]>
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 N |
289 | int cpu; |
290 | ||
0ca69886 N |
291 | if (mddev == NULL || mddev->pers == NULL |
292 | || !mddev->ready) { | |
409c57f3 N |
293 | bio_io_error(bio); |
294 | return 0; | |
295 | } | |
0ca69886 | 296 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 297 | rcu_read_lock(); |
e9c7469b | 298 | if (mddev->suspended) { |
409c57f3 N |
299 | DEFINE_WAIT(__wait); |
300 | for (;;) { | |
301 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
302 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 303 | if (!mddev->suspended) |
409c57f3 N |
304 | break; |
305 | rcu_read_unlock(); | |
306 | schedule(); | |
307 | rcu_read_lock(); | |
308 | } | |
309 | finish_wait(&mddev->sb_wait, &__wait); | |
310 | } | |
311 | atomic_inc(&mddev->active_io); | |
312 | rcu_read_unlock(); | |
49077326 | 313 | |
21a52c6d | 314 | rv = mddev->pers->make_request(mddev, bio); |
49077326 N |
315 | |
316 | cpu = part_stat_lock(); | |
317 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
318 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
319 | bio_sectors(bio)); | |
320 | part_stat_unlock(); | |
321 | ||
409c57f3 N |
322 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
323 | wake_up(&mddev->sb_wait); | |
324 | ||
325 | return rv; | |
326 | } | |
327 | ||
9e35b99c N |
328 | /* mddev_suspend makes sure no new requests are submitted |
329 | * to the device, and that any requests that have been submitted | |
330 | * are completely handled. | |
331 | * Once ->stop is called and completes, the module will be completely | |
332 | * unused. | |
333 | */ | |
390ee602 | 334 | void mddev_suspend(mddev_t *mddev) |
409c57f3 N |
335 | { |
336 | BUG_ON(mddev->suspended); | |
337 | mddev->suspended = 1; | |
338 | synchronize_rcu(); | |
339 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
340 | mddev->pers->quiesce(mddev, 1); | |
409c57f3 | 341 | } |
390ee602 | 342 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 343 | |
390ee602 | 344 | void mddev_resume(mddev_t *mddev) |
409c57f3 N |
345 | { |
346 | mddev->suspended = 0; | |
347 | wake_up(&mddev->sb_wait); | |
348 | mddev->pers->quiesce(mddev, 0); | |
1da177e4 | 349 | } |
390ee602 | 350 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 351 | |
3fa841d7 N |
352 | int mddev_congested(mddev_t *mddev, int bits) |
353 | { | |
354 | return mddev->suspended; | |
355 | } | |
356 | EXPORT_SYMBOL(mddev_congested); | |
357 | ||
a2826aa9 | 358 | /* |
e9c7469b | 359 | * Generic flush handling for md |
a2826aa9 N |
360 | */ |
361 | ||
e9c7469b | 362 | static void md_end_flush(struct bio *bio, int err) |
a2826aa9 N |
363 | { |
364 | mdk_rdev_t *rdev = bio->bi_private; | |
365 | mddev_t *mddev = rdev->mddev; | |
a2826aa9 N |
366 | |
367 | rdev_dec_pending(rdev, mddev); | |
368 | ||
369 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 370 | /* The pre-request flush has finished */ |
e804ac78 | 371 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
372 | } |
373 | bio_put(bio); | |
374 | } | |
375 | ||
a7a07e69 N |
376 | static void md_submit_flush_data(struct work_struct *ws); |
377 | ||
a035fc3e | 378 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 379 | { |
a035fc3e | 380 | mddev_t *mddev = container_of(ws, mddev_t, flush_work); |
a2826aa9 N |
381 | mdk_rdev_t *rdev; |
382 | ||
a7a07e69 N |
383 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
384 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 N |
385 | rcu_read_lock(); |
386 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
387 | if (rdev->raid_disk >= 0 && | |
388 | !test_bit(Faulty, &rdev->flags)) { | |
389 | /* Take two references, one is dropped | |
390 | * when request finishes, one after | |
391 | * we reclaim rcu_read_lock | |
392 | */ | |
393 | struct bio *bi; | |
394 | atomic_inc(&rdev->nr_pending); | |
395 | atomic_inc(&rdev->nr_pending); | |
396 | rcu_read_unlock(); | |
a167f663 | 397 | bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev); |
e9c7469b | 398 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
399 | bi->bi_private = rdev; |
400 | bi->bi_bdev = rdev->bdev; | |
401 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 402 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
403 | rcu_read_lock(); |
404 | rdev_dec_pending(rdev, mddev); | |
405 | } | |
406 | rcu_read_unlock(); | |
a7a07e69 N |
407 | if (atomic_dec_and_test(&mddev->flush_pending)) |
408 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
409 | } |
410 | ||
e9c7469b | 411 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 412 | { |
e9c7469b TH |
413 | mddev_t *mddev = container_of(ws, mddev_t, flush_work); |
414 | struct bio *bio = mddev->flush_bio; | |
a2826aa9 | 415 | |
e9c7469b | 416 | if (bio->bi_size == 0) |
a2826aa9 N |
417 | /* an empty barrier - all done */ |
418 | bio_endio(bio, 0); | |
419 | else { | |
e9c7469b | 420 | bio->bi_rw &= ~REQ_FLUSH; |
21a52c6d | 421 | if (mddev->pers->make_request(mddev, bio)) |
a2826aa9 | 422 | generic_make_request(bio); |
a2826aa9 | 423 | } |
2b74e12e N |
424 | |
425 | mddev->flush_bio = NULL; | |
426 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
427 | } |
428 | ||
e9c7469b | 429 | void md_flush_request(mddev_t *mddev, struct bio *bio) |
a2826aa9 N |
430 | { |
431 | spin_lock_irq(&mddev->write_lock); | |
432 | wait_event_lock_irq(mddev->sb_wait, | |
e9c7469b | 433 | !mddev->flush_bio, |
a2826aa9 | 434 | mddev->write_lock, /*nothing*/); |
e9c7469b | 435 | mddev->flush_bio = bio; |
a2826aa9 N |
436 | spin_unlock_irq(&mddev->write_lock); |
437 | ||
a035fc3e N |
438 | INIT_WORK(&mddev->flush_work, submit_flushes); |
439 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 440 | } |
e9c7469b | 441 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 442 | |
2ac87401 N |
443 | /* Support for plugging. |
444 | * This mirrors the plugging support in request_queue, but does not | |
445 | * require having a whole queue | |
446 | */ | |
447 | static void plugger_work(struct work_struct *work) | |
448 | { | |
449 | struct plug_handle *plug = | |
450 | container_of(work, struct plug_handle, unplug_work); | |
451 | plug->unplug_fn(plug); | |
452 | } | |
453 | static void plugger_timeout(unsigned long data) | |
454 | { | |
455 | struct plug_handle *plug = (void *)data; | |
456 | kblockd_schedule_work(NULL, &plug->unplug_work); | |
457 | } | |
458 | void plugger_init(struct plug_handle *plug, | |
459 | void (*unplug_fn)(struct plug_handle *)) | |
460 | { | |
461 | plug->unplug_flag = 0; | |
462 | plug->unplug_fn = unplug_fn; | |
463 | init_timer(&plug->unplug_timer); | |
464 | plug->unplug_timer.function = plugger_timeout; | |
465 | plug->unplug_timer.data = (unsigned long)plug; | |
466 | INIT_WORK(&plug->unplug_work, plugger_work); | |
467 | } | |
468 | EXPORT_SYMBOL_GPL(plugger_init); | |
469 | ||
470 | void plugger_set_plug(struct plug_handle *plug) | |
471 | { | |
472 | if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag)) | |
473 | mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1); | |
474 | } | |
475 | EXPORT_SYMBOL_GPL(plugger_set_plug); | |
476 | ||
477 | int plugger_remove_plug(struct plug_handle *plug) | |
478 | { | |
479 | if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) { | |
480 | del_timer(&plug->unplug_timer); | |
481 | return 1; | |
482 | } else | |
483 | return 0; | |
484 | } | |
485 | EXPORT_SYMBOL_GPL(plugger_remove_plug); | |
486 | ||
487 | ||
1da177e4 LT |
488 | static inline mddev_t *mddev_get(mddev_t *mddev) |
489 | { | |
490 | atomic_inc(&mddev->active); | |
491 | return mddev; | |
492 | } | |
493 | ||
5fd3a17e | 494 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 495 | |
1da177e4 LT |
496 | static void mddev_put(mddev_t *mddev) |
497 | { | |
a167f663 N |
498 | struct bio_set *bs = NULL; |
499 | ||
1da177e4 LT |
500 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
501 | return; | |
d3374825 | 502 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
503 | mddev->ctime == 0 && !mddev->hold_active) { |
504 | /* Array is not configured at all, and not held active, | |
505 | * so destroy it */ | |
1da177e4 | 506 | list_del(&mddev->all_mddevs); |
a167f663 N |
507 | bs = mddev->bio_set; |
508 | mddev->bio_set = NULL; | |
d3374825 | 509 | if (mddev->gendisk) { |
e804ac78 TH |
510 | /* We did a probe so need to clean up. Call |
511 | * queue_work inside the spinlock so that | |
512 | * flush_workqueue() after mddev_find will | |
513 | * succeed in waiting for the work to be done. | |
d3374825 N |
514 | */ |
515 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 516 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
517 | } else |
518 | kfree(mddev); | |
519 | } | |
520 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
521 | if (bs) |
522 | bioset_free(bs); | |
1da177e4 LT |
523 | } |
524 | ||
390ee602 | 525 | void mddev_init(mddev_t *mddev) |
fafd7fb0 N |
526 | { |
527 | mutex_init(&mddev->open_mutex); | |
528 | mutex_init(&mddev->reconfig_mutex); | |
529 | mutex_init(&mddev->bitmap_info.mutex); | |
530 | INIT_LIST_HEAD(&mddev->disks); | |
531 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
532 | init_timer(&mddev->safemode_timer); | |
533 | atomic_set(&mddev->active, 1); | |
534 | atomic_set(&mddev->openers, 0); | |
535 | atomic_set(&mddev->active_io, 0); | |
536 | spin_lock_init(&mddev->write_lock); | |
537 | atomic_set(&mddev->flush_pending, 0); | |
538 | init_waitqueue_head(&mddev->sb_wait); | |
539 | init_waitqueue_head(&mddev->recovery_wait); | |
540 | mddev->reshape_position = MaxSector; | |
541 | mddev->resync_min = 0; | |
542 | mddev->resync_max = MaxSector; | |
543 | mddev->level = LEVEL_NONE; | |
544 | } | |
390ee602 | 545 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 546 | |
1da177e4 LT |
547 | static mddev_t * mddev_find(dev_t unit) |
548 | { | |
549 | mddev_t *mddev, *new = NULL; | |
550 | ||
551 | retry: | |
552 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
553 | |
554 | if (unit) { | |
555 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
556 | if (mddev->unit == unit) { | |
557 | mddev_get(mddev); | |
558 | spin_unlock(&all_mddevs_lock); | |
559 | kfree(new); | |
560 | return mddev; | |
561 | } | |
562 | ||
563 | if (new) { | |
564 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 565 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
566 | new->hold_active = UNTIL_IOCTL; |
567 | return new; | |
1da177e4 | 568 | } |
efeb53c0 N |
569 | } else if (new) { |
570 | /* find an unused unit number */ | |
571 | static int next_minor = 512; | |
572 | int start = next_minor; | |
573 | int is_free = 0; | |
574 | int dev = 0; | |
575 | while (!is_free) { | |
576 | dev = MKDEV(MD_MAJOR, next_minor); | |
577 | next_minor++; | |
578 | if (next_minor > MINORMASK) | |
579 | next_minor = 0; | |
580 | if (next_minor == start) { | |
581 | /* Oh dear, all in use. */ | |
582 | spin_unlock(&all_mddevs_lock); | |
583 | kfree(new); | |
584 | return NULL; | |
585 | } | |
586 | ||
587 | is_free = 1; | |
588 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
589 | if (mddev->unit == dev) { | |
590 | is_free = 0; | |
591 | break; | |
592 | } | |
593 | } | |
594 | new->unit = dev; | |
595 | new->md_minor = MINOR(dev); | |
596 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
597 | list_add(&new->all_mddevs, &all_mddevs); |
598 | spin_unlock(&all_mddevs_lock); | |
599 | return new; | |
600 | } | |
601 | spin_unlock(&all_mddevs_lock); | |
602 | ||
9ffae0cf | 603 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
604 | if (!new) |
605 | return NULL; | |
606 | ||
1da177e4 LT |
607 | new->unit = unit; |
608 | if (MAJOR(unit) == MD_MAJOR) | |
609 | new->md_minor = MINOR(unit); | |
610 | else | |
611 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
612 | ||
fafd7fb0 | 613 | mddev_init(new); |
1da177e4 | 614 | |
1da177e4 LT |
615 | goto retry; |
616 | } | |
617 | ||
618 | static inline int mddev_lock(mddev_t * mddev) | |
619 | { | |
df5b89b3 | 620 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
621 | } |
622 | ||
b522adcd DW |
623 | static inline int mddev_is_locked(mddev_t *mddev) |
624 | { | |
625 | return mutex_is_locked(&mddev->reconfig_mutex); | |
626 | } | |
627 | ||
1da177e4 LT |
628 | static inline int mddev_trylock(mddev_t * mddev) |
629 | { | |
df5b89b3 | 630 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
631 | } |
632 | ||
b6eb127d N |
633 | static struct attribute_group md_redundancy_group; |
634 | ||
a64c876f | 635 | static void mddev_unlock(mddev_t * mddev) |
1da177e4 | 636 | { |
a64c876f | 637 | if (mddev->to_remove) { |
b6eb127d N |
638 | /* These cannot be removed under reconfig_mutex as |
639 | * an access to the files will try to take reconfig_mutex | |
640 | * while holding the file unremovable, which leads to | |
641 | * a deadlock. | |
bb4f1e9d N |
642 | * So hold set sysfs_active while the remove in happeing, |
643 | * and anything else which might set ->to_remove or my | |
644 | * otherwise change the sysfs namespace will fail with | |
645 | * -EBUSY if sysfs_active is still set. | |
646 | * We set sysfs_active under reconfig_mutex and elsewhere | |
647 | * test it under the same mutex to ensure its correct value | |
648 | * is seen. | |
b6eb127d | 649 | */ |
a64c876f N |
650 | struct attribute_group *to_remove = mddev->to_remove; |
651 | mddev->to_remove = NULL; | |
bb4f1e9d | 652 | mddev->sysfs_active = 1; |
b6eb127d N |
653 | mutex_unlock(&mddev->reconfig_mutex); |
654 | ||
00bcb4ac N |
655 | if (mddev->kobj.sd) { |
656 | if (to_remove != &md_redundancy_group) | |
657 | sysfs_remove_group(&mddev->kobj, to_remove); | |
658 | if (mddev->pers == NULL || | |
659 | mddev->pers->sync_request == NULL) { | |
660 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
661 | if (mddev->sysfs_action) | |
662 | sysfs_put(mddev->sysfs_action); | |
663 | mddev->sysfs_action = NULL; | |
664 | } | |
a64c876f | 665 | } |
bb4f1e9d | 666 | mddev->sysfs_active = 0; |
b6eb127d N |
667 | } else |
668 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 669 | |
005eca5e | 670 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
671 | } |
672 | ||
2989ddbd | 673 | static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) |
1da177e4 | 674 | { |
159ec1fc | 675 | mdk_rdev_t *rdev; |
1da177e4 | 676 | |
159ec1fc | 677 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
678 | if (rdev->desc_nr == nr) |
679 | return rdev; | |
159ec1fc | 680 | |
1da177e4 LT |
681 | return NULL; |
682 | } | |
683 | ||
684 | static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) | |
685 | { | |
1da177e4 LT |
686 | mdk_rdev_t *rdev; |
687 | ||
159ec1fc | 688 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
689 | if (rdev->bdev->bd_dev == dev) |
690 | return rdev; | |
159ec1fc | 691 | |
1da177e4 LT |
692 | return NULL; |
693 | } | |
694 | ||
d9d166c2 | 695 | static struct mdk_personality *find_pers(int level, char *clevel) |
2604b703 N |
696 | { |
697 | struct mdk_personality *pers; | |
d9d166c2 N |
698 | list_for_each_entry(pers, &pers_list, list) { |
699 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 700 | return pers; |
d9d166c2 N |
701 | if (strcmp(pers->name, clevel)==0) |
702 | return pers; | |
703 | } | |
2604b703 N |
704 | return NULL; |
705 | } | |
706 | ||
b73df2d3 | 707 | /* return the offset of the super block in 512byte sectors */ |
57b2caa3 | 708 | static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev) |
1da177e4 | 709 | { |
57b2caa3 | 710 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 711 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
712 | } |
713 | ||
1da177e4 LT |
714 | static int alloc_disk_sb(mdk_rdev_t * rdev) |
715 | { | |
716 | if (rdev->sb_page) | |
717 | MD_BUG(); | |
718 | ||
719 | rdev->sb_page = alloc_page(GFP_KERNEL); | |
720 | if (!rdev->sb_page) { | |
721 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 722 | return -ENOMEM; |
1da177e4 LT |
723 | } |
724 | ||
725 | return 0; | |
726 | } | |
727 | ||
728 | static void free_disk_sb(mdk_rdev_t * rdev) | |
729 | { | |
730 | if (rdev->sb_page) { | |
2d1f3b5d | 731 | put_page(rdev->sb_page); |
1da177e4 LT |
732 | rdev->sb_loaded = 0; |
733 | rdev->sb_page = NULL; | |
0f420358 | 734 | rdev->sb_start = 0; |
dd8ac336 | 735 | rdev->sectors = 0; |
1da177e4 LT |
736 | } |
737 | } | |
738 | ||
739 | ||
6712ecf8 | 740 | static void super_written(struct bio *bio, int error) |
7bfa19f2 N |
741 | { |
742 | mdk_rdev_t *rdev = bio->bi_private; | |
a9701a30 | 743 | mddev_t *mddev = rdev->mddev; |
7bfa19f2 | 744 | |
3a0f5bbb N |
745 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
746 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
747 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
748 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 749 | md_error(mddev, rdev); |
3a0f5bbb | 750 | } |
7bfa19f2 | 751 | |
a9701a30 N |
752 | if (atomic_dec_and_test(&mddev->pending_writes)) |
753 | wake_up(&mddev->sb_wait); | |
f8b58edf | 754 | bio_put(bio); |
7bfa19f2 N |
755 | } |
756 | ||
757 | void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, | |
758 | sector_t sector, int size, struct page *page) | |
759 | { | |
760 | /* write first size bytes of page to sector of rdev | |
761 | * Increment mddev->pending_writes before returning | |
762 | * and decrement it on completion, waking up sb_wait | |
763 | * if zero is reached. | |
764 | * If an error occurred, call md_error | |
765 | */ | |
a167f663 | 766 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 767 | |
a6ff7e08 | 768 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
7bfa19f2 N |
769 | bio->bi_sector = sector; |
770 | bio_add_page(bio, page, size, 0); | |
771 | bio->bi_private = rdev; | |
772 | bio->bi_end_io = super_written; | |
a9701a30 | 773 | |
7bfa19f2 | 774 | atomic_inc(&mddev->pending_writes); |
e9c7469b TH |
775 | submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA, |
776 | bio); | |
a9701a30 N |
777 | } |
778 | ||
779 | void md_super_wait(mddev_t *mddev) | |
780 | { | |
e9c7469b | 781 | /* wait for all superblock writes that were scheduled to complete */ |
a9701a30 N |
782 | DEFINE_WAIT(wq); |
783 | for(;;) { | |
784 | prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE); | |
785 | if (atomic_read(&mddev->pending_writes)==0) | |
786 | break; | |
a9701a30 N |
787 | schedule(); |
788 | } | |
789 | finish_wait(&mddev->sb_wait, &wq); | |
7bfa19f2 N |
790 | } |
791 | ||
6712ecf8 | 792 | static void bi_complete(struct bio *bio, int error) |
1da177e4 | 793 | { |
1da177e4 | 794 | complete((struct completion*)bio->bi_private); |
1da177e4 LT |
795 | } |
796 | ||
2b193363 | 797 | int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size, |
ccebd4c4 | 798 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 799 | { |
a167f663 | 800 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
801 | struct completion event; |
802 | int ret; | |
803 | ||
7b6d91da | 804 | rw |= REQ_SYNC | REQ_UNPLUG; |
1da177e4 | 805 | |
a6ff7e08 JB |
806 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
807 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 JB |
808 | if (metadata_op) |
809 | bio->bi_sector = sector + rdev->sb_start; | |
810 | else | |
811 | bio->bi_sector = sector + rdev->data_offset; | |
1da177e4 LT |
812 | bio_add_page(bio, page, size, 0); |
813 | init_completion(&event); | |
814 | bio->bi_private = &event; | |
815 | bio->bi_end_io = bi_complete; | |
816 | submit_bio(rw, bio); | |
817 | wait_for_completion(&event); | |
818 | ||
819 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
820 | bio_put(bio); | |
821 | return ret; | |
822 | } | |
a8745db2 | 823 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 824 | |
0002b271 | 825 | static int read_disk_sb(mdk_rdev_t * rdev, int size) |
1da177e4 LT |
826 | { |
827 | char b[BDEVNAME_SIZE]; | |
828 | if (!rdev->sb_page) { | |
829 | MD_BUG(); | |
830 | return -EINVAL; | |
831 | } | |
832 | if (rdev->sb_loaded) | |
833 | return 0; | |
834 | ||
835 | ||
ccebd4c4 | 836 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
837 | goto fail; |
838 | rdev->sb_loaded = 1; | |
839 | return 0; | |
840 | ||
841 | fail: | |
842 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
843 | bdevname(rdev->bdev,b)); | |
844 | return -EINVAL; | |
845 | } | |
846 | ||
847 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
848 | { | |
05710466 AN |
849 | return sb1->set_uuid0 == sb2->set_uuid0 && |
850 | sb1->set_uuid1 == sb2->set_uuid1 && | |
851 | sb1->set_uuid2 == sb2->set_uuid2 && | |
852 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
853 | } |
854 | ||
1da177e4 LT |
855 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
856 | { | |
857 | int ret; | |
858 | mdp_super_t *tmp1, *tmp2; | |
859 | ||
860 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
861 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
862 | ||
863 | if (!tmp1 || !tmp2) { | |
864 | ret = 0; | |
35020f1a | 865 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
866 | goto abort; |
867 | } | |
868 | ||
869 | *tmp1 = *sb1; | |
870 | *tmp2 = *sb2; | |
871 | ||
872 | /* | |
873 | * nr_disks is not constant | |
874 | */ | |
875 | tmp1->nr_disks = 0; | |
876 | tmp2->nr_disks = 0; | |
877 | ||
ce0c8e05 | 878 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 879 | abort: |
990a8baf JJ |
880 | kfree(tmp1); |
881 | kfree(tmp2); | |
1da177e4 LT |
882 | return ret; |
883 | } | |
884 | ||
4d167f09 N |
885 | |
886 | static u32 md_csum_fold(u32 csum) | |
887 | { | |
888 | csum = (csum & 0xffff) + (csum >> 16); | |
889 | return (csum & 0xffff) + (csum >> 16); | |
890 | } | |
891 | ||
1da177e4 LT |
892 | static unsigned int calc_sb_csum(mdp_super_t * sb) |
893 | { | |
4d167f09 N |
894 | u64 newcsum = 0; |
895 | u32 *sb32 = (u32*)sb; | |
896 | int i; | |
1da177e4 LT |
897 | unsigned int disk_csum, csum; |
898 | ||
899 | disk_csum = sb->sb_csum; | |
900 | sb->sb_csum = 0; | |
4d167f09 N |
901 | |
902 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
903 | newcsum += sb32[i]; | |
904 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
905 | ||
906 | ||
907 | #ifdef CONFIG_ALPHA | |
908 | /* This used to use csum_partial, which was wrong for several | |
909 | * reasons including that different results are returned on | |
910 | * different architectures. It isn't critical that we get exactly | |
911 | * the same return value as before (we always csum_fold before | |
912 | * testing, and that removes any differences). However as we | |
913 | * know that csum_partial always returned a 16bit value on | |
914 | * alphas, do a fold to maximise conformity to previous behaviour. | |
915 | */ | |
916 | sb->sb_csum = md_csum_fold(disk_csum); | |
917 | #else | |
1da177e4 | 918 | sb->sb_csum = disk_csum; |
4d167f09 | 919 | #endif |
1da177e4 LT |
920 | return csum; |
921 | } | |
922 | ||
923 | ||
924 | /* | |
925 | * Handle superblock details. | |
926 | * We want to be able to handle multiple superblock formats | |
927 | * so we have a common interface to them all, and an array of | |
928 | * different handlers. | |
929 | * We rely on user-space to write the initial superblock, and support | |
930 | * reading and updating of superblocks. | |
931 | * Interface methods are: | |
932 | * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) | |
933 | * loads and validates a superblock on dev. | |
934 | * if refdev != NULL, compare superblocks on both devices | |
935 | * Return: | |
936 | * 0 - dev has a superblock that is compatible with refdev | |
937 | * 1 - dev has a superblock that is compatible and newer than refdev | |
938 | * so dev should be used as the refdev in future | |
939 | * -EINVAL superblock incompatible or invalid | |
940 | * -othererror e.g. -EIO | |
941 | * | |
942 | * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) | |
943 | * Verify that dev is acceptable into mddev. | |
944 | * The first time, mddev->raid_disks will be 0, and data from | |
945 | * dev should be merged in. Subsequent calls check that dev | |
946 | * is new enough. Return 0 or -EINVAL | |
947 | * | |
948 | * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) | |
949 | * Update the superblock for rdev with data in mddev | |
950 | * This does not write to disc. | |
951 | * | |
952 | */ | |
953 | ||
954 | struct super_type { | |
0cd17fec CW |
955 | char *name; |
956 | struct module *owner; | |
957 | int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, | |
958 | int minor_version); | |
959 | int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
960 | void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); | |
961 | unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev, | |
15f4a5fd | 962 | sector_t num_sectors); |
1da177e4 LT |
963 | }; |
964 | ||
0894cc30 AN |
965 | /* |
966 | * Check that the given mddev has no bitmap. | |
967 | * | |
968 | * This function is called from the run method of all personalities that do not | |
969 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
970 | * has a bitmap. Otherwise, it returns 0. | |
971 | * | |
972 | */ | |
973 | int md_check_no_bitmap(mddev_t *mddev) | |
974 | { | |
c3d9714e | 975 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
976 | return 0; |
977 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
978 | mdname(mddev), mddev->pers->name); | |
979 | return 1; | |
980 | } | |
981 | EXPORT_SYMBOL(md_check_no_bitmap); | |
982 | ||
1da177e4 LT |
983 | /* |
984 | * load_super for 0.90.0 | |
985 | */ | |
986 | static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
987 | { | |
988 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
989 | mdp_super_t *sb; | |
990 | int ret; | |
1da177e4 LT |
991 | |
992 | /* | |
0f420358 | 993 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
994 | * it's at the end of the disk. |
995 | * | |
996 | * It also happens to be a multiple of 4Kb. | |
997 | */ | |
57b2caa3 | 998 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 999 | |
0002b271 | 1000 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
1001 | if (ret) return ret; |
1002 | ||
1003 | ret = -EINVAL; | |
1004 | ||
1005 | bdevname(rdev->bdev, b); | |
1006 | sb = (mdp_super_t*)page_address(rdev->sb_page); | |
1007 | ||
1008 | if (sb->md_magic != MD_SB_MAGIC) { | |
1009 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
1010 | b); | |
1011 | goto abort; | |
1012 | } | |
1013 | ||
1014 | if (sb->major_version != 0 || | |
f6705578 N |
1015 | sb->minor_version < 90 || |
1016 | sb->minor_version > 91) { | |
1da177e4 LT |
1017 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
1018 | sb->major_version, sb->minor_version, | |
1019 | b); | |
1020 | goto abort; | |
1021 | } | |
1022 | ||
1023 | if (sb->raid_disks <= 0) | |
1024 | goto abort; | |
1025 | ||
4d167f09 | 1026 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
1027 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
1028 | b); | |
1029 | goto abort; | |
1030 | } | |
1031 | ||
1032 | rdev->preferred_minor = sb->md_minor; | |
1033 | rdev->data_offset = 0; | |
0002b271 | 1034 | rdev->sb_size = MD_SB_BYTES; |
1da177e4 LT |
1035 | |
1036 | if (sb->level == LEVEL_MULTIPATH) | |
1037 | rdev->desc_nr = -1; | |
1038 | else | |
1039 | rdev->desc_nr = sb->this_disk.number; | |
1040 | ||
9a7b2b0f | 1041 | if (!refdev) { |
1da177e4 | 1042 | ret = 1; |
9a7b2b0f | 1043 | } else { |
1da177e4 LT |
1044 | __u64 ev1, ev2; |
1045 | mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); | |
1046 | if (!uuid_equal(refsb, sb)) { | |
1047 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1048 | b, bdevname(refdev->bdev,b2)); | |
1049 | goto abort; | |
1050 | } | |
1051 | if (!sb_equal(refsb, sb)) { | |
1052 | printk(KERN_WARNING "md: %s has same UUID" | |
1053 | " but different superblock to %s\n", | |
1054 | b, bdevname(refdev->bdev, b2)); | |
1055 | goto abort; | |
1056 | } | |
1057 | ev1 = md_event(sb); | |
1058 | ev2 = md_event(refsb); | |
1059 | if (ev1 > ev2) | |
1060 | ret = 1; | |
1061 | else | |
1062 | ret = 0; | |
1063 | } | |
8190e754 | 1064 | rdev->sectors = rdev->sb_start; |
1da177e4 | 1065 | |
dd8ac336 | 1066 | if (rdev->sectors < sb->size * 2 && sb->level > 1) |
2bf071bf N |
1067 | /* "this cannot possibly happen" ... */ |
1068 | ret = -EINVAL; | |
1069 | ||
1da177e4 LT |
1070 | abort: |
1071 | return ret; | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * validate_super for 0.90.0 | |
1076 | */ | |
1077 | static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1078 | { | |
1079 | mdp_disk_t *desc; | |
1080 | mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); | |
07d84d10 | 1081 | __u64 ev1 = md_event(sb); |
1da177e4 | 1082 | |
41158c7e | 1083 | rdev->raid_disk = -1; |
c5d79adb N |
1084 | clear_bit(Faulty, &rdev->flags); |
1085 | clear_bit(In_sync, &rdev->flags); | |
1086 | clear_bit(WriteMostly, &rdev->flags); | |
c5d79adb | 1087 | |
1da177e4 LT |
1088 | if (mddev->raid_disks == 0) { |
1089 | mddev->major_version = 0; | |
1090 | mddev->minor_version = sb->minor_version; | |
1091 | mddev->patch_version = sb->patch_version; | |
e691063a | 1092 | mddev->external = 0; |
9d8f0363 | 1093 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1094 | mddev->ctime = sb->ctime; |
1095 | mddev->utime = sb->utime; | |
1096 | mddev->level = sb->level; | |
d9d166c2 | 1097 | mddev->clevel[0] = 0; |
1da177e4 LT |
1098 | mddev->layout = sb->layout; |
1099 | mddev->raid_disks = sb->raid_disks; | |
58c0fed4 | 1100 | mddev->dev_sectors = sb->size * 2; |
07d84d10 | 1101 | mddev->events = ev1; |
c3d9714e N |
1102 | mddev->bitmap_info.offset = 0; |
1103 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; | |
1da177e4 | 1104 | |
f6705578 N |
1105 | if (mddev->minor_version >= 91) { |
1106 | mddev->reshape_position = sb->reshape_position; | |
1107 | mddev->delta_disks = sb->delta_disks; | |
1108 | mddev->new_level = sb->new_level; | |
1109 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1110 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
f6705578 N |
1111 | } else { |
1112 | mddev->reshape_position = MaxSector; | |
1113 | mddev->delta_disks = 0; | |
1114 | mddev->new_level = mddev->level; | |
1115 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1116 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1117 | } |
1118 | ||
1da177e4 LT |
1119 | if (sb->state & (1<<MD_SB_CLEAN)) |
1120 | mddev->recovery_cp = MaxSector; | |
1121 | else { | |
1122 | if (sb->events_hi == sb->cp_events_hi && | |
1123 | sb->events_lo == sb->cp_events_lo) { | |
1124 | mddev->recovery_cp = sb->recovery_cp; | |
1125 | } else | |
1126 | mddev->recovery_cp = 0; | |
1127 | } | |
1128 | ||
1129 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1130 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1131 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1132 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1133 | ||
1134 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1135 | |
1136 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
c3d9714e N |
1137 | mddev->bitmap_info.file == NULL) |
1138 | mddev->bitmap_info.offset = | |
1139 | mddev->bitmap_info.default_offset; | |
a654b9d8 | 1140 | |
41158c7e | 1141 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1142 | /* Insist on good event counter while assembling, except |
1143 | * for spares (which don't need an event count) */ | |
1da177e4 | 1144 | ++ev1; |
be6800a7 N |
1145 | if (sb->disks[rdev->desc_nr].state & ( |
1146 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
1147 | if (ev1 < mddev->events) | |
1148 | return -EINVAL; | |
41158c7e N |
1149 | } else if (mddev->bitmap) { |
1150 | /* if adding to array with a bitmap, then we can accept an | |
1151 | * older device ... but not too old. | |
1152 | */ | |
41158c7e N |
1153 | if (ev1 < mddev->bitmap->events_cleared) |
1154 | return 0; | |
07d84d10 N |
1155 | } else { |
1156 | if (ev1 < mddev->events) | |
1157 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1158 | return 0; | |
1159 | } | |
41158c7e | 1160 | |
1da177e4 | 1161 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1162 | desc = sb->disks + rdev->desc_nr; |
1163 | ||
1164 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1165 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1166 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1167 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1168 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1169 | rdev->raid_disk = desc->raid_disk; |
0261cd9f N |
1170 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1171 | /* active but not in sync implies recovery up to | |
1172 | * reshape position. We don't know exactly where | |
1173 | * that is, so set to zero for now */ | |
1174 | if (mddev->minor_version >= 91) { | |
1175 | rdev->recovery_offset = 0; | |
1176 | rdev->raid_disk = desc->raid_disk; | |
1177 | } | |
1da177e4 | 1178 | } |
8ddf9efe N |
1179 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1180 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1181 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1182 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1183 | return 0; |
1184 | } | |
1185 | ||
1186 | /* | |
1187 | * sync_super for 0.90.0 | |
1188 | */ | |
1189 | static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1190 | { | |
1191 | mdp_super_t *sb; | |
1da177e4 LT |
1192 | mdk_rdev_t *rdev2; |
1193 | int next_spare = mddev->raid_disks; | |
19133a42 | 1194 | |
1da177e4 LT |
1195 | |
1196 | /* make rdev->sb match mddev data.. | |
1197 | * | |
1198 | * 1/ zero out disks | |
1199 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1200 | * 3/ any empty disks < next_spare become removed | |
1201 | * | |
1202 | * disks[0] gets initialised to REMOVED because | |
1203 | * we cannot be sure from other fields if it has | |
1204 | * been initialised or not. | |
1205 | */ | |
1206 | int i; | |
1207 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1208 | ||
61181565 N |
1209 | rdev->sb_size = MD_SB_BYTES; |
1210 | ||
1da177e4 LT |
1211 | sb = (mdp_super_t*)page_address(rdev->sb_page); |
1212 | ||
1213 | memset(sb, 0, sizeof(*sb)); | |
1214 | ||
1215 | sb->md_magic = MD_SB_MAGIC; | |
1216 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1217 | sb->patch_version = mddev->patch_version; |
1218 | sb->gvalid_words = 0; /* ignored */ | |
1219 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1220 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1221 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1222 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1223 | ||
1224 | sb->ctime = mddev->ctime; | |
1225 | sb->level = mddev->level; | |
58c0fed4 | 1226 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1227 | sb->raid_disks = mddev->raid_disks; |
1228 | sb->md_minor = mddev->md_minor; | |
e691063a | 1229 | sb->not_persistent = 0; |
1da177e4 LT |
1230 | sb->utime = mddev->utime; |
1231 | sb->state = 0; | |
1232 | sb->events_hi = (mddev->events>>32); | |
1233 | sb->events_lo = (u32)mddev->events; | |
1234 | ||
f6705578 N |
1235 | if (mddev->reshape_position == MaxSector) |
1236 | sb->minor_version = 90; | |
1237 | else { | |
1238 | sb->minor_version = 91; | |
1239 | sb->reshape_position = mddev->reshape_position; | |
1240 | sb->new_level = mddev->new_level; | |
1241 | sb->delta_disks = mddev->delta_disks; | |
1242 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1243 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1244 | } |
1245 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1246 | if (mddev->in_sync) |
1247 | { | |
1248 | sb->recovery_cp = mddev->recovery_cp; | |
1249 | sb->cp_events_hi = (mddev->events>>32); | |
1250 | sb->cp_events_lo = (u32)mddev->events; | |
1251 | if (mddev->recovery_cp == MaxSector) | |
1252 | sb->state = (1<< MD_SB_CLEAN); | |
1253 | } else | |
1254 | sb->recovery_cp = 0; | |
1255 | ||
1256 | sb->layout = mddev->layout; | |
9d8f0363 | 1257 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1258 | |
c3d9714e | 1259 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1260 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1261 | ||
1da177e4 | 1262 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
159ec1fc | 1263 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1264 | mdp_disk_t *d; |
86e6ffdd | 1265 | int desc_nr; |
0261cd9f N |
1266 | int is_active = test_bit(In_sync, &rdev2->flags); |
1267 | ||
1268 | if (rdev2->raid_disk >= 0 && | |
1269 | sb->minor_version >= 91) | |
1270 | /* we have nowhere to store the recovery_offset, | |
1271 | * but if it is not below the reshape_position, | |
1272 | * we can piggy-back on that. | |
1273 | */ | |
1274 | is_active = 1; | |
1275 | if (rdev2->raid_disk < 0 || | |
1276 | test_bit(Faulty, &rdev2->flags)) | |
1277 | is_active = 0; | |
1278 | if (is_active) | |
86e6ffdd | 1279 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1280 | else |
86e6ffdd | 1281 | desc_nr = next_spare++; |
19133a42 | 1282 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1283 | d = &sb->disks[rdev2->desc_nr]; |
1284 | nr_disks++; | |
1285 | d->number = rdev2->desc_nr; | |
1286 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1287 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1288 | if (is_active) |
1da177e4 LT |
1289 | d->raid_disk = rdev2->raid_disk; |
1290 | else | |
1291 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1292 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1293 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1294 | else if (is_active) { |
1da177e4 | 1295 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1296 | if (test_bit(In_sync, &rdev2->flags)) |
1297 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1298 | active++; |
1299 | working++; | |
1300 | } else { | |
1301 | d->state = 0; | |
1302 | spare++; | |
1303 | working++; | |
1304 | } | |
8ddf9efe N |
1305 | if (test_bit(WriteMostly, &rdev2->flags)) |
1306 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1307 | } |
1da177e4 LT |
1308 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1309 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1310 | mdp_disk_t *d = &sb->disks[i]; | |
1311 | if (d->state == 0 && d->number == 0) { | |
1312 | d->number = i; | |
1313 | d->raid_disk = i; | |
1314 | d->state = (1<<MD_DISK_REMOVED); | |
1315 | d->state |= (1<<MD_DISK_FAULTY); | |
1316 | failed++; | |
1317 | } | |
1318 | } | |
1319 | sb->nr_disks = nr_disks; | |
1320 | sb->active_disks = active; | |
1321 | sb->working_disks = working; | |
1322 | sb->failed_disks = failed; | |
1323 | sb->spare_disks = spare; | |
1324 | ||
1325 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1326 | sb->sb_csum = calc_sb_csum(sb); | |
1327 | } | |
1328 | ||
0cd17fec CW |
1329 | /* |
1330 | * rdev_size_change for 0.90.0 | |
1331 | */ | |
1332 | static unsigned long long | |
15f4a5fd | 1333 | super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec | 1334 | { |
58c0fed4 | 1335 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1336 | return 0; /* component must fit device */ |
c3d9714e | 1337 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1338 | return 0; /* can't move bitmap */ |
57b2caa3 | 1339 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1340 | if (!num_sectors || num_sectors > rdev->sb_start) |
1341 | num_sectors = rdev->sb_start; | |
0f420358 | 1342 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1343 | rdev->sb_page); |
1344 | md_super_wait(rdev->mddev); | |
c26a44ed | 1345 | return num_sectors; |
0cd17fec CW |
1346 | } |
1347 | ||
1348 | ||
1da177e4 LT |
1349 | /* |
1350 | * version 1 superblock | |
1351 | */ | |
1352 | ||
1c05b4bc | 1353 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb) |
1da177e4 | 1354 | { |
1c05b4bc N |
1355 | __le32 disk_csum; |
1356 | u32 csum; | |
1da177e4 LT |
1357 | unsigned long long newcsum; |
1358 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1359 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1360 | int i; |
1361 | ||
1362 | disk_csum = sb->sb_csum; | |
1363 | sb->sb_csum = 0; | |
1364 | newcsum = 0; | |
1365 | for (i=0; size>=4; size -= 4 ) | |
1366 | newcsum += le32_to_cpu(*isuper++); | |
1367 | ||
1368 | if (size == 2) | |
1c05b4bc | 1369 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1370 | |
1371 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1372 | sb->sb_csum = disk_csum; | |
1373 | return cpu_to_le32(csum); | |
1374 | } | |
1375 | ||
1376 | static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) | |
1377 | { | |
1378 | struct mdp_superblock_1 *sb; | |
1379 | int ret; | |
0f420358 | 1380 | sector_t sb_start; |
1da177e4 | 1381 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1382 | int bmask; |
1da177e4 LT |
1383 | |
1384 | /* | |
0f420358 | 1385 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1386 | * It is always aligned to a 4K boundary and |
1387 | * depeding on minor_version, it can be: | |
1388 | * 0: At least 8K, but less than 12K, from end of device | |
1389 | * 1: At start of device | |
1390 | * 2: 4K from start of device. | |
1391 | */ | |
1392 | switch(minor_version) { | |
1393 | case 0: | |
77304d2a | 1394 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1395 | sb_start -= 8*2; |
1396 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1397 | break; |
1398 | case 1: | |
0f420358 | 1399 | sb_start = 0; |
1da177e4 LT |
1400 | break; |
1401 | case 2: | |
0f420358 | 1402 | sb_start = 8; |
1da177e4 LT |
1403 | break; |
1404 | default: | |
1405 | return -EINVAL; | |
1406 | } | |
0f420358 | 1407 | rdev->sb_start = sb_start; |
1da177e4 | 1408 | |
0002b271 N |
1409 | /* superblock is rarely larger than 1K, but it can be larger, |
1410 | * and it is safe to read 4k, so we do that | |
1411 | */ | |
1412 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1413 | if (ret) return ret; |
1414 | ||
1415 | ||
1416 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1417 | ||
1418 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1419 | sb->major_version != cpu_to_le32(1) || | |
1420 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1421 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1422 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1423 | return -EINVAL; |
1424 | ||
1425 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1426 | printk("md: invalid superblock checksum on %s\n", | |
1427 | bdevname(rdev->bdev,b)); | |
1428 | return -EINVAL; | |
1429 | } | |
1430 | if (le64_to_cpu(sb->data_size) < 10) { | |
1431 | printk("md: data_size too small on %s\n", | |
1432 | bdevname(rdev->bdev,b)); | |
1433 | return -EINVAL; | |
1434 | } | |
e11e93fa | 1435 | |
1da177e4 LT |
1436 | rdev->preferred_minor = 0xffff; |
1437 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
4dbcdc75 | 1438 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1439 | |
0002b271 | 1440 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1441 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1442 | if (rdev->sb_size & bmask) |
a1801f85 N |
1443 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1444 | ||
1445 | if (minor_version | |
0f420358 | 1446 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1447 | return -EINVAL; |
0002b271 | 1448 | |
31b65a0d N |
1449 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1450 | rdev->desc_nr = -1; | |
1451 | else | |
1452 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1453 | ||
9a7b2b0f | 1454 | if (!refdev) { |
8ed75463 | 1455 | ret = 1; |
9a7b2b0f | 1456 | } else { |
1da177e4 LT |
1457 | __u64 ev1, ev2; |
1458 | struct mdp_superblock_1 *refsb = | |
1459 | (struct mdp_superblock_1*)page_address(refdev->sb_page); | |
1460 | ||
1461 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1462 | sb->level != refsb->level || | |
1463 | sb->layout != refsb->layout || | |
1464 | sb->chunksize != refsb->chunksize) { | |
1465 | printk(KERN_WARNING "md: %s has strangely different" | |
1466 | " superblock to %s\n", | |
1467 | bdevname(rdev->bdev,b), | |
1468 | bdevname(refdev->bdev,b2)); | |
1469 | return -EINVAL; | |
1470 | } | |
1471 | ev1 = le64_to_cpu(sb->events); | |
1472 | ev2 = le64_to_cpu(refsb->events); | |
1473 | ||
1474 | if (ev1 > ev2) | |
8ed75463 N |
1475 | ret = 1; |
1476 | else | |
1477 | ret = 0; | |
1da177e4 | 1478 | } |
a1801f85 | 1479 | if (minor_version) |
77304d2a | 1480 | rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 1481 | le64_to_cpu(sb->data_offset); |
1da177e4 | 1482 | else |
dd8ac336 AN |
1483 | rdev->sectors = rdev->sb_start; |
1484 | if (rdev->sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1485 | return -EINVAL; |
dd8ac336 | 1486 | rdev->sectors = le64_to_cpu(sb->data_size); |
dd8ac336 | 1487 | if (le64_to_cpu(sb->size) > rdev->sectors) |
2bf071bf | 1488 | return -EINVAL; |
8ed75463 | 1489 | return ret; |
1da177e4 LT |
1490 | } |
1491 | ||
1492 | static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
1493 | { | |
1494 | struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
07d84d10 | 1495 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1496 | |
41158c7e | 1497 | rdev->raid_disk = -1; |
c5d79adb N |
1498 | clear_bit(Faulty, &rdev->flags); |
1499 | clear_bit(In_sync, &rdev->flags); | |
1500 | clear_bit(WriteMostly, &rdev->flags); | |
c5d79adb | 1501 | |
1da177e4 LT |
1502 | if (mddev->raid_disks == 0) { |
1503 | mddev->major_version = 1; | |
1504 | mddev->patch_version = 0; | |
e691063a | 1505 | mddev->external = 0; |
9d8f0363 | 1506 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1507 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1508 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1509 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1510 | mddev->clevel[0] = 0; |
1da177e4 LT |
1511 | mddev->layout = le32_to_cpu(sb->layout); |
1512 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1513 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1514 | mddev->events = ev1; |
c3d9714e N |
1515 | mddev->bitmap_info.offset = 0; |
1516 | mddev->bitmap_info.default_offset = 1024 >> 9; | |
1da177e4 LT |
1517 | |
1518 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); | |
1519 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1520 | ||
1521 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1522 | |
71c0805c | 1523 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
c3d9714e N |
1524 | mddev->bitmap_info.file == NULL ) |
1525 | mddev->bitmap_info.offset = | |
1526 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
e11e93fa | 1527 | |
f6705578 N |
1528 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1529 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1530 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1531 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1532 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1533 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
f6705578 N |
1534 | } else { |
1535 | mddev->reshape_position = MaxSector; | |
1536 | mddev->delta_disks = 0; | |
1537 | mddev->new_level = mddev->level; | |
1538 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1539 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1540 | } |
1541 | ||
41158c7e | 1542 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1543 | /* Insist of good event counter while assembling, except for |
1544 | * spares (which don't need an event count) */ | |
1da177e4 | 1545 | ++ev1; |
be6800a7 N |
1546 | if (rdev->desc_nr >= 0 && |
1547 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1548 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1549 | if (ev1 < mddev->events) | |
1550 | return -EINVAL; | |
41158c7e N |
1551 | } else if (mddev->bitmap) { |
1552 | /* If adding to array with a bitmap, then we can accept an | |
1553 | * older device, but not too old. | |
1554 | */ | |
41158c7e N |
1555 | if (ev1 < mddev->bitmap->events_cleared) |
1556 | return 0; | |
07d84d10 N |
1557 | } else { |
1558 | if (ev1 < mddev->events) | |
1559 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1560 | return 0; | |
1561 | } | |
1da177e4 LT |
1562 | if (mddev->level != LEVEL_MULTIPATH) { |
1563 | int role; | |
3673f305 N |
1564 | if (rdev->desc_nr < 0 || |
1565 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1566 | role = 0xffff; | |
1567 | rdev->desc_nr = -1; | |
1568 | } else | |
1569 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1570 | switch(role) { |
1571 | case 0xffff: /* spare */ | |
1da177e4 LT |
1572 | break; |
1573 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1574 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1575 | break; |
1576 | default: | |
5fd6c1dc N |
1577 | if ((le32_to_cpu(sb->feature_map) & |
1578 | MD_FEATURE_RECOVERY_OFFSET)) | |
1579 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); | |
1580 | else | |
1581 | set_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
1582 | rdev->raid_disk = role; |
1583 | break; | |
1584 | } | |
8ddf9efe N |
1585 | if (sb->devflags & WriteMostly1) |
1586 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1587 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1588 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1589 | |
1da177e4 LT |
1590 | return 0; |
1591 | } | |
1592 | ||
1593 | static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
1594 | { | |
1595 | struct mdp_superblock_1 *sb; | |
1da177e4 LT |
1596 | mdk_rdev_t *rdev2; |
1597 | int max_dev, i; | |
1598 | /* make rdev->sb match mddev and rdev data. */ | |
1599 | ||
1600 | sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); | |
1601 | ||
1602 | sb->feature_map = 0; | |
1603 | sb->pad0 = 0; | |
5fd6c1dc | 1604 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1605 | memset(sb->pad1, 0, sizeof(sb->pad1)); |
1606 | memset(sb->pad2, 0, sizeof(sb->pad2)); | |
1607 | memset(sb->pad3, 0, sizeof(sb->pad3)); | |
1608 | ||
1609 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1610 | sb->events = cpu_to_le64(mddev->events); | |
1611 | if (mddev->in_sync) | |
1612 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1613 | else | |
1614 | sb->resync_offset = cpu_to_le64(0); | |
1615 | ||
1c05b4bc | 1616 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1617 | |
f0ca340c | 1618 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1619 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1620 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1621 | sb->level = cpu_to_le32(mddev->level); |
1622 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1623 | |
c3d9714e N |
1624 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1625 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1626 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1627 | } |
5fd6c1dc N |
1628 | |
1629 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1630 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1631 | sb->feature_map |= |
1632 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1633 | sb->recovery_offset = | |
1634 | cpu_to_le64(rdev->recovery_offset); | |
5fd6c1dc N |
1635 | } |
1636 | ||
f6705578 N |
1637 | if (mddev->reshape_position != MaxSector) { |
1638 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1639 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1640 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1641 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1642 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1643 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
f6705578 | 1644 | } |
a654b9d8 | 1645 | |
1da177e4 | 1646 | max_dev = 0; |
159ec1fc | 1647 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
1da177e4 LT |
1648 | if (rdev2->desc_nr+1 > max_dev) |
1649 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1650 | |
70471daf N |
1651 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1652 | int bmask; | |
a778b73f | 1653 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1654 | rdev->sb_size = max_dev * 2 + 256; |
1655 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1656 | if (rdev->sb_size & bmask) | |
1657 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1658 | } else |
1659 | max_dev = le32_to_cpu(sb->max_dev); | |
1660 | ||
1da177e4 LT |
1661 | for (i=0; i<max_dev;i++) |
1662 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
1663 | ||
159ec1fc | 1664 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
1da177e4 | 1665 | i = rdev2->desc_nr; |
b2d444d7 | 1666 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1667 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1668 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1669 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1670 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1671 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1672 | else |
1673 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1674 | } | |
1675 | ||
1da177e4 LT |
1676 | sb->sb_csum = calc_sb_1_csum(sb); |
1677 | } | |
1678 | ||
0cd17fec | 1679 | static unsigned long long |
15f4a5fd | 1680 | super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) |
0cd17fec CW |
1681 | { |
1682 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1683 | sector_t max_sectors; |
58c0fed4 | 1684 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1685 | return 0; /* component must fit device */ |
0f420358 | 1686 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1687 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1688 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1689 | max_sectors -= rdev->data_offset; |
1690 | if (!num_sectors || num_sectors > max_sectors) | |
1691 | num_sectors = max_sectors; | |
c3d9714e | 1692 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1693 | /* minor version 0 with bitmap we can't move */ |
1694 | return 0; | |
1695 | } else { | |
1696 | /* minor version 0; superblock after data */ | |
0f420358 | 1697 | sector_t sb_start; |
77304d2a | 1698 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1699 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1700 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1701 | if (!num_sectors || num_sectors > max_sectors) |
1702 | num_sectors = max_sectors; | |
0f420358 | 1703 | rdev->sb_start = sb_start; |
0cd17fec CW |
1704 | } |
1705 | sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page); | |
15f4a5fd | 1706 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1707 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1708 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1709 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1710 | rdev->sb_page); |
1711 | md_super_wait(rdev->mddev); | |
c26a44ed | 1712 | return num_sectors; |
0cd17fec | 1713 | } |
1da177e4 | 1714 | |
75c96f85 | 1715 | static struct super_type super_types[] = { |
1da177e4 LT |
1716 | [0] = { |
1717 | .name = "0.90.0", | |
1718 | .owner = THIS_MODULE, | |
0cd17fec CW |
1719 | .load_super = super_90_load, |
1720 | .validate_super = super_90_validate, | |
1721 | .sync_super = super_90_sync, | |
1722 | .rdev_size_change = super_90_rdev_size_change, | |
1da177e4 LT |
1723 | }, |
1724 | [1] = { | |
1725 | .name = "md-1", | |
1726 | .owner = THIS_MODULE, | |
0cd17fec CW |
1727 | .load_super = super_1_load, |
1728 | .validate_super = super_1_validate, | |
1729 | .sync_super = super_1_sync, | |
1730 | .rdev_size_change = super_1_rdev_size_change, | |
1da177e4 LT |
1731 | }, |
1732 | }; | |
1da177e4 LT |
1733 | |
1734 | static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) | |
1735 | { | |
7dd5e7c3 | 1736 | mdk_rdev_t *rdev, *rdev2; |
1da177e4 | 1737 | |
4b80991c N |
1738 | rcu_read_lock(); |
1739 | rdev_for_each_rcu(rdev, mddev1) | |
1740 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1741 | if (rdev->bdev->bd_contains == |
4b80991c N |
1742 | rdev2->bdev->bd_contains) { |
1743 | rcu_read_unlock(); | |
7dd5e7c3 | 1744 | return 1; |
4b80991c N |
1745 | } |
1746 | rcu_read_unlock(); | |
1da177e4 LT |
1747 | return 0; |
1748 | } | |
1749 | ||
1750 | static LIST_HEAD(pending_raid_disks); | |
1751 | ||
ac5e7113 AN |
1752 | /* |
1753 | * Try to register data integrity profile for an mddev | |
1754 | * | |
1755 | * This is called when an array is started and after a disk has been kicked | |
1756 | * from the array. It only succeeds if all working and active component devices | |
1757 | * are integrity capable with matching profiles. | |
1758 | */ | |
1759 | int md_integrity_register(mddev_t *mddev) | |
1760 | { | |
1761 | mdk_rdev_t *rdev, *reference = NULL; | |
1762 | ||
1763 | if (list_empty(&mddev->disks)) | |
1764 | return 0; /* nothing to do */ | |
1765 | if (blk_get_integrity(mddev->gendisk)) | |
1766 | return 0; /* already registered */ | |
1767 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
1768 | /* skip spares and non-functional disks */ | |
1769 | if (test_bit(Faulty, &rdev->flags)) | |
1770 | continue; | |
1771 | if (rdev->raid_disk < 0) | |
1772 | continue; | |
1773 | /* | |
1774 | * If at least one rdev is not integrity capable, we can not | |
1775 | * enable data integrity for the md device. | |
1776 | */ | |
1777 | if (!bdev_get_integrity(rdev->bdev)) | |
1778 | return -EINVAL; | |
1779 | if (!reference) { | |
1780 | /* Use the first rdev as the reference */ | |
1781 | reference = rdev; | |
1782 | continue; | |
1783 | } | |
1784 | /* does this rdev's profile match the reference profile? */ | |
1785 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1786 | rdev->bdev->bd_disk) < 0) | |
1787 | return -EINVAL; | |
1788 | } | |
1789 | /* | |
1790 | * All component devices are integrity capable and have matching | |
1791 | * profiles, register the common profile for the md device. | |
1792 | */ | |
1793 | if (blk_integrity_register(mddev->gendisk, | |
1794 | bdev_get_integrity(reference->bdev)) != 0) { | |
1795 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1796 | mdname(mddev)); | |
1797 | return -EINVAL; | |
1798 | } | |
1799 | printk(KERN_NOTICE "md: data integrity on %s enabled\n", | |
1800 | mdname(mddev)); | |
1801 | return 0; | |
1802 | } | |
1803 | EXPORT_SYMBOL(md_integrity_register); | |
1804 | ||
1805 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
1806 | void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev) | |
3f9d99c1 | 1807 | { |
3f9d99c1 | 1808 | struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev); |
ac5e7113 | 1809 | struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk); |
3f9d99c1 | 1810 | |
ac5e7113 | 1811 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1812 | return; |
ac5e7113 | 1813 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1814 | return; |
ac5e7113 AN |
1815 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1816 | rdev->bdev->bd_disk) >= 0) | |
1817 | return; | |
1818 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1819 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1820 | } |
ac5e7113 | 1821 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1822 | |
1da177e4 LT |
1823 | static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) |
1824 | { | |
7dd5e7c3 | 1825 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 1826 | struct kobject *ko; |
1edf80d3 | 1827 | char *s; |
5e55e2f5 | 1828 | int err; |
1da177e4 LT |
1829 | |
1830 | if (rdev->mddev) { | |
1831 | MD_BUG(); | |
1832 | return -EINVAL; | |
1833 | } | |
11e2ede0 DW |
1834 | |
1835 | /* prevent duplicates */ | |
1836 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
1837 | return -EEXIST; | |
1838 | ||
dd8ac336 AN |
1839 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
1840 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
1841 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
1842 | if (mddev->pers) { |
1843 | /* Cannot change size, so fail | |
1844 | * If mddev->level <= 0, then we don't care | |
1845 | * about aligning sizes (e.g. linear) | |
1846 | */ | |
1847 | if (mddev->level > 0) | |
1848 | return -ENOSPC; | |
1849 | } else | |
dd8ac336 | 1850 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 1851 | } |
1da177e4 LT |
1852 | |
1853 | /* Verify rdev->desc_nr is unique. | |
1854 | * If it is -1, assign a free number, else | |
1855 | * check number is not in use | |
1856 | */ | |
1857 | if (rdev->desc_nr < 0) { | |
1858 | int choice = 0; | |
1859 | if (mddev->pers) choice = mddev->raid_disks; | |
1860 | while (find_rdev_nr(mddev, choice)) | |
1861 | choice++; | |
1862 | rdev->desc_nr = choice; | |
1863 | } else { | |
1864 | if (find_rdev_nr(mddev, rdev->desc_nr)) | |
1865 | return -EBUSY; | |
1866 | } | |
de01dfad N |
1867 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
1868 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
1869 | mdname(mddev), mddev->max_disks); | |
1870 | return -EBUSY; | |
1871 | } | |
19133a42 | 1872 | bdevname(rdev->bdev,b); |
649316b2 | 1873 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 1874 | *s = '!'; |
649316b2 | 1875 | |
1da177e4 | 1876 | rdev->mddev = mddev; |
19133a42 | 1877 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 1878 | |
b2d6db58 | 1879 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 1880 | goto fail; |
86e6ffdd | 1881 | |
0762b8bd | 1882 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
1883 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
1884 | /* failure here is OK */; | |
1885 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 1886 | |
4b80991c | 1887 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 1888 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
1889 | |
1890 | /* May as well allow recovery to be retried once */ | |
1891 | mddev->recovery_disabled = 0; | |
3f9d99c1 | 1892 | |
1da177e4 | 1893 | return 0; |
5e55e2f5 N |
1894 | |
1895 | fail: | |
1896 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
1897 | b, mdname(mddev)); | |
1898 | return err; | |
1da177e4 LT |
1899 | } |
1900 | ||
177a99b2 | 1901 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 N |
1902 | { |
1903 | mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work); | |
1904 | kobject_del(&rdev->kobj); | |
177a99b2 | 1905 | kobject_put(&rdev->kobj); |
5792a285 N |
1906 | } |
1907 | ||
1da177e4 LT |
1908 | static void unbind_rdev_from_array(mdk_rdev_t * rdev) |
1909 | { | |
1910 | char b[BDEVNAME_SIZE]; | |
1911 | if (!rdev->mddev) { | |
1912 | MD_BUG(); | |
1913 | return; | |
1914 | } | |
49731baa | 1915 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 1916 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
1917 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
1918 | rdev->mddev = NULL; | |
86e6ffdd | 1919 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
1920 | sysfs_put(rdev->sysfs_state); |
1921 | rdev->sysfs_state = NULL; | |
5792a285 | 1922 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
1923 | * writing to 'remove' to "dev/state". We also need |
1924 | * to delay it due to rcu usage. | |
5792a285 | 1925 | */ |
4b80991c | 1926 | synchronize_rcu(); |
177a99b2 N |
1927 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
1928 | kobject_get(&rdev->kobj); | |
e804ac78 | 1929 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
1930 | } |
1931 | ||
1932 | /* | |
1933 | * prevent the device from being mounted, repartitioned or | |
1934 | * otherwise reused by a RAID array (or any other kernel | |
1935 | * subsystem), by bd_claiming the device. | |
1936 | */ | |
c5d79adb | 1937 | static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared) |
1da177e4 LT |
1938 | { |
1939 | int err = 0; | |
1940 | struct block_device *bdev; | |
1941 | char b[BDEVNAME_SIZE]; | |
1942 | ||
d4d77629 TH |
1943 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
1944 | shared ? (mdk_rdev_t *)lock_rdev : rdev); | |
1da177e4 LT |
1945 | if (IS_ERR(bdev)) { |
1946 | printk(KERN_ERR "md: could not open %s.\n", | |
1947 | __bdevname(dev, b)); | |
1948 | return PTR_ERR(bdev); | |
1949 | } | |
1da177e4 LT |
1950 | rdev->bdev = bdev; |
1951 | return err; | |
1952 | } | |
1953 | ||
1954 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1955 | { | |
1956 | struct block_device *bdev = rdev->bdev; | |
1957 | rdev->bdev = NULL; | |
1958 | if (!bdev) | |
1959 | MD_BUG(); | |
e525fd89 | 1960 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
1961 | } |
1962 | ||
1963 | void md_autodetect_dev(dev_t dev); | |
1964 | ||
1965 | static void export_rdev(mdk_rdev_t * rdev) | |
1966 | { | |
1967 | char b[BDEVNAME_SIZE]; | |
1968 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1969 | bdevname(rdev->bdev,b)); | |
1970 | if (rdev->mddev) | |
1971 | MD_BUG(); | |
1972 | free_disk_sb(rdev); | |
1da177e4 | 1973 | #ifndef MODULE |
d0fae18f N |
1974 | if (test_bit(AutoDetected, &rdev->flags)) |
1975 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1976 | #endif |
1977 | unlock_rdev(rdev); | |
86e6ffdd | 1978 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1979 | } |
1980 | ||
1981 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1982 | { | |
1983 | unbind_rdev_from_array(rdev); | |
1984 | export_rdev(rdev); | |
1985 | } | |
1986 | ||
1987 | static void export_array(mddev_t *mddev) | |
1988 | { | |
159ec1fc | 1989 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 1990 | |
d089c6af | 1991 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1992 | if (!rdev->mddev) { |
1993 | MD_BUG(); | |
1994 | continue; | |
1995 | } | |
1996 | kick_rdev_from_array(rdev); | |
1997 | } | |
1998 | if (!list_empty(&mddev->disks)) | |
1999 | MD_BUG(); | |
2000 | mddev->raid_disks = 0; | |
2001 | mddev->major_version = 0; | |
2002 | } | |
2003 | ||
2004 | static void print_desc(mdp_disk_t *desc) | |
2005 | { | |
2006 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
2007 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
2008 | } | |
2009 | ||
cd2ac932 | 2010 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
2011 | { |
2012 | int i; | |
2013 | ||
2014 | printk(KERN_INFO | |
2015 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
2016 | sb->major_version, sb->minor_version, sb->patch_version, | |
2017 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
2018 | sb->ctime); | |
2019 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
2020 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
2021 | sb->md_minor, sb->layout, sb->chunk_size); | |
2022 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
2023 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
2024 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
2025 | sb->failed_disks, sb->spare_disks, | |
2026 | sb->sb_csum, (unsigned long)sb->events_lo); | |
2027 | ||
2028 | printk(KERN_INFO); | |
2029 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2030 | mdp_disk_t *desc; | |
2031 | ||
2032 | desc = sb->disks + i; | |
2033 | if (desc->number || desc->major || desc->minor || | |
2034 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
2035 | printk(" D %2d: ", i); | |
2036 | print_desc(desc); | |
2037 | } | |
2038 | } | |
2039 | printk(KERN_INFO "md: THIS: "); | |
2040 | print_desc(&sb->this_disk); | |
cd2ac932 | 2041 | } |
1da177e4 | 2042 | |
cd2ac932 CR |
2043 | static void print_sb_1(struct mdp_superblock_1 *sb) |
2044 | { | |
2045 | __u8 *uuid; | |
2046 | ||
2047 | uuid = sb->set_uuid; | |
ad361c98 | 2048 | printk(KERN_INFO |
7b75c2f8 | 2049 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 2050 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
2051 | le32_to_cpu(sb->major_version), |
2052 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 2053 | uuid, |
cd2ac932 CR |
2054 | sb->set_name, |
2055 | (unsigned long long)le64_to_cpu(sb->ctime) | |
2056 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
2057 | ||
2058 | uuid = sb->device_uuid; | |
ad361c98 JP |
2059 | printk(KERN_INFO |
2060 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 2061 | " RO:%llu\n" |
7b75c2f8 | 2062 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
2063 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
2064 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
2065 | le32_to_cpu(sb->level), |
2066 | (unsigned long long)le64_to_cpu(sb->size), | |
2067 | le32_to_cpu(sb->raid_disks), | |
2068 | le32_to_cpu(sb->layout), | |
2069 | le32_to_cpu(sb->chunksize), | |
2070 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2071 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2072 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2073 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2074 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2075 | uuid, |
cd2ac932 CR |
2076 | sb->devflags, |
2077 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2078 | (unsigned long long)le64_to_cpu(sb->events), | |
2079 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2080 | le32_to_cpu(sb->sb_csum), | |
2081 | le32_to_cpu(sb->max_dev) | |
2082 | ); | |
1da177e4 LT |
2083 | } |
2084 | ||
cd2ac932 | 2085 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
2086 | { |
2087 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2088 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2089 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2090 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2091 | rdev->desc_nr); | |
1da177e4 | 2092 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2093 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2094 | switch (major_version) { | |
2095 | case 0: | |
2096 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
2097 | break; | |
2098 | case 1: | |
2099 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
2100 | break; | |
2101 | } | |
1da177e4 LT |
2102 | } else |
2103 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2104 | } | |
2105 | ||
5e56341d | 2106 | static void md_print_devices(void) |
1da177e4 | 2107 | { |
159ec1fc | 2108 | struct list_head *tmp; |
1da177e4 LT |
2109 | mdk_rdev_t *rdev; |
2110 | mddev_t *mddev; | |
2111 | char b[BDEVNAME_SIZE]; | |
2112 | ||
2113 | printk("\n"); | |
2114 | printk("md: **********************************\n"); | |
2115 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2116 | printk("md: **********************************\n"); | |
29ac4aa3 | 2117 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2118 | |
32a7627c N |
2119 | if (mddev->bitmap) |
2120 | bitmap_print_sb(mddev->bitmap); | |
2121 | else | |
2122 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 2123 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
2124 | printk("<%s>", bdevname(rdev->bdev,b)); |
2125 | printk("\n"); | |
2126 | ||
159ec1fc | 2127 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 2128 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2129 | } |
2130 | printk("md: **********************************\n"); | |
2131 | printk("\n"); | |
2132 | } | |
2133 | ||
2134 | ||
42543769 | 2135 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 2136 | { |
42543769 N |
2137 | /* Update each superblock (in-memory image), but |
2138 | * if we are allowed to, skip spares which already | |
2139 | * have the right event counter, or have one earlier | |
2140 | * (which would mean they aren't being marked as dirty | |
2141 | * with the rest of the array) | |
2142 | */ | |
1da177e4 | 2143 | mdk_rdev_t *rdev; |
159ec1fc | 2144 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
2145 | if (rdev->sb_events == mddev->events || |
2146 | (nospares && | |
2147 | rdev->raid_disk < 0 && | |
42543769 N |
2148 | rdev->sb_events+1 == mddev->events)) { |
2149 | /* Don't update this superblock */ | |
2150 | rdev->sb_loaded = 2; | |
2151 | } else { | |
2152 | super_types[mddev->major_version]. | |
2153 | sync_super(mddev, rdev); | |
2154 | rdev->sb_loaded = 1; | |
2155 | } | |
1da177e4 LT |
2156 | } |
2157 | } | |
2158 | ||
850b2b42 | 2159 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 2160 | { |
1da177e4 | 2161 | mdk_rdev_t *rdev; |
06d91a5f | 2162 | int sync_req; |
42543769 | 2163 | int nospares = 0; |
1da177e4 | 2164 | |
1da177e4 | 2165 | repeat: |
3a3a5ddb N |
2166 | /* First make sure individual recovery_offsets are correct */ |
2167 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
2168 | if (rdev->raid_disk >= 0 && | |
2169 | mddev->delta_disks >= 0 && | |
2170 | !test_bit(In_sync, &rdev->flags) && | |
2171 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2172 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2173 | ||
2174 | } | |
bd52b746 | 2175 | if (!mddev->persistent) { |
070dc6dd | 2176 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2177 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
d97a41dc N |
2178 | if (!mddev->external) |
2179 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
3a3a5ddb N |
2180 | wake_up(&mddev->sb_wait); |
2181 | return; | |
2182 | } | |
2183 | ||
a9701a30 | 2184 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2185 | |
3a3a5ddb N |
2186 | mddev->utime = get_seconds(); |
2187 | ||
850b2b42 N |
2188 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2189 | force_change = 1; | |
2190 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2191 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2192 | * though if events isn't the right even/odd, we will have to do | |
2193 | * spares after all | |
2194 | */ | |
2195 | nospares = 1; | |
2196 | if (force_change) | |
2197 | nospares = 0; | |
2198 | if (mddev->degraded) | |
84692195 N |
2199 | /* If the array is degraded, then skipping spares is both |
2200 | * dangerous and fairly pointless. | |
2201 | * Dangerous because a device that was removed from the array | |
2202 | * might have a event_count that still looks up-to-date, | |
2203 | * so it can be re-added without a resync. | |
2204 | * Pointless because if there are any spares to skip, | |
2205 | * then a recovery will happen and soon that array won't | |
2206 | * be degraded any more and the spare can go back to sleep then. | |
2207 | */ | |
850b2b42 | 2208 | nospares = 0; |
84692195 | 2209 | |
06d91a5f | 2210 | sync_req = mddev->in_sync; |
42543769 N |
2211 | |
2212 | /* If this is just a dirty<->clean transition, and the array is clean | |
2213 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2214 | if (nospares |
42543769 | 2215 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2216 | && mddev->can_decrease_events |
2217 | && mddev->events != 1) { | |
42543769 | 2218 | mddev->events--; |
a8707c08 N |
2219 | mddev->can_decrease_events = 0; |
2220 | } else { | |
42543769 N |
2221 | /* otherwise we have to go forward and ... */ |
2222 | mddev->events ++; | |
a8707c08 | 2223 | mddev->can_decrease_events = nospares; |
42543769 | 2224 | } |
1da177e4 LT |
2225 | |
2226 | if (!mddev->events) { | |
2227 | /* | |
2228 | * oops, this 64-bit counter should never wrap. | |
2229 | * Either we are in around ~1 trillion A.C., assuming | |
2230 | * 1 reboot per second, or we have a bug: | |
2231 | */ | |
2232 | MD_BUG(); | |
2233 | mddev->events --; | |
2234 | } | |
e691063a | 2235 | sync_sbs(mddev, nospares); |
a9701a30 | 2236 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2237 | |
2238 | dprintk(KERN_INFO | |
2239 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2240 | mdname(mddev),mddev->in_sync); | |
2241 | ||
4ad13663 | 2242 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2243 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2244 | char b[BDEVNAME_SIZE]; |
2245 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2246 | if (rdev->sb_loaded != 1) |
2247 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2248 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2249 | dprintk("(skipping faulty "); |
2250 | ||
2251 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2252 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2253 | md_super_write(mddev,rdev, |
0f420358 | 2254 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2255 | rdev->sb_page); |
2256 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2257 | bdevname(rdev->bdev,b), | |
0f420358 | 2258 | (unsigned long long)rdev->sb_start); |
42543769 | 2259 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2260 | |
1da177e4 LT |
2261 | } else |
2262 | dprintk(")\n"); | |
7bfa19f2 | 2263 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2264 | /* only need to write one superblock... */ |
2265 | break; | |
2266 | } | |
a9701a30 | 2267 | md_super_wait(mddev); |
850b2b42 | 2268 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2269 | |
a9701a30 | 2270 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2271 | if (mddev->in_sync != sync_req || |
2272 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2273 | /* have to write it out again */ |
a9701a30 | 2274 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2275 | goto repeat; |
2276 | } | |
850b2b42 | 2277 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2278 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2279 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2280 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2281 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2282 | |
1da177e4 LT |
2283 | } |
2284 | ||
7f6ce769 | 2285 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2286 | * We want to accept with case. For this we use cmd_match. |
2287 | */ | |
2288 | static int cmd_match(const char *cmd, const char *str) | |
2289 | { | |
2290 | /* See if cmd, written into a sysfs file, matches | |
2291 | * str. They must either be the same, or cmd can | |
2292 | * have a trailing newline | |
2293 | */ | |
2294 | while (*cmd && *str && *cmd == *str) { | |
2295 | cmd++; | |
2296 | str++; | |
2297 | } | |
2298 | if (*cmd == '\n') | |
2299 | cmd++; | |
2300 | if (*str || *cmd) | |
2301 | return 0; | |
2302 | return 1; | |
2303 | } | |
2304 | ||
86e6ffdd N |
2305 | struct rdev_sysfs_entry { |
2306 | struct attribute attr; | |
2307 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2308 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2309 | }; | |
2310 | ||
2311 | static ssize_t | |
96de1e66 | 2312 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2313 | { |
2314 | char *sep = ""; | |
20a49ff6 | 2315 | size_t len = 0; |
86e6ffdd | 2316 | |
b2d444d7 | 2317 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2318 | len+= sprintf(page+len, "%sfaulty",sep); |
2319 | sep = ","; | |
2320 | } | |
b2d444d7 | 2321 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2322 | len += sprintf(page+len, "%sin_sync",sep); |
2323 | sep = ","; | |
2324 | } | |
f655675b N |
2325 | if (test_bit(WriteMostly, &rdev->flags)) { |
2326 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2327 | sep = ","; | |
2328 | } | |
6bfe0b49 DW |
2329 | if (test_bit(Blocked, &rdev->flags)) { |
2330 | len += sprintf(page+len, "%sblocked", sep); | |
2331 | sep = ","; | |
2332 | } | |
b2d444d7 N |
2333 | if (!test_bit(Faulty, &rdev->flags) && |
2334 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2335 | len += sprintf(page+len, "%sspare", sep); |
2336 | sep = ","; | |
2337 | } | |
2338 | return len+sprintf(page+len, "\n"); | |
2339 | } | |
2340 | ||
45dc2de1 N |
2341 | static ssize_t |
2342 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2343 | { | |
2344 | /* can write | |
2345 | * faulty - simulates and error | |
2346 | * remove - disconnects the device | |
f655675b N |
2347 | * writemostly - sets write_mostly |
2348 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2349 | * blocked - sets the Blocked flag |
2350 | * -blocked - clears the Blocked flag | |
6d56e278 | 2351 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2352 | */ |
2353 | int err = -EINVAL; | |
2354 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2355 | md_error(rdev->mddev, rdev); | |
2356 | err = 0; | |
2357 | } else if (cmd_match(buf, "remove")) { | |
2358 | if (rdev->raid_disk >= 0) | |
2359 | err = -EBUSY; | |
2360 | else { | |
2361 | mddev_t *mddev = rdev->mddev; | |
2362 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2363 | if (mddev->pers) |
2364 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2365 | md_new_event(mddev); |
2366 | err = 0; | |
2367 | } | |
f655675b N |
2368 | } else if (cmd_match(buf, "writemostly")) { |
2369 | set_bit(WriteMostly, &rdev->flags); | |
2370 | err = 0; | |
2371 | } else if (cmd_match(buf, "-writemostly")) { | |
2372 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2373 | err = 0; |
2374 | } else if (cmd_match(buf, "blocked")) { | |
2375 | set_bit(Blocked, &rdev->flags); | |
2376 | err = 0; | |
2377 | } else if (cmd_match(buf, "-blocked")) { | |
2378 | clear_bit(Blocked, &rdev->flags); | |
2379 | wake_up(&rdev->blocked_wait); | |
2380 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2381 | md_wakeup_thread(rdev->mddev->thread); | |
2382 | ||
6d56e278 N |
2383 | err = 0; |
2384 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2385 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2386 | err = 0; |
45dc2de1 | 2387 | } |
00bcb4ac N |
2388 | if (!err) |
2389 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2390 | return err ? err : len; |
2391 | } | |
80ca3a44 N |
2392 | static struct rdev_sysfs_entry rdev_state = |
2393 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2394 | |
4dbcdc75 N |
2395 | static ssize_t |
2396 | errors_show(mdk_rdev_t *rdev, char *page) | |
2397 | { | |
2398 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2399 | } | |
2400 | ||
2401 | static ssize_t | |
2402 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2403 | { | |
2404 | char *e; | |
2405 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2406 | if (*buf && (*e == 0 || *e == '\n')) { | |
2407 | atomic_set(&rdev->corrected_errors, n); | |
2408 | return len; | |
2409 | } | |
2410 | return -EINVAL; | |
2411 | } | |
2412 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2413 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2414 | |
014236d2 N |
2415 | static ssize_t |
2416 | slot_show(mdk_rdev_t *rdev, char *page) | |
2417 | { | |
2418 | if (rdev->raid_disk < 0) | |
2419 | return sprintf(page, "none\n"); | |
2420 | else | |
2421 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2422 | } | |
2423 | ||
2424 | static ssize_t | |
2425 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2426 | { | |
2427 | char *e; | |
c303da6d N |
2428 | int err; |
2429 | char nm[20]; | |
014236d2 N |
2430 | int slot = simple_strtoul(buf, &e, 10); |
2431 | if (strncmp(buf, "none", 4)==0) | |
2432 | slot = -1; | |
2433 | else if (e==buf || (*e && *e!= '\n')) | |
2434 | return -EINVAL; | |
6c2fce2e | 2435 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2436 | /* Setting 'slot' on an active array requires also |
2437 | * updating the 'rd%d' link, and communicating | |
2438 | * with the personality with ->hot_*_disk. | |
2439 | * For now we only support removing | |
2440 | * failed/spare devices. This normally happens automatically, | |
2441 | * but not when the metadata is externally managed. | |
2442 | */ | |
c303da6d N |
2443 | if (rdev->raid_disk == -1) |
2444 | return -EEXIST; | |
2445 | /* personality does all needed checks */ | |
2446 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2447 | return -EINVAL; | |
2448 | err = rdev->mddev->pers-> | |
2449 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2450 | if (err) | |
2451 | return err; | |
2452 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2453 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
b7103107 | 2454 | rdev->raid_disk = -1; |
c303da6d N |
2455 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2456 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2457 | } else if (rdev->mddev->pers) { |
2458 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2459 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2460 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2461 | */ |
2462 | ||
2463 | if (rdev->raid_disk != -1) | |
2464 | return -EBUSY; | |
2465 | ||
2466 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2467 | return -EINVAL; | |
2468 | ||
159ec1fc | 2469 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2470 | if (rdev2->raid_disk == slot) |
2471 | return -EEXIST; | |
2472 | ||
ba1b41b6 N |
2473 | if (slot >= rdev->mddev->raid_disks && |
2474 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2475 | return -ENOSPC; | |
2476 | ||
6c2fce2e NB |
2477 | rdev->raid_disk = slot; |
2478 | if (test_bit(In_sync, &rdev->flags)) | |
2479 | rdev->saved_raid_disk = slot; | |
2480 | else | |
2481 | rdev->saved_raid_disk = -1; | |
2482 | err = rdev->mddev->pers-> | |
2483 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2484 | if (err) { |
6c2fce2e | 2485 | rdev->raid_disk = -1; |
6c2fce2e | 2486 | return err; |
52664732 | 2487 | } else |
00bcb4ac | 2488 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6c2fce2e NB |
2489 | sprintf(nm, "rd%d", rdev->raid_disk); |
2490 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
00bcb4ac | 2491 | /* failure here is OK */; |
6c2fce2e | 2492 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2493 | } else { |
ba1b41b6 N |
2494 | if (slot >= rdev->mddev->raid_disks && |
2495 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2496 | return -ENOSPC; |
2497 | rdev->raid_disk = slot; | |
2498 | /* assume it is working */ | |
c5d79adb N |
2499 | clear_bit(Faulty, &rdev->flags); |
2500 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2501 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2502 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2503 | } |
014236d2 N |
2504 | return len; |
2505 | } | |
2506 | ||
2507 | ||
2508 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2509 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2510 | |
93c8cad0 N |
2511 | static ssize_t |
2512 | offset_show(mdk_rdev_t *rdev, char *page) | |
2513 | { | |
6961ece4 | 2514 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2515 | } |
2516 | ||
2517 | static ssize_t | |
2518 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2519 | { | |
2520 | char *e; | |
2521 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2522 | if (e==buf || (*e && *e != '\n')) | |
2523 | return -EINVAL; | |
8ed0a521 | 2524 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2525 | return -EBUSY; |
dd8ac336 | 2526 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2527 | /* Must set offset before size, so overlap checks |
2528 | * can be sane */ | |
2529 | return -EBUSY; | |
93c8cad0 N |
2530 | rdev->data_offset = offset; |
2531 | return len; | |
2532 | } | |
2533 | ||
2534 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2535 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2536 | |
83303b61 N |
2537 | static ssize_t |
2538 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2539 | { | |
dd8ac336 | 2540 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2541 | } |
2542 | ||
c5d79adb N |
2543 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2544 | { | |
2545 | /* check if two start/length pairs overlap */ | |
2546 | if (s1+l1 <= s2) | |
2547 | return 0; | |
2548 | if (s2+l2 <= s1) | |
2549 | return 0; | |
2550 | return 1; | |
2551 | } | |
2552 | ||
b522adcd DW |
2553 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2554 | { | |
2555 | unsigned long long blocks; | |
2556 | sector_t new; | |
2557 | ||
2558 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2559 | return -EINVAL; | |
2560 | ||
2561 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2562 | return -EINVAL; /* sector conversion overflow */ | |
2563 | ||
2564 | new = blocks * 2; | |
2565 | if (new != blocks * 2) | |
2566 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2567 | ||
2568 | *sectors = new; | |
2569 | return 0; | |
2570 | } | |
2571 | ||
83303b61 N |
2572 | static ssize_t |
2573 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2574 | { | |
27c529bb | 2575 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2576 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2577 | sector_t sectors; |
27c529bb | 2578 | |
b522adcd | 2579 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2580 | return -EINVAL; |
0cd17fec | 2581 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2582 | if (my_mddev->persistent) { |
dd8ac336 AN |
2583 | sectors = super_types[my_mddev->major_version]. |
2584 | rdev_size_change(rdev, sectors); | |
2585 | if (!sectors) | |
0cd17fec | 2586 | return -EBUSY; |
dd8ac336 | 2587 | } else if (!sectors) |
77304d2a | 2588 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2589 | rdev->data_offset; |
0cd17fec | 2590 | } |
dd8ac336 | 2591 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2592 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2593 | |
dd8ac336 AN |
2594 | rdev->sectors = sectors; |
2595 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2596 | /* need to check that all other rdevs with the same ->bdev |
2597 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2598 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2599 | * we have to change it back, we will have the lock again. |
2600 | */ | |
2601 | mddev_t *mddev; | |
2602 | int overlap = 0; | |
159ec1fc | 2603 | struct list_head *tmp; |
c5d79adb | 2604 | |
27c529bb | 2605 | mddev_unlock(my_mddev); |
29ac4aa3 | 2606 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2607 | mdk_rdev_t *rdev2; |
2608 | ||
2609 | mddev_lock(mddev); | |
159ec1fc | 2610 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
f21e9ff7 N |
2611 | if (rdev->bdev == rdev2->bdev && |
2612 | rdev != rdev2 && | |
2613 | overlaps(rdev->data_offset, rdev->sectors, | |
2614 | rdev2->data_offset, | |
2615 | rdev2->sectors)) { | |
c5d79adb N |
2616 | overlap = 1; |
2617 | break; | |
2618 | } | |
2619 | mddev_unlock(mddev); | |
2620 | if (overlap) { | |
2621 | mddev_put(mddev); | |
2622 | break; | |
2623 | } | |
2624 | } | |
27c529bb | 2625 | mddev_lock(my_mddev); |
c5d79adb N |
2626 | if (overlap) { |
2627 | /* Someone else could have slipped in a size | |
2628 | * change here, but doing so is just silly. | |
dd8ac336 | 2629 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2630 | * safe, and trust userspace not to race with |
2631 | * itself | |
2632 | */ | |
dd8ac336 | 2633 | rdev->sectors = oldsectors; |
c5d79adb N |
2634 | return -EBUSY; |
2635 | } | |
2636 | } | |
83303b61 N |
2637 | return len; |
2638 | } | |
2639 | ||
2640 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2641 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2642 | |
06e3c817 DW |
2643 | |
2644 | static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) | |
2645 | { | |
2646 | unsigned long long recovery_start = rdev->recovery_offset; | |
2647 | ||
2648 | if (test_bit(In_sync, &rdev->flags) || | |
2649 | recovery_start == MaxSector) | |
2650 | return sprintf(page, "none\n"); | |
2651 | ||
2652 | return sprintf(page, "%llu\n", recovery_start); | |
2653 | } | |
2654 | ||
2655 | static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2656 | { | |
2657 | unsigned long long recovery_start; | |
2658 | ||
2659 | if (cmd_match(buf, "none")) | |
2660 | recovery_start = MaxSector; | |
2661 | else if (strict_strtoull(buf, 10, &recovery_start)) | |
2662 | return -EINVAL; | |
2663 | ||
2664 | if (rdev->mddev->pers && | |
2665 | rdev->raid_disk >= 0) | |
2666 | return -EBUSY; | |
2667 | ||
2668 | rdev->recovery_offset = recovery_start; | |
2669 | if (recovery_start == MaxSector) | |
2670 | set_bit(In_sync, &rdev->flags); | |
2671 | else | |
2672 | clear_bit(In_sync, &rdev->flags); | |
2673 | return len; | |
2674 | } | |
2675 | ||
2676 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2677 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2678 | ||
86e6ffdd N |
2679 | static struct attribute *rdev_default_attrs[] = { |
2680 | &rdev_state.attr, | |
4dbcdc75 | 2681 | &rdev_errors.attr, |
014236d2 | 2682 | &rdev_slot.attr, |
93c8cad0 | 2683 | &rdev_offset.attr, |
83303b61 | 2684 | &rdev_size.attr, |
06e3c817 | 2685 | &rdev_recovery_start.attr, |
86e6ffdd N |
2686 | NULL, |
2687 | }; | |
2688 | static ssize_t | |
2689 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2690 | { | |
2691 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2692 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2693 | mddev_t *mddev = rdev->mddev; |
2694 | ssize_t rv; | |
86e6ffdd N |
2695 | |
2696 | if (!entry->show) | |
2697 | return -EIO; | |
27c529bb N |
2698 | |
2699 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2700 | if (!rv) { | |
2701 | if (rdev->mddev == NULL) | |
2702 | rv = -EBUSY; | |
2703 | else | |
2704 | rv = entry->show(rdev, page); | |
2705 | mddev_unlock(mddev); | |
2706 | } | |
2707 | return rv; | |
86e6ffdd N |
2708 | } |
2709 | ||
2710 | static ssize_t | |
2711 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2712 | const char *page, size_t length) | |
2713 | { | |
2714 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2715 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2716 | ssize_t rv; |
2717 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2718 | |
2719 | if (!entry->store) | |
2720 | return -EIO; | |
67463acb N |
2721 | if (!capable(CAP_SYS_ADMIN)) |
2722 | return -EACCES; | |
27c529bb | 2723 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2724 | if (!rv) { |
27c529bb N |
2725 | if (rdev->mddev == NULL) |
2726 | rv = -EBUSY; | |
2727 | else | |
2728 | rv = entry->store(rdev, page, length); | |
6a51830e | 2729 | mddev_unlock(mddev); |
ca388059 N |
2730 | } |
2731 | return rv; | |
86e6ffdd N |
2732 | } |
2733 | ||
2734 | static void rdev_free(struct kobject *ko) | |
2735 | { | |
2736 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2737 | kfree(rdev); | |
2738 | } | |
52cf25d0 | 2739 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
2740 | .show = rdev_attr_show, |
2741 | .store = rdev_attr_store, | |
2742 | }; | |
2743 | static struct kobj_type rdev_ktype = { | |
2744 | .release = rdev_free, | |
2745 | .sysfs_ops = &rdev_sysfs_ops, | |
2746 | .default_attrs = rdev_default_attrs, | |
2747 | }; | |
2748 | ||
e8bb9a83 N |
2749 | void md_rdev_init(mdk_rdev_t *rdev) |
2750 | { | |
2751 | rdev->desc_nr = -1; | |
2752 | rdev->saved_raid_disk = -1; | |
2753 | rdev->raid_disk = -1; | |
2754 | rdev->flags = 0; | |
2755 | rdev->data_offset = 0; | |
2756 | rdev->sb_events = 0; | |
2757 | rdev->last_read_error.tv_sec = 0; | |
2758 | rdev->last_read_error.tv_nsec = 0; | |
2759 | atomic_set(&rdev->nr_pending, 0); | |
2760 | atomic_set(&rdev->read_errors, 0); | |
2761 | atomic_set(&rdev->corrected_errors, 0); | |
2762 | ||
2763 | INIT_LIST_HEAD(&rdev->same_set); | |
2764 | init_waitqueue_head(&rdev->blocked_wait); | |
2765 | } | |
2766 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
2767 | /* |
2768 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2769 | * | |
2770 | * mark the device faulty if: | |
2771 | * | |
2772 | * - the device is nonexistent (zero size) | |
2773 | * - the device has no valid superblock | |
2774 | * | |
2775 | * a faulty rdev _never_ has rdev->sb set. | |
2776 | */ | |
2777 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2778 | { | |
2779 | char b[BDEVNAME_SIZE]; | |
2780 | int err; | |
2781 | mdk_rdev_t *rdev; | |
2782 | sector_t size; | |
2783 | ||
9ffae0cf | 2784 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2785 | if (!rdev) { |
2786 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2787 | return ERR_PTR(-ENOMEM); | |
2788 | } | |
1da177e4 | 2789 | |
e8bb9a83 | 2790 | md_rdev_init(rdev); |
1da177e4 LT |
2791 | if ((err = alloc_disk_sb(rdev))) |
2792 | goto abort_free; | |
2793 | ||
c5d79adb | 2794 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2795 | if (err) |
2796 | goto abort_free; | |
2797 | ||
f9cb074b | 2798 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2799 | |
77304d2a | 2800 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 LT |
2801 | if (!size) { |
2802 | printk(KERN_WARNING | |
2803 | "md: %s has zero or unknown size, marking faulty!\n", | |
2804 | bdevname(rdev->bdev,b)); | |
2805 | err = -EINVAL; | |
2806 | goto abort_free; | |
2807 | } | |
2808 | ||
2809 | if (super_format >= 0) { | |
2810 | err = super_types[super_format]. | |
2811 | load_super(rdev, NULL, super_minor); | |
2812 | if (err == -EINVAL) { | |
df968c4e N |
2813 | printk(KERN_WARNING |
2814 | "md: %s does not have a valid v%d.%d " | |
2815 | "superblock, not importing!\n", | |
2816 | bdevname(rdev->bdev,b), | |
2817 | super_format, super_minor); | |
1da177e4 LT |
2818 | goto abort_free; |
2819 | } | |
2820 | if (err < 0) { | |
2821 | printk(KERN_WARNING | |
2822 | "md: could not read %s's sb, not importing!\n", | |
2823 | bdevname(rdev->bdev,b)); | |
2824 | goto abort_free; | |
2825 | } | |
2826 | } | |
6bfe0b49 | 2827 | |
1da177e4 LT |
2828 | return rdev; |
2829 | ||
2830 | abort_free: | |
2831 | if (rdev->sb_page) { | |
2832 | if (rdev->bdev) | |
2833 | unlock_rdev(rdev); | |
2834 | free_disk_sb(rdev); | |
2835 | } | |
2836 | kfree(rdev); | |
2837 | return ERR_PTR(err); | |
2838 | } | |
2839 | ||
2840 | /* | |
2841 | * Check a full RAID array for plausibility | |
2842 | */ | |
2843 | ||
2844 | ||
a757e64c | 2845 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2846 | { |
2847 | int i; | |
159ec1fc | 2848 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2849 | char b[BDEVNAME_SIZE]; |
2850 | ||
2851 | freshest = NULL; | |
d089c6af | 2852 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2853 | switch (super_types[mddev->major_version]. |
2854 | load_super(rdev, freshest, mddev->minor_version)) { | |
2855 | case 1: | |
2856 | freshest = rdev; | |
2857 | break; | |
2858 | case 0: | |
2859 | break; | |
2860 | default: | |
2861 | printk( KERN_ERR \ | |
2862 | "md: fatal superblock inconsistency in %s" | |
2863 | " -- removing from array\n", | |
2864 | bdevname(rdev->bdev,b)); | |
2865 | kick_rdev_from_array(rdev); | |
2866 | } | |
2867 | ||
2868 | ||
2869 | super_types[mddev->major_version]. | |
2870 | validate_super(mddev, freshest); | |
2871 | ||
2872 | i = 0; | |
d089c6af | 2873 | rdev_for_each(rdev, tmp, mddev) { |
233fca36 N |
2874 | if (mddev->max_disks && |
2875 | (rdev->desc_nr >= mddev->max_disks || | |
2876 | i > mddev->max_disks)) { | |
de01dfad N |
2877 | printk(KERN_WARNING |
2878 | "md: %s: %s: only %d devices permitted\n", | |
2879 | mdname(mddev), bdevname(rdev->bdev, b), | |
2880 | mddev->max_disks); | |
2881 | kick_rdev_from_array(rdev); | |
2882 | continue; | |
2883 | } | |
1da177e4 LT |
2884 | if (rdev != freshest) |
2885 | if (super_types[mddev->major_version]. | |
2886 | validate_super(mddev, rdev)) { | |
2887 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2888 | " from array!\n", | |
2889 | bdevname(rdev->bdev,b)); | |
2890 | kick_rdev_from_array(rdev); | |
2891 | continue; | |
2892 | } | |
2893 | if (mddev->level == LEVEL_MULTIPATH) { | |
2894 | rdev->desc_nr = i++; | |
2895 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2896 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 2897 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
2898 | rdev->raid_disk = -1; |
2899 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2900 | } |
2901 | } | |
1da177e4 LT |
2902 | } |
2903 | ||
72e02075 N |
2904 | /* Read a fixed-point number. |
2905 | * Numbers in sysfs attributes should be in "standard" units where | |
2906 | * possible, so time should be in seconds. | |
2907 | * However we internally use a a much smaller unit such as | |
2908 | * milliseconds or jiffies. | |
2909 | * This function takes a decimal number with a possible fractional | |
2910 | * component, and produces an integer which is the result of | |
2911 | * multiplying that number by 10^'scale'. | |
2912 | * all without any floating-point arithmetic. | |
2913 | */ | |
2914 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
2915 | { | |
2916 | unsigned long result = 0; | |
2917 | long decimals = -1; | |
2918 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
2919 | if (*cp == '.') | |
2920 | decimals = 0; | |
2921 | else if (decimals < scale) { | |
2922 | unsigned int value; | |
2923 | value = *cp - '0'; | |
2924 | result = result * 10 + value; | |
2925 | if (decimals >= 0) | |
2926 | decimals++; | |
2927 | } | |
2928 | cp++; | |
2929 | } | |
2930 | if (*cp == '\n') | |
2931 | cp++; | |
2932 | if (*cp) | |
2933 | return -EINVAL; | |
2934 | if (decimals < 0) | |
2935 | decimals = 0; | |
2936 | while (decimals < scale) { | |
2937 | result *= 10; | |
2938 | decimals ++; | |
2939 | } | |
2940 | *res = result; | |
2941 | return 0; | |
2942 | } | |
2943 | ||
2944 | ||
19052c0e N |
2945 | static void md_safemode_timeout(unsigned long data); |
2946 | ||
16f17b39 N |
2947 | static ssize_t |
2948 | safe_delay_show(mddev_t *mddev, char *page) | |
2949 | { | |
2950 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2951 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2952 | } | |
2953 | static ssize_t | |
2954 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2955 | { | |
16f17b39 | 2956 | unsigned long msec; |
97ce0a7f | 2957 | |
72e02075 | 2958 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 2959 | return -EINVAL; |
16f17b39 N |
2960 | if (msec == 0) |
2961 | mddev->safemode_delay = 0; | |
2962 | else { | |
19052c0e | 2963 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2964 | mddev->safemode_delay = (msec*HZ)/1000; |
2965 | if (mddev->safemode_delay == 0) | |
2966 | mddev->safemode_delay = 1; | |
19052c0e N |
2967 | if (mddev->safemode_delay < old_delay) |
2968 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2969 | } |
2970 | return len; | |
2971 | } | |
2972 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2973 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2974 | |
eae1701f | 2975 | static ssize_t |
96de1e66 | 2976 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2977 | { |
2604b703 | 2978 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2979 | if (p) |
eae1701f | 2980 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2981 | else if (mddev->clevel[0]) |
2982 | return sprintf(page, "%s\n", mddev->clevel); | |
2983 | else if (mddev->level != LEVEL_NONE) | |
2984 | return sprintf(page, "%d\n", mddev->level); | |
2985 | else | |
2986 | return 0; | |
eae1701f N |
2987 | } |
2988 | ||
d9d166c2 N |
2989 | static ssize_t |
2990 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2991 | { | |
f2859af6 | 2992 | char clevel[16]; |
20a49ff6 | 2993 | ssize_t rv = len; |
245f46c2 | 2994 | struct mdk_personality *pers; |
f2859af6 | 2995 | long level; |
245f46c2 | 2996 | void *priv; |
3a981b03 | 2997 | mdk_rdev_t *rdev; |
245f46c2 N |
2998 | |
2999 | if (mddev->pers == NULL) { | |
3000 | if (len == 0) | |
3001 | return 0; | |
3002 | if (len >= sizeof(mddev->clevel)) | |
3003 | return -ENOSPC; | |
3004 | strncpy(mddev->clevel, buf, len); | |
3005 | if (mddev->clevel[len-1] == '\n') | |
3006 | len--; | |
3007 | mddev->clevel[len] = 0; | |
3008 | mddev->level = LEVEL_NONE; | |
3009 | return rv; | |
3010 | } | |
3011 | ||
3012 | /* request to change the personality. Need to ensure: | |
3013 | * - array is not engaged in resync/recovery/reshape | |
3014 | * - old personality can be suspended | |
3015 | * - new personality will access other array. | |
3016 | */ | |
3017 | ||
bb4f1e9d N |
3018 | if (mddev->sync_thread || |
3019 | mddev->reshape_position != MaxSector || | |
3020 | mddev->sysfs_active) | |
d9d166c2 | 3021 | return -EBUSY; |
245f46c2 N |
3022 | |
3023 | if (!mddev->pers->quiesce) { | |
3024 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3025 | mdname(mddev), mddev->pers->name); | |
3026 | return -EINVAL; | |
3027 | } | |
3028 | ||
3029 | /* Now find the new personality */ | |
f2859af6 | 3030 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3031 | return -EINVAL; |
f2859af6 DW |
3032 | strncpy(clevel, buf, len); |
3033 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3034 | len--; |
f2859af6 DW |
3035 | clevel[len] = 0; |
3036 | if (strict_strtol(clevel, 10, &level)) | |
3037 | level = LEVEL_NONE; | |
245f46c2 | 3038 | |
f2859af6 DW |
3039 | if (request_module("md-%s", clevel) != 0) |
3040 | request_module("md-level-%s", clevel); | |
245f46c2 | 3041 | spin_lock(&pers_lock); |
f2859af6 | 3042 | pers = find_pers(level, clevel); |
245f46c2 N |
3043 | if (!pers || !try_module_get(pers->owner)) { |
3044 | spin_unlock(&pers_lock); | |
f2859af6 | 3045 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3046 | return -EINVAL; |
3047 | } | |
3048 | spin_unlock(&pers_lock); | |
3049 | ||
3050 | if (pers == mddev->pers) { | |
3051 | /* Nothing to do! */ | |
3052 | module_put(pers->owner); | |
3053 | return rv; | |
3054 | } | |
3055 | if (!pers->takeover) { | |
3056 | module_put(pers->owner); | |
3057 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3058 | mdname(mddev), clevel); |
245f46c2 N |
3059 | return -EINVAL; |
3060 | } | |
3061 | ||
e93f68a1 N |
3062 | list_for_each_entry(rdev, &mddev->disks, same_set) |
3063 | rdev->new_raid_disk = rdev->raid_disk; | |
3064 | ||
245f46c2 N |
3065 | /* ->takeover must set new_* and/or delta_disks |
3066 | * if it succeeds, and may set them when it fails. | |
3067 | */ | |
3068 | priv = pers->takeover(mddev); | |
3069 | if (IS_ERR(priv)) { | |
3070 | mddev->new_level = mddev->level; | |
3071 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3072 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3073 | mddev->raid_disks -= mddev->delta_disks; |
3074 | mddev->delta_disks = 0; | |
3075 | module_put(pers->owner); | |
3076 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3077 | mdname(mddev), clevel); |
245f46c2 N |
3078 | return PTR_ERR(priv); |
3079 | } | |
3080 | ||
3081 | /* Looks like we have a winner */ | |
3082 | mddev_suspend(mddev); | |
3083 | mddev->pers->stop(mddev); | |
a64c876f N |
3084 | |
3085 | if (mddev->pers->sync_request == NULL && | |
3086 | pers->sync_request != NULL) { | |
3087 | /* need to add the md_redundancy_group */ | |
3088 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3089 | printk(KERN_WARNING | |
3090 | "md: cannot register extra attributes for %s\n", | |
3091 | mdname(mddev)); | |
19fdb9ee | 3092 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); |
a64c876f N |
3093 | } |
3094 | if (mddev->pers->sync_request != NULL && | |
3095 | pers->sync_request == NULL) { | |
3096 | /* need to remove the md_redundancy_group */ | |
3097 | if (mddev->to_remove == NULL) | |
3098 | mddev->to_remove = &md_redundancy_group; | |
3099 | } | |
3100 | ||
54071b38 TM |
3101 | if (mddev->pers->sync_request == NULL && |
3102 | mddev->external) { | |
3103 | /* We are converting from a no-redundancy array | |
3104 | * to a redundancy array and metadata is managed | |
3105 | * externally so we need to be sure that writes | |
3106 | * won't block due to a need to transition | |
3107 | * clean->dirty | |
3108 | * until external management is started. | |
3109 | */ | |
3110 | mddev->in_sync = 0; | |
3111 | mddev->safemode_delay = 0; | |
3112 | mddev->safemode = 0; | |
3113 | } | |
3114 | ||
e93f68a1 N |
3115 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
3116 | char nm[20]; | |
3117 | if (rdev->raid_disk < 0) | |
3118 | continue; | |
bf2cb0da | 3119 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3120 | rdev->new_raid_disk = -1; |
3121 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3122 | continue; | |
3123 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3124 | sysfs_remove_link(&mddev->kobj, nm); | |
3125 | } | |
3126 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
3127 | if (rdev->raid_disk < 0) | |
3128 | continue; | |
3129 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3130 | continue; | |
3131 | rdev->raid_disk = rdev->new_raid_disk; | |
3132 | if (rdev->raid_disk < 0) | |
3a981b03 | 3133 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 N |
3134 | else { |
3135 | char nm[20]; | |
3136 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3137 | if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) | |
3138 | printk("md: cannot register %s for %s after level change\n", | |
3139 | nm, mdname(mddev)); | |
3a981b03 | 3140 | } |
e93f68a1 N |
3141 | } |
3142 | ||
3143 | module_put(mddev->pers->owner); | |
245f46c2 N |
3144 | mddev->pers = pers; |
3145 | mddev->private = priv; | |
3146 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3147 | mddev->level = mddev->new_level; | |
3148 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3149 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3150 | mddev->delta_disks = 0; |
9af204cf TM |
3151 | if (mddev->pers->sync_request == NULL) { |
3152 | /* this is now an array without redundancy, so | |
3153 | * it must always be in_sync | |
3154 | */ | |
3155 | mddev->in_sync = 1; | |
3156 | del_timer_sync(&mddev->safemode_timer); | |
3157 | } | |
245f46c2 N |
3158 | pers->run(mddev); |
3159 | mddev_resume(mddev); | |
3160 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3161 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3162 | md_wakeup_thread(mddev->thread); | |
5cac7861 | 3163 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3164 | md_new_event(mddev); |
d9d166c2 N |
3165 | return rv; |
3166 | } | |
3167 | ||
3168 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3169 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3170 | |
d4dbd025 N |
3171 | |
3172 | static ssize_t | |
3173 | layout_show(mddev_t *mddev, char *page) | |
3174 | { | |
3175 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3176 | if (mddev->reshape_position != MaxSector && |
3177 | mddev->layout != mddev->new_layout) | |
3178 | return sprintf(page, "%d (%d)\n", | |
3179 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3180 | return sprintf(page, "%d\n", mddev->layout); |
3181 | } | |
3182 | ||
3183 | static ssize_t | |
3184 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
3185 | { | |
3186 | char *e; | |
3187 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3188 | |
3189 | if (!*buf || (*e && *e != '\n')) | |
3190 | return -EINVAL; | |
3191 | ||
b3546035 N |
3192 | if (mddev->pers) { |
3193 | int err; | |
50ac168a | 3194 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3195 | return -EBUSY; |
597a711b | 3196 | mddev->new_layout = n; |
50ac168a | 3197 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3198 | if (err) { |
3199 | mddev->new_layout = mddev->layout; | |
b3546035 | 3200 | return err; |
597a711b | 3201 | } |
b3546035 | 3202 | } else { |
08a02ecd | 3203 | mddev->new_layout = n; |
b3546035 N |
3204 | if (mddev->reshape_position == MaxSector) |
3205 | mddev->layout = n; | |
3206 | } | |
d4dbd025 N |
3207 | return len; |
3208 | } | |
3209 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3210 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3211 | |
3212 | ||
eae1701f | 3213 | static ssize_t |
96de1e66 | 3214 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 3215 | { |
bb636547 N |
3216 | if (mddev->raid_disks == 0) |
3217 | return 0; | |
08a02ecd N |
3218 | if (mddev->reshape_position != MaxSector && |
3219 | mddev->delta_disks != 0) | |
3220 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3221 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3222 | return sprintf(page, "%d\n", mddev->raid_disks); |
3223 | } | |
3224 | ||
da943b99 N |
3225 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
3226 | ||
3227 | static ssize_t | |
3228 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
3229 | { | |
da943b99 N |
3230 | char *e; |
3231 | int rv = 0; | |
3232 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3233 | ||
3234 | if (!*buf || (*e && *e != '\n')) | |
3235 | return -EINVAL; | |
3236 | ||
3237 | if (mddev->pers) | |
3238 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
3239 | else if (mddev->reshape_position != MaxSector) { |
3240 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
3241 | mddev->delta_disks = n - olddisks; | |
3242 | mddev->raid_disks = n; | |
3243 | } else | |
da943b99 N |
3244 | mddev->raid_disks = n; |
3245 | return rv ? rv : len; | |
3246 | } | |
3247 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3248 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3249 | |
3b34380a N |
3250 | static ssize_t |
3251 | chunk_size_show(mddev_t *mddev, char *page) | |
3252 | { | |
08a02ecd | 3253 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3254 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3255 | return sprintf(page, "%d (%d)\n", | |
3256 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3257 | mddev->chunk_sectors << 9); |
3258 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3259 | } |
3260 | ||
3261 | static ssize_t | |
3262 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3263 | { | |
3b34380a N |
3264 | char *e; |
3265 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3266 | ||
3b34380a N |
3267 | if (!*buf || (*e && *e != '\n')) |
3268 | return -EINVAL; | |
3269 | ||
b3546035 N |
3270 | if (mddev->pers) { |
3271 | int err; | |
50ac168a | 3272 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3273 | return -EBUSY; |
597a711b | 3274 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3275 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3276 | if (err) { |
3277 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3278 | return err; |
597a711b | 3279 | } |
b3546035 | 3280 | } else { |
664e7c41 | 3281 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3282 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3283 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3284 | } |
3b34380a N |
3285 | return len; |
3286 | } | |
3287 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3288 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3289 | |
a94213b1 N |
3290 | static ssize_t |
3291 | resync_start_show(mddev_t *mddev, char *page) | |
3292 | { | |
d1a7c503 N |
3293 | if (mddev->recovery_cp == MaxSector) |
3294 | return sprintf(page, "none\n"); | |
a94213b1 N |
3295 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3296 | } | |
3297 | ||
3298 | static ssize_t | |
3299 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
3300 | { | |
a94213b1 N |
3301 | char *e; |
3302 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3303 | ||
3304 | if (mddev->pers) | |
3305 | return -EBUSY; | |
06e3c817 DW |
3306 | if (cmd_match(buf, "none")) |
3307 | n = MaxSector; | |
3308 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3309 | return -EINVAL; |
3310 | ||
3311 | mddev->recovery_cp = n; | |
3312 | return len; | |
3313 | } | |
3314 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3315 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3316 | |
9e653b63 N |
3317 | /* |
3318 | * The array state can be: | |
3319 | * | |
3320 | * clear | |
3321 | * No devices, no size, no level | |
3322 | * Equivalent to STOP_ARRAY ioctl | |
3323 | * inactive | |
3324 | * May have some settings, but array is not active | |
3325 | * all IO results in error | |
3326 | * When written, doesn't tear down array, but just stops it | |
3327 | * suspended (not supported yet) | |
3328 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3329 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3330 | * readonly |
3331 | * no resync can happen. no superblocks get written. | |
3332 | * write requests fail | |
3333 | * read-auto | |
3334 | * like readonly, but behaves like 'clean' on a write request. | |
3335 | * | |
3336 | * clean - no pending writes, but otherwise active. | |
3337 | * When written to inactive array, starts without resync | |
3338 | * If a write request arrives then | |
3339 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3340 | * if not known, block and switch to write-pending | |
3341 | * If written to an active array that has pending writes, then fails. | |
3342 | * active | |
3343 | * fully active: IO and resync can be happening. | |
3344 | * When written to inactive array, starts with resync | |
3345 | * | |
3346 | * write-pending | |
3347 | * clean, but writes are blocked waiting for 'active' to be written. | |
3348 | * | |
3349 | * active-idle | |
3350 | * like active, but no writes have been seen for a while (100msec). | |
3351 | * | |
3352 | */ | |
3353 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3354 | write_pending, active_idle, bad_word}; | |
05381954 | 3355 | static char *array_states[] = { |
9e653b63 N |
3356 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3357 | "write-pending", "active-idle", NULL }; | |
3358 | ||
3359 | static int match_word(const char *word, char **list) | |
3360 | { | |
3361 | int n; | |
3362 | for (n=0; list[n]; n++) | |
3363 | if (cmd_match(word, list[n])) | |
3364 | break; | |
3365 | return n; | |
3366 | } | |
3367 | ||
3368 | static ssize_t | |
3369 | array_state_show(mddev_t *mddev, char *page) | |
3370 | { | |
3371 | enum array_state st = inactive; | |
3372 | ||
3373 | if (mddev->pers) | |
3374 | switch(mddev->ro) { | |
3375 | case 1: | |
3376 | st = readonly; | |
3377 | break; | |
3378 | case 2: | |
3379 | st = read_auto; | |
3380 | break; | |
3381 | case 0: | |
3382 | if (mddev->in_sync) | |
3383 | st = clean; | |
070dc6dd | 3384 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3385 | st = write_pending; |
9e653b63 N |
3386 | else if (mddev->safemode) |
3387 | st = active_idle; | |
3388 | else | |
3389 | st = active; | |
3390 | } | |
3391 | else { | |
3392 | if (list_empty(&mddev->disks) && | |
3393 | mddev->raid_disks == 0 && | |
58c0fed4 | 3394 | mddev->dev_sectors == 0) |
9e653b63 N |
3395 | st = clear; |
3396 | else | |
3397 | st = inactive; | |
3398 | } | |
3399 | return sprintf(page, "%s\n", array_states[st]); | |
3400 | } | |
3401 | ||
df5b20cf | 3402 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
a4bd82d0 | 3403 | static int md_set_readonly(mddev_t * mddev, int is_open); |
9e653b63 N |
3404 | static int do_md_run(mddev_t * mddev); |
3405 | static int restart_array(mddev_t *mddev); | |
3406 | ||
3407 | static ssize_t | |
3408 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3409 | { | |
3410 | int err = -EINVAL; | |
3411 | enum array_state st = match_word(buf, array_states); | |
3412 | switch(st) { | |
3413 | case bad_word: | |
3414 | break; | |
3415 | case clear: | |
3416 | /* stopping an active array */ | |
f2ea68cf | 3417 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3418 | return -EBUSY; |
df5b20cf | 3419 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3420 | break; |
3421 | case inactive: | |
3422 | /* stopping an active array */ | |
3423 | if (mddev->pers) { | |
f2ea68cf | 3424 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3425 | return -EBUSY; |
df5b20cf | 3426 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3427 | } else |
3428 | err = 0; /* already inactive */ | |
9e653b63 N |
3429 | break; |
3430 | case suspended: | |
3431 | break; /* not supported yet */ | |
3432 | case readonly: | |
3433 | if (mddev->pers) | |
a4bd82d0 | 3434 | err = md_set_readonly(mddev, 0); |
9e653b63 N |
3435 | else { |
3436 | mddev->ro = 1; | |
648b629e | 3437 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3438 | err = do_md_run(mddev); |
3439 | } | |
3440 | break; | |
3441 | case read_auto: | |
9e653b63 | 3442 | if (mddev->pers) { |
80268ee9 | 3443 | if (mddev->ro == 0) |
a4bd82d0 | 3444 | err = md_set_readonly(mddev, 0); |
80268ee9 | 3445 | else if (mddev->ro == 1) |
648b629e N |
3446 | err = restart_array(mddev); |
3447 | if (err == 0) { | |
3448 | mddev->ro = 2; | |
3449 | set_disk_ro(mddev->gendisk, 0); | |
3450 | } | |
9e653b63 N |
3451 | } else { |
3452 | mddev->ro = 2; | |
3453 | err = do_md_run(mddev); | |
3454 | } | |
3455 | break; | |
3456 | case clean: | |
3457 | if (mddev->pers) { | |
3458 | restart_array(mddev); | |
3459 | spin_lock_irq(&mddev->write_lock); | |
3460 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3461 | if (mddev->in_sync == 0) { |
3462 | mddev->in_sync = 1; | |
31a59e34 N |
3463 | if (mddev->safemode == 1) |
3464 | mddev->safemode = 0; | |
070dc6dd | 3465 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3466 | } |
3467 | err = 0; | |
3468 | } else | |
3469 | err = -EBUSY; | |
9e653b63 | 3470 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3471 | } else |
3472 | err = -EINVAL; | |
9e653b63 N |
3473 | break; |
3474 | case active: | |
3475 | if (mddev->pers) { | |
3476 | restart_array(mddev); | |
070dc6dd | 3477 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3478 | wake_up(&mddev->sb_wait); |
3479 | err = 0; | |
3480 | } else { | |
3481 | mddev->ro = 0; | |
648b629e | 3482 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3483 | err = do_md_run(mddev); |
3484 | } | |
3485 | break; | |
3486 | case write_pending: | |
3487 | case active_idle: | |
3488 | /* these cannot be set */ | |
3489 | break; | |
3490 | } | |
3491 | if (err) | |
3492 | return err; | |
0fd62b86 | 3493 | else { |
00bcb4ac | 3494 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3495 | return len; |
0fd62b86 | 3496 | } |
9e653b63 | 3497 | } |
80ca3a44 N |
3498 | static struct md_sysfs_entry md_array_state = |
3499 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3500 | |
1e50915f RB |
3501 | static ssize_t |
3502 | max_corrected_read_errors_show(mddev_t *mddev, char *page) { | |
3503 | return sprintf(page, "%d\n", | |
3504 | atomic_read(&mddev->max_corr_read_errors)); | |
3505 | } | |
3506 | ||
3507 | static ssize_t | |
3508 | max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) | |
3509 | { | |
3510 | char *e; | |
3511 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3512 | ||
3513 | if (*buf && (*e == 0 || *e == '\n')) { | |
3514 | atomic_set(&mddev->max_corr_read_errors, n); | |
3515 | return len; | |
3516 | } | |
3517 | return -EINVAL; | |
3518 | } | |
3519 | ||
3520 | static struct md_sysfs_entry max_corr_read_errors = | |
3521 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3522 | max_corrected_read_errors_store); | |
3523 | ||
6d7ff738 N |
3524 | static ssize_t |
3525 | null_show(mddev_t *mddev, char *page) | |
3526 | { | |
3527 | return -EINVAL; | |
3528 | } | |
3529 | ||
3530 | static ssize_t | |
3531 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3532 | { | |
3533 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3534 | /* The new device is added to the array. | |
3535 | * If the array has a persistent superblock, we read the | |
3536 | * superblock to initialise info and check validity. | |
3537 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3538 | * which mainly checks size. | |
3539 | */ | |
3540 | char *e; | |
3541 | int major = simple_strtoul(buf, &e, 10); | |
3542 | int minor; | |
3543 | dev_t dev; | |
3544 | mdk_rdev_t *rdev; | |
3545 | int err; | |
3546 | ||
3547 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3548 | return -EINVAL; | |
3549 | minor = simple_strtoul(e+1, &e, 10); | |
3550 | if (*e && *e != '\n') | |
3551 | return -EINVAL; | |
3552 | dev = MKDEV(major, minor); | |
3553 | if (major != MAJOR(dev) || | |
3554 | minor != MINOR(dev)) | |
3555 | return -EOVERFLOW; | |
3556 | ||
3557 | ||
3558 | if (mddev->persistent) { | |
3559 | rdev = md_import_device(dev, mddev->major_version, | |
3560 | mddev->minor_version); | |
3561 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3562 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3563 | mdk_rdev_t, same_set); | |
3564 | err = super_types[mddev->major_version] | |
3565 | .load_super(rdev, rdev0, mddev->minor_version); | |
3566 | if (err < 0) | |
3567 | goto out; | |
3568 | } | |
c5d79adb N |
3569 | } else if (mddev->external) |
3570 | rdev = md_import_device(dev, -2, -1); | |
3571 | else | |
6d7ff738 N |
3572 | rdev = md_import_device(dev, -1, -1); |
3573 | ||
3574 | if (IS_ERR(rdev)) | |
3575 | return PTR_ERR(rdev); | |
3576 | err = bind_rdev_to_array(rdev, mddev); | |
3577 | out: | |
3578 | if (err) | |
3579 | export_rdev(rdev); | |
3580 | return err ? err : len; | |
3581 | } | |
3582 | ||
3583 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3584 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3585 | |
9b1d1dac PC |
3586 | static ssize_t |
3587 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3588 | { | |
3589 | char *end; | |
3590 | unsigned long chunk, end_chunk; | |
3591 | ||
3592 | if (!mddev->bitmap) | |
3593 | goto out; | |
3594 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3595 | while (*buf) { | |
3596 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3597 | if (buf == end) break; | |
3598 | if (*end == '-') { /* range */ | |
3599 | buf = end + 1; | |
3600 | end_chunk = simple_strtoul(buf, &end, 0); | |
3601 | if (buf == end) break; | |
3602 | } | |
3603 | if (*end && !isspace(*end)) break; | |
3604 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3605 | buf = skip_spaces(end); |
9b1d1dac PC |
3606 | } |
3607 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3608 | out: | |
3609 | return len; | |
3610 | } | |
3611 | ||
3612 | static struct md_sysfs_entry md_bitmap = | |
3613 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3614 | ||
a35b0d69 N |
3615 | static ssize_t |
3616 | size_show(mddev_t *mddev, char *page) | |
3617 | { | |
58c0fed4 AN |
3618 | return sprintf(page, "%llu\n", |
3619 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3620 | } |
3621 | ||
d71f9f88 | 3622 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3623 | |
3624 | static ssize_t | |
3625 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3626 | { | |
3627 | /* If array is inactive, we can reduce the component size, but | |
3628 | * not increase it (except from 0). | |
3629 | * If array is active, we can try an on-line resize | |
3630 | */ | |
b522adcd DW |
3631 | sector_t sectors; |
3632 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3633 | |
58c0fed4 AN |
3634 | if (err < 0) |
3635 | return err; | |
a35b0d69 | 3636 | if (mddev->pers) { |
58c0fed4 | 3637 | err = update_size(mddev, sectors); |
850b2b42 | 3638 | md_update_sb(mddev, 1); |
a35b0d69 | 3639 | } else { |
58c0fed4 AN |
3640 | if (mddev->dev_sectors == 0 || |
3641 | mddev->dev_sectors > sectors) | |
3642 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3643 | else |
3644 | err = -ENOSPC; | |
3645 | } | |
3646 | return err ? err : len; | |
3647 | } | |
3648 | ||
3649 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3650 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3651 | |
8bb93aac N |
3652 | |
3653 | /* Metdata version. | |
e691063a N |
3654 | * This is one of |
3655 | * 'none' for arrays with no metadata (good luck...) | |
3656 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3657 | * or N.M for internally known formats |
3658 | */ | |
3659 | static ssize_t | |
3660 | metadata_show(mddev_t *mddev, char *page) | |
3661 | { | |
3662 | if (mddev->persistent) | |
3663 | return sprintf(page, "%d.%d\n", | |
3664 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3665 | else if (mddev->external) |
3666 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3667 | else |
3668 | return sprintf(page, "none\n"); | |
3669 | } | |
3670 | ||
3671 | static ssize_t | |
3672 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3673 | { | |
3674 | int major, minor; | |
3675 | char *e; | |
ea43ddd8 N |
3676 | /* Changing the details of 'external' metadata is |
3677 | * always permitted. Otherwise there must be | |
3678 | * no devices attached to the array. | |
3679 | */ | |
3680 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3681 | ; | |
3682 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3683 | return -EBUSY; |
3684 | ||
3685 | if (cmd_match(buf, "none")) { | |
3686 | mddev->persistent = 0; | |
e691063a N |
3687 | mddev->external = 0; |
3688 | mddev->major_version = 0; | |
3689 | mddev->minor_version = 90; | |
3690 | return len; | |
3691 | } | |
3692 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3693 | size_t namelen = len-9; |
e691063a N |
3694 | if (namelen >= sizeof(mddev->metadata_type)) |
3695 | namelen = sizeof(mddev->metadata_type)-1; | |
3696 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3697 | mddev->metadata_type[namelen] = 0; | |
3698 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3699 | mddev->metadata_type[--namelen] = 0; | |
3700 | mddev->persistent = 0; | |
3701 | mddev->external = 1; | |
8bb93aac N |
3702 | mddev->major_version = 0; |
3703 | mddev->minor_version = 90; | |
3704 | return len; | |
3705 | } | |
3706 | major = simple_strtoul(buf, &e, 10); | |
3707 | if (e==buf || *e != '.') | |
3708 | return -EINVAL; | |
3709 | buf = e+1; | |
3710 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3711 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3712 | return -EINVAL; |
50511da3 | 3713 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3714 | return -ENOENT; |
3715 | mddev->major_version = major; | |
3716 | mddev->minor_version = minor; | |
3717 | mddev->persistent = 1; | |
e691063a | 3718 | mddev->external = 0; |
8bb93aac N |
3719 | return len; |
3720 | } | |
3721 | ||
3722 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3723 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3724 | |
24dd469d | 3725 | static ssize_t |
7eec314d | 3726 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3727 | { |
7eec314d | 3728 | char *type = "idle"; |
b6a9ce68 N |
3729 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3730 | type = "frozen"; | |
3731 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3732 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3733 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3734 | type = "reshape"; | |
3735 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3736 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3737 | type = "resync"; | |
3738 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3739 | type = "check"; | |
3740 | else | |
3741 | type = "repair"; | |
72a23c21 | 3742 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3743 | type = "recover"; |
3744 | } | |
3745 | return sprintf(page, "%s\n", type); | |
3746 | } | |
3747 | ||
7ebc0be7 N |
3748 | static void reap_sync_thread(mddev_t *mddev); |
3749 | ||
24dd469d | 3750 | static ssize_t |
7eec314d | 3751 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3752 | { |
7eec314d N |
3753 | if (!mddev->pers || !mddev->pers->sync_request) |
3754 | return -EINVAL; | |
3755 | ||
b6a9ce68 N |
3756 | if (cmd_match(page, "frozen")) |
3757 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3758 | else | |
3759 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3760 | ||
3761 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3762 | if (mddev->sync_thread) { |
3763 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 3764 | reap_sync_thread(mddev); |
7eec314d | 3765 | } |
03c902e1 N |
3766 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3767 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3768 | return -EBUSY; |
72a23c21 NB |
3769 | else if (cmd_match(page, "resync")) |
3770 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3771 | else if (cmd_match(page, "recover")) { | |
3772 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3773 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3774 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3775 | int err; |
3776 | if (mddev->pers->start_reshape == NULL) | |
3777 | return -EINVAL; | |
3778 | err = mddev->pers->start_reshape(mddev); | |
3779 | if (err) | |
3780 | return err; | |
a99ac971 | 3781 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3782 | } else { |
bce74dac | 3783 | if (cmd_match(page, "check")) |
7eec314d | 3784 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3785 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3786 | return -EINVAL; |
3787 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3788 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3789 | } |
03c902e1 | 3790 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3791 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 3792 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
3793 | return len; |
3794 | } | |
3795 | ||
9d88883e | 3796 | static ssize_t |
96de1e66 | 3797 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3798 | { |
3799 | return sprintf(page, "%llu\n", | |
3800 | (unsigned long long) mddev->resync_mismatches); | |
3801 | } | |
3802 | ||
80ca3a44 N |
3803 | static struct md_sysfs_entry md_scan_mode = |
3804 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3805 | |
96de1e66 | 3806 | |
80ca3a44 | 3807 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3808 | |
88202a0c N |
3809 | static ssize_t |
3810 | sync_min_show(mddev_t *mddev, char *page) | |
3811 | { | |
3812 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3813 | mddev->sync_speed_min ? "local": "system"); | |
3814 | } | |
3815 | ||
3816 | static ssize_t | |
3817 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3818 | { | |
3819 | int min; | |
3820 | char *e; | |
3821 | if (strncmp(buf, "system", 6)==0) { | |
3822 | mddev->sync_speed_min = 0; | |
3823 | return len; | |
3824 | } | |
3825 | min = simple_strtoul(buf, &e, 10); | |
3826 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3827 | return -EINVAL; | |
3828 | mddev->sync_speed_min = min; | |
3829 | return len; | |
3830 | } | |
3831 | ||
3832 | static struct md_sysfs_entry md_sync_min = | |
3833 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3834 | ||
3835 | static ssize_t | |
3836 | sync_max_show(mddev_t *mddev, char *page) | |
3837 | { | |
3838 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3839 | mddev->sync_speed_max ? "local": "system"); | |
3840 | } | |
3841 | ||
3842 | static ssize_t | |
3843 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3844 | { | |
3845 | int max; | |
3846 | char *e; | |
3847 | if (strncmp(buf, "system", 6)==0) { | |
3848 | mddev->sync_speed_max = 0; | |
3849 | return len; | |
3850 | } | |
3851 | max = simple_strtoul(buf, &e, 10); | |
3852 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3853 | return -EINVAL; | |
3854 | mddev->sync_speed_max = max; | |
3855 | return len; | |
3856 | } | |
3857 | ||
3858 | static struct md_sysfs_entry md_sync_max = | |
3859 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3860 | ||
d7f3d291 IP |
3861 | static ssize_t |
3862 | degraded_show(mddev_t *mddev, char *page) | |
3863 | { | |
3864 | return sprintf(page, "%d\n", mddev->degraded); | |
3865 | } | |
3866 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3867 | |
90b08710 BS |
3868 | static ssize_t |
3869 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3870 | { | |
3871 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3872 | } | |
3873 | ||
3874 | static ssize_t | |
3875 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3876 | { | |
3877 | long n; | |
3878 | ||
3879 | if (strict_strtol(buf, 10, &n)) | |
3880 | return -EINVAL; | |
3881 | ||
3882 | if (n != 0 && n != 1) | |
3883 | return -EINVAL; | |
3884 | ||
3885 | mddev->parallel_resync = n; | |
3886 | ||
3887 | if (mddev->sync_thread) | |
3888 | wake_up(&resync_wait); | |
3889 | ||
3890 | return len; | |
3891 | } | |
3892 | ||
3893 | /* force parallel resync, even with shared block devices */ | |
3894 | static struct md_sysfs_entry md_sync_force_parallel = | |
3895 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3896 | sync_force_parallel_show, sync_force_parallel_store); | |
3897 | ||
88202a0c N |
3898 | static ssize_t |
3899 | sync_speed_show(mddev_t *mddev, char *page) | |
3900 | { | |
3901 | unsigned long resync, dt, db; | |
d1a7c503 N |
3902 | if (mddev->curr_resync == 0) |
3903 | return sprintf(page, "none\n"); | |
9687a60c AN |
3904 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3905 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3906 | if (!dt) dt++; |
9687a60c AN |
3907 | db = resync - mddev->resync_mark_cnt; |
3908 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3909 | } |
3910 | ||
80ca3a44 | 3911 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3912 | |
3913 | static ssize_t | |
3914 | sync_completed_show(mddev_t *mddev, char *page) | |
3915 | { | |
13ae864b | 3916 | unsigned long long max_sectors, resync; |
88202a0c | 3917 | |
acb180b0 N |
3918 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3919 | return sprintf(page, "none\n"); | |
3920 | ||
88202a0c | 3921 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3922 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3923 | else |
58c0fed4 | 3924 | max_sectors = mddev->dev_sectors; |
88202a0c | 3925 | |
acb180b0 | 3926 | resync = mddev->curr_resync_completed; |
13ae864b | 3927 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
3928 | } |
3929 | ||
80ca3a44 | 3930 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3931 | |
5e96ee65 NB |
3932 | static ssize_t |
3933 | min_sync_show(mddev_t *mddev, char *page) | |
3934 | { | |
3935 | return sprintf(page, "%llu\n", | |
3936 | (unsigned long long)mddev->resync_min); | |
3937 | } | |
3938 | static ssize_t | |
3939 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3940 | { | |
3941 | unsigned long long min; | |
3942 | if (strict_strtoull(buf, 10, &min)) | |
3943 | return -EINVAL; | |
3944 | if (min > mddev->resync_max) | |
3945 | return -EINVAL; | |
3946 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3947 | return -EBUSY; | |
3948 | ||
3949 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3950 | if (mddev->chunk_sectors) { |
2ac06c33 | 3951 | sector_t temp = min; |
9d8f0363 | 3952 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3953 | return -EINVAL; |
3954 | } | |
3955 | mddev->resync_min = min; | |
3956 | ||
3957 | return len; | |
3958 | } | |
3959 | ||
3960 | static struct md_sysfs_entry md_min_sync = | |
3961 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3962 | ||
c6207277 N |
3963 | static ssize_t |
3964 | max_sync_show(mddev_t *mddev, char *page) | |
3965 | { | |
3966 | if (mddev->resync_max == MaxSector) | |
3967 | return sprintf(page, "max\n"); | |
3968 | else | |
3969 | return sprintf(page, "%llu\n", | |
3970 | (unsigned long long)mddev->resync_max); | |
3971 | } | |
3972 | static ssize_t | |
3973 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3974 | { | |
3975 | if (strncmp(buf, "max", 3) == 0) | |
3976 | mddev->resync_max = MaxSector; | |
3977 | else { | |
5e96ee65 NB |
3978 | unsigned long long max; |
3979 | if (strict_strtoull(buf, 10, &max)) | |
3980 | return -EINVAL; | |
3981 | if (max < mddev->resync_min) | |
c6207277 N |
3982 | return -EINVAL; |
3983 | if (max < mddev->resync_max && | |
4d484a4a | 3984 | mddev->ro == 0 && |
c6207277 N |
3985 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3986 | return -EBUSY; | |
3987 | ||
3988 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3989 | if (mddev->chunk_sectors) { |
2ac06c33 | 3990 | sector_t temp = max; |
9d8f0363 | 3991 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
3992 | return -EINVAL; |
3993 | } | |
3994 | mddev->resync_max = max; | |
3995 | } | |
3996 | wake_up(&mddev->recovery_wait); | |
3997 | return len; | |
3998 | } | |
3999 | ||
4000 | static struct md_sysfs_entry md_max_sync = | |
4001 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4002 | ||
e464eafd N |
4003 | static ssize_t |
4004 | suspend_lo_show(mddev_t *mddev, char *page) | |
4005 | { | |
4006 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4007 | } | |
4008 | ||
4009 | static ssize_t | |
4010 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
4011 | { | |
4012 | char *e; | |
4013 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4014 | unsigned long long old = mddev->suspend_lo; |
e464eafd | 4015 | |
b8d966ef N |
4016 | if (mddev->pers == NULL || |
4017 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4018 | return -EINVAL; |
4019 | if (buf == e || (*e && *e != '\n')) | |
4020 | return -EINVAL; | |
23ddff37 N |
4021 | |
4022 | mddev->suspend_lo = new; | |
4023 | if (new >= old) | |
4024 | /* Shrinking suspended region */ | |
e464eafd | 4025 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4026 | else { |
4027 | /* Expanding suspended region - need to wait */ | |
4028 | mddev->pers->quiesce(mddev, 1); | |
4029 | mddev->pers->quiesce(mddev, 0); | |
4030 | } | |
4031 | return len; | |
e464eafd N |
4032 | } |
4033 | static struct md_sysfs_entry md_suspend_lo = | |
4034 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4035 | ||
4036 | ||
4037 | static ssize_t | |
4038 | suspend_hi_show(mddev_t *mddev, char *page) | |
4039 | { | |
4040 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4041 | } | |
4042 | ||
4043 | static ssize_t | |
4044 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
4045 | { | |
4046 | char *e; | |
4047 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4048 | unsigned long long old = mddev->suspend_hi; |
e464eafd | 4049 | |
b8d966ef N |
4050 | if (mddev->pers == NULL || |
4051 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4052 | return -EINVAL; |
4053 | if (buf == e || (*e && *e != '\n')) | |
4054 | return -EINVAL; | |
23ddff37 N |
4055 | |
4056 | mddev->suspend_hi = new; | |
4057 | if (new <= old) | |
4058 | /* Shrinking suspended region */ | |
4059 | mddev->pers->quiesce(mddev, 2); | |
4060 | else { | |
4061 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4062 | mddev->pers->quiesce(mddev, 1); |
4063 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 N |
4064 | } |
4065 | return len; | |
e464eafd N |
4066 | } |
4067 | static struct md_sysfs_entry md_suspend_hi = | |
4068 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4069 | ||
08a02ecd N |
4070 | static ssize_t |
4071 | reshape_position_show(mddev_t *mddev, char *page) | |
4072 | { | |
4073 | if (mddev->reshape_position != MaxSector) | |
4074 | return sprintf(page, "%llu\n", | |
4075 | (unsigned long long)mddev->reshape_position); | |
4076 | strcpy(page, "none\n"); | |
4077 | return 5; | |
4078 | } | |
4079 | ||
4080 | static ssize_t | |
4081 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
4082 | { | |
4083 | char *e; | |
4084 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4085 | if (mddev->pers) | |
4086 | return -EBUSY; | |
4087 | if (buf == e || (*e && *e != '\n')) | |
4088 | return -EINVAL; | |
4089 | mddev->reshape_position = new; | |
4090 | mddev->delta_disks = 0; | |
4091 | mddev->new_level = mddev->level; | |
4092 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4093 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
4094 | return len; |
4095 | } | |
4096 | ||
4097 | static struct md_sysfs_entry md_reshape_position = | |
4098 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4099 | reshape_position_store); | |
4100 | ||
b522adcd DW |
4101 | static ssize_t |
4102 | array_size_show(mddev_t *mddev, char *page) | |
4103 | { | |
4104 | if (mddev->external_size) | |
4105 | return sprintf(page, "%llu\n", | |
4106 | (unsigned long long)mddev->array_sectors/2); | |
4107 | else | |
4108 | return sprintf(page, "default\n"); | |
4109 | } | |
4110 | ||
4111 | static ssize_t | |
4112 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
4113 | { | |
4114 | sector_t sectors; | |
4115 | ||
4116 | if (strncmp(buf, "default", 7) == 0) { | |
4117 | if (mddev->pers) | |
4118 | sectors = mddev->pers->size(mddev, 0, 0); | |
4119 | else | |
4120 | sectors = mddev->array_sectors; | |
4121 | ||
4122 | mddev->external_size = 0; | |
4123 | } else { | |
4124 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4125 | return -EINVAL; | |
4126 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4127 | return -E2BIG; |
b522adcd DW |
4128 | |
4129 | mddev->external_size = 1; | |
4130 | } | |
4131 | ||
4132 | mddev->array_sectors = sectors; | |
4133 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e N |
4134 | if (mddev->pers) |
4135 | revalidate_disk(mddev->gendisk); | |
b522adcd DW |
4136 | |
4137 | return len; | |
4138 | } | |
4139 | ||
4140 | static struct md_sysfs_entry md_array_size = | |
4141 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4142 | array_size_store); | |
e464eafd | 4143 | |
eae1701f N |
4144 | static struct attribute *md_default_attrs[] = { |
4145 | &md_level.attr, | |
d4dbd025 | 4146 | &md_layout.attr, |
eae1701f | 4147 | &md_raid_disks.attr, |
3b34380a | 4148 | &md_chunk_size.attr, |
a35b0d69 | 4149 | &md_size.attr, |
a94213b1 | 4150 | &md_resync_start.attr, |
8bb93aac | 4151 | &md_metadata.attr, |
6d7ff738 | 4152 | &md_new_device.attr, |
16f17b39 | 4153 | &md_safe_delay.attr, |
9e653b63 | 4154 | &md_array_state.attr, |
08a02ecd | 4155 | &md_reshape_position.attr, |
b522adcd | 4156 | &md_array_size.attr, |
1e50915f | 4157 | &max_corr_read_errors.attr, |
411036fa N |
4158 | NULL, |
4159 | }; | |
4160 | ||
4161 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4162 | &md_scan_mode.attr, |
9d88883e | 4163 | &md_mismatches.attr, |
88202a0c N |
4164 | &md_sync_min.attr, |
4165 | &md_sync_max.attr, | |
4166 | &md_sync_speed.attr, | |
90b08710 | 4167 | &md_sync_force_parallel.attr, |
88202a0c | 4168 | &md_sync_completed.attr, |
5e96ee65 | 4169 | &md_min_sync.attr, |
c6207277 | 4170 | &md_max_sync.attr, |
e464eafd N |
4171 | &md_suspend_lo.attr, |
4172 | &md_suspend_hi.attr, | |
9b1d1dac | 4173 | &md_bitmap.attr, |
d7f3d291 | 4174 | &md_degraded.attr, |
eae1701f N |
4175 | NULL, |
4176 | }; | |
411036fa N |
4177 | static struct attribute_group md_redundancy_group = { |
4178 | .name = NULL, | |
4179 | .attrs = md_redundancy_attrs, | |
4180 | }; | |
4181 | ||
eae1701f N |
4182 | |
4183 | static ssize_t | |
4184 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4185 | { | |
4186 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4187 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4188 | ssize_t rv; |
eae1701f N |
4189 | |
4190 | if (!entry->show) | |
4191 | return -EIO; | |
5dc5cf7d IM |
4192 | rv = mddev_lock(mddev); |
4193 | if (!rv) { | |
4194 | rv = entry->show(mddev, page); | |
4195 | mddev_unlock(mddev); | |
4196 | } | |
96de1e66 | 4197 | return rv; |
eae1701f N |
4198 | } |
4199 | ||
4200 | static ssize_t | |
4201 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4202 | const char *page, size_t length) | |
4203 | { | |
4204 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4205 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4206 | ssize_t rv; |
eae1701f N |
4207 | |
4208 | if (!entry->store) | |
4209 | return -EIO; | |
67463acb N |
4210 | if (!capable(CAP_SYS_ADMIN)) |
4211 | return -EACCES; | |
5dc5cf7d | 4212 | rv = mddev_lock(mddev); |
d3374825 N |
4213 | if (mddev->hold_active == UNTIL_IOCTL) |
4214 | mddev->hold_active = 0; | |
5dc5cf7d IM |
4215 | if (!rv) { |
4216 | rv = entry->store(mddev, page, length); | |
4217 | mddev_unlock(mddev); | |
4218 | } | |
96de1e66 | 4219 | return rv; |
eae1701f N |
4220 | } |
4221 | ||
4222 | static void md_free(struct kobject *ko) | |
4223 | { | |
4224 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
4225 | |
4226 | if (mddev->sysfs_state) | |
4227 | sysfs_put(mddev->sysfs_state); | |
4228 | ||
4229 | if (mddev->gendisk) { | |
4230 | del_gendisk(mddev->gendisk); | |
4231 | put_disk(mddev->gendisk); | |
4232 | } | |
4233 | if (mddev->queue) | |
4234 | blk_cleanup_queue(mddev->queue); | |
4235 | ||
eae1701f N |
4236 | kfree(mddev); |
4237 | } | |
4238 | ||
52cf25d0 | 4239 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4240 | .show = md_attr_show, |
4241 | .store = md_attr_store, | |
4242 | }; | |
4243 | static struct kobj_type md_ktype = { | |
4244 | .release = md_free, | |
4245 | .sysfs_ops = &md_sysfs_ops, | |
4246 | .default_attrs = md_default_attrs, | |
4247 | }; | |
4248 | ||
1da177e4 LT |
4249 | int mdp_major = 0; |
4250 | ||
5fd3a17e DW |
4251 | static void mddev_delayed_delete(struct work_struct *ws) |
4252 | { | |
4253 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
4254 | ||
43a70507 | 4255 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4256 | kobject_del(&mddev->kobj); |
4257 | kobject_put(&mddev->kobj); | |
4258 | } | |
4259 | ||
efeb53c0 | 4260 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4261 | { |
48c9c27b | 4262 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
4263 | mddev_t *mddev = mddev_find(dev); |
4264 | struct gendisk *disk; | |
efeb53c0 N |
4265 | int partitioned; |
4266 | int shift; | |
4267 | int unit; | |
3830c62f | 4268 | int error; |
1da177e4 LT |
4269 | |
4270 | if (!mddev) | |
efeb53c0 N |
4271 | return -ENODEV; |
4272 | ||
4273 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4274 | shift = partitioned ? MdpMinorShift : 0; | |
4275 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4276 | |
e804ac78 TH |
4277 | /* wait for any previous instance of this device to be |
4278 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4279 | */ |
e804ac78 | 4280 | flush_workqueue(md_misc_wq); |
d3374825 | 4281 | |
48c9c27b | 4282 | mutex_lock(&disks_mutex); |
0909dc44 N |
4283 | error = -EEXIST; |
4284 | if (mddev->gendisk) | |
4285 | goto abort; | |
efeb53c0 N |
4286 | |
4287 | if (name) { | |
4288 | /* Need to ensure that 'name' is not a duplicate. | |
4289 | */ | |
4290 | mddev_t *mddev2; | |
4291 | spin_lock(&all_mddevs_lock); | |
4292 | ||
4293 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4294 | if (mddev2->gendisk && | |
4295 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4296 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4297 | goto abort; |
efeb53c0 N |
4298 | } |
4299 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4300 | } |
8b765398 | 4301 | |
0909dc44 | 4302 | error = -ENOMEM; |
8b765398 | 4303 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4304 | if (!mddev->queue) |
4305 | goto abort; | |
409c57f3 N |
4306 | mddev->queue->queuedata = mddev; |
4307 | ||
409c57f3 | 4308 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 4309 | |
1da177e4 LT |
4310 | disk = alloc_disk(1 << shift); |
4311 | if (!disk) { | |
8b765398 N |
4312 | blk_cleanup_queue(mddev->queue); |
4313 | mddev->queue = NULL; | |
0909dc44 | 4314 | goto abort; |
1da177e4 | 4315 | } |
efeb53c0 | 4316 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4317 | disk->first_minor = unit << shift; |
efeb53c0 N |
4318 | if (name) |
4319 | strcpy(disk->disk_name, name); | |
4320 | else if (partitioned) | |
1da177e4 | 4321 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4322 | else |
1da177e4 | 4323 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4324 | disk->fops = &md_fops; |
4325 | disk->private_data = mddev; | |
4326 | disk->queue = mddev->queue; | |
92850bbd | 4327 | /* Allow extended partitions. This makes the |
d3374825 | 4328 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4329 | * remove it now. |
4330 | */ | |
4331 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 LT |
4332 | add_disk(disk); |
4333 | mddev->gendisk = disk; | |
ed9e1982 TH |
4334 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4335 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4336 | if (error) { |
4337 | /* This isn't possible, but as kobject_init_and_add is marked | |
4338 | * __must_check, we must do something with the result | |
4339 | */ | |
5e55e2f5 N |
4340 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4341 | disk->disk_name); | |
0909dc44 N |
4342 | error = 0; |
4343 | } | |
00bcb4ac N |
4344 | if (mddev->kobj.sd && |
4345 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4346 | printk(KERN_DEBUG "pointless warning\n"); |
be20e6c6 DW |
4347 | |
4348 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); | |
0909dc44 N |
4349 | abort: |
4350 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4351 | if (!error && mddev->kobj.sd) { |
3830c62f | 4352 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4353 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4354 | } |
d3374825 | 4355 | mddev_put(mddev); |
0909dc44 | 4356 | return error; |
efeb53c0 N |
4357 | } |
4358 | ||
4359 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4360 | { | |
4361 | md_alloc(dev, NULL); | |
1da177e4 LT |
4362 | return NULL; |
4363 | } | |
4364 | ||
efeb53c0 N |
4365 | static int add_named_array(const char *val, struct kernel_param *kp) |
4366 | { | |
4367 | /* val must be "md_*" where * is not all digits. | |
4368 | * We allocate an array with a large free minor number, and | |
4369 | * set the name to val. val must not already be an active name. | |
4370 | */ | |
4371 | int len = strlen(val); | |
4372 | char buf[DISK_NAME_LEN]; | |
4373 | ||
4374 | while (len && val[len-1] == '\n') | |
4375 | len--; | |
4376 | if (len >= DISK_NAME_LEN) | |
4377 | return -E2BIG; | |
4378 | strlcpy(buf, val, len+1); | |
4379 | if (strncmp(buf, "md_", 3) != 0) | |
4380 | return -EINVAL; | |
4381 | return md_alloc(0, buf); | |
4382 | } | |
4383 | ||
1da177e4 LT |
4384 | static void md_safemode_timeout(unsigned long data) |
4385 | { | |
4386 | mddev_t *mddev = (mddev_t *) data; | |
4387 | ||
0fd62b86 NB |
4388 | if (!atomic_read(&mddev->writes_pending)) { |
4389 | mddev->safemode = 1; | |
4390 | if (mddev->external) | |
00bcb4ac | 4391 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4392 | } |
1da177e4 LT |
4393 | md_wakeup_thread(mddev->thread); |
4394 | } | |
4395 | ||
6ff8d8ec | 4396 | static int start_dirty_degraded; |
1da177e4 | 4397 | |
390ee602 | 4398 | int md_run(mddev_t *mddev) |
1da177e4 | 4399 | { |
2604b703 | 4400 | int err; |
1da177e4 | 4401 | mdk_rdev_t *rdev; |
2604b703 | 4402 | struct mdk_personality *pers; |
1da177e4 | 4403 | |
a757e64c N |
4404 | if (list_empty(&mddev->disks)) |
4405 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4406 | return -EINVAL; |
1da177e4 LT |
4407 | |
4408 | if (mddev->pers) | |
4409 | return -EBUSY; | |
bb4f1e9d N |
4410 | /* Cannot run until previous stop completes properly */ |
4411 | if (mddev->sysfs_active) | |
4412 | return -EBUSY; | |
b6eb127d | 4413 | |
1da177e4 LT |
4414 | /* |
4415 | * Analyze all RAID superblock(s) | |
4416 | */ | |
1ec4a939 N |
4417 | if (!mddev->raid_disks) { |
4418 | if (!mddev->persistent) | |
4419 | return -EINVAL; | |
a757e64c | 4420 | analyze_sbs(mddev); |
1ec4a939 | 4421 | } |
1da177e4 | 4422 | |
d9d166c2 N |
4423 | if (mddev->level != LEVEL_NONE) |
4424 | request_module("md-level-%d", mddev->level); | |
4425 | else if (mddev->clevel[0]) | |
4426 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4427 | |
4428 | /* | |
4429 | * Drop all container device buffers, from now on | |
4430 | * the only valid external interface is through the md | |
4431 | * device. | |
1da177e4 | 4432 | */ |
159ec1fc | 4433 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 4434 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4435 | continue; |
4436 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4437 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4438 | |
4439 | /* perform some consistency tests on the device. | |
4440 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4441 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4442 | */ |
a6ff7e08 JB |
4443 | if (rdev->meta_bdev) { |
4444 | /* Nothing to check */; | |
4445 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
4446 | if (mddev->dev_sectors && |
4447 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4448 | > rdev->sb_start) { |
f0d76d70 N |
4449 | printk("md: %s: data overlaps metadata\n", |
4450 | mdname(mddev)); | |
4451 | return -EINVAL; | |
4452 | } | |
4453 | } else { | |
0f420358 | 4454 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4455 | > rdev->data_offset) { |
4456 | printk("md: %s: metadata overlaps data\n", | |
4457 | mdname(mddev)); | |
4458 | return -EINVAL; | |
4459 | } | |
4460 | } | |
00bcb4ac | 4461 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4462 | } |
4463 | ||
a167f663 N |
4464 | if (mddev->bio_set == NULL) |
4465 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev)); | |
4466 | ||
1da177e4 | 4467 | spin_lock(&pers_lock); |
d9d166c2 | 4468 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4469 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4470 | spin_unlock(&pers_lock); |
d9d166c2 N |
4471 | if (mddev->level != LEVEL_NONE) |
4472 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4473 | mddev->level); | |
4474 | else | |
4475 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4476 | mddev->clevel); | |
1da177e4 LT |
4477 | return -EINVAL; |
4478 | } | |
2604b703 | 4479 | mddev->pers = pers; |
1da177e4 | 4480 | spin_unlock(&pers_lock); |
34817e8c N |
4481 | if (mddev->level != pers->level) { |
4482 | mddev->level = pers->level; | |
4483 | mddev->new_level = pers->level; | |
4484 | } | |
d9d166c2 | 4485 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4486 | |
f6705578 | 4487 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4488 | pers->start_reshape == NULL) { |
f6705578 N |
4489 | /* This personality cannot handle reshaping... */ |
4490 | mddev->pers = NULL; | |
4491 | module_put(pers->owner); | |
4492 | return -EINVAL; | |
4493 | } | |
4494 | ||
7dd5e7c3 N |
4495 | if (pers->sync_request) { |
4496 | /* Warn if this is a potentially silly | |
4497 | * configuration. | |
4498 | */ | |
4499 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4500 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4501 | int warned = 0; |
159ec1fc CR |
4502 | |
4503 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4504 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4505 | if (rdev < rdev2 && |
4506 | rdev->bdev->bd_contains == | |
4507 | rdev2->bdev->bd_contains) { | |
4508 | printk(KERN_WARNING | |
4509 | "%s: WARNING: %s appears to be" | |
4510 | " on the same physical disk as" | |
4511 | " %s.\n", | |
4512 | mdname(mddev), | |
4513 | bdevname(rdev->bdev,b), | |
4514 | bdevname(rdev2->bdev,b2)); | |
4515 | warned = 1; | |
4516 | } | |
4517 | } | |
159ec1fc | 4518 | |
7dd5e7c3 N |
4519 | if (warned) |
4520 | printk(KERN_WARNING | |
4521 | "True protection against single-disk" | |
4522 | " failure might be compromised.\n"); | |
4523 | } | |
4524 | ||
657390d2 | 4525 | mddev->recovery = 0; |
58c0fed4 AN |
4526 | /* may be over-ridden by personality */ |
4527 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4528 | ||
6ff8d8ec | 4529 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4530 | |
0f9552b5 | 4531 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
4532 | mddev->ro = 2; /* read-only, but switch on first write */ |
4533 | ||
b15c2e57 | 4534 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4535 | if (err) |
4536 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4537 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4538 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4539 | " but 'external_size' not in effect?\n", __func__); | |
4540 | printk(KERN_ERR | |
4541 | "md: invalid array_size %llu > default size %llu\n", | |
4542 | (unsigned long long)mddev->array_sectors / 2, | |
4543 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4544 | err = -EINVAL; | |
4545 | mddev->pers->stop(mddev); | |
4546 | } | |
4547 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4548 | err = bitmap_create(mddev); |
4549 | if (err) { | |
4550 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4551 | mdname(mddev), err); | |
4552 | mddev->pers->stop(mddev); | |
4553 | } | |
4554 | } | |
1da177e4 | 4555 | if (err) { |
1da177e4 LT |
4556 | module_put(mddev->pers->owner); |
4557 | mddev->pers = NULL; | |
32a7627c N |
4558 | bitmap_destroy(mddev); |
4559 | return err; | |
1da177e4 | 4560 | } |
5e55e2f5 | 4561 | if (mddev->pers->sync_request) { |
00bcb4ac N |
4562 | if (mddev->kobj.sd && |
4563 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
4564 | printk(KERN_WARNING |
4565 | "md: cannot register extra attributes for %s\n", | |
4566 | mdname(mddev)); | |
00bcb4ac | 4567 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4568 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4569 | mddev->ro = 0; |
4570 | ||
1da177e4 | 4571 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
4572 | atomic_set(&mddev->max_corr_read_errors, |
4573 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
4574 | mddev->safemode = 0; |
4575 | mddev->safemode_timer.function = md_safemode_timeout; | |
4576 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4577 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4578 | mddev->in_sync = 1; |
0ca69886 N |
4579 | smp_wmb(); |
4580 | mddev->ready = 1; | |
159ec1fc | 4581 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4582 | if (rdev->raid_disk >= 0) { |
4583 | char nm[20]; | |
4584 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 | 4585 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
00bcb4ac | 4586 | /* failure here is OK */; |
86e6ffdd | 4587 | } |
1da177e4 LT |
4588 | |
4589 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4590 | ||
850b2b42 N |
4591 | if (mddev->flags) |
4592 | md_update_sb(mddev, 0); | |
1da177e4 | 4593 | |
0b8c9de0 N |
4594 | md_wakeup_thread(mddev->thread); |
4595 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4596 | |
d7603b7e | 4597 | md_new_event(mddev); |
00bcb4ac N |
4598 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
4599 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 4600 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
4601 | return 0; |
4602 | } | |
390ee602 | 4603 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 4604 | |
fe60b014 N |
4605 | static int do_md_run(mddev_t *mddev) |
4606 | { | |
4607 | int err; | |
4608 | ||
4609 | err = md_run(mddev); | |
4610 | if (err) | |
4611 | goto out; | |
69e51b44 N |
4612 | err = bitmap_load(mddev); |
4613 | if (err) { | |
4614 | bitmap_destroy(mddev); | |
4615 | goto out; | |
4616 | } | |
fe60b014 N |
4617 | set_capacity(mddev->gendisk, mddev->array_sectors); |
4618 | revalidate_disk(mddev->gendisk); | |
4619 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); | |
4620 | out: | |
4621 | return err; | |
4622 | } | |
4623 | ||
1da177e4 LT |
4624 | static int restart_array(mddev_t *mddev) |
4625 | { | |
4626 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4627 | |
80fab1d7 | 4628 | /* Complain if it has no devices */ |
1da177e4 | 4629 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4630 | return -ENXIO; |
4631 | if (!mddev->pers) | |
4632 | return -EINVAL; | |
4633 | if (!mddev->ro) | |
4634 | return -EBUSY; | |
4635 | mddev->safemode = 0; | |
4636 | mddev->ro = 0; | |
4637 | set_disk_ro(disk, 0); | |
4638 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4639 | mdname(mddev)); | |
4640 | /* Kick recovery or resync if necessary */ | |
4641 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4642 | md_wakeup_thread(mddev->thread); | |
4643 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 4644 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 4645 | return 0; |
1da177e4 LT |
4646 | } |
4647 | ||
acc55e22 N |
4648 | /* similar to deny_write_access, but accounts for our holding a reference |
4649 | * to the file ourselves */ | |
4650 | static int deny_bitmap_write_access(struct file * file) | |
4651 | { | |
4652 | struct inode *inode = file->f_mapping->host; | |
4653 | ||
4654 | spin_lock(&inode->i_lock); | |
4655 | if (atomic_read(&inode->i_writecount) > 1) { | |
4656 | spin_unlock(&inode->i_lock); | |
4657 | return -ETXTBSY; | |
4658 | } | |
4659 | atomic_set(&inode->i_writecount, -1); | |
4660 | spin_unlock(&inode->i_lock); | |
4661 | ||
4662 | return 0; | |
4663 | } | |
4664 | ||
43a70507 | 4665 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
4666 | { |
4667 | struct inode *inode = file->f_mapping->host; | |
4668 | ||
4669 | spin_lock(&inode->i_lock); | |
4670 | atomic_set(&inode->i_writecount, 1); | |
4671 | spin_unlock(&inode->i_lock); | |
4672 | } | |
4673 | ||
6177b472 N |
4674 | static void md_clean(mddev_t *mddev) |
4675 | { | |
4676 | mddev->array_sectors = 0; | |
4677 | mddev->external_size = 0; | |
4678 | mddev->dev_sectors = 0; | |
4679 | mddev->raid_disks = 0; | |
4680 | mddev->recovery_cp = 0; | |
4681 | mddev->resync_min = 0; | |
4682 | mddev->resync_max = MaxSector; | |
4683 | mddev->reshape_position = MaxSector; | |
4684 | mddev->external = 0; | |
4685 | mddev->persistent = 0; | |
4686 | mddev->level = LEVEL_NONE; | |
4687 | mddev->clevel[0] = 0; | |
4688 | mddev->flags = 0; | |
4689 | mddev->ro = 0; | |
4690 | mddev->metadata_type[0] = 0; | |
4691 | mddev->chunk_sectors = 0; | |
4692 | mddev->ctime = mddev->utime = 0; | |
4693 | mddev->layout = 0; | |
4694 | mddev->max_disks = 0; | |
4695 | mddev->events = 0; | |
a8707c08 | 4696 | mddev->can_decrease_events = 0; |
6177b472 N |
4697 | mddev->delta_disks = 0; |
4698 | mddev->new_level = LEVEL_NONE; | |
4699 | mddev->new_layout = 0; | |
4700 | mddev->new_chunk_sectors = 0; | |
4701 | mddev->curr_resync = 0; | |
4702 | mddev->resync_mismatches = 0; | |
4703 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4704 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4705 | mddev->recovery = 0; | |
4706 | mddev->in_sync = 0; | |
4707 | mddev->degraded = 0; | |
6177b472 N |
4708 | mddev->safemode = 0; |
4709 | mddev->bitmap_info.offset = 0; | |
4710 | mddev->bitmap_info.default_offset = 0; | |
4711 | mddev->bitmap_info.chunksize = 0; | |
4712 | mddev->bitmap_info.daemon_sleep = 0; | |
4713 | mddev->bitmap_info.max_write_behind = 0; | |
252ac522 | 4714 | mddev->plug = NULL; |
6177b472 N |
4715 | } |
4716 | ||
defad61a | 4717 | static void __md_stop_writes(mddev_t *mddev) |
a047e125 N |
4718 | { |
4719 | if (mddev->sync_thread) { | |
4720 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4721 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 4722 | reap_sync_thread(mddev); |
a047e125 N |
4723 | } |
4724 | ||
4725 | del_timer_sync(&mddev->safemode_timer); | |
4726 | ||
4727 | bitmap_flush(mddev); | |
4728 | md_super_wait(mddev); | |
4729 | ||
4730 | if (!mddev->in_sync || mddev->flags) { | |
4731 | /* mark array as shutdown cleanly */ | |
4732 | mddev->in_sync = 1; | |
4733 | md_update_sb(mddev, 1); | |
4734 | } | |
4735 | } | |
defad61a N |
4736 | |
4737 | void md_stop_writes(mddev_t *mddev) | |
4738 | { | |
4739 | mddev_lock(mddev); | |
4740 | __md_stop_writes(mddev); | |
4741 | mddev_unlock(mddev); | |
4742 | } | |
390ee602 | 4743 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 4744 | |
390ee602 | 4745 | void md_stop(mddev_t *mddev) |
6177b472 | 4746 | { |
0ca69886 | 4747 | mddev->ready = 0; |
6177b472 N |
4748 | mddev->pers->stop(mddev); |
4749 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
4750 | mddev->to_remove = &md_redundancy_group; | |
4751 | module_put(mddev->pers->owner); | |
4752 | mddev->pers = NULL; | |
cca9cf90 | 4753 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 4754 | } |
390ee602 | 4755 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 4756 | |
a4bd82d0 N |
4757 | static int md_set_readonly(mddev_t *mddev, int is_open) |
4758 | { | |
4759 | int err = 0; | |
4760 | mutex_lock(&mddev->open_mutex); | |
4761 | if (atomic_read(&mddev->openers) > is_open) { | |
4762 | printk("md: %s still in use.\n",mdname(mddev)); | |
4763 | err = -EBUSY; | |
4764 | goto out; | |
4765 | } | |
4766 | if (mddev->pers) { | |
defad61a | 4767 | __md_stop_writes(mddev); |
a4bd82d0 N |
4768 | |
4769 | err = -ENXIO; | |
4770 | if (mddev->ro==1) | |
4771 | goto out; | |
4772 | mddev->ro = 1; | |
4773 | set_disk_ro(mddev->gendisk, 1); | |
4774 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 4775 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
a4bd82d0 N |
4776 | err = 0; |
4777 | } | |
4778 | out: | |
4779 | mutex_unlock(&mddev->open_mutex); | |
4780 | return err; | |
4781 | } | |
4782 | ||
9e653b63 N |
4783 | /* mode: |
4784 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
4785 | * 2 - stop but do not disassemble array |
4786 | */ | |
df5b20cf | 4787 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 | 4788 | { |
1da177e4 | 4789 | struct gendisk *disk = mddev->gendisk; |
c4647292 | 4790 | mdk_rdev_t *rdev; |
1da177e4 | 4791 | |
c8c00a69 | 4792 | mutex_lock(&mddev->open_mutex); |
bb4f1e9d N |
4793 | if (atomic_read(&mddev->openers) > is_open || |
4794 | mddev->sysfs_active) { | |
df5b20cf | 4795 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 N |
4796 | mutex_unlock(&mddev->open_mutex); |
4797 | return -EBUSY; | |
4798 | } | |
1da177e4 | 4799 | |
6e17b027 | 4800 | if (mddev->pers) { |
a4bd82d0 N |
4801 | if (mddev->ro) |
4802 | set_disk_ro(disk, 0); | |
409c57f3 | 4803 | |
defad61a | 4804 | __md_stop_writes(mddev); |
a4bd82d0 N |
4805 | md_stop(mddev); |
4806 | mddev->queue->merge_bvec_fn = NULL; | |
4807 | mddev->queue->unplug_fn = NULL; | |
4808 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
6177b472 | 4809 | |
a4bd82d0 | 4810 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 4811 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 4812 | |
a4bd82d0 N |
4813 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4814 | if (rdev->raid_disk >= 0) { | |
4815 | char nm[20]; | |
4816 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4817 | sysfs_remove_link(&mddev->kobj, nm); | |
4818 | } | |
c4647292 | 4819 | |
a4bd82d0 | 4820 | set_capacity(disk, 0); |
6e17b027 | 4821 | mutex_unlock(&mddev->open_mutex); |
a4bd82d0 | 4822 | revalidate_disk(disk); |
0d4ca600 | 4823 | |
a4bd82d0 N |
4824 | if (mddev->ro) |
4825 | mddev->ro = 0; | |
6e17b027 N |
4826 | } else |
4827 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
4828 | /* |
4829 | * Free resources if final stop | |
4830 | */ | |
9e653b63 | 4831 | if (mode == 0) { |
1da177e4 LT |
4832 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4833 | ||
978f946b | 4834 | bitmap_destroy(mddev); |
c3d9714e N |
4835 | if (mddev->bitmap_info.file) { |
4836 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
4837 | fput(mddev->bitmap_info.file); | |
4838 | mddev->bitmap_info.file = NULL; | |
978f946b | 4839 | } |
c3d9714e | 4840 | mddev->bitmap_info.offset = 0; |
978f946b | 4841 | |
1da177e4 LT |
4842 | export_array(mddev); |
4843 | ||
6177b472 | 4844 | md_clean(mddev); |
934d9c23 | 4845 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4846 | if (mddev->hold_active == UNTIL_STOP) |
4847 | mddev->hold_active = 0; | |
a4bd82d0 | 4848 | } |
3f9d99c1 | 4849 | blk_integrity_unregister(disk); |
d7603b7e | 4850 | md_new_event(mddev); |
00bcb4ac | 4851 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 4852 | return 0; |
1da177e4 LT |
4853 | } |
4854 | ||
fdee8ae4 | 4855 | #ifndef MODULE |
1da177e4 LT |
4856 | static void autorun_array(mddev_t *mddev) |
4857 | { | |
4858 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4859 | int err; |
4860 | ||
a757e64c | 4861 | if (list_empty(&mddev->disks)) |
1da177e4 | 4862 | return; |
1da177e4 LT |
4863 | |
4864 | printk(KERN_INFO "md: running: "); | |
4865 | ||
159ec1fc | 4866 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4867 | char b[BDEVNAME_SIZE]; |
4868 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4869 | } | |
4870 | printk("\n"); | |
4871 | ||
d710e138 | 4872 | err = do_md_run(mddev); |
1da177e4 LT |
4873 | if (err) { |
4874 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4875 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4876 | } |
4877 | } | |
4878 | ||
4879 | /* | |
4880 | * lets try to run arrays based on all disks that have arrived | |
4881 | * until now. (those are in pending_raid_disks) | |
4882 | * | |
4883 | * the method: pick the first pending disk, collect all disks with | |
4884 | * the same UUID, remove all from the pending list and put them into | |
4885 | * the 'same_array' list. Then order this list based on superblock | |
4886 | * update time (freshest comes first), kick out 'old' disks and | |
4887 | * compare superblocks. If everything's fine then run it. | |
4888 | * | |
4889 | * If "unit" is allocated, then bump its reference count | |
4890 | */ | |
4891 | static void autorun_devices(int part) | |
4892 | { | |
159ec1fc | 4893 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4894 | mddev_t *mddev; |
4895 | char b[BDEVNAME_SIZE]; | |
4896 | ||
4897 | printk(KERN_INFO "md: autorun ...\n"); | |
4898 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4899 | int unit; |
1da177e4 | 4900 | dev_t dev; |
ad01c9e3 | 4901 | LIST_HEAD(candidates); |
1da177e4 LT |
4902 | rdev0 = list_entry(pending_raid_disks.next, |
4903 | mdk_rdev_t, same_set); | |
4904 | ||
4905 | printk(KERN_INFO "md: considering %s ...\n", | |
4906 | bdevname(rdev0->bdev,b)); | |
4907 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4908 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4909 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4910 | printk(KERN_INFO "md: adding %s ...\n", | |
4911 | bdevname(rdev->bdev,b)); | |
4912 | list_move(&rdev->same_set, &candidates); | |
4913 | } | |
4914 | /* | |
4915 | * now we have a set of devices, with all of them having | |
4916 | * mostly sane superblocks. It's time to allocate the | |
4917 | * mddev. | |
4918 | */ | |
e8703fe1 N |
4919 | if (part) { |
4920 | dev = MKDEV(mdp_major, | |
4921 | rdev0->preferred_minor << MdpMinorShift); | |
4922 | unit = MINOR(dev) >> MdpMinorShift; | |
4923 | } else { | |
4924 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4925 | unit = MINOR(dev); | |
4926 | } | |
4927 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4928 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4929 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4930 | break; | |
4931 | } | |
1da177e4 LT |
4932 | |
4933 | md_probe(dev, NULL, NULL); | |
4934 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4935 | if (!mddev || !mddev->gendisk) { |
4936 | if (mddev) | |
4937 | mddev_put(mddev); | |
4938 | printk(KERN_ERR | |
1da177e4 LT |
4939 | "md: cannot allocate memory for md drive.\n"); |
4940 | break; | |
4941 | } | |
4942 | if (mddev_lock(mddev)) | |
4943 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4944 | mdname(mddev)); | |
4945 | else if (mddev->raid_disks || mddev->major_version | |
4946 | || !list_empty(&mddev->disks)) { | |
4947 | printk(KERN_WARNING | |
4948 | "md: %s already running, cannot run %s\n", | |
4949 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4950 | mddev_unlock(mddev); | |
4951 | } else { | |
4952 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4953 | mddev->persistent = 1; |
159ec1fc | 4954 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4955 | list_del_init(&rdev->same_set); |
4956 | if (bind_rdev_to_array(rdev, mddev)) | |
4957 | export_rdev(rdev); | |
4958 | } | |
4959 | autorun_array(mddev); | |
4960 | mddev_unlock(mddev); | |
4961 | } | |
4962 | /* on success, candidates will be empty, on error | |
4963 | * it won't... | |
4964 | */ | |
159ec1fc | 4965 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4966 | list_del_init(&rdev->same_set); |
1da177e4 | 4967 | export_rdev(rdev); |
4b80991c | 4968 | } |
1da177e4 LT |
4969 | mddev_put(mddev); |
4970 | } | |
4971 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4972 | } | |
fdee8ae4 | 4973 | #endif /* !MODULE */ |
1da177e4 | 4974 | |
1da177e4 LT |
4975 | static int get_version(void __user * arg) |
4976 | { | |
4977 | mdu_version_t ver; | |
4978 | ||
4979 | ver.major = MD_MAJOR_VERSION; | |
4980 | ver.minor = MD_MINOR_VERSION; | |
4981 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4982 | ||
4983 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4984 | return -EFAULT; | |
4985 | ||
4986 | return 0; | |
4987 | } | |
4988 | ||
4989 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4990 | { | |
4991 | mdu_array_info_t info; | |
a9f326eb | 4992 | int nr,working,insync,failed,spare; |
1da177e4 | 4993 | mdk_rdev_t *rdev; |
1da177e4 | 4994 | |
a9f326eb | 4995 | nr=working=insync=failed=spare=0; |
159ec1fc | 4996 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4997 | nr++; |
b2d444d7 | 4998 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4999 | failed++; |
5000 | else { | |
5001 | working++; | |
b2d444d7 | 5002 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 5003 | insync++; |
1da177e4 LT |
5004 | else |
5005 | spare++; | |
5006 | } | |
5007 | } | |
5008 | ||
5009 | info.major_version = mddev->major_version; | |
5010 | info.minor_version = mddev->minor_version; | |
5011 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5012 | info.ctime = mddev->ctime; | |
5013 | info.level = mddev->level; | |
58c0fed4 AN |
5014 | info.size = mddev->dev_sectors / 2; |
5015 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5016 | info.size = -1; |
1da177e4 LT |
5017 | info.nr_disks = nr; |
5018 | info.raid_disks = mddev->raid_disks; | |
5019 | info.md_minor = mddev->md_minor; | |
5020 | info.not_persistent= !mddev->persistent; | |
5021 | ||
5022 | info.utime = mddev->utime; | |
5023 | info.state = 0; | |
5024 | if (mddev->in_sync) | |
5025 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5026 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5027 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5028 | info.active_disks = insync; |
1da177e4 LT |
5029 | info.working_disks = working; |
5030 | info.failed_disks = failed; | |
5031 | info.spare_disks = spare; | |
5032 | ||
5033 | info.layout = mddev->layout; | |
9d8f0363 | 5034 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5035 | |
5036 | if (copy_to_user(arg, &info, sizeof(info))) | |
5037 | return -EFAULT; | |
5038 | ||
5039 | return 0; | |
5040 | } | |
5041 | ||
87162a28 | 5042 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
5043 | { |
5044 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5045 | char *ptr, *buf = NULL; | |
5046 | int err = -ENOMEM; | |
5047 | ||
b5470dc5 DW |
5048 | if (md_allow_write(mddev)) |
5049 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
5050 | else | |
5051 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 5052 | |
32a7627c N |
5053 | if (!file) |
5054 | goto out; | |
5055 | ||
5056 | /* bitmap disabled, zero the first byte and copy out */ | |
5057 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
5058 | file->pathname[0] = '\0'; | |
5059 | goto copy_out; | |
5060 | } | |
5061 | ||
5062 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5063 | if (!buf) | |
5064 | goto out; | |
5065 | ||
6bcfd601 CH |
5066 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
5067 | if (IS_ERR(ptr)) | |
32a7627c N |
5068 | goto out; |
5069 | ||
5070 | strcpy(file->pathname, ptr); | |
5071 | ||
5072 | copy_out: | |
5073 | err = 0; | |
5074 | if (copy_to_user(arg, file, sizeof(*file))) | |
5075 | err = -EFAULT; | |
5076 | out: | |
5077 | kfree(buf); | |
5078 | kfree(file); | |
5079 | return err; | |
5080 | } | |
5081 | ||
1da177e4 LT |
5082 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
5083 | { | |
5084 | mdu_disk_info_t info; | |
1da177e4 LT |
5085 | mdk_rdev_t *rdev; |
5086 | ||
5087 | if (copy_from_user(&info, arg, sizeof(info))) | |
5088 | return -EFAULT; | |
5089 | ||
26ef379f | 5090 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
5091 | if (rdev) { |
5092 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5093 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5094 | info.raid_disk = rdev->raid_disk; | |
5095 | info.state = 0; | |
b2d444d7 | 5096 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5097 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5098 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5099 | info.state |= (1<<MD_DISK_ACTIVE); |
5100 | info.state |= (1<<MD_DISK_SYNC); | |
5101 | } | |
8ddf9efe N |
5102 | if (test_bit(WriteMostly, &rdev->flags)) |
5103 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5104 | } else { |
5105 | info.major = info.minor = 0; | |
5106 | info.raid_disk = -1; | |
5107 | info.state = (1<<MD_DISK_REMOVED); | |
5108 | } | |
5109 | ||
5110 | if (copy_to_user(arg, &info, sizeof(info))) | |
5111 | return -EFAULT; | |
5112 | ||
5113 | return 0; | |
5114 | } | |
5115 | ||
5116 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
5117 | { | |
5118 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
5119 | mdk_rdev_t *rdev; | |
5120 | dev_t dev = MKDEV(info->major,info->minor); | |
5121 | ||
5122 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5123 | return -EOVERFLOW; | |
5124 | ||
5125 | if (!mddev->raid_disks) { | |
5126 | int err; | |
5127 | /* expecting a device which has a superblock */ | |
5128 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5129 | if (IS_ERR(rdev)) { | |
5130 | printk(KERN_WARNING | |
5131 | "md: md_import_device returned %ld\n", | |
5132 | PTR_ERR(rdev)); | |
5133 | return PTR_ERR(rdev); | |
5134 | } | |
5135 | if (!list_empty(&mddev->disks)) { | |
5136 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
5137 | mdk_rdev_t, same_set); | |
a9f326eb | 5138 | err = super_types[mddev->major_version] |
1da177e4 LT |
5139 | .load_super(rdev, rdev0, mddev->minor_version); |
5140 | if (err < 0) { | |
5141 | printk(KERN_WARNING | |
5142 | "md: %s has different UUID to %s\n", | |
5143 | bdevname(rdev->bdev,b), | |
5144 | bdevname(rdev0->bdev,b2)); | |
5145 | export_rdev(rdev); | |
5146 | return -EINVAL; | |
5147 | } | |
5148 | } | |
5149 | err = bind_rdev_to_array(rdev, mddev); | |
5150 | if (err) | |
5151 | export_rdev(rdev); | |
5152 | return err; | |
5153 | } | |
5154 | ||
5155 | /* | |
5156 | * add_new_disk can be used once the array is assembled | |
5157 | * to add "hot spares". They must already have a superblock | |
5158 | * written | |
5159 | */ | |
5160 | if (mddev->pers) { | |
5161 | int err; | |
5162 | if (!mddev->pers->hot_add_disk) { | |
5163 | printk(KERN_WARNING | |
5164 | "%s: personality does not support diskops!\n", | |
5165 | mdname(mddev)); | |
5166 | return -EINVAL; | |
5167 | } | |
7b1e35f6 N |
5168 | if (mddev->persistent) |
5169 | rdev = md_import_device(dev, mddev->major_version, | |
5170 | mddev->minor_version); | |
5171 | else | |
5172 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5173 | if (IS_ERR(rdev)) { |
5174 | printk(KERN_WARNING | |
5175 | "md: md_import_device returned %ld\n", | |
5176 | PTR_ERR(rdev)); | |
5177 | return PTR_ERR(rdev); | |
5178 | } | |
1a855a06 | 5179 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5180 | if (!mddev->persistent) { |
5181 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5182 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5183 | rdev->raid_disk = info->raid_disk; |
bf572541 N |
5184 | set_bit(In_sync, &rdev->flags); |
5185 | } else | |
41158c7e N |
5186 | rdev->raid_disk = -1; |
5187 | } else | |
5188 | super_types[mddev->major_version]. | |
5189 | validate_super(mddev, rdev); | |
1a855a06 N |
5190 | if (test_bit(In_sync, &rdev->flags)) |
5191 | rdev->saved_raid_disk = rdev->raid_disk; | |
5192 | else | |
5193 | rdev->saved_raid_disk = -1; | |
41158c7e | 5194 | |
b2d444d7 | 5195 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5196 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5197 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5198 | else |
5199 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5200 | |
1da177e4 LT |
5201 | rdev->raid_disk = -1; |
5202 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5203 | if (!err && !mddev->pers->hot_remove_disk) { |
5204 | /* If there is hot_add_disk but no hot_remove_disk | |
5205 | * then added disks for geometry changes, | |
5206 | * and should be added immediately. | |
5207 | */ | |
5208 | super_types[mddev->major_version]. | |
5209 | validate_super(mddev, rdev); | |
5210 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5211 | if (err) | |
5212 | unbind_rdev_from_array(rdev); | |
5213 | } | |
1da177e4 LT |
5214 | if (err) |
5215 | export_rdev(rdev); | |
52664732 | 5216 | else |
00bcb4ac | 5217 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5218 | |
17571284 | 5219 | md_update_sb(mddev, 1); |
72a23c21 NB |
5220 | if (mddev->degraded) |
5221 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5222 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 5223 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5224 | return err; |
5225 | } | |
5226 | ||
5227 | /* otherwise, add_new_disk is only allowed | |
5228 | * for major_version==0 superblocks | |
5229 | */ | |
5230 | if (mddev->major_version != 0) { | |
5231 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5232 | mdname(mddev)); | |
5233 | return -EINVAL; | |
5234 | } | |
5235 | ||
5236 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5237 | int err; | |
d710e138 | 5238 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5239 | if (IS_ERR(rdev)) { |
5240 | printk(KERN_WARNING | |
5241 | "md: error, md_import_device() returned %ld\n", | |
5242 | PTR_ERR(rdev)); | |
5243 | return PTR_ERR(rdev); | |
5244 | } | |
5245 | rdev->desc_nr = info->number; | |
5246 | if (info->raid_disk < mddev->raid_disks) | |
5247 | rdev->raid_disk = info->raid_disk; | |
5248 | else | |
5249 | rdev->raid_disk = -1; | |
5250 | ||
1da177e4 | 5251 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5252 | if (info->state & (1<<MD_DISK_SYNC)) |
5253 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5254 | |
8ddf9efe N |
5255 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5256 | set_bit(WriteMostly, &rdev->flags); | |
5257 | ||
1da177e4 LT |
5258 | if (!mddev->persistent) { |
5259 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5260 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5261 | } else | |
57b2caa3 | 5262 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5263 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5264 | |
2bf071bf N |
5265 | err = bind_rdev_to_array(rdev, mddev); |
5266 | if (err) { | |
5267 | export_rdev(rdev); | |
5268 | return err; | |
5269 | } | |
1da177e4 LT |
5270 | } |
5271 | ||
5272 | return 0; | |
5273 | } | |
5274 | ||
5275 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
5276 | { | |
5277 | char b[BDEVNAME_SIZE]; | |
5278 | mdk_rdev_t *rdev; | |
5279 | ||
1da177e4 LT |
5280 | rdev = find_rdev(mddev, dev); |
5281 | if (!rdev) | |
5282 | return -ENXIO; | |
5283 | ||
5284 | if (rdev->raid_disk >= 0) | |
5285 | goto busy; | |
5286 | ||
5287 | kick_rdev_from_array(rdev); | |
850b2b42 | 5288 | md_update_sb(mddev, 1); |
d7603b7e | 5289 | md_new_event(mddev); |
1da177e4 LT |
5290 | |
5291 | return 0; | |
5292 | busy: | |
fdefa4d8 | 5293 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5294 | bdevname(rdev->bdev,b), mdname(mddev)); |
5295 | return -EBUSY; | |
5296 | } | |
5297 | ||
5298 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
5299 | { | |
5300 | char b[BDEVNAME_SIZE]; | |
5301 | int err; | |
1da177e4 LT |
5302 | mdk_rdev_t *rdev; |
5303 | ||
5304 | if (!mddev->pers) | |
5305 | return -ENODEV; | |
5306 | ||
5307 | if (mddev->major_version != 0) { | |
5308 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5309 | " version-0 superblocks.\n", | |
5310 | mdname(mddev)); | |
5311 | return -EINVAL; | |
5312 | } | |
5313 | if (!mddev->pers->hot_add_disk) { | |
5314 | printk(KERN_WARNING | |
5315 | "%s: personality does not support diskops!\n", | |
5316 | mdname(mddev)); | |
5317 | return -EINVAL; | |
5318 | } | |
5319 | ||
d710e138 | 5320 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5321 | if (IS_ERR(rdev)) { |
5322 | printk(KERN_WARNING | |
5323 | "md: error, md_import_device() returned %ld\n", | |
5324 | PTR_ERR(rdev)); | |
5325 | return -EINVAL; | |
5326 | } | |
5327 | ||
5328 | if (mddev->persistent) | |
57b2caa3 | 5329 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5330 | else |
77304d2a | 5331 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5332 | |
8190e754 | 5333 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5334 | |
b2d444d7 | 5335 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5336 | printk(KERN_WARNING |
5337 | "md: can not hot-add faulty %s disk to %s!\n", | |
5338 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5339 | err = -EINVAL; | |
5340 | goto abort_export; | |
5341 | } | |
b2d444d7 | 5342 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5343 | rdev->desc_nr = -1; |
5842730d | 5344 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5345 | err = bind_rdev_to_array(rdev, mddev); |
5346 | if (err) | |
5347 | goto abort_export; | |
1da177e4 LT |
5348 | |
5349 | /* | |
5350 | * The rest should better be atomic, we can have disk failures | |
5351 | * noticed in interrupt contexts ... | |
5352 | */ | |
5353 | ||
1da177e4 LT |
5354 | rdev->raid_disk = -1; |
5355 | ||
850b2b42 | 5356 | md_update_sb(mddev, 1); |
1da177e4 LT |
5357 | |
5358 | /* | |
5359 | * Kick recovery, maybe this spare has to be added to the | |
5360 | * array immediately. | |
5361 | */ | |
5362 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5363 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5364 | md_new_event(mddev); |
1da177e4 LT |
5365 | return 0; |
5366 | ||
1da177e4 LT |
5367 | abort_export: |
5368 | export_rdev(rdev); | |
5369 | return err; | |
5370 | } | |
5371 | ||
32a7627c N |
5372 | static int set_bitmap_file(mddev_t *mddev, int fd) |
5373 | { | |
5374 | int err; | |
5375 | ||
36fa3063 N |
5376 | if (mddev->pers) { |
5377 | if (!mddev->pers->quiesce) | |
5378 | return -EBUSY; | |
5379 | if (mddev->recovery || mddev->sync_thread) | |
5380 | return -EBUSY; | |
5381 | /* we should be able to change the bitmap.. */ | |
5382 | } | |
32a7627c | 5383 | |
32a7627c | 5384 | |
36fa3063 N |
5385 | if (fd >= 0) { |
5386 | if (mddev->bitmap) | |
5387 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5388 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5389 | |
c3d9714e | 5390 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5391 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5392 | mdname(mddev)); | |
5393 | return -EBADF; | |
5394 | } | |
5395 | ||
c3d9714e | 5396 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
5397 | if (err) { |
5398 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5399 | mdname(mddev)); | |
c3d9714e N |
5400 | fput(mddev->bitmap_info.file); |
5401 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5402 | return err; |
5403 | } | |
c3d9714e | 5404 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5405 | } else if (mddev->bitmap == NULL) |
5406 | return -ENOENT; /* cannot remove what isn't there */ | |
5407 | err = 0; | |
5408 | if (mddev->pers) { | |
5409 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 5410 | if (fd >= 0) { |
36fa3063 | 5411 | err = bitmap_create(mddev); |
69e51b44 N |
5412 | if (!err) |
5413 | err = bitmap_load(mddev); | |
5414 | } | |
d7375ab3 | 5415 | if (fd < 0 || err) { |
36fa3063 | 5416 | bitmap_destroy(mddev); |
d7375ab3 N |
5417 | fd = -1; /* make sure to put the file */ |
5418 | } | |
36fa3063 | 5419 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5420 | } |
5421 | if (fd < 0) { | |
c3d9714e N |
5422 | if (mddev->bitmap_info.file) { |
5423 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5424 | fput(mddev->bitmap_info.file); | |
acc55e22 | 5425 | } |
c3d9714e | 5426 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
5427 | } |
5428 | ||
32a7627c N |
5429 | return err; |
5430 | } | |
5431 | ||
1da177e4 LT |
5432 | /* |
5433 | * set_array_info is used two different ways | |
5434 | * The original usage is when creating a new array. | |
5435 | * In this usage, raid_disks is > 0 and it together with | |
5436 | * level, size, not_persistent,layout,chunksize determine the | |
5437 | * shape of the array. | |
5438 | * This will always create an array with a type-0.90.0 superblock. | |
5439 | * The newer usage is when assembling an array. | |
5440 | * In this case raid_disks will be 0, and the major_version field is | |
5441 | * use to determine which style super-blocks are to be found on the devices. | |
5442 | * The minor and patch _version numbers are also kept incase the | |
5443 | * super_block handler wishes to interpret them. | |
5444 | */ | |
5445 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5446 | { | |
5447 | ||
5448 | if (info->raid_disks == 0) { | |
5449 | /* just setting version number for superblock loading */ | |
5450 | if (info->major_version < 0 || | |
50511da3 | 5451 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5452 | super_types[info->major_version].name == NULL) { |
5453 | /* maybe try to auto-load a module? */ | |
5454 | printk(KERN_INFO | |
5455 | "md: superblock version %d not known\n", | |
5456 | info->major_version); | |
5457 | return -EINVAL; | |
5458 | } | |
5459 | mddev->major_version = info->major_version; | |
5460 | mddev->minor_version = info->minor_version; | |
5461 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5462 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
5463 | /* ensure mddev_put doesn't delete this now that there |
5464 | * is some minimal configuration. | |
5465 | */ | |
5466 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
5467 | return 0; |
5468 | } | |
5469 | mddev->major_version = MD_MAJOR_VERSION; | |
5470 | mddev->minor_version = MD_MINOR_VERSION; | |
5471 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5472 | mddev->ctime = get_seconds(); | |
5473 | ||
5474 | mddev->level = info->level; | |
17115e03 | 5475 | mddev->clevel[0] = 0; |
58c0fed4 | 5476 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5477 | mddev->raid_disks = info->raid_disks; |
5478 | /* don't set md_minor, it is determined by which /dev/md* was | |
5479 | * openned | |
5480 | */ | |
5481 | if (info->state & (1<<MD_SB_CLEAN)) | |
5482 | mddev->recovery_cp = MaxSector; | |
5483 | else | |
5484 | mddev->recovery_cp = 0; | |
5485 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5486 | mddev->external = 0; |
1da177e4 LT |
5487 | |
5488 | mddev->layout = info->layout; | |
9d8f0363 | 5489 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5490 | |
5491 | mddev->max_disks = MD_SB_DISKS; | |
5492 | ||
e691063a N |
5493 | if (mddev->persistent) |
5494 | mddev->flags = 0; | |
850b2b42 | 5495 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5496 | |
c3d9714e N |
5497 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
5498 | mddev->bitmap_info.offset = 0; | |
b2a2703c | 5499 | |
f6705578 N |
5500 | mddev->reshape_position = MaxSector; |
5501 | ||
1da177e4 LT |
5502 | /* |
5503 | * Generate a 128 bit UUID | |
5504 | */ | |
5505 | get_random_bytes(mddev->uuid, 16); | |
5506 | ||
f6705578 | 5507 | mddev->new_level = mddev->level; |
664e7c41 | 5508 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5509 | mddev->new_layout = mddev->layout; |
5510 | mddev->delta_disks = 0; | |
5511 | ||
1da177e4 LT |
5512 | return 0; |
5513 | } | |
5514 | ||
1f403624 DW |
5515 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5516 | { | |
b522adcd DW |
5517 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5518 | ||
5519 | if (mddev->external_size) | |
5520 | return; | |
5521 | ||
1f403624 DW |
5522 | mddev->array_sectors = array_sectors; |
5523 | } | |
5524 | EXPORT_SYMBOL(md_set_array_sectors); | |
5525 | ||
d71f9f88 | 5526 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5527 | { |
159ec1fc | 5528 | mdk_rdev_t *rdev; |
a35b0d69 | 5529 | int rv; |
d71f9f88 | 5530 | int fit = (num_sectors == 0); |
a35b0d69 N |
5531 | |
5532 | if (mddev->pers->resize == NULL) | |
5533 | return -EINVAL; | |
d71f9f88 AN |
5534 | /* The "num_sectors" is the number of sectors of each device that |
5535 | * is used. This can only make sense for arrays with redundancy. | |
5536 | * linear and raid0 always use whatever space is available. We can only | |
5537 | * consider changing this number if no resync or reconstruction is | |
5538 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5539 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5540 | * of each device. If num_sectors is zero, we find the largest size |
5541 | * that fits. | |
a35b0d69 N |
5542 | */ |
5543 | if (mddev->sync_thread) | |
5544 | return -EBUSY; | |
dba034ee N |
5545 | if (mddev->bitmap) |
5546 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5547 | * grow, and re-add. | |
5548 | */ | |
5549 | return -EBUSY; | |
159ec1fc | 5550 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5551 | sector_t avail = rdev->sectors; |
01ab5662 | 5552 | |
d71f9f88 AN |
5553 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5554 | num_sectors = avail; | |
5555 | if (avail < num_sectors) | |
a35b0d69 N |
5556 | return -ENOSPC; |
5557 | } | |
d71f9f88 | 5558 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5559 | if (!rv) |
5560 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5561 | return rv; |
5562 | } | |
5563 | ||
da943b99 N |
5564 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5565 | { | |
5566 | int rv; | |
5567 | /* change the number of raid disks */ | |
63c70c4f | 5568 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5569 | return -EINVAL; |
5570 | if (raid_disks <= 0 || | |
233fca36 | 5571 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 5572 | return -EINVAL; |
63c70c4f | 5573 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5574 | return -EBUSY; |
63c70c4f N |
5575 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5576 | ||
5577 | rv = mddev->pers->check_reshape(mddev); | |
de171cb9 N |
5578 | if (rv < 0) |
5579 | mddev->delta_disks = 0; | |
da943b99 N |
5580 | return rv; |
5581 | } | |
5582 | ||
5583 | ||
1da177e4 LT |
5584 | /* |
5585 | * update_array_info is used to change the configuration of an | |
5586 | * on-line array. | |
5587 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5588 | * fields in the info are checked against the array. | |
5589 | * Any differences that cannot be handled will cause an error. | |
5590 | * Normally, only one change can be managed at a time. | |
5591 | */ | |
5592 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5593 | { | |
5594 | int rv = 0; | |
5595 | int cnt = 0; | |
36fa3063 N |
5596 | int state = 0; |
5597 | ||
5598 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 5599 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5600 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
5601 | |
5602 | if (mddev->major_version != info->major_version || | |
5603 | mddev->minor_version != info->minor_version || | |
5604 | /* mddev->patch_version != info->patch_version || */ | |
5605 | mddev->ctime != info->ctime || | |
5606 | mddev->level != info->level || | |
5607 | /* mddev->layout != info->layout || */ | |
5608 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5609 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5610 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5611 | ((state^info->state) & 0xfffffe00) | |
5612 | ) | |
1da177e4 LT |
5613 | return -EINVAL; |
5614 | /* Check there is only one change */ | |
58c0fed4 AN |
5615 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5616 | cnt++; | |
5617 | if (mddev->raid_disks != info->raid_disks) | |
5618 | cnt++; | |
5619 | if (mddev->layout != info->layout) | |
5620 | cnt++; | |
5621 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5622 | cnt++; | |
5623 | if (cnt == 0) | |
5624 | return 0; | |
5625 | if (cnt > 1) | |
5626 | return -EINVAL; | |
1da177e4 LT |
5627 | |
5628 | if (mddev->layout != info->layout) { | |
5629 | /* Change layout | |
5630 | * we don't need to do anything at the md level, the | |
5631 | * personality will take care of it all. | |
5632 | */ | |
50ac168a | 5633 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5634 | return -EINVAL; |
597a711b N |
5635 | else { |
5636 | mddev->new_layout = info->layout; | |
50ac168a | 5637 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5638 | if (rv) |
5639 | mddev->new_layout = mddev->layout; | |
5640 | return rv; | |
5641 | } | |
1da177e4 | 5642 | } |
58c0fed4 | 5643 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5644 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5645 | |
da943b99 N |
5646 | if (mddev->raid_disks != info->raid_disks) |
5647 | rv = update_raid_disks(mddev, info->raid_disks); | |
5648 | ||
36fa3063 N |
5649 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5650 | if (mddev->pers->quiesce == NULL) | |
5651 | return -EINVAL; | |
5652 | if (mddev->recovery || mddev->sync_thread) | |
5653 | return -EBUSY; | |
5654 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5655 | /* add the bitmap */ | |
5656 | if (mddev->bitmap) | |
5657 | return -EEXIST; | |
c3d9714e | 5658 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 5659 | return -EINVAL; |
c3d9714e N |
5660 | mddev->bitmap_info.offset = |
5661 | mddev->bitmap_info.default_offset; | |
36fa3063 N |
5662 | mddev->pers->quiesce(mddev, 1); |
5663 | rv = bitmap_create(mddev); | |
69e51b44 N |
5664 | if (!rv) |
5665 | rv = bitmap_load(mddev); | |
36fa3063 N |
5666 | if (rv) |
5667 | bitmap_destroy(mddev); | |
5668 | mddev->pers->quiesce(mddev, 0); | |
5669 | } else { | |
5670 | /* remove the bitmap */ | |
5671 | if (!mddev->bitmap) | |
5672 | return -ENOENT; | |
5673 | if (mddev->bitmap->file) | |
5674 | return -EINVAL; | |
5675 | mddev->pers->quiesce(mddev, 1); | |
5676 | bitmap_destroy(mddev); | |
5677 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 5678 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
5679 | } |
5680 | } | |
850b2b42 | 5681 | md_update_sb(mddev, 1); |
1da177e4 LT |
5682 | return rv; |
5683 | } | |
5684 | ||
5685 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5686 | { | |
5687 | mdk_rdev_t *rdev; | |
5688 | ||
5689 | if (mddev->pers == NULL) | |
5690 | return -ENODEV; | |
5691 | ||
5692 | rdev = find_rdev(mddev, dev); | |
5693 | if (!rdev) | |
5694 | return -ENODEV; | |
5695 | ||
5696 | md_error(mddev, rdev); | |
5697 | return 0; | |
5698 | } | |
5699 | ||
2f9618ce AN |
5700 | /* |
5701 | * We have a problem here : there is no easy way to give a CHS | |
5702 | * virtual geometry. We currently pretend that we have a 2 heads | |
5703 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5704 | * dosfs just mad... ;-) | |
5705 | */ | |
a885c8c4 CH |
5706 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5707 | { | |
5708 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5709 | ||
5710 | geo->heads = 2; | |
5711 | geo->sectors = 4; | |
49ce6cea | 5712 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
5713 | return 0; |
5714 | } | |
5715 | ||
a39907fa | 5716 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5717 | unsigned int cmd, unsigned long arg) |
5718 | { | |
5719 | int err = 0; | |
5720 | void __user *argp = (void __user *)arg; | |
1da177e4 | 5721 | mddev_t *mddev = NULL; |
e2218350 | 5722 | int ro; |
1da177e4 LT |
5723 | |
5724 | if (!capable(CAP_SYS_ADMIN)) | |
5725 | return -EACCES; | |
5726 | ||
5727 | /* | |
5728 | * Commands dealing with the RAID driver but not any | |
5729 | * particular array: | |
5730 | */ | |
5731 | switch (cmd) | |
5732 | { | |
5733 | case RAID_VERSION: | |
5734 | err = get_version(argp); | |
5735 | goto done; | |
5736 | ||
5737 | case PRINT_RAID_DEBUG: | |
5738 | err = 0; | |
5739 | md_print_devices(); | |
5740 | goto done; | |
5741 | ||
5742 | #ifndef MODULE | |
5743 | case RAID_AUTORUN: | |
5744 | err = 0; | |
5745 | autostart_arrays(arg); | |
5746 | goto done; | |
5747 | #endif | |
5748 | default:; | |
5749 | } | |
5750 | ||
5751 | /* | |
5752 | * Commands creating/starting a new array: | |
5753 | */ | |
5754 | ||
a39907fa | 5755 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5756 | |
5757 | if (!mddev) { | |
5758 | BUG(); | |
5759 | goto abort; | |
5760 | } | |
5761 | ||
1da177e4 LT |
5762 | err = mddev_lock(mddev); |
5763 | if (err) { | |
5764 | printk(KERN_INFO | |
5765 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5766 | err, cmd); | |
5767 | goto abort; | |
5768 | } | |
5769 | ||
5770 | switch (cmd) | |
5771 | { | |
5772 | case SET_ARRAY_INFO: | |
5773 | { | |
5774 | mdu_array_info_t info; | |
5775 | if (!arg) | |
5776 | memset(&info, 0, sizeof(info)); | |
5777 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5778 | err = -EFAULT; | |
5779 | goto abort_unlock; | |
5780 | } | |
5781 | if (mddev->pers) { | |
5782 | err = update_array_info(mddev, &info); | |
5783 | if (err) { | |
5784 | printk(KERN_WARNING "md: couldn't update" | |
5785 | " array info. %d\n", err); | |
5786 | goto abort_unlock; | |
5787 | } | |
5788 | goto done_unlock; | |
5789 | } | |
5790 | if (!list_empty(&mddev->disks)) { | |
5791 | printk(KERN_WARNING | |
5792 | "md: array %s already has disks!\n", | |
5793 | mdname(mddev)); | |
5794 | err = -EBUSY; | |
5795 | goto abort_unlock; | |
5796 | } | |
5797 | if (mddev->raid_disks) { | |
5798 | printk(KERN_WARNING | |
5799 | "md: array %s already initialised!\n", | |
5800 | mdname(mddev)); | |
5801 | err = -EBUSY; | |
5802 | goto abort_unlock; | |
5803 | } | |
5804 | err = set_array_info(mddev, &info); | |
5805 | if (err) { | |
5806 | printk(KERN_WARNING "md: couldn't set" | |
5807 | " array info. %d\n", err); | |
5808 | goto abort_unlock; | |
5809 | } | |
5810 | } | |
5811 | goto done_unlock; | |
5812 | ||
5813 | default:; | |
5814 | } | |
5815 | ||
5816 | /* | |
5817 | * Commands querying/configuring an existing array: | |
5818 | */ | |
32a7627c | 5819 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5820 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5821 | if ((!mddev->raid_disks && !mddev->external) |
5822 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5823 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5824 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5825 | err = -ENODEV; |
5826 | goto abort_unlock; | |
5827 | } | |
5828 | ||
5829 | /* | |
5830 | * Commands even a read-only array can execute: | |
5831 | */ | |
5832 | switch (cmd) | |
5833 | { | |
5834 | case GET_ARRAY_INFO: | |
5835 | err = get_array_info(mddev, argp); | |
5836 | goto done_unlock; | |
5837 | ||
32a7627c | 5838 | case GET_BITMAP_FILE: |
87162a28 | 5839 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5840 | goto done_unlock; |
5841 | ||
1da177e4 LT |
5842 | case GET_DISK_INFO: |
5843 | err = get_disk_info(mddev, argp); | |
5844 | goto done_unlock; | |
5845 | ||
5846 | case RESTART_ARRAY_RW: | |
5847 | err = restart_array(mddev); | |
5848 | goto done_unlock; | |
5849 | ||
5850 | case STOP_ARRAY: | |
d710e138 | 5851 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5852 | goto done_unlock; |
5853 | ||
5854 | case STOP_ARRAY_RO: | |
a4bd82d0 | 5855 | err = md_set_readonly(mddev, 1); |
1da177e4 LT |
5856 | goto done_unlock; |
5857 | ||
e2218350 DW |
5858 | case BLKROSET: |
5859 | if (get_user(ro, (int __user *)(arg))) { | |
5860 | err = -EFAULT; | |
5861 | goto done_unlock; | |
5862 | } | |
5863 | err = -EINVAL; | |
5864 | ||
5865 | /* if the bdev is going readonly the value of mddev->ro | |
5866 | * does not matter, no writes are coming | |
5867 | */ | |
5868 | if (ro) | |
5869 | goto done_unlock; | |
5870 | ||
5871 | /* are we are already prepared for writes? */ | |
5872 | if (mddev->ro != 1) | |
5873 | goto done_unlock; | |
5874 | ||
5875 | /* transitioning to readauto need only happen for | |
5876 | * arrays that call md_write_start | |
5877 | */ | |
5878 | if (mddev->pers) { | |
5879 | err = restart_array(mddev); | |
5880 | if (err == 0) { | |
5881 | mddev->ro = 2; | |
5882 | set_disk_ro(mddev->gendisk, 0); | |
5883 | } | |
5884 | } | |
5885 | goto done_unlock; | |
1da177e4 LT |
5886 | } |
5887 | ||
5888 | /* | |
5889 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5890 | * superblock, so we do not allow them on read-only arrays. |
5891 | * However non-MD ioctls (e.g. get-size) will still come through | |
5892 | * here and hit the 'default' below, so only disallow | |
5893 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5894 | */ |
bb57fc64 | 5895 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5896 | if (mddev->ro == 2) { |
5897 | mddev->ro = 0; | |
00bcb4ac | 5898 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 NB |
5899 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5900 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5901 | } else { |
5902 | err = -EROFS; | |
5903 | goto abort_unlock; | |
5904 | } | |
1da177e4 LT |
5905 | } |
5906 | ||
5907 | switch (cmd) | |
5908 | { | |
5909 | case ADD_NEW_DISK: | |
5910 | { | |
5911 | mdu_disk_info_t info; | |
5912 | if (copy_from_user(&info, argp, sizeof(info))) | |
5913 | err = -EFAULT; | |
5914 | else | |
5915 | err = add_new_disk(mddev, &info); | |
5916 | goto done_unlock; | |
5917 | } | |
5918 | ||
5919 | case HOT_REMOVE_DISK: | |
5920 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5921 | goto done_unlock; | |
5922 | ||
5923 | case HOT_ADD_DISK: | |
5924 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5925 | goto done_unlock; | |
5926 | ||
5927 | case SET_DISK_FAULTY: | |
5928 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5929 | goto done_unlock; | |
5930 | ||
5931 | case RUN_ARRAY: | |
d710e138 | 5932 | err = do_md_run(mddev); |
1da177e4 LT |
5933 | goto done_unlock; |
5934 | ||
32a7627c N |
5935 | case SET_BITMAP_FILE: |
5936 | err = set_bitmap_file(mddev, (int)arg); | |
5937 | goto done_unlock; | |
5938 | ||
1da177e4 | 5939 | default: |
1da177e4 LT |
5940 | err = -EINVAL; |
5941 | goto abort_unlock; | |
5942 | } | |
5943 | ||
5944 | done_unlock: | |
5945 | abort_unlock: | |
d3374825 N |
5946 | if (mddev->hold_active == UNTIL_IOCTL && |
5947 | err != -EINVAL) | |
5948 | mddev->hold_active = 0; | |
1da177e4 LT |
5949 | mddev_unlock(mddev); |
5950 | ||
5951 | return err; | |
5952 | done: | |
5953 | if (err) | |
5954 | MD_BUG(); | |
5955 | abort: | |
5956 | return err; | |
5957 | } | |
aa98aa31 AB |
5958 | #ifdef CONFIG_COMPAT |
5959 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
5960 | unsigned int cmd, unsigned long arg) | |
5961 | { | |
5962 | switch (cmd) { | |
5963 | case HOT_REMOVE_DISK: | |
5964 | case HOT_ADD_DISK: | |
5965 | case SET_DISK_FAULTY: | |
5966 | case SET_BITMAP_FILE: | |
5967 | /* These take in integer arg, do not convert */ | |
5968 | break; | |
5969 | default: | |
5970 | arg = (unsigned long)compat_ptr(arg); | |
5971 | break; | |
5972 | } | |
5973 | ||
5974 | return md_ioctl(bdev, mode, cmd, arg); | |
5975 | } | |
5976 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 5977 | |
a39907fa | 5978 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
5979 | { |
5980 | /* | |
5981 | * Succeed if we can lock the mddev, which confirms that | |
5982 | * it isn't being stopped right now. | |
5983 | */ | |
d3374825 | 5984 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
5985 | int err; |
5986 | ||
d3374825 N |
5987 | if (mddev->gendisk != bdev->bd_disk) { |
5988 | /* we are racing with mddev_put which is discarding this | |
5989 | * bd_disk. | |
5990 | */ | |
5991 | mddev_put(mddev); | |
5992 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 5993 | flush_workqueue(md_misc_wq); |
d3374825 N |
5994 | /* Then retry the open from the top */ |
5995 | return -ERESTARTSYS; | |
5996 | } | |
5997 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
5998 | ||
c8c00a69 | 5999 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6000 | goto out; |
6001 | ||
6002 | err = 0; | |
f2ea68cf | 6003 | atomic_inc(&mddev->openers); |
c8c00a69 | 6004 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6005 | |
f3b99be1 | 6006 | check_disk_size_change(mddev->gendisk, bdev); |
1da177e4 LT |
6007 | out: |
6008 | return err; | |
6009 | } | |
6010 | ||
a39907fa | 6011 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6012 | { |
a39907fa | 6013 | mddev_t *mddev = disk->private_data; |
1da177e4 | 6014 | |
52e5f9d1 | 6015 | BUG_ON(!mddev); |
f2ea68cf | 6016 | atomic_dec(&mddev->openers); |
1da177e4 LT |
6017 | mddev_put(mddev); |
6018 | ||
6019 | return 0; | |
6020 | } | |
83d5cde4 | 6021 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6022 | { |
6023 | .owner = THIS_MODULE, | |
a39907fa AV |
6024 | .open = md_open, |
6025 | .release = md_release, | |
b492b852 | 6026 | .ioctl = md_ioctl, |
aa98aa31 AB |
6027 | #ifdef CONFIG_COMPAT |
6028 | .compat_ioctl = md_compat_ioctl, | |
6029 | #endif | |
a885c8c4 | 6030 | .getgeo = md_getgeo, |
1da177e4 LT |
6031 | }; |
6032 | ||
75c96f85 | 6033 | static int md_thread(void * arg) |
1da177e4 LT |
6034 | { |
6035 | mdk_thread_t *thread = arg; | |
6036 | ||
1da177e4 LT |
6037 | /* |
6038 | * md_thread is a 'system-thread', it's priority should be very | |
6039 | * high. We avoid resource deadlocks individually in each | |
6040 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6041 | * the very same RT priority as kswapd, thus we will never get | |
6042 | * into a priority inversion deadlock. | |
6043 | * | |
6044 | * we definitely have to have equal or higher priority than | |
6045 | * bdflush, otherwise bdflush will deadlock if there are too | |
6046 | * many dirty RAID5 blocks. | |
6047 | */ | |
1da177e4 | 6048 | |
6985c43f | 6049 | allow_signal(SIGKILL); |
a6fb0934 | 6050 | while (!kthread_should_stop()) { |
1da177e4 | 6051 | |
93588e22 N |
6052 | /* We need to wait INTERRUPTIBLE so that |
6053 | * we don't add to the load-average. | |
6054 | * That means we need to be sure no signals are | |
6055 | * pending | |
6056 | */ | |
6057 | if (signal_pending(current)) | |
6058 | flush_signals(current); | |
6059 | ||
6060 | wait_event_interruptible_timeout | |
6061 | (thread->wqueue, | |
6062 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6063 | || kthread_should_stop(), | |
6064 | thread->timeout); | |
1da177e4 | 6065 | |
6c987910 N |
6066 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6067 | if (!kthread_should_stop()) | |
589a594b | 6068 | thread->run(thread->mddev); |
1da177e4 | 6069 | } |
a6fb0934 | 6070 | |
1da177e4 LT |
6071 | return 0; |
6072 | } | |
6073 | ||
6074 | void md_wakeup_thread(mdk_thread_t *thread) | |
6075 | { | |
6076 | if (thread) { | |
6077 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
6078 | set_bit(THREAD_WAKEUP, &thread->flags); | |
6079 | wake_up(&thread->wqueue); | |
6080 | } | |
6081 | } | |
6082 | ||
6083 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
6084 | const char *name) | |
6085 | { | |
6086 | mdk_thread_t *thread; | |
1da177e4 | 6087 | |
9ffae0cf | 6088 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
6089 | if (!thread) |
6090 | return NULL; | |
6091 | ||
1da177e4 LT |
6092 | init_waitqueue_head(&thread->wqueue); |
6093 | ||
1da177e4 LT |
6094 | thread->run = run; |
6095 | thread->mddev = mddev; | |
32a7627c | 6096 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6097 | thread->tsk = kthread_run(md_thread, thread, |
6098 | "%s_%s", | |
6099 | mdname(thread->mddev), | |
6100 | name ?: mddev->pers->name); | |
a6fb0934 | 6101 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6102 | kfree(thread); |
6103 | return NULL; | |
6104 | } | |
1da177e4 LT |
6105 | return thread; |
6106 | } | |
6107 | ||
1da177e4 LT |
6108 | void md_unregister_thread(mdk_thread_t *thread) |
6109 | { | |
e0cf8f04 N |
6110 | if (!thread) |
6111 | return; | |
ba25f9dc | 6112 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
6113 | |
6114 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6115 | kfree(thread); |
6116 | } | |
6117 | ||
6118 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
6119 | { | |
6120 | if (!mddev) { | |
6121 | MD_BUG(); | |
6122 | return; | |
6123 | } | |
6124 | ||
b2d444d7 | 6125 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6126 | return; |
6bfe0b49 DW |
6127 | |
6128 | if (mddev->external) | |
6129 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 6130 | /* |
1da177e4 LT |
6131 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
6132 | mdname(mddev), | |
6133 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
6134 | __builtin_return_address(0),__builtin_return_address(1), | |
6135 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 6136 | */ |
d0a0a5ee AM |
6137 | if (!mddev->pers) |
6138 | return; | |
1da177e4 LT |
6139 | if (!mddev->pers->error_handler) |
6140 | return; | |
6141 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6142 | if (mddev->degraded) |
6143 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6144 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6145 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6146 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6147 | md_wakeup_thread(mddev->thread); | |
768a418d | 6148 | if (mddev->event_work.func) |
e804ac78 | 6149 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6150 | md_new_event_inintr(mddev); |
1da177e4 LT |
6151 | } |
6152 | ||
6153 | /* seq_file implementation /proc/mdstat */ | |
6154 | ||
6155 | static void status_unused(struct seq_file *seq) | |
6156 | { | |
6157 | int i = 0; | |
6158 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6159 | |
6160 | seq_printf(seq, "unused devices: "); | |
6161 | ||
159ec1fc | 6162 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6163 | char b[BDEVNAME_SIZE]; |
6164 | i++; | |
6165 | seq_printf(seq, "%s ", | |
6166 | bdevname(rdev->bdev,b)); | |
6167 | } | |
6168 | if (!i) | |
6169 | seq_printf(seq, "<none>"); | |
6170 | ||
6171 | seq_printf(seq, "\n"); | |
6172 | } | |
6173 | ||
6174 | ||
6175 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
6176 | { | |
dd71cf6b N |
6177 | sector_t max_sectors, resync, res; |
6178 | unsigned long dt, db; | |
6179 | sector_t rt; | |
4588b42e N |
6180 | int scale; |
6181 | unsigned int per_milli; | |
1da177e4 | 6182 | |
dd71cf6b | 6183 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6184 | |
6185 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 6186 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6187 | else |
dd71cf6b | 6188 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6189 | |
6190 | /* | |
6191 | * Should not happen. | |
6192 | */ | |
dd71cf6b | 6193 | if (!max_sectors) { |
1da177e4 LT |
6194 | MD_BUG(); |
6195 | return; | |
6196 | } | |
4588b42e | 6197 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6198 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6199 | * u32, as those are the requirements for sector_div. |
6200 | * Thus 'scale' must be at least 10 | |
6201 | */ | |
6202 | scale = 10; | |
6203 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6204 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6205 | scale++; |
6206 | } | |
6207 | res = (resync>>scale)*1000; | |
dd71cf6b | 6208 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6209 | |
6210 | per_milli = res; | |
1da177e4 | 6211 | { |
4588b42e | 6212 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6213 | seq_printf(seq, "["); |
6214 | for (i = 0; i < x; i++) | |
6215 | seq_printf(seq, "="); | |
6216 | seq_printf(seq, ">"); | |
6217 | for (i = 0; i < y; i++) | |
6218 | seq_printf(seq, "."); | |
6219 | seq_printf(seq, "] "); | |
6220 | } | |
4588b42e | 6221 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6222 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6223 | "reshape" : | |
61df9d91 N |
6224 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6225 | "check" : | |
6226 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6227 | "resync" : "recovery"))), | |
6228 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6229 | (unsigned long long) resync/2, |
6230 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6231 | |
6232 | /* | |
1da177e4 LT |
6233 | * dt: time from mark until now |
6234 | * db: blocks written from mark until now | |
6235 | * rt: remaining time | |
dd71cf6b N |
6236 | * |
6237 | * rt is a sector_t, so could be 32bit or 64bit. | |
6238 | * So we divide before multiply in case it is 32bit and close | |
6239 | * to the limit. | |
6240 | * We scale the divisor (db) by 32 to avoid loosing precision | |
6241 | * near the end of resync when the number of remaining sectors | |
6242 | * is close to 'db'. | |
6243 | * We then divide rt by 32 after multiplying by db to compensate. | |
6244 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6245 | */ |
6246 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6247 | if (!dt) dt++; | |
ff4e8d9a N |
6248 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6249 | - mddev->resync_mark_cnt; | |
1da177e4 | 6250 | |
dd71cf6b N |
6251 | rt = max_sectors - resync; /* number of remaining sectors */ |
6252 | sector_div(rt, db/32+1); | |
6253 | rt *= dt; | |
6254 | rt >>= 5; | |
6255 | ||
6256 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6257 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6258 | |
ff4e8d9a | 6259 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6260 | } |
6261 | ||
6262 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6263 | { | |
6264 | struct list_head *tmp; | |
6265 | loff_t l = *pos; | |
6266 | mddev_t *mddev; | |
6267 | ||
6268 | if (l >= 0x10000) | |
6269 | return NULL; | |
6270 | if (!l--) | |
6271 | /* header */ | |
6272 | return (void*)1; | |
6273 | ||
6274 | spin_lock(&all_mddevs_lock); | |
6275 | list_for_each(tmp,&all_mddevs) | |
6276 | if (!l--) { | |
6277 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
6278 | mddev_get(mddev); | |
6279 | spin_unlock(&all_mddevs_lock); | |
6280 | return mddev; | |
6281 | } | |
6282 | spin_unlock(&all_mddevs_lock); | |
6283 | if (!l--) | |
6284 | return (void*)2;/* tail */ | |
6285 | return NULL; | |
6286 | } | |
6287 | ||
6288 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6289 | { | |
6290 | struct list_head *tmp; | |
6291 | mddev_t *next_mddev, *mddev = v; | |
6292 | ||
6293 | ++*pos; | |
6294 | if (v == (void*)2) | |
6295 | return NULL; | |
6296 | ||
6297 | spin_lock(&all_mddevs_lock); | |
6298 | if (v == (void*)1) | |
6299 | tmp = all_mddevs.next; | |
6300 | else | |
6301 | tmp = mddev->all_mddevs.next; | |
6302 | if (tmp != &all_mddevs) | |
6303 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
6304 | else { | |
6305 | next_mddev = (void*)2; | |
6306 | *pos = 0x10000; | |
6307 | } | |
6308 | spin_unlock(&all_mddevs_lock); | |
6309 | ||
6310 | if (v != (void*)1) | |
6311 | mddev_put(mddev); | |
6312 | return next_mddev; | |
6313 | ||
6314 | } | |
6315 | ||
6316 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6317 | { | |
6318 | mddev_t *mddev = v; | |
6319 | ||
6320 | if (mddev && v != (void*)1 && v != (void*)2) | |
6321 | mddev_put(mddev); | |
6322 | } | |
6323 | ||
d7603b7e N |
6324 | struct mdstat_info { |
6325 | int event; | |
6326 | }; | |
6327 | ||
1da177e4 LT |
6328 | static int md_seq_show(struct seq_file *seq, void *v) |
6329 | { | |
6330 | mddev_t *mddev = v; | |
dd8ac336 | 6331 | sector_t sectors; |
1da177e4 | 6332 | mdk_rdev_t *rdev; |
d7603b7e | 6333 | struct mdstat_info *mi = seq->private; |
32a7627c | 6334 | struct bitmap *bitmap; |
1da177e4 LT |
6335 | |
6336 | if (v == (void*)1) { | |
2604b703 | 6337 | struct mdk_personality *pers; |
1da177e4 LT |
6338 | seq_printf(seq, "Personalities : "); |
6339 | spin_lock(&pers_lock); | |
2604b703 N |
6340 | list_for_each_entry(pers, &pers_list, list) |
6341 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6342 | |
6343 | spin_unlock(&pers_lock); | |
6344 | seq_printf(seq, "\n"); | |
d7603b7e | 6345 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
6346 | return 0; |
6347 | } | |
6348 | if (v == (void*)2) { | |
6349 | status_unused(seq); | |
6350 | return 0; | |
6351 | } | |
6352 | ||
5dc5cf7d | 6353 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6354 | return -EINTR; |
5dc5cf7d | 6355 | |
1da177e4 LT |
6356 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
6357 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
6358 | mddev->pers ? "" : "in"); | |
6359 | if (mddev->pers) { | |
f91de92e | 6360 | if (mddev->ro==1) |
1da177e4 | 6361 | seq_printf(seq, " (read-only)"); |
f91de92e | 6362 | if (mddev->ro==2) |
52720ae7 | 6363 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
6364 | seq_printf(seq, " %s", mddev->pers->name); |
6365 | } | |
6366 | ||
dd8ac336 | 6367 | sectors = 0; |
159ec1fc | 6368 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
6369 | char b[BDEVNAME_SIZE]; |
6370 | seq_printf(seq, " %s[%d]", | |
6371 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
6372 | if (test_bit(WriteMostly, &rdev->flags)) |
6373 | seq_printf(seq, "(W)"); | |
b2d444d7 | 6374 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6375 | seq_printf(seq, "(F)"); |
6376 | continue; | |
b325a32e N |
6377 | } else if (rdev->raid_disk < 0) |
6378 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 6379 | sectors += rdev->sectors; |
1da177e4 LT |
6380 | } |
6381 | ||
6382 | if (!list_empty(&mddev->disks)) { | |
6383 | if (mddev->pers) | |
6384 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
6385 | (unsigned long long) |
6386 | mddev->array_sectors / 2); | |
1da177e4 LT |
6387 | else |
6388 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 6389 | (unsigned long long)sectors / 2); |
1da177e4 | 6390 | } |
1cd6bf19 N |
6391 | if (mddev->persistent) { |
6392 | if (mddev->major_version != 0 || | |
6393 | mddev->minor_version != 90) { | |
6394 | seq_printf(seq," super %d.%d", | |
6395 | mddev->major_version, | |
6396 | mddev->minor_version); | |
6397 | } | |
e691063a N |
6398 | } else if (mddev->external) |
6399 | seq_printf(seq, " super external:%s", | |
6400 | mddev->metadata_type); | |
6401 | else | |
1cd6bf19 | 6402 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
6403 | |
6404 | if (mddev->pers) { | |
d710e138 | 6405 | mddev->pers->status(seq, mddev); |
1da177e4 | 6406 | seq_printf(seq, "\n "); |
8e1b39d6 N |
6407 | if (mddev->pers->sync_request) { |
6408 | if (mddev->curr_resync > 2) { | |
d710e138 | 6409 | status_resync(seq, mddev); |
8e1b39d6 N |
6410 | seq_printf(seq, "\n "); |
6411 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
6412 | seq_printf(seq, "\tresync=DELAYED\n "); | |
6413 | else if (mddev->recovery_cp < MaxSector) | |
6414 | seq_printf(seq, "\tresync=PENDING\n "); | |
6415 | } | |
32a7627c N |
6416 | } else |
6417 | seq_printf(seq, "\n "); | |
6418 | ||
6419 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
6420 | unsigned long chunk_kb; |
6421 | unsigned long flags; | |
32a7627c | 6422 | spin_lock_irqsave(&bitmap->lock, flags); |
42a04b50 | 6423 | chunk_kb = mddev->bitmap_info.chunksize >> 10; |
32a7627c N |
6424 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " |
6425 | "%lu%s chunk", | |
6426 | bitmap->pages - bitmap->missing_pages, | |
6427 | bitmap->pages, | |
6428 | (bitmap->pages - bitmap->missing_pages) | |
6429 | << (PAGE_SHIFT - 10), | |
42a04b50 | 6430 | chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, |
32a7627c | 6431 | chunk_kb ? "KB" : "B"); |
78d742d8 N |
6432 | if (bitmap->file) { |
6433 | seq_printf(seq, ", file: "); | |
c32c2f63 | 6434 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 6435 | } |
78d742d8 | 6436 | |
32a7627c N |
6437 | seq_printf(seq, "\n"); |
6438 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
6439 | } |
6440 | ||
6441 | seq_printf(seq, "\n"); | |
6442 | } | |
6443 | mddev_unlock(mddev); | |
6444 | ||
6445 | return 0; | |
6446 | } | |
6447 | ||
110518bc | 6448 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6449 | .start = md_seq_start, |
6450 | .next = md_seq_next, | |
6451 | .stop = md_seq_stop, | |
6452 | .show = md_seq_show, | |
6453 | }; | |
6454 | ||
6455 | static int md_seq_open(struct inode *inode, struct file *file) | |
6456 | { | |
6457 | int error; | |
d7603b7e N |
6458 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6459 | if (mi == NULL) | |
6460 | return -ENOMEM; | |
1da177e4 LT |
6461 | |
6462 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6463 | if (error) |
6464 | kfree(mi); | |
6465 | else { | |
6466 | struct seq_file *p = file->private_data; | |
6467 | p->private = mi; | |
6468 | mi->event = atomic_read(&md_event_count); | |
6469 | } | |
1da177e4 LT |
6470 | return error; |
6471 | } | |
6472 | ||
d7603b7e N |
6473 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6474 | { | |
6475 | struct seq_file *m = filp->private_data; | |
6476 | struct mdstat_info *mi = m->private; | |
6477 | int mask; | |
6478 | ||
6479 | poll_wait(filp, &md_event_waiters, wait); | |
6480 | ||
6481 | /* always allow read */ | |
6482 | mask = POLLIN | POLLRDNORM; | |
6483 | ||
6484 | if (mi->event != atomic_read(&md_event_count)) | |
6485 | mask |= POLLERR | POLLPRI; | |
6486 | return mask; | |
6487 | } | |
6488 | ||
fa027c2a | 6489 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6490 | .owner = THIS_MODULE, |
1da177e4 LT |
6491 | .open = md_seq_open, |
6492 | .read = seq_read, | |
6493 | .llseek = seq_lseek, | |
c3f94b40 | 6494 | .release = seq_release_private, |
d7603b7e | 6495 | .poll = mdstat_poll, |
1da177e4 LT |
6496 | }; |
6497 | ||
2604b703 | 6498 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6499 | { |
1da177e4 | 6500 | spin_lock(&pers_lock); |
2604b703 N |
6501 | list_add_tail(&p->list, &pers_list); |
6502 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6503 | spin_unlock(&pers_lock); |
6504 | return 0; | |
6505 | } | |
6506 | ||
2604b703 | 6507 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6508 | { |
2604b703 | 6509 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6510 | spin_lock(&pers_lock); |
2604b703 | 6511 | list_del_init(&p->list); |
1da177e4 LT |
6512 | spin_unlock(&pers_lock); |
6513 | return 0; | |
6514 | } | |
6515 | ||
eea1bf38 | 6516 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6517 | { |
6518 | mdk_rdev_t * rdev; | |
1da177e4 | 6519 | int idle; |
eea1bf38 | 6520 | int curr_events; |
1da177e4 LT |
6521 | |
6522 | idle = 1; | |
4b80991c N |
6523 | rcu_read_lock(); |
6524 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6525 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6526 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6527 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6528 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6529 | /* sync IO will cause sync_io to increase before the disk_stats |
6530 | * as sync_io is counted when a request starts, and | |
6531 | * disk_stats is counted when it completes. | |
6532 | * So resync activity will cause curr_events to be smaller than | |
6533 | * when there was no such activity. | |
6534 | * non-sync IO will cause disk_stat to increase without | |
6535 | * increasing sync_io so curr_events will (eventually) | |
6536 | * be larger than it was before. Once it becomes | |
6537 | * substantially larger, the test below will cause | |
6538 | * the array to appear non-idle, and resync will slow | |
6539 | * down. | |
6540 | * If there is a lot of outstanding resync activity when | |
6541 | * we set last_event to curr_events, then all that activity | |
6542 | * completing might cause the array to appear non-idle | |
6543 | * and resync will be slowed down even though there might | |
6544 | * not have been non-resync activity. This will only | |
6545 | * happen once though. 'last_events' will soon reflect | |
6546 | * the state where there is little or no outstanding | |
6547 | * resync requests, and further resync activity will | |
6548 | * always make curr_events less than last_events. | |
c0e48521 | 6549 | * |
1da177e4 | 6550 | */ |
eea1bf38 | 6551 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6552 | rdev->last_events = curr_events; |
6553 | idle = 0; | |
6554 | } | |
6555 | } | |
4b80991c | 6556 | rcu_read_unlock(); |
1da177e4 LT |
6557 | return idle; |
6558 | } | |
6559 | ||
6560 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6561 | { | |
6562 | /* another "blocks" (512byte) blocks have been synced */ | |
6563 | atomic_sub(blocks, &mddev->recovery_active); | |
6564 | wake_up(&mddev->recovery_wait); | |
6565 | if (!ok) { | |
dfc70645 | 6566 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6567 | md_wakeup_thread(mddev->thread); |
6568 | // stop recovery, signal do_sync .... | |
6569 | } | |
6570 | } | |
6571 | ||
6572 | ||
06d91a5f N |
6573 | /* md_write_start(mddev, bi) |
6574 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6575 | * in superblock) before writing, schedule a superblock update |
6576 | * and wait for it to complete. | |
06d91a5f | 6577 | */ |
3d310eb7 | 6578 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6579 | { |
0fd62b86 | 6580 | int did_change = 0; |
06d91a5f | 6581 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6582 | return; |
06d91a5f | 6583 | |
f91de92e N |
6584 | BUG_ON(mddev->ro == 1); |
6585 | if (mddev->ro == 2) { | |
6586 | /* need to switch to read/write */ | |
6587 | mddev->ro = 0; | |
6588 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6589 | md_wakeup_thread(mddev->thread); | |
25156198 | 6590 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6591 | did_change = 1; |
f91de92e | 6592 | } |
06d91a5f | 6593 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6594 | if (mddev->safemode == 1) |
6595 | mddev->safemode = 0; | |
06d91a5f | 6596 | if (mddev->in_sync) { |
a9701a30 | 6597 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6598 | if (mddev->in_sync) { |
6599 | mddev->in_sync = 0; | |
850b2b42 | 6600 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 6601 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 6602 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6603 | did_change = 1; |
3d310eb7 | 6604 | } |
a9701a30 | 6605 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6606 | } |
0fd62b86 | 6607 | if (did_change) |
00bcb4ac | 6608 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 6609 | wait_event(mddev->sb_wait, |
09a44cc1 | 6610 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 LT |
6611 | } |
6612 | ||
6613 | void md_write_end(mddev_t *mddev) | |
6614 | { | |
6615 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6616 | if (mddev->safemode == 2) | |
6617 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6618 | else if (mddev->safemode_delay) |
1da177e4 LT |
6619 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6620 | } | |
6621 | } | |
6622 | ||
2a2275d6 N |
6623 | /* md_allow_write(mddev) |
6624 | * Calling this ensures that the array is marked 'active' so that writes | |
6625 | * may proceed without blocking. It is important to call this before | |
6626 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6627 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6628 | * |
6629 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6630 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6631 | */ |
b5470dc5 | 6632 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6633 | { |
6634 | if (!mddev->pers) | |
b5470dc5 | 6635 | return 0; |
2a2275d6 | 6636 | if (mddev->ro) |
b5470dc5 | 6637 | return 0; |
1a0fd497 | 6638 | if (!mddev->pers->sync_request) |
b5470dc5 | 6639 | return 0; |
2a2275d6 N |
6640 | |
6641 | spin_lock_irq(&mddev->write_lock); | |
6642 | if (mddev->in_sync) { | |
6643 | mddev->in_sync = 0; | |
6644 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 6645 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
6646 | if (mddev->safemode_delay && |
6647 | mddev->safemode == 0) | |
6648 | mddev->safemode = 1; | |
6649 | spin_unlock_irq(&mddev->write_lock); | |
6650 | md_update_sb(mddev, 0); | |
00bcb4ac | 6651 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
6652 | } else |
6653 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 | 6654 | |
070dc6dd | 6655 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
6656 | return -EAGAIN; |
6657 | else | |
6658 | return 0; | |
2a2275d6 N |
6659 | } |
6660 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6661 | ||
b63d7c2e | 6662 | void md_unplug(mddev_t *mddev) |
252ac522 N |
6663 | { |
6664 | if (mddev->queue) | |
6665 | blk_unplug(mddev->queue); | |
6666 | if (mddev->plug) | |
6667 | mddev->plug->unplug_fn(mddev->plug); | |
6668 | } | |
6669 | ||
1da177e4 LT |
6670 | #define SYNC_MARKS 10 |
6671 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6672 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6673 | { |
6674 | mddev_t *mddev2; | |
6675 | unsigned int currspeed = 0, | |
6676 | window; | |
57afd89f | 6677 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6678 | unsigned long mark[SYNC_MARKS]; |
6679 | sector_t mark_cnt[SYNC_MARKS]; | |
6680 | int last_mark,m; | |
6681 | struct list_head *tmp; | |
6682 | sector_t last_check; | |
57afd89f | 6683 | int skipped = 0; |
5fd6c1dc | 6684 | mdk_rdev_t *rdev; |
61df9d91 | 6685 | char *desc; |
1da177e4 LT |
6686 | |
6687 | /* just incase thread restarts... */ | |
6688 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6689 | return; | |
5fd6c1dc N |
6690 | if (mddev->ro) /* never try to sync a read-only array */ |
6691 | return; | |
1da177e4 | 6692 | |
61df9d91 N |
6693 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6694 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6695 | desc = "data-check"; | |
6696 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6697 | desc = "requested-resync"; | |
6698 | else | |
6699 | desc = "resync"; | |
6700 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6701 | desc = "reshape"; | |
6702 | else | |
6703 | desc = "recovery"; | |
6704 | ||
1da177e4 LT |
6705 | /* we overload curr_resync somewhat here. |
6706 | * 0 == not engaged in resync at all | |
6707 | * 2 == checking that there is no conflict with another sync | |
6708 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6709 | * commense | |
6710 | * other == active in resync - this many blocks | |
6711 | * | |
6712 | * Before starting a resync we must have set curr_resync to | |
6713 | * 2, and then checked that every "conflicting" array has curr_resync | |
6714 | * less than ours. When we find one that is the same or higher | |
6715 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6716 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6717 | * This will mean we have to start checking from the beginning again. | |
6718 | * | |
6719 | */ | |
6720 | ||
6721 | do { | |
6722 | mddev->curr_resync = 2; | |
6723 | ||
6724 | try_again: | |
404e4b43 | 6725 | if (kthread_should_stop()) |
6985c43f | 6726 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
404e4b43 N |
6727 | |
6728 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
1da177e4 | 6729 | goto skip; |
29ac4aa3 | 6730 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6731 | if (mddev2 == mddev) |
6732 | continue; | |
90b08710 BS |
6733 | if (!mddev->parallel_resync |
6734 | && mddev2->curr_resync | |
6735 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6736 | DEFINE_WAIT(wq); |
6737 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6738 | /* arbitrarily yield */ | |
6739 | mddev->curr_resync = 1; | |
6740 | wake_up(&resync_wait); | |
6741 | } | |
6742 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6743 | /* no need to wait here, we can wait the next | |
6744 | * time 'round when curr_resync == 2 | |
6745 | */ | |
6746 | continue; | |
9744197c N |
6747 | /* We need to wait 'interruptible' so as not to |
6748 | * contribute to the load average, and not to | |
6749 | * be caught by 'softlockup' | |
6750 | */ | |
6751 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6752 | if (!kthread_should_stop() && |
8712e553 | 6753 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6754 | printk(KERN_INFO "md: delaying %s of %s" |
6755 | " until %s has finished (they" | |
1da177e4 | 6756 | " share one or more physical units)\n", |
61df9d91 | 6757 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6758 | mddev_put(mddev2); |
9744197c N |
6759 | if (signal_pending(current)) |
6760 | flush_signals(current); | |
1da177e4 LT |
6761 | schedule(); |
6762 | finish_wait(&resync_wait, &wq); | |
6763 | goto try_again; | |
6764 | } | |
6765 | finish_wait(&resync_wait, &wq); | |
6766 | } | |
6767 | } | |
6768 | } while (mddev->curr_resync < 2); | |
6769 | ||
5fd6c1dc | 6770 | j = 0; |
9d88883e | 6771 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6772 | /* resync follows the size requested by the personality, |
57afd89f | 6773 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6774 | */ |
6775 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6776 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6777 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6778 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6779 | j = mddev->resync_min; | |
6780 | else if (!mddev->bitmap) | |
5fd6c1dc | 6781 | j = mddev->recovery_cp; |
5e96ee65 | 6782 | |
ccfcc3c1 | 6783 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6784 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6785 | else { |
1da177e4 | 6786 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6787 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6788 | j = MaxSector; |
4e59ca7d DW |
6789 | rcu_read_lock(); |
6790 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6791 | if (rdev->raid_disk >= 0 && |
6792 | !test_bit(Faulty, &rdev->flags) && | |
6793 | !test_bit(In_sync, &rdev->flags) && | |
6794 | rdev->recovery_offset < j) | |
6795 | j = rdev->recovery_offset; | |
4e59ca7d | 6796 | rcu_read_unlock(); |
5fd6c1dc | 6797 | } |
1da177e4 | 6798 | |
61df9d91 N |
6799 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6800 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6801 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6802 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6803 | "(but not more than %d KB/sec) for %s.\n", |
6804 | speed_max(mddev), desc); | |
1da177e4 | 6805 | |
eea1bf38 | 6806 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6807 | |
57afd89f | 6808 | io_sectors = 0; |
1da177e4 LT |
6809 | for (m = 0; m < SYNC_MARKS; m++) { |
6810 | mark[m] = jiffies; | |
57afd89f | 6811 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6812 | } |
6813 | last_mark = 0; | |
6814 | mddev->resync_mark = mark[last_mark]; | |
6815 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6816 | ||
6817 | /* | |
6818 | * Tune reconstruction: | |
6819 | */ | |
6820 | window = 32*(PAGE_SIZE/512); | |
6821 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6822 | window/2,(unsigned long long) max_sectors/2); | |
6823 | ||
6824 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6825 | last_check = 0; |
6826 | ||
6827 | if (j>2) { | |
6828 | printk(KERN_INFO | |
61df9d91 N |
6829 | "md: resuming %s of %s from checkpoint.\n", |
6830 | desc, mdname(mddev)); | |
1da177e4 LT |
6831 | mddev->curr_resync = j; |
6832 | } | |
75d3da43 | 6833 | mddev->curr_resync_completed = j; |
1da177e4 LT |
6834 | |
6835 | while (j < max_sectors) { | |
57afd89f | 6836 | sector_t sectors; |
1da177e4 | 6837 | |
57afd89f | 6838 | skipped = 0; |
97e4f42d | 6839 | |
7a91ee1f N |
6840 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6841 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6842 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6843 | > (max_sectors >> 4)) || | |
6844 | (j - mddev->curr_resync_completed)*2 | |
6845 | >= mddev->resync_max - mddev->curr_resync_completed | |
6846 | )) { | |
97e4f42d | 6847 | /* time to update curr_resync_completed */ |
252ac522 | 6848 | md_unplug(mddev); |
97e4f42d N |
6849 | wait_event(mddev->recovery_wait, |
6850 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 6851 | mddev->curr_resync_completed = j; |
070dc6dd | 6852 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 6853 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6854 | } |
acb180b0 | 6855 | |
e62e58a5 N |
6856 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6857 | /* As this condition is controlled by user-space, | |
6858 | * we can block indefinitely, so use '_interruptible' | |
6859 | * to avoid triggering warnings. | |
6860 | */ | |
6861 | flush_signals(current); /* just in case */ | |
6862 | wait_event_interruptible(mddev->recovery_wait, | |
6863 | mddev->resync_max > j | |
6864 | || kthread_should_stop()); | |
6865 | } | |
acb180b0 N |
6866 | |
6867 | if (kthread_should_stop()) | |
6868 | goto interrupted; | |
6869 | ||
57afd89f | 6870 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6871 | currspeed < speed_min(mddev)); |
57afd89f | 6872 | if (sectors == 0) { |
dfc70645 | 6873 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6874 | goto out; |
6875 | } | |
57afd89f N |
6876 | |
6877 | if (!skipped) { /* actual IO requested */ | |
6878 | io_sectors += sectors; | |
6879 | atomic_add(sectors, &mddev->recovery_active); | |
6880 | } | |
6881 | ||
1da177e4 LT |
6882 | j += sectors; |
6883 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6884 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6885 | if (last_check == 0) |
6886 | /* this is the earliers that rebuilt will be | |
6887 | * visible in /proc/mdstat | |
6888 | */ | |
6889 | md_new_event(mddev); | |
57afd89f N |
6890 | |
6891 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6892 | continue; |
6893 | ||
57afd89f | 6894 | last_check = io_sectors; |
1da177e4 | 6895 | |
dfc70645 | 6896 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6897 | break; |
6898 | ||
6899 | repeat: | |
6900 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6901 | /* step marks */ | |
6902 | int next = (last_mark+1) % SYNC_MARKS; | |
6903 | ||
6904 | mddev->resync_mark = mark[next]; | |
6905 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6906 | mark[next] = jiffies; | |
57afd89f | 6907 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6908 | last_mark = next; |
6909 | } | |
6910 | ||
6911 | ||
c6207277 N |
6912 | if (kthread_should_stop()) |
6913 | goto interrupted; | |
6914 | ||
1da177e4 LT |
6915 | |
6916 | /* | |
6917 | * this loop exits only if either when we are slower than | |
6918 | * the 'hard' speed limit, or the system was IO-idle for | |
6919 | * a jiffy. | |
6920 | * the system might be non-idle CPU-wise, but we only care | |
6921 | * about not overloading the IO subsystem. (things like an | |
6922 | * e2fsck being done on the RAID array should execute fast) | |
6923 | */ | |
252ac522 | 6924 | md_unplug(mddev); |
1da177e4 LT |
6925 | cond_resched(); |
6926 | ||
57afd89f N |
6927 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6928 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6929 | |
88202a0c N |
6930 | if (currspeed > speed_min(mddev)) { |
6931 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6932 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6933 | msleep(500); |
1da177e4 LT |
6934 | goto repeat; |
6935 | } | |
6936 | } | |
6937 | } | |
61df9d91 | 6938 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6939 | /* |
6940 | * this also signals 'finished resyncing' to md_stop | |
6941 | */ | |
6942 | out: | |
252ac522 | 6943 | md_unplug(mddev); |
1da177e4 LT |
6944 | |
6945 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
6946 | ||
6947 | /* tell personality that we are finished */ | |
57afd89f | 6948 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6949 | |
dfc70645 | 6950 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6951 | mddev->curr_resync > 2) { |
6952 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6953 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6954 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6955 | printk(KERN_INFO | |
61df9d91 N |
6956 | "md: checkpointing %s of %s.\n", |
6957 | desc, mdname(mddev)); | |
5fd6c1dc N |
6958 | mddev->recovery_cp = mddev->curr_resync; |
6959 | } | |
6960 | } else | |
6961 | mddev->recovery_cp = MaxSector; | |
6962 | } else { | |
6963 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6964 | mddev->curr_resync = MaxSector; | |
4e59ca7d DW |
6965 | rcu_read_lock(); |
6966 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc | 6967 | if (rdev->raid_disk >= 0 && |
70fffd0b | 6968 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
6969 | !test_bit(Faulty, &rdev->flags) && |
6970 | !test_bit(In_sync, &rdev->flags) && | |
6971 | rdev->recovery_offset < mddev->curr_resync) | |
6972 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 6973 | rcu_read_unlock(); |
5fd6c1dc | 6974 | } |
1da177e4 | 6975 | } |
17571284 | 6976 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6977 | |
1da177e4 | 6978 | skip: |
c07b70ad N |
6979 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
6980 | /* We completed so min/max setting can be forgotten if used. */ | |
6981 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6982 | mddev->resync_min = 0; | |
6983 | mddev->resync_max = MaxSector; | |
6984 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6985 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 | 6986 | mddev->curr_resync = 0; |
efa59339 N |
6987 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
6988 | mddev->curr_resync_completed = 0; | |
c6207277 | 6989 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
1da177e4 LT |
6990 | wake_up(&resync_wait); |
6991 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
6992 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
6993 | return; |
6994 | ||
6995 | interrupted: | |
6996 | /* | |
6997 | * got a signal, exit. | |
6998 | */ | |
6999 | printk(KERN_INFO | |
7000 | "md: md_do_sync() got signal ... exiting\n"); | |
7001 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7002 | goto out; | |
7003 | ||
1da177e4 | 7004 | } |
29269553 | 7005 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
7006 | |
7007 | ||
b4c4c7b8 N |
7008 | static int remove_and_add_spares(mddev_t *mddev) |
7009 | { | |
7010 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
7011 | int spares = 0; |
7012 | ||
97e4f42d N |
7013 | mddev->curr_resync_completed = 0; |
7014 | ||
159ec1fc | 7015 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 7016 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 7017 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7018 | (test_bit(Faulty, &rdev->flags) || |
7019 | ! test_bit(In_sync, &rdev->flags)) && | |
7020 | atomic_read(&rdev->nr_pending)==0) { | |
7021 | if (mddev->pers->hot_remove_disk( | |
7022 | mddev, rdev->raid_disk)==0) { | |
7023 | char nm[20]; | |
7024 | sprintf(nm,"rd%d", rdev->raid_disk); | |
7025 | sysfs_remove_link(&mddev->kobj, nm); | |
7026 | rdev->raid_disk = -1; | |
7027 | } | |
7028 | } | |
7029 | ||
4044ba58 | 7030 | if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { |
159ec1fc | 7031 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 7032 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
7033 | !test_bit(In_sync, &rdev->flags) && |
7034 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 7035 | spares++; |
b4c4c7b8 N |
7036 | if (rdev->raid_disk < 0 |
7037 | && !test_bit(Faulty, &rdev->flags)) { | |
7038 | rdev->recovery_offset = 0; | |
199050ea NB |
7039 | if (mddev->pers-> |
7040 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
7041 | char nm[20]; |
7042 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
7043 | if (sysfs_create_link(&mddev->kobj, |
7044 | &rdev->kobj, nm)) | |
00bcb4ac | 7045 | /* failure here is OK */; |
b4c4c7b8 N |
7046 | spares++; |
7047 | md_new_event(mddev); | |
93be75ff | 7048 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
b4c4c7b8 N |
7049 | } else |
7050 | break; | |
7051 | } | |
dfc70645 | 7052 | } |
b4c4c7b8 N |
7053 | } |
7054 | return spares; | |
7055 | } | |
7ebc0be7 N |
7056 | |
7057 | static void reap_sync_thread(mddev_t *mddev) | |
7058 | { | |
7059 | mdk_rdev_t *rdev; | |
7060 | ||
7061 | /* resync has finished, collect result */ | |
7062 | md_unregister_thread(mddev->sync_thread); | |
7063 | mddev->sync_thread = NULL; | |
7064 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
7065 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
7066 | /* success...*/ | |
7067 | /* activate any spares */ | |
7068 | if (mddev->pers->spare_active(mddev)) | |
7069 | sysfs_notify(&mddev->kobj, NULL, | |
7070 | "degraded"); | |
7071 | } | |
7072 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
7073 | mddev->pers->finish_reshape) | |
7074 | mddev->pers->finish_reshape(mddev); | |
7075 | md_update_sb(mddev, 1); | |
7076 | ||
7077 | /* if array is no-longer degraded, then any saved_raid_disk | |
7078 | * information must be scrapped | |
7079 | */ | |
7080 | if (!mddev->degraded) | |
7081 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
7082 | rdev->saved_raid_disk = -1; | |
7083 | ||
7084 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7085 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7086 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7087 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7088 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7089 | /* flag recovery needed just to double check */ | |
7090 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7091 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7092 | md_new_event(mddev); | |
7093 | } | |
7094 | ||
1da177e4 LT |
7095 | /* |
7096 | * This routine is regularly called by all per-raid-array threads to | |
7097 | * deal with generic issues like resync and super-block update. | |
7098 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7099 | * need this as they never do any recovery or update the superblock. | |
7100 | * | |
7101 | * It does not do any resync itself, but rather "forks" off other threads | |
7102 | * to do that as needed. | |
7103 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7104 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7105 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7106 | * and wakeups up this thread which will reap the thread and finish up. |
7107 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7108 | * | |
7109 | * The overall approach is: | |
7110 | * 1/ if the superblock needs updating, update it. | |
7111 | * 2/ If a recovery thread is running, don't do anything else. | |
7112 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7113 | * 4/ If there are any faulty devices, remove them. | |
7114 | * 5/ If array is degraded, try to add spares devices | |
7115 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7116 | */ | |
7117 | void md_check_recovery(mddev_t *mddev) | |
7118 | { | |
5f40402d | 7119 | if (mddev->bitmap) |
aa5cbd10 | 7120 | bitmap_daemon_work(mddev); |
1da177e4 LT |
7121 | |
7122 | if (mddev->ro) | |
7123 | return; | |
fca4d848 N |
7124 | |
7125 | if (signal_pending(current)) { | |
31a59e34 | 7126 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7127 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7128 | mdname(mddev)); | |
7129 | mddev->safemode = 2; | |
7130 | } | |
7131 | flush_signals(current); | |
7132 | } | |
7133 | ||
c89a8eee N |
7134 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7135 | return; | |
1da177e4 | 7136 | if ( ! ( |
126925c0 | 7137 | (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7138 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7139 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7140 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7141 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7142 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7143 | )) |
7144 | return; | |
fca4d848 | 7145 | |
df5b89b3 | 7146 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7147 | int spares = 0; |
fca4d848 | 7148 | |
c89a8eee N |
7149 | if (mddev->ro) { |
7150 | /* Only thing we do on a ro array is remove | |
7151 | * failed devices. | |
7152 | */ | |
7153 | remove_and_add_spares(mddev); | |
7154 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7155 | goto unlock; | |
7156 | } | |
7157 | ||
31a59e34 | 7158 | if (!mddev->external) { |
0fd62b86 | 7159 | int did_change = 0; |
31a59e34 N |
7160 | spin_lock_irq(&mddev->write_lock); |
7161 | if (mddev->safemode && | |
7162 | !atomic_read(&mddev->writes_pending) && | |
7163 | !mddev->in_sync && | |
7164 | mddev->recovery_cp == MaxSector) { | |
7165 | mddev->in_sync = 1; | |
0fd62b86 | 7166 | did_change = 1; |
070dc6dd | 7167 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7168 | } |
7169 | if (mddev->safemode == 1) | |
7170 | mddev->safemode = 0; | |
7171 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7172 | if (did_change) |
00bcb4ac | 7173 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7174 | } |
fca4d848 | 7175 | |
850b2b42 N |
7176 | if (mddev->flags) |
7177 | md_update_sb(mddev, 0); | |
06d91a5f | 7178 | |
1da177e4 LT |
7179 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7180 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7181 | /* resync/recovery still happening */ | |
7182 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7183 | goto unlock; | |
7184 | } | |
7185 | if (mddev->sync_thread) { | |
7ebc0be7 | 7186 | reap_sync_thread(mddev); |
1da177e4 LT |
7187 | goto unlock; |
7188 | } | |
72a23c21 NB |
7189 | /* Set RUNNING before clearing NEEDED to avoid |
7190 | * any transients in the value of "sync_action". | |
7191 | */ | |
7192 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7193 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
7194 | /* Clear some bits that don't mean anything, but |
7195 | * might be left set | |
7196 | */ | |
24dd469d N |
7197 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7198 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7199 | |
5fd6c1dc N |
7200 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
7201 | goto unlock; | |
1da177e4 LT |
7202 | /* no recovery is running. |
7203 | * remove any failed drives, then | |
7204 | * add spares if possible. | |
7205 | * Spare are also removed and re-added, to allow | |
7206 | * the personality to fail the re-add. | |
7207 | */ | |
1da177e4 | 7208 | |
b4c4c7b8 | 7209 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7210 | if (mddev->pers->check_reshape == NULL || |
7211 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7212 | /* Cannot proceed */ |
7213 | goto unlock; | |
7214 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7215 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 7216 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
7217 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7218 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7219 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7220 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7221 | } else if (mddev->recovery_cp < MaxSector) { |
7222 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7223 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7224 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7225 | /* nothing to be done ... */ | |
1da177e4 | 7226 | goto unlock; |
24dd469d | 7227 | |
1da177e4 | 7228 | if (mddev->pers->sync_request) { |
a654b9d8 N |
7229 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
7230 | /* We are adding a device or devices to an array | |
7231 | * which has the bitmap stored on all devices. | |
7232 | * So make sure all bitmap pages get written | |
7233 | */ | |
7234 | bitmap_write_all(mddev->bitmap); | |
7235 | } | |
1da177e4 LT |
7236 | mddev->sync_thread = md_register_thread(md_do_sync, |
7237 | mddev, | |
0da3c619 | 7238 | "resync"); |
1da177e4 LT |
7239 | if (!mddev->sync_thread) { |
7240 | printk(KERN_ERR "%s: could not start resync" | |
7241 | " thread...\n", | |
7242 | mdname(mddev)); | |
7243 | /* leave the spares where they are, it shouldn't hurt */ | |
7ebc0be7 N |
7244 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
7245 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7246 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7247 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7248 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
d7603b7e | 7249 | } else |
1da177e4 | 7250 | md_wakeup_thread(mddev->sync_thread); |
00bcb4ac | 7251 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7252 | md_new_event(mddev); |
1da177e4 LT |
7253 | } |
7254 | unlock: | |
72a23c21 NB |
7255 | if (!mddev->sync_thread) { |
7256 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7257 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7258 | &mddev->recovery)) | |
0c3573f1 | 7259 | if (mddev->sysfs_action) |
00bcb4ac | 7260 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7261 | } |
1da177e4 LT |
7262 | mddev_unlock(mddev); |
7263 | } | |
7264 | } | |
7265 | ||
6bfe0b49 DW |
7266 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
7267 | { | |
00bcb4ac | 7268 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 DW |
7269 | wait_event_timeout(rdev->blocked_wait, |
7270 | !test_bit(Blocked, &rdev->flags), | |
7271 | msecs_to_jiffies(5000)); | |
7272 | rdev_dec_pending(rdev, mddev); | |
7273 | } | |
7274 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7275 | ||
75c96f85 AB |
7276 | static int md_notify_reboot(struct notifier_block *this, |
7277 | unsigned long code, void *x) | |
1da177e4 LT |
7278 | { |
7279 | struct list_head *tmp; | |
7280 | mddev_t *mddev; | |
7281 | ||
7282 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
7283 | ||
7284 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
7285 | ||
29ac4aa3 | 7286 | for_each_mddev(mddev, tmp) |
c71d4887 | 7287 | if (mddev_trylock(mddev)) { |
2b25000b N |
7288 | /* Force a switch to readonly even array |
7289 | * appears to still be in use. Hence | |
7290 | * the '100'. | |
7291 | */ | |
a4bd82d0 | 7292 | md_set_readonly(mddev, 100); |
c71d4887 NB |
7293 | mddev_unlock(mddev); |
7294 | } | |
1da177e4 LT |
7295 | /* |
7296 | * certain more exotic SCSI devices are known to be | |
7297 | * volatile wrt too early system reboots. While the | |
7298 | * right place to handle this issue is the given | |
7299 | * driver, we do want to have a safe RAID driver ... | |
7300 | */ | |
7301 | mdelay(1000*1); | |
7302 | } | |
7303 | return NOTIFY_DONE; | |
7304 | } | |
7305 | ||
75c96f85 | 7306 | static struct notifier_block md_notifier = { |
1da177e4 LT |
7307 | .notifier_call = md_notify_reboot, |
7308 | .next = NULL, | |
7309 | .priority = INT_MAX, /* before any real devices */ | |
7310 | }; | |
7311 | ||
7312 | static void md_geninit(void) | |
7313 | { | |
1da177e4 LT |
7314 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
7315 | ||
c7705f34 | 7316 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
7317 | } |
7318 | ||
75c96f85 | 7319 | static int __init md_init(void) |
1da177e4 | 7320 | { |
e804ac78 TH |
7321 | int ret = -ENOMEM; |
7322 | ||
7323 | md_wq = alloc_workqueue("md", WQ_RESCUER, 0); | |
7324 | if (!md_wq) | |
7325 | goto err_wq; | |
7326 | ||
7327 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
7328 | if (!md_misc_wq) | |
7329 | goto err_misc_wq; | |
7330 | ||
7331 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
7332 | goto err_md; | |
7333 | ||
7334 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
7335 | goto err_mdp; | |
7336 | mdp_major = ret; | |
7337 | ||
3dbd8c2e | 7338 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
7339 | md_probe, NULL, NULL); |
7340 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
7341 | md_probe, NULL, NULL); |
7342 | ||
1da177e4 | 7343 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 7344 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
7345 | |
7346 | md_geninit(); | |
d710e138 | 7347 | return 0; |
1da177e4 | 7348 | |
e804ac78 TH |
7349 | err_mdp: |
7350 | unregister_blkdev(MD_MAJOR, "md"); | |
7351 | err_md: | |
7352 | destroy_workqueue(md_misc_wq); | |
7353 | err_misc_wq: | |
7354 | destroy_workqueue(md_wq); | |
7355 | err_wq: | |
7356 | return ret; | |
7357 | } | |
1da177e4 LT |
7358 | |
7359 | #ifndef MODULE | |
7360 | ||
7361 | /* | |
7362 | * Searches all registered partitions for autorun RAID arrays | |
7363 | * at boot time. | |
7364 | */ | |
4d936ec1 ME |
7365 | |
7366 | static LIST_HEAD(all_detected_devices); | |
7367 | struct detected_devices_node { | |
7368 | struct list_head list; | |
7369 | dev_t dev; | |
7370 | }; | |
1da177e4 LT |
7371 | |
7372 | void md_autodetect_dev(dev_t dev) | |
7373 | { | |
4d936ec1 ME |
7374 | struct detected_devices_node *node_detected_dev; |
7375 | ||
7376 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
7377 | if (node_detected_dev) { | |
7378 | node_detected_dev->dev = dev; | |
7379 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
7380 | } else { | |
7381 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
7382 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
7383 | } | |
1da177e4 LT |
7384 | } |
7385 | ||
7386 | ||
7387 | static void autostart_arrays(int part) | |
7388 | { | |
7389 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
7390 | struct detected_devices_node *node_detected_dev; |
7391 | dev_t dev; | |
7392 | int i_scanned, i_passed; | |
1da177e4 | 7393 | |
4d936ec1 ME |
7394 | i_scanned = 0; |
7395 | i_passed = 0; | |
1da177e4 | 7396 | |
4d936ec1 | 7397 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 7398 | |
4d936ec1 ME |
7399 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
7400 | i_scanned++; | |
7401 | node_detected_dev = list_entry(all_detected_devices.next, | |
7402 | struct detected_devices_node, list); | |
7403 | list_del(&node_detected_dev->list); | |
7404 | dev = node_detected_dev->dev; | |
7405 | kfree(node_detected_dev); | |
df968c4e | 7406 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
7407 | if (IS_ERR(rdev)) |
7408 | continue; | |
7409 | ||
b2d444d7 | 7410 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7411 | MD_BUG(); |
7412 | continue; | |
7413 | } | |
d0fae18f | 7414 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 7415 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 7416 | i_passed++; |
1da177e4 | 7417 | } |
4d936ec1 ME |
7418 | |
7419 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
7420 | i_scanned, i_passed); | |
1da177e4 LT |
7421 | |
7422 | autorun_devices(part); | |
7423 | } | |
7424 | ||
fdee8ae4 | 7425 | #endif /* !MODULE */ |
1da177e4 LT |
7426 | |
7427 | static __exit void md_exit(void) | |
7428 | { | |
7429 | mddev_t *mddev; | |
7430 | struct list_head *tmp; | |
8ab5e4c1 | 7431 | |
3dbd8c2e | 7432 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 7433 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 7434 | |
3dbd8c2e | 7435 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
7436 | unregister_blkdev(mdp_major, "mdp"); |
7437 | unregister_reboot_notifier(&md_notifier); | |
7438 | unregister_sysctl_table(raid_table_header); | |
7439 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 7440 | for_each_mddev(mddev, tmp) { |
1da177e4 | 7441 | export_array(mddev); |
d3374825 | 7442 | mddev->hold_active = 0; |
1da177e4 | 7443 | } |
e804ac78 TH |
7444 | destroy_workqueue(md_misc_wq); |
7445 | destroy_workqueue(md_wq); | |
1da177e4 LT |
7446 | } |
7447 | ||
685784aa | 7448 | subsys_initcall(md_init); |
1da177e4 LT |
7449 | module_exit(md_exit) |
7450 | ||
f91de92e N |
7451 | static int get_ro(char *buffer, struct kernel_param *kp) |
7452 | { | |
7453 | return sprintf(buffer, "%d", start_readonly); | |
7454 | } | |
7455 | static int set_ro(const char *val, struct kernel_param *kp) | |
7456 | { | |
7457 | char *e; | |
7458 | int num = simple_strtoul(val, &e, 10); | |
7459 | if (*val && (*e == '\0' || *e == '\n')) { | |
7460 | start_readonly = num; | |
4dbcdc75 | 7461 | return 0; |
f91de92e N |
7462 | } |
7463 | return -EINVAL; | |
7464 | } | |
7465 | ||
80ca3a44 N |
7466 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
7467 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 7468 | |
efeb53c0 | 7469 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 7470 | |
1da177e4 LT |
7471 | EXPORT_SYMBOL(register_md_personality); |
7472 | EXPORT_SYMBOL(unregister_md_personality); | |
7473 | EXPORT_SYMBOL(md_error); | |
7474 | EXPORT_SYMBOL(md_done_sync); | |
7475 | EXPORT_SYMBOL(md_write_start); | |
7476 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
7477 | EXPORT_SYMBOL(md_register_thread); |
7478 | EXPORT_SYMBOL(md_unregister_thread); | |
7479 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
7480 | EXPORT_SYMBOL(md_check_recovery); |
7481 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7482 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 7483 | MODULE_ALIAS("md"); |
72008652 | 7484 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |