |
4616 | |
4617 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4618 | ||
850b2b42 N |
4619 | if (mddev->flags) |
4620 | md_update_sb(mddev, 0); | |
1da177e4 | 4621 | |
0b8c9de0 N |
4622 | md_wakeup_thread(mddev->thread); |
4623 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4624 | |
d7603b7e | 4625 | md_new_event(mddev); |
00bcb4ac N |
4626 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
4627 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 4628 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
4629 | return 0; |
4630 | } | |
390ee602 | 4631 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 4632 | |
fe60b014 N |
4633 | static int do_md_run(mddev_t *mddev) |
4634 | { | |
4635 | int err; | |
4636 | ||
4637 | err = md_run(mddev); | |
4638 | if (err) | |
4639 | goto out; | |
69e51b44 N |
4640 | err = bitmap_load(mddev); |
4641 | if (err) { | |
4642 | bitmap_destroy(mddev); | |
4643 | goto out; | |
4644 | } | |
fe60b014 N |
4645 | set_capacity(mddev->gendisk, mddev->array_sectors); |
4646 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 4647 | mddev->changed = 1; |
fe60b014 N |
4648 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
4649 | out: | |
4650 | return err; | |
4651 | } | |
4652 | ||
1da177e4 LT |
4653 | static int restart_array(mddev_t *mddev) |
4654 | { | |
4655 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4656 | |
80fab1d7 | 4657 | /* Complain if it has no devices */ |
1da177e4 | 4658 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4659 | return -ENXIO; |
4660 | if (!mddev->pers) | |
4661 | return -EINVAL; | |
4662 | if (!mddev->ro) | |
4663 | return -EBUSY; | |
4664 | mddev->safemode = 0; | |
4665 | mddev->ro = 0; | |
4666 | set_disk_ro(disk, 0); | |
4667 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4668 | mdname(mddev)); | |
4669 | /* Kick recovery or resync if necessary */ | |
4670 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4671 | md_wakeup_thread(mddev->thread); | |
4672 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 4673 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 4674 | return 0; |
1da177e4 LT |
4675 | } |
4676 | ||
acc55e22 N |
4677 | /* similar to deny_write_access, but accounts for our holding a reference |
4678 | * to the file ourselves */ | |
4679 | static int deny_bitmap_write_access(struct file * file) | |
4680 | { | |
4681 | struct inode *inode = file->f_mapping->host; | |
4682 | ||
4683 | spin_lock(&inode->i_lock); | |
4684 | if (atomic_read(&inode->i_writecount) > 1) { | |
4685 | spin_unlock(&inode->i_lock); | |
4686 | return -ETXTBSY; | |
4687 | } | |
4688 | atomic_set(&inode->i_writecount, -1); | |
4689 | spin_unlock(&inode->i_lock); | |
4690 | ||
4691 | return 0; | |
4692 | } | |
4693 | ||
43a70507 | 4694 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
4695 | { |
4696 | struct inode *inode = file->f_mapping->host; | |
4697 | ||
4698 | spin_lock(&inode->i_lock); | |
4699 | atomic_set(&inode->i_writecount, 1); | |
4700 | spin_unlock(&inode->i_lock); | |
4701 | } | |
4702 | ||
6177b472 N |
4703 | static void md_clean(mddev_t *mddev) |
4704 | { | |
4705 | mddev->array_sectors = 0; | |
4706 | mddev->external_size = 0; | |
4707 | mddev->dev_sectors = 0; | |
4708 | mddev->raid_disks = 0; | |
4709 | mddev->recovery_cp = 0; | |
4710 | mddev->resync_min = 0; | |
4711 | mddev->resync_max = MaxSector; | |
4712 | mddev->reshape_position = MaxSector; | |
4713 | mddev->external = 0; | |
4714 | mddev->persistent = 0; | |
4715 | mddev->level = LEVEL_NONE; | |
4716 | mddev->clevel[0] = 0; | |
4717 | mddev->flags = 0; | |
4718 | mddev->ro = 0; | |
4719 | mddev->metadata_type[0] = 0; | |
4720 | mddev->chunk_sectors = 0; | |
4721 | mddev->ctime = mddev->utime = 0; | |
4722 | mddev->layout = 0; | |
4723 | mddev->max_disks = 0; | |
4724 | mddev->events = 0; | |
a8707c08 | 4725 | mddev->can_decrease_events = 0; |
6177b472 N |
4726 | mddev->delta_disks = 0; |
4727 | mddev->new_level = LEVEL_NONE; | |
4728 | mddev->new_layout = 0; | |
4729 | mddev->new_chunk_sectors = 0; | |
4730 | mddev->curr_resync = 0; | |
4731 | mddev->resync_mismatches = 0; | |
4732 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4733 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4734 | mddev->recovery = 0; | |
4735 | mddev->in_sync = 0; | |
f0b4f7e2 | 4736 | mddev->changed = 0; |
6177b472 | 4737 | mddev->degraded = 0; |
6177b472 N |
4738 | mddev->safemode = 0; |
4739 | mddev->bitmap_info.offset = 0; | |
4740 | mddev->bitmap_info.default_offset = 0; | |
4741 | mddev->bitmap_info.chunksize = 0; | |
4742 | mddev->bitmap_info.daemon_sleep = 0; | |
4743 | mddev->bitmap_info.max_write_behind = 0; | |
4744 | } | |
4745 | ||
defad61a | 4746 | static void __md_stop_writes(mddev_t *mddev) |
a047e125 N |
4747 | { |
4748 | if (mddev->sync_thread) { | |
4749 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4750 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 4751 | reap_sync_thread(mddev); |
a047e125 N |
4752 | } |
4753 | ||
4754 | del_timer_sync(&mddev->safemode_timer); | |
4755 | ||
4756 | bitmap_flush(mddev); | |
4757 | md_super_wait(mddev); | |
4758 | ||
4759 | if (!mddev->in_sync || mddev->flags) { | |
4760 | /* mark array as shutdown cleanly */ | |
4761 | mddev->in_sync = 1; | |
4762 | md_update_sb(mddev, 1); | |
4763 | } | |
4764 | } | |
defad61a N |
4765 | |
4766 | void md_stop_writes(mddev_t *mddev) | |
4767 | { | |
4768 | mddev_lock(mddev); | |
4769 | __md_stop_writes(mddev); | |
4770 | mddev_unlock(mddev); | |
4771 | } | |
390ee602 | 4772 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 4773 | |
390ee602 | 4774 | void md_stop(mddev_t *mddev) |
6177b472 | 4775 | { |
0ca69886 | 4776 | mddev->ready = 0; |
6177b472 N |
4777 | mddev->pers->stop(mddev); |
4778 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
4779 | mddev->to_remove = &md_redundancy_group; | |
4780 | module_put(mddev->pers->owner); | |
4781 | mddev->pers = NULL; | |
cca9cf90 | 4782 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 4783 | } |
390ee602 | 4784 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 4785 | |
a4bd82d0 N |
4786 | static int md_set_readonly(mddev_t *mddev, int is_open) |
4787 | { | |
4788 | int err = 0; | |
4789 | mutex_lock(&mddev->open_mutex); | |
4790 | if (atomic_read(&mddev->openers) > is_open) { | |
4791 | printk("md: %s still in use.\n",mdname(mddev)); | |
4792 | err = -EBUSY; | |
4793 | goto out; | |
4794 | } | |
4795 | if (mddev->pers) { | |
defad61a | 4796 | __md_stop_writes(mddev); |
a4bd82d0 N |
4797 | |
4798 | err = -ENXIO; | |
4799 | if (mddev->ro==1) | |
4800 | goto out; | |
4801 | mddev->ro = 1; | |
4802 | set_disk_ro(mddev->gendisk, 1); | |
4803 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 4804 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
a4bd82d0 N |
4805 | err = 0; |
4806 | } | |
4807 | out: | |
4808 | mutex_unlock(&mddev->open_mutex); | |
4809 | return err; | |
4810 | } | |
4811 | ||
9e653b63 N |
4812 | /* mode: |
4813 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
4814 | * 2 - stop but do not disassemble array |
4815 | */ | |
df5b20cf | 4816 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 | 4817 | { |
1da177e4 | 4818 | struct gendisk *disk = mddev->gendisk; |
c4647292 | 4819 | mdk_rdev_t *rdev; |
1da177e4 | 4820 | |
c8c00a69 | 4821 | mutex_lock(&mddev->open_mutex); |
bb4f1e9d N |
4822 | if (atomic_read(&mddev->openers) > is_open || |
4823 | mddev->sysfs_active) { | |
df5b20cf | 4824 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 N |
4825 | mutex_unlock(&mddev->open_mutex); |
4826 | return -EBUSY; | |
4827 | } | |
1da177e4 | 4828 | |
6e17b027 | 4829 | if (mddev->pers) { |
a4bd82d0 N |
4830 | if (mddev->ro) |
4831 | set_disk_ro(disk, 0); | |
409c57f3 | 4832 | |
defad61a | 4833 | __md_stop_writes(mddev); |
a4bd82d0 N |
4834 | md_stop(mddev); |
4835 | mddev->queue->merge_bvec_fn = NULL; | |
a4bd82d0 | 4836 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 4837 | |
a4bd82d0 | 4838 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 4839 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 4840 | |
a4bd82d0 N |
4841 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4842 | if (rdev->raid_disk >= 0) { | |
4843 | char nm[20]; | |
4844 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4845 | sysfs_remove_link(&mddev->kobj, nm); | |
4846 | } | |
c4647292 | 4847 | |
a4bd82d0 | 4848 | set_capacity(disk, 0); |
6e17b027 | 4849 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 4850 | mddev->changed = 1; |
a4bd82d0 | 4851 | revalidate_disk(disk); |
0d4ca600 | 4852 | |
a4bd82d0 N |
4853 | if (mddev->ro) |
4854 | mddev->ro = 0; | |
6e17b027 N |
4855 | } else |
4856 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
4857 | /* |
4858 | * Free resources if final stop | |
4859 | */ | |
9e653b63 | 4860 | if (mode == 0) { |
1da177e4 LT |
4861 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4862 | ||
978f946b | 4863 | bitmap_destroy(mddev); |
c3d9714e N |
4864 | if (mddev->bitmap_info.file) { |
4865 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
4866 | fput(mddev->bitmap_info.file); | |
4867 | mddev->bitmap_info.file = NULL; | |
978f946b | 4868 | } |
c3d9714e | 4869 | mddev->bitmap_info.offset = 0; |
978f946b | 4870 | |
1da177e4 LT |
4871 | export_array(mddev); |
4872 | ||
6177b472 | 4873 | md_clean(mddev); |
934d9c23 | 4874 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4875 | if (mddev->hold_active == UNTIL_STOP) |
4876 | mddev->hold_active = 0; | |
a4bd82d0 | 4877 | } |
3f9d99c1 | 4878 | blk_integrity_unregister(disk); |
d7603b7e | 4879 | md_new_event(mddev); |
00bcb4ac | 4880 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 4881 | return 0; |
1da177e4 LT |
4882 | } |
4883 | ||
fdee8ae4 | 4884 | #ifndef MODULE |
1da177e4 LT |
4885 | static void autorun_array(mddev_t *mddev) |
4886 | { | |
4887 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4888 | int err; |
4889 | ||
a757e64c | 4890 | if (list_empty(&mddev->disks)) |
1da177e4 | 4891 | return; |
1da177e4 LT |
4892 | |
4893 | printk(KERN_INFO "md: running: "); | |
4894 | ||
159ec1fc | 4895 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4896 | char b[BDEVNAME_SIZE]; |
4897 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4898 | } | |
4899 | printk("\n"); | |
4900 | ||
d710e138 | 4901 | err = do_md_run(mddev); |
1da177e4 LT |
4902 | if (err) { |
4903 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4904 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4905 | } |
4906 | } | |
4907 | ||
4908 | /* | |
4909 | * lets try to run arrays based on all disks that have arrived | |
4910 | * until now. (those are in pending_raid_disks) | |
4911 | * | |
4912 | * the method: pick the first pending disk, collect all disks with | |
4913 | * the same UUID, remove all from the pending list and put them into | |
4914 | * the 'same_array' list. Then order this list based on superblock | |
4915 | * update time (freshest comes first), kick out 'old' disks and | |
4916 | * compare superblocks. If everything's fine then run it. | |
4917 | * | |
4918 | * If "unit" is allocated, then bump its reference count | |
4919 | */ | |
4920 | static void autorun_devices(int part) | |
4921 | { | |
159ec1fc | 4922 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4923 | mddev_t *mddev; |
4924 | char b[BDEVNAME_SIZE]; | |
4925 | ||
4926 | printk(KERN_INFO "md: autorun ...\n"); | |
4927 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4928 | int unit; |
1da177e4 | 4929 | dev_t dev; |
ad01c9e3 | 4930 | LIST_HEAD(candidates); |
1da177e4 LT |
4931 | rdev0 = list_entry(pending_raid_disks.next, |
4932 | mdk_rdev_t, same_set); | |
4933 | ||
4934 | printk(KERN_INFO "md: considering %s ...\n", | |
4935 | bdevname(rdev0->bdev,b)); | |
4936 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4937 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4938 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4939 | printk(KERN_INFO "md: adding %s ...\n", | |
4940 | bdevname(rdev->bdev,b)); | |
4941 | list_move(&rdev->same_set, &candidates); | |
4942 | } | |
4943 | /* | |
4944 | * now we have a set of devices, with all of them having | |
4945 | * mostly sane superblocks. It's time to allocate the | |
4946 | * mddev. | |
4947 | */ | |
e8703fe1 N |
4948 | if (part) { |
4949 | dev = MKDEV(mdp_major, | |
4950 | rdev0->preferred_minor << MdpMinorShift); | |
4951 | unit = MINOR(dev) >> MdpMinorShift; | |
4952 | } else { | |
4953 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4954 | unit = MINOR(dev); | |
4955 | } | |
4956 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4957 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4958 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4959 | break; | |
4960 | } | |
1da177e4 LT |
4961 | |
4962 | md_probe(dev, NULL, NULL); | |
4963 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4964 | if (!mddev || !mddev->gendisk) { |
4965 | if (mddev) | |
4966 | mddev_put(mddev); | |
4967 | printk(KERN_ERR | |
1da177e4 LT |
4968 | "md: cannot allocate memory for md drive.\n"); |
4969 | break; | |
4970 | } | |
4971 | if (mddev_lock(mddev)) | |
4972 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4973 | mdname(mddev)); | |
4974 | else if (mddev->raid_disks || mddev->major_version | |
4975 | || !list_empty(&mddev->disks)) { | |
4976 | printk(KERN_WARNING | |
4977 | "md: %s already running, cannot run %s\n", | |
4978 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4979 | mddev_unlock(mddev); | |
4980 | } else { | |
4981 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4982 | mddev->persistent = 1; |
159ec1fc | 4983 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4984 | list_del_init(&rdev->same_set); |
4985 | if (bind_rdev_to_array(rdev, mddev)) | |
4986 | export_rdev(rdev); | |
4987 | } | |
4988 | autorun_array(mddev); | |
4989 | mddev_unlock(mddev); | |
4990 | } | |
4991 | /* on success, candidates will be empty, on error | |
4992 | * it won't... | |
4993 | */ | |
159ec1fc | 4994 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4995 | list_del_init(&rdev->same_set); |
1da177e4 | 4996 | export_rdev(rdev); |
4b80991c | 4997 | } |
1da177e4 LT |
4998 | mddev_put(mddev); |
4999 | } | |
5000 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5001 | } | |
fdee8ae4 | 5002 | #endif /* !MODULE */ |
1da177e4 | 5003 | |
1da177e4 LT |
5004 | static int get_version(void __user * arg) |
5005 | { | |
5006 | mdu_version_t ver; | |
5007 | ||
5008 | ver.major = MD_MAJOR_VERSION; | |
5009 | ver.minor = MD_MINOR_VERSION; | |
5010 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5011 | ||
5012 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5013 | return -EFAULT; | |
5014 | ||
5015 | return 0; | |
5016 | } | |
5017 | ||
5018 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
5019 | { | |
5020 | mdu_array_info_t info; | |
a9f326eb | 5021 | int nr,working,insync,failed,spare; |
1da177e4 | 5022 | mdk_rdev_t *rdev; |
1da177e4 | 5023 | |
a9f326eb | 5024 | nr=working=insync=failed=spare=0; |
159ec1fc | 5025 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 5026 | nr++; |
b2d444d7 | 5027 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5028 | failed++; |
5029 | else { | |
5030 | working++; | |
b2d444d7 | 5031 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 5032 | insync++; |
1da177e4 LT |
5033 | else |
5034 | spare++; | |
5035 | } | |
5036 | } | |
5037 | ||
5038 | info.major_version = mddev->major_version; | |
5039 | info.minor_version = mddev->minor_version; | |
5040 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5041 | info.ctime = mddev->ctime; | |
5042 | info.level = mddev->level; | |
58c0fed4 AN |
5043 | info.size = mddev->dev_sectors / 2; |
5044 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5045 | info.size = -1; |
1da177e4 LT |
5046 | info.nr_disks = nr; |
5047 | info.raid_disks = mddev->raid_disks; | |
5048 | info.md_minor = mddev->md_minor; | |
5049 | info.not_persistent= !mddev->persistent; | |
5050 | ||
5051 | info.utime = mddev->utime; | |
5052 | info.state = 0; | |
5053 | if (mddev->in_sync) | |
5054 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5055 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5056 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5057 | info.active_disks = insync; |
1da177e4 LT |
5058 | info.working_disks = working; |
5059 | info.failed_disks = failed; | |
5060 | info.spare_disks = spare; | |
5061 | ||
5062 | info.layout = mddev->layout; | |
9d8f0363 | 5063 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5064 | |
5065 | if (copy_to_user(arg, &info, sizeof(info))) | |
5066 | return -EFAULT; | |
5067 | ||
5068 | return 0; | |
5069 | } | |
5070 | ||
87162a28 | 5071 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
5072 | { |
5073 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5074 | char *ptr, *buf = NULL; | |
5075 | int err = -ENOMEM; | |
5076 | ||
b5470dc5 DW |
5077 | if (md_allow_write(mddev)) |
5078 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
5079 | else | |
5080 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 5081 | |
32a7627c N |
5082 | if (!file) |
5083 | goto out; | |
5084 | ||
5085 | /* bitmap disabled, zero the first byte and copy out */ | |
5086 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
5087 | file->pathname[0] = '\0'; | |
5088 | goto copy_out; | |
5089 | } | |
5090 | ||
5091 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5092 | if (!buf) | |
5093 | goto out; | |
5094 | ||
6bcfd601 CH |
5095 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
5096 | if (IS_ERR(ptr)) | |
32a7627c N |
5097 | goto out; |
5098 | ||
5099 | strcpy(file->pathname, ptr); | |
5100 | ||
5101 | copy_out: | |
5102 | err = 0; | |
5103 | if (copy_to_user(arg, file, sizeof(*file))) | |
5104 | err = -EFAULT; | |
5105 | out: | |
5106 | kfree(buf); | |
5107 | kfree(file); | |
5108 | return err; | |
5109 | } | |
5110 | ||
1da177e4 LT |
5111 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
5112 | { | |
5113 | mdu_disk_info_t info; | |
1da177e4 LT |
5114 | mdk_rdev_t *rdev; |
5115 | ||
5116 | if (copy_from_user(&info, arg, sizeof(info))) | |
5117 | return -EFAULT; | |
5118 | ||
26ef379f | 5119 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
5120 | if (rdev) { |
5121 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5122 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5123 | info.raid_disk = rdev->raid_disk; | |
5124 | info.state = 0; | |
b2d444d7 | 5125 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5126 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5127 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5128 | info.state |= (1<<MD_DISK_ACTIVE); |
5129 | info.state |= (1<<MD_DISK_SYNC); | |
5130 | } | |
8ddf9efe N |
5131 | if (test_bit(WriteMostly, &rdev->flags)) |
5132 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5133 | } else { |
5134 | info.major = info.minor = 0; | |
5135 | info.raid_disk = -1; | |
5136 | info.state = (1<<MD_DISK_REMOVED); | |
5137 | } | |
5138 | ||
5139 | if (copy_to_user(arg, &info, sizeof(info))) | |
5140 | return -EFAULT; | |
5141 | ||
5142 | return 0; | |
5143 | } | |
5144 | ||
5145 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
5146 | { | |
5147 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
5148 | mdk_rdev_t *rdev; | |
5149 | dev_t dev = MKDEV(info->major,info->minor); | |
5150 | ||
5151 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5152 | return -EOVERFLOW; | |
5153 | ||
5154 | if (!mddev->raid_disks) { | |
5155 | int err; | |
5156 | /* expecting a device which has a superblock */ | |
5157 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5158 | if (IS_ERR(rdev)) { | |
5159 | printk(KERN_WARNING | |
5160 | "md: md_import_device returned %ld\n", | |
5161 | PTR_ERR(rdev)); | |
5162 | return PTR_ERR(rdev); | |
5163 | } | |
5164 | if (!list_empty(&mddev->disks)) { | |
5165 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
5166 | mdk_rdev_t, same_set); | |
a9f326eb | 5167 | err = super_types[mddev->major_version] |
1da177e4 LT |
5168 | .load_super(rdev, rdev0, mddev->minor_version); |
5169 | if (err < 0) { | |
5170 | printk(KERN_WARNING | |
5171 | "md: %s has different UUID to %s\n", | |
5172 | bdevname(rdev->bdev,b), | |
5173 | bdevname(rdev0->bdev,b2)); | |
5174 | export_rdev(rdev); | |
5175 | return -EINVAL; | |
5176 | } | |
5177 | } | |
5178 | err = bind_rdev_to_array(rdev, mddev); | |
5179 | if (err) | |
5180 | export_rdev(rdev); | |
5181 | return err; | |
5182 | } | |
5183 | ||
5184 | /* | |
5185 | * add_new_disk can be used once the array is assembled | |
5186 | * to add "hot spares". They must already have a superblock | |
5187 | * written | |
5188 | */ | |
5189 | if (mddev->pers) { | |
5190 | int err; | |
5191 | if (!mddev->pers->hot_add_disk) { | |
5192 | printk(KERN_WARNING | |
5193 | "%s: personality does not support diskops!\n", | |
5194 | mdname(mddev)); | |
5195 | return -EINVAL; | |
5196 | } | |
7b1e35f6 N |
5197 | if (mddev->persistent) |
5198 | rdev = md_import_device(dev, mddev->major_version, | |
5199 | mddev->minor_version); | |
5200 | else | |
5201 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5202 | if (IS_ERR(rdev)) { |
5203 | printk(KERN_WARNING | |
5204 | "md: md_import_device returned %ld\n", | |
5205 | PTR_ERR(rdev)); | |
5206 | return PTR_ERR(rdev); | |
5207 | } | |
1a855a06 | 5208 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5209 | if (!mddev->persistent) { |
5210 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5211 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5212 | rdev->raid_disk = info->raid_disk; |
bf572541 N |
5213 | set_bit(In_sync, &rdev->flags); |
5214 | } else | |
41158c7e N |
5215 | rdev->raid_disk = -1; |
5216 | } else | |
5217 | super_types[mddev->major_version]. | |
5218 | validate_super(mddev, rdev); | |
1a855a06 N |
5219 | if (test_bit(In_sync, &rdev->flags)) |
5220 | rdev->saved_raid_disk = rdev->raid_disk; | |
5221 | else | |
5222 | rdev->saved_raid_disk = -1; | |
41158c7e | 5223 | |
b2d444d7 | 5224 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5225 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5226 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5227 | else |
5228 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5229 | |
1da177e4 LT |
5230 | rdev->raid_disk = -1; |
5231 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5232 | if (!err && !mddev->pers->hot_remove_disk) { |
5233 | /* If there is hot_add_disk but no hot_remove_disk | |
5234 | * then added disks for geometry changes, | |
5235 | * and should be added immediately. | |
5236 | */ | |
5237 | super_types[mddev->major_version]. | |
5238 | validate_super(mddev, rdev); | |
5239 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5240 | if (err) | |
5241 | unbind_rdev_from_array(rdev); | |
5242 | } | |
1da177e4 LT |
5243 | if (err) |
5244 | export_rdev(rdev); | |
52664732 | 5245 | else |
00bcb4ac | 5246 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5247 | |
17571284 | 5248 | md_update_sb(mddev, 1); |
72a23c21 NB |
5249 | if (mddev->degraded) |
5250 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5251 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 5252 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5253 | return err; |
5254 | } | |
5255 | ||
5256 | /* otherwise, add_new_disk is only allowed | |
5257 | * for major_version==0 superblocks | |
5258 | */ | |
5259 | if (mddev->major_version != 0) { | |
5260 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5261 | mdname(mddev)); | |
5262 | return -EINVAL; | |
5263 | } | |
5264 | ||
5265 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5266 | int err; | |
d710e138 | 5267 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5268 | if (IS_ERR(rdev)) { |
5269 | printk(KERN_WARNING | |
5270 | "md: error, md_import_device() returned %ld\n", | |
5271 | PTR_ERR(rdev)); | |
5272 | return PTR_ERR(rdev); | |
5273 | } | |
5274 | rdev->desc_nr = info->number; | |
5275 | if (info->raid_disk < mddev->raid_disks) | |
5276 | rdev->raid_disk = info->raid_disk; | |
5277 | else | |
5278 | rdev->raid_disk = -1; | |
5279 | ||
1da177e4 | 5280 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5281 | if (info->state & (1<<MD_DISK_SYNC)) |
5282 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5283 | |
8ddf9efe N |
5284 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5285 | set_bit(WriteMostly, &rdev->flags); | |
5286 | ||
1da177e4 LT |
5287 | if (!mddev->persistent) { |
5288 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5289 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5290 | } else | |
57b2caa3 | 5291 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5292 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5293 | |
2bf071bf N |
5294 | err = bind_rdev_to_array(rdev, mddev); |
5295 | if (err) { | |
5296 | export_rdev(rdev); | |
5297 | return err; | |
5298 | } | |
1da177e4 LT |
5299 | } |
5300 | ||
5301 | return 0; | |
5302 | } | |
5303 | ||
5304 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
5305 | { | |
5306 | char b[BDEVNAME_SIZE]; | |
5307 | mdk_rdev_t *rdev; | |
5308 | ||
1da177e4 LT |
5309 | rdev = find_rdev(mddev, dev); |
5310 | if (!rdev) | |
5311 | return -ENXIO; | |
5312 | ||
5313 | if (rdev->raid_disk >= 0) | |
5314 | goto busy; | |
5315 | ||
5316 | kick_rdev_from_array(rdev); | |
850b2b42 | 5317 | md_update_sb(mddev, 1); |
d7603b7e | 5318 | md_new_event(mddev); |
1da177e4 LT |
5319 | |
5320 | return 0; | |
5321 | busy: | |
fdefa4d8 | 5322 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5323 | bdevname(rdev->bdev,b), mdname(mddev)); |
5324 | return -EBUSY; | |
5325 | } | |
5326 | ||
5327 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
5328 | { | |
5329 | char b[BDEVNAME_SIZE]; | |
5330 | int err; | |
1da177e4 LT |
5331 | mdk_rdev_t *rdev; |
5332 | ||
5333 | if (!mddev->pers) | |
5334 | return -ENODEV; | |
5335 | ||
5336 | if (mddev->major_version != 0) { | |
5337 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5338 | " version-0 superblocks.\n", | |
5339 | mdname(mddev)); | |
5340 | return -EINVAL; | |
5341 | } | |
5342 | if (!mddev->pers->hot_add_disk) { | |
5343 | printk(KERN_WARNING | |
5344 | "%s: personality does not support diskops!\n", | |
5345 | mdname(mddev)); | |
5346 | return -EINVAL; | |
5347 | } | |
5348 | ||
d710e138 | 5349 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5350 | if (IS_ERR(rdev)) { |
5351 | printk(KERN_WARNING | |
5352 | "md: error, md_import_device() returned %ld\n", | |
5353 | PTR_ERR(rdev)); | |
5354 | return -EINVAL; | |
5355 | } | |
5356 | ||
5357 | if (mddev->persistent) | |
57b2caa3 | 5358 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5359 | else |
77304d2a | 5360 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5361 | |
8190e754 | 5362 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5363 | |
b2d444d7 | 5364 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5365 | printk(KERN_WARNING |
5366 | "md: can not hot-add faulty %s disk to %s!\n", | |
5367 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5368 | err = -EINVAL; | |
5369 | goto abort_export; | |
5370 | } | |
b2d444d7 | 5371 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5372 | rdev->desc_nr = -1; |
5842730d | 5373 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5374 | err = bind_rdev_to_array(rdev, mddev); |
5375 | if (err) | |
5376 | goto abort_export; | |
1da177e4 LT |
5377 | |
5378 | /* | |
5379 | * The rest should better be atomic, we can have disk failures | |
5380 | * noticed in interrupt contexts ... | |
5381 | */ | |
5382 | ||
1da177e4 LT |
5383 | rdev->raid_disk = -1; |
5384 | ||
850b2b42 | 5385 | md_update_sb(mddev, 1); |
1da177e4 LT |
5386 | |
5387 | /* | |
5388 | * Kick recovery, maybe this spare has to be added to the | |
5389 | * array immediately. | |
5390 | */ | |
5391 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5392 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5393 | md_new_event(mddev); |
1da177e4 LT |
5394 | return 0; |
5395 | ||
1da177e4 LT |
5396 | abort_export: |
5397 | export_rdev(rdev); | |
5398 | return err; | |
5399 | } | |
5400 | ||
32a7627c N |
5401 | static int set_bitmap_file(mddev_t *mddev, int fd) |
5402 | { | |
5403 | int err; | |
5404 | ||
36fa3063 N |
5405 | if (mddev->pers) { |
5406 | if (!mddev->pers->quiesce) | |
5407 | return -EBUSY; | |
5408 | if (mddev->recovery || mddev->sync_thread) | |
5409 | return -EBUSY; | |
5410 | /* we should be able to change the bitmap.. */ | |
5411 | } | |
32a7627c | 5412 | |
32a7627c | 5413 | |
36fa3063 N |
5414 | if (fd >= 0) { |
5415 | if (mddev->bitmap) | |
5416 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5417 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5418 | |
c3d9714e | 5419 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5420 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5421 | mdname(mddev)); | |
5422 | return -EBADF; | |
5423 | } | |
5424 | ||
c3d9714e | 5425 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
5426 | if (err) { |
5427 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5428 | mdname(mddev)); | |
c3d9714e N |
5429 | fput(mddev->bitmap_info.file); |
5430 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5431 | return err; |
5432 | } | |
c3d9714e | 5433 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5434 | } else if (mddev->bitmap == NULL) |
5435 | return -ENOENT; /* cannot remove what isn't there */ | |
5436 | err = 0; | |
5437 | if (mddev->pers) { | |
5438 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 5439 | if (fd >= 0) { |
36fa3063 | 5440 | err = bitmap_create(mddev); |
69e51b44 N |
5441 | if (!err) |
5442 | err = bitmap_load(mddev); | |
5443 | } | |
d7375ab3 | 5444 | if (fd < 0 || err) { |
36fa3063 | 5445 | bitmap_destroy(mddev); |
d7375ab3 N |
5446 | fd = -1; /* make sure to put the file */ |
5447 | } | |
36fa3063 | 5448 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5449 | } |
5450 | if (fd < 0) { | |
c3d9714e N |
5451 | if (mddev->bitmap_info.file) { |
5452 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5453 | fput(mddev->bitmap_info.file); | |
acc55e22 | 5454 | } |
c3d9714e | 5455 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
5456 | } |
5457 | ||
32a7627c N |
5458 | return err; |
5459 | } | |
5460 | ||
1da177e4 LT |
5461 | /* |
5462 | * set_array_info is used two different ways | |
5463 | * The original usage is when creating a new array. | |
5464 | * In this usage, raid_disks is > 0 and it together with | |
5465 | * level, size, not_persistent,layout,chunksize determine the | |
5466 | * shape of the array. | |
5467 | * This will always create an array with a type-0.90.0 superblock. | |
5468 | * The newer usage is when assembling an array. | |
5469 | * In this case raid_disks will be 0, and the major_version field is | |
5470 | * use to determine which style super-blocks are to be found on the devices. | |
5471 | * The minor and patch _version numbers are also kept incase the | |
5472 | * super_block handler wishes to interpret them. | |
5473 | */ | |
5474 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5475 | { | |
5476 | ||
5477 | if (info->raid_disks == 0) { | |
5478 | /* just setting version number for superblock loading */ | |
5479 | if (info->major_version < 0 || | |
50511da3 | 5480 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5481 | super_types[info->major_version].name == NULL) { |
5482 | /* maybe try to auto-load a module? */ | |
5483 | printk(KERN_INFO | |
5484 | "md: superblock version %d not known\n", | |
5485 | info->major_version); | |
5486 | return -EINVAL; | |
5487 | } | |
5488 | mddev->major_version = info->major_version; | |
5489 | mddev->minor_version = info->minor_version; | |
5490 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5491 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
5492 | /* ensure mddev_put doesn't delete this now that there |
5493 | * is some minimal configuration. | |
5494 | */ | |
5495 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
5496 | return 0; |
5497 | } | |
5498 | mddev->major_version = MD_MAJOR_VERSION; | |
5499 | mddev->minor_version = MD_MINOR_VERSION; | |
5500 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5501 | mddev->ctime = get_seconds(); | |
5502 | ||
5503 | mddev->level = info->level; | |
17115e03 | 5504 | mddev->clevel[0] = 0; |
58c0fed4 | 5505 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5506 | mddev->raid_disks = info->raid_disks; |
5507 | /* don't set md_minor, it is determined by which /dev/md* was | |
5508 | * openned | |
5509 | */ | |
5510 | if (info->state & (1<<MD_SB_CLEAN)) | |
5511 | mddev->recovery_cp = MaxSector; | |
5512 | else | |
5513 | mddev->recovery_cp = 0; | |
5514 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5515 | mddev->external = 0; |
1da177e4 LT |
5516 | |
5517 | mddev->layout = info->layout; | |
9d8f0363 | 5518 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5519 | |
5520 | mddev->max_disks = MD_SB_DISKS; | |
5521 | ||
e691063a N |
5522 | if (mddev->persistent) |
5523 | mddev->flags = 0; | |
850b2b42 | 5524 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5525 | |
c3d9714e N |
5526 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
5527 | mddev->bitmap_info.offset = 0; | |
b2a2703c | 5528 | |
f6705578 N |
5529 | mddev->reshape_position = MaxSector; |
5530 | ||
1da177e4 LT |
5531 | /* |
5532 | * Generate a 128 bit UUID | |
5533 | */ | |
5534 | get_random_bytes(mddev->uuid, 16); | |
5535 | ||
f6705578 | 5536 | mddev->new_level = mddev->level; |
664e7c41 | 5537 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5538 | mddev->new_layout = mddev->layout; |
5539 | mddev->delta_disks = 0; | |
5540 | ||
1da177e4 LT |
5541 | return 0; |
5542 | } | |
5543 | ||
1f403624 DW |
5544 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5545 | { | |
b522adcd DW |
5546 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5547 | ||
5548 | if (mddev->external_size) | |
5549 | return; | |
5550 | ||
1f403624 DW |
5551 | mddev->array_sectors = array_sectors; |
5552 | } | |
5553 | EXPORT_SYMBOL(md_set_array_sectors); | |
5554 | ||
d71f9f88 | 5555 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5556 | { |
159ec1fc | 5557 | mdk_rdev_t *rdev; |
a35b0d69 | 5558 | int rv; |
d71f9f88 | 5559 | int fit = (num_sectors == 0); |
a35b0d69 N |
5560 | |
5561 | if (mddev->pers->resize == NULL) | |
5562 | return -EINVAL; | |
d71f9f88 AN |
5563 | /* The "num_sectors" is the number of sectors of each device that |
5564 | * is used. This can only make sense for arrays with redundancy. | |
5565 | * linear and raid0 always use whatever space is available. We can only | |
5566 | * consider changing this number if no resync or reconstruction is | |
5567 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5568 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5569 | * of each device. If num_sectors is zero, we find the largest size |
5570 | * that fits. | |
a35b0d69 N |
5571 | */ |
5572 | if (mddev->sync_thread) | |
5573 | return -EBUSY; | |
dba034ee N |
5574 | if (mddev->bitmap) |
5575 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5576 | * grow, and re-add. | |
5577 | */ | |
5578 | return -EBUSY; | |
159ec1fc | 5579 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5580 | sector_t avail = rdev->sectors; |
01ab5662 | 5581 | |
d71f9f88 AN |
5582 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5583 | num_sectors = avail; | |
5584 | if (avail < num_sectors) | |
a35b0d69 N |
5585 | return -ENOSPC; |
5586 | } | |
d71f9f88 | 5587 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5588 | if (!rv) |
5589 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5590 | return rv; |
5591 | } | |
5592 | ||
da943b99 N |
5593 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5594 | { | |
5595 | int rv; | |
5596 | /* change the number of raid disks */ | |
63c70c4f | 5597 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5598 | return -EINVAL; |
5599 | if (raid_disks <= 0 || | |
233fca36 | 5600 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 5601 | return -EINVAL; |
63c70c4f | 5602 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5603 | return -EBUSY; |
63c70c4f N |
5604 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5605 | ||
5606 | rv = mddev->pers->check_reshape(mddev); | |
de171cb9 N |
5607 | if (rv < 0) |
5608 | mddev->delta_disks = 0; | |
da943b99 N |
5609 | return rv; |
5610 | } | |
5611 | ||
5612 | ||
1da177e4 LT |
5613 | /* |
5614 | * update_array_info is used to change the configuration of an | |
5615 | * on-line array. | |
5616 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5617 | * fields in the info are checked against the array. | |
5618 | * Any differences that cannot be handled will cause an error. | |
5619 | * Normally, only one change can be managed at a time. | |
5620 | */ | |
5621 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5622 | { | |
5623 | int rv = 0; | |
5624 | int cnt = 0; | |
36fa3063 N |
5625 | int state = 0; |
5626 | ||
5627 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 5628 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5629 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
5630 | |
5631 | if (mddev->major_version != info->major_version || | |
5632 | mddev->minor_version != info->minor_version || | |
5633 | /* mddev->patch_version != info->patch_version || */ | |
5634 | mddev->ctime != info->ctime || | |
5635 | mddev->level != info->level || | |
5636 | /* mddev->layout != info->layout || */ | |
5637 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5638 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5639 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5640 | ((state^info->state) & 0xfffffe00) | |
5641 | ) | |
1da177e4 LT |
5642 | return -EINVAL; |
5643 | /* Check there is only one change */ | |
58c0fed4 AN |
5644 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5645 | cnt++; | |
5646 | if (mddev->raid_disks != info->raid_disks) | |
5647 | cnt++; | |
5648 | if (mddev->layout != info->layout) | |
5649 | cnt++; | |
5650 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5651 | cnt++; | |
5652 | if (cnt == 0) | |
5653 | return 0; | |
5654 | if (cnt > 1) | |
5655 | return -EINVAL; | |
1da177e4 LT |
5656 | |
5657 | if (mddev->layout != info->layout) { | |
5658 | /* Change layout | |
5659 | * we don't need to do anything at the md level, the | |
5660 | * personality will take care of it all. | |
5661 | */ | |
50ac168a | 5662 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5663 | return -EINVAL; |
597a711b N |
5664 | else { |
5665 | mddev->new_layout = info->layout; | |
50ac168a | 5666 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5667 | if (rv) |
5668 | mddev->new_layout = mddev->layout; | |
5669 | return rv; | |
5670 | } | |
1da177e4 | 5671 | } |
58c0fed4 | 5672 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5673 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5674 | |
da943b99 N |
5675 | if (mddev->raid_disks != info->raid_disks) |
5676 | rv = update_raid_disks(mddev, info->raid_disks); | |
5677 | ||
36fa3063 N |
5678 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5679 | if (mddev->pers->quiesce == NULL) | |
5680 | return -EINVAL; | |
5681 | if (mddev->recovery || mddev->sync_thread) | |
5682 | return -EBUSY; | |
5683 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5684 | /* add the bitmap */ | |
5685 | if (mddev->bitmap) | |
5686 | return -EEXIST; | |
c3d9714e | 5687 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 5688 | return -EINVAL; |
c3d9714e N |
5689 | mddev->bitmap_info.offset = |
5690 | mddev->bitmap_info.default_offset; | |
36fa3063 N |
5691 | mddev->pers->quiesce(mddev, 1); |
5692 | rv = bitmap_create(mddev); | |
69e51b44 N |
5693 | if (!rv) |
5694 | rv = bitmap_load(mddev); | |
36fa3063 N |
5695 | if (rv) |
5696 | bitmap_destroy(mddev); | |
5697 | mddev->pers->quiesce(mddev, 0); | |
5698 | } else { | |
5699 | /* remove the bitmap */ | |
5700 | if (!mddev->bitmap) | |
5701 | return -ENOENT; | |
5702 | if (mddev->bitmap->file) | |
5703 | return -EINVAL; | |
5704 | mddev->pers->quiesce(mddev, 1); | |
5705 | bitmap_destroy(mddev); | |
5706 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 5707 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
5708 | } |
5709 | } | |
850b2b42 | 5710 | md_update_sb(mddev, 1); |
1da177e4 LT |
5711 | return rv; |
5712 | } | |
5713 | ||
5714 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5715 | { | |
5716 | mdk_rdev_t *rdev; | |
5717 | ||
5718 | if (mddev->pers == NULL) | |
5719 | return -ENODEV; | |
5720 | ||
5721 | rdev = find_rdev(mddev, dev); | |
5722 | if (!rdev) | |
5723 | return -ENODEV; | |
5724 | ||
5725 | md_error(mddev, rdev); | |
5726 | return 0; | |
5727 | } | |
5728 | ||
2f9618ce AN |
5729 | /* |
5730 | * We have a problem here : there is no easy way to give a CHS | |
5731 | * virtual geometry. We currently pretend that we have a 2 heads | |
5732 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5733 | * dosfs just mad... ;-) | |
5734 | */ | |
a885c8c4 CH |
5735 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5736 | { | |
5737 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5738 | ||
5739 | geo->heads = 2; | |
5740 | geo->sectors = 4; | |
49ce6cea | 5741 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
5742 | return 0; |
5743 | } | |
5744 | ||
a39907fa | 5745 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5746 | unsigned int cmd, unsigned long arg) |
5747 | { | |
5748 | int err = 0; | |
5749 | void __user *argp = (void __user *)arg; | |
1da177e4 | 5750 | mddev_t *mddev = NULL; |
e2218350 | 5751 | int ro; |
1da177e4 LT |
5752 | |
5753 | if (!capable(CAP_SYS_ADMIN)) | |
5754 | return -EACCES; | |
5755 | ||
5756 | /* | |
5757 | * Commands dealing with the RAID driver but not any | |
5758 | * particular array: | |
5759 | */ | |
5760 | switch (cmd) | |
5761 | { | |
5762 | case RAID_VERSION: | |
5763 | err = get_version(argp); | |
5764 | goto done; | |
5765 | ||
5766 | case PRINT_RAID_DEBUG: | |
5767 | err = 0; | |
5768 | md_print_devices(); | |
5769 | goto done; | |
5770 | ||
5771 | #ifndef MODULE | |
5772 | case RAID_AUTORUN: | |
5773 | err = 0; | |
5774 | autostart_arrays(arg); | |
5775 | goto done; | |
5776 | #endif | |
5777 | default:; | |
5778 | } | |
5779 | ||
5780 | /* | |
5781 | * Commands creating/starting a new array: | |
5782 | */ | |
5783 | ||
a39907fa | 5784 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5785 | |
5786 | if (!mddev) { | |
5787 | BUG(); | |
5788 | goto abort; | |
5789 | } | |
5790 | ||
1da177e4 LT |
5791 | err = mddev_lock(mddev); |
5792 | if (err) { | |
5793 | printk(KERN_INFO | |
5794 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5795 | err, cmd); | |
5796 | goto abort; | |
5797 | } | |
5798 | ||
5799 | switch (cmd) | |
5800 | { | |
5801 | case SET_ARRAY_INFO: | |
5802 | { | |
5803 | mdu_array_info_t info; | |
5804 | if (!arg) | |
5805 | memset(&info, 0, sizeof(info)); | |
5806 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5807 | err = -EFAULT; | |
5808 | goto abort_unlock; | |
5809 | } | |
5810 | if (mddev->pers) { | |
5811 | err = update_array_info(mddev, &info); | |
5812 | if (err) { | |
5813 | printk(KERN_WARNING "md: couldn't update" | |
5814 | " array info. %d\n", err); | |
5815 | goto abort_unlock; | |
5816 | } | |
5817 | goto done_unlock; | |
5818 | } | |
5819 | if (!list_empty(&mddev->disks)) { | |
5820 | printk(KERN_WARNING | |
5821 | "md: array %s already has disks!\n", | |
5822 | mdname(mddev)); | |
5823 | err = -EBUSY; | |
5824 | goto abort_unlock; | |
5825 | } | |
5826 | if (mddev->raid_disks) { | |
5827 | printk(KERN_WARNING | |
5828 | "md: array %s already initialised!\n", | |
5829 | mdname(mddev)); | |
5830 | err = -EBUSY; | |
5831 | goto abort_unlock; | |
5832 | } | |
5833 | err = set_array_info(mddev, &info); | |
5834 | if (err) { | |
5835 | printk(KERN_WARNING "md: couldn't set" | |
5836 | " array info. %d\n", err); | |
5837 | goto abort_unlock; | |
5838 | } | |
5839 | } | |
5840 | goto done_unlock; | |
5841 | ||
5842 | default:; | |
5843 | } | |
5844 | ||
5845 | /* | |
5846 | * Commands querying/configuring an existing array: | |
5847 | */ | |
32a7627c | 5848 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5849 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5850 | if ((!mddev->raid_disks && !mddev->external) |
5851 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5852 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5853 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5854 | err = -ENODEV; |
5855 | goto abort_unlock; | |
5856 | } | |
5857 | ||
5858 | /* | |
5859 | * Commands even a read-only array can execute: | |
5860 | */ | |
5861 | switch (cmd) | |
5862 | { | |
5863 | case GET_ARRAY_INFO: | |
5864 | err = get_array_info(mddev, argp); | |
5865 | goto done_unlock; | |
5866 | ||
32a7627c | 5867 | case GET_BITMAP_FILE: |
87162a28 | 5868 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5869 | goto done_unlock; |
5870 | ||
1da177e4 LT |
5871 | case GET_DISK_INFO: |
5872 | err = get_disk_info(mddev, argp); | |
5873 | goto done_unlock; | |
5874 | ||
5875 | case RESTART_ARRAY_RW: | |
5876 | err = restart_array(mddev); | |
5877 | goto done_unlock; | |
5878 | ||
5879 | case STOP_ARRAY: | |
d710e138 | 5880 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5881 | goto done_unlock; |
5882 | ||
5883 | case STOP_ARRAY_RO: | |
a4bd82d0 | 5884 | err = md_set_readonly(mddev, 1); |
1da177e4 LT |
5885 | goto done_unlock; |
5886 | ||
e2218350 DW |
5887 | case BLKROSET: |
5888 | if (get_user(ro, (int __user *)(arg))) { | |
5889 | err = -EFAULT; | |
5890 | goto done_unlock; | |
5891 | } | |
5892 | err = -EINVAL; | |
5893 | ||
5894 | /* if the bdev is going readonly the value of mddev->ro | |
5895 | * does not matter, no writes are coming | |
5896 | */ | |
5897 | if (ro) | |
5898 | goto done_unlock; | |
5899 | ||
5900 | /* are we are already prepared for writes? */ | |
5901 | if (mddev->ro != 1) | |
5902 | goto done_unlock; | |
5903 | ||
5904 | /* transitioning to readauto need only happen for | |
5905 | * arrays that call md_write_start | |
5906 | */ | |
5907 | if (mddev->pers) { | |
5908 | err = restart_array(mddev); | |
5909 | if (err == 0) { | |
5910 | mddev->ro = 2; | |
5911 | set_disk_ro(mddev->gendisk, 0); | |
5912 | } | |
5913 | } | |
5914 | goto done_unlock; | |
1da177e4 LT |
5915 | } |
5916 | ||
5917 | /* | |
5918 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5919 | * superblock, so we do not allow them on read-only arrays. |
5920 | * However non-MD ioctls (e.g. get-size) will still come through | |
5921 | * here and hit the 'default' below, so only disallow | |
5922 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5923 | */ |
bb57fc64 | 5924 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5925 | if (mddev->ro == 2) { |
5926 | mddev->ro = 0; | |
00bcb4ac | 5927 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 NB |
5928 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5929 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5930 | } else { |
5931 | err = -EROFS; | |
5932 | goto abort_unlock; | |
5933 | } | |
1da177e4 LT |
5934 | } |
5935 | ||
5936 | switch (cmd) | |
5937 | { | |
5938 | case ADD_NEW_DISK: | |
5939 | { | |
5940 | mdu_disk_info_t info; | |
5941 | if (copy_from_user(&info, argp, sizeof(info))) | |
5942 | err = -EFAULT; | |
5943 | else | |
5944 | err = add_new_disk(mddev, &info); | |
5945 | goto done_unlock; | |
5946 | } | |
5947 | ||
5948 | case HOT_REMOVE_DISK: | |
5949 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5950 | goto done_unlock; | |
5951 | ||
5952 | case HOT_ADD_DISK: | |
5953 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5954 | goto done_unlock; | |
5955 | ||
5956 | case SET_DISK_FAULTY: | |
5957 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5958 | goto done_unlock; | |
5959 | ||
5960 | case RUN_ARRAY: | |
d710e138 | 5961 | err = do_md_run(mddev); |
1da177e4 LT |
5962 | goto done_unlock; |
5963 | ||
32a7627c N |
5964 | case SET_BITMAP_FILE: |
5965 | err = set_bitmap_file(mddev, (int)arg); | |
5966 | goto done_unlock; | |
5967 | ||
1da177e4 | 5968 | default: |
1da177e4 LT |
5969 | err = -EINVAL; |
5970 | goto abort_unlock; | |
5971 | } | |
5972 | ||
5973 | done_unlock: | |
5974 | abort_unlock: | |
d3374825 N |
5975 | if (mddev->hold_active == UNTIL_IOCTL && |
5976 | err != -EINVAL) | |
5977 | mddev->hold_active = 0; | |
1da177e4 LT |
5978 | mddev_unlock(mddev); |
5979 | ||
5980 | return err; | |
5981 | done: | |
5982 | if (err) | |
5983 | MD_BUG(); | |
5984 | abort: | |
5985 | return err; | |
5986 | } | |
aa98aa31 AB |
5987 | #ifdef CONFIG_COMPAT |
5988 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
5989 | unsigned int cmd, unsigned long arg) | |
5990 | { | |
5991 | switch (cmd) { | |
5992 | case HOT_REMOVE_DISK: | |
5993 | case HOT_ADD_DISK: | |
5994 | case SET_DISK_FAULTY: | |
5995 | case SET_BITMAP_FILE: | |
5996 | /* These take in integer arg, do not convert */ | |
5997 | break; | |
5998 | default: | |
5999 | arg = (unsigned long)compat_ptr(arg); | |
6000 | break; | |
6001 | } | |
6002 | ||
6003 | return md_ioctl(bdev, mode, cmd, arg); | |
6004 | } | |
6005 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6006 | |
a39907fa | 6007 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6008 | { |
6009 | /* | |
6010 | * Succeed if we can lock the mddev, which confirms that | |
6011 | * it isn't being stopped right now. | |
6012 | */ | |
d3374825 | 6013 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6014 | int err; |
6015 | ||
d3374825 N |
6016 | if (mddev->gendisk != bdev->bd_disk) { |
6017 | /* we are racing with mddev_put which is discarding this | |
6018 | * bd_disk. | |
6019 | */ | |
6020 | mddev_put(mddev); | |
6021 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6022 | flush_workqueue(md_misc_wq); |
d3374825 N |
6023 | /* Then retry the open from the top */ |
6024 | return -ERESTARTSYS; | |
6025 | } | |
6026 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6027 | ||
c8c00a69 | 6028 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6029 | goto out; |
6030 | ||
6031 | err = 0; | |
f2ea68cf | 6032 | atomic_inc(&mddev->openers); |
c8c00a69 | 6033 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6034 | |
f0b4f7e2 | 6035 | check_disk_change(bdev); |
1da177e4 LT |
6036 | out: |
6037 | return err; | |
6038 | } | |
6039 | ||
a39907fa | 6040 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6041 | { |
a39907fa | 6042 | mddev_t *mddev = disk->private_data; |
1da177e4 | 6043 | |
52e5f9d1 | 6044 | BUG_ON(!mddev); |
f2ea68cf | 6045 | atomic_dec(&mddev->openers); |
1da177e4 LT |
6046 | mddev_put(mddev); |
6047 | ||
6048 | return 0; | |
6049 | } | |
f0b4f7e2 N |
6050 | |
6051 | static int md_media_changed(struct gendisk *disk) | |
6052 | { | |
6053 | mddev_t *mddev = disk->private_data; | |
6054 | ||
6055 | return mddev->changed; | |
6056 | } | |
6057 | ||
6058 | static int md_revalidate(struct gendisk *disk) | |
6059 | { | |
6060 | mddev_t *mddev = disk->private_data; | |
6061 | ||
6062 | mddev->changed = 0; | |
6063 | return 0; | |
6064 | } | |
83d5cde4 | 6065 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6066 | { |
6067 | .owner = THIS_MODULE, | |
a39907fa AV |
6068 | .open = md_open, |
6069 | .release = md_release, | |
b492b852 | 6070 | .ioctl = md_ioctl, |
aa98aa31 AB |
6071 | #ifdef CONFIG_COMPAT |
6072 | .compat_ioctl = md_compat_ioctl, | |
6073 | #endif | |
a885c8c4 | 6074 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6075 | .media_changed = md_media_changed, |
6076 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6077 | }; |
6078 | ||
75c96f85 | 6079 | static int md_thread(void * arg) |
1da177e4 LT |
6080 | { |
6081 | mdk_thread_t *thread = arg; | |
6082 | ||
1da177e4 LT |
6083 | /* |
6084 | * md_thread is a 'system-thread', it's priority should be very | |
6085 | * high. We avoid resource deadlocks individually in each | |
6086 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6087 | * the very same RT priority as kswapd, thus we will never get | |
6088 | * into a priority inversion deadlock. | |
6089 | * | |
6090 | * we definitely have to have equal or higher priority than | |
6091 | * bdflush, otherwise bdflush will deadlock if there are too | |
6092 | * many dirty RAID5 blocks. | |
6093 | */ | |
1da177e4 | 6094 | |
6985c43f | 6095 | allow_signal(SIGKILL); |
a6fb0934 | 6096 | while (!kthread_should_stop()) { |
1da177e4 | 6097 | |
93588e22 N |
6098 | /* We need to wait INTERRUPTIBLE so that |
6099 | * we don't add to the load-average. | |
6100 | * That means we need to be sure no signals are | |
6101 | * pending | |
6102 | */ | |
6103 | if (signal_pending(current)) | |
6104 | flush_signals(current); | |
6105 | ||
6106 | wait_event_interruptible_timeout | |
6107 | (thread->wqueue, | |
6108 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6109 | || kthread_should_stop(), | |
6110 | thread->timeout); | |
1da177e4 | 6111 | |
6c987910 N |
6112 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6113 | if (!kthread_should_stop()) | |
589a594b | 6114 | thread->run(thread->mddev); |
1da177e4 | 6115 | } |
a6fb0934 | 6116 | |
1da177e4 LT |
6117 | return 0; |
6118 | } | |
6119 | ||
6120 | void md_wakeup_thread(mdk_thread_t *thread) | |
6121 | { | |
6122 | if (thread) { | |
6123 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
6124 | set_bit(THREAD_WAKEUP, &thread->flags); | |
6125 | wake_up(&thread->wqueue); | |
6126 | } | |
6127 | } | |
6128 | ||
6129 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
6130 | const char *name) | |
6131 | { | |
6132 | mdk_thread_t *thread; | |
1da177e4 | 6133 | |
9ffae0cf | 6134 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
6135 | if (!thread) |
6136 | return NULL; | |
6137 | ||
1da177e4 LT |
6138 | init_waitqueue_head(&thread->wqueue); |
6139 | ||
1da177e4 LT |
6140 | thread->run = run; |
6141 | thread->mddev = mddev; | |
32a7627c | 6142 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6143 | thread->tsk = kthread_run(md_thread, thread, |
6144 | "%s_%s", | |
6145 | mdname(thread->mddev), | |
6146 | name ?: mddev->pers->name); | |
a6fb0934 | 6147 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6148 | kfree(thread); |
6149 | return NULL; | |
6150 | } | |
1da177e4 LT |
6151 | return thread; |
6152 | } | |
6153 | ||
1da177e4 LT |
6154 | void md_unregister_thread(mdk_thread_t *thread) |
6155 | { | |
e0cf8f04 N |
6156 | if (!thread) |
6157 | return; | |
ba25f9dc | 6158 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
6159 | |
6160 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6161 | kfree(thread); |
6162 | } | |
6163 | ||
6164 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
6165 | { | |
6166 | if (!mddev) { | |
6167 | MD_BUG(); | |
6168 | return; | |
6169 | } | |
6170 | ||
b2d444d7 | 6171 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6172 | return; |
6bfe0b49 DW |
6173 | |
6174 | if (mddev->external) | |
6175 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 6176 | /* |
1da177e4 LT |
6177 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
6178 | mdname(mddev), | |
6179 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
6180 | __builtin_return_address(0),__builtin_return_address(1), | |
6181 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 6182 | */ |
d0a0a5ee AM |
6183 | if (!mddev->pers) |
6184 | return; | |
1da177e4 LT |
6185 | if (!mddev->pers->error_handler) |
6186 | return; | |
6187 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6188 | if (mddev->degraded) |
6189 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6190 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6191 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6192 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6193 | md_wakeup_thread(mddev->thread); | |
768a418d | 6194 | if (mddev->event_work.func) |
e804ac78 | 6195 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6196 | md_new_event_inintr(mddev); |
1da177e4 LT |
6197 | } |
6198 | ||
6199 | /* seq_file implementation /proc/mdstat */ | |
6200 | ||
6201 | static void status_unused(struct seq_file *seq) | |
6202 | { | |
6203 | int i = 0; | |
6204 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6205 | |
6206 | seq_printf(seq, "unused devices: "); | |
6207 | ||
159ec1fc | 6208 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6209 | char b[BDEVNAME_SIZE]; |
6210 | i++; | |
6211 | seq_printf(seq, "%s ", | |
6212 | bdevname(rdev->bdev,b)); | |
6213 | } | |
6214 | if (!i) | |
6215 | seq_printf(seq, "<none>"); | |
6216 | ||
6217 | seq_printf(seq, "\n"); | |
6218 | } | |
6219 | ||
6220 | ||
6221 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
6222 | { | |
dd71cf6b N |
6223 | sector_t max_sectors, resync, res; |
6224 | unsigned long dt, db; | |
6225 | sector_t rt; | |
4588b42e N |
6226 | int scale; |
6227 | unsigned int per_milli; | |
1da177e4 | 6228 | |
dd71cf6b | 6229 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6230 | |
6231 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 6232 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6233 | else |
dd71cf6b | 6234 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6235 | |
6236 | /* | |
6237 | * Should not happen. | |
6238 | */ | |
dd71cf6b | 6239 | if (!max_sectors) { |
1da177e4 LT |
6240 | MD_BUG(); |
6241 | return; | |
6242 | } | |
4588b42e | 6243 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6244 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6245 | * u32, as those are the requirements for sector_div. |
6246 | * Thus 'scale' must be at least 10 | |
6247 | */ | |
6248 | scale = 10; | |
6249 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6250 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6251 | scale++; |
6252 | } | |
6253 | res = (resync>>scale)*1000; | |
dd71cf6b | 6254 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6255 | |
6256 | per_milli = res; | |
1da177e4 | 6257 | { |
4588b42e | 6258 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6259 | seq_printf(seq, "["); |
6260 | for (i = 0; i < x; i++) | |
6261 | seq_printf(seq, "="); | |
6262 | seq_printf(seq, ">"); | |
6263 | for (i = 0; i < y; i++) | |
6264 | seq_printf(seq, "."); | |
6265 | seq_printf(seq, "] "); | |
6266 | } | |
4588b42e | 6267 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6268 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6269 | "reshape" : | |
61df9d91 N |
6270 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6271 | "check" : | |
6272 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6273 | "resync" : "recovery"))), | |
6274 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6275 | (unsigned long long) resync/2, |
6276 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6277 | |
6278 | /* | |
1da177e4 LT |
6279 | * dt: time from mark until now |
6280 | * db: blocks written from mark until now | |
6281 | * rt: remaining time | |
dd71cf6b N |
6282 | * |
6283 | * rt is a sector_t, so could be 32bit or 64bit. | |
6284 | * So we divide before multiply in case it is 32bit and close | |
6285 | * to the limit. | |
25985edc | 6286 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
6287 | * near the end of resync when the number of remaining sectors |
6288 | * is close to 'db'. | |
6289 | * We then divide rt by 32 after multiplying by db to compensate. | |
6290 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6291 | */ |
6292 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6293 | if (!dt) dt++; | |
ff4e8d9a N |
6294 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6295 | - mddev->resync_mark_cnt; | |
1da177e4 | 6296 | |
dd71cf6b N |
6297 | rt = max_sectors - resync; /* number of remaining sectors */ |
6298 | sector_div(rt, db/32+1); | |
6299 | rt *= dt; | |
6300 | rt >>= 5; | |
6301 | ||
6302 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6303 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6304 | |
ff4e8d9a | 6305 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6306 | } |
6307 | ||
6308 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6309 | { | |
6310 | struct list_head *tmp; | |
6311 | loff_t l = *pos; | |
6312 | mddev_t *mddev; | |
6313 | ||
6314 | if (l >= 0x10000) | |
6315 | return NULL; | |
6316 | if (!l--) | |
6317 | /* header */ | |
6318 | return (void*)1; | |
6319 | ||
6320 | spin_lock(&all_mddevs_lock); | |
6321 | list_for_each(tmp,&all_mddevs) | |
6322 | if (!l--) { | |
6323 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
6324 | mddev_get(mddev); | |
6325 | spin_unlock(&all_mddevs_lock); | |
6326 | return mddev; | |
6327 | } | |
6328 | spin_unlock(&all_mddevs_lock); | |
6329 | if (!l--) | |
6330 | return (void*)2;/* tail */ | |
6331 | return NULL; | |
6332 | } | |
6333 | ||
6334 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6335 | { | |
6336 | struct list_head *tmp; | |
6337 | mddev_t *next_mddev, *mddev = v; | |
6338 | ||
6339 | ++*pos; | |
6340 | if (v == (void*)2) | |
6341 | return NULL; | |
6342 | ||
6343 | spin_lock(&all_mddevs_lock); | |
6344 | if (v == (void*)1) | |
6345 | tmp = all_mddevs.next; | |
6346 | else | |
6347 | tmp = mddev->all_mddevs.next; | |
6348 | if (tmp != &all_mddevs) | |
6349 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
6350 | else { | |
6351 | next_mddev = (void*)2; | |
6352 | *pos = 0x10000; | |
6353 | } | |
6354 | spin_unlock(&all_mddevs_lock); | |
6355 | ||
6356 | if (v != (void*)1) | |
6357 | mddev_put(mddev); | |
6358 | return next_mddev; | |
6359 | ||
6360 | } | |
6361 | ||
6362 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6363 | { | |
6364 | mddev_t *mddev = v; | |
6365 | ||
6366 | if (mddev && v != (void*)1 && v != (void*)2) | |
6367 | mddev_put(mddev); | |
6368 | } | |
6369 | ||
d7603b7e N |
6370 | struct mdstat_info { |
6371 | int event; | |
6372 | }; | |
6373 | ||
1da177e4 LT |
6374 | static int md_seq_show(struct seq_file *seq, void *v) |
6375 | { | |
6376 | mddev_t *mddev = v; | |
dd8ac336 | 6377 | sector_t sectors; |
1da177e4 | 6378 | mdk_rdev_t *rdev; |
d7603b7e | 6379 | struct mdstat_info *mi = seq->private; |
32a7627c | 6380 | struct bitmap *bitmap; |
1da177e4 LT |
6381 | |
6382 | if (v == (void*)1) { | |
2604b703 | 6383 | struct mdk_personality *pers; |
1da177e4 LT |
6384 | seq_printf(seq, "Personalities : "); |
6385 | spin_lock(&pers_lock); | |
2604b703 N |
6386 | list_for_each_entry(pers, &pers_list, list) |
6387 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6388 | |
6389 | spin_unlock(&pers_lock); | |
6390 | seq_printf(seq, "\n"); | |
d7603b7e | 6391 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
6392 | return 0; |
6393 | } | |
6394 | if (v == (void*)2) { | |
6395 | status_unused(seq); | |
6396 | return 0; | |
6397 | } | |
6398 | ||
5dc5cf7d | 6399 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6400 | return -EINTR; |
5dc5cf7d | 6401 | |
1da177e4 LT |
6402 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
6403 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
6404 | mddev->pers ? "" : "in"); | |
6405 | if (mddev->pers) { | |
f91de92e | 6406 | if (mddev->ro==1) |
1da177e4 | 6407 | seq_printf(seq, " (read-only)"); |
f91de92e | 6408 | if (mddev->ro==2) |
52720ae7 | 6409 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
6410 | seq_printf(seq, " %s", mddev->pers->name); |
6411 | } | |
6412 | ||
dd8ac336 | 6413 | sectors = 0; |
159ec1fc | 6414 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
6415 | char b[BDEVNAME_SIZE]; |
6416 | seq_printf(seq, " %s[%d]", | |
6417 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
6418 | if (test_bit(WriteMostly, &rdev->flags)) |
6419 | seq_printf(seq, "(W)"); | |
b2d444d7 | 6420 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6421 | seq_printf(seq, "(F)"); |
6422 | continue; | |
b325a32e N |
6423 | } else if (rdev->raid_disk < 0) |
6424 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 6425 | sectors += rdev->sectors; |
1da177e4 LT |
6426 | } |
6427 | ||
6428 | if (!list_empty(&mddev->disks)) { | |
6429 | if (mddev->pers) | |
6430 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
6431 | (unsigned long long) |
6432 | mddev->array_sectors / 2); | |
1da177e4 LT |
6433 | else |
6434 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 6435 | (unsigned long long)sectors / 2); |
1da177e4 | 6436 | } |
1cd6bf19 N |
6437 | if (mddev->persistent) { |
6438 | if (mddev->major_version != 0 || | |
6439 | mddev->minor_version != 90) { | |
6440 | seq_printf(seq," super %d.%d", | |
6441 | mddev->major_version, | |
6442 | mddev->minor_version); | |
6443 | } | |
e691063a N |
6444 | } else if (mddev->external) |
6445 | seq_printf(seq, " super external:%s", | |
6446 | mddev->metadata_type); | |
6447 | else | |
1cd6bf19 | 6448 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
6449 | |
6450 | if (mddev->pers) { | |
d710e138 | 6451 | mddev->pers->status(seq, mddev); |
1da177e4 | 6452 | seq_printf(seq, "\n "); |
8e1b39d6 N |
6453 | if (mddev->pers->sync_request) { |
6454 | if (mddev->curr_resync > 2) { | |
d710e138 | 6455 | status_resync(seq, mddev); |
8e1b39d6 N |
6456 | seq_printf(seq, "\n "); |
6457 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
6458 | seq_printf(seq, "\tresync=DELAYED\n "); | |
6459 | else if (mddev->recovery_cp < MaxSector) | |
6460 | seq_printf(seq, "\tresync=PENDING\n "); | |
6461 | } | |
32a7627c N |
6462 | } else |
6463 | seq_printf(seq, "\n "); | |
6464 | ||
6465 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
6466 | unsigned long chunk_kb; |
6467 | unsigned long flags; | |
32a7627c | 6468 | spin_lock_irqsave(&bitmap->lock, flags); |
42a04b50 | 6469 | chunk_kb = mddev->bitmap_info.chunksize >> 10; |
32a7627c N |
6470 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " |
6471 | "%lu%s chunk", | |
6472 | bitmap->pages - bitmap->missing_pages, | |
6473 | bitmap->pages, | |
6474 | (bitmap->pages - bitmap->missing_pages) | |
6475 | << (PAGE_SHIFT - 10), | |
42a04b50 | 6476 | chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, |
32a7627c | 6477 | chunk_kb ? "KB" : "B"); |
78d742d8 N |
6478 | if (bitmap->file) { |
6479 | seq_printf(seq, ", file: "); | |
c32c2f63 | 6480 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 6481 | } |
78d742d8 | 6482 | |
32a7627c N |
6483 | seq_printf(seq, "\n"); |
6484 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
6485 | } |
6486 | ||
6487 | seq_printf(seq, "\n"); | |
6488 | } | |
6489 | mddev_unlock(mddev); | |
6490 | ||
6491 | return 0; | |
6492 | } | |
6493 | ||
110518bc | 6494 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6495 | .start = md_seq_start, |
6496 | .next = md_seq_next, | |
6497 | .stop = md_seq_stop, | |
6498 | .show = md_seq_show, | |
6499 | }; | |
6500 | ||
6501 | static int md_seq_open(struct inode *inode, struct file *file) | |
6502 | { | |
6503 | int error; | |
d7603b7e N |
6504 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6505 | if (mi == NULL) | |
6506 | return -ENOMEM; | |
1da177e4 LT |
6507 | |
6508 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6509 | if (error) |
6510 | kfree(mi); | |
6511 | else { | |
6512 | struct seq_file *p = file->private_data; | |
6513 | p->private = mi; | |
6514 | mi->event = atomic_read(&md_event_count); | |
6515 | } | |
1da177e4 LT |
6516 | return error; |
6517 | } | |
6518 | ||
d7603b7e N |
6519 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6520 | { | |
6521 | struct seq_file *m = filp->private_data; | |
6522 | struct mdstat_info *mi = m->private; | |
6523 | int mask; | |
6524 | ||
6525 | poll_wait(filp, &md_event_waiters, wait); | |
6526 | ||
6527 | /* always allow read */ | |
6528 | mask = POLLIN | POLLRDNORM; | |
6529 | ||
6530 | if (mi->event != atomic_read(&md_event_count)) | |
6531 | mask |= POLLERR | POLLPRI; | |
6532 | return mask; | |
6533 | } | |
6534 | ||
fa027c2a | 6535 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6536 | .owner = THIS_MODULE, |
1da177e4 LT |
6537 | .open = md_seq_open, |
6538 | .read = seq_read, | |
6539 | .llseek = seq_lseek, | |
c3f94b40 | 6540 | .release = seq_release_private, |
d7603b7e | 6541 | .poll = mdstat_poll, |
1da177e4 LT |
6542 | }; |
6543 | ||
2604b703 | 6544 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6545 | { |
1da177e4 | 6546 | spin_lock(&pers_lock); |
2604b703 N |
6547 | list_add_tail(&p->list, &pers_list); |
6548 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6549 | spin_unlock(&pers_lock); |
6550 | return 0; | |
6551 | } | |
6552 | ||
2604b703 | 6553 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6554 | { |
2604b703 | 6555 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6556 | spin_lock(&pers_lock); |
2604b703 | 6557 | list_del_init(&p->list); |
1da177e4 LT |
6558 | spin_unlock(&pers_lock); |
6559 | return 0; | |
6560 | } | |
6561 | ||
eea1bf38 | 6562 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6563 | { |
6564 | mdk_rdev_t * rdev; | |
1da177e4 | 6565 | int idle; |
eea1bf38 | 6566 | int curr_events; |
1da177e4 LT |
6567 | |
6568 | idle = 1; | |
4b80991c N |
6569 | rcu_read_lock(); |
6570 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6571 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6572 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6573 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6574 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6575 | /* sync IO will cause sync_io to increase before the disk_stats |
6576 | * as sync_io is counted when a request starts, and | |
6577 | * disk_stats is counted when it completes. | |
6578 | * So resync activity will cause curr_events to be smaller than | |
6579 | * when there was no such activity. | |
6580 | * non-sync IO will cause disk_stat to increase without | |
6581 | * increasing sync_io so curr_events will (eventually) | |
6582 | * be larger than it was before. Once it becomes | |
6583 | * substantially larger, the test below will cause | |
6584 | * the array to appear non-idle, and resync will slow | |
6585 | * down. | |
6586 | * If there is a lot of outstanding resync activity when | |
6587 | * we set last_event to curr_events, then all that activity | |
6588 | * completing might cause the array to appear non-idle | |
6589 | * and resync will be slowed down even though there might | |
6590 | * not have been non-resync activity. This will only | |
6591 | * happen once though. 'last_events' will soon reflect | |
6592 | * the state where there is little or no outstanding | |
6593 | * resync requests, and further resync activity will | |
6594 | * always make curr_events less than last_events. | |
c0e48521 | 6595 | * |
1da177e4 | 6596 | */ |
eea1bf38 | 6597 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6598 | rdev->last_events = curr_events; |
6599 | idle = 0; | |
6600 | } | |
6601 | } | |
4b80991c | 6602 | rcu_read_unlock(); |
1da177e4 LT |
6603 | return idle; |
6604 | } | |
6605 | ||
6606 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6607 | { | |
6608 | /* another "blocks" (512byte) blocks have been synced */ | |
6609 | atomic_sub(blocks, &mddev->recovery_active); | |
6610 | wake_up(&mddev->recovery_wait); | |
6611 | if (!ok) { | |
dfc70645 | 6612 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6613 | md_wakeup_thread(mddev->thread); |
6614 | // stop recovery, signal do_sync .... | |
6615 | } | |
6616 | } | |
6617 | ||
6618 | ||
06d91a5f N |
6619 | /* md_write_start(mddev, bi) |
6620 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6621 | * in superblock) before writing, schedule a superblock update |
6622 | * and wait for it to complete. | |
06d91a5f | 6623 | */ |
3d310eb7 | 6624 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6625 | { |
0fd62b86 | 6626 | int did_change = 0; |
06d91a5f | 6627 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6628 | return; |
06d91a5f | 6629 | |
f91de92e N |
6630 | BUG_ON(mddev->ro == 1); |
6631 | if (mddev->ro == 2) { | |
6632 | /* need to switch to read/write */ | |
6633 | mddev->ro = 0; | |
6634 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6635 | md_wakeup_thread(mddev->thread); | |
25156198 | 6636 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6637 | did_change = 1; |
f91de92e | 6638 | } |
06d91a5f | 6639 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6640 | if (mddev->safemode == 1) |
6641 | mddev->safemode = 0; | |
06d91a5f | 6642 | if (mddev->in_sync) { |
a9701a30 | 6643 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6644 | if (mddev->in_sync) { |
6645 | mddev->in_sync = 0; | |
850b2b42 | 6646 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 6647 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 6648 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6649 | did_change = 1; |
3d310eb7 | 6650 | } |
a9701a30 | 6651 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6652 | } |
0fd62b86 | 6653 | if (did_change) |
00bcb4ac | 6654 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 6655 | wait_event(mddev->sb_wait, |
09a44cc1 | 6656 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 LT |
6657 | } |
6658 | ||
6659 | void md_write_end(mddev_t *mddev) | |
6660 | { | |
6661 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6662 | if (mddev->safemode == 2) | |
6663 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6664 | else if (mddev->safemode_delay) |
1da177e4 LT |
6665 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6666 | } | |
6667 | } | |
6668 | ||
2a2275d6 N |
6669 | /* md_allow_write(mddev) |
6670 | * Calling this ensures that the array is marked 'active' so that writes | |
6671 | * may proceed without blocking. It is important to call this before | |
6672 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6673 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6674 | * |
6675 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6676 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6677 | */ |
b5470dc5 | 6678 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6679 | { |
6680 | if (!mddev->pers) | |
b5470dc5 | 6681 | return 0; |
2a2275d6 | 6682 | if (mddev->ro) |
b5470dc5 | 6683 | return 0; |
1a0fd497 | 6684 | if (!mddev->pers->sync_request) |
b5470dc5 | 6685 | return 0; |
2a2275d6 N |
6686 | |
6687 | spin_lock_irq(&mddev->write_lock); | |
6688 | if (mddev->in_sync) { | |
6689 | mddev->in_sync = 0; | |
6690 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 6691 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
6692 | if (mddev->safemode_delay && |
6693 | mddev->safemode == 0) | |
6694 | mddev->safemode = 1; | |
6695 | spin_unlock_irq(&mddev->write_lock); | |
6696 | md_update_sb(mddev, 0); | |
00bcb4ac | 6697 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
6698 | } else |
6699 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 | 6700 | |
070dc6dd | 6701 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
6702 | return -EAGAIN; |
6703 | else | |
6704 | return 0; | |
2a2275d6 N |
6705 | } |
6706 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6707 | ||
1da177e4 LT |
6708 | #define SYNC_MARKS 10 |
6709 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6710 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6711 | { |
6712 | mddev_t *mddev2; | |
6713 | unsigned int currspeed = 0, | |
6714 | window; | |
57afd89f | 6715 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6716 | unsigned long mark[SYNC_MARKS]; |
6717 | sector_t mark_cnt[SYNC_MARKS]; | |
6718 | int last_mark,m; | |
6719 | struct list_head *tmp; | |
6720 | sector_t last_check; | |
57afd89f | 6721 | int skipped = 0; |
5fd6c1dc | 6722 | mdk_rdev_t *rdev; |
61df9d91 | 6723 | char *desc; |
1da177e4 LT |
6724 | |
6725 | /* just incase thread restarts... */ | |
6726 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6727 | return; | |
5fd6c1dc N |
6728 | if (mddev->ro) /* never try to sync a read-only array */ |
6729 | return; | |
1da177e4 | 6730 | |
61df9d91 N |
6731 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6732 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6733 | desc = "data-check"; | |
6734 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6735 | desc = "requested-resync"; | |
6736 | else | |
6737 | desc = "resync"; | |
6738 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6739 | desc = "reshape"; | |
6740 | else | |
6741 | desc = "recovery"; | |
6742 | ||
1da177e4 LT |
6743 | /* we overload curr_resync somewhat here. |
6744 | * 0 == not engaged in resync at all | |
6745 | * 2 == checking that there is no conflict with another sync | |
6746 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6747 | * commense | |
6748 | * other == active in resync - this many blocks | |
6749 | * | |
6750 | * Before starting a resync we must have set curr_resync to | |
6751 | * 2, and then checked that every "conflicting" array has curr_resync | |
6752 | * less than ours. When we find one that is the same or higher | |
6753 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6754 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6755 | * This will mean we have to start checking from the beginning again. | |
6756 | * | |
6757 | */ | |
6758 | ||
6759 | do { | |
6760 | mddev->curr_resync = 2; | |
6761 | ||
6762 | try_again: | |
404e4b43 | 6763 | if (kthread_should_stop()) |
6985c43f | 6764 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
404e4b43 N |
6765 | |
6766 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
1da177e4 | 6767 | goto skip; |
29ac4aa3 | 6768 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6769 | if (mddev2 == mddev) |
6770 | continue; | |
90b08710 BS |
6771 | if (!mddev->parallel_resync |
6772 | && mddev2->curr_resync | |
6773 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6774 | DEFINE_WAIT(wq); |
6775 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6776 | /* arbitrarily yield */ | |
6777 | mddev->curr_resync = 1; | |
6778 | wake_up(&resync_wait); | |
6779 | } | |
6780 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6781 | /* no need to wait here, we can wait the next | |
6782 | * time 'round when curr_resync == 2 | |
6783 | */ | |
6784 | continue; | |
9744197c N |
6785 | /* We need to wait 'interruptible' so as not to |
6786 | * contribute to the load average, and not to | |
6787 | * be caught by 'softlockup' | |
6788 | */ | |
6789 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6790 | if (!kthread_should_stop() && |
8712e553 | 6791 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6792 | printk(KERN_INFO "md: delaying %s of %s" |
6793 | " until %s has finished (they" | |
1da177e4 | 6794 | " share one or more physical units)\n", |
61df9d91 | 6795 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6796 | mddev_put(mddev2); |
9744197c N |
6797 | if (signal_pending(current)) |
6798 | flush_signals(current); | |
1da177e4 LT |
6799 | schedule(); |
6800 | finish_wait(&resync_wait, &wq); | |
6801 | goto try_again; | |
6802 | } | |
6803 | finish_wait(&resync_wait, &wq); | |
6804 | } | |
6805 | } | |
6806 | } while (mddev->curr_resync < 2); | |
6807 | ||
5fd6c1dc | 6808 | j = 0; |
9d88883e | 6809 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6810 | /* resync follows the size requested by the personality, |
57afd89f | 6811 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6812 | */ |
6813 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6814 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6815 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6816 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6817 | j = mddev->resync_min; | |
6818 | else if (!mddev->bitmap) | |
5fd6c1dc | 6819 | j = mddev->recovery_cp; |
5e96ee65 | 6820 | |
ccfcc3c1 | 6821 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6822 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6823 | else { |
1da177e4 | 6824 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6825 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6826 | j = MaxSector; |
4e59ca7d DW |
6827 | rcu_read_lock(); |
6828 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6829 | if (rdev->raid_disk >= 0 && |
6830 | !test_bit(Faulty, &rdev->flags) && | |
6831 | !test_bit(In_sync, &rdev->flags) && | |
6832 | rdev->recovery_offset < j) | |
6833 | j = rdev->recovery_offset; | |
4e59ca7d | 6834 | rcu_read_unlock(); |
5fd6c1dc | 6835 | } |
1da177e4 | 6836 | |
61df9d91 N |
6837 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6838 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6839 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6840 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6841 | "(but not more than %d KB/sec) for %s.\n", |
6842 | speed_max(mddev), desc); | |
1da177e4 | 6843 | |
eea1bf38 | 6844 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6845 | |
57afd89f | 6846 | io_sectors = 0; |
1da177e4 LT |
6847 | for (m = 0; m < SYNC_MARKS; m++) { |
6848 | mark[m] = jiffies; | |
57afd89f | 6849 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6850 | } |
6851 | last_mark = 0; | |
6852 | mddev->resync_mark = mark[last_mark]; | |
6853 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6854 | ||
6855 | /* | |
6856 | * Tune reconstruction: | |
6857 | */ | |
6858 | window = 32*(PAGE_SIZE/512); | |
6859 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6860 | window/2,(unsigned long long) max_sectors/2); | |
6861 | ||
6862 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6863 | last_check = 0; |
6864 | ||
6865 | if (j>2) { | |
6866 | printk(KERN_INFO | |
61df9d91 N |
6867 | "md: resuming %s of %s from checkpoint.\n", |
6868 | desc, mdname(mddev)); | |
1da177e4 LT |
6869 | mddev->curr_resync = j; |
6870 | } | |
75d3da43 | 6871 | mddev->curr_resync_completed = j; |
1da177e4 LT |
6872 | |
6873 | while (j < max_sectors) { | |
57afd89f | 6874 | sector_t sectors; |
1da177e4 | 6875 | |
57afd89f | 6876 | skipped = 0; |
97e4f42d | 6877 | |
7a91ee1f N |
6878 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6879 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6880 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6881 | > (max_sectors >> 4)) || | |
6882 | (j - mddev->curr_resync_completed)*2 | |
6883 | >= mddev->resync_max - mddev->curr_resync_completed | |
6884 | )) { | |
97e4f42d | 6885 | /* time to update curr_resync_completed */ |
97e4f42d N |
6886 | wait_event(mddev->recovery_wait, |
6887 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 6888 | mddev->curr_resync_completed = j; |
070dc6dd | 6889 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 6890 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6891 | } |
acb180b0 | 6892 | |
e62e58a5 N |
6893 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6894 | /* As this condition is controlled by user-space, | |
6895 | * we can block indefinitely, so use '_interruptible' | |
6896 | * to avoid triggering warnings. | |
6897 | */ | |
6898 | flush_signals(current); /* just in case */ | |
6899 | wait_event_interruptible(mddev->recovery_wait, | |
6900 | mddev->resync_max > j | |
6901 | || kthread_should_stop()); | |
6902 | } | |
acb180b0 N |
6903 | |
6904 | if (kthread_should_stop()) | |
6905 | goto interrupted; | |
6906 | ||
57afd89f | 6907 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6908 | currspeed < speed_min(mddev)); |
57afd89f | 6909 | if (sectors == 0) { |
dfc70645 | 6910 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6911 | goto out; |
6912 | } | |
57afd89f N |
6913 | |
6914 | if (!skipped) { /* actual IO requested */ | |
6915 | io_sectors += sectors; | |
6916 | atomic_add(sectors, &mddev->recovery_active); | |
6917 | } | |
6918 | ||
1da177e4 LT |
6919 | j += sectors; |
6920 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6921 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6922 | if (last_check == 0) |
6923 | /* this is the earliers that rebuilt will be | |
6924 | * visible in /proc/mdstat | |
6925 | */ | |
6926 | md_new_event(mddev); | |
57afd89f N |
6927 | |
6928 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6929 | continue; |
6930 | ||
57afd89f | 6931 | last_check = io_sectors; |
1da177e4 | 6932 | |
dfc70645 | 6933 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6934 | break; |
6935 | ||
6936 | repeat: | |
6937 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6938 | /* step marks */ | |
6939 | int next = (last_mark+1) % SYNC_MARKS; | |
6940 | ||
6941 | mddev->resync_mark = mark[next]; | |
6942 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6943 | mark[next] = jiffies; | |
57afd89f | 6944 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6945 | last_mark = next; |
6946 | } | |
6947 | ||
6948 | ||
c6207277 N |
6949 | if (kthread_should_stop()) |
6950 | goto interrupted; | |
6951 | ||
1da177e4 LT |
6952 | |
6953 | /* | |
6954 | * this loop exits only if either when we are slower than | |
6955 | * the 'hard' speed limit, or the system was IO-idle for | |
6956 | * a jiffy. | |
6957 | * the system might be non-idle CPU-wise, but we only care | |
6958 | * about not overloading the IO subsystem. (things like an | |
6959 | * e2fsck being done on the RAID array should execute fast) | |
6960 | */ | |
1da177e4 LT |
6961 | cond_resched(); |
6962 | ||
57afd89f N |
6963 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6964 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6965 | |
88202a0c N |
6966 | if (currspeed > speed_min(mddev)) { |
6967 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6968 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6969 | msleep(500); |
1da177e4 LT |
6970 | goto repeat; |
6971 | } | |
6972 | } | |
6973 | } | |
61df9d91 | 6974 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6975 | /* |
6976 | * this also signals 'finished resyncing' to md_stop | |
6977 | */ | |
6978 | out: | |
1da177e4 LT |
6979 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
6980 | ||
6981 | /* tell personality that we are finished */ | |
57afd89f | 6982 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6983 | |
dfc70645 | 6984 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6985 | mddev->curr_resync > 2) { |
6986 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6987 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6988 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6989 | printk(KERN_INFO | |
61df9d91 N |
6990 | "md: checkpointing %s of %s.\n", |
6991 | desc, mdname(mddev)); | |
5fd6c1dc N |
6992 | mddev->recovery_cp = mddev->curr_resync; |
6993 | } | |
6994 | } else | |
6995 | mddev->recovery_cp = MaxSector; | |
6996 | } else { | |
6997 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6998 | mddev->curr_resync = MaxSector; | |
4e59ca7d DW |
6999 | rcu_read_lock(); |
7000 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc | 7001 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7002 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7003 | !test_bit(Faulty, &rdev->flags) && |
7004 | !test_bit(In_sync, &rdev->flags) && | |
7005 | rdev->recovery_offset < mddev->curr_resync) | |
7006 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7007 | rcu_read_unlock(); |
5fd6c1dc | 7008 | } |
1da177e4 | 7009 | } |
17571284 | 7010 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7011 | |
1da177e4 | 7012 | skip: |
c07b70ad N |
7013 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7014 | /* We completed so min/max setting can be forgotten if used. */ | |
7015 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7016 | mddev->resync_min = 0; | |
7017 | mddev->resync_max = MaxSector; | |
7018 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7019 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 LT |
7020 | mddev->curr_resync = 0; |
7021 | wake_up(&resync_wait); | |
7022 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
7023 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
7024 | return; |
7025 | ||
7026 | interrupted: | |
7027 | /* | |
7028 | * got a signal, exit. | |
7029 | */ | |
7030 | printk(KERN_INFO | |
7031 | "md: md_do_sync() got signal ... exiting\n"); | |
7032 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7033 | goto out; | |
7034 | ||
1da177e4 | 7035 | } |
29269553 | 7036 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
7037 | |
7038 | ||
b4c4c7b8 N |
7039 | static int remove_and_add_spares(mddev_t *mddev) |
7040 | { | |
7041 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
7042 | int spares = 0; |
7043 | ||
97e4f42d N |
7044 | mddev->curr_resync_completed = 0; |
7045 | ||
159ec1fc | 7046 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 7047 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 7048 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7049 | (test_bit(Faulty, &rdev->flags) || |
7050 | ! test_bit(In_sync, &rdev->flags)) && | |
7051 | atomic_read(&rdev->nr_pending)==0) { | |
7052 | if (mddev->pers->hot_remove_disk( | |
7053 | mddev, rdev->raid_disk)==0) { | |
7054 | char nm[20]; | |
7055 | sprintf(nm,"rd%d", rdev->raid_disk); | |
7056 | sysfs_remove_link(&mddev->kobj, nm); | |
7057 | rdev->raid_disk = -1; | |
7058 | } | |
7059 | } | |
7060 | ||
a8c42c7f | 7061 | if (mddev->degraded && !mddev->recovery_disabled) { |
159ec1fc | 7062 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 7063 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
7064 | !test_bit(In_sync, &rdev->flags) && |
7065 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 7066 | spares++; |
b4c4c7b8 N |
7067 | if (rdev->raid_disk < 0 |
7068 | && !test_bit(Faulty, &rdev->flags)) { | |
7069 | rdev->recovery_offset = 0; | |
199050ea NB |
7070 | if (mddev->pers-> |
7071 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
7072 | char nm[20]; |
7073 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
7074 | if (sysfs_create_link(&mddev->kobj, |
7075 | &rdev->kobj, nm)) | |
00bcb4ac | 7076 | /* failure here is OK */; |
b4c4c7b8 N |
7077 | spares++; |
7078 | md_new_event(mddev); | |
93be75ff | 7079 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
b4c4c7b8 N |
7080 | } else |
7081 | break; | |
7082 | } | |
dfc70645 | 7083 | } |
b4c4c7b8 N |
7084 | } |
7085 | return spares; | |
7086 | } | |
7ebc0be7 N |
7087 | |
7088 | static void reap_sync_thread(mddev_t *mddev) | |
7089 | { | |
7090 | mdk_rdev_t *rdev; | |
7091 | ||
7092 | /* resync has finished, collect result */ | |
7093 | md_unregister_thread(mddev->sync_thread); | |
7094 | mddev->sync_thread = NULL; | |
7095 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
7096 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
7097 | /* success...*/ | |
7098 | /* activate any spares */ | |
7099 | if (mddev->pers->spare_active(mddev)) | |
7100 | sysfs_notify(&mddev->kobj, NULL, | |
7101 | "degraded"); | |
7102 | } | |
7103 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
7104 | mddev->pers->finish_reshape) | |
7105 | mddev->pers->finish_reshape(mddev); | |
7106 | md_update_sb(mddev, 1); | |
7107 | ||
7108 | /* if array is no-longer degraded, then any saved_raid_disk | |
7109 | * information must be scrapped | |
7110 | */ | |
7111 | if (!mddev->degraded) | |
7112 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
7113 | rdev->saved_raid_disk = -1; | |
7114 | ||
7115 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7116 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7117 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7118 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7119 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7120 | /* flag recovery needed just to double check */ | |
7121 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7122 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7123 | md_new_event(mddev); | |
7124 | } | |
7125 | ||
1da177e4 LT |
7126 | /* |
7127 | * This routine is regularly called by all per-raid-array threads to | |
7128 | * deal with generic issues like resync and super-block update. | |
7129 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7130 | * need this as they never do any recovery or update the superblock. | |
7131 | * | |
7132 | * It does not do any resync itself, but rather "forks" off other threads | |
7133 | * to do that as needed. | |
7134 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7135 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7136 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7137 | * and wakeups up this thread which will reap the thread and finish up. |
7138 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7139 | * | |
7140 | * The overall approach is: | |
7141 | * 1/ if the superblock needs updating, update it. | |
7142 | * 2/ If a recovery thread is running, don't do anything else. | |
7143 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7144 | * 4/ If there are any faulty devices, remove them. | |
7145 | * 5/ If array is degraded, try to add spares devices | |
7146 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7147 | */ | |
7148 | void md_check_recovery(mddev_t *mddev) | |
7149 | { | |
5f40402d | 7150 | if (mddev->bitmap) |
aa5cbd10 | 7151 | bitmap_daemon_work(mddev); |
1da177e4 LT |
7152 | |
7153 | if (mddev->ro) | |
7154 | return; | |
fca4d848 N |
7155 | |
7156 | if (signal_pending(current)) { | |
31a59e34 | 7157 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7158 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7159 | mdname(mddev)); | |
7160 | mddev->safemode = 2; | |
7161 | } | |
7162 | flush_signals(current); | |
7163 | } | |
7164 | ||
c89a8eee N |
7165 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7166 | return; | |
1da177e4 | 7167 | if ( ! ( |
126925c0 | 7168 | (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7169 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7170 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7171 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7172 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7173 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7174 | )) |
7175 | return; | |
fca4d848 | 7176 | |
df5b89b3 | 7177 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7178 | int spares = 0; |
fca4d848 | 7179 | |
c89a8eee N |
7180 | if (mddev->ro) { |
7181 | /* Only thing we do on a ro array is remove | |
7182 | * failed devices. | |
7183 | */ | |
a8c42c7f N |
7184 | mdk_rdev_t *rdev; |
7185 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
7186 | if (rdev->raid_disk >= 0 && | |
7187 | !test_bit(Blocked, &rdev->flags) && | |
7188 | test_bit(Faulty, &rdev->flags) && | |
7189 | atomic_read(&rdev->nr_pending)==0) { | |
7190 | if (mddev->pers->hot_remove_disk( | |
7191 | mddev, rdev->raid_disk)==0) { | |
7192 | char nm[20]; | |
7193 | sprintf(nm,"rd%d", rdev->raid_disk); | |
7194 | sysfs_remove_link(&mddev->kobj, nm); | |
7195 | rdev->raid_disk = -1; | |
7196 | } | |
7197 | } | |
c89a8eee N |
7198 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7199 | goto unlock; | |
7200 | } | |
7201 | ||
31a59e34 | 7202 | if (!mddev->external) { |
0fd62b86 | 7203 | int did_change = 0; |
31a59e34 N |
7204 | spin_lock_irq(&mddev->write_lock); |
7205 | if (mddev->safemode && | |
7206 | !atomic_read(&mddev->writes_pending) && | |
7207 | !mddev->in_sync && | |
7208 | mddev->recovery_cp == MaxSector) { | |
7209 | mddev->in_sync = 1; | |
0fd62b86 | 7210 | did_change = 1; |
070dc6dd | 7211 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7212 | } |
7213 | if (mddev->safemode == 1) | |
7214 | mddev->safemode = 0; | |
7215 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7216 | if (did_change) |
00bcb4ac | 7217 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7218 | } |
fca4d848 | 7219 | |
850b2b42 N |
7220 | if (mddev->flags) |
7221 | md_update_sb(mddev, 0); | |
06d91a5f | 7222 | |
1da177e4 LT |
7223 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7224 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7225 | /* resync/recovery still happening */ | |
7226 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7227 | goto unlock; | |
7228 | } | |
7229 | if (mddev->sync_thread) { | |
7ebc0be7 | 7230 | reap_sync_thread(mddev); |
1da177e4 LT |
7231 | goto unlock; |
7232 | } | |
72a23c21 NB |
7233 | /* Set RUNNING before clearing NEEDED to avoid |
7234 | * any transients in the value of "sync_action". | |
7235 | */ | |
7236 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7237 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
7238 | /* Clear some bits that don't mean anything, but |
7239 | * might be left set | |
7240 | */ | |
24dd469d N |
7241 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7242 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7243 | |
5fd6c1dc N |
7244 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
7245 | goto unlock; | |
1da177e4 LT |
7246 | /* no recovery is running. |
7247 | * remove any failed drives, then | |
7248 | * add spares if possible. | |
7249 | * Spare are also removed and re-added, to allow | |
7250 | * the personality to fail the re-add. | |
7251 | */ | |
1da177e4 | 7252 | |
b4c4c7b8 | 7253 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7254 | if (mddev->pers->check_reshape == NULL || |
7255 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7256 | /* Cannot proceed */ |
7257 | goto unlock; | |
7258 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7259 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 7260 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
7261 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7262 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7263 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7264 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7265 | } else if (mddev->recovery_cp < MaxSector) { |
7266 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7267 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7268 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7269 | /* nothing to be done ... */ | |
1da177e4 | 7270 | goto unlock; |
24dd469d | 7271 | |
1da177e4 | 7272 | if (mddev->pers->sync_request) { |
a654b9d8 N |
7273 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
7274 | /* We are adding a device or devices to an array | |
7275 | * which has the bitmap stored on all devices. | |
7276 | * So make sure all bitmap pages get written | |
7277 | */ | |
7278 | bitmap_write_all(mddev->bitmap); | |
7279 | } | |
1da177e4 LT |
7280 | mddev->sync_thread = md_register_thread(md_do_sync, |
7281 | mddev, | |
0da3c619 | 7282 | "resync"); |
1da177e4 LT |
7283 | if (!mddev->sync_thread) { |
7284 | printk(KERN_ERR "%s: could not start resync" | |
7285 | " thread...\n", | |
7286 | mdname(mddev)); | |
7287 | /* leave the spares where they are, it shouldn't hurt */ | |
7ebc0be7 N |
7288 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
7289 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7290 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7291 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7292 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
d7603b7e | 7293 | } else |
1da177e4 | 7294 | md_wakeup_thread(mddev->sync_thread); |
00bcb4ac | 7295 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7296 | md_new_event(mddev); |
1da177e4 LT |
7297 | } |
7298 | unlock: | |
72a23c21 NB |
7299 | if (!mddev->sync_thread) { |
7300 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7301 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7302 | &mddev->recovery)) | |
0c3573f1 | 7303 | if (mddev->sysfs_action) |
00bcb4ac | 7304 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7305 | } |
1da177e4 LT |
7306 | mddev_unlock(mddev); |
7307 | } | |
7308 | } | |
7309 | ||
6bfe0b49 DW |
7310 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
7311 | { | |
00bcb4ac | 7312 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 DW |
7313 | wait_event_timeout(rdev->blocked_wait, |
7314 | !test_bit(Blocked, &rdev->flags), | |
7315 | msecs_to_jiffies(5000)); | |
7316 | rdev_dec_pending(rdev, mddev); | |
7317 | } | |
7318 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7319 | ||
75c96f85 AB |
7320 | static int md_notify_reboot(struct notifier_block *this, |
7321 | unsigned long code, void *x) | |
1da177e4 LT |
7322 | { |
7323 | struct list_head *tmp; | |
7324 | mddev_t *mddev; | |
7325 | ||
7326 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
7327 | ||
7328 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
7329 | ||
29ac4aa3 | 7330 | for_each_mddev(mddev, tmp) |
c71d4887 | 7331 | if (mddev_trylock(mddev)) { |
2b25000b N |
7332 | /* Force a switch to readonly even array |
7333 | * appears to still be in use. Hence | |
7334 | * the '100'. | |
7335 | */ | |
a4bd82d0 | 7336 | md_set_readonly(mddev, 100); |
c71d4887 NB |
7337 | mddev_unlock(mddev); |
7338 | } | |
1da177e4 LT |
7339 | /* |
7340 | * certain more exotic SCSI devices are known to be | |
7341 | * volatile wrt too early system reboots. While the | |
7342 | * right place to handle this issue is the given | |
7343 | * driver, we do want to have a safe RAID driver ... | |
7344 | */ | |
7345 | mdelay(1000*1); | |
7346 | } | |
7347 | return NOTIFY_DONE; | |
7348 | } | |
7349 | ||
75c96f85 | 7350 | static struct notifier_block md_notifier = { |
1da177e4 LT |
7351 | .notifier_call = md_notify_reboot, |
7352 | .next = NULL, | |
7353 | .priority = INT_MAX, /* before any real devices */ | |
7354 | }; | |
7355 | ||
7356 | static void md_geninit(void) | |
7357 | { | |
1da177e4 LT |
7358 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
7359 | ||
c7705f34 | 7360 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
7361 | } |
7362 | ||
75c96f85 | 7363 | static int __init md_init(void) |
1da177e4 | 7364 | { |
e804ac78 TH |
7365 | int ret = -ENOMEM; |
7366 | ||
ada609ee | 7367 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
7368 | if (!md_wq) |
7369 | goto err_wq; | |
7370 | ||
7371 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
7372 | if (!md_misc_wq) | |
7373 | goto err_misc_wq; | |
7374 | ||
7375 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
7376 | goto err_md; | |
7377 | ||
7378 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
7379 | goto err_mdp; | |
7380 | mdp_major = ret; | |
7381 | ||
3dbd8c2e | 7382 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
7383 | md_probe, NULL, NULL); |
7384 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
7385 | md_probe, NULL, NULL); |
7386 | ||
1da177e4 | 7387 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 7388 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
7389 | |
7390 | md_geninit(); | |
d710e138 | 7391 | return 0; |
1da177e4 | 7392 | |
e804ac78 TH |
7393 | err_mdp: |
7394 | unregister_blkdev(MD_MAJOR, "md"); | |
7395 | err_md: | |
7396 | destroy_workqueue(md_misc_wq); | |
7397 | err_misc_wq: | |
7398 | destroy_workqueue(md_wq); | |
7399 | err_wq: | |
7400 | return ret; | |
7401 | } | |
1da177e4 LT |
7402 | |
7403 | #ifndef MODULE | |
7404 | ||
7405 | /* | |
7406 | * Searches all registered partitions for autorun RAID arrays | |
7407 | * at boot time. | |
7408 | */ | |
4d936ec1 ME |
7409 | |
7410 | static LIST_HEAD(all_detected_devices); | |
7411 | struct detected_devices_node { | |
7412 | struct list_head list; | |
7413 | dev_t dev; | |
7414 | }; | |
1da177e4 LT |
7415 | |
7416 | void md_autodetect_dev(dev_t dev) | |
7417 | { | |
4d936ec1 ME |
7418 | struct detected_devices_node *node_detected_dev; |
7419 | ||
7420 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
7421 | if (node_detected_dev) { | |
7422 | node_detected_dev->dev = dev; | |
7423 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
7424 | } else { | |
7425 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
7426 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
7427 | } | |
1da177e4 LT |
7428 | } |
7429 | ||
7430 | ||
7431 | static void autostart_arrays(int part) | |
7432 | { | |
7433 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
7434 | struct detected_devices_node *node_detected_dev; |
7435 | dev_t dev; | |
7436 | int i_scanned, i_passed; | |
1da177e4 | 7437 | |
4d936ec1 ME |
7438 | i_scanned = 0; |
7439 | i_passed = 0; | |
1da177e4 | 7440 | |
4d936ec1 | 7441 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 7442 | |
4d936ec1 ME |
7443 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
7444 | i_scanned++; | |
7445 | node_detected_dev = list_entry(all_detected_devices.next, | |
7446 | struct detected_devices_node, list); | |
7447 | list_del(&node_detected_dev->list); | |
7448 | dev = node_detected_dev->dev; | |
7449 | kfree(node_detected_dev); | |
df968c4e | 7450 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
7451 | if (IS_ERR(rdev)) |
7452 | continue; | |
7453 | ||
b2d444d7 | 7454 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7455 | MD_BUG(); |
7456 | continue; | |
7457 | } | |
d0fae18f | 7458 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 7459 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 7460 | i_passed++; |
1da177e4 | 7461 | } |
4d936ec1 ME |
7462 | |
7463 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
7464 | i_scanned, i_passed); | |
1da177e4 LT |
7465 | |
7466 | autorun_devices(part); | |
7467 | } | |
7468 | ||
fdee8ae4 | 7469 | #endif /* !MODULE */ |
1da177e4 LT |
7470 | |
7471 | static __exit void md_exit(void) | |
7472 | { | |
7473 | mddev_t *mddev; | |
7474 | struct list_head *tmp; | |
8ab5e4c1 | 7475 | |
3dbd8c2e | 7476 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 7477 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 7478 | |
3dbd8c2e | 7479 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
7480 | unregister_blkdev(mdp_major, "mdp"); |
7481 | unregister_reboot_notifier(&md_notifier); | |
7482 | unregister_sysctl_table(raid_table_header); | |
7483 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 7484 | for_each_mddev(mddev, tmp) { |
1da177e4 | 7485 | export_array(mddev); |
d3374825 | 7486 | mddev->hold_active = 0; |
1da177e4 | 7487 | } |
e804ac78 TH |
7488 | destroy_workqueue(md_misc_wq); |
7489 | destroy_workqueue(md_wq); | |
1da177e4 LT |
7490 | } |
7491 | ||
685784aa | 7492 | subsys_initcall(md_init); |
1da177e4 LT |
7493 | module_exit(md_exit) |
7494 | ||
f91de92e N |
7495 | static int get_ro(char *buffer, struct kernel_param *kp) |
7496 | { | |
7497 | return sprintf(buffer, "%d", start_readonly); | |
7498 | } | |
7499 | static int set_ro(const char *val, struct kernel_param *kp) | |
7500 | { | |
7501 | char *e; | |
7502 | int num = simple_strtoul(val, &e, 10); | |
7503 | if (*val && (*e == '\0' || *e == '\n')) { | |
7504 | start_readonly = num; | |
4dbcdc75 | 7505 | return 0; |
f91de92e N |
7506 | } |
7507 | return -EINVAL; | |
7508 | } | |
7509 | ||
80ca3a44 N |
7510 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
7511 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 7512 | |
efeb53c0 | 7513 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 7514 | |
1da177e4 LT |
7515 | EXPORT_SYMBOL(register_md_personality); |
7516 | EXPORT_SYMBOL(unregister_md_personality); | |
7517 | EXPORT_SYMBOL(md_error); | |
7518 | EXPORT_SYMBOL(md_done_sync); | |
7519 | EXPORT_SYMBOL(md_write_start); | |
7520 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
7521 | EXPORT_SYMBOL(md_register_thread); |
7522 | EXPORT_SYMBOL(md_unregister_thread); | |
7523 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
7524 | EXPORT_SYMBOL(md_check_recovery); |
7525 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7526 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 7527 | MODULE_ALIAS("md"); |
72008652 | 7528 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |