]>
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 | } | |
c5d79adb N |
1950 | if (!shared) |
1951 | set_bit(AllReserved, &rdev->flags); | |
1da177e4 LT |
1952 | rdev->bdev = bdev; |
1953 | return err; | |
1954 | } | |
1955 | ||
1956 | static void unlock_rdev(mdk_rdev_t *rdev) | |
1957 | { | |
1958 | struct block_device *bdev = rdev->bdev; | |
1959 | rdev->bdev = NULL; | |
1960 | if (!bdev) | |
1961 | MD_BUG(); | |
e525fd89 | 1962 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
1963 | } |
1964 | ||
1965 | void md_autodetect_dev(dev_t dev); | |
1966 | ||
1967 | static void export_rdev(mdk_rdev_t * rdev) | |
1968 | { | |
1969 | char b[BDEVNAME_SIZE]; | |
1970 | printk(KERN_INFO "md: export_rdev(%s)\n", | |
1971 | bdevname(rdev->bdev,b)); | |
1972 | if (rdev->mddev) | |
1973 | MD_BUG(); | |
1974 | free_disk_sb(rdev); | |
1da177e4 | 1975 | #ifndef MODULE |
d0fae18f N |
1976 | if (test_bit(AutoDetected, &rdev->flags)) |
1977 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
1978 | #endif |
1979 | unlock_rdev(rdev); | |
86e6ffdd | 1980 | kobject_put(&rdev->kobj); |
1da177e4 LT |
1981 | } |
1982 | ||
1983 | static void kick_rdev_from_array(mdk_rdev_t * rdev) | |
1984 | { | |
1985 | unbind_rdev_from_array(rdev); | |
1986 | export_rdev(rdev); | |
1987 | } | |
1988 | ||
1989 | static void export_array(mddev_t *mddev) | |
1990 | { | |
159ec1fc | 1991 | mdk_rdev_t *rdev, *tmp; |
1da177e4 | 1992 | |
d089c6af | 1993 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
1994 | if (!rdev->mddev) { |
1995 | MD_BUG(); | |
1996 | continue; | |
1997 | } | |
1998 | kick_rdev_from_array(rdev); | |
1999 | } | |
2000 | if (!list_empty(&mddev->disks)) | |
2001 | MD_BUG(); | |
2002 | mddev->raid_disks = 0; | |
2003 | mddev->major_version = 0; | |
2004 | } | |
2005 | ||
2006 | static void print_desc(mdp_disk_t *desc) | |
2007 | { | |
2008 | printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, | |
2009 | desc->major,desc->minor,desc->raid_disk,desc->state); | |
2010 | } | |
2011 | ||
cd2ac932 | 2012 | static void print_sb_90(mdp_super_t *sb) |
1da177e4 LT |
2013 | { |
2014 | int i; | |
2015 | ||
2016 | printk(KERN_INFO | |
2017 | "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", | |
2018 | sb->major_version, sb->minor_version, sb->patch_version, | |
2019 | sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, | |
2020 | sb->ctime); | |
2021 | printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", | |
2022 | sb->level, sb->size, sb->nr_disks, sb->raid_disks, | |
2023 | sb->md_minor, sb->layout, sb->chunk_size); | |
2024 | printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" | |
2025 | " FD:%d SD:%d CSUM:%08x E:%08lx\n", | |
2026 | sb->utime, sb->state, sb->active_disks, sb->working_disks, | |
2027 | sb->failed_disks, sb->spare_disks, | |
2028 | sb->sb_csum, (unsigned long)sb->events_lo); | |
2029 | ||
2030 | printk(KERN_INFO); | |
2031 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2032 | mdp_disk_t *desc; | |
2033 | ||
2034 | desc = sb->disks + i; | |
2035 | if (desc->number || desc->major || desc->minor || | |
2036 | desc->raid_disk || (desc->state && (desc->state != 4))) { | |
2037 | printk(" D %2d: ", i); | |
2038 | print_desc(desc); | |
2039 | } | |
2040 | } | |
2041 | printk(KERN_INFO "md: THIS: "); | |
2042 | print_desc(&sb->this_disk); | |
cd2ac932 | 2043 | } |
1da177e4 | 2044 | |
cd2ac932 CR |
2045 | static void print_sb_1(struct mdp_superblock_1 *sb) |
2046 | { | |
2047 | __u8 *uuid; | |
2048 | ||
2049 | uuid = sb->set_uuid; | |
ad361c98 | 2050 | printk(KERN_INFO |
7b75c2f8 | 2051 | "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n" |
ad361c98 | 2052 | "md: Name: \"%s\" CT:%llu\n", |
cd2ac932 CR |
2053 | le32_to_cpu(sb->major_version), |
2054 | le32_to_cpu(sb->feature_map), | |
7b75c2f8 | 2055 | uuid, |
cd2ac932 CR |
2056 | sb->set_name, |
2057 | (unsigned long long)le64_to_cpu(sb->ctime) | |
2058 | & MD_SUPERBLOCK_1_TIME_SEC_MASK); | |
2059 | ||
2060 | uuid = sb->device_uuid; | |
ad361c98 JP |
2061 | printk(KERN_INFO |
2062 | "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu" | |
cd2ac932 | 2063 | " RO:%llu\n" |
7b75c2f8 | 2064 | "md: Dev:%08x UUID: %pU\n" |
ad361c98 JP |
2065 | "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n" |
2066 | "md: (MaxDev:%u) \n", | |
cd2ac932 CR |
2067 | le32_to_cpu(sb->level), |
2068 | (unsigned long long)le64_to_cpu(sb->size), | |
2069 | le32_to_cpu(sb->raid_disks), | |
2070 | le32_to_cpu(sb->layout), | |
2071 | le32_to_cpu(sb->chunksize), | |
2072 | (unsigned long long)le64_to_cpu(sb->data_offset), | |
2073 | (unsigned long long)le64_to_cpu(sb->data_size), | |
2074 | (unsigned long long)le64_to_cpu(sb->super_offset), | |
2075 | (unsigned long long)le64_to_cpu(sb->recovery_offset), | |
2076 | le32_to_cpu(sb->dev_number), | |
7b75c2f8 | 2077 | uuid, |
cd2ac932 CR |
2078 | sb->devflags, |
2079 | (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK, | |
2080 | (unsigned long long)le64_to_cpu(sb->events), | |
2081 | (unsigned long long)le64_to_cpu(sb->resync_offset), | |
2082 | le32_to_cpu(sb->sb_csum), | |
2083 | le32_to_cpu(sb->max_dev) | |
2084 | ); | |
1da177e4 LT |
2085 | } |
2086 | ||
cd2ac932 | 2087 | static void print_rdev(mdk_rdev_t *rdev, int major_version) |
1da177e4 LT |
2088 | { |
2089 | char b[BDEVNAME_SIZE]; | |
dd8ac336 AN |
2090 | printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n", |
2091 | bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors, | |
b2d444d7 N |
2092 | test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags), |
2093 | rdev->desc_nr); | |
1da177e4 | 2094 | if (rdev->sb_loaded) { |
cd2ac932 CR |
2095 | printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version); |
2096 | switch (major_version) { | |
2097 | case 0: | |
2098 | print_sb_90((mdp_super_t*)page_address(rdev->sb_page)); | |
2099 | break; | |
2100 | case 1: | |
2101 | print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page)); | |
2102 | break; | |
2103 | } | |
1da177e4 LT |
2104 | } else |
2105 | printk(KERN_INFO "md: no rdev superblock!\n"); | |
2106 | } | |
2107 | ||
5e56341d | 2108 | static void md_print_devices(void) |
1da177e4 | 2109 | { |
159ec1fc | 2110 | struct list_head *tmp; |
1da177e4 LT |
2111 | mdk_rdev_t *rdev; |
2112 | mddev_t *mddev; | |
2113 | char b[BDEVNAME_SIZE]; | |
2114 | ||
2115 | printk("\n"); | |
2116 | printk("md: **********************************\n"); | |
2117 | printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); | |
2118 | printk("md: **********************************\n"); | |
29ac4aa3 | 2119 | for_each_mddev(mddev, tmp) { |
1da177e4 | 2120 | |
32a7627c N |
2121 | if (mddev->bitmap) |
2122 | bitmap_print_sb(mddev->bitmap); | |
2123 | else | |
2124 | printk("%s: ", mdname(mddev)); | |
159ec1fc | 2125 | list_for_each_entry(rdev, &mddev->disks, same_set) |
1da177e4 LT |
2126 | printk("<%s>", bdevname(rdev->bdev,b)); |
2127 | printk("\n"); | |
2128 | ||
159ec1fc | 2129 | list_for_each_entry(rdev, &mddev->disks, same_set) |
cd2ac932 | 2130 | print_rdev(rdev, mddev->major_version); |
1da177e4 LT |
2131 | } |
2132 | printk("md: **********************************\n"); | |
2133 | printk("\n"); | |
2134 | } | |
2135 | ||
2136 | ||
42543769 | 2137 | static void sync_sbs(mddev_t * mddev, int nospares) |
1da177e4 | 2138 | { |
42543769 N |
2139 | /* Update each superblock (in-memory image), but |
2140 | * if we are allowed to, skip spares which already | |
2141 | * have the right event counter, or have one earlier | |
2142 | * (which would mean they aren't being marked as dirty | |
2143 | * with the rest of the array) | |
2144 | */ | |
1da177e4 | 2145 | mdk_rdev_t *rdev; |
159ec1fc | 2146 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
42543769 N |
2147 | if (rdev->sb_events == mddev->events || |
2148 | (nospares && | |
2149 | rdev->raid_disk < 0 && | |
42543769 N |
2150 | rdev->sb_events+1 == mddev->events)) { |
2151 | /* Don't update this superblock */ | |
2152 | rdev->sb_loaded = 2; | |
2153 | } else { | |
2154 | super_types[mddev->major_version]. | |
2155 | sync_super(mddev, rdev); | |
2156 | rdev->sb_loaded = 1; | |
2157 | } | |
1da177e4 LT |
2158 | } |
2159 | } | |
2160 | ||
850b2b42 | 2161 | static void md_update_sb(mddev_t * mddev, int force_change) |
1da177e4 | 2162 | { |
1da177e4 | 2163 | mdk_rdev_t *rdev; |
06d91a5f | 2164 | int sync_req; |
42543769 | 2165 | int nospares = 0; |
1da177e4 | 2166 | |
1da177e4 | 2167 | repeat: |
3a3a5ddb N |
2168 | /* First make sure individual recovery_offsets are correct */ |
2169 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
2170 | if (rdev->raid_disk >= 0 && | |
2171 | mddev->delta_disks >= 0 && | |
2172 | !test_bit(In_sync, &rdev->flags) && | |
2173 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2174 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2175 | ||
2176 | } | |
bd52b746 | 2177 | if (!mddev->persistent) { |
070dc6dd | 2178 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2179 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
d97a41dc N |
2180 | if (!mddev->external) |
2181 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
3a3a5ddb N |
2182 | wake_up(&mddev->sb_wait); |
2183 | return; | |
2184 | } | |
2185 | ||
a9701a30 | 2186 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2187 | |
3a3a5ddb N |
2188 | mddev->utime = get_seconds(); |
2189 | ||
850b2b42 N |
2190 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2191 | force_change = 1; | |
2192 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2193 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2194 | * though if events isn't the right even/odd, we will have to do | |
2195 | * spares after all | |
2196 | */ | |
2197 | nospares = 1; | |
2198 | if (force_change) | |
2199 | nospares = 0; | |
2200 | if (mddev->degraded) | |
84692195 N |
2201 | /* If the array is degraded, then skipping spares is both |
2202 | * dangerous and fairly pointless. | |
2203 | * Dangerous because a device that was removed from the array | |
2204 | * might have a event_count that still looks up-to-date, | |
2205 | * so it can be re-added without a resync. | |
2206 | * Pointless because if there are any spares to skip, | |
2207 | * then a recovery will happen and soon that array won't | |
2208 | * be degraded any more and the spare can go back to sleep then. | |
2209 | */ | |
850b2b42 | 2210 | nospares = 0; |
84692195 | 2211 | |
06d91a5f | 2212 | sync_req = mddev->in_sync; |
42543769 N |
2213 | |
2214 | /* If this is just a dirty<->clean transition, and the array is clean | |
2215 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2216 | if (nospares |
42543769 | 2217 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2218 | && mddev->can_decrease_events |
2219 | && mddev->events != 1) { | |
42543769 | 2220 | mddev->events--; |
a8707c08 N |
2221 | mddev->can_decrease_events = 0; |
2222 | } else { | |
42543769 N |
2223 | /* otherwise we have to go forward and ... */ |
2224 | mddev->events ++; | |
a8707c08 | 2225 | mddev->can_decrease_events = nospares; |
42543769 | 2226 | } |
1da177e4 LT |
2227 | |
2228 | if (!mddev->events) { | |
2229 | /* | |
2230 | * oops, this 64-bit counter should never wrap. | |
2231 | * Either we are in around ~1 trillion A.C., assuming | |
2232 | * 1 reboot per second, or we have a bug: | |
2233 | */ | |
2234 | MD_BUG(); | |
2235 | mddev->events --; | |
2236 | } | |
e691063a | 2237 | sync_sbs(mddev, nospares); |
a9701a30 | 2238 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 LT |
2239 | |
2240 | dprintk(KERN_INFO | |
2241 | "md: updating %s RAID superblock on device (in sync %d)\n", | |
2242 | mdname(mddev),mddev->in_sync); | |
2243 | ||
4ad13663 | 2244 | bitmap_update_sb(mddev->bitmap); |
159ec1fc | 2245 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
2246 | char b[BDEVNAME_SIZE]; |
2247 | dprintk(KERN_INFO "md: "); | |
42543769 N |
2248 | if (rdev->sb_loaded != 1) |
2249 | continue; /* no noise on spare devices */ | |
b2d444d7 | 2250 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2251 | dprintk("(skipping faulty "); |
2252 | ||
2253 | dprintk("%s ", bdevname(rdev->bdev,b)); | |
b2d444d7 | 2254 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2255 | md_super_write(mddev,rdev, |
0f420358 | 2256 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 N |
2257 | rdev->sb_page); |
2258 | dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", | |
2259 | bdevname(rdev->bdev,b), | |
0f420358 | 2260 | (unsigned long long)rdev->sb_start); |
42543769 | 2261 | rdev->sb_events = mddev->events; |
7bfa19f2 | 2262 | |
1da177e4 LT |
2263 | } else |
2264 | dprintk(")\n"); | |
7bfa19f2 | 2265 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2266 | /* only need to write one superblock... */ |
2267 | break; | |
2268 | } | |
a9701a30 | 2269 | md_super_wait(mddev); |
850b2b42 | 2270 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2271 | |
a9701a30 | 2272 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2273 | if (mddev->in_sync != sync_req || |
2274 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2275 | /* have to write it out again */ |
a9701a30 | 2276 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2277 | goto repeat; |
2278 | } | |
850b2b42 | 2279 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2280 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2281 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2282 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2283 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2284 | |
1da177e4 LT |
2285 | } |
2286 | ||
7f6ce769 | 2287 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2288 | * We want to accept with case. For this we use cmd_match. |
2289 | */ | |
2290 | static int cmd_match(const char *cmd, const char *str) | |
2291 | { | |
2292 | /* See if cmd, written into a sysfs file, matches | |
2293 | * str. They must either be the same, or cmd can | |
2294 | * have a trailing newline | |
2295 | */ | |
2296 | while (*cmd && *str && *cmd == *str) { | |
2297 | cmd++; | |
2298 | str++; | |
2299 | } | |
2300 | if (*cmd == '\n') | |
2301 | cmd++; | |
2302 | if (*str || *cmd) | |
2303 | return 0; | |
2304 | return 1; | |
2305 | } | |
2306 | ||
86e6ffdd N |
2307 | struct rdev_sysfs_entry { |
2308 | struct attribute attr; | |
2309 | ssize_t (*show)(mdk_rdev_t *, char *); | |
2310 | ssize_t (*store)(mdk_rdev_t *, const char *, size_t); | |
2311 | }; | |
2312 | ||
2313 | static ssize_t | |
96de1e66 | 2314 | state_show(mdk_rdev_t *rdev, char *page) |
86e6ffdd N |
2315 | { |
2316 | char *sep = ""; | |
20a49ff6 | 2317 | size_t len = 0; |
86e6ffdd | 2318 | |
b2d444d7 | 2319 | if (test_bit(Faulty, &rdev->flags)) { |
86e6ffdd N |
2320 | len+= sprintf(page+len, "%sfaulty",sep); |
2321 | sep = ","; | |
2322 | } | |
b2d444d7 | 2323 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2324 | len += sprintf(page+len, "%sin_sync",sep); |
2325 | sep = ","; | |
2326 | } | |
f655675b N |
2327 | if (test_bit(WriteMostly, &rdev->flags)) { |
2328 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2329 | sep = ","; | |
2330 | } | |
6bfe0b49 DW |
2331 | if (test_bit(Blocked, &rdev->flags)) { |
2332 | len += sprintf(page+len, "%sblocked", sep); | |
2333 | sep = ","; | |
2334 | } | |
b2d444d7 N |
2335 | if (!test_bit(Faulty, &rdev->flags) && |
2336 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2337 | len += sprintf(page+len, "%sspare", sep); |
2338 | sep = ","; | |
2339 | } | |
2340 | return len+sprintf(page+len, "\n"); | |
2341 | } | |
2342 | ||
45dc2de1 N |
2343 | static ssize_t |
2344 | state_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2345 | { | |
2346 | /* can write | |
2347 | * faulty - simulates and error | |
2348 | * remove - disconnects the device | |
f655675b N |
2349 | * writemostly - sets write_mostly |
2350 | * -writemostly - clears write_mostly | |
6bfe0b49 DW |
2351 | * blocked - sets the Blocked flag |
2352 | * -blocked - clears the Blocked flag | |
6d56e278 | 2353 | * insync - sets Insync providing device isn't active |
45dc2de1 N |
2354 | */ |
2355 | int err = -EINVAL; | |
2356 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2357 | md_error(rdev->mddev, rdev); | |
2358 | err = 0; | |
2359 | } else if (cmd_match(buf, "remove")) { | |
2360 | if (rdev->raid_disk >= 0) | |
2361 | err = -EBUSY; | |
2362 | else { | |
2363 | mddev_t *mddev = rdev->mddev; | |
2364 | kick_rdev_from_array(rdev); | |
3f9d7b0d N |
2365 | if (mddev->pers) |
2366 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2367 | md_new_event(mddev); |
2368 | err = 0; | |
2369 | } | |
f655675b N |
2370 | } else if (cmd_match(buf, "writemostly")) { |
2371 | set_bit(WriteMostly, &rdev->flags); | |
2372 | err = 0; | |
2373 | } else if (cmd_match(buf, "-writemostly")) { | |
2374 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2375 | err = 0; |
2376 | } else if (cmd_match(buf, "blocked")) { | |
2377 | set_bit(Blocked, &rdev->flags); | |
2378 | err = 0; | |
2379 | } else if (cmd_match(buf, "-blocked")) { | |
2380 | clear_bit(Blocked, &rdev->flags); | |
2381 | wake_up(&rdev->blocked_wait); | |
2382 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2383 | md_wakeup_thread(rdev->mddev->thread); | |
2384 | ||
6d56e278 N |
2385 | err = 0; |
2386 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2387 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2388 | err = 0; |
45dc2de1 | 2389 | } |
00bcb4ac N |
2390 | if (!err) |
2391 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2392 | return err ? err : len; |
2393 | } | |
80ca3a44 N |
2394 | static struct rdev_sysfs_entry rdev_state = |
2395 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2396 | |
4dbcdc75 N |
2397 | static ssize_t |
2398 | errors_show(mdk_rdev_t *rdev, char *page) | |
2399 | { | |
2400 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2401 | } | |
2402 | ||
2403 | static ssize_t | |
2404 | errors_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2405 | { | |
2406 | char *e; | |
2407 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2408 | if (*buf && (*e == 0 || *e == '\n')) { | |
2409 | atomic_set(&rdev->corrected_errors, n); | |
2410 | return len; | |
2411 | } | |
2412 | return -EINVAL; | |
2413 | } | |
2414 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2415 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2416 | |
014236d2 N |
2417 | static ssize_t |
2418 | slot_show(mdk_rdev_t *rdev, char *page) | |
2419 | { | |
2420 | if (rdev->raid_disk < 0) | |
2421 | return sprintf(page, "none\n"); | |
2422 | else | |
2423 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2424 | } | |
2425 | ||
2426 | static ssize_t | |
2427 | slot_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2428 | { | |
2429 | char *e; | |
c303da6d N |
2430 | int err; |
2431 | char nm[20]; | |
014236d2 N |
2432 | int slot = simple_strtoul(buf, &e, 10); |
2433 | if (strncmp(buf, "none", 4)==0) | |
2434 | slot = -1; | |
2435 | else if (e==buf || (*e && *e!= '\n')) | |
2436 | return -EINVAL; | |
6c2fce2e | 2437 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2438 | /* Setting 'slot' on an active array requires also |
2439 | * updating the 'rd%d' link, and communicating | |
2440 | * with the personality with ->hot_*_disk. | |
2441 | * For now we only support removing | |
2442 | * failed/spare devices. This normally happens automatically, | |
2443 | * but not when the metadata is externally managed. | |
2444 | */ | |
c303da6d N |
2445 | if (rdev->raid_disk == -1) |
2446 | return -EEXIST; | |
2447 | /* personality does all needed checks */ | |
2448 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2449 | return -EINVAL; | |
2450 | err = rdev->mddev->pers-> | |
2451 | hot_remove_disk(rdev->mddev, rdev->raid_disk); | |
2452 | if (err) | |
2453 | return err; | |
2454 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2455 | sysfs_remove_link(&rdev->mddev->kobj, nm); | |
b7103107 | 2456 | rdev->raid_disk = -1; |
c303da6d N |
2457 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2458 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e NB |
2459 | } else if (rdev->mddev->pers) { |
2460 | mdk_rdev_t *rdev2; | |
6c2fce2e | 2461 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2462 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2463 | */ |
2464 | ||
2465 | if (rdev->raid_disk != -1) | |
2466 | return -EBUSY; | |
2467 | ||
2468 | if (rdev->mddev->pers->hot_add_disk == NULL) | |
2469 | return -EINVAL; | |
2470 | ||
159ec1fc | 2471 | list_for_each_entry(rdev2, &rdev->mddev->disks, same_set) |
6c2fce2e NB |
2472 | if (rdev2->raid_disk == slot) |
2473 | return -EEXIST; | |
2474 | ||
ba1b41b6 N |
2475 | if (slot >= rdev->mddev->raid_disks && |
2476 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2477 | return -ENOSPC; | |
2478 | ||
6c2fce2e NB |
2479 | rdev->raid_disk = slot; |
2480 | if (test_bit(In_sync, &rdev->flags)) | |
2481 | rdev->saved_raid_disk = slot; | |
2482 | else | |
2483 | rdev->saved_raid_disk = -1; | |
2484 | err = rdev->mddev->pers-> | |
2485 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2486 | if (err) { |
6c2fce2e | 2487 | rdev->raid_disk = -1; |
6c2fce2e | 2488 | return err; |
52664732 | 2489 | } else |
00bcb4ac | 2490 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6c2fce2e NB |
2491 | sprintf(nm, "rd%d", rdev->raid_disk); |
2492 | if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm)) | |
00bcb4ac | 2493 | /* failure here is OK */; |
6c2fce2e | 2494 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2495 | } else { |
ba1b41b6 N |
2496 | if (slot >= rdev->mddev->raid_disks && |
2497 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2498 | return -ENOSPC; |
2499 | rdev->raid_disk = slot; | |
2500 | /* assume it is working */ | |
c5d79adb N |
2501 | clear_bit(Faulty, &rdev->flags); |
2502 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2503 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2504 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2505 | } |
014236d2 N |
2506 | return len; |
2507 | } | |
2508 | ||
2509 | ||
2510 | static struct rdev_sysfs_entry rdev_slot = | |
80ca3a44 | 2511 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2512 | |
93c8cad0 N |
2513 | static ssize_t |
2514 | offset_show(mdk_rdev_t *rdev, char *page) | |
2515 | { | |
6961ece4 | 2516 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2517 | } |
2518 | ||
2519 | static ssize_t | |
2520 | offset_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2521 | { | |
2522 | char *e; | |
2523 | unsigned long long offset = simple_strtoull(buf, &e, 10); | |
2524 | if (e==buf || (*e && *e != '\n')) | |
2525 | return -EINVAL; | |
8ed0a521 | 2526 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2527 | return -EBUSY; |
dd8ac336 | 2528 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2529 | /* Must set offset before size, so overlap checks |
2530 | * can be sane */ | |
2531 | return -EBUSY; | |
93c8cad0 N |
2532 | rdev->data_offset = offset; |
2533 | return len; | |
2534 | } | |
2535 | ||
2536 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2537 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2538 | |
83303b61 N |
2539 | static ssize_t |
2540 | rdev_size_show(mdk_rdev_t *rdev, char *page) | |
2541 | { | |
dd8ac336 | 2542 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2543 | } |
2544 | ||
c5d79adb N |
2545 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2546 | { | |
2547 | /* check if two start/length pairs overlap */ | |
2548 | if (s1+l1 <= s2) | |
2549 | return 0; | |
2550 | if (s2+l2 <= s1) | |
2551 | return 0; | |
2552 | return 1; | |
2553 | } | |
2554 | ||
b522adcd DW |
2555 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2556 | { | |
2557 | unsigned long long blocks; | |
2558 | sector_t new; | |
2559 | ||
2560 | if (strict_strtoull(buf, 10, &blocks) < 0) | |
2561 | return -EINVAL; | |
2562 | ||
2563 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2564 | return -EINVAL; /* sector conversion overflow */ | |
2565 | ||
2566 | new = blocks * 2; | |
2567 | if (new != blocks * 2) | |
2568 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2569 | ||
2570 | *sectors = new; | |
2571 | return 0; | |
2572 | } | |
2573 | ||
83303b61 N |
2574 | static ssize_t |
2575 | rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2576 | { | |
27c529bb | 2577 | mddev_t *my_mddev = rdev->mddev; |
dd8ac336 | 2578 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2579 | sector_t sectors; |
27c529bb | 2580 | |
b522adcd | 2581 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2582 | return -EINVAL; |
0cd17fec | 2583 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2584 | if (my_mddev->persistent) { |
dd8ac336 AN |
2585 | sectors = super_types[my_mddev->major_version]. |
2586 | rdev_size_change(rdev, sectors); | |
2587 | if (!sectors) | |
0cd17fec | 2588 | return -EBUSY; |
dd8ac336 | 2589 | } else if (!sectors) |
77304d2a | 2590 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2591 | rdev->data_offset; |
0cd17fec | 2592 | } |
dd8ac336 | 2593 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2594 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2595 | |
dd8ac336 AN |
2596 | rdev->sectors = sectors; |
2597 | if (sectors > oldsectors && my_mddev->external) { | |
c5d79adb N |
2598 | /* need to check that all other rdevs with the same ->bdev |
2599 | * do not overlap. We need to unlock the mddev to avoid | |
dd8ac336 | 2600 | * a deadlock. We have already changed rdev->sectors, and if |
c5d79adb N |
2601 | * we have to change it back, we will have the lock again. |
2602 | */ | |
2603 | mddev_t *mddev; | |
2604 | int overlap = 0; | |
159ec1fc | 2605 | struct list_head *tmp; |
c5d79adb | 2606 | |
27c529bb | 2607 | mddev_unlock(my_mddev); |
29ac4aa3 | 2608 | for_each_mddev(mddev, tmp) { |
c5d79adb N |
2609 | mdk_rdev_t *rdev2; |
2610 | ||
2611 | mddev_lock(mddev); | |
159ec1fc | 2612 | list_for_each_entry(rdev2, &mddev->disks, same_set) |
c5d79adb N |
2613 | if (test_bit(AllReserved, &rdev2->flags) || |
2614 | (rdev->bdev == rdev2->bdev && | |
2615 | rdev != rdev2 && | |
dd8ac336 | 2616 | overlaps(rdev->data_offset, rdev->sectors, |
d07bd3bc | 2617 | rdev2->data_offset, |
dd8ac336 | 2618 | rdev2->sectors))) { |
c5d79adb N |
2619 | overlap = 1; |
2620 | break; | |
2621 | } | |
2622 | mddev_unlock(mddev); | |
2623 | if (overlap) { | |
2624 | mddev_put(mddev); | |
2625 | break; | |
2626 | } | |
2627 | } | |
27c529bb | 2628 | mddev_lock(my_mddev); |
c5d79adb N |
2629 | if (overlap) { |
2630 | /* Someone else could have slipped in a size | |
2631 | * change here, but doing so is just silly. | |
dd8ac336 | 2632 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2633 | * safe, and trust userspace not to race with |
2634 | * itself | |
2635 | */ | |
dd8ac336 | 2636 | rdev->sectors = oldsectors; |
c5d79adb N |
2637 | return -EBUSY; |
2638 | } | |
2639 | } | |
83303b61 N |
2640 | return len; |
2641 | } | |
2642 | ||
2643 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2644 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2645 | |
06e3c817 DW |
2646 | |
2647 | static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page) | |
2648 | { | |
2649 | unsigned long long recovery_start = rdev->recovery_offset; | |
2650 | ||
2651 | if (test_bit(In_sync, &rdev->flags) || | |
2652 | recovery_start == MaxSector) | |
2653 | return sprintf(page, "none\n"); | |
2654 | ||
2655 | return sprintf(page, "%llu\n", recovery_start); | |
2656 | } | |
2657 | ||
2658 | static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len) | |
2659 | { | |
2660 | unsigned long long recovery_start; | |
2661 | ||
2662 | if (cmd_match(buf, "none")) | |
2663 | recovery_start = MaxSector; | |
2664 | else if (strict_strtoull(buf, 10, &recovery_start)) | |
2665 | return -EINVAL; | |
2666 | ||
2667 | if (rdev->mddev->pers && | |
2668 | rdev->raid_disk >= 0) | |
2669 | return -EBUSY; | |
2670 | ||
2671 | rdev->recovery_offset = recovery_start; | |
2672 | if (recovery_start == MaxSector) | |
2673 | set_bit(In_sync, &rdev->flags); | |
2674 | else | |
2675 | clear_bit(In_sync, &rdev->flags); | |
2676 | return len; | |
2677 | } | |
2678 | ||
2679 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2680 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2681 | ||
86e6ffdd N |
2682 | static struct attribute *rdev_default_attrs[] = { |
2683 | &rdev_state.attr, | |
4dbcdc75 | 2684 | &rdev_errors.attr, |
014236d2 | 2685 | &rdev_slot.attr, |
93c8cad0 | 2686 | &rdev_offset.attr, |
83303b61 | 2687 | &rdev_size.attr, |
06e3c817 | 2688 | &rdev_recovery_start.attr, |
86e6ffdd N |
2689 | NULL, |
2690 | }; | |
2691 | static ssize_t | |
2692 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2693 | { | |
2694 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2695 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2696 | mddev_t *mddev = rdev->mddev; |
2697 | ssize_t rv; | |
86e6ffdd N |
2698 | |
2699 | if (!entry->show) | |
2700 | return -EIO; | |
27c529bb N |
2701 | |
2702 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2703 | if (!rv) { | |
2704 | if (rdev->mddev == NULL) | |
2705 | rv = -EBUSY; | |
2706 | else | |
2707 | rv = entry->show(rdev, page); | |
2708 | mddev_unlock(mddev); | |
2709 | } | |
2710 | return rv; | |
86e6ffdd N |
2711 | } |
2712 | ||
2713 | static ssize_t | |
2714 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2715 | const char *page, size_t length) | |
2716 | { | |
2717 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
2718 | mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj); | |
27c529bb N |
2719 | ssize_t rv; |
2720 | mddev_t *mddev = rdev->mddev; | |
86e6ffdd N |
2721 | |
2722 | if (!entry->store) | |
2723 | return -EIO; | |
67463acb N |
2724 | if (!capable(CAP_SYS_ADMIN)) |
2725 | return -EACCES; | |
27c529bb | 2726 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2727 | if (!rv) { |
27c529bb N |
2728 | if (rdev->mddev == NULL) |
2729 | rv = -EBUSY; | |
2730 | else | |
2731 | rv = entry->store(rdev, page, length); | |
6a51830e | 2732 | mddev_unlock(mddev); |
ca388059 N |
2733 | } |
2734 | return rv; | |
86e6ffdd N |
2735 | } |
2736 | ||
2737 | static void rdev_free(struct kobject *ko) | |
2738 | { | |
2739 | mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj); | |
2740 | kfree(rdev); | |
2741 | } | |
52cf25d0 | 2742 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
2743 | .show = rdev_attr_show, |
2744 | .store = rdev_attr_store, | |
2745 | }; | |
2746 | static struct kobj_type rdev_ktype = { | |
2747 | .release = rdev_free, | |
2748 | .sysfs_ops = &rdev_sysfs_ops, | |
2749 | .default_attrs = rdev_default_attrs, | |
2750 | }; | |
2751 | ||
e8bb9a83 N |
2752 | void md_rdev_init(mdk_rdev_t *rdev) |
2753 | { | |
2754 | rdev->desc_nr = -1; | |
2755 | rdev->saved_raid_disk = -1; | |
2756 | rdev->raid_disk = -1; | |
2757 | rdev->flags = 0; | |
2758 | rdev->data_offset = 0; | |
2759 | rdev->sb_events = 0; | |
2760 | rdev->last_read_error.tv_sec = 0; | |
2761 | rdev->last_read_error.tv_nsec = 0; | |
2762 | atomic_set(&rdev->nr_pending, 0); | |
2763 | atomic_set(&rdev->read_errors, 0); | |
2764 | atomic_set(&rdev->corrected_errors, 0); | |
2765 | ||
2766 | INIT_LIST_HEAD(&rdev->same_set); | |
2767 | init_waitqueue_head(&rdev->blocked_wait); | |
2768 | } | |
2769 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
2770 | /* |
2771 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
2772 | * | |
2773 | * mark the device faulty if: | |
2774 | * | |
2775 | * - the device is nonexistent (zero size) | |
2776 | * - the device has no valid superblock | |
2777 | * | |
2778 | * a faulty rdev _never_ has rdev->sb set. | |
2779 | */ | |
2780 | static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) | |
2781 | { | |
2782 | char b[BDEVNAME_SIZE]; | |
2783 | int err; | |
2784 | mdk_rdev_t *rdev; | |
2785 | sector_t size; | |
2786 | ||
9ffae0cf | 2787 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
2788 | if (!rdev) { |
2789 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
2790 | return ERR_PTR(-ENOMEM); | |
2791 | } | |
1da177e4 | 2792 | |
e8bb9a83 | 2793 | md_rdev_init(rdev); |
1da177e4 LT |
2794 | if ((err = alloc_disk_sb(rdev))) |
2795 | goto abort_free; | |
2796 | ||
c5d79adb | 2797 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
2798 | if (err) |
2799 | goto abort_free; | |
2800 | ||
f9cb074b | 2801 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 2802 | |
77304d2a | 2803 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 LT |
2804 | if (!size) { |
2805 | printk(KERN_WARNING | |
2806 | "md: %s has zero or unknown size, marking faulty!\n", | |
2807 | bdevname(rdev->bdev,b)); | |
2808 | err = -EINVAL; | |
2809 | goto abort_free; | |
2810 | } | |
2811 | ||
2812 | if (super_format >= 0) { | |
2813 | err = super_types[super_format]. | |
2814 | load_super(rdev, NULL, super_minor); | |
2815 | if (err == -EINVAL) { | |
df968c4e N |
2816 | printk(KERN_WARNING |
2817 | "md: %s does not have a valid v%d.%d " | |
2818 | "superblock, not importing!\n", | |
2819 | bdevname(rdev->bdev,b), | |
2820 | super_format, super_minor); | |
1da177e4 LT |
2821 | goto abort_free; |
2822 | } | |
2823 | if (err < 0) { | |
2824 | printk(KERN_WARNING | |
2825 | "md: could not read %s's sb, not importing!\n", | |
2826 | bdevname(rdev->bdev,b)); | |
2827 | goto abort_free; | |
2828 | } | |
2829 | } | |
6bfe0b49 | 2830 | |
1da177e4 LT |
2831 | return rdev; |
2832 | ||
2833 | abort_free: | |
2834 | if (rdev->sb_page) { | |
2835 | if (rdev->bdev) | |
2836 | unlock_rdev(rdev); | |
2837 | free_disk_sb(rdev); | |
2838 | } | |
2839 | kfree(rdev); | |
2840 | return ERR_PTR(err); | |
2841 | } | |
2842 | ||
2843 | /* | |
2844 | * Check a full RAID array for plausibility | |
2845 | */ | |
2846 | ||
2847 | ||
a757e64c | 2848 | static void analyze_sbs(mddev_t * mddev) |
1da177e4 LT |
2849 | { |
2850 | int i; | |
159ec1fc | 2851 | mdk_rdev_t *rdev, *freshest, *tmp; |
1da177e4 LT |
2852 | char b[BDEVNAME_SIZE]; |
2853 | ||
2854 | freshest = NULL; | |
d089c6af | 2855 | rdev_for_each(rdev, tmp, mddev) |
1da177e4 LT |
2856 | switch (super_types[mddev->major_version]. |
2857 | load_super(rdev, freshest, mddev->minor_version)) { | |
2858 | case 1: | |
2859 | freshest = rdev; | |
2860 | break; | |
2861 | case 0: | |
2862 | break; | |
2863 | default: | |
2864 | printk( KERN_ERR \ | |
2865 | "md: fatal superblock inconsistency in %s" | |
2866 | " -- removing from array\n", | |
2867 | bdevname(rdev->bdev,b)); | |
2868 | kick_rdev_from_array(rdev); | |
2869 | } | |
2870 | ||
2871 | ||
2872 | super_types[mddev->major_version]. | |
2873 | validate_super(mddev, freshest); | |
2874 | ||
2875 | i = 0; | |
d089c6af | 2876 | rdev_for_each(rdev, tmp, mddev) { |
233fca36 N |
2877 | if (mddev->max_disks && |
2878 | (rdev->desc_nr >= mddev->max_disks || | |
2879 | i > mddev->max_disks)) { | |
de01dfad N |
2880 | printk(KERN_WARNING |
2881 | "md: %s: %s: only %d devices permitted\n", | |
2882 | mdname(mddev), bdevname(rdev->bdev, b), | |
2883 | mddev->max_disks); | |
2884 | kick_rdev_from_array(rdev); | |
2885 | continue; | |
2886 | } | |
1da177e4 LT |
2887 | if (rdev != freshest) |
2888 | if (super_types[mddev->major_version]. | |
2889 | validate_super(mddev, rdev)) { | |
2890 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
2891 | " from array!\n", | |
2892 | bdevname(rdev->bdev,b)); | |
2893 | kick_rdev_from_array(rdev); | |
2894 | continue; | |
2895 | } | |
2896 | if (mddev->level == LEVEL_MULTIPATH) { | |
2897 | rdev->desc_nr = i++; | |
2898 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 2899 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 2900 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
2901 | rdev->raid_disk = -1; |
2902 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
2903 | } |
2904 | } | |
1da177e4 LT |
2905 | } |
2906 | ||
72e02075 N |
2907 | /* Read a fixed-point number. |
2908 | * Numbers in sysfs attributes should be in "standard" units where | |
2909 | * possible, so time should be in seconds. | |
2910 | * However we internally use a a much smaller unit such as | |
2911 | * milliseconds or jiffies. | |
2912 | * This function takes a decimal number with a possible fractional | |
2913 | * component, and produces an integer which is the result of | |
2914 | * multiplying that number by 10^'scale'. | |
2915 | * all without any floating-point arithmetic. | |
2916 | */ | |
2917 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
2918 | { | |
2919 | unsigned long result = 0; | |
2920 | long decimals = -1; | |
2921 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
2922 | if (*cp == '.') | |
2923 | decimals = 0; | |
2924 | else if (decimals < scale) { | |
2925 | unsigned int value; | |
2926 | value = *cp - '0'; | |
2927 | result = result * 10 + value; | |
2928 | if (decimals >= 0) | |
2929 | decimals++; | |
2930 | } | |
2931 | cp++; | |
2932 | } | |
2933 | if (*cp == '\n') | |
2934 | cp++; | |
2935 | if (*cp) | |
2936 | return -EINVAL; | |
2937 | if (decimals < 0) | |
2938 | decimals = 0; | |
2939 | while (decimals < scale) { | |
2940 | result *= 10; | |
2941 | decimals ++; | |
2942 | } | |
2943 | *res = result; | |
2944 | return 0; | |
2945 | } | |
2946 | ||
2947 | ||
19052c0e N |
2948 | static void md_safemode_timeout(unsigned long data); |
2949 | ||
16f17b39 N |
2950 | static ssize_t |
2951 | safe_delay_show(mddev_t *mddev, char *page) | |
2952 | { | |
2953 | int msec = (mddev->safemode_delay*1000)/HZ; | |
2954 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
2955 | } | |
2956 | static ssize_t | |
2957 | safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len) | |
2958 | { | |
16f17b39 | 2959 | unsigned long msec; |
97ce0a7f | 2960 | |
72e02075 | 2961 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 2962 | return -EINVAL; |
16f17b39 N |
2963 | if (msec == 0) |
2964 | mddev->safemode_delay = 0; | |
2965 | else { | |
19052c0e | 2966 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
2967 | mddev->safemode_delay = (msec*HZ)/1000; |
2968 | if (mddev->safemode_delay == 0) | |
2969 | mddev->safemode_delay = 1; | |
19052c0e N |
2970 | if (mddev->safemode_delay < old_delay) |
2971 | md_safemode_timeout((unsigned long)mddev); | |
16f17b39 N |
2972 | } |
2973 | return len; | |
2974 | } | |
2975 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 2976 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 2977 | |
eae1701f | 2978 | static ssize_t |
96de1e66 | 2979 | level_show(mddev_t *mddev, char *page) |
eae1701f | 2980 | { |
2604b703 | 2981 | struct mdk_personality *p = mddev->pers; |
d9d166c2 | 2982 | if (p) |
eae1701f | 2983 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
2984 | else if (mddev->clevel[0]) |
2985 | return sprintf(page, "%s\n", mddev->clevel); | |
2986 | else if (mddev->level != LEVEL_NONE) | |
2987 | return sprintf(page, "%d\n", mddev->level); | |
2988 | else | |
2989 | return 0; | |
eae1701f N |
2990 | } |
2991 | ||
d9d166c2 N |
2992 | static ssize_t |
2993 | level_store(mddev_t *mddev, const char *buf, size_t len) | |
2994 | { | |
f2859af6 | 2995 | char clevel[16]; |
20a49ff6 | 2996 | ssize_t rv = len; |
245f46c2 | 2997 | struct mdk_personality *pers; |
f2859af6 | 2998 | long level; |
245f46c2 | 2999 | void *priv; |
3a981b03 | 3000 | mdk_rdev_t *rdev; |
245f46c2 N |
3001 | |
3002 | if (mddev->pers == NULL) { | |
3003 | if (len == 0) | |
3004 | return 0; | |
3005 | if (len >= sizeof(mddev->clevel)) | |
3006 | return -ENOSPC; | |
3007 | strncpy(mddev->clevel, buf, len); | |
3008 | if (mddev->clevel[len-1] == '\n') | |
3009 | len--; | |
3010 | mddev->clevel[len] = 0; | |
3011 | mddev->level = LEVEL_NONE; | |
3012 | return rv; | |
3013 | } | |
3014 | ||
3015 | /* request to change the personality. Need to ensure: | |
3016 | * - array is not engaged in resync/recovery/reshape | |
3017 | * - old personality can be suspended | |
3018 | * - new personality will access other array. | |
3019 | */ | |
3020 | ||
bb4f1e9d N |
3021 | if (mddev->sync_thread || |
3022 | mddev->reshape_position != MaxSector || | |
3023 | mddev->sysfs_active) | |
d9d166c2 | 3024 | return -EBUSY; |
245f46c2 N |
3025 | |
3026 | if (!mddev->pers->quiesce) { | |
3027 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3028 | mdname(mddev), mddev->pers->name); | |
3029 | return -EINVAL; | |
3030 | } | |
3031 | ||
3032 | /* Now find the new personality */ | |
f2859af6 | 3033 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3034 | return -EINVAL; |
f2859af6 DW |
3035 | strncpy(clevel, buf, len); |
3036 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3037 | len--; |
f2859af6 DW |
3038 | clevel[len] = 0; |
3039 | if (strict_strtol(clevel, 10, &level)) | |
3040 | level = LEVEL_NONE; | |
245f46c2 | 3041 | |
f2859af6 DW |
3042 | if (request_module("md-%s", clevel) != 0) |
3043 | request_module("md-level-%s", clevel); | |
245f46c2 | 3044 | spin_lock(&pers_lock); |
f2859af6 | 3045 | pers = find_pers(level, clevel); |
245f46c2 N |
3046 | if (!pers || !try_module_get(pers->owner)) { |
3047 | spin_unlock(&pers_lock); | |
f2859af6 | 3048 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3049 | return -EINVAL; |
3050 | } | |
3051 | spin_unlock(&pers_lock); | |
3052 | ||
3053 | if (pers == mddev->pers) { | |
3054 | /* Nothing to do! */ | |
3055 | module_put(pers->owner); | |
3056 | return rv; | |
3057 | } | |
3058 | if (!pers->takeover) { | |
3059 | module_put(pers->owner); | |
3060 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3061 | mdname(mddev), clevel); |
245f46c2 N |
3062 | return -EINVAL; |
3063 | } | |
3064 | ||
e93f68a1 N |
3065 | list_for_each_entry(rdev, &mddev->disks, same_set) |
3066 | rdev->new_raid_disk = rdev->raid_disk; | |
3067 | ||
245f46c2 N |
3068 | /* ->takeover must set new_* and/or delta_disks |
3069 | * if it succeeds, and may set them when it fails. | |
3070 | */ | |
3071 | priv = pers->takeover(mddev); | |
3072 | if (IS_ERR(priv)) { | |
3073 | mddev->new_level = mddev->level; | |
3074 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3075 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3076 | mddev->raid_disks -= mddev->delta_disks; |
3077 | mddev->delta_disks = 0; | |
3078 | module_put(pers->owner); | |
3079 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3080 | mdname(mddev), clevel); |
245f46c2 N |
3081 | return PTR_ERR(priv); |
3082 | } | |
3083 | ||
3084 | /* Looks like we have a winner */ | |
3085 | mddev_suspend(mddev); | |
3086 | mddev->pers->stop(mddev); | |
a64c876f N |
3087 | |
3088 | if (mddev->pers->sync_request == NULL && | |
3089 | pers->sync_request != NULL) { | |
3090 | /* need to add the md_redundancy_group */ | |
3091 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3092 | printk(KERN_WARNING | |
3093 | "md: cannot register extra attributes for %s\n", | |
3094 | mdname(mddev)); | |
19fdb9ee | 3095 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action"); |
a64c876f N |
3096 | } |
3097 | if (mddev->pers->sync_request != NULL && | |
3098 | pers->sync_request == NULL) { | |
3099 | /* need to remove the md_redundancy_group */ | |
3100 | if (mddev->to_remove == NULL) | |
3101 | mddev->to_remove = &md_redundancy_group; | |
3102 | } | |
3103 | ||
54071b38 TM |
3104 | if (mddev->pers->sync_request == NULL && |
3105 | mddev->external) { | |
3106 | /* We are converting from a no-redundancy array | |
3107 | * to a redundancy array and metadata is managed | |
3108 | * externally so we need to be sure that writes | |
3109 | * won't block due to a need to transition | |
3110 | * clean->dirty | |
3111 | * until external management is started. | |
3112 | */ | |
3113 | mddev->in_sync = 0; | |
3114 | mddev->safemode_delay = 0; | |
3115 | mddev->safemode = 0; | |
3116 | } | |
3117 | ||
e93f68a1 N |
3118 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
3119 | char nm[20]; | |
3120 | if (rdev->raid_disk < 0) | |
3121 | continue; | |
bf2cb0da | 3122 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3123 | rdev->new_raid_disk = -1; |
3124 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3125 | continue; | |
3126 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3127 | sysfs_remove_link(&mddev->kobj, nm); | |
3128 | } | |
3129 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
3130 | if (rdev->raid_disk < 0) | |
3131 | continue; | |
3132 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3133 | continue; | |
3134 | rdev->raid_disk = rdev->new_raid_disk; | |
3135 | if (rdev->raid_disk < 0) | |
3a981b03 | 3136 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 N |
3137 | else { |
3138 | char nm[20]; | |
3139 | sprintf(nm, "rd%d", rdev->raid_disk); | |
3140 | if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) | |
3141 | printk("md: cannot register %s for %s after level change\n", | |
3142 | nm, mdname(mddev)); | |
3a981b03 | 3143 | } |
e93f68a1 N |
3144 | } |
3145 | ||
3146 | module_put(mddev->pers->owner); | |
245f46c2 N |
3147 | mddev->pers = pers; |
3148 | mddev->private = priv; | |
3149 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3150 | mddev->level = mddev->new_level; | |
3151 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3152 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3153 | mddev->delta_disks = 0; |
9af204cf TM |
3154 | if (mddev->pers->sync_request == NULL) { |
3155 | /* this is now an array without redundancy, so | |
3156 | * it must always be in_sync | |
3157 | */ | |
3158 | mddev->in_sync = 1; | |
3159 | del_timer_sync(&mddev->safemode_timer); | |
3160 | } | |
245f46c2 N |
3161 | pers->run(mddev); |
3162 | mddev_resume(mddev); | |
3163 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3164 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3165 | md_wakeup_thread(mddev->thread); | |
5cac7861 | 3166 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3167 | md_new_event(mddev); |
d9d166c2 N |
3168 | return rv; |
3169 | } | |
3170 | ||
3171 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3172 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3173 | |
d4dbd025 N |
3174 | |
3175 | static ssize_t | |
3176 | layout_show(mddev_t *mddev, char *page) | |
3177 | { | |
3178 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3179 | if (mddev->reshape_position != MaxSector && |
3180 | mddev->layout != mddev->new_layout) | |
3181 | return sprintf(page, "%d (%d)\n", | |
3182 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3183 | return sprintf(page, "%d\n", mddev->layout); |
3184 | } | |
3185 | ||
3186 | static ssize_t | |
3187 | layout_store(mddev_t *mddev, const char *buf, size_t len) | |
3188 | { | |
3189 | char *e; | |
3190 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3191 | |
3192 | if (!*buf || (*e && *e != '\n')) | |
3193 | return -EINVAL; | |
3194 | ||
b3546035 N |
3195 | if (mddev->pers) { |
3196 | int err; | |
50ac168a | 3197 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3198 | return -EBUSY; |
597a711b | 3199 | mddev->new_layout = n; |
50ac168a | 3200 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3201 | if (err) { |
3202 | mddev->new_layout = mddev->layout; | |
b3546035 | 3203 | return err; |
597a711b | 3204 | } |
b3546035 | 3205 | } else { |
08a02ecd | 3206 | mddev->new_layout = n; |
b3546035 N |
3207 | if (mddev->reshape_position == MaxSector) |
3208 | mddev->layout = n; | |
3209 | } | |
d4dbd025 N |
3210 | return len; |
3211 | } | |
3212 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3213 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 N |
3214 | |
3215 | ||
eae1701f | 3216 | static ssize_t |
96de1e66 | 3217 | raid_disks_show(mddev_t *mddev, char *page) |
eae1701f | 3218 | { |
bb636547 N |
3219 | if (mddev->raid_disks == 0) |
3220 | return 0; | |
08a02ecd N |
3221 | if (mddev->reshape_position != MaxSector && |
3222 | mddev->delta_disks != 0) | |
3223 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3224 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3225 | return sprintf(page, "%d\n", mddev->raid_disks); |
3226 | } | |
3227 | ||
da943b99 N |
3228 | static int update_raid_disks(mddev_t *mddev, int raid_disks); |
3229 | ||
3230 | static ssize_t | |
3231 | raid_disks_store(mddev_t *mddev, const char *buf, size_t len) | |
3232 | { | |
da943b99 N |
3233 | char *e; |
3234 | int rv = 0; | |
3235 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3236 | ||
3237 | if (!*buf || (*e && *e != '\n')) | |
3238 | return -EINVAL; | |
3239 | ||
3240 | if (mddev->pers) | |
3241 | rv = update_raid_disks(mddev, n); | |
08a02ecd N |
3242 | else if (mddev->reshape_position != MaxSector) { |
3243 | int olddisks = mddev->raid_disks - mddev->delta_disks; | |
3244 | mddev->delta_disks = n - olddisks; | |
3245 | mddev->raid_disks = n; | |
3246 | } else | |
da943b99 N |
3247 | mddev->raid_disks = n; |
3248 | return rv ? rv : len; | |
3249 | } | |
3250 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3251 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3252 | |
3b34380a N |
3253 | static ssize_t |
3254 | chunk_size_show(mddev_t *mddev, char *page) | |
3255 | { | |
08a02ecd | 3256 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3257 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3258 | return sprintf(page, "%d (%d)\n", | |
3259 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3260 | mddev->chunk_sectors << 9); |
3261 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3262 | } |
3263 | ||
3264 | static ssize_t | |
3265 | chunk_size_store(mddev_t *mddev, const char *buf, size_t len) | |
3266 | { | |
3b34380a N |
3267 | char *e; |
3268 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3269 | ||
3b34380a N |
3270 | if (!*buf || (*e && *e != '\n')) |
3271 | return -EINVAL; | |
3272 | ||
b3546035 N |
3273 | if (mddev->pers) { |
3274 | int err; | |
50ac168a | 3275 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3276 | return -EBUSY; |
597a711b | 3277 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3278 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3279 | if (err) { |
3280 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3281 | return err; |
597a711b | 3282 | } |
b3546035 | 3283 | } else { |
664e7c41 | 3284 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3285 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3286 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3287 | } |
3b34380a N |
3288 | return len; |
3289 | } | |
3290 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3291 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3292 | |
a94213b1 N |
3293 | static ssize_t |
3294 | resync_start_show(mddev_t *mddev, char *page) | |
3295 | { | |
d1a7c503 N |
3296 | if (mddev->recovery_cp == MaxSector) |
3297 | return sprintf(page, "none\n"); | |
a94213b1 N |
3298 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3299 | } | |
3300 | ||
3301 | static ssize_t | |
3302 | resync_start_store(mddev_t *mddev, const char *buf, size_t len) | |
3303 | { | |
a94213b1 N |
3304 | char *e; |
3305 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3306 | ||
3307 | if (mddev->pers) | |
3308 | return -EBUSY; | |
06e3c817 DW |
3309 | if (cmd_match(buf, "none")) |
3310 | n = MaxSector; | |
3311 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3312 | return -EINVAL; |
3313 | ||
3314 | mddev->recovery_cp = n; | |
3315 | return len; | |
3316 | } | |
3317 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3318 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3319 | |
9e653b63 N |
3320 | /* |
3321 | * The array state can be: | |
3322 | * | |
3323 | * clear | |
3324 | * No devices, no size, no level | |
3325 | * Equivalent to STOP_ARRAY ioctl | |
3326 | * inactive | |
3327 | * May have some settings, but array is not active | |
3328 | * all IO results in error | |
3329 | * When written, doesn't tear down array, but just stops it | |
3330 | * suspended (not supported yet) | |
3331 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3332 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3333 | * readonly |
3334 | * no resync can happen. no superblocks get written. | |
3335 | * write requests fail | |
3336 | * read-auto | |
3337 | * like readonly, but behaves like 'clean' on a write request. | |
3338 | * | |
3339 | * clean - no pending writes, but otherwise active. | |
3340 | * When written to inactive array, starts without resync | |
3341 | * If a write request arrives then | |
3342 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3343 | * if not known, block and switch to write-pending | |
3344 | * If written to an active array that has pending writes, then fails. | |
3345 | * active | |
3346 | * fully active: IO and resync can be happening. | |
3347 | * When written to inactive array, starts with resync | |
3348 | * | |
3349 | * write-pending | |
3350 | * clean, but writes are blocked waiting for 'active' to be written. | |
3351 | * | |
3352 | * active-idle | |
3353 | * like active, but no writes have been seen for a while (100msec). | |
3354 | * | |
3355 | */ | |
3356 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3357 | write_pending, active_idle, bad_word}; | |
05381954 | 3358 | static char *array_states[] = { |
9e653b63 N |
3359 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3360 | "write-pending", "active-idle", NULL }; | |
3361 | ||
3362 | static int match_word(const char *word, char **list) | |
3363 | { | |
3364 | int n; | |
3365 | for (n=0; list[n]; n++) | |
3366 | if (cmd_match(word, list[n])) | |
3367 | break; | |
3368 | return n; | |
3369 | } | |
3370 | ||
3371 | static ssize_t | |
3372 | array_state_show(mddev_t *mddev, char *page) | |
3373 | { | |
3374 | enum array_state st = inactive; | |
3375 | ||
3376 | if (mddev->pers) | |
3377 | switch(mddev->ro) { | |
3378 | case 1: | |
3379 | st = readonly; | |
3380 | break; | |
3381 | case 2: | |
3382 | st = read_auto; | |
3383 | break; | |
3384 | case 0: | |
3385 | if (mddev->in_sync) | |
3386 | st = clean; | |
070dc6dd | 3387 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3388 | st = write_pending; |
9e653b63 N |
3389 | else if (mddev->safemode) |
3390 | st = active_idle; | |
3391 | else | |
3392 | st = active; | |
3393 | } | |
3394 | else { | |
3395 | if (list_empty(&mddev->disks) && | |
3396 | mddev->raid_disks == 0 && | |
58c0fed4 | 3397 | mddev->dev_sectors == 0) |
9e653b63 N |
3398 | st = clear; |
3399 | else | |
3400 | st = inactive; | |
3401 | } | |
3402 | return sprintf(page, "%s\n", array_states[st]); | |
3403 | } | |
3404 | ||
df5b20cf | 3405 | static int do_md_stop(mddev_t * mddev, int ro, int is_open); |
a4bd82d0 | 3406 | static int md_set_readonly(mddev_t * mddev, int is_open); |
9e653b63 N |
3407 | static int do_md_run(mddev_t * mddev); |
3408 | static int restart_array(mddev_t *mddev); | |
3409 | ||
3410 | static ssize_t | |
3411 | array_state_store(mddev_t *mddev, const char *buf, size_t len) | |
3412 | { | |
3413 | int err = -EINVAL; | |
3414 | enum array_state st = match_word(buf, array_states); | |
3415 | switch(st) { | |
3416 | case bad_word: | |
3417 | break; | |
3418 | case clear: | |
3419 | /* stopping an active array */ | |
f2ea68cf | 3420 | if (atomic_read(&mddev->openers) > 0) |
e691063a | 3421 | return -EBUSY; |
df5b20cf | 3422 | err = do_md_stop(mddev, 0, 0); |
9e653b63 N |
3423 | break; |
3424 | case inactive: | |
3425 | /* stopping an active array */ | |
3426 | if (mddev->pers) { | |
f2ea68cf | 3427 | if (atomic_read(&mddev->openers) > 0) |
9e653b63 | 3428 | return -EBUSY; |
df5b20cf | 3429 | err = do_md_stop(mddev, 2, 0); |
e691063a N |
3430 | } else |
3431 | err = 0; /* already inactive */ | |
9e653b63 N |
3432 | break; |
3433 | case suspended: | |
3434 | break; /* not supported yet */ | |
3435 | case readonly: | |
3436 | if (mddev->pers) | |
a4bd82d0 | 3437 | err = md_set_readonly(mddev, 0); |
9e653b63 N |
3438 | else { |
3439 | mddev->ro = 1; | |
648b629e | 3440 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3441 | err = do_md_run(mddev); |
3442 | } | |
3443 | break; | |
3444 | case read_auto: | |
9e653b63 | 3445 | if (mddev->pers) { |
80268ee9 | 3446 | if (mddev->ro == 0) |
a4bd82d0 | 3447 | err = md_set_readonly(mddev, 0); |
80268ee9 | 3448 | else if (mddev->ro == 1) |
648b629e N |
3449 | err = restart_array(mddev); |
3450 | if (err == 0) { | |
3451 | mddev->ro = 2; | |
3452 | set_disk_ro(mddev->gendisk, 0); | |
3453 | } | |
9e653b63 N |
3454 | } else { |
3455 | mddev->ro = 2; | |
3456 | err = do_md_run(mddev); | |
3457 | } | |
3458 | break; | |
3459 | case clean: | |
3460 | if (mddev->pers) { | |
3461 | restart_array(mddev); | |
3462 | spin_lock_irq(&mddev->write_lock); | |
3463 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3464 | if (mddev->in_sync == 0) { |
3465 | mddev->in_sync = 1; | |
31a59e34 N |
3466 | if (mddev->safemode == 1) |
3467 | mddev->safemode = 0; | |
070dc6dd | 3468 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3469 | } |
3470 | err = 0; | |
3471 | } else | |
3472 | err = -EBUSY; | |
9e653b63 | 3473 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3474 | } else |
3475 | err = -EINVAL; | |
9e653b63 N |
3476 | break; |
3477 | case active: | |
3478 | if (mddev->pers) { | |
3479 | restart_array(mddev); | |
070dc6dd | 3480 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3481 | wake_up(&mddev->sb_wait); |
3482 | err = 0; | |
3483 | } else { | |
3484 | mddev->ro = 0; | |
648b629e | 3485 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3486 | err = do_md_run(mddev); |
3487 | } | |
3488 | break; | |
3489 | case write_pending: | |
3490 | case active_idle: | |
3491 | /* these cannot be set */ | |
3492 | break; | |
3493 | } | |
3494 | if (err) | |
3495 | return err; | |
0fd62b86 | 3496 | else { |
00bcb4ac | 3497 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3498 | return len; |
0fd62b86 | 3499 | } |
9e653b63 | 3500 | } |
80ca3a44 N |
3501 | static struct md_sysfs_entry md_array_state = |
3502 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3503 | |
1e50915f RB |
3504 | static ssize_t |
3505 | max_corrected_read_errors_show(mddev_t *mddev, char *page) { | |
3506 | return sprintf(page, "%d\n", | |
3507 | atomic_read(&mddev->max_corr_read_errors)); | |
3508 | } | |
3509 | ||
3510 | static ssize_t | |
3511 | max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len) | |
3512 | { | |
3513 | char *e; | |
3514 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3515 | ||
3516 | if (*buf && (*e == 0 || *e == '\n')) { | |
3517 | atomic_set(&mddev->max_corr_read_errors, n); | |
3518 | return len; | |
3519 | } | |
3520 | return -EINVAL; | |
3521 | } | |
3522 | ||
3523 | static struct md_sysfs_entry max_corr_read_errors = | |
3524 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3525 | max_corrected_read_errors_store); | |
3526 | ||
6d7ff738 N |
3527 | static ssize_t |
3528 | null_show(mddev_t *mddev, char *page) | |
3529 | { | |
3530 | return -EINVAL; | |
3531 | } | |
3532 | ||
3533 | static ssize_t | |
3534 | new_dev_store(mddev_t *mddev, const char *buf, size_t len) | |
3535 | { | |
3536 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3537 | /* The new device is added to the array. | |
3538 | * If the array has a persistent superblock, we read the | |
3539 | * superblock to initialise info and check validity. | |
3540 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3541 | * which mainly checks size. | |
3542 | */ | |
3543 | char *e; | |
3544 | int major = simple_strtoul(buf, &e, 10); | |
3545 | int minor; | |
3546 | dev_t dev; | |
3547 | mdk_rdev_t *rdev; | |
3548 | int err; | |
3549 | ||
3550 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3551 | return -EINVAL; | |
3552 | minor = simple_strtoul(e+1, &e, 10); | |
3553 | if (*e && *e != '\n') | |
3554 | return -EINVAL; | |
3555 | dev = MKDEV(major, minor); | |
3556 | if (major != MAJOR(dev) || | |
3557 | minor != MINOR(dev)) | |
3558 | return -EOVERFLOW; | |
3559 | ||
3560 | ||
3561 | if (mddev->persistent) { | |
3562 | rdev = md_import_device(dev, mddev->major_version, | |
3563 | mddev->minor_version); | |
3564 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3565 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
3566 | mdk_rdev_t, same_set); | |
3567 | err = super_types[mddev->major_version] | |
3568 | .load_super(rdev, rdev0, mddev->minor_version); | |
3569 | if (err < 0) | |
3570 | goto out; | |
3571 | } | |
c5d79adb N |
3572 | } else if (mddev->external) |
3573 | rdev = md_import_device(dev, -2, -1); | |
3574 | else | |
6d7ff738 N |
3575 | rdev = md_import_device(dev, -1, -1); |
3576 | ||
3577 | if (IS_ERR(rdev)) | |
3578 | return PTR_ERR(rdev); | |
3579 | err = bind_rdev_to_array(rdev, mddev); | |
3580 | out: | |
3581 | if (err) | |
3582 | export_rdev(rdev); | |
3583 | return err ? err : len; | |
3584 | } | |
3585 | ||
3586 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3587 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3588 | |
9b1d1dac PC |
3589 | static ssize_t |
3590 | bitmap_store(mddev_t *mddev, const char *buf, size_t len) | |
3591 | { | |
3592 | char *end; | |
3593 | unsigned long chunk, end_chunk; | |
3594 | ||
3595 | if (!mddev->bitmap) | |
3596 | goto out; | |
3597 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3598 | while (*buf) { | |
3599 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3600 | if (buf == end) break; | |
3601 | if (*end == '-') { /* range */ | |
3602 | buf = end + 1; | |
3603 | end_chunk = simple_strtoul(buf, &end, 0); | |
3604 | if (buf == end) break; | |
3605 | } | |
3606 | if (*end && !isspace(*end)) break; | |
3607 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3608 | buf = skip_spaces(end); |
9b1d1dac PC |
3609 | } |
3610 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3611 | out: | |
3612 | return len; | |
3613 | } | |
3614 | ||
3615 | static struct md_sysfs_entry md_bitmap = | |
3616 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3617 | ||
a35b0d69 N |
3618 | static ssize_t |
3619 | size_show(mddev_t *mddev, char *page) | |
3620 | { | |
58c0fed4 AN |
3621 | return sprintf(page, "%llu\n", |
3622 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3623 | } |
3624 | ||
d71f9f88 | 3625 | static int update_size(mddev_t *mddev, sector_t num_sectors); |
a35b0d69 N |
3626 | |
3627 | static ssize_t | |
3628 | size_store(mddev_t *mddev, const char *buf, size_t len) | |
3629 | { | |
3630 | /* If array is inactive, we can reduce the component size, but | |
3631 | * not increase it (except from 0). | |
3632 | * If array is active, we can try an on-line resize | |
3633 | */ | |
b522adcd DW |
3634 | sector_t sectors; |
3635 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3636 | |
58c0fed4 AN |
3637 | if (err < 0) |
3638 | return err; | |
a35b0d69 | 3639 | if (mddev->pers) { |
58c0fed4 | 3640 | err = update_size(mddev, sectors); |
850b2b42 | 3641 | md_update_sb(mddev, 1); |
a35b0d69 | 3642 | } else { |
58c0fed4 AN |
3643 | if (mddev->dev_sectors == 0 || |
3644 | mddev->dev_sectors > sectors) | |
3645 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3646 | else |
3647 | err = -ENOSPC; | |
3648 | } | |
3649 | return err ? err : len; | |
3650 | } | |
3651 | ||
3652 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3653 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3654 | |
8bb93aac N |
3655 | |
3656 | /* Metdata version. | |
e691063a N |
3657 | * This is one of |
3658 | * 'none' for arrays with no metadata (good luck...) | |
3659 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3660 | * or N.M for internally known formats |
3661 | */ | |
3662 | static ssize_t | |
3663 | metadata_show(mddev_t *mddev, char *page) | |
3664 | { | |
3665 | if (mddev->persistent) | |
3666 | return sprintf(page, "%d.%d\n", | |
3667 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3668 | else if (mddev->external) |
3669 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3670 | else |
3671 | return sprintf(page, "none\n"); | |
3672 | } | |
3673 | ||
3674 | static ssize_t | |
3675 | metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
3676 | { | |
3677 | int major, minor; | |
3678 | char *e; | |
ea43ddd8 N |
3679 | /* Changing the details of 'external' metadata is |
3680 | * always permitted. Otherwise there must be | |
3681 | * no devices attached to the array. | |
3682 | */ | |
3683 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3684 | ; | |
3685 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3686 | return -EBUSY; |
3687 | ||
3688 | if (cmd_match(buf, "none")) { | |
3689 | mddev->persistent = 0; | |
e691063a N |
3690 | mddev->external = 0; |
3691 | mddev->major_version = 0; | |
3692 | mddev->minor_version = 90; | |
3693 | return len; | |
3694 | } | |
3695 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3696 | size_t namelen = len-9; |
e691063a N |
3697 | if (namelen >= sizeof(mddev->metadata_type)) |
3698 | namelen = sizeof(mddev->metadata_type)-1; | |
3699 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3700 | mddev->metadata_type[namelen] = 0; | |
3701 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3702 | mddev->metadata_type[--namelen] = 0; | |
3703 | mddev->persistent = 0; | |
3704 | mddev->external = 1; | |
8bb93aac N |
3705 | mddev->major_version = 0; |
3706 | mddev->minor_version = 90; | |
3707 | return len; | |
3708 | } | |
3709 | major = simple_strtoul(buf, &e, 10); | |
3710 | if (e==buf || *e != '.') | |
3711 | return -EINVAL; | |
3712 | buf = e+1; | |
3713 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3714 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3715 | return -EINVAL; |
50511da3 | 3716 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3717 | return -ENOENT; |
3718 | mddev->major_version = major; | |
3719 | mddev->minor_version = minor; | |
3720 | mddev->persistent = 1; | |
e691063a | 3721 | mddev->external = 0; |
8bb93aac N |
3722 | return len; |
3723 | } | |
3724 | ||
3725 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3726 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3727 | |
24dd469d | 3728 | static ssize_t |
7eec314d | 3729 | action_show(mddev_t *mddev, char *page) |
24dd469d | 3730 | { |
7eec314d | 3731 | char *type = "idle"; |
b6a9ce68 N |
3732 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3733 | type = "frozen"; | |
3734 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 3735 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
3736 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3737 | type = "reshape"; | |
3738 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
3739 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
3740 | type = "resync"; | |
3741 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
3742 | type = "check"; | |
3743 | else | |
3744 | type = "repair"; | |
72a23c21 | 3745 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
3746 | type = "recover"; |
3747 | } | |
3748 | return sprintf(page, "%s\n", type); | |
3749 | } | |
3750 | ||
7ebc0be7 N |
3751 | static void reap_sync_thread(mddev_t *mddev); |
3752 | ||
24dd469d | 3753 | static ssize_t |
7eec314d | 3754 | action_store(mddev_t *mddev, const char *page, size_t len) |
24dd469d | 3755 | { |
7eec314d N |
3756 | if (!mddev->pers || !mddev->pers->sync_request) |
3757 | return -EINVAL; | |
3758 | ||
b6a9ce68 N |
3759 | if (cmd_match(page, "frozen")) |
3760 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3761 | else | |
3762 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
3763 | ||
3764 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
3765 | if (mddev->sync_thread) { |
3766 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 3767 | reap_sync_thread(mddev); |
7eec314d | 3768 | } |
03c902e1 N |
3769 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
3770 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 3771 | return -EBUSY; |
72a23c21 NB |
3772 | else if (cmd_match(page, "resync")) |
3773 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3774 | else if (cmd_match(page, "recover")) { | |
3775 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 3776 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 3777 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
3778 | int err; |
3779 | if (mddev->pers->start_reshape == NULL) | |
3780 | return -EINVAL; | |
3781 | err = mddev->pers->start_reshape(mddev); | |
3782 | if (err) | |
3783 | return err; | |
a99ac971 | 3784 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 3785 | } else { |
bce74dac | 3786 | if (cmd_match(page, "check")) |
7eec314d | 3787 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 3788 | else if (!cmd_match(page, "repair")) |
7eec314d N |
3789 | return -EINVAL; |
3790 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
3791 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 3792 | } |
03c902e1 | 3793 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 3794 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 3795 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
3796 | return len; |
3797 | } | |
3798 | ||
9d88883e | 3799 | static ssize_t |
96de1e66 | 3800 | mismatch_cnt_show(mddev_t *mddev, char *page) |
9d88883e N |
3801 | { |
3802 | return sprintf(page, "%llu\n", | |
3803 | (unsigned long long) mddev->resync_mismatches); | |
3804 | } | |
3805 | ||
80ca3a44 N |
3806 | static struct md_sysfs_entry md_scan_mode = |
3807 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
24dd469d | 3808 | |
96de1e66 | 3809 | |
80ca3a44 | 3810 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 3811 | |
88202a0c N |
3812 | static ssize_t |
3813 | sync_min_show(mddev_t *mddev, char *page) | |
3814 | { | |
3815 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
3816 | mddev->sync_speed_min ? "local": "system"); | |
3817 | } | |
3818 | ||
3819 | static ssize_t | |
3820 | sync_min_store(mddev_t *mddev, const char *buf, size_t len) | |
3821 | { | |
3822 | int min; | |
3823 | char *e; | |
3824 | if (strncmp(buf, "system", 6)==0) { | |
3825 | mddev->sync_speed_min = 0; | |
3826 | return len; | |
3827 | } | |
3828 | min = simple_strtoul(buf, &e, 10); | |
3829 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
3830 | return -EINVAL; | |
3831 | mddev->sync_speed_min = min; | |
3832 | return len; | |
3833 | } | |
3834 | ||
3835 | static struct md_sysfs_entry md_sync_min = | |
3836 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
3837 | ||
3838 | static ssize_t | |
3839 | sync_max_show(mddev_t *mddev, char *page) | |
3840 | { | |
3841 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
3842 | mddev->sync_speed_max ? "local": "system"); | |
3843 | } | |
3844 | ||
3845 | static ssize_t | |
3846 | sync_max_store(mddev_t *mddev, const char *buf, size_t len) | |
3847 | { | |
3848 | int max; | |
3849 | char *e; | |
3850 | if (strncmp(buf, "system", 6)==0) { | |
3851 | mddev->sync_speed_max = 0; | |
3852 | return len; | |
3853 | } | |
3854 | max = simple_strtoul(buf, &e, 10); | |
3855 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
3856 | return -EINVAL; | |
3857 | mddev->sync_speed_max = max; | |
3858 | return len; | |
3859 | } | |
3860 | ||
3861 | static struct md_sysfs_entry md_sync_max = | |
3862 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
3863 | ||
d7f3d291 IP |
3864 | static ssize_t |
3865 | degraded_show(mddev_t *mddev, char *page) | |
3866 | { | |
3867 | return sprintf(page, "%d\n", mddev->degraded); | |
3868 | } | |
3869 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 3870 | |
90b08710 BS |
3871 | static ssize_t |
3872 | sync_force_parallel_show(mddev_t *mddev, char *page) | |
3873 | { | |
3874 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
3875 | } | |
3876 | ||
3877 | static ssize_t | |
3878 | sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len) | |
3879 | { | |
3880 | long n; | |
3881 | ||
3882 | if (strict_strtol(buf, 10, &n)) | |
3883 | return -EINVAL; | |
3884 | ||
3885 | if (n != 0 && n != 1) | |
3886 | return -EINVAL; | |
3887 | ||
3888 | mddev->parallel_resync = n; | |
3889 | ||
3890 | if (mddev->sync_thread) | |
3891 | wake_up(&resync_wait); | |
3892 | ||
3893 | return len; | |
3894 | } | |
3895 | ||
3896 | /* force parallel resync, even with shared block devices */ | |
3897 | static struct md_sysfs_entry md_sync_force_parallel = | |
3898 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
3899 | sync_force_parallel_show, sync_force_parallel_store); | |
3900 | ||
88202a0c N |
3901 | static ssize_t |
3902 | sync_speed_show(mddev_t *mddev, char *page) | |
3903 | { | |
3904 | unsigned long resync, dt, db; | |
d1a7c503 N |
3905 | if (mddev->curr_resync == 0) |
3906 | return sprintf(page, "none\n"); | |
9687a60c AN |
3907 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
3908 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 3909 | if (!dt) dt++; |
9687a60c AN |
3910 | db = resync - mddev->resync_mark_cnt; |
3911 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
3912 | } |
3913 | ||
80ca3a44 | 3914 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
3915 | |
3916 | static ssize_t | |
3917 | sync_completed_show(mddev_t *mddev, char *page) | |
3918 | { | |
13ae864b | 3919 | unsigned long long max_sectors, resync; |
88202a0c | 3920 | |
acb180b0 N |
3921 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3922 | return sprintf(page, "none\n"); | |
3923 | ||
88202a0c | 3924 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
58c0fed4 | 3925 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 3926 | else |
58c0fed4 | 3927 | max_sectors = mddev->dev_sectors; |
88202a0c | 3928 | |
acb180b0 | 3929 | resync = mddev->curr_resync_completed; |
13ae864b | 3930 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
3931 | } |
3932 | ||
80ca3a44 | 3933 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 3934 | |
5e96ee65 NB |
3935 | static ssize_t |
3936 | min_sync_show(mddev_t *mddev, char *page) | |
3937 | { | |
3938 | return sprintf(page, "%llu\n", | |
3939 | (unsigned long long)mddev->resync_min); | |
3940 | } | |
3941 | static ssize_t | |
3942 | min_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3943 | { | |
3944 | unsigned long long min; | |
3945 | if (strict_strtoull(buf, 10, &min)) | |
3946 | return -EINVAL; | |
3947 | if (min > mddev->resync_max) | |
3948 | return -EINVAL; | |
3949 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3950 | return -EBUSY; | |
3951 | ||
3952 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3953 | if (mddev->chunk_sectors) { |
2ac06c33 | 3954 | sector_t temp = min; |
9d8f0363 | 3955 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
3956 | return -EINVAL; |
3957 | } | |
3958 | mddev->resync_min = min; | |
3959 | ||
3960 | return len; | |
3961 | } | |
3962 | ||
3963 | static struct md_sysfs_entry md_min_sync = | |
3964 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
3965 | ||
c6207277 N |
3966 | static ssize_t |
3967 | max_sync_show(mddev_t *mddev, char *page) | |
3968 | { | |
3969 | if (mddev->resync_max == MaxSector) | |
3970 | return sprintf(page, "max\n"); | |
3971 | else | |
3972 | return sprintf(page, "%llu\n", | |
3973 | (unsigned long long)mddev->resync_max); | |
3974 | } | |
3975 | static ssize_t | |
3976 | max_sync_store(mddev_t *mddev, const char *buf, size_t len) | |
3977 | { | |
3978 | if (strncmp(buf, "max", 3) == 0) | |
3979 | mddev->resync_max = MaxSector; | |
3980 | else { | |
5e96ee65 NB |
3981 | unsigned long long max; |
3982 | if (strict_strtoull(buf, 10, &max)) | |
3983 | return -EINVAL; | |
3984 | if (max < mddev->resync_min) | |
c6207277 N |
3985 | return -EINVAL; |
3986 | if (max < mddev->resync_max && | |
4d484a4a | 3987 | mddev->ro == 0 && |
c6207277 N |
3988 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
3989 | return -EBUSY; | |
3990 | ||
3991 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 3992 | if (mddev->chunk_sectors) { |
2ac06c33 | 3993 | sector_t temp = max; |
9d8f0363 | 3994 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
3995 | return -EINVAL; |
3996 | } | |
3997 | mddev->resync_max = max; | |
3998 | } | |
3999 | wake_up(&mddev->recovery_wait); | |
4000 | return len; | |
4001 | } | |
4002 | ||
4003 | static struct md_sysfs_entry md_max_sync = | |
4004 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4005 | ||
e464eafd N |
4006 | static ssize_t |
4007 | suspend_lo_show(mddev_t *mddev, char *page) | |
4008 | { | |
4009 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4010 | } | |
4011 | ||
4012 | static ssize_t | |
4013 | suspend_lo_store(mddev_t *mddev, const char *buf, size_t len) | |
4014 | { | |
4015 | char *e; | |
4016 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4017 | unsigned long long old = mddev->suspend_lo; |
e464eafd | 4018 | |
b8d966ef N |
4019 | if (mddev->pers == NULL || |
4020 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4021 | return -EINVAL; |
4022 | if (buf == e || (*e && *e != '\n')) | |
4023 | return -EINVAL; | |
23ddff37 N |
4024 | |
4025 | mddev->suspend_lo = new; | |
4026 | if (new >= old) | |
4027 | /* Shrinking suspended region */ | |
e464eafd | 4028 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4029 | else { |
4030 | /* Expanding suspended region - need to wait */ | |
4031 | mddev->pers->quiesce(mddev, 1); | |
4032 | mddev->pers->quiesce(mddev, 0); | |
4033 | } | |
4034 | return len; | |
e464eafd N |
4035 | } |
4036 | static struct md_sysfs_entry md_suspend_lo = | |
4037 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4038 | ||
4039 | ||
4040 | static ssize_t | |
4041 | suspend_hi_show(mddev_t *mddev, char *page) | |
4042 | { | |
4043 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4044 | } | |
4045 | ||
4046 | static ssize_t | |
4047 | suspend_hi_store(mddev_t *mddev, const char *buf, size_t len) | |
4048 | { | |
4049 | char *e; | |
4050 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4051 | unsigned long long old = mddev->suspend_hi; |
e464eafd | 4052 | |
b8d966ef N |
4053 | if (mddev->pers == NULL || |
4054 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4055 | return -EINVAL; |
4056 | if (buf == e || (*e && *e != '\n')) | |
4057 | return -EINVAL; | |
23ddff37 N |
4058 | |
4059 | mddev->suspend_hi = new; | |
4060 | if (new <= old) | |
4061 | /* Shrinking suspended region */ | |
4062 | mddev->pers->quiesce(mddev, 2); | |
4063 | else { | |
4064 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4065 | mddev->pers->quiesce(mddev, 1); |
4066 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 N |
4067 | } |
4068 | return len; | |
e464eafd N |
4069 | } |
4070 | static struct md_sysfs_entry md_suspend_hi = | |
4071 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4072 | ||
08a02ecd N |
4073 | static ssize_t |
4074 | reshape_position_show(mddev_t *mddev, char *page) | |
4075 | { | |
4076 | if (mddev->reshape_position != MaxSector) | |
4077 | return sprintf(page, "%llu\n", | |
4078 | (unsigned long long)mddev->reshape_position); | |
4079 | strcpy(page, "none\n"); | |
4080 | return 5; | |
4081 | } | |
4082 | ||
4083 | static ssize_t | |
4084 | reshape_position_store(mddev_t *mddev, const char *buf, size_t len) | |
4085 | { | |
4086 | char *e; | |
4087 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4088 | if (mddev->pers) | |
4089 | return -EBUSY; | |
4090 | if (buf == e || (*e && *e != '\n')) | |
4091 | return -EINVAL; | |
4092 | mddev->reshape_position = new; | |
4093 | mddev->delta_disks = 0; | |
4094 | mddev->new_level = mddev->level; | |
4095 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4096 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
08a02ecd N |
4097 | return len; |
4098 | } | |
4099 | ||
4100 | static struct md_sysfs_entry md_reshape_position = | |
4101 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4102 | reshape_position_store); | |
4103 | ||
b522adcd DW |
4104 | static ssize_t |
4105 | array_size_show(mddev_t *mddev, char *page) | |
4106 | { | |
4107 | if (mddev->external_size) | |
4108 | return sprintf(page, "%llu\n", | |
4109 | (unsigned long long)mddev->array_sectors/2); | |
4110 | else | |
4111 | return sprintf(page, "default\n"); | |
4112 | } | |
4113 | ||
4114 | static ssize_t | |
4115 | array_size_store(mddev_t *mddev, const char *buf, size_t len) | |
4116 | { | |
4117 | sector_t sectors; | |
4118 | ||
4119 | if (strncmp(buf, "default", 7) == 0) { | |
4120 | if (mddev->pers) | |
4121 | sectors = mddev->pers->size(mddev, 0, 0); | |
4122 | else | |
4123 | sectors = mddev->array_sectors; | |
4124 | ||
4125 | mddev->external_size = 0; | |
4126 | } else { | |
4127 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4128 | return -EINVAL; | |
4129 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4130 | return -E2BIG; |
b522adcd DW |
4131 | |
4132 | mddev->external_size = 1; | |
4133 | } | |
4134 | ||
4135 | mddev->array_sectors = sectors; | |
4136 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e N |
4137 | if (mddev->pers) |
4138 | revalidate_disk(mddev->gendisk); | |
b522adcd DW |
4139 | |
4140 | return len; | |
4141 | } | |
4142 | ||
4143 | static struct md_sysfs_entry md_array_size = | |
4144 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4145 | array_size_store); | |
e464eafd | 4146 | |
eae1701f N |
4147 | static struct attribute *md_default_attrs[] = { |
4148 | &md_level.attr, | |
d4dbd025 | 4149 | &md_layout.attr, |
eae1701f | 4150 | &md_raid_disks.attr, |
3b34380a | 4151 | &md_chunk_size.attr, |
a35b0d69 | 4152 | &md_size.attr, |
a94213b1 | 4153 | &md_resync_start.attr, |
8bb93aac | 4154 | &md_metadata.attr, |
6d7ff738 | 4155 | &md_new_device.attr, |
16f17b39 | 4156 | &md_safe_delay.attr, |
9e653b63 | 4157 | &md_array_state.attr, |
08a02ecd | 4158 | &md_reshape_position.attr, |
b522adcd | 4159 | &md_array_size.attr, |
1e50915f | 4160 | &max_corr_read_errors.attr, |
411036fa N |
4161 | NULL, |
4162 | }; | |
4163 | ||
4164 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4165 | &md_scan_mode.attr, |
9d88883e | 4166 | &md_mismatches.attr, |
88202a0c N |
4167 | &md_sync_min.attr, |
4168 | &md_sync_max.attr, | |
4169 | &md_sync_speed.attr, | |
90b08710 | 4170 | &md_sync_force_parallel.attr, |
88202a0c | 4171 | &md_sync_completed.attr, |
5e96ee65 | 4172 | &md_min_sync.attr, |
c6207277 | 4173 | &md_max_sync.attr, |
e464eafd N |
4174 | &md_suspend_lo.attr, |
4175 | &md_suspend_hi.attr, | |
9b1d1dac | 4176 | &md_bitmap.attr, |
d7f3d291 | 4177 | &md_degraded.attr, |
eae1701f N |
4178 | NULL, |
4179 | }; | |
411036fa N |
4180 | static struct attribute_group md_redundancy_group = { |
4181 | .name = NULL, | |
4182 | .attrs = md_redundancy_attrs, | |
4183 | }; | |
4184 | ||
eae1701f N |
4185 | |
4186 | static ssize_t | |
4187 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4188 | { | |
4189 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4190 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4191 | ssize_t rv; |
eae1701f N |
4192 | |
4193 | if (!entry->show) | |
4194 | return -EIO; | |
5dc5cf7d IM |
4195 | rv = mddev_lock(mddev); |
4196 | if (!rv) { | |
4197 | rv = entry->show(mddev, page); | |
4198 | mddev_unlock(mddev); | |
4199 | } | |
96de1e66 | 4200 | return rv; |
eae1701f N |
4201 | } |
4202 | ||
4203 | static ssize_t | |
4204 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4205 | const char *page, size_t length) | |
4206 | { | |
4207 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
4208 | mddev_t *mddev = container_of(kobj, struct mddev_s, kobj); | |
96de1e66 | 4209 | ssize_t rv; |
eae1701f N |
4210 | |
4211 | if (!entry->store) | |
4212 | return -EIO; | |
67463acb N |
4213 | if (!capable(CAP_SYS_ADMIN)) |
4214 | return -EACCES; | |
5dc5cf7d | 4215 | rv = mddev_lock(mddev); |
d3374825 N |
4216 | if (mddev->hold_active == UNTIL_IOCTL) |
4217 | mddev->hold_active = 0; | |
5dc5cf7d IM |
4218 | if (!rv) { |
4219 | rv = entry->store(mddev, page, length); | |
4220 | mddev_unlock(mddev); | |
4221 | } | |
96de1e66 | 4222 | return rv; |
eae1701f N |
4223 | } |
4224 | ||
4225 | static void md_free(struct kobject *ko) | |
4226 | { | |
4227 | mddev_t *mddev = container_of(ko, mddev_t, kobj); | |
a21d1504 N |
4228 | |
4229 | if (mddev->sysfs_state) | |
4230 | sysfs_put(mddev->sysfs_state); | |
4231 | ||
4232 | if (mddev->gendisk) { | |
4233 | del_gendisk(mddev->gendisk); | |
4234 | put_disk(mddev->gendisk); | |
4235 | } | |
4236 | if (mddev->queue) | |
4237 | blk_cleanup_queue(mddev->queue); | |
4238 | ||
eae1701f N |
4239 | kfree(mddev); |
4240 | } | |
4241 | ||
52cf25d0 | 4242 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4243 | .show = md_attr_show, |
4244 | .store = md_attr_store, | |
4245 | }; | |
4246 | static struct kobj_type md_ktype = { | |
4247 | .release = md_free, | |
4248 | .sysfs_ops = &md_sysfs_ops, | |
4249 | .default_attrs = md_default_attrs, | |
4250 | }; | |
4251 | ||
1da177e4 LT |
4252 | int mdp_major = 0; |
4253 | ||
5fd3a17e DW |
4254 | static void mddev_delayed_delete(struct work_struct *ws) |
4255 | { | |
4256 | mddev_t *mddev = container_of(ws, mddev_t, del_work); | |
4257 | ||
43a70507 | 4258 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4259 | kobject_del(&mddev->kobj); |
4260 | kobject_put(&mddev->kobj); | |
4261 | } | |
4262 | ||
efeb53c0 | 4263 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4264 | { |
48c9c27b | 4265 | static DEFINE_MUTEX(disks_mutex); |
1da177e4 LT |
4266 | mddev_t *mddev = mddev_find(dev); |
4267 | struct gendisk *disk; | |
efeb53c0 N |
4268 | int partitioned; |
4269 | int shift; | |
4270 | int unit; | |
3830c62f | 4271 | int error; |
1da177e4 LT |
4272 | |
4273 | if (!mddev) | |
efeb53c0 N |
4274 | return -ENODEV; |
4275 | ||
4276 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4277 | shift = partitioned ? MdpMinorShift : 0; | |
4278 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4279 | |
e804ac78 TH |
4280 | /* wait for any previous instance of this device to be |
4281 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4282 | */ |
e804ac78 | 4283 | flush_workqueue(md_misc_wq); |
d3374825 | 4284 | |
48c9c27b | 4285 | mutex_lock(&disks_mutex); |
0909dc44 N |
4286 | error = -EEXIST; |
4287 | if (mddev->gendisk) | |
4288 | goto abort; | |
efeb53c0 N |
4289 | |
4290 | if (name) { | |
4291 | /* Need to ensure that 'name' is not a duplicate. | |
4292 | */ | |
4293 | mddev_t *mddev2; | |
4294 | spin_lock(&all_mddevs_lock); | |
4295 | ||
4296 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4297 | if (mddev2->gendisk && | |
4298 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4299 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4300 | goto abort; |
efeb53c0 N |
4301 | } |
4302 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4303 | } |
8b765398 | 4304 | |
0909dc44 | 4305 | error = -ENOMEM; |
8b765398 | 4306 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4307 | if (!mddev->queue) |
4308 | goto abort; | |
409c57f3 N |
4309 | mddev->queue->queuedata = mddev; |
4310 | ||
409c57f3 | 4311 | blk_queue_make_request(mddev->queue, md_make_request); |
8b765398 | 4312 | |
1da177e4 LT |
4313 | disk = alloc_disk(1 << shift); |
4314 | if (!disk) { | |
8b765398 N |
4315 | blk_cleanup_queue(mddev->queue); |
4316 | mddev->queue = NULL; | |
0909dc44 | 4317 | goto abort; |
1da177e4 | 4318 | } |
efeb53c0 | 4319 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4320 | disk->first_minor = unit << shift; |
efeb53c0 N |
4321 | if (name) |
4322 | strcpy(disk->disk_name, name); | |
4323 | else if (partitioned) | |
1da177e4 | 4324 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4325 | else |
1da177e4 | 4326 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4327 | disk->fops = &md_fops; |
4328 | disk->private_data = mddev; | |
4329 | disk->queue = mddev->queue; | |
92850bbd | 4330 | /* Allow extended partitions. This makes the |
d3374825 | 4331 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4332 | * remove it now. |
4333 | */ | |
4334 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 LT |
4335 | add_disk(disk); |
4336 | mddev->gendisk = disk; | |
ed9e1982 TH |
4337 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4338 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4339 | if (error) { |
4340 | /* This isn't possible, but as kobject_init_and_add is marked | |
4341 | * __must_check, we must do something with the result | |
4342 | */ | |
5e55e2f5 N |
4343 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4344 | disk->disk_name); | |
0909dc44 N |
4345 | error = 0; |
4346 | } | |
00bcb4ac N |
4347 | if (mddev->kobj.sd && |
4348 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4349 | printk(KERN_DEBUG "pointless warning\n"); |
be20e6c6 DW |
4350 | |
4351 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); | |
0909dc44 N |
4352 | abort: |
4353 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4354 | if (!error && mddev->kobj.sd) { |
3830c62f | 4355 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4356 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4357 | } |
d3374825 | 4358 | mddev_put(mddev); |
0909dc44 | 4359 | return error; |
efeb53c0 N |
4360 | } |
4361 | ||
4362 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4363 | { | |
4364 | md_alloc(dev, NULL); | |
1da177e4 LT |
4365 | return NULL; |
4366 | } | |
4367 | ||
efeb53c0 N |
4368 | static int add_named_array(const char *val, struct kernel_param *kp) |
4369 | { | |
4370 | /* val must be "md_*" where * is not all digits. | |
4371 | * We allocate an array with a large free minor number, and | |
4372 | * set the name to val. val must not already be an active name. | |
4373 | */ | |
4374 | int len = strlen(val); | |
4375 | char buf[DISK_NAME_LEN]; | |
4376 | ||
4377 | while (len && val[len-1] == '\n') | |
4378 | len--; | |
4379 | if (len >= DISK_NAME_LEN) | |
4380 | return -E2BIG; | |
4381 | strlcpy(buf, val, len+1); | |
4382 | if (strncmp(buf, "md_", 3) != 0) | |
4383 | return -EINVAL; | |
4384 | return md_alloc(0, buf); | |
4385 | } | |
4386 | ||
1da177e4 LT |
4387 | static void md_safemode_timeout(unsigned long data) |
4388 | { | |
4389 | mddev_t *mddev = (mddev_t *) data; | |
4390 | ||
0fd62b86 NB |
4391 | if (!atomic_read(&mddev->writes_pending)) { |
4392 | mddev->safemode = 1; | |
4393 | if (mddev->external) | |
00bcb4ac | 4394 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4395 | } |
1da177e4 LT |
4396 | md_wakeup_thread(mddev->thread); |
4397 | } | |
4398 | ||
6ff8d8ec | 4399 | static int start_dirty_degraded; |
1da177e4 | 4400 | |
390ee602 | 4401 | int md_run(mddev_t *mddev) |
1da177e4 | 4402 | { |
2604b703 | 4403 | int err; |
1da177e4 | 4404 | mdk_rdev_t *rdev; |
2604b703 | 4405 | struct mdk_personality *pers; |
1da177e4 | 4406 | |
a757e64c N |
4407 | if (list_empty(&mddev->disks)) |
4408 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4409 | return -EINVAL; |
1da177e4 LT |
4410 | |
4411 | if (mddev->pers) | |
4412 | return -EBUSY; | |
bb4f1e9d N |
4413 | /* Cannot run until previous stop completes properly */ |
4414 | if (mddev->sysfs_active) | |
4415 | return -EBUSY; | |
b6eb127d | 4416 | |
1da177e4 LT |
4417 | /* |
4418 | * Analyze all RAID superblock(s) | |
4419 | */ | |
1ec4a939 N |
4420 | if (!mddev->raid_disks) { |
4421 | if (!mddev->persistent) | |
4422 | return -EINVAL; | |
a757e64c | 4423 | analyze_sbs(mddev); |
1ec4a939 | 4424 | } |
1da177e4 | 4425 | |
d9d166c2 N |
4426 | if (mddev->level != LEVEL_NONE) |
4427 | request_module("md-level-%d", mddev->level); | |
4428 | else if (mddev->clevel[0]) | |
4429 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4430 | |
4431 | /* | |
4432 | * Drop all container device buffers, from now on | |
4433 | * the only valid external interface is through the md | |
4434 | * device. | |
1da177e4 | 4435 | */ |
159ec1fc | 4436 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 | 4437 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4438 | continue; |
4439 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4440 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4441 | |
4442 | /* perform some consistency tests on the device. | |
4443 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4444 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4445 | */ |
a6ff7e08 JB |
4446 | if (rdev->meta_bdev) { |
4447 | /* Nothing to check */; | |
4448 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
4449 | if (mddev->dev_sectors && |
4450 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4451 | > rdev->sb_start) { |
f0d76d70 N |
4452 | printk("md: %s: data overlaps metadata\n", |
4453 | mdname(mddev)); | |
4454 | return -EINVAL; | |
4455 | } | |
4456 | } else { | |
0f420358 | 4457 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4458 | > rdev->data_offset) { |
4459 | printk("md: %s: metadata overlaps data\n", | |
4460 | mdname(mddev)); | |
4461 | return -EINVAL; | |
4462 | } | |
4463 | } | |
00bcb4ac | 4464 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4465 | } |
4466 | ||
a167f663 N |
4467 | if (mddev->bio_set == NULL) |
4468 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev)); | |
4469 | ||
1da177e4 | 4470 | spin_lock(&pers_lock); |
d9d166c2 | 4471 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4472 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4473 | spin_unlock(&pers_lock); |
d9d166c2 N |
4474 | if (mddev->level != LEVEL_NONE) |
4475 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4476 | mddev->level); | |
4477 | else | |
4478 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4479 | mddev->clevel); | |
1da177e4 LT |
4480 | return -EINVAL; |
4481 | } | |
2604b703 | 4482 | mddev->pers = pers; |
1da177e4 | 4483 | spin_unlock(&pers_lock); |
34817e8c N |
4484 | if (mddev->level != pers->level) { |
4485 | mddev->level = pers->level; | |
4486 | mddev->new_level = pers->level; | |
4487 | } | |
d9d166c2 | 4488 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4489 | |
f6705578 | 4490 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4491 | pers->start_reshape == NULL) { |
f6705578 N |
4492 | /* This personality cannot handle reshaping... */ |
4493 | mddev->pers = NULL; | |
4494 | module_put(pers->owner); | |
4495 | return -EINVAL; | |
4496 | } | |
4497 | ||
7dd5e7c3 N |
4498 | if (pers->sync_request) { |
4499 | /* Warn if this is a potentially silly | |
4500 | * configuration. | |
4501 | */ | |
4502 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
4503 | mdk_rdev_t *rdev2; | |
7dd5e7c3 | 4504 | int warned = 0; |
159ec1fc CR |
4505 | |
4506 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4507 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
7dd5e7c3 N |
4508 | if (rdev < rdev2 && |
4509 | rdev->bdev->bd_contains == | |
4510 | rdev2->bdev->bd_contains) { | |
4511 | printk(KERN_WARNING | |
4512 | "%s: WARNING: %s appears to be" | |
4513 | " on the same physical disk as" | |
4514 | " %s.\n", | |
4515 | mdname(mddev), | |
4516 | bdevname(rdev->bdev,b), | |
4517 | bdevname(rdev2->bdev,b2)); | |
4518 | warned = 1; | |
4519 | } | |
4520 | } | |
159ec1fc | 4521 | |
7dd5e7c3 N |
4522 | if (warned) |
4523 | printk(KERN_WARNING | |
4524 | "True protection against single-disk" | |
4525 | " failure might be compromised.\n"); | |
4526 | } | |
4527 | ||
657390d2 | 4528 | mddev->recovery = 0; |
58c0fed4 AN |
4529 | /* may be over-ridden by personality */ |
4530 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4531 | ||
6ff8d8ec | 4532 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4533 | |
0f9552b5 | 4534 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
4535 | mddev->ro = 2; /* read-only, but switch on first write */ |
4536 | ||
b15c2e57 | 4537 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4538 | if (err) |
4539 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4540 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4541 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4542 | " but 'external_size' not in effect?\n", __func__); | |
4543 | printk(KERN_ERR | |
4544 | "md: invalid array_size %llu > default size %llu\n", | |
4545 | (unsigned long long)mddev->array_sectors / 2, | |
4546 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4547 | err = -EINVAL; | |
4548 | mddev->pers->stop(mddev); | |
4549 | } | |
4550 | if (err == 0 && mddev->pers->sync_request) { | |
b15c2e57 N |
4551 | err = bitmap_create(mddev); |
4552 | if (err) { | |
4553 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4554 | mdname(mddev), err); | |
4555 | mddev->pers->stop(mddev); | |
4556 | } | |
4557 | } | |
1da177e4 | 4558 | if (err) { |
1da177e4 LT |
4559 | module_put(mddev->pers->owner); |
4560 | mddev->pers = NULL; | |
32a7627c N |
4561 | bitmap_destroy(mddev); |
4562 | return err; | |
1da177e4 | 4563 | } |
5e55e2f5 | 4564 | if (mddev->pers->sync_request) { |
00bcb4ac N |
4565 | if (mddev->kobj.sd && |
4566 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
4567 | printk(KERN_WARNING |
4568 | "md: cannot register extra attributes for %s\n", | |
4569 | mdname(mddev)); | |
00bcb4ac | 4570 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4571 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4572 | mddev->ro = 0; |
4573 | ||
1da177e4 | 4574 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
4575 | atomic_set(&mddev->max_corr_read_errors, |
4576 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
4577 | mddev->safemode = 0; |
4578 | mddev->safemode_timer.function = md_safemode_timeout; | |
4579 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4580 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4581 | mddev->in_sync = 1; |
0ca69886 N |
4582 | smp_wmb(); |
4583 | mddev->ready = 1; | |
159ec1fc | 4584 | list_for_each_entry(rdev, &mddev->disks, same_set) |
86e6ffdd N |
4585 | if (rdev->raid_disk >= 0) { |
4586 | char nm[20]; | |
4587 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 | 4588 | if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm)) |
00bcb4ac | 4589 | /* failure here is OK */; |
86e6ffdd | 4590 | } |
1da177e4 LT |
4591 | |
4592 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4593 | ||
850b2b42 N |
4594 | if (mddev->flags) |
4595 | md_update_sb(mddev, 0); | |
1da177e4 | 4596 | |
0b8c9de0 N |
4597 | md_wakeup_thread(mddev->thread); |
4598 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5fd6c1dc | 4599 | |
d7603b7e | 4600 | md_new_event(mddev); |
00bcb4ac N |
4601 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
4602 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 4603 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
4604 | return 0; |
4605 | } | |
390ee602 | 4606 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 4607 | |
fe60b014 N |
4608 | static int do_md_run(mddev_t *mddev) |
4609 | { | |
4610 | int err; | |
4611 | ||
4612 | err = md_run(mddev); | |
4613 | if (err) | |
4614 | goto out; | |
69e51b44 N |
4615 | err = bitmap_load(mddev); |
4616 | if (err) { | |
4617 | bitmap_destroy(mddev); | |
4618 | goto out; | |
4619 | } | |
fe60b014 N |
4620 | set_capacity(mddev->gendisk, mddev->array_sectors); |
4621 | revalidate_disk(mddev->gendisk); | |
4622 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); | |
4623 | out: | |
4624 | return err; | |
4625 | } | |
4626 | ||
1da177e4 LT |
4627 | static int restart_array(mddev_t *mddev) |
4628 | { | |
4629 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4630 | |
80fab1d7 | 4631 | /* Complain if it has no devices */ |
1da177e4 | 4632 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4633 | return -ENXIO; |
4634 | if (!mddev->pers) | |
4635 | return -EINVAL; | |
4636 | if (!mddev->ro) | |
4637 | return -EBUSY; | |
4638 | mddev->safemode = 0; | |
4639 | mddev->ro = 0; | |
4640 | set_disk_ro(disk, 0); | |
4641 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4642 | mdname(mddev)); | |
4643 | /* Kick recovery or resync if necessary */ | |
4644 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4645 | md_wakeup_thread(mddev->thread); | |
4646 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 4647 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 4648 | return 0; |
1da177e4 LT |
4649 | } |
4650 | ||
acc55e22 N |
4651 | /* similar to deny_write_access, but accounts for our holding a reference |
4652 | * to the file ourselves */ | |
4653 | static int deny_bitmap_write_access(struct file * file) | |
4654 | { | |
4655 | struct inode *inode = file->f_mapping->host; | |
4656 | ||
4657 | spin_lock(&inode->i_lock); | |
4658 | if (atomic_read(&inode->i_writecount) > 1) { | |
4659 | spin_unlock(&inode->i_lock); | |
4660 | return -ETXTBSY; | |
4661 | } | |
4662 | atomic_set(&inode->i_writecount, -1); | |
4663 | spin_unlock(&inode->i_lock); | |
4664 | ||
4665 | return 0; | |
4666 | } | |
4667 | ||
43a70507 | 4668 | void restore_bitmap_write_access(struct file *file) |
acc55e22 N |
4669 | { |
4670 | struct inode *inode = file->f_mapping->host; | |
4671 | ||
4672 | spin_lock(&inode->i_lock); | |
4673 | atomic_set(&inode->i_writecount, 1); | |
4674 | spin_unlock(&inode->i_lock); | |
4675 | } | |
4676 | ||
6177b472 N |
4677 | static void md_clean(mddev_t *mddev) |
4678 | { | |
4679 | mddev->array_sectors = 0; | |
4680 | mddev->external_size = 0; | |
4681 | mddev->dev_sectors = 0; | |
4682 | mddev->raid_disks = 0; | |
4683 | mddev->recovery_cp = 0; | |
4684 | mddev->resync_min = 0; | |
4685 | mddev->resync_max = MaxSector; | |
4686 | mddev->reshape_position = MaxSector; | |
4687 | mddev->external = 0; | |
4688 | mddev->persistent = 0; | |
4689 | mddev->level = LEVEL_NONE; | |
4690 | mddev->clevel[0] = 0; | |
4691 | mddev->flags = 0; | |
4692 | mddev->ro = 0; | |
4693 | mddev->metadata_type[0] = 0; | |
4694 | mddev->chunk_sectors = 0; | |
4695 | mddev->ctime = mddev->utime = 0; | |
4696 | mddev->layout = 0; | |
4697 | mddev->max_disks = 0; | |
4698 | mddev->events = 0; | |
a8707c08 | 4699 | mddev->can_decrease_events = 0; |
6177b472 N |
4700 | mddev->delta_disks = 0; |
4701 | mddev->new_level = LEVEL_NONE; | |
4702 | mddev->new_layout = 0; | |
4703 | mddev->new_chunk_sectors = 0; | |
4704 | mddev->curr_resync = 0; | |
4705 | mddev->resync_mismatches = 0; | |
4706 | mddev->suspend_lo = mddev->suspend_hi = 0; | |
4707 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
4708 | mddev->recovery = 0; | |
4709 | mddev->in_sync = 0; | |
4710 | mddev->degraded = 0; | |
6177b472 N |
4711 | mddev->safemode = 0; |
4712 | mddev->bitmap_info.offset = 0; | |
4713 | mddev->bitmap_info.default_offset = 0; | |
4714 | mddev->bitmap_info.chunksize = 0; | |
4715 | mddev->bitmap_info.daemon_sleep = 0; | |
4716 | mddev->bitmap_info.max_write_behind = 0; | |
252ac522 | 4717 | mddev->plug = NULL; |
6177b472 N |
4718 | } |
4719 | ||
defad61a | 4720 | static void __md_stop_writes(mddev_t *mddev) |
a047e125 N |
4721 | { |
4722 | if (mddev->sync_thread) { | |
4723 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4724 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7ebc0be7 | 4725 | reap_sync_thread(mddev); |
a047e125 N |
4726 | } |
4727 | ||
4728 | del_timer_sync(&mddev->safemode_timer); | |
4729 | ||
4730 | bitmap_flush(mddev); | |
4731 | md_super_wait(mddev); | |
4732 | ||
4733 | if (!mddev->in_sync || mddev->flags) { | |
4734 | /* mark array as shutdown cleanly */ | |
4735 | mddev->in_sync = 1; | |
4736 | md_update_sb(mddev, 1); | |
4737 | } | |
4738 | } | |
defad61a N |
4739 | |
4740 | void md_stop_writes(mddev_t *mddev) | |
4741 | { | |
4742 | mddev_lock(mddev); | |
4743 | __md_stop_writes(mddev); | |
4744 | mddev_unlock(mddev); | |
4745 | } | |
390ee602 | 4746 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 4747 | |
390ee602 | 4748 | void md_stop(mddev_t *mddev) |
6177b472 | 4749 | { |
0ca69886 | 4750 | mddev->ready = 0; |
6177b472 N |
4751 | mddev->pers->stop(mddev); |
4752 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
4753 | mddev->to_remove = &md_redundancy_group; | |
4754 | module_put(mddev->pers->owner); | |
4755 | mddev->pers = NULL; | |
cca9cf90 | 4756 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 4757 | } |
390ee602 | 4758 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 4759 | |
a4bd82d0 N |
4760 | static int md_set_readonly(mddev_t *mddev, int is_open) |
4761 | { | |
4762 | int err = 0; | |
4763 | mutex_lock(&mddev->open_mutex); | |
4764 | if (atomic_read(&mddev->openers) > is_open) { | |
4765 | printk("md: %s still in use.\n",mdname(mddev)); | |
4766 | err = -EBUSY; | |
4767 | goto out; | |
4768 | } | |
4769 | if (mddev->pers) { | |
defad61a | 4770 | __md_stop_writes(mddev); |
a4bd82d0 N |
4771 | |
4772 | err = -ENXIO; | |
4773 | if (mddev->ro==1) | |
4774 | goto out; | |
4775 | mddev->ro = 1; | |
4776 | set_disk_ro(mddev->gendisk, 1); | |
4777 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 4778 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
a4bd82d0 N |
4779 | err = 0; |
4780 | } | |
4781 | out: | |
4782 | mutex_unlock(&mddev->open_mutex); | |
4783 | return err; | |
4784 | } | |
4785 | ||
9e653b63 N |
4786 | /* mode: |
4787 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
4788 | * 2 - stop but do not disassemble array |
4789 | */ | |
df5b20cf | 4790 | static int do_md_stop(mddev_t * mddev, int mode, int is_open) |
1da177e4 | 4791 | { |
1da177e4 | 4792 | struct gendisk *disk = mddev->gendisk; |
c4647292 | 4793 | mdk_rdev_t *rdev; |
1da177e4 | 4794 | |
c8c00a69 | 4795 | mutex_lock(&mddev->open_mutex); |
bb4f1e9d N |
4796 | if (atomic_read(&mddev->openers) > is_open || |
4797 | mddev->sysfs_active) { | |
df5b20cf | 4798 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 N |
4799 | mutex_unlock(&mddev->open_mutex); |
4800 | return -EBUSY; | |
4801 | } | |
1da177e4 | 4802 | |
6e17b027 | 4803 | if (mddev->pers) { |
a4bd82d0 N |
4804 | if (mddev->ro) |
4805 | set_disk_ro(disk, 0); | |
409c57f3 | 4806 | |
defad61a | 4807 | __md_stop_writes(mddev); |
a4bd82d0 N |
4808 | md_stop(mddev); |
4809 | mddev->queue->merge_bvec_fn = NULL; | |
4810 | mddev->queue->unplug_fn = NULL; | |
4811 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
6177b472 | 4812 | |
a4bd82d0 | 4813 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 4814 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 4815 | |
a4bd82d0 N |
4816 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4817 | if (rdev->raid_disk >= 0) { | |
4818 | char nm[20]; | |
4819 | sprintf(nm, "rd%d", rdev->raid_disk); | |
4820 | sysfs_remove_link(&mddev->kobj, nm); | |
4821 | } | |
c4647292 | 4822 | |
a4bd82d0 | 4823 | set_capacity(disk, 0); |
6e17b027 | 4824 | mutex_unlock(&mddev->open_mutex); |
a4bd82d0 | 4825 | revalidate_disk(disk); |
0d4ca600 | 4826 | |
a4bd82d0 N |
4827 | if (mddev->ro) |
4828 | mddev->ro = 0; | |
6e17b027 N |
4829 | } else |
4830 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
4831 | /* |
4832 | * Free resources if final stop | |
4833 | */ | |
9e653b63 | 4834 | if (mode == 0) { |
1da177e4 LT |
4835 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
4836 | ||
978f946b | 4837 | bitmap_destroy(mddev); |
c3d9714e N |
4838 | if (mddev->bitmap_info.file) { |
4839 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
4840 | fput(mddev->bitmap_info.file); | |
4841 | mddev->bitmap_info.file = NULL; | |
978f946b | 4842 | } |
c3d9714e | 4843 | mddev->bitmap_info.offset = 0; |
978f946b | 4844 | |
1da177e4 LT |
4845 | export_array(mddev); |
4846 | ||
6177b472 | 4847 | md_clean(mddev); |
934d9c23 | 4848 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
4849 | if (mddev->hold_active == UNTIL_STOP) |
4850 | mddev->hold_active = 0; | |
a4bd82d0 | 4851 | } |
3f9d99c1 | 4852 | blk_integrity_unregister(disk); |
d7603b7e | 4853 | md_new_event(mddev); |
00bcb4ac | 4854 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 4855 | return 0; |
1da177e4 LT |
4856 | } |
4857 | ||
fdee8ae4 | 4858 | #ifndef MODULE |
1da177e4 LT |
4859 | static void autorun_array(mddev_t *mddev) |
4860 | { | |
4861 | mdk_rdev_t *rdev; | |
1da177e4 LT |
4862 | int err; |
4863 | ||
a757e64c | 4864 | if (list_empty(&mddev->disks)) |
1da177e4 | 4865 | return; |
1da177e4 LT |
4866 | |
4867 | printk(KERN_INFO "md: running: "); | |
4868 | ||
159ec1fc | 4869 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
4870 | char b[BDEVNAME_SIZE]; |
4871 | printk("<%s>", bdevname(rdev->bdev,b)); | |
4872 | } | |
4873 | printk("\n"); | |
4874 | ||
d710e138 | 4875 | err = do_md_run(mddev); |
1da177e4 LT |
4876 | if (err) { |
4877 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
d710e138 | 4878 | do_md_stop(mddev, 0, 0); |
1da177e4 LT |
4879 | } |
4880 | } | |
4881 | ||
4882 | /* | |
4883 | * lets try to run arrays based on all disks that have arrived | |
4884 | * until now. (those are in pending_raid_disks) | |
4885 | * | |
4886 | * the method: pick the first pending disk, collect all disks with | |
4887 | * the same UUID, remove all from the pending list and put them into | |
4888 | * the 'same_array' list. Then order this list based on superblock | |
4889 | * update time (freshest comes first), kick out 'old' disks and | |
4890 | * compare superblocks. If everything's fine then run it. | |
4891 | * | |
4892 | * If "unit" is allocated, then bump its reference count | |
4893 | */ | |
4894 | static void autorun_devices(int part) | |
4895 | { | |
159ec1fc | 4896 | mdk_rdev_t *rdev0, *rdev, *tmp; |
1da177e4 LT |
4897 | mddev_t *mddev; |
4898 | char b[BDEVNAME_SIZE]; | |
4899 | ||
4900 | printk(KERN_INFO "md: autorun ...\n"); | |
4901 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 4902 | int unit; |
1da177e4 | 4903 | dev_t dev; |
ad01c9e3 | 4904 | LIST_HEAD(candidates); |
1da177e4 LT |
4905 | rdev0 = list_entry(pending_raid_disks.next, |
4906 | mdk_rdev_t, same_set); | |
4907 | ||
4908 | printk(KERN_INFO "md: considering %s ...\n", | |
4909 | bdevname(rdev0->bdev,b)); | |
4910 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 4911 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
4912 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
4913 | printk(KERN_INFO "md: adding %s ...\n", | |
4914 | bdevname(rdev->bdev,b)); | |
4915 | list_move(&rdev->same_set, &candidates); | |
4916 | } | |
4917 | /* | |
4918 | * now we have a set of devices, with all of them having | |
4919 | * mostly sane superblocks. It's time to allocate the | |
4920 | * mddev. | |
4921 | */ | |
e8703fe1 N |
4922 | if (part) { |
4923 | dev = MKDEV(mdp_major, | |
4924 | rdev0->preferred_minor << MdpMinorShift); | |
4925 | unit = MINOR(dev) >> MdpMinorShift; | |
4926 | } else { | |
4927 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
4928 | unit = MINOR(dev); | |
4929 | } | |
4930 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
4931 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
4932 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
4933 | break; | |
4934 | } | |
1da177e4 LT |
4935 | |
4936 | md_probe(dev, NULL, NULL); | |
4937 | mddev = mddev_find(dev); | |
9bbbca3a NB |
4938 | if (!mddev || !mddev->gendisk) { |
4939 | if (mddev) | |
4940 | mddev_put(mddev); | |
4941 | printk(KERN_ERR | |
1da177e4 LT |
4942 | "md: cannot allocate memory for md drive.\n"); |
4943 | break; | |
4944 | } | |
4945 | if (mddev_lock(mddev)) | |
4946 | printk(KERN_WARNING "md: %s locked, cannot run\n", | |
4947 | mdname(mddev)); | |
4948 | else if (mddev->raid_disks || mddev->major_version | |
4949 | || !list_empty(&mddev->disks)) { | |
4950 | printk(KERN_WARNING | |
4951 | "md: %s already running, cannot run %s\n", | |
4952 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
4953 | mddev_unlock(mddev); | |
4954 | } else { | |
4955 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 4956 | mddev->persistent = 1; |
159ec1fc | 4957 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
4958 | list_del_init(&rdev->same_set); |
4959 | if (bind_rdev_to_array(rdev, mddev)) | |
4960 | export_rdev(rdev); | |
4961 | } | |
4962 | autorun_array(mddev); | |
4963 | mddev_unlock(mddev); | |
4964 | } | |
4965 | /* on success, candidates will be empty, on error | |
4966 | * it won't... | |
4967 | */ | |
159ec1fc | 4968 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 4969 | list_del_init(&rdev->same_set); |
1da177e4 | 4970 | export_rdev(rdev); |
4b80991c | 4971 | } |
1da177e4 LT |
4972 | mddev_put(mddev); |
4973 | } | |
4974 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
4975 | } | |
fdee8ae4 | 4976 | #endif /* !MODULE */ |
1da177e4 | 4977 | |
1da177e4 LT |
4978 | static int get_version(void __user * arg) |
4979 | { | |
4980 | mdu_version_t ver; | |
4981 | ||
4982 | ver.major = MD_MAJOR_VERSION; | |
4983 | ver.minor = MD_MINOR_VERSION; | |
4984 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
4985 | ||
4986 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
4987 | return -EFAULT; | |
4988 | ||
4989 | return 0; | |
4990 | } | |
4991 | ||
4992 | static int get_array_info(mddev_t * mddev, void __user * arg) | |
4993 | { | |
4994 | mdu_array_info_t info; | |
a9f326eb | 4995 | int nr,working,insync,failed,spare; |
1da177e4 | 4996 | mdk_rdev_t *rdev; |
1da177e4 | 4997 | |
a9f326eb | 4998 | nr=working=insync=failed=spare=0; |
159ec1fc | 4999 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 5000 | nr++; |
b2d444d7 | 5001 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5002 | failed++; |
5003 | else { | |
5004 | working++; | |
b2d444d7 | 5005 | if (test_bit(In_sync, &rdev->flags)) |
a9f326eb | 5006 | insync++; |
1da177e4 LT |
5007 | else |
5008 | spare++; | |
5009 | } | |
5010 | } | |
5011 | ||
5012 | info.major_version = mddev->major_version; | |
5013 | info.minor_version = mddev->minor_version; | |
5014 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5015 | info.ctime = mddev->ctime; | |
5016 | info.level = mddev->level; | |
58c0fed4 AN |
5017 | info.size = mddev->dev_sectors / 2; |
5018 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5019 | info.size = -1; |
1da177e4 LT |
5020 | info.nr_disks = nr; |
5021 | info.raid_disks = mddev->raid_disks; | |
5022 | info.md_minor = mddev->md_minor; | |
5023 | info.not_persistent= !mddev->persistent; | |
5024 | ||
5025 | info.utime = mddev->utime; | |
5026 | info.state = 0; | |
5027 | if (mddev->in_sync) | |
5028 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5029 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5030 | info.state = (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5031 | info.active_disks = insync; |
1da177e4 LT |
5032 | info.working_disks = working; |
5033 | info.failed_disks = failed; | |
5034 | info.spare_disks = spare; | |
5035 | ||
5036 | info.layout = mddev->layout; | |
9d8f0363 | 5037 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5038 | |
5039 | if (copy_to_user(arg, &info, sizeof(info))) | |
5040 | return -EFAULT; | |
5041 | ||
5042 | return 0; | |
5043 | } | |
5044 | ||
87162a28 | 5045 | static int get_bitmap_file(mddev_t * mddev, void __user * arg) |
32a7627c N |
5046 | { |
5047 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5048 | char *ptr, *buf = NULL; | |
5049 | int err = -ENOMEM; | |
5050 | ||
b5470dc5 DW |
5051 | if (md_allow_write(mddev)) |
5052 | file = kmalloc(sizeof(*file), GFP_NOIO); | |
5053 | else | |
5054 | file = kmalloc(sizeof(*file), GFP_KERNEL); | |
2a2275d6 | 5055 | |
32a7627c N |
5056 | if (!file) |
5057 | goto out; | |
5058 | ||
5059 | /* bitmap disabled, zero the first byte and copy out */ | |
5060 | if (!mddev->bitmap || !mddev->bitmap->file) { | |
5061 | file->pathname[0] = '\0'; | |
5062 | goto copy_out; | |
5063 | } | |
5064 | ||
5065 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5066 | if (!buf) | |
5067 | goto out; | |
5068 | ||
6bcfd601 CH |
5069 | ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname)); |
5070 | if (IS_ERR(ptr)) | |
32a7627c N |
5071 | goto out; |
5072 | ||
5073 | strcpy(file->pathname, ptr); | |
5074 | ||
5075 | copy_out: | |
5076 | err = 0; | |
5077 | if (copy_to_user(arg, file, sizeof(*file))) | |
5078 | err = -EFAULT; | |
5079 | out: | |
5080 | kfree(buf); | |
5081 | kfree(file); | |
5082 | return err; | |
5083 | } | |
5084 | ||
1da177e4 LT |
5085 | static int get_disk_info(mddev_t * mddev, void __user * arg) |
5086 | { | |
5087 | mdu_disk_info_t info; | |
1da177e4 LT |
5088 | mdk_rdev_t *rdev; |
5089 | ||
5090 | if (copy_from_user(&info, arg, sizeof(info))) | |
5091 | return -EFAULT; | |
5092 | ||
26ef379f | 5093 | rdev = find_rdev_nr(mddev, info.number); |
1da177e4 LT |
5094 | if (rdev) { |
5095 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5096 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5097 | info.raid_disk = rdev->raid_disk; | |
5098 | info.state = 0; | |
b2d444d7 | 5099 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5100 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5101 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5102 | info.state |= (1<<MD_DISK_ACTIVE); |
5103 | info.state |= (1<<MD_DISK_SYNC); | |
5104 | } | |
8ddf9efe N |
5105 | if (test_bit(WriteMostly, &rdev->flags)) |
5106 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5107 | } else { |
5108 | info.major = info.minor = 0; | |
5109 | info.raid_disk = -1; | |
5110 | info.state = (1<<MD_DISK_REMOVED); | |
5111 | } | |
5112 | ||
5113 | if (copy_to_user(arg, &info, sizeof(info))) | |
5114 | return -EFAULT; | |
5115 | ||
5116 | return 0; | |
5117 | } | |
5118 | ||
5119 | static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) | |
5120 | { | |
5121 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
5122 | mdk_rdev_t *rdev; | |
5123 | dev_t dev = MKDEV(info->major,info->minor); | |
5124 | ||
5125 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5126 | return -EOVERFLOW; | |
5127 | ||
5128 | if (!mddev->raid_disks) { | |
5129 | int err; | |
5130 | /* expecting a device which has a superblock */ | |
5131 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5132 | if (IS_ERR(rdev)) { | |
5133 | printk(KERN_WARNING | |
5134 | "md: md_import_device returned %ld\n", | |
5135 | PTR_ERR(rdev)); | |
5136 | return PTR_ERR(rdev); | |
5137 | } | |
5138 | if (!list_empty(&mddev->disks)) { | |
5139 | mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, | |
5140 | mdk_rdev_t, same_set); | |
a9f326eb | 5141 | err = super_types[mddev->major_version] |
1da177e4 LT |
5142 | .load_super(rdev, rdev0, mddev->minor_version); |
5143 | if (err < 0) { | |
5144 | printk(KERN_WARNING | |
5145 | "md: %s has different UUID to %s\n", | |
5146 | bdevname(rdev->bdev,b), | |
5147 | bdevname(rdev0->bdev,b2)); | |
5148 | export_rdev(rdev); | |
5149 | return -EINVAL; | |
5150 | } | |
5151 | } | |
5152 | err = bind_rdev_to_array(rdev, mddev); | |
5153 | if (err) | |
5154 | export_rdev(rdev); | |
5155 | return err; | |
5156 | } | |
5157 | ||
5158 | /* | |
5159 | * add_new_disk can be used once the array is assembled | |
5160 | * to add "hot spares". They must already have a superblock | |
5161 | * written | |
5162 | */ | |
5163 | if (mddev->pers) { | |
5164 | int err; | |
5165 | if (!mddev->pers->hot_add_disk) { | |
5166 | printk(KERN_WARNING | |
5167 | "%s: personality does not support diskops!\n", | |
5168 | mdname(mddev)); | |
5169 | return -EINVAL; | |
5170 | } | |
7b1e35f6 N |
5171 | if (mddev->persistent) |
5172 | rdev = md_import_device(dev, mddev->major_version, | |
5173 | mddev->minor_version); | |
5174 | else | |
5175 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 LT |
5176 | if (IS_ERR(rdev)) { |
5177 | printk(KERN_WARNING | |
5178 | "md: md_import_device returned %ld\n", | |
5179 | PTR_ERR(rdev)); | |
5180 | return PTR_ERR(rdev); | |
5181 | } | |
1a855a06 | 5182 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5183 | if (!mddev->persistent) { |
5184 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5185 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5186 | rdev->raid_disk = info->raid_disk; |
bf572541 N |
5187 | set_bit(In_sync, &rdev->flags); |
5188 | } else | |
41158c7e N |
5189 | rdev->raid_disk = -1; |
5190 | } else | |
5191 | super_types[mddev->major_version]. | |
5192 | validate_super(mddev, rdev); | |
1a855a06 N |
5193 | if (test_bit(In_sync, &rdev->flags)) |
5194 | rdev->saved_raid_disk = rdev->raid_disk; | |
5195 | else | |
5196 | rdev->saved_raid_disk = -1; | |
41158c7e | 5197 | |
b2d444d7 | 5198 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5199 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5200 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5201 | else |
5202 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5203 | |
1da177e4 LT |
5204 | rdev->raid_disk = -1; |
5205 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5206 | if (!err && !mddev->pers->hot_remove_disk) { |
5207 | /* If there is hot_add_disk but no hot_remove_disk | |
5208 | * then added disks for geometry changes, | |
5209 | * and should be added immediately. | |
5210 | */ | |
5211 | super_types[mddev->major_version]. | |
5212 | validate_super(mddev, rdev); | |
5213 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5214 | if (err) | |
5215 | unbind_rdev_from_array(rdev); | |
5216 | } | |
1da177e4 LT |
5217 | if (err) |
5218 | export_rdev(rdev); | |
52664732 | 5219 | else |
00bcb4ac | 5220 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5221 | |
17571284 | 5222 | md_update_sb(mddev, 1); |
72a23c21 NB |
5223 | if (mddev->degraded) |
5224 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5225 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
005eca5e | 5226 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5227 | return err; |
5228 | } | |
5229 | ||
5230 | /* otherwise, add_new_disk is only allowed | |
5231 | * for major_version==0 superblocks | |
5232 | */ | |
5233 | if (mddev->major_version != 0) { | |
5234 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5235 | mdname(mddev)); | |
5236 | return -EINVAL; | |
5237 | } | |
5238 | ||
5239 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5240 | int err; | |
d710e138 | 5241 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5242 | if (IS_ERR(rdev)) { |
5243 | printk(KERN_WARNING | |
5244 | "md: error, md_import_device() returned %ld\n", | |
5245 | PTR_ERR(rdev)); | |
5246 | return PTR_ERR(rdev); | |
5247 | } | |
5248 | rdev->desc_nr = info->number; | |
5249 | if (info->raid_disk < mddev->raid_disks) | |
5250 | rdev->raid_disk = info->raid_disk; | |
5251 | else | |
5252 | rdev->raid_disk = -1; | |
5253 | ||
1da177e4 | 5254 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5255 | if (info->state & (1<<MD_DISK_SYNC)) |
5256 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5257 | |
8ddf9efe N |
5258 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5259 | set_bit(WriteMostly, &rdev->flags); | |
5260 | ||
1da177e4 LT |
5261 | if (!mddev->persistent) { |
5262 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5263 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5264 | } else | |
57b2caa3 | 5265 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5266 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5267 | |
2bf071bf N |
5268 | err = bind_rdev_to_array(rdev, mddev); |
5269 | if (err) { | |
5270 | export_rdev(rdev); | |
5271 | return err; | |
5272 | } | |
1da177e4 LT |
5273 | } |
5274 | ||
5275 | return 0; | |
5276 | } | |
5277 | ||
5278 | static int hot_remove_disk(mddev_t * mddev, dev_t dev) | |
5279 | { | |
5280 | char b[BDEVNAME_SIZE]; | |
5281 | mdk_rdev_t *rdev; | |
5282 | ||
1da177e4 LT |
5283 | rdev = find_rdev(mddev, dev); |
5284 | if (!rdev) | |
5285 | return -ENXIO; | |
5286 | ||
5287 | if (rdev->raid_disk >= 0) | |
5288 | goto busy; | |
5289 | ||
5290 | kick_rdev_from_array(rdev); | |
850b2b42 | 5291 | md_update_sb(mddev, 1); |
d7603b7e | 5292 | md_new_event(mddev); |
1da177e4 LT |
5293 | |
5294 | return 0; | |
5295 | busy: | |
fdefa4d8 | 5296 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5297 | bdevname(rdev->bdev,b), mdname(mddev)); |
5298 | return -EBUSY; | |
5299 | } | |
5300 | ||
5301 | static int hot_add_disk(mddev_t * mddev, dev_t dev) | |
5302 | { | |
5303 | char b[BDEVNAME_SIZE]; | |
5304 | int err; | |
1da177e4 LT |
5305 | mdk_rdev_t *rdev; |
5306 | ||
5307 | if (!mddev->pers) | |
5308 | return -ENODEV; | |
5309 | ||
5310 | if (mddev->major_version != 0) { | |
5311 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5312 | " version-0 superblocks.\n", | |
5313 | mdname(mddev)); | |
5314 | return -EINVAL; | |
5315 | } | |
5316 | if (!mddev->pers->hot_add_disk) { | |
5317 | printk(KERN_WARNING | |
5318 | "%s: personality does not support diskops!\n", | |
5319 | mdname(mddev)); | |
5320 | return -EINVAL; | |
5321 | } | |
5322 | ||
d710e138 | 5323 | rdev = md_import_device(dev, -1, 0); |
1da177e4 LT |
5324 | if (IS_ERR(rdev)) { |
5325 | printk(KERN_WARNING | |
5326 | "md: error, md_import_device() returned %ld\n", | |
5327 | PTR_ERR(rdev)); | |
5328 | return -EINVAL; | |
5329 | } | |
5330 | ||
5331 | if (mddev->persistent) | |
57b2caa3 | 5332 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5333 | else |
77304d2a | 5334 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5335 | |
8190e754 | 5336 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5337 | |
b2d444d7 | 5338 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
5339 | printk(KERN_WARNING |
5340 | "md: can not hot-add faulty %s disk to %s!\n", | |
5341 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5342 | err = -EINVAL; | |
5343 | goto abort_export; | |
5344 | } | |
b2d444d7 | 5345 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5346 | rdev->desc_nr = -1; |
5842730d | 5347 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5348 | err = bind_rdev_to_array(rdev, mddev); |
5349 | if (err) | |
5350 | goto abort_export; | |
1da177e4 LT |
5351 | |
5352 | /* | |
5353 | * The rest should better be atomic, we can have disk failures | |
5354 | * noticed in interrupt contexts ... | |
5355 | */ | |
5356 | ||
1da177e4 LT |
5357 | rdev->raid_disk = -1; |
5358 | ||
850b2b42 | 5359 | md_update_sb(mddev, 1); |
1da177e4 LT |
5360 | |
5361 | /* | |
5362 | * Kick recovery, maybe this spare has to be added to the | |
5363 | * array immediately. | |
5364 | */ | |
5365 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5366 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5367 | md_new_event(mddev); |
1da177e4 LT |
5368 | return 0; |
5369 | ||
1da177e4 LT |
5370 | abort_export: |
5371 | export_rdev(rdev); | |
5372 | return err; | |
5373 | } | |
5374 | ||
32a7627c N |
5375 | static int set_bitmap_file(mddev_t *mddev, int fd) |
5376 | { | |
5377 | int err; | |
5378 | ||
36fa3063 N |
5379 | if (mddev->pers) { |
5380 | if (!mddev->pers->quiesce) | |
5381 | return -EBUSY; | |
5382 | if (mddev->recovery || mddev->sync_thread) | |
5383 | return -EBUSY; | |
5384 | /* we should be able to change the bitmap.. */ | |
5385 | } | |
32a7627c | 5386 | |
32a7627c | 5387 | |
36fa3063 N |
5388 | if (fd >= 0) { |
5389 | if (mddev->bitmap) | |
5390 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5391 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5392 | |
c3d9714e | 5393 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5394 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5395 | mdname(mddev)); | |
5396 | return -EBADF; | |
5397 | } | |
5398 | ||
c3d9714e | 5399 | err = deny_bitmap_write_access(mddev->bitmap_info.file); |
36fa3063 N |
5400 | if (err) { |
5401 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", | |
5402 | mdname(mddev)); | |
c3d9714e N |
5403 | fput(mddev->bitmap_info.file); |
5404 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5405 | return err; |
5406 | } | |
c3d9714e | 5407 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5408 | } else if (mddev->bitmap == NULL) |
5409 | return -ENOENT; /* cannot remove what isn't there */ | |
5410 | err = 0; | |
5411 | if (mddev->pers) { | |
5412 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 5413 | if (fd >= 0) { |
36fa3063 | 5414 | err = bitmap_create(mddev); |
69e51b44 N |
5415 | if (!err) |
5416 | err = bitmap_load(mddev); | |
5417 | } | |
d7375ab3 | 5418 | if (fd < 0 || err) { |
36fa3063 | 5419 | bitmap_destroy(mddev); |
d7375ab3 N |
5420 | fd = -1; /* make sure to put the file */ |
5421 | } | |
36fa3063 | 5422 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5423 | } |
5424 | if (fd < 0) { | |
c3d9714e N |
5425 | if (mddev->bitmap_info.file) { |
5426 | restore_bitmap_write_access(mddev->bitmap_info.file); | |
5427 | fput(mddev->bitmap_info.file); | |
acc55e22 | 5428 | } |
c3d9714e | 5429 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
5430 | } |
5431 | ||
32a7627c N |
5432 | return err; |
5433 | } | |
5434 | ||
1da177e4 LT |
5435 | /* |
5436 | * set_array_info is used two different ways | |
5437 | * The original usage is when creating a new array. | |
5438 | * In this usage, raid_disks is > 0 and it together with | |
5439 | * level, size, not_persistent,layout,chunksize determine the | |
5440 | * shape of the array. | |
5441 | * This will always create an array with a type-0.90.0 superblock. | |
5442 | * The newer usage is when assembling an array. | |
5443 | * In this case raid_disks will be 0, and the major_version field is | |
5444 | * use to determine which style super-blocks are to be found on the devices. | |
5445 | * The minor and patch _version numbers are also kept incase the | |
5446 | * super_block handler wishes to interpret them. | |
5447 | */ | |
5448 | static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) | |
5449 | { | |
5450 | ||
5451 | if (info->raid_disks == 0) { | |
5452 | /* just setting version number for superblock loading */ | |
5453 | if (info->major_version < 0 || | |
50511da3 | 5454 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5455 | super_types[info->major_version].name == NULL) { |
5456 | /* maybe try to auto-load a module? */ | |
5457 | printk(KERN_INFO | |
5458 | "md: superblock version %d not known\n", | |
5459 | info->major_version); | |
5460 | return -EINVAL; | |
5461 | } | |
5462 | mddev->major_version = info->major_version; | |
5463 | mddev->minor_version = info->minor_version; | |
5464 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5465 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
5466 | /* ensure mddev_put doesn't delete this now that there |
5467 | * is some minimal configuration. | |
5468 | */ | |
5469 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
5470 | return 0; |
5471 | } | |
5472 | mddev->major_version = MD_MAJOR_VERSION; | |
5473 | mddev->minor_version = MD_MINOR_VERSION; | |
5474 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5475 | mddev->ctime = get_seconds(); | |
5476 | ||
5477 | mddev->level = info->level; | |
17115e03 | 5478 | mddev->clevel[0] = 0; |
58c0fed4 | 5479 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5480 | mddev->raid_disks = info->raid_disks; |
5481 | /* don't set md_minor, it is determined by which /dev/md* was | |
5482 | * openned | |
5483 | */ | |
5484 | if (info->state & (1<<MD_SB_CLEAN)) | |
5485 | mddev->recovery_cp = MaxSector; | |
5486 | else | |
5487 | mddev->recovery_cp = 0; | |
5488 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5489 | mddev->external = 0; |
1da177e4 LT |
5490 | |
5491 | mddev->layout = info->layout; | |
9d8f0363 | 5492 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5493 | |
5494 | mddev->max_disks = MD_SB_DISKS; | |
5495 | ||
e691063a N |
5496 | if (mddev->persistent) |
5497 | mddev->flags = 0; | |
850b2b42 | 5498 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5499 | |
c3d9714e N |
5500 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
5501 | mddev->bitmap_info.offset = 0; | |
b2a2703c | 5502 | |
f6705578 N |
5503 | mddev->reshape_position = MaxSector; |
5504 | ||
1da177e4 LT |
5505 | /* |
5506 | * Generate a 128 bit UUID | |
5507 | */ | |
5508 | get_random_bytes(mddev->uuid, 16); | |
5509 | ||
f6705578 | 5510 | mddev->new_level = mddev->level; |
664e7c41 | 5511 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5512 | mddev->new_layout = mddev->layout; |
5513 | mddev->delta_disks = 0; | |
5514 | ||
1da177e4 LT |
5515 | return 0; |
5516 | } | |
5517 | ||
1f403624 DW |
5518 | void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors) |
5519 | { | |
b522adcd DW |
5520 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5521 | ||
5522 | if (mddev->external_size) | |
5523 | return; | |
5524 | ||
1f403624 DW |
5525 | mddev->array_sectors = array_sectors; |
5526 | } | |
5527 | EXPORT_SYMBOL(md_set_array_sectors); | |
5528 | ||
d71f9f88 | 5529 | static int update_size(mddev_t *mddev, sector_t num_sectors) |
a35b0d69 | 5530 | { |
159ec1fc | 5531 | mdk_rdev_t *rdev; |
a35b0d69 | 5532 | int rv; |
d71f9f88 | 5533 | int fit = (num_sectors == 0); |
a35b0d69 N |
5534 | |
5535 | if (mddev->pers->resize == NULL) | |
5536 | return -EINVAL; | |
d71f9f88 AN |
5537 | /* The "num_sectors" is the number of sectors of each device that |
5538 | * is used. This can only make sense for arrays with redundancy. | |
5539 | * linear and raid0 always use whatever space is available. We can only | |
5540 | * consider changing this number if no resync or reconstruction is | |
5541 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5542 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5543 | * of each device. If num_sectors is zero, we find the largest size |
5544 | * that fits. | |
a35b0d69 N |
5545 | */ |
5546 | if (mddev->sync_thread) | |
5547 | return -EBUSY; | |
dba034ee N |
5548 | if (mddev->bitmap) |
5549 | /* Sorry, cannot grow a bitmap yet, just remove it, | |
5550 | * grow, and re-add. | |
5551 | */ | |
5552 | return -EBUSY; | |
159ec1fc | 5553 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dd8ac336 | 5554 | sector_t avail = rdev->sectors; |
01ab5662 | 5555 | |
d71f9f88 AN |
5556 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5557 | num_sectors = avail; | |
5558 | if (avail < num_sectors) | |
a35b0d69 N |
5559 | return -ENOSPC; |
5560 | } | |
d71f9f88 | 5561 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5562 | if (!rv) |
5563 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5564 | return rv; |
5565 | } | |
5566 | ||
da943b99 N |
5567 | static int update_raid_disks(mddev_t *mddev, int raid_disks) |
5568 | { | |
5569 | int rv; | |
5570 | /* change the number of raid disks */ | |
63c70c4f | 5571 | if (mddev->pers->check_reshape == NULL) |
da943b99 N |
5572 | return -EINVAL; |
5573 | if (raid_disks <= 0 || | |
233fca36 | 5574 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 5575 | return -EINVAL; |
63c70c4f | 5576 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5577 | return -EBUSY; |
63c70c4f N |
5578 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
5579 | ||
5580 | rv = mddev->pers->check_reshape(mddev); | |
de171cb9 N |
5581 | if (rv < 0) |
5582 | mddev->delta_disks = 0; | |
da943b99 N |
5583 | return rv; |
5584 | } | |
5585 | ||
5586 | ||
1da177e4 LT |
5587 | /* |
5588 | * update_array_info is used to change the configuration of an | |
5589 | * on-line array. | |
5590 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
5591 | * fields in the info are checked against the array. | |
5592 | * Any differences that cannot be handled will cause an error. | |
5593 | * Normally, only one change can be managed at a time. | |
5594 | */ | |
5595 | static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) | |
5596 | { | |
5597 | int rv = 0; | |
5598 | int cnt = 0; | |
36fa3063 N |
5599 | int state = 0; |
5600 | ||
5601 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 5602 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 5603 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
5604 | |
5605 | if (mddev->major_version != info->major_version || | |
5606 | mddev->minor_version != info->minor_version || | |
5607 | /* mddev->patch_version != info->patch_version || */ | |
5608 | mddev->ctime != info->ctime || | |
5609 | mddev->level != info->level || | |
5610 | /* mddev->layout != info->layout || */ | |
5611 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 5612 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
5613 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
5614 | ((state^info->state) & 0xfffffe00) | |
5615 | ) | |
1da177e4 LT |
5616 | return -EINVAL; |
5617 | /* Check there is only one change */ | |
58c0fed4 AN |
5618 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
5619 | cnt++; | |
5620 | if (mddev->raid_disks != info->raid_disks) | |
5621 | cnt++; | |
5622 | if (mddev->layout != info->layout) | |
5623 | cnt++; | |
5624 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
5625 | cnt++; | |
5626 | if (cnt == 0) | |
5627 | return 0; | |
5628 | if (cnt > 1) | |
5629 | return -EINVAL; | |
1da177e4 LT |
5630 | |
5631 | if (mddev->layout != info->layout) { | |
5632 | /* Change layout | |
5633 | * we don't need to do anything at the md level, the | |
5634 | * personality will take care of it all. | |
5635 | */ | |
50ac168a | 5636 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 5637 | return -EINVAL; |
597a711b N |
5638 | else { |
5639 | mddev->new_layout = info->layout; | |
50ac168a | 5640 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
5641 | if (rv) |
5642 | mddev->new_layout = mddev->layout; | |
5643 | return rv; | |
5644 | } | |
1da177e4 | 5645 | } |
58c0fed4 | 5646 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 5647 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 5648 | |
da943b99 N |
5649 | if (mddev->raid_disks != info->raid_disks) |
5650 | rv = update_raid_disks(mddev, info->raid_disks); | |
5651 | ||
36fa3063 N |
5652 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
5653 | if (mddev->pers->quiesce == NULL) | |
5654 | return -EINVAL; | |
5655 | if (mddev->recovery || mddev->sync_thread) | |
5656 | return -EBUSY; | |
5657 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
5658 | /* add the bitmap */ | |
5659 | if (mddev->bitmap) | |
5660 | return -EEXIST; | |
c3d9714e | 5661 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 5662 | return -EINVAL; |
c3d9714e N |
5663 | mddev->bitmap_info.offset = |
5664 | mddev->bitmap_info.default_offset; | |
36fa3063 N |
5665 | mddev->pers->quiesce(mddev, 1); |
5666 | rv = bitmap_create(mddev); | |
69e51b44 N |
5667 | if (!rv) |
5668 | rv = bitmap_load(mddev); | |
36fa3063 N |
5669 | if (rv) |
5670 | bitmap_destroy(mddev); | |
5671 | mddev->pers->quiesce(mddev, 0); | |
5672 | } else { | |
5673 | /* remove the bitmap */ | |
5674 | if (!mddev->bitmap) | |
5675 | return -ENOENT; | |
5676 | if (mddev->bitmap->file) | |
5677 | return -EINVAL; | |
5678 | mddev->pers->quiesce(mddev, 1); | |
5679 | bitmap_destroy(mddev); | |
5680 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 5681 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
5682 | } |
5683 | } | |
850b2b42 | 5684 | md_update_sb(mddev, 1); |
1da177e4 LT |
5685 | return rv; |
5686 | } | |
5687 | ||
5688 | static int set_disk_faulty(mddev_t *mddev, dev_t dev) | |
5689 | { | |
5690 | mdk_rdev_t *rdev; | |
5691 | ||
5692 | if (mddev->pers == NULL) | |
5693 | return -ENODEV; | |
5694 | ||
5695 | rdev = find_rdev(mddev, dev); | |
5696 | if (!rdev) | |
5697 | return -ENODEV; | |
5698 | ||
5699 | md_error(mddev, rdev); | |
5700 | return 0; | |
5701 | } | |
5702 | ||
2f9618ce AN |
5703 | /* |
5704 | * We have a problem here : there is no easy way to give a CHS | |
5705 | * virtual geometry. We currently pretend that we have a 2 heads | |
5706 | * 4 sectors (with a BIG number of cylinders...). This drives | |
5707 | * dosfs just mad... ;-) | |
5708 | */ | |
a885c8c4 CH |
5709 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
5710 | { | |
5711 | mddev_t *mddev = bdev->bd_disk->private_data; | |
5712 | ||
5713 | geo->heads = 2; | |
5714 | geo->sectors = 4; | |
49ce6cea | 5715 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
5716 | return 0; |
5717 | } | |
5718 | ||
a39907fa | 5719 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
5720 | unsigned int cmd, unsigned long arg) |
5721 | { | |
5722 | int err = 0; | |
5723 | void __user *argp = (void __user *)arg; | |
1da177e4 | 5724 | mddev_t *mddev = NULL; |
e2218350 | 5725 | int ro; |
1da177e4 LT |
5726 | |
5727 | if (!capable(CAP_SYS_ADMIN)) | |
5728 | return -EACCES; | |
5729 | ||
5730 | /* | |
5731 | * Commands dealing with the RAID driver but not any | |
5732 | * particular array: | |
5733 | */ | |
5734 | switch (cmd) | |
5735 | { | |
5736 | case RAID_VERSION: | |
5737 | err = get_version(argp); | |
5738 | goto done; | |
5739 | ||
5740 | case PRINT_RAID_DEBUG: | |
5741 | err = 0; | |
5742 | md_print_devices(); | |
5743 | goto done; | |
5744 | ||
5745 | #ifndef MODULE | |
5746 | case RAID_AUTORUN: | |
5747 | err = 0; | |
5748 | autostart_arrays(arg); | |
5749 | goto done; | |
5750 | #endif | |
5751 | default:; | |
5752 | } | |
5753 | ||
5754 | /* | |
5755 | * Commands creating/starting a new array: | |
5756 | */ | |
5757 | ||
a39907fa | 5758 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
5759 | |
5760 | if (!mddev) { | |
5761 | BUG(); | |
5762 | goto abort; | |
5763 | } | |
5764 | ||
1da177e4 LT |
5765 | err = mddev_lock(mddev); |
5766 | if (err) { | |
5767 | printk(KERN_INFO | |
5768 | "md: ioctl lock interrupted, reason %d, cmd %d\n", | |
5769 | err, cmd); | |
5770 | goto abort; | |
5771 | } | |
5772 | ||
5773 | switch (cmd) | |
5774 | { | |
5775 | case SET_ARRAY_INFO: | |
5776 | { | |
5777 | mdu_array_info_t info; | |
5778 | if (!arg) | |
5779 | memset(&info, 0, sizeof(info)); | |
5780 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
5781 | err = -EFAULT; | |
5782 | goto abort_unlock; | |
5783 | } | |
5784 | if (mddev->pers) { | |
5785 | err = update_array_info(mddev, &info); | |
5786 | if (err) { | |
5787 | printk(KERN_WARNING "md: couldn't update" | |
5788 | " array info. %d\n", err); | |
5789 | goto abort_unlock; | |
5790 | } | |
5791 | goto done_unlock; | |
5792 | } | |
5793 | if (!list_empty(&mddev->disks)) { | |
5794 | printk(KERN_WARNING | |
5795 | "md: array %s already has disks!\n", | |
5796 | mdname(mddev)); | |
5797 | err = -EBUSY; | |
5798 | goto abort_unlock; | |
5799 | } | |
5800 | if (mddev->raid_disks) { | |
5801 | printk(KERN_WARNING | |
5802 | "md: array %s already initialised!\n", | |
5803 | mdname(mddev)); | |
5804 | err = -EBUSY; | |
5805 | goto abort_unlock; | |
5806 | } | |
5807 | err = set_array_info(mddev, &info); | |
5808 | if (err) { | |
5809 | printk(KERN_WARNING "md: couldn't set" | |
5810 | " array info. %d\n", err); | |
5811 | goto abort_unlock; | |
5812 | } | |
5813 | } | |
5814 | goto done_unlock; | |
5815 | ||
5816 | default:; | |
5817 | } | |
5818 | ||
5819 | /* | |
5820 | * Commands querying/configuring an existing array: | |
5821 | */ | |
32a7627c | 5822 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 5823 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
5824 | if ((!mddev->raid_disks && !mddev->external) |
5825 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
5826 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
5827 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 LT |
5828 | err = -ENODEV; |
5829 | goto abort_unlock; | |
5830 | } | |
5831 | ||
5832 | /* | |
5833 | * Commands even a read-only array can execute: | |
5834 | */ | |
5835 | switch (cmd) | |
5836 | { | |
5837 | case GET_ARRAY_INFO: | |
5838 | err = get_array_info(mddev, argp); | |
5839 | goto done_unlock; | |
5840 | ||
32a7627c | 5841 | case GET_BITMAP_FILE: |
87162a28 | 5842 | err = get_bitmap_file(mddev, argp); |
32a7627c N |
5843 | goto done_unlock; |
5844 | ||
1da177e4 LT |
5845 | case GET_DISK_INFO: |
5846 | err = get_disk_info(mddev, argp); | |
5847 | goto done_unlock; | |
5848 | ||
5849 | case RESTART_ARRAY_RW: | |
5850 | err = restart_array(mddev); | |
5851 | goto done_unlock; | |
5852 | ||
5853 | case STOP_ARRAY: | |
d710e138 | 5854 | err = do_md_stop(mddev, 0, 1); |
1da177e4 LT |
5855 | goto done_unlock; |
5856 | ||
5857 | case STOP_ARRAY_RO: | |
a4bd82d0 | 5858 | err = md_set_readonly(mddev, 1); |
1da177e4 LT |
5859 | goto done_unlock; |
5860 | ||
e2218350 DW |
5861 | case BLKROSET: |
5862 | if (get_user(ro, (int __user *)(arg))) { | |
5863 | err = -EFAULT; | |
5864 | goto done_unlock; | |
5865 | } | |
5866 | err = -EINVAL; | |
5867 | ||
5868 | /* if the bdev is going readonly the value of mddev->ro | |
5869 | * does not matter, no writes are coming | |
5870 | */ | |
5871 | if (ro) | |
5872 | goto done_unlock; | |
5873 | ||
5874 | /* are we are already prepared for writes? */ | |
5875 | if (mddev->ro != 1) | |
5876 | goto done_unlock; | |
5877 | ||
5878 | /* transitioning to readauto need only happen for | |
5879 | * arrays that call md_write_start | |
5880 | */ | |
5881 | if (mddev->pers) { | |
5882 | err = restart_array(mddev); | |
5883 | if (err == 0) { | |
5884 | mddev->ro = 2; | |
5885 | set_disk_ro(mddev->gendisk, 0); | |
5886 | } | |
5887 | } | |
5888 | goto done_unlock; | |
1da177e4 LT |
5889 | } |
5890 | ||
5891 | /* | |
5892 | * The remaining ioctls are changing the state of the | |
f91de92e N |
5893 | * superblock, so we do not allow them on read-only arrays. |
5894 | * However non-MD ioctls (e.g. get-size) will still come through | |
5895 | * here and hit the 'default' below, so only disallow | |
5896 | * 'md' ioctls, and switch to rw mode if started auto-readonly. | |
1da177e4 | 5897 | */ |
bb57fc64 | 5898 | if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) { |
f91de92e N |
5899 | if (mddev->ro == 2) { |
5900 | mddev->ro = 0; | |
00bcb4ac | 5901 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 NB |
5902 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5903 | md_wakeup_thread(mddev->thread); | |
f91de92e N |
5904 | } else { |
5905 | err = -EROFS; | |
5906 | goto abort_unlock; | |
5907 | } | |
1da177e4 LT |
5908 | } |
5909 | ||
5910 | switch (cmd) | |
5911 | { | |
5912 | case ADD_NEW_DISK: | |
5913 | { | |
5914 | mdu_disk_info_t info; | |
5915 | if (copy_from_user(&info, argp, sizeof(info))) | |
5916 | err = -EFAULT; | |
5917 | else | |
5918 | err = add_new_disk(mddev, &info); | |
5919 | goto done_unlock; | |
5920 | } | |
5921 | ||
5922 | case HOT_REMOVE_DISK: | |
5923 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
5924 | goto done_unlock; | |
5925 | ||
5926 | case HOT_ADD_DISK: | |
5927 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
5928 | goto done_unlock; | |
5929 | ||
5930 | case SET_DISK_FAULTY: | |
5931 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
5932 | goto done_unlock; | |
5933 | ||
5934 | case RUN_ARRAY: | |
d710e138 | 5935 | err = do_md_run(mddev); |
1da177e4 LT |
5936 | goto done_unlock; |
5937 | ||
32a7627c N |
5938 | case SET_BITMAP_FILE: |
5939 | err = set_bitmap_file(mddev, (int)arg); | |
5940 | goto done_unlock; | |
5941 | ||
1da177e4 | 5942 | default: |
1da177e4 LT |
5943 | err = -EINVAL; |
5944 | goto abort_unlock; | |
5945 | } | |
5946 | ||
5947 | done_unlock: | |
5948 | abort_unlock: | |
d3374825 N |
5949 | if (mddev->hold_active == UNTIL_IOCTL && |
5950 | err != -EINVAL) | |
5951 | mddev->hold_active = 0; | |
1da177e4 LT |
5952 | mddev_unlock(mddev); |
5953 | ||
5954 | return err; | |
5955 | done: | |
5956 | if (err) | |
5957 | MD_BUG(); | |
5958 | abort: | |
5959 | return err; | |
5960 | } | |
aa98aa31 AB |
5961 | #ifdef CONFIG_COMPAT |
5962 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
5963 | unsigned int cmd, unsigned long arg) | |
5964 | { | |
5965 | switch (cmd) { | |
5966 | case HOT_REMOVE_DISK: | |
5967 | case HOT_ADD_DISK: | |
5968 | case SET_DISK_FAULTY: | |
5969 | case SET_BITMAP_FILE: | |
5970 | /* These take in integer arg, do not convert */ | |
5971 | break; | |
5972 | default: | |
5973 | arg = (unsigned long)compat_ptr(arg); | |
5974 | break; | |
5975 | } | |
5976 | ||
5977 | return md_ioctl(bdev, mode, cmd, arg); | |
5978 | } | |
5979 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 5980 | |
a39907fa | 5981 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
5982 | { |
5983 | /* | |
5984 | * Succeed if we can lock the mddev, which confirms that | |
5985 | * it isn't being stopped right now. | |
5986 | */ | |
d3374825 | 5987 | mddev_t *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
5988 | int err; |
5989 | ||
d3374825 N |
5990 | if (mddev->gendisk != bdev->bd_disk) { |
5991 | /* we are racing with mddev_put which is discarding this | |
5992 | * bd_disk. | |
5993 | */ | |
5994 | mddev_put(mddev); | |
5995 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 5996 | flush_workqueue(md_misc_wq); |
d3374825 N |
5997 | /* Then retry the open from the top */ |
5998 | return -ERESTARTSYS; | |
5999 | } | |
6000 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6001 | ||
c8c00a69 | 6002 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6003 | goto out; |
6004 | ||
6005 | err = 0; | |
f2ea68cf | 6006 | atomic_inc(&mddev->openers); |
c8c00a69 | 6007 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6008 | |
f3b99be1 | 6009 | check_disk_size_change(mddev->gendisk, bdev); |
1da177e4 LT |
6010 | out: |
6011 | return err; | |
6012 | } | |
6013 | ||
a39907fa | 6014 | static int md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6015 | { |
a39907fa | 6016 | mddev_t *mddev = disk->private_data; |
1da177e4 | 6017 | |
52e5f9d1 | 6018 | BUG_ON(!mddev); |
f2ea68cf | 6019 | atomic_dec(&mddev->openers); |
1da177e4 LT |
6020 | mddev_put(mddev); |
6021 | ||
6022 | return 0; | |
6023 | } | |
83d5cde4 | 6024 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6025 | { |
6026 | .owner = THIS_MODULE, | |
a39907fa AV |
6027 | .open = md_open, |
6028 | .release = md_release, | |
b492b852 | 6029 | .ioctl = md_ioctl, |
aa98aa31 AB |
6030 | #ifdef CONFIG_COMPAT |
6031 | .compat_ioctl = md_compat_ioctl, | |
6032 | #endif | |
a885c8c4 | 6033 | .getgeo = md_getgeo, |
1da177e4 LT |
6034 | }; |
6035 | ||
75c96f85 | 6036 | static int md_thread(void * arg) |
1da177e4 LT |
6037 | { |
6038 | mdk_thread_t *thread = arg; | |
6039 | ||
1da177e4 LT |
6040 | /* |
6041 | * md_thread is a 'system-thread', it's priority should be very | |
6042 | * high. We avoid resource deadlocks individually in each | |
6043 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6044 | * the very same RT priority as kswapd, thus we will never get | |
6045 | * into a priority inversion deadlock. | |
6046 | * | |
6047 | * we definitely have to have equal or higher priority than | |
6048 | * bdflush, otherwise bdflush will deadlock if there are too | |
6049 | * many dirty RAID5 blocks. | |
6050 | */ | |
1da177e4 | 6051 | |
6985c43f | 6052 | allow_signal(SIGKILL); |
a6fb0934 | 6053 | while (!kthread_should_stop()) { |
1da177e4 | 6054 | |
93588e22 N |
6055 | /* We need to wait INTERRUPTIBLE so that |
6056 | * we don't add to the load-average. | |
6057 | * That means we need to be sure no signals are | |
6058 | * pending | |
6059 | */ | |
6060 | if (signal_pending(current)) | |
6061 | flush_signals(current); | |
6062 | ||
6063 | wait_event_interruptible_timeout | |
6064 | (thread->wqueue, | |
6065 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6066 | || kthread_should_stop(), | |
6067 | thread->timeout); | |
1da177e4 | 6068 | |
6c987910 N |
6069 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6070 | if (!kthread_should_stop()) | |
589a594b | 6071 | thread->run(thread->mddev); |
1da177e4 | 6072 | } |
a6fb0934 | 6073 | |
1da177e4 LT |
6074 | return 0; |
6075 | } | |
6076 | ||
6077 | void md_wakeup_thread(mdk_thread_t *thread) | |
6078 | { | |
6079 | if (thread) { | |
6080 | dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); | |
6081 | set_bit(THREAD_WAKEUP, &thread->flags); | |
6082 | wake_up(&thread->wqueue); | |
6083 | } | |
6084 | } | |
6085 | ||
6086 | mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, | |
6087 | const char *name) | |
6088 | { | |
6089 | mdk_thread_t *thread; | |
1da177e4 | 6090 | |
9ffae0cf | 6091 | thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL); |
1da177e4 LT |
6092 | if (!thread) |
6093 | return NULL; | |
6094 | ||
1da177e4 LT |
6095 | init_waitqueue_head(&thread->wqueue); |
6096 | ||
1da177e4 LT |
6097 | thread->run = run; |
6098 | thread->mddev = mddev; | |
32a7627c | 6099 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6100 | thread->tsk = kthread_run(md_thread, thread, |
6101 | "%s_%s", | |
6102 | mdname(thread->mddev), | |
6103 | name ?: mddev->pers->name); | |
a6fb0934 | 6104 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6105 | kfree(thread); |
6106 | return NULL; | |
6107 | } | |
1da177e4 LT |
6108 | return thread; |
6109 | } | |
6110 | ||
1da177e4 LT |
6111 | void md_unregister_thread(mdk_thread_t *thread) |
6112 | { | |
e0cf8f04 N |
6113 | if (!thread) |
6114 | return; | |
ba25f9dc | 6115 | dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
a6fb0934 N |
6116 | |
6117 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6118 | kfree(thread); |
6119 | } | |
6120 | ||
6121 | void md_error(mddev_t *mddev, mdk_rdev_t *rdev) | |
6122 | { | |
6123 | if (!mddev) { | |
6124 | MD_BUG(); | |
6125 | return; | |
6126 | } | |
6127 | ||
b2d444d7 | 6128 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6129 | return; |
6bfe0b49 DW |
6130 | |
6131 | if (mddev->external) | |
6132 | set_bit(Blocked, &rdev->flags); | |
32a7627c | 6133 | /* |
1da177e4 LT |
6134 | dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", |
6135 | mdname(mddev), | |
6136 | MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), | |
6137 | __builtin_return_address(0),__builtin_return_address(1), | |
6138 | __builtin_return_address(2),__builtin_return_address(3)); | |
32a7627c | 6139 | */ |
d0a0a5ee AM |
6140 | if (!mddev->pers) |
6141 | return; | |
1da177e4 LT |
6142 | if (!mddev->pers->error_handler) |
6143 | return; | |
6144 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6145 | if (mddev->degraded) |
6146 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6147 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6148 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6149 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6150 | md_wakeup_thread(mddev->thread); | |
768a418d | 6151 | if (mddev->event_work.func) |
e804ac78 | 6152 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6153 | md_new_event_inintr(mddev); |
1da177e4 LT |
6154 | } |
6155 | ||
6156 | /* seq_file implementation /proc/mdstat */ | |
6157 | ||
6158 | static void status_unused(struct seq_file *seq) | |
6159 | { | |
6160 | int i = 0; | |
6161 | mdk_rdev_t *rdev; | |
1da177e4 LT |
6162 | |
6163 | seq_printf(seq, "unused devices: "); | |
6164 | ||
159ec1fc | 6165 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6166 | char b[BDEVNAME_SIZE]; |
6167 | i++; | |
6168 | seq_printf(seq, "%s ", | |
6169 | bdevname(rdev->bdev,b)); | |
6170 | } | |
6171 | if (!i) | |
6172 | seq_printf(seq, "<none>"); | |
6173 | ||
6174 | seq_printf(seq, "\n"); | |
6175 | } | |
6176 | ||
6177 | ||
6178 | static void status_resync(struct seq_file *seq, mddev_t * mddev) | |
6179 | { | |
dd71cf6b N |
6180 | sector_t max_sectors, resync, res; |
6181 | unsigned long dt, db; | |
6182 | sector_t rt; | |
4588b42e N |
6183 | int scale; |
6184 | unsigned int per_milli; | |
1da177e4 | 6185 | |
dd71cf6b | 6186 | resync = mddev->curr_resync - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6187 | |
6188 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
dd71cf6b | 6189 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6190 | else |
dd71cf6b | 6191 | max_sectors = mddev->dev_sectors; |
1da177e4 LT |
6192 | |
6193 | /* | |
6194 | * Should not happen. | |
6195 | */ | |
dd71cf6b | 6196 | if (!max_sectors) { |
1da177e4 LT |
6197 | MD_BUG(); |
6198 | return; | |
6199 | } | |
4588b42e | 6200 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6201 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6202 | * u32, as those are the requirements for sector_div. |
6203 | * Thus 'scale' must be at least 10 | |
6204 | */ | |
6205 | scale = 10; | |
6206 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6207 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6208 | scale++; |
6209 | } | |
6210 | res = (resync>>scale)*1000; | |
dd71cf6b | 6211 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6212 | |
6213 | per_milli = res; | |
1da177e4 | 6214 | { |
4588b42e | 6215 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6216 | seq_printf(seq, "["); |
6217 | for (i = 0; i < x; i++) | |
6218 | seq_printf(seq, "="); | |
6219 | seq_printf(seq, ">"); | |
6220 | for (i = 0; i < y; i++) | |
6221 | seq_printf(seq, "."); | |
6222 | seq_printf(seq, "] "); | |
6223 | } | |
4588b42e | 6224 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6225 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6226 | "reshape" : | |
61df9d91 N |
6227 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6228 | "check" : | |
6229 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6230 | "resync" : "recovery"))), | |
6231 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6232 | (unsigned long long) resync/2, |
6233 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6234 | |
6235 | /* | |
1da177e4 LT |
6236 | * dt: time from mark until now |
6237 | * db: blocks written from mark until now | |
6238 | * rt: remaining time | |
dd71cf6b N |
6239 | * |
6240 | * rt is a sector_t, so could be 32bit or 64bit. | |
6241 | * So we divide before multiply in case it is 32bit and close | |
6242 | * to the limit. | |
6243 | * We scale the divisor (db) by 32 to avoid loosing precision | |
6244 | * near the end of resync when the number of remaining sectors | |
6245 | * is close to 'db'. | |
6246 | * We then divide rt by 32 after multiplying by db to compensate. | |
6247 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6248 | */ |
6249 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6250 | if (!dt) dt++; | |
ff4e8d9a N |
6251 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6252 | - mddev->resync_mark_cnt; | |
1da177e4 | 6253 | |
dd71cf6b N |
6254 | rt = max_sectors - resync; /* number of remaining sectors */ |
6255 | sector_div(rt, db/32+1); | |
6256 | rt *= dt; | |
6257 | rt >>= 5; | |
6258 | ||
6259 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6260 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6261 | |
ff4e8d9a | 6262 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6263 | } |
6264 | ||
6265 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6266 | { | |
6267 | struct list_head *tmp; | |
6268 | loff_t l = *pos; | |
6269 | mddev_t *mddev; | |
6270 | ||
6271 | if (l >= 0x10000) | |
6272 | return NULL; | |
6273 | if (!l--) | |
6274 | /* header */ | |
6275 | return (void*)1; | |
6276 | ||
6277 | spin_lock(&all_mddevs_lock); | |
6278 | list_for_each(tmp,&all_mddevs) | |
6279 | if (!l--) { | |
6280 | mddev = list_entry(tmp, mddev_t, all_mddevs); | |
6281 | mddev_get(mddev); | |
6282 | spin_unlock(&all_mddevs_lock); | |
6283 | return mddev; | |
6284 | } | |
6285 | spin_unlock(&all_mddevs_lock); | |
6286 | if (!l--) | |
6287 | return (void*)2;/* tail */ | |
6288 | return NULL; | |
6289 | } | |
6290 | ||
6291 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6292 | { | |
6293 | struct list_head *tmp; | |
6294 | mddev_t *next_mddev, *mddev = v; | |
6295 | ||
6296 | ++*pos; | |
6297 | if (v == (void*)2) | |
6298 | return NULL; | |
6299 | ||
6300 | spin_lock(&all_mddevs_lock); | |
6301 | if (v == (void*)1) | |
6302 | tmp = all_mddevs.next; | |
6303 | else | |
6304 | tmp = mddev->all_mddevs.next; | |
6305 | if (tmp != &all_mddevs) | |
6306 | next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); | |
6307 | else { | |
6308 | next_mddev = (void*)2; | |
6309 | *pos = 0x10000; | |
6310 | } | |
6311 | spin_unlock(&all_mddevs_lock); | |
6312 | ||
6313 | if (v != (void*)1) | |
6314 | mddev_put(mddev); | |
6315 | return next_mddev; | |
6316 | ||
6317 | } | |
6318 | ||
6319 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6320 | { | |
6321 | mddev_t *mddev = v; | |
6322 | ||
6323 | if (mddev && v != (void*)1 && v != (void*)2) | |
6324 | mddev_put(mddev); | |
6325 | } | |
6326 | ||
d7603b7e N |
6327 | struct mdstat_info { |
6328 | int event; | |
6329 | }; | |
6330 | ||
1da177e4 LT |
6331 | static int md_seq_show(struct seq_file *seq, void *v) |
6332 | { | |
6333 | mddev_t *mddev = v; | |
dd8ac336 | 6334 | sector_t sectors; |
1da177e4 | 6335 | mdk_rdev_t *rdev; |
d7603b7e | 6336 | struct mdstat_info *mi = seq->private; |
32a7627c | 6337 | struct bitmap *bitmap; |
1da177e4 LT |
6338 | |
6339 | if (v == (void*)1) { | |
2604b703 | 6340 | struct mdk_personality *pers; |
1da177e4 LT |
6341 | seq_printf(seq, "Personalities : "); |
6342 | spin_lock(&pers_lock); | |
2604b703 N |
6343 | list_for_each_entry(pers, &pers_list, list) |
6344 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6345 | |
6346 | spin_unlock(&pers_lock); | |
6347 | seq_printf(seq, "\n"); | |
d7603b7e | 6348 | mi->event = atomic_read(&md_event_count); |
1da177e4 LT |
6349 | return 0; |
6350 | } | |
6351 | if (v == (void*)2) { | |
6352 | status_unused(seq); | |
6353 | return 0; | |
6354 | } | |
6355 | ||
5dc5cf7d | 6356 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6357 | return -EINTR; |
5dc5cf7d | 6358 | |
1da177e4 LT |
6359 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
6360 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
6361 | mddev->pers ? "" : "in"); | |
6362 | if (mddev->pers) { | |
f91de92e | 6363 | if (mddev->ro==1) |
1da177e4 | 6364 | seq_printf(seq, " (read-only)"); |
f91de92e | 6365 | if (mddev->ro==2) |
52720ae7 | 6366 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
6367 | seq_printf(seq, " %s", mddev->pers->name); |
6368 | } | |
6369 | ||
dd8ac336 | 6370 | sectors = 0; |
159ec1fc | 6371 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
6372 | char b[BDEVNAME_SIZE]; |
6373 | seq_printf(seq, " %s[%d]", | |
6374 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
6375 | if (test_bit(WriteMostly, &rdev->flags)) |
6376 | seq_printf(seq, "(W)"); | |
b2d444d7 | 6377 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6378 | seq_printf(seq, "(F)"); |
6379 | continue; | |
b325a32e N |
6380 | } else if (rdev->raid_disk < 0) |
6381 | seq_printf(seq, "(S)"); /* spare */ | |
dd8ac336 | 6382 | sectors += rdev->sectors; |
1da177e4 LT |
6383 | } |
6384 | ||
6385 | if (!list_empty(&mddev->disks)) { | |
6386 | if (mddev->pers) | |
6387 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
6388 | (unsigned long long) |
6389 | mddev->array_sectors / 2); | |
1da177e4 LT |
6390 | else |
6391 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 6392 | (unsigned long long)sectors / 2); |
1da177e4 | 6393 | } |
1cd6bf19 N |
6394 | if (mddev->persistent) { |
6395 | if (mddev->major_version != 0 || | |
6396 | mddev->minor_version != 90) { | |
6397 | seq_printf(seq," super %d.%d", | |
6398 | mddev->major_version, | |
6399 | mddev->minor_version); | |
6400 | } | |
e691063a N |
6401 | } else if (mddev->external) |
6402 | seq_printf(seq, " super external:%s", | |
6403 | mddev->metadata_type); | |
6404 | else | |
1cd6bf19 | 6405 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
6406 | |
6407 | if (mddev->pers) { | |
d710e138 | 6408 | mddev->pers->status(seq, mddev); |
1da177e4 | 6409 | seq_printf(seq, "\n "); |
8e1b39d6 N |
6410 | if (mddev->pers->sync_request) { |
6411 | if (mddev->curr_resync > 2) { | |
d710e138 | 6412 | status_resync(seq, mddev); |
8e1b39d6 N |
6413 | seq_printf(seq, "\n "); |
6414 | } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) | |
6415 | seq_printf(seq, "\tresync=DELAYED\n "); | |
6416 | else if (mddev->recovery_cp < MaxSector) | |
6417 | seq_printf(seq, "\tresync=PENDING\n "); | |
6418 | } | |
32a7627c N |
6419 | } else |
6420 | seq_printf(seq, "\n "); | |
6421 | ||
6422 | if ((bitmap = mddev->bitmap)) { | |
32a7627c N |
6423 | unsigned long chunk_kb; |
6424 | unsigned long flags; | |
32a7627c | 6425 | spin_lock_irqsave(&bitmap->lock, flags); |
42a04b50 | 6426 | chunk_kb = mddev->bitmap_info.chunksize >> 10; |
32a7627c N |
6427 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " |
6428 | "%lu%s chunk", | |
6429 | bitmap->pages - bitmap->missing_pages, | |
6430 | bitmap->pages, | |
6431 | (bitmap->pages - bitmap->missing_pages) | |
6432 | << (PAGE_SHIFT - 10), | |
42a04b50 | 6433 | chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize, |
32a7627c | 6434 | chunk_kb ? "KB" : "B"); |
78d742d8 N |
6435 | if (bitmap->file) { |
6436 | seq_printf(seq, ", file: "); | |
c32c2f63 | 6437 | seq_path(seq, &bitmap->file->f_path, " \t\n"); |
32a7627c | 6438 | } |
78d742d8 | 6439 | |
32a7627c N |
6440 | seq_printf(seq, "\n"); |
6441 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1da177e4 LT |
6442 | } |
6443 | ||
6444 | seq_printf(seq, "\n"); | |
6445 | } | |
6446 | mddev_unlock(mddev); | |
6447 | ||
6448 | return 0; | |
6449 | } | |
6450 | ||
110518bc | 6451 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6452 | .start = md_seq_start, |
6453 | .next = md_seq_next, | |
6454 | .stop = md_seq_stop, | |
6455 | .show = md_seq_show, | |
6456 | }; | |
6457 | ||
6458 | static int md_seq_open(struct inode *inode, struct file *file) | |
6459 | { | |
6460 | int error; | |
d7603b7e N |
6461 | struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL); |
6462 | if (mi == NULL) | |
6463 | return -ENOMEM; | |
1da177e4 LT |
6464 | |
6465 | error = seq_open(file, &md_seq_ops); | |
d7603b7e N |
6466 | if (error) |
6467 | kfree(mi); | |
6468 | else { | |
6469 | struct seq_file *p = file->private_data; | |
6470 | p->private = mi; | |
6471 | mi->event = atomic_read(&md_event_count); | |
6472 | } | |
1da177e4 LT |
6473 | return error; |
6474 | } | |
6475 | ||
d7603b7e N |
6476 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6477 | { | |
6478 | struct seq_file *m = filp->private_data; | |
6479 | struct mdstat_info *mi = m->private; | |
6480 | int mask; | |
6481 | ||
6482 | poll_wait(filp, &md_event_waiters, wait); | |
6483 | ||
6484 | /* always allow read */ | |
6485 | mask = POLLIN | POLLRDNORM; | |
6486 | ||
6487 | if (mi->event != atomic_read(&md_event_count)) | |
6488 | mask |= POLLERR | POLLPRI; | |
6489 | return mask; | |
6490 | } | |
6491 | ||
fa027c2a | 6492 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6493 | .owner = THIS_MODULE, |
1da177e4 LT |
6494 | .open = md_seq_open, |
6495 | .read = seq_read, | |
6496 | .llseek = seq_lseek, | |
c3f94b40 | 6497 | .release = seq_release_private, |
d7603b7e | 6498 | .poll = mdstat_poll, |
1da177e4 LT |
6499 | }; |
6500 | ||
2604b703 | 6501 | int register_md_personality(struct mdk_personality *p) |
1da177e4 | 6502 | { |
1da177e4 | 6503 | spin_lock(&pers_lock); |
2604b703 N |
6504 | list_add_tail(&p->list, &pers_list); |
6505 | printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level); | |
1da177e4 LT |
6506 | spin_unlock(&pers_lock); |
6507 | return 0; | |
6508 | } | |
6509 | ||
2604b703 | 6510 | int unregister_md_personality(struct mdk_personality *p) |
1da177e4 | 6511 | { |
2604b703 | 6512 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 6513 | spin_lock(&pers_lock); |
2604b703 | 6514 | list_del_init(&p->list); |
1da177e4 LT |
6515 | spin_unlock(&pers_lock); |
6516 | return 0; | |
6517 | } | |
6518 | ||
eea1bf38 | 6519 | static int is_mddev_idle(mddev_t *mddev, int init) |
1da177e4 LT |
6520 | { |
6521 | mdk_rdev_t * rdev; | |
1da177e4 | 6522 | int idle; |
eea1bf38 | 6523 | int curr_events; |
1da177e4 LT |
6524 | |
6525 | idle = 1; | |
4b80991c N |
6526 | rcu_read_lock(); |
6527 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 6528 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
6529 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
6530 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
6531 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
6532 | /* sync IO will cause sync_io to increase before the disk_stats |
6533 | * as sync_io is counted when a request starts, and | |
6534 | * disk_stats is counted when it completes. | |
6535 | * So resync activity will cause curr_events to be smaller than | |
6536 | * when there was no such activity. | |
6537 | * non-sync IO will cause disk_stat to increase without | |
6538 | * increasing sync_io so curr_events will (eventually) | |
6539 | * be larger than it was before. Once it becomes | |
6540 | * substantially larger, the test below will cause | |
6541 | * the array to appear non-idle, and resync will slow | |
6542 | * down. | |
6543 | * If there is a lot of outstanding resync activity when | |
6544 | * we set last_event to curr_events, then all that activity | |
6545 | * completing might cause the array to appear non-idle | |
6546 | * and resync will be slowed down even though there might | |
6547 | * not have been non-resync activity. This will only | |
6548 | * happen once though. 'last_events' will soon reflect | |
6549 | * the state where there is little or no outstanding | |
6550 | * resync requests, and further resync activity will | |
6551 | * always make curr_events less than last_events. | |
c0e48521 | 6552 | * |
1da177e4 | 6553 | */ |
eea1bf38 | 6554 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
6555 | rdev->last_events = curr_events; |
6556 | idle = 0; | |
6557 | } | |
6558 | } | |
4b80991c | 6559 | rcu_read_unlock(); |
1da177e4 LT |
6560 | return idle; |
6561 | } | |
6562 | ||
6563 | void md_done_sync(mddev_t *mddev, int blocks, int ok) | |
6564 | { | |
6565 | /* another "blocks" (512byte) blocks have been synced */ | |
6566 | atomic_sub(blocks, &mddev->recovery_active); | |
6567 | wake_up(&mddev->recovery_wait); | |
6568 | if (!ok) { | |
dfc70645 | 6569 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6570 | md_wakeup_thread(mddev->thread); |
6571 | // stop recovery, signal do_sync .... | |
6572 | } | |
6573 | } | |
6574 | ||
6575 | ||
06d91a5f N |
6576 | /* md_write_start(mddev, bi) |
6577 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
6578 | * in superblock) before writing, schedule a superblock update |
6579 | * and wait for it to complete. | |
06d91a5f | 6580 | */ |
3d310eb7 | 6581 | void md_write_start(mddev_t *mddev, struct bio *bi) |
1da177e4 | 6582 | { |
0fd62b86 | 6583 | int did_change = 0; |
06d91a5f | 6584 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 6585 | return; |
06d91a5f | 6586 | |
f91de92e N |
6587 | BUG_ON(mddev->ro == 1); |
6588 | if (mddev->ro == 2) { | |
6589 | /* need to switch to read/write */ | |
6590 | mddev->ro = 0; | |
6591 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6592 | md_wakeup_thread(mddev->thread); | |
25156198 | 6593 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 6594 | did_change = 1; |
f91de92e | 6595 | } |
06d91a5f | 6596 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
6597 | if (mddev->safemode == 1) |
6598 | mddev->safemode = 0; | |
06d91a5f | 6599 | if (mddev->in_sync) { |
a9701a30 | 6600 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
6601 | if (mddev->in_sync) { |
6602 | mddev->in_sync = 0; | |
850b2b42 | 6603 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 6604 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 6605 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 6606 | did_change = 1; |
3d310eb7 | 6607 | } |
a9701a30 | 6608 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 6609 | } |
0fd62b86 | 6610 | if (did_change) |
00bcb4ac | 6611 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 6612 | wait_event(mddev->sb_wait, |
09a44cc1 | 6613 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 LT |
6614 | } |
6615 | ||
6616 | void md_write_end(mddev_t *mddev) | |
6617 | { | |
6618 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
6619 | if (mddev->safemode == 2) | |
6620 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 6621 | else if (mddev->safemode_delay) |
1da177e4 LT |
6622 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
6623 | } | |
6624 | } | |
6625 | ||
2a2275d6 N |
6626 | /* md_allow_write(mddev) |
6627 | * Calling this ensures that the array is marked 'active' so that writes | |
6628 | * may proceed without blocking. It is important to call this before | |
6629 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
6630 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
6631 | * |
6632 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
6633 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 6634 | */ |
b5470dc5 | 6635 | int md_allow_write(mddev_t *mddev) |
2a2275d6 N |
6636 | { |
6637 | if (!mddev->pers) | |
b5470dc5 | 6638 | return 0; |
2a2275d6 | 6639 | if (mddev->ro) |
b5470dc5 | 6640 | return 0; |
1a0fd497 | 6641 | if (!mddev->pers->sync_request) |
b5470dc5 | 6642 | return 0; |
2a2275d6 N |
6643 | |
6644 | spin_lock_irq(&mddev->write_lock); | |
6645 | if (mddev->in_sync) { | |
6646 | mddev->in_sync = 0; | |
6647 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 6648 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
6649 | if (mddev->safemode_delay && |
6650 | mddev->safemode == 0) | |
6651 | mddev->safemode = 1; | |
6652 | spin_unlock_irq(&mddev->write_lock); | |
6653 | md_update_sb(mddev, 0); | |
00bcb4ac | 6654 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
6655 | } else |
6656 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 | 6657 | |
070dc6dd | 6658 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
6659 | return -EAGAIN; |
6660 | else | |
6661 | return 0; | |
2a2275d6 N |
6662 | } |
6663 | EXPORT_SYMBOL_GPL(md_allow_write); | |
6664 | ||
b63d7c2e | 6665 | void md_unplug(mddev_t *mddev) |
252ac522 N |
6666 | { |
6667 | if (mddev->queue) | |
6668 | blk_unplug(mddev->queue); | |
6669 | if (mddev->plug) | |
6670 | mddev->plug->unplug_fn(mddev->plug); | |
6671 | } | |
6672 | ||
1da177e4 LT |
6673 | #define SYNC_MARKS 10 |
6674 | #define SYNC_MARK_STEP (3*HZ) | |
29269553 | 6675 | void md_do_sync(mddev_t *mddev) |
1da177e4 LT |
6676 | { |
6677 | mddev_t *mddev2; | |
6678 | unsigned int currspeed = 0, | |
6679 | window; | |
57afd89f | 6680 | sector_t max_sectors,j, io_sectors; |
1da177e4 LT |
6681 | unsigned long mark[SYNC_MARKS]; |
6682 | sector_t mark_cnt[SYNC_MARKS]; | |
6683 | int last_mark,m; | |
6684 | struct list_head *tmp; | |
6685 | sector_t last_check; | |
57afd89f | 6686 | int skipped = 0; |
5fd6c1dc | 6687 | mdk_rdev_t *rdev; |
61df9d91 | 6688 | char *desc; |
1da177e4 LT |
6689 | |
6690 | /* just incase thread restarts... */ | |
6691 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
6692 | return; | |
5fd6c1dc N |
6693 | if (mddev->ro) /* never try to sync a read-only array */ |
6694 | return; | |
1da177e4 | 6695 | |
61df9d91 N |
6696 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
6697 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
6698 | desc = "data-check"; | |
6699 | else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6700 | desc = "requested-resync"; | |
6701 | else | |
6702 | desc = "resync"; | |
6703 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
6704 | desc = "reshape"; | |
6705 | else | |
6706 | desc = "recovery"; | |
6707 | ||
1da177e4 LT |
6708 | /* we overload curr_resync somewhat here. |
6709 | * 0 == not engaged in resync at all | |
6710 | * 2 == checking that there is no conflict with another sync | |
6711 | * 1 == like 2, but have yielded to allow conflicting resync to | |
6712 | * commense | |
6713 | * other == active in resync - this many blocks | |
6714 | * | |
6715 | * Before starting a resync we must have set curr_resync to | |
6716 | * 2, and then checked that every "conflicting" array has curr_resync | |
6717 | * less than ours. When we find one that is the same or higher | |
6718 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
6719 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
6720 | * This will mean we have to start checking from the beginning again. | |
6721 | * | |
6722 | */ | |
6723 | ||
6724 | do { | |
6725 | mddev->curr_resync = 2; | |
6726 | ||
6727 | try_again: | |
404e4b43 | 6728 | if (kthread_should_stop()) |
6985c43f | 6729 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
404e4b43 N |
6730 | |
6731 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
1da177e4 | 6732 | goto skip; |
29ac4aa3 | 6733 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
6734 | if (mddev2 == mddev) |
6735 | continue; | |
90b08710 BS |
6736 | if (!mddev->parallel_resync |
6737 | && mddev2->curr_resync | |
6738 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
6739 | DEFINE_WAIT(wq); |
6740 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
6741 | /* arbitrarily yield */ | |
6742 | mddev->curr_resync = 1; | |
6743 | wake_up(&resync_wait); | |
6744 | } | |
6745 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
6746 | /* no need to wait here, we can wait the next | |
6747 | * time 'round when curr_resync == 2 | |
6748 | */ | |
6749 | continue; | |
9744197c N |
6750 | /* We need to wait 'interruptible' so as not to |
6751 | * contribute to the load average, and not to | |
6752 | * be caught by 'softlockup' | |
6753 | */ | |
6754 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
787453c2 | 6755 | if (!kthread_should_stop() && |
8712e553 | 6756 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
6757 | printk(KERN_INFO "md: delaying %s of %s" |
6758 | " until %s has finished (they" | |
1da177e4 | 6759 | " share one or more physical units)\n", |
61df9d91 | 6760 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 6761 | mddev_put(mddev2); |
9744197c N |
6762 | if (signal_pending(current)) |
6763 | flush_signals(current); | |
1da177e4 LT |
6764 | schedule(); |
6765 | finish_wait(&resync_wait, &wq); | |
6766 | goto try_again; | |
6767 | } | |
6768 | finish_wait(&resync_wait, &wq); | |
6769 | } | |
6770 | } | |
6771 | } while (mddev->curr_resync < 2); | |
6772 | ||
5fd6c1dc | 6773 | j = 0; |
9d88883e | 6774 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 6775 | /* resync follows the size requested by the personality, |
57afd89f | 6776 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
6777 | */ |
6778 | max_sectors = mddev->resync_max_sectors; | |
9d88883e | 6779 | mddev->resync_mismatches = 0; |
5fd6c1dc | 6780 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
6781 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
6782 | j = mddev->resync_min; | |
6783 | else if (!mddev->bitmap) | |
5fd6c1dc | 6784 | j = mddev->recovery_cp; |
5e96ee65 | 6785 | |
ccfcc3c1 | 6786 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
58c0fed4 | 6787 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6788 | else { |
1da177e4 | 6789 | /* recovery follows the physical size of devices */ |
58c0fed4 | 6790 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 6791 | j = MaxSector; |
4e59ca7d DW |
6792 | rcu_read_lock(); |
6793 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc N |
6794 | if (rdev->raid_disk >= 0 && |
6795 | !test_bit(Faulty, &rdev->flags) && | |
6796 | !test_bit(In_sync, &rdev->flags) && | |
6797 | rdev->recovery_offset < j) | |
6798 | j = rdev->recovery_offset; | |
4e59ca7d | 6799 | rcu_read_unlock(); |
5fd6c1dc | 6800 | } |
1da177e4 | 6801 | |
61df9d91 N |
6802 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
6803 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
6804 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 6805 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
6806 | "(but not more than %d KB/sec) for %s.\n", |
6807 | speed_max(mddev), desc); | |
1da177e4 | 6808 | |
eea1bf38 | 6809 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 6810 | |
57afd89f | 6811 | io_sectors = 0; |
1da177e4 LT |
6812 | for (m = 0; m < SYNC_MARKS; m++) { |
6813 | mark[m] = jiffies; | |
57afd89f | 6814 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
6815 | } |
6816 | last_mark = 0; | |
6817 | mddev->resync_mark = mark[last_mark]; | |
6818 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
6819 | ||
6820 | /* | |
6821 | * Tune reconstruction: | |
6822 | */ | |
6823 | window = 32*(PAGE_SIZE/512); | |
6824 | printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", | |
6825 | window/2,(unsigned long long) max_sectors/2); | |
6826 | ||
6827 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
6828 | last_check = 0; |
6829 | ||
6830 | if (j>2) { | |
6831 | printk(KERN_INFO | |
61df9d91 N |
6832 | "md: resuming %s of %s from checkpoint.\n", |
6833 | desc, mdname(mddev)); | |
1da177e4 LT |
6834 | mddev->curr_resync = j; |
6835 | } | |
75d3da43 | 6836 | mddev->curr_resync_completed = j; |
1da177e4 LT |
6837 | |
6838 | while (j < max_sectors) { | |
57afd89f | 6839 | sector_t sectors; |
1da177e4 | 6840 | |
57afd89f | 6841 | skipped = 0; |
97e4f42d | 6842 | |
7a91ee1f N |
6843 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
6844 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
6845 | (mddev->curr_resync - mddev->curr_resync_completed) | |
6846 | > (max_sectors >> 4)) || | |
6847 | (j - mddev->curr_resync_completed)*2 | |
6848 | >= mddev->resync_max - mddev->curr_resync_completed | |
6849 | )) { | |
97e4f42d | 6850 | /* time to update curr_resync_completed */ |
252ac522 | 6851 | md_unplug(mddev); |
97e4f42d N |
6852 | wait_event(mddev->recovery_wait, |
6853 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 6854 | mddev->curr_resync_completed = j; |
070dc6dd | 6855 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 6856 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 6857 | } |
acb180b0 | 6858 | |
e62e58a5 N |
6859 | while (j >= mddev->resync_max && !kthread_should_stop()) { |
6860 | /* As this condition is controlled by user-space, | |
6861 | * we can block indefinitely, so use '_interruptible' | |
6862 | * to avoid triggering warnings. | |
6863 | */ | |
6864 | flush_signals(current); /* just in case */ | |
6865 | wait_event_interruptible(mddev->recovery_wait, | |
6866 | mddev->resync_max > j | |
6867 | || kthread_should_stop()); | |
6868 | } | |
acb180b0 N |
6869 | |
6870 | if (kthread_should_stop()) | |
6871 | goto interrupted; | |
6872 | ||
57afd89f | 6873 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 6874 | currspeed < speed_min(mddev)); |
57afd89f | 6875 | if (sectors == 0) { |
dfc70645 | 6876 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 LT |
6877 | goto out; |
6878 | } | |
57afd89f N |
6879 | |
6880 | if (!skipped) { /* actual IO requested */ | |
6881 | io_sectors += sectors; | |
6882 | atomic_add(sectors, &mddev->recovery_active); | |
6883 | } | |
6884 | ||
1da177e4 LT |
6885 | j += sectors; |
6886 | if (j>1) mddev->curr_resync = j; | |
ff4e8d9a | 6887 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e N |
6888 | if (last_check == 0) |
6889 | /* this is the earliers that rebuilt will be | |
6890 | * visible in /proc/mdstat | |
6891 | */ | |
6892 | md_new_event(mddev); | |
57afd89f N |
6893 | |
6894 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
6895 | continue; |
6896 | ||
57afd89f | 6897 | last_check = io_sectors; |
1da177e4 | 6898 | |
dfc70645 | 6899 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 LT |
6900 | break; |
6901 | ||
6902 | repeat: | |
6903 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
6904 | /* step marks */ | |
6905 | int next = (last_mark+1) % SYNC_MARKS; | |
6906 | ||
6907 | mddev->resync_mark = mark[next]; | |
6908 | mddev->resync_mark_cnt = mark_cnt[next]; | |
6909 | mark[next] = jiffies; | |
57afd89f | 6910 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
6911 | last_mark = next; |
6912 | } | |
6913 | ||
6914 | ||
c6207277 N |
6915 | if (kthread_should_stop()) |
6916 | goto interrupted; | |
6917 | ||
1da177e4 LT |
6918 | |
6919 | /* | |
6920 | * this loop exits only if either when we are slower than | |
6921 | * the 'hard' speed limit, or the system was IO-idle for | |
6922 | * a jiffy. | |
6923 | * the system might be non-idle CPU-wise, but we only care | |
6924 | * about not overloading the IO subsystem. (things like an | |
6925 | * e2fsck being done on the RAID array should execute fast) | |
6926 | */ | |
252ac522 | 6927 | md_unplug(mddev); |
1da177e4 LT |
6928 | cond_resched(); |
6929 | ||
57afd89f N |
6930 | currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 |
6931 | /((jiffies-mddev->resync_mark)/HZ +1) +1; | |
1da177e4 | 6932 | |
88202a0c N |
6933 | if (currspeed > speed_min(mddev)) { |
6934 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 6935 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 6936 | msleep(500); |
1da177e4 LT |
6937 | goto repeat; |
6938 | } | |
6939 | } | |
6940 | } | |
61df9d91 | 6941 | printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc); |
1da177e4 LT |
6942 | /* |
6943 | * this also signals 'finished resyncing' to md_stop | |
6944 | */ | |
6945 | out: | |
252ac522 | 6946 | md_unplug(mddev); |
1da177e4 LT |
6947 | |
6948 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); | |
6949 | ||
6950 | /* tell personality that we are finished */ | |
57afd89f | 6951 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 6952 | |
dfc70645 | 6953 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
6954 | mddev->curr_resync > 2) { |
6955 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
6956 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6957 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
6958 | printk(KERN_INFO | |
61df9d91 N |
6959 | "md: checkpointing %s of %s.\n", |
6960 | desc, mdname(mddev)); | |
5fd6c1dc N |
6961 | mddev->recovery_cp = mddev->curr_resync; |
6962 | } | |
6963 | } else | |
6964 | mddev->recovery_cp = MaxSector; | |
6965 | } else { | |
6966 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6967 | mddev->curr_resync = MaxSector; | |
4e59ca7d DW |
6968 | rcu_read_lock(); |
6969 | list_for_each_entry_rcu(rdev, &mddev->disks, same_set) | |
5fd6c1dc | 6970 | if (rdev->raid_disk >= 0 && |
70fffd0b | 6971 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
6972 | !test_bit(Faulty, &rdev->flags) && |
6973 | !test_bit(In_sync, &rdev->flags) && | |
6974 | rdev->recovery_offset < mddev->curr_resync) | |
6975 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 6976 | rcu_read_unlock(); |
5fd6c1dc | 6977 | } |
1da177e4 | 6978 | } |
17571284 | 6979 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6980 | |
1da177e4 | 6981 | skip: |
c07b70ad N |
6982 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
6983 | /* We completed so min/max setting can be forgotten if used. */ | |
6984 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6985 | mddev->resync_min = 0; | |
6986 | mddev->resync_max = MaxSector; | |
6987 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6988 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 | 6989 | mddev->curr_resync = 0; |
efa59339 N |
6990 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
6991 | mddev->curr_resync_completed = 0; | |
c6207277 | 6992 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
1da177e4 LT |
6993 | wake_up(&resync_wait); |
6994 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
6995 | md_wakeup_thread(mddev->thread); | |
c6207277 N |
6996 | return; |
6997 | ||
6998 | interrupted: | |
6999 | /* | |
7000 | * got a signal, exit. | |
7001 | */ | |
7002 | printk(KERN_INFO | |
7003 | "md: md_do_sync() got signal ... exiting\n"); | |
7004 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
7005 | goto out; | |
7006 | ||
1da177e4 | 7007 | } |
29269553 | 7008 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 LT |
7009 | |
7010 | ||
b4c4c7b8 N |
7011 | static int remove_and_add_spares(mddev_t *mddev) |
7012 | { | |
7013 | mdk_rdev_t *rdev; | |
b4c4c7b8 N |
7014 | int spares = 0; |
7015 | ||
97e4f42d N |
7016 | mddev->curr_resync_completed = 0; |
7017 | ||
159ec1fc | 7018 | list_for_each_entry(rdev, &mddev->disks, same_set) |
b4c4c7b8 | 7019 | if (rdev->raid_disk >= 0 && |
6bfe0b49 | 7020 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7021 | (test_bit(Faulty, &rdev->flags) || |
7022 | ! test_bit(In_sync, &rdev->flags)) && | |
7023 | atomic_read(&rdev->nr_pending)==0) { | |
7024 | if (mddev->pers->hot_remove_disk( | |
7025 | mddev, rdev->raid_disk)==0) { | |
7026 | char nm[20]; | |
7027 | sprintf(nm,"rd%d", rdev->raid_disk); | |
7028 | sysfs_remove_link(&mddev->kobj, nm); | |
7029 | rdev->raid_disk = -1; | |
7030 | } | |
7031 | } | |
7032 | ||
4044ba58 | 7033 | if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) { |
159ec1fc | 7034 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
dfc70645 | 7035 | if (rdev->raid_disk >= 0 && |
e5427135 DW |
7036 | !test_bit(In_sync, &rdev->flags) && |
7037 | !test_bit(Blocked, &rdev->flags)) | |
dfc70645 | 7038 | spares++; |
b4c4c7b8 N |
7039 | if (rdev->raid_disk < 0 |
7040 | && !test_bit(Faulty, &rdev->flags)) { | |
7041 | rdev->recovery_offset = 0; | |
199050ea NB |
7042 | if (mddev->pers-> |
7043 | hot_add_disk(mddev, rdev) == 0) { | |
b4c4c7b8 N |
7044 | char nm[20]; |
7045 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5e55e2f5 N |
7046 | if (sysfs_create_link(&mddev->kobj, |
7047 | &rdev->kobj, nm)) | |
00bcb4ac | 7048 | /* failure here is OK */; |
b4c4c7b8 N |
7049 | spares++; |
7050 | md_new_event(mddev); | |
93be75ff | 7051 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
b4c4c7b8 N |
7052 | } else |
7053 | break; | |
7054 | } | |
dfc70645 | 7055 | } |
b4c4c7b8 N |
7056 | } |
7057 | return spares; | |
7058 | } | |
7ebc0be7 N |
7059 | |
7060 | static void reap_sync_thread(mddev_t *mddev) | |
7061 | { | |
7062 | mdk_rdev_t *rdev; | |
7063 | ||
7064 | /* resync has finished, collect result */ | |
7065 | md_unregister_thread(mddev->sync_thread); | |
7066 | mddev->sync_thread = NULL; | |
7067 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
7068 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
7069 | /* success...*/ | |
7070 | /* activate any spares */ | |
7071 | if (mddev->pers->spare_active(mddev)) | |
7072 | sysfs_notify(&mddev->kobj, NULL, | |
7073 | "degraded"); | |
7074 | } | |
7075 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
7076 | mddev->pers->finish_reshape) | |
7077 | mddev->pers->finish_reshape(mddev); | |
7078 | md_update_sb(mddev, 1); | |
7079 | ||
7080 | /* if array is no-longer degraded, then any saved_raid_disk | |
7081 | * information must be scrapped | |
7082 | */ | |
7083 | if (!mddev->degraded) | |
7084 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
7085 | rdev->saved_raid_disk = -1; | |
7086 | ||
7087 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7088 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7089 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7090 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7091 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7092 | /* flag recovery needed just to double check */ | |
7093 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7094 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7095 | md_new_event(mddev); | |
7096 | } | |
7097 | ||
1da177e4 LT |
7098 | /* |
7099 | * This routine is regularly called by all per-raid-array threads to | |
7100 | * deal with generic issues like resync and super-block update. | |
7101 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7102 | * need this as they never do any recovery or update the superblock. | |
7103 | * | |
7104 | * It does not do any resync itself, but rather "forks" off other threads | |
7105 | * to do that as needed. | |
7106 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7107 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7108 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7109 | * and wakeups up this thread which will reap the thread and finish up. |
7110 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7111 | * | |
7112 | * The overall approach is: | |
7113 | * 1/ if the superblock needs updating, update it. | |
7114 | * 2/ If a recovery thread is running, don't do anything else. | |
7115 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7116 | * 4/ If there are any faulty devices, remove them. | |
7117 | * 5/ If array is degraded, try to add spares devices | |
7118 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7119 | */ | |
7120 | void md_check_recovery(mddev_t *mddev) | |
7121 | { | |
5f40402d | 7122 | if (mddev->bitmap) |
aa5cbd10 | 7123 | bitmap_daemon_work(mddev); |
1da177e4 LT |
7124 | |
7125 | if (mddev->ro) | |
7126 | return; | |
fca4d848 N |
7127 | |
7128 | if (signal_pending(current)) { | |
31a59e34 | 7129 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7130 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7131 | mdname(mddev)); | |
7132 | mddev->safemode = 2; | |
7133 | } | |
7134 | flush_signals(current); | |
7135 | } | |
7136 | ||
c89a8eee N |
7137 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7138 | return; | |
1da177e4 | 7139 | if ( ! ( |
126925c0 | 7140 | (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7141 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7142 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7143 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7144 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7145 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7146 | )) |
7147 | return; | |
fca4d848 | 7148 | |
df5b89b3 | 7149 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7150 | int spares = 0; |
fca4d848 | 7151 | |
c89a8eee N |
7152 | if (mddev->ro) { |
7153 | /* Only thing we do on a ro array is remove | |
7154 | * failed devices. | |
7155 | */ | |
7156 | remove_and_add_spares(mddev); | |
7157 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7158 | goto unlock; | |
7159 | } | |
7160 | ||
31a59e34 | 7161 | if (!mddev->external) { |
0fd62b86 | 7162 | int did_change = 0; |
31a59e34 N |
7163 | spin_lock_irq(&mddev->write_lock); |
7164 | if (mddev->safemode && | |
7165 | !atomic_read(&mddev->writes_pending) && | |
7166 | !mddev->in_sync && | |
7167 | mddev->recovery_cp == MaxSector) { | |
7168 | mddev->in_sync = 1; | |
0fd62b86 | 7169 | did_change = 1; |
070dc6dd | 7170 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7171 | } |
7172 | if (mddev->safemode == 1) | |
7173 | mddev->safemode = 0; | |
7174 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7175 | if (did_change) |
00bcb4ac | 7176 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7177 | } |
fca4d848 | 7178 | |
850b2b42 N |
7179 | if (mddev->flags) |
7180 | md_update_sb(mddev, 0); | |
06d91a5f | 7181 | |
1da177e4 LT |
7182 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7183 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7184 | /* resync/recovery still happening */ | |
7185 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7186 | goto unlock; | |
7187 | } | |
7188 | if (mddev->sync_thread) { | |
7ebc0be7 | 7189 | reap_sync_thread(mddev); |
1da177e4 LT |
7190 | goto unlock; |
7191 | } | |
72a23c21 NB |
7192 | /* Set RUNNING before clearing NEEDED to avoid |
7193 | * any transients in the value of "sync_action". | |
7194 | */ | |
7195 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7196 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
24dd469d N |
7197 | /* Clear some bits that don't mean anything, but |
7198 | * might be left set | |
7199 | */ | |
24dd469d N |
7200 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7201 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7202 | |
5fd6c1dc N |
7203 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
7204 | goto unlock; | |
1da177e4 LT |
7205 | /* no recovery is running. |
7206 | * remove any failed drives, then | |
7207 | * add spares if possible. | |
7208 | * Spare are also removed and re-added, to allow | |
7209 | * the personality to fail the re-add. | |
7210 | */ | |
1da177e4 | 7211 | |
b4c4c7b8 | 7212 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7213 | if (mddev->pers->check_reshape == NULL || |
7214 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 N |
7215 | /* Cannot proceed */ |
7216 | goto unlock; | |
7217 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
72a23c21 | 7218 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
b4c4c7b8 | 7219 | } else if ((spares = remove_and_add_spares(mddev))) { |
24dd469d N |
7220 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7221 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7222 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7223 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7224 | } else if (mddev->recovery_cp < MaxSector) { |
7225 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7226 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7227 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7228 | /* nothing to be done ... */ | |
1da177e4 | 7229 | goto unlock; |
24dd469d | 7230 | |
1da177e4 | 7231 | if (mddev->pers->sync_request) { |
a654b9d8 N |
7232 | if (spares && mddev->bitmap && ! mddev->bitmap->file) { |
7233 | /* We are adding a device or devices to an array | |
7234 | * which has the bitmap stored on all devices. | |
7235 | * So make sure all bitmap pages get written | |
7236 | */ | |
7237 | bitmap_write_all(mddev->bitmap); | |
7238 | } | |
1da177e4 LT |
7239 | mddev->sync_thread = md_register_thread(md_do_sync, |
7240 | mddev, | |
0da3c619 | 7241 | "resync"); |
1da177e4 LT |
7242 | if (!mddev->sync_thread) { |
7243 | printk(KERN_ERR "%s: could not start resync" | |
7244 | " thread...\n", | |
7245 | mdname(mddev)); | |
7246 | /* leave the spares where they are, it shouldn't hurt */ | |
7ebc0be7 N |
7247 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
7248 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7249 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7250 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7251 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
d7603b7e | 7252 | } else |
1da177e4 | 7253 | md_wakeup_thread(mddev->sync_thread); |
00bcb4ac | 7254 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
d7603b7e | 7255 | md_new_event(mddev); |
1da177e4 LT |
7256 | } |
7257 | unlock: | |
72a23c21 NB |
7258 | if (!mddev->sync_thread) { |
7259 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7260 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7261 | &mddev->recovery)) | |
0c3573f1 | 7262 | if (mddev->sysfs_action) |
00bcb4ac | 7263 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7264 | } |
1da177e4 LT |
7265 | mddev_unlock(mddev); |
7266 | } | |
7267 | } | |
7268 | ||
6bfe0b49 DW |
7269 | void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
7270 | { | |
00bcb4ac | 7271 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 DW |
7272 | wait_event_timeout(rdev->blocked_wait, |
7273 | !test_bit(Blocked, &rdev->flags), | |
7274 | msecs_to_jiffies(5000)); | |
7275 | rdev_dec_pending(rdev, mddev); | |
7276 | } | |
7277 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7278 | ||
75c96f85 AB |
7279 | static int md_notify_reboot(struct notifier_block *this, |
7280 | unsigned long code, void *x) | |
1da177e4 LT |
7281 | { |
7282 | struct list_head *tmp; | |
7283 | mddev_t *mddev; | |
7284 | ||
7285 | if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { | |
7286 | ||
7287 | printk(KERN_INFO "md: stopping all md devices.\n"); | |
7288 | ||
29ac4aa3 | 7289 | for_each_mddev(mddev, tmp) |
c71d4887 | 7290 | if (mddev_trylock(mddev)) { |
2b25000b N |
7291 | /* Force a switch to readonly even array |
7292 | * appears to still be in use. Hence | |
7293 | * the '100'. | |
7294 | */ | |
a4bd82d0 | 7295 | md_set_readonly(mddev, 100); |
c71d4887 NB |
7296 | mddev_unlock(mddev); |
7297 | } | |
1da177e4 LT |
7298 | /* |
7299 | * certain more exotic SCSI devices are known to be | |
7300 | * volatile wrt too early system reboots. While the | |
7301 | * right place to handle this issue is the given | |
7302 | * driver, we do want to have a safe RAID driver ... | |
7303 | */ | |
7304 | mdelay(1000*1); | |
7305 | } | |
7306 | return NOTIFY_DONE; | |
7307 | } | |
7308 | ||
75c96f85 | 7309 | static struct notifier_block md_notifier = { |
1da177e4 LT |
7310 | .notifier_call = md_notify_reboot, |
7311 | .next = NULL, | |
7312 | .priority = INT_MAX, /* before any real devices */ | |
7313 | }; | |
7314 | ||
7315 | static void md_geninit(void) | |
7316 | { | |
1da177e4 LT |
7317 | dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
7318 | ||
c7705f34 | 7319 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
7320 | } |
7321 | ||
75c96f85 | 7322 | static int __init md_init(void) |
1da177e4 | 7323 | { |
e804ac78 TH |
7324 | int ret = -ENOMEM; |
7325 | ||
7326 | md_wq = alloc_workqueue("md", WQ_RESCUER, 0); | |
7327 | if (!md_wq) | |
7328 | goto err_wq; | |
7329 | ||
7330 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
7331 | if (!md_misc_wq) | |
7332 | goto err_misc_wq; | |
7333 | ||
7334 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
7335 | goto err_md; | |
7336 | ||
7337 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
7338 | goto err_mdp; | |
7339 | mdp_major = ret; | |
7340 | ||
3dbd8c2e | 7341 | blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE, |
e8703fe1 N |
7342 | md_probe, NULL, NULL); |
7343 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
7344 | md_probe, NULL, NULL); |
7345 | ||
1da177e4 | 7346 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 7347 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
7348 | |
7349 | md_geninit(); | |
d710e138 | 7350 | return 0; |
1da177e4 | 7351 | |
e804ac78 TH |
7352 | err_mdp: |
7353 | unregister_blkdev(MD_MAJOR, "md"); | |
7354 | err_md: | |
7355 | destroy_workqueue(md_misc_wq); | |
7356 | err_misc_wq: | |
7357 | destroy_workqueue(md_wq); | |
7358 | err_wq: | |
7359 | return ret; | |
7360 | } | |
1da177e4 LT |
7361 | |
7362 | #ifndef MODULE | |
7363 | ||
7364 | /* | |
7365 | * Searches all registered partitions for autorun RAID arrays | |
7366 | * at boot time. | |
7367 | */ | |
4d936ec1 ME |
7368 | |
7369 | static LIST_HEAD(all_detected_devices); | |
7370 | struct detected_devices_node { | |
7371 | struct list_head list; | |
7372 | dev_t dev; | |
7373 | }; | |
1da177e4 LT |
7374 | |
7375 | void md_autodetect_dev(dev_t dev) | |
7376 | { | |
4d936ec1 ME |
7377 | struct detected_devices_node *node_detected_dev; |
7378 | ||
7379 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
7380 | if (node_detected_dev) { | |
7381 | node_detected_dev->dev = dev; | |
7382 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
7383 | } else { | |
7384 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
7385 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
7386 | } | |
1da177e4 LT |
7387 | } |
7388 | ||
7389 | ||
7390 | static void autostart_arrays(int part) | |
7391 | { | |
7392 | mdk_rdev_t *rdev; | |
4d936ec1 ME |
7393 | struct detected_devices_node *node_detected_dev; |
7394 | dev_t dev; | |
7395 | int i_scanned, i_passed; | |
1da177e4 | 7396 | |
4d936ec1 ME |
7397 | i_scanned = 0; |
7398 | i_passed = 0; | |
1da177e4 | 7399 | |
4d936ec1 | 7400 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 7401 | |
4d936ec1 ME |
7402 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
7403 | i_scanned++; | |
7404 | node_detected_dev = list_entry(all_detected_devices.next, | |
7405 | struct detected_devices_node, list); | |
7406 | list_del(&node_detected_dev->list); | |
7407 | dev = node_detected_dev->dev; | |
7408 | kfree(node_detected_dev); | |
df968c4e | 7409 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
7410 | if (IS_ERR(rdev)) |
7411 | continue; | |
7412 | ||
b2d444d7 | 7413 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7414 | MD_BUG(); |
7415 | continue; | |
7416 | } | |
d0fae18f | 7417 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 7418 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 7419 | i_passed++; |
1da177e4 | 7420 | } |
4d936ec1 ME |
7421 | |
7422 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
7423 | i_scanned, i_passed); | |
1da177e4 LT |
7424 | |
7425 | autorun_devices(part); | |
7426 | } | |
7427 | ||
fdee8ae4 | 7428 | #endif /* !MODULE */ |
1da177e4 LT |
7429 | |
7430 | static __exit void md_exit(void) | |
7431 | { | |
7432 | mddev_t *mddev; | |
7433 | struct list_head *tmp; | |
8ab5e4c1 | 7434 | |
3dbd8c2e | 7435 | blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS); |
e8703fe1 | 7436 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 7437 | |
3dbd8c2e | 7438 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
7439 | unregister_blkdev(mdp_major, "mdp"); |
7440 | unregister_reboot_notifier(&md_notifier); | |
7441 | unregister_sysctl_table(raid_table_header); | |
7442 | remove_proc_entry("mdstat", NULL); | |
29ac4aa3 | 7443 | for_each_mddev(mddev, tmp) { |
1da177e4 | 7444 | export_array(mddev); |
d3374825 | 7445 | mddev->hold_active = 0; |
1da177e4 | 7446 | } |
e804ac78 TH |
7447 | destroy_workqueue(md_misc_wq); |
7448 | destroy_workqueue(md_wq); | |
1da177e4 LT |
7449 | } |
7450 | ||
685784aa | 7451 | subsys_initcall(md_init); |
1da177e4 LT |
7452 | module_exit(md_exit) |
7453 | ||
f91de92e N |
7454 | static int get_ro(char *buffer, struct kernel_param *kp) |
7455 | { | |
7456 | return sprintf(buffer, "%d", start_readonly); | |
7457 | } | |
7458 | static int set_ro(const char *val, struct kernel_param *kp) | |
7459 | { | |
7460 | char *e; | |
7461 | int num = simple_strtoul(val, &e, 10); | |
7462 | if (*val && (*e == '\0' || *e == '\n')) { | |
7463 | start_readonly = num; | |
4dbcdc75 | 7464 | return 0; |
f91de92e N |
7465 | } |
7466 | return -EINVAL; | |
7467 | } | |
7468 | ||
80ca3a44 N |
7469 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
7470 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
6ff8d8ec | 7471 | |
efeb53c0 | 7472 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 7473 | |
1da177e4 LT |
7474 | EXPORT_SYMBOL(register_md_personality); |
7475 | EXPORT_SYMBOL(unregister_md_personality); | |
7476 | EXPORT_SYMBOL(md_error); | |
7477 | EXPORT_SYMBOL(md_done_sync); | |
7478 | EXPORT_SYMBOL(md_write_start); | |
7479 | EXPORT_SYMBOL(md_write_end); | |
1da177e4 LT |
7480 | EXPORT_SYMBOL(md_register_thread); |
7481 | EXPORT_SYMBOL(md_unregister_thread); | |
7482 | EXPORT_SYMBOL(md_wakeup_thread); | |
1da177e4 LT |
7483 | EXPORT_SYMBOL(md_check_recovery); |
7484 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7485 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 7486 | MODULE_ALIAS("md"); |
72008652 | 7487 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |